gas/
[binutils-gdb.git] / gas / config / tc-i386.c
1 /* i386.c -- Assemble code for the Intel 80386
2 Copyright 1989, 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
3 2000, 2001, 2002, 2003, 2004, 2005, 2006
4 Free Software Foundation, Inc.
5
6 This file is part of GAS, the GNU Assembler.
7
8 GAS is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 2, or (at your option)
11 any later version.
12
13 GAS is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
17
18 You should have received a copy of the GNU General Public License
19 along with GAS; see the file COPYING. If not, write to the Free
20 Software Foundation, 51 Franklin Street - Fifth Floor, Boston, MA
21 02110-1301, USA. */
22
23 /* Intel 80386 machine specific gas.
24 Written by Eliot Dresselhaus (eliot@mgm.mit.edu).
25 x86_64 support by Jan Hubicka (jh@suse.cz)
26 VIA PadLock support by Michal Ludvig (mludvig@suse.cz)
27 Bugs & suggestions are completely welcome. This is free software.
28 Please help us make it better. */
29
30 #include "as.h"
31 #include "safe-ctype.h"
32 #include "subsegs.h"
33 #include "dwarf2dbg.h"
34 #include "dw2gencfi.h"
35 #include "opcode/i386.h"
36 #include "elf/x86-64.h"
37
38 #ifndef REGISTER_WARNINGS
39 #define REGISTER_WARNINGS 1
40 #endif
41
42 #ifndef INFER_ADDR_PREFIX
43 #define INFER_ADDR_PREFIX 1
44 #endif
45
46 #ifndef SCALE1_WHEN_NO_INDEX
47 /* Specifying a scale factor besides 1 when there is no index is
48 futile. eg. `mov (%ebx,2),%al' does exactly the same as
49 `mov (%ebx),%al'. To slavishly follow what the programmer
50 specified, set SCALE1_WHEN_NO_INDEX to 0. */
51 #define SCALE1_WHEN_NO_INDEX 1
52 #endif
53
54 #ifndef DEFAULT_ARCH
55 #define DEFAULT_ARCH "i386"
56 #endif
57
58 #ifndef INLINE
59 #if __GNUC__ >= 2
60 #define INLINE __inline__
61 #else
62 #define INLINE
63 #endif
64 #endif
65
66 static INLINE unsigned int mode_from_disp_size PARAMS ((unsigned int));
67 static INLINE int fits_in_signed_byte PARAMS ((offsetT));
68 static INLINE int fits_in_unsigned_byte PARAMS ((offsetT));
69 static INLINE int fits_in_unsigned_word PARAMS ((offsetT));
70 static INLINE int fits_in_signed_word PARAMS ((offsetT));
71 static INLINE int fits_in_unsigned_long PARAMS ((offsetT));
72 static INLINE int fits_in_signed_long PARAMS ((offsetT));
73 static int smallest_imm_type PARAMS ((offsetT));
74 static offsetT offset_in_range PARAMS ((offsetT, int));
75 static int add_prefix PARAMS ((unsigned int));
76 static void set_code_flag PARAMS ((int));
77 static void set_16bit_gcc_code_flag PARAMS ((int));
78 static void set_intel_syntax PARAMS ((int));
79 static void set_cpu_arch PARAMS ((int));
80 #ifdef TE_PE
81 static void pe_directive_secrel PARAMS ((int));
82 #endif
83 static void signed_cons PARAMS ((int));
84 static char *output_invalid PARAMS ((int c));
85 static int i386_operand PARAMS ((char *operand_string));
86 static int i386_intel_operand PARAMS ((char *operand_string, int got_a_float));
87 static const reg_entry *parse_register PARAMS ((char *reg_string,
88 char **end_op));
89 static char *parse_insn PARAMS ((char *, char *));
90 static char *parse_operands PARAMS ((char *, const char *));
91 static void swap_operands PARAMS ((void));
92 static void swap_imm_operands PARAMS ((void));
93 static void optimize_imm PARAMS ((void));
94 static void optimize_disp PARAMS ((void));
95 static int match_template PARAMS ((void));
96 static int check_string PARAMS ((void));
97 static int process_suffix PARAMS ((void));
98 static int check_byte_reg PARAMS ((void));
99 static int check_long_reg PARAMS ((void));
100 static int check_qword_reg PARAMS ((void));
101 static int check_word_reg PARAMS ((void));
102 static int finalize_imm PARAMS ((void));
103 static int process_operands PARAMS ((void));
104 static const seg_entry *build_modrm_byte PARAMS ((void));
105 static void output_insn PARAMS ((void));
106 static void output_branch PARAMS ((void));
107 static void output_jump PARAMS ((void));
108 static void output_interseg_jump PARAMS ((void));
109 static void output_imm PARAMS ((fragS *insn_start_frag,
110 offsetT insn_start_off));
111 static void output_disp PARAMS ((fragS *insn_start_frag,
112 offsetT insn_start_off));
113 #ifndef I386COFF
114 static void s_bss PARAMS ((int));
115 #endif
116 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
117 static void handle_large_common (int small ATTRIBUTE_UNUSED);
118 #endif
119
120 static const char *default_arch = DEFAULT_ARCH;
121
122 /* 'md_assemble ()' gathers together information and puts it into a
123 i386_insn. */
124
125 union i386_op
126 {
127 expressionS *disps;
128 expressionS *imms;
129 const reg_entry *regs;
130 };
131
132 struct _i386_insn
133 {
134 /* TM holds the template for the insn were currently assembling. */
135 template tm;
136
137 /* SUFFIX holds the instruction mnemonic suffix if given.
138 (e.g. 'l' for 'movl') */
139 char suffix;
140
141 /* OPERANDS gives the number of given operands. */
142 unsigned int operands;
143
144 /* REG_OPERANDS, DISP_OPERANDS, MEM_OPERANDS, IMM_OPERANDS give the number
145 of given register, displacement, memory operands and immediate
146 operands. */
147 unsigned int reg_operands, disp_operands, mem_operands, imm_operands;
148
149 /* TYPES [i] is the type (see above #defines) which tells us how to
150 use OP[i] for the corresponding operand. */
151 unsigned int types[MAX_OPERANDS];
152
153 /* Displacement expression, immediate expression, or register for each
154 operand. */
155 union i386_op op[MAX_OPERANDS];
156
157 /* Flags for operands. */
158 unsigned int flags[MAX_OPERANDS];
159 #define Operand_PCrel 1
160
161 /* Relocation type for operand */
162 enum bfd_reloc_code_real reloc[MAX_OPERANDS];
163
164 /* BASE_REG, INDEX_REG, and LOG2_SCALE_FACTOR are used to encode
165 the base index byte below. */
166 const reg_entry *base_reg;
167 const reg_entry *index_reg;
168 unsigned int log2_scale_factor;
169
170 /* SEG gives the seg_entries of this insn. They are zero unless
171 explicit segment overrides are given. */
172 const seg_entry *seg[2];
173
174 /* PREFIX holds all the given prefix opcodes (usually null).
175 PREFIXES is the number of prefix opcodes. */
176 unsigned int prefixes;
177 unsigned char prefix[MAX_PREFIXES];
178
179 /* RM and SIB are the modrm byte and the sib byte where the
180 addressing modes of this insn are encoded. */
181
182 modrm_byte rm;
183 rex_byte rex;
184 sib_byte sib;
185 };
186
187 typedef struct _i386_insn i386_insn;
188
189 /* List of chars besides those in app.c:symbol_chars that can start an
190 operand. Used to prevent the scrubber eating vital white-space. */
191 const char extra_symbol_chars[] = "*%-(["
192 #ifdef LEX_AT
193 "@"
194 #endif
195 #ifdef LEX_QM
196 "?"
197 #endif
198 ;
199
200 #if (defined (TE_I386AIX) \
201 || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) \
202 && !defined (TE_GNU) \
203 && !defined (TE_LINUX) \
204 && !defined (TE_NETWARE) \
205 && !defined (TE_FreeBSD) \
206 && !defined (TE_NetBSD)))
207 /* This array holds the chars that always start a comment. If the
208 pre-processor is disabled, these aren't very useful. The option
209 --divide will remove '/' from this list. */
210 const char *i386_comment_chars = "#/";
211 #define SVR4_COMMENT_CHARS 1
212 #define PREFIX_SEPARATOR '\\'
213
214 #else
215 const char *i386_comment_chars = "#";
216 #define PREFIX_SEPARATOR '/'
217 #endif
218
219 /* This array holds the chars that only start a comment at the beginning of
220 a line. If the line seems to have the form '# 123 filename'
221 .line and .file directives will appear in the pre-processed output.
222 Note that input_file.c hand checks for '#' at the beginning of the
223 first line of the input file. This is because the compiler outputs
224 #NO_APP at the beginning of its output.
225 Also note that comments started like this one will always work if
226 '/' isn't otherwise defined. */
227 const char line_comment_chars[] = "#/";
228
229 const char line_separator_chars[] = ";";
230
231 /* Chars that can be used to separate mant from exp in floating point
232 nums. */
233 const char EXP_CHARS[] = "eE";
234
235 /* Chars that mean this number is a floating point constant
236 As in 0f12.456
237 or 0d1.2345e12. */
238 const char FLT_CHARS[] = "fFdDxX";
239
240 /* Tables for lexical analysis. */
241 static char mnemonic_chars[256];
242 static char register_chars[256];
243 static char operand_chars[256];
244 static char identifier_chars[256];
245 static char digit_chars[256];
246
247 /* Lexical macros. */
248 #define is_mnemonic_char(x) (mnemonic_chars[(unsigned char) x])
249 #define is_operand_char(x) (operand_chars[(unsigned char) x])
250 #define is_register_char(x) (register_chars[(unsigned char) x])
251 #define is_space_char(x) ((x) == ' ')
252 #define is_identifier_char(x) (identifier_chars[(unsigned char) x])
253 #define is_digit_char(x) (digit_chars[(unsigned char) x])
254
255 /* All non-digit non-letter characters that may occur in an operand. */
256 static char operand_special_chars[] = "%$-+(,)*._~/<>|&^!:[@]";
257
258 /* md_assemble() always leaves the strings it's passed unaltered. To
259 effect this we maintain a stack of saved characters that we've smashed
260 with '\0's (indicating end of strings for various sub-fields of the
261 assembler instruction). */
262 static char save_stack[32];
263 static char *save_stack_p;
264 #define END_STRING_AND_SAVE(s) \
265 do { *save_stack_p++ = *(s); *(s) = '\0'; } while (0)
266 #define RESTORE_END_STRING(s) \
267 do { *(s) = *--save_stack_p; } while (0)
268
269 /* The instruction we're assembling. */
270 static i386_insn i;
271
272 /* Possible templates for current insn. */
273 static const templates *current_templates;
274
275 /* Per instruction expressionS buffers: 2 displacements & 2 immediate max. */
276 static expressionS disp_expressions[2], im_expressions[2];
277
278 /* Current operand we are working on. */
279 static int this_operand;
280
281 /* We support four different modes. FLAG_CODE variable is used to distinguish
282 these. */
283
284 enum flag_code {
285 CODE_32BIT,
286 CODE_16BIT,
287 CODE_64BIT };
288 #define NUM_FLAG_CODE ((int) CODE_64BIT + 1)
289
290 static enum flag_code flag_code;
291 static unsigned int object_64bit;
292 static int use_rela_relocations = 0;
293
294 /* The names used to print error messages. */
295 static const char *flag_code_names[] =
296 {
297 "32",
298 "16",
299 "64"
300 };
301
302 /* 1 for intel syntax,
303 0 if att syntax. */
304 static int intel_syntax = 0;
305
306 /* 1 if register prefix % not required. */
307 static int allow_naked_reg = 0;
308
309 /* Used in 16 bit gcc mode to add an l suffix to call, ret, enter,
310 leave, push, and pop instructions so that gcc has the same stack
311 frame as in 32 bit mode. */
312 static char stackop_size = '\0';
313
314 /* Non-zero to optimize code alignment. */
315 int optimize_align_code = 1;
316
317 /* Non-zero to quieten some warnings. */
318 static int quiet_warnings = 0;
319
320 /* CPU name. */
321 static const char *cpu_arch_name = NULL;
322 static const char *cpu_sub_arch_name = NULL;
323
324 /* CPU feature flags. */
325 static unsigned int cpu_arch_flags = CpuUnknownFlags | CpuNo64;
326
327 /* If we have selected a cpu we are generating instructions for. */
328 static int cpu_arch_tune_set = 0;
329
330 /* Cpu we are generating instructions for. */
331 static enum processor_type cpu_arch_tune = PROCESSOR_UNKNOWN;
332
333 /* CPU feature flags of cpu we are generating instructions for. */
334 static unsigned int cpu_arch_tune_flags = 0;
335
336 /* CPU instruction set architecture used. */
337 static enum processor_type cpu_arch_isa = PROCESSOR_UNKNOWN;
338
339 /* CPU feature flags of instruction set architecture used. */
340 static unsigned int cpu_arch_isa_flags = 0;
341
342 /* If set, conditional jumps are not automatically promoted to handle
343 larger than a byte offset. */
344 static unsigned int no_cond_jump_promotion = 0;
345
346 /* Pre-defined "_GLOBAL_OFFSET_TABLE_". */
347 static symbolS *GOT_symbol;
348
349 /* The dwarf2 return column, adjusted for 32 or 64 bit. */
350 unsigned int x86_dwarf2_return_column;
351
352 /* The dwarf2 data alignment, adjusted for 32 or 64 bit. */
353 int x86_cie_data_alignment;
354
355 /* Interface to relax_segment.
356 There are 3 major relax states for 386 jump insns because the
357 different types of jumps add different sizes to frags when we're
358 figuring out what sort of jump to choose to reach a given label. */
359
360 /* Types. */
361 #define UNCOND_JUMP 0
362 #define COND_JUMP 1
363 #define COND_JUMP86 2
364
365 /* Sizes. */
366 #define CODE16 1
367 #define SMALL 0
368 #define SMALL16 (SMALL | CODE16)
369 #define BIG 2
370 #define BIG16 (BIG | CODE16)
371
372 #ifndef INLINE
373 #ifdef __GNUC__
374 #define INLINE __inline__
375 #else
376 #define INLINE
377 #endif
378 #endif
379
380 #define ENCODE_RELAX_STATE(type, size) \
381 ((relax_substateT) (((type) << 2) | (size)))
382 #define TYPE_FROM_RELAX_STATE(s) \
383 ((s) >> 2)
384 #define DISP_SIZE_FROM_RELAX_STATE(s) \
385 ((((s) & 3) == BIG ? 4 : (((s) & 3) == BIG16 ? 2 : 1)))
386
387 /* This table is used by relax_frag to promote short jumps to long
388 ones where necessary. SMALL (short) jumps may be promoted to BIG
389 (32 bit long) ones, and SMALL16 jumps to BIG16 (16 bit long). We
390 don't allow a short jump in a 32 bit code segment to be promoted to
391 a 16 bit offset jump because it's slower (requires data size
392 prefix), and doesn't work, unless the destination is in the bottom
393 64k of the code segment (The top 16 bits of eip are zeroed). */
394
395 const relax_typeS md_relax_table[] =
396 {
397 /* The fields are:
398 1) most positive reach of this state,
399 2) most negative reach of this state,
400 3) how many bytes this mode will have in the variable part of the frag
401 4) which index into the table to try if we can't fit into this one. */
402
403 /* UNCOND_JUMP states. */
404 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG)},
405 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16)},
406 /* dword jmp adds 4 bytes to frag:
407 0 extra opcode bytes, 4 displacement bytes. */
408 {0, 0, 4, 0},
409 /* word jmp adds 2 byte2 to frag:
410 0 extra opcode bytes, 2 displacement bytes. */
411 {0, 0, 2, 0},
412
413 /* COND_JUMP states. */
414 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG)},
415 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG16)},
416 /* dword conditionals adds 5 bytes to frag:
417 1 extra opcode byte, 4 displacement bytes. */
418 {0, 0, 5, 0},
419 /* word conditionals add 3 bytes to frag:
420 1 extra opcode byte, 2 displacement bytes. */
421 {0, 0, 3, 0},
422
423 /* COND_JUMP86 states. */
424 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG)},
425 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG16)},
426 /* dword conditionals adds 5 bytes to frag:
427 1 extra opcode byte, 4 displacement bytes. */
428 {0, 0, 5, 0},
429 /* word conditionals add 4 bytes to frag:
430 1 displacement byte and a 3 byte long branch insn. */
431 {0, 0, 4, 0}
432 };
433
434 static const arch_entry cpu_arch[] =
435 {
436 {"generic32", PROCESSOR_GENERIC32,
437 Cpu186|Cpu286|Cpu386},
438 {"generic64", PROCESSOR_GENERIC64,
439 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
440 |CpuMMX2|CpuSSE|CpuSSE2},
441 {"i8086", PROCESSOR_UNKNOWN,
442 0},
443 {"i186", PROCESSOR_UNKNOWN,
444 Cpu186},
445 {"i286", PROCESSOR_UNKNOWN,
446 Cpu186|Cpu286},
447 {"i386", PROCESSOR_GENERIC32,
448 Cpu186|Cpu286|Cpu386},
449 {"i486", PROCESSOR_I486,
450 Cpu186|Cpu286|Cpu386|Cpu486},
451 {"i586", PROCESSOR_PENTIUM,
452 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586},
453 {"i686", PROCESSOR_PENTIUMPRO,
454 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686},
455 {"pentium", PROCESSOR_PENTIUM,
456 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586},
457 {"pentiumpro",PROCESSOR_PENTIUMPRO,
458 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686},
459 {"pentiumii", PROCESSOR_PENTIUMPRO,
460 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX},
461 {"pentiumiii",PROCESSOR_PENTIUMPRO,
462 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuMMX2|CpuSSE},
463 {"pentium4", PROCESSOR_PENTIUM4,
464 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
465 |CpuMMX2|CpuSSE|CpuSSE2},
466 {"prescott", PROCESSOR_NOCONA,
467 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
468 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
469 {"nocona", PROCESSOR_NOCONA,
470 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
471 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
472 {"yonah", PROCESSOR_CORE,
473 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
474 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
475 {"core", PROCESSOR_CORE,
476 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
477 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
478 {"merom", PROCESSOR_CORE2,
479 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
480 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSSE3},
481 {"core2", PROCESSOR_CORE2,
482 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
483 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSSE3},
484 {"k6", PROCESSOR_K6,
485 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX},
486 {"k6_2", PROCESSOR_K6,
487 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX|Cpu3dnow},
488 {"athlon", PROCESSOR_ATHLON,
489 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6
490 |CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA},
491 {"sledgehammer", PROCESSOR_K8,
492 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6
493 |CpuSledgehammer|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2},
494 {"opteron", PROCESSOR_K8,
495 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6
496 |CpuSledgehammer|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2},
497 {"k8", PROCESSOR_K8,
498 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6
499 |CpuSledgehammer|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2},
500 {"amdfam10", PROCESSOR_AMDFAM10,
501 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuSledgehammer
502 |CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2|CpuSSE3|CpuSSE4a
503 |CpuABM},
504 {".mmx", PROCESSOR_UNKNOWN,
505 CpuMMX},
506 {".sse", PROCESSOR_UNKNOWN,
507 CpuMMX|CpuMMX2|CpuSSE},
508 {".sse2", PROCESSOR_UNKNOWN,
509 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2},
510 {".sse3", PROCESSOR_UNKNOWN,
511 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
512 {".ssse3", PROCESSOR_UNKNOWN,
513 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSSE3},
514 {".3dnow", PROCESSOR_UNKNOWN,
515 CpuMMX|Cpu3dnow},
516 {".3dnowa", PROCESSOR_UNKNOWN,
517 CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA},
518 {".padlock", PROCESSOR_UNKNOWN,
519 CpuPadLock},
520 {".pacifica", PROCESSOR_UNKNOWN,
521 CpuSVME},
522 {".svme", PROCESSOR_UNKNOWN,
523 CpuSVME},
524 {".sse4a", PROCESSOR_UNKNOWN,
525 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSE4a},
526 {".abm", PROCESSOR_UNKNOWN,
527 CpuABM}
528 };
529
530 const pseudo_typeS md_pseudo_table[] =
531 {
532 #if !defined(OBJ_AOUT) && !defined(USE_ALIGN_PTWO)
533 {"align", s_align_bytes, 0},
534 #else
535 {"align", s_align_ptwo, 0},
536 #endif
537 {"arch", set_cpu_arch, 0},
538 #ifndef I386COFF
539 {"bss", s_bss, 0},
540 #endif
541 {"ffloat", float_cons, 'f'},
542 {"dfloat", float_cons, 'd'},
543 {"tfloat", float_cons, 'x'},
544 {"value", cons, 2},
545 {"slong", signed_cons, 4},
546 {"noopt", s_ignore, 0},
547 {"optim", s_ignore, 0},
548 {"code16gcc", set_16bit_gcc_code_flag, CODE_16BIT},
549 {"code16", set_code_flag, CODE_16BIT},
550 {"code32", set_code_flag, CODE_32BIT},
551 {"code64", set_code_flag, CODE_64BIT},
552 {"intel_syntax", set_intel_syntax, 1},
553 {"att_syntax", set_intel_syntax, 0},
554 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
555 {"largecomm", handle_large_common, 0},
556 #else
557 {"file", (void (*) PARAMS ((int))) dwarf2_directive_file, 0},
558 {"loc", dwarf2_directive_loc, 0},
559 {"loc_mark_labels", dwarf2_directive_loc_mark_labels, 0},
560 #endif
561 #ifdef TE_PE
562 {"secrel32", pe_directive_secrel, 0},
563 #endif
564 {0, 0, 0}
565 };
566
567 /* For interface with expression (). */
568 extern char *input_line_pointer;
569
570 /* Hash table for instruction mnemonic lookup. */
571 static struct hash_control *op_hash;
572
573 /* Hash table for register lookup. */
574 static struct hash_control *reg_hash;
575 \f
576 void
577 i386_align_code (fragP, count)
578 fragS *fragP;
579 int count;
580 {
581 /* Various efficient no-op patterns for aligning code labels.
582 Note: Don't try to assemble the instructions in the comments.
583 0L and 0w are not legal. */
584 static const char f32_1[] =
585 {0x90}; /* nop */
586 static const char f32_2[] =
587 {0x66,0x90}; /* xchg %ax,%ax */
588 static const char f32_3[] =
589 {0x8d,0x76,0x00}; /* leal 0(%esi),%esi */
590 static const char f32_4[] =
591 {0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
592 static const char f32_5[] =
593 {0x90, /* nop */
594 0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
595 static const char f32_6[] =
596 {0x8d,0xb6,0x00,0x00,0x00,0x00}; /* leal 0L(%esi),%esi */
597 static const char f32_7[] =
598 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
599 static const char f32_8[] =
600 {0x90, /* nop */
601 0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
602 static const char f32_9[] =
603 {0x89,0xf6, /* movl %esi,%esi */
604 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
605 static const char f32_10[] =
606 {0x8d,0x76,0x00, /* leal 0(%esi),%esi */
607 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
608 static const char f32_11[] =
609 {0x8d,0x74,0x26,0x00, /* leal 0(%esi,1),%esi */
610 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
611 static const char f32_12[] =
612 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
613 0x8d,0xbf,0x00,0x00,0x00,0x00}; /* leal 0L(%edi),%edi */
614 static const char f32_13[] =
615 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
616 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
617 static const char f32_14[] =
618 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00, /* leal 0L(%esi,1),%esi */
619 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
620 static const char f32_15[] =
621 {0xeb,0x0d,0x90,0x90,0x90,0x90,0x90, /* jmp .+15; lotsa nops */
622 0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90};
623 static const char f16_3[] =
624 {0x8d,0x74,0x00}; /* lea 0(%esi),%esi */
625 static const char f16_4[] =
626 {0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
627 static const char f16_5[] =
628 {0x90, /* nop */
629 0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
630 static const char f16_6[] =
631 {0x89,0xf6, /* mov %si,%si */
632 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
633 static const char f16_7[] =
634 {0x8d,0x74,0x00, /* lea 0(%si),%si */
635 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
636 static const char f16_8[] =
637 {0x8d,0xb4,0x00,0x00, /* lea 0w(%si),%si */
638 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
639 static const char *const f32_patt[] = {
640 f32_1, f32_2, f32_3, f32_4, f32_5, f32_6, f32_7, f32_8,
641 f32_9, f32_10, f32_11, f32_12, f32_13, f32_14, f32_15
642 };
643 static const char *const f16_patt[] = {
644 f32_1, f32_2, f16_3, f16_4, f16_5, f16_6, f16_7, f16_8,
645 f32_15, f32_15, f32_15, f32_15, f32_15, f32_15, f32_15
646 };
647 /* nopl (%[re]ax) */
648 static const char alt_3[] =
649 {0x0f,0x1f,0x00};
650 /* nopl 0(%[re]ax) */
651 static const char alt_4[] =
652 {0x0f,0x1f,0x40,0x00};
653 /* nopl 0(%[re]ax,%[re]ax,1) */
654 static const char alt_5[] =
655 {0x0f,0x1f,0x44,0x00,0x00};
656 /* nopw 0(%[re]ax,%[re]ax,1) */
657 static const char alt_6[] =
658 {0x66,0x0f,0x1f,0x44,0x00,0x00};
659 /* nopl 0L(%[re]ax) */
660 static const char alt_7[] =
661 {0x0f,0x1f,0x80,0x00,0x00,0x00,0x00};
662 /* nopl 0L(%[re]ax,%[re]ax,1) */
663 static const char alt_8[] =
664 {0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
665 /* nopw 0L(%[re]ax,%[re]ax,1) */
666 static const char alt_9[] =
667 {0x66,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
668 /* nopw %cs:0L(%[re]ax,%[re]ax,1) */
669 static const char alt_10[] =
670 {0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
671 /* data16
672 nopw %cs:0L(%[re]ax,%[re]ax,1) */
673 static const char alt_long_11[] =
674 {0x66,
675 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
676 /* data16
677 data16
678 nopw %cs:0L(%[re]ax,%[re]ax,1) */
679 static const char alt_long_12[] =
680 {0x66,
681 0x66,
682 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
683 /* data16
684 data16
685 data16
686 nopw %cs:0L(%[re]ax,%[re]ax,1) */
687 static const char alt_long_13[] =
688 {0x66,
689 0x66,
690 0x66,
691 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
692 /* data16
693 data16
694 data16
695 data16
696 nopw %cs:0L(%[re]ax,%[re]ax,1) */
697 static const char alt_long_14[] =
698 {0x66,
699 0x66,
700 0x66,
701 0x66,
702 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
703 /* data16
704 data16
705 data16
706 data16
707 data16
708 nopw %cs:0L(%[re]ax,%[re]ax,1) */
709 static const char alt_long_15[] =
710 {0x66,
711 0x66,
712 0x66,
713 0x66,
714 0x66,
715 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
716 /* nopl 0(%[re]ax,%[re]ax,1)
717 nopw 0(%[re]ax,%[re]ax,1) */
718 static const char alt_short_11[] =
719 {0x0f,0x1f,0x44,0x00,0x00,
720 0x66,0x0f,0x1f,0x44,0x00,0x00};
721 /* nopw 0(%[re]ax,%[re]ax,1)
722 nopw 0(%[re]ax,%[re]ax,1) */
723 static const char alt_short_12[] =
724 {0x66,0x0f,0x1f,0x44,0x00,0x00,
725 0x66,0x0f,0x1f,0x44,0x00,0x00};
726 /* nopw 0(%[re]ax,%[re]ax,1)
727 nopl 0L(%[re]ax) */
728 static const char alt_short_13[] =
729 {0x66,0x0f,0x1f,0x44,0x00,0x00,
730 0x0f,0x1f,0x80,0x00,0x00,0x00,0x00};
731 /* nopl 0L(%[re]ax)
732 nopl 0L(%[re]ax) */
733 static const char alt_short_14[] =
734 {0x0f,0x1f,0x80,0x00,0x00,0x00,0x00,
735 0x0f,0x1f,0x80,0x00,0x00,0x00,0x00};
736 /* nopl 0L(%[re]ax)
737 nopl 0L(%[re]ax,%[re]ax,1) */
738 static const char alt_short_15[] =
739 {0x0f,0x1f,0x80,0x00,0x00,0x00,0x00,
740 0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
741 static const char *const alt_short_patt[] = {
742 f32_1, f32_2, alt_3, alt_4, alt_5, alt_6, alt_7, alt_8,
743 alt_9, alt_10, alt_short_11, alt_short_12, alt_short_13,
744 alt_short_14, alt_short_15
745 };
746 static const char *const alt_long_patt[] = {
747 f32_1, f32_2, alt_3, alt_4, alt_5, alt_6, alt_7, alt_8,
748 alt_9, alt_10, alt_long_11, alt_long_12, alt_long_13,
749 alt_long_14, alt_long_15
750 };
751
752 if (count <= 0 || count > 15)
753 return;
754
755 /* We need to decide which NOP sequence to use for 32bit and
756 64bit. When -mtune= is used:
757
758 1. For PROCESSOR_I486, PROCESSOR_PENTIUM and PROCESSOR_GENERIC32,
759 f32_patt will be used.
760 2. For PROCESSOR_K8 and PROCESSOR_AMDFAM10 in 64bit, NOPs with 0x66 prefix will be used.
761 3. For PROCESSOR_CORE2, alt_long_patt will be used.
762 4. For PROCESSOR_PENTIUMPRO, PROCESSOR_PENTIUM4, PROCESSOR_NOCONA,
763 PROCESSOR_CORE, PROCESSOR_CORE2, PROCESSOR_K6, PROCESSOR_ATHLON
764 and PROCESSOR_GENERIC64, alt_short_patt will be used.
765
766 When -mtune= isn't used, alt_short_patt will be used if
767 cpu_arch_isa_flags has Cpu686. Otherwise, f32_patt will be used.
