079e9d88660c6ae2f45bcd7f3aa22b0c6350d6ac
[deliverable/binutils-gdb.git] / gas / config / tc-i386.c
1 /* tc-i386.c -- Assemble code for the Intel 80386
2 Copyright 1989, 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
3 2000, 2001, 2002, 2003, 2004, 2005, 2006
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: max displacements & immediates. */
276 static expressionS disp_expressions[MAX_MEMORY_OPERANDS];
277 static expressionS im_expressions[MAX_IMMEDIATE_OPERANDS];
278
279 /* Current operand we are working on. */
280 static int this_operand;
281
282 /* We support four different modes. FLAG_CODE variable is used to distinguish
283 these. */
284
285 enum flag_code {
286 CODE_32BIT,
287 CODE_16BIT,
288 CODE_64BIT };
289 #define NUM_FLAG_CODE ((int) CODE_64BIT + 1)
290
291 static enum flag_code flag_code;
292 static unsigned int object_64bit;
293 static int use_rela_relocations = 0;
294
295 /* The names used to print error messages. */
296 static const char *flag_code_names[] =
297 {
298 "32",
299 "16",
300 "64"
301 };
302
303 /* 1 for intel syntax,
304 0 if att syntax. */
305 static int intel_syntax = 0;
306
307 /* 1 if register prefix % not required. */
308 static int allow_naked_reg = 0;
309
310 /* Used in 16 bit gcc mode to add an l suffix to call, ret, enter,
311 leave, push, and pop instructions so that gcc has the same stack
312 frame as in 32 bit mode. */
313 static char stackop_size = '\0';
314
315 /* Non-zero to optimize code alignment. */
316 int optimize_align_code = 1;
317
318 /* Non-zero to quieten some warnings. */
319 static int quiet_warnings = 0;
320
321 /* CPU name. */
322 static const char *cpu_arch_name = NULL;
323 static const char *cpu_sub_arch_name = NULL;
324
325 /* CPU feature flags. */
326 static unsigned int cpu_arch_flags = CpuUnknownFlags | CpuNo64;
327
328 /* If we have selected a cpu we are generating instructions for. */
329 static int cpu_arch_tune_set = 0;
330
331 /* Cpu we are generating instructions for. */
332 static enum processor_type cpu_arch_tune = PROCESSOR_UNKNOWN;
333
334 /* CPU feature flags of cpu we are generating instructions for. */
335 static unsigned int cpu_arch_tune_flags = 0;
336
337 /* CPU instruction set architecture used. */
338 static enum processor_type cpu_arch_isa = PROCESSOR_UNKNOWN;
339
340 /* CPU feature flags of instruction set architecture used. */
341 static unsigned int cpu_arch_isa_flags = 0;
342
343 /* If set, conditional jumps are not automatically promoted to handle
344 larger than a byte offset. */
345 static unsigned int no_cond_jump_promotion = 0;
346
347 /* Pre-defined "_GLOBAL_OFFSET_TABLE_". */
348 static symbolS *GOT_symbol;
349
350 /* The dwarf2 return column, adjusted for 32 or 64 bit. */
351 unsigned int x86_dwarf2_return_column;
352
353 /* The dwarf2 data alignment, adjusted for 32 or 64 bit. */
354 int x86_cie_data_alignment;
355
356 /* Interface to relax_segment.
357 There are 3 major relax states for 386 jump insns because the
358 different types of jumps add different sizes to frags when we're
359 figuring out what sort of jump to choose to reach a given label. */
360
361 /* Types. */
362 #define UNCOND_JUMP 0
363 #define COND_JUMP 1
364 #define COND_JUMP86 2
365
366 /* Sizes. */
367 #define CODE16 1
368 #define SMALL 0
369 #define SMALL16 (SMALL | CODE16)
370 #define BIG 2
371 #define BIG16 (BIG | CODE16)
372
373 #ifndef INLINE
374 #ifdef __GNUC__
375 #define INLINE __inline__
376 #else
377 #define INLINE
378 #endif
379 #endif
380
381 #define ENCODE_RELAX_STATE(type, size) \
382 ((relax_substateT) (((type) << 2) | (size)))
383 #define TYPE_FROM_RELAX_STATE(s) \
384 ((s) >> 2)
385 #define DISP_SIZE_FROM_RELAX_STATE(s) \
386 ((((s) & 3) == BIG ? 4 : (((s) & 3) == BIG16 ? 2 : 1)))
387
388 /* This table is used by relax_frag to promote short jumps to long
389 ones where necessary. SMALL (short) jumps may be promoted to BIG
390 (32 bit long) ones, and SMALL16 jumps to BIG16 (16 bit long). We
391 don't allow a short jump in a 32 bit code segment to be promoted to
392 a 16 bit offset jump because it's slower (requires data size
393 prefix), and doesn't work, unless the destination is in the bottom
394 64k of the code segment (The top 16 bits of eip are zeroed). */
395
396 const relax_typeS md_relax_table[] =
397 {
398 /* The fields are:
399 1) most positive reach of this state,
400 2) most negative reach of this state,
401 3) how many bytes this mode will have in the variable part of the frag
402 4) which index into the table to try if we can't fit into this one. */
403
404 /* UNCOND_JUMP states. */
405 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG)},
406 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16)},
407 /* dword jmp adds 4 bytes to frag:
408 0 extra opcode bytes, 4 displacement bytes. */
409 {0, 0, 4, 0},
410 /* word jmp adds 2 byte2 to frag:
411 0 extra opcode bytes, 2 displacement bytes. */
412 {0, 0, 2, 0},
413
414 /* COND_JUMP states. */
415 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG)},
416 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG16)},
417 /* dword conditionals adds 5 bytes to frag:
418 1 extra opcode byte, 4 displacement bytes. */
419 {0, 0, 5, 0},
420 /* word conditionals add 3 bytes to frag:
421 1 extra opcode byte, 2 displacement bytes. */
422 {0, 0, 3, 0},
423
424 /* COND_JUMP86 states. */
425 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG)},
426 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG16)},
427 /* dword conditionals adds 5 bytes to frag:
428 1 extra opcode byte, 4 displacement bytes. */
429 {0, 0, 5, 0},
430 /* word conditionals add 4 bytes to frag:
431 1 displacement byte and a 3 byte long branch insn. */
432 {0, 0, 4, 0}
433 };
434
435 static const arch_entry cpu_arch[] =
436 {
437 {"generic32", PROCESSOR_GENERIC32,
438 Cpu186|Cpu286|Cpu386},
439 {"generic64", PROCESSOR_GENERIC64,
440 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
441 |CpuMMX2|CpuSSE|CpuSSE2},
442 {"i8086", PROCESSOR_UNKNOWN,
443 0},
444 {"i186", PROCESSOR_UNKNOWN,
445 Cpu186},
446 {"i286", PROCESSOR_UNKNOWN,
447 Cpu186|Cpu286},
448 {"i386", PROCESSOR_GENERIC32,
449 Cpu186|Cpu286|Cpu386},
450 {"i486", PROCESSOR_I486,
451 Cpu186|Cpu286|Cpu386|Cpu486},
452 {"i586", PROCESSOR_PENTIUM,
453 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586},
454 {"i686", PROCESSOR_PENTIUMPRO,
455 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686},
456 {"pentium", PROCESSOR_PENTIUM,
457 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586},
458 {"pentiumpro",PROCESSOR_PENTIUMPRO,
459 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686},
460 {"pentiumii", PROCESSOR_PENTIUMPRO,
461 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX},
462 {"pentiumiii",PROCESSOR_PENTIUMPRO,
463 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuMMX2|CpuSSE},
464 {"pentium4", PROCESSOR_PENTIUM4,
465 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
466 |CpuMMX2|CpuSSE|CpuSSE2},
467 {"prescott", PROCESSOR_NOCONA,
468 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
469 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
470 {"nocona", PROCESSOR_NOCONA,
471 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
472 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
473 {"yonah", PROCESSOR_CORE,
474 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
475 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
476 {"core", PROCESSOR_CORE,
477 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
478 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
479 {"merom", PROCESSOR_CORE2,
480 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
481 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSSE3},
482 {"core2", PROCESSOR_CORE2,
483 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
484 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSSE3},
485 {"k6", PROCESSOR_K6,
486 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX},
487 {"k6_2", PROCESSOR_K6,
488 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX|Cpu3dnow},
489 {"athlon", PROCESSOR_ATHLON,
490 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6
491 |CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA},
492 {"sledgehammer", PROCESSOR_K8,
493 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6
494 |CpuSledgehammer|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2},
495 {"opteron", PROCESSOR_K8,
496 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6
497 |CpuSledgehammer|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2},
498 {"k8", PROCESSOR_K8,
499 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6
500 |CpuSledgehammer|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2},
501 {"amdfam10", PROCESSOR_AMDFAM10,
502 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuSledgehammer
503 |CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2|CpuSSE3|CpuSSE4a
504 |CpuABM},
505 {".mmx", PROCESSOR_UNKNOWN,
506 CpuMMX},
507 {".sse", PROCESSOR_UNKNOWN,
508 CpuMMX|CpuMMX2|CpuSSE},
509 {".sse2", PROCESSOR_UNKNOWN,
510 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2},
511 {".sse3", PROCESSOR_UNKNOWN,
512 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
513 {".ssse3", PROCESSOR_UNKNOWN,
514 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSSE3},
515 {".3dnow", PROCESSOR_UNKNOWN,
516 CpuMMX|Cpu3dnow},
517 {".3dnowa", PROCESSOR_UNKNOWN,
518 CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA},
519 {".padlock", PROCESSOR_UNKNOWN,
520 CpuPadLock},
521 {".pacifica", PROCESSOR_UNKNOWN,
522 CpuSVME},
523 {".svme", PROCESSOR_UNKNOWN,
524 CpuSVME},
525 {".sse4a", PROCESSOR_UNKNOWN,
526 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSE4a},
527 {".abm", PROCESSOR_UNKNOWN,
528 CpuABM}
529 };
530
531 const pseudo_typeS md_pseudo_table[] =
532 {
533 #if !defined(OBJ_AOUT) && !defined(USE_ALIGN_PTWO)
534 {"align", s_align_bytes, 0},
535 #else
536 {"align", s_align_ptwo, 0},
537 #endif
538 {"arch", set_cpu_arch, 0},
539 #ifndef I386COFF
540 {"bss", s_bss, 0},
541 #endif
542 {"ffloat", float_cons, 'f'},
543 {"dfloat", float_cons, 'd'},
544 {"tfloat", float_cons, 'x'},
545 {"value", cons, 2},
546 {"slong", signed_cons, 4},
547 {"noopt", s_ignore, 0},
548 {"optim", s_ignore, 0},
549 {"code16gcc", set_16bit_gcc_code_flag, CODE_16BIT},
550 {"code16", set_code_flag, CODE_16BIT},
551 {"code32", set_code_flag, CODE_32BIT},
552 {"code64", set_code_flag, CODE_64BIT},
553 {"intel_syntax", set_intel_syntax, 1},
554 {"att_syntax", set_intel_syntax, 0},
555 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
556 {"largecomm", handle_large_common, 0},
557 #else
558 {"file", (void (*) PARAMS ((int))) dwarf2_directive_file, 0},
559 {"loc", dwarf2_directive_loc, 0},
560 {"loc_mark_labels", dwarf2_directive_loc_mark_labels, 0},
561 #endif
562 #ifdef TE_PE
563 {"secrel32", pe_directive_secrel, 0},
564 #endif
565 {0, 0, 0}
566 };
567
568 /* For interface with expression (). */
569 extern char *input_line_pointer;
570
571 /* Hash table for instruction mnemonic lookup. */
572 static struct hash_control *op_hash;
573
574 /* Hash table for register lookup. */
575 static struct hash_control *reg_hash;
576 \f
577 void
578 i386_align_code (fragP, count)
579 fragS *fragP;
580 int count;
581 {
582 /* Various efficient no-op patterns for aligning code labels.
583 Note: Don't try to assemble the instructions in the comments.
584 0L and 0w are not legal. */
585 static const char f32_1[] =
586 {0x90}; /* nop */
587 static const char f32_2[] =
588 {0x66,0x90}; /* xchg %ax,%ax */
589 static const char f32_3[] =
590 {0x8d,0x76,0x00}; /* leal 0(%esi),%esi */
591 static const char f32_4[] =
592 {0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
593 static const char f32_5[] =
594 {0x90, /* nop */
595 0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
596 static const char f32_6[] =
597 {0x8d,0xb6,0x00,0x00,0x00,0x00}; /* leal 0L(%esi),%esi */
598 static const char f32_7[] =
599 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
600 static const char f32_8[] =
601 {0x90, /* nop */
602 0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
603 static const char f32_9[] =
604 {0x89,0xf6, /* movl %esi,%esi */
605 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
606 static const char f32_10[] =
607 {0x8d,0x76,0x00, /* leal 0(%esi),%esi */
608 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
609 static const char f32_11[] =
610 {0x8d,0x74,0x26,0x00, /* leal 0(%esi,1),%esi */
611 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
612 static const char f32_12[] =
613 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
614 0x8d,0xbf,0x00,0x00,0x00,0x00}; /* leal 0L(%edi),%edi */
615 static const char f32_13[] =
616 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
617 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
618 static const char f32_14[] =
619 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00, /* leal 0L(%esi,1),%esi */
620 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
621 static const char f32_15[] =
622 {0xeb,0x0d,0x90,0x90,0x90,0x90,0x90, /* jmp .+15; lotsa nops */
623 0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90};
624 static const char f16_3[] =
625 {0x8d,0x74,0x00}; /* lea 0(%esi),%esi */
626 static const char f16_4[] =
627 {0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
628 static const char f16_5[] =
629 {0x90, /* nop */
630 0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
631 static const char f16_6[] =
632 {0x89,0xf6, /* mov %si,%si */
633 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
634 static const char f16_7[] =
635 {0x8d,0x74,0x00, /* lea 0(%si),%si */
636 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
637 static const char f16_8[] =
638 {0x8d,0xb4,0x00,0x00, /* lea 0w(%si),%si */
639 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
640 static const char *const f32_patt[] = {
641 f32_1, f32_2, f32_3, f32_4, f32_5, f32_6, f32_7, f32_8,
642 f32_9, f32_10, f32_11, f32_12, f32_13, f32_14, f32_15
643 };
644 static const char *const f16_patt[] = {
645 f32_1, f32_2, f16_3, f16_4, f16_5, f16_6, f16_7, f16_8,
646 f32_15, f32_15, f32_15, f32_15, f32_15, f32_15, f32_15
647 };
648 /* nopl (%[re]ax) */
649 static const char alt_3[] =
650 {0x0f,0x1f,0x00};
651 /* nopl 0(%[re]ax) */
652 static const char alt_4[] =
653 {0x0f,0x1f,0x40,0x00};
654 /* nopl 0(%[re]ax,%[re]ax,1) */
655 static const char alt_5[] =
656 {0x0f,0x1f,0x44,0x00,0x00};
657 /* nopw 0(%[re]ax,%[re]ax,1) */
658 static const char alt_6[] =
659 {0x66,0x0f,0x1f,0x44,0x00,0x00};
660 /* nopl 0L(%[re]ax) */
661 static const char alt_7[] =
662 {0x0f,0x1f,0x80,0x00,0x00,0x00,0x00};
663 /* nopl 0L(%[re]ax,%[re]ax,1) */
664 static const char alt_8[] =
665 {0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
666 /* nopw 0L(%[re]ax,%[re]ax,1) */
667 static const char alt_9[] =
668 {0x66,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
669 /* nopw %cs:0L(%[re]ax,%[re]ax,1) */
670 static const char alt_10[] =
671 {0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
672 /* data16
673 nopw %cs:0L(%[re]ax,%[re]ax,1) */
674 static const char alt_long_11[] =
675 {0x66,
676 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
677 /* data16
678 data16
679 nopw %cs:0L(%[re]ax,%[re]ax,1) */
680 static const char alt_long_12[] =
681 {0x66,
682 0x66,
683 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
684 /* data16
685 data16
686 data16
687 nopw %cs:0L(%[re]ax,%[re]ax,1) */
688 static const char alt_long_13[] =
689 {0x66,
690 0x66,
691 0x66,
692 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
693 /* data16
694 data16
695 data16
696 data16
697 nopw %cs:0L(%[re]ax,%[re]ax,1) */
698 static const char alt_long_14[] =
699 {0x66,
700 0x66,
701 0x66,
702 0x66,
703 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
704 /* data16
705 data16
706 data16
707 data16
708 data16
709 nopw %cs:0L(%[re]ax,%[re]ax,1) */
710 static const char alt_long_15[] =
711 {0x66,
712 0x66,
713 0x66,
714 0x66,
715 0x66,
716 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
717 /* nopl 0(%[re]ax,%[re]ax,1)
718 nopw 0(%[re]ax,%[re]ax,1) */
719 static const char alt_short_11[] =
720 {0x0f,0x1f,0x44,0x00,0x00,
721 0x66,0x0f,0x1f,0x44,0x00,0x00};
722 /* nopw 0(%[re]ax,%[re]ax,1)
723 nopw 0(%[re]ax,%[re]ax,1) */
724 static const char alt_short_12[] =
725 {0x66,0x0f,0x1f,0x44,0x00,0x00,
726 0x66,0x0f,0x1f,0x44,0x00,0x00};
727 /* nopw 0(%[re]ax,%[re]ax,1)
728 nopl 0L(%[re]ax) */
729 static const char alt_short_13[] =
730 {0x66,0x0f,0x1f,0x44,0x00,0x00,
731 0x0f,0x1f,0x80,0x00,0x00,0x00,0x00};
732 /* nopl 0L(%[re]ax)
733 nopl 0L(%[re]ax) */
734 static const char alt_short_14[] =
735 {0x0f,0x1f,0x80,0x00,0x00,0x00,0x00,
736 0x0f,0x1f,0x80,0x00,0x00,0x00,0x00};
737 /* nopl 0L(%[re]ax)
738 nopl 0L(%[re]ax,%[re]ax,1) */
739 static const char alt_short_15[] =
740 {0x0f,0x1f,0x80,0x00,0x00,0x00,0x00,
741 0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
742 static const char *const alt_short_patt[] = {
743 f32_1, f32_2, alt_3, alt_4, alt_5, alt_6, alt_7, alt_8,
744 alt_9, alt_10, alt_short_11, alt_short_12, alt_short_13,
745 alt_short_14, alt_short_15
746 };
747 static const char *const alt_long_patt[] = {
748 f32_1, f32_2, alt_3, alt_4, alt_5, alt_6, alt_7, alt_8,
749 alt_9, alt_10, alt_long_11, alt_long_12, alt_long_13,
750 alt_long_14, alt_long_15
751 };
752
753 if (count <= 0 || count > 15)
754 return;
755
756 /* We need to decide which NOP sequence to use for 32bit and
757 64bit. When -mtune= is used:
758
759 1. For PROCESSOR_I486, PROCESSOR_PENTIUM and PROCESSOR_GENERIC32,
760 f32_patt will be used.
761 2. For PROCESSOR_K8 and PROCESSOR_AMDFAM10 in 64bit, NOPs with 0x66 prefix will be used.
762 3. For PROCESSOR_CORE2, alt_long_patt will be used.
763 4. For PROCESSOR_PENTIUMPRO, PROCESSOR_PENTIUM4, PROCESSOR_NOCONA,
764 PROCESSOR_CORE, PROCESSOR_CORE2, PROCESSOR_K6, PROCESSOR_ATHLON
765 and PROCESSOR_GENERIC64, alt_short_patt will be used.
766
767 When -mtune= isn't used, alt_short_patt will be used if
768 cpu_arch_isa_flags has Cpu686. Otherwise, f32_patt will be used.
769
770 When -march= or .arch is used, we can't use anything beyond
771 cpu_arch_isa_flags. */
772
773 if (flag_code == CODE_16BIT)
774 {
775 memcpy (fragP->fr_literal + fragP->fr_fix,
776 f16_patt[count - 1], count);
777 if (count > 8)
778 /* Adjust jump offset. */
779 fragP->fr_literal[fragP->fr_fix + 1] = count - 2;
780 }
781 else if (flag_code == CODE_64BIT && cpu_arch_tune == PROCESSOR_K8)
782 {
783 int i;
784 int nnops = (count + 3) / 4;
785 int len = count / nnops;
786 int remains = count - nnops * len;
787 int pos = 0;
788
789 /* The recommended way to pad 64bit code is to use NOPs preceded
790 by maximally four 0x66 prefixes. Balance the size of nops. */
791 for (i = 0; i < remains; i++)
792 {
793 memset (fragP->fr_literal + fragP->fr_fix + pos, 0x66, len);
794 fragP->fr_literal[fragP->fr_fix + pos + len] = 0x90;
795 pos += len + 1;
796 }
797 for (; i < nnops; i++)
798 {
799 memset (fragP->fr_literal + fragP->fr_fix + pos, 0x66, len - 1);
800 fragP->fr_literal[fragP->fr_fix + pos + len - 1] = 0x90;
801 pos += len;
802 }
803 }
804 else
805 {
806 const char *const *patt = NULL;
807
808 if (cpu_arch_isa == PROCESSOR_UNKNOWN)
809 {
810 /* PROCESSOR_UNKNOWN means that all ISAs may be used. */
811 switch (cpu_arch_tune)
812 {
813 case PROCESSOR_UNKNOWN:
814 /* We use cpu_arch_isa_flags to check if we SHOULD
815 optimize for Cpu686. */
816 if ((cpu_arch_isa_flags & Cpu686) != 0)
817 patt = alt_short_patt;
818 else
819 patt = f32_patt;
820 break;
821 case PROCESSOR_CORE2:
822 patt = alt_long_patt;
823 break;
824 case PROCESSOR_PENTIUMPRO:
825 case PROCESSOR_PENTIUM4:
826 case PROCESSOR_NOCONA:
827 case PROCESSOR_CORE:
828 case PROCESSOR_K6:
829 case PROCESSOR_ATHLON:
830 case PROCESSOR_K8:
831 case PROCESSOR_GENERIC64:
832 case PROCESSOR_AMDFAM10:
833 patt = alt_short_patt;
834 break;
835 case PROCESSOR_I486:
836 case PROCESSOR_PENTIUM:
837 case PROCESSOR_GENERIC32:
838 patt = f32_patt;
839 break;
840 }
841 }
842 else
843 {
844 switch (cpu_arch_tune)
845 {
846 case PROCESSOR_UNKNOWN:
847 /* When cpu_arch_isa is net, cpu_arch_tune shouldn't be
848 PROCESSOR_UNKNOWN. */
849 abort ();
850 break;
851
852 case PROCESSOR_I486:
853 case PROCESSOR_PENTIUM:
854 case PROCESSOR_PENTIUMPRO:
855 case PROCESSOR_PENTIUM4:
856 case PROCESSOR_NOCONA:
857 case PROCESSOR_CORE:
858 case PROCESSOR_K6:
859 case PROCESSOR_ATHLON:
860 case PROCESSOR_K8:
861 case PROCESSOR_AMDFAM10:
862 case PROCESSOR_GENERIC32:
863 /* We use cpu_arch_isa_flags to check if we CAN optimize
864 for Cpu686. */
865 if ((cpu_arch_isa_flags & Cpu686) != 0)
866 patt = alt_short_patt;
867 else
868 patt = f32_patt;
869 break;
870 case PROCESSOR_CORE2:
871 if ((cpu_arch_isa_flags & Cpu686) != 0)
872 patt = alt_long_patt;
873 else
874 patt = f32_patt;
875 break;
876 case PROCESSOR_GENERIC64:
877 patt = alt_short_patt;
878 break;
879 }
880 }
881
882 memcpy (fragP->fr_literal + fragP->fr_fix,
883 patt[count - 1], count);
884 }
885 fragP->fr_var = count;
886 }
887
888 static INLINE unsigned int
889 mode_from_disp_size (t)
890 unsigned int t;
891 {
892 return (t & Disp8) ? 1 : (t & (Disp16 | Disp32 | Disp32S)) ? 2 : 0;
893 }
894
895 static INLINE int
896 fits_in_signed_byte (num)
897 offsetT num;
898 {
899 return (num >= -128) && (num <= 127);
900 }
901
902 static INLINE int
903 fits_in_unsigned_byte (num)
904 offsetT num;
905 {
906 return (num & 0xff) == num;
907 }
908
909 static INLINE int
910 fits_in_unsigned_word (num)
911 offsetT num;
912 {
913 return (num & 0xffff) == num;
914 }
915
916 static INLINE int
917 fits_in_signed_word (num)
918 offsetT num;
919 {
920 return (-32768 <= num) && (num <= 32767);
921 }
922
923 static INLINE int
924 fits_in_signed_long (num)
925 offsetT num ATTRIBUTE_UNUSED;
926 {
927 #ifndef BFD64
928 return 1;
929 #else
930 return (!(((offsetT) -1 << 31) & num)
931 || (((offsetT) -1 << 31) & num) == ((offsetT) -1 << 31));
932 #endif
933 } /* fits_in_signed_long() */
934
935 static INLINE int
936 fits_in_unsigned_long (num)
937 offsetT num ATTRIBUTE_UNUSED;
938 {
939 #ifndef BFD64
940 return 1;
941 #else
942 return (num & (((offsetT) 2 << 31) - 1)) == num;
943 #endif
944 } /* fits_in_unsigned_long() */
945
946 static int
947 smallest_imm_type (num)
948 offsetT num;
949 {
950 if (cpu_arch_flags != (Cpu186 | Cpu286 | Cpu386 | Cpu486 | CpuNo64))
951 {
952 /* This code is disabled on the 486 because all the Imm1 forms
953 in the opcode table are slower on the i486. They're the
954 versions with the implicitly specified single-position
955 displacement, which has another syntax if you really want to
956 use that form. */
957 if (num == 1)
958 return Imm1 | Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64;
959 }
960 return (fits_in_signed_byte (num)
961 ? (Imm8S | Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
962 : fits_in_unsigned_byte (num)
963 ? (Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
964 : (fits_in_signed_word (num) || fits_in_unsigned_word (num))
965 ? (Imm16 | Imm32 | Imm32S | Imm64)
966 : fits_in_signed_long (num)
967 ? (Imm32 | Imm32S | Imm64)
968 : fits_in_unsigned_long (num)
969 ? (Imm32 | Imm64)
970 : Imm64);
971 }
972
973 static offsetT
974 offset_in_range (val, size)
975 offsetT val;
976 int size;
977 {
978 addressT mask;
979
980 switch (size)
981 {
982 case 1: mask = ((addressT) 1 << 8) - 1; break;
983 case 2: mask = ((addressT) 1 << 16) - 1; break;
984 case 4: mask = ((addressT) 2 << 31) - 1; break;
985 #ifdef BFD64
986 case 8: mask = ((addressT) 2 << 63) - 1; break;
987 #endif
988 default: abort ();
989 }
990
991 /* If BFD64, sign extend val. */
992 if (!use_rela_relocations)
993 if ((val & ~(((addressT) 2 << 31) - 1)) == 0)
994 val = (val ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
995
996 if ((val & ~mask) != 0 && (val & ~mask) != ~mask)
997 {
998 char buf1[40], buf2[40];
999
1000 sprint_value (buf1, val);
1001 sprint_value (buf2, val & mask);
1002 as_warn (_("%s shortened to %s"), buf1, buf2);
1003 }
1004 return val & mask;
1005 }
1006
1007 /* Returns 0 if attempting to add a prefix where one from the same
1008 class already exists, 1 if non rep/repne added, 2 if rep/repne
1009 added. */
1010 static int
1011 add_prefix (prefix)
1012 unsigned int prefix;
1013 {
1014 int ret = 1;
1015 unsigned int q;
1016
1017 if (prefix >= REX_OPCODE && prefix < REX_OPCODE + 16
1018 && flag_code == CODE_64BIT)
1019 {
1020 if ((i.prefix[REX_PREFIX] & prefix & REX_MODE64)
1021 || ((i.prefix[REX_PREFIX] & (REX_EXTX | REX_EXTY | REX_EXTZ))
1022 && (prefix & (REX_EXTX | REX_EXTY | REX_EXTZ))))
1023 ret = 0;
1024 q = REX_PREFIX;
1025 }
1026 else
1027 {
1028 switch (prefix)
1029 {
1030 default:
1031 abort ();
1032
1033 case CS_PREFIX_OPCODE:
1034 case DS_PREFIX_OPCODE:
1035 case ES_PREFIX_OPCODE:
1036 case FS_PREFIX_OPCODE:
1037 case GS_PREFIX_OPCODE:
1038 case SS_PREFIX_OPCODE:
1039 q = SEG_PREFIX;
1040 break;
1041
1042 case REPNE_PREFIX_OPCODE:
1043 case REPE_PREFIX_OPCODE:
1044 ret = 2;
1045 /* fall thru */
1046 case LOCK_PREFIX_OPCODE:
1047 q = LOCKREP_PREFIX;
1048 break;
1049
1050 case FWAIT_OPCODE:
1051 q = WAIT_PREFIX;
1052 break;
1053
1054 case ADDR_PREFIX_OPCODE:
1055 q = ADDR_PREFIX;
1056 break;
1057
1058 case DATA_PREFIX_OPCODE:
1059 q = DATA_PREFIX;
1060 break;
1061 }
1062 if (i.prefix[q] != 0)
1063 ret = 0;
1064 }
1065
1066 if (ret)
1067 {
1068 if (!i.prefix[q])
1069 ++i.prefixes;
1070 i.prefix[q] |= prefix;
1071 }
1072 else
1073 as_bad (_("same type of prefix used twice"));
1074
1075 return ret;
1076 }
1077
1078 static void
1079 set_code_flag (value)
1080 int value;
1081 {
1082 flag_code = value;
1083 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
1084 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
1085 if (value == CODE_64BIT && !(cpu_arch_flags & CpuSledgehammer))
1086 {
1087 as_bad (_("64bit mode not supported on this CPU."));
1088 }
1089 if (value == CODE_32BIT && !(cpu_arch_flags & Cpu386))
1090 {
1091 as_bad (_("32bit mode not supported on this CPU."));
1092 }
1093 stackop_size = '\0';
1094 }
1095
1096 static void
1097 set_16bit_gcc_code_flag (new_code_flag)
1098 int new_code_flag;
1099 {
1100 flag_code = new_code_flag;
1101 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
1102 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
1103 stackop_size = LONG_MNEM_SUFFIX;
1104 }
1105
1106 static void
1107 set_intel_syntax (syntax_flag)
1108 int syntax_flag;
1109 {
1110 /* Find out if register prefixing is specified. */
1111 int ask_naked_reg = 0;
1112
1113 SKIP_WHITESPACE ();
1114 if (!is_end_of_line[(unsigned char) *input_line_pointer])
1115 {
1116 char *string = input_line_pointer;
1117 int e = get_symbol_end ();
1118
1119 if (strcmp (string, "prefix") == 0)
1120 ask_naked_reg = 1;
1121 else if (strcmp (string, "noprefix") == 0)
1122 ask_naked_reg = -1;
1123 else
1124 as_bad (_("bad argument to syntax directive."));
1125 *input_line_pointer = e;
1126 }
1127 demand_empty_rest_of_line ();
1128
1129 intel_syntax = syntax_flag;
1130
1131 if (ask_naked_reg == 0)
1132 allow_naked_reg = (intel_syntax
1133 && (bfd_get_symbol_leading_char (stdoutput) != '\0'));
1134 else
1135 allow_naked_reg = (ask_naked_reg < 0);
1136
1137 identifier_chars['%'] = intel_syntax && allow_naked_reg ? '%' : 0;
1138 identifier_chars['$'] = intel_syntax ? '$' : 0;
1139 }
1140
1141 static void
1142 set_cpu_arch (dummy)
1143 int dummy ATTRIBUTE_UNUSED;
1144 {
1145 SKIP_WHITESPACE ();
1146
1147 if (!is_end_of_line[(unsigned char) *input_line_pointer])
1148 {
1149 char *string = input_line_pointer;
1150 int e = get_symbol_end ();
1151 unsigned int i;
1152
1153 for (i = 0; i < ARRAY_SIZE (cpu_arch); i++)
1154 {
1155 if (strcmp (string, cpu_arch[i].name) == 0)
1156 {
1157 if (*string != '.')
