2007-09-04 H.J. Lu <hongjiu.lu@intel.com>
[deliverable/binutils-gdb.git] / gas / config / tc-i386.c
1 /* tc-i386.c -- Assemble code for the Intel 80386
2 Copyright 1989, 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
3 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007
4 Free Software Foundation, Inc.
5
6 This file is part of GAS, the GNU Assembler.
7
8 GAS is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 3, or (at your option)
11 any later version.
12
13 GAS is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
17
18 You should have received a copy of the GNU General Public License
19 along with GAS; see the file COPYING. If not, write to the Free
20 Software Foundation, 51 Franklin Street - Fifth Floor, Boston, MA
21 02110-1301, USA. */
22
23 /* Intel 80386 machine specific gas.
24 Written by Eliot Dresselhaus (eliot@mgm.mit.edu).
25 x86_64 support by Jan Hubicka (jh@suse.cz)
26 VIA PadLock support by Michal Ludvig (mludvig@suse.cz)
27 Bugs & suggestions are completely welcome. This is free software.
28 Please help us make it better. */
29
30 #include "as.h"
31 #include "safe-ctype.h"
32 #include "subsegs.h"
33 #include "dwarf2dbg.h"
34 #include "dw2gencfi.h"
35 #include "elf/x86-64.h"
36
37 #ifndef REGISTER_WARNINGS
38 #define REGISTER_WARNINGS 1
39 #endif
40
41 #ifndef INFER_ADDR_PREFIX
42 #define INFER_ADDR_PREFIX 1
43 #endif
44
45 #ifndef SCALE1_WHEN_NO_INDEX
46 /* Specifying a scale factor besides 1 when there is no index is
47 futile. eg. `mov (%ebx,2),%al' does exactly the same as
48 `mov (%ebx),%al'. To slavishly follow what the programmer
49 specified, set SCALE1_WHEN_NO_INDEX to 0. */
50 #define SCALE1_WHEN_NO_INDEX 1
51 #endif
52
53 #ifndef DEFAULT_ARCH
54 #define DEFAULT_ARCH "i386"
55 #endif
56
57 #ifndef INLINE
58 #if __GNUC__ >= 2
59 #define INLINE __inline__
60 #else
61 #define INLINE
62 #endif
63 #endif
64
65 static void set_code_flag (int);
66 static void set_16bit_gcc_code_flag (int);
67 static void set_intel_syntax (int);
68 static void set_cpu_arch (int);
69 #ifdef TE_PE
70 static void pe_directive_secrel (int);
71 #endif
72 static void signed_cons (int);
73 static char *output_invalid (int c);
74 static int i386_operand (char *);
75 static int i386_intel_operand (char *, int);
76 static const reg_entry *parse_register (char *, char **);
77 static char *parse_insn (char *, char *);
78 static char *parse_operands (char *, const char *);
79 static void swap_operands (void);
80 static void swap_2_operands (int, int);
81 static void optimize_imm (void);
82 static void optimize_disp (void);
83 static int match_template (void);
84 static int check_string (void);
85 static int process_suffix (void);
86 static int check_byte_reg (void);
87 static int check_long_reg (void);
88 static int check_qword_reg (void);
89 static int check_word_reg (void);
90 static int finalize_imm (void);
91 static int process_operands (void);
92 static const seg_entry *build_modrm_byte (void);
93 static void output_insn (void);
94 static void output_imm (fragS *, offsetT);
95 static void output_disp (fragS *, offsetT);
96 #ifndef I386COFF
97 static void s_bss (int);
98 #endif
99 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
100 static void handle_large_common (int small ATTRIBUTE_UNUSED);
101 #endif
102
103 static const char *default_arch = DEFAULT_ARCH;
104
105 /* 'md_assemble ()' gathers together information and puts it into a
106 i386_insn. */
107
108 union i386_op
109 {
110 expressionS *disps;
111 expressionS *imms;
112 const reg_entry *regs;
113 };
114
115 struct _i386_insn
116 {
117 /* TM holds the template for the insn were currently assembling. */
118 template tm;
119
120 /* SUFFIX holds the instruction mnemonic suffix if given.
121 (e.g. 'l' for 'movl') */
122 char suffix;
123
124 /* OPERANDS gives the number of given operands. */
125 unsigned int operands;
126
127 /* REG_OPERANDS, DISP_OPERANDS, MEM_OPERANDS, IMM_OPERANDS give the number
128 of given register, displacement, memory operands and immediate
129 operands. */
130 unsigned int reg_operands, disp_operands, mem_operands, imm_operands;
131
132 /* TYPES [i] is the type (see above #defines) which tells us how to
133 use OP[i] for the corresponding operand. */
134 unsigned int types[MAX_OPERANDS];
135
136 /* Displacement expression, immediate expression, or register for each
137 operand. */
138 union i386_op op[MAX_OPERANDS];
139
140 /* Flags for operands. */
141 unsigned int flags[MAX_OPERANDS];
142 #define Operand_PCrel 1
143
144 /* Relocation type for operand */
145 enum bfd_reloc_code_real reloc[MAX_OPERANDS];
146
147 /* BASE_REG, INDEX_REG, and LOG2_SCALE_FACTOR are used to encode
148 the base index byte below. */
149 const reg_entry *base_reg;
150 const reg_entry *index_reg;
151 unsigned int log2_scale_factor;
152
153 /* SEG gives the seg_entries of this insn. They are zero unless
154 explicit segment overrides are given. */
155 const seg_entry *seg[2];
156
157 /* PREFIX holds all the given prefix opcodes (usually null).
158 PREFIXES is the number of prefix opcodes. */
159 unsigned int prefixes;
160 unsigned char prefix[MAX_PREFIXES];
161
162 /* RM and SIB are the modrm byte and the sib byte where the
163 addressing modes of this insn are encoded. */
164
165 modrm_byte rm;
166 rex_byte rex;
167 sib_byte sib;
168 };
169
170 typedef struct _i386_insn i386_insn;
171
172 /* List of chars besides those in app.c:symbol_chars that can start an
173 operand. Used to prevent the scrubber eating vital white-space. */
174 const char extra_symbol_chars[] = "*%-(["
175 #ifdef LEX_AT
176 "@"
177 #endif
178 #ifdef LEX_QM
179 "?"
180 #endif
181 ;
182
183 #if (defined (TE_I386AIX) \
184 || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) \
185 && !defined (TE_GNU) \
186 && !defined (TE_LINUX) \
187 && !defined (TE_NETWARE) \
188 && !defined (TE_FreeBSD) \
189 && !defined (TE_NetBSD)))
190 /* This array holds the chars that always start a comment. If the
191 pre-processor is disabled, these aren't very useful. The option
192 --divide will remove '/' from this list. */
193 const char *i386_comment_chars = "#/";
194 #define SVR4_COMMENT_CHARS 1
195 #define PREFIX_SEPARATOR '\\'
196
197 #else
198 const char *i386_comment_chars = "#";
199 #define PREFIX_SEPARATOR '/'
200 #endif
201
202 /* This array holds the chars that only start a comment at the beginning of
203 a line. If the line seems to have the form '# 123 filename'
204 .line and .file directives will appear in the pre-processed output.
205 Note that input_file.c hand checks for '#' at the beginning of the
206 first line of the input file. This is because the compiler outputs
207 #NO_APP at the beginning of its output.
208 Also note that comments started like this one will always work if
209 '/' isn't otherwise defined. */
210 const char line_comment_chars[] = "#/";
211
212 const char line_separator_chars[] = ";";
213
214 /* Chars that can be used to separate mant from exp in floating point
215 nums. */
216 const char EXP_CHARS[] = "eE";
217
218 /* Chars that mean this number is a floating point constant
219 As in 0f12.456
220 or 0d1.2345e12. */
221 const char FLT_CHARS[] = "fFdDxX";
222
223 /* Tables for lexical analysis. */
224 static char mnemonic_chars[256];
225 static char register_chars[256];
226 static char operand_chars[256];
227 static char identifier_chars[256];
228 static char digit_chars[256];
229
230 /* Lexical macros. */
231 #define is_mnemonic_char(x) (mnemonic_chars[(unsigned char) x])
232 #define is_operand_char(x) (operand_chars[(unsigned char) x])
233 #define is_register_char(x) (register_chars[(unsigned char) x])
234 #define is_space_char(x) ((x) == ' ')
235 #define is_identifier_char(x) (identifier_chars[(unsigned char) x])
236 #define is_digit_char(x) (digit_chars[(unsigned char) x])
237
238 /* All non-digit non-letter characters that may occur in an operand. */
239 static char operand_special_chars[] = "%$-+(,)*._~/<>|&^!:[@]";
240
241 /* md_assemble() always leaves the strings it's passed unaltered. To
242 effect this we maintain a stack of saved characters that we've smashed
243 with '\0's (indicating end of strings for various sub-fields of the
244 assembler instruction). */
245 static char save_stack[32];
246 static char *save_stack_p;
247 #define END_STRING_AND_SAVE(s) \
248 do { *save_stack_p++ = *(s); *(s) = '\0'; } while (0)
249 #define RESTORE_END_STRING(s) \
250 do { *(s) = *--save_stack_p; } while (0)
251
252 /* The instruction we're assembling. */
253 static i386_insn i;
254
255 /* Possible templates for current insn. */
256 static const templates *current_templates;
257
258 /* Per instruction expressionS buffers: max displacements & immediates. */
259 static expressionS disp_expressions[MAX_MEMORY_OPERANDS];
260 static expressionS im_expressions[MAX_IMMEDIATE_OPERANDS];
261
262 /* Current operand we are working on. */
263 static int this_operand;
264
265 /* We support four different modes. FLAG_CODE variable is used to distinguish
266 these. */
267
268 enum flag_code {
269 CODE_32BIT,
270 CODE_16BIT,
271 CODE_64BIT };
272 #define NUM_FLAG_CODE ((int) CODE_64BIT + 1)
273
274 static enum flag_code flag_code;
275 static unsigned int object_64bit;
276 static int use_rela_relocations = 0;
277
278 /* The names used to print error messages. */
279 static const char *flag_code_names[] =
280 {
281 "32",
282 "16",
283 "64"
284 };
285
286 /* 1 for intel syntax,
287 0 if att syntax. */
288 static int intel_syntax = 0;
289
290 /* 1 if register prefix % not required. */
291 static int allow_naked_reg = 0;
292
293 /* Register prefix used for error message. */
294 static const char *register_prefix = "%";
295
296 /* Used in 16 bit gcc mode to add an l suffix to call, ret, enter,
297 leave, push, and pop instructions so that gcc has the same stack
298 frame as in 32 bit mode. */
299 static char stackop_size = '\0';
300
301 /* Non-zero to optimize code alignment. */
302 int optimize_align_code = 1;
303
304 /* Non-zero to quieten some warnings. */
305 static int quiet_warnings = 0;
306
307 /* CPU name. */
308 static const char *cpu_arch_name = NULL;
309 static const char *cpu_sub_arch_name = NULL;
310
311 /* CPU feature flags. */
312 static unsigned int cpu_arch_flags = CpuUnknownFlags | CpuNo64;
313
314 /* If we have selected a cpu we are generating instructions for. */
315 static int cpu_arch_tune_set = 0;
316
317 /* Cpu we are generating instructions for. */
318 static enum processor_type cpu_arch_tune = PROCESSOR_UNKNOWN;
319
320 /* CPU feature flags of cpu we are generating instructions for. */
321 static unsigned int cpu_arch_tune_flags = 0;
322
323 /* CPU instruction set architecture used. */
324 static enum processor_type cpu_arch_isa = PROCESSOR_UNKNOWN;
325
326 /* CPU feature flags of instruction set architecture used. */
327 static unsigned int cpu_arch_isa_flags = 0;
328
329 /* If set, conditional jumps are not automatically promoted to handle
330 larger than a byte offset. */
331 static unsigned int no_cond_jump_promotion = 0;
332
333 /* Pre-defined "_GLOBAL_OFFSET_TABLE_". */
334 static symbolS *GOT_symbol;
335
336 /* The dwarf2 return column, adjusted for 32 or 64 bit. */
337 unsigned int x86_dwarf2_return_column;
338
339 /* The dwarf2 data alignment, adjusted for 32 or 64 bit. */
340 int x86_cie_data_alignment;
341
342 /* Interface to relax_segment.
343 There are 3 major relax states for 386 jump insns because the
344 different types of jumps add different sizes to frags when we're
345 figuring out what sort of jump to choose to reach a given label. */
346
347 /* Types. */
348 #define UNCOND_JUMP 0
349 #define COND_JUMP 1
350 #define COND_JUMP86 2
351
352 /* Sizes. */
353 #define CODE16 1
354 #define SMALL 0
355 #define SMALL16 (SMALL | CODE16)
356 #define BIG 2
357 #define BIG16 (BIG | CODE16)
358
359 #ifndef INLINE
360 #ifdef __GNUC__
361 #define INLINE __inline__
362 #else
363 #define INLINE
364 #endif
365 #endif
366
367 #define ENCODE_RELAX_STATE(type, size) \
368 ((relax_substateT) (((type) << 2) | (size)))
369 #define TYPE_FROM_RELAX_STATE(s) \
370 ((s) >> 2)
371 #define DISP_SIZE_FROM_RELAX_STATE(s) \
372 ((((s) & 3) == BIG ? 4 : (((s) & 3) == BIG16 ? 2 : 1)))
373
374 /* This table is used by relax_frag to promote short jumps to long
375 ones where necessary. SMALL (short) jumps may be promoted to BIG
376 (32 bit long) ones, and SMALL16 jumps to BIG16 (16 bit long). We
377 don't allow a short jump in a 32 bit code segment to be promoted to
378 a 16 bit offset jump because it's slower (requires data size
379 prefix), and doesn't work, unless the destination is in the bottom
380 64k of the code segment (The top 16 bits of eip are zeroed). */
381
382 const relax_typeS md_relax_table[] =
383 {
384 /* The fields are:
385 1) most positive reach of this state,
386 2) most negative reach of this state,
387 3) how many bytes this mode will have in the variable part of the frag
388 4) which index into the table to try if we can't fit into this one. */
389
390 /* UNCOND_JUMP states. */
391 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG)},
392 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16)},
393 /* dword jmp adds 4 bytes to frag:
394 0 extra opcode bytes, 4 displacement bytes. */
395 {0, 0, 4, 0},
396 /* word jmp adds 2 byte2 to frag:
397 0 extra opcode bytes, 2 displacement bytes. */
398 {0, 0, 2, 0},
399
400 /* COND_JUMP states. */
401 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG)},
402 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG16)},
403 /* dword conditionals adds 5 bytes to frag:
404 1 extra opcode byte, 4 displacement bytes. */
405 {0, 0, 5, 0},
406 /* word conditionals add 3 bytes to frag:
407 1 extra opcode byte, 2 displacement bytes. */
408 {0, 0, 3, 0},
409
410 /* COND_JUMP86 states. */
411 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG)},
412 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG16)},
413 /* dword conditionals adds 5 bytes to frag:
414 1 extra opcode byte, 4 displacement bytes. */
415 {0, 0, 5, 0},
416 /* word conditionals add 4 bytes to frag:
417 1 displacement byte and a 3 byte long branch insn. */
418 {0, 0, 4, 0}
419 };
420
421 static const arch_entry cpu_arch[] =
422 {
423 {"generic32", PROCESSOR_GENERIC32,
424 Cpu186|Cpu286|Cpu386},
425 {"generic64", PROCESSOR_GENERIC64,
426 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
427 |CpuMMX2|CpuSSE|CpuSSE2},
428 {"i8086", PROCESSOR_UNKNOWN,
429 0},
430 {"i186", PROCESSOR_UNKNOWN,
431 Cpu186},
432 {"i286", PROCESSOR_UNKNOWN,
433 Cpu186|Cpu286},
434 {"i386", PROCESSOR_I386,
435 Cpu186|Cpu286|Cpu386},
436 {"i486", PROCESSOR_I486,
437 Cpu186|Cpu286|Cpu386|Cpu486},
438 {"i586", PROCESSOR_PENTIUM,
439 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586},
440 {"i686", PROCESSOR_PENTIUMPRO,
441 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686},
442 {"pentium", PROCESSOR_PENTIUM,
443 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586},
444 {"pentiumpro",PROCESSOR_PENTIUMPRO,
445 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686},
446 {"pentiumii", PROCESSOR_PENTIUMPRO,
447 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX},
448 {"pentiumiii",PROCESSOR_PENTIUMPRO,
449 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuMMX2|CpuSSE},
450 {"pentium4", PROCESSOR_PENTIUM4,
451 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
452 |CpuMMX2|CpuSSE|CpuSSE2},
453 {"prescott", PROCESSOR_NOCONA,
454 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
455 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
456 {"nocona", PROCESSOR_NOCONA,
457 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
458 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
459 {"yonah", PROCESSOR_CORE,
460 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
461 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
462 {"core", PROCESSOR_CORE,
463 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
464 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
465 {"merom", PROCESSOR_CORE2,
466 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
467 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSSE3},
468 {"core2", PROCESSOR_CORE2,
469 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
470 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSSE3},
471 {"k6", PROCESSOR_K6,
472 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX},
473 {"k6_2", PROCESSOR_K6,
474 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX|Cpu3dnow},
475 {"athlon", PROCESSOR_ATHLON,
476 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6
477 |CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA},
478 {"sledgehammer", PROCESSOR_K8,
479 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6
480 |CpuSledgehammer|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2},
481 {"opteron", PROCESSOR_K8,
482 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6
483 |CpuSledgehammer|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2},
484 {"k8", PROCESSOR_K8,
485 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6
486 |CpuSledgehammer|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2},
487 {"amdfam10", PROCESSOR_AMDFAM10,
488 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuSledgehammer
489 |CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2|CpuSSE3|CpuSSE4a
490 |CpuABM},
491 {".mmx", PROCESSOR_UNKNOWN,
492 CpuMMX},
493 {".sse", PROCESSOR_UNKNOWN,
494 CpuMMX|CpuMMX2|CpuSSE},
495 {".sse2", PROCESSOR_UNKNOWN,
496 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2},
497 {".sse3", PROCESSOR_UNKNOWN,
498 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
499 {".ssse3", PROCESSOR_UNKNOWN,
500 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSSE3},
501 {".sse4.1", PROCESSOR_UNKNOWN,
502 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSSE3|CpuSSE4_1},
503 {".sse4.2", PROCESSOR_UNKNOWN,
504 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSSE3|CpuSSE4},
505 {".sse4", PROCESSOR_UNKNOWN,
506 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSSE3|CpuSSE4},
507 {".3dnow", PROCESSOR_UNKNOWN,
508 CpuMMX|Cpu3dnow},
509 {".3dnowa", PROCESSOR_UNKNOWN,
510 CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA},
511 {".padlock", PROCESSOR_UNKNOWN,
512 CpuPadLock},
513 {".pacifica", PROCESSOR_UNKNOWN,
514 CpuSVME},
515 {".svme", PROCESSOR_UNKNOWN,
516 CpuSVME},
517 {".sse4a", PROCESSOR_UNKNOWN,
518 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSE4a},
519 {".abm", PROCESSOR_UNKNOWN,
520 CpuABM}
521 };
522
523 const pseudo_typeS md_pseudo_table[] =
524 {
525 #if !defined(OBJ_AOUT) && !defined(USE_ALIGN_PTWO)
526 {"align", s_align_bytes, 0},
527 #else
528 {"align", s_align_ptwo, 0},
529 #endif
530 {"arch", set_cpu_arch, 0},
531 #ifndef I386COFF
532 {"bss", s_bss, 0},
533 #endif
534 {"ffloat", float_cons, 'f'},
535 {"dfloat", float_cons, 'd'},
536 {"tfloat", float_cons, 'x'},
537 {"value", cons, 2},
538 {"slong", signed_cons, 4},
539 {"noopt", s_ignore, 0},
540 {"optim", s_ignore, 0},
541 {"code16gcc", set_16bit_gcc_code_flag, CODE_16BIT},
542 {"code16", set_code_flag, CODE_16BIT},
543 {"code32", set_code_flag, CODE_32BIT},
544 {"code64", set_code_flag, CODE_64BIT},
545 {"intel_syntax", set_intel_syntax, 1},
546 {"att_syntax", set_intel_syntax, 0},
547 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
548 {"largecomm", handle_large_common, 0},
549 #else
550 {"file", (void (*) (int)) dwarf2_directive_file, 0},
551 {"loc", dwarf2_directive_loc, 0},
552 {"loc_mark_labels", dwarf2_directive_loc_mark_labels, 0},
553 #endif
554 #ifdef TE_PE
555 {"secrel32", pe_directive_secrel, 0},
556 #endif
557 {0, 0, 0}
558 };
559
560 /* For interface with expression (). */
561 extern char *input_line_pointer;
562
563 /* Hash table for instruction mnemonic lookup. */
564 static struct hash_control *op_hash;
565
566 /* Hash table for register lookup. */
567 static struct hash_control *reg_hash;
568 \f
569 void
570 i386_align_code (fragS *fragP, int count)
571 {
572 /* Various efficient no-op patterns for aligning code labels.
573 Note: Don't try to assemble the instructions in the comments.
574 0L and 0w are not legal. */
575 static const char f32_1[] =
576 {0x90}; /* nop */
577 static const char f32_2[] =
578 {0x66,0x90}; /* xchg %ax,%ax */
579 static const char f32_3[] =
580 {0x8d,0x76,0x00}; /* leal 0(%esi),%esi */
581 static const char f32_4[] =
582 {0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
583 static const char f32_5[] =
584 {0x90, /* nop */
585 0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
586 static const char f32_6[] =
587 {0x8d,0xb6,0x00,0x00,0x00,0x00}; /* leal 0L(%esi),%esi */
588 static const char f32_7[] =
589 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
590 static const char f32_8[] =
591 {0x90, /* nop */
592 0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
593 static const char f32_9[] =
594 {0x89,0xf6, /* movl %esi,%esi */
595 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
596 static const char f32_10[] =
597 {0x8d,0x76,0x00, /* leal 0(%esi),%esi */
598 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
599 static const char f32_11[] =
600 {0x8d,0x74,0x26,0x00, /* leal 0(%esi,1),%esi */
601 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
602 static const char f32_12[] =
603 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
604 0x8d,0xbf,0x00,0x00,0x00,0x00}; /* leal 0L(%edi),%edi */
605 static const char f32_13[] =
606 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
607 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
608 static const char f32_14[] =
609 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00, /* leal 0L(%esi,1),%esi */
610 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
611 static const char f16_3[] =
612 {0x8d,0x74,0x00}; /* lea 0(%esi),%esi */
613 static const char f16_4[] =
614 {0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
615 static const char f16_5[] =
616 {0x90, /* nop */
617 0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
618 static const char f16_6[] =
619 {0x89,0xf6, /* mov %si,%si */
620 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
621 static const char f16_7[] =
622 {0x8d,0x74,0x00, /* lea 0(%si),%si */
623 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
624 static const char f16_8[] =
625 {0x8d,0xb4,0x00,0x00, /* lea 0w(%si),%si */
626 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
627 static const char jump_31[] =
628 {0xeb,0x1d,0x90,0x90,0x90,0x90,0x90, /* jmp .+31; lotsa nops */
629 0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90,
630 0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90,
631 0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90};
632 static const char *const f32_patt[] = {
633 f32_1, f32_2, f32_3, f32_4, f32_5, f32_6, f32_7, f32_8,
634 f32_9, f32_10, f32_11, f32_12, f32_13, f32_14
635 };
636 static const char *const f16_patt[] = {
637 f32_1, f32_2, f16_3, f16_4, f16_5, f16_6, f16_7, f16_8
638 };
639 /* nopl (%[re]ax) */
640 static const char alt_3[] =
641 {0x0f,0x1f,0x00};
642 /* nopl 0(%[re]ax) */
643 static const char alt_4[] =
644 {0x0f,0x1f,0x40,0x00};
645 /* nopl 0(%[re]ax,%[re]ax,1) */
646 static const char alt_5[] =
647 {0x0f,0x1f,0x44,0x00,0x00};
648 /* nopw 0(%[re]ax,%[re]ax,1) */
649 static const char alt_6[] =
650 {0x66,0x0f,0x1f,0x44,0x00,0x00};
651 /* nopl 0L(%[re]ax) */
652 static const char alt_7[] =
653 {0x0f,0x1f,0x80,0x00,0x00,0x00,0x00};
654 /* nopl 0L(%[re]ax,%[re]ax,1) */
655 static const char alt_8[] =
656 {0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
657 /* nopw 0L(%[re]ax,%[re]ax,1) */
658 static const char alt_9[] =
659 {0x66,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
660 /* nopw %cs:0L(%[re]ax,%[re]ax,1) */
661 static const char alt_10[] =
662 {0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
663 /* data16
664 nopw %cs:0L(%[re]ax,%[re]ax,1) */
665 static const char alt_long_11[] =
666 {0x66,
667 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
668 /* data16
669 data16
670 nopw %cs:0L(%[re]ax,%[re]ax,1) */
671 static const char alt_long_12[] =
672 {0x66,
673 0x66,
674 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
675 /* data16
676 data16
677 data16
678 nopw %cs:0L(%[re]ax,%[re]ax,1) */
679 static const char alt_long_13[] =
680 {0x66,
681 0x66,
682 0x66,
683 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
684 /* data16
685 data16
686 data16
687 data16
688 nopw %cs:0L(%[re]ax,%[re]ax,1) */
689 static const char alt_long_14[] =
690 {0x66,
691 0x66,
692 0x66,
693 0x66,
694 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
695 /* data16
696 data16
697 data16
698 data16
699 data16
700 nopw %cs:0L(%[re]ax,%[re]ax,1) */
701 static const char alt_long_15[] =
702 {0x66,
703 0x66,
704 0x66,
705 0x66,
706 0x66,
707 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
708 /* nopl 0(%[re]ax,%[re]ax,1)
709 nopw 0(%[re]ax,%[re]ax,1) */
710 static const char alt_short_11[] =
711 {0x0f,0x1f,0x44,0x00,0x00,
712 0x66,0x0f,0x1f,0x44,0x00,0x00};
713 /* nopw 0(%[re]ax,%[re]ax,1)
714 nopw 0(%[re]ax,%[re]ax,1) */
715 static const char alt_short_12[] =
716 {0x66,0x0f,0x1f,0x44,0x00,0x00,
717 0x66,0x0f,0x1f,0x44,0x00,0x00};
718 /* nopw 0(%[re]ax,%[re]ax,1)
719 nopl 0L(%[re]ax) */
720 static const char alt_short_13[] =
721 {0x66,0x0f,0x1f,0x44,0x00,0x00,
722 0x0f,0x1f,0x80,0x00,0x00,0x00,0x00};
723 /* nopl 0L(%[re]ax)
724 nopl 0L(%[re]ax) */
725 static const char alt_short_14[] =
726 {0x0f,0x1f,0x80,0x00,0x00,0x00,0x00,
727 0x0f,0x1f,0x80,0x00,0x00,0x00,0x00};
728 /* nopl 0L(%[re]ax)
729 nopl 0L(%[re]ax,%[re]ax,1) */
730 static const char alt_short_15[] =
731 {0x0f,0x1f,0x80,0x00,0x00,0x00,0x00,
732 0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
733 static const char *const alt_short_patt[] = {
734 f32_1, f32_2, alt_3, alt_4, alt_5, alt_6, alt_7, alt_8,
735 alt_9, alt_10, alt_short_11, alt_short_12, alt_short_13,
736 alt_short_14, alt_short_15
737 };
738 static const char *const alt_long_patt[] = {
739 f32_1, f32_2, alt_3, alt_4, alt_5, alt_6, alt_7, alt_8,
740 alt_9, alt_10, alt_long_11, alt_long_12, alt_long_13,
741 alt_long_14, alt_long_15
742 };
743
744 /* Only align for at least a positive non-zero boundary. */
745 if (count <= 0 || count > MAX_MEM_FOR_RS_ALIGN_CODE)
746 return;
747
748 /* We need to decide which NOP sequence to use for 32bit and
749 64bit. When -mtune= is used:
750
751 1. For PROCESSOR_I386, PROCESSOR_I486, PROCESSOR_PENTIUM and
752 PROCESSOR_GENERIC32, f32_patt will be used.
753 2. For PROCESSOR_PENTIUMPRO, PROCESSOR_PENTIUM4, PROCESSOR_NOCONA,
754 PROCESSOR_CORE, PROCESSOR_CORE2, and PROCESSOR_GENERIC64,
755 alt_long_patt will be used.
756 3. For PROCESSOR_ATHLON, PROCESSOR_K6, PROCESSOR_K8 and
757 PROCESSOR_AMDFAM10, alt_short_patt will be used.
758
759 When -mtune= isn't used, alt_long_patt will be used if
760 cpu_arch_isa_flags has Cpu686. Otherwise, f32_patt will
761 be used.