768
769 When -march= or .arch is used, we can't use anything beyond
770 cpu_arch_isa_flags. */
771
772 if (flag_code == CODE_16BIT)
773 {
774 memcpy (fragP->fr_literal + fragP->fr_fix,
775 f16_patt[count - 1], count);
776 if (count > 8)
777 /* Adjust jump offset. */
778 fragP->fr_literal[fragP->fr_fix + 1] = count - 2;
779 }
780 else if (flag_code == CODE_64BIT && cpu_arch_tune == PROCESSOR_K8)
781 {
782 int i;
783 int nnops = (count + 3) / 4;
784 int len = count / nnops;
785 int remains = count - nnops * len;
786 int pos = 0;
787
788 /* The recommended way to pad 64bit code is to use NOPs preceded
789 by maximally four 0x66 prefixes. Balance the size of nops. */
790 for (i = 0; i < remains; i++)
791 {
792 memset (fragP->fr_literal + fragP->fr_fix + pos, 0x66, len);
793 fragP->fr_literal[fragP->fr_fix + pos + len] = 0x90;
794 pos += len + 1;
795 }
796 for (; i < nnops; i++)
797 {
798 memset (fragP->fr_literal + fragP->fr_fix + pos, 0x66, len - 1);
799 fragP->fr_literal[fragP->fr_fix + pos + len - 1] = 0x90;
800 pos += len;
801 }
802 }
803 else
804 {
805 const char *const *patt = NULL;
806
807 if (cpu_arch_isa == PROCESSOR_UNKNOWN)
808 {
809 /* PROCESSOR_UNKNOWN means that all ISAs may be used. */
810 switch (cpu_arch_tune)
811 {
812 case PROCESSOR_UNKNOWN:
813 /* We use cpu_arch_isa_flags to check if we SHOULD
814 optimize for Cpu686. */
815 if ((cpu_arch_isa_flags & Cpu686) != 0)
816 patt = alt_short_patt;
817 else
818 patt = f32_patt;
819 break;
820 case PROCESSOR_CORE2:
821 patt = alt_long_patt;
822 break;
823 case PROCESSOR_PENTIUMPRO:
824 case PROCESSOR_PENTIUM4:
825 case PROCESSOR_NOCONA:
826 case PROCESSOR_CORE:
827 case PROCESSOR_K6:
828 case PROCESSOR_ATHLON:
829 case PROCESSOR_K8:
830 case PROCESSOR_GENERIC64:
831 case PROCESSOR_AMDFAM10:
832 patt = alt_short_patt;
833 break;
834 case PROCESSOR_I486:
835 case PROCESSOR_PENTIUM:
836 case PROCESSOR_GENERIC32:
837 patt = f32_patt;
838 break;
839 }
840 }
841 else
842 {
843 switch (cpu_arch_tune)
844 {
845 case PROCESSOR_UNKNOWN:
846 /* When cpu_arch_isa is net, cpu_arch_tune shouldn't be
847 PROCESSOR_UNKNOWN. */
848 abort ();
849 break;
850
851 case PROCESSOR_I486:
852 case PROCESSOR_PENTIUM:
853 case PROCESSOR_PENTIUMPRO:
854 case PROCESSOR_PENTIUM4:
855 case PROCESSOR_NOCONA:
856 case PROCESSOR_CORE:
857 case PROCESSOR_K6:
858 case PROCESSOR_ATHLON:
859 case PROCESSOR_K8:
860 case PROCESSOR_AMDFAM10:
861 case PROCESSOR_GENERIC32:
862 /* We use cpu_arch_isa_flags to check if we CAN optimize
863 for Cpu686. */
864 if ((cpu_arch_isa_flags & Cpu686) != 0)
865 patt = alt_short_patt;
866 else
867 patt = f32_patt;
868 break;
869 case PROCESSOR_CORE2:
870 if ((cpu_arch_isa_flags & Cpu686) != 0)
871 patt = alt_long_patt;
872 else
873 patt = f32_patt;
874 break;
875 case PROCESSOR_GENERIC64:
876 patt = alt_short_patt;
877 break;
878 }
879 }
880
881 memcpy (fragP->fr_literal + fragP->fr_fix,
882 patt[count - 1], count);
883 }
884 fragP->fr_var = count;
885 }
886
887 static INLINE unsigned int
888 mode_from_disp_size (t)
889 unsigned int t;
890 {
891 return (t & Disp8) ? 1 : (t & (Disp16 | Disp32 | Disp32S)) ? 2 : 0;
892 }
893
894 static INLINE int
895 fits_in_signed_byte (num)
896 offsetT num;
897 {
898 return (num >= -128) && (num <= 127);
899 }
900
901 static INLINE int
902 fits_in_unsigned_byte (num)
903 offsetT num;
904 {
905 return (num & 0xff) == num;
906 }
907
908 static INLINE int
909 fits_in_unsigned_word (num)
910 offsetT num;
911 {
912 return (num & 0xffff) == num;
913 }
914
915 static INLINE int
916 fits_in_signed_word (num)
917 offsetT num;
918 {
919 return (-32768 <= num) && (num <= 32767);
920 }
921 static INLINE int
922 fits_in_signed_long (num)
923 offsetT num ATTRIBUTE_UNUSED;
924 {
925 #ifndef BFD64
926 return 1;
927 #else
928 return (!(((offsetT) -1 << 31) & num)
929 || (((offsetT) -1 << 31) & num) == ((offsetT) -1 << 31));
930 #endif
931 } /* fits_in_signed_long() */
932 static INLINE int
933 fits_in_unsigned_long (num)
934 offsetT num ATTRIBUTE_UNUSED;
935 {
936 #ifndef BFD64
937 return 1;
938 #else
939 return (num & (((offsetT) 2 << 31) - 1)) == num;
940 #endif
941 } /* fits_in_unsigned_long() */
942
943 static int
944 smallest_imm_type (num)
945 offsetT num;
946 {
947 if (cpu_arch_flags != (Cpu186 | Cpu286 | Cpu386 | Cpu486 | CpuNo64))
948 {
949 /* This code is disabled on the 486 because all the Imm1 forms
950 in the opcode table are slower on the i486. They're the
951 versions with the implicitly specified single-position
952 displacement, which has another syntax if you really want to
953 use that form. */
954 if (num == 1)
955 return Imm1 | Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64;
956 }
957 return (fits_in_signed_byte (num)
958 ? (Imm8S | Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
959 : fits_in_unsigned_byte (num)
960 ? (Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
961 : (fits_in_signed_word (num) || fits_in_unsigned_word (num))
962 ? (Imm16 | Imm32 | Imm32S | Imm64)
963 : fits_in_signed_long (num)
964 ? (Imm32 | Imm32S | Imm64)
965 : fits_in_unsigned_long (num)
966 ? (Imm32 | Imm64)
967 : Imm64);
968 }
969
970 static offsetT
971 offset_in_range (val, size)
972 offsetT val;
973 int size;
974 {
975 addressT mask;
976
977 switch (size)
978 {
979 case 1: mask = ((addressT) 1 << 8) - 1; break;
980 case 2: mask = ((addressT) 1 << 16) - 1; break;
981 case 4: mask = ((addressT) 2 << 31) - 1; break;
982 #ifdef BFD64
983 case 8: mask = ((addressT) 2 << 63) - 1; break;
984 #endif
985 default: abort ();
986 }
987
988 /* If BFD64, sign extend val. */
989 if (!use_rela_relocations)
990 if ((val & ~(((addressT) 2 << 31) - 1)) == 0)
991 val = (val ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
992
993 if ((val & ~mask) != 0 && (val & ~mask) != ~mask)
994 {
995 char buf1[40], buf2[40];
996
997 sprint_value (buf1, val);
998 sprint_value (buf2, val & mask);
999 as_warn (_("%s shortened to %s"), buf1, buf2);
1000 }
1001 return val & mask;
1002 }
1003
1004 /* Returns 0 if attempting to add a prefix where one from the same
1005 class already exists, 1 if non rep/repne added, 2 if rep/repne
1006 added. */
1007 static int
1008 add_prefix (prefix)
1009 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_MODE64)
1018 || ((i.prefix[REX_PREFIX] & (REX_EXTX | REX_EXTY | REX_EXTZ))
1019 && (prefix & (REX_EXTX | REX_EXTY | REX_EXTZ))))
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 (value)
1077 int value;
1078 {
1079 flag_code = value;
1080 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
1081 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
1082 if (value == CODE_64BIT && !(cpu_arch_flags & CpuSledgehammer))
1083 {
1084 as_bad (_("64bit mode not supported on this CPU."));
1085 }
1086 if (value == CODE_32BIT && !(cpu_arch_flags & Cpu386))
1087 {
1088 as_bad (_("32bit mode not supported on this CPU."));
1089 }
1090 stackop_size = '\0';
1091 }
1092
1093 static void
1094 set_16bit_gcc_code_flag (new_code_flag)
1095 int new_code_flag;
1096 {
1097 flag_code = new_code_flag;
1098 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
1099 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
1100 stackop_size = LONG_MNEM_SUFFIX;
1101 }
1102
1103 static void
1104 set_intel_syntax (syntax_flag)
1105 int syntax_flag;
1106 {
1107 /* Find out if register prefixing is specified. */
1108 int ask_naked_reg = 0;
1109
1110 SKIP_WHITESPACE ();
1111 if (!is_end_of_line[(unsigned char) *input_line_pointer])
1112 {
1113 char *string = input_line_pointer;
1114 int e = get_symbol_end ();
1115
1116 if (strcmp (string, "prefix") == 0)
1117 ask_naked_reg = 1;
1118 else if (strcmp (string, "noprefix") == 0)
1119 ask_naked_reg = -1;
1120 else
1121 as_bad (_("bad argument to syntax directive."));
1122 *input_line_pointer = e;
1123 }
1124 demand_empty_rest_of_line ();
1125
1126 intel_syntax = syntax_flag;
1127
1128 if (ask_naked_reg == 0)
1129 allow_naked_reg = (intel_syntax
1130 && (bfd_get_symbol_leading_char (stdoutput) != '\0'));
1131 else
1132 allow_naked_reg = (ask_naked_reg < 0);
1133
1134 identifier_chars['%'] = intel_syntax && allow_naked_reg ? '%' : 0;
1135 identifier_chars['$'] = intel_syntax ? '$' : 0;
1136 }
1137
1138 static void
1139 set_cpu_arch (dummy)
1140 int dummy ATTRIBUTE_UNUSED;
1141 {
1142 SKIP_WHITESPACE ();
1143
1144 if (!is_end_of_line[(unsigned char) *input_line_pointer])
1145 {
1146 char *string = input_line_pointer;
1147 int e = get_symbol_end ();
1148 unsigned int i;
1149
1150 for (i = 0; i < ARRAY_SIZE (cpu_arch); i++)
1151 {
1152 if (strcmp (string, cpu_arch[i].name) == 0)
1153 {
1154 if (*string != '.')
1155 {
1156 cpu_arch_name = cpu_arch[i].name;
1157 cpu_sub_arch_name = NULL;
1158 cpu_arch_flags = (cpu_arch[i].flags
1159 | (flag_code == CODE_64BIT ? Cpu64 : CpuNo64));
1160 cpu_arch_isa = cpu_arch[i].type;
1161 cpu_arch_isa_flags = cpu_arch[i].flags;
1162 if (!cpu_arch_tune_set)
1163 {
1164 cpu_arch_tune = cpu_arch_isa;
1165 cpu_arch_tune_flags = cpu_arch_isa_flags;
1166 }
1167 break;
1168 }
1169 if ((cpu_arch_flags | cpu_arch[i].flags) != cpu_arch_flags)
1170 {
1171 cpu_sub_arch_name = cpu_arch[i].name;
1172 cpu_arch_flags |= cpu_arch[i].flags;
1173 }
1174 *input_line_pointer = e;
1175 demand_empty_rest_of_line ();
1176 return;
1177 }
1178 }
1179 if (i >= ARRAY_SIZE (cpu_arch))
1180 as_bad (_("no such architecture: `%s'"), string);
1181
1182 *input_line_pointer = e;
1183 }
1184 else
1185 as_bad (_("missing cpu architecture"));
1186
1187 no_cond_jump_promotion = 0;
1188 if (*input_line_pointer == ','
1189 && !is_end_of_line[(unsigned char) input_line_pointer[1]])
1190 {
1191 char *string = ++input_line_pointer;
1192 int e = get_symbol_end ();
1193
1194 if (strcmp (string, "nojumps") == 0)
1195 no_cond_jump_promotion = 1;
1196 else if (strcmp (string, "jumps") == 0)
1197 ;
1198 else
1199 as_bad (_("no such architecture modifier: `%s'"), string);
1200
1201 *input_line_pointer = e;
1202 }
1203
1204 demand_empty_rest_of_line ();
1205 }
1206
1207 unsigned long
1208 i386_mach ()
1209 {
1210 if (!strcmp (default_arch, "x86_64"))
1211 return bfd_mach_x86_64;
1212 else if (!strcmp (default_arch, "i386"))
1213 return bfd_mach_i386_i386;
1214 else
1215 as_fatal (_("Unknown architecture"));
1216 }
1217 \f
1218 void
1219 md_begin ()
1220 {
1221 const char *hash_err;
1222
1223 /* Initialize op_hash hash table. */
1224 op_hash = hash_new ();
1225
1226 {
1227 const template *optab;
1228 templates *core_optab;
1229
1230 /* Setup for loop. */
1231 optab = i386_optab;
1232 core_optab = (templates *) xmalloc (sizeof (templates));
1233 core_optab->start = optab;
1234
1235 while (1)
1236 {
1237 ++optab;
1238 if (optab->name == NULL
1239 || strcmp (optab->name, (optab - 1)->name) != 0)
1240 {
1241 /* different name --> ship out current template list;
1242 add to hash table; & begin anew. */
1243 core_optab->end = optab;
1244 hash_err = hash_insert (op_hash,
1245 (optab - 1)->name,
1246 (PTR) core_optab);
1247 if (hash_err)
1248 {
1249 as_fatal (_("Internal Error: Can't hash %s: %s"),
1250 (optab - 1)->name,
1251 hash_err);
1252 }
1253 if (optab->name == NULL)
1254 break;
1255 core_optab = (templates *) xmalloc (sizeof (templates));
1256 core_optab->start = optab;
1257 }
1258 }
1259 }
1260
1261 /* Initialize reg_hash hash table. */
1262 reg_hash = hash_new ();
1263 {
1264 const reg_entry *regtab;
1265
1266 for (regtab = i386_regtab;
1267 regtab < i386_regtab + sizeof (i386_regtab) / sizeof (i386_regtab[0]);
1268 regtab++)
1269 {
1270 hash_err = hash_insert (reg_hash, regtab->reg_name, (PTR) regtab);
1271 if (hash_err)
1272 as_fatal (_("Internal Error: Can't hash %s: %s"),
1273 regtab->reg_name,
1274 hash_err);
1275 }
1276 }
1277
1278 /* Fill in lexical tables: mnemonic_chars, operand_chars. */
1279 {
1280 int c;
1281 char *p;
1282
1283 for (c = 0; c < 256; c++)
1284 {
1285 if (ISDIGIT (c))
1286 {
1287 digit_chars[c] = c;
1288 mnemonic_chars[c] = c;
1289 register_chars[c] = c;
1290 operand_chars[c] = c;
1291 }
1292 else if (ISLOWER (c))
1293 {
1294 mnemonic_chars[c] = c;
1295 register_chars[c] = c;
1296 operand_chars[c] = c;
1297 }
1298 else if (ISUPPER (c))
1299 {
1300 mnemonic_chars[c] = TOLOWER (c);
1301 register_chars[c] = mnemonic_chars[c];
1302 operand_chars[c] = c;
1303 }
1304
1305 if (ISALPHA (c) || ISDIGIT (c))
1306 identifier_chars[c] = c;
1307 else if (c >= 128)
1308 {
1309 identifier_chars[c] = c;
1310 operand_chars[c] = c;
1311 }
1312 }
1313
1314 #ifdef LEX_AT
1315 identifier_chars['@'] = '@';
1316 #endif
1317 #ifdef LEX_QM
1318 identifier_chars['?'] = '?';
1319 operand_chars['?'] = '?';
1320 #endif
1321 digit_chars['-'] = '-';
1322 mnemonic_chars['-'] = '-';
1323 identifier_chars['_'] = '_';
1324 identifier_chars['.'] = '.';
1325
1326 for (p = operand_special_chars; *p != '\0'; p++)
1327 operand_chars[(unsigned char) *p] = *p;
1328 }
1329
1330 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1331 if (IS_ELF)
1332 {
1333 record_alignment (text_section, 2);
1334 record_alignment (data_section, 2);
1335 record_alignment (bss_section, 2);
1336 }
1337 #endif
1338
1339 if (flag_code == CODE_64BIT)
1340 {
1341 x86_dwarf2_return_column = 16;
1342 x86_cie_data_alignment = -8;
1343 }
1344 else
1345 {
1346 x86_dwarf2_return_column = 8;
1347 x86_cie_data_alignment = -4;
1348 }
1349 }
1350
1351 void
1352 i386_print_statistics (file)
1353 FILE *file;
1354 {
1355 hash_print_statistics (file, "i386 opcode", op_hash);
1356 hash_print_statistics (file, "i386 register", reg_hash);
1357 }
1358 \f
1359 #ifdef DEBUG386
1360
1361 /* Debugging routines for md_assemble. */
1362 static void pi PARAMS ((char *, i386_insn *));
1363 static void pte PARAMS ((template *));
1364 static void pt PARAMS ((unsigned int));
1365 static void pe PARAMS ((expressionS *));
1366 static void ps PARAMS ((symbolS *));
1367
1368 static void
1369 pi (line, x)
1370 char *line;
1371 i386_insn *x;
1372 {
1373 unsigned int i;
1374
1375 fprintf (stdout, "%s: template ", line);
1376 pte (&x->tm);
1377 fprintf (stdout, " address: base %s index %s scale %x\n",
1378 x->base_reg ? x->base_reg->reg_name : "none",
1379 x->index_reg ? x->index_reg->reg_name : "none",
1380 x->log2_scale_factor);
1381 fprintf (stdout, " modrm: mode %x reg %x reg/mem %x\n",
1382 x->rm.mode, x->rm.reg, x->rm.regmem);
1383 fprintf (stdout, " sib: base %x index %x scale %x\n",
1384 x->sib.base, x->sib.index, x->sib.scale);
1385 fprintf (stdout, " rex: 64bit %x extX %x extY %x extZ %x\n",
1386 (x->rex & REX_MODE64) != 0,
1387 (x->rex & REX_EXTX) != 0,
1388 (x->rex & REX_EXTY) != 0,
1389 (x->rex & REX_EXTZ) != 0);
1390 for (i = 0; i < x->operands; i++)
1391 {
1392 fprintf (stdout, " #%d: ", i + 1);
1393 pt (x->types[i]);
1394 fprintf (stdout, "\n");
1395 if (x->types[i]
1396 & (Reg | SReg2 | SReg3 | Control | Debug | Test | RegMMX | RegXMM))
1397 fprintf (stdout, "%s\n", x->op[i].regs->reg_name);
1398 if (x->types[i] & Imm)
1399 pe (x->op[i].imms);
1400 if (x->types[i] & Disp)
1401 pe (x->op[i].disps);
1402 }
1403 }
1404
1405 static void
1406 pte (t)
1407 template *t;
1408 {
1409 unsigned int i;
1410 fprintf (stdout, " %d operands ", t->operands);
1411 fprintf (stdout, "opcode %x ", t->base_opcode);
1412 if (t->extension_opcode != None)
1413 fprintf (stdout, "ext %x ", t->extension_opcode);
1414 if (t->opcode_modifier & D)
1415 fprintf (stdout, "D");
1416 if (t->opcode_modifier & W)
1417 fprintf (stdout, "W");
1418 fprintf (stdout, "\n");
1419 for (i = 0; i < t->operands; i++)
1420 {
1421 fprintf (stdout, " #%d type ", i + 1);
1422 pt (t->operand_types[i]);
1423 fprintf (stdout, "\n");
1424 }
1425 }
1426
1427 static void
1428 pe (e)
1429 expressionS *e;
1430 {
1431 fprintf (stdout, " operation %d\n", e->X_op);
1432 fprintf (stdout, " add_number %ld (%lx)\n",
1433 (long) e->X_add_number, (long) e->X_add_number);
1434 if (e->X_add_symbol)
1435 {
1436 fprintf (stdout, " add_symbol ");
1437 ps (e->X_add_symbol);
1438 fprintf (stdout, "\n");
1439 }
1440 if (e->X_op_symbol)
1441 {
1442 fprintf (stdout, " op_symbol ");
1443 ps (e->X_op_symbol);
1444 fprintf (stdout, "\n");
1445 }
1446 }
1447
1448 static void
1449 ps (s)
1450 symbolS *s;
1451 {
1452 fprintf (stdout, "%s type %s%s",
1453 S_GET_NAME (s),
1454 S_IS_EXTERNAL (s) ? "EXTERNAL " : "",
1455 segment_name (S_GET_SEGMENT (s)));
1456 }
1457
1458 static struct type_name
1459 {
1460 unsigned int mask;
1461 char *tname;
1462 }
1463 const type_names[] =
1464 {
1465 { Reg8, "r8" },
1466 { Reg16, "r16" },
1467 { Reg32, "r32" },
1468 { Reg64, "r64" },
1469 { Imm8, "i8" },
1470 { Imm8S, "i8s" },
1471 { Imm16, "i16" },
1472 { Imm32, "i32" },
1473 { Imm32S, "i32s" },
1474 { Imm64, "i64" },
1475 { Imm1, "i1" },
1476 { BaseIndex, "BaseIndex" },
1477 { Disp8, "d8" },
1478 { Disp16, "d16" },
1479 { Disp32, "d32" },
1480 { Disp32S, "d32s" },
1481 { Disp64, "d64" },
1482 { InOutPortReg, "InOutPortReg" },
1483 { ShiftCount, "ShiftCount" },
1484 { Control, "control reg" },
1485 { Test, "test reg" },
1486 { Debug, "debug reg" },
1487 { FloatReg, "FReg" },
1488 { FloatAcc, "FAcc" },
1489 { SReg2, "SReg2" },
1490 { SReg3, "SReg3" },
1491 { Acc, "Acc" },
1492 { JumpAbsolute, "Jump Absolute" },
1493 { RegMMX, "rMMX" },
1494 { RegXMM, "rXMM" },
1495 { EsSeg, "es" },
1496 { 0, "" }
1497 };
1498
1499 static void
1500 pt (t)
1501 unsigned int t;
1502 {
1503 const struct type_name *ty;
1504
1505 for (ty = type_names; ty->mask; ty++)
1506 if (t & ty->mask)
1507 fprintf (stdout, "%s, ", ty->tname);
1508 fflush (stdout);
1509 }
1510
1511 #endif /* DEBUG386 */
1512 \f
1513 static bfd_reloc_code_real_type
1514 reloc (unsigned int size,
1515 int pcrel,
1516 int sign,
1517 bfd_reloc_code_real_type other)
1518 {
1519 if (other != NO_RELOC)
1520 {
1521 reloc_howto_type *reloc;
1522
1523 if (size == 8)
1524 switch (other)
1525 {
1526 case BFD_RELOC_X86_64_GOT32:
1527 return BFD_RELOC_X86_64_GOT64;
1528 break;
1529 case BFD_RELOC_X86_64_PLTOFF64:
1530 return BFD_RELOC_X86_64_PLTOFF64;
1531 break;
1532 case BFD_RELOC_X86_64_GOTPC32:
1533 other = BFD_RELOC_X86_64_GOTPC64;
1534 break;
1535 case BFD_RELOC_X86_64_GOTPCREL:
1536 other = BFD_RELOC_X86_64_GOTPCREL64;
1537 break;
1538 case BFD_RELOC_X86_64_TPOFF32:
1539 other = BFD_RELOC_X86_64_TPOFF64;
1540 break;
1541 case BFD_RELOC_X86_64_DTPOFF32:
1542 other = BFD_RELOC_X86_64_DTPOFF64;
1543 break;
1544 default:
1545 break;
1546 }
1547
1548 /* Sign-checking 4-byte relocations in 16-/32-bit code is pointless. */
1549 if (size == 4 && flag_code != CODE_64BIT)
1550 sign = -1;
1551
1552 reloc = bfd_reloc_type_lookup (stdoutput, other);
1553 if (!reloc)
1554 as_bad (_("unknown relocation (%u)"), other);
1555 else if (size != bfd_get_reloc_size (reloc))
1556 as_bad (_("%u-byte relocation cannot be applied to %u-byte field"),
1557 bfd_get_reloc_size (reloc),
1558 size);
1559 else if (pcrel && !reloc->pc_relative)
1560 as_bad (_("non-pc-relative relocation for pc-relative field"));
1561 else if ((reloc->complain_on_overflow == complain_overflow_signed
1562 && !sign)
1563 || (reloc->complain_on_overflow == complain_overflow_unsigned
1564 && sign > 0))
1565 as_bad (_("relocated field and relocation type differ in signedness"));
1566 else
1567 return other;
1568 return NO_RELOC;
1569 }
1570
1571 if (pcrel)
1572 {
1573 if (!sign)
1574 as_bad (_("there are no unsigned pc-relative relocations"));
1575 switch (size)
1576 {
1577 case 1: return BFD_RELOC_8_PCREL;
1578 case 2: return BFD_RELOC_16_PCREL;
1579 case 4: return BFD_RELOC_32_PCREL;
1580 case 8: return BFD_RELOC_64_PCREL;
1581 }
1582 as_bad (_("cannot do %u byte pc-relative relocation"), size);
1583 }
1584 else
1585 {
1586 if (sign > 0)
1587 switch (size)
1588 {
1589 case 4: return BFD_RELOC_X86_64_32S;
1590 }
1591 else
1592 switch (size)
1593 {
1594 case 1: return BFD_RELOC_8;
1595 case 2: return BFD_RELOC_16;
1596 case 4: return BFD_RELOC_32;
1597 case 8: return BFD_RELOC_64;
1598 }
1599 as_bad (_("cannot do %s %u byte relocation"),
1600 sign > 0 ? "signed" : "unsigned", size);
1601 }
1602
1603 abort ();
1604 return BFD_RELOC_NONE;
1605 }
1606
1607 /* Here we decide which fixups can be adjusted to make them relative to
1608 the beginning of the section instead of the symbol. Basically we need
1609 to make sure that the dynamic relocations are done correctly, so in
1610 some cases we force the original symbol to be used. */
1611
1612 int
1613 tc_i386_fix_adjustable (fixP)
1614 fixS *fixP ATTRIBUTE_UNUSED;
1615 {
1616 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1617 if (!IS_ELF)
1618 return 1;
1619
1620 /* Don't adjust pc-relative references to merge sections in 64-bit
1621 mode. */
1622 if (use_rela_relocations
1623 && (S_GET_SEGMENT (fixP->fx_addsy)->flags & SEC_MERGE) != 0
1624 && fixP->fx_pcrel)
1625 return 0;
1626
1627 /* The x86_64 GOTPCREL are represented as 32bit PCrel relocations
1628 and changed later by validate_fix. */
1629 if (GOT_symbol && fixP->fx_subsy == GOT_symbol
1630 && fixP->fx_r_type == BFD_RELOC_32_PCREL)
1631 return 0;
1632
1633 /* adjust_reloc_syms doesn't know about the GOT. */
1634 if (fixP->fx_r_type == BFD_RELOC_386_GOTOFF
1635 || fixP->fx_r_type == BFD_RELOC_386_PLT32
1636 || fixP->fx_r_type == BFD_RELOC_386_GOT32
1637 || fixP->fx_r_type == BFD_RELOC_386_TLS_GD
1638 || fixP->fx_r_type == BFD_RELOC_386_TLS_LDM
1639 || fixP->fx_r_type == BFD_RELOC_386_TLS_LDO_32
1640 || fixP->fx_r_type == BFD_RELOC_386_TLS_IE_32
1641 || fixP->fx_r_type == BFD_RELOC_386_TLS_IE
1642 || fixP->fx_r_type == BFD_RELOC_386_TLS_GOTIE
1643 || fixP->fx_r_type == BFD_RELOC_386_TLS_LE_32
1644 || fixP->fx_r_type == BFD_RELOC_386_TLS_LE
1645 || fixP->fx_r_type == BFD_RELOC_386_TLS_GOTDESC
1646 || fixP->fx_r_type == BFD_RELOC_386_TLS_DESC_CALL
1647 || fixP->fx_r_type == BFD_RELOC_X86_64_PLT32
1648 || fixP->fx_r_type == BFD_RELOC_X86_64_GOT32
1649 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPCREL
1650 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSGD
1651 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSLD
1652 || fixP->fx_r_type == BFD_RELOC_X86_64_DTPOFF32
1653 || fixP->fx_r_type == BFD_RELOC_X86_64_DTPOFF64
1654 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTTPOFF
1655 || fixP->fx_r_type == BFD_RELOC_X86_64_TPOFF32
1656 || fixP->fx_r_type == BFD_RELOC_X86_64_TPOFF64
1657 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTOFF64
1658 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPC32_TLSDESC
1659 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSDESC_CALL
1660 || fixP->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1661 || fixP->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1662 return 0;
1663 #endif
1664 return 1;
1665 }
1666
1667 static int intel_float_operand PARAMS ((const char *mnemonic));
1668
1669 static int
1670 intel_float_operand (mnemonic)
1671 const char *mnemonic;
1672 {
1673 /* Note that the value returned is meaningful only for opcodes with (memory)
1674 operands, hence the code here is free to improperly handle opcodes that
1675 have no operands (for better performance and smaller code). */
1676
1677 if (mnemonic[0] != 'f')
1678 return 0; /* non-math */
1679
1680 switch (mnemonic[1])
1681 {
1682 /* fclex, fdecstp, fdisi, femms, feni, fincstp, finit, fsetpm, and
1683 the fs segment override prefix not currently handled because no
1684 call path can make opcodes without operands get here */
1685 case 'i':
1686 return 2 /* integer op */;
1687 case 'l':
1688 if (mnemonic[2] == 'd' && (mnemonic[3] == 'c' || mnemonic[3] == 'e'))
1689 return 3; /* fldcw/fldenv */
1690 break;
1691 case 'n':
1692 if (mnemonic[2] != 'o' /* fnop */)
1693 return 3; /* non-waiting control op */
1694 break;
1695 case 'r':
1696 if (mnemonic[2] == 's')
1697 return 3; /* frstor/frstpm */
1698 break;
1699 case 's':
1700 if (mnemonic[2] == 'a')
1701 return 3; /* fsave */
1702 if (mnemonic[2] == 't')
1703 {
1704 switch (mnemonic[3])
1705 {
1706 case 'c': /* fstcw */
1707 case 'd': /* fstdw */
1708 case 'e': /* fstenv */
1709 case 's': /* fsts[gw] */
1710 return 3;
1711 }
1712 }
1713 break;
1714 case 'x':
1715 if (mnemonic[2] == 'r' || mnemonic[2] == 's')
1716 return 0; /* fxsave/fxrstor are not really math ops */
1717 break;
1718 }
1719
1720 return 1;
1721 }
1722
1723 /* This is the guts of the machine-dependent assembler. LINE points to a
1724 machine dependent instruction. This function is supposed to emit
1725 the frags/bytes it assembles to. */
1726
1727 void
1728 md_assemble (line)
1729 char *line;
1730 {
1731 int j;
1732 char mnemonic[MAX_MNEM_SIZE];
1733
1734 /* Initialize globals. */
1735 memset (&i, '\0', sizeof (i));
1736 for (j = 0; j < MAX_OPERANDS; j++)
1737 i.reloc[j] = NO_RELOC;
1738 memset (disp_expressions, '\0', sizeof (disp_expressions));
1739 memset (im_expressions, '\0', sizeof (im_expressions));
1740 save_stack_p = save_stack;
1741
1742 /* First parse an instruction mnemonic & call i386_operand for the operands.
1743 We assume that the scrubber has arranged it so that line[0] is the valid
1744 start of a (possibly prefixed) mnemonic. */
1745
1746 line = parse_insn (line, mnemonic);
1747 if (line == NULL)
1748 return;
1749
1750 line = parse_operands (line, mnemonic);
1751 if (line == NULL)
1752 return;
1753
1754 /* The order of the immediates should be reversed
1755 for 2 immediates extrq and insertq instructions */
1756 if ((i.imm_operands == 2) &&
1757 ((strcmp (mnemonic, "extrq") == 0)
1758 || (strcmp (mnemonic, "insertq") == 0)))
1759 {
1760 swap_imm_operands ();
1761 /* "extrq" and insertq" are the only two instructions whose operands
1762 have to be reversed even though they have two immediate operands.
1763 */
1764 if (intel_syntax)
1765 swap_operands ();
1766 }
1767
1768 /* Now we've parsed the mnemonic into a set of templates, and have the
1769 operands at hand. */
1770
1771 /* All intel opcodes have reversed operands except for "bound" and
1772 "enter". We also don't reverse intersegment "jmp" and "call"
1773 instructions with 2 immediate operands so that the immediate segment
1774 precedes the offset, as it does when in AT&T mode. */
1775 if (intel_syntax && i.operands > 1
1776 && (strcmp (mnemonic, "bound") != 0)
1777 && (strcmp (mnemonic, "invlpga") != 0)
1778 && !((i.types[0] & Imm) && (i.types[1] & Imm)))
1779 swap_operands ();
1780
1781 if (i.imm_operands)
1782 optimize_imm ();
1783
1784 /* Don't optimize displacement for movabs since it only takes 64bit
1785 displacement. */
1786 if (i.disp_operands
1787 && (flag_code != CODE_64BIT
1788 || strcmp (mnemonic, "movabs") != 0))
1789 optimize_disp ();
1790
1791 /* Next, we find a template that matches the given insn,
1792 making sure the overlap of the given operands types is consistent
1793 with the template operand types. */
1794
1795 if (!match_template ())
1796 return;
1797
1798 if (intel_syntax)
1799 {
1800 /* Undo SYSV386_COMPAT brokenness when in Intel mode. See i386.h */
1801 if (SYSV386_COMPAT
1802 && (i.tm.base_opcode & 0xfffffde0) == 0xdce0)
1803 i.tm.base_opcode ^= FloatR;
1804
1805 /* Zap movzx and movsx suffix. The suffix may have been set from
1806 "word ptr" or "byte ptr" on the source operand, but we'll use
1807 the suffix later to choose the destination register. */
1808 if ((i.tm.base_opcode & ~9) == 0x0fb6)
1809 {
1810 if (i.reg_operands < 2
1811 && !i.suffix
1812 && (~i.tm.opcode_modifier
1813 & (No_bSuf
1814 | No_wSuf
1815 | No_lSuf
1816 | No_sSuf
1817 | No_xSuf
1818 | No_qSuf)))
1819 as_bad (_("ambiguous operand size for `%s'"), i.tm.name);
1820
1821 i.suffix = 0;
1822 }
1823 }
1824
1825 if (i.tm.opcode_modifier & FWait)
1826 if (!add_prefix (FWAIT_OPCODE))
1827 return;
1828
1829 /* Check string instruction segment overrides. */
1830 if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0)
1831 {
1832 if (!check_string ())
1833 return;
1834 }
1835
1836 if (!process_suffix ())
1837 return;
1838
1839 /* Make still unresolved immediate matches conform to size of immediate
1840 given in i.suffix. */
1841 if (!finalize_imm ())
1842 return;
1843
1844 if (i.types[0] & Imm1)
1845 i.imm_operands = 0; /* kludge for shift insns. */
1846 if (i.types[0] & ImplicitRegister)
1847 i.reg_operands--;
1848 if (i.types[1] & ImplicitRegister)
1849 i.reg_operands--;
1850 if (i.types[2] & ImplicitRegister)
1851 i.reg_operands--;
1852
1853 if (i.tm.opcode_modifier & ImmExt)
1854 {
1855 expressionS *exp;
1856
1857 if ((i.tm.cpu_flags & CpuSSE3) && i.operands > 0)
1858 {
1859 /* Streaming SIMD extensions 3 Instructions have the fixed
1860 operands with an opcode suffix which is coded in the same
1861 place as an 8-bit immediate field would be. Here we check
1862 those operands and remove them afterwards. */
1863 unsigned int x;
1864
1865 for (x = 0; x < i.operands; x++)
1866 if (i.op[x].regs->reg_num != x)
1867 as_bad (_("can't use register '%%%s' as operand %d in '%s'."),
1868 i.op[x].regs->reg_name, x + 1, i.tm.name);
1869 i.operands = 0;
1870 }
1871
1872 /* These AMD 3DNow! and Intel Katmai New Instructions have an
1873 opcode suffix which is coded in the same place as an 8-bit
1874 immediate field would be. Here we fake an 8-bit immediate
1875 operand from the opcode suffix stored in tm.extension_opcode. */
1876
1877 assert (i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS);
1878
1879 exp = &im_expressions[i.imm_operands++];
1880 i.op[i.operands].imms = exp;
1881 i.types[i.operands++] = Imm8;
1882 exp->X_op = O_constant;
1883 exp->X_add_number = i.tm.extension_opcode;
1884 i.tm.extension_opcode = None;
1885 }
1886
1887 /* For insns with operands there are more diddles to do to the opcode. */
1888 if (i.operands)
1889 {
1890 if (!process_operands ())
1891 return;
1892 }
1893 else if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
1894 {
1895 /* UnixWare fsub no args is alias for fsubp, fadd -> faddp, etc. */
1896 as_warn (_("translating to `%sp'"), i.tm.name);
1897 }
1898
1899 /* Handle conversion of 'int $3' --> special int3 insn. */
1900 if (i.tm.base_opcode == INT_OPCODE && i.op[0].imms->X_add_number == 3)
1901 {
1902 i.tm.base_opcode = INT3_OPCODE;
1903 i.imm_operands = 0;
1904 }
1905
1906 if ((i.tm.opcode_modifier & (Jump | JumpByte | JumpDword))
1907 && i.op[0].disps->X_op == O_constant)
1908 {
1909 /* Convert "jmp constant" (and "call constant") to a jump (call) to
1910 the absolute address given by the constant. Since ix86 jumps and
1911 calls are pc relative, we need to generate a reloc. */
1912 i.op[0].disps->X_add_symbol = &abs_symbol;
1913 i.op[0].disps->X_op = O_symbol;
1914 }
1915
1916 if ((i.tm.opcode_modifier & Rex64) != 0)
1917 i.rex |= REX_MODE64;
1918
1919 /* For 8 bit registers we need an empty rex prefix. Also if the
1920 instruction already has a prefix, we need to convert old
1921 registers to new ones. */
1922
1923 if (((i.types[0] & Reg8) != 0
1924 && (i.op[0].regs->reg_flags & RegRex64) != 0)
1925 || ((i.types[1] & Reg8) != 0
1926 && (i.op[1].regs->reg_flags & RegRex64) != 0)
1927 || (((i.types[0] & Reg8) != 0 || (i.types[1] & Reg8) != 0)
1928 && i.rex != 0))
1929 {
1930 int x;
1931
1932 i.rex |= REX_OPCODE;
1933 for (x = 0; x < 2; x++)
1934 {
1935 /* Look for 8 bit operand that uses old registers. */
1936 if ((i.types[x] & Reg8) != 0
1937 && (i.op[x].regs->reg_flags & RegRex64) == 0)
1938 {
1939 /* In case it is "hi" register, give up. */
1940 if (i.op[x].regs->reg_num > 3)
1941 as_bad (_("can't encode register '%%%s' in an instruction requiring REX prefix."),
1942 i.op[x].regs->reg_name);
1943
1944 /* Otherwise it is equivalent to the extended register.