1158 {
1159 cpu_arch_name = cpu_arch[i].name;
1160 cpu_sub_arch_name = NULL;
1161 cpu_arch_flags = (cpu_arch[i].flags
1162 | (flag_code == CODE_64BIT ? Cpu64 : CpuNo64));
1163 cpu_arch_isa = cpu_arch[i].type;
1164 cpu_arch_isa_flags = cpu_arch[i].flags;
1165 if (!cpu_arch_tune_set)
1166 {
1167 cpu_arch_tune = cpu_arch_isa;
1168 cpu_arch_tune_flags = cpu_arch_isa_flags;
1169 }
1170 break;
1171 }
1172 if ((cpu_arch_flags | cpu_arch[i].flags) != cpu_arch_flags)
1173 {
1174 cpu_sub_arch_name = cpu_arch[i].name;
1175 cpu_arch_flags |= cpu_arch[i].flags;
1176 }
1177 *input_line_pointer = e;
1178 demand_empty_rest_of_line ();
1179 return;
1180 }
1181 }
1182 if (i >= ARRAY_SIZE (cpu_arch))
1183 as_bad (_("no such architecture: `%s'"), string);
1184
1185 *input_line_pointer = e;
1186 }
1187 else
1188 as_bad (_("missing cpu architecture"));
1189
1190 no_cond_jump_promotion = 0;
1191 if (*input_line_pointer == ','
1192 && !is_end_of_line[(unsigned char) input_line_pointer[1]])
1193 {
1194 char *string = ++input_line_pointer;
1195 int e = get_symbol_end ();
1196
1197 if (strcmp (string, "nojumps") == 0)
1198 no_cond_jump_promotion = 1;
1199 else if (strcmp (string, "jumps") == 0)
1200 ;
1201 else
1202 as_bad (_("no such architecture modifier: `%s'"), string);
1203
1204 *input_line_pointer = e;
1205 }
1206
1207 demand_empty_rest_of_line ();
1208 }
1209
1210 unsigned long
1211 i386_mach ()
1212 {
1213 if (!strcmp (default_arch, "x86_64"))
1214 return bfd_mach_x86_64;
1215 else if (!strcmp (default_arch, "i386"))
1216 return bfd_mach_i386_i386;
1217 else
1218 as_fatal (_("Unknown architecture"));
1219 }
1220 \f
1221 void
1222 md_begin ()
1223 {
1224 const char *hash_err;
1225
1226 /* Initialize op_hash hash table. */
1227 op_hash = hash_new ();
1228
1229 {
1230 const template *optab;
1231 templates *core_optab;
1232
1233 /* Setup for loop. */
1234 optab = i386_optab;
1235 core_optab = (templates *) xmalloc (sizeof (templates));
1236 core_optab->start = optab;
1237
1238 while (1)
1239 {
1240 ++optab;
1241 if (optab->name == NULL
1242 || strcmp (optab->name, (optab - 1)->name) != 0)
1243 {
1244 /* different name --> ship out current template list;
1245 add to hash table; & begin anew. */
1246 core_optab->end = optab;
1247 hash_err = hash_insert (op_hash,
1248 (optab - 1)->name,
1249 (PTR) core_optab);
1250 if (hash_err)
1251 {
1252 as_fatal (_("Internal Error: Can't hash %s: %s"),
1253 (optab - 1)->name,
1254 hash_err);
1255 }
1256 if (optab->name == NULL)
1257 break;
1258 core_optab = (templates *) xmalloc (sizeof (templates));
1259 core_optab->start = optab;
1260 }
1261 }
1262 }
1263
1264 /* Initialize reg_hash hash table. */
1265 reg_hash = hash_new ();
1266 {
1267 const reg_entry *regtab;
1268
1269 for (regtab = i386_regtab;
1270 regtab < i386_regtab + sizeof (i386_regtab) / sizeof (i386_regtab[0]);
1271 regtab++)
1272 {
1273 hash_err = hash_insert (reg_hash, regtab->reg_name, (PTR) regtab);
1274 if (hash_err)
1275 as_fatal (_("Internal Error: Can't hash %s: %s"),
1276 regtab->reg_name,
1277 hash_err);
1278 }
1279 }
1280
1281 /* Fill in lexical tables: mnemonic_chars, operand_chars. */
1282 {
1283 int c;
1284 char *p;
1285
1286 for (c = 0; c < 256; c++)
1287 {
1288 if (ISDIGIT (c))
1289 {
1290 digit_chars[c] = c;
1291 mnemonic_chars[c] = c;
1292 register_chars[c] = c;
1293 operand_chars[c] = c;
1294 }
1295 else if (ISLOWER (c))
1296 {
1297 mnemonic_chars[c] = c;
1298 register_chars[c] = c;
1299 operand_chars[c] = c;
1300 }
1301 else if (ISUPPER (c))
1302 {
1303 mnemonic_chars[c] = TOLOWER (c);
1304 register_chars[c] = mnemonic_chars[c];
1305 operand_chars[c] = c;
1306 }
1307
1308 if (ISALPHA (c) || ISDIGIT (c))
1309 identifier_chars[c] = c;
1310 else if (c >= 128)
1311 {
1312 identifier_chars[c] = c;
1313 operand_chars[c] = c;
1314 }
1315 }
1316
1317 #ifdef LEX_AT
1318 identifier_chars['@'] = '@';
1319 #endif
1320 #ifdef LEX_QM
1321 identifier_chars['?'] = '?';
1322 operand_chars['?'] = '?';
1323 #endif
1324 digit_chars['-'] = '-';
1325 mnemonic_chars['-'] = '-';
1326 identifier_chars['_'] = '_';
1327 identifier_chars['.'] = '.';
1328
1329 for (p = operand_special_chars; *p != '\0'; p++)
1330 operand_chars[(unsigned char) *p] = *p;
1331 }
1332
1333 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1334 if (IS_ELF)
1335 {
1336 record_alignment (text_section, 2);
1337 record_alignment (data_section, 2);
1338 record_alignment (bss_section, 2);
1339 }
1340 #endif
1341
1342 if (flag_code == CODE_64BIT)
1343 {
1344 x86_dwarf2_return_column = 16;
1345 x86_cie_data_alignment = -8;
1346 }
1347 else
1348 {
1349 x86_dwarf2_return_column = 8;
1350 x86_cie_data_alignment = -4;
1351 }
1352 }
1353
1354 void
1355 i386_print_statistics (file)
1356 FILE *file;
1357 {
1358 hash_print_statistics (file, "i386 opcode", op_hash);
1359 hash_print_statistics (file, "i386 register", reg_hash);
1360 }
1361 \f
1362 #ifdef DEBUG386
1363
1364 /* Debugging routines for md_assemble. */
1365 static void pi PARAMS ((char *, i386_insn *));
1366 static void pte PARAMS ((template *));
1367 static void pt PARAMS ((unsigned int));
1368 static void pe PARAMS ((expressionS *));
1369 static void ps PARAMS ((symbolS *));
1370
1371 static void
1372 pi (line, x)
1373 char *line;
1374 i386_insn *x;
1375 {
1376 unsigned int i;
1377
1378 fprintf (stdout, "%s: template ", line);
1379 pte (&x->tm);
1380 fprintf (stdout, " address: base %s index %s scale %x\n",
1381 x->base_reg ? x->base_reg->reg_name : "none",
1382 x->index_reg ? x->index_reg->reg_name : "none",
1383 x->log2_scale_factor);
1384 fprintf (stdout, " modrm: mode %x reg %x reg/mem %x\n",
1385 x->rm.mode, x->rm.reg, x->rm.regmem);
1386 fprintf (stdout, " sib: base %x index %x scale %x\n",
1387 x->sib.base, x->sib.index, x->sib.scale);
1388 fprintf (stdout, " rex: 64bit %x extX %x extY %x extZ %x\n",
1389 (x->rex & REX_MODE64) != 0,
1390 (x->rex & REX_EXTX) != 0,
1391 (x->rex & REX_EXTY) != 0,
1392 (x->rex & REX_EXTZ) != 0);
1393 for (i = 0; i < x->operands; i++)
1394 {
1395 fprintf (stdout, " #%d: ", i + 1);
1396 pt (x->types[i]);
1397 fprintf (stdout, "\n");
1398 if (x->types[i]
1399 & (Reg | SReg2 | SReg3 | Control | Debug | Test | RegMMX | RegXMM))
1400 fprintf (stdout, "%s\n", x->op[i].regs->reg_name);
1401 if (x->types[i] & Imm)
1402 pe (x->op[i].imms);
1403 if (x->types[i] & Disp)
1404 pe (x->op[i].disps);
1405 }
1406 }
1407
1408 static void
1409 pte (t)
1410 template *t;
1411 {
1412 unsigned int i;
1413 fprintf (stdout, " %d operands ", t->operands);
1414 fprintf (stdout, "opcode %x ", t->base_opcode);
1415 if (t->extension_opcode != None)
1416 fprintf (stdout, "ext %x ", t->extension_opcode);
1417 if (t->opcode_modifier & D)
1418 fprintf (stdout, "D");
1419 if (t->opcode_modifier & W)
1420 fprintf (stdout, "W");
1421 fprintf (stdout, "\n");
1422 for (i = 0; i < t->operands; i++)
1423 {
1424 fprintf (stdout, " #%d type ", i + 1);
1425 pt (t->operand_types[i]);
1426 fprintf (stdout, "\n");
1427 }
1428 }
1429
1430 static void
1431 pe (e)
1432 expressionS *e;
1433 {
1434 fprintf (stdout, " operation %d\n", e->X_op);
1435 fprintf (stdout, " add_number %ld (%lx)\n",
1436 (long) e->X_add_number, (long) e->X_add_number);
1437 if (e->X_add_symbol)
1438 {
1439 fprintf (stdout, " add_symbol ");
1440 ps (e->X_add_symbol);
1441 fprintf (stdout, "\n");
1442 }
1443 if (e->X_op_symbol)
1444 {
1445 fprintf (stdout, " op_symbol ");
1446 ps (e->X_op_symbol);
1447 fprintf (stdout, "\n");
1448 }
1449 }
1450
1451 static void
1452 ps (s)
1453 symbolS *s;
1454 {
1455 fprintf (stdout, "%s type %s%s",
1456 S_GET_NAME (s),
1457 S_IS_EXTERNAL (s) ? "EXTERNAL " : "",
1458 segment_name (S_GET_SEGMENT (s)));
1459 }
1460
1461 static struct type_name
1462 {
1463 unsigned int mask;
1464 char *tname;
1465 }
1466 const type_names[] =
1467 {
1468 { Reg8, "r8" },
1469 { Reg16, "r16" },
1470 { Reg32, "r32" },
1471 { Reg64, "r64" },
1472 { Imm8, "i8" },
1473 { Imm8S, "i8s" },
1474 { Imm16, "i16" },
1475 { Imm32, "i32" },
1476 { Imm32S, "i32s" },
1477 { Imm64, "i64" },
1478 { Imm1, "i1" },
1479 { BaseIndex, "BaseIndex" },
1480 { Disp8, "d8" },
1481 { Disp16, "d16" },
1482 { Disp32, "d32" },
1483 { Disp32S, "d32s" },
1484 { Disp64, "d64" },
1485 { InOutPortReg, "InOutPortReg" },
1486 { ShiftCount, "ShiftCount" },
1487 { Control, "control reg" },
1488 { Test, "test reg" },
1489 { Debug, "debug reg" },
1490 { FloatReg, "FReg" },
1491 { FloatAcc, "FAcc" },
1492 { SReg2, "SReg2" },
1493 { SReg3, "SReg3" },
1494 { Acc, "Acc" },
1495 { JumpAbsolute, "Jump Absolute" },
1496 { RegMMX, "rMMX" },
1497 { RegXMM, "rXMM" },
1498 { EsSeg, "es" },
1499 { 0, "" }
1500 };
1501
1502 static void
1503 pt (t)
1504 unsigned int t;
1505 {
1506 const struct type_name *ty;
1507
1508 for (ty = type_names; ty->mask; ty++)
1509 if (t & ty->mask)
1510 fprintf (stdout, "%s, ", ty->tname);
1511 fflush (stdout);
1512 }
1513
1514 #endif /* DEBUG386 */
1515 \f
1516 static bfd_reloc_code_real_type
1517 reloc (unsigned int size,
1518 int pcrel,
1519 int sign,
1520 bfd_reloc_code_real_type other)
1521 {
1522 if (other != NO_RELOC)
1523 {
1524 reloc_howto_type *reloc;
1525
1526 if (size == 8)
1527 switch (other)
1528 {
1529 case BFD_RELOC_X86_64_GOT32:
1530 return BFD_RELOC_X86_64_GOT64;
1531 break;
1532 case BFD_RELOC_X86_64_PLTOFF64:
1533 return BFD_RELOC_X86_64_PLTOFF64;
1534 break;
1535 case BFD_RELOC_X86_64_GOTPC32:
1536 other = BFD_RELOC_X86_64_GOTPC64;
1537 break;
1538 case BFD_RELOC_X86_64_GOTPCREL:
1539 other = BFD_RELOC_X86_64_GOTPCREL64;
1540 break;
1541 case BFD_RELOC_X86_64_TPOFF32:
1542 other = BFD_RELOC_X86_64_TPOFF64;
1543 break;
1544 case BFD_RELOC_X86_64_DTPOFF32:
1545 other = BFD_RELOC_X86_64_DTPOFF64;
1546 break;
1547 default:
1548 break;
1549 }
1550
1551 /* Sign-checking 4-byte relocations in 16-/32-bit code is pointless. */
1552 if (size == 4 && flag_code != CODE_64BIT)
1553 sign = -1;
1554
1555 reloc = bfd_reloc_type_lookup (stdoutput, other);
1556 if (!reloc)
1557 as_bad (_("unknown relocation (%u)"), other);
1558 else if (size != bfd_get_reloc_size (reloc))
1559 as_bad (_("%u-byte relocation cannot be applied to %u-byte field"),
1560 bfd_get_reloc_size (reloc),
1561 size);
1562 else if (pcrel && !reloc->pc_relative)
1563 as_bad (_("non-pc-relative relocation for pc-relative field"));
1564 else if ((reloc->complain_on_overflow == complain_overflow_signed
1565 && !sign)
1566 || (reloc->complain_on_overflow == complain_overflow_unsigned
1567 && sign > 0))
1568 as_bad (_("relocated field and relocation type differ in signedness"));
1569 else
1570 return other;
1571 return NO_RELOC;
1572 }
1573
1574 if (pcrel)
1575 {
1576 if (!sign)
1577 as_bad (_("there are no unsigned pc-relative relocations"));
1578 switch (size)
1579 {
1580 case 1: return BFD_RELOC_8_PCREL;
1581 case 2: return BFD_RELOC_16_PCREL;
1582 case 4: return BFD_RELOC_32_PCREL;
1583 case 8: return BFD_RELOC_64_PCREL;
1584 }
1585 as_bad (_("cannot do %u byte pc-relative relocation"), size);
1586 }
1587 else
1588 {
1589 if (sign > 0)
1590 switch (size)
1591 {
1592 case 4: return BFD_RELOC_X86_64_32S;
1593 }
1594 else
1595 switch (size)
1596 {
1597 case 1: return BFD_RELOC_8;
1598 case 2: return BFD_RELOC_16;
1599 case 4: return BFD_RELOC_32;
1600 case 8: return BFD_RELOC_64;
1601 }
1602 as_bad (_("cannot do %s %u byte relocation"),
1603 sign > 0 ? "signed" : "unsigned", size);
1604 }
1605
1606 abort ();
1607 return BFD_RELOC_NONE;
1608 }
1609
1610 /* Here we decide which fixups can be adjusted to make them relative to
1611 the beginning of the section instead of the symbol. Basically we need
1612 to make sure that the dynamic relocations are done correctly, so in
1613 some cases we force the original symbol to be used. */
1614
1615 int
1616 tc_i386_fix_adjustable (fixP)
1617 fixS *fixP ATTRIBUTE_UNUSED;
1618 {
1619 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1620 if (!IS_ELF)
1621 return 1;
1622
1623 /* Don't adjust pc-relative references to merge sections in 64-bit
1624 mode. */
1625 if (use_rela_relocations
1626 && (S_GET_SEGMENT (fixP->fx_addsy)->flags & SEC_MERGE) != 0
1627 && fixP->fx_pcrel)
1628 return 0;
1629
1630 /* The x86_64 GOTPCREL are represented as 32bit PCrel relocations
1631 and changed later by validate_fix. */
1632 if (GOT_symbol && fixP->fx_subsy == GOT_symbol
1633 && fixP->fx_r_type == BFD_RELOC_32_PCREL)
1634 return 0;
1635
1636 /* adjust_reloc_syms doesn't know about the GOT. */
1637 if (fixP->fx_r_type == BFD_RELOC_386_GOTOFF
1638 || fixP->fx_r_type == BFD_RELOC_386_PLT32
1639 || fixP->fx_r_type == BFD_RELOC_386_GOT32
1640 || fixP->fx_r_type == BFD_RELOC_386_TLS_GD
1641 || fixP->fx_r_type == BFD_RELOC_386_TLS_LDM
1642 || fixP->fx_r_type == BFD_RELOC_386_TLS_LDO_32
1643 || fixP->fx_r_type == BFD_RELOC_386_TLS_IE_32
1644 || fixP->fx_r_type == BFD_RELOC_386_TLS_IE
1645 || fixP->fx_r_type == BFD_RELOC_386_TLS_GOTIE
1646 || fixP->fx_r_type == BFD_RELOC_386_TLS_LE_32
1647 || fixP->fx_r_type == BFD_RELOC_386_TLS_LE
1648 || fixP->fx_r_type == BFD_RELOC_386_TLS_GOTDESC
1649 || fixP->fx_r_type == BFD_RELOC_386_TLS_DESC_CALL
1650 || fixP->fx_r_type == BFD_RELOC_X86_64_PLT32
1651 || fixP->fx_r_type == BFD_RELOC_X86_64_GOT32
1652 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPCREL
1653 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSGD
1654 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSLD
1655 || fixP->fx_r_type == BFD_RELOC_X86_64_DTPOFF32
1656 || fixP->fx_r_type == BFD_RELOC_X86_64_DTPOFF64
1657 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTTPOFF
1658 || fixP->fx_r_type == BFD_RELOC_X86_64_TPOFF32
1659 || fixP->fx_r_type == BFD_RELOC_X86_64_TPOFF64
1660 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTOFF64
1661 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPC32_TLSDESC
1662 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSDESC_CALL
1663 || fixP->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1664 || fixP->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1665 return 0;
1666 #endif
1667 return 1;
1668 }
1669
1670 static int intel_float_operand PARAMS ((const char *mnemonic));
1671
1672 static int
1673 intel_float_operand (mnemonic)
1674 const char *mnemonic;
1675 {
1676 /* Note that the value returned is meaningful only for opcodes with (memory)
1677 operands, hence the code here is free to improperly handle opcodes that
1678 have no operands (for better performance and smaller code). */
1679
1680 if (mnemonic[0] != 'f')
1681 return 0; /* non-math */
1682
1683 switch (mnemonic[1])
1684 {
1685 /* fclex, fdecstp, fdisi, femms, feni, fincstp, finit, fsetpm, and
1686 the fs segment override prefix not currently handled because no
1687 call path can make opcodes without operands get here */
1688 case 'i':
1689 return 2 /* integer op */;
1690 case 'l':
1691 if (mnemonic[2] == 'd' && (mnemonic[3] == 'c' || mnemonic[3] == 'e'))
1692 return 3; /* fldcw/fldenv */
1693 break;
1694 case 'n':
1695 if (mnemonic[2] != 'o' /* fnop */)
1696 return 3; /* non-waiting control op */
1697 break;
1698 case 'r':
1699 if (mnemonic[2] == 's')
1700 return 3; /* frstor/frstpm */
1701 break;
1702 case 's':
1703 if (mnemonic[2] == 'a')
1704 return 3; /* fsave */
1705 if (mnemonic[2] == 't')
1706 {
1707 switch (mnemonic[3])
1708 {
1709 case 'c': /* fstcw */
1710 case 'd': /* fstdw */
1711 case 'e': /* fstenv */
1712 case 's': /* fsts[gw] */
1713 return 3;
1714 }
1715 }
1716 break;
1717 case 'x':
1718 if (mnemonic[2] == 'r' || mnemonic[2] == 's')
1719 return 0; /* fxsave/fxrstor are not really math ops */
1720 break;
1721 }
1722
1723 return 1;
1724 }
1725
1726 /* This is the guts of the machine-dependent assembler. LINE points to a
1727 machine dependent instruction. This function is supposed to emit
1728 the frags/bytes it assembles to. */
1729
1730 void
1731 md_assemble (line)
1732 char *line;
1733 {
1734 int j;
1735 char mnemonic[MAX_MNEM_SIZE];
1736
1737 /* Initialize globals. */
1738 memset (&i, '\0', sizeof (i));
1739 for (j = 0; j < MAX_OPERANDS; j++)
1740 i.reloc[j] = NO_RELOC;
1741 memset (disp_expressions, '\0', sizeof (disp_expressions));
1742 memset (im_expressions, '\0', sizeof (im_expressions));
1743 save_stack_p = save_stack;
1744
1745 /* First parse an instruction mnemonic & call i386_operand for the operands.
1746 We assume that the scrubber has arranged it so that line[0] is the valid
1747 start of a (possibly prefixed) mnemonic. */
1748
1749 line = parse_insn (line, mnemonic);
1750 if (line == NULL)
1751 return;
1752
1753 line = parse_operands (line, mnemonic);
1754 if (line == NULL)
1755 return;
1756
1757 /* The order of the immediates should be reversed
1758 for 2 immediates extrq and insertq instructions */
1759 if ((i.imm_operands == 2) &&
1760 ((strcmp (mnemonic, "extrq") == 0)
1761 || (strcmp (mnemonic, "insertq") == 0)))
1762 {
1763 swap_imm_operands ();
1764 /* "extrq" and insertq" are the only two instructions whose operands
1765 have to be reversed even though they have two immediate operands.
1766 */
1767 if (intel_syntax)
1768 swap_operands ();
1769 }
1770
1771 /* Now we've parsed the mnemonic into a set of templates, and have the
1772 operands at hand. */
1773
1774 /* All intel opcodes have reversed operands except for "bound" and
1775 "enter". We also don't reverse intersegment "jmp" and "call"
1776 instructions with 2 immediate operands so that the immediate segment
1777 precedes the offset, as it does when in AT&T mode. */
1778 if (intel_syntax && i.operands > 1
1779 && (strcmp (mnemonic, "bound") != 0)
1780 && (strcmp (mnemonic, "invlpga") != 0)
1781 && !((i.types[0] & Imm) && (i.types[1] & Imm)))
1782 swap_operands ();
1783
1784 if (i.imm_operands)
1785 optimize_imm ();
1786
1787 /* Don't optimize displacement for movabs since it only takes 64bit
1788 displacement. */
1789 if (i.disp_operands
1790 && (flag_code != CODE_64BIT
1791 || strcmp (mnemonic, "movabs") != 0))
1792 optimize_disp ();
1793
1794 /* Next, we find a template that matches the given insn,
1795 making sure the overlap of the given operands types is consistent
1796 with the template operand types. */
1797
1798 if (!match_template ())
1799 return;
1800
1801 if (intel_syntax)
1802 {
1803 /* Undo SYSV386_COMPAT brokenness when in Intel mode. See i386.h */
1804 if (SYSV386_COMPAT
1805 && (i.tm.base_opcode & 0xfffffde0) == 0xdce0)
1806 i.tm.base_opcode ^= FloatR;
1807
1808 /* Zap movzx and movsx suffix. The suffix may have been set from
1809 "word ptr" or "byte ptr" on the source operand, but we'll use
1810 the suffix later to choose the destination register. */
1811 if ((i.tm.base_opcode & ~9) == 0x0fb6)
1812 {
1813 if (i.reg_operands < 2
1814 && !i.suffix
1815 && (~i.tm.opcode_modifier
1816 & (No_bSuf
1817 | No_wSuf
1818 | No_lSuf
1819 | No_sSuf
1820 | No_xSuf
1821 | No_qSuf)))
1822 as_bad (_("ambiguous operand size for `%s'"), i.tm.name);
1823
1824 i.suffix = 0;
1825 }
1826 }
1827
1828 if (i.tm.opcode_modifier & FWait)
1829 if (!add_prefix (FWAIT_OPCODE))
1830 return;
1831
1832 /* Check string instruction segment overrides. */
1833 if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0)
1834 {
1835 if (!check_string ())
1836 return;
1837 }
1838
1839 if (!process_suffix ())
1840 return;
1841
1842 /* Make still unresolved immediate matches conform to size of immediate
1843 given in i.suffix. */
1844 if (!finalize_imm ())
1845 return;
1846
1847 if (i.types[0] & Imm1)
1848 i.imm_operands = 0; /* kludge for shift insns. */
1849 if (i.types[0] & ImplicitRegister)
1850 i.reg_operands--;
1851 if (i.types[1] & ImplicitRegister)
1852 i.reg_operands--;
1853 if (i.types[2] & ImplicitRegister)
1854 i.reg_operands--;
1855
1856 if (i.tm.opcode_modifier & ImmExt)
1857 {
1858 expressionS *exp;
1859
1860 if ((i.tm.cpu_flags & CpuSSE3) && i.operands > 0)
1861 {
1862 /* Streaming SIMD extensions 3 Instructions have the fixed
1863 operands with an opcode suffix which is coded in the same
1864 place as an 8-bit immediate field would be. Here we check
1865 those operands and remove them afterwards. */
1866 unsigned int x;
1867
1868 for (x = 0; x < i.operands; x++)
1869 if (i.op[x].regs->reg_num != x)
1870 as_bad (_("can't use register '%%%s' as operand %d in '%s'."),
1871 i.op[x].regs->reg_name, x + 1, i.tm.name);
1872 i.operands = 0;
1873 }
1874
1875 /* These AMD 3DNow! and Intel Katmai New Instructions have an
1876 opcode suffix which is coded in the same place as an 8-bit
1877 immediate field would be. Here we fake an 8-bit immediate
1878 operand from the opcode suffix stored in tm.extension_opcode. */
1879
1880 assert (i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS);
1881
1882 exp = &im_expressions[i.imm_operands++];
1883 i.op[i.operands].imms = exp;
1884 i.types[i.operands++] = Imm8;
1885 exp->X_op = O_constant;
1886 exp->X_add_number = i.tm.extension_opcode;
1887 i.tm.extension_opcode = None;
1888 }
1889
1890 /* For insns with operands there are more diddles to do to the opcode. */
1891 if (i.operands)
1892 {
1893 if (!process_operands ())
1894 return;
1895 }
1896 else if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
1897 {
1898 /* UnixWare fsub no args is alias for fsubp, fadd -> faddp, etc. */
1899 as_warn (_("translating to `%sp'"), i.tm.name);
1900 }
1901
1902 /* Handle conversion of 'int $3' --> special int3 insn. */
1903 if (i.tm.base_opcode == INT_OPCODE && i.op[0].imms->X_add_number == 3)
1904 {
1905 i.tm.base_opcode = INT3_OPCODE;
1906 i.imm_operands = 0;
1907 }
1908
1909 if ((i.tm.opcode_modifier & (Jump | JumpByte | JumpDword))
1910 && i.op[0].disps->X_op == O_constant)
1911 {
1912 /* Convert "jmp constant" (and "call constant") to a jump (call) to
1913 the absolute address given by the constant. Since ix86 jumps and
1914 calls are pc relative, we need to generate a reloc. */
1915 i.op[0].disps->X_add_symbol = &abs_symbol;
1916 i.op[0].disps->X_op = O_symbol;
1917 }
1918
1919 if ((i.tm.opcode_modifier & Rex64) != 0)
1920 i.rex |= REX_MODE64;
1921
1922 /* For 8 bit registers we need an empty rex prefix. Also if the
1923 instruction already has a prefix, we need to convert old
1924 registers to new ones. */
1925
1926 if (((i.types[0] & Reg8) != 0
1927 && (i.op[0].regs->reg_flags & RegRex64) != 0)
1928 || ((i.types[1] & Reg8) != 0
1929 && (i.op[1].regs->reg_flags & RegRex64) != 0)
1930 || (((i.types[0] & Reg8) != 0 || (i.types[1] & Reg8) != 0)
1931 && i.rex != 0))
1932 {
1933 int x;
1934
1935 i.rex |= REX_OPCODE;
1936 for (x = 0; x < 2; x++)
1937 {
1938 /* Look for 8 bit operand that uses old registers. */
1939 if ((i.types[x] & Reg8) != 0
1940 && (i.op[x].regs->reg_flags & RegRex64) == 0)
1941 {
1942 /* In case it is "hi" register, give up. */
1943 if (i.op[x].regs->reg_num > 3)
1944 as_bad (_("can't encode register '%%%s' in an instruction requiring REX prefix."),
1945 i.op[x].regs->reg_name);
1946
1947 /* Otherwise it is equivalent to the extended register.
1948 Since the encoding doesn't change this is merely
1949 cosmetic cleanup for debug output. */
1950
1951 i.op[x].regs = i.op[x].regs + 8;
1952 }
1953 }
1954 }
1955
1956 if (i.rex != 0)
1957 add_prefix (REX_OPCODE | i.rex);
1958
1959 /* We are ready to output the insn. */
1960 output_insn ();
1961 }
1962
1963 static char *
1964 parse_insn (line, mnemonic)
1965 char *line;
1966 char *mnemonic;
1967 {
1968 char *l = line;
1969 char *token_start = l;
1970 char *mnem_p;
1971 int supported;
1972 const template *t;
1973
1974 /* Non-zero if we found a prefix only acceptable with string insns. */
1975 const char *expecting_string_instruction = NULL;
1976
1977 while (1)
1978 {
1979 mnem_p = mnemonic;
1980 while ((*mnem_p = mnemonic_chars[(unsigned char) *l]) != 0)
1981 {
1982 mnem_p++;
1983 if (mnem_p >= mnemonic + MAX_MNEM_SIZE)
1984 {
1985 as_bad (_("no such instruction: `%s'"), token_start);
1986 return NULL;
1987 }
1988 l++;
1989 }
1990 if (!is_space_char (*l)
1991 && *l != END_OF_INSN
1992 && (intel_syntax
1993 || (*l != PREFIX_SEPARATOR
1994 && *l != ',')))
1995 {
1996 as_bad (_("invalid character %s in mnemonic"),
1997 output_invalid (*l));
1998 return NULL;
1999 }
2000 if (token_start == l)
2001 {
2002 if (!intel_syntax && *l == PREFIX_SEPARATOR)
2003 as_bad (_("expecting prefix; got nothing"));
2004 else
2005 as_bad (_("expecting mnemonic; got nothing"));
2006 return NULL;
2007 }
2008
2009 /* Look up instruction (or prefix) via hash table. */
2010 current_templates = hash_find (op_hash, mnemonic);
2011
2012 if (*l != END_OF_INSN
2013 && (!is_space_char (*l) || l[1] != END_OF_INSN)
2014 && current_templates
2015 && (current_templates->start->opcode_modifier & IsPrefix))
2016 {
2017 if (current_templates->start->cpu_flags
2018 & (flag_code != CODE_64BIT ? Cpu64 : CpuNo64))
2019 {
2020 as_bad ((flag_code != CODE_64BIT
2021 ? _("`%s' is only supported in 64-bit mode")
2022 : _("`%s' is not supported in 64-bit mode")),
2023 current_templates->start->name);
2024 return NULL;
2025 }
2026 /* If we are in 16-bit mode, do not allow addr16 or data16.
2027 Similarly, in 32-bit mode, do not allow addr32 or data32. */
2028 if ((current_templates->start->opcode_modifier & (Size16 | Size32))
2029 && flag_code != CODE_64BIT
2030 && (((current_templates->start->opcode_modifier & Size32) != 0)
2031 ^ (flag_code == CODE_16BIT)))
2032 {
2033 as_bad (_("redundant %s prefix"),
2034 current_templates->start->name);
2035 return NULL;
2036 }
2037 /* Add prefix, checking for repeated prefixes. */
2038 switch (add_prefix (current_templates->start->base_opcode))
2039 {
2040 case 0:
2041 return NULL;
2042 case 2:
2043 expecting_string_instruction = current_templates->start->name;
2044 break;
2045 }
2046 /* Skip past PREFIX_SEPARATOR and reset token_start. */
2047 token_start = ++l;
2048 }
2049 else
2050 break;
2051 }
2052
2053 if (!current_templates)
2054 {
2055 /* See if we can get a match by trimming off a suffix. */
2056 switch (mnem_p[-1])
2057 {
2058 case WORD_MNEM_SUFFIX:
2059 if (intel_syntax && (intel_float_operand (mnemonic) & 2))
2060 i.suffix = SHORT_MNEM_SUFFIX;
2061 else
2062 case BYTE_MNEM_SUFFIX:
2063 case QWORD_MNEM_SUFFIX:
2064 i.suffix = mnem_p[-1];
2065 mnem_p[-1] = '\0';
2066 current_templates = hash_find (op_hash, mnemonic);
2067 break;
2068 case SHORT_MNEM_SUFFIX:
2069 case LONG_MNEM_SUFFIX:
2070 if (!intel_syntax)
2071 {
2072 i.suffix = mnem_p[-1];
2073 mnem_p[-1] = '\0';
2074 current_templates = hash_find (op_hash, mnemonic);
2075 }
2076 break;
2077
2078 /* Intel Syntax. */
2079 case 'd':
2080 if (intel_syntax)
2081 {
2082 if (intel_float_operand (mnemonic) == 1)
2083 i.suffix = SHORT_MNEM_SUFFIX;
2084 else
2085 i.suffix = LONG_MNEM_SUFFIX;
2086 mnem_p[-1] = '\0';
2087 current_templates = hash_find (op_hash, mnemonic);
2088 }
2089 break;
2090 }
2091 if (!current_templates)
2092 {
2093 as_bad (_("no such instruction: `%s'"), token_start);
2094 return NULL;
2095 }
2096 }
2097
2098 if (current_templates->start->opcode_modifier & (Jump | JumpByte))
2099 {
2100 /* Check for a branch hint. We allow ",pt" and ",pn" for
2101 predict taken and predict not taken respectively.