762
763 When -march= or .arch is used, we can't use anything beyond
764 cpu_arch_isa_flags. */
765
766 if (flag_code == CODE_16BIT)
767 {
768 if (count > 8)
769 {
770 memcpy (fragP->fr_literal + fragP->fr_fix,
771 jump_31, count);
772 /* Adjust jump offset. */
773 fragP->fr_literal[fragP->fr_fix + 1] = count - 2;
774 }
775 else
776 memcpy (fragP->fr_literal + fragP->fr_fix,
777 f16_patt[count - 1], count);
778 }
779 else
780 {
781 const char *const *patt = NULL;
782
783 if (cpu_arch_isa == PROCESSOR_UNKNOWN)
784 {
785 /* PROCESSOR_UNKNOWN means that all ISAs may be used. */
786 switch (cpu_arch_tune)
787 {
788 case PROCESSOR_UNKNOWN:
789 /* We use cpu_arch_isa_flags to check if we SHOULD
790 optimize for Cpu686. */
791 if ((cpu_arch_isa_flags & Cpu686) != 0)
792 patt = alt_long_patt;
793 else
794 patt = f32_patt;
795 break;
796 case PROCESSOR_PENTIUMPRO:
797 case PROCESSOR_PENTIUM4:
798 case PROCESSOR_NOCONA:
799 case PROCESSOR_CORE:
800 case PROCESSOR_CORE2:
801 case PROCESSOR_GENERIC64:
802 patt = alt_long_patt;
803 break;
804 case PROCESSOR_K6:
805 case PROCESSOR_ATHLON:
806 case PROCESSOR_K8:
807 case PROCESSOR_AMDFAM10:
808 patt = alt_short_patt;
809 break;
810 case PROCESSOR_I386:
811 case PROCESSOR_I486:
812 case PROCESSOR_PENTIUM:
813 case PROCESSOR_GENERIC32:
814 patt = f32_patt;
815 break;
816 }
817 }
818 else
819 {
820 switch (cpu_arch_tune)
821 {
822 case PROCESSOR_UNKNOWN:
823 /* When cpu_arch_isa is net, cpu_arch_tune shouldn't be
824 PROCESSOR_UNKNOWN. */
825 abort ();
826 break;
827
828 case PROCESSOR_I386:
829 case PROCESSOR_I486:
830 case PROCESSOR_PENTIUM:
831 case PROCESSOR_K6:
832 case PROCESSOR_ATHLON:
833 case PROCESSOR_K8:
834 case PROCESSOR_AMDFAM10:
835 case PROCESSOR_GENERIC32:
836 /* We use cpu_arch_isa_flags to check if we CAN optimize
837 for Cpu686. */
838 if ((cpu_arch_isa_flags & Cpu686) != 0)
839 patt = alt_short_patt;
840 else
841 patt = f32_patt;
842 break;
843 case PROCESSOR_PENTIUMPRO:
844 case PROCESSOR_PENTIUM4:
845 case PROCESSOR_NOCONA:
846 case PROCESSOR_CORE:
847 case PROCESSOR_CORE2:
848 if ((cpu_arch_isa_flags & Cpu686) != 0)
849 patt = alt_long_patt;
850 else
851 patt = f32_patt;
852 break;
853 case PROCESSOR_GENERIC64:
854 patt = alt_long_patt;
855 break;
856 }
857 }
858
859 if (patt == f32_patt)
860 {
861 /* If the padding is less than 15 bytes, we use the normal
862 ones. Otherwise, we use a jump instruction and adjust
863 its offset. */
864 if (count < 15)
865 memcpy (fragP->fr_literal + fragP->fr_fix,
866 patt[count - 1], count);
867 else
868 {
869 memcpy (fragP->fr_literal + fragP->fr_fix,
870 jump_31, count);
871 /* Adjust jump offset. */
872 fragP->fr_literal[fragP->fr_fix + 1] = count - 2;
873 }
874 }
875 else
876 {
877 /* Maximum length of an instruction is 15 byte. If the
878 padding is greater than 15 bytes and we don't use jump,
879 we have to break it into smaller pieces. */
880 int padding = count;
881 while (padding > 15)
882 {
883 padding -= 15;
884 memcpy (fragP->fr_literal + fragP->fr_fix + padding,
885 patt [14], 15);
886 }
887
888 if (padding)
889 memcpy (fragP->fr_literal + fragP->fr_fix,
890 patt [padding - 1], padding);
891 }
892 }
893 fragP->fr_var = count;
894 }
895
896 static INLINE unsigned int
897 mode_from_disp_size (unsigned int t)
898 {
899 return (t & Disp8) ? 1 : (t & (Disp16 | Disp32 | Disp32S)) ? 2 : 0;
900 }
901
902 static INLINE int
903 fits_in_signed_byte (offsetT num)
904 {
905 return (num >= -128) && (num <= 127);
906 }
907
908 static INLINE int
909 fits_in_unsigned_byte (offsetT num)
910 {
911 return (num & 0xff) == num;
912 }
913
914 static INLINE int
915 fits_in_unsigned_word (offsetT num)
916 {
917 return (num & 0xffff) == num;
918 }
919
920 static INLINE int
921 fits_in_signed_word (offsetT num)
922 {
923 return (-32768 <= num) && (num <= 32767);
924 }
925
926 static INLINE int
927 fits_in_signed_long (offsetT num ATTRIBUTE_UNUSED)
928 {
929 #ifndef BFD64
930 return 1;
931 #else
932 return (!(((offsetT) -1 << 31) & num)
933 || (((offsetT) -1 << 31) & num) == ((offsetT) -1 << 31));
934 #endif
935 } /* fits_in_signed_long() */
936
937 static INLINE int
938 fits_in_unsigned_long (offsetT num ATTRIBUTE_UNUSED)
939 {
940 #ifndef BFD64
941 return 1;
942 #else
943 return (num & (((offsetT) 2 << 31) - 1)) == num;
944 #endif
945 } /* fits_in_unsigned_long() */
946
947 static unsigned int
948 smallest_imm_type (offsetT num)
949 {
950 if (cpu_arch_flags != (Cpu186 | Cpu286 | Cpu386 | Cpu486 | CpuNo64))
951 {
952 /* This code is disabled on the 486 because all the Imm1 forms
953 in the opcode table are slower on the i486. They're the
954 versions with the implicitly specified single-position
955 displacement, which has another syntax if you really want to
956 use that form. */
957 if (num == 1)
958 return Imm1 | Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64;
959 }
960 return (fits_in_signed_byte (num)
961 ? (Imm8S | Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
962 : fits_in_unsigned_byte (num)
963 ? (Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
964 : (fits_in_signed_word (num) || fits_in_unsigned_word (num))
965 ? (Imm16 | Imm32 | Imm32S | Imm64)
966 : fits_in_signed_long (num)
967 ? (Imm32 | Imm32S | Imm64)
968 : fits_in_unsigned_long (num)
969 ? (Imm32 | Imm64)
970 : Imm64);
971 }
972
973 static offsetT
974 offset_in_range (offsetT val, int size)
975 {
976 addressT mask;
977
978 switch (size)
979 {
980 case 1: mask = ((addressT) 1 << 8) - 1; break;
981 case 2: mask = ((addressT) 1 << 16) - 1; break;
982 case 4: mask = ((addressT) 2 << 31) - 1; break;
983 #ifdef BFD64
984 case 8: mask = ((addressT) 2 << 63) - 1; break;
985 #endif
986 default: abort ();
987 }
988
989 /* If BFD64, sign extend val. */
990 if (!use_rela_relocations)
991 if ((val & ~(((addressT) 2 << 31) - 1)) == 0)
992 val = (val ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
993
994 if ((val & ~mask) != 0 && (val & ~mask) != ~mask)
995 {
996 char buf1[40], buf2[40];
997
998 sprint_value (buf1, val);
999 sprint_value (buf2, val & mask);
1000 as_warn (_("%s shortened to %s"), buf1, buf2);
1001 }
1002 return val & mask;
1003 }
1004
1005 /* Returns 0 if attempting to add a prefix where one from the same
1006 class already exists, 1 if non rep/repne added, 2 if rep/repne
1007 added. */
1008 static int
1009 add_prefix (unsigned int prefix)
1010 {
1011 int ret = 1;
1012 unsigned int q;
1013
1014 if (prefix >= REX_OPCODE && prefix < REX_OPCODE + 16
1015 && flag_code == CODE_64BIT)
1016 {
1017 if ((i.prefix[REX_PREFIX] & prefix & REX_W)
1018 || ((i.prefix[REX_PREFIX] & (REX_R | REX_X | REX_B))
1019 && (prefix & (REX_R | REX_X | REX_B))))
1020 ret = 0;
1021 q = REX_PREFIX;
1022 }
1023 else
1024 {
1025 switch (prefix)
1026 {
1027 default:
1028 abort ();
1029
1030 case CS_PREFIX_OPCODE:
1031 case DS_PREFIX_OPCODE:
1032 case ES_PREFIX_OPCODE:
1033 case FS_PREFIX_OPCODE:
1034 case GS_PREFIX_OPCODE:
1035 case SS_PREFIX_OPCODE:
1036 q = SEG_PREFIX;
1037 break;
1038
1039 case REPNE_PREFIX_OPCODE:
1040 case REPE_PREFIX_OPCODE:
1041 ret = 2;
1042 /* fall thru */
1043 case LOCK_PREFIX_OPCODE:
1044 q = LOCKREP_PREFIX;
1045 break;
1046
1047 case FWAIT_OPCODE:
1048 q = WAIT_PREFIX;
1049 break;
1050
1051 case ADDR_PREFIX_OPCODE:
1052 q = ADDR_PREFIX;
1053 break;
1054
1055 case DATA_PREFIX_OPCODE:
1056 q = DATA_PREFIX;
1057 break;
1058 }
1059 if (i.prefix[q] != 0)
1060 ret = 0;
1061 }
1062
1063 if (ret)
1064 {
1065 if (!i.prefix[q])
1066 ++i.prefixes;
1067 i.prefix[q] |= prefix;
1068 }
1069 else
1070 as_bad (_("same type of prefix used twice"));
1071
1072 return ret;
1073 }
1074
1075 static void
1076 set_code_flag (int value)
1077 {
1078 flag_code = value;
1079 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
1080 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
1081 if (value == CODE_64BIT && !(cpu_arch_flags & CpuSledgehammer))
1082 {
1083 as_bad (_("64bit mode not supported on this CPU."));
1084 }
1085 if (value == CODE_32BIT && !(cpu_arch_flags & Cpu386))
1086 {
1087 as_bad (_("32bit mode not supported on this CPU."));
1088 }
1089 stackop_size = '\0';
1090 }
1091
1092 static void
1093 set_16bit_gcc_code_flag (int new_code_flag)
1094 {
1095 flag_code = new_code_flag;
1096 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
1097 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
1098 stackop_size = LONG_MNEM_SUFFIX;
1099 }
1100
1101 static void
1102 set_intel_syntax (int syntax_flag)
1103 {
1104 /* Find out if register prefixing is specified. */
1105 int ask_naked_reg = 0;
1106
1107 SKIP_WHITESPACE ();
1108 if (!is_end_of_line[(unsigned char) *input_line_pointer])
1109 {
1110 char *string = input_line_pointer;
1111 int e = get_symbol_end ();
1112
1113 if (strcmp (string, "prefix") == 0)
1114 ask_naked_reg = 1;
1115 else if (strcmp (string, "noprefix") == 0)
1116 ask_naked_reg = -1;
1117 else
1118 as_bad (_("bad argument to syntax directive."));
1119 *input_line_pointer = e;
1120 }
1121 demand_empty_rest_of_line ();
1122
1123 intel_syntax = syntax_flag;
1124
1125 if (ask_naked_reg == 0)
1126 allow_naked_reg = (intel_syntax
1127 && (bfd_get_symbol_leading_char (stdoutput) != '\0'));
1128 else
1129 allow_naked_reg = (ask_naked_reg < 0);
1130
1131 identifier_chars['%'] = intel_syntax && allow_naked_reg ? '%' : 0;
1132 identifier_chars['$'] = intel_syntax ? '$' : 0;
1133 register_prefix = allow_naked_reg ? "" : "%";
1134 }
1135
1136 static void
1137 set_cpu_arch (int dummy ATTRIBUTE_UNUSED)
1138 {
1139 SKIP_WHITESPACE ();
1140
1141 if (!is_end_of_line[(unsigned char) *input_line_pointer])
1142 {
1143 char *string = input_line_pointer;
1144 int e = get_symbol_end ();
1145 unsigned int i;
1146
1147 for (i = 0; i < ARRAY_SIZE (cpu_arch); i++)
1148 {
1149 if (strcmp (string, cpu_arch[i].name) == 0)
1150 {
1151 if (*string != '.')
1152 {
1153 cpu_arch_name = cpu_arch[i].name;
1154 cpu_sub_arch_name = NULL;
1155 cpu_arch_flags = (cpu_arch[i].flags
1156 | (flag_code == CODE_64BIT
1157 ? Cpu64 : CpuNo64));
1158 cpu_arch_isa = cpu_arch[i].type;
1159 cpu_arch_isa_flags = cpu_arch[i].flags;
1160 if (!cpu_arch_tune_set)
1161 {
1162 cpu_arch_tune = cpu_arch_isa;
1163 cpu_arch_tune_flags = cpu_arch_isa_flags;
1164 }
1165 break;
1166 }
1167 if ((cpu_arch_flags | cpu_arch[i].flags) != cpu_arch_flags)
1168 {
1169 cpu_sub_arch_name = cpu_arch[i].name;
1170 cpu_arch_flags |= cpu_arch[i].flags;
1171 }
1172 *input_line_pointer = e;
1173 demand_empty_rest_of_line ();
1174 return;
1175 }
1176 }
1177 if (i >= ARRAY_SIZE (cpu_arch))
1178 as_bad (_("no such architecture: `%s'"), string);
1179
1180 *input_line_pointer = e;
1181 }
1182 else
1183 as_bad (_("missing cpu architecture"));
1184
1185 no_cond_jump_promotion = 0;
1186 if (*input_line_pointer == ','
1187 && !is_end_of_line[(unsigned char) input_line_pointer[1]])
1188 {
1189 char *string = ++input_line_pointer;
1190 int e = get_symbol_end ();
1191
1192 if (strcmp (string, "nojumps") == 0)
1193 no_cond_jump_promotion = 1;
1194 else if (strcmp (string, "jumps") == 0)
1195 ;
1196 else
1197 as_bad (_("no such architecture modifier: `%s'"), string);
1198
1199 *input_line_pointer = e;
1200 }
1201
1202 demand_empty_rest_of_line ();
1203 }
1204
1205 unsigned long
1206 i386_mach ()
1207 {
1208 if (!strcmp (default_arch, "x86_64"))
1209 return bfd_mach_x86_64;
1210 else if (!strcmp (default_arch, "i386"))
1211 return bfd_mach_i386_i386;
1212 else
1213 as_fatal (_("Unknown architecture"));
1214 }
1215 \f
1216 void
1217 md_begin ()
1218 {
1219 const char *hash_err;
1220
1221 /* Initialize op_hash hash table. */
1222 op_hash = hash_new ();
1223
1224 {
1225 const template *optab;
1226 templates *core_optab;
1227
1228 /* Setup for loop. */
1229 optab = i386_optab;
1230 core_optab = (templates *) xmalloc (sizeof (templates));
1231 core_optab->start = optab;
1232
1233 while (1)
1234 {
1235 ++optab;
1236 if (optab->name == NULL
1237 || strcmp (optab->name, (optab - 1)->name) != 0)
1238 {
1239 /* different name --> ship out current template list;
1240 add to hash table; & begin anew. */
1241 core_optab->end = optab;
1242 hash_err = hash_insert (op_hash,
1243 (optab - 1)->name,
1244 (PTR) core_optab);
1245 if (hash_err)
1246 {
1247 as_fatal (_("Internal Error: Can't hash %s: %s"),
1248 (optab - 1)->name,
1249 hash_err);
1250 }
1251 if (optab->name == NULL)
1252 break;
1253 core_optab = (templates *) xmalloc (sizeof (templates));
1254 core_optab->start = optab;
1255 }
1256 }
1257 }
1258
1259 /* Initialize reg_hash hash table. */
1260 reg_hash = hash_new ();
1261 {
1262 const reg_entry *regtab;
1263 unsigned int regtab_size = i386_regtab_size;
1264
1265 for (regtab = i386_regtab; regtab_size--; regtab++)
1266 {
1267 hash_err = hash_insert (reg_hash, regtab->reg_name, (PTR) regtab);
1268 if (hash_err)
1269 as_fatal (_("Internal Error: Can't hash %s: %s"),
1270 regtab->reg_name,
1271 hash_err);
1272 }
1273 }
1274
1275 /* Fill in lexical tables: mnemonic_chars, operand_chars. */
1276 {
1277 int c;
1278 char *p;
1279
1280 for (c = 0; c < 256; c++)
1281 {
1282 if (ISDIGIT (c))
1283 {
1284 digit_chars[c] = c;
1285 mnemonic_chars[c] = c;
1286 register_chars[c] = c;
1287 operand_chars[c] = c;
1288 }
1289 else if (ISLOWER (c))
1290 {
1291 mnemonic_chars[c] = c;
1292 register_chars[c] = c;
1293 operand_chars[c] = c;
1294 }
1295 else if (ISUPPER (c))
1296 {
1297 mnemonic_chars[c] = TOLOWER (c);
1298 register_chars[c] = mnemonic_chars[c];
1299 operand_chars[c] = c;
1300 }
1301
1302 if (ISALPHA (c) || ISDIGIT (c))
1303 identifier_chars[c] = c;
1304 else if (c >= 128)
1305 {
1306 identifier_chars[c] = c;
1307 operand_chars[c] = c;
1308 }
1309 }
1310
1311 #ifdef LEX_AT
1312 identifier_chars['@'] = '@';
1313 #endif
1314 #ifdef LEX_QM
1315 identifier_chars['?'] = '?';
1316 operand_chars['?'] = '?';
1317 #endif
1318 digit_chars['-'] = '-';
1319 mnemonic_chars['-'] = '-';
1320 mnemonic_chars['.'] = '.';
1321 identifier_chars['_'] = '_';
1322 identifier_chars['.'] = '.';
1323
1324 for (p = operand_special_chars; *p != '\0'; p++)
1325 operand_chars[(unsigned char) *p] = *p;
1326 }
1327
1328 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1329 if (IS_ELF)
1330 {
1331 record_alignment (text_section, 2);
1332 record_alignment (data_section, 2);
1333 record_alignment (bss_section, 2);
1334 }
1335 #endif
1336
1337 if (flag_code == CODE_64BIT)
1338 {
1339 x86_dwarf2_return_column = 16;
1340 x86_cie_data_alignment = -8;
1341 }
1342 else
1343 {
1344 x86_dwarf2_return_column = 8;
1345 x86_cie_data_alignment = -4;
1346 }
1347 }
1348
1349 void
1350 i386_print_statistics (FILE *file)
1351 {
1352 hash_print_statistics (file, "i386 opcode", op_hash);
1353 hash_print_statistics (file, "i386 register", reg_hash);
1354 }
1355 \f
1356 #ifdef DEBUG386
1357
1358 /* Debugging routines for md_assemble. */
1359 static void pte (template *);
1360 static void pt (unsigned int);
1361 static void pe (expressionS *);
1362 static void ps (symbolS *);
1363
1364 static void
1365 pi (char *line, i386_insn *x)
1366 {
1367 unsigned int i;
1368
1369 fprintf (stdout, "%s: template ", line);
1370 pte (&x->tm);
1371 fprintf (stdout, " address: base %s index %s scale %x\n",
1372 x->base_reg ? x->base_reg->reg_name : "none",
1373 x->index_reg ? x->index_reg->reg_name : "none",
1374 x->log2_scale_factor);
1375 fprintf (stdout, " modrm: mode %x reg %x reg/mem %x\n",
1376 x->rm.mode, x->rm.reg, x->rm.regmem);
1377 fprintf (stdout, " sib: base %x index %x scale %x\n",
1378 x->sib.base, x->sib.index, x->sib.scale);
1379 fprintf (stdout, " rex: 64bit %x extX %x extY %x extZ %x\n",
1380 (x->rex & REX_W) != 0,
1381 (x->rex & REX_R) != 0,
1382 (x->rex & REX_X) != 0,
1383 (x->rex & REX_B) != 0);
1384 for (i = 0; i < x->operands; i++)
1385 {
1386 fprintf (stdout, " #%d: ", i + 1);
1387 pt (x->types[i]);
1388 fprintf (stdout, "\n");
1389 if (x->types[i]
1390 & (Reg | SReg2 | SReg3 | Control | Debug | Test | RegMMX | RegXMM))
1391 fprintf (stdout, "%s\n", x->op[i].regs->reg_name);
1392 if (x->types[i] & Imm)
1393 pe (x->op[i].imms);
1394 if (x->types[i] & Disp)
1395 pe (x->op[i].disps);
1396 }
1397 }
1398
1399 static void
1400 pte (template *t)
1401 {
1402 unsigned int i;
1403 fprintf (stdout, " %d operands ", t->operands);
1404 fprintf (stdout, "opcode %x ", t->base_opcode);
1405 if (t->extension_opcode != None)
1406 fprintf (stdout, "ext %x ", t->extension_opcode);
1407 if (t->opcode_modifier & D)
1408 fprintf (stdout, "D");
1409 if (t->opcode_modifier & W)
1410 fprintf (stdout, "W");
1411 fprintf (stdout, "\n");
1412 for (i = 0; i < t->operands; i++)
1413 {
1414 fprintf (stdout, " #%d type ", i + 1);
1415 pt (t->operand_types[i]);
1416 fprintf (stdout, "\n");
1417 }
1418 }
1419
1420 static void
1421 pe (expressionS *e)
1422 {
1423 fprintf (stdout, " operation %d\n", e->X_op);
1424 fprintf (stdout, " add_number %ld (%lx)\n",
1425 (long) e->X_add_number, (long) e->X_add_number);
1426 if (e->X_add_symbol)
1427 {
1428 fprintf (stdout, " add_symbol ");
1429 ps (e->X_add_symbol);
1430 fprintf (stdout, "\n");
1431 }
1432 if (e->X_op_symbol)
1433 {
1434 fprintf (stdout, " op_symbol ");
1435 ps (e->X_op_symbol);
1436 fprintf (stdout, "\n");
1437 }
1438 }
1439
1440 static void
1441 ps (symbolS *s)
1442 {
1443 fprintf (stdout, "%s type %s%s",
1444 S_GET_NAME (s),
1445 S_IS_EXTERNAL (s) ? "EXTERNAL " : "",
1446 segment_name (S_GET_SEGMENT (s)));
1447 }
1448
1449 static struct type_name
1450 {
1451 unsigned int mask;
1452 char *tname;
1453 }
1454 const type_names[] =
1455 {
1456 { Reg8, "r8" },
1457 { Reg16, "r16" },
1458 { Reg32, "r32" },
1459 { Reg64, "r64" },
1460 { Imm8, "i8" },
1461 { Imm8S, "i8s" },
1462 { Imm16, "i16" },
1463 { Imm32, "i32" },
1464 { Imm32S, "i32s" },
1465 { Imm64, "i64" },
1466 { Imm1, "i1" },
1467 { BaseIndex, "BaseIndex" },
1468 { Disp8, "d8" },
1469 { Disp16, "d16" },
1470 { Disp32, "d32" },
1471 { Disp32S, "d32s" },
1472 { Disp64, "d64" },
1473 { InOutPortReg, "InOutPortReg" },
1474 { ShiftCount, "ShiftCount" },
1475 { Control, "control reg" },
1476 { Test, "test reg" },
1477 { Debug, "debug reg" },
1478 { FloatReg, "FReg" },
1479 { FloatAcc, "FAcc" },
1480 { SReg2, "SReg2" },
1481 { SReg3, "SReg3" },
1482 { Acc, "Acc" },
1483 { JumpAbsolute, "Jump Absolute" },
1484 { RegMMX, "rMMX" },
1485 { RegXMM, "rXMM" },
1486 { EsSeg, "es" },
1487 { 0, "" }
1488 };
1489
1490 static void
1491 pt (t)
1492 unsigned int t;
1493 {
1494 const struct type_name *ty;
1495
1496 for (ty = type_names; ty->mask; ty++)
1497 if (t & ty->mask)
1498 fprintf (stdout, "%s, ", ty->tname);
1499 fflush (stdout);
1500 }
1501
1502 #endif /* DEBUG386 */
1503 \f
1504 static bfd_reloc_code_real_type
1505 reloc (unsigned int size,
1506 int pcrel,
1507 int sign,
1508 bfd_reloc_code_real_type other)
1509 {
1510 if (other != NO_RELOC)
1511 {
1512 reloc_howto_type *reloc;
1513
1514 if (size == 8)
1515 switch (other)
1516 {
1517 case BFD_RELOC_X86_64_GOT32:
1518 return BFD_RELOC_X86_64_GOT64;
1519 break;
1520 case BFD_RELOC_X86_64_PLTOFF64:
1521 return BFD_RELOC_X86_64_PLTOFF64;
1522 break;
1523 case BFD_RELOC_X86_64_GOTPC32:
1524 other = BFD_RELOC_X86_64_GOTPC64;
1525 break;
1526 case BFD_RELOC_X86_64_GOTPCREL:
1527 other = BFD_RELOC_X86_64_GOTPCREL64;
1528 break;
1529 case BFD_RELOC_X86_64_TPOFF32:
1530 other = BFD_RELOC_X86_64_TPOFF64;
1531 break;
1532 case BFD_RELOC_X86_64_DTPOFF32:
1533 other = BFD_RELOC_X86_64_DTPOFF64;
1534 break;
1535 default:
1536 break;
1537 }
1538
1539 /* Sign-checking 4-byte relocations in 16-/32-bit code is pointless. */
1540 if (size == 4 && flag_code != CODE_64BIT)
1541 sign = -1;
1542
1543 reloc = bfd_reloc_type_lookup (stdoutput, other);
1544 if (!reloc)
1545 as_bad (_("unknown relocation (%u)"), other);
1546 else if (size != bfd_get_reloc_size (reloc))
1547 as_bad (_("%u-byte relocation cannot be applied to %u-byte field"),
1548 bfd_get_reloc_size (reloc),
1549 size);
1550 else if (pcrel && !reloc->pc_relative)
1551 as_bad (_("non-pc-relative relocation for pc-relative field"));
1552 else if ((reloc->complain_on_overflow == complain_overflow_signed
1553 && !sign)
1554 || (reloc->complain_on_overflow == complain_overflow_unsigned
1555 && sign > 0))
1556 as_bad (_("relocated field and relocation type differ in signedness"));
1557 else
1558 return other;
1559 return NO_RELOC;
1560 }
1561
1562 if (pcrel)
1563 {
1564 if (!sign)
1565 as_bad (_("there are no unsigned pc-relative relocations"));
1566 switch (size)
1567 {
1568 case 1: return BFD_RELOC_8_PCREL;
1569 case 2: return BFD_RELOC_16_PCREL;
1570 case 4: return BFD_RELOC_32_PCREL;
1571 case 8: return BFD_RELOC_64_PCREL;
1572 }
1573 as_bad (_("cannot do %u byte pc-relative relocation"), size);
1574 }
1575 else
1576 {
1577 if (sign > 0)
1578 switch (size)
1579 {
1580 case 4: return BFD_RELOC_X86_64_32S;
1581 }
1582 else
1583 switch (size)
1584 {
1585 case 1: return BFD_RELOC_8;
1586 case 2: return BFD_RELOC_16;
1587 case 4: return BFD_RELOC_32;
1588 case 8: return BFD_RELOC_64;
1589 }
1590 as_bad (_("cannot do %s %u byte relocation"),
1591 sign > 0 ? "signed" : "unsigned", size);
1592 }
1593
1594 abort ();
1595 return BFD_RELOC_NONE;
1596 }
1597
1598 /* Here we decide which fixups can be adjusted to make them relative to
1599 the beginning of the section instead of the symbol. Basically we need
1600 to make sure that the dynamic relocations are done correctly, so in
1601 some cases we force the original symbol to be used. */
1602
1603 int
1604 tc_i386_fix_adjustable (fixS *fixP ATTRIBUTE_UNUSED)
1605 {
1606 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1607 if (!IS_ELF)
1608 return 1;
1609
1610 /* Don't adjust pc-relative references to merge sections in 64-bit
1611 mode. */
1612 if (use_rela_relocations
1613 && (S_GET_SEGMENT (fixP->fx_addsy)->flags & SEC_MERGE) != 0
1614 && fixP->fx_pcrel)
1615 return 0;
1616
1617 /* The x86_64 GOTPCREL are represented as 32bit PCrel relocations
1618 and changed later by validate_fix. */
1619 if (GOT_symbol && fixP->fx_subsy == GOT_symbol
1620 && fixP->fx_r_type == BFD_RELOC_32_PCREL)
1621 return 0;
1622
1623 /* adjust_reloc_syms doesn't know about the GOT. */
1624 if (fixP->fx_r_type == BFD_RELOC_386_GOTOFF
1625 || fixP->fx_r_type == BFD_RELOC_386_PLT32
1626 || fixP->fx_r_type == BFD_RELOC_386_GOT32
1627 || fixP->fx_r_type == BFD_RELOC_386_TLS_GD
1628 || fixP->fx_r_type == BFD_RELOC_386_TLS_LDM
1629 || fixP->fx_r_type == BFD_RELOC_386_TLS_LDO_32
1630 || fixP->fx_r_type == BFD_RELOC_386_TLS_IE_32
1631 || fixP->fx_r_type == BFD_RELOC_386_TLS_IE
1632 || fixP->fx_r_type == BFD_RELOC_386_TLS_GOTIE
1633 || fixP->fx_r_type == BFD_RELOC_386_TLS_LE_32
1634 || fixP->fx_r_type == BFD_RELOC_386_TLS_LE
1635 || fixP->fx_r_type == BFD_RELOC_386_TLS_GOTDESC
1636 || fixP->fx_r_type == BFD_RELOC_386_TLS_DESC_CALL
1637 || fixP->fx_r_type == BFD_RELOC_X86_64_PLT32
1638 || fixP->fx_r_type == BFD_RELOC_X86_64_GOT32
1639 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPCREL
1640 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSGD
1641 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSLD
1642 || fixP->fx_r_type == BFD_RELOC_X86_64_DTPOFF32
1643 || fixP->fx_r_type == BFD_RELOC_X86_64_DTPOFF64
1644 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTTPOFF
1645 || fixP->fx_r_type == BFD_RELOC_X86_64_TPOFF32
1646 || fixP->fx_r_type == BFD_RELOC_X86_64_TPOFF64
1647 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTOFF64
1648 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPC32_TLSDESC
1649 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSDESC_CALL
1650 || fixP->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1651 || fixP->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1652 return 0;
1653 #endif
1654 return 1;
1655 }
1656
1657 static int
1658 intel_float_operand (const char *mnemonic)
1659 {
1660 /* Note that the value returned is meaningful only for opcodes with (memory)
1661 operands, hence the code here is free to improperly handle opcodes that
1662 have no operands (for better performance and smaller code). */
1663
1664 if (mnemonic[0] != 'f')
1665 return 0; /* non-math */
1666
1667 switch (mnemonic[1])
1668 {
1669 /* fclex, fdecstp, fdisi, femms, feni, fincstp, finit, fsetpm, and
1670 the fs segment override prefix not currently handled because no
1671 call path can make opcodes without operands get here */
1672 case 'i':
1673 return 2 /* integer op */;
1674 case 'l':
1675 if (mnemonic[2] == 'd' && (mnemonic[3] == 'c' || mnemonic[3] == 'e'))
1676 return 3; /* fldcw/fldenv */
1677 break;
1678 case 'n':
1679 if (mnemonic[2] != 'o' /* fnop */)
1680 return 3; /* non-waiting control op */
1681 break;
1682 case 'r':
1683 if (mnemonic[2] == 's')
1684 return 3; /* frstor/frstpm */
1685 break;
1686 case 's':
1687 if (mnemonic[2] == 'a')
1688 return 3; /* fsave */
1689 if (mnemonic[2] == 't')
1690 {
1691 switch (mnemonic[3])
1692 {
1693 case 'c': /* fstcw */
1694 case 'd': /* fstdw */
1695 case 'e': /* fstenv */
1696 case 's': /* fsts[gw] */
1697 return 3;
1698 }
1699 }
1700 break;
1701 case 'x':
1702 if (mnemonic[2] == 'r' || mnemonic[2] == 's')
1703 return 0; /* fxsave/fxrstor are not really math ops */
1704 break;
1705 }
1706
1707 return 1;
1708 }
1709
1710 /* This is the guts of the machine-dependent assembler. LINE points to a
1711 machine dependent instruction. This function is supposed to emit
1712 the frags/bytes it assembles to. */
1713
1714 void
1715 md_assemble (line)
1716 char *line;
1717 {
1718 int j;
1719 char mnemonic[MAX_MNEM_SIZE];
1720
1721 /* Initialize globals. */
1722 memset (&i, '\0', sizeof (i));
1723 for (j = 0; j < MAX_OPERANDS; j++)
1724 i.reloc[j] = NO_RELOC;
1725 memset (disp_expressions, '\0', sizeof (disp_expressions));
1726 memset (im_expressions, '\0', sizeof (im_expressions));
1727 save_stack_p = save_stack;
1728
1729 /* First parse an instruction mnemonic & call i386_operand for the operands.
1730 We assume that the scrubber has arranged it so that line[0] is the valid
1731 start of a (possibly prefixed) mnemonic. */
1732
1733 line = parse_insn (line, mnemonic);
1734 if (line == NULL)
1735 return;
1736
1737 line = parse_operands (line, mnemonic);
1738 if (line == NULL)
1739 return;
1740
1741 /* The order of the immediates should be reversed
1742 for 2 immediates extrq and insertq instructions */
1743 if ((i.imm_operands == 2)
1744 && ((strcmp (mnemonic, "extrq") == 0)
1745 || (strcmp (mnemonic, "insertq") == 0)))
1746 {
1747 swap_2_operands (0, 1);
1748 /* "extrq" and insertq" are the only two instructions whose operands
1749 have to be reversed even though they have two immediate operands.
1750 */
1751 if (intel_syntax)
1752 swap_operands ();
1753 }
1754
1755 /* Now we've parsed the mnemonic into a set of templates, and have the
1756 operands at hand. */
1757
1758 /* All intel opcodes have reversed operands except for "bound" and
1759 "enter". We also don't reverse intersegment "jmp" and "call"
1760 instructions with 2 immediate operands so that the immediate segment
1761 precedes the offset, as it does when in AT&T mode. */
1762 if (intel_syntax
1763 && i.operands > 1
1764 && (strcmp (mnemonic, "bound") != 0)
1765 && (strcmp (mnemonic, "invlpga") != 0)
1766 && !((i.types[0] & Imm) && (i.types[1] & Imm)))
1767 swap_operands ();
1768
1769 if (i.imm_operands)
1770 optimize_imm ();
1771
1772 /* Don't optimize displacement for movabs since it only takes 64bit
1773 displacement. */
1774 if (i.disp_operands
1775 && (flag_code != CODE_64BIT
1776 || strcmp (mnemonic, "movabs") != 0))
1777 optimize_disp ();
1778
1779 /* Next, we find a template that matches the given insn,
1780 making sure the overlap of the given operands types is consistent
1781 with the template operand types. */
1782
1783 if (!match_template ())
1784 return;
1785
1786 if (intel_syntax)
1787 {
1788 /* Undo SYSV386_COMPAT brokenness when in Intel mode. See i386.h */
1789 if (SYSV386_COMPAT
1790 && (i.tm.base_opcode & 0xfffffde0) == 0xdce0)
1791 i.tm.base_opcode ^= Opcode_FloatR;
1792
1793 /* Zap movzx and movsx suffix. The suffix may have been set from
1794 "word ptr" or "byte ptr" on the source operand, but we'll use
1795 the suffix later to choose the destination register. */
1796 if ((i.tm.base_opcode & ~9) == 0x0fb6)
1797 {
1798 if (i.reg_operands < 2
1799 && !i.suffix
1800 && (~i.tm.opcode_modifier
1801 & (No_bSuf
1802 | No_wSuf
1803 | No_lSuf
1804 | No_sSuf
1805 | No_xSuf
1806 | No_qSuf)))
1807 as_bad (_("ambiguous operand size for `%s'"), i.tm.name);
1808
1809 i.suffix = 0;
1810 }
1811 }
1812
1813 if (i.tm.opcode_modifier & FWait)
1814 if (!add_prefix (FWAIT_OPCODE))
1815 return;
1816
1817 /* Check string instruction segment overrides. */
1818 if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0)
1819 {
1820 if (!check_string ())
1821 return;
1822 }
1823
1824 if (!process_suffix ())
1825 return;
1826
1827 /* Make still unresolved immediate matches conform to size of immediate
1828 given in i.suffix. */
1829 if (!finalize_imm ())
1830 return;
1831
1832 if (i.types[0] & Imm1)
1833 i.imm_operands = 0; /* kludge for shift insns. */
1834 if (i.types[0] & ImplicitRegister)
1835 i.reg_operands--;
1836 if (i.types[1] & ImplicitRegister)
1837 i.reg_operands--;
1838 if (i.types[2] & ImplicitRegister)
1839 i.reg_operands--;
1840
1841 if (i.tm.opcode_modifier & ImmExt)
1842 {
1843 expressionS *exp;
1844
1845 if ((i.tm.cpu_flags & CpuSSE3) && i.operands > 0)
1846 {
1847 /* Streaming SIMD extensions 3 Instructions have the fixed
1848 operands with an opcode suffix which is coded in the same
1849 place as an 8-bit immediate field would be. Here we check
1850 those operands and remove them afterwards. */
1851 unsigned int x;
1852
1853 for (x = 0; x < i.operands; x++)
1854 if (i.op[x].regs->reg_num != x)
1855 as_bad (_("can't use register '%s%s' as operand %d in '%s'."),
1856 register_prefix,
1857 i.op[x].regs->reg_name,
1858 x + 1,
1859 i.tm.name);
1860 i.operands = 0;
1861 }
1862
1863 /* These AMD 3DNow! and Intel Katmai New Instructions have an
1864 opcode suffix which is coded in the same place as an 8-bit
1865 immediate field would be. Here we fake an 8-bit immediate
1866 operand from the opcode suffix stored in tm.extension_opcode. */
1867
1868 assert (i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS);
1869
1870 exp = &im_expressions[i.imm_operands++];
1871 i.op[i.operands].imms = exp;
1872 i.types[i.operands++] = Imm8;
1873 exp->X_op = O_constant;
1874 exp->X_add_number = i.tm.extension_opcode;
1875 i.tm.extension_opcode = None;
1876 }
1877
1878 /* For insns with operands there are more diddles to do to the opcode. */
1879 if (i.operands)
1880 {
1881 if (!process_operands ())
1882 return;
1883 }
1884 else if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
1885 {
1886 /* UnixWare fsub no args is alias for fsubp, fadd -> faddp, etc. */
1887 as_warn (_("translating to `%sp'"), i.tm.name);
1888 }
1889
1890 /* Handle conversion of 'int $3' --> special int3 insn. */
1891 if (i.tm.base_opcode == INT_OPCODE && i.op[0].imms->X_add_number == 3)
1892 {
1893 i.tm.base_opcode = INT3_OPCODE;
1894 i.imm_operands = 0;
1895 }
1896
1897 if ((i.tm.opcode_modifier & (Jump | JumpByte | JumpDword))
1898 && i.op[0].disps->X_op == O_constant)
1899 {
1900 /* Convert "jmp constant" (and "call constant") to a jump (call) to
1901 the absolute address given by the constant. Since ix86 jumps and
1902 calls are pc relative, we need to generate a reloc. */
1903 i.op[0].disps->X_add_symbol = &abs_symbol;
1904 i.op[0].disps->X_op = O_symbol;
1905 }
1906
1907 if ((i.tm.opcode_modifier & Rex64) != 0)
1908 i.rex |= REX_W;
1909
1910 /* For 8 bit registers we need an empty rex prefix. Also if the
1911 instruction already has a prefix, we need to convert old
1912 registers to new ones. */
1913
1914 if (((i.types[0] & Reg8) != 0
1915 && (i.op[0].regs->reg_flags & RegRex64) != 0)
1916 || ((i.types[1] & Reg8) != 0
1917 && (i.op[1].regs->reg_flags & RegRex64) != 0)
1918 || (((i.types[0] & Reg8) != 0 || (i.types[1] & Reg8) != 0)
1919 && i.rex != 0))
1920 {
1921 int x;
1922
1923 i.rex |= REX_OPCODE;
1924 for (x = 0; x < 2; x++)
1925 {
1926 /* Look for 8 bit operand that uses old registers. */
1927 if ((i.types[x] & Reg8) != 0
1928 && (i.op[x].regs->reg_flags & RegRex64) == 0)
1929 {
1930 /* In case it is "hi" register, give up. */
1931 if (i.op[x].regs->reg_num > 3)
1932 as_bad (_("can't encode register '%s%s' in an "
1933 "instruction requiring REX prefix."),
1934 register_prefix, i.op[x].regs->reg_name);
1935
1936 /* Otherwise it is equivalent to the extended register.