1945 Since the encoding doesn't change this is merely
1946 cosmetic cleanup for debug output. */
1947
1948 i.op[x].regs = i.op[x].regs + 8;
1949 }
1950 }
1951 }
1952
1953 if (i.rex != 0)
1954 add_prefix (REX_OPCODE | i.rex);
1955
1956 /* We are ready to output the insn. */
1957 output_insn ();
1958 }
1959
1960 static char *
1961 parse_insn (line, mnemonic)
1962 char *line;
1963 char *mnemonic;
1964 {
1965 char *l = line;
1966 char *token_start = l;
1967 char *mnem_p;
1968 int supported;
1969 const template *t;
1970
1971 /* Non-zero if we found a prefix only acceptable with string insns. */
1972 const char *expecting_string_instruction = NULL;
1973
1974 while (1)
1975 {
1976 mnem_p = mnemonic;
1977 while ((*mnem_p = mnemonic_chars[(unsigned char) *l]) != 0)
1978 {
1979 mnem_p++;
1980 if (mnem_p >= mnemonic + MAX_MNEM_SIZE)
1981 {
1982 as_bad (_("no such instruction: `%s'"), token_start);
1983 return NULL;
1984 }
1985 l++;
1986 }
1987 if (!is_space_char (*l)
1988 && *l != END_OF_INSN
1989 && (intel_syntax
1990 || (*l != PREFIX_SEPARATOR
1991 && *l != ',')))
1992 {
1993 as_bad (_("invalid character %s in mnemonic"),
1994 output_invalid (*l));
1995 return NULL;
1996 }
1997 if (token_start == l)
1998 {
1999 if (!intel_syntax && *l == PREFIX_SEPARATOR)
2000 as_bad (_("expecting prefix; got nothing"));
2001 else
2002 as_bad (_("expecting mnemonic; got nothing"));
2003 return NULL;
2004 }
2005
2006 /* Look up instruction (or prefix) via hash table. */
2007 current_templates = hash_find (op_hash, mnemonic);
2008
2009 if (*l != END_OF_INSN
2010 && (!is_space_char (*l) || l[1] != END_OF_INSN)
2011 && current_templates
2012 && (current_templates->start->opcode_modifier & IsPrefix))
2013 {
2014 if (current_templates->start->cpu_flags
2015 & (flag_code != CODE_64BIT ? Cpu64 : CpuNo64))
2016 {
2017 as_bad ((flag_code != CODE_64BIT
2018 ? _("`%s' is only supported in 64-bit mode")
2019 : _("`%s' is not supported in 64-bit mode")),
2020 current_templates->start->name);
2021 return NULL;
2022 }
2023 /* If we are in 16-bit mode, do not allow addr16 or data16.
2024 Similarly, in 32-bit mode, do not allow addr32 or data32. */
2025 if ((current_templates->start->opcode_modifier & (Size16 | Size32))
2026 && flag_code != CODE_64BIT
2027 && (((current_templates->start->opcode_modifier & Size32) != 0)
2028 ^ (flag_code == CODE_16BIT)))
2029 {
2030 as_bad (_("redundant %s prefix"),
2031 current_templates->start->name);
2032 return NULL;
2033 }
2034 /* Add prefix, checking for repeated prefixes. */
2035 switch (add_prefix (current_templates->start->base_opcode))
2036 {
2037 case 0:
2038 return NULL;
2039 case 2:
2040 expecting_string_instruction = current_templates->start->name;
2041 break;
2042 }
2043 /* Skip past PREFIX_SEPARATOR and reset token_start. */
2044 token_start = ++l;
2045 }
2046 else
2047 break;
2048 }
2049
2050 if (!current_templates)
2051 {
2052 /* See if we can get a match by trimming off a suffix. */
2053 switch (mnem_p[-1])
2054 {
2055 case WORD_MNEM_SUFFIX:
2056 if (intel_syntax && (intel_float_operand (mnemonic) & 2))
2057 i.suffix = SHORT_MNEM_SUFFIX;
2058 else
2059 case BYTE_MNEM_SUFFIX:
2060 case QWORD_MNEM_SUFFIX:
2061 i.suffix = mnem_p[-1];
2062 mnem_p[-1] = '\0';
2063 current_templates = hash_find (op_hash, mnemonic);
2064 break;
2065 case SHORT_MNEM_SUFFIX:
2066 case LONG_MNEM_SUFFIX:
2067 if (!intel_syntax)
2068 {
2069 i.suffix = mnem_p[-1];
2070 mnem_p[-1] = '\0';
2071 current_templates = hash_find (op_hash, mnemonic);
2072 }
2073 break;
2074
2075 /* Intel Syntax. */
2076 case 'd':
2077 if (intel_syntax)
2078 {
2079 if (intel_float_operand (mnemonic) == 1)
2080 i.suffix = SHORT_MNEM_SUFFIX;
2081 else
2082 i.suffix = LONG_MNEM_SUFFIX;
2083 mnem_p[-1] = '\0';
2084 current_templates = hash_find (op_hash, mnemonic);
2085 }
2086 break;
2087 }
2088 if (!current_templates)
2089 {
2090 as_bad (_("no such instruction: `%s'"), token_start);
2091 return NULL;
2092 }
2093 }
2094
2095 if (current_templates->start->opcode_modifier & (Jump | JumpByte))
2096 {
2097 /* Check for a branch hint. We allow ",pt" and ",pn" for
2098 predict taken and predict not taken respectively.
2099 I'm not sure that branch hints actually do anything on loop
2100 and jcxz insns (JumpByte) for current Pentium4 chips. They
2101 may work in the future and it doesn't hurt to accept them
2102 now. */
2103 if (l[0] == ',' && l[1] == 'p')
2104 {
2105 if (l[2] == 't')
2106 {
2107 if (!add_prefix (DS_PREFIX_OPCODE))
2108 return NULL;
2109 l += 3;
2110 }
2111 else if (l[2] == 'n')
2112 {
2113 if (!add_prefix (CS_PREFIX_OPCODE))
2114 return NULL;
2115 l += 3;
2116 }
2117 }
2118 }
2119 /* Any other comma loses. */
2120 if (*l == ',')
2121 {
2122 as_bad (_("invalid character %s in mnemonic"),
2123 output_invalid (*l));
2124 return NULL;
2125 }
2126
2127 /* Check if instruction is supported on specified architecture. */
2128 supported = 0;
2129 for (t = current_templates->start; t < current_templates->end; ++t)
2130 {
2131 if (!((t->cpu_flags & ~(Cpu64 | CpuNo64))
2132 & ~(cpu_arch_flags & ~(Cpu64 | CpuNo64))))
2133 supported |= 1;
2134 if (!(t->cpu_flags & (flag_code == CODE_64BIT ? CpuNo64 : Cpu64)))
2135 supported |= 2;
2136 }
2137 if (!(supported & 2))
2138 {
2139 as_bad (flag_code == CODE_64BIT
2140 ? _("`%s' is not supported in 64-bit mode")
2141 : _("`%s' is only supported in 64-bit mode"),
2142 current_templates->start->name);
2143 return NULL;
2144 }
2145 if (!(supported & 1))
2146 {
2147 as_warn (_("`%s' is not supported on `%s%s'"),
2148 current_templates->start->name,
2149 cpu_arch_name,
2150 cpu_sub_arch_name ? cpu_sub_arch_name : "");
2151 }
2152 else if ((Cpu386 & ~cpu_arch_flags) && (flag_code != CODE_16BIT))
2153 {
2154 as_warn (_("use .code16 to ensure correct addressing mode"));
2155 }
2156
2157 /* Check for rep/repne without a string instruction. */
2158 if (expecting_string_instruction)
2159 {
2160 static templates override;
2161
2162 for (t = current_templates->start; t < current_templates->end; ++t)
2163 if (t->opcode_modifier & IsString)
2164 break;
2165 if (t >= current_templates->end)
2166 {
2167 as_bad (_("expecting string instruction after `%s'"),
2168 expecting_string_instruction);
2169 return NULL;
2170 }
2171 for (override.start = t; t < current_templates->end; ++t)
2172 if (!(t->opcode_modifier & IsString))
2173 break;
2174 override.end = t;
2175 current_templates = &override;
2176 }
2177
2178 return l;
2179 }
2180
2181 static char *
2182 parse_operands (l, mnemonic)
2183 char *l;
2184 const char *mnemonic;
2185 {
2186 char *token_start;
2187
2188 /* 1 if operand is pending after ','. */
2189 unsigned int expecting_operand = 0;
2190
2191 /* Non-zero if operand parens not balanced. */
2192 unsigned int paren_not_balanced;
2193
2194 while (*l != END_OF_INSN)
2195 {
2196 /* Skip optional white space before operand. */
2197 if (is_space_char (*l))
2198 ++l;
2199 if (!is_operand_char (*l) && *l != END_OF_INSN)
2200 {
2201 as_bad (_("invalid character %s before operand %d"),
2202 output_invalid (*l),
2203 i.operands + 1);
2204 return NULL;
2205 }
2206 token_start = l; /* after white space */
2207 paren_not_balanced = 0;
2208 while (paren_not_balanced || *l != ',')
2209 {
2210 if (*l == END_OF_INSN)
2211 {
2212 if (paren_not_balanced)
2213 {
2214 if (!intel_syntax)
2215 as_bad (_("unbalanced parenthesis in operand %d."),
2216 i.operands + 1);
2217 else
2218 as_bad (_("unbalanced brackets in operand %d."),
2219 i.operands + 1);
2220 return NULL;
2221 }
2222 else
2223 break; /* we are done */
2224 }
2225 else if (!is_operand_char (*l) && !is_space_char (*l))
2226 {
2227 as_bad (_("invalid character %s in operand %d"),
2228 output_invalid (*l),
2229 i.operands + 1);
2230 return NULL;
2231 }
2232 if (!intel_syntax)
2233 {
2234 if (*l == '(')
2235 ++paren_not_balanced;
2236 if (*l == ')')
2237 --paren_not_balanced;
2238 }
2239 else
2240 {
2241 if (*l == '[')
2242 ++paren_not_balanced;
2243 if (*l == ']')
2244 --paren_not_balanced;
2245 }
2246 l++;
2247 }
2248 if (l != token_start)
2249 { /* Yes, we've read in another operand. */
2250 unsigned int operand_ok;
2251 this_operand = i.operands++;
2252 if (i.operands > MAX_OPERANDS)
2253 {
2254 as_bad (_("spurious operands; (%d operands/instruction max)"),
2255 MAX_OPERANDS);
2256 return NULL;
2257 }
2258 /* Now parse operand adding info to 'i' as we go along. */
2259 END_STRING_AND_SAVE (l);
2260
2261 if (intel_syntax)
2262 operand_ok =
2263 i386_intel_operand (token_start,
2264 intel_float_operand (mnemonic));
2265 else
2266 operand_ok = i386_operand (token_start);
2267
2268 RESTORE_END_STRING (l);
2269 if (!operand_ok)
2270 return NULL;
2271 }
2272 else
2273 {
2274 if (expecting_operand)
2275 {
2276 expecting_operand_after_comma:
2277 as_bad (_("expecting operand after ','; got nothing"));
2278 return NULL;
2279 }
2280 if (*l == ',')
2281 {
2282 as_bad (_("expecting operand before ','; got nothing"));
2283 return NULL;
2284 }
2285 }
2286
2287 /* Now *l must be either ',' or END_OF_INSN. */
2288 if (*l == ',')
2289 {
2290 if (*++l == END_OF_INSN)
2291 {
2292 /* Just skip it, if it's \n complain. */
2293 goto expecting_operand_after_comma;
2294 }
2295 expecting_operand = 1;
2296 }
2297 }
2298 return l;
2299 }
2300
2301 static void
2302 swap_imm_operands ()
2303 {
2304 union i386_op temp_op;
2305 unsigned int temp_type;
2306 enum bfd_reloc_code_real temp_reloc;
2307 int xchg1 = 0;
2308 int xchg2 = 1;
2309
2310 temp_type = i.types[xchg2];
2311 i.types[xchg2] = i.types[xchg1];
2312 i.types[xchg1] = temp_type;
2313 temp_op = i.op[xchg2];
2314 i.op[xchg2] = i.op[xchg1];
2315 i.op[xchg1] = temp_op;
2316 temp_reloc = i.reloc[xchg2];
2317 i.reloc[xchg2] = i.reloc[xchg1];
2318 i.reloc[xchg1] = temp_reloc;
2319 }
2320
2321
2322 static void
2323 swap_operands ()
2324 {
2325 union i386_op temp_op;
2326 unsigned int temp_type;
2327 enum bfd_reloc_code_real temp_reloc;
2328 int xchg1 = 0;
2329 int xchg2 = 0;
2330
2331 if (i.operands == 4)
2332 /* There will be two exchanges in a 4 operand instruction.
2333 First exchange is the done inside this block.(1st and 4rth operand)
2334 The next exchange is done outside this block.(2nd and 3rd operand) */
2335 {
2336 xchg1 = 0;
2337 xchg2 = 3;
2338 temp_type = i.types[xchg2];
2339 i.types[xchg2] = i.types[xchg1];
2340 i.types[xchg1] = temp_type;
2341 temp_op = i.op[xchg2];
2342 i.op[xchg2] = i.op[xchg1];
2343 i.op[xchg1] = temp_op;
2344 temp_reloc = i.reloc[xchg2];
2345 i.reloc[xchg2] = i.reloc[xchg1];
2346 i.reloc[xchg1] = temp_reloc;
2347 xchg1 = 1;
2348 xchg2 = 2;
2349 }
2350
2351 if (i.operands == 2)
2352 {
2353 xchg1 = 0;
2354 xchg2 = 1;
2355 }
2356 else if (i.operands == 3)
2357 {
2358 xchg1 = 0;
2359 xchg2 = 2;
2360 }
2361 temp_type = i.types[xchg2];
2362 i.types[xchg2] = i.types[xchg1];
2363 i.types[xchg1] = temp_type;
2364 temp_op = i.op[xchg2];
2365 i.op[xchg2] = i.op[xchg1];
2366 i.op[xchg1] = temp_op;
2367 temp_reloc = i.reloc[xchg2];
2368 i.reloc[xchg2] = i.reloc[xchg1];
2369 i.reloc[xchg1] = temp_reloc;
2370
2371 if (i.mem_operands == 2)
2372 {
2373 const seg_entry *temp_seg;
2374 temp_seg = i.seg[0];
2375 i.seg[0] = i.seg[1];
2376 i.seg[1] = temp_seg;
2377 }
2378 }
2379
2380 /* Try to ensure constant immediates are represented in the smallest
2381 opcode possible. */
2382 static void
2383 optimize_imm ()
2384 {
2385 char guess_suffix = 0;
2386 int op;
2387
2388 if (i.suffix)
2389 guess_suffix = i.suffix;
2390 else if (i.reg_operands)
2391 {
2392 /* Figure out a suffix from the last register operand specified.
2393 We can't do this properly yet, ie. excluding InOutPortReg,
2394 but the following works for instructions with immediates.
2395 In any case, we can't set i.suffix yet. */
2396 for (op = i.operands; --op >= 0;)
2397 if (i.types[op] & Reg)
2398 {
2399 if (i.types[op] & Reg8)
2400 guess_suffix = BYTE_MNEM_SUFFIX;
2401 else if (i.types[op] & Reg16)
2402 guess_suffix = WORD_MNEM_SUFFIX;
2403 else if (i.types[op] & Reg32)
2404 guess_suffix = LONG_MNEM_SUFFIX;
2405 else if (i.types[op] & Reg64)
2406 guess_suffix = QWORD_MNEM_SUFFIX;
2407 break;
2408 }
2409 }
2410 else if ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0))
2411 guess_suffix = WORD_MNEM_SUFFIX;
2412
2413 for (op = i.operands; --op >= 0;)
2414 if (i.types[op] & Imm)
2415 {
2416 switch (i.op[op].imms->X_op)
2417 {
2418 case O_constant:
2419 /* If a suffix is given, this operand may be shortened. */
2420 switch (guess_suffix)
2421 {
2422 case LONG_MNEM_SUFFIX:
2423 i.types[op] |= Imm32 | Imm64;
2424 break;
2425 case WORD_MNEM_SUFFIX:
2426 i.types[op] |= Imm16 | Imm32S | Imm32 | Imm64;
2427 break;
2428 case BYTE_MNEM_SUFFIX:
2429 i.types[op] |= Imm16 | Imm8 | Imm8S | Imm32S | Imm32 | Imm64;
2430 break;
2431 }
2432
2433 /* If this operand is at most 16 bits, convert it
2434 to a signed 16 bit number before trying to see
2435 whether it will fit in an even smaller size.
2436 This allows a 16-bit operand such as $0xffe0 to
2437 be recognised as within Imm8S range. */
2438 if ((i.types[op] & Imm16)
2439 && (i.op[op].imms->X_add_number & ~(offsetT) 0xffff) == 0)
2440 {
2441 i.op[op].imms->X_add_number =
2442 (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000);
2443 }
2444 if ((i.types[op] & Imm32)
2445 && ((i.op[op].imms->X_add_number & ~(((offsetT) 2 << 31) - 1))
2446 == 0))
2447 {
2448 i.op[op].imms->X_add_number = ((i.op[op].imms->X_add_number
2449 ^ ((offsetT) 1 << 31))
2450 - ((offsetT) 1 << 31));
2451 }
2452 i.types[op] |= smallest_imm_type (i.op[op].imms->X_add_number);
2453
2454 /* We must avoid matching of Imm32 templates when 64bit
2455 only immediate is available. */
2456 if (guess_suffix == QWORD_MNEM_SUFFIX)
2457 i.types[op] &= ~Imm32;
2458 break;
2459
2460 case O_absent:
2461 case O_register:
2462 abort ();
2463
2464 /* Symbols and expressions. */
2465 default:
2466 /* Convert symbolic operand to proper sizes for matching, but don't
2467 prevent matching a set of insns that only supports sizes other
2468 than those matching the insn suffix. */
2469 {
2470 unsigned int mask, allowed = 0;
2471 const template *t;
2472
2473 for (t = current_templates->start; t < current_templates->end; ++t)
2474 allowed |= t->operand_types[op];
2475 switch (guess_suffix)
2476 {
2477 case QWORD_MNEM_SUFFIX:
2478 mask = Imm64 | Imm32S;
2479 break;
2480 case LONG_MNEM_SUFFIX:
2481 mask = Imm32;
2482 break;
2483 case WORD_MNEM_SUFFIX:
2484 mask = Imm16;
2485 break;
2486 case BYTE_MNEM_SUFFIX:
2487 mask = Imm8;
2488 break;
2489 default:
2490 mask = 0;
2491 break;
2492 }
2493 if (mask & allowed)
2494 i.types[op] &= mask;
2495 }
2496 break;
2497 }
2498 }
2499 }
2500
2501 /* Try to use the smallest displacement type too. */
2502 static void
2503 optimize_disp ()
2504 {
2505 int op;
2506
2507 for (op = i.operands; --op >= 0;)
2508 if (i.types[op] & Disp)
2509 {
2510 if (i.op[op].disps->X_op == O_constant)
2511 {
2512 offsetT disp = i.op[op].disps->X_add_number;
2513
2514 if ((i.types[op] & Disp16)
2515 && (disp & ~(offsetT) 0xffff) == 0)
2516 {
2517 /* If this operand is at most 16 bits, convert
2518 to a signed 16 bit number and don't use 64bit
2519 displacement. */
2520 disp = (((disp & 0xffff) ^ 0x8000) - 0x8000);
2521 i.types[op] &= ~Disp64;
2522 }
2523 if ((i.types[op] & Disp32)
2524 && (disp & ~(((offsetT) 2 << 31) - 1)) == 0)
2525 {
2526 /* If this operand is at most 32 bits, convert
2527 to a signed 32 bit number and don't use 64bit
2528 displacement. */
2529 disp &= (((offsetT) 2 << 31) - 1);
2530 disp = (disp ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
2531 i.types[op] &= ~Disp64;
2532 }
2533 if (!disp && (i.types[op] & BaseIndex))
2534 {
2535 i.types[op] &= ~Disp;
2536 i.op[op].disps = 0;
2537 i.disp_operands--;
2538 }
2539 else if (flag_code == CODE_64BIT)
2540 {
2541 if (fits_in_signed_long (disp))
2542 {
2543 i.types[op] &= ~Disp64;
2544 i.types[op] |= Disp32S;
2545 }
2546 if (fits_in_unsigned_long (disp))
2547 i.types[op] |= Disp32;
2548 }
2549 if ((i.types[op] & (Disp32 | Disp32S | Disp16))
2550 && fits_in_signed_byte (disp))
2551 i.types[op] |= Disp8;
2552 }
2553 else if (i.reloc[op] == BFD_RELOC_386_TLS_DESC_CALL
2554 || i.reloc[op] == BFD_RELOC_X86_64_TLSDESC_CALL)
2555 {
2556 fix_new_exp (frag_now, frag_more (0) - frag_now->fr_literal, 0,
2557 i.op[op].disps, 0, i.reloc[op]);
2558 i.types[op] &= ~Disp;
2559 }
2560 else
2561 /* We only support 64bit displacement on constants. */
2562 i.types[op] &= ~Disp64;
2563 }
2564 }
2565
2566 static int
2567 match_template ()
2568 {
2569 /* Points to template once we've found it. */
2570 const template *t;
2571 unsigned int overlap0, overlap1, overlap2, overlap3;
2572 unsigned int found_reverse_match;
2573 int suffix_check;
2574 unsigned int operand_types [MAX_OPERANDS];
2575 int addr_prefix_disp;
2576
2577 #if MAX_OPERANDS != 4
2578 # error "MAX_OPERANDS must be 4."
2579 #endif
2580
2581 #define MATCH(overlap, given, template) \
2582 ((overlap & ~JumpAbsolute) \
2583 && (((given) & (BaseIndex | JumpAbsolute)) \
2584 == ((overlap) & (BaseIndex | JumpAbsolute))))
2585
2586 /* If given types r0 and r1 are registers they must be of the same type
2587 unless the expected operand type register overlap is null.
2588 Note that Acc in a template matches every size of reg. */
2589 #define CONSISTENT_REGISTER_MATCH(m0, g0, t0, m1, g1, t1) \
2590 (((g0) & Reg) == 0 || ((g1) & Reg) == 0 \
2591 || ((g0) & Reg) == ((g1) & Reg) \
2592 || ((((m0) & Acc) ? Reg : (t0)) & (((m1) & Acc) ? Reg : (t1)) & Reg) == 0 )
2593
2594 overlap0 = 0;
2595 overlap1 = 0;
2596 overlap2 = 0;
2597 overlap3 = 0;
2598 found_reverse_match = 0;
2599 operand_types [0] = 0;
2600 operand_types [1] = 0;
2601 operand_types [2] = 0;
2602 operand_types [3] = 0;
2603 addr_prefix_disp = -1;
2604 suffix_check = (i.suffix == BYTE_MNEM_SUFFIX
2605 ? No_bSuf
2606 : (i.suffix == WORD_MNEM_SUFFIX
2607 ? No_wSuf
2608 : (i.suffix == SHORT_MNEM_SUFFIX
2609 ? No_sSuf
2610 : (i.suffix == LONG_MNEM_SUFFIX
2611 ? No_lSuf
2612 : (i.suffix == QWORD_MNEM_SUFFIX
2613 ? No_qSuf
2614 : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX
2615 ? No_xSuf : 0))))));
2616
2617 for (t = current_templates->start; t < current_templates->end; t++)
2618 {
2619 addr_prefix_disp = -1;
2620
2621 /* Must have right number of operands. */
2622 if (i.operands != t->operands)
2623 continue;
2624
2625 /* Check the suffix, except for some instructions in intel mode. */
2626 if ((t->opcode_modifier & suffix_check)
2627 && !(intel_syntax
2628 && (t->opcode_modifier & IgnoreSize)))
2629 continue;
2630
2631 operand_types [0] = t->operand_types [0];
2632 operand_types [1] = t->operand_types [1];
2633 operand_types [2] = t->operand_types [2];
2634 operand_types [3] = t->operand_types [3];
2635
2636 /* In general, don't allow 64-bit operands in 32-bit mode. */
2637 if (i.suffix == QWORD_MNEM_SUFFIX
2638 && flag_code != CODE_64BIT
2639 && (intel_syntax
2640 ? (!(t->opcode_modifier & IgnoreSize)
2641 && !intel_float_operand (t->name))
2642 : intel_float_operand (t->name) != 2)
2643 && (!(operand_types[0] & (RegMMX | RegXMM))
2644 || !(operand_types[t->operands > 1] & (RegMMX | RegXMM)))
2645 && (t->base_opcode != 0x0fc7
2646 || t->extension_opcode != 1 /* cmpxchg8b */))
2647 continue;
2648
2649 /* Do not verify operands when there are none. */
2650 else if (!t->operands)
2651 {
2652 if (t->cpu_flags & ~cpu_arch_flags)
2653 continue;
2654 /* We've found a match; break out of loop. */
2655 break;
2656 }
2657
2658 /* Address size prefix will turn Disp64/Disp32/Disp16 operand
2659 into Disp32/Disp16/Disp32 operand. */
2660 if (i.prefix[ADDR_PREFIX] != 0)
2661 {
2662 unsigned int j, DispOn = 0, DispOff = 0;
2663
2664 switch (flag_code)
2665 {
2666 case CODE_16BIT:
2667 DispOn = Disp32;
2668 DispOff = Disp16;
2669 break;
2670 case CODE_32BIT:
2671 DispOn = Disp16;
2672 DispOff = Disp32;
2673 break;
2674 case CODE_64BIT:
2675 DispOn = Disp32;
2676 DispOff = Disp64;
2677 break;
2678 }
2679
2680 for (j = 0; j < MAX_OPERANDS; j++)
2681 {
2682 /* There should be only one Disp operand. */
2683 if ((operand_types[j] & DispOff))
2684 {
2685 addr_prefix_disp = j;
2686 operand_types[j] |= DispOn;
2687 operand_types[j] &= ~DispOff;
2688 break;
2689 }
2690 }
2691 }
2692
2693 overlap0 = i.types[0] & operand_types[0];
2694 switch (t->operands)
2695 {
2696 case 1:
2697 if (!MATCH (overlap0, i.types[0], operand_types[0]))
2698 continue;
2699 break;
2700 case 2:
2701 case 3:
2702 case 4:
2703 overlap1 = i.types[1] & operand_types[1];
2704 if (!MATCH (overlap0, i.types[0], operand_types[0])
2705 || !MATCH (overlap1, i.types[1], operand_types[1])
2706 /* monitor in SSE3 is a very special case. The first
2707 register and the second register may have different
2708 sizes. */
2709 || !((t->base_opcode == 0x0f01
2710 && t->extension_opcode == 0xc8)
2711 || CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2712 operand_types[0],
2713 overlap1, i.types[1],
2714 operand_types[1])))
2715 {
2716 /* Check if other direction is valid ... */
2717 if ((t->opcode_modifier & (D | FloatD)) == 0)
2718 continue;
2719
2720 /* Try reversing direction of operands. */
2721 overlap0 = i.types[0] & operand_types[1];
2722 overlap1 = i.types[1] & operand_types[0];
2723 if (!MATCH (overlap0, i.types[0], operand_types[1])
2724 || !MATCH (overlap1, i.types[1], operand_types[0])
2725 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2726 operand_types[1],
2727 overlap1, i.types[1],
2728 operand_types[0]))
2729 {
2730 /* Does not match either direction. */
2731 continue;
2732 }
2733 /* found_reverse_match holds which of D or FloatDR
2734 we've found. */
2735 found_reverse_match = t->opcode_modifier & (D | FloatDR);
2736 }
2737 else
2738 {
2739 /* Found a forward 2 operand match here. */
2740 if (t->operands > 2)
2741 overlap2 = i.types[2] & operand_types[2];
2742 if (t->operands > 3)
2743 overlap3 = i.types[3] & operand_types[3];
2744
2745 switch (t->operands)
2746 {
2747 case 4:
2748 if (!MATCH (overlap3, i.types[3], operand_types[3])
2749 || !CONSISTENT_REGISTER_MATCH (overlap2,
2750 i.types[2],
2751 operand_types[2],
2752 overlap3,
2753 i.types[3],
2754 operand_types[3]))
2755 continue;
2756 case 3:
2757 /* Here we make use of the fact that there are no
2758 reverse match 3 operand instructions, and all 3
2759 operand instructions only need to be checked for
2760 register consistency between operands 2 and 3. */
2761 if (!MATCH (overlap2, i.types[2], operand_types[2])
2762 || !CONSISTENT_REGISTER_MATCH (overlap1,
2763 i.types[1],
2764 operand_types[1],
2765 overlap2,
2766 i.types[2],
2767 operand_types[2]))
2768 continue;
2769 break;
2770 }
2771 }
2772 /* Found either forward/reverse 2, 3 or 4 operand match here:
2773 slip through to break. */
2774 }
2775 if (t->cpu_flags & ~cpu_arch_flags)
2776 {
2777 found_reverse_match = 0;
2778 continue;
2779 }
2780 /* We've found a match; break out of loop. */
2781 break;
2782 }
2783
2784 if (t == current_templates->end)
2785 {
2786 /* We found no match. */
2787 as_bad (_("suffix or operands invalid for `%s'"),
2788 current_templates->start->name);
2789 return 0;
2790 }
2791
2792 if (!quiet_warnings)
2793 {
2794 if (!intel_syntax
2795 && ((i.types[0] & JumpAbsolute)
2796 != (operand_types[0] & JumpAbsolute)))
2797 {
2798 as_warn (_("indirect %s without `*'"), t->name);
2799 }
2800
2801 if ((t->opcode_modifier & (IsPrefix | IgnoreSize))
2802 == (IsPrefix | IgnoreSize))
2803 {
2804 /* Warn them that a data or address size prefix doesn't
2805 affect assembly of the next line of code. */
2806 as_warn (_("stand-alone `%s' prefix"), t->name);
2807 }
2808 }
2809
2810 /* Copy the template we found. */
2811 i.tm = *t;
2812
2813 if (addr_prefix_disp != -1)
2814 i.tm.operand_types[addr_prefix_disp]
2815 = operand_types[addr_prefix_disp];
2816
2817 if (found_reverse_match)
2818 {
2819 /* If we found a reverse match we must alter the opcode
2820 direction bit. found_reverse_match holds bits to change
2821 (different for int & float insns). */
2822
2823 i.tm.base_opcode ^= found_reverse_match;
2824
2825 i.tm.operand_types[0] = operand_types[1];
2826 i.tm.operand_types[1] = operand_types[0];
2827 }
2828
2829 return 1;
2830 }
2831
2832 static int
2833 check_string ()
2834 {
2835 int mem_op = (i.types[0] & AnyMem) ? 0 : 1;
2836 if ((i.tm.operand_types[mem_op] & EsSeg) != 0)
2837 {
2838 if (i.seg[0] != NULL && i.seg[0] != &es)
2839 {
2840 as_bad (_("`%s' operand %d must use `%%es' segment"),
2841 i.tm.name,
2842 mem_op + 1);
2843 return 0;
2844 }
2845 /* There's only ever one segment override allowed per instruction.