2102 I'm not sure that branch hints actually do anything on loop
2103 and jcxz insns (JumpByte) for current Pentium4 chips. They
2104 may work in the future and it doesn't hurt to accept them
2105 now. */
2106 if (l[0] == ',' && l[1] == 'p')
2107 {
2108 if (l[2] == 't')
2109 {
2110 if (!add_prefix (DS_PREFIX_OPCODE))
2111 return NULL;
2112 l += 3;
2113 }
2114 else if (l[2] == 'n')
2115 {
2116 if (!add_prefix (CS_PREFIX_OPCODE))
2117 return NULL;
2118 l += 3;
2119 }
2120 }
2121 }
2122 /* Any other comma loses. */
2123 if (*l == ',')
2124 {
2125 as_bad (_("invalid character %s in mnemonic"),
2126 output_invalid (*l));
2127 return NULL;
2128 }
2129
2130 /* Check if instruction is supported on specified architecture. */
2131 supported = 0;
2132 for (t = current_templates->start; t < current_templates->end; ++t)
2133 {
2134 if (!((t->cpu_flags & ~(Cpu64 | CpuNo64))
2135 & ~(cpu_arch_flags & ~(Cpu64 | CpuNo64))))
2136 supported |= 1;
2137 if (!(t->cpu_flags & (flag_code == CODE_64BIT ? CpuNo64 : Cpu64)))
2138 supported |= 2;
2139 }
2140 if (!(supported & 2))
2141 {
2142 as_bad (flag_code == CODE_64BIT
2143 ? _("`%s' is not supported in 64-bit mode")
2144 : _("`%s' is only supported in 64-bit mode"),
2145 current_templates->start->name);
2146 return NULL;
2147 }
2148 if (!(supported & 1))
2149 {
2150 as_warn (_("`%s' is not supported on `%s%s'"),
2151 current_templates->start->name,
2152 cpu_arch_name,
2153 cpu_sub_arch_name ? cpu_sub_arch_name : "");
2154 }
2155 else if ((Cpu386 & ~cpu_arch_flags) && (flag_code != CODE_16BIT))
2156 {
2157 as_warn (_("use .code16 to ensure correct addressing mode"));
2158 }
2159
2160 /* Check for rep/repne without a string instruction. */
2161 if (expecting_string_instruction)
2162 {
2163 static templates override;
2164
2165 for (t = current_templates->start; t < current_templates->end; ++t)
2166 if (t->opcode_modifier & IsString)
2167 break;
2168 if (t >= current_templates->end)
2169 {
2170 as_bad (_("expecting string instruction after `%s'"),
2171 expecting_string_instruction);
2172 return NULL;
2173 }
2174 for (override.start = t; t < current_templates->end; ++t)
2175 if (!(t->opcode_modifier & IsString))
2176 break;
2177 override.end = t;
2178 current_templates = &override;
2179 }
2180
2181 return l;
2182 }
2183
2184 static char *
2185 parse_operands (l, mnemonic)
2186 char *l;
2187 const char *mnemonic;
2188 {
2189 char *token_start;
2190
2191 /* 1 if operand is pending after ','. */
2192 unsigned int expecting_operand = 0;
2193
2194 /* Non-zero if operand parens not balanced. */
2195 unsigned int paren_not_balanced;
2196
2197 while (*l != END_OF_INSN)
2198 {
2199 /* Skip optional white space before operand. */
2200 if (is_space_char (*l))
2201 ++l;
2202 if (!is_operand_char (*l) && *l != END_OF_INSN)
2203 {
2204 as_bad (_("invalid character %s before operand %d"),
2205 output_invalid (*l),
2206 i.operands + 1);
2207 return NULL;
2208 }
2209 token_start = l; /* after white space */
2210 paren_not_balanced = 0;
2211 while (paren_not_balanced || *l != ',')
2212 {
2213 if (*l == END_OF_INSN)
2214 {
2215 if (paren_not_balanced)
2216 {
2217 if (!intel_syntax)
2218 as_bad (_("unbalanced parenthesis in operand %d."),
2219 i.operands + 1);
2220 else
2221 as_bad (_("unbalanced brackets in operand %d."),
2222 i.operands + 1);
2223 return NULL;
2224 }
2225 else
2226 break; /* we are done */
2227 }
2228 else if (!is_operand_char (*l) && !is_space_char (*l))
2229 {
2230 as_bad (_("invalid character %s in operand %d"),
2231 output_invalid (*l),
2232 i.operands + 1);
2233 return NULL;
2234 }
2235 if (!intel_syntax)
2236 {
2237 if (*l == '(')
2238 ++paren_not_balanced;
2239 if (*l == ')')
2240 --paren_not_balanced;
2241 }
2242 else
2243 {
2244 if (*l == '[')
2245 ++paren_not_balanced;
2246 if (*l == ']')
2247 --paren_not_balanced;
2248 }
2249 l++;
2250 }
2251 if (l != token_start)
2252 { /* Yes, we've read in another operand. */
2253 unsigned int operand_ok;
2254 this_operand = i.operands++;
2255 if (i.operands > MAX_OPERANDS)
2256 {
2257 as_bad (_("spurious operands; (%d operands/instruction max)"),
2258 MAX_OPERANDS);
2259 return NULL;
2260 }
2261 /* Now parse operand adding info to 'i' as we go along. */
2262 END_STRING_AND_SAVE (l);
2263
2264 if (intel_syntax)
2265 operand_ok =
2266 i386_intel_operand (token_start,
2267 intel_float_operand (mnemonic));
2268 else
2269 operand_ok = i386_operand (token_start);
2270
2271 RESTORE_END_STRING (l);
2272 if (!operand_ok)
2273 return NULL;
2274 }
2275 else
2276 {
2277 if (expecting_operand)
2278 {
2279 expecting_operand_after_comma:
2280 as_bad (_("expecting operand after ','; got nothing"));
2281 return NULL;
2282 }
2283 if (*l == ',')
2284 {
2285 as_bad (_("expecting operand before ','; got nothing"));
2286 return NULL;
2287 }
2288 }
2289
2290 /* Now *l must be either ',' or END_OF_INSN. */
2291 if (*l == ',')
2292 {
2293 if (*++l == END_OF_INSN)
2294 {
2295 /* Just skip it, if it's \n complain. */
2296 goto expecting_operand_after_comma;
2297 }
2298 expecting_operand = 1;
2299 }
2300 }
2301 return l;
2302 }
2303
2304 static void
2305 swap_imm_operands ()
2306 {
2307 union i386_op temp_op;
2308 unsigned int temp_type;
2309 enum bfd_reloc_code_real temp_reloc;
2310 int xchg1 = 0;
2311 int xchg2 = 1;
2312
2313 temp_type = i.types[xchg2];
2314 i.types[xchg2] = i.types[xchg1];
2315 i.types[xchg1] = temp_type;
2316 temp_op = i.op[xchg2];
2317 i.op[xchg2] = i.op[xchg1];
2318 i.op[xchg1] = temp_op;
2319 temp_reloc = i.reloc[xchg2];
2320 i.reloc[xchg2] = i.reloc[xchg1];
2321 i.reloc[xchg1] = temp_reloc;
2322 }
2323
2324
2325 static void
2326 swap_operands ()
2327 {
2328 union i386_op temp_op;
2329 unsigned int temp_type;
2330 enum bfd_reloc_code_real temp_reloc;
2331 int xchg1 = 0;
2332 int xchg2 = 0;
2333
2334 if (i.operands == 4)
2335 /* There will be two exchanges in a 4 operand instruction.
2336 First exchange is the done inside this block.(1st and 4rth operand)
2337 The next exchange is done outside this block.(2nd and 3rd operand) */
2338 {
2339 xchg1 = 0;
2340 xchg2 = 3;
2341 temp_type = i.types[xchg2];
2342 i.types[xchg2] = i.types[xchg1];
2343 i.types[xchg1] = temp_type;
2344 temp_op = i.op[xchg2];
2345 i.op[xchg2] = i.op[xchg1];
2346 i.op[xchg1] = temp_op;
2347 temp_reloc = i.reloc[xchg2];
2348 i.reloc[xchg2] = i.reloc[xchg1];
2349 i.reloc[xchg1] = temp_reloc;
2350 xchg1 = 1;
2351 xchg2 = 2;
2352 }
2353
2354 if (i.operands == 2)
2355 {
2356 xchg1 = 0;
2357 xchg2 = 1;
2358 }
2359 else if (i.operands == 3)
2360 {
2361 xchg1 = 0;
2362 xchg2 = 2;
2363 }
2364 temp_type = i.types[xchg2];
2365 i.types[xchg2] = i.types[xchg1];
2366 i.types[xchg1] = temp_type;
2367 temp_op = i.op[xchg2];
2368 i.op[xchg2] = i.op[xchg1];
2369 i.op[xchg1] = temp_op;
2370 temp_reloc = i.reloc[xchg2];
2371 i.reloc[xchg2] = i.reloc[xchg1];
2372 i.reloc[xchg1] = temp_reloc;
2373
2374 if (i.mem_operands == 2)
2375 {
2376 const seg_entry *temp_seg;
2377 temp_seg = i.seg[0];
2378 i.seg[0] = i.seg[1];
2379 i.seg[1] = temp_seg;
2380 }
2381 }
2382
2383 /* Try to ensure constant immediates are represented in the smallest
2384 opcode possible. */
2385 static void
2386 optimize_imm ()
2387 {
2388 char guess_suffix = 0;
2389 int op;
2390
2391 if (i.suffix)
2392 guess_suffix = i.suffix;
2393 else if (i.reg_operands)
2394 {
2395 /* Figure out a suffix from the last register operand specified.
2396 We can't do this properly yet, ie. excluding InOutPortReg,
2397 but the following works for instructions with immediates.
2398 In any case, we can't set i.suffix yet. */
2399 for (op = i.operands; --op >= 0;)
2400 if (i.types[op] & Reg)
2401 {
2402 if (i.types[op] & Reg8)
2403 guess_suffix = BYTE_MNEM_SUFFIX;
2404 else if (i.types[op] & Reg16)
2405 guess_suffix = WORD_MNEM_SUFFIX;
2406 else if (i.types[op] & Reg32)
2407 guess_suffix = LONG_MNEM_SUFFIX;
2408 else if (i.types[op] & Reg64)
2409 guess_suffix = QWORD_MNEM_SUFFIX;
2410 break;
2411 }
2412 }
2413 else if ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0))
2414 guess_suffix = WORD_MNEM_SUFFIX;
2415
2416 for (op = i.operands; --op >= 0;)
2417 if (i.types[op] & Imm)
2418 {
2419 switch (i.op[op].imms->X_op)
2420 {
2421 case O_constant:
2422 /* If a suffix is given, this operand may be shortened. */
2423 switch (guess_suffix)
2424 {
2425 case LONG_MNEM_SUFFIX:
2426 i.types[op] |= Imm32 | Imm64;
2427 break;
2428 case WORD_MNEM_SUFFIX:
2429 i.types[op] |= Imm16 | Imm32S | Imm32 | Imm64;
2430 break;
2431 case BYTE_MNEM_SUFFIX:
2432 i.types[op] |= Imm16 | Imm8 | Imm8S | Imm32S | Imm32 | Imm64;
2433 break;
2434 }
2435
2436 /* If this operand is at most 16 bits, convert it
2437 to a signed 16 bit number before trying to see
2438 whether it will fit in an even smaller size.
2439 This allows a 16-bit operand such as $0xffe0 to
2440 be recognised as within Imm8S range. */
2441 if ((i.types[op] & Imm16)
2442 && (i.op[op].imms->X_add_number & ~(offsetT) 0xffff) == 0)
2443 {
2444 i.op[op].imms->X_add_number =
2445 (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000);
2446 }
2447 if ((i.types[op] & Imm32)
2448 && ((i.op[op].imms->X_add_number & ~(((offsetT) 2 << 31) - 1))
2449 == 0))
2450 {
2451 i.op[op].imms->X_add_number = ((i.op[op].imms->X_add_number
2452 ^ ((offsetT) 1 << 31))
2453 - ((offsetT) 1 << 31));
2454 }
2455 i.types[op] |= smallest_imm_type (i.op[op].imms->X_add_number);
2456
2457 /* We must avoid matching of Imm32 templates when 64bit
2458 only immediate is available. */
2459 if (guess_suffix == QWORD_MNEM_SUFFIX)
2460 i.types[op] &= ~Imm32;
2461 break;
2462
2463 case O_absent:
2464 case O_register:
2465 abort ();
2466
2467 /* Symbols and expressions. */
2468 default:
2469 /* Convert symbolic operand to proper sizes for matching, but don't
2470 prevent matching a set of insns that only supports sizes other
2471 than those matching the insn suffix. */
2472 {
2473 unsigned int mask, allowed = 0;
2474 const template *t;
2475
2476 for (t = current_templates->start; t < current_templates->end; ++t)
2477 allowed |= t->operand_types[op];
2478 switch (guess_suffix)
2479 {
2480 case QWORD_MNEM_SUFFIX:
2481 mask = Imm64 | Imm32S;
2482 break;
2483 case LONG_MNEM_SUFFIX:
2484 mask = Imm32;
2485 break;
2486 case WORD_MNEM_SUFFIX:
2487 mask = Imm16;
2488 break;
2489 case BYTE_MNEM_SUFFIX:
2490 mask = Imm8;
2491 break;
2492 default:
2493 mask = 0;
2494 break;
2495 }
2496 if (mask & allowed)
2497 i.types[op] &= mask;
2498 }
2499 break;
2500 }
2501 }
2502 }
2503
2504 /* Try to use the smallest displacement type too. */
2505 static void
2506 optimize_disp ()
2507 {
2508 int op;
2509
2510 for (op = i.operands; --op >= 0;)
2511 if (i.types[op] & Disp)
2512 {
2513 if (i.op[op].disps->X_op == O_constant)
2514 {
2515 offsetT disp = i.op[op].disps->X_add_number;
2516
2517 if ((i.types[op] & Disp16)
2518 && (disp & ~(offsetT) 0xffff) == 0)
2519 {
2520 /* If this operand is at most 16 bits, convert
2521 to a signed 16 bit number and don't use 64bit
2522 displacement. */
2523 disp = (((disp & 0xffff) ^ 0x8000) - 0x8000);
2524 i.types[op] &= ~Disp64;
2525 }
2526 if ((i.types[op] & Disp32)
2527 && (disp & ~(((offsetT) 2 << 31) - 1)) == 0)
2528 {
2529 /* If this operand is at most 32 bits, convert
2530 to a signed 32 bit number and don't use 64bit
2531 displacement. */
2532 disp &= (((offsetT) 2 << 31) - 1);
2533 disp = (disp ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
2534 i.types[op] &= ~Disp64;
2535 }
2536 if (!disp && (i.types[op] & BaseIndex))
2537 {
2538 i.types[op] &= ~Disp;
2539 i.op[op].disps = 0;
2540 i.disp_operands--;
2541 }
2542 else if (flag_code == CODE_64BIT)
2543 {
2544 if (fits_in_signed_long (disp))
2545 {
2546 i.types[op] &= ~Disp64;
2547 i.types[op] |= Disp32S;
2548 }
2549 if (fits_in_unsigned_long (disp))
2550 i.types[op] |= Disp32;
2551 }
2552 if ((i.types[op] & (Disp32 | Disp32S | Disp16))
2553 && fits_in_signed_byte (disp))
2554 i.types[op] |= Disp8;
2555 }
2556 else if (i.reloc[op] == BFD_RELOC_386_TLS_DESC_CALL
2557 || i.reloc[op] == BFD_RELOC_X86_64_TLSDESC_CALL)
2558 {
2559 fix_new_exp (frag_now, frag_more (0) - frag_now->fr_literal, 0,
2560 i.op[op].disps, 0, i.reloc[op]);
2561 i.types[op] &= ~Disp;
2562 }
2563 else
2564 /* We only support 64bit displacement on constants. */
2565 i.types[op] &= ~Disp64;
2566 }
2567 }
2568
2569 static int
2570 match_template ()
2571 {
2572 /* Points to template once we've found it. */
2573 const template *t;
2574 unsigned int overlap0, overlap1, overlap2, overlap3;
2575 unsigned int found_reverse_match;
2576 int suffix_check;
2577 unsigned int operand_types [MAX_OPERANDS];
2578 int addr_prefix_disp;
2579 unsigned int j;
2580
2581 #if MAX_OPERANDS != 4
2582 # error "MAX_OPERANDS must be 4."
2583 #endif
2584
2585 #define MATCH(overlap, given, template) \
2586 ((overlap & ~JumpAbsolute) \
2587 && (((given) & (BaseIndex | JumpAbsolute)) \
2588 == ((overlap) & (BaseIndex | JumpAbsolute))))
2589
2590 /* If given types r0 and r1 are registers they must be of the same type
2591 unless the expected operand type register overlap is null.
2592 Note that Acc in a template matches every size of reg. */
2593 #define CONSISTENT_REGISTER_MATCH(m0, g0, t0, m1, g1, t1) \
2594 (((g0) & Reg) == 0 || ((g1) & Reg) == 0 \
2595 || ((g0) & Reg) == ((g1) & Reg) \
2596 || ((((m0) & Acc) ? Reg : (t0)) & (((m1) & Acc) ? Reg : (t1)) & Reg) == 0 )
2597
2598 overlap0 = 0;
2599 overlap1 = 0;
2600 overlap2 = 0;
2601 overlap3 = 0;
2602 found_reverse_match = 0;
2603 for (j = 0; j < MAX_OPERANDS; j++)
2604 operand_types [j] = 0;
2605 addr_prefix_disp = -1;
2606 suffix_check = (i.suffix == BYTE_MNEM_SUFFIX
2607 ? No_bSuf
2608 : (i.suffix == WORD_MNEM_SUFFIX
2609 ? No_wSuf
2610 : (i.suffix == SHORT_MNEM_SUFFIX
2611 ? No_sSuf
2612 : (i.suffix == LONG_MNEM_SUFFIX
2613 ? No_lSuf
2614 : (i.suffix == QWORD_MNEM_SUFFIX
2615 ? No_qSuf
2616 : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX
2617 ? No_xSuf : 0))))));
2618
2619 for (t = current_templates->start; t < current_templates->end; t++)
2620 {
2621 addr_prefix_disp = -1;
2622
2623 /* Must have right number of operands. */
2624 if (i.operands != t->operands)
2625 continue;
2626
2627 /* Check the suffix, except for some instructions in intel mode. */
2628 if ((t->opcode_modifier & suffix_check)
2629 && !(intel_syntax
2630 && (t->opcode_modifier & IgnoreSize)))
2631 continue;
2632
2633 for (j = 0; j < MAX_OPERANDS; j++)
2634 operand_types [j] = t->operand_types [j];
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 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 switch (t->operands)
2741 {
2742 case 4:
2743 overlap3 = i.types[3] & operand_types[3];
2744 case 3:
2745 overlap2 = i.types[2] & operand_types[2];
2746 break;
2747 }
2748
2749 switch (t->operands)
2750 {
2751 case 4:
2752 if (!MATCH (overlap3, i.types[3], operand_types[3])
2753 || !CONSISTENT_REGISTER_MATCH (overlap2,
2754 i.types[2],
2755 operand_types[2],
2756 overlap3,
2757 i.types[3],
2758 operand_types[3]))
2759 continue;
2760 case 3:
2761 /* Here we make use of the fact that there are no
2762 reverse match 3 operand instructions, and all 3
2763 operand instructions only need to be checked for
2764 register consistency between operands 2 and 3. */
2765 if (!MATCH (overlap2, i.types[2], operand_types[2])
2766 || !CONSISTENT_REGISTER_MATCH (overlap1,
2767 i.types[1],
2768 operand_types[1],
2769 overlap2,
2770 i.types[2],
2771 operand_types[2]))
2772 continue;
2773 break;
2774 }
2775 }
2776 /* Found either forward/reverse 2, 3 or 4 operand match here:
2777 slip through to break. */
2778 }
2779 if (t->cpu_flags & ~cpu_arch_flags)
2780 {
2781 found_reverse_match = 0;
2782 continue;
2783 }
2784 /* We've found a match; break out of loop. */
2785 break;
2786 }
2787
2788 if (t == current_templates->end)
2789 {
2790 /* We found no match. */
2791 as_bad (_("suffix or operands invalid for `%s'"),
2792 current_templates->start->name);
2793 return 0;
2794 }
2795
2796 if (!quiet_warnings)
2797 {
2798 if (!intel_syntax
2799 && ((i.types[0] & JumpAbsolute)
2800 != (operand_types[0] & JumpAbsolute)))
2801 {
2802 as_warn (_("indirect %s without `*'"), t->name);
2803 }
2804
2805 if ((t->opcode_modifier & (IsPrefix | IgnoreSize))
2806 == (IsPrefix | IgnoreSize))
2807 {
2808 /* Warn them that a data or address size prefix doesn't
2809 affect assembly of the next line of code. */
2810 as_warn (_("stand-alone `%s' prefix"), t->name);
2811 }
2812 }
2813
2814 /* Copy the template we found. */
2815 i.tm = *t;
2816
2817 if (addr_prefix_disp != -1)
2818 i.tm.operand_types[addr_prefix_disp]
2819 = operand_types[addr_prefix_disp];
2820
2821 if (found_reverse_match)
2822 {
2823 /* If we found a reverse match we must alter the opcode
2824 direction bit. found_reverse_match holds bits to change
2825 (different for int & float insns). */
2826
2827 i.tm.base_opcode ^= found_reverse_match;
2828
2829 i.tm.operand_types[0] = operand_types[1];
2830 i.tm.operand_types[1] = operand_types[0];
2831 }
2832
2833 return 1;
2834 }
2835
2836 static int
2837 check_string ()
2838 {
2839 int mem_op = (i.types[0] & AnyMem) ? 0 : 1;
2840 if ((i.tm.operand_types[mem_op] & EsSeg) != 0)
2841 {
2842 if (i.seg[0] != NULL && i.seg[0] != &es)
2843 {
2844 as_bad (_("`%s' operand %d must use `%%es' segment"),
2845 i.tm.name,
2846 mem_op + 1);
2847 return 0;
2848 }
2849 /* There's only ever one segment override allowed per instruction.
2850 This instruction possibly has a legal segment override on the
2851 second operand, so copy the segment to where non-string
2852 instructions store it, allowing common code. */
2853 i.seg[0] = i.seg[1];
2854 }
2855 else if ((i.tm.operand_types[mem_op + 1] & EsSeg) != 0)
2856 {
2857 if (i.seg[1] != NULL && i.seg[1] != &es)
2858 {
2859 as_bad (_("`%s' operand %d must use `%%es' segment"),
2860 i.tm.name,
2861 mem_op + 2);
2862 return 0;
2863 }
2864 }
2865 return 1;
2866 }
2867
2868 static int
2869 process_suffix (void)
2870 {
2871 /* If matched instruction specifies an explicit instruction mnemonic
2872 suffix, use it. */
2873 if (i.tm.opcode_modifier & (Size16 | Size32 | Size64))
2874 {
2875 if (i.tm.opcode_modifier & Size16)
2876 i.suffix = WORD_MNEM_SUFFIX;
2877 else if (i.tm.opcode_modifier & Size64)
2878 i.suffix = QWORD_MNEM_SUFFIX;
2879 else
2880 i.suffix = LONG_MNEM_SUFFIX;
2881 }
2882 else if (i.reg_operands)
2883 {
2884 /* If there's no instruction mnemonic suffix we try to invent one
2885 based on register operands. */
2886 if (!i.suffix)
2887 {
2888 /* We take i.suffix from the last register operand specified,
2889 Destination register type is more significant than source
2890 register type. */
2891 int op;
2892
2893 for (op = i.operands; --op >= 0;)
2894 if ((i.types[op] & Reg)
2895 && !(i.tm.operand_types[op] & InOutPortReg))
2896 {
2897 i.suffix = ((i.types[op] & Reg8) ? BYTE_MNEM_SUFFIX :
2898 (i.types[op] & Reg16) ? WORD_MNEM_SUFFIX :
2899 (i.types[op] & Reg64) ? QWORD_MNEM_SUFFIX :
2900 LONG_MNEM_SUFFIX);
2901 break;
2902 }
2903 }
2904 else if (i.suffix == BYTE_MNEM_SUFFIX)
2905 {
2906 if (!check_byte_reg ())
2907 return 0;
2908 }
2909 else if (i.suffix == LONG_MNEM_SUFFIX)
2910 {
2911 if (!check_long_reg ())
2912 return 0;
2913 }
2914 else if (i.suffix == QWORD_MNEM_SUFFIX)
2915 {
2916 if (!check_qword_reg ())
2917 return 0;
2918 }
2919 else if (i.suffix == WORD_MNEM_SUFFIX)
2920 {
2921 if (!check_word_reg ())
2922 return 0;
2923 }
2924 else if (intel_syntax && (i.tm.opcode_modifier & IgnoreSize))
2925 /* Do nothing if the instruction is going to ignore the prefix. */
2926 ;
2927 else
2928 abort ();
2929 }
2930 else if ((i.tm.opcode_modifier & DefaultSize)
2931 && !i.suffix
2932 /* exclude fldenv/frstor/fsave/fstenv */
2933 && (i.tm.opcode_modifier & No_sSuf))
2934 {
2935 i.suffix = stackop_size;
2936 }
2937 else if (intel_syntax
2938 && !i.suffix
2939 && ((i.tm.operand_types[0] & JumpAbsolute)
2940 || (i.tm.opcode_modifier & (JumpByte|JumpInterSegment))
2941 || (i.tm.base_opcode == 0x0f01 /* [ls][gi]dt */
2942 && i.tm.extension_opcode <= 3)))
2943 {
2944 switch (flag_code)
2945 {
2946 case CODE_64BIT:
2947 if (!(i.tm.opcode_modifier & No_qSuf))
2948 {
2949 i.suffix = QWORD_MNEM_SUFFIX;
2950 break;
2951 }
2952 case CODE_32BIT:
2953 if (!(i.tm.opcode_modifier & No_lSuf))
2954 i.suffix = LONG_MNEM_SUFFIX;
2955 break;
2956 case CODE_16BIT:
2957 if (!(i.tm.opcode_modifier & No_wSuf))
2958 i.suffix = WORD_MNEM_SUFFIX;
2959 break;
2960 }
2961 }
2962
2963 if (!i.suffix)
2964 {
2965 if (!intel_syntax)
2966 {
2967 if (i.tm.opcode_modifier & W)
2968 {
2969 as_bad (_("no instruction mnemonic suffix given and no register operands; can't size instruction"));
2970 return 0;
2971 }
2972 }
2973 else
2974 {
2975 unsigned int suffixes = (~i.tm.opcode_modifier
2976 & (No_bSuf
2977 | No_wSuf
2978 | No_lSuf
2979 | No_sSuf
2980 | No_xSuf
2981 | No_qSuf));
2982
2983 if ((i.tm.opcode_modifier & W)
2984 || ((suffixes & (suffixes - 1))
2985 && !(i.tm.opcode_modifier & (DefaultSize | IgnoreSize))))
2986 {
2987 as_bad (_("ambiguous operand size for `%s'"), i.tm.name);
2988 return 0;
2989 }
2990 }
2991 }
2992
2993 /* Change the opcode based on the operand size given by i.suffix;
2994 We don't need to change things for byte insns. */
2995
2996 if (i.suffix && i.suffix != BYTE_MNEM_SUFFIX)
2997 {
2998 /* It's not a byte, select word/dword operation. */
2999 if (i.tm.opcode_modifier & W)
3000 {
3001 if (i.tm.opcode_modifier & ShortForm)
3002 i.tm.base_opcode |= 8;
3003 else
3004 i.tm.base_opcode |= 1;
3005 }
3006
3007 /* Now select between word & dword operations via the operand
3008 size prefix, except for instructions that will ignore this
3009 prefix anyway. */
3010 if (i.tm.base_opcode == 0x0f01 && i.tm.extension_opcode == 0xc8)
3011 {
3012 /* monitor in SSE3 is a very special case. The default size
3013 of AX is the size of mode. The address size override
3014 prefix will change the size of AX. */
3015 if (i.op->regs[0].reg_type &
3016 (flag_code == CODE_32BIT ? Reg16 : Reg32))
3017 if (!add_prefix (ADDR_PREFIX_OPCODE))
3018 return 0;
3019 }
3020 else if (i.suffix != QWORD_MNEM_SUFFIX
3021 && i.suffix != LONG_DOUBLE_MNEM_SUFFIX
3022 && !(i.tm.opcode_modifier & (IgnoreSize | FloatMF))
3023 && ((i.suffix == LONG_MNEM_SUFFIX) == (flag_code == CODE_16BIT)
3024 || (flag_code == CODE_64BIT
3025 && (i.tm.opcode_modifier & JumpByte))))
3026 {
3027 unsigned int prefix = DATA_PREFIX_OPCODE;
3028
3029 if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */
3030 prefix = ADDR_PREFIX_OPCODE;
3031
3032 if (!add_prefix (prefix))
3033 return 0;
3034 }
3035
3036 /* Set mode64 for an operand. */
3037 if (i.suffix == QWORD_MNEM_SUFFIX
3038 && flag_code == CODE_64BIT
3039 && (i.tm.opcode_modifier & NoRex64) == 0)
3040 {
3041 /* Special case for xchg %rax,%rax. It is NOP and doesn't
3042 need rex64. */
3043 if (i.operands != 2
3044 || i.types [0] != (Acc | Reg64)
3045 || i.types [1] != (Acc | Reg64)
3046 || strcmp (i.tm.name, "xchg") != 0)
3047 i.rex |= REX_MODE64;
3048 }
3049
3050 /* Size floating point instruction. */
3051 if (i.suffix == LONG_MNEM_SUFFIX)
3052 if (i.tm.opcode_modifier & FloatMF)
3053 i.tm.base_opcode ^= 4;
3054 }
3055
3056 return 1;
3057 }
3058
3059 static int
3060 check_byte_reg (void)
3061 {
3062 int op;
3063
3064 for (op = i.operands; --op >= 0;)
3065 {
3066 /* If this is an eight bit register, it's OK. If it's the 16 or
3067 32 bit version of an eight bit register, we will just use the
3068 low portion, and that's OK too. */
3069 if (i.types[op] & Reg8)
3070 continue;
3071
3072 /* movzx and movsx should not generate this warning. */
3073 if (intel_syntax
3074 && (i.tm.base_opcode == 0xfb7
3075 || i.tm.base_opcode == 0xfb6
3076 || i.tm.base_opcode == 0x63
3077 || i.tm.base_opcode == 0xfbe
3078 || i.tm.base_opcode == 0xfbf))
3079 continue;
3080
3081 if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4)
3082 {
3083 /* Prohibit these changes in the 64bit mode, since the
3084 lowering is more complicated. */
3085 if (flag_code == CODE_64BIT
3086 && (i.tm.operand_types[op] & InOutPortReg) == 0)
3087 {
3088 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
3089 i.op[op].regs->reg_name,
3090 i.suffix);
3091 return 0;
3092 }
3093 #if REGISTER_WARNINGS
3094 if (!quiet_warnings
3095 && (i.tm.operand_types[op] & InOutPortReg) == 0)
3096 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
3097 (i.op[op].regs + (i.types[op] & Reg16
3098 ? REGNAM_AL - REGNAM_AX
3099 : REGNAM_AL - REGNAM_EAX))->reg_name,
3100 i.op[op].regs->reg_name,
3101 i.suffix);
3102 #endif
3103 continue;
3104 }
3105 /* Any other register is bad. */
3106 if (i.types[op] & (Reg | RegMMX | RegXMM
3107 | SReg2 | SReg3
3108 | Control | Debug | Test
3109 | FloatReg | FloatAcc))
3110 {
3111 as_bad (_("`%%%s' not allowed with `%s%c'"),
3112 i.op[op].regs->reg_name,
3113 i.tm.name,
3114 i.suffix);
3115 return 0;
3116 }
3117 }
3118 return 1;
3119 }
3120
3121 static int
3122 check_long_reg ()
3123 {
3124 int op;
3125
3126 for (op = i.operands; --op >= 0;)
3127 /* Reject eight bit registers, except where the template requires
3128 them. (eg. movzb) */
3129 if ((i.types[op] & Reg8) != 0
3130 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
3131 {
3132 as_bad (_("`%%%s' not allowed with `%s%c'"),
3133 i.op[op].regs->reg_name,
3134 i.tm.name,
3135 i.suffix);
3136 return 0;
3137 }
3138 /* Warn if the e prefix on a general reg is missing. */
3139 else if ((!quiet_warnings || flag_code == CODE_64BIT)
3140 && (i.types[op] & Reg16) != 0
3141 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
3142 {
3143 /* Prohibit these changes in the 64bit mode, since the
3144 lowering is more complicated. */
3145 if (flag_code == CODE_64BIT)
3146 {
3147 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
3148 i.op[op].regs->reg_name,
3149 i.suffix);
3150 return 0;
3151 }
3152 #if REGISTER_WARNINGS
3153 else
3154 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
3155 (i.op[op].regs + REGNAM_EAX - REGNAM_AX)->reg_name,
3156 i.op[op].regs->reg_name,
3157 i.suffix);
3158 #endif
3159 }
3160 /* Warn if the r prefix on a general reg is missing. */
3161 else if ((i.types[op] & Reg64) != 0
3162 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
3163 {
3164 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
3165 i.op[op].regs->reg_name,
3166 i.suffix);
3167 return 0;
3168 }
3169 return 1;
3170 }
3171
3172 static int
3173 check_qword_reg ()
3174 {
3175 int op;
3176
3177 for (op = i.operands; --op >= 0; )
3178 /* Reject eight bit registers, except where the template requires
3179 them. (eg. movzb) */
3180 if ((i.types[op] & Reg8) != 0
3181 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
3182 {
3183 as_bad (_("`%%%s' not allowed with `%s%c'"),
3184 i.op[op].regs->reg_name,
3185 i.tm.name,
3186 i.suffix);
3187 return 0;
3188 }
3189 /* Warn if the e prefix on a general reg is missing. */
3190 else if (((i.types[op] & Reg16) != 0
3191 || (i.types[op] & Reg32) != 0)
3192 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
3193 {
3194 /* Prohibit these changes in the 64bit mode, since the
3195 lowering is more complicated. */
3196 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
3197 i.op[op].regs->reg_name,
3198 i.suffix);
3199 return 0;
3200 }
3201 return 1;
3202 }
3203
3204 static int
3205 check_word_reg ()
3206 {
3207 int op;
3208 for (op = i.operands; --op >= 0;)
3209 /* Reject eight bit registers, except where the template requires
3210 them. (eg. movzb) */
3211 if ((i.types[op] & Reg8) != 0
3212 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
3213 {
3214 as_bad (_("`%%%s' not allowed with `%s%c'"),
3215 i.op[op].regs->reg_name,
3216 i.tm.name,
3217 i.suffix);
3218 return 0;
3219 }
3220 /* Warn if the e prefix on a general reg is present. */
3221 else if ((!quiet_warnings || flag_code == CODE_64BIT)
3222 && (i.types[op] & Reg32) != 0
3223 && (i.tm.operand_types[op] & (Reg16 | Acc)) != 0)
3224 {
3225 /* Prohibit these changes in the 64bit mode, since the
3226 lowering is more complicated. */
3227 if (flag_code == CODE_64BIT)
3228 {
3229 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
3230 i.op[op].regs->reg_name,
3231 i.suffix);
3232 return 0;
3233 }
3234 else
3235 #if REGISTER_WARNINGS
3236 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
3237 (i.op[op].regs + REGNAM_AX - REGNAM_EAX)->reg_name,
3238 i.op[op].regs->reg_name,
3239 i.suffix);
3240 #endif
3241 }
3242 return 1;
3243 }
3244
3245 static int
3246 finalize_imm ()
3247 {
3248 unsigned int overlap0, overlap1, overlap2;
3249
3250 overlap0 = i.types[0] & i.tm.operand_types[0];
3251 if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64))
3252 && overlap0 != Imm8 && overlap0 != Imm8S
3253 && overlap0 != Imm16 && overlap0 != Imm32S
3254 && overlap0 != Imm32 && overlap0 != Imm64)
3255 {
3256 if (i.suffix)
3257 {
3258 overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX
3259 ? Imm8 | Imm8S
3260 : (i.suffix == WORD_MNEM_SUFFIX
3261 ? Imm16
3262 : (i.suffix == QWORD_MNEM_SUFFIX
3263 ? Imm64 | Imm32S
3264 : Imm32)));
3265 }
3266 else if (overlap0 == (Imm16 | Imm32S | Imm32)
3267 || overlap0 == (Imm16 | Imm32)
3268 || overlap0 == (Imm16 | Imm32S))
3269 {
3270 overlap0 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
3271 ? Imm16 : Imm32S);
3272 }
3273 if (overlap0 != Imm8 && overlap0 != Imm8S
3274 && overlap0 != Imm16 && overlap0 != Imm32S
3275 && overlap0 != Imm32 && overlap0 != Imm64)
3276 {
3277 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size"));
3278 return 0;
3279 }
3280 }
3281 i.types[0] = overlap0;
3282
3283 overlap1 = i.types[1] & i.tm.operand_types[1];
3284 if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32 | Imm64))
3285 && overlap1 != Imm8 && overlap1 != Imm8S
3286 && overlap1 != Imm16 && overlap1 != Imm32S
3287 && overlap1 != Imm32 && overlap1 != Imm64)
3288 {
3289 if (i.suffix)
3290 {
3291 overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX
3292 ? Imm8 | Imm8S
3293 : (i.suffix == WORD_MNEM_SUFFIX
3294 ? Imm16
3295 : (i.suffix == QWORD_MNEM_SUFFIX
3296 ? Imm64 | Imm32S
3297 : Imm32)));
3298 }
3299 else if (overlap1 == (Imm16 | Imm32 | Imm32S)
3300 || overlap1 == (Imm16 | Imm32)
3301 || overlap1 == (Imm16 | Imm32S))
3302 {
3303 overlap1 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
3304 ? Imm16 : Imm32S);
3305 }
3306 if (overlap1 != Imm8 && overlap1 != Imm8S
3307 && overlap1 != Imm16 && overlap1 != Imm32S
3308 && overlap1 != Imm32 && overlap1 != Imm64)
3309 {
3310 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size %x %c"),overlap1, i.suffix);
3311 return 0;
3312 }
3313 }
3314 i.types[1] = overlap1;
3315
3316 overlap2 = i.types[2] & i.tm.operand_types[2];
3317 assert ((overlap2 & Imm) == 0);
3318 i.types[2] = overlap2;
3319
3320 return 1;
3321 }
3322
3323 static int
3324 process_operands ()
3325 {
3326 /* Default segment register this instruction will use for memory
3327 accesses. 0 means unknown. This is only for optimizing out
3328 unnecessary segment overrides. */
3329 const seg_entry *default_seg = 0;
3330
3331 /* The imul $imm, %reg instruction is converted into
3332 imul $imm, %reg, %reg, and the clr %reg instruction
3333 is converted into xor %reg, %reg. */
3334 if (i.tm.opcode_modifier & regKludge)
3335 {
3336 unsigned int first_reg_op = (i.types[0] & Reg) ? 0 : 1;
3337 /* Pretend we saw the extra register operand. */
3338 assert (i.reg_operands == 1
3339 && i.op[first_reg_op + 1].regs == 0);
3340 i.op[first_reg_op + 1].regs = i.op[first_reg_op].regs;
3341 i.types[first_reg_op + 1] = i.types[first_reg_op];
3342 i.operands++;
3343 i.reg_operands++;
3344 }
3345
3346 if (i.tm.opcode_modifier & ShortForm)
3347 {
3348 /* The register or float register operand is in operand 0 or 1. */
3349 unsigned int op = (i.types[0] & (Reg | FloatReg)) ? 0 : 1;
3350 /* Register goes in low 3 bits of opcode. */
3351 i.tm.base_opcode |= i.op[op].regs->reg_num;
3352 if ((i.op[op].regs->reg_flags & RegRex) != 0)
3353 i.rex |= REX_EXTZ;
3354 if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
3355 {
3356 /* Warn about some common errors, but press on regardless.