1937 Since the encoding doesn't change this is merely
1938 cosmetic cleanup for debug output. */
1939
1940 i.op[x].regs = i.op[x].regs + 8;
1941 }
1942 }
1943 }
1944
1945 if (i.rex != 0)
1946 add_prefix (REX_OPCODE | i.rex);
1947
1948 /* We are ready to output the insn. */
1949 output_insn ();
1950 }
1951
1952 static char *
1953 parse_insn (char *line, char *mnemonic)
1954 {
1955 char *l = line;
1956 char *token_start = l;
1957 char *mnem_p;
1958 int supported;
1959 const template *t;
1960
1961 /* Non-zero if we found a prefix only acceptable with string insns. */
1962 const char *expecting_string_instruction = NULL;
1963
1964 while (1)
1965 {
1966 mnem_p = mnemonic;
1967 while ((*mnem_p = mnemonic_chars[(unsigned char) *l]) != 0)
1968 {
1969 mnem_p++;
1970 if (mnem_p >= mnemonic + MAX_MNEM_SIZE)
1971 {
1972 as_bad (_("no such instruction: `%s'"), token_start);
1973 return NULL;
1974 }
1975 l++;
1976 }
1977 if (!is_space_char (*l)
1978 && *l != END_OF_INSN
1979 && (intel_syntax
1980 || (*l != PREFIX_SEPARATOR
1981 && *l != ',')))
1982 {
1983 as_bad (_("invalid character %s in mnemonic"),
1984 output_invalid (*l));
1985 return NULL;
1986 }
1987 if (token_start == l)
1988 {
1989 if (!intel_syntax && *l == PREFIX_SEPARATOR)
1990 as_bad (_("expecting prefix; got nothing"));
1991 else
1992 as_bad (_("expecting mnemonic; got nothing"));
1993 return NULL;
1994 }
1995
1996 /* Look up instruction (or prefix) via hash table. */
1997 current_templates = hash_find (op_hash, mnemonic);
1998
1999 if (*l != END_OF_INSN
2000 && (!is_space_char (*l) || l[1] != END_OF_INSN)
2001 && current_templates
2002 && (current_templates->start->opcode_modifier & IsPrefix))
2003 {
2004 if (current_templates->start->cpu_flags
2005 & (flag_code != CODE_64BIT ? Cpu64 : CpuNo64))
2006 {
2007 as_bad ((flag_code != CODE_64BIT
2008 ? _("`%s' is only supported in 64-bit mode")
2009 : _("`%s' is not supported in 64-bit mode")),
2010 current_templates->start->name);
2011 return NULL;
2012 }
2013 /* If we are in 16-bit mode, do not allow addr16 or data16.
2014 Similarly, in 32-bit mode, do not allow addr32 or data32. */
2015 if ((current_templates->start->opcode_modifier & (Size16 | Size32))
2016 && flag_code != CODE_64BIT
2017 && (((current_templates->start->opcode_modifier & Size32) != 0)
2018 ^ (flag_code == CODE_16BIT)))
2019 {
2020 as_bad (_("redundant %s prefix"),
2021 current_templates->start->name);
2022 return NULL;
2023 }
2024 /* Add prefix, checking for repeated prefixes. */
2025 switch (add_prefix (current_templates->start->base_opcode))
2026 {
2027 case 0:
2028 return NULL;
2029 case 2:
2030 expecting_string_instruction = current_templates->start->name;
2031 break;
2032 }
2033 /* Skip past PREFIX_SEPARATOR and reset token_start. */
2034 token_start = ++l;
2035 }
2036 else
2037 break;
2038 }
2039
2040 if (!current_templates)
2041 {
2042 /* See if we can get a match by trimming off a suffix. */
2043 switch (mnem_p[-1])
2044 {
2045 case WORD_MNEM_SUFFIX:
2046 if (intel_syntax && (intel_float_operand (mnemonic) & 2))
2047 i.suffix = SHORT_MNEM_SUFFIX;
2048 else
2049 case BYTE_MNEM_SUFFIX:
2050 case QWORD_MNEM_SUFFIX:
2051 i.suffix = mnem_p[-1];
2052 mnem_p[-1] = '\0';
2053 current_templates = hash_find (op_hash, mnemonic);
2054 break;
2055 case SHORT_MNEM_SUFFIX:
2056 case LONG_MNEM_SUFFIX:
2057 if (!intel_syntax)
2058 {
2059 i.suffix = mnem_p[-1];
2060 mnem_p[-1] = '\0';
2061 current_templates = hash_find (op_hash, mnemonic);
2062 }
2063 break;
2064
2065 /* Intel Syntax. */
2066 case 'd':
2067 if (intel_syntax)
2068 {
2069 if (intel_float_operand (mnemonic) == 1)
2070 i.suffix = SHORT_MNEM_SUFFIX;
2071 else
2072 i.suffix = LONG_MNEM_SUFFIX;
2073 mnem_p[-1] = '\0';
2074 current_templates = hash_find (op_hash, mnemonic);
2075 }
2076 break;
2077 }
2078 if (!current_templates)
2079 {
2080 as_bad (_("no such instruction: `%s'"), token_start);
2081 return NULL;
2082 }
2083 }
2084
2085 if (current_templates->start->opcode_modifier & (Jump | JumpByte))
2086 {
2087 /* Check for a branch hint. We allow ",pt" and ",pn" for
2088 predict taken and predict not taken respectively.
2089 I'm not sure that branch hints actually do anything on loop
2090 and jcxz insns (JumpByte) for current Pentium4 chips. They
2091 may work in the future and it doesn't hurt to accept them
2092 now. */
2093 if (l[0] == ',' && l[1] == 'p')
2094 {
2095 if (l[2] == 't')
2096 {
2097 if (!add_prefix (DS_PREFIX_OPCODE))
2098 return NULL;
2099 l += 3;
2100 }
2101 else if (l[2] == 'n')
2102 {
2103 if (!add_prefix (CS_PREFIX_OPCODE))
2104 return NULL;
2105 l += 3;
2106 }
2107 }
2108 }
2109 /* Any other comma loses. */
2110 if (*l == ',')
2111 {
2112 as_bad (_("invalid character %s in mnemonic"),
2113 output_invalid (*l));
2114 return NULL;
2115 }
2116
2117 /* Check if instruction is supported on specified architecture. */
2118 supported = 0;
2119 for (t = current_templates->start; t < current_templates->end; ++t)
2120 {
2121 if (!((t->cpu_flags & ~(Cpu64 | CpuNo64))
2122 & ~(cpu_arch_flags & ~(Cpu64 | CpuNo64))))
2123 supported |= 1;
2124 if (!(t->cpu_flags & (flag_code == CODE_64BIT ? CpuNo64 : Cpu64)))
2125 supported |= 2;
2126 }
2127 if (!(supported & 2))
2128 {
2129 as_bad (flag_code == CODE_64BIT
2130 ? _("`%s' is not supported in 64-bit mode")
2131 : _("`%s' is only supported in 64-bit mode"),
2132 current_templates->start->name);
2133 return NULL;
2134 }
2135 if (!(supported & 1))
2136 {
2137 as_warn (_("`%s' is not supported on `%s%s'"),
2138 current_templates->start->name,
2139 cpu_arch_name,
2140 cpu_sub_arch_name ? cpu_sub_arch_name : "");
2141 }
2142 else if ((Cpu386 & ~cpu_arch_flags) && (flag_code != CODE_16BIT))
2143 {
2144 as_warn (_("use .code16 to ensure correct addressing mode"));
2145 }
2146
2147 /* Check for rep/repne without a string instruction. */
2148 if (expecting_string_instruction)
2149 {
2150 static templates override;
2151
2152 for (t = current_templates->start; t < current_templates->end; ++t)
2153 if (t->opcode_modifier & IsString)
2154 break;
2155 if (t >= current_templates->end)
2156 {
2157 as_bad (_("expecting string instruction after `%s'"),
2158 expecting_string_instruction);
2159 return NULL;
2160 }
2161 for (override.start = t; t < current_templates->end; ++t)
2162 if (!(t->opcode_modifier & IsString))
2163 break;
2164 override.end = t;
2165 current_templates = &override;
2166 }
2167
2168 return l;
2169 }
2170
2171 static char *
2172 parse_operands (char *l, const char *mnemonic)
2173 {
2174 char *token_start;
2175
2176 /* 1 if operand is pending after ','. */
2177 unsigned int expecting_operand = 0;
2178
2179 /* Non-zero if operand parens not balanced. */
2180 unsigned int paren_not_balanced;
2181
2182 while (*l != END_OF_INSN)
2183 {
2184 /* Skip optional white space before operand. */
2185 if (is_space_char (*l))
2186 ++l;
2187 if (!is_operand_char (*l) && *l != END_OF_INSN)
2188 {
2189 as_bad (_("invalid character %s before operand %d"),
2190 output_invalid (*l),
2191 i.operands + 1);
2192 return NULL;
2193 }
2194 token_start = l; /* after white space */
2195 paren_not_balanced = 0;
2196 while (paren_not_balanced || *l != ',')
2197 {
2198 if (*l == END_OF_INSN)
2199 {
2200 if (paren_not_balanced)
2201 {
2202 if (!intel_syntax)
2203 as_bad (_("unbalanced parenthesis in operand %d."),
2204 i.operands + 1);
2205 else
2206 as_bad (_("unbalanced brackets in operand %d."),
2207 i.operands + 1);
2208 return NULL;
2209 }
2210 else
2211 break; /* we are done */
2212 }
2213 else if (!is_operand_char (*l) && !is_space_char (*l))
2214 {
2215 as_bad (_("invalid character %s in operand %d"),
2216 output_invalid (*l),
2217 i.operands + 1);
2218 return NULL;
2219 }
2220 if (!intel_syntax)
2221 {
2222 if (*l == '(')
2223 ++paren_not_balanced;
2224 if (*l == ')')
2225 --paren_not_balanced;
2226 }
2227 else
2228 {
2229 if (*l == '[')
2230 ++paren_not_balanced;
2231 if (*l == ']')
2232 --paren_not_balanced;
2233 }
2234 l++;
2235 }
2236 if (l != token_start)
2237 { /* Yes, we've read in another operand. */
2238 unsigned int operand_ok;
2239 this_operand = i.operands++;
2240 if (i.operands > MAX_OPERANDS)
2241 {
2242 as_bad (_("spurious operands; (%d operands/instruction max)"),
2243 MAX_OPERANDS);
2244 return NULL;
2245 }
2246 /* Now parse operand adding info to 'i' as we go along. */
2247 END_STRING_AND_SAVE (l);
2248
2249 if (intel_syntax)
2250 operand_ok =
2251 i386_intel_operand (token_start,
2252 intel_float_operand (mnemonic));
2253 else
2254 operand_ok = i386_operand (token_start);
2255
2256 RESTORE_END_STRING (l);
2257 if (!operand_ok)
2258 return NULL;
2259 }
2260 else
2261 {
2262 if (expecting_operand)
2263 {
2264 expecting_operand_after_comma:
2265 as_bad (_("expecting operand after ','; got nothing"));
2266 return NULL;
2267 }
2268 if (*l == ',')
2269 {
2270 as_bad (_("expecting operand before ','; got nothing"));
2271 return NULL;
2272 }
2273 }
2274
2275 /* Now *l must be either ',' or END_OF_INSN. */
2276 if (*l == ',')
2277 {
2278 if (*++l == END_OF_INSN)
2279 {
2280 /* Just skip it, if it's \n complain. */
2281 goto expecting_operand_after_comma;
2282 }
2283 expecting_operand = 1;
2284 }
2285 }
2286 return l;
2287 }
2288
2289 static void
2290 swap_2_operands (int xchg1, int xchg2)
2291 {
2292 union i386_op temp_op;
2293 unsigned int temp_type;
2294 enum bfd_reloc_code_real temp_reloc;
2295
2296 temp_type = i.types[xchg2];
2297 i.types[xchg2] = i.types[xchg1];
2298 i.types[xchg1] = temp_type;
2299 temp_op = i.op[xchg2];
2300 i.op[xchg2] = i.op[xchg1];
2301 i.op[xchg1] = temp_op;
2302 temp_reloc = i.reloc[xchg2];
2303 i.reloc[xchg2] = i.reloc[xchg1];
2304 i.reloc[xchg1] = temp_reloc;
2305 }
2306
2307 static void
2308 swap_operands (void)
2309 {
2310 switch (i.operands)
2311 {
2312 case 4:
2313 swap_2_operands (1, i.operands - 2);
2314 case 3:
2315 case 2:
2316 swap_2_operands (0, i.operands - 1);
2317 break;
2318 default:
2319 abort ();
2320 }
2321
2322 if (i.mem_operands == 2)
2323 {
2324 const seg_entry *temp_seg;
2325 temp_seg = i.seg[0];
2326 i.seg[0] = i.seg[1];
2327 i.seg[1] = temp_seg;
2328 }
2329 }
2330
2331 /* Try to ensure constant immediates are represented in the smallest
2332 opcode possible. */
2333 static void
2334 optimize_imm (void)
2335 {
2336 char guess_suffix = 0;
2337 int op;
2338
2339 if (i.suffix)
2340 guess_suffix = i.suffix;
2341 else if (i.reg_operands)
2342 {
2343 /* Figure out a suffix from the last register operand specified.
2344 We can't do this properly yet, ie. excluding InOutPortReg,
2345 but the following works for instructions with immediates.
2346 In any case, we can't set i.suffix yet. */
2347 for (op = i.operands; --op >= 0;)
2348 if (i.types[op] & Reg)
2349 {
2350 if (i.types[op] & Reg8)
2351 guess_suffix = BYTE_MNEM_SUFFIX;
2352 else if (i.types[op] & Reg16)
2353 guess_suffix = WORD_MNEM_SUFFIX;
2354 else if (i.types[op] & Reg32)
2355 guess_suffix = LONG_MNEM_SUFFIX;
2356 else if (i.types[op] & Reg64)
2357 guess_suffix = QWORD_MNEM_SUFFIX;
2358 break;
2359 }
2360 }
2361 else if ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0))
2362 guess_suffix = WORD_MNEM_SUFFIX;
2363
2364 for (op = i.operands; --op >= 0;)
2365 if (i.types[op] & Imm)
2366 {
2367 switch (i.op[op].imms->X_op)
2368 {
2369 case O_constant:
2370 /* If a suffix is given, this operand may be shortened. */
2371 switch (guess_suffix)
2372 {
2373 case LONG_MNEM_SUFFIX:
2374 i.types[op] |= Imm32 | Imm64;
2375 break;
2376 case WORD_MNEM_SUFFIX:
2377 i.types[op] |= Imm16 | Imm32S | Imm32 | Imm64;
2378 break;
2379 case BYTE_MNEM_SUFFIX:
2380 i.types[op] |= Imm16 | Imm8 | Imm8S | Imm32S | Imm32 | Imm64;
2381 break;
2382 }
2383
2384 /* If this operand is at most 16 bits, convert it
2385 to a signed 16 bit number before trying to see
2386 whether it will fit in an even smaller size.
2387 This allows a 16-bit operand such as $0xffe0 to
2388 be recognised as within Imm8S range. */
2389 if ((i.types[op] & Imm16)
2390 && (i.op[op].imms->X_add_number & ~(offsetT) 0xffff) == 0)
2391 {
2392 i.op[op].imms->X_add_number =
2393 (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000);
2394 }
2395 if ((i.types[op] & Imm32)
2396 && ((i.op[op].imms->X_add_number & ~(((offsetT) 2 << 31) - 1))
2397 == 0))
2398 {
2399 i.op[op].imms->X_add_number = ((i.op[op].imms->X_add_number
2400 ^ ((offsetT) 1 << 31))
2401 - ((offsetT) 1 << 31));
2402 }
2403 i.types[op] |= smallest_imm_type (i.op[op].imms->X_add_number);
2404
2405 /* We must avoid matching of Imm32 templates when 64bit
2406 only immediate is available. */
2407 if (guess_suffix == QWORD_MNEM_SUFFIX)
2408 i.types[op] &= ~Imm32;
2409 break;
2410
2411 case O_absent:
2412 case O_register:
2413 abort ();
2414
2415 /* Symbols and expressions. */
2416 default:
2417 /* Convert symbolic operand to proper sizes for matching, but don't
2418 prevent matching a set of insns that only supports sizes other
2419 than those matching the insn suffix. */
2420 {
2421 unsigned int mask, allowed = 0;
2422 const template *t;
2423
2424 for (t = current_templates->start;
2425 t < current_templates->end;
2426 ++t)
2427 allowed |= t->operand_types[op];
2428 switch (guess_suffix)
2429 {
2430 case QWORD_MNEM_SUFFIX:
2431 mask = Imm64 | Imm32S;
2432 break;
2433 case LONG_MNEM_SUFFIX:
2434 mask = Imm32;
2435 break;
2436 case WORD_MNEM_SUFFIX:
2437 mask = Imm16;
2438 break;
2439 case BYTE_MNEM_SUFFIX:
2440 mask = Imm8;
2441 break;
2442 default:
2443 mask = 0;
2444 break;
2445 }
2446 if (mask & allowed)
2447 i.types[op] &= mask;
2448 }
2449 break;
2450 }
2451 }
2452 }
2453
2454 /* Try to use the smallest displacement type too. */
2455 static void
2456 optimize_disp (void)
2457 {
2458 int op;
2459
2460 for (op = i.operands; --op >= 0;)
2461 if (i.types[op] & Disp)
2462 {
2463 if (i.op[op].disps->X_op == O_constant)
2464 {
2465 offsetT disp = i.op[op].disps->X_add_number;
2466
2467 if ((i.types[op] & Disp16)
2468 && (disp & ~(offsetT) 0xffff) == 0)
2469 {
2470 /* If this operand is at most 16 bits, convert
2471 to a signed 16 bit number and don't use 64bit
2472 displacement. */
2473 disp = (((disp & 0xffff) ^ 0x8000) - 0x8000);
2474 i.types[op] &= ~Disp64;
2475 }
2476 if ((i.types[op] & Disp32)
2477 && (disp & ~(((offsetT) 2 << 31) - 1)) == 0)
2478 {
2479 /* If this operand is at most 32 bits, convert
2480 to a signed 32 bit number and don't use 64bit
2481 displacement. */
2482 disp &= (((offsetT) 2 << 31) - 1);
2483 disp = (disp ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
2484 i.types[op] &= ~Disp64;
2485 }
2486 if (!disp && (i.types[op] & BaseIndex))
2487 {
2488 i.types[op] &= ~Disp;
2489 i.op[op].disps = 0;
2490 i.disp_operands--;
2491 }
2492 else if (flag_code == CODE_64BIT)
2493 {
2494 if (fits_in_signed_long (disp))
2495 {
2496 i.types[op] &= ~Disp64;
2497 i.types[op] |= Disp32S;
2498 }
2499 if (fits_in_unsigned_long (disp))
2500 i.types[op] |= Disp32;
2501 }
2502 if ((i.types[op] & (Disp32 | Disp32S | Disp16))
2503 && fits_in_signed_byte (disp))
2504 i.types[op] |= Disp8;
2505 }
2506 else if (i.reloc[op] == BFD_RELOC_386_TLS_DESC_CALL
2507 || i.reloc[op] == BFD_RELOC_X86_64_TLSDESC_CALL)
2508 {
2509 fix_new_exp (frag_now, frag_more (0) - frag_now->fr_literal, 0,
2510 i.op[op].disps, 0, i.reloc[op]);
2511 i.types[op] &= ~Disp;
2512 }
2513 else
2514 /* We only support 64bit displacement on constants. */
2515 i.types[op] &= ~Disp64;
2516 }
2517 }
2518
2519 static int
2520 match_template (void)
2521 {
2522 /* Points to template once we've found it. */
2523 const template *t;
2524 unsigned int overlap0, overlap1, overlap2, overlap3;
2525 unsigned int found_reverse_match;
2526 int suffix_check;
2527 unsigned int operand_types [MAX_OPERANDS];
2528 int addr_prefix_disp;
2529 unsigned int j;
2530
2531 #if MAX_OPERANDS != 4
2532 # error "MAX_OPERANDS must be 4."
2533 #endif
2534
2535 #define MATCH(overlap, given, template) \
2536 ((overlap & ~JumpAbsolute) \
2537 && (((given) & (BaseIndex | JumpAbsolute)) \
2538 == ((overlap) & (BaseIndex | JumpAbsolute))))
2539
2540 /* If given types r0 and r1 are registers they must be of the same type
2541 unless the expected operand type register overlap is null.
2542 Note that Acc in a template matches every size of reg. */
2543 #define CONSISTENT_REGISTER_MATCH(m0, g0, t0, m1, g1, t1) \
2544 (((g0) & Reg) == 0 || ((g1) & Reg) == 0 \
2545 || ((g0) & Reg) == ((g1) & Reg) \
2546 || ((((m0) & Acc) ? Reg : (t0)) & (((m1) & Acc) ? Reg : (t1)) & Reg) == 0 )
2547
2548 overlap0 = 0;
2549 overlap1 = 0;
2550 overlap2 = 0;
2551 overlap3 = 0;
2552 found_reverse_match = 0;
2553 for (j = 0; j < MAX_OPERANDS; j++)
2554 operand_types [j] = 0;
2555 addr_prefix_disp = -1;
2556 suffix_check = (i.suffix == BYTE_MNEM_SUFFIX
2557 ? No_bSuf
2558 : (i.suffix == WORD_MNEM_SUFFIX
2559 ? No_wSuf
2560 : (i.suffix == SHORT_MNEM_SUFFIX
2561 ? No_sSuf
2562 : (i.suffix == LONG_MNEM_SUFFIX
2563 ? No_lSuf
2564 : (i.suffix == QWORD_MNEM_SUFFIX
2565 ? No_qSuf
2566 : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX
2567 ? No_xSuf : 0))))));
2568
2569 for (t = current_templates->start; t < current_templates->end; t++)
2570 {
2571 addr_prefix_disp = -1;
2572
2573 /* Must have right number of operands. */
2574 if (i.operands != t->operands)
2575 continue;
2576
2577 /* Check the suffix, except for some instructions in intel mode. */
2578 if ((t->opcode_modifier & suffix_check)
2579 && !(intel_syntax
2580 && (t->opcode_modifier & IgnoreSize)))
2581 continue;
2582
2583 for (j = 0; j < MAX_OPERANDS; j++)
2584 operand_types [j] = t->operand_types [j];
2585
2586 /* In general, don't allow 64-bit operands in 32-bit mode. */
2587 if (i.suffix == QWORD_MNEM_SUFFIX
2588 && flag_code != CODE_64BIT
2589 && (intel_syntax
2590 ? (!(t->opcode_modifier & IgnoreSize)
2591 && !intel_float_operand (t->name))
2592 : intel_float_operand (t->name) != 2)
2593 && (!(operand_types[0] & (RegMMX | RegXMM))
2594 || !(operand_types[t->operands > 1] & (RegMMX | RegXMM)))
2595 && (t->base_opcode != 0x0fc7
2596 || t->extension_opcode != 1 /* cmpxchg8b */))
2597 continue;
2598
2599 /* Do not verify operands when there are none. */
2600 else if (!t->operands)
2601 {
2602 if (t->cpu_flags & ~cpu_arch_flags)
2603 continue;
2604 /* We've found a match; break out of loop. */
2605 break;
2606 }
2607
2608 /* Address size prefix will turn Disp64/Disp32/Disp16 operand
2609 into Disp32/Disp16/Disp32 operand. */
2610 if (i.prefix[ADDR_PREFIX] != 0)
2611 {
2612 unsigned int DispOn = 0, DispOff = 0;
2613
2614 switch (flag_code)
2615 {
2616 case CODE_16BIT:
2617 DispOn = Disp32;
2618 DispOff = Disp16;
2619 break;
2620 case CODE_32BIT:
2621 DispOn = Disp16;
2622 DispOff = Disp32;
2623 break;
2624 case CODE_64BIT:
2625 DispOn = Disp32;
2626 DispOff = Disp64;
2627 break;
2628 }
2629
2630 for (j = 0; j < MAX_OPERANDS; j++)
2631 {
2632 /* There should be only one Disp operand. */
2633 if ((operand_types[j] & DispOff))
2634 {
2635 addr_prefix_disp = j;
2636 operand_types[j] |= DispOn;
2637 operand_types[j] &= ~DispOff;
2638 break;
2639 }
2640 }
2641 }
2642
2643 overlap0 = i.types[0] & operand_types[0];
2644 switch (t->operands)
2645 {
2646 case 1:
2647 if (!MATCH (overlap0, i.types[0], operand_types[0]))
2648 continue;
2649 break;
2650 case 2:
2651 /* xchg %eax, %eax is a special case. It is an aliase for nop
2652 only in 32bit mode and we can use opcode 0x90. In 64bit
2653 mode, we can't use 0x90 for xchg %eax, %eax since it should
2654 zero-extend %eax to %rax. */
2655 if (flag_code == CODE_64BIT
2656 && t->base_opcode == 0x90
2657 && i.types [0] == (Acc | Reg32)
2658 && i.types [1] == (Acc | Reg32))
2659 continue;
2660 case 3:
2661 case 4:
2662 overlap1 = i.types[1] & operand_types[1];
2663 if (!MATCH (overlap0, i.types[0], operand_types[0])
2664 || !MATCH (overlap1, i.types[1], operand_types[1])
2665 /* monitor in SSE3 is a very special case. The first
2666 register and the second register may have different
2667 sizes. The same applies to crc32 in SSE4.2. */
2668 || !((t->base_opcode == 0x0f01
2669 && t->extension_opcode == 0xc8)
2670 || t->base_opcode == 0xf20f38f1
2671 || CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2672 operand_types[0],
2673 overlap1, i.types[1],
2674 operand_types[1])))
2675 {
2676 /* Check if other direction is valid ... */
2677 if ((t->opcode_modifier & (D | FloatD)) == 0)
2678 continue;
2679
2680 /* Try reversing direction of operands. */
2681 overlap0 = i.types[0] & operand_types[1];
2682 overlap1 = i.types[1] & operand_types[0];
2683 if (!MATCH (overlap0, i.types[0], operand_types[1])
2684 || !MATCH (overlap1, i.types[1], operand_types[0])
2685 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2686 operand_types[1],
2687 overlap1, i.types[1],
2688 operand_types[0]))
2689 {
2690 /* Does not match either direction. */
2691 continue;
2692 }
2693 /* found_reverse_match holds which of D or FloatDR
2694 we've found. */
2695 if ((t->opcode_modifier & D))
2696 found_reverse_match = Opcode_D;
2697 else if ((t->opcode_modifier & FloatD))
2698 found_reverse_match = Opcode_FloatD;
2699 else
2700 found_reverse_match = 0;
2701 if ((t->opcode_modifier & FloatR))
2702 found_reverse_match |= Opcode_FloatR;
2703 }
2704 else
2705 {
2706 /* Found a forward 2 operand match here. */
2707 switch (t->operands)
2708 {
2709 case 4:
2710 overlap3 = i.types[3] & operand_types[3];
2711 case 3:
2712 overlap2 = i.types[2] & operand_types[2];
2713 break;
2714 }
2715
2716 switch (t->operands)
2717 {
2718 case 4:
2719 if (!MATCH (overlap3, i.types[3], operand_types[3])
2720 || !CONSISTENT_REGISTER_MATCH (overlap2,
2721 i.types[2],
2722 operand_types[2],
2723 overlap3,
2724 i.types[3],
2725 operand_types[3]))
2726 continue;
2727 case 3:
2728 /* Here we make use of the fact that there are no
2729 reverse match 3 operand instructions, and all 3
2730 operand instructions only need to be checked for
2731 register consistency between operands 2 and 3. */
2732 if (!MATCH (overlap2, i.types[2], operand_types[2])
2733 || !CONSISTENT_REGISTER_MATCH (overlap1,
2734 i.types[1],
2735 operand_types[1],
2736 overlap2,
2737 i.types[2],
2738 operand_types[2]))
2739 continue;
2740 break;
2741 }
2742 }
2743 /* Found either forward/reverse 2, 3 or 4 operand match here:
2744 slip through to break. */
2745 }
2746 if (t->cpu_flags & ~cpu_arch_flags)
2747 {
2748 found_reverse_match = 0;
2749 continue;
2750 }
2751 /* We've found a match; break out of loop. */
2752 break;
2753 }
2754
2755 if (t == current_templates->end)
2756 {
2757 /* We found no match. */
2758 as_bad (_("suffix or operands invalid for `%s'"),
2759 current_templates->start->name);
2760 return 0;
2761 }
2762
2763 if (!quiet_warnings)
2764 {
2765 if (!intel_syntax
2766 && ((i.types[0] & JumpAbsolute)
2767 != (operand_types[0] & JumpAbsolute)))
2768 {
2769 as_warn (_("indirect %s without `*'"), t->name);
2770 }
2771
2772 if ((t->opcode_modifier & (IsPrefix | IgnoreSize))
2773 == (IsPrefix | IgnoreSize))
2774 {
2775 /* Warn them that a data or address size prefix doesn't
2776 affect assembly of the next line of code. */
2777 as_warn (_("stand-alone `%s' prefix"), t->name);
2778 }
2779 }
2780
2781 /* Copy the template we found. */
2782 i.tm = *t;
2783
2784 if (addr_prefix_disp != -1)
2785 i.tm.operand_types[addr_prefix_disp]
2786 = operand_types[addr_prefix_disp];
2787
2788 if (found_reverse_match)
2789 {
2790 /* If we found a reverse match we must alter the opcode
2791 direction bit. found_reverse_match holds bits to change
2792 (different for int & float insns). */
2793
2794 i.tm.base_opcode ^= found_reverse_match;
2795
2796 i.tm.operand_types[0] = operand_types[1];
2797 i.tm.operand_types[1] = operand_types[0];
2798 }
2799
2800 return 1;
2801 }
2802
2803 static int
2804 check_string (void)
2805 {
2806 int mem_op = (i.types[0] & AnyMem) ? 0 : 1;
2807 if ((i.tm.operand_types[mem_op] & EsSeg) != 0)
2808 {
2809 if (i.seg[0] != NULL && i.seg[0] != &es)
2810 {
2811 as_bad (_("`%s' operand %d must use `%%es' segment"),
2812 i.tm.name,
2813 mem_op + 1);
2814 return 0;
2815 }
2816 /* There's only ever one segment override allowed per instruction.