2846 This instruction possibly has a legal segment override on the
2847 second operand, so copy the segment to where non-string
2848 instructions store it, allowing common code. */
2849 i.seg[0] = i.seg[1];
2850 }
2851 else if ((i.tm.operand_types[mem_op + 1] & EsSeg) != 0)
2852 {
2853 if (i.seg[1] != NULL && i.seg[1] != &es)
2854 {
2855 as_bad (_("`%s' operand %d must use `%%es' segment"),
2856 i.tm.name,
2857 mem_op + 2);
2858 return 0;
2859 }
2860 }
2861 return 1;
2862 }
2863
2864 static int
2865 process_suffix (void)
2866 {
2867 /* If matched instruction specifies an explicit instruction mnemonic
2868 suffix, use it. */
2869 if (i.tm.opcode_modifier & (Size16 | Size32 | Size64))
2870 {
2871 if (i.tm.opcode_modifier & Size16)
2872 i.suffix = WORD_MNEM_SUFFIX;
2873 else if (i.tm.opcode_modifier & Size64)
2874 i.suffix = QWORD_MNEM_SUFFIX;
2875 else
2876 i.suffix = LONG_MNEM_SUFFIX;
2877 }
2878 else if (i.reg_operands)
2879 {
2880 /* If there's no instruction mnemonic suffix we try to invent one
2881 based on register operands. */
2882 if (!i.suffix)
2883 {
2884 /* We take i.suffix from the last register operand specified,
2885 Destination register type is more significant than source
2886 register type. */
2887 int op;
2888
2889 for (op = i.operands; --op >= 0;)
2890 if ((i.types[op] & Reg)
2891 && !(i.tm.operand_types[op] & InOutPortReg))
2892 {
2893 i.suffix = ((i.types[op] & Reg8) ? BYTE_MNEM_SUFFIX :
2894 (i.types[op] & Reg16) ? WORD_MNEM_SUFFIX :
2895 (i.types[op] & Reg64) ? QWORD_MNEM_SUFFIX :
2896 LONG_MNEM_SUFFIX);
2897 break;
2898 }
2899 }
2900 else if (i.suffix == BYTE_MNEM_SUFFIX)
2901 {
2902 if (!check_byte_reg ())
2903 return 0;
2904 }
2905 else if (i.suffix == LONG_MNEM_SUFFIX)
2906 {
2907 if (!check_long_reg ())
2908 return 0;
2909 }
2910 else if (i.suffix == QWORD_MNEM_SUFFIX)
2911 {
2912 if (!check_qword_reg ())
2913 return 0;
2914 }
2915 else if (i.suffix == WORD_MNEM_SUFFIX)
2916 {
2917 if (!check_word_reg ())
2918 return 0;
2919 }
2920 else if (intel_syntax && (i.tm.opcode_modifier & IgnoreSize))
2921 /* Do nothing if the instruction is going to ignore the prefix. */
2922 ;
2923 else
2924 abort ();
2925 }
2926 else if ((i.tm.opcode_modifier & DefaultSize)
2927 && !i.suffix
2928 /* exclude fldenv/frstor/fsave/fstenv */
2929 && (i.tm.opcode_modifier & No_sSuf))
2930 {
2931 i.suffix = stackop_size;
2932 }
2933 else if (intel_syntax
2934 && !i.suffix
2935 && ((i.tm.operand_types[0] & JumpAbsolute)
2936 || (i.tm.opcode_modifier & (JumpByte|JumpInterSegment))
2937 || (i.tm.base_opcode == 0x0f01 /* [ls][gi]dt */
2938 && i.tm.extension_opcode <= 3)))
2939 {
2940 switch (flag_code)
2941 {
2942 case CODE_64BIT:
2943 if (!(i.tm.opcode_modifier & No_qSuf))
2944 {
2945 i.suffix = QWORD_MNEM_SUFFIX;
2946 break;
2947 }
2948 case CODE_32BIT:
2949 if (!(i.tm.opcode_modifier & No_lSuf))
2950 i.suffix = LONG_MNEM_SUFFIX;
2951 break;
2952 case CODE_16BIT:
2953 if (!(i.tm.opcode_modifier & No_wSuf))
2954 i.suffix = WORD_MNEM_SUFFIX;
2955 break;
2956 }
2957 }
2958
2959 if (!i.suffix)
2960 {
2961 if (!intel_syntax)
2962 {
2963 if (i.tm.opcode_modifier & W)
2964 {
2965 as_bad (_("no instruction mnemonic suffix given and no register operands; can't size instruction"));
2966 return 0;
2967 }
2968 }
2969 else
2970 {
2971 unsigned int suffixes = (~i.tm.opcode_modifier
2972 & (No_bSuf
2973 | No_wSuf
2974 | No_lSuf
2975 | No_sSuf
2976 | No_xSuf
2977 | No_qSuf));
2978
2979 if ((i.tm.opcode_modifier & W)
2980 || ((suffixes & (suffixes - 1))
2981 && !(i.tm.opcode_modifier & (DefaultSize | IgnoreSize))))
2982 {
2983 as_bad (_("ambiguous operand size for `%s'"), i.tm.name);
2984 return 0;
2985 }
2986 }
2987 }
2988
2989 /* Change the opcode based on the operand size given by i.suffix;
2990 We don't need to change things for byte insns. */
2991
2992 if (i.suffix && i.suffix != BYTE_MNEM_SUFFIX)
2993 {
2994 /* It's not a byte, select word/dword operation. */
2995 if (i.tm.opcode_modifier & W)
2996 {
2997 if (i.tm.opcode_modifier & ShortForm)
2998 i.tm.base_opcode |= 8;
2999 else
3000 i.tm.base_opcode |= 1;
3001 }
3002
3003 /* Now select between word & dword operations via the operand
3004 size prefix, except for instructions that will ignore this
3005 prefix anyway. */
3006 if (i.tm.base_opcode == 0x0f01 && i.tm.extension_opcode == 0xc8)
3007 {
3008 /* monitor in SSE3 is a very special case. The default size
3009 of AX is the size of mode. The address size override
3010 prefix will change the size of AX. */
3011 if (i.op->regs[0].reg_type &
3012 (flag_code == CODE_32BIT ? Reg16 : Reg32))
3013 if (!add_prefix (ADDR_PREFIX_OPCODE))
3014 return 0;
3015 }
3016 else if (i.suffix != QWORD_MNEM_SUFFIX
3017 && i.suffix != LONG_DOUBLE_MNEM_SUFFIX
3018 && !(i.tm.opcode_modifier & (IgnoreSize | FloatMF))
3019 && ((i.suffix == LONG_MNEM_SUFFIX) == (flag_code == CODE_16BIT)
3020 || (flag_code == CODE_64BIT
3021 && (i.tm.opcode_modifier & JumpByte))))
3022 {
3023 unsigned int prefix = DATA_PREFIX_OPCODE;
3024
3025 if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */
3026 prefix = ADDR_PREFIX_OPCODE;
3027
3028 if (!add_prefix (prefix))
3029 return 0;
3030 }
3031
3032 /* Set mode64 for an operand. */
3033 if (i.suffix == QWORD_MNEM_SUFFIX
3034 && flag_code == CODE_64BIT
3035 && (i.tm.opcode_modifier & NoRex64) == 0)
3036 {
3037 /* Special case for xchg %rax,%rax. It is NOP and doesn't
3038 need rex64. */
3039 if (i.operands != 2
3040 || i.types [0] != (Acc | Reg64)
3041 || i.types [1] != (Acc | Reg64)
3042 || strcmp (i.tm.name, "xchg") != 0)
3043 i.rex |= REX_MODE64;
3044 }
3045
3046 /* Size floating point instruction. */
3047 if (i.suffix == LONG_MNEM_SUFFIX)
3048 if (i.tm.opcode_modifier & FloatMF)
3049 i.tm.base_opcode ^= 4;
3050 }
3051
3052 return 1;
3053 }
3054
3055 static int
3056 check_byte_reg (void)
3057 {
3058 int op;
3059
3060 for (op = i.operands; --op >= 0;)
3061 {
3062 /* If this is an eight bit register, it's OK. If it's the 16 or
3063 32 bit version of an eight bit register, we will just use the
3064 low portion, and that's OK too. */
3065 if (i.types[op] & Reg8)
3066 continue;
3067
3068 /* movzx and movsx should not generate this warning. */
3069 if (intel_syntax
3070 && (i.tm.base_opcode == 0xfb7
3071 || i.tm.base_opcode == 0xfb6
3072 || i.tm.base_opcode == 0x63
3073 || i.tm.base_opcode == 0xfbe
3074 || i.tm.base_opcode == 0xfbf))
3075 continue;
3076
3077 if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4)
3078 {
3079 /* Prohibit these changes in the 64bit mode, since the
3080 lowering is more complicated. */
3081 if (flag_code == CODE_64BIT
3082 && (i.tm.operand_types[op] & InOutPortReg) == 0)
3083 {
3084 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
3085 i.op[op].regs->reg_name,
3086 i.suffix);
3087 return 0;
3088 }
3089 #if REGISTER_WARNINGS
3090 if (!quiet_warnings
3091 && (i.tm.operand_types[op] & InOutPortReg) == 0)
3092 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
3093 (i.op[op].regs + (i.types[op] & Reg16
3094 ? REGNAM_AL - REGNAM_AX
3095 : REGNAM_AL - REGNAM_EAX))->reg_name,
3096 i.op[op].regs->reg_name,
3097 i.suffix);
3098 #endif
3099 continue;
3100 }
3101 /* Any other register is bad. */
3102 if (i.types[op] & (Reg | RegMMX | RegXMM
3103 | SReg2 | SReg3
3104 | Control | Debug | Test
3105 | FloatReg | FloatAcc))
3106 {
3107 as_bad (_("`%%%s' not allowed with `%s%c'"),
3108 i.op[op].regs->reg_name,
3109 i.tm.name,
3110 i.suffix);
3111 return 0;
3112 }
3113 }
3114 return 1;
3115 }
3116
3117 static int
3118 check_long_reg ()
3119 {
3120 int op;
3121
3122 for (op = i.operands; --op >= 0;)
3123 /* Reject eight bit registers, except where the template requires
3124 them. (eg. movzb) */
3125 if ((i.types[op] & Reg8) != 0
3126 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
3127 {
3128 as_bad (_("`%%%s' not allowed with `%s%c'"),
3129 i.op[op].regs->reg_name,
3130 i.tm.name,
3131 i.suffix);
3132 return 0;
3133 }
3134 /* Warn if the e prefix on a general reg is missing. */
3135 else if ((!quiet_warnings || flag_code == CODE_64BIT)
3136 && (i.types[op] & Reg16) != 0
3137 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
3138 {
3139 /* Prohibit these changes in the 64bit mode, since the
3140 lowering is more complicated. */
3141 if (flag_code == CODE_64BIT)
3142 {
3143 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
3144 i.op[op].regs->reg_name,
3145 i.suffix);
3146 return 0;
3147 }
3148 #if REGISTER_WARNINGS
3149 else
3150 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
3151 (i.op[op].regs + REGNAM_EAX - REGNAM_AX)->reg_name,
3152 i.op[op].regs->reg_name,
3153 i.suffix);
3154 #endif
3155 }
3156 /* Warn if the r prefix on a general reg is missing. */
3157 else if ((i.types[op] & Reg64) != 0
3158 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
3159 {
3160 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
3161 i.op[op].regs->reg_name,
3162 i.suffix);
3163 return 0;
3164 }
3165 return 1;
3166 }
3167
3168 static int
3169 check_qword_reg ()
3170 {
3171 int op;
3172
3173 for (op = i.operands; --op >= 0; )
3174 /* Reject eight bit registers, except where the template requires
3175 them. (eg. movzb) */
3176 if ((i.types[op] & Reg8) != 0
3177 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
3178 {
3179 as_bad (_("`%%%s' not allowed with `%s%c'"),
3180 i.op[op].regs->reg_name,
3181 i.tm.name,
3182 i.suffix);
3183 return 0;
3184 }
3185 /* Warn if the e prefix on a general reg is missing. */
3186 else if (((i.types[op] & Reg16) != 0
3187 || (i.types[op] & Reg32) != 0)
3188 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
3189 {
3190 /* Prohibit these changes in the 64bit mode, since the
3191 lowering is more complicated. */
3192 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
3193 i.op[op].regs->reg_name,
3194 i.suffix);
3195 return 0;
3196 }
3197 return 1;
3198 }
3199
3200 static int
3201 check_word_reg ()
3202 {
3203 int op;
3204 for (op = i.operands; --op >= 0;)
3205 /* Reject eight bit registers, except where the template requires
3206 them. (eg. movzb) */
3207 if ((i.types[op] & Reg8) != 0
3208 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
3209 {
3210 as_bad (_("`%%%s' not allowed with `%s%c'"),
3211 i.op[op].regs->reg_name,
3212 i.tm.name,
3213 i.suffix);
3214 return 0;
3215 }
3216 /* Warn if the e prefix on a general reg is present. */
3217 else if ((!quiet_warnings || flag_code == CODE_64BIT)
3218 && (i.types[op] & Reg32) != 0
3219 && (i.tm.operand_types[op] & (Reg16 | Acc)) != 0)
3220 {
3221 /* Prohibit these changes in the 64bit mode, since the
3222 lowering is more complicated. */
3223 if (flag_code == CODE_64BIT)
3224 {
3225 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
3226 i.op[op].regs->reg_name,
3227 i.suffix);
3228 return 0;
3229 }
3230 else
3231 #if REGISTER_WARNINGS
3232 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
3233 (i.op[op].regs + REGNAM_AX - REGNAM_EAX)->reg_name,
3234 i.op[op].regs->reg_name,
3235 i.suffix);
3236 #endif
3237 }
3238 return 1;
3239 }
3240
3241 static int
3242 finalize_imm ()
3243 {
3244 unsigned int overlap0, overlap1, overlap2;
3245
3246 overlap0 = i.types[0] & i.tm.operand_types[0];
3247 if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64))
3248 && overlap0 != Imm8 && overlap0 != Imm8S
3249 && overlap0 != Imm16 && overlap0 != Imm32S
3250 && overlap0 != Imm32 && overlap0 != Imm64)
3251 {
3252 if (i.suffix)
3253 {
3254 overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX
3255 ? Imm8 | Imm8S
3256 : (i.suffix == WORD_MNEM_SUFFIX
3257 ? Imm16
3258 : (i.suffix == QWORD_MNEM_SUFFIX
3259 ? Imm64 | Imm32S
3260 : Imm32)));
3261 }
3262 else if (overlap0 == (Imm16 | Imm32S | Imm32)
3263 || overlap0 == (Imm16 | Imm32)
3264 || overlap0 == (Imm16 | Imm32S))
3265 {
3266 overlap0 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
3267 ? Imm16 : Imm32S);
3268 }
3269 if (overlap0 != Imm8 && overlap0 != Imm8S
3270 && overlap0 != Imm16 && overlap0 != Imm32S
3271 && overlap0 != Imm32 && overlap0 != Imm64)
3272 {
3273 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size"));
3274 return 0;
3275 }
3276 }
3277 i.types[0] = overlap0;
3278
3279 overlap1 = i.types[1] & i.tm.operand_types[1];
3280 if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32 | Imm64))
3281 && overlap1 != Imm8 && overlap1 != Imm8S
3282 && overlap1 != Imm16 && overlap1 != Imm32S
3283 && overlap1 != Imm32 && overlap1 != Imm64)
3284 {
3285 if (i.suffix)
3286 {
3287 overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX
3288 ? Imm8 | Imm8S
3289 : (i.suffix == WORD_MNEM_SUFFIX
3290 ? Imm16
3291 : (i.suffix == QWORD_MNEM_SUFFIX
3292 ? Imm64 | Imm32S
3293 : Imm32)));
3294 }
3295 else if (overlap1 == (Imm16 | Imm32 | Imm32S)
3296 || overlap1 == (Imm16 | Imm32)
3297 || overlap1 == (Imm16 | Imm32S))
3298 {
3299 overlap1 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
3300 ? Imm16 : Imm32S);
3301 }
3302 if (overlap1 != Imm8 && overlap1 != Imm8S
3303 && overlap1 != Imm16 && overlap1 != Imm32S
3304 && overlap1 != Imm32 && overlap1 != Imm64)
3305 {
3306 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size %x %c"),overlap1, i.suffix);
3307 return 0;
3308 }
3309 }
3310 i.types[1] = overlap1;
3311
3312 overlap2 = i.types[2] & i.tm.operand_types[2];
3313 assert ((overlap2 & Imm) == 0);
3314 i.types[2] = overlap2;
3315
3316 return 1;
3317 }
3318
3319 static int
3320 process_operands ()
3321 {
3322 /* Default segment register this instruction will use for memory
3323 accesses. 0 means unknown. This is only for optimizing out
3324 unnecessary segment overrides. */
3325 const seg_entry *default_seg = 0;
3326
3327 /* The imul $imm, %reg instruction is converted into
3328 imul $imm, %reg, %reg, and the clr %reg instruction
3329 is converted into xor %reg, %reg. */
3330 if (i.tm.opcode_modifier & regKludge)
3331 {
3332 unsigned int first_reg_op = (i.types[0] & Reg) ? 0 : 1;
3333 /* Pretend we saw the extra register operand. */
3334 assert (i.op[first_reg_op + 1].regs == 0);
3335 i.op[first_reg_op + 1].regs = i.op[first_reg_op].regs;
3336 i.types[first_reg_op + 1] = i.types[first_reg_op];
3337 i.reg_operands = 2;
3338 }
3339
3340 if (i.tm.opcode_modifier & ShortForm)
3341 {
3342 /* The register or float register operand is in operand 0 or 1. */
3343 unsigned int op = (i.types[0] & (Reg | FloatReg)) ? 0 : 1;
3344 /* Register goes in low 3 bits of opcode. */
3345 i.tm.base_opcode |= i.op[op].regs->reg_num;
3346 if ((i.op[op].regs->reg_flags & RegRex) != 0)
3347 i.rex |= REX_EXTZ;
3348 if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
3349 {
3350 /* Warn about some common errors, but press on regardless.
3351 The first case can be generated by gcc (<= 2.8.1). */
3352 if (i.operands == 2)
3353 {
3354 /* Reversed arguments on faddp, fsubp, etc. */
3355 as_warn (_("translating to `%s %%%s,%%%s'"), i.tm.name,
3356 i.op[1].regs->reg_name,
3357 i.op[0].regs->reg_name);
3358 }
3359 else
3360 {
3361 /* Extraneous `l' suffix on fp insn. */
3362 as_warn (_("translating to `%s %%%s'"), i.tm.name,
3363 i.op[0].regs->reg_name);
3364 }
3365 }
3366 }
3367 else if (i.tm.opcode_modifier & Modrm)
3368 {
3369 /* The opcode is completed (modulo i.tm.extension_opcode which
3370 must be put into the modrm byte). Now, we make the modrm and
3371 index base bytes based on all the info we've collected. */
3372
3373 default_seg = build_modrm_byte ();
3374 }
3375 else if (i.tm.opcode_modifier & (Seg2ShortForm | Seg3ShortForm))
3376 {
3377 if (i.tm.base_opcode == POP_SEG_SHORT
3378 && i.op[0].regs->reg_num == 1)
3379 {
3380 as_bad (_("you can't `pop %%cs'"));
3381 return 0;
3382 }
3383 i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
3384 if ((i.op[0].regs->reg_flags & RegRex) != 0)
3385 i.rex |= REX_EXTZ;
3386 }
3387 else if ((i.tm.base_opcode & ~(D | W)) == MOV_AX_DISP32)
3388 {
3389 default_seg = &ds;
3390 }
3391 else if ((i.tm.opcode_modifier & IsString) != 0)
3392 {
3393 /* For the string instructions that allow a segment override
3394 on one of their operands, the default segment is ds. */
3395 default_seg = &ds;
3396 }
3397
3398 if ((i.tm.base_opcode == 0x8d /* lea */
3399 || (i.tm.cpu_flags & CpuSVME))
3400 && i.seg[0] && !quiet_warnings)
3401 as_warn (_("segment override on `%s' is ineffectual"), i.tm.name);
3402
3403 /* If a segment was explicitly specified, and the specified segment
3404 is not the default, use an opcode prefix to select it. If we
3405 never figured out what the default segment is, then default_seg
3406 will be zero at this point, and the specified segment prefix will
3407 always be used. */
3408 if ((i.seg[0]) && (i.seg[0] != default_seg))
3409 {
3410 if (!add_prefix (i.seg[0]->seg_prefix))
3411 return 0;
3412 }
3413 return 1;
3414 }
3415
3416 static const seg_entry *
3417 build_modrm_byte ()
3418 {
3419 const seg_entry *default_seg = 0;
3420
3421 /* i.reg_operands MUST be the number of real register operands;
3422 implicit registers do not count. */
3423 if (i.reg_operands == 2)
3424 {
3425 unsigned int source, dest;
3426 source = ((i.types[0]
3427 & (Reg | RegMMX | RegXMM
3428 | SReg2 | SReg3
3429 | Control | Debug | Test))
3430 ? 0 : 1);
3431
3432 /* In 4 operands instructions with 2 immediate operands, the first two are immediate
3433 bytes and hence source operand will be in the next byte after the immediates */
3434 if ((i.operands == 4)&&(i.imm_operands=2)) source++;
3435 dest = source + 1;
3436
3437 i.rm.mode = 3;
3438 /* One of the register operands will be encoded in the i.tm.reg
3439 field, the other in the combined i.tm.mode and i.tm.regmem
3440 fields. If no form of this instruction supports a memory
3441 destination operand, then we assume the source operand may
3442 sometimes be a memory operand and so we need to store the
3443 destination in the i.rm.reg field. */
3444 if ((i.tm.operand_types[dest] & AnyMem) == 0)
3445 {
3446 i.rm.reg = i.op[dest].regs->reg_num;
3447 i.rm.regmem = i.op[source].regs->reg_num;
3448 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
3449 i.rex |= REX_EXTX;
3450 if ((i.op[source].regs->reg_flags & RegRex) != 0)
3451 i.rex |= REX_EXTZ;
3452 }
3453 else
3454 {
3455 i.rm.reg = i.op[source].regs->reg_num;
3456 i.rm.regmem = i.op[dest].regs->reg_num;
3457 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
3458 i.rex |= REX_EXTZ;
3459 if ((i.op[source].regs->reg_flags & RegRex) != 0)
3460 i.rex |= REX_EXTX;
3461 }
3462 if (flag_code != CODE_64BIT && (i.rex & (REX_EXTX | REX_EXTZ)))
3463 {
3464 if (!((i.types[0] | i.types[1]) & Control))
3465 abort ();
3466 i.rex &= ~(REX_EXTX | REX_EXTZ);
3467 add_prefix (LOCK_PREFIX_OPCODE);
3468 }
3469 }
3470 else
3471 { /* If it's not 2 reg operands... */
3472 if (i.mem_operands)
3473 {
3474 unsigned int fake_zero_displacement = 0;
3475 unsigned int op = ((i.types[0] & AnyMem)
3476 ? 0
3477 : (i.types[1] & AnyMem) ? 1 : 2);
3478
3479 default_seg = &ds;
3480
3481 if (i.base_reg == 0)
3482 {
3483 i.rm.mode = 0;
3484 if (!i.disp_operands)
3485 fake_zero_displacement = 1;
3486 if (i.index_reg == 0)
3487 {
3488 /* Operand is just <disp> */
3489 if (flag_code == CODE_64BIT)
3490 {
3491 /* 64bit mode overwrites the 32bit absolute
3492 addressing by RIP relative addressing and
3493 absolute addressing is encoded by one of the
3494 redundant SIB forms. */
3495 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3496 i.sib.base = NO_BASE_REGISTER;
3497 i.sib.index = NO_INDEX_REGISTER;
3498 i.types[op] = ((i.prefix[ADDR_PREFIX] == 0) ? Disp32S : Disp32);
3499 }
3500 else if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3501 {
3502 i.rm.regmem = NO_BASE_REGISTER_16;
3503 i.types[op] = Disp16;
3504 }
3505 else
3506 {
3507 i.rm.regmem = NO_BASE_REGISTER;
3508 i.types[op] = Disp32;
3509 }
3510 }
3511 else /* !i.base_reg && i.index_reg */
3512 {
3513 i.sib.index = i.index_reg->reg_num;
3514 i.sib.base = NO_BASE_REGISTER;
3515 i.sib.scale = i.log2_scale_factor;
3516 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3517 i.types[op] &= ~Disp;
3518 if (flag_code != CODE_64BIT)
3519 i.types[op] |= Disp32; /* Must be 32 bit */
3520 else
3521 i.types[op] |= Disp32S;
3522 if ((i.index_reg->reg_flags & RegRex) != 0)
3523 i.rex |= REX_EXTY;
3524 }
3525 }
3526 /* RIP addressing for 64bit mode. */
3527 else if (i.base_reg->reg_type == BaseIndex)
3528 {
3529 i.rm.regmem = NO_BASE_REGISTER;
3530 i.types[op] &= ~ Disp;
3531 i.types[op] |= Disp32S;
3532 i.flags[op] = Operand_PCrel;
3533 if (! i.disp_operands)
3534 fake_zero_displacement = 1;
3535 }
3536 else if (i.base_reg->reg_type & Reg16)
3537 {
3538 switch (i.base_reg->reg_num)
3539 {
3540 case 3: /* (%bx) */
3541 if (i.index_reg == 0)
3542 i.rm.regmem = 7;
3543 else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */
3544 i.rm.regmem = i.index_reg->reg_num - 6;
3545 break;
3546 case 5: /* (%bp) */
3547 default_seg = &ss;
3548 if (i.index_reg == 0)
3549 {
3550 i.rm.regmem = 6;
3551 if ((i.types[op] & Disp) == 0)
3552 {
3553 /* fake (%bp) into 0(%bp) */
3554 i.types[op] |= Disp8;
3555 fake_zero_displacement = 1;
3556 }
3557 }
3558 else /* (%bp,%si) -> 2, or (%bp,%di) -> 3 */
3559 i.rm.regmem = i.index_reg->reg_num - 6 + 2;
3560 break;
3561 default: /* (%si) -> 4 or (%di) -> 5 */
3562 i.rm.regmem = i.base_reg->reg_num - 6 + 4;
3563 }
3564 i.rm.mode = mode_from_disp_size (i.types[op]);
3565 }
3566 else /* i.base_reg and 32/64 bit mode */
3567 {
3568 if (flag_code == CODE_64BIT
3569 && (i.types[op] & Disp))
3570 i.types[op] = (i.types[op] & Disp8) | (i.prefix[ADDR_PREFIX] == 0 ? Disp32S : Disp32);
3571
3572 i.rm.regmem = i.base_reg->reg_num;
3573 if ((i.base_reg->reg_flags & RegRex) != 0)
3574 i.rex |= REX_EXTZ;
3575 i.sib.base = i.base_reg->reg_num;
3576 /* x86-64 ignores REX prefix bit here to avoid decoder
3577 complications. */
3578 if ((i.base_reg->reg_num & 7) == EBP_REG_NUM)
3579 {
3580 default_seg = &ss;
3581 if (i.disp_operands == 0)
3582 {
3583 fake_zero_displacement = 1;
3584 i.types[op] |= Disp8;
3585 }
3586 }
3587 else if (i.base_reg->reg_num == ESP_REG_NUM)
3588 {
3589 default_seg = &ss;
3590 }
3591 i.sib.scale = i.log2_scale_factor;
3592 if (i.index_reg == 0)
3593 {
3594 /* <disp>(%esp) becomes two byte modrm with no index
3595 register. We've already stored the code for esp
3596 in i.rm.regmem ie. ESCAPE_TO_TWO_BYTE_ADDRESSING.
3597 Any base register besides %esp will not use the
3598 extra modrm byte. */
3599 i.sib.index = NO_INDEX_REGISTER;
3600 #if !SCALE1_WHEN_NO_INDEX
3601 /* Another case where we force the second modrm byte. */
3602 if (i.log2_scale_factor)
3603 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3604 #endif
3605 }
3606 else
3607 {
3608 i.sib.index = i.index_reg->reg_num;
3609 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3610 if ((i.index_reg->reg_flags & RegRex) != 0)
3611 i.rex |= REX_EXTY;
3612 }
3613
3614 if (i.disp_operands
3615 && (i.reloc[op] == BFD_RELOC_386_TLS_DESC_CALL
3616 || i.reloc[op] == BFD_RELOC_X86_64_TLSDESC_CALL))
3617 i.rm.mode = 0;
3618 else
3619 i.rm.mode = mode_from_disp_size (i.types[op]);
3620 }
3621
3622 if (fake_zero_displacement)
3623 {
3624 /* Fakes a zero displacement assuming that i.types[op]
3625 holds the correct displacement size. */
3626 expressionS *exp;
3627
3628 assert (i.op[op].disps == 0);
3629 exp = &disp_expressions[i.disp_operands++];
3630 i.op[op].disps = exp;
3631 exp->X_op = O_constant;
3632 exp->X_add_number = 0;
3633 exp->X_add_symbol = (symbolS *) 0;
3634 exp->X_op_symbol = (symbolS *) 0;
3635 }
3636 }
3637
3638 /* Fill in i.rm.reg or i.rm.regmem field with register operand
3639 (if any) based on i.tm.extension_opcode. Again, we must be
3640 careful to make sure that segment/control/debug/test/MMX
3641 registers are coded into the i.rm.reg field. */
3642 if (i.reg_operands)
3643 {
3644 unsigned int op =
3645 ((i.types[0]
3646 & (Reg | RegMMX | RegXMM
3647 | SReg2 | SReg3
3648 | Control | Debug | Test))
3649 ? 0
3650 : ((i.types[1]
3651 & (Reg | RegMMX | RegXMM
3652 | SReg2 | SReg3
3653 | Control | Debug | Test))
3654 ? 1
3655 : 2));
3656 /* If there is an extension opcode to put here, the register
3657 number must be put into the regmem field. */
3658 if (i.tm.extension_opcode != None)
3659 {
3660 i.rm.regmem = i.op[op].regs->reg_num;
3661 if ((i.op[op].regs->reg_flags & RegRex) != 0)
3662 i.rex |= REX_EXTZ;
3663 }
3664 else
3665 {
3666 i.rm.reg = i.op[op].regs->reg_num;
3667 if ((i.op[op].regs->reg_flags & RegRex) != 0)
3668 i.rex |= REX_EXTX;
3669 }
3670
3671 /* Now, if no memory operand has set i.rm.mode = 0, 1, 2 we
3672 must set it to 3 to indicate this is a register operand
3673 in the regmem field. */
3674 if (!i.mem_operands)
3675 i.rm.mode = 3;
3676 }
3677
3678 /* Fill in i.rm.reg field with extension opcode (if any). */
3679 if (i.tm.extension_opcode != None)
3680 i.rm.reg = i.tm.extension_opcode;
3681 }
3682 return default_seg;
3683 }
3684
3685 static void
3686 output_branch ()
3687 {
3688 char *p;
3689 int code16;
3690 int prefix;
3691 relax_substateT subtype;
3692 symbolS *sym;
3693 offsetT off;
3694
3695 code16 = 0;
3696 if (flag_code == CODE_16BIT)
3697 code16 = CODE16;
3698
3699 prefix = 0;
3700 if (i.prefix[DATA_PREFIX] != 0)
3701 {
3702 prefix = 1;
3703 i.prefixes -= 1;
3704 code16 ^= CODE16;
3705 }
3706 /* Pentium4 branch hints. */
3707 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
3708 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
3709 {
3710 prefix++;
3711 i.prefixes--;
3712 }
3713 if (i.prefix[REX_PREFIX] != 0)
3714 {
3715 prefix++;
3716 i.prefixes--;
3717 }
3718
3719 if (i.prefixes != 0 && !intel_syntax)
3720 as_warn (_("skipping prefixes on this instruction"));
3721
3722 /* It's always a symbol; End frag & setup for relax.
3723 Make sure there is enough room in this frag for the largest
3724 instruction we may generate in md_convert_frag. This is 2
3725 bytes for the opcode and room for the prefix and largest
3726 displacement. */
3727 frag_grow (prefix + 2 + 4);
3728 /* Prefix and 1 opcode byte go in fr_fix. */
3729 p = frag_more (prefix + 1);
3730 if (i.prefix[DATA_PREFIX] != 0)
3731 *p++ = DATA_PREFIX_OPCODE;
3732 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE
3733 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE)
3734 *p++ = i.prefix[SEG_PREFIX];
3735 if (i.prefix[REX_PREFIX] != 0)
3736 *p++ = i.prefix[REX_PREFIX];
3737 *p = i.tm.base_opcode;
3738
3739 if ((unsigned char) *p == JUMP_PC_RELATIVE)
3740 subtype = ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL);
3741 else if ((cpu_arch_flags & Cpu386) != 0)
3742 subtype = ENCODE_RELAX_STATE (COND_JUMP, SMALL);
3743 else
3744 subtype = ENCODE_RELAX_STATE (COND_JUMP86, SMALL);
3745 subtype |= code16;
3746
3747 sym = i.op[0].disps->X_add_symbol;
3748 off = i.op[0].disps->X_add_number;
3749
3750 if (i.op[0].disps->X_op != O_constant
3751 && i.op[0].disps->X_op != O_symbol)
3752 {
3753 /* Handle complex expressions. */
3754 sym = make_expr_symbol (i.op[0].disps);
3755 off = 0;
3756 }
3757
3758 /* 1 possible extra opcode + 4 byte displacement go in var part.
