Switch to GPLv3
[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_GENERIC32,
435 Cpu186|Cpu286|Cpu386},
436 {"i486", PROCESSOR_I486,
437 Cpu186|Cpu286|Cpu386|Cpu486},
438 {"i586", PROCESSOR_PENTIUM,
439 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586},
440 {"i686", PROCESSOR_PENTIUMPRO,
441 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686},
442 {"pentium", PROCESSOR_PENTIUM,
443 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586},
444 {"pentiumpro",PROCESSOR_PENTIUMPRO,
445 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686},
446 {"pentiumii", PROCESSOR_PENTIUMPRO,
447 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX},
448 {"pentiumiii",PROCESSOR_PENTIUMPRO,
449 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuMMX2|CpuSSE},
450 {"pentium4", PROCESSOR_PENTIUM4,
451 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
452 |CpuMMX2|CpuSSE|CpuSSE2},
453 {"prescott", PROCESSOR_NOCONA,
454 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
455 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
456 {"nocona", PROCESSOR_NOCONA,
457 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
458 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
459 {"yonah", PROCESSOR_CORE,
460 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
461 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
462 {"core", PROCESSOR_CORE,
463 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
464 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
465 {"merom", PROCESSOR_CORE2,
466 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
467 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSSE3},
468 {"core2", PROCESSOR_CORE2,
469 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
470 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSSE3},
471 {"k6", PROCESSOR_K6,
472 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX},
473 {"k6_2", PROCESSOR_K6,
474 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX|Cpu3dnow},
475 {"athlon", PROCESSOR_ATHLON,
476 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6
477 |CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA},
478 {"sledgehammer", PROCESSOR_K8,
479 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6
480 |CpuSledgehammer|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2},
481 {"opteron", PROCESSOR_K8,
482 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6
483 |CpuSledgehammer|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2},
484 {"k8", PROCESSOR_K8,
485 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6
486 |CpuSledgehammer|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2},
487 {"amdfam10", PROCESSOR_AMDFAM10,
488 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuSledgehammer
489 |CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2|CpuSSE3|CpuSSE4a
490 |CpuABM},
491 {".mmx", PROCESSOR_UNKNOWN,
492 CpuMMX},
493 {".sse", PROCESSOR_UNKNOWN,
494 CpuMMX|CpuMMX2|CpuSSE},
495 {".sse2", PROCESSOR_UNKNOWN,
496 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2},
497 {".sse3", PROCESSOR_UNKNOWN,
498 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
499 {".ssse3", PROCESSOR_UNKNOWN,
500 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSSE3},
501 {".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 f32_15[] =
612 {0xeb,0x0d,0x90,0x90,0x90,0x90,0x90, /* jmp .+15; lotsa nops */
613 0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90};
614 static const char f16_3[] =
615 {0x8d,0x74,0x00}; /* lea 0(%esi),%esi */
616 static const char f16_4[] =
617 {0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
618 static const char f16_5[] =
619 {0x90, /* nop */
620 0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
621 static const char f16_6[] =
622 {0x89,0xf6, /* mov %si,%si */
623 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
624 static const char f16_7[] =
625 {0x8d,0x74,0x00, /* lea 0(%si),%si */
626 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
627 static const char f16_8[] =
628 {0x8d,0xb4,0x00,0x00, /* lea 0w(%si),%si */
629 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