2817 This instruction possibly has a legal segment override on the
2818 second operand, so copy the segment to where non-string
2819 instructions store it, allowing common code. */
2820 i.seg[0] = i.seg[1];
2821 }
2822 else if ((i.tm.operand_types[mem_op + 1] & EsSeg) != 0)
2823 {
2824 if (i.seg[1] != NULL && i.seg[1] != &es)
2825 {
2826 as_bad (_("`%s' operand %d must use `%%es' segment"),
2827 i.tm.name,
2828 mem_op + 2);
2829 return 0;
2830 }
2831 }
2832 return 1;
2833 }
2834
2835 static int
2836 process_suffix (void)
2837 {
2838 /* If matched instruction specifies an explicit instruction mnemonic
2839 suffix, use it. */
2840 if (i.tm.opcode_modifier & (Size16 | Size32 | Size64))
2841 {
2842 if (i.tm.opcode_modifier & Size16)
2843 i.suffix = WORD_MNEM_SUFFIX;
2844 else if (i.tm.opcode_modifier & Size64)
2845 i.suffix = QWORD_MNEM_SUFFIX;
2846 else
2847 i.suffix = LONG_MNEM_SUFFIX;
2848 }
2849 else if (i.reg_operands)
2850 {
2851 /* If there's no instruction mnemonic suffix we try to invent one
2852 based on register operands. */
2853 if (!i.suffix)
2854 {
2855 /* We take i.suffix from the last register operand specified,
2856 Destination register type is more significant than source
2857 register type. crc32 in SSE4.2 prefers source register
2858 type. */
2859 if (i.tm.base_opcode == 0xf20f38f1)
2860 {
2861 if ((i.types[0] & Reg))
2862 i.suffix = ((i.types[0] & Reg16) ? WORD_MNEM_SUFFIX :
2863 LONG_MNEM_SUFFIX);
2864 }
2865 else if (i.tm.base_opcode == 0xf20f38f0)
2866 {
2867 if ((i.types[0] & Reg8))
2868 i.suffix = BYTE_MNEM_SUFFIX;
2869 }
2870
2871 if (!i.suffix)
2872 {
2873 int op;
2874
2875 if (i.tm.base_opcode == 0xf20f38f1
2876 || i.tm.base_opcode == 0xf20f38f0)
2877 {
2878 /* We have to know the operand size for crc32. */
2879 as_bad (_("ambiguous memory operand size for `%s`"),
2880 i.tm.name);
2881 return 0;
2882 }
2883
2884 for (op = i.operands; --op >= 0;)
2885 if ((i.types[op] & Reg)
2886 && !(i.tm.operand_types[op] & InOutPortReg))
2887 {
2888 i.suffix = ((i.types[op] & Reg8) ? BYTE_MNEM_SUFFIX :
2889 (i.types[op] & Reg16) ? WORD_MNEM_SUFFIX :
2890 (i.types[op] & Reg64) ? QWORD_MNEM_SUFFIX :
2891 LONG_MNEM_SUFFIX);
2892 break;
2893 }
2894 }
2895 }
2896 else if (i.suffix == BYTE_MNEM_SUFFIX)
2897 {
2898 if (!check_byte_reg ())
2899 return 0;
2900 }
2901 else if (i.suffix == LONG_MNEM_SUFFIX)
2902 {
2903 if (!check_long_reg ())
2904 return 0;
2905 }
2906 else if (i.suffix == QWORD_MNEM_SUFFIX)
2907 {
2908 if (!check_qword_reg ())
2909 return 0;
2910 }
2911 else if (i.suffix == WORD_MNEM_SUFFIX)
2912 {
2913 if (!check_word_reg ())
2914 return 0;
2915 }
2916 else if (intel_syntax && (i.tm.opcode_modifier & IgnoreSize))
2917 /* Do nothing if the instruction is going to ignore the prefix. */
2918 ;
2919 else
2920 abort ();
2921 }
2922 else if ((i.tm.opcode_modifier & DefaultSize)
2923 && !i.suffix
2924 /* exclude fldenv/frstor/fsave/fstenv */
2925 && (i.tm.opcode_modifier & No_sSuf))
2926 {
2927 i.suffix = stackop_size;
2928 }
2929 else if (intel_syntax
2930 && !i.suffix
2931 && ((i.tm.operand_types[0] & JumpAbsolute)
2932 || (i.tm.opcode_modifier & (JumpByte|JumpInterSegment))
2933 || (i.tm.base_opcode == 0x0f01 /* [ls][gi]dt */
2934 && i.tm.extension_opcode <= 3)))
2935 {
2936 switch (flag_code)
2937 {
2938 case CODE_64BIT:
2939 if (!(i.tm.opcode_modifier & No_qSuf))
2940 {
2941 i.suffix = QWORD_MNEM_SUFFIX;
2942 break;
2943 }
2944 case CODE_32BIT:
2945 if (!(i.tm.opcode_modifier & No_lSuf))
2946 i.suffix = LONG_MNEM_SUFFIX;
2947 break;
2948 case CODE_16BIT:
2949 if (!(i.tm.opcode_modifier & No_wSuf))
2950 i.suffix = WORD_MNEM_SUFFIX;
2951 break;
2952 }
2953 }
2954
2955 if (!i.suffix)
2956 {
2957 if (!intel_syntax)
2958 {
2959 if (i.tm.opcode_modifier & W)
2960 {
2961 as_bad (_("no instruction mnemonic suffix given and "
2962 "no register operands; can't size instruction"));
2963 return 0;
2964 }
2965 }
2966 else
2967 {
2968 unsigned int suffixes = (~i.tm.opcode_modifier
2969 & (No_bSuf
2970 | No_wSuf
2971 | No_lSuf
2972 | No_sSuf
2973 | No_xSuf
2974 | No_qSuf));
2975
2976 if ((i.tm.opcode_modifier & W)
2977 || ((suffixes & (suffixes - 1))
2978 && !(i.tm.opcode_modifier & (DefaultSize | IgnoreSize))))
2979 {
2980 as_bad (_("ambiguous operand size for `%s'"), i.tm.name);
2981 return 0;
2982 }
2983 }
2984 }
2985
2986 /* Change the opcode based on the operand size given by i.suffix;
2987 We don't need to change things for byte insns. */
2988
2989 if (i.suffix && i.suffix != BYTE_MNEM_SUFFIX)
2990 {
2991 /* It's not a byte, select word/dword operation. */
2992 if (i.tm.opcode_modifier & W)
2993 {
2994 if (i.tm.opcode_modifier & ShortForm)
2995 i.tm.base_opcode |= 8;
2996 else
2997 i.tm.base_opcode |= 1;
2998 }
2999
3000 /* Now select between word & dword operations via the operand
3001 size prefix, except for instructions that will ignore this
3002 prefix anyway. */
3003 if (i.tm.base_opcode == 0x0f01 && i.tm.extension_opcode == 0xc8)
3004 {
3005 /* monitor in SSE3 is a very special case. The default size
3006 of AX is the size of mode. The address size override
3007 prefix will change the size of AX. */
3008 if (i.op->regs[0].reg_type &
3009 (flag_code == CODE_32BIT ? Reg16 : Reg32))
3010 if (!add_prefix (ADDR_PREFIX_OPCODE))
3011 return 0;
3012 }
3013 else if (i.suffix != QWORD_MNEM_SUFFIX
3014 && i.suffix != LONG_DOUBLE_MNEM_SUFFIX
3015 && !(i.tm.opcode_modifier & (IgnoreSize | FloatMF))
3016 && ((i.suffix == LONG_MNEM_SUFFIX) == (flag_code == CODE_16BIT)
3017 || (flag_code == CODE_64BIT
3018 && (i.tm.opcode_modifier & JumpByte))))
3019 {
3020 unsigned int prefix = DATA_PREFIX_OPCODE;
3021
3022 if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */
3023 prefix = ADDR_PREFIX_OPCODE;
3024
3025 if (!add_prefix (prefix))
3026 return 0;
3027 }
3028
3029 /* Set mode64 for an operand. */
3030 if (i.suffix == QWORD_MNEM_SUFFIX
3031 && flag_code == CODE_64BIT
3032 && (i.tm.opcode_modifier & NoRex64) == 0)
3033 {
3034 /* Special case for xchg %rax,%rax. It is NOP and doesn't
3035 need rex64. cmpxchg8b is also a special case. */
3036 if (! (i.operands == 2
3037 && i.tm.base_opcode == 0x90
3038 && i.tm.extension_opcode == None
3039 && i.types [0] == (Acc | Reg64)
3040 && i.types [1] == (Acc | Reg64))
3041 && ! (i.operands == 1
3042 && i.tm.base_opcode == 0xfc7
3043 && i.tm.extension_opcode == 1
3044 && (i.types [0] & Reg) == 0
3045 && (i.types [0] & AnyMem) != 0))
3046 i.rex |= REX_W;
3047 }
3048
3049 /* Size floating point instruction. */
3050 if (i.suffix == LONG_MNEM_SUFFIX)
3051 if (i.tm.opcode_modifier & FloatMF)
3052 i.tm.base_opcode ^= 4;
3053 }
3054
3055 return 1;
3056 }
3057
3058 static int
3059 check_byte_reg (void)
3060 {
3061 int op;
3062
3063 for (op = i.operands; --op >= 0;)
3064 {
3065 /* If this is an eight bit register, it's OK. If it's the 16 or
3066 32 bit version of an eight bit register, we will just use the
3067 low portion, and that's OK too. */
3068 if (i.types[op] & Reg8)
3069 continue;
3070
3071 /* movzx, movsx, pextrb and pinsrb should not generate this
3072 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 || i.tm.base_opcode == 0x660f3a14
3080 || i.tm.base_opcode == 0x660f3a20))
3081 continue;
3082
3083 /* crc32 doesn't generate this warning. */
3084 if (i.tm.base_opcode == 0xf20f38f0)
3085 continue;
3086
3087 if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4)
3088 {
3089 /* Prohibit these changes in the 64bit mode, since the
3090 lowering is more complicated. */
3091 if (flag_code == CODE_64BIT
3092 && (i.tm.operand_types[op] & InOutPortReg) == 0)
3093 {
3094 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3095 register_prefix, i.op[op].regs->reg_name,
3096 i.suffix);
3097 return 0;
3098 }
3099 #if REGISTER_WARNINGS
3100 if (!quiet_warnings
3101 && (i.tm.operand_types[op] & InOutPortReg) == 0)
3102 as_warn (_("using `%s%s' instead of `%s%s' due to `%c' suffix"),
3103 register_prefix,
3104 (i.op[op].regs + (i.types[op] & Reg16
3105 ? REGNAM_AL - REGNAM_AX
3106 : REGNAM_AL - REGNAM_EAX))->reg_name,
3107 register_prefix,
3108 i.op[op].regs->reg_name,
3109 i.suffix);
3110 #endif
3111 continue;
3112 }
3113 /* Any other register is bad. */
3114 if (i.types[op] & (Reg | RegMMX | RegXMM
3115 | SReg2 | SReg3
3116 | Control | Debug | Test
3117 | FloatReg | FloatAcc))
3118 {
3119 as_bad (_("`%s%s' not allowed with `%s%c'"),
3120 register_prefix,
3121 i.op[op].regs->reg_name,
3122 i.tm.name,
3123 i.suffix);
3124 return 0;
3125 }
3126 }
3127 return 1;
3128 }
3129
3130 static int
3131 check_long_reg (void)
3132 {
3133 int op;
3134
3135 for (op = i.operands; --op >= 0;)
3136 /* Reject eight bit registers, except where the template requires
3137 them. (eg. movzb) */
3138 if ((i.types[op] & Reg8) != 0
3139 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
3140 {
3141 as_bad (_("`%s%s' not allowed with `%s%c'"),
3142 register_prefix,
3143 i.op[op].regs->reg_name,
3144 i.tm.name,
3145 i.suffix);
3146 return 0;
3147 }
3148 /* Warn if the e prefix on a general reg is missing. */
3149 else if ((!quiet_warnings || flag_code == CODE_64BIT)
3150 && (i.types[op] & Reg16) != 0
3151 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
3152 {
3153 /* Prohibit these changes in the 64bit mode, since the
3154 lowering is more complicated. */
3155 if (flag_code == CODE_64BIT)
3156 {
3157 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3158 register_prefix, i.op[op].regs->reg_name,
3159 i.suffix);
3160 return 0;
3161 }
3162 #if REGISTER_WARNINGS
3163 else
3164 as_warn (_("using `%s%s' instead of `%s%s' due to `%c' suffix"),
3165 register_prefix,
3166 (i.op[op].regs + REGNAM_EAX - REGNAM_AX)->reg_name,
3167 register_prefix,
3168 i.op[op].regs->reg_name,
3169 i.suffix);
3170 #endif
3171 }
3172 /* Warn if the r prefix on a general reg is missing. */
3173 else if ((i.types[op] & Reg64) != 0
3174 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
3175 {
3176 if (intel_syntax
3177 && i.tm.base_opcode == 0xf30f2d
3178 && (i.types[0] & RegXMM) == 0)
3179 {
3180 /* cvtss2si converts DWORD memory to Reg64. We want
3181 REX byte. */
3182 i.suffix = QWORD_MNEM_SUFFIX;
3183 }
3184 else
3185 {
3186 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3187 register_prefix, i.op[op].regs->reg_name,
3188 i.suffix);
3189 return 0;
3190 }
3191 }
3192 return 1;
3193 }
3194
3195 static int
3196 check_qword_reg (void)
3197 {
3198 int op;
3199
3200 for (op = i.operands; --op >= 0; )
3201 /* Reject eight bit registers, except where the template requires
3202 them. (eg. movzb) */
3203 if ((i.types[op] & Reg8) != 0
3204 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
3205 {
3206 as_bad (_("`%s%s' not allowed with `%s%c'"),
3207 register_prefix,
3208 i.op[op].regs->reg_name,
3209 i.tm.name,
3210 i.suffix);
3211 return 0;
3212 }
3213 /* Warn if the e prefix on a general reg is missing. */
3214 else if ((i.types[op] & (Reg16 | Reg32)) != 0
3215 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
3216 {
3217 /* Prohibit these changes in the 64bit mode, since the
3218 lowering is more complicated. */
3219 if (intel_syntax
3220 && i.tm.base_opcode == 0xf20f2d
3221 && (i.types[0] & RegXMM) == 0)
3222 {
3223 /* cvtsd2si converts QWORD memory to Reg32. We don't want
3224 REX byte. */
3225 i.suffix = LONG_MNEM_SUFFIX;
3226 }
3227 else
3228 {
3229 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3230 register_prefix, i.op[op].regs->reg_name,
3231 i.suffix);
3232 return 0;
3233 }
3234 }
3235 return 1;
3236 }
3237
3238 static int
3239 check_word_reg (void)
3240 {
3241 int op;
3242 for (op = i.operands; --op >= 0;)
3243 /* Reject eight bit registers, except where the template requires
3244 them. (eg. movzb) */
3245 if ((i.types[op] & Reg8) != 0
3246 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
3247 {
3248 as_bad (_("`%s%s' not allowed with `%s%c'"),
3249 register_prefix,
3250 i.op[op].regs->reg_name,
3251 i.tm.name,
3252 i.suffix);
3253 return 0;
3254 }
3255 /* Warn if the e prefix on a general reg is present. */
3256 else if ((!quiet_warnings || flag_code == CODE_64BIT)
3257 && (i.types[op] & Reg32) != 0
3258 && (i.tm.operand_types[op] & (Reg16 | Acc)) != 0)
3259 {
3260 /* Prohibit these changes in the 64bit mode, since the
3261 lowering is more complicated. */
3262 if (flag_code == CODE_64BIT)
3263 {
3264 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3265 register_prefix, i.op[op].regs->reg_name,
3266 i.suffix);
3267 return 0;
3268 }
3269 else
3270 #if REGISTER_WARNINGS
3271 as_warn (_("using `%s%s' instead of `%s%s' due to `%c' suffix"),
3272 register_prefix,
3273 (i.op[op].regs + REGNAM_AX - REGNAM_EAX)->reg_name,
3274 register_prefix,
3275 i.op[op].regs->reg_name,
3276 i.suffix);
3277 #endif
3278 }
3279 return 1;
3280 }
3281
3282 static int
3283 finalize_imm (void)
3284 {
3285 unsigned int overlap0, overlap1, overlap2;
3286
3287 overlap0 = i.types[0] & i.tm.operand_types[0];
3288 if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64))
3289 && overlap0 != Imm8 && overlap0 != Imm8S
3290 && overlap0 != Imm16 && overlap0 != Imm32S
3291 && overlap0 != Imm32 && overlap0 != Imm64)
3292 {
3293 if (i.suffix)
3294 {
3295 overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX
3296 ? Imm8 | Imm8S
3297 : (i.suffix == WORD_MNEM_SUFFIX
3298 ? Imm16
3299 : (i.suffix == QWORD_MNEM_SUFFIX
3300 ? Imm64 | Imm32S
3301 : Imm32)));
3302 }
3303 else if (overlap0 == (Imm16 | Imm32S | Imm32)
3304 || overlap0 == (Imm16 | Imm32)
3305 || overlap0 == (Imm16 | Imm32S))
3306 {
3307 overlap0 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
3308 ? Imm16 : Imm32S);
3309 }
3310 if (overlap0 != Imm8 && overlap0 != Imm8S
3311 && overlap0 != Imm16 && overlap0 != Imm32S
3312 && overlap0 != Imm32 && overlap0 != Imm64)
3313 {
3314 as_bad (_("no instruction mnemonic suffix given; "
3315 "can't determine immediate size"));
3316 return 0;
3317 }
3318 }
3319 i.types[0] = overlap0;
3320
3321 overlap1 = i.types[1] & i.tm.operand_types[1];
3322 if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32 | Imm64))
3323 && overlap1 != Imm8 && overlap1 != Imm8S
3324 && overlap1 != Imm16 && overlap1 != Imm32S
3325 && overlap1 != Imm32 && overlap1 != Imm64)
3326 {
3327 if (i.suffix)
3328 {
3329 overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX
3330 ? Imm8 | Imm8S
3331 : (i.suffix == WORD_MNEM_SUFFIX
3332 ? Imm16
3333 : (i.suffix == QWORD_MNEM_SUFFIX
3334 ? Imm64 | Imm32S
3335 : Imm32)));
3336 }
3337 else if (overlap1 == (Imm16 | Imm32 | Imm32S)
3338 || overlap1 == (Imm16 | Imm32)
3339 || overlap1 == (Imm16 | Imm32S))
3340 {
3341 overlap1 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
3342 ? Imm16 : Imm32S);
3343 }
3344 if (overlap1 != Imm8 && overlap1 != Imm8S
3345 && overlap1 != Imm16 && overlap1 != Imm32S
3346 && overlap1 != Imm32 && overlap1 != Imm64)
3347 {
3348 as_bad (_("no instruction mnemonic suffix given; "
3349 "can't determine immediate size %x %c"),
3350 overlap1, i.suffix);
3351 return 0;
3352 }
3353 }
3354 i.types[1] = overlap1;
3355
3356 overlap2 = i.types[2] & i.tm.operand_types[2];
3357 assert ((overlap2 & Imm) == 0);
3358 i.types[2] = overlap2;
3359
3360 return 1;
3361 }
3362
3363 static int
3364 process_operands (void)
3365 {
3366 /* Default segment register this instruction will use for memory
3367 accesses. 0 means unknown. This is only for optimizing out
3368 unnecessary segment overrides. */
3369 const seg_entry *default_seg = 0;
3370
3371 /* The imul $imm, %reg instruction is converted into
3372 imul $imm, %reg, %reg, and the clr %reg instruction
3373 is converted into xor %reg, %reg. */
3374 if (i.tm.opcode_modifier & RegKludge)
3375 {
3376 if ((i.tm.cpu_flags & CpuSSE4_1))
3377 {
3378 /* The first operand in instruction blendvpd, blendvps and
3379 pblendvb in SSE4.1 is implicit and must be xmm0. */
3380 assert (i.operands == 3
3381 && i.reg_operands >= 2
3382 && i.types[0] == RegXMM);
3383 if (i.op[0].regs->reg_num != 0)
3384 {
3385 if (intel_syntax)
3386 as_bad (_("the last operand of `%s' must be `%sxmm0'"),
3387 i.tm.name, register_prefix);
3388 else
3389 as_bad (_("the first operand of `%s' must be `%sxmm0'"),
3390 i.tm.name, register_prefix);
3391 return 0;
3392 }
3393 i.op[0] = i.op[1];
3394 i.op[1] = i.op[2];
3395 i.types[0] = i.types[1];
3396 i.types[1] = i.types[2];
3397 i.operands--;
3398 i.reg_operands--;
3399
3400 /* We need to adjust fields in i.tm since they are used by
3401 build_modrm_byte. */
3402 i.tm.operand_types [0] = i.tm.operand_types [1];
3403 i.tm.operand_types [1] = i.tm.operand_types [2];
3404 i.tm.operands--;
3405 }
3406 else
3407 {
3408 unsigned int first_reg_op = (i.types[0] & Reg) ? 0 : 1;
3409 /* Pretend we saw the extra register operand. */
3410 assert (i.reg_operands == 1
3411 && i.op[first_reg_op + 1].regs == 0);
3412 i.op[first_reg_op + 1].regs = i.op[first_reg_op].regs;
3413 i.types[first_reg_op + 1] = i.types[first_reg_op];
3414 i.operands++;
3415 i.reg_operands++;
3416 }
3417 }
3418
3419 if (i.tm.opcode_modifier & ShortForm)
3420 {
3421 if (i.types[0] & (SReg2 | SReg3))
3422 {
3423 if (i.tm.base_opcode == POP_SEG_SHORT
3424 && i.op[0].regs->reg_num == 1)
3425 {
3426 as_bad (_("you can't `pop %%cs'"));
3427 return 0;
3428 }
3429 i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
3430 if ((i.op[0].regs->reg_flags & RegRex) != 0)
3431 i.rex |= REX_B;
3432 }
3433 else
3434 {
3435 /* The register or float register operand is in operand 0 or 1. */
3436 unsigned int op = (i.types[0] & (Reg | FloatReg)) ? 0 : 1;
3437 /* Register goes in low 3 bits of opcode. */
3438 i.tm.base_opcode |= i.op[op].regs->reg_num;
3439 if ((i.op[op].regs->reg_flags & RegRex) != 0)
3440 i.rex |= REX_B;
3441 if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
3442 {
3443 /* Warn about some common errors, but press on regardless.
3444 The first case can be generated by gcc (<= 2.8.1). */
3445 if (i.operands == 2)
3446 {
3447 /* Reversed arguments on faddp, fsubp, etc. */
3448 as_warn (_("translating to `%s %s%s,%s%s'"), i.tm.name,
3449 register_prefix, i.op[1].regs->reg_name,
3450 register_prefix, i.op[0].regs->reg_name);
3451 }
3452 else
3453 {
3454 /* Extraneous `l' suffix on fp insn. */
3455 as_warn (_("translating to `%s %s%s'"), i.tm.name,
3456 register_prefix, i.op[0].regs->reg_name);
3457 }
3458 }
3459 }
3460 }
3461 else if (i.tm.opcode_modifier & Modrm)
3462 {
3463 /* The opcode is completed (modulo i.tm.extension_opcode which
3464 must be put into the modrm byte). Now, we make the modrm and
3465 index base bytes based on all the info we've collected. */
3466
3467 default_seg = build_modrm_byte ();
3468 }
3469 else if ((i.tm.base_opcode & ~0x3) == MOV_AX_DISP32)
3470 {
3471 default_seg = &ds;
3472 }
3473 else if ((i.tm.opcode_modifier & IsString) != 0)
3474 {
3475 /* For the string instructions that allow a segment override
3476 on one of their operands, the default segment is ds. */
3477 default_seg = &ds;
3478 }
3479
3480 if (i.tm.base_opcode == 0x8d /* lea */
3481 && i.seg[0]
3482 && !quiet_warnings)
3483 as_warn (_("segment override on `%s' is ineffectual"), i.tm.name);
3484
3485 /* If a segment was explicitly specified, and the specified segment
3486 is not the default, use an opcode prefix to select it. If we
3487 never figured out what the default segment is, then default_seg
3488 will be zero at this point, and the specified segment prefix will
3489 always be used. */
3490 if ((i.seg[0]) && (i.seg[0] != default_seg))
3491 {
3492 if (!add_prefix (i.seg[0]->seg_prefix))
3493 return 0;
3494 }
3495 return 1;
3496 }
3497
3498 static const seg_entry *
3499 build_modrm_byte (void)
3500 {
3501 const seg_entry *default_seg = 0;
3502
3503 /* i.reg_operands MUST be the number of real register operands;
3504 implicit registers do not count. */
3505 if (i.reg_operands == 2)
3506 {
3507 unsigned int source, dest;
3508
3509 switch (i.operands)
3510 {
3511 case 2:
3512 source = 0;
3513 break;
3514 case 3:
3515 /* When there are 3 operands, one of them may be immediate,
3516 which may be the first or the last operand. Otherwise,
3517 the first operand must be shift count register (cl). */
3518 assert (i.imm_operands == 1
3519 || (i.imm_operands == 0
3520 && (i.types[0] & ShiftCount)));
3521 source = (i.types[0] & (Imm | ShiftCount)) ? 1 : 0;
3522 break;
3523 case 4:
3524 /* When there are 4 operands, the first two must be immediate
3525 operands. The source operand will be the 3rd one. */
3526 assert (i.imm_operands == 2
3527 && (i.types[0] & Imm)
3528 && (i.types[1] & Imm));
3529 source = 2;
3530 break;
3531 default:
3532 abort ();
3533 }
3534
3535 dest = source + 1;
3536
3537 i.rm.mode = 3;
3538 /* One of the register operands will be encoded in the i.tm.reg
3539 field, the other in the combined i.tm.mode and i.tm.regmem
3540 fields. If no form of this instruction supports a memory
3541 destination operand, then we assume the source operand may
3542 sometimes be a memory operand and so we need to store the
3543 destination in the i.rm.reg field. */
3544 if ((i.tm.operand_types[dest] & (AnyMem | RegMem)) == 0)
3545 {
3546 i.rm.reg = i.op[dest].regs->reg_num;
3547 i.rm.regmem = i.op[source].regs->reg_num;
3548 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
3549 i.rex |= REX_R;
3550 if ((i.op[source].regs->reg_flags & RegRex) != 0)
3551 i.rex |= REX_B;
3552 }
3553 else
3554 {
3555 i.rm.reg = i.op[source].regs->reg_num;
3556 i.rm.regmem = i.op[dest].regs->reg_num;
3557 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
3558 i.rex |= REX_B;
3559 if ((i.op[source].regs->reg_flags & RegRex) != 0)
3560 i.rex |= REX_R;
3561 }
3562 if (flag_code != CODE_64BIT && (i.rex & (REX_R | REX_B)))
3563 {
3564 if (!((i.types[0] | i.types[1]) & Control))
3565 abort ();
3566 i.rex &= ~(REX_R | REX_B);
3567 add_prefix (LOCK_PREFIX_OPCODE);
3568 }
3569 }
3570 else
3571 { /* If it's not 2 reg operands... */
3572 if (i.mem_operands)
3573 {
3574 unsigned int fake_zero_displacement = 0;
3575 unsigned int op;
3576
3577 for (op = 0; op < i.operands; op++)
3578 if ((i.types[op] & AnyMem))
3579 break;
3580 assert (op < i.operands);
3581
3582 default_seg = &ds;
3583
3584 if (i.base_reg == 0)
3585 {
3586 i.rm.mode = 0;
3587 if (!i.disp_operands)
3588 fake_zero_displacement = 1;
3589 if (i.index_reg == 0)
3590 {
3591 /* Operand is just <disp> */
3592 if (flag_code == CODE_64BIT)
3593 {
3594 /* 64bit mode overwrites the 32bit absolute
3595 addressing by RIP relative addressing and
3596 absolute addressing is encoded by one of the
3597 redundant SIB forms. */
3598 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3599 i.sib.base = NO_BASE_REGISTER;
3600 i.sib.index = NO_INDEX_REGISTER;
3601 i.types[op] = ((i.prefix[ADDR_PREFIX] == 0)
3602 ? Disp32S : Disp32);
3603 }
3604 else if ((flag_code == CODE_16BIT)
3605 ^ (i.prefix[ADDR_PREFIX] != 0))
3606 {
3607 i.rm.regmem = NO_BASE_REGISTER_16;
3608 i.types[op] = Disp16;
3609 }
3610 else
3611 {
3612 i.rm.regmem = NO_BASE_REGISTER;
3613 i.types[op] = Disp32;
3614 }
3615 }
3616 else /* !i.base_reg && i.index_reg */
3617 {
3618 i.sib.index = i.index_reg->reg_num;
3619 i.sib.base = NO_BASE_REGISTER;
3620 i.sib.scale = i.log2_scale_factor;
3621 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3622 i.types[op] &= ~Disp;
3623 if (flag_code != CODE_64BIT)
3624 i.types[op] |= Disp32; /* Must be 32 bit */
3625 else
3626 i.types[op] |= Disp32S;
3627 if ((i.index_reg->reg_flags & RegRex) != 0)
3628 i.rex |= REX_X;
3629 }
3630 }
3631 /* RIP addressing for 64bit mode. */
3632 else if (i.base_reg->reg_type == BaseIndex)
3633 {
3634 i.rm.regmem = NO_BASE_REGISTER;
3635 i.types[op] &= ~ Disp;
3636 i.types[op] |= Disp32S;
3637 i.flags[op] |= Operand_PCrel;
3638 if (! i.disp_operands)
3639 fake_zero_displacement = 1;
3640 }
3641 else if (i.base_reg->reg_type & Reg16)
3642 {
3643 switch (i.base_reg->reg_num)
3644 {
3645 case 3: /* (%bx) */
3646 if (i.index_reg == 0)
3647 i.rm.regmem = 7;
3648 else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */
3649 i.rm.regmem = i.index_reg->reg_num - 6;
3650 break;
3651 case 5: /* (%bp) */
3652 default_seg = &ss;
3653 if (i.index_reg == 0)
3654 {
3655 i.rm.regmem = 6;
3656 if ((i.types[op] & Disp) == 0)
3657 {
3658 /* fake (%bp) into 0(%bp) */
3659 i.types[op] |= Disp8;
3660 fake_zero_displacement = 1;
3661 }
3662 }
3663 else /* (%bp,%si) -> 2, or (%bp,%di) -> 3 */
3664 i.rm.regmem = i.index_reg->reg_num - 6 + 2;
3665 break;
3666 default: /* (%si) -> 4 or (%di) -> 5 */
3667 i.rm.regmem = i.base_reg->reg_num - 6 + 4;
3668 }
3669 i.rm.mode = mode_from_disp_size (i.types[op]);
3670 }
3671 else /* i.base_reg and 32/64 bit mode */
3672 {
3673 if (flag_code == CODE_64BIT
3674 && (i.types[op] & Disp))
3675 i.types[op] = ((i.types[op] & Disp8)
3676 | (i.prefix[ADDR_PREFIX] == 0
3677 ? Disp32S : Disp32));
3678
3679 i.rm.regmem = i.base_reg->reg_num;
3680 if ((i.base_reg->reg_flags & RegRex) != 0)
3681 i.rex |= REX_B;
3682 i.sib.base = i.base_reg->reg_num;
3683 /* x86-64 ignores REX prefix bit here to avoid decoder
3684 complications. */
3685 if ((i.base_reg->reg_num & 7) == EBP_REG_NUM)
3686 {
3687 default_seg = &ss;
3688 if (i.disp_operands == 0)
3689 {
3690 fake_zero_displacement = 1;
3691 i.types[op] |= Disp8;
3692 }
3693 }
3694 else if (i.base_reg->reg_num == ESP_REG_NUM)
3695 {
3696 default_seg = &ss;
3697 }
3698 i.sib.scale = i.log2_scale_factor;
3699 if (i.index_reg == 0)
3700 {
3701 /* <disp>(%esp) becomes two byte modrm with no index
3702 register. We've already stored the code for esp
3703 in i.rm.regmem ie. ESCAPE_TO_TWO_BYTE_ADDRESSING.
3704 Any base register besides %esp will not use the
3705 extra modrm byte. */
3706 i.sib.index = NO_INDEX_REGISTER;
3707 #if !SCALE1_WHEN_NO_INDEX
3708 /* Another case where we force the second modrm byte. */
3709 if (i.log2_scale_factor)
3710 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3711 #endif
3712 }
3713 else
3714 {
3715 i.sib.index = i.index_reg->reg_num;
3716 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3717 if ((i.index_reg->reg_flags & RegRex) != 0)
3718 i.rex |= REX_X;
3719 }
3720
3721 if (i.disp_operands
3722 && (i.reloc[op] == BFD_RELOC_386_TLS_DESC_CALL
3723 || i.reloc[op] == BFD_RELOC_X86_64_TLSDESC_CALL))
3724 i.rm.mode = 0;
3725 else
3726 i.rm.mode = mode_from_disp_size (i.types[op]);
3727 }
3728
3729 if (fake_zero_displacement)
3730 {
3731 /* Fakes a zero displacement assuming that i.types[op]
3732 holds the correct displacement size. */
3733 expressionS *exp;
3734
3735 assert (i.op[op].disps == 0);
3736 exp = &disp_expressions[i.disp_operands++];
3737 i.op[op].disps = exp;
3738 exp->X_op = O_constant;
3739 exp->X_add_number = 0;
3740 exp->X_add_symbol = (symbolS *) 0;
3741 exp->X_op_symbol = (symbolS *) 0;
3742 }
3743 }
3744
3745 /* Fill in i.rm.reg or i.rm.regmem field with register operand
3746 (if any) based on i.tm.extension_opcode. Again, we must be
3747 careful to make sure that segment/control/debug/test/MMX
3748 registers are coded into the i.rm.reg field. */
3749 if (i.reg_operands)
3750 {
3751 unsigned int op;
3752
3753 for (op = 0; op < i.operands; op++)
3754 if ((i.types[op] & (Reg | RegMMX | RegXMM
3755 | SReg2 | SReg3
3756 | Control | Debug | Test)))
3757 break;
3758 assert (op < i.operands);
3759
3760 /* If there is an extension opcode to put here, the register
3761 number must be put into the regmem field. */
3762 if (i.tm.extension_opcode != None)
3763 {
3764 i.rm.regmem = i.op[op].regs->reg_num;
3765 if ((i.op[op].regs->reg_flags & RegRex) != 0)
3766 i.rex |= REX_B;
3767 }
3768 else
3769 {
3770 i.rm.reg = i.op[op].regs->reg_num;
3771 if ((i.op[op].regs->reg_flags & RegRex) != 0)
3772 i.rex |= REX_R;
3773 }
3774
3775 /* Now, if no memory operand has set i.rm.mode = 0, 1, 2 we
3776 must set it to 3 to indicate this is a register operand
3777 in the regmem field. */
3778 if (!i.mem_operands)
3779 i.rm.mode = 3;
3780 }
3781
3782 /* Fill in i.rm.reg field with extension opcode (if any). */
3783 if (i.tm.extension_opcode != None)
3784 i.rm.reg = i.tm.extension_opcode;
3785 }
3786 return default_seg;
3787 }
3788
3789 static void
3790 output_branch (void)
3791 {
3792 char *p;
3793 int code16;
3794 int prefix;
3795 relax_substateT subtype;
3796 symbolS *sym;
3797 offsetT off;
3798
3799 code16 = 0;
3800 if (flag_code == CODE_16BIT)
3801 code16 = CODE16;
3802
3803 prefix = 0;
3804 if (i.prefix[DATA_PREFIX] != 0)
3805 {
3806 prefix = 1;
3807 i.prefixes -= 1;
3808 code16 ^= CODE16;
3809 }
3810 /* Pentium4 branch hints. */
3811 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
3812 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
3813 {
3814 prefix++;
3815 i.prefixes--;
3816 }
3817 if (i.prefix[REX_PREFIX] != 0)
3818 {
3819 prefix++;
3820 i.prefixes--;
3821 }
3822
3823 if (i.prefixes != 0 && !intel_syntax)
3824 as_warn (_("skipping prefixes on this instruction"));
3825
3826 /* It's always a symbol; End frag & setup for relax.
3827 Make sure there is enough room in this frag for the largest
3828 instruction we may generate in md_convert_frag. This is 2
3829 bytes for the opcode and room for the prefix and largest
3830 displacement. */
3831 frag_grow (prefix + 2 + 4);
3832 /* Prefix and 1 opcode byte go in fr_fix. */
3833 p = frag_more (prefix + 1);
3834 if (i.prefix[DATA_PREFIX] != 0)
3835 *p++ = DATA_PREFIX_OPCODE;
3836 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE
3837 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE)
3838 *p++ = i.prefix[SEG_PREFIX];
3839 if (i.prefix[REX_PREFIX] != 0)
3840 *p++ = i.prefix[REX_PREFIX];
3841 *p = i.tm.base_opcode;
3842
3843 if ((unsigned char) *p == JUMP_PC_RELATIVE)
3844 subtype = ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL);
3845 else if ((cpu_arch_flags & Cpu386) != 0)
3846 subtype = ENCODE_RELAX_STATE (COND_JUMP, SMALL);
3847 else
3848 subtype = ENCODE_RELAX_STATE (COND_JUMP86, SMALL);
3849 subtype |= code16;
3850
3851 sym = i.op[0].disps->X_add_symbol;
3852 off = i.op[0].disps->X_add_number;
3853
3854 if (i.op[0].disps->X_op != O_constant
3855 && i.op[0].disps->X_op != O_symbol)
3856 {
3857 /* Handle complex expressions. */
3858 sym = make_expr_symbol (i.op[0].disps);
3859 off = 0;
3860 }
3861
3862 /* 1 possible extra opcode + 4 byte displacement go in var part.