3759 Pass reloc in fr_var. */
3760 frag_var (rs_machine_dependent, 5, i.reloc[0], subtype, sym, off, p);
3761 }
3762
3763 static void
3764 output_jump ()
3765 {
3766 char *p;
3767 int size;
3768 fixS *fixP;
3769
3770 if (i.tm.opcode_modifier & JumpByte)
3771 {
3772 /* This is a loop or jecxz type instruction. */
3773 size = 1;
3774 if (i.prefix[ADDR_PREFIX] != 0)
3775 {
3776 FRAG_APPEND_1_CHAR (ADDR_PREFIX_OPCODE);
3777 i.prefixes -= 1;
3778 }
3779 /* Pentium4 branch hints. */
3780 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
3781 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
3782 {
3783 FRAG_APPEND_1_CHAR (i.prefix[SEG_PREFIX]);
3784 i.prefixes--;
3785 }
3786 }
3787 else
3788 {
3789 int code16;
3790
3791 code16 = 0;
3792 if (flag_code == CODE_16BIT)
3793 code16 = CODE16;
3794
3795 if (i.prefix[DATA_PREFIX] != 0)
3796 {
3797 FRAG_APPEND_1_CHAR (DATA_PREFIX_OPCODE);
3798 i.prefixes -= 1;
3799 code16 ^= CODE16;
3800 }
3801
3802 size = 4;
3803 if (code16)
3804 size = 2;
3805 }
3806
3807 if (i.prefix[REX_PREFIX] != 0)
3808 {
3809 FRAG_APPEND_1_CHAR (i.prefix[REX_PREFIX]);
3810 i.prefixes -= 1;
3811 }
3812
3813 if (i.prefixes != 0 && !intel_syntax)
3814 as_warn (_("skipping prefixes on this instruction"));
3815
3816 p = frag_more (1 + size);
3817 *p++ = i.tm.base_opcode;
3818
3819 fixP = fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3820 i.op[0].disps, 1, reloc (size, 1, 1, i.reloc[0]));
3821
3822 /* All jumps handled here are signed, but don't use a signed limit
3823 check for 32 and 16 bit jumps as we want to allow wrap around at
3824 4G and 64k respectively. */
3825 if (size == 1)
3826 fixP->fx_signed = 1;
3827 }
3828
3829 static void
3830 output_interseg_jump ()
3831 {
3832 char *p;
3833 int size;
3834 int prefix;
3835 int code16;
3836
3837 code16 = 0;
3838 if (flag_code == CODE_16BIT)
3839 code16 = CODE16;
3840
3841 prefix = 0;
3842 if (i.prefix[DATA_PREFIX] != 0)
3843 {
3844 prefix = 1;
3845 i.prefixes -= 1;
3846 code16 ^= CODE16;
3847 }
3848 if (i.prefix[REX_PREFIX] != 0)
3849 {
3850 prefix++;
3851 i.prefixes -= 1;
3852 }
3853
3854 size = 4;
3855 if (code16)
3856 size = 2;
3857
3858 if (i.prefixes != 0 && !intel_syntax)
3859 as_warn (_("skipping prefixes on this instruction"));
3860
3861 /* 1 opcode; 2 segment; offset */
3862 p = frag_more (prefix + 1 + 2 + size);
3863
3864 if (i.prefix[DATA_PREFIX] != 0)
3865 *p++ = DATA_PREFIX_OPCODE;
3866
3867 if (i.prefix[REX_PREFIX] != 0)
3868 *p++ = i.prefix[REX_PREFIX];
3869
3870 *p++ = i.tm.base_opcode;
3871 if (i.op[1].imms->X_op == O_constant)
3872 {
3873 offsetT n = i.op[1].imms->X_add_number;
3874
3875 if (size == 2
3876 && !fits_in_unsigned_word (n)
3877 && !fits_in_signed_word (n))
3878 {
3879 as_bad (_("16-bit jump out of range"));
3880 return;
3881 }
3882 md_number_to_chars (p, n, size);
3883 }
3884 else
3885 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3886 i.op[1].imms, 0, reloc (size, 0, 0, i.reloc[1]));
3887 if (i.op[0].imms->X_op != O_constant)
3888 as_bad (_("can't handle non absolute segment in `%s'"),
3889 i.tm.name);
3890 md_number_to_chars (p + size, (valueT) i.op[0].imms->X_add_number, 2);
3891 }
3892
3893 static void
3894 output_insn ()
3895 {
3896 fragS *insn_start_frag;
3897 offsetT insn_start_off;
3898
3899 /* Tie dwarf2 debug info to the address at the start of the insn.
3900 We can't do this after the insn has been output as the current
3901 frag may have been closed off. eg. by frag_var. */
3902 dwarf2_emit_insn (0);
3903
3904 insn_start_frag = frag_now;
3905 insn_start_off = frag_now_fix ();
3906
3907 /* Output jumps. */
3908 if (i.tm.opcode_modifier & Jump)
3909 output_branch ();
3910 else if (i.tm.opcode_modifier & (JumpByte | JumpDword))
3911 output_jump ();
3912 else if (i.tm.opcode_modifier & JumpInterSegment)
3913 output_interseg_jump ();
3914 else
3915 {
3916 /* Output normal instructions here. */
3917 char *p;
3918 unsigned char *q;
3919 unsigned int prefix;
3920
3921 /* All opcodes on i386 have either 1 or 2 bytes. Supplemental
3922 Streaming SIMD extensions 3 Instructions have 3 bytes. We may
3923 use one more higher byte to specify a prefix the instruction
3924 requires. */
3925 if ((i.tm.cpu_flags & CpuSSSE3) != 0)
3926 {
3927 if (i.tm.base_opcode & 0xff000000)
3928 {
3929 prefix = (i.tm.base_opcode >> 24) & 0xff;
3930 goto check_prefix;
3931 }
3932 }
3933 else if ((i.tm.base_opcode & 0xff0000) != 0)
3934 {
3935 prefix = (i.tm.base_opcode >> 16) & 0xff;
3936 if ((i.tm.cpu_flags & CpuPadLock) != 0)
3937 {
3938 check_prefix:
3939 if (prefix != REPE_PREFIX_OPCODE
3940 || i.prefix[LOCKREP_PREFIX] != REPE_PREFIX_OPCODE)
3941 add_prefix (prefix);
3942 }
3943 else
3944 add_prefix (prefix);
3945 }
3946
3947 /* The prefix bytes. */
3948 for (q = i.prefix;
3949 q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
3950 q++)
3951 {
3952 if (*q)
3953 {
3954 p = frag_more (1);
3955 md_number_to_chars (p, (valueT) *q, 1);
3956 }
3957 }
3958
3959 /* Now the opcode; be careful about word order here! */
3960 if (fits_in_unsigned_byte (i.tm.base_opcode))
3961 {
3962 FRAG_APPEND_1_CHAR (i.tm.base_opcode);
3963 }
3964 else
3965 {
3966 if ((i.tm.cpu_flags & CpuSSSE3) != 0)
3967 {
3968 p = frag_more (3);
3969 *p++ = (i.tm.base_opcode >> 16) & 0xff;
3970 }
3971 else
3972 p = frag_more (2);
3973
3974 /* Put out high byte first: can't use md_number_to_chars! */
3975 *p++ = (i.tm.base_opcode >> 8) & 0xff;
3976 *p = i.tm.base_opcode & 0xff;
3977 }
3978
3979 /* Now the modrm byte and sib byte (if present). */
3980 if (i.tm.opcode_modifier & Modrm)
3981 {
3982 p = frag_more (1);
3983 md_number_to_chars (p,
3984 (valueT) (i.rm.regmem << 0
3985 | i.rm.reg << 3
3986 | i.rm.mode << 6),
3987 1);
3988 /* If i.rm.regmem == ESP (4)
3989 && i.rm.mode != (Register mode)
3990 && not 16 bit
3991 ==> need second modrm byte. */
3992 if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING
3993 && i.rm.mode != 3
3994 && !(i.base_reg && (i.base_reg->reg_type & Reg16) != 0))
3995 {
3996 p = frag_more (1);
3997 md_number_to_chars (p,
3998 (valueT) (i.sib.base << 0
3999 | i.sib.index << 3
4000 | i.sib.scale << 6),
4001 1);
4002 }
4003 }
4004
4005 if (i.disp_operands)
4006 output_disp (insn_start_frag, insn_start_off);
4007
4008 if (i.imm_operands)
4009 output_imm (insn_start_frag, insn_start_off);
4010 }
4011
4012 #ifdef DEBUG386
4013 if (flag_debug)
4014 {
4015 pi ("" /*line*/, &i);
4016 }
4017 #endif /* DEBUG386 */
4018 }
4019
4020 static void
4021 output_disp (fragS *insn_start_frag, offsetT insn_start_off)
4022 {
4023 char *p;
4024 unsigned int n;
4025
4026 for (n = 0; n < i.operands; n++)
4027 {
4028 if (i.types[n] & Disp)
4029 {
4030 if (i.op[n].disps->X_op == O_constant)
4031 {
4032 int size;
4033 offsetT val;
4034
4035 size = 4;
4036 if (i.types[n] & (Disp8 | Disp16 | Disp64))
4037 {
4038 size = 2;
4039 if (i.types[n] & Disp8)
4040 size = 1;
4041 if (i.types[n] & Disp64)
4042 size = 8;
4043 }
4044 val = offset_in_range (i.op[n].disps->X_add_number,
4045 size);
4046 p = frag_more (size);
4047 md_number_to_chars (p, val, size);
4048 }
4049 else
4050 {
4051 enum bfd_reloc_code_real reloc_type;
4052 int size = 4;
4053 int sign = 0;
4054 int pcrel = (i.flags[n] & Operand_PCrel) != 0;
4055
4056 /* The PC relative address is computed relative
4057 to the instruction boundary, so in case immediate
4058 fields follows, we need to adjust the value. */
4059 if (pcrel && i.imm_operands)
4060 {
4061 int imm_size = 4;
4062 unsigned int n1;
4063
4064 for (n1 = 0; n1 < i.operands; n1++)
4065 if (i.types[n1] & Imm)
4066 {
4067 if (i.types[n1] & (Imm8 | Imm8S | Imm16 | Imm64))
4068 {
4069 imm_size = 2;
4070 if (i.types[n1] & (Imm8 | Imm8S))
4071 imm_size = 1;
4072 if (i.types[n1] & Imm64)
4073 imm_size = 8;
4074 }
4075 break;
4076 }
4077 /* We should find the immediate. */
4078 if (n1 == i.operands)
4079 abort ();
4080 i.op[n].disps->X_add_number -= imm_size;
4081 }
4082
4083 if (i.types[n] & Disp32S)
4084 sign = 1;
4085
4086 if (i.types[n] & (Disp16 | Disp64))
4087 {
4088 size = 2;
4089 if (i.types[n] & Disp64)
4090 size = 8;
4091 }
4092
4093 p = frag_more (size);
4094 reloc_type = reloc (size, pcrel, sign, i.reloc[n]);
4095 if (GOT_symbol
4096 && GOT_symbol == i.op[n].disps->X_add_symbol
4097 && (((reloc_type == BFD_RELOC_32
4098 || reloc_type == BFD_RELOC_X86_64_32S
4099 || (reloc_type == BFD_RELOC_64
4100 && object_64bit))
4101 && (i.op[n].disps->X_op == O_symbol
4102 || (i.op[n].disps->X_op == O_add
4103 && ((symbol_get_value_expression
4104 (i.op[n].disps->X_op_symbol)->X_op)
4105 == O_subtract))))
4106 || reloc_type == BFD_RELOC_32_PCREL))
4107 {
4108 offsetT add;
4109
4110 if (insn_start_frag == frag_now)
4111 add = (p - frag_now->fr_literal) - insn_start_off;
4112 else
4113 {
4114 fragS *fr;
4115
4116 add = insn_start_frag->fr_fix - insn_start_off;
4117 for (fr = insn_start_frag->fr_next;
4118 fr && fr != frag_now; fr = fr->fr_next)
4119 add += fr->fr_fix;
4120 add += p - frag_now->fr_literal;
4121 }
4122
4123 if (!object_64bit)
4124 {
4125 reloc_type = BFD_RELOC_386_GOTPC;
4126 i.op[n].imms->X_add_number += add;
4127 }
4128 else if (reloc_type == BFD_RELOC_64)
4129 reloc_type = BFD_RELOC_X86_64_GOTPC64;
4130 else
4131 /* Don't do the adjustment for x86-64, as there
4132 the pcrel addressing is relative to the _next_
4133 insn, and that is taken care of in other code. */
4134 reloc_type = BFD_RELOC_X86_64_GOTPC32;
4135 }
4136 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
4137 i.op[n].disps, pcrel, reloc_type);
4138 }
4139 }
4140 }
4141 }
4142
4143 static void
4144 output_imm (fragS *insn_start_frag, offsetT insn_start_off)
4145 {
4146 char *p;
4147 unsigned int n;
4148
4149 for (n = 0; n < i.operands; n++)
4150 {
4151 if (i.types[n] & Imm)
4152 {
4153 if (i.op[n].imms->X_op == O_constant)
4154 {
4155 int size;
4156 offsetT val;
4157
4158 size = 4;
4159 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
4160 {
4161 size = 2;
4162 if (i.types[n] & (Imm8 | Imm8S))
4163 size = 1;
4164 else if (i.types[n] & Imm64)
4165 size = 8;
4166 }
4167 val = offset_in_range (i.op[n].imms->X_add_number,
4168 size);
4169 p = frag_more (size);
4170 md_number_to_chars (p, val, size);
4171 }
4172 else
4173 {
4174 /* Not absolute_section.
4175 Need a 32-bit fixup (don't support 8bit
4176 non-absolute imms). Try to support other
4177 sizes ... */
4178 enum bfd_reloc_code_real reloc_type;
4179 int size = 4;
4180 int sign = 0;
4181
4182 if ((i.types[n] & (Imm32S))
4183 && (i.suffix == QWORD_MNEM_SUFFIX
4184 || (!i.suffix && (i.tm.opcode_modifier & No_lSuf))))
4185 sign = 1;
4186 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
4187 {
4188 size = 2;
4189 if (i.types[n] & (Imm8 | Imm8S))
4190 size = 1;
4191 if (i.types[n] & Imm64)
4192 size = 8;
4193 }
4194
4195 p = frag_more (size);
4196 reloc_type = reloc (size, 0, sign, i.reloc[n]);
4197
4198 /* This is tough to explain. We end up with this one if we
4199 * have operands that look like
4200 * "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal here is to
4201 * obtain the absolute address of the GOT, and it is strongly
4202 * preferable from a performance point of view to avoid using
4203 * a runtime relocation for this. The actual sequence of
4204 * instructions often look something like:
4205 *
4206 * call .L66
4207 * .L66:
4208 * popl %ebx
4209 * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
4210 *
4211 * The call and pop essentially return the absolute address
4212 * of the label .L66 and store it in %ebx. The linker itself
4213 * will ultimately change the first operand of the addl so
4214 * that %ebx points to the GOT, but to keep things simple, the
4215 * .o file must have this operand set so that it generates not
4216 * the absolute address of .L66, but the absolute address of
4217 * itself. This allows the linker itself simply treat a GOTPC
4218 * relocation as asking for a pcrel offset to the GOT to be
4219 * added in, and the addend of the relocation is stored in the
4220 * operand field for the instruction itself.
4221 *
4222 * Our job here is to fix the operand so that it would add
4223 * the correct offset so that %ebx would point to itself. The
4224 * thing that is tricky is that .-.L66 will point to the
4225 * beginning of the instruction, so we need to further modify
4226 * the operand so that it will point to itself. There are
4227 * other cases where you have something like:
4228 *
4229 * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
4230 *
4231 * and here no correction would be required. Internally in
4232 * the assembler we treat operands of this form as not being
4233 * pcrel since the '.' is explicitly mentioned, and I wonder
4234 * whether it would simplify matters to do it this way. Who
4235 * knows. In earlier versions of the PIC patches, the
4236 * pcrel_adjust field was used to store the correction, but
4237 * since the expression is not pcrel, I felt it would be
4238 * confusing to do it this way. */
4239
4240 if ((reloc_type == BFD_RELOC_32
4241 || reloc_type == BFD_RELOC_X86_64_32S
4242 || reloc_type == BFD_RELOC_64)
4243 && GOT_symbol
4244 && GOT_symbol == i.op[n].imms->X_add_symbol
4245 && (i.op[n].imms->X_op == O_symbol
4246 || (i.op[n].imms->X_op == O_add
4247 && ((symbol_get_value_expression
4248 (i.op[n].imms->X_op_symbol)->X_op)
4249 == O_subtract))))
4250 {
4251 offsetT add;
4252
4253 if (insn_start_frag == frag_now)
4254 add = (p - frag_now->fr_literal) - insn_start_off;
4255 else
4256 {
4257 fragS *fr;
4258
4259 add = insn_start_frag->fr_fix - insn_start_off;
4260 for (fr = insn_start_frag->fr_next;
4261 fr && fr != frag_now; fr = fr->fr_next)
4262 add += fr->fr_fix;
4263 add += p - frag_now->fr_literal;
4264 }
4265
4266 if (!object_64bit)
4267 reloc_type = BFD_RELOC_386_GOTPC;
4268 else if (size == 4)
4269 reloc_type = BFD_RELOC_X86_64_GOTPC32;
4270 else if (size == 8)
4271 reloc_type = BFD_RELOC_X86_64_GOTPC64;
4272 i.op[n].imms->X_add_number += add;
4273 }
4274 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
4275 i.op[n].imms, 0, reloc_type);
4276 }
4277 }
4278 }
4279 }
4280 \f
4281 /* x86_cons_fix_new is called via the expression parsing code when a
4282 reloc is needed. We use this hook to get the correct .got reloc. */
4283 static enum bfd_reloc_code_real got_reloc = NO_RELOC;
4284 static int cons_sign = -1;
4285
4286 void
4287 x86_cons_fix_new (fragS *frag,
4288 unsigned int off,
4289 unsigned int len,
4290 expressionS *exp)
4291 {
4292 enum bfd_reloc_code_real r = reloc (len, 0, cons_sign, got_reloc);
4293
4294 got_reloc = NO_RELOC;
4295
4296 #ifdef TE_PE
4297 if (exp->X_op == O_secrel)
4298 {
4299 exp->X_op = O_symbol;
4300 r = BFD_RELOC_32_SECREL;
4301 }
4302 #endif
4303
4304 fix_new_exp (frag, off, len, exp, 0, r);
4305 }
4306
4307 #if (!defined (OBJ_ELF) && !defined (OBJ_MAYBE_ELF)) || defined (LEX_AT)
4308 # define lex_got(reloc, adjust, types) NULL
4309 #else
4310 /* Parse operands of the form
4311 <symbol>@GOTOFF+<nnn>
4312 and similar .plt or .got references.
4313
4314 If we find one, set up the correct relocation in RELOC and copy the
4315 input string, minus the `@GOTOFF' into a malloc'd buffer for
4316 parsing by the calling routine. Return this buffer, and if ADJUST
4317 is non-null set it to the length of the string we removed from the
4318 input line. Otherwise return NULL. */
4319 static char *
4320 lex_got (enum bfd_reloc_code_real *reloc,
4321 int *adjust,
4322 unsigned int *types)
4323 {
4324 /* Some of the relocations depend on the size of what field is to
4325 be relocated. But in our callers i386_immediate and i386_displacement
4326 we don't yet know the operand size (this will be set by insn
4327 matching). Hence we record the word32 relocation here,
4328 and adjust the reloc according to the real size in reloc(). */
4329 static const struct {
4330 const char *str;
4331 const enum bfd_reloc_code_real rel[2];
4332 const unsigned int types64;
4333 } gotrel[] = {
4334 { "PLTOFF", { 0, BFD_RELOC_X86_64_PLTOFF64 }, Imm64 },
4335 { "PLT", { BFD_RELOC_386_PLT32, BFD_RELOC_X86_64_PLT32 }, Imm32|Imm32S|Disp32 },
4336 { "GOTPLT", { 0, BFD_RELOC_X86_64_GOTPLT64 }, Imm64|Disp64 },
4337 { "GOTOFF", { BFD_RELOC_386_GOTOFF, BFD_RELOC_X86_64_GOTOFF64 }, Imm64|Disp64 },
4338 { "GOTPCREL", { 0, BFD_RELOC_X86_64_GOTPCREL }, Imm32|Imm32S|Disp32 },
4339 { "TLSGD", { BFD_RELOC_386_TLS_GD, BFD_RELOC_X86_64_TLSGD }, Imm32|Imm32S|Disp32 },
4340 { "TLSLDM", { BFD_RELOC_386_TLS_LDM, 0 }, 0 },
4341 { "TLSLD", { 0, BFD_RELOC_X86_64_TLSLD }, Imm32|Imm32S|Disp32 },
4342 { "GOTTPOFF", { BFD_RELOC_386_TLS_IE_32, BFD_RELOC_X86_64_GOTTPOFF }, Imm32|Imm32S|Disp32 },
4343 { "TPOFF", { BFD_RELOC_386_TLS_LE_32, BFD_RELOC_X86_64_TPOFF32 }, Imm32|Imm32S|Imm64|Disp32|Disp64 },
4344 { "NTPOFF", { BFD_RELOC_386_TLS_LE, 0 }, 0 },
4345 { "DTPOFF", { BFD_RELOC_386_TLS_LDO_32, BFD_RELOC_X86_64_DTPOFF32 }, Imm32|Imm32S|Imm64|Disp32|Disp64 },
4346 { "GOTNTPOFF",{ BFD_RELOC_386_TLS_GOTIE, 0 }, 0 },
4347 { "INDNTPOFF",{ BFD_RELOC_386_TLS_IE, 0 }, 0 },
4348 { "GOT", { BFD_RELOC_386_GOT32, BFD_RELOC_X86_64_GOT32 }, Imm32|Imm32S|Disp32|Imm64 },
4349 { "TLSDESC", { BFD_RELOC_386_TLS_GOTDESC, BFD_RELOC_X86_64_GOTPC32_TLSDESC }, Imm32|Imm32S|Disp32 },
4350 { "TLSCALL", { BFD_RELOC_386_TLS_DESC_CALL, BFD_RELOC_X86_64_TLSDESC_CALL }, Imm32|Imm32S|Disp32 }
4351 };
4352 char *cp;
4353 unsigned int j;
4354
4355 if (!IS_ELF)
4356 return NULL;
4357
4358 for (cp = input_line_pointer; *cp != '@'; cp++)
4359 if (is_end_of_line[(unsigned char) *cp])
4360 return NULL;
4361
4362 for (j = 0; j < sizeof (gotrel) / sizeof (gotrel[0]); j++)
4363 {
4364 int len;
4365
4366 len = strlen (gotrel[j].str);
4367 if (strncasecmp (cp + 1, gotrel[j].str, len) == 0)
4368 {
4369 if (gotrel[j].rel[object_64bit] != 0)
4370 {
4371 int first, second;
4372 char *tmpbuf, *past_reloc;
4373
4374 *reloc = gotrel[j].rel[object_64bit];
4375 if (adjust)
4376 *adjust = len;
4377
4378 if (types)
4379 {
4380 if (flag_code != CODE_64BIT)
4381 *types = Imm32|Disp32;
4382 else
4383 *types = gotrel[j].types64;
4384 }
4385
4386 if (GOT_symbol == NULL)
4387 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
4388
4389 /* Replace the relocation token with ' ', so that
4390 errors like foo@GOTOFF1 will be detected. */
4391
4392 /* The length of the first part of our input line. */
4393 first = cp - input_line_pointer;
4394
4395 /* The second part goes from after the reloc token until
4396 (and including) an end_of_line char. Don't use strlen
4397 here as the end_of_line char may not be a NUL. */
4398 past_reloc = cp + 1 + len;
4399 for (cp = past_reloc; !is_end_of_line[(unsigned char) *cp++]; )
4400 ;
4401 second = cp - past_reloc;
4402
4403 /* Allocate and copy string. The trailing NUL shouldn't
4404 be necessary, but be safe. */
4405 tmpbuf = xmalloc (first + second + 2);
4406 memcpy (tmpbuf, input_line_pointer, first);
4407 tmpbuf[first] = ' ';
4408 memcpy (tmpbuf + first + 1, past_reloc, second);
4409 tmpbuf[first + second + 1] = '\0';
4410 return tmpbuf;
4411 }
4412
4413 as_bad (_("@%s reloc is not supported with %d-bit output format"),
4414 gotrel[j].str, 1 << (5 + object_64bit));
4415 return NULL;
4416 }
4417 }
4418
4419 /* Might be a symbol version string. Don't as_bad here. */
4420 return NULL;
4421 }
4422
4423 void
4424 x86_cons (exp, size)
4425 expressionS *exp;
4426 int size;
4427 {
4428 if (size == 4 || (object_64bit && size == 8))
4429 {
4430 /* Handle @GOTOFF and the like in an expression. */
4431 char *save;
4432 char *gotfree_input_line;
4433 int adjust;
4434
4435 save = input_line_pointer;
4436 gotfree_input_line = lex_got (&got_reloc, &adjust, NULL);
4437 if (gotfree_input_line)
4438 input_line_pointer = gotfree_input_line;
4439
4440 expression (exp);
4441
4442 if (gotfree_input_line)
4443 {
4444 /* expression () has merrily parsed up to the end of line,
4445 or a comma - in the wrong buffer. Transfer how far
4446 input_line_pointer has moved to the right buffer. */
4447 input_line_pointer = (save
4448 + (input_line_pointer - gotfree_input_line)
4449 + adjust);
4450 free (gotfree_input_line);
4451 }
4452 }
4453 else
4454 expression (exp);
4455 }
4456 #endif
4457
4458 static void signed_cons (int size)
4459 {
4460 if (flag_code == CODE_64BIT)
4461 cons_sign = 1;
4462 cons (size);
4463 cons_sign = -1;
4464 }
4465
4466 #ifdef TE_PE
4467 static void
4468 pe_directive_secrel (dummy)
4469 int dummy ATTRIBUTE_UNUSED;
4470 {
4471 expressionS exp;
4472
4473 do
4474 {
4475 expression (&exp);
4476 if (exp.X_op == O_symbol)
4477 exp.X_op = O_secrel;
4478
4479 emit_expr (&exp, 4);
4480 }
4481 while (*input_line_pointer++ == ',');
4482
4483 input_line_pointer--;
4484 demand_empty_rest_of_line ();
4485 }
4486 #endif
4487
4488 static int i386_immediate PARAMS ((char *));
4489
4490 static int
4491 i386_immediate (imm_start)
4492 char *imm_start;
4493 {
4494 char *save_input_line_pointer;
4495 char *gotfree_input_line;
4496 segT exp_seg = 0;
4497 expressionS *exp;
4498 unsigned int types = ~0U;
4499
4500 if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
4501 {
4502 as_bad (_("only 1 or 2 immediate operands are allowed"));
4503 return 0;
4504 }
4505
4506 exp = &im_expressions[i.imm_operands++];
4507 i.op[this_operand].imms = exp;
4508
4509 if (is_space_char (*imm_start))
4510 ++imm_start;
4511
4512 save_input_line_pointer = input_line_pointer;
4513 input_line_pointer = imm_start;
4514
4515 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL, &types);
4516 if (gotfree_input_line)
4517 input_line_pointer = gotfree_input_line;
4518
4519 exp_seg = expression (exp);
4520
4521 SKIP_WHITESPACE ();
4522 if (*input_line_pointer)
4523 as_bad (_("junk `%s' after expression"), input_line_pointer);
4524
4525 input_line_pointer = save_input_line_pointer;
4526 if (gotfree_input_line)
4527 free (gotfree_input_line);
4528
4529 if (exp->X_op == O_absent || exp->X_op == O_big)
4530 {
4531 /* Missing or bad expr becomes absolute 0. */
4532 as_bad (_("missing or invalid immediate expression `%s' taken as 0"),
4533 imm_start);
4534 exp->X_op = O_constant;
4535 exp->X_add_number = 0;
4536 exp->X_add_symbol = (symbolS *) 0;
4537 exp->X_op_symbol = (symbolS *) 0;
4538 }
4539 else if (exp->X_op == O_constant)
4540 {
4541 /* Size it properly later. */
4542 i.types[this_operand] |= Imm64;
4543 /* If BFD64, sign extend val. */
4544 if (!use_rela_relocations)
4545 if ((exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0)
4546 exp->X_add_number = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
4547 }
4548 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4549 else if (OUTPUT_FLAVOR == bfd_target_aout_flavour
4550 && exp_seg != absolute_section
4551 && exp_seg != text_section
4552 && exp_seg != data_section
4553 && exp_seg != bss_section
4554 && exp_seg != undefined_section
4555 && !bfd_is_com_section (exp_seg))
4556 {
4557 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
4558 return 0;
4559 }
4560 #endif
4561 else if (!intel_syntax && exp->X_op == O_register)
4562 {
4563 as_bad (_("illegal immediate register operand %s"), imm_start);
4564 return 0;
4565 }
4566 else
4567 {
4568 /* This is an address. The size of the address will be
4569 determined later, depending on destination register,
4570 suffix, or the default for the section. */
4571 i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64;
4572 i.types[this_operand] &= types;
4573 }
4574
4575 return 1;
4576 }
4577
4578 static char *i386_scale PARAMS ((char *));
4579
4580 static char *
4581 i386_scale (scale)
4582 char *scale;
4583 {
4584 offsetT val;
4585 char *save = input_line_pointer;
4586
4587 input_line_pointer = scale;
4588 val = get_absolute_expression ();
4589
4590 switch (val)
4591 {
4592 case 1:
4593 i.log2_scale_factor = 0;
4594 break;
4595 case 2:
4596 i.log2_scale_factor = 1;
4597 break;
4598 case 4:
4599 i.log2_scale_factor = 2;
4600 break;
4601 case 8:
4602 i.log2_scale_factor = 3;
4603 break;
4604 default:
4605 {
4606 char sep = *input_line_pointer;
4607
4608 *input_line_pointer = '\0';
4609 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
4610 scale);
4611 *input_line_pointer = sep;
4612 input_line_pointer = save;
4613 return NULL;
4614 }
4615 }
4616 if (i.log2_scale_factor != 0 && i.index_reg == 0)
4617 {
4618 as_warn (_("scale factor of %d without an index register"),
4619 1 << i.log2_scale_factor);
4620 #if SCALE1_WHEN_NO_INDEX
4621 i.log2_scale_factor = 0;
4622 #endif
4623 }
4624 scale = input_line_pointer;
4625 input_line_pointer = save;
4626 return scale;
4627 }
4628
4629 static int i386_displacement PARAMS ((char *, char *));
4630
4631 static int
4632 i386_displacement (disp_start, disp_end)
4633 char *disp_start;
4634 char *disp_end;
4635 {
4636 expressionS *exp;
4637 segT exp_seg = 0;
4638 char *save_input_line_pointer;
4639 char *gotfree_input_line;
4640 int bigdisp, override;
4641 unsigned int types = Disp;
4642
4643 if ((i.types[this_operand] & JumpAbsolute)
4644 || !(current_templates->start->opcode_modifier & (Jump | JumpDword)))
4645 {
4646 bigdisp = Disp32;
4647 override = (i.prefix[ADDR_PREFIX] != 0);
4648 }
4649 else
4650 {
4651 /* For PC-relative branches, the width of the displacement
4652 is dependent upon data size, not address size. */
4653 bigdisp = 0;
4654 override = (i.prefix[DATA_PREFIX] != 0);
4655 }
4656 if (flag_code == CODE_64BIT)
4657 {
4658 if (!bigdisp)
4659 bigdisp = ((override || i.suffix == WORD_MNEM_SUFFIX)
4660 ? Disp16
4661 : Disp32S | Disp32);
4662 else if (!override)
4663 bigdisp = Disp64 | Disp32S | Disp32;
4664 }
4665 else
4666 {
4667 if (!bigdisp)
4668 {
4669 if (!override)
4670 override = (i.suffix == (flag_code != CODE_16BIT
4671 ? WORD_MNEM_SUFFIX
4672 : LONG_MNEM_SUFFIX));
4673 bigdisp = Disp32;
4674 }
4675 if ((flag_code == CODE_16BIT) ^ override)
4676 bigdisp = Disp16;
4677 }
4678 i.types[this_operand] |= bigdisp;
4679
4680 exp = &disp_expressions[i.disp_operands];
4681 i.op[this_operand].disps = exp;
4682 i.disp_operands++;
4683 save_input_line_pointer = input_line_pointer;
4684 input_line_pointer = disp_start;
4685 END_STRING_AND_SAVE (disp_end);
4686
4687 #ifndef GCC_ASM_O_HACK
4688 #define GCC_ASM_O_HACK 0
4689 #endif
4690 #if GCC_ASM_O_HACK
4691 END_STRING_AND_SAVE (disp_end + 1);
4692 if ((i.types[this_operand] & BaseIndex) != 0
4693 && displacement_string_end[-1] == '+')
4694 {
4695 /* This hack is to avoid a warning when using the "o"
4696 constraint within gcc asm statements.
4697 For instance:
4698
4699 #define _set_tssldt_desc(n,addr,limit,type) \
4700 __asm__ __volatile__ ( \
4701 "movw %w2,%0\n\t" \
4702 "movw %w1,2+%0\n\t" \
4703 "rorl $16,%1\n\t" \
4704 "movb %b1,4+%0\n\t" \
4705 "movb %4,5+%0\n\t" \
4706 "movb $0,6+%0\n\t" \
4707 "movb %h1,7+%0\n\t" \
4708 "rorl $16,%1" \
4709 : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
4710
4711 This works great except that the output assembler ends
4712 up looking a bit weird if it turns out that there is
4713 no offset. You end up producing code that looks like:
4714
4715 #APP
4716 movw $235,(%eax)
4717 movw %dx,2+(%eax)
4718 rorl $16,%edx
4719 movb %dl,4+(%eax)
4720 movb $137,5+(%eax)
4721 movb $0,6+(%eax)
4722 movb %dh,7+(%eax)
4723 rorl $16,%edx
4724 #NO_APP
4725
4726 So here we provide the missing zero. */
4727
4728 *displacement_string_end = '0';
4729 }
4730 #endif
4731 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL, &types);
4732 if (gotfree_input_line)
4733 input_line_pointer = gotfree_input_line;
4734
4735 exp_seg = expression (exp);
4736
4737 SKIP_WHITESPACE ();
4738 if (*input_line_pointer)
4739 as_bad (_("junk `%s' after expression"), input_line_pointer);
4740 #if GCC_ASM_O_HACK
4741 RESTORE_END_STRING (disp_end + 1);
4742 #endif
4743 RESTORE_END_STRING (disp_end);
4744 input_line_pointer = save_input_line_pointer;
4745 if (gotfree_input_line)
4746 free (gotfree_input_line);
4747
4748 /* We do this to make sure that the section symbol is in
4749 the symbol table. We will ultimately change the relocation
4750 to be relative to the beginning of the section. */
4751 if (i.reloc[this_operand] == BFD_RELOC_386_GOTOFF
4752 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
4753 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTOFF64)
4754 {
4755 if (exp->X_op != O_symbol)
4756 {
4757 as_bad (_("bad expression used with @%s"),
4758 (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
4759 ? "GOTPCREL"
4760 : "GOTOFF"));
4761 return 0;
4762 }
4763
4764 if (S_IS_LOCAL (exp->X_add_symbol)
4765 && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
4766 section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
4767 exp->X_op = O_subtract;
4768 exp->X_op_symbol = GOT_symbol;
4769 if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
4770 i.reloc[this_operand] = BFD_RELOC_32_PCREL;
4771 else if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTOFF64)
4772 i.reloc[this_operand] = BFD_RELOC_64;
4773 else
4774 i.reloc[this_operand] = BFD_RELOC_32;
4775 }
4776
4777 if (exp->X_op == O_absent || exp->X_op == O_big)
4778 {
4779 /* Missing or bad expr becomes absolute 0. */
4780 as_bad (_("missing or invalid displacement expression `%s' taken as 0"),
4781 disp_start);
4782 exp->X_op = O_constant;
4783 exp->X_add_number = 0;
4784 exp->X_add_symbol = (symbolS *) 0;
4785 exp->X_op_symbol = (symbolS *) 0;
4786 }
4787
4788 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4789 if (exp->X_op != O_constant
4790 && OUTPUT_FLAVOR == bfd_target_aout_flavour
4791 && exp_seg != absolute_section
4792 && exp_seg != text_section
4793 && exp_seg != data_section
4794 && exp_seg != bss_section
4795 && exp_seg != undefined_section
4796 && !bfd_is_com_section (exp_seg))
4797 {
4798 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
4799 return 0;
4800 }
4801 #endif
4802
4803 if (!(i.types[this_operand] & ~Disp))
4804 i.types[this_operand] &= types;
4805
4806 return 1;
4807 }
4808
4809 static int i386_index_check PARAMS ((const char *));
4810
4811 /* Make sure the memory operand we've been dealt is valid.