630 static const char *const f32_patt[] = {
631 f32_1, f32_2, f32_3, f32_4, f32_5, f32_6, f32_7, f32_8,
632 f32_9, f32_10, f32_11, f32_12, f32_13, f32_14, f32_15
633 };
634 static const char *const f16_patt[] = {
635 f32_1, f32_2, f16_3, f16_4, f16_5, f16_6, f16_7, f16_8,
636 f32_15, f32_15, f32_15, f32_15, f32_15, f32_15, f32_15
637 };
638 /* nopl (%[re]ax) */
639 static const char alt_3[] =
640 {0x0f,0x1f,0x00};
641 /* nopl 0(%[re]ax) */
642 static const char alt_4[] =
643 {0x0f,0x1f,0x40,0x00};
644 /* nopl 0(%[re]ax,%[re]ax,1) */
645 static const char alt_5[] =
646 {0x0f,0x1f,0x44,0x00,0x00};
647 /* nopw 0(%[re]ax,%[re]ax,1) */
648 static const char alt_6[] =
649 {0x66,0x0f,0x1f,0x44,0x00,0x00};
650 /* nopl 0L(%[re]ax) */
651 static const char alt_7[] =
652 {0x0f,0x1f,0x80,0x00,0x00,0x00,0x00};
653 /* nopl 0L(%[re]ax,%[re]ax,1) */
654 static const char alt_8[] =
655 {0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
656 /* nopw 0L(%[re]ax,%[re]ax,1) */
657 static const char alt_9[] =
658 {0x66,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
659 /* nopw %cs:0L(%[re]ax,%[re]ax,1) */
660 static const char alt_10[] =
661 {0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
662 /* data16
663 nopw %cs:0L(%[re]ax,%[re]ax,1) */
664 static const char alt_long_11[] =
665 {0x66,
666 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
667 /* data16
668 data16
669 nopw %cs:0L(%[re]ax,%[re]ax,1) */
670 static const char alt_long_12[] =
671 {0x66,
672 0x66,
673 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
674 /* data16
675 data16
676 data16
677 nopw %cs:0L(%[re]ax,%[re]ax,1) */
678 static const char alt_long_13[] =
679 {0x66,
680 0x66,
681 0x66,
682 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
683 /* data16
684 data16
685 data16
686 data16
687 nopw %cs:0L(%[re]ax,%[re]ax,1) */
688 static const char alt_long_14[] =
689 {0x66,
690 0x66,
691 0x66,
692 0x66,
693 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
694 /* data16
695 data16
696 data16
697 data16
698 data16
699 nopw %cs:0L(%[re]ax,%[re]ax,1) */
700 static const char alt_long_15[] =
701 {0x66,
702 0x66,
703 0x66,
704 0x66,
705 0x66,
706 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
707 /* nopl 0(%[re]ax,%[re]ax,1)
708 nopw 0(%[re]ax,%[re]ax,1) */
709 static const char alt_short_11[] =
710 {0x0f,0x1f,0x44,0x00,0x00,
711 0x66,0x0f,0x1f,0x44,0x00,0x00};
712 /* nopw 0(%[re]ax,%[re]ax,1)
713 nopw 0(%[re]ax,%[re]ax,1) */
714 static const char alt_short_12[] =
715 {0x66,0x0f,0x1f,0x44,0x00,0x00,
716 0x66,0x0f,0x1f,0x44,0x00,0x00};
717 /* nopw 0(%[re]ax,%[re]ax,1)
718 nopl 0L(%[re]ax) */
719 static const char alt_short_13[] =
720 {0x66,0x0f,0x1f,0x44,0x00,0x00,
721 0x0f,0x1f,0x80,0x00,0x00,0x00,0x00};
722 /* nopl 0L(%[re]ax)
723 nopl 0L(%[re]ax) */
724 static const char alt_short_14[] =
725 {0x0f,0x1f,0x80,0x00,0x00,0x00,0x00,
726 0x0f,0x1f,0x80,0x00,0x00,0x00,0x00};
727 /* nopl 0L(%[re]ax)
728 nopl 0L(%[re]ax,%[re]ax,1) */
729 static const char alt_short_15[] =
730 {0x0f,0x1f,0x80,0x00,0x00,0x00,0x00,
731 0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
732 static const char *const alt_short_patt[] = {
733 f32_1, f32_2, alt_3, alt_4, alt_5, alt_6, alt_7, alt_8,
734 alt_9, alt_10, alt_short_11, alt_short_12, alt_short_13,
735 alt_short_14, alt_short_15
736 };
737 static const char *const alt_long_patt[] = {
738 f32_1, f32_2, alt_3, alt_4, alt_5, alt_6, alt_7, alt_8,
739 alt_9, alt_10, alt_long_11, alt_long_12, alt_long_13,
740 alt_long_14, alt_long_15
741 };
742
743 if (count <= 0 || count > 15)
744 return;
745
746 /* We need to decide which NOP sequence to use for 32bit and
747 64bit. When -mtune= is used:
748
749 1. For PROCESSOR_I486, PROCESSOR_PENTIUM and PROCESSOR_GENERIC32,
750 f32_patt will be used.
751 2. For PROCESSOR_K8 and PROCESSOR_AMDFAM10 in 64bit, NOPs with
752 0x66 prefix will be used.
753 3. For PROCESSOR_CORE2, alt_long_patt will be used.
754 4. For PROCESSOR_PENTIUMPRO, PROCESSOR_PENTIUM4, PROCESSOR_NOCONA,
755 PROCESSOR_CORE, PROCESSOR_CORE2, PROCESSOR_K6, PROCESSOR_ATHLON
756 and PROCESSOR_GENERIC64, alt_short_patt will be used.