3357 The first case can be generated by gcc (<= 2.8.1). */
3358 if (i.operands == 2)
3359 {
3360 /* Reversed arguments on faddp, fsubp, etc. */
3361 as_warn (_("translating to `%s %%%s,%%%s'"), i.tm.name,
3362 i.op[1].regs->reg_name,
3363 i.op[0].regs->reg_name);
3364 }
3365 else
3366 {
3367 /* Extraneous `l' suffix on fp insn. */
3368 as_warn (_("translating to `%s %%%s'"), i.tm.name,
3369 i.op[0].regs->reg_name);
3370 }
3371 }
3372 }
3373 else if (i.tm.opcode_modifier & Modrm)
3374 {
3375 /* The opcode is completed (modulo i.tm.extension_opcode which
3376 must be put into the modrm byte). Now, we make the modrm and
3377 index base bytes based on all the info we've collected. */
3378
3379 default_seg = build_modrm_byte ();
3380 }
3381 else if (i.tm.opcode_modifier & (Seg2ShortForm | Seg3ShortForm))
3382 {
3383 if (i.tm.base_opcode == POP_SEG_SHORT
3384 && i.op[0].regs->reg_num == 1)
3385 {
3386 as_bad (_("you can't `pop %%cs'"));
3387 return 0;
3388 }
3389 i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
3390 if ((i.op[0].regs->reg_flags & RegRex) != 0)
3391 i.rex |= REX_EXTZ;
3392 }
3393 else if ((i.tm.base_opcode & ~(D | W)) == MOV_AX_DISP32)
3394 {
3395 default_seg = &ds;
3396 }
3397 else if ((i.tm.opcode_modifier & IsString) != 0)
3398 {
3399 /* For the string instructions that allow a segment override
3400 on one of their operands, the default segment is ds. */
3401 default_seg = &ds;
3402 }
3403
3404 if ((i.tm.base_opcode == 0x8d /* lea */
3405 || (i.tm.cpu_flags & CpuSVME))
3406 && i.seg[0] && !quiet_warnings)
3407 as_warn (_("segment override on `%s' is ineffectual"), i.tm.name);
3408
3409 /* If a segment was explicitly specified, and the specified segment
3410 is not the default, use an opcode prefix to select it. If we
3411 never figured out what the default segment is, then default_seg
3412 will be zero at this point, and the specified segment prefix will
3413 always be used. */
3414 if ((i.seg[0]) && (i.seg[0] != default_seg))
3415 {
3416 if (!add_prefix (i.seg[0]->seg_prefix))
3417 return 0;
3418 }
3419 return 1;
3420 }
3421
3422 static const seg_entry *
3423 build_modrm_byte ()
3424 {
3425 const seg_entry *default_seg = 0;
3426
3427 /* i.reg_operands MUST be the number of real register operands;
3428 implicit registers do not count. */
3429 if (i.reg_operands == 2)
3430 {
3431 unsigned int source, dest;
3432
3433 switch (i.operands)
3434 {
3435 case 2:
3436 source = 0;
3437 break;
3438 case 3:
3439 /* When there are 3 operands, one of them must be immediate,
3440 which may be the first or the last operand. */
3441 assert (i.imm_operands == 1);
3442 source = (i.types[0] & Imm) ? 1 : 0;
3443 break;
3444 case 4:
3445 /* When there are 4 operands, the first two must be immediate
3446 operands. The source operand will be the 3rd one. */
3447 assert (i.imm_operands == 2
3448 && (i.types[0] & Imm)
3449 && (i.types[1] & Imm));
3450 source = 2;
3451 break;
3452 default:
3453 abort ();
3454 }
3455
3456 dest = source + 1;
3457
3458 i.rm.mode = 3;
3459 /* One of the register operands will be encoded in the i.tm.reg
3460 field, the other in the combined i.tm.mode and i.tm.regmem
3461 fields. If no form of this instruction supports a memory
3462 destination operand, then we assume the source operand may
3463 sometimes be a memory operand and so we need to store the
3464 destination in the i.rm.reg field. */
3465 if ((i.tm.operand_types[dest] & AnyMem) == 0)
3466 {
3467 i.rm.reg = i.op[dest].regs->reg_num;
3468 i.rm.regmem = i.op[source].regs->reg_num;
3469 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
3470 i.rex |= REX_EXTX;
3471 if ((i.op[source].regs->reg_flags & RegRex) != 0)
3472 i.rex |= REX_EXTZ;
3473 }
3474 else
3475 {
3476 i.rm.reg = i.op[source].regs->reg_num;
3477 i.rm.regmem = i.op[dest].regs->reg_num;
3478 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
3479 i.rex |= REX_EXTZ;
3480 if ((i.op[source].regs->reg_flags & RegRex) != 0)
3481 i.rex |= REX_EXTX;
3482 }
3483 if (flag_code != CODE_64BIT && (i.rex & (REX_EXTX | REX_EXTZ)))
3484 {
3485 if (!((i.types[0] | i.types[1]) & Control))
3486 abort ();
3487 i.rex &= ~(REX_EXTX | REX_EXTZ);
3488 add_prefix (LOCK_PREFIX_OPCODE);
3489 }
3490 }
3491 else
3492 { /* If it's not 2 reg operands... */
3493 if (i.mem_operands)
3494 {
3495 unsigned int fake_zero_displacement = 0;
3496 unsigned int op = ((i.types[0] & AnyMem)
3497 ? 0
3498 : (i.types[1] & AnyMem) ? 1 : 2);
3499
3500 default_seg = &ds;
3501
3502 if (i.base_reg == 0)
3503 {
3504 i.rm.mode = 0;
3505 if (!i.disp_operands)
3506 fake_zero_displacement = 1;
3507 if (i.index_reg == 0)
3508 {
3509 /* Operand is just <disp> */
3510 if (flag_code == CODE_64BIT)
3511 {
3512 /* 64bit mode overwrites the 32bit absolute
3513 addressing by RIP relative addressing and
3514 absolute addressing is encoded by one of the
3515 redundant SIB forms. */
3516 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3517 i.sib.base = NO_BASE_REGISTER;
3518 i.sib.index = NO_INDEX_REGISTER;
3519 i.types[op] = ((i.prefix[ADDR_PREFIX] == 0)
3520 ? Disp32S : Disp32);
3521 }
3522 else if ((flag_code == CODE_16BIT)
3523 ^ (i.prefix[ADDR_PREFIX] != 0))
3524 {
3525 i.rm.regmem = NO_BASE_REGISTER_16;
3526 i.types[op] = Disp16;
3527 }
3528 else
3529 {
3530 i.rm.regmem = NO_BASE_REGISTER;
3531 i.types[op] = Disp32;
3532 }
3533 }
3534 else /* !i.base_reg && i.index_reg */
3535 {
3536 i.sib.index = i.index_reg->reg_num;
3537 i.sib.base = NO_BASE_REGISTER;
3538 i.sib.scale = i.log2_scale_factor;
3539 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3540 i.types[op] &= ~Disp;
3541 if (flag_code != CODE_64BIT)
3542 i.types[op] |= Disp32; /* Must be 32 bit */
3543 else
3544 i.types[op] |= Disp32S;
3545 if ((i.index_reg->reg_flags & RegRex) != 0)
3546 i.rex |= REX_EXTY;
3547 }
3548 }
3549 /* RIP addressing for 64bit mode. */
3550 else if (i.base_reg->reg_type == BaseIndex)
3551 {
3552 i.rm.regmem = NO_BASE_REGISTER;
3553 i.types[op] &= ~ Disp;
3554 i.types[op] |= Disp32S;
3555 i.flags[op] |= Operand_PCrel;
3556 if (! i.disp_operands)
3557 fake_zero_displacement = 1;
3558 }
3559 else if (i.base_reg->reg_type & Reg16)
3560 {
3561 switch (i.base_reg->reg_num)
3562 {
3563 case 3: /* (%bx) */
3564 if (i.index_reg == 0)
3565 i.rm.regmem = 7;
3566 else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */
3567 i.rm.regmem = i.index_reg->reg_num - 6;
3568 break;
3569 case 5: /* (%bp) */
3570 default_seg = &ss;
3571 if (i.index_reg == 0)
3572 {
3573 i.rm.regmem = 6;
3574 if ((i.types[op] & Disp) == 0)
3575 {
3576 /* fake (%bp) into 0(%bp) */
3577 i.types[op] |= Disp8;
3578 fake_zero_displacement = 1;
3579 }
3580 }
3581 else /* (%bp,%si) -> 2, or (%bp,%di) -> 3 */
3582 i.rm.regmem = i.index_reg->reg_num - 6 + 2;
3583 break;
3584 default: /* (%si) -> 4 or (%di) -> 5 */
3585 i.rm.regmem = i.base_reg->reg_num - 6 + 4;
3586 }
3587 i.rm.mode = mode_from_disp_size (i.types[op]);
3588 }
3589 else /* i.base_reg and 32/64 bit mode */
3590 {
3591 if (flag_code == CODE_64BIT
3592 && (i.types[op] & Disp))
3593 i.types[op] = ((i.types[op] & Disp8)
3594 | (i.prefix[ADDR_PREFIX] == 0
3595 ? Disp32S : Disp32));
3596
3597 i.rm.regmem = i.base_reg->reg_num;
3598 if ((i.base_reg->reg_flags & RegRex) != 0)
3599 i.rex |= REX_EXTZ;
3600 i.sib.base = i.base_reg->reg_num;
3601 /* x86-64 ignores REX prefix bit here to avoid decoder
3602 complications. */
3603 if ((i.base_reg->reg_num & 7) == EBP_REG_NUM)
3604 {
3605 default_seg = &ss;
3606 if (i.disp_operands == 0)
3607 {
3608 fake_zero_displacement = 1;
3609 i.types[op] |= Disp8;
3610 }
3611 }
3612 else if (i.base_reg->reg_num == ESP_REG_NUM)
3613 {
3614 default_seg = &ss;
3615 }
3616 i.sib.scale = i.log2_scale_factor;
3617 if (i.index_reg == 0)
3618 {
3619 /* <disp>(%esp) becomes two byte modrm with no index
3620 register. We've already stored the code for esp
3621 in i.rm.regmem ie. ESCAPE_TO_TWO_BYTE_ADDRESSING.
3622 Any base register besides %esp will not use the
3623 extra modrm byte. */
3624 i.sib.index = NO_INDEX_REGISTER;
3625 #if !SCALE1_WHEN_NO_INDEX
3626 /* Another case where we force the second modrm byte. */
3627 if (i.log2_scale_factor)
3628 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3629 #endif
3630 }
3631 else
3632 {
3633 i.sib.index = i.index_reg->reg_num;
3634 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3635 if ((i.index_reg->reg_flags & RegRex) != 0)
3636 i.rex |= REX_EXTY;
3637 }
3638
3639 if (i.disp_operands
3640 && (i.reloc[op] == BFD_RELOC_386_TLS_DESC_CALL
3641 || i.reloc[op] == BFD_RELOC_X86_64_TLSDESC_CALL))
3642 i.rm.mode = 0;
3643 else
3644 i.rm.mode = mode_from_disp_size (i.types[op]);
3645 }
3646
3647 if (fake_zero_displacement)
3648 {
3649 /* Fakes a zero displacement assuming that i.types[op]
3650 holds the correct displacement size. */
3651 expressionS *exp;
3652
3653 assert (i.op[op].disps == 0);
3654 exp = &disp_expressions[i.disp_operands++];
3655 i.op[op].disps = exp;
3656 exp->X_op = O_constant;
3657 exp->X_add_number = 0;
3658 exp->X_add_symbol = (symbolS *) 0;
3659 exp->X_op_symbol = (symbolS *) 0;
3660 }
3661 }
3662
3663 /* Fill in i.rm.reg or i.rm.regmem field with register operand
3664 (if any) based on i.tm.extension_opcode. Again, we must be
3665 careful to make sure that segment/control/debug/test/MMX
3666 registers are coded into the i.rm.reg field. */
3667 if (i.reg_operands)
3668 {
3669 unsigned int op =
3670 ((i.types[0]
3671 & (Reg | RegMMX | RegXMM
3672 | SReg2 | SReg3
3673 | Control | Debug | Test))
3674 ? 0
3675 : ((i.types[1]
3676 & (Reg | RegMMX | RegXMM
3677 | SReg2 | SReg3
3678 | Control | Debug | Test))
3679 ? 1
3680 : 2));
3681 /* If there is an extension opcode to put here, the register
3682 number must be put into the regmem field. */
3683 if (i.tm.extension_opcode != None)
3684 {
3685 i.rm.regmem = i.op[op].regs->reg_num;
3686 if ((i.op[op].regs->reg_flags & RegRex) != 0)
3687 i.rex |= REX_EXTZ;
3688 }
3689 else
3690 {
3691 i.rm.reg = i.op[op].regs->reg_num;
3692 if ((i.op[op].regs->reg_flags & RegRex) != 0)
3693 i.rex |= REX_EXTX;
3694 }
3695
3696 /* Now, if no memory operand has set i.rm.mode = 0, 1, 2 we
3697 must set it to 3 to indicate this is a register operand
3698 in the regmem field. */
3699 if (!i.mem_operands)
3700 i.rm.mode = 3;
3701 }
3702
3703 /* Fill in i.rm.reg field with extension opcode (if any). */
3704 if (i.tm.extension_opcode != None)
3705 i.rm.reg = i.tm.extension_opcode;
3706 }
3707 return default_seg;
3708 }
3709
3710 static void
3711 output_branch ()
3712 {
3713 char *p;
3714 int code16;
3715 int prefix;
3716 relax_substateT subtype;
3717 symbolS *sym;
3718 offsetT off;
3719
3720 code16 = 0;
3721 if (flag_code == CODE_16BIT)
3722 code16 = CODE16;
3723
3724 prefix = 0;
3725 if (i.prefix[DATA_PREFIX] != 0)
3726 {
3727 prefix = 1;
3728 i.prefixes -= 1;
3729 code16 ^= CODE16;
3730 }
3731 /* Pentium4 branch hints. */
3732 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
3733 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
3734 {
3735 prefix++;
3736 i.prefixes--;
3737 }
3738 if (i.prefix[REX_PREFIX] != 0)
3739 {
3740 prefix++;
3741 i.prefixes--;
3742 }
3743
3744 if (i.prefixes != 0 && !intel_syntax)
3745 as_warn (_("skipping prefixes on this instruction"));
3746
3747 /* It's always a symbol; End frag & setup for relax.
3748 Make sure there is enough room in this frag for the largest
3749 instruction we may generate in md_convert_frag. This is 2
3750 bytes for the opcode and room for the prefix and largest
3751 displacement. */
3752 frag_grow (prefix + 2 + 4);
3753 /* Prefix and 1 opcode byte go in fr_fix. */
3754 p = frag_more (prefix + 1);
3755 if (i.prefix[DATA_PREFIX] != 0)
3756 *p++ = DATA_PREFIX_OPCODE;
3757 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE
3758 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE)
3759 *p++ = i.prefix[SEG_PREFIX];
3760 if (i.prefix[REX_PREFIX] != 0)
3761 *p++ = i.prefix[REX_PREFIX];
3762 *p = i.tm.base_opcode;
3763
3764 if ((unsigned char) *p == JUMP_PC_RELATIVE)
3765 subtype = ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL);
3766 else if ((cpu_arch_flags & Cpu386) != 0)
3767 subtype = ENCODE_RELAX_STATE (COND_JUMP, SMALL);
3768 else
3769 subtype = ENCODE_RELAX_STATE (COND_JUMP86, SMALL);
3770 subtype |= code16;
3771
3772 sym = i.op[0].disps->X_add_symbol;
3773 off = i.op[0].disps->X_add_number;
3774
3775 if (i.op[0].disps->X_op != O_constant
3776 && i.op[0].disps->X_op != O_symbol)
3777 {
3778 /* Handle complex expressions. */
3779 sym = make_expr_symbol (i.op[0].disps);
3780 off = 0;
3781 }
3782
3783 /* 1 possible extra opcode + 4 byte displacement go in var part.
3784 Pass reloc in fr_var. */
3785 frag_var (rs_machine_dependent, 5, i.reloc[0], subtype, sym, off, p);
3786 }
3787
3788 static void
3789 output_jump ()
3790 {
3791 char *p;
3792 int size;
3793 fixS *fixP;
3794
3795 if (i.tm.opcode_modifier & JumpByte)
3796 {
3797 /* This is a loop or jecxz type instruction. */
3798 size = 1;
3799 if (i.prefix[ADDR_PREFIX] != 0)
3800 {
3801 FRAG_APPEND_1_CHAR (ADDR_PREFIX_OPCODE);
3802 i.prefixes -= 1;
3803 }
3804 /* Pentium4 branch hints. */
3805 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
3806 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
3807 {
3808 FRAG_APPEND_1_CHAR (i.prefix[SEG_PREFIX]);
3809 i.prefixes--;
3810 }
3811 }
3812 else
3813 {
3814 int code16;
3815
3816 code16 = 0;
3817 if (flag_code == CODE_16BIT)
3818 code16 = CODE16;
3819
3820 if (i.prefix[DATA_PREFIX] != 0)
3821 {
3822 FRAG_APPEND_1_CHAR (DATA_PREFIX_OPCODE);
3823 i.prefixes -= 1;
3824 code16 ^= CODE16;
3825 }
3826
3827 size = 4;
3828 if (code16)
3829 size = 2;
3830 }
3831
3832 if (i.prefix[REX_PREFIX] != 0)
3833 {
3834 FRAG_APPEND_1_CHAR (i.prefix[REX_PREFIX]);
3835 i.prefixes -= 1;
3836 }
3837
3838 if (i.prefixes != 0 && !intel_syntax)
3839 as_warn (_("skipping prefixes on this instruction"));
3840
3841 p = frag_more (1 + size);
3842 *p++ = i.tm.base_opcode;
3843
3844 fixP = fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3845 i.op[0].disps, 1, reloc (size, 1, 1, i.reloc[0]));
3846
3847 /* All jumps handled here are signed, but don't use a signed limit
3848 check for 32 and 16 bit jumps as we want to allow wrap around at
3849 4G and 64k respectively. */
3850 if (size == 1)
3851 fixP->fx_signed = 1;
3852 }
3853
3854 static void
3855 output_interseg_jump ()
3856 {
3857 char *p;
3858 int size;
3859 int prefix;
3860 int code16;
3861
3862 code16 = 0;
3863 if (flag_code == CODE_16BIT)
3864 code16 = CODE16;
3865
3866 prefix = 0;
3867 if (i.prefix[DATA_PREFIX] != 0)
3868 {
3869 prefix = 1;
3870 i.prefixes -= 1;
3871 code16 ^= CODE16;
3872 }
3873 if (i.prefix[REX_PREFIX] != 0)
3874 {
3875 prefix++;
3876 i.prefixes -= 1;
3877 }
3878
3879 size = 4;
3880 if (code16)
3881 size = 2;
3882
3883 if (i.prefixes != 0 && !intel_syntax)
3884 as_warn (_("skipping prefixes on this instruction"));
3885
3886 /* 1 opcode; 2 segment; offset */
3887 p = frag_more (prefix + 1 + 2 + size);
3888
3889 if (i.prefix[DATA_PREFIX] != 0)
3890 *p++ = DATA_PREFIX_OPCODE;
3891
3892 if (i.prefix[REX_PREFIX] != 0)
3893 *p++ = i.prefix[REX_PREFIX];
3894
3895 *p++ = i.tm.base_opcode;
3896 if (i.op[1].imms->X_op == O_constant)
3897 {
3898 offsetT n = i.op[1].imms->X_add_number;
3899
3900 if (size == 2
3901 && !fits_in_unsigned_word (n)
3902 && !fits_in_signed_word (n))
3903 {
3904 as_bad (_("16-bit jump out of range"));
3905 return;
3906 }
3907 md_number_to_chars (p, n, size);
3908 }
3909 else
3910 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3911 i.op[1].imms, 0, reloc (size, 0, 0, i.reloc[1]));
3912 if (i.op[0].imms->X_op != O_constant)
3913 as_bad (_("can't handle non absolute segment in `%s'"),
3914 i.tm.name);
3915 md_number_to_chars (p + size, (valueT) i.op[0].imms->X_add_number, 2);
3916 }
3917
3918 static void
3919 output_insn ()
3920 {
3921 fragS *insn_start_frag;
3922 offsetT insn_start_off;
3923
3924 /* Tie dwarf2 debug info to the address at the start of the insn.