4812 Return 1 on success, 0 on a failure. */
4813
4814 static int
4815 i386_index_check (operand_string)
4816 const char *operand_string;
4817 {
4818 int ok;
4819 #if INFER_ADDR_PREFIX
4820 int fudged = 0;
4821
4822 tryprefix:
4823 #endif
4824 ok = 1;
4825 if ((current_templates->start->cpu_flags & CpuSVME)
4826 && current_templates->end[-1].operand_types[0] == AnyMem)
4827 {
4828 /* Memory operands of SVME insns are special in that they only allow
4829 rAX as their memory address and ignore any segment override. */
4830 unsigned RegXX;
4831
4832 /* SKINIT is even more restrictive: it always requires EAX. */
4833 if (strcmp (current_templates->start->name, "skinit") == 0)
4834 RegXX = Reg32;
4835 else if (flag_code == CODE_64BIT)
4836 RegXX = i.prefix[ADDR_PREFIX] == 0 ? Reg64 : Reg32;
4837 else
4838 RegXX = ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0)
4839 ? Reg16
4840 : Reg32);
4841 if (!i.base_reg
4842 || !(i.base_reg->reg_type & Acc)
4843 || !(i.base_reg->reg_type & RegXX)
4844 || i.index_reg
4845 || (i.types[0] & Disp))
4846 ok = 0;
4847 }
4848 else if (flag_code == CODE_64BIT)
4849 {
4850 unsigned RegXX = (i.prefix[ADDR_PREFIX] == 0 ? Reg64 : Reg32);
4851
4852 if ((i.base_reg
4853 && ((i.base_reg->reg_type & RegXX) == 0)
4854 && (i.base_reg->reg_type != BaseIndex
4855 || i.index_reg))
4856 || (i.index_reg
4857 && ((i.index_reg->reg_type & (RegXX | BaseIndex))
4858 != (RegXX | BaseIndex))))
4859 ok = 0;
4860 }
4861 else
4862 {
4863 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
4864 {
4865 /* 16bit checks. */
4866 if ((i.base_reg
4867 && ((i.base_reg->reg_type & (Reg16 | BaseIndex | RegRex))
4868 != (Reg16 | BaseIndex)))
4869 || (i.index_reg
4870 && (((i.index_reg->reg_type & (Reg16 | BaseIndex))
4871 != (Reg16 | BaseIndex))
4872 || !(i.base_reg
4873 && i.base_reg->reg_num < 6
4874 && i.index_reg->reg_num >= 6
4875 && i.log2_scale_factor == 0))))
4876 ok = 0;
4877 }
4878 else
4879 {
4880 /* 32bit checks. */
4881 if ((i.base_reg
4882 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
4883 || (i.index_reg
4884 && ((i.index_reg->reg_type & (Reg32 | BaseIndex | RegRex))
4885 != (Reg32 | BaseIndex))))
4886 ok = 0;
4887 }
4888 }
4889 if (!ok)
4890 {
4891 #if INFER_ADDR_PREFIX
4892 if (i.prefix[ADDR_PREFIX] == 0)
4893 {
4894 i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
4895 i.prefixes += 1;
4896 /* Change the size of any displacement too. At most one of
4897 Disp16 or Disp32 is set.
4898 FIXME. There doesn't seem to be any real need for separate
4899 Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
4900 Removing them would probably clean up the code quite a lot. */
4901 if (flag_code != CODE_64BIT && (i.types[this_operand] & (Disp16 | Disp32)))
4902 i.types[this_operand] ^= (Disp16 | Disp32);
4903 fudged = 1;
4904 goto tryprefix;
4905 }
4906 if (fudged)
4907 as_bad (_("`%s' is not a valid base/index expression"),
4908 operand_string);
4909 else
4910 #endif
4911 as_bad (_("`%s' is not a valid %s bit base/index expression"),
4912 operand_string,
4913 flag_code_names[flag_code]);
4914 }
4915 return ok;
4916 }
4917
4918 /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
4919 on error. */
4920
4921 static int
4922 i386_operand (operand_string)
4923 char *operand_string;
4924 {
4925 const reg_entry *r;
4926 char *end_op;
4927 char *op_string = operand_string;
4928
4929 if (is_space_char (*op_string))
4930 ++op_string;
4931
4932 /* We check for an absolute prefix (differentiating,
4933 for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
4934 if (*op_string == ABSOLUTE_PREFIX)
4935 {
4936 ++op_string;
4937 if (is_space_char (*op_string))
4938 ++op_string;
4939 i.types[this_operand] |= JumpAbsolute;
4940 }
4941
4942 /* Check if operand is a register. */
4943 if ((r = parse_register (op_string, &end_op)) != NULL)
4944 {
4945 /* Check for a segment override by searching for ':' after a
4946 segment register. */
4947 op_string = end_op;
4948 if (is_space_char (*op_string))
4949 ++op_string;
4950 if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
4951 {
4952 switch (r->reg_num)
4953 {
4954 case 0:
4955 i.seg[i.mem_operands] = &es;
4956 break;
4957 case 1:
4958 i.seg[i.mem_operands] = &cs;
4959 break;
4960 case 2:
4961 i.seg[i.mem_operands] = &ss;
4962 break;
4963 case 3:
4964 i.seg[i.mem_operands] = &ds;
4965 break;
4966 case 4:
4967 i.seg[i.mem_operands] = &fs;
4968 break;
4969 case 5:
4970 i.seg[i.mem_operands] = &gs;
4971 break;
4972 }
4973
4974 /* Skip the ':' and whitespace. */
4975 ++op_string;
4976 if (is_space_char (*op_string))
4977 ++op_string;
4978
4979 if (!is_digit_char (*op_string)
4980 && !is_identifier_char (*op_string)
4981 && *op_string != '('
4982 && *op_string != ABSOLUTE_PREFIX)
4983 {
4984 as_bad (_("bad memory operand `%s'"), op_string);
4985 return 0;
4986 }
4987 /* Handle case of %es:*foo. */
4988 if (*op_string == ABSOLUTE_PREFIX)
4989 {
4990 ++op_string;
4991 if (is_space_char (*op_string))
4992 ++op_string;
4993 i.types[this_operand] |= JumpAbsolute;
4994 }
4995 goto do_memory_reference;
4996 }
4997 if (*op_string)
4998 {
4999 as_bad (_("junk `%s' after register"), op_string);
5000 return 0;
5001 }
5002 i.types[this_operand] |= r->reg_type & ~BaseIndex;
5003 i.op[this_operand].regs = r;
5004 i.reg_operands++;
5005 }
5006 else if (*op_string == REGISTER_PREFIX)
5007 {
5008 as_bad (_("bad register name `%s'"), op_string);
5009 return 0;
5010 }
5011 else if (*op_string == IMMEDIATE_PREFIX)
5012 {
5013 ++op_string;
5014 if (i.types[this_operand] & JumpAbsolute)
5015 {
5016 as_bad (_("immediate operand illegal with absolute jump"));
5017 return 0;
5018 }
5019 if (!i386_immediate (op_string))
5020 return 0;
5021 }
5022 else if (is_digit_char (*op_string)
5023 || is_identifier_char (*op_string)
5024 || *op_string == '(')
5025 {
5026 /* This is a memory reference of some sort. */
5027 char *base_string;
5028
5029 /* Start and end of displacement string expression (if found). */
5030 char *displacement_string_start;
5031 char *displacement_string_end;
5032
5033 do_memory_reference:
5034 if ((i.mem_operands == 1
5035 && (current_templates->start->opcode_modifier & IsString) == 0)
5036 || i.mem_operands == 2)
5037 {
5038 as_bad (_("too many memory references for `%s'"),
5039 current_templates->start->name);
5040 return 0;
5041 }
5042
5043 /* Check for base index form. We detect the base index form by
5044 looking for an ')' at the end of the operand, searching
5045 for the '(' matching it, and finding a REGISTER_PREFIX or ','
5046 after the '('. */
5047 base_string = op_string + strlen (op_string);
5048
5049 --base_string;
5050 if (is_space_char (*base_string))
5051 --base_string;
5052
5053 /* If we only have a displacement, set-up for it to be parsed later. */
5054 displacement_string_start = op_string;
5055 displacement_string_end = base_string + 1;
5056
5057 if (*base_string == ')')
5058 {
5059 char *temp_string;
5060 unsigned int parens_balanced = 1;
5061 /* We've already checked that the number of left & right ()'s are
5062 equal, so this loop will not be infinite. */
5063 do
5064 {
5065 base_string--;
5066 if (*base_string == ')')
5067 parens_balanced++;
5068 if (*base_string == '(')
5069 parens_balanced--;
5070 }
5071 while (parens_balanced);
5072
5073 temp_string = base_string;
5074
5075 /* Skip past '(' and whitespace. */
5076 ++base_string;
5077 if (is_space_char (*base_string))
5078 ++base_string;
5079
5080 if (*base_string == ','
5081 || ((i.base_reg = parse_register (base_string, &end_op)) != NULL))
5082 {
5083 displacement_string_end = temp_string;
5084
5085 i.types[this_operand] |= BaseIndex;
5086
5087 if (i.base_reg)
5088 {
5089 base_string = end_op;
5090 if (is_space_char (*base_string))
5091 ++base_string;
5092 }
5093
5094 /* There may be an index reg or scale factor here. */
5095 if (*base_string == ',')
5096 {
5097 ++base_string;
5098 if (is_space_char (*base_string))
5099 ++base_string;
5100
5101 if ((i.index_reg = parse_register (base_string, &end_op)) != NULL)
5102 {
5103 base_string = end_op;
5104 if (is_space_char (*base_string))
5105 ++base_string;
5106 if (*base_string == ',')
5107 {
5108 ++base_string;
5109 if (is_space_char (*base_string))
5110 ++base_string;
5111 }
5112 else if (*base_string != ')')
5113 {
5114 as_bad (_("expecting `,' or `)' after index register in `%s'"),
5115 operand_string);
5116 return 0;
5117 }
5118 }
5119 else if (*base_string == REGISTER_PREFIX)
5120 {
5121 as_bad (_("bad register name `%s'"), base_string);
5122 return 0;
5123 }
5124
5125 /* Check for scale factor. */
5126 if (*base_string != ')')
5127 {
5128 char *end_scale = i386_scale (base_string);
5129
5130 if (!end_scale)
5131 return 0;
5132
5133 base_string = end_scale;
5134 if (is_space_char (*base_string))
5135 ++base_string;
5136 if (*base_string != ')')
5137 {
5138 as_bad (_("expecting `)' after scale factor in `%s'"),
5139 operand_string);
5140 return 0;
5141 }
5142 }
5143 else if (!i.index_reg)
5144 {
5145 as_bad (_("expecting index register or scale factor after `,'; got '%c'"),
5146 *base_string);
5147 return 0;
5148 }
5149 }
5150 else if (*base_string != ')')
5151 {
5152 as_bad (_("expecting `,' or `)' after base register in `%s'"),
5153 operand_string);
5154 return 0;
5155 }
5156 }
5157 else if (*base_string == REGISTER_PREFIX)
5158 {
5159 as_bad (_("bad register name `%s'"), base_string);
5160 return 0;
5161 }
5162 }
5163
5164 /* If there's an expression beginning the operand, parse it,
5165 assuming displacement_string_start and
5166 displacement_string_end are meaningful. */
5167 if (displacement_string_start != displacement_string_end)
5168 {
5169 if (!i386_displacement (displacement_string_start,
5170 displacement_string_end))
5171 return 0;
5172 }
5173
5174 /* Special case for (%dx) while doing input/output op. */
5175 if (i.base_reg
5176 && i.base_reg->reg_type == (Reg16 | InOutPortReg)
5177 && i.index_reg == 0
5178 && i.log2_scale_factor == 0
5179 && i.seg[i.mem_operands] == 0
5180 && (i.types[this_operand] & Disp) == 0)
5181 {
5182 i.types[this_operand] = InOutPortReg;
5183 return 1;
5184 }
5185
5186 if (i386_index_check (operand_string) == 0)
5187 return 0;
5188 i.mem_operands++;
5189 }
5190 else
5191 {
5192 /* It's not a memory operand; argh! */
5193 as_bad (_("invalid char %s beginning operand %d `%s'"),
5194 output_invalid (*op_string),
5195 this_operand + 1,
5196 op_string);
5197 return 0;
5198 }
5199 return 1; /* Normal return. */
5200 }
5201 \f
5202 /* md_estimate_size_before_relax()
5203
5204 Called just before relax() for rs_machine_dependent frags. The x86
5205 assembler uses these frags to handle variable size jump
5206 instructions.
5207
5208 Any symbol that is now undefined will not become defined.
5209 Return the correct fr_subtype in the frag.
5210 Return the initial "guess for variable size of frag" to caller.
5211 The guess is actually the growth beyond the fixed part. Whatever
5212 we do to grow the fixed or variable part contributes to our
5213 returned value. */
5214
5215 int
5216 md_estimate_size_before_relax (fragP, segment)
5217 fragS *fragP;
5218 segT segment;
5219 {
5220 /* We've already got fragP->fr_subtype right; all we have to do is
5221 check for un-relaxable symbols. On an ELF system, we can't relax
5222 an externally visible symbol, because it may be overridden by a
5223 shared library. */
5224 if (S_GET_SEGMENT (fragP->fr_symbol) != segment
5225 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5226 || (IS_ELF
5227 && (S_IS_EXTERNAL (fragP->fr_symbol)
5228 || S_IS_WEAK (fragP->fr_symbol)))
5229 #endif
5230 )
5231 {
5232 /* Symbol is undefined in this segment, or we need to keep a
5233 reloc so that weak symbols can be overridden. */
5234 int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
5235 enum bfd_reloc_code_real reloc_type;
5236 unsigned char *opcode;
5237 int old_fr_fix;
5238
5239 if (fragP->fr_var != NO_RELOC)
5240 reloc_type = fragP->fr_var;
5241 else if (size == 2)
5242 reloc_type = BFD_RELOC_16_PCREL;
5243 else
5244 reloc_type = BFD_RELOC_32_PCREL;
5245
5246 old_fr_fix = fragP->fr_fix;
5247 opcode = (unsigned char *) fragP->fr_opcode;
5248
5249 switch (TYPE_FROM_RELAX_STATE (fragP->fr_subtype))
5250 {
5251 case UNCOND_JUMP:
5252 /* Make jmp (0xeb) a (d)word displacement jump. */
5253 opcode[0] = 0xe9;
5254 fragP->fr_fix += size;
5255 fix_new (fragP, old_fr_fix, size,
5256 fragP->fr_symbol,
5257 fragP->fr_offset, 1,
5258 reloc_type);
5259 break;
5260
5261 case COND_JUMP86:
5262 if (size == 2
5263 && (!no_cond_jump_promotion || fragP->fr_var != NO_RELOC))
5264 {
5265 /* Negate the condition, and branch past an
5266 unconditional jump. */
5267 opcode[0] ^= 1;
5268 opcode[1] = 3;
5269 /* Insert an unconditional jump. */
5270 opcode[2] = 0xe9;
5271 /* We added two extra opcode bytes, and have a two byte
5272 offset. */
5273 fragP->fr_fix += 2 + 2;
5274 fix_new (fragP, old_fr_fix + 2, 2,
5275 fragP->fr_symbol,
5276 fragP->fr_offset, 1,
5277 reloc_type);
5278 break;
5279 }
5280 /* Fall through. */
5281
5282 case COND_JUMP:
5283 if (no_cond_jump_promotion && fragP->fr_var == NO_RELOC)
5284 {
5285 fixS *fixP;
5286
5287 fragP->fr_fix += 1;
5288 fixP = fix_new (fragP, old_fr_fix, 1,
5289 fragP->fr_symbol,
5290 fragP->fr_offset, 1,
5291 BFD_RELOC_8_PCREL);
5292 fixP->fx_signed = 1;
5293 break;
5294 }
5295
5296 /* This changes the byte-displacement jump 0x7N
5297 to the (d)word-displacement jump 0x0f,0x8N. */
5298 opcode[1] = opcode[0] + 0x10;
5299 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
5300 /* We've added an opcode byte. */
5301 fragP->fr_fix += 1 + size;
5302 fix_new (fragP, old_fr_fix + 1, size,
5303 fragP->fr_symbol,
5304 fragP->fr_offset, 1,
5305 reloc_type);
5306 break;
5307
5308 default:
5309 BAD_CASE (fragP->fr_subtype);
5310 break;
5311 }
5312 frag_wane (fragP);
5313 return fragP->fr_fix - old_fr_fix;
5314 }
5315
5316 /* Guess size depending on current relax state. Initially the relax
5317 state will correspond to a short jump and we return 1, because
5318 the variable part of the frag (the branch offset) is one byte
5319 long. However, we can relax a section more than once and in that
5320 case we must either set fr_subtype back to the unrelaxed state,
5321 or return the value for the appropriate branch. */
5322 return md_relax_table[fragP->fr_subtype].rlx_length;
5323 }
5324
5325 /* Called after relax() is finished.
5326
5327 In: Address of frag.
5328 fr_type == rs_machine_dependent.
5329 fr_subtype is what the address relaxed to.
5330
5331 Out: Any fixSs and constants are set up.
5332 Caller will turn frag into a ".space 0". */
5333
5334 void
5335 md_convert_frag (abfd, sec, fragP)
5336 bfd *abfd ATTRIBUTE_UNUSED;
5337 segT sec ATTRIBUTE_UNUSED;
5338 fragS *fragP;
5339 {
5340 unsigned char *opcode;
5341 unsigned char *where_to_put_displacement = NULL;
5342 offsetT target_address;
5343 offsetT opcode_address;
5344 unsigned int extension = 0;
5345 offsetT displacement_from_opcode_start;
5346
5347 opcode = (unsigned char *) fragP->fr_opcode;
5348
5349 /* Address we want to reach in file space. */
5350 target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
5351
5352 /* Address opcode resides at in file space. */
5353 opcode_address = fragP->fr_address + fragP->fr_fix;
5354
5355 /* Displacement from opcode start to fill into instruction. */
5356 displacement_from_opcode_start = target_address - opcode_address;
5357
5358 if ((fragP->fr_subtype & BIG) == 0)
5359 {
5360 /* Don't have to change opcode. */
5361 extension = 1; /* 1 opcode + 1 displacement */
5362 where_to_put_displacement = &opcode[1];
5363 }
5364 else
5365 {
5366 if (no_cond_jump_promotion
5367 && TYPE_FROM_RELAX_STATE (fragP->fr_subtype) != UNCOND_JUMP)
5368 as_warn_where (fragP->fr_file, fragP->fr_line, _("long jump required"));
5369
5370 switch (fragP->fr_subtype)
5371 {
5372 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
5373 extension = 4; /* 1 opcode + 4 displacement */
5374 opcode[0] = 0xe9;
5375 where_to_put_displacement = &opcode[1];
5376 break;
5377
5378 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
5379 extension = 2; /* 1 opcode + 2 displacement */
5380 opcode[0] = 0xe9;
5381 where_to_put_displacement = &opcode[1];
5382 break;
5383
5384 case ENCODE_RELAX_STATE (COND_JUMP, BIG):
5385 case ENCODE_RELAX_STATE (COND_JUMP86, BIG):
5386 extension = 5; /* 2 opcode + 4 displacement */
5387 opcode[1] = opcode[0] + 0x10;
5388 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
5389 where_to_put_displacement = &opcode[2];
5390 break;
5391
5392 case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
5393 extension = 3; /* 2 opcode + 2 displacement */
5394 opcode[1] = opcode[0] + 0x10;
5395 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
5396 where_to_put_displacement = &opcode[2];
5397 break;
5398
5399 case ENCODE_RELAX_STATE (COND_JUMP86, BIG16):
5400 extension = 4;
5401 opcode[0] ^= 1;
5402 opcode[1] = 3;
5403 opcode[2] = 0xe9;
5404 where_to_put_displacement = &opcode[3];
5405 break;
5406
5407 default:
5408 BAD_CASE (fragP->fr_subtype);
5409 break;
5410 }
5411 }
5412
5413 /* If size if less then four we are sure that the operand fits,
5414 but if it's 4, then it could be that the displacement is larger
5415 then -/+ 2GB. */
5416 if (DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype) == 4
5417 && object_64bit
5418 && ((addressT) (displacement_from_opcode_start - extension
5419 + ((addressT) 1 << 31))
5420 > (((addressT) 2 << 31) - 1)))
5421 {
5422 as_bad_where (fragP->fr_file, fragP->fr_line,
5423 _("jump target out of range"));
5424 /* Make us emit 0. */
5425 displacement_from_opcode_start = extension;
5426 }
5427 /* Now put displacement after opcode. */
5428 md_number_to_chars ((char *) where_to_put_displacement,
5429 (valueT) (displacement_from_opcode_start - extension),
5430 DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
5431 fragP->fr_fix += extension;
5432 }
5433 \f
5434 /* Size of byte displacement jmp. */
5435 int md_short_jump_size = 2;
5436
5437 /* Size of dword displacement jmp. */
5438 int md_long_jump_size = 5;
5439
5440 void
5441 md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
5442 char *ptr;
5443 addressT from_addr, to_addr;
5444 fragS *frag ATTRIBUTE_UNUSED;
5445 symbolS *to_symbol ATTRIBUTE_UNUSED;
5446 {
5447 offsetT offset;
5448
5449 offset = to_addr - (from_addr + 2);
5450 /* Opcode for byte-disp jump. */
5451 md_number_to_chars (ptr, (valueT) 0xeb, 1);
5452 md_number_to_chars (ptr + 1, (valueT) offset, 1);
5453 }
5454
5455 void
5456 md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
5457 char *ptr;
5458 addressT from_addr, to_addr;
5459 fragS *frag ATTRIBUTE_UNUSED;
5460 symbolS *to_symbol ATTRIBUTE_UNUSED;
5461 {
5462 offsetT offset;
5463
5464 offset = to_addr - (from_addr + 5);
5465 md_number_to_chars (ptr, (valueT) 0xe9, 1);
5466 md_number_to_chars (ptr + 1, (valueT) offset, 4);
5467 }
5468 \f
5469 /* Apply a fixup (fixS) to segment data, once it has been determined
5470 by our caller that we have all the info we need to fix it up.
5471
5472 On the 386, immediates, displacements, and data pointers are all in
5473 the same (little-endian) format, so we don't need to care about which
5474 we are handling. */
5475
5476 void
5477 md_apply_fix (fixP, valP, seg)
5478 /* The fix we're to put in. */
5479 fixS *fixP;
5480 /* Pointer to the value of the bits. */
5481 valueT *valP;
5482 /* Segment fix is from. */
5483 segT seg ATTRIBUTE_UNUSED;
5484 {
5485 char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
5486 valueT value = *valP;
5487
5488 #if !defined (TE_Mach)
5489 if (fixP->fx_pcrel)
5490 {
5491 switch (fixP->fx_r_type)
5492 {
5493 default:
5494 break;
5495
5496 case BFD_RELOC_64:
5497 fixP->fx_r_type = BFD_RELOC_64_PCREL;
5498 break;
5499 case BFD_RELOC_32:
5500 case BFD_RELOC_X86_64_32S:
5501 fixP->fx_r_type = BFD_RELOC_32_PCREL;
5502 break;
5503 case BFD_RELOC_16:
5504 fixP->fx_r_type = BFD_RELOC_16_PCREL;
5505 break;
5506 case BFD_RELOC_8:
5507 fixP->fx_r_type = BFD_RELOC_8_PCREL;
5508 break;
5509 }
5510 }
5511
5512 if (fixP->fx_addsy != NULL
5513 && (fixP->fx_r_type == BFD_RELOC_32_PCREL
5514 || fixP->fx_r_type == BFD_RELOC_64_PCREL
5515 || fixP->fx_r_type == BFD_RELOC_16_PCREL
5516 || fixP->fx_r_type == BFD_RELOC_8_PCREL)
5517 && !use_rela_relocations)
5518 {
5519 /* This is a hack. There should be a better way to handle this.
5520 This covers for the fact that bfd_install_relocation will
5521 subtract the current location (for partial_inplace, PC relative
5522 relocations); see more below. */
5523 #ifndef OBJ_AOUT
5524 if (IS_ELF
5525 #ifdef TE_PE
5526 || OUTPUT_FLAVOR == bfd_target_coff_flavour
5527 #endif
5528 )
5529 value += fixP->fx_where + fixP->fx_frag->fr_address;
5530 #endif
5531 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5532 if (IS_ELF)
5533 {
5534 segT sym_seg = S_GET_SEGMENT (fixP->fx_addsy);
5535
5536 if ((sym_seg == seg
5537 || (symbol_section_p (fixP->fx_addsy)
5538 && sym_seg != absolute_section))
5539 && !generic_force_reloc (fixP))
5540 {
5541 /* Yes, we add the values in twice. This is because
5542 bfd_install_relocation subtracts them out again. I think
5543 bfd_install_relocation is broken, but I don't dare change
5544 it. FIXME. */
5545 value += fixP->fx_where + fixP->fx_frag->fr_address;
5546 }
5547 }
5548 #endif
5549 #if defined (OBJ_COFF) && defined (TE_PE)
5550 /* For some reason, the PE format does not store a
5551 section address offset for a PC relative symbol. */
5552 if (S_GET_SEGMENT (fixP->fx_addsy) != seg
5553 || S_IS_WEAK (fixP->fx_addsy))
5554 value += md_pcrel_from (fixP);
5555 #endif
5556 }
5557
5558 /* Fix a few things - the dynamic linker expects certain values here,
5559 and we must not disappoint it. */
5560 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5561 if (IS_ELF && fixP->fx_addsy)
5562 switch (fixP->fx_r_type)
5563 {
5564 case BFD_RELOC_386_PLT32:
5565 case BFD_RELOC_X86_64_PLT32:
5566 /* Make the jump instruction point to the address of the operand. At
5567 runtime we merely add the offset to the actual PLT entry. */
5568 value = -4;
5569 break;
5570
5571 case BFD_RELOC_386_TLS_GD:
5572 case BFD_RELOC_386_TLS_LDM:
5573 case BFD_RELOC_386_TLS_IE_32:
5574 case BFD_RELOC_386_TLS_IE:
5575 case BFD_RELOC_386_TLS_GOTIE:
5576 case BFD_RELOC_386_TLS_GOTDESC:
5577 case BFD_RELOC_X86_64_TLSGD:
5578 case BFD_RELOC_X86_64_TLSLD:
5579 case BFD_RELOC_X86_64_GOTTPOFF:
5580 case BFD_RELOC_X86_64_GOTPC32_TLSDESC:
5581 value = 0; /* Fully resolved at runtime. No addend. */
5582 /* Fallthrough */
5583 case BFD_RELOC_386_TLS_LE:
5584 case BFD_RELOC_386_TLS_LDO_32:
5585 case BFD_RELOC_386_TLS_LE_32:
5586 case BFD_RELOC_X86_64_DTPOFF32:
5587 case BFD_RELOC_X86_64_DTPOFF64:
5588 case BFD_RELOC_X86_64_TPOFF32:
5589 case BFD_RELOC_X86_64_TPOFF64:
5590 S_SET_THREAD_LOCAL (fixP->fx_addsy);
5591 break;
5592
5593 case BFD_RELOC_386_TLS_DESC_CALL:
5594 case BFD_RELOC_X86_64_TLSDESC_CALL:
5595 value = 0; /* Fully resolved at runtime. No addend. */
5596 S_SET_THREAD_LOCAL (fixP->fx_addsy);
5597 fixP->fx_done = 0;
5598 return;
5599
5600 case BFD_RELOC_386_GOT32:
5601 case BFD_RELOC_X86_64_GOT32:
5602 value = 0; /* Fully resolved at runtime. No addend. */
5603 break;
5604
5605 case BFD_RELOC_VTABLE_INHERIT:
5606 case BFD_RELOC_VTABLE_ENTRY:
5607 fixP->fx_done = 0;
5608 return;
5609
5610 default:
5611 break;
5612 }
5613 #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
5614 *valP = value;
5615 #endif /* !defined (TE_Mach) */
5616
5617 /* Are we finished with this relocation now? */
5618 if (fixP->fx_addsy == NULL)
5619 fixP->fx_done = 1;
5620 else if (use_rela_relocations)
5621 {
5622 fixP->fx_no_overflow = 1;
5623 /* Remember value for tc_gen_reloc. */
5624 fixP->fx_addnumber = value;
5625 value = 0;
5626 }
5627
5628 md_number_to_chars (p, value, fixP->fx_size);
5629 }
5630 \f
5631 #define MAX_LITTLENUMS 6
5632
5633 /* Turn the string pointed to by litP into a floating point constant
5634 of type TYPE, and emit the appropriate bytes. The number of
5635 LITTLENUMS emitted is stored in *SIZEP. An error message is
5636 returned, or NULL on OK. */
5637
5638 char *
5639 md_atof (type, litP, sizeP)
5640 int type;
5641 char *litP;
5642 int *sizeP;
5643 {
5644 int prec;
5645 LITTLENUM_TYPE words[MAX_LITTLENUMS];
5646 LITTLENUM_TYPE *wordP;
5647 char *t;
5648
5649 switch (type)
5650 {
5651 case 'f':
5652 case 'F':
5653 prec = 2;
5654 break;
5655
5656 case 'd':
5657 case 'D':
5658 prec = 4;
5659 break;
5660
5661 case 'x':
5662 case 'X':
5663 prec = 5;
5664 break;
5665
5666 default:
5667 *sizeP = 0;
5668 return _("Bad call to md_atof ()");
5669 }
5670 t = atof_ieee (input_line_pointer, type, words);
5671 if (t)
5672 input_line_pointer = t;
5673
5674 *sizeP = prec * sizeof (LITTLENUM_TYPE);
5675 /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
5676 the bigendian 386. */
5677 for (wordP = words + prec - 1; prec--;)
5678 {
5679 md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
5680 litP += sizeof (LITTLENUM_TYPE);
5681 }
5682 return 0;
5683 }
5684 \f
5685 static char output_invalid_buf[sizeof (unsigned char) * 2 + 6];
5686
5687 static char *
5688 output_invalid (c)
5689 int c;
5690 {
5691 if (ISPRINT (c))
5692 snprintf (output_invalid_buf, sizeof (output_invalid_buf),
5693 "'%c'", c);
5694 else
5695 snprintf (output_invalid_buf, sizeof (output_invalid_buf),
5696 "(0x%x)", (unsigned char) c);
5697 return output_invalid_buf;
5698 }
5699
5700 /* REG_STRING starts *before* REGISTER_PREFIX. */
5701
5702 static const reg_entry *
5703 parse_real_register (char *reg_string, char **end_op)
5704 {
5705 char *s = reg_string;
5706 char *p;
5707 char reg_name_given[MAX_REG_NAME_SIZE + 1];
5708 const reg_entry *r;
5709
5710 /* Skip possible REGISTER_PREFIX and possible whitespace. */
5711 if (*s == REGISTER_PREFIX)
5712 ++s;
5713
5714 if (is_space_char (*s))
5715 ++s;
5716
5717 p = reg_name_given;
5718 while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
5719 {
5720 if (p >= reg_name_given + MAX_REG_NAME_SIZE)
5721 return (const reg_entry *) NULL;
5722 s++;
5723 }
5724
5725 /* For naked regs, make sure that we are not dealing with an identifier.