3863 Pass reloc in fr_var. */
3864 frag_var (rs_machine_dependent, 5, i.reloc[0], subtype, sym, off, p);
3865 }
3866
3867 static void
3868 output_jump (void)
3869 {
3870 char *p;
3871 int size;
3872 fixS *fixP;
3873
3874 if (i.tm.opcode_modifier & JumpByte)
3875 {
3876 /* This is a loop or jecxz type instruction. */
3877 size = 1;
3878 if (i.prefix[ADDR_PREFIX] != 0)
3879 {
3880 FRAG_APPEND_1_CHAR (ADDR_PREFIX_OPCODE);
3881 i.prefixes -= 1;
3882 }
3883 /* Pentium4 branch hints. */
3884 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
3885 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
3886 {
3887 FRAG_APPEND_1_CHAR (i.prefix[SEG_PREFIX]);
3888 i.prefixes--;
3889 }
3890 }
3891 else
3892 {
3893 int code16;
3894
3895 code16 = 0;
3896 if (flag_code == CODE_16BIT)
3897 code16 = CODE16;
3898
3899 if (i.prefix[DATA_PREFIX] != 0)
3900 {
3901 FRAG_APPEND_1_CHAR (DATA_PREFIX_OPCODE);
3902 i.prefixes -= 1;
3903 code16 ^= CODE16;
3904 }
3905
3906 size = 4;
3907 if (code16)
3908 size = 2;
3909 }
3910
3911 if (i.prefix[REX_PREFIX] != 0)
3912 {
3913 FRAG_APPEND_1_CHAR (i.prefix[REX_PREFIX]);
3914 i.prefixes -= 1;
3915 }
3916
3917 if (i.prefixes != 0 && !intel_syntax)
3918 as_warn (_("skipping prefixes on this instruction"));
3919
3920 p = frag_more (1 + size);
3921 *p++ = i.tm.base_opcode;
3922
3923 fixP = fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3924 i.op[0].disps, 1, reloc (size, 1, 1, i.reloc[0]));
3925
3926 /* All jumps handled here are signed, but don't use a signed limit
3927 check for 32 and 16 bit jumps as we want to allow wrap around at
3928 4G and 64k respectively. */
3929 if (size == 1)
3930 fixP->fx_signed = 1;
3931 }
3932
3933 static void
3934 output_interseg_jump (void)
3935 {
3936 char *p;
3937 int size;
3938 int prefix;
3939 int code16;
3940
3941 code16 = 0;
3942 if (flag_code == CODE_16BIT)
3943 code16 = CODE16;
3944
3945 prefix = 0;
3946 if (i.prefix[DATA_PREFIX] != 0)
3947 {
3948 prefix = 1;
3949 i.prefixes -= 1;
3950 code16 ^= CODE16;
3951 }
3952 if (i.prefix[REX_PREFIX] != 0)
3953 {
3954 prefix++;
3955 i.prefixes -= 1;
3956 }
3957
3958 size = 4;
3959 if (code16)
3960 size = 2;
3961
3962 if (i.prefixes != 0 && !intel_syntax)
3963 as_warn (_("skipping prefixes on this instruction"));
3964
3965 /* 1 opcode; 2 segment; offset */
3966 p = frag_more (prefix + 1 + 2 + size);
3967
3968 if (i.prefix[DATA_PREFIX] != 0)
3969 *p++ = DATA_PREFIX_OPCODE;
3970
3971 if (i.prefix[REX_PREFIX] != 0)
3972 *p++ = i.prefix[REX_PREFIX];
3973
3974 *p++ = i.tm.base_opcode;
3975 if (i.op[1].imms->X_op == O_constant)
3976 {
3977 offsetT n = i.op[1].imms->X_add_number;
3978
3979 if (size == 2
3980 && !fits_in_unsigned_word (n)
3981 && !fits_in_signed_word (n))
3982 {
3983 as_bad (_("16-bit jump out of range"));
3984 return;
3985 }
3986 md_number_to_chars (p, n, size);
3987 }
3988 else
3989 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3990 i.op[1].imms, 0, reloc (size, 0, 0, i.reloc[1]));
3991 if (i.op[0].imms->X_op != O_constant)
3992 as_bad (_("can't handle non absolute segment in `%s'"),
3993 i.tm.name);
3994 md_number_to_chars (p + size, (valueT) i.op[0].imms->X_add_number, 2);
3995 }
3996
3997 static void
3998 output_insn (void)
3999 {
4000 fragS *insn_start_frag;
4001 offsetT insn_start_off;
4002
4003 /* Tie dwarf2 debug info to the address at the start of the insn.
4004 We can't do this after the insn has been output as the current
4005 frag may have been closed off. eg. by frag_var. */
4006 dwarf2_emit_insn (0);
4007
4008 insn_start_frag = frag_now;
4009 insn_start_off = frag_now_fix ();
4010
4011 /* Output jumps. */
4012 if (i.tm.opcode_modifier & Jump)
4013 output_branch ();
4014 else if (i.tm.opcode_modifier & (JumpByte | JumpDword))
4015 output_jump ();
4016 else if (i.tm.opcode_modifier & JumpInterSegment)
4017 output_interseg_jump ();
4018 else
4019 {
4020 /* Output normal instructions here. */
4021 char *p;
4022 unsigned char *q;
4023 unsigned int prefix;
4024
4025 /* All opcodes on i386 have either 1 or 2 bytes. SSSE3 and
4026 SSE4 instructions have 3 bytes. We may use one more higher
4027 byte to specify a prefix the instruction requires. Exclude
4028 instructions which are in both SSE4 and ABM. */
4029 if ((i.tm.cpu_flags & (CpuSSSE3 | CpuSSE4)) != 0
4030 && (i.tm.cpu_flags & CpuABM) == 0)
4031 {
4032 if (i.tm.base_opcode & 0xff000000)
4033 {
4034 prefix = (i.tm.base_opcode >> 24) & 0xff;
4035 goto check_prefix;
4036 }
4037 }
4038 else if ((i.tm.base_opcode & 0xff0000) != 0)
4039 {
4040 prefix = (i.tm.base_opcode >> 16) & 0xff;
4041 if ((i.tm.cpu_flags & CpuPadLock) != 0)
4042 {
4043 check_prefix:
4044 if (prefix != REPE_PREFIX_OPCODE
4045 || i.prefix[LOCKREP_PREFIX] != REPE_PREFIX_OPCODE)
4046 add_prefix (prefix);
4047 }
4048 else
4049 add_prefix (prefix);
4050 }
4051
4052 /* The prefix bytes. */
4053 for (q = i.prefix;
4054 q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
4055 q++)
4056 {
4057 if (*q)
4058 {
4059 p = frag_more (1);
4060 md_number_to_chars (p, (valueT) *q, 1);
4061 }
4062 }
4063
4064 /* Now the opcode; be careful about word order here! */
4065 if (fits_in_unsigned_byte (i.tm.base_opcode))
4066 {
4067 FRAG_APPEND_1_CHAR (i.tm.base_opcode);
4068 }
4069 else
4070 {
4071 if ((i.tm.cpu_flags & (CpuSSSE3 | CpuSSE4)) != 0
4072 && (i.tm.cpu_flags & CpuABM) == 0)
4073 {
4074 p = frag_more (3);
4075 *p++ = (i.tm.base_opcode >> 16) & 0xff;
4076 }
4077 else
4078 p = frag_more (2);
4079
4080 /* Put out high byte first: can't use md_number_to_chars! */
4081 *p++ = (i.tm.base_opcode >> 8) & 0xff;
4082 *p = i.tm.base_opcode & 0xff;
4083 }
4084
4085 /* Now the modrm byte and sib byte (if present). */
4086 if (i.tm.opcode_modifier & Modrm)
4087 {
4088 p = frag_more (1);
4089 md_number_to_chars (p,
4090 (valueT) (i.rm.regmem << 0
4091 | i.rm.reg << 3
4092 | i.rm.mode << 6),
4093 1);
4094 /* If i.rm.regmem == ESP (4)
4095 && i.rm.mode != (Register mode)
4096 && not 16 bit
4097 ==> need second modrm byte. */
4098 if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING
4099 && i.rm.mode != 3
4100 && !(i.base_reg && (i.base_reg->reg_type & Reg16) != 0))
4101 {
4102 p = frag_more (1);
4103 md_number_to_chars (p,
4104 (valueT) (i.sib.base << 0
4105 | i.sib.index << 3
4106 | i.sib.scale << 6),
4107 1);
4108 }
4109 }
4110
4111 if (i.disp_operands)
4112 output_disp (insn_start_frag, insn_start_off);
4113
4114 if (i.imm_operands)
4115 output_imm (insn_start_frag, insn_start_off);
4116 }
4117
4118 #ifdef DEBUG386
4119 if (flag_debug)
4120 {
4121 pi ("" /*line*/, &i);
4122 }
4123 #endif /* DEBUG386 */
4124 }
4125
4126 /* Return the size of the displacement operand N. */
4127
4128 static int
4129 disp_size (unsigned int n)
4130 {
4131 int size = 4;
4132 if (i.types[n] & (Disp8 | Disp16 | Disp64))
4133 {
4134 size = 2;
4135 if (i.types[n] & Disp8)
4136 size = 1;
4137 if (i.types[n] & Disp64)
4138 size = 8;
4139 }
4140 return size;
4141 }
4142
4143 /* Return the size of the immediate operand N. */
4144
4145 static int
4146 imm_size (unsigned int n)
4147 {
4148 int size = 4;
4149 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
4150 {
4151 size = 2;
4152 if (i.types[n] & (Imm8 | Imm8S))
4153 size = 1;
4154 if (i.types[n] & Imm64)
4155 size = 8;
4156 }
4157 return size;
4158 }
4159
4160 static void
4161 output_disp (fragS *insn_start_frag, offsetT insn_start_off)
4162 {
4163 char *p;
4164 unsigned int n;
4165
4166 for (n = 0; n < i.operands; n++)
4167 {
4168 if (i.types[n] & Disp)
4169 {
4170 if (i.op[n].disps->X_op == O_constant)
4171 {
4172 int size = disp_size (n);
4173 offsetT val;
4174
4175 val = offset_in_range (i.op[n].disps->X_add_number,
4176 size);
4177 p = frag_more (size);
4178 md_number_to_chars (p, val, size);
4179 }
4180 else
4181 {
4182 enum bfd_reloc_code_real reloc_type;
4183 int size = disp_size (n);
4184 int sign = (i.types[n] & Disp32S) != 0;
4185 int pcrel = (i.flags[n] & Operand_PCrel) != 0;
4186
4187 /* We can't have 8 bit displacement here. */
4188 assert ((i.types[n] & Disp8) == 0);
4189
4190 /* The PC relative address is computed relative
4191 to the instruction boundary, so in case immediate
4192 fields follows, we need to adjust the value. */
4193 if (pcrel && i.imm_operands)
4194 {
4195 unsigned int n1;
4196 int sz = 0;
4197
4198 for (n1 = 0; n1 < i.operands; n1++)
4199 if (i.types[n1] & Imm)
4200 {
4201 /* Only one immediate is allowed for PC
4202 relative address. */
4203 assert (sz == 0);
4204 sz = imm_size (n1);
4205 i.op[n].disps->X_add_number -= sz;
4206 }
4207 /* We should find the immediate. */
4208 assert (sz != 0);
4209 }
4210
4211 p = frag_more (size);
4212 reloc_type = reloc (size, pcrel, sign, i.reloc[n]);
4213 if (GOT_symbol
4214 && GOT_symbol == i.op[n].disps->X_add_symbol
4215 && (((reloc_type == BFD_RELOC_32
4216 || reloc_type == BFD_RELOC_X86_64_32S
4217 || (reloc_type == BFD_RELOC_64
4218 && object_64bit))
4219 && (i.op[n].disps->X_op == O_symbol
4220 || (i.op[n].disps->X_op == O_add
4221 && ((symbol_get_value_expression
4222 (i.op[n].disps->X_op_symbol)->X_op)
4223 == O_subtract))))
4224 || reloc_type == BFD_RELOC_32_PCREL))
4225 {
4226 offsetT add;
4227
4228 if (insn_start_frag == frag_now)
4229 add = (p - frag_now->fr_literal) - insn_start_off;
4230 else
4231 {
4232 fragS *fr;
4233
4234 add = insn_start_frag->fr_fix - insn_start_off;
4235 for (fr = insn_start_frag->fr_next;
4236 fr && fr != frag_now; fr = fr->fr_next)
4237 add += fr->fr_fix;
4238 add += p - frag_now->fr_literal;
4239 }
4240
4241 if (!object_64bit)
4242 {
4243 reloc_type = BFD_RELOC_386_GOTPC;
4244 i.op[n].imms->X_add_number += add;
4245 }
4246 else if (reloc_type == BFD_RELOC_64)
4247 reloc_type = BFD_RELOC_X86_64_GOTPC64;
4248 else
4249 /* Don't do the adjustment for x86-64, as there
4250 the pcrel addressing is relative to the _next_
4251 insn, and that is taken care of in other code. */
4252 reloc_type = BFD_RELOC_X86_64_GOTPC32;
4253 }
4254 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
4255 i.op[n].disps, pcrel, reloc_type);
4256 }
4257 }
4258 }
4259 }
4260
4261 static void
4262 output_imm (fragS *insn_start_frag, offsetT insn_start_off)
4263 {
4264 char *p;
4265 unsigned int n;
4266
4267 for (n = 0; n < i.operands; n++)
4268 {
4269 if (i.types[n] & Imm)
4270 {
4271 if (i.op[n].imms->X_op == O_constant)
4272 {
4273 int size = imm_size (n);
4274 offsetT val;
4275
4276 val = offset_in_range (i.op[n].imms->X_add_number,
4277 size);
4278 p = frag_more (size);
4279 md_number_to_chars (p, val, size);
4280 }
4281 else
4282 {
4283 /* Not absolute_section.
4284 Need a 32-bit fixup (don't support 8bit
4285 non-absolute imms). Try to support other
4286 sizes ... */
4287 enum bfd_reloc_code_real reloc_type;
4288 int size = imm_size (n);
4289 int sign;
4290
4291 if ((i.types[n] & (Imm32S))
4292 && (i.suffix == QWORD_MNEM_SUFFIX
4293 || (!i.suffix && (i.tm.opcode_modifier & No_lSuf))))
4294 sign = 1;
4295 else
4296 sign = 0;
4297
4298 p = frag_more (size);
4299 reloc_type = reloc (size, 0, sign, i.reloc[n]);
4300
4301 /* This is tough to explain. We end up with this one if we
4302 * have operands that look like
4303 * "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal here is to
4304 * obtain the absolute address of the GOT, and it is strongly
4305 * preferable from a performance point of view to avoid using
4306 * a runtime relocation for this. The actual sequence of
4307 * instructions often look something like:
4308 *
4309 * call .L66
4310 * .L66:
4311 * popl %ebx
4312 * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
4313 *
4314 * The call and pop essentially return the absolute address
4315 * of the label .L66 and store it in %ebx. The linker itself
4316 * will ultimately change the first operand of the addl so
4317 * that %ebx points to the GOT, but to keep things simple, the
4318 * .o file must have this operand set so that it generates not
4319 * the absolute address of .L66, but the absolute address of
4320 * itself. This allows the linker itself simply treat a GOTPC
4321 * relocation as asking for a pcrel offset to the GOT to be
4322 * added in, and the addend of the relocation is stored in the
4323 * operand field for the instruction itself.
4324 *
4325 * Our job here is to fix the operand so that it would add
4326 * the correct offset so that %ebx would point to itself. The
4327 * thing that is tricky is that .-.L66 will point to the
4328 * beginning of the instruction, so we need to further modify
4329 * the operand so that it will point to itself. There are
4330 * other cases where you have something like:
4331 *
4332 * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
4333 *
4334 * and here no correction would be required. Internally in
4335 * the assembler we treat operands of this form as not being
4336 * pcrel since the '.' is explicitly mentioned, and I wonder
4337 * whether it would simplify matters to do it this way. Who
4338 * knows. In earlier versions of the PIC patches, the
4339 * pcrel_adjust field was used to store the correction, but
4340 * since the expression is not pcrel, I felt it would be
4341 * confusing to do it this way. */
4342
4343 if ((reloc_type == BFD_RELOC_32
4344 || reloc_type == BFD_RELOC_X86_64_32S
4345 || reloc_type == BFD_RELOC_64)
4346 && GOT_symbol
4347 && GOT_symbol == i.op[n].imms->X_add_symbol
4348 && (i.op[n].imms->X_op == O_symbol
4349 || (i.op[n].imms->X_op == O_add
4350 && ((symbol_get_value_expression
4351 (i.op[n].imms->X_op_symbol)->X_op)
4352 == O_subtract))))
4353 {
4354 offsetT add;
4355
4356 if (insn_start_frag == frag_now)
4357 add = (p - frag_now->fr_literal) - insn_start_off;
4358 else
4359 {
4360 fragS *fr;
4361
4362 add = insn_start_frag->fr_fix - insn_start_off;
4363 for (fr = insn_start_frag->fr_next;
4364 fr && fr != frag_now; fr = fr->fr_next)
4365 add += fr->fr_fix;
4366 add += p - frag_now->fr_literal;
4367 }
4368
4369 if (!object_64bit)
4370 reloc_type = BFD_RELOC_386_GOTPC;
4371 else if (size == 4)
4372 reloc_type = BFD_RELOC_X86_64_GOTPC32;
4373 else if (size == 8)
4374 reloc_type = BFD_RELOC_X86_64_GOTPC64;
4375 i.op[n].imms->X_add_number += add;
4376 }
4377 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
4378 i.op[n].imms, 0, reloc_type);
4379 }
4380 }
4381 }
4382 }
4383 \f
4384 /* x86_cons_fix_new is called via the expression parsing code when a
4385 reloc is needed. We use this hook to get the correct .got reloc. */
4386 static enum bfd_reloc_code_real got_reloc = NO_RELOC;
4387 static int cons_sign = -1;
4388
4389 void
4390 x86_cons_fix_new (fragS *frag, unsigned int off, unsigned int len,
4391 expressionS *exp)
4392 {
4393 enum bfd_reloc_code_real r = reloc (len, 0, cons_sign, got_reloc);
4394
4395 got_reloc = NO_RELOC;
4396
4397 #ifdef TE_PE
4398 if (exp->X_op == O_secrel)
4399 {
4400 exp->X_op = O_symbol;
4401 r = BFD_RELOC_32_SECREL;
4402 }
4403 #endif
4404
4405 fix_new_exp (frag, off, len, exp, 0, r);
4406 }
4407
4408 #if (!defined (OBJ_ELF) && !defined (OBJ_MAYBE_ELF)) || defined (LEX_AT)
4409 # define lex_got(reloc, adjust, types) NULL
4410 #else
4411 /* Parse operands of the form
4412 <symbol>@GOTOFF+<nnn>
4413 and similar .plt or .got references.
4414
4415 If we find one, set up the correct relocation in RELOC and copy the
4416 input string, minus the `@GOTOFF' into a malloc'd buffer for
4417 parsing by the calling routine. Return this buffer, and if ADJUST
4418 is non-null set it to the length of the string we removed from the
4419 input line. Otherwise return NULL. */
4420 static char *
4421 lex_got (enum bfd_reloc_code_real *reloc,
4422 int *adjust,
4423 unsigned int *types)
4424 {
4425 /* Some of the relocations depend on the size of what field is to
4426 be relocated. But in our callers i386_immediate and i386_displacement
4427 we don't yet know the operand size (this will be set by insn
4428 matching). Hence we record the word32 relocation here,
4429 and adjust the reloc according to the real size in reloc(). */
4430 static const struct {
4431 const char *str;
4432 const enum bfd_reloc_code_real rel[2];
4433 const unsigned int types64;
4434 } gotrel[] = {
4435 { "PLTOFF", { 0,
4436 BFD_RELOC_X86_64_PLTOFF64 },
4437 Imm64 },
4438 { "PLT", { BFD_RELOC_386_PLT32,
4439 BFD_RELOC_X86_64_PLT32 },
4440 Imm32 | Imm32S | Disp32 },
4441 { "GOTPLT", { 0,
4442 BFD_RELOC_X86_64_GOTPLT64 },
4443 Imm64 | Disp64 },
4444 { "GOTOFF", { BFD_RELOC_386_GOTOFF,
4445 BFD_RELOC_X86_64_GOTOFF64 },
4446 Imm64 | Disp64 },
4447 { "GOTPCREL", { 0,
4448 BFD_RELOC_X86_64_GOTPCREL },
4449 Imm32 | Imm32S | Disp32 },
4450 { "TLSGD", { BFD_RELOC_386_TLS_GD,
4451 BFD_RELOC_X86_64_TLSGD },
4452 Imm32 | Imm32S | Disp32 },
4453 { "TLSLDM", { BFD_RELOC_386_TLS_LDM,
4454 0 },
4455 0 },
4456 { "TLSLD", { 0,
4457 BFD_RELOC_X86_64_TLSLD },
4458 Imm32 | Imm32S | Disp32 },
4459 { "GOTTPOFF", { BFD_RELOC_386_TLS_IE_32,
4460 BFD_RELOC_X86_64_GOTTPOFF },
4461 Imm32 | Imm32S | Disp32 },
4462 { "TPOFF", { BFD_RELOC_386_TLS_LE_32,
4463 BFD_RELOC_X86_64_TPOFF32 },
4464 Imm32 | Imm32S | Imm64 | Disp32 | Disp64 },
4465 { "NTPOFF", { BFD_RELOC_386_TLS_LE,
4466 0 },
4467 0 },
4468 { "DTPOFF", { BFD_RELOC_386_TLS_LDO_32,
4469 BFD_RELOC_X86_64_DTPOFF32 },
4470 Imm32 | Imm32S | Imm64 | Disp32 | Disp64 },
4471 { "GOTNTPOFF",{ BFD_RELOC_386_TLS_GOTIE,
4472 0 },
4473 0 },
4474 { "INDNTPOFF",{ BFD_RELOC_386_TLS_IE,
4475 0 },
4476 0 },
4477 { "GOT", { BFD_RELOC_386_GOT32,
4478 BFD_RELOC_X86_64_GOT32 },
4479 Imm32 | Imm32S | Disp32 | Imm64 },
4480 { "TLSDESC", { BFD_RELOC_386_TLS_GOTDESC,
4481 BFD_RELOC_X86_64_GOTPC32_TLSDESC },
4482 Imm32 | Imm32S | Disp32 },
4483 { "TLSCALL", { BFD_RELOC_386_TLS_DESC_CALL,
4484 BFD_RELOC_X86_64_TLSDESC_CALL },
4485 Imm32 | Imm32S | Disp32 }
4486 };
4487 char *cp;
4488 unsigned int j;
4489
4490 if (!IS_ELF)
4491 return NULL;
4492
4493 for (cp = input_line_pointer; *cp != '@'; cp++)
4494 if (is_end_of_line[(unsigned char) *cp] || *cp == ',')
4495 return NULL;
4496
4497 for (j = 0; j < sizeof (gotrel) / sizeof (gotrel[0]); j++)
4498 {
4499 int len;
4500
4501 len = strlen (gotrel[j].str);
4502 if (strncasecmp (cp + 1, gotrel[j].str, len) == 0)
4503 {
4504 if (gotrel[j].rel[object_64bit] != 0)
4505 {
4506 int first, second;
4507 char *tmpbuf, *past_reloc;
4508
4509 *reloc = gotrel[j].rel[object_64bit];
4510 if (adjust)
4511 *adjust = len;
4512
4513 if (types)
4514 {
4515 if (flag_code != CODE_64BIT)
4516 *types = Imm32 | Disp32;
4517 else
4518 *types = gotrel[j].types64;
4519 }
4520
4521 if (GOT_symbol == NULL)
4522 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
4523
4524 /* The length of the first part of our input line. */
4525 first = cp - input_line_pointer;
4526
4527 /* The second part goes from after the reloc token until
4528 (and including) an end_of_line char or comma. */
4529 past_reloc = cp + 1 + len;
4530 cp = past_reloc;
4531 while (!is_end_of_line[(unsigned char) *cp] && *cp != ',')
4532 ++cp;
4533 second = cp + 1 - past_reloc;
4534
4535 /* Allocate and copy string. The trailing NUL shouldn't
4536 be necessary, but be safe. */
4537 tmpbuf = xmalloc (first + second + 2);
4538 memcpy (tmpbuf, input_line_pointer, first);
4539 if (second != 0 && *past_reloc != ' ')
4540 /* Replace the relocation token with ' ', so that
4541 errors like foo@GOTOFF1 will be detected. */
4542 tmpbuf[first++] = ' ';
4543 memcpy (tmpbuf + first, past_reloc, second);
4544 tmpbuf[first + second] = '\0';
4545 return tmpbuf;
4546 }
4547
4548 as_bad (_("@%s reloc is not supported with %d-bit output format"),
4549 gotrel[j].str, 1 << (5 + object_64bit));
4550 return NULL;
4551 }
4552 }
4553
4554 /* Might be a symbol version string. Don't as_bad here. */
4555 return NULL;
4556 }
4557
4558 void
4559 x86_cons (expressionS *exp, int size)
4560 {
4561 if (size == 4 || (object_64bit && size == 8))
4562 {
4563 /* Handle @GOTOFF and the like in an expression. */
4564 char *save;
4565 char *gotfree_input_line;
4566 int adjust;
4567
4568 save = input_line_pointer;
4569 gotfree_input_line = lex_got (&got_reloc, &adjust, NULL);
4570 if (gotfree_input_line)
4571 input_line_pointer = gotfree_input_line;
4572
4573 expression (exp);
4574
4575 if (gotfree_input_line)
4576 {
4577 /* expression () has merrily parsed up to the end of line,
4578 or a comma - in the wrong buffer. Transfer how far
4579 input_line_pointer has moved to the right buffer. */
4580 input_line_pointer = (save
4581 + (input_line_pointer - gotfree_input_line)
4582 + adjust);
4583 free (gotfree_input_line);
4584 if (exp->X_op == O_constant
4585 || exp->X_op == O_absent
4586 || exp->X_op == O_illegal
4587 || exp->X_op == O_register
4588 || exp->X_op == O_big)
4589 {
4590 char c = *input_line_pointer;
4591 *input_line_pointer = 0;
4592 as_bad (_("missing or invalid expression `%s'"), save);
4593 *input_line_pointer = c;
4594 }
4595 }
4596 }
4597 else
4598 expression (exp);
4599 }
4600 #endif
4601
4602 static void signed_cons (int size)
4603 {
4604 if (flag_code == CODE_64BIT)
4605 cons_sign = 1;
4606 cons (size);
4607 cons_sign = -1;
4608 }
4609
4610 #ifdef TE_PE
4611 static void
4612 pe_directive_secrel (dummy)
4613 int dummy ATTRIBUTE_UNUSED;
4614 {
4615 expressionS exp;
4616
4617 do
4618 {
4619 expression (&exp);
4620 if (exp.X_op == O_symbol)
4621 exp.X_op = O_secrel;
4622
4623 emit_expr (&exp, 4);
4624 }
4625 while (*input_line_pointer++ == ',');
4626
4627 input_line_pointer--;
4628 demand_empty_rest_of_line ();
4629 }
4630 #endif
4631
4632 static int
4633 i386_immediate (char *imm_start)
4634 {
4635 char *save_input_line_pointer;
4636 char *gotfree_input_line;
4637 segT exp_seg = 0;
4638 expressionS *exp;
4639 unsigned int types = ~0U;
4640
4641 if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
4642 {
4643 as_bad (_("at most %d immediate operands are allowed"),
4644 MAX_IMMEDIATE_OPERANDS);
4645 return 0;
4646 }
4647
4648 exp = &im_expressions[i.imm_operands++];
4649 i.op[this_operand].imms = exp;
4650
4651 if (is_space_char (*imm_start))
4652 ++imm_start;
4653
4654 save_input_line_pointer = input_line_pointer;
4655 input_line_pointer = imm_start;
4656
4657 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL, &types);
4658 if (gotfree_input_line)
4659 input_line_pointer = gotfree_input_line;
4660
4661 exp_seg = expression (exp);
4662
4663 SKIP_WHITESPACE ();
4664 if (*input_line_pointer)
4665 as_bad (_("junk `%s' after expression"), input_line_pointer);
4666
4667 input_line_pointer = save_input_line_pointer;
4668 if (gotfree_input_line)
4669 free (gotfree_input_line);
4670
4671 if (exp->X_op == O_absent
4672 || exp->X_op == O_illegal
4673 || exp->X_op == O_big
4674 || (gotfree_input_line
4675 && (exp->X_op == O_constant
4676 || exp->X_op == O_register)))
4677 {
4678 as_bad (_("missing or invalid immediate expression `%s'"),
4679 imm_start);
4680 return 0;
4681 }
4682 else if (exp->X_op == O_constant)
4683 {
4684 /* Size it properly later. */
4685 i.types[this_operand] |= Imm64;
4686 /* If BFD64, sign extend val. */
4687 if (!use_rela_relocations
4688 && (exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0)
4689 exp->X_add_number
4690 = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
4691 }
4692 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4693 else if (OUTPUT_FLAVOR == bfd_target_aout_flavour
4694 && exp_seg != absolute_section
4695 && exp_seg != text_section
4696 && exp_seg != data_section
4697 && exp_seg != bss_section
4698 && exp_seg != undefined_section
4699 && !bfd_is_com_section (exp_seg))
4700 {
4701 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
4702 return 0;
4703 }
4704 #endif
4705 else if (!intel_syntax && exp->X_op == O_register)
4706 {
4707 as_bad (_("illegal immediate register operand %s"), imm_start);
4708 return 0;
4709 }
4710 else
4711 {
4712 /* This is an address. The size of the address will be
4713 determined later, depending on destination register,
4714 suffix, or the default for the section. */
4715 i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64;
4716 i.types[this_operand] &= types;
4717 }
4718
4719 return 1;
4720 }
4721
4722 static char *
4723 i386_scale (char *scale)
4724 {
4725 offsetT val;
4726 char *save = input_line_pointer;
4727
4728 input_line_pointer = scale;
4729 val = get_absolute_expression ();
4730
4731 switch (val)
4732 {
4733 case 1:
4734 i.log2_scale_factor = 0;
4735 break;
4736 case 2:
4737 i.log2_scale_factor = 1;
4738 break;
4739 case 4:
4740 i.log2_scale_factor = 2;
4741 break;
4742 case 8:
4743 i.log2_scale_factor = 3;
4744 break;
4745 default:
4746 {
4747 char sep = *input_line_pointer;
4748
4749 *input_line_pointer = '\0';
4750 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
4751 scale);
4752 *input_line_pointer = sep;
4753 input_line_pointer = save;
4754 return NULL;
4755 }
4756 }
4757 if (i.log2_scale_factor != 0 && i.index_reg == 0)
4758 {
4759 as_warn (_("scale factor of %d without an index register"),
4760 1 << i.log2_scale_factor);
4761 #if SCALE1_WHEN_NO_INDEX
4762 i.log2_scale_factor = 0;
4763 #endif
4764 }
4765 scale = input_line_pointer;
4766 input_line_pointer = save;
4767 return scale;
4768 }
4769
4770 static int
4771 i386_displacement (char *disp_start, char *disp_end)
4772 {
4773 expressionS *exp;
4774 segT exp_seg = 0;
4775 char *save_input_line_pointer;
4776 char *gotfree_input_line;
4777 int bigdisp, override;
4778 unsigned int types = Disp;
4779 int ret;
4780
4781 if (i.disp_operands == MAX_MEMORY_OPERANDS)
4782 {
4783 as_bad (_("at most %d displacement operands are allowed"),
4784 MAX_MEMORY_OPERANDS);
4785 return 0;
4786 }
4787
4788 if ((i.types[this_operand] & JumpAbsolute)
4789 || !(current_templates->start->opcode_modifier & (Jump | JumpDword)))
4790 {
4791 bigdisp = Disp32;
4792 override = (i.prefix[ADDR_PREFIX] != 0);
4793 }
4794 else
4795 {
4796 /* For PC-relative branches, the width of the displacement
4797 is dependent upon data size, not address size. */
4798 bigdisp = 0;
4799 override = (i.prefix[DATA_PREFIX] != 0);
4800 }
4801 if (flag_code == CODE_64BIT)
4802 {
4803 if (!bigdisp)
4804 bigdisp = ((override || i.suffix == WORD_MNEM_SUFFIX)
4805 ? Disp16
4806 : Disp32S | Disp32);
4807 else if (!override)
4808 bigdisp = Disp64 | Disp32S | Disp32;
4809 }
4810 else
4811 {
4812 if (!bigdisp)
4813 {
4814 if (!override)
4815 override = (i.suffix == (flag_code != CODE_16BIT
4816 ? WORD_MNEM_SUFFIX
4817 : LONG_MNEM_SUFFIX));
4818 bigdisp = Disp32;
4819 }
4820 if ((flag_code == CODE_16BIT) ^ override)
4821 bigdisp = Disp16;
4822 }
4823 i.types[this_operand] |= bigdisp;
4824
4825 exp = &disp_expressions[i.disp_operands];
4826 i.op[this_operand].disps = exp;
4827 i.disp_operands++;
4828 save_input_line_pointer = input_line_pointer;
4829 input_line_pointer = disp_start;
4830 END_STRING_AND_SAVE (disp_end);
4831
4832 #ifndef GCC_ASM_O_HACK
4833 #define GCC_ASM_O_HACK 0
4834 #endif
4835 #if GCC_ASM_O_HACK
4836 END_STRING_AND_SAVE (disp_end + 1);
4837 if ((i.types[this_operand] & BaseIndex) != 0
4838 && displacement_string_end[-1] == '+')
4839 {
4840 /* This hack is to avoid a warning when using the "o"
4841 constraint within gcc asm statements.