757
758 When -mtune= isn't used, alt_short_patt will be used if
759 cpu_arch_isa_flags has Cpu686. Otherwise, f32_patt will be used.
760
761 When -march= or .arch is used, we can't use anything beyond
762 cpu_arch_isa_flags. */
763
764 if (flag_code == CODE_16BIT)
765 {
766 memcpy (fragP->fr_literal + fragP->fr_fix,
767 f16_patt[count - 1], count);
768 if (count > 8)
769 /* Adjust jump offset. */
770 fragP->fr_literal[fragP->fr_fix + 1] = count - 2;
771 }
772 else if (flag_code == CODE_64BIT && cpu_arch_tune == PROCESSOR_K8)
773 {
774 int i;
775 int nnops = (count + 3) / 4;
776 int len = count / nnops;
777 int remains = count - nnops * len;
778 int pos = 0;
779
780 /* The recommended way to pad 64bit code is to use NOPs preceded
781 by maximally four 0x66 prefixes. Balance the size of nops. */
782 for (i = 0; i < remains; i++)
783 {
784 memset (fragP->fr_literal + fragP->fr_fix + pos, 0x66, len);
785 fragP->fr_literal[fragP->fr_fix + pos + len] = 0x90;
786 pos += len + 1;
787 }
788 for (; i < nnops; i++)
789 {
790 memset (fragP->fr_literal + fragP->fr_fix + pos, 0x66, len - 1);
791 fragP->fr_literal[fragP->fr_fix + pos + len - 1] = 0x90;
792 pos += len;
793 }
794 }
795 else
796 {
797 const char *const *patt = NULL;
798
799 if (cpu_arch_isa == PROCESSOR_UNKNOWN)
800 {
801 /* PROCESSOR_UNKNOWN means that all ISAs may be used. */
802 switch (cpu_arch_tune)
803 {
804 case PROCESSOR_UNKNOWN:
805 /* We use cpu_arch_isa_flags to check if we SHOULD
806 optimize for Cpu686. */
807 if ((cpu_arch_isa_flags & Cpu686) != 0)
808 patt = alt_short_patt;
809 else
810 patt = f32_patt;
811 break;
812 case PROCESSOR_CORE2:
813 patt = alt_long_patt;
814 break;
815 case PROCESSOR_PENTIUMPRO:
816 case PROCESSOR_PENTIUM4:
817 case PROCESSOR_NOCONA:
818 case PROCESSOR_CORE:
819 case PROCESSOR_K6:
820 case PROCESSOR_ATHLON:
821 case PROCESSOR_K8:
822 case PROCESSOR_GENERIC64:
823 case PROCESSOR_AMDFAM10:
824 patt = alt_short_patt;
825 break;
826 case PROCESSOR_I486:
827 case PROCESSOR_PENTIUM:
828 case PROCESSOR_GENERIC32:
829 patt = f32_patt;
830 break;
831 }
832 }
833 else
834 {
835 switch (cpu_arch_tune)
836 {
837 case PROCESSOR_UNKNOWN:
838 /* When cpu_arch_isa is net, cpu_arch_tune shouldn't be
839 PROCESSOR_UNKNOWN. */
840 abort ();
841 break;
842
843 case PROCESSOR_I486:
844 case PROCESSOR_PENTIUM:
845 case PROCESSOR_PENTIUMPRO:
846 case PROCESSOR_PENTIUM4:
847 case PROCESSOR_NOCONA:
848 case PROCESSOR_CORE:
849 case PROCESSOR_K6:
850 case PROCESSOR_ATHLON:
851 case PROCESSOR_K8:
852 case PROCESSOR_AMDFAM10:
853 case PROCESSOR_GENERIC32:
854 /* We use cpu_arch_isa_flags to check if we CAN optimize
855 for Cpu686. */
856 if ((cpu_arch_isa_flags & Cpu686) != 0)
857 patt = alt_short_patt;
858 else
859 patt = f32_patt;
860 break;
861 case PROCESSOR_CORE2:
862 if ((cpu_arch_isa_flags & Cpu686) != 0)
863 patt = alt_long_patt;
864 else
865 patt = f32_patt;
866 break;
867 case PROCESSOR_GENERIC64:
868 patt = alt_short_patt;
869 break;
870 }
871 }
872
873 memcpy (fragP->fr_literal + fragP->fr_fix,
874 patt[count - 1], count);
875 }
876 fragP->fr_var = count;
877 }
878
879 static INLINE unsigned int
880 mode_from_disp_size (unsigned int t)
881 {
882 return (t & Disp8) ? 1 : (t & (Disp16 | Disp32 | Disp32S)) ? 2 : 0;