3925 We can't do this after the insn has been output as the current
3926 frag may have been closed off. eg. by frag_var. */
3927 dwarf2_emit_insn (0);
3928
3929 insn_start_frag = frag_now;
3930 insn_start_off = frag_now_fix ();
3931
3932 /* Output jumps. */
3933 if (i.tm.opcode_modifier & Jump)
3934 output_branch ();
3935 else if (i.tm.opcode_modifier & (JumpByte | JumpDword))
3936 output_jump ();
3937 else if (i.tm.opcode_modifier & JumpInterSegment)
3938 output_interseg_jump ();
3939 else
3940 {
3941 /* Output normal instructions here. */
3942 char *p;
3943 unsigned char *q;
3944 unsigned int prefix;
3945
3946 /* All opcodes on i386 have either 1 or 2 bytes. Supplemental
3947 Streaming SIMD extensions 3 Instructions have 3 bytes. We may
3948 use one more higher byte to specify a prefix the instruction
3949 requires. */
3950 if ((i.tm.cpu_flags & CpuSSSE3) != 0)
3951 {
3952 if (i.tm.base_opcode & 0xff000000)
3953 {
3954 prefix = (i.tm.base_opcode >> 24) & 0xff;
3955 goto check_prefix;
3956 }
3957 }
3958 else if ((i.tm.base_opcode & 0xff0000) != 0)
3959 {
3960 prefix = (i.tm.base_opcode >> 16) & 0xff;
3961 if ((i.tm.cpu_flags & CpuPadLock) != 0)
3962 {
3963 check_prefix:
3964 if (prefix != REPE_PREFIX_OPCODE
3965 || i.prefix[LOCKREP_PREFIX] != REPE_PREFIX_OPCODE)
3966 add_prefix (prefix);
3967 }
3968 else
3969 add_prefix (prefix);
3970 }
3971
3972 /* The prefix bytes. */
3973 for (q = i.prefix;
3974 q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
3975 q++)
3976 {
3977 if (*q)
3978 {
3979 p = frag_more (1);
3980 md_number_to_chars (p, (valueT) *q, 1);
3981 }
3982 }
3983
3984 /* Now the opcode; be careful about word order here! */
3985 if (fits_in_unsigned_byte (i.tm.base_opcode))
3986 {
3987 FRAG_APPEND_1_CHAR (i.tm.base_opcode);
3988 }
3989 else
3990 {
3991 if ((i.tm.cpu_flags & CpuSSSE3) != 0)
3992 {
3993 p = frag_more (3);
3994 *p++ = (i.tm.base_opcode >> 16) & 0xff;
3995 }
3996 else
3997 p = frag_more (2);
3998
3999 /* Put out high byte first: can't use md_number_to_chars! */
4000 *p++ = (i.tm.base_opcode >> 8) & 0xff;
4001 *p = i.tm.base_opcode & 0xff;
4002 }
4003
4004 /* Now the modrm byte and sib byte (if present). */
4005 if (i.tm.opcode_modifier & Modrm)
4006 {
4007 p = frag_more (1);
4008 md_number_to_chars (p,
4009 (valueT) (i.rm.regmem << 0
4010 | i.rm.reg << 3
4011 | i.rm.mode << 6),
4012 1);
4013 /* If i.rm.regmem == ESP (4)
4014 && i.rm.mode != (Register mode)
4015 && not 16 bit
4016 ==> need second modrm byte. */
4017 if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING
4018 && i.rm.mode != 3
4019 && !(i.base_reg && (i.base_reg->reg_type & Reg16) != 0))
4020 {
4021 p = frag_more (1);
4022 md_number_to_chars (p,
4023 (valueT) (i.sib.base << 0
4024 | i.sib.index << 3
4025 | i.sib.scale << 6),
4026 1);
4027 }
4028 }
4029
4030 if (i.disp_operands)
4031 output_disp (insn_start_frag, insn_start_off);
4032
4033 if (i.imm_operands)
4034 output_imm (insn_start_frag, insn_start_off);
4035 }
4036
4037 #ifdef DEBUG386
4038 if (flag_debug)
4039 {
4040 pi ("" /*line*/, &i);
4041 }
4042 #endif /* DEBUG386 */
4043 }
4044
4045 static void
4046 output_disp (fragS *insn_start_frag, offsetT insn_start_off)
4047 {
4048 char *p;
4049 unsigned int n;
4050
4051 for (n = 0; n < i.operands; n++)
4052 {
4053 if (i.types[n] & Disp)
4054 {
4055 if (i.op[n].disps->X_op == O_constant)
4056 {
4057 int size;
4058 offsetT val;
4059
4060 size = 4;
4061 if (i.types[n] & (Disp8 | Disp16 | Disp64))
4062 {
4063 size = 2;
4064 if (i.types[n] & Disp8)
4065 size = 1;
4066 if (i.types[n] & Disp64)
4067 size = 8;
4068 }
4069 val = offset_in_range (i.op[n].disps->X_add_number,
4070 size);
4071 p = frag_more (size);
4072 md_number_to_chars (p, val, size);
4073 }
4074 else
4075 {
4076 enum bfd_reloc_code_real reloc_type;
4077 int size = 4;
4078 int sign = 0;
4079 int pcrel = (i.flags[n] & Operand_PCrel) != 0;
4080
4081 /* The PC relative address is computed relative
4082 to the instruction boundary, so in case immediate
4083 fields follows, we need to adjust the value. */
4084 if (pcrel && i.imm_operands)
4085 {
4086 int imm_size = 4;
4087 unsigned int n1;
4088
4089 for (n1 = 0; n1 < i.operands; n1++)
4090 if (i.types[n1] & Imm)
4091 {
4092 if (i.types[n1] & (Imm8 | Imm8S | Imm16 | Imm64))
4093 {
4094 imm_size = 2;
4095 if (i.types[n1] & (Imm8 | Imm8S))
4096 imm_size = 1;
4097 if (i.types[n1] & Imm64)
4098 imm_size = 8;
4099 }
4100 break;
4101 }
4102 /* We should find the immediate. */
4103 if (n1 == i.operands)
4104 abort ();
4105 i.op[n].disps->X_add_number -= imm_size;
4106 }
4107
4108 if (i.types[n] & Disp32S)
4109 sign = 1;
4110
4111 if (i.types[n] & (Disp16 | Disp64))
4112 {
4113 size = 2;
4114 if (i.types[n] & Disp64)
4115 size = 8;
4116 }
4117
4118 p = frag_more (size);
4119 reloc_type = reloc (size, pcrel, sign, i.reloc[n]);
4120 if (GOT_symbol
4121 && GOT_symbol == i.op[n].disps->X_add_symbol
4122 && (((reloc_type == BFD_RELOC_32
4123 || reloc_type == BFD_RELOC_X86_64_32S
4124 || (reloc_type == BFD_RELOC_64
4125 && object_64bit))
4126 && (i.op[n].disps->X_op == O_symbol
4127 || (i.op[n].disps->X_op == O_add
4128 && ((symbol_get_value_expression
4129 (i.op[n].disps->X_op_symbol)->X_op)
4130 == O_subtract))))
4131 || reloc_type == BFD_RELOC_32_PCREL))
4132 {
4133 offsetT add;
4134
4135 if (insn_start_frag == frag_now)
4136 add = (p - frag_now->fr_literal) - insn_start_off;
4137 else
4138 {
4139 fragS *fr;
4140
4141 add = insn_start_frag->fr_fix - insn_start_off;
4142 for (fr = insn_start_frag->fr_next;
4143 fr && fr != frag_now; fr = fr->fr_next)
4144 add += fr->fr_fix;
4145 add += p - frag_now->fr_literal;
4146 }
4147
4148 if (!object_64bit)
4149 {
4150 reloc_type = BFD_RELOC_386_GOTPC;
4151 i.op[n].imms->X_add_number += add;
4152 }
4153 else if (reloc_type == BFD_RELOC_64)
4154 reloc_type = BFD_RELOC_X86_64_GOTPC64;
4155 else
4156 /* Don't do the adjustment for x86-64, as there
4157 the pcrel addressing is relative to the _next_
4158 insn, and that is taken care of in other code. */
4159 reloc_type = BFD_RELOC_X86_64_GOTPC32;
4160 }
4161 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
4162 i.op[n].disps, pcrel, reloc_type);
4163 }
4164 }
4165 }
4166 }
4167
4168 static void
4169 output_imm (fragS *insn_start_frag, offsetT insn_start_off)
4170 {
4171 char *p;
4172 unsigned int n;
4173
4174 for (n = 0; n < i.operands; n++)
4175 {
4176 if (i.types[n] & Imm)
4177 {
4178 if (i.op[n].imms->X_op == O_constant)
4179 {
4180 int size;
4181 offsetT val;
4182
4183 size = 4;
4184 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
4185 {
4186 size = 2;
4187 if (i.types[n] & (Imm8 | Imm8S))
4188 size = 1;
4189 else if (i.types[n] & Imm64)
4190 size = 8;
4191 }
4192 val = offset_in_range (i.op[n].imms->X_add_number,
4193 size);
4194 p = frag_more (size);
4195 md_number_to_chars (p, val, size);
4196 }
4197 else
4198 {
4199 /* Not absolute_section.
4200 Need a 32-bit fixup (don't support 8bit
4201 non-absolute imms). Try to support other
4202 sizes ... */
4203 enum bfd_reloc_code_real reloc_type;
4204 int size = 4;
4205 int sign = 0;
4206
4207 if ((i.types[n] & (Imm32S))
4208 && (i.suffix == QWORD_MNEM_SUFFIX
4209 || (!i.suffix && (i.tm.opcode_modifier & No_lSuf))))
4210 sign = 1;
4211 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
4212 {
4213 size = 2;
4214 if (i.types[n] & (Imm8 | Imm8S))
4215 size = 1;
4216 if (i.types[n] & Imm64)
4217 size = 8;
4218 }
4219
4220 p = frag_more (size);
4221 reloc_type = reloc (size, 0, sign, i.reloc[n]);
4222
4223 /* This is tough to explain. We end up with this one if we
4224 * have operands that look like
4225 * "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal here is to
4226 * obtain the absolute address of the GOT, and it is strongly
4227 * preferable from a performance point of view to avoid using
4228 * a runtime relocation for this. The actual sequence of
4229 * instructions often look something like:
4230 *
4231 * call .L66
4232 * .L66:
4233 * popl %ebx
4234 * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
4235 *
4236 * The call and pop essentially return the absolute address
4237 * of the label .L66 and store it in %ebx. The linker itself
4238 * will ultimately change the first operand of the addl so
4239 * that %ebx points to the GOT, but to keep things simple, the
4240 * .o file must have this operand set so that it generates not
4241 * the absolute address of .L66, but the absolute address of
4242 * itself. This allows the linker itself simply treat a GOTPC
4243 * relocation as asking for a pcrel offset to the GOT to be
4244 * added in, and the addend of the relocation is stored in the
4245 * operand field for the instruction itself.
4246 *
4247 * Our job here is to fix the operand so that it would add
4248 * the correct offset so that %ebx would point to itself. The
4249 * thing that is tricky is that .-.L66 will point to the
4250 * beginning of the instruction, so we need to further modify
4251 * the operand so that it will point to itself. There are
4252 * other cases where you have something like:
4253 *
4254 * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
4255 *
4256 * and here no correction would be required. Internally in
4257 * the assembler we treat operands of this form as not being
4258 * pcrel since the '.' is explicitly mentioned, and I wonder
4259 * whether it would simplify matters to do it this way. Who
4260 * knows. In earlier versions of the PIC patches, the
4261 * pcrel_adjust field was used to store the correction, but
4262 * since the expression is not pcrel, I felt it would be
4263 * confusing to do it this way. */
4264
4265 if ((reloc_type == BFD_RELOC_32
4266 || reloc_type == BFD_RELOC_X86_64_32S
4267 || reloc_type == BFD_RELOC_64)
4268 && GOT_symbol
4269 && GOT_symbol == i.op[n].imms->X_add_symbol
4270 && (i.op[n].imms->X_op == O_symbol
4271 || (i.op[n].imms->X_op == O_add
4272 && ((symbol_get_value_expression
4273 (i.op[n].imms->X_op_symbol)->X_op)
4274 == O_subtract))))
4275 {
4276 offsetT add;
4277
4278 if (insn_start_frag == frag_now)
4279 add = (p - frag_now->fr_literal) - insn_start_off;
4280 else
4281 {
4282 fragS *fr;
4283
4284 add = insn_start_frag->fr_fix - insn_start_off;
4285 for (fr = insn_start_frag->fr_next;
4286 fr && fr != frag_now; fr = fr->fr_next)
4287 add += fr->fr_fix;
4288 add += p - frag_now->fr_literal;
4289 }
4290
4291 if (!object_64bit)
4292 reloc_type = BFD_RELOC_386_GOTPC;
4293 else if (size == 4)
4294 reloc_type = BFD_RELOC_X86_64_GOTPC32;
4295 else if (size == 8)
4296 reloc_type = BFD_RELOC_X86_64_GOTPC64;
4297 i.op[n].imms->X_add_number += add;
4298 }
4299 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
4300 i.op[n].imms, 0, reloc_type);
4301 }
4302 }
4303 }
4304 }
4305 \f
4306 /* x86_cons_fix_new is called via the expression parsing code when a
4307 reloc is needed. We use this hook to get the correct .got reloc. */
4308 static enum bfd_reloc_code_real got_reloc = NO_RELOC;
4309 static int cons_sign = -1;
4310
4311 void
4312 x86_cons_fix_new (fragS *frag,
4313 unsigned int off,
4314 unsigned int len,
4315 expressionS *exp)
4316 {
4317 enum bfd_reloc_code_real r = reloc (len, 0, cons_sign, got_reloc);
4318
4319 got_reloc = NO_RELOC;
4320
4321 #ifdef TE_PE
4322 if (exp->X_op == O_secrel)
4323 {
4324 exp->X_op = O_symbol;
4325 r = BFD_RELOC_32_SECREL;
4326 }
4327 #endif
4328
4329 fix_new_exp (frag, off, len, exp, 0, r);
4330 }
4331
4332 #if (!defined (OBJ_ELF) && !defined (OBJ_MAYBE_ELF)) || defined (LEX_AT)
4333 # define lex_got(reloc, adjust, types) NULL
4334 #else
4335 /* Parse operands of the form
4336 <symbol>@GOTOFF+<nnn>
4337 and similar .plt or .got references.
4338
4339 If we find one, set up the correct relocation in RELOC and copy the
4340 input string, minus the `@GOTOFF' into a malloc'd buffer for
4341 parsing by the calling routine. Return this buffer, and if ADJUST
4342 is non-null set it to the length of the string we removed from the
4343 input line. Otherwise return NULL. */
4344 static char *
4345 lex_got (enum bfd_reloc_code_real *reloc,
4346 int *adjust,
4347 unsigned int *types)
4348 {
4349 /* Some of the relocations depend on the size of what field is to
4350 be relocated. But in our callers i386_immediate and i386_displacement
4351 we don't yet know the operand size (this will be set by insn
4352 matching). Hence we record the word32 relocation here,
4353 and adjust the reloc according to the real size in reloc(). */
4354 static const struct {
4355 const char *str;
4356 const enum bfd_reloc_code_real rel[2];
4357 const unsigned int types64;
4358 } gotrel[] = {
4359 { "PLTOFF", { 0, BFD_RELOC_X86_64_PLTOFF64 }, Imm64 },
4360 { "PLT", { BFD_RELOC_386_PLT32, BFD_RELOC_X86_64_PLT32 }, Imm32|Imm32S|Disp32 },
4361 { "GOTPLT", { 0, BFD_RELOC_X86_64_GOTPLT64 }, Imm64|Disp64 },
4362 { "GOTOFF", { BFD_RELOC_386_GOTOFF, BFD_RELOC_X86_64_GOTOFF64 }, Imm64|Disp64 },
4363 { "GOTPCREL", { 0, BFD_RELOC_X86_64_GOTPCREL }, Imm32|Imm32S|Disp32 },
4364 { "TLSGD", { BFD_RELOC_386_TLS_GD, BFD_RELOC_X86_64_TLSGD }, Imm32|Imm32S|Disp32 },
4365 { "TLSLDM", { BFD_RELOC_386_TLS_LDM, 0 }, 0 },
4366 { "TLSLD", { 0, BFD_RELOC_X86_64_TLSLD }, Imm32|Imm32S|Disp32 },
4367 { "GOTTPOFF", { BFD_RELOC_386_TLS_IE_32, BFD_RELOC_X86_64_GOTTPOFF }, Imm32|Imm32S|Disp32 },
4368 { "TPOFF", { BFD_RELOC_386_TLS_LE_32, BFD_RELOC_X86_64_TPOFF32 }, Imm32|Imm32S|Imm64|Disp32|Disp64 },
4369 { "NTPOFF", { BFD_RELOC_386_TLS_LE, 0 }, 0 },
4370 { "DTPOFF", { BFD_RELOC_386_TLS_LDO_32, BFD_RELOC_X86_64_DTPOFF32 }, Imm32|Imm32S|Imm64|Disp32|Disp64 },
4371 { "GOTNTPOFF",{ BFD_RELOC_386_TLS_GOTIE, 0 }, 0 },
4372 { "INDNTPOFF",{ BFD_RELOC_386_TLS_IE, 0 }, 0 },
4373 { "GOT", { BFD_RELOC_386_GOT32, BFD_RELOC_X86_64_GOT32 }, Imm32|Imm32S|Disp32|Imm64 },
4374 { "TLSDESC", { BFD_RELOC_386_TLS_GOTDESC, BFD_RELOC_X86_64_GOTPC32_TLSDESC }, Imm32|Imm32S|Disp32 },
4375 { "TLSCALL", { BFD_RELOC_386_TLS_DESC_CALL, BFD_RELOC_X86_64_TLSDESC_CALL }, Imm32|Imm32S|Disp32 }
4376 };
4377 char *cp;
4378 unsigned int j;
4379
4380 if (!IS_ELF)
4381 return NULL;
4382
4383 for (cp = input_line_pointer; *cp != '@'; cp++)
4384 if (is_end_of_line[(unsigned char) *cp])
4385 return NULL;
4386
4387 for (j = 0; j < sizeof (gotrel) / sizeof (gotrel[0]); j++)
4388 {
4389 int len;
4390
4391 len = strlen (gotrel[j].str);
4392 if (strncasecmp (cp + 1, gotrel[j].str, len) == 0)
4393 {
4394 if (gotrel[j].rel[object_64bit] != 0)
4395 {
4396 int first, second;
4397 char *tmpbuf, *past_reloc;
4398
4399 *reloc = gotrel[j].rel[object_64bit];
4400 if (adjust)
4401 *adjust = len;
4402
4403 if (types)
4404 {
4405 if (flag_code != CODE_64BIT)
4406 *types = Imm32|Disp32;
4407 else
4408 *types = gotrel[j].types64;
4409 }
4410
4411 if (GOT_symbol == NULL)
4412 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
4413
4414 /* Replace the relocation token with ' ', so that
4415 errors like foo@GOTOFF1 will be detected. */
4416
4417 /* The length of the first part of our input line. */
4418 first = cp - input_line_pointer;
4419
4420 /* The second part goes from after the reloc token until
4421 (and including) an end_of_line char. Don't use strlen
4422 here as the end_of_line char may not be a NUL. */
4423 past_reloc = cp + 1 + len;
4424 for (cp = past_reloc; !is_end_of_line[(unsigned char) *cp++]; )
4425 ;
4426 second = cp - past_reloc;
4427
4428 /* Allocate and copy string. The trailing NUL shouldn't
4429 be necessary, but be safe. */
4430 tmpbuf = xmalloc (first + second + 2);
4431 memcpy (tmpbuf, input_line_pointer, first);
4432 tmpbuf[first] = ' ';
4433 memcpy (tmpbuf + first + 1, past_reloc, second);
4434 tmpbuf[first + second + 1] = '\0';
4435 return tmpbuf;
4436 }
4437
4438 as_bad (_("@%s reloc is not supported with %d-bit output format"),
4439 gotrel[j].str, 1 << (5 + object_64bit));
4440 return NULL;
4441 }
4442 }
4443
4444 /* Might be a symbol version string. Don't as_bad here. */
4445 return NULL;
4446 }
4447
4448 void
4449 x86_cons (exp, size)
4450 expressionS *exp;
4451 int size;
4452 {
4453 if (size == 4 || (object_64bit && size == 8))
4454 {
4455 /* Handle @GOTOFF and the like in an expression. */
4456 char *save;
4457 char *gotfree_input_line;
4458 int adjust;
4459
4460 save = input_line_pointer;
4461 gotfree_input_line = lex_got (&got_reloc, &adjust, NULL);
4462 if (gotfree_input_line)
4463 input_line_pointer = gotfree_input_line;
4464
4465 expression (exp);
4466
4467 if (gotfree_input_line)
4468 {
4469 /* expression () has merrily parsed up to the end of line,
4470 or a comma - in the wrong buffer. Transfer how far
4471 input_line_pointer has moved to the right buffer. */
4472 input_line_pointer = (save
4473 + (input_line_pointer - gotfree_input_line)
4474 + adjust);
4475 free (gotfree_input_line);
4476 }
4477 }
4478 else
4479 expression (exp);
4480 }
4481 #endif
4482
4483 static void signed_cons (int size)
4484 {
4485 if (flag_code == CODE_64BIT)
4486 cons_sign = 1;
4487 cons (size);
4488 cons_sign = -1;
4489 }
4490
4491 #ifdef TE_PE
4492 static void
4493 pe_directive_secrel (dummy)
4494 int dummy ATTRIBUTE_UNUSED;
4495 {
4496 expressionS exp;
4497
4498 do
4499 {
4500 expression (&exp);
4501 if (exp.X_op == O_symbol)
4502 exp.X_op = O_secrel;
4503
4504 emit_expr (&exp, 4);
4505 }
4506 while (*input_line_pointer++ == ',');
4507
4508 input_line_pointer--;
4509 demand_empty_rest_of_line ();
4510 }
4511 #endif
4512
4513 static int
4514 i386_immediate (char *imm_start)
4515 {
4516 char *save_input_line_pointer;
4517 char *gotfree_input_line;
4518 segT exp_seg = 0;
4519 expressionS *exp;
4520 unsigned int types = ~0U;
4521
4522 if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
4523 {
4524 as_bad (_("at most %d immediate operands are allowed"),
4525 MAX_IMMEDIATE_OPERANDS);
4526 return 0;
4527 }
4528
4529 exp = &im_expressions[i.imm_operands++];
4530 i.op[this_operand].imms = exp;
4531
4532 if (is_space_char (*imm_start))
4533 ++imm_start;
4534
4535 save_input_line_pointer = input_line_pointer;
4536 input_line_pointer = imm_start;
4537
4538 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL, &types);
4539 if (gotfree_input_line)
4540 input_line_pointer = gotfree_input_line;
4541
4542 exp_seg = expression (exp);
4543
4544 SKIP_WHITESPACE ();
4545 if (*input_line_pointer)
4546 as_bad (_("junk `%s' after expression"), input_line_pointer);
4547
4548 input_line_pointer = save_input_line_pointer;
4549 if (gotfree_input_line)
4550 free (gotfree_input_line);
4551
4552 if (exp->X_op == O_absent || exp->X_op == O_big)
4553 {
4554 /* Missing or bad expr becomes absolute 0. */
4555 as_bad (_("missing or invalid immediate expression `%s' taken as 0"),
4556 imm_start);
4557 exp->X_op = O_constant;
4558 exp->X_add_number = 0;
4559 exp->X_add_symbol = (symbolS *) 0;
4560 exp->X_op_symbol = (symbolS *) 0;
4561 }
4562 else if (exp->X_op == O_constant)
4563 {
4564 /* Size it properly later. */
4565 i.types[this_operand] |= Imm64;
4566 /* If BFD64, sign extend val. */
4567 if (!use_rela_relocations)
4568 if ((exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0)
4569 exp->X_add_number = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
4570 }
4571 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4572 else if (OUTPUT_FLAVOR == bfd_target_aout_flavour
4573 && exp_seg != absolute_section
4574 && exp_seg != text_section
4575 && exp_seg != data_section
4576 && exp_seg != bss_section
4577 && exp_seg != undefined_section
4578 && !bfd_is_com_section (exp_seg))
4579 {
4580 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
4581 return 0;
4582 }
4583 #endif
4584 else if (!intel_syntax && exp->X_op == O_register)
4585 {
4586 as_bad (_("illegal immediate register operand %s"), imm_start);
4587 return 0;
4588 }
4589 else
4590 {
4591 /* This is an address. The size of the address will be
4592 determined later, depending on destination register,
4593 suffix, or the default for the section. */
4594 i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64;
4595 i.types[this_operand] &= types;
4596 }
4597
4598 return 1;
4599 }
4600
4601 static char *i386_scale PARAMS ((char *));
4602
4603 static char *
4604 i386_scale (scale)
4605 char *scale;
4606 {
4607 offsetT val;
4608 char *save = input_line_pointer;
4609
4610 input_line_pointer = scale;
4611 val = get_absolute_expression ();
4612
4613 switch (val)
4614 {
4615 case 1:
4616 i.log2_scale_factor = 0;
4617 break;
4618 case 2:
4619 i.log2_scale_factor = 1;
4620 break;
4621 case 4:
4622 i.log2_scale_factor = 2;
4623 break;
4624 case 8:
4625 i.log2_scale_factor = 3;
4626 break;
4627 default:
4628 {
4629 char sep = *input_line_pointer;
4630
4631 *input_line_pointer = '\0';
4632 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
4633 scale);
4634 *input_line_pointer = sep;
4635 input_line_pointer = save;
4636 return NULL;
4637 }
4638 }
4639 if (i.log2_scale_factor != 0 && i.index_reg == 0)
4640 {
4641 as_warn (_("scale factor of %d without an index register"),
4642 1 << i.log2_scale_factor);
4643 #if SCALE1_WHEN_NO_INDEX
4644 i.log2_scale_factor = 0;
4645 #endif
4646 }
4647 scale = input_line_pointer;
4648 input_line_pointer = save;
4649 return scale;
4650 }
4651
4652 static int i386_displacement PARAMS ((char *, char *));
4653
4654 static int
4655 i386_displacement (disp_start, disp_end)
4656 char *disp_start;
4657 char *disp_end;
4658 {
4659 expressionS *exp;
4660 segT exp_seg = 0;
4661 char *save_input_line_pointer;
4662 char *gotfree_input_line;
4663 int bigdisp, override;
4664 unsigned int types = Disp;
4665
4666 if (i.disp_operands == MAX_MEMORY_OPERANDS)
4667 {
4668 as_bad (_("at most %d displacement operands are allowed"),
4669 MAX_MEMORY_OPERANDS);
4670 return 0;
4671 }
4672
4673 if ((i.types[this_operand] & JumpAbsolute)
4674 || !(current_templates->start->opcode_modifier & (Jump | JumpDword)))
4675 {
4676 bigdisp = Disp32;
4677 override = (i.prefix[ADDR_PREFIX] != 0);
4678 }
4679 else
4680 {
4681 /* For PC-relative branches, the width of the displacement
4682 is dependent upon data size, not address size. */
4683 bigdisp = 0;
4684 override = (i.prefix[DATA_PREFIX] != 0);
4685 }
4686 if (flag_code == CODE_64BIT)
4687 {
4688 if (!bigdisp)
4689 bigdisp = ((override || i.suffix == WORD_MNEM_SUFFIX)
4690 ? Disp16
4691 : Disp32S | Disp32);
4692 else if (!override)
4693 bigdisp = Disp64 | Disp32S | Disp32;
4694 }
4695 else
4696 {
4697 if (!bigdisp)
4698 {
4699 if (!override)
4700 override = (i.suffix == (flag_code != CODE_16BIT
4701 ? WORD_MNEM_SUFFIX
4702 : LONG_MNEM_SUFFIX));
4703 bigdisp = Disp32;
4704 }
4705 if ((flag_code == CODE_16BIT) ^ override)
4706 bigdisp = Disp16;
4707 }
4708 i.types[this_operand] |= bigdisp;
4709
4710 exp = &disp_expressions[i.disp_operands];
4711 i.op[this_operand].disps = exp;
4712 i.disp_operands++;
4713 save_input_line_pointer = input_line_pointer;
4714 input_line_pointer = disp_start;
4715 END_STRING_AND_SAVE (disp_end);
4716
4717 #ifndef GCC_ASM_O_HACK
4718 #define GCC_ASM_O_HACK 0
4719 #endif
4720 #if GCC_ASM_O_HACK
4721 END_STRING_AND_SAVE (disp_end + 1);
4722 if ((i.types[this_operand] & BaseIndex) != 0
4723 && displacement_string_end[-1] == '+')
4724 {
4725 /* This hack is to avoid a warning when using the "o"
4726 constraint within gcc asm statements.
4727 For instance:
4728
4729 #define _set_tssldt_desc(n,addr,limit,type) \
4730 __asm__ __volatile__ ( \
4731 "movw %w2,%0\n\t" \
4732 "movw %w1,2+%0\n\t" \
4733 "rorl $16,%1\n\t" \
4734 "movb %b1,4+%0\n\t" \
4735 "movb %4,5+%0\n\t" \
4736 "movb $0,6+%0\n\t" \
4737 "movb %h1,7+%0\n\t" \
4738 "rorl $16,%1" \
4739 : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
4740
4741 This works great except that the output assembler ends
4742 up looking a bit weird if it turns out that there is
4743 no offset. You end up producing code that looks like:
4744
4745 #APP
4746 movw $235,(%eax)
4747 movw %dx,2+(%eax)
4748 rorl $16,%edx
4749 movb %dl,4+(%eax)
4750 movb $137,5+(%eax)
4751 movb $0,6+(%eax)
4752 movb %dh,7+(%eax)
4753 rorl $16,%edx
4754 #NO_APP
4755
4756 So here we provide the missing zero. */
4757
4758 *displacement_string_end = '0';
4759 }
4760 #endif
4761 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL, &types);
4762 if (gotfree_input_line)
4763 input_line_pointer = gotfree_input_line;
4764
4765 exp_seg = expression (exp);
4766
4767 SKIP_WHITESPACE ();
4768 if (*input_line_pointer)
4769 as_bad (_("junk `%s' after expression"), input_line_pointer);
4770 #if GCC_ASM_O_HACK
4771 RESTORE_END_STRING (disp_end + 1);
4772 #endif
4773 RESTORE_END_STRING (disp_end);
4774 input_line_pointer = save_input_line_pointer;
4775 if (gotfree_input_line)
4776 free (gotfree_input_line);
4777
4778 /* We do this to make sure that the section symbol is in
4779 the symbol table. We will ultimately change the relocation
4780 to be relative to the beginning of the section. */
4781 if (i.reloc[this_operand] == BFD_RELOC_386_GOTOFF
4782 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
4783 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTOFF64)
4784 {
4785 if (exp->X_op != O_symbol)
4786 {
4787 as_bad (_("bad expression used with @%s"),
4788 (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
4789 ? "GOTPCREL"
4790 : "GOTOFF"));
4791 return 0;
4792 }
4793
4794 if (S_IS_LOCAL (exp->X_add_symbol)
4795 && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
4796 section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
4797 exp->X_op = O_subtract;
4798 exp->X_op_symbol = GOT_symbol;
4799 if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
4800 i.reloc[this_operand] = BFD_RELOC_32_PCREL;
4801 else if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTOFF64)
4802 i.reloc[this_operand] = BFD_RELOC_64;
4803 else
4804 i.reloc[this_operand] = BFD_RELOC_32;
4805 }
4806
4807 if (exp->X_op == O_absent || exp->X_op == O_big)
4808 {
4809 /* Missing or bad expr becomes absolute 0. */
4810 as_bad (_("missing or invalid displacement expression `%s' taken as 0"),
4811 disp_start);
4812 exp->X_op = O_constant;
4813 exp->X_add_number = 0;
4814 exp->X_add_symbol = (symbolS *) 0;
4815 exp->X_op_symbol = (symbolS *) 0;
4816 }
4817
4818 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4819 if (exp->X_op != O_constant
4820 && OUTPUT_FLAVOR == bfd_target_aout_flavour
4821 && exp_seg != absolute_section
4822 && exp_seg != text_section
4823 && exp_seg != data_section
4824 && exp_seg != bss_section
4825 && exp_seg != undefined_section
4826 && !bfd_is_com_section (exp_seg))
4827 {
4828 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
4829 return 0;
4830 }
4831 #endif
4832
4833 if (!(i.types[this_operand] & ~Disp))
4834 i.types[this_operand] &= types;
4835
4836 return 1;
4837 }
4838
4839 static int i386_index_check PARAMS ((const char *));
4840
4841 /* Make sure the memory operand we've been dealt is valid.