4842 For instance:
4843
4844 #define _set_tssldt_desc(n,addr,limit,type) \
4845 __asm__ __volatile__ ( \
4846 "movw %w2,%0\n\t" \
4847 "movw %w1,2+%0\n\t" \
4848 "rorl $16,%1\n\t" \
4849 "movb %b1,4+%0\n\t" \
4850 "movb %4,5+%0\n\t" \
4851 "movb $0,6+%0\n\t" \
4852 "movb %h1,7+%0\n\t" \
4853 "rorl $16,%1" \
4854 : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
4855
4856 This works great except that the output assembler ends
4857 up looking a bit weird if it turns out that there is
4858 no offset. You end up producing code that looks like:
4859
4860 #APP
4861 movw $235,(%eax)
4862 movw %dx,2+(%eax)
4863 rorl $16,%edx
4864 movb %dl,4+(%eax)
4865 movb $137,5+(%eax)
4866 movb $0,6+(%eax)
4867 movb %dh,7+(%eax)
4868 rorl $16,%edx
4869 #NO_APP
4870
4871 So here we provide the missing zero. */
4872
4873 *displacement_string_end = '0';
4874 }
4875 #endif
4876 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL, &types);
4877 if (gotfree_input_line)
4878 input_line_pointer = gotfree_input_line;
4879
4880 exp_seg = expression (exp);
4881
4882 SKIP_WHITESPACE ();
4883 if (*input_line_pointer)
4884 as_bad (_("junk `%s' after expression"), input_line_pointer);
4885 #if GCC_ASM_O_HACK
4886 RESTORE_END_STRING (disp_end + 1);
4887 #endif
4888 input_line_pointer = save_input_line_pointer;
4889 if (gotfree_input_line)
4890 free (gotfree_input_line);
4891 ret = 1;
4892
4893 /* We do this to make sure that the section symbol is in
4894 the symbol table. We will ultimately change the relocation
4895 to be relative to the beginning of the section. */
4896 if (i.reloc[this_operand] == BFD_RELOC_386_GOTOFF
4897 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
4898 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTOFF64)
4899 {
4900 if (exp->X_op != O_symbol)
4901 goto inv_disp;
4902
4903 if (S_IS_LOCAL (exp->X_add_symbol)
4904 && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
4905 section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
4906 exp->X_op = O_subtract;
4907 exp->X_op_symbol = GOT_symbol;
4908 if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
4909 i.reloc[this_operand] = BFD_RELOC_32_PCREL;
4910 else if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTOFF64)
4911 i.reloc[this_operand] = BFD_RELOC_64;
4912 else
4913 i.reloc[this_operand] = BFD_RELOC_32;
4914 }
4915
4916 else if (exp->X_op == O_absent
4917 || exp->X_op == O_illegal
4918 || exp->X_op == O_big
4919 || (gotfree_input_line
4920 && (exp->X_op == O_constant
4921 || exp->X_op == O_register)))
4922 {
4923 inv_disp:
4924 as_bad (_("missing or invalid displacement expression `%s'"),
4925 disp_start);
4926 ret = 0;
4927 }
4928
4929 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4930 else if (exp->X_op != O_constant
4931 && OUTPUT_FLAVOR == bfd_target_aout_flavour
4932 && exp_seg != absolute_section
4933 && exp_seg != text_section
4934 && exp_seg != data_section
4935 && exp_seg != bss_section
4936 && exp_seg != undefined_section
4937 && !bfd_is_com_section (exp_seg))
4938 {
4939 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
4940 ret = 0;
4941 }
4942 #endif
4943
4944 RESTORE_END_STRING (disp_end);
4945
4946 if (!(i.types[this_operand] & ~Disp))
4947 i.types[this_operand] &= types;
4948
4949 return ret;
4950 }
4951
4952 /* Make sure the memory operand we've been dealt is valid.
4953 Return 1 on success, 0 on a failure. */
4954
4955 static int
4956 i386_index_check (const char *operand_string)
4957 {
4958 int ok;
4959 #if INFER_ADDR_PREFIX
4960 int fudged = 0;
4961
4962 tryprefix:
4963 #endif
4964 ok = 1;
4965 if (flag_code == CODE_64BIT)
4966 {
4967 unsigned RegXX = (i.prefix[ADDR_PREFIX] == 0 ? Reg64 : Reg32);
4968
4969 if ((i.base_reg
4970 && ((i.base_reg->reg_type & RegXX) == 0)
4971 && (i.base_reg->reg_type != BaseIndex
4972 || i.index_reg))
4973 || (i.index_reg
4974 && ((i.index_reg->reg_type & (RegXX | BaseIndex))
4975 != (RegXX | BaseIndex))))
4976 ok = 0;
4977 }
4978 else
4979 {
4980 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
4981 {
4982 /* 16bit checks. */
4983 if ((i.base_reg
4984 && ((i.base_reg->reg_type & (Reg16 | BaseIndex | RegRex))
4985 != (Reg16 | BaseIndex)))
4986 || (i.index_reg
4987 && (((i.index_reg->reg_type & (Reg16 | BaseIndex))
4988 != (Reg16 | BaseIndex))
4989 || !(i.base_reg
4990 && i.base_reg->reg_num < 6
4991 && i.index_reg->reg_num >= 6
4992 && i.log2_scale_factor == 0))))
4993 ok = 0;
4994 }
4995 else
4996 {
4997 /* 32bit checks. */
4998 if ((i.base_reg
4999 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
5000 || (i.index_reg
5001 && ((i.index_reg->reg_type & (Reg32 | BaseIndex | RegRex))
5002 != (Reg32 | BaseIndex))))
5003 ok = 0;
5004 }
5005 }
5006 if (!ok)
5007 {
5008 #if INFER_ADDR_PREFIX
5009 if (i.prefix[ADDR_PREFIX] == 0)
5010 {
5011 i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
5012 i.prefixes += 1;
5013 /* Change the size of any displacement too. At most one of
5014 Disp16 or Disp32 is set.
5015 FIXME. There doesn't seem to be any real need for separate
5016 Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
5017 Removing them would probably clean up the code quite a lot. */
5018 if (flag_code != CODE_64BIT
5019 && (i.types[this_operand] & (Disp16 | Disp32)))
5020 i.types[this_operand] ^= (Disp16 | Disp32);
5021 fudged = 1;
5022 goto tryprefix;
5023 }
5024 if (fudged)
5025 as_bad (_("`%s' is not a valid base/index expression"),
5026 operand_string);
5027 else
5028 #endif
5029 as_bad (_("`%s' is not a valid %s bit base/index expression"),
5030 operand_string,
5031 flag_code_names[flag_code]);
5032 }
5033 return ok;
5034 }
5035
5036 /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
5037 on error. */
5038
5039 static int
5040 i386_operand (char *operand_string)
5041 {
5042 const reg_entry *r;
5043 char *end_op;
5044 char *op_string = operand_string;
5045
5046 if (is_space_char (*op_string))
5047 ++op_string;
5048
5049 /* We check for an absolute prefix (differentiating,
5050 for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
5051 if (*op_string == ABSOLUTE_PREFIX)
5052 {
5053 ++op_string;
5054 if (is_space_char (*op_string))
5055 ++op_string;
5056 i.types[this_operand] |= JumpAbsolute;
5057 }
5058
5059 /* Check if operand is a register. */
5060 if ((r = parse_register (op_string, &end_op)) != NULL)
5061 {
5062 /* Check for a segment override by searching for ':' after a
5063 segment register. */
5064 op_string = end_op;
5065 if (is_space_char (*op_string))
5066 ++op_string;
5067 if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
5068 {
5069 switch (r->reg_num)
5070 {
5071 case 0:
5072 i.seg[i.mem_operands] = &es;
5073 break;
5074 case 1:
5075 i.seg[i.mem_operands] = &cs;
5076 break;
5077 case 2:
5078 i.seg[i.mem_operands] = &ss;
5079 break;
5080 case 3:
5081 i.seg[i.mem_operands] = &ds;
5082 break;
5083 case 4:
5084 i.seg[i.mem_operands] = &fs;
5085 break;
5086 case 5:
5087 i.seg[i.mem_operands] = &gs;
5088 break;
5089 }
5090
5091 /* Skip the ':' and whitespace. */
5092 ++op_string;
5093 if (is_space_char (*op_string))
5094 ++op_string;
5095
5096 if (!is_digit_char (*op_string)
5097 && !is_identifier_char (*op_string)
5098 && *op_string != '('
5099 && *op_string != ABSOLUTE_PREFIX)
5100 {
5101 as_bad (_("bad memory operand `%s'"), op_string);
5102 return 0;
5103 }
5104 /* Handle case of %es:*foo. */
5105 if (*op_string == ABSOLUTE_PREFIX)
5106 {
5107 ++op_string;
5108 if (is_space_char (*op_string))
5109 ++op_string;
5110 i.types[this_operand] |= JumpAbsolute;
5111 }
5112 goto do_memory_reference;
5113 }
5114 if (*op_string)
5115 {
5116 as_bad (_("junk `%s' after register"), op_string);
5117 return 0;
5118 }
5119 i.types[this_operand] |= r->reg_type & ~BaseIndex;
5120 i.op[this_operand].regs = r;
5121 i.reg_operands++;
5122 }
5123 else if (*op_string == REGISTER_PREFIX)
5124 {
5125 as_bad (_("bad register name `%s'"), op_string);
5126 return 0;
5127 }
5128 else if (*op_string == IMMEDIATE_PREFIX)
5129 {
5130 ++op_string;
5131 if (i.types[this_operand] & JumpAbsolute)
5132 {
5133 as_bad (_("immediate operand illegal with absolute jump"));
5134 return 0;
5135 }
5136 if (!i386_immediate (op_string))
5137 return 0;
5138 }
5139 else if (is_digit_char (*op_string)
5140 || is_identifier_char (*op_string)
5141 || *op_string == '(')
5142 {
5143 /* This is a memory reference of some sort. */
5144 char *base_string;
5145
5146 /* Start and end of displacement string expression (if found). */
5147 char *displacement_string_start;
5148 char *displacement_string_end;
5149
5150 do_memory_reference:
5151 if ((i.mem_operands == 1
5152 && (current_templates->start->opcode_modifier & IsString) == 0)
5153 || i.mem_operands == 2)
5154 {
5155 as_bad (_("too many memory references for `%s'"),
5156 current_templates->start->name);
5157 return 0;
5158 }
5159
5160 /* Check for base index form. We detect the base index form by
5161 looking for an ')' at the end of the operand, searching
5162 for the '(' matching it, and finding a REGISTER_PREFIX or ','
5163 after the '('. */
5164 base_string = op_string + strlen (op_string);
5165
5166 --base_string;
5167 if (is_space_char (*base_string))
5168 --base_string;
5169
5170 /* If we only have a displacement, set-up for it to be parsed later. */
5171 displacement_string_start = op_string;
5172 displacement_string_end = base_string + 1;
5173
5174 if (*base_string == ')')
5175 {
5176 char *temp_string;
5177 unsigned int parens_balanced = 1;
5178 /* We've already checked that the number of left & right ()'s are
5179 equal, so this loop will not be infinite. */
5180 do
5181 {
5182 base_string--;
5183 if (*base_string == ')')
5184 parens_balanced++;
5185 if (*base_string == '(')
5186 parens_balanced--;
5187 }
5188 while (parens_balanced);
5189
5190 temp_string = base_string;
5191
5192 /* Skip past '(' and whitespace. */
5193 ++base_string;
5194 if (is_space_char (*base_string))
5195 ++base_string;
5196
5197 if (*base_string == ','
5198 || ((i.base_reg = parse_register (base_string, &end_op))
5199 != NULL))
5200 {
5201 displacement_string_end = temp_string;
5202
5203 i.types[this_operand] |= BaseIndex;
5204
5205 if (i.base_reg)
5206 {
5207 base_string = end_op;
5208 if (is_space_char (*base_string))
5209 ++base_string;
5210 }
5211
5212 /* There may be an index reg or scale factor here. */
5213 if (*base_string == ',')
5214 {
5215 ++base_string;
5216 if (is_space_char (*base_string))
5217 ++base_string;
5218
5219 if ((i.index_reg = parse_register (base_string, &end_op))
5220 != NULL)
5221 {
5222 base_string = end_op;
5223 if (is_space_char (*base_string))
5224 ++base_string;
5225 if (*base_string == ',')
5226 {
5227 ++base_string;
5228 if (is_space_char (*base_string))
5229 ++base_string;
5230 }
5231 else if (*base_string != ')')
5232 {
5233 as_bad (_("expecting `,' or `)' "
5234 "after index register in `%s'"),
5235 operand_string);
5236 return 0;
5237 }
5238 }
5239 else if (*base_string == REGISTER_PREFIX)
5240 {
5241 as_bad (_("bad register name `%s'"), base_string);
5242 return 0;
5243 }
5244
5245 /* Check for scale factor. */
5246 if (*base_string != ')')
5247 {
5248 char *end_scale = i386_scale (base_string);
5249
5250 if (!end_scale)
5251 return 0;
5252
5253 base_string = end_scale;
5254 if (is_space_char (*base_string))
5255 ++base_string;
5256 if (*base_string != ')')
5257 {
5258 as_bad (_("expecting `)' "
5259 "after scale factor in `%s'"),
5260 operand_string);
5261 return 0;
5262 }
5263 }
5264 else if (!i.index_reg)
5265 {
5266 as_bad (_("expecting index register or scale factor "
5267 "after `,'; got '%c'"),
5268 *base_string);
5269 return 0;
5270 }
5271 }
5272 else if (*base_string != ')')
5273 {
5274 as_bad (_("expecting `,' or `)' "
5275 "after base register in `%s'"),
5276 operand_string);
5277 return 0;
5278 }
5279 }
5280 else if (*base_string == REGISTER_PREFIX)
5281 {
5282 as_bad (_("bad register name `%s'"), base_string);
5283 return 0;
5284 }
5285 }
5286
5287 /* If there's an expression beginning the operand, parse it,
5288 assuming displacement_string_start and
5289 displacement_string_end are meaningful. */
5290 if (displacement_string_start != displacement_string_end)
5291 {
5292 if (!i386_displacement (displacement_string_start,
5293 displacement_string_end))
5294 return 0;
5295 }
5296
5297 /* Special case for (%dx) while doing input/output op. */
5298 if (i.base_reg
5299 && i.base_reg->reg_type == (Reg16 | InOutPortReg)
5300 && i.index_reg == 0
5301 && i.log2_scale_factor == 0
5302 && i.seg[i.mem_operands] == 0
5303 && (i.types[this_operand] & Disp) == 0)
5304 {
5305 i.types[this_operand] = InOutPortReg;
5306 return 1;
5307 }
5308
5309 if (i386_index_check (operand_string) == 0)
5310 return 0;
5311 i.mem_operands++;
5312 }
5313 else
5314 {
5315 /* It's not a memory operand; argh! */
5316 as_bad (_("invalid char %s beginning operand %d `%s'"),
5317 output_invalid (*op_string),
5318 this_operand + 1,
5319 op_string);
5320 return 0;
5321 }
5322 return 1; /* Normal return. */
5323 }
5324 \f
5325 /* md_estimate_size_before_relax()
5326
5327 Called just before relax() for rs_machine_dependent frags. The x86
5328 assembler uses these frags to handle variable size jump
5329 instructions.
5330
5331 Any symbol that is now undefined will not become defined.
5332 Return the correct fr_subtype in the frag.
5333 Return the initial "guess for variable size of frag" to caller.
5334 The guess is actually the growth beyond the fixed part. Whatever
5335 we do to grow the fixed or variable part contributes to our
5336 returned value. */
5337
5338 int
5339 md_estimate_size_before_relax (fragP, segment)
5340 fragS *fragP;
5341 segT segment;
5342 {
5343 /* We've already got fragP->fr_subtype right; all we have to do is
5344 check for un-relaxable symbols. On an ELF system, we can't relax
5345 an externally visible symbol, because it may be overridden by a
5346 shared library. */
5347 if (S_GET_SEGMENT (fragP->fr_symbol) != segment
5348 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5349 || (IS_ELF
5350 && (S_IS_EXTERNAL (fragP->fr_symbol)
5351 || S_IS_WEAK (fragP->fr_symbol)))
5352 #endif
5353 )
5354 {
5355 /* Symbol is undefined in this segment, or we need to keep a
5356 reloc so that weak symbols can be overridden. */
5357 int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
5358 enum bfd_reloc_code_real reloc_type;
5359 unsigned char *opcode;
5360 int old_fr_fix;
5361
5362 if (fragP->fr_var != NO_RELOC)
5363 reloc_type = fragP->fr_var;
5364 else if (size == 2)
5365 reloc_type = BFD_RELOC_16_PCREL;
5366 else
5367 reloc_type = BFD_RELOC_32_PCREL;
5368
5369 old_fr_fix = fragP->fr_fix;
5370 opcode = (unsigned char *) fragP->fr_opcode;
5371
5372 switch (TYPE_FROM_RELAX_STATE (fragP->fr_subtype))
5373 {
5374 case UNCOND_JUMP:
5375 /* Make jmp (0xeb) a (d)word displacement jump. */
5376 opcode[0] = 0xe9;
5377 fragP->fr_fix += size;
5378 fix_new (fragP, old_fr_fix, size,
5379 fragP->fr_symbol,
5380 fragP->fr_offset, 1,
5381 reloc_type);
5382 break;
5383
5384 case COND_JUMP86:
5385 if (size == 2
5386 && (!no_cond_jump_promotion || fragP->fr_var != NO_RELOC))
5387 {
5388 /* Negate the condition, and branch past an
5389 unconditional jump. */
5390 opcode[0] ^= 1;
5391 opcode[1] = 3;
5392 /* Insert an unconditional jump. */
5393 opcode[2] = 0xe9;
5394 /* We added two extra opcode bytes, and have a two byte
5395 offset. */
5396 fragP->fr_fix += 2 + 2;
5397 fix_new (fragP, old_fr_fix + 2, 2,
5398 fragP->fr_symbol,
5399 fragP->fr_offset, 1,
5400 reloc_type);
5401 break;
5402 }
5403 /* Fall through. */
5404
5405 case COND_JUMP:
5406 if (no_cond_jump_promotion && fragP->fr_var == NO_RELOC)
5407 {
5408 fixS *fixP;
5409
5410 fragP->fr_fix += 1;
5411 fixP = fix_new (fragP, old_fr_fix, 1,
5412 fragP->fr_symbol,
5413 fragP->fr_offset, 1,
5414 BFD_RELOC_8_PCREL);
5415 fixP->fx_signed = 1;
5416 break;
5417 }
5418
5419 /* This changes the byte-displacement jump 0x7N
5420 to the (d)word-displacement jump 0x0f,0x8N. */
5421 opcode[1] = opcode[0] + 0x10;
5422 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
5423 /* We've added an opcode byte. */
5424 fragP->fr_fix += 1 + size;
5425 fix_new (fragP, old_fr_fix + 1, size,
5426 fragP->fr_symbol,
5427 fragP->fr_offset, 1,
5428 reloc_type);
5429 break;
5430
5431 default:
5432 BAD_CASE (fragP->fr_subtype);
5433 break;
5434 }
5435 frag_wane (fragP);
5436 return fragP->fr_fix - old_fr_fix;
5437 }
5438
5439 /* Guess size depending on current relax state. Initially the relax
5440 state will correspond to a short jump and we return 1, because
5441 the variable part of the frag (the branch offset) is one byte
5442 long. However, we can relax a section more than once and in that
5443 case we must either set fr_subtype back to the unrelaxed state,
5444 or return the value for the appropriate branch. */
5445 return md_relax_table[fragP->fr_subtype].rlx_length;
5446 }
5447
5448 /* Called after relax() is finished.
5449
5450 In: Address of frag.
5451 fr_type == rs_machine_dependent.
5452 fr_subtype is what the address relaxed to.
5453
5454 Out: Any fixSs and constants are set up.
5455 Caller will turn frag into a ".space 0". */
5456
5457 void
5458 md_convert_frag (abfd, sec, fragP)
5459 bfd *abfd ATTRIBUTE_UNUSED;
5460 segT sec ATTRIBUTE_UNUSED;
5461 fragS *fragP;
5462 {
5463 unsigned char *opcode;
5464 unsigned char *where_to_put_displacement = NULL;
5465 offsetT target_address;
5466 offsetT opcode_address;
5467 unsigned int extension = 0;
5468 offsetT displacement_from_opcode_start;
5469
5470 opcode = (unsigned char *) fragP->fr_opcode;
5471
5472 /* Address we want to reach in file space. */
5473 target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
5474
5475 /* Address opcode resides at in file space. */
5476 opcode_address = fragP->fr_address + fragP->fr_fix;
5477
5478 /* Displacement from opcode start to fill into instruction. */
5479 displacement_from_opcode_start = target_address - opcode_address;
5480
5481 if ((fragP->fr_subtype & BIG) == 0)
5482 {
5483 /* Don't have to change opcode. */
5484 extension = 1; /* 1 opcode + 1 displacement */
5485 where_to_put_displacement = &opcode[1];
5486 }
5487 else
5488 {
5489 if (no_cond_jump_promotion
5490 && TYPE_FROM_RELAX_STATE (fragP->fr_subtype) != UNCOND_JUMP)
5491 as_warn_where (fragP->fr_file, fragP->fr_line,
5492 _("long jump required"));
5493
5494 switch (fragP->fr_subtype)
5495 {
5496 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
5497 extension = 4; /* 1 opcode + 4 displacement */
5498 opcode[0] = 0xe9;
5499 where_to_put_displacement = &opcode[1];
5500 break;
5501
5502 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
5503 extension = 2; /* 1 opcode + 2 displacement */
5504 opcode[0] = 0xe9;
5505 where_to_put_displacement = &opcode[1];
5506 break;
5507
5508 case ENCODE_RELAX_STATE (COND_JUMP, BIG):
5509 case ENCODE_RELAX_STATE (COND_JUMP86, BIG):
5510 extension = 5; /* 2 opcode + 4 displacement */
5511 opcode[1] = opcode[0] + 0x10;
5512 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
5513 where_to_put_displacement = &opcode[2];
5514 break;
5515
5516 case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
5517 extension = 3; /* 2 opcode + 2 displacement */
5518 opcode[1] = opcode[0] + 0x10;
5519 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
5520 where_to_put_displacement = &opcode[2];
5521 break;
5522
5523 case ENCODE_RELAX_STATE (COND_JUMP86, BIG16):
5524 extension = 4;
5525 opcode[0] ^= 1;
5526 opcode[1] = 3;
5527 opcode[2] = 0xe9;
5528 where_to_put_displacement = &opcode[3];
5529 break;
5530
5531 default:
5532 BAD_CASE (fragP->fr_subtype);
5533 break;
5534 }
5535 }
5536
5537 /* If size if less then four we are sure that the operand fits,
5538 but if it's 4, then it could be that the displacement is larger
5539 then -/+ 2GB. */
5540 if (DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype) == 4
5541 && object_64bit
5542 && ((addressT) (displacement_from_opcode_start - extension
5543 + ((addressT) 1 << 31))
5544 > (((addressT) 2 << 31) - 1)))
5545 {
5546 as_bad_where (fragP->fr_file, fragP->fr_line,
5547 _("jump target out of range"));
5548 /* Make us emit 0. */
5549 displacement_from_opcode_start = extension;
5550 }
5551 /* Now put displacement after opcode. */
5552 md_number_to_chars ((char *) where_to_put_displacement,
5553 (valueT) (displacement_from_opcode_start - extension),
5554 DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
5555 fragP->fr_fix += extension;
5556 }
5557 \f
5558 /* Size of byte displacement jmp. */
5559 int md_short_jump_size = 2;
5560
5561 /* Size of dword displacement jmp. */
5562 int md_long_jump_size = 5;
5563
5564 void
5565 md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
5566 char *ptr;
5567 addressT from_addr, to_addr;
5568 fragS *frag ATTRIBUTE_UNUSED;
5569 symbolS *to_symbol ATTRIBUTE_UNUSED;
5570 {
5571 offsetT offset;
5572
5573 offset = to_addr - (from_addr + 2);
5574 /* Opcode for byte-disp jump. */
5575 md_number_to_chars (ptr, (valueT) 0xeb, 1);
5576 md_number_to_chars (ptr + 1, (valueT) offset, 1);
5577 }
5578
5579 void
5580 md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
5581 char *ptr;
5582 addressT from_addr, to_addr;
5583 fragS *frag ATTRIBUTE_UNUSED;
5584 symbolS *to_symbol ATTRIBUTE_UNUSED;
5585 {
5586 offsetT offset;
5587
5588 offset = to_addr - (from_addr + 5);
5589 md_number_to_chars (ptr, (valueT) 0xe9, 1);
5590 md_number_to_chars (ptr + 1, (valueT) offset, 4);
5591 }
5592 \f
5593 /* Apply a fixup (fixS) to segment data, once it has been determined
5594 by our caller that we have all the info we need to fix it up.
5595
5596 On the 386, immediates, displacements, and data pointers are all in
5597 the same (little-endian) format, so we don't need to care about which
5598 we are handling. */
5599
5600 void
5601 md_apply_fix (fixP, valP, seg)
5602 /* The fix we're to put in. */
5603 fixS *fixP;
5604 /* Pointer to the value of the bits. */
5605 valueT *valP;
5606 /* Segment fix is from. */
5607 segT seg ATTRIBUTE_UNUSED;
5608 {
5609 char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
5610 valueT value = *valP;
5611
5612 #if !defined (TE_Mach)
5613 if (fixP->fx_pcrel)
5614 {
5615 switch (fixP->fx_r_type)
5616 {
5617 default:
5618 break;
5619
5620 case BFD_RELOC_64:
5621 fixP->fx_r_type = BFD_RELOC_64_PCREL;
5622 break;
5623 case BFD_RELOC_32:
5624 case BFD_RELOC_X86_64_32S:
5625 fixP->fx_r_type = BFD_RELOC_32_PCREL;
5626 break;
5627 case BFD_RELOC_16:
5628 fixP->fx_r_type = BFD_RELOC_16_PCREL;
5629 break;
5630 case BFD_RELOC_8:
5631 fixP->fx_r_type = BFD_RELOC_8_PCREL;
5632 break;
5633 }
5634 }
5635
5636 if (fixP->fx_addsy != NULL
5637 && (fixP->fx_r_type == BFD_RELOC_32_PCREL
5638 || fixP->fx_r_type == BFD_RELOC_64_PCREL
5639 || fixP->fx_r_type == BFD_RELOC_16_PCREL
5640 || fixP->fx_r_type == BFD_RELOC_8_PCREL)
5641 && !use_rela_relocations)
5642 {
5643 /* This is a hack. There should be a better way to handle this.
5644 This covers for the fact that bfd_install_relocation will
5645 subtract the current location (for partial_inplace, PC relative
5646 relocations); see more below. */
5647 #ifndef OBJ_AOUT
5648 if (IS_ELF
5649 #ifdef TE_PE
5650 || OUTPUT_FLAVOR == bfd_target_coff_flavour
5651 #endif
5652 )
5653 value += fixP->fx_where + fixP->fx_frag->fr_address;
5654 #endif
5655 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5656 if (IS_ELF)
5657 {
5658 segT sym_seg = S_GET_SEGMENT (fixP->fx_addsy);
5659
5660 if ((sym_seg == seg
5661 || (symbol_section_p (fixP->fx_addsy)
5662 && sym_seg != absolute_section))
5663 && !generic_force_reloc (fixP))
5664 {
5665 /* Yes, we add the values in twice. This is because
5666 bfd_install_relocation subtracts them out again. I think
5667 bfd_install_relocation is broken, but I don't dare change
5668 it. FIXME. */
5669 value += fixP->fx_where + fixP->fx_frag->fr_address;
5670 }
5671 }
5672 #endif
5673 #if defined (OBJ_COFF) && defined (TE_PE)
5674 /* For some reason, the PE format does not store a
5675 section address offset for a PC relative symbol. */
5676 if (S_GET_SEGMENT (fixP->fx_addsy) != seg
5677 || S_IS_WEAK (fixP->fx_addsy))
5678 value += md_pcrel_from (fixP);
5679 #endif
5680 }
5681
5682 /* Fix a few things - the dynamic linker expects certain values here,
5683 and we must not disappoint it. */
5684 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5685 if (IS_ELF && fixP->fx_addsy)
5686 switch (fixP->fx_r_type)
5687 {
5688 case BFD_RELOC_386_PLT32:
5689 case BFD_RELOC_X86_64_PLT32:
5690 /* Make the jump instruction point to the address of the operand. At
5691 runtime we merely add the offset to the actual PLT entry. */
5692 value = -4;
5693 break;
5694
5695 case BFD_RELOC_386_TLS_GD:
5696 case BFD_RELOC_386_TLS_LDM:
5697 case BFD_RELOC_386_TLS_IE_32:
5698 case BFD_RELOC_386_TLS_IE:
5699 case BFD_RELOC_386_TLS_GOTIE:
5700 case BFD_RELOC_386_TLS_GOTDESC:
5701 case BFD_RELOC_X86_64_TLSGD:
5702 case BFD_RELOC_X86_64_TLSLD:
5703 case BFD_RELOC_X86_64_GOTTPOFF:
5704 case BFD_RELOC_X86_64_GOTPC32_TLSDESC:
5705 value = 0; /* Fully resolved at runtime. No addend. */
5706 /* Fallthrough */
5707 case BFD_RELOC_386_TLS_LE:
5708 case BFD_RELOC_386_TLS_LDO_32:
5709 case BFD_RELOC_386_TLS_LE_32:
5710 case BFD_RELOC_X86_64_DTPOFF32:
5711 case BFD_RELOC_X86_64_DTPOFF64:
5712 case BFD_RELOC_X86_64_TPOFF32:
5713 case BFD_RELOC_X86_64_TPOFF64:
5714 S_SET_THREAD_LOCAL (fixP->fx_addsy);
5715 break;
5716
5717 case BFD_RELOC_386_TLS_DESC_CALL:
5718 case BFD_RELOC_X86_64_TLSDESC_CALL:
5719 value = 0; /* Fully resolved at runtime. No addend. */
5720 S_SET_THREAD_LOCAL (fixP->fx_addsy);
5721 fixP->fx_done = 0;
5722 return;
5723
5724 case BFD_RELOC_386_GOT32:
5725 case BFD_RELOC_X86_64_GOT32:
5726 value = 0; /* Fully resolved at runtime. No addend. */
5727 break;
5728
5729 case BFD_RELOC_VTABLE_INHERIT:
5730 case BFD_RELOC_VTABLE_ENTRY:
5731 fixP->fx_done = 0;
5732 return;
5733
5734 default:
5735 break;
5736 }
5737 #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
5738 *valP = value;
5739 #endif /* !defined (TE_Mach) */
5740
5741 /* Are we finished with this relocation now? */
5742 if (fixP->fx_addsy == NULL)
5743 fixP->fx_done = 1;
5744 else if (use_rela_relocations)
5745 {
5746 fixP->fx_no_overflow = 1;
5747 /* Remember value for tc_gen_reloc. */
5748 fixP->fx_addnumber = value;
5749 value = 0;
5750 }
5751
5752 md_number_to_chars (p, value, fixP->fx_size);
5753 }
5754 \f
5755 #define MAX_LITTLENUMS 6
5756
5757 /* Turn the string pointed to by litP into a floating point constant
5758 of type TYPE, and emit the appropriate bytes. The number of
5759 LITTLENUMS emitted is stored in *SIZEP. An error message is
5760 returned, or NULL on OK. */
5761
5762 char *
5763 md_atof (type, litP, sizeP)
5764 int type;
5765 char *litP;
5766 int *sizeP;
5767 {
5768 int prec;
5769 LITTLENUM_TYPE words[MAX_LITTLENUMS];
5770 LITTLENUM_TYPE *wordP;
5771 char *t;
5772
5773 switch (type)
5774 {
5775 case 'f':
5776 case 'F':
5777 prec = 2;
5778 break;
5779
5780 case 'd':
5781 case 'D':
5782 prec = 4;
5783 break;
5784
5785 case 'x':
5786 case 'X':
5787 prec = 5;
5788 break;
5789
5790 default:
5791 *sizeP = 0;
5792 return _("Bad call to md_atof ()");
5793 }
5794 t = atof_ieee (input_line_pointer, type, words);
5795 if (t)
5796 input_line_pointer = t;
5797
5798 *sizeP = prec * sizeof (LITTLENUM_TYPE);
5799 /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
5800 the bigendian 386. */
5801 for (wordP = words + prec - 1; prec--;)
5802 {
5803 md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
5804 litP += sizeof (LITTLENUM_TYPE);
5805 }
5806 return 0;
5807 }
5808 \f
5809 static char output_invalid_buf[sizeof (unsigned char) * 2 + 6];
5810
5811 static char *
5812 output_invalid (int c)
5813 {
5814 if (ISPRINT (c))
5815 snprintf (output_invalid_buf, sizeof (output_invalid_buf),
5816 "'%c'", c);
5817 else
5818 snprintf (output_invalid_buf, sizeof (output_invalid_buf),
5819 "(0x%x)", (unsigned char) c);
5820 return output_invalid_buf;
5821 }
5822
5823 /* REG_STRING starts *before* REGISTER_PREFIX. */
5824
5825 static const reg_entry *
5826 parse_real_register (char *reg_string, char **end_op)
5827 {
5828 char *s = reg_string;
5829 char *p;
5830 char reg_name_given[MAX_REG_NAME_SIZE + 1];
5831 const reg_entry *r;
5832
5833 /* Skip possible REGISTER_PREFIX and possible whitespace. */
5834 if (*s == REGISTER_PREFIX)
5835 ++s;
5836
5837 if (is_space_char (*s))
5838 ++s;
5839
5840 p = reg_name_given;
5841 while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
5842 {
5843 if (p >= reg_name_given + MAX_REG_NAME_SIZE)
5844 return (const reg_entry *) NULL;
5845 s++;
5846 }
5847
5848 /* For naked regs, make sure that we are not dealing with an identifier.