883 }
884
885 static INLINE int
886 fits_in_signed_byte (offsetT num)
887 {
888 return (num >= -128) && (num <= 127);
889 }
890
891 static INLINE int
892 fits_in_unsigned_byte (offsetT num)
893 {
894 return (num & 0xff) == num;
895 }
896
897 static INLINE int
898 fits_in_unsigned_word (offsetT num)
899 {
900 return (num & 0xffff) == num;
901 }
902
903 static INLINE int
904 fits_in_signed_word (offsetT num)
905 {
906 return (-32768 <= num) && (num <= 32767);
907 }
908
909 static INLINE int
910 fits_in_signed_long (offsetT num ATTRIBUTE_UNUSED)
911 {
912 #ifndef BFD64
913 return 1;
914 #else
915 return (!(((offsetT) -1 << 31) & num)
916 || (((offsetT) -1 << 31) & num) == ((offsetT) -1 << 31));
917 #endif
918 } /* fits_in_signed_long() */
919
920 static INLINE int
921 fits_in_unsigned_long (offsetT num ATTRIBUTE_UNUSED)
922 {
923 #ifndef BFD64
924 return 1;
925 #else
926 return (num & (((offsetT) 2 << 31) - 1)) == num;
927 #endif
928 } /* fits_in_unsigned_long() */
929
930 static unsigned int
931 smallest_imm_type (offsetT num)
932 {
933 if (cpu_arch_flags != (Cpu186 | Cpu286 | Cpu386 | Cpu486 | CpuNo64))
934 {
935 /* This code is disabled on the 486 because all the Imm1 forms
936 in the opcode table are slower on the i486. They're the
937 versions with the implicitly specified single-position
938 displacement, which has another syntax if you really want to
939 use that form. */
940 if (num == 1)
941 return Imm1 | Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64;
942 }
943 return (fits_in_signed_byte (num)
944 ? (Imm8S | Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
945 : fits_in_unsigned_byte (num)
946 ? (Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
947 : (fits_in_signed_word (num) || fits_in_unsigned_word (num))
948 ? (Imm16 | Imm32 | Imm32S | Imm64)
949 : fits_in_signed_long (num)
950 ? (Imm32 | Imm32S | Imm64)
951 : fits_in_unsigned_long (num)
952 ? (Imm32 | Imm64)
953 : Imm64);
954 }
955
956 static offsetT
957 offset_in_range (offsetT val, int size)
958 {
959 addressT mask;
960
961 switch (size)
962 {
963 case 1: mask = ((addressT) 1 << 8) - 1; break;
964 case 2: mask = ((addressT) 1 << 16) - 1; break;
965 case 4: mask = ((addressT) 2 << 31) - 1; break;
966 #ifdef BFD64
967 case 8: mask = ((addressT) 2 << 63) - 1; break;
968 #endif
969 default: abort ();
970 }
971
972 /* If BFD64, sign extend val. */
973 if (!use_rela_relocations)
974 if ((val & ~(((addressT) 2 << 31) - 1)) == 0)
975 val = (val ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
976
977 if ((val & ~mask) != 0 && (val & ~mask) != ~mask)
978 {
979 char buf1[40], buf2[40];
980
981 sprint_value (buf1, val);
982 sprint_value (buf2, val & mask);
983 as_warn (_("%s shortened to %s"), buf1, buf2);
984 }
985 return val & mask;
986 }
987
988 /* Returns 0 if attempting to add a prefix where one from the same
989 class already exists, 1 if non rep/repne added, 2 if rep/repne
990 added. */
991 static int
992 add_prefix (unsigned int prefix)
993 {
994 int ret = 1;
995 unsigned int q;
996
997 if (prefix >= REX_OPCODE && prefix < REX_OPCODE + 16
998 && flag_code == CODE_64BIT)
999 {
1000 if ((i.prefix[REX_PREFIX] & prefix & REX_W)