4842 Return 1 on success, 0 on a failure. */
4843
4844 static int
4845 i386_index_check (operand_string)
4846 const char *operand_string;
4847 {
4848 int ok;
4849 #if INFER_ADDR_PREFIX
4850 int fudged = 0;
4851
4852 tryprefix:
4853 #endif
4854 ok = 1;
4855 if ((current_templates->start->cpu_flags & CpuSVME)
4856 && current_templates->end[-1].operand_types[0] == AnyMem)
4857 {
4858 /* Memory operands of SVME insns are special in that they only allow
4859 rAX as their memory address and ignore any segment override. */
4860 unsigned RegXX;
4861
4862 /* SKINIT is even more restrictive: it always requires EAX. */
4863 if (strcmp (current_templates->start->name, "skinit") == 0)
4864 RegXX = Reg32;
4865 else if (flag_code == CODE_64BIT)
4866 RegXX = i.prefix[ADDR_PREFIX] == 0 ? Reg64 : Reg32;
4867 else
4868 RegXX = ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0)
4869 ? Reg16
4870 : Reg32);
4871 if (!i.base_reg
4872 || !(i.base_reg->reg_type & Acc)
4873 || !(i.base_reg->reg_type & RegXX)
4874 || i.index_reg
4875 || (i.types[0] & Disp))
4876 ok = 0;
4877 }
4878 else if (flag_code == CODE_64BIT)
4879 {
4880 unsigned RegXX = (i.prefix[ADDR_PREFIX] == 0 ? Reg64 : Reg32);
4881
4882 if ((i.base_reg
4883 && ((i.base_reg->reg_type & RegXX) == 0)
4884 && (i.base_reg->reg_type != BaseIndex
4885 || i.index_reg))
4886 || (i.index_reg
4887 && ((i.index_reg->reg_type & (RegXX | BaseIndex))
4888 != (RegXX | BaseIndex))))
4889 ok = 0;
4890 }
4891 else
4892 {
4893 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
4894 {
4895 /* 16bit checks. */
4896 if ((i.base_reg
4897 && ((i.base_reg->reg_type & (Reg16 | BaseIndex | RegRex))
4898 != (Reg16 | BaseIndex)))
4899 || (i.index_reg
4900 && (((i.index_reg->reg_type & (Reg16 | BaseIndex))
4901 != (Reg16 | BaseIndex))
4902 || !(i.base_reg
4903 && i.base_reg->reg_num < 6
4904 && i.index_reg->reg_num >= 6
4905 && i.log2_scale_factor == 0))))
4906 ok = 0;
4907 }
4908 else
4909 {
4910 /* 32bit checks. */
4911 if ((i.base_reg
4912 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
4913 || (i.index_reg
4914 && ((i.index_reg->reg_type & (Reg32 | BaseIndex | RegRex))
4915 != (Reg32 | BaseIndex))))
4916 ok = 0;
4917 }
4918 }
4919 if (!ok)
4920 {
4921 #if INFER_ADDR_PREFIX
4922 if (i.prefix[ADDR_PREFIX] == 0)
4923 {
4924 i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
4925 i.prefixes += 1;
4926 /* Change the size of any displacement too. At most one of
4927 Disp16 or Disp32 is set.
4928 FIXME. There doesn't seem to be any real need for separate
4929 Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
4930 Removing them would probably clean up the code quite a lot. */
4931 if (flag_code != CODE_64BIT && (i.types[this_operand] & (Disp16 | Disp32)))
4932 i.types[this_operand] ^= (Disp16 | Disp32);
4933 fudged = 1;
4934 goto tryprefix;
4935 }
4936 if (fudged)
4937 as_bad (_("`%s' is not a valid base/index expression"),
4938 operand_string);
4939 else
4940 #endif
4941 as_bad (_("`%s' is not a valid %s bit base/index expression"),
4942 operand_string,
4943 flag_code_names[flag_code]);
4944 }
4945 return ok;
4946 }
4947
4948 /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
4949 on error. */
4950
4951 static int
4952 i386_operand (operand_string)
4953 char *operand_string;
4954 {
4955 const reg_entry *r;
4956 char *end_op;
4957 char *op_string = operand_string;
4958
4959 if (is_space_char (*op_string))
4960 ++op_string;
4961
4962 /* We check for an absolute prefix (differentiating,
4963 for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
4964 if (*op_string == ABSOLUTE_PREFIX)
4965 {
4966 ++op_string;
4967 if (is_space_char (*op_string))
4968 ++op_string;
4969 i.types[this_operand] |= JumpAbsolute;
4970 }
4971
4972 /* Check if operand is a register. */
4973 if ((r = parse_register (op_string, &end_op)) != NULL)
4974 {
4975 /* Check for a segment override by searching for ':' after a
4976 segment register. */
4977 op_string = end_op;
4978 if (is_space_char (*op_string))
4979 ++op_string;
4980 if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
4981 {
4982 switch (r->reg_num)
4983 {
4984 case 0:
4985 i.seg[i.mem_operands] = &es;
4986 break;
4987 case 1:
4988 i.seg[i.mem_operands] = &cs;
4989 break;
4990 case 2:
4991 i.seg[i.mem_operands] = &ss;
4992 break;
4993 case 3:
4994 i.seg[i.mem_operands] = &ds;
4995 break;
4996 case 4:
4997 i.seg[i.mem_operands] = &fs;
4998 break;
4999 case 5:
5000 i.seg[i.mem_operands] = &gs;
5001 break;
5002 }
5003
5004 /* Skip the ':' and whitespace. */
5005 ++op_string;
5006 if (is_space_char (*op_string))
5007 ++op_string;
5008
5009 if (!is_digit_char (*op_string)
5010 && !is_identifier_char (*op_string)
5011 && *op_string != '('
5012 && *op_string != ABSOLUTE_PREFIX)
5013 {
5014 as_bad (_("bad memory operand `%s'"), op_string);
5015 return 0;
5016 }
5017 /* Handle case of %es:*foo. */
5018 if (*op_string == ABSOLUTE_PREFIX)
5019 {
5020 ++op_string;
5021 if (is_space_char (*op_string))
5022 ++op_string;
5023 i.types[this_operand] |= JumpAbsolute;
5024 }
5025 goto do_memory_reference;
5026 }
5027 if (*op_string)
5028 {
5029 as_bad (_("junk `%s' after register"), op_string);
5030 return 0;
5031 }
5032 i.types[this_operand] |= r->reg_type & ~BaseIndex;
5033 i.op[this_operand].regs = r;
5034 i.reg_operands++;
5035 }
5036 else if (*op_string == REGISTER_PREFIX)
5037 {
5038 as_bad (_("bad register name `%s'"), op_string);
5039 return 0;
5040 }
5041 else if (*op_string == IMMEDIATE_PREFIX)
5042 {
5043 ++op_string;
5044 if (i.types[this_operand] & JumpAbsolute)
5045 {
5046 as_bad (_("immediate operand illegal with absolute jump"));
5047 return 0;
5048 }
5049 if (!i386_immediate (op_string))
5050 return 0;
5051 }
5052 else if (is_digit_char (*op_string)
5053 || is_identifier_char (*op_string)
5054 || *op_string == '(')
5055 {
5056 /* This is a memory reference of some sort. */
5057 char *base_string;
5058
5059 /* Start and end of displacement string expression (if found). */
5060 char *displacement_string_start;
5061 char *displacement_string_end;
5062
5063 do_memory_reference:
5064 if ((i.mem_operands == 1
5065 && (current_templates->start->opcode_modifier & IsString) == 0)
5066 || i.mem_operands == 2)
5067 {
5068 as_bad (_("too many memory references for `%s'"),
5069 current_templates->start->name);
5070 return 0;
5071 }
5072
5073 /* Check for base index form. We detect the base index form by
5074 looking for an ')' at the end of the operand, searching
5075 for the '(' matching it, and finding a REGISTER_PREFIX or ','
5076 after the '('. */
5077 base_string = op_string + strlen (op_string);
5078
5079 --base_string;
5080 if (is_space_char (*base_string))
5081 --base_string;
5082
5083 /* If we only have a displacement, set-up for it to be parsed later. */
5084 displacement_string_start = op_string;
5085 displacement_string_end = base_string + 1;
5086
5087 if (*base_string == ')')
5088 {
5089 char *temp_string;
5090 unsigned int parens_balanced = 1;
5091 /* We've already checked that the number of left & right ()'s are
5092 equal, so this loop will not be infinite. */
5093 do
5094 {
5095 base_string--;
5096 if (*base_string == ')')
5097 parens_balanced++;
5098 if (*base_string == '(')
5099 parens_balanced--;
5100 }
5101 while (parens_balanced);
5102
5103 temp_string = base_string;
5104
5105 /* Skip past '(' and whitespace. */
5106 ++base_string;
5107 if (is_space_char (*base_string))
5108 ++base_string;
5109
5110 if (*base_string == ','
5111 || ((i.base_reg = parse_register (base_string, &end_op)) != NULL))
5112 {
5113 displacement_string_end = temp_string;
5114
5115 i.types[this_operand] |= BaseIndex;
5116
5117 if (i.base_reg)
5118 {
5119 base_string = end_op;
5120 if (is_space_char (*base_string))
5121 ++base_string;
5122 }
5123
5124 /* There may be an index reg or scale factor here. */
5125 if (*base_string == ',')
5126 {
5127 ++base_string;
5128 if (is_space_char (*base_string))
5129 ++base_string;
5130
5131 if ((i.index_reg = parse_register (base_string, &end_op)) != NULL)
5132 {
5133 base_string = end_op;
5134 if (is_space_char (*base_string))
5135 ++base_string;
5136 if (*base_string == ',')
5137 {
5138 ++base_string;
5139 if (is_space_char (*base_string))
5140 ++base_string;
5141 }
5142 else if (*base_string != ')')
5143 {
5144 as_bad (_("expecting `,' or `)' after index register in `%s'"),
5145 operand_string);
5146 return 0;
5147 }
5148 }
5149 else if (*base_string == REGISTER_PREFIX)
5150 {
5151 as_bad (_("bad register name `%s'"), base_string);
5152 return 0;
5153 }
5154
5155 /* Check for scale factor. */
5156 if (*base_string != ')')
5157 {
5158 char *end_scale = i386_scale (base_string);
5159
5160 if (!end_scale)
5161 return 0;
5162
5163 base_string = end_scale;
5164 if (is_space_char (*base_string))
5165 ++base_string;
5166 if (*base_string != ')')
5167 {
5168 as_bad (_("expecting `)' after scale factor in `%s'"),
5169 operand_string);
5170 return 0;
5171 }
5172 }
5173 else if (!i.index_reg)
5174 {
5175 as_bad (_("expecting index register or scale factor after `,'; got '%c'"),
5176 *base_string);
5177 return 0;
5178 }
5179 }
5180 else if (*base_string != ')')
5181 {
5182 as_bad (_("expecting `,' or `)' after base register in `%s'"),
5183 operand_string);
5184 return 0;
5185 }
5186 }
5187 else if (*base_string == REGISTER_PREFIX)
5188 {
5189 as_bad (_("bad register name `%s'"), base_string);
5190 return 0;
5191 }
5192 }
5193
5194 /* If there's an expression beginning the operand, parse it,
5195 assuming displacement_string_start and
5196 displacement_string_end are meaningful. */
5197 if (displacement_string_start != displacement_string_end)
5198 {
5199 if (!i386_displacement (displacement_string_start,
5200 displacement_string_end))
5201 return 0;
5202 }
5203
5204 /* Special case for (%dx) while doing input/output op. */
5205 if (i.base_reg
5206 && i.base_reg->reg_type == (Reg16 | InOutPortReg)
5207 && i.index_reg == 0
5208 && i.log2_scale_factor == 0
5209 && i.seg[i.mem_operands] == 0
5210 && (i.types[this_operand] & Disp) == 0)
5211 {
5212 i.types[this_operand] = InOutPortReg;
5213 return 1;
5214 }
5215
5216 if (i386_index_check (operand_string) == 0)
5217 return 0;
5218 i.mem_operands++;
5219 }
5220 else
5221 {
5222 /* It's not a memory operand; argh! */
5223 as_bad (_("invalid char %s beginning operand %d `%s'"),
5224 output_invalid (*op_string),
5225 this_operand + 1,
5226 op_string);
5227 return 0;
5228 }
5229 return 1; /* Normal return. */
5230 }
5231 \f
5232 /* md_estimate_size_before_relax()
5233
5234 Called just before relax() for rs_machine_dependent frags. The x86
5235 assembler uses these frags to handle variable size jump
5236 instructions.
5237
5238 Any symbol that is now undefined will not become defined.
5239 Return the correct fr_subtype in the frag.
5240 Return the initial "guess for variable size of frag" to caller.
5241 The guess is actually the growth beyond the fixed part. Whatever
5242 we do to grow the fixed or variable part contributes to our
5243 returned value. */
5244
5245 int
5246 md_estimate_size_before_relax (fragP, segment)
5247 fragS *fragP;
5248 segT segment;
5249 {
5250 /* We've already got fragP->fr_subtype right; all we have to do is
5251 check for un-relaxable symbols. On an ELF system, we can't relax
5252 an externally visible symbol, because it may be overridden by a
5253 shared library. */
5254 if (S_GET_SEGMENT (fragP->fr_symbol) != segment
5255 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5256 || (IS_ELF
5257 && (S_IS_EXTERNAL (fragP->fr_symbol)
5258 || S_IS_WEAK (fragP->fr_symbol)))
5259 #endif
5260 )
5261 {
5262 /* Symbol is undefined in this segment, or we need to keep a
5263 reloc so that weak symbols can be overridden. */
5264 int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
5265 enum bfd_reloc_code_real reloc_type;
5266 unsigned char *opcode;
5267 int old_fr_fix;
5268
5269 if (fragP->fr_var != NO_RELOC)
5270 reloc_type = fragP->fr_var;
5271 else if (size == 2)
5272 reloc_type = BFD_RELOC_16_PCREL;
5273 else
5274 reloc_type = BFD_RELOC_32_PCREL;
5275
5276 old_fr_fix = fragP->fr_fix;
5277 opcode = (unsigned char *) fragP->fr_opcode;
5278
5279 switch (TYPE_FROM_RELAX_STATE (fragP->fr_subtype))
5280 {
5281 case UNCOND_JUMP:
5282 /* Make jmp (0xeb) a (d)word displacement jump. */
5283 opcode[0] = 0xe9;
5284 fragP->fr_fix += size;
5285 fix_new (fragP, old_fr_fix, size,
5286 fragP->fr_symbol,
5287 fragP->fr_offset, 1,
5288 reloc_type);
5289 break;
5290
5291 case COND_JUMP86:
5292 if (size == 2
5293 && (!no_cond_jump_promotion || fragP->fr_var != NO_RELOC))
5294 {
5295 /* Negate the condition, and branch past an
5296 unconditional jump. */
5297 opcode[0] ^= 1;
5298 opcode[1] = 3;
5299 /* Insert an unconditional jump. */
5300 opcode[2] = 0xe9;
5301 /* We added two extra opcode bytes, and have a two byte
5302 offset. */
5303 fragP->fr_fix += 2 + 2;
5304 fix_new (fragP, old_fr_fix + 2, 2,
5305 fragP->fr_symbol,
5306 fragP->fr_offset, 1,
5307 reloc_type);
5308 break;
5309 }
5310 /* Fall through. */
5311
5312 case COND_JUMP:
5313 if (no_cond_jump_promotion && fragP->fr_var == NO_RELOC)
5314 {
5315 fixS *fixP;
5316
5317 fragP->fr_fix += 1;
5318 fixP = fix_new (fragP, old_fr_fix, 1,
5319 fragP->fr_symbol,
5320 fragP->fr_offset, 1,
5321 BFD_RELOC_8_PCREL);
5322 fixP->fx_signed = 1;
5323 break;
5324 }
5325
5326 /* This changes the byte-displacement jump 0x7N
5327 to the (d)word-displacement jump 0x0f,0x8N. */
5328 opcode[1] = opcode[0] + 0x10;
5329 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
5330 /* We've added an opcode byte. */
5331 fragP->fr_fix += 1 + size;
5332 fix_new (fragP, old_fr_fix + 1, size,
5333 fragP->fr_symbol,
5334 fragP->fr_offset, 1,
5335 reloc_type);
5336 break;
5337
5338 default:
5339 BAD_CASE (fragP->fr_subtype);
5340 break;
5341 }
5342 frag_wane (fragP);
5343 return fragP->fr_fix - old_fr_fix;
5344 }
5345
5346 /* Guess size depending on current relax state. Initially the relax
5347 state will correspond to a short jump and we return 1, because
5348 the variable part of the frag (the branch offset) is one byte
5349 long. However, we can relax a section more than once and in that
5350 case we must either set fr_subtype back to the unrelaxed state,
5351 or return the value for the appropriate branch. */
5352 return md_relax_table[fragP->fr_subtype].rlx_length;
5353 }
5354
5355 /* Called after relax() is finished.
5356
5357 In: Address of frag.
5358 fr_type == rs_machine_dependent.
5359 fr_subtype is what the address relaxed to.
5360
5361 Out: Any fixSs and constants are set up.
5362 Caller will turn frag into a ".space 0". */
5363
5364 void
5365 md_convert_frag (abfd, sec, fragP)
5366 bfd *abfd ATTRIBUTE_UNUSED;
5367 segT sec ATTRIBUTE_UNUSED;
5368 fragS *fragP;
5369 {
5370 unsigned char *opcode;
5371 unsigned char *where_to_put_displacement = NULL;
5372 offsetT target_address;
5373 offsetT opcode_address;
5374 unsigned int extension = 0;
5375 offsetT displacement_from_opcode_start;
5376
5377 opcode = (unsigned char *) fragP->fr_opcode;
5378
5379 /* Address we want to reach in file space. */
5380 target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
5381
5382 /* Address opcode resides at in file space. */
5383 opcode_address = fragP->fr_address + fragP->fr_fix;
5384
5385 /* Displacement from opcode start to fill into instruction. */
5386 displacement_from_opcode_start = target_address - opcode_address;
5387
5388 if ((fragP->fr_subtype & BIG) == 0)
5389 {
5390 /* Don't have to change opcode. */
5391 extension = 1; /* 1 opcode + 1 displacement */
5392 where_to_put_displacement = &opcode[1];
5393 }
5394 else
5395 {
5396 if (no_cond_jump_promotion
5397 && TYPE_FROM_RELAX_STATE (fragP->fr_subtype) != UNCOND_JUMP)
5398 as_warn_where (fragP->fr_file, fragP->fr_line, _("long jump required"));
5399
5400 switch (fragP->fr_subtype)
5401 {
5402 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
5403 extension = 4; /* 1 opcode + 4 displacement */
5404 opcode[0] = 0xe9;
5405 where_to_put_displacement = &opcode[1];
5406 break;
5407
5408 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
5409 extension = 2; /* 1 opcode + 2 displacement */
5410 opcode[0] = 0xe9;
5411 where_to_put_displacement = &opcode[1];
5412 break;
5413
5414 case ENCODE_RELAX_STATE (COND_JUMP, BIG):
5415 case ENCODE_RELAX_STATE (COND_JUMP86, BIG):
5416 extension = 5; /* 2 opcode + 4 displacement */
5417 opcode[1] = opcode[0] + 0x10;
5418 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
5419 where_to_put_displacement = &opcode[2];
5420 break;
5421
5422 case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
5423 extension = 3; /* 2 opcode + 2 displacement */
5424 opcode[1] = opcode[0] + 0x10;
5425 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
5426 where_to_put_displacement = &opcode[2];
5427 break;
5428
5429 case ENCODE_RELAX_STATE (COND_JUMP86, BIG16):
5430 extension = 4;
5431 opcode[0] ^= 1;
5432 opcode[1] = 3;
5433 opcode[2] = 0xe9;
5434 where_to_put_displacement = &opcode[3];
5435 break;
5436
5437 default:
5438 BAD_CASE (fragP->fr_subtype);
5439 break;
5440 }
5441 }
5442
5443 /* If size if less then four we are sure that the operand fits,
5444 but if it's 4, then it could be that the displacement is larger
5445 then -/+ 2GB. */
5446 if (DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype) == 4
5447 && object_64bit
5448 && ((addressT) (displacement_from_opcode_start - extension
5449 + ((addressT) 1 << 31))
5450 > (((addressT) 2 << 31) - 1)))
5451 {
5452 as_bad_where (fragP->fr_file, fragP->fr_line,
5453 _("jump target out of range"));
5454 /* Make us emit 0. */
5455 displacement_from_opcode_start = extension;
5456 }
5457 /* Now put displacement after opcode. */
5458 md_number_to_chars ((char *) where_to_put_displacement,
5459 (valueT) (displacement_from_opcode_start - extension),
5460 DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
5461 fragP->fr_fix += extension;
5462 }
5463 \f
5464 /* Size of byte displacement jmp. */
5465 int md_short_jump_size = 2;
5466
5467 /* Size of dword displacement jmp. */
5468 int md_long_jump_size = 5;
5469
5470 void
5471 md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
5472 char *ptr;
5473 addressT from_addr, to_addr;
5474 fragS *frag ATTRIBUTE_UNUSED;
5475 symbolS *to_symbol ATTRIBUTE_UNUSED;
5476 {
5477 offsetT offset;
5478
5479 offset = to_addr - (from_addr + 2);
5480 /* Opcode for byte-disp jump. */
5481 md_number_to_chars (ptr, (valueT) 0xeb, 1);
5482 md_number_to_chars (ptr + 1, (valueT) offset, 1);
5483 }
5484
5485 void
5486 md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
5487 char *ptr;
5488 addressT from_addr, to_addr;
5489 fragS *frag ATTRIBUTE_UNUSED;
5490 symbolS *to_symbol ATTRIBUTE_UNUSED;
5491 {
5492 offsetT offset;
5493
5494 offset = to_addr - (from_addr + 5);
5495 md_number_to_chars (ptr, (valueT) 0xe9, 1);
5496 md_number_to_chars (ptr + 1, (valueT) offset, 4);
5497 }
5498 \f
5499 /* Apply a fixup (fixS) to segment data, once it has been determined
5500 by our caller that we have all the info we need to fix it up.
5501
5502 On the 386, immediates, displacements, and data pointers are all in
5503 the same (little-endian) format, so we don't need to care about which
5504 we are handling. */
5505
5506 void
5507 md_apply_fix (fixP, valP, seg)
5508 /* The fix we're to put in. */
5509 fixS *fixP;
5510 /* Pointer to the value of the bits. */
5511 valueT *valP;
5512 /* Segment fix is from. */
5513 segT seg ATTRIBUTE_UNUSED;
5514 {
5515 char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
5516 valueT value = *valP;
5517
5518 #if !defined (TE_Mach)
5519 if (fixP->fx_pcrel)
5520 {
5521 switch (fixP->fx_r_type)
5522 {
5523 default:
5524 break;
5525
5526 case BFD_RELOC_64:
5527 fixP->fx_r_type = BFD_RELOC_64_PCREL;
5528 break;
5529 case BFD_RELOC_32:
5530 case BFD_RELOC_X86_64_32S:
5531 fixP->fx_r_type = BFD_RELOC_32_PCREL;
5532 break;
5533 case BFD_RELOC_16:
5534 fixP->fx_r_type = BFD_RELOC_16_PCREL;
5535 break;
5536 case BFD_RELOC_8:
5537 fixP->fx_r_type = BFD_RELOC_8_PCREL;
5538 break;
5539 }
5540 }
5541
5542 if (fixP->fx_addsy != NULL
5543 && (fixP->fx_r_type == BFD_RELOC_32_PCREL
5544 || fixP->fx_r_type == BFD_RELOC_64_PCREL
5545 || fixP->fx_r_type == BFD_RELOC_16_PCREL
5546 || fixP->fx_r_type == BFD_RELOC_8_PCREL)
5547 && !use_rela_relocations)
5548 {
5549 /* This is a hack. There should be a better way to handle this.
5550 This covers for the fact that bfd_install_relocation will
5551 subtract the current location (for partial_inplace, PC relative
5552 relocations); see more below. */
5553 #ifndef OBJ_AOUT
5554 if (IS_ELF
5555 #ifdef TE_PE
5556 || OUTPUT_FLAVOR == bfd_target_coff_flavour
5557 #endif
5558 )
5559 value += fixP->fx_where + fixP->fx_frag->fr_address;
5560 #endif
5561 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5562 if (IS_ELF)
5563 {
5564 segT sym_seg = S_GET_SEGMENT (fixP->fx_addsy);
5565
5566 if ((sym_seg == seg
5567 || (symbol_section_p (fixP->fx_addsy)
5568 && sym_seg != absolute_section))
5569 && !generic_force_reloc (fixP))
5570 {
5571 /* Yes, we add the values in twice. This is because
5572 bfd_install_relocation subtracts them out again. I think
5573 bfd_install_relocation is broken, but I don't dare change
5574 it. FIXME. */
5575 value += fixP->fx_where + fixP->fx_frag->fr_address;
5576 }
5577 }
5578 #endif
5579 #if defined (OBJ_COFF) && defined (TE_PE)
5580 /* For some reason, the PE format does not store a
5581 section address offset for a PC relative symbol. */
5582 if (S_GET_SEGMENT (fixP->fx_addsy) != seg
5583 || S_IS_WEAK (fixP->fx_addsy))
5584 value += md_pcrel_from (fixP);
5585 #endif
5586 }
5587
5588 /* Fix a few things - the dynamic linker expects certain values here,
5589 and we must not disappoint it. */
5590 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5591 if (IS_ELF && fixP->fx_addsy)
5592 switch (fixP->fx_r_type)
5593 {
5594 case BFD_RELOC_386_PLT32:
5595 case BFD_RELOC_X86_64_PLT32:
5596 /* Make the jump instruction point to the address of the operand. At
5597 runtime we merely add the offset to the actual PLT entry. */
5598 value = -4;
5599 break;
5600
5601 case BFD_RELOC_386_TLS_GD:
5602 case BFD_RELOC_386_TLS_LDM:
5603 case BFD_RELOC_386_TLS_IE_32:
5604 case BFD_RELOC_386_TLS_IE:
5605 case BFD_RELOC_386_TLS_GOTIE:
5606 case BFD_RELOC_386_TLS_GOTDESC:
5607 case BFD_RELOC_X86_64_TLSGD:
5608 case BFD_RELOC_X86_64_TLSLD:
5609 case BFD_RELOC_X86_64_GOTTPOFF:
5610 case BFD_RELOC_X86_64_GOTPC32_TLSDESC:
5611 value = 0; /* Fully resolved at runtime. No addend. */
5612 /* Fallthrough */
5613 case BFD_RELOC_386_TLS_LE:
5614 case BFD_RELOC_386_TLS_LDO_32:
5615 case BFD_RELOC_386_TLS_LE_32:
5616 case BFD_RELOC_X86_64_DTPOFF32:
5617 case BFD_RELOC_X86_64_DTPOFF64:
5618 case BFD_RELOC_X86_64_TPOFF32:
5619 case BFD_RELOC_X86_64_TPOFF64:
5620 S_SET_THREAD_LOCAL (fixP->fx_addsy);
5621 break;
5622
5623 case BFD_RELOC_386_TLS_DESC_CALL:
5624 case BFD_RELOC_X86_64_TLSDESC_CALL:
5625 value = 0; /* Fully resolved at runtime. No addend. */
5626 S_SET_THREAD_LOCAL (fixP->fx_addsy);
5627 fixP->fx_done = 0;
5628 return;
5629
5630 case BFD_RELOC_386_GOT32:
5631 case BFD_RELOC_X86_64_GOT32:
5632 value = 0; /* Fully resolved at runtime. No addend. */
5633 break;
5634
5635 case BFD_RELOC_VTABLE_INHERIT:
5636 case BFD_RELOC_VTABLE_ENTRY:
5637 fixP->fx_done = 0;
5638 return;
5639
5640 default:
5641 break;
5642 }
5643 #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
5644 *valP = value;
5645 #endif /* !defined (TE_Mach) */
5646
5647 /* Are we finished with this relocation now? */
5648 if (fixP->fx_addsy == NULL)
5649 fixP->fx_done = 1;
5650 else if (use_rela_relocations)
5651 {
5652 fixP->fx_no_overflow = 1;
5653 /* Remember value for tc_gen_reloc. */
5654 fixP->fx_addnumber = value;
5655 value = 0;
5656 }
5657
5658 md_number_to_chars (p, value, fixP->fx_size);
5659 }
5660 \f
5661 #define MAX_LITTLENUMS 6
5662
5663 /* Turn the string pointed to by litP into a floating point constant
5664 of type TYPE, and emit the appropriate bytes. The number of
5665 LITTLENUMS emitted is stored in *SIZEP. An error message is
5666 returned, or NULL on OK. */
5667
5668 char *
5669 md_atof (type, litP, sizeP)
5670 int type;
5671 char *litP;
5672 int *sizeP;
5673 {
5674 int prec;
5675 LITTLENUM_TYPE words[MAX_LITTLENUMS];
5676 LITTLENUM_TYPE *wordP;
5677 char *t;
5678
5679 switch (type)
5680 {
5681 case 'f':
5682 case 'F':
5683 prec = 2;
5684 break;
5685
5686 case 'd':
5687 case 'D':
5688 prec = 4;
5689 break;
5690
5691 case 'x':
5692 case 'X':
5693 prec = 5;
5694 break;
5695
5696 default:
5697 *sizeP = 0;
5698 return _("Bad call to md_atof ()");
5699 }
5700 t = atof_ieee (input_line_pointer, type, words);
5701 if (t)
5702 input_line_pointer = t;
5703
5704 *sizeP = prec * sizeof (LITTLENUM_TYPE);
5705 /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
5706 the bigendian 386. */
5707 for (wordP = words + prec - 1; prec--;)
5708 {
5709 md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
5710 litP += sizeof (LITTLENUM_TYPE);
5711 }
5712 return 0;
5713 }
5714 \f
5715 static char output_invalid_buf[sizeof (unsigned char) * 2 + 6];
5716
5717 static char *
5718 output_invalid (c)
5719 int c;
5720 {
5721 if (ISPRINT (c))
5722 snprintf (output_invalid_buf, sizeof (output_invalid_buf),
5723 "'%c'", c);
5724 else
5725 snprintf (output_invalid_buf, sizeof (output_invalid_buf),
5726 "(0x%x)", (unsigned char) c);
5727 return output_invalid_buf;
5728 }
5729
5730 /* REG_STRING starts *before* REGISTER_PREFIX. */
5731
5732 static const reg_entry *
5733 parse_real_register (char *reg_string, char **end_op)
5734 {
5735 char *s = reg_string;
5736 char *p;
5737 char reg_name_given[MAX_REG_NAME_SIZE + 1];
5738 const reg_entry *r;
5739
5740 /* Skip possible REGISTER_PREFIX and possible whitespace. */
5741 if (*s == REGISTER_PREFIX)
5742 ++s;
5743
5744 if (is_space_char (*s))
5745 ++s;
5746
5747 p = reg_name_given;
5748 while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
5749 {
5750 if (p >= reg_name_given + MAX_REG_NAME_SIZE)
5751 return (const reg_entry *) NULL;
5752 s++;
5753 }
5754
5755 /* For naked regs, make sure that we are not dealing with an identifier.
5756 This prevents confusing an identifier like `eax_var' with register
5757 `eax'. */
5758 if (allow_naked_reg && identifier_chars[(unsigned char) *s])
5759 return (const reg_entry *) NULL;
5760
5761 *end_op = s;
5762
5763 r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
5764
5765 /* Handle floating point regs, allowing spaces in the (i) part. */
5766 if (r == i386_regtab /* %st is first entry of table */)
5767 {
5768 if (is_space_char (*s))
5769 ++s;
5770 if (*s == '(')
5771 {
5772 ++s;
5773 if (is_space_char (*s))
5774 ++s;
5775 if (*s >= '0' && *s <= '7')
5776 {
5777 r = &i386_float_regtab[*s - '0'];
5778 ++s;
5779 if (is_space_char (*s))
5780 ++s;
5781 if (*s == ')')
5782 {
5783 *end_op = s + 1;
5784 return r;
5785 }
5786 }
5787 /* We have "%st(" then garbage. */
5788 return (const reg_entry *) NULL;
5789 }
5790 }
5791
5792 if (r != NULL
5793 && ((r->reg_flags & (RegRex64 | RegRex)) | (r->reg_type & Reg64)) != 0
5794 && (r->reg_type != Control || !(cpu_arch_flags & CpuSledgehammer))
5795 && flag_code != CODE_64BIT)
5796 return (const reg_entry *) NULL;
5797
5798 return r;
5799 }
5800
5801 /* REG_STRING starts *before* REGISTER_PREFIX. */
5802
5803 static const reg_entry *
5804 parse_register (char *reg_string, char **end_op)
5805 {
5806 const reg_entry *r;
5807
5808 if (*reg_string == REGISTER_PREFIX || allow_naked_reg)
5809 r = parse_real_register (reg_string, end_op);
5810 else
5811 r = NULL;
5812 if (!r)
5813 {
5814 char *save = input_line_pointer;
5815 char c;
5816 symbolS *symbolP;
5817
5818 input_line_pointer = reg_string;
5819 c = get_symbol_end ();
5820 symbolP = symbol_find (reg_string);
5821 if (symbolP && S_GET_SEGMENT (symbolP) == reg_section)
5822 {
5823 const expressionS *e = symbol_get_value_expression (symbolP);
5824
5825 know (e->X_op == O_register);
5826 know (e->X_add_number >= 0 && (valueT) e->X_add_number < ARRAY_SIZE (i386_regtab));
5827 r = i386_regtab + e->X_add_number;
5828 *end_op = input_line_pointer;
5829 }
5830 *input_line_pointer = c;
5831 input_line_pointer = save;
5832 }
5833 return r;
5834 }
5835
5836 int
5837 i386_parse_name (char *name, expressionS *e, char *nextcharP)
5838 {
5839 const reg_entry *r;
5840 char *end = input_line_pointer;
5841
5842 *end = *nextcharP;
5843 r = parse_register (name, &input_line_pointer);
5844 if (r && end <= input_line_pointer)
5845 {
5846 *nextcharP = *input_line_pointer;
5847 *input_line_pointer = 0;
5848 e->X_op = O_register;
5849 e->X_add_number = r - i386_regtab;
5850 return 1;
5851 }
5852 input_line_pointer = end;
5853 *end = 0;
5854 return 0;
5855 }
5856
5857 void
5858 md_operand (expressionS *e)
5859 {
5860 if (*input_line_pointer == REGISTER_PREFIX)
5861 {
5862 char *end;
5863 const reg_entry *r = parse_real_register (input_line_pointer, &end);
5864
5865 if (r)
5866 {
5867 e->X_op = O_register;
5868 e->X_add_number = r - i386_regtab;
5869 input_line_pointer = end;
5870 }
5871 }
5872 }
5873
5874 \f
5875 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5876 const char *md_shortopts = "kVQ:sqn";
5877 #else
5878 const char *md_shortopts = "qn";
5879 #endif
5880
5881 #define OPTION_32 (OPTION_MD_BASE + 0)
5882 #define OPTION_64 (OPTION_MD_BASE + 1)
5883 #define OPTION_DIVIDE (OPTION_MD_BASE + 2)
5884 #define OPTION_MARCH (OPTION_MD_BASE + 3)
5885 #define OPTION_MTUNE (OPTION_MD_BASE + 4)
5886
5887 struct option md_longopts[] =
5888 {
5889 {"32", no_argument, NULL, OPTION_32},
5890 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) || defined(TE_PEP)
5891 {"64", no_argument, NULL, OPTION_64},
5892 #endif
5893 {"divide", no_argument, NULL, OPTION_DIVIDE},
5894 {"march", required_argument, NULL, OPTION_MARCH},
5895 {"mtune", required_argument, NULL, OPTION_MTUNE},
5896 {NULL, no_argument, NULL, 0}
5897 };
5898 size_t md_longopts_size = sizeof (md_longopts);
5899
5900 int
5901 md_parse_option (int c, char *arg)
5902 {
5903 unsigned int i;
5904
5905 switch (c)
5906 {
5907 case 'n':
5908 optimize_align_code = 0;
5909 break;
5910
5911 case 'q':
5912 quiet_warnings = 1;
5913 break;
5914
5915 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5916 /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
5917 should be emitted or not. FIXME: Not implemented. */
5918 case 'Q':
5919 break;
5920
5921 /* -V: SVR4 argument to print version ID. */
5922 case 'V':
5923 print_version_id ();
5924 break;
5925
5926 /* -k: Ignore for FreeBSD compatibility. */
5927 case 'k':
5928 break;
5929
5930 case 's':
5931 /* -s: On i386 Solaris, this tells the native assembler to use
5932 .stab instead of .stab.excl. We always use .stab anyhow. */
5933 break;
5934 #endif
5935 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) || defined(TE_PEP)
5936 case OPTION_64:
5937 {
5938 const char **list, **l;
5939
5940 list = bfd_target_list ();
5941 for (l = list; *l != NULL; l++)
5942 if (CONST_STRNEQ (*l, "elf64-x86-64")
5943 || strcmp (*l, "coff-x86-64") == 0
5944 || strcmp (*l, "pe-x86-64") == 0
5945 || strcmp (*l, "pei-x86-64") == 0)
5946 {
5947 default_arch = "x86_64";
5948 break;
5949 }
5950 if (*l == NULL)
5951 as_fatal (_("No compiled in support for x86_64"));
5952 free (list);
5953 }
5954 break;
5955 #endif
5956
5957 case OPTION_32:
5958 default_arch = "i386";
5959 break;
5960
5961 case OPTION_DIVIDE:
5962 #ifdef SVR4_COMMENT_CHARS
5963 {
5964 char *n, *t;
5965 const char *s;
5966
5967 n = (char *) xmalloc (strlen (i386_comment_chars) + 1);
5968 t = n;
5969 for (s = i386_comment_chars; *s != '\0'; s++)
5970 if (*s != '/')
5971 *t++ = *s;
5972 *t = '\0';
5973 i386_comment_chars = n;
5974 }
5975 #endif
5976 break;
5977
5978 case OPTION_MARCH:
5979 if (*arg == '.')