5726 This prevents confusing an identifier like `eax_var' with register
5727 `eax'. */
5728 if (allow_naked_reg && identifier_chars[(unsigned char) *s])
5729 return (const reg_entry *) NULL;
5730
5731 *end_op = s;
5732
5733 r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
5734
5735 /* Handle floating point regs, allowing spaces in the (i) part. */
5736 if (r == i386_regtab /* %st is first entry of table */)
5737 {
5738 if (is_space_char (*s))
5739 ++s;
5740 if (*s == '(')
5741 {
5742 ++s;
5743 if (is_space_char (*s))
5744 ++s;
5745 if (*s >= '0' && *s <= '7')
5746 {
5747 r = &i386_float_regtab[*s - '0'];
5748 ++s;
5749 if (is_space_char (*s))
5750 ++s;
5751 if (*s == ')')
5752 {
5753 *end_op = s + 1;
5754 return r;
5755 }
5756 }
5757 /* We have "%st(" then garbage. */
5758 return (const reg_entry *) NULL;
5759 }
5760 }
5761
5762 if (r != NULL
5763 && ((r->reg_flags & (RegRex64 | RegRex)) | (r->reg_type & Reg64)) != 0
5764 && (r->reg_type != Control || !(cpu_arch_flags & CpuSledgehammer))
5765 && flag_code != CODE_64BIT)
5766 return (const reg_entry *) NULL;
5767
5768 return r;
5769 }
5770
5771 /* REG_STRING starts *before* REGISTER_PREFIX. */
5772
5773 static const reg_entry *
5774 parse_register (char *reg_string, char **end_op)
5775 {
5776 const reg_entry *r;
5777
5778 if (*reg_string == REGISTER_PREFIX || allow_naked_reg)
5779 r = parse_real_register (reg_string, end_op);
5780 else
5781 r = NULL;
5782 if (!r)
5783 {
5784 char *save = input_line_pointer;
5785 char c;
5786 symbolS *symbolP;
5787
5788 input_line_pointer = reg_string;
5789 c = get_symbol_end ();
5790 symbolP = symbol_find (reg_string);
5791 if (symbolP && S_GET_SEGMENT (symbolP) == reg_section)
5792 {
5793 const expressionS *e = symbol_get_value_expression (symbolP);
5794
5795 know (e->X_op == O_register);
5796 know (e->X_add_number >= 0 && (valueT) e->X_add_number < ARRAY_SIZE (i386_regtab));
5797 r = i386_regtab + e->X_add_number;
5798 *end_op = input_line_pointer;
5799 }
5800 *input_line_pointer = c;
5801 input_line_pointer = save;
5802 }
5803 return r;
5804 }
5805
5806 int
5807 i386_parse_name (char *name, expressionS *e, char *nextcharP)
5808 {
5809 const reg_entry *r;
5810 char *end = input_line_pointer;
5811
5812 *end = *nextcharP;
5813 r = parse_register (name, &input_line_pointer);
5814 if (r && end <= input_line_pointer)
5815 {
5816 *nextcharP = *input_line_pointer;
5817 *input_line_pointer = 0;
5818 e->X_op = O_register;
5819 e->X_add_number = r - i386_regtab;
5820 return 1;
5821 }
5822 input_line_pointer = end;
5823 *end = 0;
5824 return 0;
5825 }
5826
5827 void
5828 md_operand (expressionS *e)
5829 {
5830 if (*input_line_pointer == REGISTER_PREFIX)
5831 {
5832 char *end;
5833 const reg_entry *r = parse_real_register (input_line_pointer, &end);
5834
5835 if (r)
5836 {
5837 e->X_op = O_register;
5838 e->X_add_number = r - i386_regtab;
5839 input_line_pointer = end;
5840 }
5841 }
5842 }
5843
5844 \f
5845 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5846 const char *md_shortopts = "kVQ:sqn";
5847 #else
5848 const char *md_shortopts = "qn";
5849 #endif
5850
5851 #define OPTION_32 (OPTION_MD_BASE + 0)
5852 #define OPTION_64 (OPTION_MD_BASE + 1)
5853 #define OPTION_DIVIDE (OPTION_MD_BASE + 2)
5854 #define OPTION_MARCH (OPTION_MD_BASE + 3)
5855 #define OPTION_MTUNE (OPTION_MD_BASE + 4)
5856
5857 struct option md_longopts[] =
5858 {
5859 {"32", no_argument, NULL, OPTION_32},
5860 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) || defined(TE_PEP)
5861 {"64", no_argument, NULL, OPTION_64},
5862 #endif
5863 {"divide", no_argument, NULL, OPTION_DIVIDE},
5864 {"march", required_argument, NULL, OPTION_MARCH},
5865 {"mtune", required_argument, NULL, OPTION_MTUNE},
5866 {NULL, no_argument, NULL, 0}
5867 };
5868 size_t md_longopts_size = sizeof (md_longopts);
5869
5870 int
5871 md_parse_option (int c, char *arg)
5872 {
5873 unsigned int i;
5874
5875 switch (c)
5876 {
5877 case 'n':
5878 optimize_align_code = 0;
5879 break;
5880
5881 case 'q':
5882 quiet_warnings = 1;
5883 break;
5884
5885 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5886 /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
5887 should be emitted or not. FIXME: Not implemented. */
5888 case 'Q':
5889 break;
5890
5891 /* -V: SVR4 argument to print version ID. */
5892 case 'V':
5893 print_version_id ();
5894 break;
5895
5896 /* -k: Ignore for FreeBSD compatibility. */
5897 case 'k':
5898 break;
5899
5900 case 's':
5901 /* -s: On i386 Solaris, this tells the native assembler to use
5902 .stab instead of .stab.excl. We always use .stab anyhow. */
5903 break;
5904 #endif
5905 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) || defined(TE_PEP)
5906 case OPTION_64:
5907 {
5908 const char **list, **l;
5909
5910 list = bfd_target_list ();
5911 for (l = list; *l != NULL; l++)
5912 if (CONST_STRNEQ (*l, "elf64-x86-64")
5913 || strcmp (*l, "coff-x86-64") == 0
5914 || strcmp (*l, "pe-x86-64") == 0
5915 || strcmp (*l, "pei-x86-64") == 0)
5916 {
5917 default_arch = "x86_64";
5918 break;
5919 }
5920 if (*l == NULL)
5921 as_fatal (_("No compiled in support for x86_64"));
5922 free (list);
5923 }
5924 break;
5925 #endif
5926
5927 case OPTION_32:
5928 default_arch = "i386";
5929 break;
5930
5931 case OPTION_DIVIDE:
5932 #ifdef SVR4_COMMENT_CHARS
5933 {
5934 char *n, *t;
5935 const char *s;
5936
5937 n = (char *) xmalloc (strlen (i386_comment_chars) + 1);
5938 t = n;
5939 for (s = i386_comment_chars; *s != '\0'; s++)
5940 if (*s != '/')
5941 *t++ = *s;
5942 *t = '\0';
5943 i386_comment_chars = n;
5944 }
5945 #endif
5946 break;
5947
5948 case OPTION_MARCH:
5949 if (*arg == '.')
5950 as_fatal (_("Invalid -march= option: `%s'"), arg);
5951 for (i = 0; i < ARRAY_SIZE (cpu_arch); i++)
5952 {
5953 if (strcmp (arg, cpu_arch [i].name) == 0)
5954 {
5955 cpu_arch_isa = cpu_arch[i].type;
5956 cpu_arch_isa_flags = cpu_arch[i].flags;
5957 if (!cpu_arch_tune_set)
5958 {
5959 cpu_arch_tune = cpu_arch_isa;
5960 cpu_arch_tune_flags = cpu_arch_isa_flags;
5961 }
5962 break;
5963 }
5964 }
5965 if (i >= ARRAY_SIZE (cpu_arch))
5966 as_fatal (_("Invalid -march= option: `%s'"), arg);
5967 break;
5968
5969 case OPTION_MTUNE:
5970 if (*arg == '.')
5971 as_fatal (_("Invalid -mtune= option: `%s'"), arg);
5972 for (i = 0; i < ARRAY_SIZE (cpu_arch); i++)
5973 {
5974 if (strcmp (arg, cpu_arch [i].name) == 0)
5975 {
5976 cpu_arch_tune_set = 1;
5977 cpu_arch_tune = cpu_arch [i].type;
5978 cpu_arch_tune_flags = cpu_arch[i].flags;
5979 break;
5980 }
5981 }
5982 if (i >= ARRAY_SIZE (cpu_arch))
5983 as_fatal (_("Invalid -mtune= option: `%s'"), arg);
5984 break;
5985
5986 default:
5987 return 0;
5988 }
5989 return 1;
5990 }
5991
5992 void
5993 md_show_usage (stream)
5994 FILE *stream;
5995 {
5996 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5997 fprintf (stream, _("\
5998 -Q ignored\n\
5999 -V print assembler version number\n\
6000 -k ignored\n"));
6001 #endif
6002 fprintf (stream, _("\
6003 -n Do not optimize code alignment\n\
6004 -q quieten some warnings\n"));
6005 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
6006 fprintf (stream, _("\
6007 -s ignored\n"));
6008 #endif
6009 #ifdef SVR4_COMMENT_CHARS
6010 fprintf (stream, _("\
6011 --divide do not treat `/' as a comment character\n"));
6012 #else
6013 fprintf (stream, _("\
6014 --divide ignored\n"));
6015 #endif
6016 fprintf (stream, _("\
6017 -march=CPU/-mtune=CPU generate code/optimize for CPU, where CPU is one of:\n\
6018 i386, i486, pentium, pentiumpro, pentium4, nocona,\n\
6019 core, core2, k6, athlon, k8, generic32, generic64\n"));
6020
6021 }
6022
6023 #if defined(TE_PEP)
6024 const char *
6025 x86_64_target_format (void)
6026 {
6027 if (strcmp (default_arch, "x86_64") == 0)
6028 {
6029 set_code_flag (CODE_64BIT);
6030 return COFF_TARGET_FORMAT;
6031 }
6032 else if (strcmp (default_arch, "i386") == 0)
6033 {
6034 set_code_flag (CODE_32BIT);
6035 return "coff-i386";
6036 }
6037
6038 as_fatal (_("Unknown architecture"));
6039 return NULL;
6040 }
6041 #endif
6042
6043 #if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
6044 || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
6045
6046 /* Pick the target format to use. */
6047
6048 const char *
6049 i386_target_format ()
6050 {
6051 if (!strcmp (default_arch, "x86_64"))
6052 {
6053 set_code_flag (CODE_64BIT);
6054 if (cpu_arch_isa_flags == 0)
6055 cpu_arch_isa_flags = Cpu186|Cpu286|Cpu386|Cpu486
6056 |Cpu586|Cpu686|CpuP4|CpuMMX|CpuMMX2
6057 |CpuSSE|CpuSSE2;
6058 if (cpu_arch_tune_flags == 0)
6059 cpu_arch_tune_flags = Cpu186|Cpu286|Cpu386|Cpu486
6060 |Cpu586|Cpu686|CpuP4|CpuMMX|CpuMMX2
6061 |CpuSSE|CpuSSE2;
6062 }
6063 else if (!strcmp (default_arch, "i386"))
6064 {
6065 set_code_flag (CODE_32BIT);
6066 if (cpu_arch_isa_flags == 0)
6067 cpu_arch_isa_flags = Cpu186|Cpu286|Cpu386;
6068 if (cpu_arch_tune_flags == 0)
6069 cpu_arch_tune_flags = Cpu186|Cpu286|Cpu386;
6070 }
6071 else
6072 as_fatal (_("Unknown architecture"));
6073 switch (OUTPUT_FLAVOR)
6074 {
6075 #ifdef OBJ_MAYBE_AOUT
6076 case bfd_target_aout_flavour:
6077 return AOUT_TARGET_FORMAT;
6078 #endif
6079 #ifdef OBJ_MAYBE_COFF
6080 case bfd_target_coff_flavour:
6081 return "coff-i386";
6082 #endif
6083 #if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF)
6084 case bfd_target_elf_flavour:
6085 {
6086 if (flag_code == CODE_64BIT)
6087 {
6088 object_64bit = 1;
6089 use_rela_relocations = 1;
6090 }
6091 return flag_code == CODE_64BIT ? ELF_TARGET_FORMAT64 : ELF_TARGET_FORMAT;
6092 }
6093 #endif
6094 default:
6095 abort ();
6096 return NULL;
6097 }
6098 }
6099
6100 #endif /* OBJ_MAYBE_ more than one */
6101
6102 #if (defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
6103 void i386_elf_emit_arch_note ()
6104 {
6105 if (IS_ELF && cpu_arch_name != NULL)
6106 {
6107 char *p;
6108 asection *seg = now_seg;
6109 subsegT subseg = now_subseg;
6110 Elf_Internal_Note i_note;
6111 Elf_External_Note e_note;
6112 asection *note_secp;
6113 int len;
6114
6115 /* Create the .note section. */
6116 note_secp = subseg_new (".note", 0);
6117 bfd_set_section_flags (stdoutput,
6118 note_secp,
6119 SEC_HAS_CONTENTS | SEC_READONLY);
6120
6121 /* Process the arch string. */
6122 len = strlen (cpu_arch_name);
6123
6124 i_note.namesz = len + 1;
6125 i_note.descsz = 0;
6126 i_note.type = NT_ARCH;
6127 p = frag_more (sizeof (e_note.namesz));
6128 md_number_to_chars (p, (valueT) i_note.namesz, sizeof (e_note.namesz));
6129 p = frag_more (sizeof (e_note.descsz));
6130 md_number_to_chars (p, (valueT) i_note.descsz, sizeof (e_note.descsz));
6131 p = frag_more (sizeof (e_note.type));
6132 md_number_to_chars (p, (valueT) i_note.type, sizeof (e_note.type));
6133 p = frag_more (len + 1);
6134 strcpy (p, cpu_arch_name);
6135
6136 frag_align (2, 0, 0);
6137
6138 subseg_set (seg, subseg);
6139 }
6140 }
6141 #endif
6142 \f
6143 symbolS *
6144 md_undefined_symbol (name)
6145 char *name;
6146 {
6147 if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
6148 && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
6149 && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
6150 && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
6151 {
6152 if (!GOT_symbol)
6153 {
6154 if (symbol_find (name))
6155 as_bad (_("GOT already in symbol table"));
6156 GOT_symbol = symbol_new (name, undefined_section,
6157 (valueT) 0, &zero_address_frag);
6158 };
6159 return GOT_symbol;
6160 }
6161 return 0;
6162 }
6163
6164 /* Round up a section size to the appropriate boundary. */
6165
6166 valueT
6167 md_section_align (segment, size)
6168 segT segment ATTRIBUTE_UNUSED;
6169 valueT size;
6170 {
6171 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
6172 if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
6173 {
6174 /* For a.out, force the section size to be aligned. If we don't do
6175 this, BFD will align it for us, but it will not write out the
6176 final bytes of the section. This may be a bug in BFD, but it is
6177 easier to fix it here since that is how the other a.out targets
6178 work. */
6179 int align;
6180
6181 align = bfd_get_section_alignment (stdoutput, segment);
6182 size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
6183 }
6184 #endif
6185
6186 return size;
6187 }
6188
6189 /* On the i386, PC-relative offsets are relative to the start of the
6190 next instruction. That is, the address of the offset, plus its
6191 size, since the offset is always the last part of the insn. */
6192
6193 long
6194 md_pcrel_from (fixP)
6195 fixS *fixP;
6196 {
6197 return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
6198 }
6199
6200 #ifndef I386COFF
6201
6202 static void
6203 s_bss (ignore)
6204 int ignore ATTRIBUTE_UNUSED;
6205 {
6206 int temp;
6207
6208 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
6209 if (IS_ELF)
6210 obj_elf_section_change_hook ();
6211 #endif
6212 temp = get_absolute_expression ();
6213 subseg_set (bss_section, (subsegT) temp);
6214 demand_empty_rest_of_line ();
6215 }
6216
6217 #endif
6218
6219 void
6220 i386_validate_fix (fixp)
6221 fixS *fixp;
6222 {
6223 if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
6224 {
6225 if (fixp->fx_r_type == BFD_RELOC_32_PCREL)
6226 {
6227 if (!object_64bit)
6228 abort ();
6229 fixp->fx_r_type = BFD_RELOC_X86_64_GOTPCREL;
6230 }
6231 else
6232 {
6233 if (!object_64bit)
6234 fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
6235 else
6236 fixp->fx_r_type = BFD_RELOC_X86_64_GOTOFF64;
6237 }
6238 fixp->fx_subsy = 0;
6239 }
6240 }
6241
6242 arelent *
6243 tc_gen_reloc (section, fixp)
6244 asection *section ATTRIBUTE_UNUSED;
6245 fixS *fixp;
6246 {
6247 arelent *rel;
6248 bfd_reloc_code_real_type code;
6249
6250 switch (fixp->fx_r_type)
6251 {
6252 case BFD_RELOC_X86_64_PLT32:
6253 case BFD_RELOC_X86_64_GOT32:
6254 case BFD_RELOC_X86_64_GOTPCREL:
6255 case BFD_RELOC_386_PLT32:
6256 case BFD_RELOC_386_GOT32:
6257 case BFD_RELOC_386_GOTOFF:
6258 case BFD_RELOC_386_GOTPC:
6259 case BFD_RELOC_386_TLS_GD:
6260 case BFD_RELOC_386_TLS_LDM:
6261 case BFD_RELOC_386_TLS_LDO_32:
6262 case BFD_RELOC_386_TLS_IE_32:
6263 case BFD_RELOC_386_TLS_IE:
6264 case BFD_RELOC_386_TLS_GOTIE:
6265 case BFD_RELOC_386_TLS_LE_32:
6266 case BFD_RELOC_386_TLS_LE:
6267 case BFD_RELOC_386_TLS_GOTDESC:
6268 case BFD_RELOC_386_TLS_DESC_CALL:
6269 case BFD_RELOC_X86_64_TLSGD:
6270 case BFD_RELOC_X86_64_TLSLD:
6271 case BFD_RELOC_X86_64_DTPOFF32:
6272 case BFD_RELOC_X86_64_DTPOFF64:
6273 case BFD_RELOC_X86_64_GOTTPOFF:
6274 case BFD_RELOC_X86_64_TPOFF32:
6275 case BFD_RELOC_X86_64_TPOFF64:
6276 case BFD_RELOC_X86_64_GOTOFF64:
6277 case BFD_RELOC_X86_64_GOTPC32:
6278 case BFD_RELOC_X86_64_GOT64:
6279 case BFD_RELOC_X86_64_GOTPCREL64:
6280 case BFD_RELOC_X86_64_GOTPC64:
6281 case BFD_RELOC_X86_64_GOTPLT64:
6282 case BFD_RELOC_X86_64_PLTOFF64:
6283 case BFD_RELOC_X86_64_GOTPC32_TLSDESC:
6284 case BFD_RELOC_X86_64_TLSDESC_CALL:
6285 case BFD_RELOC_RVA:
6286 case BFD_RELOC_VTABLE_ENTRY:
6287 case BFD_RELOC_VTABLE_INHERIT:
6288 #ifdef TE_PE
6289 case BFD_RELOC_32_SECREL:
6290 #endif
6291 code = fixp->fx_r_type;
6292 break;
6293 case BFD_RELOC_X86_64_32S:
6294 if (!fixp->fx_pcrel)
6295 {
6296 /* Don't turn BFD_RELOC_X86_64_32S into BFD_RELOC_32. */
6297 code = fixp->fx_r_type;
6298 break;
6299 }
6300 default:
6301 if (fixp->fx_pcrel)
6302 {
6303 switch (fixp->fx_size)
6304 {
6305 default:
6306 as_bad_where (fixp->fx_file, fixp->fx_line,
6307 _("can not do %d byte pc-relative relocation"),
6308 fixp->fx_size);
6309 code = BFD_RELOC_32_PCREL;
6310 break;
6311 case 1: code = BFD_RELOC_8_PCREL; break;
6312 case 2: code = BFD_RELOC_16_PCREL; break;
6313 case 4: code = BFD_RELOC_32_PCREL; break;
6314 #ifdef BFD64
6315 case 8: code = BFD_RELOC_64_PCREL; break;
6316 #endif
6317 }
6318 }
6319 else
6320 {
6321 switch (fixp->fx_size)
6322 {
6323 default:
6324 as_bad_where (fixp->fx_file, fixp->fx_line,
6325 _("can not do %d byte relocation"),
6326 fixp->fx_size);
6327 code = BFD_RELOC_32;
6328 break;
6329 case 1: code = BFD_RELOC_8; break;
6330 case 2: code = BFD_RELOC_16; break;
6331 case 4: code = BFD_RELOC_32; break;
6332 #ifdef BFD64
6333 case 8: code = BFD_RELOC_64; break;
6334 #endif
6335 }
6336 }
6337 break;
6338 }
6339
6340 if ((code == BFD_RELOC_32
6341 || code == BFD_RELOC_32_PCREL
6342 || code == BFD_RELOC_X86_64_32S)
6343 && GOT_symbol
6344 && fixp->fx_addsy == GOT_symbol)
6345 {
6346 if (!object_64bit)
6347 code = BFD_RELOC_386_GOTPC;
6348 else
6349 code = BFD_RELOC_X86_64_GOTPC32;
6350 }
6351 if ((code == BFD_RELOC_64 || code == BFD_RELOC_64_PCREL)
6352 && GOT_symbol
6353 && fixp->fx_addsy == GOT_symbol)
6354 {
6355 code = BFD_RELOC_X86_64_GOTPC64;
6356 }
6357
6358 rel = (arelent *) xmalloc (sizeof (arelent));
6359 rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
6360 *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
6361
6362 rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
6363
6364 if (!use_rela_relocations)
6365 {
6366 /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
6367 vtable entry to be used in the relocation's section offset. */
6368 if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
6369 rel->address = fixp->fx_offset;
6370
6371 rel->addend = 0;
6372 }
6373 /* Use the rela in 64bit mode. */
6374 else
6375 {
6376 if (!fixp->fx_pcrel)
6377 rel->addend = fixp->fx_offset;
6378 else
6379 switch (code)
6380 {
6381 case BFD_RELOC_X86_64_PLT32:
6382 case BFD_RELOC_X86_64_GOT32:
6383 case BFD_RELOC_X86_64_GOTPCREL:
6384 case BFD_RELOC_X86_64_TLSGD:
6385 case BFD_RELOC_X86_64_TLSLD:
6386 case BFD_RELOC_X86_64_GOTTPOFF:
6387 case BFD_RELOC_X86_64_GOTPC32_TLSDESC:
6388 case BFD_RELOC_X86_64_TLSDESC_CALL:
6389 rel->addend = fixp->fx_offset - fixp->fx_size;
6390 break;
6391 default:
6392 rel->addend = (section->vma
6393 - fixp->fx_size
6394 + fixp->fx_addnumber
6395 + md_pcrel_from (fixp));
6396 break;
6397 }
6398 }
6399
6400 rel->howto = bfd_reloc_type_lookup (stdoutput, code);
6401 if (rel->howto == NULL)
6402 {
6403 as_bad_where (fixp->fx_file, fixp->fx_line,
6404 _("cannot represent relocation type %s"),
6405 bfd_get_reloc_code_name (code));
6406 /* Set howto to a garbage value so that we can keep going. */
6407 rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
6408 assert (rel->howto != NULL);
6409 }
6410
6411 return rel;
6412 }
6413
6414 \f
6415 /* Parse operands using Intel syntax. This implements a recursive descent
6416 parser based on the BNF grammar published in Appendix B of the MASM 6.1
6417 Programmer's Guide.
6418
6419 FIXME: We do not recognize the full operand grammar defined in the MASM
6420 documentation. In particular, all the structure/union and
6421 high-level macro operands are missing.
6422
6423 Uppercase words are terminals, lower case words are non-terminals.
6424 Objects surrounded by double brackets '[[' ']]' are optional. Vertical
6425 bars '|' denote choices. Most grammar productions are implemented in
6426 functions called 'intel_<production>'.
6427
6428 Initial production is 'expr'.
6429
6430 addOp + | -
6431
6432 alpha [a-zA-Z]
6433
6434 binOp & | AND | \| | OR | ^ | XOR
6435
6436 byteRegister AL | AH | BL | BH | CL | CH | DL | DH
6437
6438 constant digits [[ radixOverride ]]
6439
6440 dataType BYTE | WORD | DWORD | FWORD | QWORD | TBYTE | OWORD | XMMWORD
6441
6442 digits decdigit
6443 | digits decdigit
6444 | digits hexdigit
6445
6446 decdigit [0-9]
6447
6448 e04 e04 addOp e05
6449 | e05
6450
6451 e05 e05 binOp e06
6452 | e06
6453
6454 e06 e06 mulOp e09
6455 | e09
6456
6457 e09 OFFSET e10
6458 | SHORT e10
6459 | + e10
6460 | - e10
6461 | ~ e10
6462 | NOT e10
6463 | e09 PTR e10
6464 | e09 : e10
6465 | e10
6466
6467 e10 e10 [ expr ]
6468 | e11
6469
6470 e11 ( expr )
6471 | [ expr ]
6472 | constant
6473 | dataType
6474 | id
6475 | $
6476 | register
6477
6478 => expr expr cmpOp e04
6479 | e04
6480
6481 gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
6482 | BP | EBP | SP | ESP | DI | EDI | SI | ESI
6483
6484 hexdigit a | b | c | d | e | f
6485 | A | B | C | D | E | F
6486
6487 id alpha
6488 | id alpha
6489 | id decdigit
6490
6491 mulOp * | / | % | MOD | << | SHL | >> | SHR
6492
6493 quote " | '
6494
6495 register specialRegister
6496 | gpRegister
6497 | byteRegister
6498
6499 segmentRegister CS | DS | ES | FS | GS | SS
6500
6501 specialRegister CR0 | CR2 | CR3 | CR4
6502 | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
6503 | TR3 | TR4 | TR5 | TR6 | TR7
6504
6505 We simplify the grammar in obvious places (e.g., register parsing is
6506 done by calling parse_register) and eliminate immediate left recursion
6507 to implement a recursive-descent parser.
6508
6509 expr e04 expr'
6510
6511 expr' cmpOp e04 expr'
6512 | Empty
6513
6514 e04 e05 e04'
6515
6516 e04' addOp e05 e04'
6517 | Empty
6518
6519 e05 e06 e05'
6520
6521 e05' binOp e06 e05'
6522 | Empty
6523
6524 e06 e09 e06'
6525
6526 e06' mulOp e09 e06'
6527 | Empty
6528
6529 e09 OFFSET e10 e09'
6530 | SHORT e10'
6531 | + e10'
6532 | - e10'
6533 | ~ e10'
6534 | NOT e10'
6535 | e10 e09'
6536
6537 e09' PTR e10 e09'
6538 | : e10 e09'
6539 | Empty
6540
6541 e10 e11 e10'
6542
6543 e10' [ expr ] e10'
6544 | Empty
6545
6546 e11 ( expr )
6547 | [ expr ]
6548 | BYTE
6549 | WORD
6550 | DWORD
6551 | FWORD
6552 | QWORD
6553 | TBYTE
6554 | OWORD
6555 | XMMWORD
6556 | .
6557 | $
6558 | register
6559 | id
6560 | constant */
6561
6562 /* Parsing structure for the intel syntax parser. Used to implement the
6563 semantic actions for the operand grammar. */
6564 struct intel_parser_s
6565 {
6566 char *op_string; /* The string being parsed. */
6567 int got_a_float; /* Whether the operand is a float. */
6568 int op_modifier; /* Operand modifier. */
6569 int is_mem; /* 1 if operand is memory reference. */
6570 int in_offset; /* >=1 if parsing operand of offset. */
6571 int in_bracket; /* >=1 if parsing operand in brackets. */
6572 const reg_entry *reg; /* Last register reference found. */
6573 char *disp; /* Displacement string being built. */
6574 char *next_operand; /* Resume point when splitting operands. */
6575 };
6576
6577 static struct intel_parser_s intel_parser;
6578
6579 /* Token structure for parsing intel syntax. */
6580 struct intel_token
6581 {
6582 int code; /* Token code. */
6583 const reg_entry *reg; /* Register entry for register tokens. */
6584 char *str; /* String representation. */
6585 };
6586
6587 static struct intel_token cur_token, prev_token;
6588
6589 /* Token codes for the intel parser. Since T_SHORT is already used
6590 by COFF, undefine it first to prevent a warning. */
6591 #define T_NIL -1
6592 #define T_CONST 1
6593 #define T_REG 2
6594 #define T_BYTE 3
6595 #define T_WORD 4
6596 #define T_DWORD 5
6597 #define T_FWORD 6
6598 #define T_QWORD 7
6599 #define T_TBYTE 8
6600 #define T_XMMWORD 9
6601 #undef T_SHORT
6602 #define T_SHORT 10
6603 #define T_OFFSET 11
6604 #define T_PTR 12
6605 #define T_ID 13
6606 #define T_SHL 14
6607 #define T_SHR 15
6608
6609 /* Prototypes for intel parser functions. */
6610 static int intel_match_token PARAMS ((int code));
6611 static void intel_get_token PARAMS ((void));
6612 static void intel_putback_token PARAMS ((void));
6613 static int intel_expr PARAMS ((void));
6614 static int intel_e04 PARAMS ((void));
6615 static int intel_e05 PARAMS ((void));
6616 static int intel_e06 PARAMS ((void));
6617 static int intel_e09 PARAMS ((void));
6618 static int intel_bracket_expr PARAMS ((void));
6619 static int intel_e10 PARAMS ((void));
6620 static int intel_e11 PARAMS ((void));
6621
6622 static int
6623 i386_intel_operand (operand_string, got_a_float)
6624 char *operand_string;
6625 int got_a_float;
6626 {
6627 int ret;
6628 char *p;
6629
6630 p = intel_parser.op_string = xstrdup (operand_string);
6631 intel_parser.disp = (char *) xmalloc (strlen (operand_string) + 1);
6632
6633 for (;;)
6634 {
6635 /* Initialize token holders. */
6636 cur_token.code = prev_token.code = T_NIL;
6637 cur_token.reg = prev_token.reg = NULL;
6638 cur_token.str = prev_token.str = NULL;
6639
6640 /* Initialize parser structure. */
6641 intel_parser.got_a_float = got_a_float;
6642 intel_parser.op_modifier = 0;
6643 intel_parser.is_mem = 0;
6644 intel_parser.in_offset = 0;
6645 intel_parser.in_bracket = 0;
6646 intel_parser.reg = NULL;
6647 intel_parser.disp[0] = '\0';
6648 intel_parser.next_operand = NULL;
6649
6650 /* Read the first token and start the parser. */
6651 intel_get_token ();
6652 ret = intel_expr ();
6653
6654 if (!ret)
6655 break;
6656
6657 if (cur_token.code != T_NIL)
6658 {
6659 as_bad (_("invalid operand for '%s' ('%s' unexpected)"),
6660 current_templates->start->name, cur_token.str);
6661 ret = 0;
6662 }
6663 /* If we found a memory reference, hand it over to i386_displacement
6664 to fill in the rest of the operand fields. */
6665 else if (intel_parser.is_mem)
6666 {
6667 if ((i.mem_operands == 1
6668 && (current_templates->start->opcode_modifier & IsString) == 0)
6669 || i.mem_operands == 2)
6670 {
6671 as_bad (_("too many memory references for '%s'"),
6672 current_templates->start->name);
6673 ret = 0;
6674 }
6675 else
6676 {
6677 char *s = intel_parser.disp;
6678 i.mem_operands++;
6679
6680 if (!quiet_warnings && intel_parser.is_mem < 0)
6681 /* See the comments in intel_bracket_expr. */
6682 as_warn (_("Treating `%s' as memory reference"), operand_string);
6683
6684 /* Add the displacement expression. */
6685 if (*s != '\0')
6686 ret = i386_displacement (s, s + strlen (s));
6687 if (ret)
6688 {
6689 /* Swap base and index in 16-bit memory operands like
6690 [si+bx]. Since i386_index_check is also used in AT&T
6691 mode we have to do that here. */
6692 if (i.base_reg
6693 && i.index_reg
6694 && (i.base_reg->reg_type & Reg16)
6695 && (i.index_reg->reg_type & Reg16)
6696 && i.base_reg->reg_num >= 6
6697 && i.index_reg->reg_num < 6)
6698 {
6699 const reg_entry *base = i.index_reg;
6700
6701 i.index_reg = i.base_reg;
6702 i.base_reg = base;
6703 }
6704 ret = i386_index_check (operand_string);
6705 }
6706 }
6707 }
6708
6709 /* Constant and OFFSET expressions are handled by i386_immediate. */
6710 else if ((intel_parser.op_modifier & (1 << T_OFFSET))
6711 || intel_parser.reg == NULL)
6712 ret = i386_immediate (intel_parser.disp);
6713
6714 if (intel_parser.next_operand && this_operand >= MAX_OPERANDS - 1)
6715 ret = 0;
6716 if (!ret || !intel_parser.next_operand)
6717 break;
6718 intel_parser.op_string = intel_parser.next_operand;
6719 this_operand = i.operands++;
6720 }
6721
6722 free (p);
6723 free (intel_parser.disp);
6724
6725 return ret;
6726 }
6727
6728 #define NUM_ADDRESS_REGS (!!i.base_reg + !!i.index_reg)
6729
6730 /* expr e04 expr'
6731
6732 expr' cmpOp e04 expr'
6733 | Empty */
6734 static int
6735 intel_expr ()
6736 {
6737 /* XXX Implement the comparison operators. */
6738 return intel_e04 ();
6739 }
6740
6741 /* e04 e05 e04'
6742
6743 e04' addOp e05 e04'
6744 | Empty */
6745 static int
6746 intel_e04 ()
6747 {
6748 int nregs = -1;
6749
6750 for (;;)
6751 {
6752 if (!intel_e05())
6753 return 0;
6754
6755 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
6756 i.base_reg = i386_regtab + REGNAM_AL; /* al is invalid as base */
6757
6758 if (cur_token.code == '+')
6759 nregs = -1;
6760 else if (cur_token.code == '-')
6761 nregs = NUM_ADDRESS_REGS;
6762 else
6763 return 1;
6764
6765 strcat (intel_parser.disp, cur_token.str);
6766 intel_match_token (cur_token.code);
6767 }
6768 }
6769
6770 /* e05 e06 e05'
6771
6772 e05' binOp e06 e05'
6773 | Empty */
6774 static int
6775 intel_e05 ()
6776 {
6777 int nregs = ~NUM_ADDRESS_REGS;
6778
6779 for (;;)
6780 {
6781 if (!intel_e06())
6782 return 0;
6783
6784 if (cur_token.code == '&' || cur_token.code == '|' || cur_token.code == '^')
6785 {
6786 char str[2];
6787
6788 str[0] = cur_token.code;
6789 str[1] = 0;
6790 strcat (intel_parser.disp, str);
6791 }
6792 else
6793 break;
6794
6795 intel_match_token (cur_token.code);
6796
6797 if (nregs < 0)
6798 nregs = ~nregs;
6799 }
6800 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
6801 i.base_reg = i386_regtab + REGNAM_AL + 1; /* cl is invalid as base */
6802 return 1;
6803 }
6804
6805 /* e06 e09 e06'
6806
6807 e06' mulOp e09 e06'
6808 | Empty */
6809 static int
6810 intel_e06 ()
6811 {
6812 int nregs = ~NUM_ADDRESS_REGS;
6813
6814 for (;;)
6815 {
6816 if (!intel_e09())
6817 return 0;
6818
6819 if (cur_token.code == '*' || cur_token.code == '/' || cur_token.code == '%')
6820 {
6821 char str[2];
6822
6823 str[0] = cur_token.code;
6824 str[1] = 0;
6825 strcat (intel_parser.disp, str);
6826 }
6827 else if (cur_token.code == T_SHL)
6828 strcat (intel_parser.disp, "<<");
6829 else if (cur_token.code == T_SHR)
6830 strcat (intel_parser.disp, ">>");
6831 else
6832 break;
6833
6834 intel_match_token (cur_token.code);
6835
6836 if (nregs < 0)
6837 nregs = ~nregs;
6838 }
6839 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
6840 i.base_reg = i386_regtab + REGNAM_AL + 2; /* dl is invalid as base */
6841 return 1;
6842 }
6843
6844 /* e09 OFFSET e09
6845 | SHORT e09
6846 | + e09
6847 | - e09
6848 | ~ e09
6849 | NOT e09
6850 | e10 e09'
6851
6852 e09' PTR e10 e09'
6853 | : e10 e09'
6854 | Empty */
6855 static int
6856 intel_e09 ()
6857 {
6858 int nregs = ~NUM_ADDRESS_REGS;
6859 int in_offset = 0;
6860
6861 for (;;)
6862 {
6863 /* Don't consume constants here. */
6864 if (cur_token.code == '+' || cur_token.code == '-')
6865 {
6866 /* Need to look one token ahead - if the next token
6867 is a constant, the current token is its sign. */
6868 int next_code;
6869
6870 intel_match_token (cur_token.code);
6871 next_code = cur_token.code;
6872 intel_putback_token ();
6873 if (next_code == T_CONST)
6874 break;
6875 }
6876
6877 /* e09 OFFSET e09 */
6878 if (cur_token.code == T_OFFSET)
6879 {
6880 if (!in_offset++)
6881 ++intel_parser.in_offset;
6882 }
6883
6884 /* e09 SHORT e09 */
6885 else if (cur_token.code == T_SHORT)
6886 intel_parser.op_modifier |= 1 << T_SHORT;
6887
6888 /* e09 + e09 */
6889 else if (cur_token.code == '+')
6890 strcat (intel_parser.disp, "+");
6891
6892 /* e09 - e09
6893 | ~ e09
6894 | NOT e09 */
6895 else if (cur_token.code == '-' || cur_token.code == '~')
6896 {
6897 char str[2];
6898
6899 if (nregs < 0)
6900 nregs = ~nregs;
6901 str[0] = cur_token.code;
6902 str[1] = 0;
6903 strcat (intel_parser.disp, str);
6904 }
6905
6906 /* e09 e10 e09' */
6907 else
6908 break;
6909
6910 intel_match_token (cur_token.code);
6911 }
6912
6913 for (;;)
6914 {
6915 if (!intel_e10 ())
6916 return 0;
6917
6918 /* e09' PTR e10 e09' */
6919 if (cur_token.code == T_PTR)
6920 {
6921 char suffix;
6922
6923 if (prev_token.code == T_BYTE)
6924 suffix = BYTE_MNEM_SUFFIX;
6925
6926 else if (prev_token.code == T_WORD)
6927 {
6928 if (current_templates->start->name[0] == 'l'
6929 && current_templates->start->name[2] == 's'
6930 && current_templates->start->name[3] == 0)
6931 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
6932 else if (intel_parser.got_a_float == 2) /* "fi..." */
6933 suffix = SHORT_MNEM_SUFFIX;
6934 else
6935 suffix = WORD_MNEM_SUFFIX;
6936 }
6937
6938 else if (prev_token.code == T_DWORD)
6939 {
6940 if (current_templates->start->name[0] == 'l'
6941 && current_templates->start->name[2] == 's'
6942 && current_templates->start->name[3] == 0)
6943 suffix = WORD_MNEM_SUFFIX;
6944 else if (flag_code == CODE_16BIT
6945 && (current_templates->start->opcode_modifier
6946 & (Jump | JumpDword)))
6947 suffix = LONG_DOUBLE_MNEM_SUFFIX;
6948 else if (intel_parser.got_a_float == 1) /* "f..." */
6949 suffix = SHORT_MNEM_SUFFIX;
6950 else
6951 suffix = LONG_MNEM_SUFFIX;
6952 }
6953
6954 else if (prev_token.code == T_FWORD)
6955 {
6956 if (current_templates->start->name[0] == 'l'
6957 && current_templates->start->name[2] == 's'
6958 && current_templates->start->name[3] == 0)
6959 suffix = LONG_MNEM_SUFFIX;
6960 else if (!intel_parser.got_a_float)
6961 {
6962 if (flag_code == CODE_16BIT)
6963 add_prefix (DATA_PREFIX_OPCODE);
6964 suffix = LONG_DOUBLE_MNEM_SUFFIX;
6965 }
6966 else
6967 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
6968 }
6969
6970 else if (prev_token.code == T_QWORD)
6971 {
6972 if (intel_parser.got_a_float == 1) /* "f..." */
6973 suffix = LONG_MNEM_SUFFIX;
6974 else
6975 suffix = QWORD_MNEM_SUFFIX;
6976 }
6977
6978 else if (prev_token.code == T_TBYTE)
6979 {
6980 if (intel_parser.got_a_float == 1)
6981 suffix = LONG_DOUBLE_MNEM_SUFFIX;
6982 else
6983 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
6984 }
6985
6986 else if (prev_token.code == T_XMMWORD)
6987 {
6988 /* XXX ignored for now, but accepted since gcc uses it */
6989 suffix = 0;
6990 }
6991
6992 else
6993 {
6994 as_bad (_("Unknown operand modifier `%s'"), prev_token.str);
6995 return 0;
6996 }
6997
6998 /* Operands for jump/call using 'ptr' notation denote absolute
6999 addresses. */
7000 if (current_templates->start->opcode_modifier & (Jump | JumpDword))
7001 i.types[this_operand] |= JumpAbsolute;
7002
7003 if (current_templates->start->base_opcode == 0x8d /* lea */)
7004 ;
7005 else if (!i.suffix)
7006 i.suffix = suffix;
7007 else if (i.suffix != suffix)
7008 {
7009 as_bad (_("Conflicting operand modifiers"));
7010 return 0;
7011 }
7012
7013 }
7014
7015 /* e09' : e10 e09' */
7016 else if (cur_token.code == ':')
7017 {
7018 if (prev_token.code != T_REG)
7019 {
7020 /* While {call,jmp} SSSS:OOOO is MASM syntax only when SSSS is a
7021 segment/group identifier (which we don't have), using comma
7022 as the operand separator there is even less consistent, since
7023 there all branches only have a single operand. */
7024 if (this_operand != 0
7025 || intel_parser.in_offset
7026 || intel_parser.in_bracket
7027 || (!(current_templates->start->opcode_modifier
7028 & (Jump|JumpDword|JumpInterSegment))
7029 && !(current_templates->start->operand_types[0]
7030 & JumpAbsolute)))
7031 return intel_match_token (T_NIL);
7032 /* Remember the start of the 2nd operand and terminate 1st
7033 operand here.