1001 || ((i.prefix[REX_PREFIX] & (REX_R | REX_X | REX_B))
1002 && (prefix & (REX_R | REX_X | REX_B))))
1003 ret = 0;
1004 q = REX_PREFIX;
1005 }
1006 else
1007 {
1008 switch (prefix)
1009 {
1010 default:
1011 abort ();
1012
1013 case CS_PREFIX_OPCODE:
1014 case DS_PREFIX_OPCODE:
1015 case ES_PREFIX_OPCODE:
1016 case FS_PREFIX_OPCODE:
1017 case GS_PREFIX_OPCODE:
1018 case SS_PREFIX_OPCODE:
1019 q = SEG_PREFIX;
1020 break;
1021
1022 case REPNE_PREFIX_OPCODE:
1023 case REPE_PREFIX_OPCODE:
1024 ret = 2;
1025 /* fall thru */
1026 case LOCK_PREFIX_OPCODE:
1027 q = LOCKREP_PREFIX;
1028 break;
1029
1030 case FWAIT_OPCODE:
1031 q = WAIT_PREFIX;
1032 break;
1033
1034 case ADDR_PREFIX_OPCODE:
1035 q = ADDR_PREFIX;
1036 break;
1037
1038 case DATA_PREFIX_OPCODE:
1039 q = DATA_PREFIX;
1040 break;
1041 }
1042 if (i.prefix[q] != 0)
1043 ret = 0;
1044 }
1045
1046 if (ret)
1047 {
1048 if (!i.prefix[q])
1049 ++i.prefixes;
1050 i.prefix[q] |= prefix;
1051 }
1052 else
1053 as_bad (_("same type of prefix used twice"));
1054
1055 return ret;
1056 }
1057
1058 static void
1059 set_code_flag (int value)
1060 {
1061 flag_code = value;
1062 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
1063 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
1064 if (value == CODE_64BIT && !(cpu_arch_flags & CpuSledgehammer))
1065 {
1066 as_bad (_("64bit mode not supported on this CPU."));
1067 }
1068 if (value == CODE_32BIT && !(cpu_arch_flags & Cpu386))
1069 {
1070 as_bad (_("32bit mode not supported on this CPU."));
1071 }
1072 stackop_size = '\0';
1073 }
1074
1075 static void
1076 set_16bit_gcc_code_flag (int new_code_flag)
1077 {
1078 flag_code = new_code_flag;
1079 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
1080 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
1081 stackop_size = LONG_MNEM_SUFFIX;
1082 }
1083
1084 static void
1085 set_intel_syntax (int syntax_flag)
1086 {
1087 /* Find out if register prefixing is specified. */
1088 int ask_naked_reg = 0;
1089
1090 SKIP_WHITESPACE ();
1091 if (!is_end_of_line[(unsigned char) *input_line_pointer])
1092 {
1093 char *string = input_line_pointer;
1094 int e = get_symbol_end ();
1095
1096 if (strcmp (string, "prefix") == 0)
1097 ask_naked_reg = 1;
1098 else if (strcmp (string, "noprefix") == 0)
1099 ask_naked_reg = -1;
1100 else
1101 as_bad (_("bad argument to syntax directive."));
1102 *input_line_pointer = e;
1103 }
1104 demand_empty_rest_of_line ();
1105
1106 intel_syntax = syntax_flag;
1107
1108 if (ask_naked_reg == 0)
1109 allow_naked_reg = (intel_syntax
1110 && (bfd_get_symbol_leading_char (stdoutput) != '\0'));
1111 else
1112 allow_naked_reg = (ask_naked_reg < 0);
1113
1114 identifier_chars['%'] = intel_syntax && allow_naked_reg ? '%' : 0;
1115 identifier_chars['$'] = intel_syntax ? '$' : 0;
1116 register_prefix = allow_naked_reg ? "" : "%";
1117 }
1118
1119 static void
1120 set_cpu_arch (int dummy ATTRIBUTE_UNUSED)
1121 {
1122 SKIP_WHITESPACE ();
1123
1124 if (!is_end_of_line[(unsigned char) *input_line_pointer])
1125 {
1126 char *string = input_line_pointer;
1127 int e = get_symbol_end ();
1128 unsigned int i;
1129
1130 for (i = 0; i < ARRAY_SIZE (cpu_arch); i++)
1131 {
1132 if (strcmp (string, cpu_arch[i].name) == 0)
1133 {
1134 if (*string != '.')