5849 This prevents confusing an identifier like `eax_var' with register
5850 `eax'. */
5851 if (allow_naked_reg && identifier_chars[(unsigned char) *s])
5852 return (const reg_entry *) NULL;
5853
5854 *end_op = s;
5855
5856 r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
5857
5858 /* Handle floating point regs, allowing spaces in the (i) part. */
5859 if (r == i386_regtab /* %st is first entry of table */)
5860 {
5861 if (is_space_char (*s))
5862 ++s;
5863 if (*s == '(')
5864 {
5865 ++s;
5866 if (is_space_char (*s))
5867 ++s;
5868 if (*s >= '0' && *s <= '7')
5869 {
5870 int fpr = *s - '0';
5871 ++s;
5872 if (is_space_char (*s))
5873 ++s;
5874 if (*s == ')')
5875 {
5876 *end_op = s + 1;
5877 r = hash_find (reg_hash, "st(0)");
5878 know (r);
5879 return r + fpr;
5880 }
5881 }
5882 /* We have "%st(" then garbage. */
5883 return (const reg_entry *) NULL;
5884 }
5885 }
5886
5887 if (r != NULL
5888 && ((r->reg_flags & (RegRex64 | RegRex)) | (r->reg_type & Reg64)) != 0
5889 && (r->reg_type != Control || !(cpu_arch_flags & CpuSledgehammer))
5890 && flag_code != CODE_64BIT)
5891 return (const reg_entry *) NULL;
5892
5893 return r;
5894 }
5895
5896 /* REG_STRING starts *before* REGISTER_PREFIX. */
5897
5898 static const reg_entry *
5899 parse_register (char *reg_string, char **end_op)
5900 {
5901 const reg_entry *r;
5902
5903 if (*reg_string == REGISTER_PREFIX || allow_naked_reg)
5904 r = parse_real_register (reg_string, end_op);
5905 else
5906 r = NULL;
5907 if (!r)
5908 {
5909 char *save = input_line_pointer;
5910 char c;
5911 symbolS *symbolP;
5912
5913 input_line_pointer = reg_string;
5914 c = get_symbol_end ();
5915 symbolP = symbol_find (reg_string);
5916 if (symbolP && S_GET_SEGMENT (symbolP) == reg_section)
5917 {
5918 const expressionS *e = symbol_get_value_expression (symbolP);
5919
5920 know (e->X_op == O_register);
5921 know (e->X_add_number >= 0
5922 && (valueT) e->X_add_number < i386_regtab_size);
5923 r = i386_regtab + e->X_add_number;
5924 *end_op = input_line_pointer;
5925 }
5926 *input_line_pointer = c;
5927 input_line_pointer = save;
5928 }
5929 return r;
5930 }
5931
5932 int
5933 i386_parse_name (char *name, expressionS *e, char *nextcharP)
5934 {
5935 const reg_entry *r;
5936 char *end = input_line_pointer;
5937
5938 *end = *nextcharP;
5939 r = parse_register (name, &input_line_pointer);
5940 if (r && end <= input_line_pointer)
5941 {
5942 *nextcharP = *input_line_pointer;
5943 *input_line_pointer = 0;
5944 e->X_op = O_register;
5945 e->X_add_number = r - i386_regtab;
5946 return 1;
5947 }
5948 input_line_pointer = end;
5949 *end = 0;
5950 return 0;
5951 }
5952
5953 void
5954 md_operand (expressionS *e)
5955 {
5956 if (*input_line_pointer == REGISTER_PREFIX)
5957 {
5958 char *end;
5959 const reg_entry *r = parse_real_register (input_line_pointer, &end);
5960
5961 if (r)
5962 {
5963 e->X_op = O_register;
5964 e->X_add_number = r - i386_regtab;
5965 input_line_pointer = end;
5966 }
5967 }
5968 }
5969
5970 \f
5971 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5972 const char *md_shortopts = "kVQ:sqn";
5973 #else
5974 const char *md_shortopts = "qn";
5975 #endif
5976
5977 #define OPTION_32 (OPTION_MD_BASE + 0)
5978 #define OPTION_64 (OPTION_MD_BASE + 1)
5979 #define OPTION_DIVIDE (OPTION_MD_BASE + 2)
5980 #define OPTION_MARCH (OPTION_MD_BASE + 3)
5981 #define OPTION_MTUNE (OPTION_MD_BASE + 4)
5982
5983 struct option md_longopts[] =
5984 {
5985 {"32", no_argument, NULL, OPTION_32},
5986 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) || defined(TE_PEP)
5987 {"64", no_argument, NULL, OPTION_64},
5988 #endif
5989 {"divide", no_argument, NULL, OPTION_DIVIDE},
5990 {"march", required_argument, NULL, OPTION_MARCH},
5991 {"mtune", required_argument, NULL, OPTION_MTUNE},
5992 {NULL, no_argument, NULL, 0}
5993 };
5994 size_t md_longopts_size = sizeof (md_longopts);
5995
5996 int
5997 md_parse_option (int c, char *arg)
5998 {
5999 unsigned int i;
6000
6001 switch (c)
6002 {
6003 case 'n':
6004 optimize_align_code = 0;
6005 break;
6006
6007 case 'q':
6008 quiet_warnings = 1;
6009 break;
6010
6011 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
6012 /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
6013 should be emitted or not. FIXME: Not implemented. */
6014 case 'Q':
6015 break;
6016
6017 /* -V: SVR4 argument to print version ID. */
6018 case 'V':
6019 print_version_id ();
6020 break;
6021
6022 /* -k: Ignore for FreeBSD compatibility. */
6023 case 'k':
6024 break;
6025
6026 case 's':
6027 /* -s: On i386 Solaris, this tells the native assembler to use
6028 .stab instead of .stab.excl. We always use .stab anyhow. */
6029 break;
6030 #endif
6031 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) || defined(TE_PEP)
6032 case OPTION_64:
6033 {
6034 const char **list, **l;
6035
6036 list = bfd_target_list ();
6037 for (l = list; *l != NULL; l++)
6038 if (CONST_STRNEQ (*l, "elf64-x86-64")
6039 || strcmp (*l, "coff-x86-64") == 0
6040 || strcmp (*l, "pe-x86-64") == 0
6041 || strcmp (*l, "pei-x86-64") == 0)
6042 {
6043 default_arch = "x86_64";
6044 break;
6045 }
6046 if (*l == NULL)
6047 as_fatal (_("No compiled in support for x86_64"));
6048 free (list);
6049 }
6050 break;
6051 #endif
6052
6053 case OPTION_32:
6054 default_arch = "i386";
6055 break;
6056
6057 case OPTION_DIVIDE:
6058 #ifdef SVR4_COMMENT_CHARS
6059 {
6060 char *n, *t;
6061 const char *s;
6062
6063 n = (char *) xmalloc (strlen (i386_comment_chars) + 1);
6064 t = n;
6065 for (s = i386_comment_chars; *s != '\0'; s++)
6066 if (*s != '/')
6067 *t++ = *s;
6068 *t = '\0';
6069 i386_comment_chars = n;
6070 }
6071 #endif
6072 break;
6073
6074 case OPTION_MARCH:
6075 if (*arg == '.')
6076 as_fatal (_("Invalid -march= option: `%s'"), arg);
6077 for (i = 0; i < ARRAY_SIZE (cpu_arch); i++)
6078 {
6079 if (strcmp (arg, cpu_arch [i].name) == 0)
6080 {
6081 cpu_arch_isa = cpu_arch[i].type;
6082 cpu_arch_isa_flags = cpu_arch[i].flags;
6083 if (!cpu_arch_tune_set)
6084 {
6085 cpu_arch_tune = cpu_arch_isa;
6086 cpu_arch_tune_flags = cpu_arch_isa_flags;
6087 }
6088 break;
6089 }
6090 }
6091 if (i >= ARRAY_SIZE (cpu_arch))
6092 as_fatal (_("Invalid -march= option: `%s'"), arg);
6093 break;
6094
6095 case OPTION_MTUNE:
6096 if (*arg == '.')
6097 as_fatal (_("Invalid -mtune= option: `%s'"), arg);
6098 for (i = 0; i < ARRAY_SIZE (cpu_arch); i++)
6099 {
6100 if (strcmp (arg, cpu_arch [i].name) == 0)
6101 {
6102 cpu_arch_tune_set = 1;
6103 cpu_arch_tune = cpu_arch [i].type;
6104 cpu_arch_tune_flags = cpu_arch[i].flags;
6105 break;
6106 }
6107 }
6108 if (i >= ARRAY_SIZE (cpu_arch))
6109 as_fatal (_("Invalid -mtune= option: `%s'"), arg);
6110 break;
6111
6112 default:
6113 return 0;
6114 }
6115 return 1;
6116 }
6117
6118 void
6119 md_show_usage (stream)
6120 FILE *stream;
6121 {
6122 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
6123 fprintf (stream, _("\
6124 -Q ignored\n\
6125 -V print assembler version number\n\
6126 -k ignored\n"));
6127 #endif
6128 fprintf (stream, _("\
6129 -n Do not optimize code alignment\n\
6130 -q quieten some warnings\n"));
6131 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
6132 fprintf (stream, _("\
6133 -s ignored\n"));
6134 #endif
6135 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) || defined(TE_PEP)
6136 fprintf (stream, _("\
6137 --32/--64 generate 32bit/64bit code\n"));
6138 #endif
6139 #ifdef SVR4_COMMENT_CHARS
6140 fprintf (stream, _("\
6141 --divide do not treat `/' as a comment character\n"));
6142 #else
6143 fprintf (stream, _("\
6144 --divide ignored\n"));
6145 #endif
6146 fprintf (stream, _("\
6147 -march=CPU/-mtune=CPU generate code/optimize for CPU, where CPU is one of:\n\
6148 i386, i486, pentium, pentiumpro, pentium4, nocona,\n\
6149 core, core2, k6, athlon, k8, generic32, generic64\n"));
6150
6151 }
6152
6153 #if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
6154 || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) || defined (TE_PEP))
6155
6156 /* Pick the target format to use. */
6157
6158 const char *
6159 i386_target_format (void)
6160 {
6161 if (!strcmp (default_arch, "x86_64"))
6162 {
6163 set_code_flag (CODE_64BIT);
6164 if (cpu_arch_isa_flags == 0)
6165 cpu_arch_isa_flags = Cpu186|Cpu286|Cpu386|Cpu486
6166 |Cpu586|Cpu686|CpuP4|CpuMMX|CpuMMX2
6167 |CpuSSE|CpuSSE2;
6168 if (cpu_arch_tune_flags == 0)
6169 cpu_arch_tune_flags = Cpu186|Cpu286|Cpu386|Cpu486
6170 |Cpu586|Cpu686|CpuP4|CpuMMX|CpuMMX2
6171 |CpuSSE|CpuSSE2;
6172 }
6173 else if (!strcmp (default_arch, "i386"))
6174 {
6175 set_code_flag (CODE_32BIT);
6176 if (cpu_arch_isa_flags == 0)
6177 cpu_arch_isa_flags = Cpu186|Cpu286|Cpu386;
6178 if (cpu_arch_tune_flags == 0)
6179 cpu_arch_tune_flags = Cpu186|Cpu286|Cpu386;
6180 }
6181 else
6182 as_fatal (_("Unknown architecture"));
6183 switch (OUTPUT_FLAVOR)
6184 {
6185 #ifdef TE_PEP
6186 case bfd_target_coff_flavour:
6187 return flag_code == CODE_64BIT ? COFF_TARGET_FORMAT : "coff-i386";
6188 break;
6189 #endif
6190 #ifdef OBJ_MAYBE_AOUT
6191 case bfd_target_aout_flavour:
6192 return AOUT_TARGET_FORMAT;
6193 #endif
6194 #ifdef OBJ_MAYBE_COFF
6195 case bfd_target_coff_flavour:
6196 return "coff-i386";
6197 #endif
6198 #if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF)
6199 case bfd_target_elf_flavour:
6200 {
6201 if (flag_code == CODE_64BIT)
6202 {
6203 object_64bit = 1;
6204 use_rela_relocations = 1;
6205 }
6206 return flag_code == CODE_64BIT ? ELF_TARGET_FORMAT64 : ELF_TARGET_FORMAT;
6207 }
6208 #endif
6209 default:
6210 abort ();
6211 return NULL;
6212 }
6213 }
6214
6215 #endif /* OBJ_MAYBE_ more than one */
6216
6217 #if (defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
6218 void
6219 i386_elf_emit_arch_note (void)
6220 {
6221 if (IS_ELF && cpu_arch_name != NULL)
6222 {
6223 char *p;
6224 asection *seg = now_seg;
6225 subsegT subseg = now_subseg;
6226 Elf_Internal_Note i_note;
6227 Elf_External_Note e_note;
6228 asection *note_secp;
6229 int len;
6230
6231 /* Create the .note section. */
6232 note_secp = subseg_new (".note", 0);
6233 bfd_set_section_flags (stdoutput,
6234 note_secp,
6235 SEC_HAS_CONTENTS | SEC_READONLY);
6236
6237 /* Process the arch string. */
6238 len = strlen (cpu_arch_name);
6239
6240 i_note.namesz = len + 1;
6241 i_note.descsz = 0;
6242 i_note.type = NT_ARCH;
6243 p = frag_more (sizeof (e_note.namesz));
6244 md_number_to_chars (p, (valueT) i_note.namesz, sizeof (e_note.namesz));
6245 p = frag_more (sizeof (e_note.descsz));
6246 md_number_to_chars (p, (valueT) i_note.descsz, sizeof (e_note.descsz));
6247 p = frag_more (sizeof (e_note.type));
6248 md_number_to_chars (p, (valueT) i_note.type, sizeof (e_note.type));
6249 p = frag_more (len + 1);
6250 strcpy (p, cpu_arch_name);
6251
6252 frag_align (2, 0, 0);
6253
6254 subseg_set (seg, subseg);
6255 }
6256 }
6257 #endif
6258 \f
6259 symbolS *
6260 md_undefined_symbol (name)
6261 char *name;
6262 {
6263 if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
6264 && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
6265 && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
6266 && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
6267 {
6268 if (!GOT_symbol)
6269 {
6270 if (symbol_find (name))
6271 as_bad (_("GOT already in symbol table"));
6272 GOT_symbol = symbol_new (name, undefined_section,
6273 (valueT) 0, &zero_address_frag);
6274 };
6275 return GOT_symbol;
6276 }
6277 return 0;
6278 }
6279
6280 /* Round up a section size to the appropriate boundary. */
6281
6282 valueT
6283 md_section_align (segment, size)
6284 segT segment ATTRIBUTE_UNUSED;
6285 valueT size;
6286 {
6287 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
6288 if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
6289 {
6290 /* For a.out, force the section size to be aligned. If we don't do
6291 this, BFD will align it for us, but it will not write out the
6292 final bytes of the section. This may be a bug in BFD, but it is
6293 easier to fix it here since that is how the other a.out targets
6294 work. */
6295 int align;
6296
6297 align = bfd_get_section_alignment (stdoutput, segment);
6298 size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
6299 }
6300 #endif
6301
6302 return size;
6303 }
6304
6305 /* On the i386, PC-relative offsets are relative to the start of the
6306 next instruction. That is, the address of the offset, plus its
6307 size, since the offset is always the last part of the insn. */
6308
6309 long
6310 md_pcrel_from (fixS *fixP)
6311 {
6312 return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
6313 }
6314
6315 #ifndef I386COFF
6316
6317 static void
6318 s_bss (int ignore ATTRIBUTE_UNUSED)
6319 {
6320 int temp;
6321
6322 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
6323 if (IS_ELF)
6324 obj_elf_section_change_hook ();
6325 #endif
6326 temp = get_absolute_expression ();
6327 subseg_set (bss_section, (subsegT) temp);
6328 demand_empty_rest_of_line ();
6329 }
6330
6331 #endif
6332
6333 void
6334 i386_validate_fix (fixS *fixp)
6335 {
6336 if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
6337 {
6338 if (fixp->fx_r_type == BFD_RELOC_32_PCREL)
6339 {
6340 if (!object_64bit)
6341 abort ();
6342 fixp->fx_r_type = BFD_RELOC_X86_64_GOTPCREL;
6343 }
6344 else
6345 {
6346 if (!object_64bit)
6347 fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
6348 else
6349 fixp->fx_r_type = BFD_RELOC_X86_64_GOTOFF64;
6350 }
6351 fixp->fx_subsy = 0;
6352 }
6353 }
6354
6355 arelent *
6356 tc_gen_reloc (section, fixp)
6357 asection *section ATTRIBUTE_UNUSED;
6358 fixS *fixp;
6359 {
6360 arelent *rel;
6361 bfd_reloc_code_real_type code;
6362
6363 switch (fixp->fx_r_type)
6364 {
6365 case BFD_RELOC_X86_64_PLT32:
6366 case BFD_RELOC_X86_64_GOT32:
6367 case BFD_RELOC_X86_64_GOTPCREL:
6368 case BFD_RELOC_386_PLT32:
6369 case BFD_RELOC_386_GOT32:
6370 case BFD_RELOC_386_GOTOFF:
6371 case BFD_RELOC_386_GOTPC:
6372 case BFD_RELOC_386_TLS_GD:
6373 case BFD_RELOC_386_TLS_LDM:
6374 case BFD_RELOC_386_TLS_LDO_32:
6375 case BFD_RELOC_386_TLS_IE_32:
6376 case BFD_RELOC_386_TLS_IE:
6377 case BFD_RELOC_386_TLS_GOTIE:
6378 case BFD_RELOC_386_TLS_LE_32:
6379 case BFD_RELOC_386_TLS_LE:
6380 case BFD_RELOC_386_TLS_GOTDESC:
6381 case BFD_RELOC_386_TLS_DESC_CALL:
6382 case BFD_RELOC_X86_64_TLSGD:
6383 case BFD_RELOC_X86_64_TLSLD:
6384 case BFD_RELOC_X86_64_DTPOFF32:
6385 case BFD_RELOC_X86_64_DTPOFF64:
6386 case BFD_RELOC_X86_64_GOTTPOFF:
6387 case BFD_RELOC_X86_64_TPOFF32:
6388 case BFD_RELOC_X86_64_TPOFF64:
6389 case BFD_RELOC_X86_64_GOTOFF64:
6390 case BFD_RELOC_X86_64_GOTPC32:
6391 case BFD_RELOC_X86_64_GOT64:
6392 case BFD_RELOC_X86_64_GOTPCREL64:
6393 case BFD_RELOC_X86_64_GOTPC64:
6394 case BFD_RELOC_X86_64_GOTPLT64:
6395 case BFD_RELOC_X86_64_PLTOFF64:
6396 case BFD_RELOC_X86_64_GOTPC32_TLSDESC:
6397 case BFD_RELOC_X86_64_TLSDESC_CALL:
6398 case BFD_RELOC_RVA:
6399 case BFD_RELOC_VTABLE_ENTRY:
6400 case BFD_RELOC_VTABLE_INHERIT:
6401 #ifdef TE_PE
6402 case BFD_RELOC_32_SECREL:
6403 #endif
6404 code = fixp->fx_r_type;
6405 break;
6406 case BFD_RELOC_X86_64_32S:
6407 if (!fixp->fx_pcrel)
6408 {
6409 /* Don't turn BFD_RELOC_X86_64_32S into BFD_RELOC_32. */
6410 code = fixp->fx_r_type;
6411 break;
6412 }
6413 default:
6414 if (fixp->fx_pcrel)
6415 {
6416 switch (fixp->fx_size)
6417 {
6418 default:
6419 as_bad_where (fixp->fx_file, fixp->fx_line,
6420 _("can not do %d byte pc-relative relocation"),
6421 fixp->fx_size);
6422 code = BFD_RELOC_32_PCREL;
6423 break;
6424 case 1: code = BFD_RELOC_8_PCREL; break;
6425 case 2: code = BFD_RELOC_16_PCREL; break;
6426 case 4: code = BFD_RELOC_32_PCREL; break;
6427 #ifdef BFD64
6428 case 8: code = BFD_RELOC_64_PCREL; break;
6429 #endif
6430 }
6431 }
6432 else
6433 {
6434 switch (fixp->fx_size)
6435 {
6436 default:
6437 as_bad_where (fixp->fx_file, fixp->fx_line,
6438 _("can not do %d byte relocation"),
6439 fixp->fx_size);
6440 code = BFD_RELOC_32;
6441 break;
6442 case 1: code = BFD_RELOC_8; break;
6443 case 2: code = BFD_RELOC_16; break;
6444 case 4: code = BFD_RELOC_32; break;
6445 #ifdef BFD64
6446 case 8: code = BFD_RELOC_64; break;
6447 #endif
6448 }
6449 }
6450 break;
6451 }
6452
6453 if ((code == BFD_RELOC_32
6454 || code == BFD_RELOC_32_PCREL
6455 || code == BFD_RELOC_X86_64_32S)
6456 && GOT_symbol
6457 && fixp->fx_addsy == GOT_symbol)
6458 {
6459 if (!object_64bit)
6460 code = BFD_RELOC_386_GOTPC;
6461 else
6462 code = BFD_RELOC_X86_64_GOTPC32;
6463 }
6464 if ((code == BFD_RELOC_64 || code == BFD_RELOC_64_PCREL)
6465 && GOT_symbol
6466 && fixp->fx_addsy == GOT_symbol)
6467 {
6468 code = BFD_RELOC_X86_64_GOTPC64;
6469 }
6470
6471 rel = (arelent *) xmalloc (sizeof (arelent));
6472 rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
6473 *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
6474
6475 rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
6476
6477 if (!use_rela_relocations)
6478 {
6479 /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
6480 vtable entry to be used in the relocation's section offset. */
6481 if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
6482 rel->address = fixp->fx_offset;
6483
6484 rel->addend = 0;
6485 }
6486 /* Use the rela in 64bit mode. */
6487 else
6488 {
6489 if (!fixp->fx_pcrel)
6490 rel->addend = fixp->fx_offset;
6491 else
6492 switch (code)
6493 {
6494 case BFD_RELOC_X86_64_PLT32:
6495 case BFD_RELOC_X86_64_GOT32:
6496 case BFD_RELOC_X86_64_GOTPCREL:
6497 case BFD_RELOC_X86_64_TLSGD:
6498 case BFD_RELOC_X86_64_TLSLD:
6499 case BFD_RELOC_X86_64_GOTTPOFF:
6500 case BFD_RELOC_X86_64_GOTPC32_TLSDESC:
6501 case BFD_RELOC_X86_64_TLSDESC_CALL:
6502 rel->addend = fixp->fx_offset - fixp->fx_size;
6503 break;
6504 default:
6505 rel->addend = (section->vma
6506 - fixp->fx_size
6507 + fixp->fx_addnumber
6508 + md_pcrel_from (fixp));
6509 break;
6510 }
6511 }
6512
6513 rel->howto = bfd_reloc_type_lookup (stdoutput, code);
6514 if (rel->howto == NULL)
6515 {
6516 as_bad_where (fixp->fx_file, fixp->fx_line,
6517 _("cannot represent relocation type %s"),
6518 bfd_get_reloc_code_name (code));
6519 /* Set howto to a garbage value so that we can keep going. */
6520 rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
6521 assert (rel->howto != NULL);
6522 }
6523
6524 return rel;
6525 }
6526
6527 \f
6528 /* Parse operands using Intel syntax. This implements a recursive descent
6529 parser based on the BNF grammar published in Appendix B of the MASM 6.1
6530 Programmer's Guide.
6531
6532 FIXME: We do not recognize the full operand grammar defined in the MASM
6533 documentation. In particular, all the structure/union and
6534 high-level macro operands are missing.
6535
6536 Uppercase words are terminals, lower case words are non-terminals.
6537 Objects surrounded by double brackets '[[' ']]' are optional. Vertical
6538 bars '|' denote choices. Most grammar productions are implemented in
6539 functions called 'intel_<production>'.
6540
6541 Initial production is 'expr'.
6542
6543 addOp + | -
6544
6545 alpha [a-zA-Z]
6546
6547 binOp & | AND | \| | OR | ^ | XOR
6548
6549 byteRegister AL | AH | BL | BH | CL | CH | DL | DH
6550
6551 constant digits [[ radixOverride ]]
6552
6553 dataType BYTE | WORD | DWORD | FWORD | QWORD | TBYTE | OWORD | XMMWORD
6554
6555 digits decdigit
6556 | digits decdigit
6557 | digits hexdigit
6558
6559 decdigit [0-9]
6560
6561 e04 e04 addOp e05
6562 | e05
6563
6564 e05 e05 binOp e06
6565 | e06
6566
6567 e06 e06 mulOp e09
6568 | e09
6569
6570 e09 OFFSET e10
6571 | SHORT e10
6572 | + e10
6573 | - e10
6574 | ~ e10
6575 | NOT e10
6576 | e09 PTR e10
6577 | e09 : e10
6578 | e10
6579
6580 e10 e10 [ expr ]
6581 | e11
6582
6583 e11 ( expr )
6584 | [ expr ]
6585 | constant
6586 | dataType
6587 | id
6588 | $
6589 | register
6590
6591 => expr expr cmpOp e04
6592 | e04
6593
6594 gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
6595 | BP | EBP | SP | ESP | DI | EDI | SI | ESI
6596
6597 hexdigit a | b | c | d | e | f
6598 | A | B | C | D | E | F
6599
6600 id alpha
6601 | id alpha
6602 | id decdigit
6603
6604 mulOp * | / | % | MOD | << | SHL | >> | SHR
6605
6606 quote " | '
6607
6608 register specialRegister
6609 | gpRegister
6610 | byteRegister
6611
6612 segmentRegister CS | DS | ES | FS | GS | SS
6613
6614 specialRegister CR0 | CR2 | CR3 | CR4
6615 | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
6616 | TR3 | TR4 | TR5 | TR6 | TR7
6617
6618 We simplify the grammar in obvious places (e.g., register parsing is
6619 done by calling parse_register) and eliminate immediate left recursion
6620 to implement a recursive-descent parser.
6621
6622 expr e04 expr'
6623
6624 expr' cmpOp e04 expr'
6625 | Empty
6626
6627 e04 e05 e04'
6628
6629 e04' addOp e05 e04'
6630 | Empty
6631
6632 e05 e06 e05'
6633
6634 e05' binOp e06 e05'
6635 | Empty
6636
6637 e06 e09 e06'
6638
6639 e06' mulOp e09 e06'
6640 | Empty
6641
6642 e09 OFFSET e10 e09'
6643 | SHORT e10'
6644 | + e10'
6645 | - e10'
6646 | ~ e10'
6647 | NOT e10'
6648 | e10 e09'
6649
6650 e09' PTR e10 e09'
6651 | : e10 e09'
6652 | Empty
6653
6654 e10 e11 e10'
6655
6656 e10' [ expr ] e10'
6657 | Empty
6658
6659 e11 ( expr )
6660 | [ expr ]
6661 | BYTE
6662 | WORD
6663 | DWORD
6664 | FWORD
6665 | QWORD
6666 | TBYTE
6667 | OWORD
6668 | XMMWORD
6669 | .
6670 | $
6671 | register
6672 | id
6673 | constant */
6674
6675 /* Parsing structure for the intel syntax parser. Used to implement the
6676 semantic actions for the operand grammar. */
6677 struct intel_parser_s
6678 {
6679 char *op_string; /* The string being parsed. */
6680 int got_a_float; /* Whether the operand is a float. */
6681 int op_modifier; /* Operand modifier. */
6682 int is_mem; /* 1 if operand is memory reference. */
6683 int in_offset; /* >=1 if parsing operand of offset. */
6684 int in_bracket; /* >=1 if parsing operand in brackets. */
6685 const reg_entry *reg; /* Last register reference found. */
6686 char *disp; /* Displacement string being built. */
6687 char *next_operand; /* Resume point when splitting operands. */
6688 };
6689
6690 static struct intel_parser_s intel_parser;
6691
6692 /* Token structure for parsing intel syntax. */
6693 struct intel_token
6694 {
6695 int code; /* Token code. */
6696 const reg_entry *reg; /* Register entry for register tokens. */
6697 char *str; /* String representation. */
6698 };
6699
6700 static struct intel_token cur_token, prev_token;
6701
6702 /* Token codes for the intel parser. Since T_SHORT is already used
6703 by COFF, undefine it first to prevent a warning. */
6704 #define T_NIL -1
6705 #define T_CONST 1
6706 #define T_REG 2
6707 #define T_BYTE 3
6708 #define T_WORD 4
6709 #define T_DWORD 5
6710 #define T_FWORD 6
6711 #define T_QWORD 7
6712 #define T_TBYTE 8
6713 #define T_XMMWORD 9
6714 #undef T_SHORT
6715 #define T_SHORT 10
6716 #define T_OFFSET 11
6717 #define T_PTR 12
6718 #define T_ID 13
6719 #define T_SHL 14
6720 #define T_SHR 15
6721
6722 /* Prototypes for intel parser functions. */
6723 static int intel_match_token (int);
6724 static void intel_putback_token (void);
6725 static void intel_get_token (void);
6726 static int intel_expr (void);
6727 static int intel_e04 (void);
6728 static int intel_e05 (void);
6729 static int intel_e06 (void);
6730 static int intel_e09 (void);
6731 static int intel_e10 (void);
6732 static int intel_e11 (void);
6733
6734 static int
6735 i386_intel_operand (char *operand_string, int got_a_float)
6736 {
6737 int ret;
6738 char *p;
6739
6740 p = intel_parser.op_string = xstrdup (operand_string);
6741 intel_parser.disp = (char *) xmalloc (strlen (operand_string) + 1);
6742
6743 for (;;)
6744 {
6745 /* Initialize token holders. */
6746 cur_token.code = prev_token.code = T_NIL;
6747 cur_token.reg = prev_token.reg = NULL;
6748 cur_token.str = prev_token.str = NULL;
6749
6750 /* Initialize parser structure. */
6751 intel_parser.got_a_float = got_a_float;
6752 intel_parser.op_modifier = 0;
6753 intel_parser.is_mem = 0;
6754 intel_parser.in_offset = 0;
6755 intel_parser.in_bracket = 0;
6756 intel_parser.reg = NULL;
6757 intel_parser.disp[0] = '\0';
6758 intel_parser.next_operand = NULL;
6759
6760 /* Read the first token and start the parser. */
6761 intel_get_token ();
6762 ret = intel_expr ();
6763
6764 if (!ret)
6765 break;
6766
6767 if (cur_token.code != T_NIL)
6768 {
6769 as_bad (_("invalid operand for '%s' ('%s' unexpected)"),
6770 current_templates->start->name, cur_token.str);
6771 ret = 0;
6772 }
6773 /* If we found a memory reference, hand it over to i386_displacement
6774 to fill in the rest of the operand fields. */
6775 else if (intel_parser.is_mem)
6776 {
6777 if ((i.mem_operands == 1
6778 && (current_templates->start->opcode_modifier & IsString) == 0)
6779 || i.mem_operands == 2)
6780 {
6781 as_bad (_("too many memory references for '%s'"),
6782 current_templates->start->name);
6783 ret = 0;
6784 }
6785 else
6786 {
6787 char *s = intel_parser.disp;
6788 i.mem_operands++;
6789
6790 if (!quiet_warnings && intel_parser.is_mem < 0)
6791 /* See the comments in intel_bracket_expr. */
6792 as_warn (_("Treating `%s' as memory reference"), operand_string);
6793
6794 /* Add the displacement expression. */
6795 if (*s != '\0')
6796 ret = i386_displacement (s, s + strlen (s));
6797 if (ret)
6798 {
6799 /* Swap base and index in 16-bit memory operands like
6800 [si+bx]. Since i386_index_check is also used in AT&T
6801 mode we have to do that here. */
6802 if (i.base_reg
6803 && i.index_reg
6804 && (i.base_reg->reg_type & Reg16)
6805 && (i.index_reg->reg_type & Reg16)
6806 && i.base_reg->reg_num >= 6
6807 && i.index_reg->reg_num < 6)
6808 {
6809 const reg_entry *base = i.index_reg;
6810
6811 i.index_reg = i.base_reg;
6812 i.base_reg = base;
6813 }
6814 ret = i386_index_check (operand_string);
6815 }
6816 }
6817 }
6818
6819 /* Constant and OFFSET expressions are handled by i386_immediate. */
6820 else if ((intel_parser.op_modifier & (1 << T_OFFSET))
6821 || intel_parser.reg == NULL)
6822 ret = i386_immediate (intel_parser.disp);
6823
6824 if (intel_parser.next_operand && this_operand >= MAX_OPERANDS - 1)
6825 ret = 0;
6826 if (!ret || !intel_parser.next_operand)
6827 break;
6828 intel_parser.op_string = intel_parser.next_operand;
6829 this_operand = i.operands++;
6830 }
6831
6832 free (p);
6833 free (intel_parser.disp);
6834
6835 return ret;
6836 }
6837
6838 #define NUM_ADDRESS_REGS (!!i.base_reg + !!i.index_reg)
6839
6840 /* expr e04 expr'
6841
6842 expr' cmpOp e04 expr'
6843 | Empty */
6844 static int
6845 intel_expr (void)
6846 {
6847 /* XXX Implement the comparison operators. */
6848 return intel_e04 ();
6849 }
6850
6851 /* e04 e05 e04'
6852
6853 e04' addOp e05 e04'
6854 | Empty */
6855 static int
6856 intel_e04 (void)
6857 {
6858 int nregs = -1;
6859
6860 for (;;)
6861 {
6862 if (!intel_e05())
6863 return 0;
6864
6865 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
6866 i.base_reg = i386_regtab + REGNAM_AL; /* al is invalid as base */
6867
6868 if (cur_token.code == '+')
6869 nregs = -1;
6870 else if (cur_token.code == '-')
6871 nregs = NUM_ADDRESS_REGS;
6872 else
6873 return 1;
6874
6875 strcat (intel_parser.disp, cur_token.str);
6876 intel_match_token (cur_token.code);
6877 }
6878 }
6879
6880 /* e05 e06 e05'
6881
6882 e05' binOp e06 e05'
6883 | Empty */
6884 static int
6885 intel_e05 (void)
6886 {
6887 int nregs = ~NUM_ADDRESS_REGS;
6888
6889 for (;;)
6890 {
6891 if (!intel_e06())
6892 return 0;
6893
6894 if (cur_token.code == '&'
6895 || cur_token.code == '|'
6896 || cur_token.code == '^')
6897 {
6898 char str[2];
6899
6900 str[0] = cur_token.code;
6901 str[1] = 0;
6902 strcat (intel_parser.disp, str);
6903 }
6904 else
6905 break;
6906
6907 intel_match_token (cur_token.code);
6908
6909 if (nregs < 0)
6910 nregs = ~nregs;
6911 }
6912 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
6913 i.base_reg = i386_regtab + REGNAM_AL + 1; /* cl is invalid as base */
6914 return 1;
6915 }
6916
6917 /* e06 e09 e06'
6918
6919 e06' mulOp e09 e06'
6920 | Empty */
6921 static int
6922 intel_e06 (void)
6923 {
6924 int nregs = ~NUM_ADDRESS_REGS;
6925
6926 for (;;)
6927 {
6928 if (!intel_e09())
6929 return 0;
6930
6931 if (cur_token.code == '*'
6932 || cur_token.code == '/'
6933 || cur_token.code == '%')
6934 {
6935 char str[2];
6936
6937 str[0] = cur_token.code;
6938 str[1] = 0;
6939 strcat (intel_parser.disp, str);
6940 }
6941 else if (cur_token.code == T_SHL)
6942 strcat (intel_parser.disp, "<<");
6943 else if (cur_token.code == T_SHR)
6944 strcat (intel_parser.disp, ">>");
6945 else
6946 break;
6947
6948 intel_match_token (cur_token.code);
6949
6950 if (nregs < 0)
6951 nregs = ~nregs;
6952 }
6953 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
6954 i.base_reg = i386_regtab + REGNAM_AL + 2; /* dl is invalid as base */
6955 return 1;
6956 }
6957
6958 /* e09 OFFSET e09
6959 | SHORT e09
6960 | + e09
6961 | - e09
6962 | ~ e09
6963 | NOT e09
6964 | e10 e09'
6965
6966 e09' PTR e10 e09'
6967 | : e10 e09'
6968 | Empty */
6969 static int
6970 intel_e09 (void)
6971 {
6972 int nregs = ~NUM_ADDRESS_REGS;
6973 int in_offset = 0;
6974
6975 for (;;)
6976 {
6977 /* Don't consume constants here. */
6978 if (cur_token.code == '+' || cur_token.code == '-')
6979 {
6980 /* Need to look one token ahead - if the next token
6981 is a constant, the current token is its sign. */
6982 int next_code;
6983
6984 intel_match_token (cur_token.code);
6985 next_code = cur_token.code;
6986 intel_putback_token ();
6987 if (next_code == T_CONST)
6988 break;
6989 }
6990
6991 /* e09 OFFSET e09 */
6992 if (cur_token.code == T_OFFSET)
6993 {
6994 if (!in_offset++)
6995 ++intel_parser.in_offset;
6996 }
6997
6998 /* e09 SHORT e09 */
6999 else if (cur_token.code == T_SHORT)
7000 intel_parser.op_modifier |= 1 << T_SHORT;
7001
7002 /* e09 + e09 */
7003 else if (cur_token.code == '+')
7004 strcat (intel_parser.disp, "+");
7005
7006 /* e09 - e09
7007 | ~ e09
7008 | NOT e09 */
7009 else if (cur_token.code == '-' || cur_token.code == '~')
7010 {
7011 char str[2];
7012
7013 if (nregs < 0)
7014 nregs = ~nregs;
7015 str[0] = cur_token.code;
7016 str[1] = 0;
7017 strcat (intel_parser.disp, str);
7018 }
7019
7020 /* e09 e10 e09' */
7021 else
7022 break;
7023
7024 intel_match_token (cur_token.code);
7025 }
7026
7027 for (;;)
7028 {
7029 if (!intel_e10 ())
7030 return 0;
7031
7032 /* e09' PTR e10 e09' */
7033 if (cur_token.code == T_PTR)
7034 {
7035 char suffix;
7036
7037 if (prev_token.code == T_BYTE)
7038 suffix = BYTE_MNEM_SUFFIX;
7039
7040 else if (prev_token.code == T_WORD)
7041 {
7042 if (current_templates->start->name[0] == 'l'
7043 && current_templates->start->name[2] == 's'
7044 && current_templates->start->name[3] == 0)
7045 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
7046 else if (intel_parser.got_a_float == 2) /* "fi..." */
7047 suffix = SHORT_MNEM_SUFFIX;
7048 else
7049 suffix = WORD_MNEM_SUFFIX;
7050 }
7051
7052 else if (prev_token.code == T_DWORD)
7053 {
7054 if (current_templates->start->name[0] == 'l'
7055 && current_templates->start->name[2] == 's'
7056 && current_templates->start->name[3] == 0)
7057 suffix = WORD_MNEM_SUFFIX;
7058 else if (flag_code == CODE_16BIT
7059 && (current_templates->start->opcode_modifier
7060 & (Jump | JumpDword)))
7061 suffix = LONG_DOUBLE_MNEM_SUFFIX;
7062 else if (intel_parser.got_a_float == 1) /* "f..." */
7063 suffix = SHORT_MNEM_SUFFIX;
7064 else
7065 suffix = LONG_MNEM_SUFFIX;
7066 }
7067
7068 else if (prev_token.code == T_FWORD)
7069 {
7070 if (current_templates->start->name[0] == 'l'
7071 && current_templates->start->name[2] == 's'
7072 && current_templates->start->name[3] == 0)
7073 suffix = LONG_MNEM_SUFFIX;
7074 else if (!intel_parser.got_a_float)
7075 {
7076 if (flag_code == CODE_16BIT)
7077 add_prefix (DATA_PREFIX_OPCODE);
7078 suffix = LONG_DOUBLE_MNEM_SUFFIX;
7079 }
7080 else
7081 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
7082 }
7083
7084 else if (prev_token.code == T_QWORD)
7085 {
7086 if (intel_parser.got_a_float == 1) /* "f..." */
7087 suffix = LONG_MNEM_SUFFIX;
7088 else
7089 suffix = QWORD_MNEM_SUFFIX;
7090 }
7091
7092 else if (prev_token.code == T_TBYTE)
7093 {
7094 if (intel_parser.got_a_float == 1)
7095 suffix = LONG_DOUBLE_MNEM_SUFFIX;
7096 else
7097 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
7098 }
7099
7100 else if (prev_token.code == T_XMMWORD)
7101 {
7102 /* XXX ignored for now, but accepted since gcc uses it */
7103 suffix = 0;
7104 }
7105
7106 else
7107 {
7108 as_bad (_("Unknown operand modifier `%s'"), prev_token.str);
7109 return 0;
7110 }
7111
7112 /* Operands for jump/call using 'ptr' notation denote absolute
7113 addresses. */
7114 if (current_templates->start->opcode_modifier & (Jump | JumpDword))
7115 i.types[this_operand] |= JumpAbsolute;
7116
7117 if (current_templates->start->base_opcode == 0x8d /* lea */)
7118 ;
7119 else if (!i.suffix)
7120 i.suffix = suffix;
7121 else if (i.suffix != suffix)
7122 {
7123 as_bad (_("Conflicting operand modifiers"));
7124 return 0;
7125 }
7126
7127 }
7128
7129 /* e09' : e10 e09' */
7130 else if (cur_token.code == ':')
7131 {
7132 if (prev_token.code != T_REG)
7133 {
7134 /* While {call,jmp} SSSS:OOOO is MASM syntax only when SSSS is a
7135 segment/group identifier (which we don't have), using comma
7136 as the operand separator there is even less consistent, since
7137 there all branches only have a single operand. */
7138 if (this_operand != 0
7139 || intel_parser.in_offset
7140 || intel_parser.in_bracket
7141 || (!(current_templates->start->opcode_modifier
7142 & (Jump|JumpDword|JumpInterSegment))
7143 && !(current_templates->start->operand_types[0]
7144 & JumpAbsolute)))
7145 return intel_match_token (T_NIL);
7146 /* Remember the start of the 2nd operand and terminate 1st
7147 operand here.