7034 XXX This isn't right, yet (when SSSS:OOOO is right operand of
7035 another expression), but it gets at least the simplest case
7036 (a plain number or symbol on the left side) right. */
7037 intel_parser.next_operand = intel_parser.op_string;
7038 *--intel_parser.op_string = '\0';
7039 return intel_match_token (':');
7040 }
7041 }
7042
7043 /* e09' Empty */
7044 else
7045 break;
7046
7047 intel_match_token (cur_token.code);
7048
7049 }
7050
7051 if (in_offset)
7052 {
7053 --intel_parser.in_offset;
7054 if (nregs < 0)
7055 nregs = ~nregs;
7056 if (NUM_ADDRESS_REGS > nregs)
7057 {
7058 as_bad (_("Invalid operand to `OFFSET'"));
7059 return 0;
7060 }
7061 intel_parser.op_modifier |= 1 << T_OFFSET;
7062 }
7063
7064 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
7065 i.base_reg = i386_regtab + REGNAM_AL + 3; /* bl is invalid as base */
7066 return 1;
7067 }
7068
7069 static int
7070 intel_bracket_expr ()
7071 {
7072 int was_offset = intel_parser.op_modifier & (1 << T_OFFSET);
7073 const char *start = intel_parser.op_string;
7074 int len;
7075
7076 if (i.op[this_operand].regs)
7077 return intel_match_token (T_NIL);
7078
7079 intel_match_token ('[');
7080
7081 /* Mark as a memory operand only if it's not already known to be an
7082 offset expression. If it's an offset expression, we need to keep
7083 the brace in. */
7084 if (!intel_parser.in_offset)
7085 {
7086 ++intel_parser.in_bracket;
7087
7088 /* Operands for jump/call inside brackets denote absolute addresses. */
7089 if (current_templates->start->opcode_modifier & (Jump | JumpDword))
7090 i.types[this_operand] |= JumpAbsolute;
7091
7092 /* Unfortunately gas always diverged from MASM in a respect that can't
7093 be easily fixed without risking to break code sequences likely to be
7094 encountered (the testsuite even check for this): MASM doesn't consider
7095 an expression inside brackets unconditionally as a memory reference.
7096 When that is e.g. a constant, an offset expression, or the sum of the
7097 two, this is still taken as a constant load. gas, however, always
7098 treated these as memory references. As a compromise, we'll try to make
7099 offset expressions inside brackets work the MASM way (since that's
7100 less likely to be found in real world code), but make constants alone
7101 continue to work the traditional gas way. In either case, issue a
7102 warning. */
7103 intel_parser.op_modifier &= ~was_offset;
7104 }
7105 else
7106 strcat (intel_parser.disp, "[");
7107
7108 /* Add a '+' to the displacement string if necessary. */
7109 if (*intel_parser.disp != '\0'
7110 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
7111 strcat (intel_parser.disp, "+");
7112
7113 if (intel_expr ()
7114 && (len = intel_parser.op_string - start - 1,
7115 intel_match_token (']')))
7116 {
7117 /* Preserve brackets when the operand is an offset expression. */
7118 if (intel_parser.in_offset)
7119 strcat (intel_parser.disp, "]");
7120 else
7121 {
7122 --intel_parser.in_bracket;
7123 if (i.base_reg || i.index_reg)
7124 intel_parser.is_mem = 1;
7125 if (!intel_parser.is_mem)
7126 {
7127 if (!(intel_parser.op_modifier & (1 << T_OFFSET)))
7128 /* Defer the warning until all of the operand was parsed. */
7129 intel_parser.is_mem = -1;
7130 else if (!quiet_warnings)
7131 as_warn (_("`[%.*s]' taken to mean just `%.*s'"), len, start, len, start);
7132 }
7133 }
7134 intel_parser.op_modifier |= was_offset;
7135
7136 return 1;
7137 }
7138 return 0;
7139 }
7140
7141 /* e10 e11 e10'
7142
7143 e10' [ expr ] e10'
7144 | Empty */
7145 static int
7146 intel_e10 ()
7147 {
7148 if (!intel_e11 ())
7149 return 0;
7150
7151 while (cur_token.code == '[')
7152 {
7153 if (!intel_bracket_expr ())
7154 return 0;
7155 }
7156
7157 return 1;
7158 }
7159
7160 /* e11 ( expr )
7161 | [ expr ]
7162 | BYTE
7163 | WORD
7164 | DWORD
7165 | FWORD
7166 | QWORD
7167 | TBYTE
7168 | OWORD
7169 | XMMWORD
7170 | $
7171 | .
7172 | register
7173 | id
7174 | constant */
7175 static int
7176 intel_e11 ()
7177 {
7178 switch (cur_token.code)
7179 {
7180 /* e11 ( expr ) */
7181 case '(':
7182 intel_match_token ('(');
7183 strcat (intel_parser.disp, "(");
7184
7185 if (intel_expr () && intel_match_token (')'))
7186 {
7187 strcat (intel_parser.disp, ")");
7188 return 1;
7189 }
7190 return 0;
7191
7192 /* e11 [ expr ] */
7193 case '[':
7194 return intel_bracket_expr ();
7195
7196 /* e11 $
7197 | . */
7198 case '.':
7199 strcat (intel_parser.disp, cur_token.str);
7200 intel_match_token (cur_token.code);
7201
7202 /* Mark as a memory operand only if it's not already known to be an
7203 offset expression. */
7204 if (!intel_parser.in_offset)
7205 intel_parser.is_mem = 1;
7206
7207 return 1;
7208
7209 /* e11 register */
7210 case T_REG:
7211 {
7212 const reg_entry *reg = intel_parser.reg = cur_token.reg;
7213
7214 intel_match_token (T_REG);
7215
7216 /* Check for segment change. */
7217 if (cur_token.code == ':')
7218 {
7219 if (!(reg->reg_type & (SReg2 | SReg3)))
7220 {
7221 as_bad (_("`%s' is not a valid segment register"), reg->reg_name);
7222 return 0;
7223 }
7224 else if (i.seg[i.mem_operands])
7225 as_warn (_("Extra segment override ignored"));
7226 else
7227 {
7228 if (!intel_parser.in_offset)
7229 intel_parser.is_mem = 1;
7230 switch (reg->reg_num)
7231 {
7232 case 0:
7233 i.seg[i.mem_operands] = &es;
7234 break;
7235 case 1:
7236 i.seg[i.mem_operands] = &cs;
7237 break;
7238 case 2:
7239 i.seg[i.mem_operands] = &ss;
7240 break;
7241 case 3:
7242 i.seg[i.mem_operands] = &ds;
7243 break;
7244 case 4:
7245 i.seg[i.mem_operands] = &fs;
7246 break;
7247 case 5:
7248 i.seg[i.mem_operands] = &gs;
7249 break;
7250 }
7251 }
7252 }
7253
7254 /* Not a segment register. Check for register scaling. */
7255 else if (cur_token.code == '*')
7256 {
7257 if (!intel_parser.in_bracket)
7258 {
7259 as_bad (_("Register scaling only allowed in memory operands"));
7260 return 0;
7261 }
7262
7263 if (reg->reg_type & Reg16) /* Disallow things like [si*1]. */
7264 reg = i386_regtab + REGNAM_AX + 4; /* sp is invalid as index */
7265 else if (i.index_reg)
7266 reg = i386_regtab + REGNAM_EAX + 4; /* esp is invalid as index */
7267
7268 /* What follows must be a valid scale. */
7269 intel_match_token ('*');
7270 i.index_reg = reg;
7271 i.types[this_operand] |= BaseIndex;
7272
7273 /* Set the scale after setting the register (otherwise,
7274 i386_scale will complain) */
7275 if (cur_token.code == '+' || cur_token.code == '-')
7276 {
7277 char *str, sign = cur_token.code;
7278 intel_match_token (cur_token.code);
7279 if (cur_token.code != T_CONST)
7280 {
7281 as_bad (_("Syntax error: Expecting a constant, got `%s'"),
7282 cur_token.str);
7283 return 0;
7284 }
7285 str = (char *) xmalloc (strlen (cur_token.str) + 2);
7286 strcpy (str + 1, cur_token.str);
7287 *str = sign;
7288 if (!i386_scale (str))
7289 return 0;
7290 free (str);
7291 }
7292 else if (!i386_scale (cur_token.str))
7293 return 0;
7294 intel_match_token (cur_token.code);
7295 }
7296
7297 /* No scaling. If this is a memory operand, the register is either a
7298 base register (first occurrence) or an index register (second
7299 occurrence). */
7300 else if (intel_parser.in_bracket)
7301 {
7302
7303 if (!i.base_reg)
7304 i.base_reg = reg;
7305 else if (!i.index_reg)
7306 i.index_reg = reg;
7307 else
7308 {
7309 as_bad (_("Too many register references in memory operand"));
7310 return 0;
7311 }
7312
7313 i.types[this_operand] |= BaseIndex;
7314 }
7315
7316 /* It's neither base nor index. */
7317 else if (!intel_parser.in_offset && !intel_parser.is_mem)
7318 {
7319 i.types[this_operand] |= reg->reg_type & ~BaseIndex;
7320 i.op[this_operand].regs = reg;
7321 i.reg_operands++;
7322 }
7323 else
7324 {
7325 as_bad (_("Invalid use of register"));
7326 return 0;
7327 }
7328
7329 /* Since registers are not part of the displacement string (except
7330 when we're parsing offset operands), we may need to remove any
7331 preceding '+' from the displacement string. */
7332 if (*intel_parser.disp != '\0'
7333 && !intel_parser.in_offset)
7334 {
7335 char *s = intel_parser.disp;
7336 s += strlen (s) - 1;
7337 if (*s == '+')
7338 *s = '\0';
7339 }
7340
7341 return 1;
7342 }
7343
7344 /* e11 BYTE
7345 | WORD
7346 | DWORD
7347 | FWORD
7348 | QWORD
7349 | TBYTE
7350 | OWORD
7351 | XMMWORD */
7352 case T_BYTE:
7353 case T_WORD:
7354 case T_DWORD:
7355 case T_FWORD:
7356 case T_QWORD:
7357 case T_TBYTE:
7358 case T_XMMWORD:
7359 intel_match_token (cur_token.code);
7360
7361 if (cur_token.code == T_PTR)
7362 return 1;
7363
7364 /* It must have been an identifier. */
7365 intel_putback_token ();
7366 cur_token.code = T_ID;
7367 /* FALLTHRU */
7368
7369 /* e11 id
7370 | constant */
7371 case T_ID:
7372 if (!intel_parser.in_offset && intel_parser.is_mem <= 0)
7373 {
7374 symbolS *symbolP;
7375
7376 /* The identifier represents a memory reference only if it's not
7377 preceded by an offset modifier and if it's not an equate. */
7378 symbolP = symbol_find(cur_token.str);
7379 if (!symbolP || S_GET_SEGMENT(symbolP) != absolute_section)
7380 intel_parser.is_mem = 1;
7381 }
7382 /* FALLTHRU */
7383
7384 case T_CONST:
7385 case '-':
7386 case '+':
7387 {
7388 char *save_str, sign = 0;
7389
7390 /* Allow constants that start with `+' or `-'. */
7391 if (cur_token.code == '-' || cur_token.code == '+')
7392 {
7393 sign = cur_token.code;
7394 intel_match_token (cur_token.code);
7395 if (cur_token.code != T_CONST)
7396 {
7397 as_bad (_("Syntax error: Expecting a constant, got `%s'"),
7398 cur_token.str);
7399 return 0;
7400 }
7401 }
7402
7403 save_str = (char *) xmalloc (strlen (cur_token.str) + 2);
7404 strcpy (save_str + !!sign, cur_token.str);
7405 if (sign)
7406 *save_str = sign;
7407
7408 /* Get the next token to check for register scaling. */
7409 intel_match_token (cur_token.code);
7410
7411 /* Check if this constant is a scaling factor for an index register. */
7412 if (cur_token.code == '*')
7413 {
7414 if (intel_match_token ('*') && cur_token.code == T_REG)
7415 {
7416 const reg_entry *reg = cur_token.reg;
7417
7418 if (!intel_parser.in_bracket)
7419 {
7420 as_bad (_("Register scaling only allowed in memory operands"));
7421 return 0;
7422 }
7423
7424 if (reg->reg_type & Reg16) /* Disallow things like [1*si]. */
7425 reg = i386_regtab + REGNAM_AX + 4; /* sp is invalid as index */
7426 else if (i.index_reg)
7427 reg = i386_regtab + REGNAM_EAX + 4; /* esp is invalid as index */
7428
7429 /* The constant is followed by `* reg', so it must be
7430 a valid scale. */
7431 i.index_reg = reg;
7432 i.types[this_operand] |= BaseIndex;
7433
7434 /* Set the scale after setting the register (otherwise,
7435 i386_scale will complain) */
7436 if (!i386_scale (save_str))
7437 return 0;
7438 intel_match_token (T_REG);
7439
7440 /* Since registers are not part of the displacement
7441 string, we may need to remove any preceding '+' from
7442 the displacement string. */
7443 if (*intel_parser.disp != '\0')
7444 {
7445 char *s = intel_parser.disp;
7446 s += strlen (s) - 1;
7447 if (*s == '+')
7448 *s = '\0';
7449 }
7450
7451 free (save_str);
7452
7453 return 1;
7454 }
7455
7456 /* The constant was not used for register scaling. Since we have
7457 already consumed the token following `*' we now need to put it
7458 back in the stream. */
7459 intel_putback_token ();
7460 }
7461
7462 /* Add the constant to the displacement string. */
7463 strcat (intel_parser.disp, save_str);
7464 free (save_str);
7465
7466 return 1;
7467 }
7468 }
7469
7470 as_bad (_("Unrecognized token '%s'"), cur_token.str);
7471 return 0;
7472 }
7473
7474 /* Match the given token against cur_token. If they match, read the next
7475 token from the operand string. */
7476 static int
7477 intel_match_token (code)
7478 int code;
7479 {
7480 if (cur_token.code == code)
7481 {
7482 intel_get_token ();
7483 return 1;
7484 }
7485 else
7486 {
7487 as_bad (_("Unexpected token `%s'"), cur_token.str);
7488 return 0;
7489 }
7490 }
7491
7492 /* Read a new token from intel_parser.op_string and store it in cur_token. */
7493 static void
7494 intel_get_token ()
7495 {
7496 char *end_op;
7497 const reg_entry *reg;
7498 struct intel_token new_token;
7499
7500 new_token.code = T_NIL;
7501 new_token.reg = NULL;
7502 new_token.str = NULL;
7503
7504 /* Free the memory allocated to the previous token and move
7505 cur_token to prev_token. */
7506 if (prev_token.str)
7507 free (prev_token.str);
7508
7509 prev_token = cur_token;
7510
7511 /* Skip whitespace. */
7512 while (is_space_char (*intel_parser.op_string))
7513 intel_parser.op_string++;
7514
7515 /* Return an empty token if we find nothing else on the line. */
7516 if (*intel_parser.op_string == '\0')
7517 {
7518 cur_token = new_token;
7519 return;
7520 }
7521
7522 /* The new token cannot be larger than the remainder of the operand
7523 string. */
7524 new_token.str = (char *) xmalloc (strlen (intel_parser.op_string) + 1);
7525 new_token.str[0] = '\0';
7526
7527 if (strchr ("0123456789", *intel_parser.op_string))
7528 {
7529 char *p = new_token.str;
7530 char *q = intel_parser.op_string;
7531 new_token.code = T_CONST;
7532
7533 /* Allow any kind of identifier char to encompass floating point and
7534 hexadecimal numbers. */
7535 while (is_identifier_char (*q))
7536 *p++ = *q++;
7537 *p = '\0';
7538
7539 /* Recognize special symbol names [0-9][bf]. */
7540 if (strlen (intel_parser.op_string) == 2
7541 && (intel_parser.op_string[1] == 'b'
7542 || intel_parser.op_string[1] == 'f'))
7543 new_token.code = T_ID;
7544 }
7545
7546 else if ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL)
7547 {
7548 size_t len = end_op - intel_parser.op_string;
7549
7550 new_token.code = T_REG;
7551 new_token.reg = reg;
7552
7553 memcpy (new_token.str, intel_parser.op_string, len);
7554 new_token.str[len] = '\0';
7555 }
7556
7557 else if (is_identifier_char (*intel_parser.op_string))
7558 {
7559 char *p = new_token.str;
7560 char *q = intel_parser.op_string;
7561
7562 /* A '.' or '$' followed by an identifier char is an identifier.
7563 Otherwise, it's operator '.' followed by an expression. */
7564 if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
7565 {
7566 new_token.code = '.';
7567 new_token.str[0] = '.';
7568 new_token.str[1] = '\0';
7569 }
7570 else
7571 {
7572 while (is_identifier_char (*q) || *q == '@')
7573 *p++ = *q++;
7574 *p = '\0';
7575
7576 if (strcasecmp (new_token.str, "NOT") == 0)
7577 new_token.code = '~';
7578
7579 else if (strcasecmp (new_token.str, "MOD") == 0)
7580 new_token.code = '%';
7581
7582 else if (strcasecmp (new_token.str, "AND") == 0)
7583 new_token.code = '&';
7584
7585 else if (strcasecmp (new_token.str, "OR") == 0)
7586 new_token.code = '|';
7587
7588 else if (strcasecmp (new_token.str, "XOR") == 0)
7589 new_token.code = '^';
7590
7591 else if (strcasecmp (new_token.str, "SHL") == 0)
7592 new_token.code = T_SHL;
7593
7594 else if (strcasecmp (new_token.str, "SHR") == 0)
7595 new_token.code = T_SHR;
7596
7597 else if (strcasecmp (new_token.str, "BYTE") == 0)
7598 new_token.code = T_BYTE;
7599
7600 else if (strcasecmp (new_token.str, "WORD") == 0)
7601 new_token.code = T_WORD;
7602
7603 else if (strcasecmp (new_token.str, "DWORD") == 0)
7604 new_token.code = T_DWORD;
7605
7606 else if (strcasecmp (new_token.str, "FWORD") == 0)
7607 new_token.code = T_FWORD;
7608
7609 else if (strcasecmp (new_token.str, "QWORD") == 0)
7610 new_token.code = T_QWORD;
7611
7612 else if (strcasecmp (new_token.str, "TBYTE") == 0
7613 /* XXX remove (gcc still uses it) */
7614 || strcasecmp (new_token.str, "XWORD") == 0)
7615 new_token.code = T_TBYTE;
7616
7617 else if (strcasecmp (new_token.str, "XMMWORD") == 0
7618 || strcasecmp (new_token.str, "OWORD") == 0)
7619 new_token.code = T_XMMWORD;
7620
7621 else if (strcasecmp (new_token.str, "PTR") == 0)
7622 new_token.code = T_PTR;
7623
7624 else if (strcasecmp (new_token.str, "SHORT") == 0)
7625 new_token.code = T_SHORT;
7626
7627 else if (strcasecmp (new_token.str, "OFFSET") == 0)
7628 {
7629 new_token.code = T_OFFSET;
7630
7631 /* ??? This is not mentioned in the MASM grammar but gcc
7632 makes use of it with -mintel-syntax. OFFSET may be
7633 followed by FLAT: */
7634 if (strncasecmp (q, " FLAT:", 6) == 0)
7635 strcat (new_token.str, " FLAT:");
7636 }
7637
7638 /* ??? This is not mentioned in the MASM grammar. */
7639 else if (strcasecmp (new_token.str, "FLAT") == 0)
7640 {
7641 new_token.code = T_OFFSET;
7642 if (*q == ':')
7643 strcat (new_token.str, ":");
7644 else
7645 as_bad (_("`:' expected"));
7646 }
7647
7648 else
7649 new_token.code = T_ID;
7650 }
7651 }
7652
7653 else if (strchr ("+-/*%|&^:[]()~", *intel_parser.op_string))
7654 {
7655 new_token.code = *intel_parser.op_string;
7656 new_token.str[0] = *intel_parser.op_string;
7657 new_token.str[1] = '\0';
7658 }
7659
7660 else if (strchr ("<>", *intel_parser.op_string)
7661 && *intel_parser.op_string == *(intel_parser.op_string + 1))
7662 {
7663 new_token.code = *intel_parser.op_string == '<' ? T_SHL : T_SHR;
7664 new_token.str[0] = *intel_parser.op_string;
7665 new_token.str[1] = *intel_parser.op_string;
7666 new_token.str[2] = '\0';
7667 }
7668
7669 else
7670 as_bad (_("Unrecognized token `%s'"), intel_parser.op_string);
7671
7672 intel_parser.op_string += strlen (new_token.str);
7673 cur_token = new_token;
7674 }
7675
7676 /* Put cur_token back into the token stream and make cur_token point to
7677 prev_token. */
7678 static void
7679 intel_putback_token ()
7680 {
7681 if (cur_token.code != T_NIL)
7682 {
7683 intel_parser.op_string -= strlen (cur_token.str);
7684 free (cur_token.str);
7685 }
7686 cur_token = prev_token;
7687
7688 /* Forget prev_token. */
7689 prev_token.code = T_NIL;
7690 prev_token.reg = NULL;
7691 prev_token.str = NULL;
7692 }
7693
7694 int
7695 tc_x86_regname_to_dw2regnum (char *regname)
7696 {
7697 unsigned int regnum;
7698 unsigned int regnames_count;
7699 static const char *const regnames_32[] =
7700 {
7701 "eax", "ecx", "edx", "ebx",
7702 "esp", "ebp", "esi", "edi",
7703 "eip", "eflags", NULL,
7704 "st0", "st1", "st2", "st3",
7705 "st4", "st5", "st6", "st7",
7706 NULL, NULL,
7707 "xmm0", "xmm1", "xmm2", "xmm3",
7708 "xmm4", "xmm5", "xmm6", "xmm7",
7709 "mm0", "mm1", "mm2", "mm3",
7710 "mm4", "mm5", "mm6", "mm7",
7711 "fcw", "fsw", "mxcsr",
7712 "es", "cs", "ss", "ds", "fs", "gs", NULL, NULL,
7713 "tr", "ldtr"
7714 };
7715 static const char *const regnames_64[] =
7716 {
7717 "rax", "rdx", "rcx", "rbx",
7718 "rsi", "rdi", "rbp", "rsp",
7719 "r8", "r9", "r10", "r11",
7720 "r12", "r13", "r14", "r15",
7721 "rip",
7722 "xmm0", "xmm1", "xmm2", "xmm3",
7723 "xmm4", "xmm5", "xmm6", "xmm7",
7724 "xmm8", "xmm9", "xmm10", "xmm11",
7725 "xmm12", "xmm13", "xmm14", "xmm15",
7726 "st0", "st1", "st2", "st3",
7727 "st4", "st5", "st6", "st7",
7728 "mm0", "mm1", "mm2", "mm3",
7729 "mm4", "mm5", "mm6", "mm7",
7730 "rflags",
7731 "es", "cs", "ss", "ds", "fs", "gs", NULL, NULL,
7732 "fs.base", "gs.base", NULL, NULL,
7733 "tr", "ldtr",
7734 "mxcsr", "fcw", "fsw"
7735 };
7736 const char *const *regnames;
7737
7738 if (flag_code == CODE_64BIT)
7739 {
7740 regnames = regnames_64;
7741 regnames_count = ARRAY_SIZE (regnames_64);
7742 }
7743 else
7744 {
7745 regnames = regnames_32;
7746 regnames_count = ARRAY_SIZE (regnames_32);
7747 }
7748
7749 for (regnum = 0; regnum < regnames_count; regnum++)
7750 if (regnames[regnum] != NULL
7751 && strcmp (regname, regnames[regnum]) == 0)
7752 return regnum;
7753
7754 return -1;
7755 }
7756
7757 void
7758 tc_x86_frame_initial_instructions (void)
7759 {
7760 static unsigned int sp_regno;
7761
7762 if (!sp_regno)
7763 sp_regno = tc_x86_regname_to_dw2regnum (flag_code == CODE_64BIT
7764 ? "rsp" : "esp");
7765
7766 cfi_add_CFA_def_cfa (sp_regno, -x86_cie_data_alignment);
7767 cfi_add_CFA_offset (x86_dwarf2_return_column, x86_cie_data_alignment);
7768 }
7769
7770 int
7771 i386_elf_section_type (const char *str, size_t len)
7772 {
7773 if (flag_code == CODE_64BIT
7774 && len == sizeof ("unwind") - 1
7775 && strncmp (str, "unwind", 6) == 0)
7776 return SHT_X86_64_UNWIND;
7777
7778 return -1;
7779 }
7780
7781 #ifdef TE_PE
7782 void
7783 tc_pe_dwarf2_emit_offset (symbolS *symbol, unsigned int size)
7784 {
7785 expressionS expr;
7786
7787 expr.X_op = O_secrel;
7788 expr.X_add_symbol = symbol;
7789 expr.X_add_number = 0;
7790 emit_expr (&expr, size);
7791 }
7792 #endif
7793
7794 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
7795 /* For ELF on x86-64, add support for SHF_X86_64_LARGE. */
7796
7797 int
7798 x86_64_section_letter (int letter, char **ptr_msg)
7799 {
7800 if (flag_code == CODE_64BIT)
7801 {
7802 if (letter == 'l')
7803 return SHF_X86_64_LARGE;
7804
7805 *ptr_msg = _("Bad .section directive: want a,l,w,x,M,S,G,T in string");
7806 }
7807 else
7808 *ptr_msg = _("Bad .section directive: want a,w,x,M,S,G,T in string");
7809 return -1;
7810 }
7811
7812 int
7813 x86_64_section_word (char *str, size_t len)
7814 {
7815 if (len == 5 && flag_code == CODE_64BIT && CONST_STRNEQ (str, "large"))
7816 return SHF_X86_64_LARGE;
7817
7818 return -1;
7819 }
7820
7821 static void
7822 handle_large_common (int small ATTRIBUTE_UNUSED)
7823 {
7824 if (flag_code != CODE_64BIT)
7825 {
7826 s_comm_internal (0, elf_common_parse);
7827 as_warn (_(".largecomm supported only in 64bit mode, producing .comm"));
7828 }
7829 else
7830 {
7831 static segT lbss_section;
7832 asection *saved_com_section_ptr = elf_com_section_ptr;
7833 asection *saved_bss_section = bss_section;
7834
7835 if (lbss_section == NULL)
7836 {
7837 flagword applicable;
7838 segT seg = now_seg;
7839 subsegT subseg = now_subseg;
7840
7841 /* The .lbss section is for local .largecomm symbols. */
7842 lbss_section = subseg_new (".lbss", 0);
7843 applicable = bfd_applicable_section_flags (stdoutput);
7844 bfd_set_section_flags (stdoutput, lbss_section,
7845 applicable & SEC_ALLOC);
7846 seg_info (lbss_section)->bss = 1;
7847
7848 subseg_set (seg, subseg);
7849 }
7850
7851 elf_com_section_ptr = &_bfd_elf_large_com_section;
7852 bss_section = lbss_section;
7853
7854 s_comm_internal (0, elf_common_parse);
7855
7856 elf_com_section_ptr = saved_com_section_ptr;
7857 bss_section = saved_bss_section;
7858 }
7859 }
7860 #endif /* OBJ_ELF || OBJ_MAYBE_ELF */