1135 {
1136 cpu_arch_name = cpu_arch[i].name;
1137 cpu_sub_arch_name = NULL;
1138 cpu_arch_flags = (cpu_arch[i].flags
1139 | (flag_code == CODE_64BIT
1140 ? Cpu64 : CpuNo64));
1141 cpu_arch_isa = cpu_arch[i].type;
1142 cpu_arch_isa_flags = cpu_arch[i].flags;
1143 if (!cpu_arch_tune_set)
1144 {
1145 cpu_arch_tune = cpu_arch_isa;
1146 cpu_arch_tune_flags = cpu_arch_isa_flags;
1147 }
1148 break;
1149 }
1150 if ((cpu_arch_flags | cpu_arch[i].flags) != cpu_arch_flags)
1151 {
1152 cpu_sub_arch_name = cpu_arch[i].name;
1153 cpu_arch_flags |= cpu_arch[i].flags;
1154 }
1155 *input_line_pointer = e;
1156 demand_empty_rest_of_line ();
1157 return;
1158 }
1159 }
1160 if (i >= ARRAY_SIZE (cpu_arch))
1161 as_bad (_("no such architecture: `%s'"), string);
1162
1163 *input_line_pointer = e;
1164 }
1165 else
1166 as_bad (_("missing cpu architecture"));
1167
1168 no_cond_jump_promotion = 0;
1169 if (*input_line_pointer == ','
1170 && !is_end_of_line[(unsigned char) input_line_pointer[1]])
1171 {
1172 char *string = ++input_line_pointer;
1173 int e = get_symbol_end ();
1174
1175 if (strcmp (string, "nojumps") == 0)
1176 no_cond_jump_promotion = 1;
1177 else if (strcmp (string, "jumps") == 0)
1178 ;
1179 else
1180 as_bad (_("no such architecture modifier: `%s'"), string);
1181
1182 *input_line_pointer = e;
1183 }
1184
1185 demand_empty_rest_of_line ();
1186 }
1187
1188 unsigned long
1189 i386_mach ()
1190 {
1191 if (!strcmp (default_arch, "x86_64"))
1192 return bfd_mach_x86_64;
1193 else if (!strcmp (default_arch, "i386"))
1194 return bfd_mach_i386_i386;
1195 else
1196 as_fatal (_("Unknown architecture"));
1197 }
1198 \f
1199 void
1200 md_begin ()
1201 {
1202 const char *hash_err;
1203
1204 /* Initialize op_hash hash table. */
1205 op_hash = hash_new ();
1206
1207 {
1208 const template *optab;
1209 templates *core_optab;
1210
1211 /* Setup for loop. */
1212 optab = i386_optab;
1213 core_optab = (templates *) xmalloc (sizeof (templates));
1214 core_optab->start = optab;
1215
1216 while (1)
1217 {
1218 ++optab;
1219 if (optab->name == NULL
1220 || strcmp (optab->name, (optab - 1)->name) != 0)
1221 {
1222 /* different name --> ship out current template list;
1223 add to hash table; & begin anew. */
1224 core_optab->end = optab;
1225 hash_err = hash_insert (op_hash,
1226 (optab - 1)->name,
1227 (PTR) core_optab);
1228 if (hash_err)
1229 {
1230 as_fatal (_("Internal Error: Can't hash %s: %s"),
1231 (optab - 1)->name,
1232 hash_err);
1233 }
1234 if (optab->name == NULL)
1235 break;
1236 core_optab = (templates *) xmalloc (sizeof (templates));
1237 core_optab->start = optab;
1238 }
1239 }
1240 }
1241
1242 /* Initialize reg_hash hash table. */
1243 reg_hash = hash_new ();
1244 {
1245 const reg_entry *regtab;
1246 unsigned int regtab_size = i386_regtab_size;
1247
1248 for (regtab = i386_regtab; regtab_size--; regtab++)
1249 {
1250 hash_err = hash_insert (reg_hash, regtab->reg_name, (PTR) regtab);
1251 if (hash_err)
1252 as_fatal (_("Internal Error: Can't hash %s: %s"),
1253 regtab->reg_name,
1254 hash_err);
1255 }
1256 }
1257
1258 /* Fill in lexical tables: mnemonic_chars, operand_chars. */
1259 {
1260 int c;
1261 char *p;
1262
1263 for (c = 0; c < 256; c++)
1264 {
1265 if (ISDIGIT (c))
1266 {
1267 digit_chars[c] = c;
1268 mnemonic_chars[c] = c;