5980 as_fatal (_("Invalid -march= option: `%s'"), arg);
5981 for (i = 0; i < ARRAY_SIZE (cpu_arch); i++)
5982 {
5983 if (strcmp (arg, cpu_arch [i].name) == 0)
5984 {
5985 cpu_arch_isa = cpu_arch[i].type;
5986 cpu_arch_isa_flags = cpu_arch[i].flags;
5987 if (!cpu_arch_tune_set)
5988 {
5989 cpu_arch_tune = cpu_arch_isa;
5990 cpu_arch_tune_flags = cpu_arch_isa_flags;
5991 }
5992 break;
5993 }
5994 }
5995 if (i >= ARRAY_SIZE (cpu_arch))
5996 as_fatal (_("Invalid -march= option: `%s'"), arg);
5997 break;
5998
5999 case OPTION_MTUNE:
6000 if (*arg == '.')
6001 as_fatal (_("Invalid -mtune= option: `%s'"), arg);
6002 for (i = 0; i < ARRAY_SIZE (cpu_arch); i++)
6003 {
6004 if (strcmp (arg, cpu_arch [i].name) == 0)
6005 {
6006 cpu_arch_tune_set = 1;
6007 cpu_arch_tune = cpu_arch [i].type;
6008 cpu_arch_tune_flags = cpu_arch[i].flags;
6009 break;
6010 }
6011 }
6012 if (i >= ARRAY_SIZE (cpu_arch))
6013 as_fatal (_("Invalid -mtune= option: `%s'"), arg);
6014 break;
6015
6016 default:
6017 return 0;
6018 }
6019 return 1;
6020 }
6021
6022 void
6023 md_show_usage (stream)
6024 FILE *stream;
6025 {
6026 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
6027 fprintf (stream, _("\
6028 -Q ignored\n\
6029 -V print assembler version number\n\
6030 -k ignored\n"));
6031 #endif
6032 fprintf (stream, _("\
6033 -n Do not optimize code alignment\n\
6034 -q quieten some warnings\n"));
6035 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
6036 fprintf (stream, _("\
6037 -s ignored\n"));
6038 #endif
6039 #ifdef SVR4_COMMENT_CHARS
6040 fprintf (stream, _("\
6041 --divide do not treat `/' as a comment character\n"));
6042 #else
6043 fprintf (stream, _("\
6044 --divide ignored\n"));
6045 #endif
6046 fprintf (stream, _("\
6047 -march=CPU/-mtune=CPU generate code/optimize for CPU, where CPU is one of:\n\
6048 i386, i486, pentium, pentiumpro, pentium4, nocona,\n\
6049 core, core2, k6, athlon, k8, generic32, generic64\n"));
6050
6051 }
6052
6053 #if defined(TE_PEP)
6054 const char *
6055 x86_64_target_format (void)
6056 {
6057 if (strcmp (default_arch, "x86_64") == 0)
6058 {
6059 set_code_flag (CODE_64BIT);
6060 return COFF_TARGET_FORMAT;
6061 }
6062 else if (strcmp (default_arch, "i386") == 0)
6063 {
6064 set_code_flag (CODE_32BIT);
6065 return "coff-i386";
6066 }
6067
6068 as_fatal (_("Unknown architecture"));
6069 return NULL;
6070 }
6071 #endif
6072
6073 #if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
6074 || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
6075
6076 /* Pick the target format to use. */
6077
6078 const char *
6079 i386_target_format ()
6080 {
6081 if (!strcmp (default_arch, "x86_64"))
6082 {
6083 set_code_flag (CODE_64BIT);
6084 if (cpu_arch_isa_flags == 0)
6085 cpu_arch_isa_flags = Cpu186|Cpu286|Cpu386|Cpu486
6086 |Cpu586|Cpu686|CpuP4|CpuMMX|CpuMMX2
6087 |CpuSSE|CpuSSE2;
6088 if (cpu_arch_tune_flags == 0)
6089 cpu_arch_tune_flags = Cpu186|Cpu286|Cpu386|Cpu486
6090 |Cpu586|Cpu686|CpuP4|CpuMMX|CpuMMX2
6091 |CpuSSE|CpuSSE2;
6092 }
6093 else if (!strcmp (default_arch, "i386"))
6094 {
6095 set_code_flag (CODE_32BIT);
6096 if (cpu_arch_isa_flags == 0)
6097 cpu_arch_isa_flags = Cpu186|Cpu286|Cpu386;
6098 if (cpu_arch_tune_flags == 0)
6099 cpu_arch_tune_flags = Cpu186|Cpu286|Cpu386;
6100 }
6101 else
6102 as_fatal (_("Unknown architecture"));
6103 switch (OUTPUT_FLAVOR)
6104 {
6105 #ifdef OBJ_MAYBE_AOUT
6106 case bfd_target_aout_flavour:
6107 return AOUT_TARGET_FORMAT;
6108 #endif
6109 #ifdef OBJ_MAYBE_COFF
6110 case bfd_target_coff_flavour:
6111 return "coff-i386";
6112 #endif
6113 #if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF)
6114 case bfd_target_elf_flavour:
6115 {
6116 if (flag_code == CODE_64BIT)
6117 {
6118 object_64bit = 1;
6119 use_rela_relocations = 1;
6120 }
6121 return flag_code == CODE_64BIT ? ELF_TARGET_FORMAT64 : ELF_TARGET_FORMAT;
6122 }
6123 #endif
6124 default:
6125 abort ();
6126 return NULL;
6127 }
6128 }
6129
6130 #endif /* OBJ_MAYBE_ more than one */
6131
6132 #if (defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
6133 void i386_elf_emit_arch_note ()
6134 {
6135 if (IS_ELF && cpu_arch_name != NULL)
6136 {
6137 char *p;
6138 asection *seg = now_seg;
6139 subsegT subseg = now_subseg;
6140 Elf_Internal_Note i_note;
6141 Elf_External_Note e_note;
6142 asection *note_secp;
6143 int len;
6144
6145 /* Create the .note section. */
6146 note_secp = subseg_new (".note", 0);
6147 bfd_set_section_flags (stdoutput,
6148 note_secp,
6149 SEC_HAS_CONTENTS | SEC_READONLY);
6150
6151 /* Process the arch string. */
6152 len = strlen (cpu_arch_name);
6153
6154 i_note.namesz = len + 1;
6155 i_note.descsz = 0;
6156 i_note.type = NT_ARCH;
6157 p = frag_more (sizeof (e_note.namesz));
6158 md_number_to_chars (p, (valueT) i_note.namesz, sizeof (e_note.namesz));
6159 p = frag_more (sizeof (e_note.descsz));
6160 md_number_to_chars (p, (valueT) i_note.descsz, sizeof (e_note.descsz));
6161 p = frag_more (sizeof (e_note.type));
6162 md_number_to_chars (p, (valueT) i_note.type, sizeof (e_note.type));
6163 p = frag_more (len + 1);
6164 strcpy (p, cpu_arch_name);
6165
6166 frag_align (2, 0, 0);
6167
6168 subseg_set (seg, subseg);
6169 }
6170 }
6171 #endif
6172 \f
6173 symbolS *
6174 md_undefined_symbol (name)
6175 char *name;
6176 {
6177 if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
6178 && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
6179 && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
6180 && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
6181 {
6182 if (!GOT_symbol)
6183 {
6184 if (symbol_find (name))
6185 as_bad (_("GOT already in symbol table"));
6186 GOT_symbol = symbol_new (name, undefined_section,
6187 (valueT) 0, &zero_address_frag);
6188 };
6189 return GOT_symbol;
6190 }
6191 return 0;
6192 }
6193
6194 /* Round up a section size to the appropriate boundary. */
6195
6196 valueT
6197 md_section_align (segment, size)
6198 segT segment ATTRIBUTE_UNUSED;
6199 valueT size;
6200 {
6201 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
6202 if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
6203 {
6204 /* For a.out, force the section size to be aligned. If we don't do
6205 this, BFD will align it for us, but it will not write out the
6206 final bytes of the section. This may be a bug in BFD, but it is
6207 easier to fix it here since that is how the other a.out targets
6208 work. */
6209 int align;
6210
6211 align = bfd_get_section_alignment (stdoutput, segment);
6212 size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
6213 }
6214 #endif
6215
6216 return size;
6217 }
6218
6219 /* On the i386, PC-relative offsets are relative to the start of the
6220 next instruction. That is, the address of the offset, plus its
6221 size, since the offset is always the last part of the insn. */
6222
6223 long
6224 md_pcrel_from (fixP)
6225 fixS *fixP;
6226 {
6227 return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
6228 }
6229
6230 #ifndef I386COFF
6231
6232 static void
6233 s_bss (ignore)
6234 int ignore ATTRIBUTE_UNUSED;
6235 {
6236 int temp;
6237
6238 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
6239 if (IS_ELF)
6240 obj_elf_section_change_hook ();
6241 #endif
6242 temp = get_absolute_expression ();
6243 subseg_set (bss_section, (subsegT) temp);
6244 demand_empty_rest_of_line ();
6245 }
6246
6247 #endif
6248
6249 void
6250 i386_validate_fix (fixp)
6251 fixS *fixp;
6252 {
6253 if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
6254 {
6255 if (fixp->fx_r_type == BFD_RELOC_32_PCREL)
6256 {
6257 if (!object_64bit)
6258 abort ();
6259 fixp->fx_r_type = BFD_RELOC_X86_64_GOTPCREL;
6260 }
6261 else
6262 {
6263 if (!object_64bit)
6264 fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
6265 else
6266 fixp->fx_r_type = BFD_RELOC_X86_64_GOTOFF64;
6267 }
6268 fixp->fx_subsy = 0;
6269 }
6270 }
6271
6272 arelent *
6273 tc_gen_reloc (section, fixp)
6274 asection *section ATTRIBUTE_UNUSED;
6275 fixS *fixp;
6276 {
6277 arelent *rel;
6278 bfd_reloc_code_real_type code;
6279
6280 switch (fixp->fx_r_type)
6281 {
6282 case BFD_RELOC_X86_64_PLT32:
6283 case BFD_RELOC_X86_64_GOT32:
6284 case BFD_RELOC_X86_64_GOTPCREL:
6285 case BFD_RELOC_386_PLT32:
6286 case BFD_RELOC_386_GOT32:
6287 case BFD_RELOC_386_GOTOFF:
6288 case BFD_RELOC_386_GOTPC:
6289 case BFD_RELOC_386_TLS_GD:
6290 case BFD_RELOC_386_TLS_LDM:
6291 case BFD_RELOC_386_TLS_LDO_32:
6292 case BFD_RELOC_386_TLS_IE_32:
6293 case BFD_RELOC_386_TLS_IE:
6294 case BFD_RELOC_386_TLS_GOTIE:
6295 case BFD_RELOC_386_TLS_LE_32:
6296 case BFD_RELOC_386_TLS_LE:
6297 case BFD_RELOC_386_TLS_GOTDESC:
6298 case BFD_RELOC_386_TLS_DESC_CALL:
6299 case BFD_RELOC_X86_64_TLSGD:
6300 case BFD_RELOC_X86_64_TLSLD:
6301 case BFD_RELOC_X86_64_DTPOFF32:
6302 case BFD_RELOC_X86_64_DTPOFF64:
6303 case BFD_RELOC_X86_64_GOTTPOFF:
6304 case BFD_RELOC_X86_64_TPOFF32:
6305 case BFD_RELOC_X86_64_TPOFF64:
6306 case BFD_RELOC_X86_64_GOTOFF64:
6307 case BFD_RELOC_X86_64_GOTPC32:
6308 case BFD_RELOC_X86_64_GOT64:
6309 case BFD_RELOC_X86_64_GOTPCREL64:
6310 case BFD_RELOC_X86_64_GOTPC64:
6311 case BFD_RELOC_X86_64_GOTPLT64:
6312 case BFD_RELOC_X86_64_PLTOFF64:
6313 case BFD_RELOC_X86_64_GOTPC32_TLSDESC:
6314 case BFD_RELOC_X86_64_TLSDESC_CALL:
6315 case BFD_RELOC_RVA:
6316 case BFD_RELOC_VTABLE_ENTRY:
6317 case BFD_RELOC_VTABLE_INHERIT:
6318 #ifdef TE_PE
6319 case BFD_RELOC_32_SECREL:
6320 #endif
6321 code = fixp->fx_r_type;
6322 break;
6323 case BFD_RELOC_X86_64_32S:
6324 if (!fixp->fx_pcrel)
6325 {
6326 /* Don't turn BFD_RELOC_X86_64_32S into BFD_RELOC_32. */
6327 code = fixp->fx_r_type;
6328 break;
6329 }
6330 default:
6331 if (fixp->fx_pcrel)
6332 {
6333 switch (fixp->fx_size)
6334 {
6335 default:
6336 as_bad_where (fixp->fx_file, fixp->fx_line,
6337 _("can not do %d byte pc-relative relocation"),
6338 fixp->fx_size);
6339 code = BFD_RELOC_32_PCREL;
6340 break;
6341 case 1: code = BFD_RELOC_8_PCREL; break;
6342 case 2: code = BFD_RELOC_16_PCREL; break;
6343 case 4: code = BFD_RELOC_32_PCREL; break;
6344 #ifdef BFD64
6345 case 8: code = BFD_RELOC_64_PCREL; break;
6346 #endif
6347 }
6348 }
6349 else
6350 {
6351 switch (fixp->fx_size)
6352 {
6353 default:
6354 as_bad_where (fixp->fx_file, fixp->fx_line,
6355 _("can not do %d byte relocation"),
6356 fixp->fx_size);
6357 code = BFD_RELOC_32;
6358 break;
6359 case 1: code = BFD_RELOC_8; break;
6360 case 2: code = BFD_RELOC_16; break;
6361 case 4: code = BFD_RELOC_32; break;
6362 #ifdef BFD64
6363 case 8: code = BFD_RELOC_64; break;
6364 #endif
6365 }
6366 }
6367 break;
6368 }
6369
6370 if ((code == BFD_RELOC_32
6371 || code == BFD_RELOC_32_PCREL
6372 || code == BFD_RELOC_X86_64_32S)
6373 && GOT_symbol
6374 && fixp->fx_addsy == GOT_symbol)
6375 {
6376 if (!object_64bit)
6377 code = BFD_RELOC_386_GOTPC;
6378 else
6379 code = BFD_RELOC_X86_64_GOTPC32;
6380 }
6381 if ((code == BFD_RELOC_64 || code == BFD_RELOC_64_PCREL)
6382 && GOT_symbol
6383 && fixp->fx_addsy == GOT_symbol)
6384 {
6385 code = BFD_RELOC_X86_64_GOTPC64;
6386 }
6387
6388 rel = (arelent *) xmalloc (sizeof (arelent));
6389 rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
6390 *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
6391
6392 rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
6393
6394 if (!use_rela_relocations)
6395 {
6396 /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
6397 vtable entry to be used in the relocation's section offset. */
6398 if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
6399 rel->address = fixp->fx_offset;
6400
6401 rel->addend = 0;
6402 }
6403 /* Use the rela in 64bit mode. */
6404 else
6405 {
6406 if (!fixp->fx_pcrel)
6407 rel->addend = fixp->fx_offset;
6408 else
6409 switch (code)
6410 {
6411 case BFD_RELOC_X86_64_PLT32:
6412 case BFD_RELOC_X86_64_GOT32:
6413 case BFD_RELOC_X86_64_GOTPCREL:
6414 case BFD_RELOC_X86_64_TLSGD:
6415 case BFD_RELOC_X86_64_TLSLD:
6416 case BFD_RELOC_X86_64_GOTTPOFF:
6417 case BFD_RELOC_X86_64_GOTPC32_TLSDESC:
6418 case BFD_RELOC_X86_64_TLSDESC_CALL:
6419 rel->addend = fixp->fx_offset - fixp->fx_size;
6420 break;
6421 default:
6422 rel->addend = (section->vma
6423 - fixp->fx_size
6424 + fixp->fx_addnumber
6425 + md_pcrel_from (fixp));
6426 break;
6427 }
6428 }
6429
6430 rel->howto = bfd_reloc_type_lookup (stdoutput, code);
6431 if (rel->howto == NULL)
6432 {
6433 as_bad_where (fixp->fx_file, fixp->fx_line,
6434 _("cannot represent relocation type %s"),
6435 bfd_get_reloc_code_name (code));
6436 /* Set howto to a garbage value so that we can keep going. */
6437 rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
6438 assert (rel->howto != NULL);
6439 }
6440
6441 return rel;
6442 }
6443
6444 \f
6445 /* Parse operands using Intel syntax. This implements a recursive descent
6446 parser based on the BNF grammar published in Appendix B of the MASM 6.1
6447 Programmer's Guide.
6448
6449 FIXME: We do not recognize the full operand grammar defined in the MASM
6450 documentation. In particular, all the structure/union and
6451 high-level macro operands are missing.
6452
6453 Uppercase words are terminals, lower case words are non-terminals.
6454 Objects surrounded by double brackets '[[' ']]' are optional. Vertical
6455 bars '|' denote choices. Most grammar productions are implemented in
6456 functions called 'intel_<production>'.
6457
6458 Initial production is 'expr'.
6459
6460 addOp + | -
6461
6462 alpha [a-zA-Z]
6463
6464 binOp & | AND | \| | OR | ^ | XOR
6465
6466 byteRegister AL | AH | BL | BH | CL | CH | DL | DH
6467
6468 constant digits [[ radixOverride ]]
6469
6470 dataType BYTE | WORD | DWORD | FWORD | QWORD | TBYTE | OWORD | XMMWORD
6471
6472 digits decdigit
6473 | digits decdigit
6474 | digits hexdigit
6475
6476 decdigit [0-9]
6477
6478 e04 e04 addOp e05
6479 | e05
6480
6481 e05 e05 binOp e06
6482 | e06
6483
6484 e06 e06 mulOp e09
6485 | e09
6486
6487 e09 OFFSET e10
6488 | SHORT e10
6489 | + e10
6490 | - e10
6491 | ~ e10
6492 | NOT e10
6493 | e09 PTR e10
6494 | e09 : e10
6495 | e10
6496
6497 e10 e10 [ expr ]
6498 | e11
6499
6500 e11 ( expr )
6501 | [ expr ]
6502 | constant
6503 | dataType
6504 | id
6505 | $
6506 | register
6507
6508 => expr expr cmpOp e04
6509 | e04
6510
6511 gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
6512 | BP | EBP | SP | ESP | DI | EDI | SI | ESI
6513
6514 hexdigit a | b | c | d | e | f
6515 | A | B | C | D | E | F
6516
6517 id alpha
6518 | id alpha
6519 | id decdigit
6520
6521 mulOp * | / | % | MOD | << | SHL | >> | SHR
6522
6523 quote " | '
6524
6525 register specialRegister
6526 | gpRegister
6527 | byteRegister
6528
6529 segmentRegister CS | DS | ES | FS | GS | SS
6530
6531 specialRegister CR0 | CR2 | CR3 | CR4
6532 | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
6533 | TR3 | TR4 | TR5 | TR6 | TR7
6534
6535 We simplify the grammar in obvious places (e.g., register parsing is
6536 done by calling parse_register) and eliminate immediate left recursion
6537 to implement a recursive-descent parser.
6538
6539 expr e04 expr'
6540
6541 expr' cmpOp e04 expr'
6542 | Empty
6543
6544 e04 e05 e04'
6545
6546 e04' addOp e05 e04'
6547 | Empty
6548
6549 e05 e06 e05'
6550
6551 e05' binOp e06 e05'
6552 | Empty
6553
6554 e06 e09 e06'
6555
6556 e06' mulOp e09 e06'
6557 | Empty
6558
6559 e09 OFFSET e10 e09'
6560 | SHORT e10'
6561 | + e10'
6562 | - e10'
6563 | ~ e10'
6564 | NOT e10'
6565 | e10 e09'
6566
6567 e09' PTR e10 e09'
6568 | : e10 e09'
6569 | Empty
6570
6571 e10 e11 e10'
6572
6573 e10' [ expr ] e10'
6574 | Empty
6575
6576 e11 ( expr )
6577 | [ expr ]
6578 | BYTE
6579 | WORD
6580 | DWORD
6581 | FWORD
6582 | QWORD
6583 | TBYTE
6584 | OWORD
6585 | XMMWORD
6586 | .
6587 | $
6588 | register
6589 | id
6590 | constant */
6591
6592 /* Parsing structure for the intel syntax parser. Used to implement the
6593 semantic actions for the operand grammar. */
6594 struct intel_parser_s
6595 {
6596 char *op_string; /* The string being parsed. */
6597 int got_a_float; /* Whether the operand is a float. */
6598 int op_modifier; /* Operand modifier. */
6599 int is_mem; /* 1 if operand is memory reference. */
6600 int in_offset; /* >=1 if parsing operand of offset. */
6601 int in_bracket; /* >=1 if parsing operand in brackets. */
6602 const reg_entry *reg; /* Last register reference found. */
6603 char *disp; /* Displacement string being built. */
6604 char *next_operand; /* Resume point when splitting operands. */
6605 };
6606
6607 static struct intel_parser_s intel_parser;
6608
6609 /* Token structure for parsing intel syntax. */
6610 struct intel_token
6611 {
6612 int code; /* Token code. */
6613 const reg_entry *reg; /* Register entry for register tokens. */
6614 char *str; /* String representation. */
6615 };
6616
6617 static struct intel_token cur_token, prev_token;
6618
6619 /* Token codes for the intel parser. Since T_SHORT is already used
6620 by COFF, undefine it first to prevent a warning. */
6621 #define T_NIL -1
6622 #define T_CONST 1
6623 #define T_REG 2
6624 #define T_BYTE 3
6625 #define T_WORD 4
6626 #define T_DWORD 5
6627 #define T_FWORD 6
6628 #define T_QWORD 7
6629 #define T_TBYTE 8
6630 #define T_XMMWORD 9
6631 #undef T_SHORT
6632 #define T_SHORT 10
6633 #define T_OFFSET 11
6634 #define T_PTR 12
6635 #define T_ID 13
6636 #define T_SHL 14
6637 #define T_SHR 15
6638
6639 /* Prototypes for intel parser functions. */
6640 static int intel_match_token PARAMS ((int code));
6641 static void intel_get_token PARAMS ((void));
6642 static void intel_putback_token PARAMS ((void));
6643 static int intel_expr PARAMS ((void));
6644 static int intel_e04 PARAMS ((void));
6645 static int intel_e05 PARAMS ((void));
6646 static int intel_e06 PARAMS ((void));
6647 static int intel_e09 PARAMS ((void));
6648 static int intel_bracket_expr PARAMS ((void));
6649 static int intel_e10 PARAMS ((void));
6650 static int intel_e11 PARAMS ((void));
6651
6652 static int
6653 i386_intel_operand (operand_string, got_a_float)
6654 char *operand_string;
6655 int got_a_float;
6656 {
6657 int ret;
6658 char *p;
6659
6660 p = intel_parser.op_string = xstrdup (operand_string);
6661 intel_parser.disp = (char *) xmalloc (strlen (operand_string) + 1);
6662
6663 for (;;)
6664 {
6665 /* Initialize token holders. */
6666 cur_token.code = prev_token.code = T_NIL;
6667 cur_token.reg = prev_token.reg = NULL;
6668 cur_token.str = prev_token.str = NULL;
6669
6670 /* Initialize parser structure. */
6671 intel_parser.got_a_float = got_a_float;
6672 intel_parser.op_modifier = 0;
6673 intel_parser.is_mem = 0;
6674 intel_parser.in_offset = 0;
6675 intel_parser.in_bracket = 0;
6676 intel_parser.reg = NULL;
6677 intel_parser.disp[0] = '\0';
6678 intel_parser.next_operand = NULL;
6679
6680 /* Read the first token and start the parser. */
6681 intel_get_token ();
6682 ret = intel_expr ();
6683
6684 if (!ret)
6685 break;
6686
6687 if (cur_token.code != T_NIL)
6688 {
6689 as_bad (_("invalid operand for '%s' ('%s' unexpected)"),
6690 current_templates->start->name, cur_token.str);
6691 ret = 0;
6692 }
6693 /* If we found a memory reference, hand it over to i386_displacement
6694 to fill in the rest of the operand fields. */
6695 else if (intel_parser.is_mem)
6696 {
6697 if ((i.mem_operands == 1
6698 && (current_templates->start->opcode_modifier & IsString) == 0)
6699 || i.mem_operands == 2)
6700 {
6701 as_bad (_("too many memory references for '%s'"),
6702 current_templates->start->name);
6703 ret = 0;
6704 }
6705 else
6706 {
6707 char *s = intel_parser.disp;
6708 i.mem_operands++;
6709
6710 if (!quiet_warnings && intel_parser.is_mem < 0)
6711 /* See the comments in intel_bracket_expr. */
6712 as_warn (_("Treating `%s' as memory reference"), operand_string);
6713
6714 /* Add the displacement expression. */
6715 if (*s != '\0')
6716 ret = i386_displacement (s, s + strlen (s));
6717 if (ret)
6718 {
6719 /* Swap base and index in 16-bit memory operands like
6720 [si+bx]. Since i386_index_check is also used in AT&T
6721 mode we have to do that here. */
6722 if (i.base_reg
6723 && i.index_reg
6724 && (i.base_reg->reg_type & Reg16)
6725 && (i.index_reg->reg_type & Reg16)
6726 && i.base_reg->reg_num >= 6
6727 && i.index_reg->reg_num < 6)
6728 {
6729 const reg_entry *base = i.index_reg;
6730
6731 i.index_reg = i.base_reg;
6732 i.base_reg = base;
6733 }
6734 ret = i386_index_check (operand_string);
6735 }
6736 }
6737 }
6738
6739 /* Constant and OFFSET expressions are handled by i386_immediate. */
6740 else if ((intel_parser.op_modifier & (1 << T_OFFSET))
6741 || intel_parser.reg == NULL)
6742 ret = i386_immediate (intel_parser.disp);
6743
6744 if (intel_parser.next_operand && this_operand >= MAX_OPERANDS - 1)
6745 ret = 0;
6746 if (!ret || !intel_parser.next_operand)
6747 break;
6748 intel_parser.op_string = intel_parser.next_operand;
6749 this_operand = i.operands++;
6750 }
6751
6752 free (p);
6753 free (intel_parser.disp);
6754
6755 return ret;
6756 }
6757
6758 #define NUM_ADDRESS_REGS (!!i.base_reg + !!i.index_reg)
6759
6760 /* expr e04 expr'
6761
6762 expr' cmpOp e04 expr'
6763 | Empty */
6764 static int
6765 intel_expr ()
6766 {
6767 /* XXX Implement the comparison operators. */
6768 return intel_e04 ();
6769 }
6770
6771 /* e04 e05 e04'
6772
6773 e04' addOp e05 e04'
6774 | Empty */
6775 static int
6776 intel_e04 ()
6777 {
6778 int nregs = -1;
6779
6780 for (;;)
6781 {
6782 if (!intel_e05())
6783 return 0;
6784
6785 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
6786 i.base_reg = i386_regtab + REGNAM_AL; /* al is invalid as base */
6787
6788 if (cur_token.code == '+')
6789 nregs = -1;
6790 else if (cur_token.code == '-')
6791 nregs = NUM_ADDRESS_REGS;
6792 else
6793 return 1;
6794
6795 strcat (intel_parser.disp, cur_token.str);
6796 intel_match_token (cur_token.code);
6797 }
6798 }
6799
6800 /* e05 e06 e05'
6801
6802 e05' binOp e06 e05'
6803 | Empty */
6804 static int
6805 intel_e05 ()
6806 {
6807 int nregs = ~NUM_ADDRESS_REGS;
6808
6809 for (;;)
6810 {
6811 if (!intel_e06())
6812 return 0;
6813
6814 if (cur_token.code == '&' || cur_token.code == '|' || cur_token.code == '^')
6815 {
6816 char str[2];
6817
6818 str[0] = cur_token.code;
6819 str[1] = 0;
6820 strcat (intel_parser.disp, str);
6821 }
6822 else
6823 break;
6824
6825 intel_match_token (cur_token.code);
6826
6827 if (nregs < 0)
6828 nregs = ~nregs;
6829 }
6830 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
6831 i.base_reg = i386_regtab + REGNAM_AL + 1; /* cl is invalid as base */
6832 return 1;
6833 }
6834
6835 /* e06 e09 e06'
6836
6837 e06' mulOp e09 e06'
6838 | Empty */
6839 static int
6840 intel_e06 ()
6841 {
6842 int nregs = ~NUM_ADDRESS_REGS;
6843
6844 for (;;)
6845 {
6846 if (!intel_e09())
6847 return 0;
6848
6849 if (cur_token.code == '*' || cur_token.code == '/' || cur_token.code == '%')
6850 {
6851 char str[2];
6852
6853 str[0] = cur_token.code;
6854 str[1] = 0;
6855 strcat (intel_parser.disp, str);
6856 }
6857 else if (cur_token.code == T_SHL)
6858 strcat (intel_parser.disp, "<<");
6859 else if (cur_token.code == T_SHR)
6860 strcat (intel_parser.disp, ">>");
6861 else
6862 break;
6863
6864 intel_match_token (cur_token.code);
6865
6866 if (nregs < 0)
6867 nregs = ~nregs;
6868 }
6869 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
6870 i.base_reg = i386_regtab + REGNAM_AL + 2; /* dl is invalid as base */
6871 return 1;
6872 }
6873
6874 /* e09 OFFSET e09
6875 | SHORT e09
6876 | + e09
6877 | - e09
6878 | ~ e09
6879 | NOT e09
6880 | e10 e09'
6881
6882 e09' PTR e10 e09'
6883 | : e10 e09'
6884 | Empty */
6885 static int
6886 intel_e09 ()
6887 {
6888 int nregs = ~NUM_ADDRESS_REGS;
6889 int in_offset = 0;
6890
6891 for (;;)
6892 {
6893 /* Don't consume constants here. */
6894 if (cur_token.code == '+' || cur_token.code == '-')
6895 {
6896 /* Need to look one token ahead - if the next token
6897 is a constant, the current token is its sign. */
6898 int next_code;
6899
6900 intel_match_token (cur_token.code);
6901 next_code = cur_token.code;
6902 intel_putback_token ();
6903 if (next_code == T_CONST)
6904 break;
6905 }
6906
6907 /* e09 OFFSET e09 */
6908 if (cur_token.code == T_OFFSET)
6909 {
6910 if (!in_offset++)
6911 ++intel_parser.in_offset;
6912 }
6913
6914 /* e09 SHORT e09 */
6915 else if (cur_token.code == T_SHORT)
6916 intel_parser.op_modifier |= 1 << T_SHORT;
6917
6918 /* e09 + e09 */
6919 else if (cur_token.code == '+')
6920 strcat (intel_parser.disp, "+");
6921
6922 /* e09 - e09
6923 | ~ e09
6924 | NOT e09 */
6925 else if (cur_token.code == '-' || cur_token.code == '~')
6926 {
6927 char str[2];
6928
6929 if (nregs < 0)
6930 nregs = ~nregs;
6931 str[0] = cur_token.code;
6932 str[1] = 0;
6933 strcat (intel_parser.disp, str);
6934 }
6935
6936 /* e09 e10 e09' */
6937 else
6938 break;
6939
6940 intel_match_token (cur_token.code);
6941 }
6942
6943 for (;;)
6944 {
6945 if (!intel_e10 ())
6946 return 0;
6947
6948 /* e09' PTR e10 e09' */
6949 if (cur_token.code == T_PTR)
6950 {
6951 char suffix;
6952
6953 if (prev_token.code == T_BYTE)
6954 suffix = BYTE_MNEM_SUFFIX;
6955
6956 else if (prev_token.code == T_WORD)
6957 {
6958 if (current_templates->start->name[0] == 'l'
6959 && current_templates->start->name[2] == 's'
6960 && current_templates->start->name[3] == 0)
6961 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
6962 else if (intel_parser.got_a_float == 2) /* "fi..." */
6963 suffix = SHORT_MNEM_SUFFIX;
6964 else
6965 suffix = WORD_MNEM_SUFFIX;
6966 }
6967
6968 else if (prev_token.code == T_DWORD)
6969 {
6970 if (current_templates->start->name[0] == 'l'
6971 && current_templates->start->name[2] == 's'
6972 && current_templates->start->name[3] == 0)
6973 suffix = WORD_MNEM_SUFFIX;
6974 else if (flag_code == CODE_16BIT
6975 && (current_templates->start->opcode_modifier
6976 & (Jump | JumpDword)))
6977 suffix = LONG_DOUBLE_MNEM_SUFFIX;
6978 else if (intel_parser.got_a_float == 1) /* "f..." */
6979 suffix = SHORT_MNEM_SUFFIX;
6980 else
6981 suffix = LONG_MNEM_SUFFIX;
6982 }
6983
6984 else if (prev_token.code == T_FWORD)
6985 {
6986 if (current_templates->start->name[0] == 'l'
6987 && current_templates->start->name[2] == 's'
6988 && current_templates->start->name[3] == 0)
6989 suffix = LONG_MNEM_SUFFIX;
6990 else if (!intel_parser.got_a_float)
6991 {
6992 if (flag_code == CODE_16BIT)
6993 add_prefix (DATA_PREFIX_OPCODE);
6994 suffix = LONG_DOUBLE_MNEM_SUFFIX;
6995 }
6996 else
6997 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
6998 }
6999
7000 else if (prev_token.code == T_QWORD)
7001 {
7002 if (intel_parser.got_a_float == 1) /* "f..." */
7003 suffix = LONG_MNEM_SUFFIX;
7004 else
7005 suffix = QWORD_MNEM_SUFFIX;
7006 }
7007
7008 else if (prev_token.code == T_TBYTE)
7009 {
7010 if (intel_parser.got_a_float == 1)
7011 suffix = LONG_DOUBLE_MNEM_SUFFIX;
7012 else
7013 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
7014 }
7015
7016 else if (prev_token.code == T_XMMWORD)
7017 {
7018 /* XXX ignored for now, but accepted since gcc uses it */
7019 suffix = 0;
7020 }
7021
7022 else
7023 {
7024 as_bad (_("Unknown operand modifier `%s'"), prev_token.str);
7025 return 0;
7026 }
7027
7028 /* Operands for jump/call using 'ptr' notation denote absolute
7029 addresses. */
7030 if (current_templates->start->opcode_modifier & (Jump | JumpDword))
7031 i.types[this_operand] |= JumpAbsolute;
7032
7033 if (current_templates->start->base_opcode == 0x8d /* lea */)
7034 ;
7035 else if (!i.suffix)
7036 i.suffix = suffix;
7037 else if (i.suffix != suffix)
7038 {
7039 as_bad (_("Conflicting operand modifiers"));
7040 return 0;
7041 }
7042
7043 }
7044
7045 /* e09' : e10 e09' */
7046 else if (cur_token.code == ':')
7047 {
7048 if (prev_token.code != T_REG)
7049 {
7050 /* While {call,jmp} SSSS:OOOO is MASM syntax only when SSSS is a
7051 segment/group identifier (which we don't have), using comma
7052 as the operand separator there is even less consistent, since
7053 there all branches only have a single operand. */
7054 if (this_operand != 0
7055 || intel_parser.in_offset
7056 || intel_parser.in_bracket
7057 || (!(current_templates->start->opcode_modifier
7058 & (Jump|JumpDword|JumpInterSegment))
7059 && !(current_templates->start->operand_types[0]
7060 & JumpAbsolute)))
7061 return intel_match_token (T_NIL);
7062 /* Remember the start of the 2nd operand and terminate 1st
7063 operand here.