7148 XXX This isn't right, yet (when SSSS:OOOO is right operand of
7149 another expression), but it gets at least the simplest case
7150 (a plain number or symbol on the left side) right. */
7151 intel_parser.next_operand = intel_parser.op_string;
7152 *--intel_parser.op_string = '\0';
7153 return intel_match_token (':');
7154 }
7155 }
7156
7157 /* e09' Empty */
7158 else
7159 break;
7160
7161 intel_match_token (cur_token.code);
7162
7163 }
7164
7165 if (in_offset)
7166 {
7167 --intel_parser.in_offset;
7168 if (nregs < 0)
7169 nregs = ~nregs;
7170 if (NUM_ADDRESS_REGS > nregs)
7171 {
7172 as_bad (_("Invalid operand to `OFFSET'"));
7173 return 0;
7174 }
7175 intel_parser.op_modifier |= 1 << T_OFFSET;
7176 }
7177
7178 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
7179 i.base_reg = i386_regtab + REGNAM_AL + 3; /* bl is invalid as base */
7180 return 1;
7181 }
7182
7183 static int
7184 intel_bracket_expr (void)
7185 {
7186 int was_offset = intel_parser.op_modifier & (1 << T_OFFSET);
7187 const char *start = intel_parser.op_string;
7188 int len;
7189
7190 if (i.op[this_operand].regs)
7191 return intel_match_token (T_NIL);
7192
7193 intel_match_token ('[');
7194
7195 /* Mark as a memory operand only if it's not already known to be an
7196 offset expression. If it's an offset expression, we need to keep
7197 the brace in. */
7198 if (!intel_parser.in_offset)
7199 {
7200 ++intel_parser.in_bracket;
7201
7202 /* Operands for jump/call inside brackets denote absolute addresses. */
7203 if (current_templates->start->opcode_modifier & (Jump | JumpDword))
7204 i.types[this_operand] |= JumpAbsolute;
7205
7206 /* Unfortunately gas always diverged from MASM in a respect that can't
7207 be easily fixed without risking to break code sequences likely to be
7208 encountered (the testsuite even check for this): MASM doesn't consider
7209 an expression inside brackets unconditionally as a memory reference.
7210 When that is e.g. a constant, an offset expression, or the sum of the
7211 two, this is still taken as a constant load. gas, however, always
7212 treated these as memory references. As a compromise, we'll try to make
7213 offset expressions inside brackets work the MASM way (since that's
7214 less likely to be found in real world code), but make constants alone
7215 continue to work the traditional gas way. In either case, issue a
7216 warning. */
7217 intel_parser.op_modifier &= ~was_offset;
7218 }
7219 else
7220 strcat (intel_parser.disp, "[");
7221
7222 /* Add a '+' to the displacement string if necessary. */
7223 if (*intel_parser.disp != '\0'
7224 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
7225 strcat (intel_parser.disp, "+");
7226
7227 if (intel_expr ()
7228 && (len = intel_parser.op_string - start - 1,
7229 intel_match_token (']')))
7230 {
7231 /* Preserve brackets when the operand is an offset expression. */
7232 if (intel_parser.in_offset)
7233 strcat (intel_parser.disp, "]");
7234 else
7235 {
7236 --intel_parser.in_bracket;
7237 if (i.base_reg || i.index_reg)
7238 intel_parser.is_mem = 1;
7239 if (!intel_parser.is_mem)
7240 {
7241 if (!(intel_parser.op_modifier & (1 << T_OFFSET)))
7242 /* Defer the warning until all of the operand was parsed. */
7243 intel_parser.is_mem = -1;
7244 else if (!quiet_warnings)
7245 as_warn (_("`[%.*s]' taken to mean just `%.*s'"),
7246 len, start, len, start);
7247 }
7248 }
7249 intel_parser.op_modifier |= was_offset;
7250
7251 return 1;
7252 }
7253 return 0;
7254 }
7255
7256 /* e10 e11 e10'
7257
7258 e10' [ expr ] e10'
7259 | Empty */
7260 static int
7261 intel_e10 (void)
7262 {
7263 if (!intel_e11 ())
7264 return 0;
7265
7266 while (cur_token.code == '[')
7267 {
7268 if (!intel_bracket_expr ())
7269 return 0;
7270 }
7271
7272 return 1;
7273 }
7274
7275 /* e11 ( expr )
7276 | [ expr ]
7277 | BYTE
7278 | WORD
7279 | DWORD
7280 | FWORD
7281 | QWORD
7282 | TBYTE
7283 | OWORD
7284 | XMMWORD
7285 | $
7286 | .
7287 | register
7288 | id
7289 | constant */
7290 static int
7291 intel_e11 (void)
7292 {
7293 switch (cur_token.code)
7294 {
7295 /* e11 ( expr ) */
7296 case '(':
7297 intel_match_token ('(');
7298 strcat (intel_parser.disp, "(");
7299
7300 if (intel_expr () && intel_match_token (')'))
7301 {
7302 strcat (intel_parser.disp, ")");
7303 return 1;
7304 }
7305 return 0;
7306
7307 /* e11 [ expr ] */
7308 case '[':
7309 return intel_bracket_expr ();
7310
7311 /* e11 $
7312 | . */
7313 case '.':
7314 strcat (intel_parser.disp, cur_token.str);
7315 intel_match_token (cur_token.code);
7316
7317 /* Mark as a memory operand only if it's not already known to be an
7318 offset expression. */
7319 if (!intel_parser.in_offset)
7320 intel_parser.is_mem = 1;
7321
7322 return 1;
7323
7324 /* e11 register */
7325 case T_REG:
7326 {
7327 const reg_entry *reg = intel_parser.reg = cur_token.reg;
7328
7329 intel_match_token (T_REG);
7330
7331 /* Check for segment change. */
7332 if (cur_token.code == ':')
7333 {
7334 if (!(reg->reg_type & (SReg2 | SReg3)))
7335 {
7336 as_bad (_("`%s' is not a valid segment register"),
7337 reg->reg_name);
7338 return 0;
7339 }
7340 else if (i.seg[i.mem_operands])
7341 as_warn (_("Extra segment override ignored"));
7342 else
7343 {
7344 if (!intel_parser.in_offset)
7345 intel_parser.is_mem = 1;
7346 switch (reg->reg_num)
7347 {
7348 case 0:
7349 i.seg[i.mem_operands] = &es;
7350 break;
7351 case 1:
7352 i.seg[i.mem_operands] = &cs;
7353 break;
7354 case 2:
7355 i.seg[i.mem_operands] = &ss;
7356 break;
7357 case 3:
7358 i.seg[i.mem_operands] = &ds;
7359 break;
7360 case 4:
7361 i.seg[i.mem_operands] = &fs;
7362 break;
7363 case 5:
7364 i.seg[i.mem_operands] = &gs;
7365 break;
7366 }
7367 }
7368 }
7369
7370 /* Not a segment register. Check for register scaling. */
7371 else if (cur_token.code == '*')
7372 {
7373 if (!intel_parser.in_bracket)
7374 {
7375 as_bad (_("Register scaling only allowed in memory operands"));
7376 return 0;
7377 }
7378
7379 if (reg->reg_type & Reg16) /* Disallow things like [si*1]. */
7380 reg = i386_regtab + REGNAM_AX + 4; /* sp is invalid as index */
7381 else if (i.index_reg)
7382 reg = i386_regtab + REGNAM_EAX + 4; /* esp is invalid as index */
7383
7384 /* What follows must be a valid scale. */
7385 intel_match_token ('*');
7386 i.index_reg = reg;
7387 i.types[this_operand] |= BaseIndex;
7388
7389 /* Set the scale after setting the register (otherwise,
7390 i386_scale will complain) */
7391 if (cur_token.code == '+' || cur_token.code == '-')
7392 {
7393 char *str, sign = cur_token.code;
7394 intel_match_token (cur_token.code);
7395 if (cur_token.code != T_CONST)
7396 {
7397 as_bad (_("Syntax error: Expecting a constant, got `%s'"),
7398 cur_token.str);
7399 return 0;
7400 }
7401 str = (char *) xmalloc (strlen (cur_token.str) + 2);
7402 strcpy (str + 1, cur_token.str);
7403 *str = sign;
7404 if (!i386_scale (str))
7405 return 0;
7406 free (str);
7407 }
7408 else if (!i386_scale (cur_token.str))
7409 return 0;
7410 intel_match_token (cur_token.code);
7411 }
7412
7413 /* No scaling. If this is a memory operand, the register is either a
7414 base register (first occurrence) or an index register (second
7415 occurrence). */
7416 else if (intel_parser.in_bracket)
7417 {
7418
7419 if (!i.base_reg)
7420 i.base_reg = reg;
7421 else if (!i.index_reg)
7422 i.index_reg = reg;
7423 else
7424 {
7425 as_bad (_("Too many register references in memory operand"));
7426 return 0;
7427 }
7428
7429 i.types[this_operand] |= BaseIndex;
7430 }
7431
7432 /* It's neither base nor index. */
7433 else if (!intel_parser.in_offset && !intel_parser.is_mem)
7434 {
7435 i.types[this_operand] |= reg->reg_type & ~BaseIndex;
7436 i.op[this_operand].regs = reg;
7437 i.reg_operands++;
7438 }
7439 else
7440 {
7441 as_bad (_("Invalid use of register"));
7442 return 0;
7443 }
7444
7445 /* Since registers are not part of the displacement string (except
7446 when we're parsing offset operands), we may need to remove any
7447 preceding '+' from the displacement string. */
7448 if (*intel_parser.disp != '\0'
7449 && !intel_parser.in_offset)
7450 {
7451 char *s = intel_parser.disp;
7452 s += strlen (s) - 1;
7453 if (*s == '+')
7454 *s = '\0';
7455 }
7456
7457 return 1;
7458 }
7459
7460 /* e11 BYTE
7461 | WORD
7462 | DWORD
7463 | FWORD
7464 | QWORD
7465 | TBYTE
7466 | OWORD
7467 | XMMWORD */
7468 case T_BYTE:
7469 case T_WORD:
7470 case T_DWORD:
7471 case T_FWORD:
7472 case T_QWORD:
7473 case T_TBYTE:
7474 case T_XMMWORD:
7475 intel_match_token (cur_token.code);
7476
7477 if (cur_token.code == T_PTR)
7478 return 1;
7479
7480 /* It must have been an identifier. */
7481 intel_putback_token ();
7482 cur_token.code = T_ID;
7483 /* FALLTHRU */
7484
7485 /* e11 id
7486 | constant */
7487 case T_ID:
7488 if (!intel_parser.in_offset && intel_parser.is_mem <= 0)
7489 {
7490 symbolS *symbolP;
7491
7492 /* The identifier represents a memory reference only if it's not
7493 preceded by an offset modifier and if it's not an equate. */
7494 symbolP = symbol_find(cur_token.str);
7495 if (!symbolP || S_GET_SEGMENT(symbolP) != absolute_section)
7496 intel_parser.is_mem = 1;
7497 }
7498 /* FALLTHRU */
7499
7500 case T_CONST:
7501 case '-':
7502 case '+':
7503 {
7504 char *save_str, sign = 0;
7505
7506 /* Allow constants that start with `+' or `-'. */
7507 if (cur_token.code == '-' || cur_token.code == '+')
7508 {
7509 sign = cur_token.code;
7510 intel_match_token (cur_token.code);
7511 if (cur_token.code != T_CONST)
7512 {
7513 as_bad (_("Syntax error: Expecting a constant, got `%s'"),
7514 cur_token.str);
7515 return 0;
7516 }
7517 }
7518
7519 save_str = (char *) xmalloc (strlen (cur_token.str) + 2);
7520 strcpy (save_str + !!sign, cur_token.str);
7521 if (sign)
7522 *save_str = sign;
7523
7524 /* Get the next token to check for register scaling. */
7525 intel_match_token (cur_token.code);
7526
7527 /* Check if this constant is a scaling factor for an
7528 index register. */
7529 if (cur_token.code == '*')
7530 {
7531 if (intel_match_token ('*') && cur_token.code == T_REG)
7532 {
7533 const reg_entry *reg = cur_token.reg;
7534
7535 if (!intel_parser.in_bracket)
7536 {
7537 as_bad (_("Register scaling only allowed "
7538 "in memory operands"));
7539 return 0;
7540 }
7541
7542 /* Disallow things like [1*si].
7543 sp and esp are invalid as index. */
7544 if (reg->reg_type & Reg16)
7545 reg = i386_regtab + REGNAM_AX + 4;
7546 else if (i.index_reg)
7547 reg = i386_regtab + REGNAM_EAX + 4;
7548
7549 /* The constant is followed by `* reg', so it must be
7550 a valid scale. */
7551 i.index_reg = reg;
7552 i.types[this_operand] |= BaseIndex;
7553
7554 /* Set the scale after setting the register (otherwise,
7555 i386_scale will complain) */
7556 if (!i386_scale (save_str))
7557 return 0;
7558 intel_match_token (T_REG);
7559
7560 /* Since registers are not part of the displacement
7561 string, we may need to remove any preceding '+' from
7562 the displacement string. */
7563 if (*intel_parser.disp != '\0')
7564 {
7565 char *s = intel_parser.disp;
7566 s += strlen (s) - 1;
7567 if (*s == '+')
7568 *s = '\0';
7569 }
7570
7571 free (save_str);
7572
7573 return 1;
7574 }
7575
7576 /* The constant was not used for register scaling. Since we have
7577 already consumed the token following `*' we now need to put it
7578 back in the stream. */
7579 intel_putback_token ();
7580 }
7581
7582 /* Add the constant to the displacement string. */
7583 strcat (intel_parser.disp, save_str);
7584 free (save_str);
7585
7586 return 1;
7587 }
7588 }
7589
7590 as_bad (_("Unrecognized token '%s'"), cur_token.str);
7591 return 0;
7592 }
7593
7594 /* Match the given token against cur_token. If they match, read the next
7595 token from the operand string. */
7596 static int
7597 intel_match_token (int code)
7598 {
7599 if (cur_token.code == code)
7600 {
7601 intel_get_token ();
7602 return 1;
7603 }
7604 else
7605 {
7606 as_bad (_("Unexpected token `%s'"), cur_token.str);
7607 return 0;
7608 }
7609 }
7610
7611 /* Read a new token from intel_parser.op_string and store it in cur_token. */
7612 static void
7613 intel_get_token (void)
7614 {
7615 char *end_op;
7616 const reg_entry *reg;
7617 struct intel_token new_token;
7618
7619 new_token.code = T_NIL;
7620 new_token.reg = NULL;
7621 new_token.str = NULL;
7622
7623 /* Free the memory allocated to the previous token and move
7624 cur_token to prev_token. */
7625 if (prev_token.str)
7626 free (prev_token.str);
7627
7628 prev_token = cur_token;
7629
7630 /* Skip whitespace. */
7631 while (is_space_char (*intel_parser.op_string))
7632 intel_parser.op_string++;
7633
7634 /* Return an empty token if we find nothing else on the line. */
7635 if (*intel_parser.op_string == '\0')
7636 {
7637 cur_token = new_token;
7638 return;
7639 }
7640
7641 /* The new token cannot be larger than the remainder of the operand
7642 string. */
7643 new_token.str = (char *) xmalloc (strlen (intel_parser.op_string) + 1);
7644 new_token.str[0] = '\0';
7645
7646 if (strchr ("0123456789", *intel_parser.op_string))
7647 {
7648 char *p = new_token.str;
7649 char *q = intel_parser.op_string;
7650 new_token.code = T_CONST;
7651
7652 /* Allow any kind of identifier char to encompass floating point and
7653 hexadecimal numbers. */
7654 while (is_identifier_char (*q))
7655 *p++ = *q++;
7656 *p = '\0';
7657
7658 /* Recognize special symbol names [0-9][bf]. */
7659 if (strlen (intel_parser.op_string) == 2
7660 && (intel_parser.op_string[1] == 'b'
7661 || intel_parser.op_string[1] == 'f'))
7662 new_token.code = T_ID;
7663 }
7664
7665 else if ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL)
7666 {
7667 size_t len = end_op - intel_parser.op_string;
7668
7669 new_token.code = T_REG;
7670 new_token.reg = reg;
7671
7672 memcpy (new_token.str, intel_parser.op_string, len);
7673 new_token.str[len] = '\0';
7674 }
7675
7676 else if (is_identifier_char (*intel_parser.op_string))
7677 {
7678 char *p = new_token.str;
7679 char *q = intel_parser.op_string;
7680
7681 /* A '.' or '$' followed by an identifier char is an identifier.
7682 Otherwise, it's operator '.' followed by an expression. */
7683 if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
7684 {
7685 new_token.code = '.';
7686 new_token.str[0] = '.';
7687 new_token.str[1] = '\0';
7688 }
7689 else
7690 {
7691 while (is_identifier_char (*q) || *q == '@')
7692 *p++ = *q++;
7693 *p = '\0';
7694
7695 if (strcasecmp (new_token.str, "NOT") == 0)
7696 new_token.code = '~';
7697
7698 else if (strcasecmp (new_token.str, "MOD") == 0)
7699 new_token.code = '%';
7700
7701 else if (strcasecmp (new_token.str, "AND") == 0)
7702 new_token.code = '&';
7703
7704 else if (strcasecmp (new_token.str, "OR") == 0)
7705 new_token.code = '|';
7706
7707 else if (strcasecmp (new_token.str, "XOR") == 0)
7708 new_token.code = '^';
7709
7710 else if (strcasecmp (new_token.str, "SHL") == 0)
7711 new_token.code = T_SHL;
7712
7713 else if (strcasecmp (new_token.str, "SHR") == 0)
7714 new_token.code = T_SHR;
7715
7716 else if (strcasecmp (new_token.str, "BYTE") == 0)
7717 new_token.code = T_BYTE;
7718
7719 else if (strcasecmp (new_token.str, "WORD") == 0)
7720 new_token.code = T_WORD;
7721
7722 else if (strcasecmp (new_token.str, "DWORD") == 0)
7723 new_token.code = T_DWORD;
7724
7725 else if (strcasecmp (new_token.str, "FWORD") == 0)
7726 new_token.code = T_FWORD;
7727
7728 else if (strcasecmp (new_token.str, "QWORD") == 0)
7729 new_token.code = T_QWORD;
7730
7731 else if (strcasecmp (new_token.str, "TBYTE") == 0
7732 /* XXX remove (gcc still uses it) */
7733 || strcasecmp (new_token.str, "XWORD") == 0)
7734 new_token.code = T_TBYTE;
7735
7736 else if (strcasecmp (new_token.str, "XMMWORD") == 0
7737 || strcasecmp (new_token.str, "OWORD") == 0)
7738 new_token.code = T_XMMWORD;
7739
7740 else if (strcasecmp (new_token.str, "PTR") == 0)
7741 new_token.code = T_PTR;
7742
7743 else if (strcasecmp (new_token.str, "SHORT") == 0)
7744 new_token.code = T_SHORT;
7745
7746 else if (strcasecmp (new_token.str, "OFFSET") == 0)
7747 {
7748 new_token.code = T_OFFSET;
7749
7750 /* ??? This is not mentioned in the MASM grammar but gcc
7751 makes use of it with -mintel-syntax. OFFSET may be
7752 followed by FLAT: */
7753 if (strncasecmp (q, " FLAT:", 6) == 0)
7754 strcat (new_token.str, " FLAT:");
7755 }
7756
7757 /* ??? This is not mentioned in the MASM grammar. */
7758 else if (strcasecmp (new_token.str, "FLAT") == 0)
7759 {
7760 new_token.code = T_OFFSET;
7761 if (*q == ':')
7762 strcat (new_token.str, ":");
7763 else
7764 as_bad (_("`:' expected"));
7765 }
7766
7767 else
7768 new_token.code = T_ID;
7769 }
7770 }
7771
7772 else if (strchr ("+-/*%|&^:[]()~", *intel_parser.op_string))
7773 {
7774 new_token.code = *intel_parser.op_string;
7775 new_token.str[0] = *intel_parser.op_string;
7776 new_token.str[1] = '\0';
7777 }
7778
7779 else if (strchr ("<>", *intel_parser.op_string)
7780 && *intel_parser.op_string == *(intel_parser.op_string + 1))
7781 {
7782 new_token.code = *intel_parser.op_string == '<' ? T_SHL : T_SHR;
7783 new_token.str[0] = *intel_parser.op_string;
7784 new_token.str[1] = *intel_parser.op_string;
7785 new_token.str[2] = '\0';
7786 }
7787
7788 else
7789 as_bad (_("Unrecognized token `%s'"), intel_parser.op_string);
7790
7791 intel_parser.op_string += strlen (new_token.str);
7792 cur_token = new_token;
7793 }
7794
7795 /* Put cur_token back into the token stream and make cur_token point to
7796 prev_token. */
7797 static void
7798 intel_putback_token (void)
7799 {
7800 if (cur_token.code != T_NIL)
7801 {
7802 intel_parser.op_string -= strlen (cur_token.str);
7803 free (cur_token.str);
7804 }
7805 cur_token = prev_token;
7806
7807 /* Forget prev_token. */
7808 prev_token.code = T_NIL;
7809 prev_token.reg = NULL;
7810 prev_token.str = NULL;
7811 }
7812
7813 int
7814 tc_x86_regname_to_dw2regnum (char *regname)
7815 {
7816 unsigned int regnum;
7817 unsigned int regnames_count;
7818 static const char *const regnames_32[] =
7819 {
7820 "eax", "ecx", "edx", "ebx",
7821 "esp", "ebp", "esi", "edi",
7822 "eip", "eflags", NULL,
7823 "st0", "st1", "st2", "st3",
7824 "st4", "st5", "st6", "st7",
7825 NULL, NULL,
7826 "xmm0", "xmm1", "xmm2", "xmm3",
7827 "xmm4", "xmm5", "xmm6", "xmm7",
7828 "mm0", "mm1", "mm2", "mm3",
7829 "mm4", "mm5", "mm6", "mm7",
7830 "fcw", "fsw", "mxcsr",
7831 "es", "cs", "ss", "ds", "fs", "gs", NULL, NULL,
7832 "tr", "ldtr"
7833 };
7834 static const char *const regnames_64[] =
7835 {
7836 "rax", "rdx", "rcx", "rbx",
7837 "rsi", "rdi", "rbp", "rsp",
7838 "r8", "r9", "r10", "r11",
7839 "r12", "r13", "r14", "r15",
7840 "rip",
7841 "xmm0", "xmm1", "xmm2", "xmm3",
7842 "xmm4", "xmm5", "xmm6", "xmm7",
7843 "xmm8", "xmm9", "xmm10", "xmm11",
7844 "xmm12", "xmm13", "xmm14", "xmm15",
7845 "st0", "st1", "st2", "st3",
7846 "st4", "st5", "st6", "st7",
7847 "mm0", "mm1", "mm2", "mm3",
7848 "mm4", "mm5", "mm6", "mm7",
7849 "rflags",
7850 "es", "cs", "ss", "ds", "fs", "gs", NULL, NULL,
7851 "fs.base", "gs.base", NULL, NULL,
7852 "tr", "ldtr",
7853 "mxcsr", "fcw", "fsw"
7854 };
7855 const char *const *regnames;
7856
7857 if (flag_code == CODE_64BIT)
7858 {
7859 regnames = regnames_64;
7860 regnames_count = ARRAY_SIZE (regnames_64);
7861 }
7862 else
7863 {
7864 regnames = regnames_32;
7865 regnames_count = ARRAY_SIZE (regnames_32);
7866 }
7867
7868 for (regnum = 0; regnum < regnames_count; regnum++)
7869 if (regnames[regnum] != NULL
7870 && strcmp (regname, regnames[regnum]) == 0)
7871 return regnum;
7872
7873 return -1;
7874 }
7875
7876 void
7877 tc_x86_frame_initial_instructions (void)
7878 {
7879 static unsigned int sp_regno;
7880
7881 if (!sp_regno)
7882 sp_regno = tc_x86_regname_to_dw2regnum (flag_code == CODE_64BIT
7883 ? "rsp" : "esp");
7884
7885 cfi_add_CFA_def_cfa (sp_regno, -x86_cie_data_alignment);
7886 cfi_add_CFA_offset (x86_dwarf2_return_column, x86_cie_data_alignment);
7887 }
7888
7889 int
7890 i386_elf_section_type (const char *str, size_t len)
7891 {
7892 if (flag_code == CODE_64BIT
7893 && len == sizeof ("unwind") - 1
7894 && strncmp (str, "unwind", 6) == 0)
7895 return SHT_X86_64_UNWIND;
7896
7897 return -1;
7898 }
7899
7900 #ifdef TE_PE
7901 void
7902 tc_pe_dwarf2_emit_offset (symbolS *symbol, unsigned int size)
7903 {
7904 expressionS expr;
7905
7906 expr.X_op = O_secrel;
7907 expr.X_add_symbol = symbol;
7908 expr.X_add_number = 0;
7909 emit_expr (&expr, size);
7910 }
7911 #endif
7912
7913 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
7914 /* For ELF on x86-64, add support for SHF_X86_64_LARGE. */
7915
7916 int
7917 x86_64_section_letter (int letter, char **ptr_msg)
7918 {
7919 if (flag_code == CODE_64BIT)
7920 {
7921 if (letter == 'l')
7922 return SHF_X86_64_LARGE;
7923
7924 *ptr_msg = _("Bad .section directive: want a,l,w,x,M,S,G,T in string");
7925 }
7926 else
7927 *ptr_msg = _("Bad .section directive: want a,w,x,M,S,G,T in string");
7928 return -1;
7929 }
7930
7931 int
7932 x86_64_section_word (char *str, size_t len)
7933 {
7934 if (len == 5 && flag_code == CODE_64BIT && CONST_STRNEQ (str, "large"))
7935 return SHF_X86_64_LARGE;
7936
7937 return -1;
7938 }
7939
7940 static void
7941 handle_large_common (int small ATTRIBUTE_UNUSED)
7942 {
7943 if (flag_code != CODE_64BIT)
7944 {
7945 s_comm_internal (0, elf_common_parse);
7946 as_warn (_(".largecomm supported only in 64bit mode, producing .comm"));
7947 }
7948 else
7949 {
7950 static segT lbss_section;
7951 asection *saved_com_section_ptr = elf_com_section_ptr;
7952 asection *saved_bss_section = bss_section;
7953
7954 if (lbss_section == NULL)
7955 {
7956 flagword applicable;
7957 segT seg = now_seg;
7958 subsegT subseg = now_subseg;
7959
7960 /* The .lbss section is for local .largecomm symbols. */
7961 lbss_section = subseg_new (".lbss", 0);
7962 applicable = bfd_applicable_section_flags (stdoutput);
7963 bfd_set_section_flags (stdoutput, lbss_section,
7964 applicable & SEC_ALLOC);
7965 seg_info (lbss_section)->bss = 1;
7966
7967 subseg_set (seg, subseg);
7968 }
7969
7970 elf_com_section_ptr = &_bfd_elf_large_com_section;
7971 bss_section = lbss_section;
7972
7973 s_comm_internal (0, elf_common_parse);
7974
7975 elf_com_section_ptr = saved_com_section_ptr;
7976 bss_section = saved_bss_section;
7977 }
7978 }
7979 #endif /* OBJ_ELF || OBJ_MAYBE_ELF */
This page took 0.259024 seconds and 4 git commands to generate.