1269 register_chars[c] = c;
1270 operand_chars[c] = c;
1271 }
1272 else if (ISLOWER (c))
1273 {
1274 mnemonic_chars[c] = c;
1275 register_chars[c] = c;
1276 operand_chars[c] = c;
1277 }
1278 else if (ISUPPER (c))
1279 {
1280 mnemonic_chars[c] = TOLOWER (c);
1281 register_chars[c] = mnemonic_chars[c];
1282 operand_chars[c] = c;
1283 }
1284
1285 if (ISALPHA (c) || ISDIGIT (c))
1286 identifier_chars[c] = c;
1287 else if (c >= 128)
1288 {
1289 identifier_chars[c] = c;
1290 operand_chars[c] = c;
1291 }
1292 }
1293
1294 #ifdef LEX_AT
1295 identifier_chars['@'] = '@';
1296 #endif
1297 #ifdef LEX_QM
1298 identifier_chars['?'] = '?';
1299 operand_chars['?'] = '?';
1300 #endif
1301 digit_chars['-'] = '-';
1302 mnemonic_chars['-'] = '-';
1303 mnemonic_chars['.'] = '.';
1304 identifier_chars['_'] = '_';
1305 identifier_chars['.'] = '.';
1306
1307 for (p = operand_special_chars; *p != '\0'; p++)
1308 operand_chars[(unsigned char) *p] = *p;
1309 }
1310
1311 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1312 if (IS_ELF)
1313 {
1314 record_alignment (text_section, 2);
1315 record_alignment (data_section, 2);
1316 record_alignment (bss_section, 2);
1317 }
1318 #endif
1319
1320 if (flag_code == CODE_64BIT)
1321 {
1322 x86_dwarf2_return_column = 16;
1323 x86_cie_data_alignment = -8;
1324 }
1325 else
1326 {
1327 x86_dwarf2_return_column = 8;
1328 x86_cie_data_alignment = -4;
1329 }
1330 }
1331
1332 void
1333 i386_print_statistics (FILE *file)
1334 {
1335 hash_print_statistics (file, "i386 opcode", op_hash);
1336 hash_print_statistics (file, "i386 register", reg_hash);
1337 }
1338 \f
1339 #ifdef DEBUG386
1340
1341 /* Debugging routines for md_assemble. */
1342 static void pte (template *);
1343 static void pt (unsigned int);
1344 static void pe (expressionS *);
1345 static void ps (symbolS *);
1346
1347 static void
1348 pi (char *line, i386_insn *x)
1349 {
1350 unsigned int i;
1351
1352 fprintf (stdout, "%s: template ", line);
1353 pte (&x->tm);
1354 fprintf (stdout, " address: base %s index %s scale %x\n",
1355 x->base_reg ? x->base_reg->reg_name : "none",
1356 x->index_reg ? x->index_reg->reg_name : "none",
1357 x->log2_scale_factor);
1358 fprintf (stdout, " modrm: mode %x reg %x reg/mem %x\n",
1359 x->rm.mode, x->rm.reg, x->rm.regmem);
1360 fprintf (stdout, " sib: base %x index %x scale %x\n",
1361 x->sib.base, x->sib.index, x->sib.scale);
1362 fprintf (stdout, " rex: 64bit %x extX %x extY %x extZ %x\n",
1363 (x->rex & REX_W) != 0,
1364 (x->rex & REX_R) != 0,
1365 (x->rex & REX_X) != 0,
1366 (x->rex & REX_B) != 0);
1367 for (i = 0; i < x->operands; i++)
1368 {
1369 fprintf (stdout, " #%d: ", i + 1);
1370 pt (x->types[i]);
1371 fprintf (stdout, "\n");
1372 if (x->types[i]
1373 & (Reg | SReg2 | SReg3 | Control | Debug | Test | RegMMX | RegXMM))
1374 fprintf (stdout, "%s\n", x->op[i].regs->reg_name);
1375 if (x->types[i] & Imm)
1376 pe (x->op[i].imms);
1377 if (x->types[i] & Disp)
1378 pe (x->op[i].disps);
1379 }
1380 }
1381
1382 static void
1383 pte (template *t)
1384 {
1385 unsigned int i;
1386 fprintf (stdout, " %d operands ", t->operands);
1387 fprintf (stdout, "opcode %x ", t->base_opcode);
1388 if (t->extension_opcode != None)
1389 fprintf (stdout, "ext %x ", t->extension_opcode);
1390 if (t->opcode_modifier & D)