7064 XXX This isn't right, yet (when SSSS:OOOO is right operand of
7065 another expression), but it gets at least the simplest case
7066 (a plain number or symbol on the left side) right. */
7067 intel_parser.next_operand = intel_parser.op_string;
7068 *--intel_parser.op_string = '\0';
7069 return intel_match_token (':');
7070 }
7071 }
7072
7073 /* e09' Empty */
7074 else
7075 break;
7076
7077 intel_match_token (cur_token.code);
7078
7079 }
7080
7081 if (in_offset)
7082 {
7083 --intel_parser.in_offset;
7084 if (nregs < 0)
7085 nregs = ~nregs;
7086 if (NUM_ADDRESS_REGS > nregs)
7087 {
7088 as_bad (_("Invalid operand to `OFFSET'"));
7089 return 0;
7090 }
7091 intel_parser.op_modifier |= 1 << T_OFFSET;
7092 }
7093
7094 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
7095 i.base_reg = i386_regtab + REGNAM_AL + 3; /* bl is invalid as base */
7096 return 1;
7097 }
7098
7099 static int
7100 intel_bracket_expr ()
7101 {
7102 int was_offset = intel_parser.op_modifier & (1 << T_OFFSET);
7103 const char *start = intel_parser.op_string;
7104 int len;
7105
7106 if (i.op[this_operand].regs)
7107 return intel_match_token (T_NIL);
7108
7109 intel_match_token ('[');
7110
7111 /* Mark as a memory operand only if it's not already known to be an
7112 offset expression. If it's an offset expression, we need to keep
7113 the brace in. */
7114 if (!intel_parser.in_offset)
7115 {
7116 ++intel_parser.in_bracket;
7117
7118 /* Operands for jump/call inside brackets denote absolute addresses. */
7119 if (current_templates->start->opcode_modifier & (Jump | JumpDword))
7120 i.types[this_operand] |= JumpAbsolute;
7121
7122 /* Unfortunately gas always diverged from MASM in a respect that can't
7123 be easily fixed without risking to break code sequences likely to be
7124 encountered (the testsuite even check for this): MASM doesn't consider
7125 an expression inside brackets unconditionally as a memory reference.
7126 When that is e.g. a constant, an offset expression, or the sum of the
7127 two, this is still taken as a constant load. gas, however, always
7128 treated these as memory references. As a compromise, we'll try to make
7129 offset expressions inside brackets work the MASM way (since that's
7130 less likely to be found in real world code), but make constants alone
7131 continue to work the traditional gas way. In either case, issue a
7132 warning. */
7133 intel_parser.op_modifier &= ~was_offset;
7134 }
7135 else
7136 strcat (intel_parser.disp, "[");
7137
7138 /* Add a '+' to the displacement string if necessary. */
7139 if (*intel_parser.disp != '\0'
7140 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
7141 strcat (intel_parser.disp, "+");
7142
7143 if (intel_expr ()
7144 && (len = intel_parser.op_string - start - 1,
7145 intel_match_token (']')))
7146 {
7147 /* Preserve brackets when the operand is an offset expression. */
7148 if (intel_parser.in_offset)
7149 strcat (intel_parser.disp, "]");
7150 else
7151 {
7152 --intel_parser.in_bracket;
7153 if (i.base_reg || i.index_reg)
7154 intel_parser.is_mem = 1;
7155 if (!intel_parser.is_mem)
7156 {
7157 if (!(intel_parser.op_modifier & (1 << T_OFFSET)))
7158 /* Defer the warning until all of the operand was parsed. */
7159 intel_parser.is_mem = -1;
7160 else if (!quiet_warnings)
7161 as_warn (_("`[%.*s]' taken to mean just `%.*s'"), len, start, len, start);
7162 }
7163 }
7164 intel_parser.op_modifier |= was_offset;
7165
7166 return 1;
7167 }
7168 return 0;
7169 }
7170
7171 /* e10 e11 e10'
7172
7173 e10' [ expr ] e10'
7174 | Empty */
7175 static int
7176 intel_e10 ()
7177 {
7178 if (!intel_e11 ())
7179 return 0;
7180
7181 while (cur_token.code == '[')
7182 {
7183 if (!intel_bracket_expr ())
7184 return 0;
7185 }
7186
7187 return 1;
7188 }
7189
7190 /* e11 ( expr )
7191 | [ expr ]
7192 | BYTE
7193 | WORD
7194 | DWORD
7195 | FWORD
7196 | QWORD
7197 | TBYTE
7198 | OWORD
7199 | XMMWORD
7200 | $
7201 | .
7202 | register
7203 | id
7204 | constant */
7205 static int
7206 intel_e11 ()
7207 {
7208 switch (cur_token.code)
7209 {
7210 /* e11 ( expr ) */
7211 case '(':
7212 intel_match_token ('(');
7213 strcat (intel_parser.disp, "(");
7214
7215 if (intel_expr () && intel_match_token (')'))
7216 {
7217 strcat (intel_parser.disp, ")");
7218 return 1;
7219 }
7220 return 0;
7221
7222 /* e11 [ expr ] */
7223 case '[':
7224 return intel_bracket_expr ();
7225
7226 /* e11 $
7227 | . */
7228 case '.':
7229 strcat (intel_parser.disp, cur_token.str);
7230 intel_match_token (cur_token.code);
7231
7232 /* Mark as a memory operand only if it's not already known to be an
7233 offset expression. */
7234 if (!intel_parser.in_offset)
7235 intel_parser.is_mem = 1;
7236
7237 return 1;
7238
7239 /* e11 register */
7240 case T_REG:
7241 {
7242 const reg_entry *reg = intel_parser.reg = cur_token.reg;
7243
7244 intel_match_token (T_REG);
7245
7246 /* Check for segment change. */
7247 if (cur_token.code == ':')
7248 {
7249 if (!(reg->reg_type & (SReg2 | SReg3)))
7250 {
7251 as_bad (_("`%s' is not a valid segment register"), reg->reg_name);
7252 return 0;
7253 }
7254 else if (i.seg[i.mem_operands])
7255 as_warn (_("Extra segment override ignored"));
7256 else
7257 {
7258 if (!intel_parser.in_offset)
7259 intel_parser.is_mem = 1;
7260 switch (reg->reg_num)
7261 {
7262 case 0:
7263 i.seg[i.mem_operands] = &es;
7264 break;
7265 case 1:
7266 i.seg[i.mem_operands] = &cs;
7267 break;
7268 case 2:
7269 i.seg[i.mem_operands] = &ss;
7270 break;
7271 case 3:
7272 i.seg[i.mem_operands] = &ds;
7273 break;
7274 case 4:
7275 i.seg[i.mem_operands] = &fs;
7276 break;
7277 case 5:
7278 i.seg[i.mem_operands] = &gs;
7279 break;
7280 }
7281 }
7282 }
7283
7284 /* Not a segment register. Check for register scaling. */
7285 else if (cur_token.code == '*')
7286 {
7287 if (!intel_parser.in_bracket)
7288 {
7289 as_bad (_("Register scaling only allowed in memory operands"));
7290 return 0;
7291 }
7292
7293 if (reg->reg_type & Reg16) /* Disallow things like [si*1]. */
7294 reg = i386_regtab + REGNAM_AX + 4; /* sp is invalid as index */
7295 else if (i.index_reg)
7296 reg = i386_regtab + REGNAM_EAX + 4; /* esp is invalid as index */
7297
7298 /* What follows must be a valid scale. */
7299 intel_match_token ('*');
7300 i.index_reg = reg;
7301 i.types[this_operand] |= BaseIndex;
7302
7303 /* Set the scale after setting the register (otherwise,
7304 i386_scale will complain) */
7305 if (cur_token.code == '+' || cur_token.code == '-')
7306 {
7307 char *str, sign = cur_token.code;
7308 intel_match_token (cur_token.code);
7309 if (cur_token.code != T_CONST)
7310 {
7311 as_bad (_("Syntax error: Expecting a constant, got `%s'"),
7312 cur_token.str);
7313 return 0;
7314 }
7315 str = (char *) xmalloc (strlen (cur_token.str) + 2);
7316 strcpy (str + 1, cur_token.str);
7317 *str = sign;
7318 if (!i386_scale (str))
7319 return 0;
7320 free (str);
7321 }
7322 else if (!i386_scale (cur_token.str))
7323 return 0;
7324 intel_match_token (cur_token.code);
7325 }
7326
7327 /* No scaling. If this is a memory operand, the register is either a
7328 base register (first occurrence) or an index register (second
7329 occurrence). */
7330 else if (intel_parser.in_bracket)
7331 {
7332
7333 if (!i.base_reg)
7334 i.base_reg = reg;
7335 else if (!i.index_reg)
7336 i.index_reg = reg;
7337 else
7338 {
7339 as_bad (_("Too many register references in memory operand"));
7340 return 0;
7341 }
7342
7343 i.types[this_operand] |= BaseIndex;
7344 }
7345
7346 /* It's neither base nor index. */
7347 else if (!intel_parser.in_offset && !intel_parser.is_mem)
7348 {
7349 i.types[this_operand] |= reg->reg_type & ~BaseIndex;
7350 i.op[this_operand].regs = reg;
7351 i.reg_operands++;
7352 }
7353 else
7354 {
7355 as_bad (_("Invalid use of register"));
7356 return 0;
7357 }
7358
7359 /* Since registers are not part of the displacement string (except
7360 when we're parsing offset operands), we may need to remove any
7361 preceding '+' from the displacement string. */
7362 if (*intel_parser.disp != '\0'
7363 && !intel_parser.in_offset)
7364 {
7365 char *s = intel_parser.disp;
7366 s += strlen (s) - 1;
7367 if (*s == '+')
7368 *s = '\0';
7369 }
7370
7371 return 1;
7372 }
7373
7374 /* e11 BYTE
7375 | WORD
7376 | DWORD
7377 | FWORD
7378 | QWORD
7379 | TBYTE
7380 | OWORD
7381 | XMMWORD */
7382 case T_BYTE:
7383 case T_WORD:
7384 case T_DWORD:
7385 case T_FWORD:
7386 case T_QWORD:
7387 case T_TBYTE:
7388 case T_XMMWORD:
7389 intel_match_token (cur_token.code);
7390
7391 if (cur_token.code == T_PTR)
7392 return 1;
7393
7394 /* It must have been an identifier. */
7395 intel_putback_token ();
7396 cur_token.code = T_ID;
7397 /* FALLTHRU */
7398
7399 /* e11 id
7400 | constant */
7401 case T_ID:
7402 if (!intel_parser.in_offset && intel_parser.is_mem <= 0)
7403 {
7404 symbolS *symbolP;
7405
7406 /* The identifier represents a memory reference only if it's not
7407 preceded by an offset modifier and if it's not an equate. */
7408 symbolP = symbol_find(cur_token.str);
7409 if (!symbolP || S_GET_SEGMENT(symbolP) != absolute_section)
7410 intel_parser.is_mem = 1;
7411 }
7412 /* FALLTHRU */
7413
7414 case T_CONST:
7415 case '-':
7416 case '+':
7417 {
7418 char *save_str, sign = 0;
7419
7420 /* Allow constants that start with `+' or `-'. */
7421 if (cur_token.code == '-' || cur_token.code == '+')
7422 {
7423 sign = cur_token.code;
7424 intel_match_token (cur_token.code);
7425 if (cur_token.code != T_CONST)
7426 {
7427 as_bad (_("Syntax error: Expecting a constant, got `%s'"),
7428 cur_token.str);
7429 return 0;
7430 }
7431 }
7432
7433 save_str = (char *) xmalloc (strlen (cur_token.str) + 2);
7434 strcpy (save_str + !!sign, cur_token.str);
7435 if (sign)
7436 *save_str = sign;
7437
7438 /* Get the next token to check for register scaling. */
7439 intel_match_token (cur_token.code);
7440
7441 /* Check if this constant is a scaling factor for an index register. */
7442 if (cur_token.code == '*')
7443 {
7444 if (intel_match_token ('*') && cur_token.code == T_REG)
7445 {
7446 const reg_entry *reg = cur_token.reg;
7447
7448 if (!intel_parser.in_bracket)
7449 {
7450 as_bad (_("Register scaling only allowed in memory operands"));
7451 return 0;
7452 }
7453
7454 if (reg->reg_type & Reg16) /* Disallow things like [1*si]. */
7455 reg = i386_regtab + REGNAM_AX + 4; /* sp is invalid as index */
7456 else if (i.index_reg)
7457 reg = i386_regtab + REGNAM_EAX + 4; /* esp is invalid as index */
7458
7459 /* The constant is followed by `* reg', so it must be
7460 a valid scale. */
7461 i.index_reg = reg;
7462 i.types[this_operand] |= BaseIndex;
7463
7464 /* Set the scale after setting the register (otherwise,
7465 i386_scale will complain) */
7466 if (!i386_scale (save_str))
7467 return 0;
7468 intel_match_token (T_REG);
7469
7470 /* Since registers are not part of the displacement
7471 string, we may need to remove any preceding '+' from
7472 the displacement string. */
7473 if (*intel_parser.disp != '\0')
7474 {
7475 char *s = intel_parser.disp;
7476 s += strlen (s) - 1;
7477 if (*s == '+')
7478 *s = '\0';
7479 }
7480
7481 free (save_str);
7482
7483 return 1;
7484 }
7485
7486 /* The constant was not used for register scaling. Since we have
7487 already consumed the token following `*' we now need to put it
7488 back in the stream. */
7489 intel_putback_token ();
7490 }
7491
7492 /* Add the constant to the displacement string. */
7493 strcat (intel_parser.disp, save_str);
7494 free (save_str);
7495
7496 return 1;
7497 }
7498 }
7499
7500 as_bad (_("Unrecognized token '%s'"), cur_token.str);
7501 return 0;
7502 }
7503
7504 /* Match the given token against cur_token. If they match, read the next
7505 token from the operand string. */
7506 static int
7507 intel_match_token (code)
7508 int code;
7509 {
7510 if (cur_token.code == code)
7511 {
7512 intel_get_token ();
7513 return 1;
7514 }
7515 else
7516 {
7517 as_bad (_("Unexpected token `%s'"), cur_token.str);
7518 return 0;
7519 }
7520 }
7521
7522 /* Read a new token from intel_parser.op_string and store it in cur_token. */
7523 static void
7524 intel_get_token ()
7525 {
7526 char *end_op;
7527 const reg_entry *reg;
7528 struct intel_token new_token;
7529
7530 new_token.code = T_NIL;
7531 new_token.reg = NULL;
7532 new_token.str = NULL;
7533
7534 /* Free the memory allocated to the previous token and move
7535 cur_token to prev_token. */
7536 if (prev_token.str)
7537 free (prev_token.str);
7538
7539 prev_token = cur_token;
7540
7541 /* Skip whitespace. */
7542 while (is_space_char (*intel_parser.op_string))
7543 intel_parser.op_string++;
7544
7545 /* Return an empty token if we find nothing else on the line. */
7546 if (*intel_parser.op_string == '\0')
7547 {
7548 cur_token = new_token;
7549 return;
7550 }
7551
7552 /* The new token cannot be larger than the remainder of the operand
7553 string. */
7554 new_token.str = (char *) xmalloc (strlen (intel_parser.op_string) + 1);
7555 new_token.str[0] = '\0';
7556
7557 if (strchr ("0123456789", *intel_parser.op_string))
7558 {
7559 char *p = new_token.str;
7560 char *q = intel_parser.op_string;
7561 new_token.code = T_CONST;
7562
7563 /* Allow any kind of identifier char to encompass floating point and
7564 hexadecimal numbers. */
7565 while (is_identifier_char (*q))
7566 *p++ = *q++;
7567 *p = '\0';
7568
7569 /* Recognize special symbol names [0-9][bf]. */
7570 if (strlen (intel_parser.op_string) == 2
7571 && (intel_parser.op_string[1] == 'b'
7572 || intel_parser.op_string[1] == 'f'))
7573 new_token.code = T_ID;
7574 }
7575
7576 else if ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL)
7577 {
7578 size_t len = end_op - intel_parser.op_string;
7579
7580 new_token.code = T_REG;
7581 new_token.reg = reg;
7582
7583 memcpy (new_token.str, intel_parser.op_string, len);
7584 new_token.str[len] = '\0';
7585 }
7586
7587 else if (is_identifier_char (*intel_parser.op_string))
7588 {
7589 char *p = new_token.str;
7590 char *q = intel_parser.op_string;
7591
7592 /* A '.' or '$' followed by an identifier char is an identifier.
7593 Otherwise, it's operator '.' followed by an expression. */
7594 if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
7595 {
7596 new_token.code = '.';
7597 new_token.str[0] = '.';
7598 new_token.str[1] = '\0';
7599 }
7600 else
7601 {
7602 while (is_identifier_char (*q) || *q == '@')
7603 *p++ = *q++;
7604 *p = '\0';
7605
7606 if (strcasecmp (new_token.str, "NOT") == 0)
7607 new_token.code = '~';
7608
7609 else if (strcasecmp (new_token.str, "MOD") == 0)
7610 new_token.code = '%';
7611
7612 else if (strcasecmp (new_token.str, "AND") == 0)
7613 new_token.code = '&';
7614
7615 else if (strcasecmp (new_token.str, "OR") == 0)
7616 new_token.code = '|';
7617
7618 else if (strcasecmp (new_token.str, "XOR") == 0)
7619 new_token.code = '^';
7620
7621 else if (strcasecmp (new_token.str, "SHL") == 0)
7622 new_token.code = T_SHL;
7623
7624 else if (strcasecmp (new_token.str, "SHR") == 0)
7625 new_token.code = T_SHR;
7626
7627 else if (strcasecmp (new_token.str, "BYTE") == 0)
7628 new_token.code = T_BYTE;
7629
7630 else if (strcasecmp (new_token.str, "WORD") == 0)
7631 new_token.code = T_WORD;
7632
7633 else if (strcasecmp (new_token.str, "DWORD") == 0)
7634 new_token.code = T_DWORD;
7635
7636 else if (strcasecmp (new_token.str, "FWORD") == 0)
7637 new_token.code = T_FWORD;
7638
7639 else if (strcasecmp (new_token.str, "QWORD") == 0)
7640 new_token.code = T_QWORD;
7641
7642 else if (strcasecmp (new_token.str, "TBYTE") == 0
7643 /* XXX remove (gcc still uses it) */
7644 || strcasecmp (new_token.str, "XWORD") == 0)
7645 new_token.code = T_TBYTE;
7646
7647 else if (strcasecmp (new_token.str, "XMMWORD") == 0
7648 || strcasecmp (new_token.str, "OWORD") == 0)
7649 new_token.code = T_XMMWORD;
7650
7651 else if (strcasecmp (new_token.str, "PTR") == 0)
7652 new_token.code = T_PTR;
7653
7654 else if (strcasecmp (new_token.str, "SHORT") == 0)
7655 new_token.code = T_SHORT;
7656
7657 else if (strcasecmp (new_token.str, "OFFSET") == 0)
7658 {
7659 new_token.code = T_OFFSET;
7660
7661 /* ??? This is not mentioned in the MASM grammar but gcc
7662 makes use of it with -mintel-syntax. OFFSET may be
7663 followed by FLAT: */
7664 if (strncasecmp (q, " FLAT:", 6) == 0)
7665 strcat (new_token.str, " FLAT:");
7666 }
7667
7668 /* ??? This is not mentioned in the MASM grammar. */
7669 else if (strcasecmp (new_token.str, "FLAT") == 0)
7670 {
7671 new_token.code = T_OFFSET;
7672 if (*q == ':')
7673 strcat (new_token.str, ":");
7674 else
7675 as_bad (_("`:' expected"));
7676 }
7677
7678 else
7679 new_token.code = T_ID;
7680 }
7681 }
7682
7683 else if (strchr ("+-/*%|&^:[]()~", *intel_parser.op_string))
7684 {
7685 new_token.code = *intel_parser.op_string;
7686 new_token.str[0] = *intel_parser.op_string;
7687 new_token.str[1] = '\0';
7688 }
7689
7690 else if (strchr ("<>", *intel_parser.op_string)
7691 && *intel_parser.op_string == *(intel_parser.op_string + 1))
7692 {
7693 new_token.code = *intel_parser.op_string == '<' ? T_SHL : T_SHR;
7694 new_token.str[0] = *intel_parser.op_string;
7695 new_token.str[1] = *intel_parser.op_string;
7696 new_token.str[2] = '\0';
7697 }
7698
7699 else
7700 as_bad (_("Unrecognized token `%s'"), intel_parser.op_string);
7701
7702 intel_parser.op_string += strlen (new_token.str);
7703 cur_token = new_token;
7704 }
7705
7706 /* Put cur_token back into the token stream and make cur_token point to
7707 prev_token. */
7708 static void
7709 intel_putback_token ()
7710 {
7711 if (cur_token.code != T_NIL)
7712 {
7713 intel_parser.op_string -= strlen (cur_token.str);
7714 free (cur_token.str);
7715 }
7716 cur_token = prev_token;
7717
7718 /* Forget prev_token. */
7719 prev_token.code = T_NIL;
7720 prev_token.reg = NULL;
7721 prev_token.str = NULL;
7722 }
7723
7724 int
7725 tc_x86_regname_to_dw2regnum (char *regname)
7726 {
7727 unsigned int regnum;
7728 unsigned int regnames_count;
7729 static const char *const regnames_32[] =
7730 {
7731 "eax", "ecx", "edx", "ebx",
7732 "esp", "ebp", "esi", "edi",
7733 "eip", "eflags", NULL,
7734 "st0", "st1", "st2", "st3",
7735 "st4", "st5", "st6", "st7",
7736 NULL, NULL,
7737 "xmm0", "xmm1", "xmm2", "xmm3",
7738 "xmm4", "xmm5", "xmm6", "xmm7",
7739 "mm0", "mm1", "mm2", "mm3",
7740 "mm4", "mm5", "mm6", "mm7",
7741 "fcw", "fsw", "mxcsr",
7742 "es", "cs", "ss", "ds", "fs", "gs", NULL, NULL,
7743 "tr", "ldtr"
7744 };
7745 static const char *const regnames_64[] =
7746 {
7747 "rax", "rdx", "rcx", "rbx",
7748 "rsi", "rdi", "rbp", "rsp",
7749 "r8", "r9", "r10", "r11",
7750 "r12", "r13", "r14", "r15",
7751 "rip",
7752 "xmm0", "xmm1", "xmm2", "xmm3",
7753 "xmm4", "xmm5", "xmm6", "xmm7",
7754 "xmm8", "xmm9", "xmm10", "xmm11",
7755 "xmm12", "xmm13", "xmm14", "xmm15",
7756 "st0", "st1", "st2", "st3",
7757 "st4", "st5", "st6", "st7",
7758 "mm0", "mm1", "mm2", "mm3",
7759 "mm4", "mm5", "mm6", "mm7",
7760 "rflags",
7761 "es", "cs", "ss", "ds", "fs", "gs", NULL, NULL,
7762 "fs.base", "gs.base", NULL, NULL,
7763 "tr", "ldtr",
7764 "mxcsr", "fcw", "fsw"
7765 };
7766 const char *const *regnames;
7767
7768 if (flag_code == CODE_64BIT)
7769 {
7770 regnames = regnames_64;
7771 regnames_count = ARRAY_SIZE (regnames_64);
7772 }
7773 else
7774 {
7775 regnames = regnames_32;
7776 regnames_count = ARRAY_SIZE (regnames_32);
7777 }
7778
7779 for (regnum = 0; regnum < regnames_count; regnum++)
7780 if (regnames[regnum] != NULL
7781 && strcmp (regname, regnames[regnum]) == 0)
7782 return regnum;
7783
7784 return -1;
7785 }
7786
7787 void
7788 tc_x86_frame_initial_instructions (void)
7789 {
7790 static unsigned int sp_regno;
7791
7792 if (!sp_regno)
7793 sp_regno = tc_x86_regname_to_dw2regnum (flag_code == CODE_64BIT
7794 ? "rsp" : "esp");
7795
7796 cfi_add_CFA_def_cfa (sp_regno, -x86_cie_data_alignment);
7797 cfi_add_CFA_offset (x86_dwarf2_return_column, x86_cie_data_alignment);
7798 }
7799
7800 int
7801 i386_elf_section_type (const char *str, size_t len)
7802 {
7803 if (flag_code == CODE_64BIT
7804 && len == sizeof ("unwind") - 1
7805 && strncmp (str, "unwind", 6) == 0)
7806 return SHT_X86_64_UNWIND;
7807
7808 return -1;
7809 }
7810
7811 #ifdef TE_PE
7812 void
7813 tc_pe_dwarf2_emit_offset (symbolS *symbol, unsigned int size)
7814 {
7815 expressionS expr;
7816
7817 expr.X_op = O_secrel;
7818 expr.X_add_symbol = symbol;
7819 expr.X_add_number = 0;
7820 emit_expr (&expr, size);
7821 }
7822 #endif
7823
7824 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
7825 /* For ELF on x86-64, add support for SHF_X86_64_LARGE. */
7826
7827 int
7828 x86_64_section_letter (int letter, char **ptr_msg)
7829 {
7830 if (flag_code == CODE_64BIT)
7831 {
7832 if (letter == 'l')
7833 return SHF_X86_64_LARGE;
7834
7835 *ptr_msg = _("Bad .section directive: want a,l,w,x,M,S,G,T in string");
7836 }
7837 else
7838 *ptr_msg = _("Bad .section directive: want a,w,x,M,S,G,T in string");
7839 return -1;
7840 }
7841
7842 int
7843 x86_64_section_word (char *str, size_t len)
7844 {
7845 if (len == 5 && flag_code == CODE_64BIT && CONST_STRNEQ (str, "large"))
7846 return SHF_X86_64_LARGE;
7847
7848 return -1;
7849 }
7850
7851 static void
7852 handle_large_common (int small ATTRIBUTE_UNUSED)
7853 {
7854 if (flag_code != CODE_64BIT)
7855 {
7856 s_comm_internal (0, elf_common_parse);
7857 as_warn (_(".largecomm supported only in 64bit mode, producing .comm"));
7858 }
7859 else
7860 {
7861 static segT lbss_section;
7862 asection *saved_com_section_ptr = elf_com_section_ptr;
7863 asection *saved_bss_section = bss_section;
7864
7865 if (lbss_section == NULL)
7866 {
7867 flagword applicable;
7868 segT seg = now_seg;
7869 subsegT subseg = now_subseg;
7870
7871 /* The .lbss section is for local .largecomm symbols. */
7872 lbss_section = subseg_new (".lbss", 0);
7873 applicable = bfd_applicable_section_flags (stdoutput);
7874 bfd_set_section_flags (stdoutput, lbss_section,
7875 applicable & SEC_ALLOC);
7876 seg_info (lbss_section)->bss = 1;
7877
7878 subseg_set (seg, subseg);
7879 }
7880
7881 elf_com_section_ptr = &_bfd_elf_large_com_section;
7882 bss_section = lbss_section;
7883
7884 s_comm_internal (0, elf_common_parse);
7885
7886 elf_com_section_ptr = saved_com_section_ptr;
7887 bss_section = saved_bss_section;
7888 }
7889 }
7890 #endif /* OBJ_ELF || OBJ_MAYBE_ELF */
This page took 0.193652 seconds and 4 git commands to generate.