1391 fprintf (stdout, "D");
1392 if (t->opcode_modifier & W)
1393 fprintf (stdout, "W");
1394 fprintf (stdout, "\n");
1395 for (i = 0; i < t->operands; i++)
1396 {
1397 fprintf (stdout, " #%d type ", i + 1);
1398 pt (t->operand_types[i]);
1399 fprintf (stdout, "\n");
1400 }
1401 }
1402
1403 static void
1404 pe (expressionS *e)
1405 {
1406 fprintf (stdout, " operation %d\n", e->X_op);
1407 fprintf (stdout, " add_number %ld (%lx)\n",
1408 (long) e->X_add_number, (long) e->X_add_number);
1409 if (e->X_add_symbol)
1410 {
1411 fprintf (stdout, " add_symbol ");
1412 ps (e->X_add_symbol);
1413 fprintf (stdout, "\n");
1414 }
1415 if (e->X_op_symbol)
1416 {
1417 fprintf (stdout, " op_symbol ");
1418 ps (e->X_op_symbol);
1419 fprintf (stdout, "\n");
1420 }
1421 }
1422
1423 static void
1424 ps (symbolS *s)
1425 {
1426 fprintf (stdout, "%s type %s%s",
1427 S_GET_NAME (s),
1428 S_IS_EXTERNAL (s) ? "EXTERNAL " : "",
1429 segment_name (S_GET_SEGMENT (s)));
1430 }
1431
1432 static struct type_name
1433 {
1434 unsigned int mask;
1435 char *tname;
1436 }
1437 const type_names[] =
1438 {
1439 { Reg8, "r8" },
1440 { Reg16, "r16" },
1441 { Reg32, "r32" },
1442 { Reg64, "r64" },
1443 { Imm8, "i8" },
1444 { Imm8S, "i8s" },
1445 { Imm16, "i16" },
1446 { Imm32, "i32" },
1447 { Imm32S, "i32s" },
1448 { Imm64, "i64" },
1449 { Imm1, "i1" },
1450 { BaseIndex, "BaseIndex" },
1451 { Disp8, "d8" },
1452 { Disp16, "d16" },
1453 { Disp32, "d32" },
1454 { Disp32S, "d32s" },
1455 { Disp64, "d64" },
1456 { InOutPortReg, "InOutPortReg" },
1457 { ShiftCount, "ShiftCount" },
1458 { Control, "control reg" },
1459 { Test, "test reg" },
1460 { Debug, "debug reg" },
1461 { FloatReg, "FReg" },
1462 { FloatAcc, "FAcc" },
1463 { SReg2, "SReg2" },
1464 { SReg3, "SReg3" },
1465 { Acc, "Acc" },
1466 { JumpAbsolute, "Jump Absolute" },
1467 { RegMMX, "rMMX" },
1468 { RegXMM, "rXMM" },
1469 { EsSeg, "es" },
1470 { 0, "" }
1471 };
1472
1473 static void
1474 pt (t)
1475 unsigned int t;
1476 {
1477 const struct type_name *ty;
1478
1479 for (ty = type_names; ty->mask; ty++)
1480 if (t & ty->mask)
1481 fprintf (stdout, "%s, ", ty->tname);
1482 fflush (stdout);
1483 }
1484
1485 #endif /* DEBUG386 */
1486 \f
1487 static bfd_reloc_code_real_type
1488 reloc (unsigned int size,
1489 int pcrel,
1490 int sign,
1491 bfd_reloc_code_real_type other)
1492 {
1493 if (other != NO_RELOC)
1494 {
1495 reloc_howto_type *reloc;
1496
1497 if (size == 8)
1498 switch (other)
1499 {
1500 case BFD_RELOC_X86_64_GOT32:
1501 return BFD_RELOC_X86_64_GOT64;
1502 break;
1503 case BFD_RELOC_X86_64_PLTOFF64:
1504 return BFD_RELOC_X86_64_PLTOFF64;
1505 break;
1506 case BFD_RELOC_X86_64_GOTPC32:
1507 other = BFD_RELOC_X86_64_GOTPC64;
1508 break;
1509 case BFD_RELOC_X86_64_GOTPCREL:
1510 other = BFD_RELOC_X86_64_GOTPCREL64;
1511 break;
1512 case BFD_RELOC_X86_64_TPOFF32:
1513 other = BFD_RELOC_X86_64_TPOFF64;
1514 break;
1515 case BFD_RELOC_X86_64_DTPOFF32:
1516 other = BFD_RELOC_X86_64_DTPOFF64;
1517 break;
1518 default:
1519 break;
1520 }
1521
1522 /* Sign-checking 4-byte relocations in 16-/32-bit code is pointless. */
1523 if (size == 4 && flag_code != CODE_64BIT)
1524 sign = -1;
1525
1526 reloc = bfd_reloc_type_lookup (stdoutput, other);
1527 if (!reloc)
1528 as_bad (_("unknown relocation (%u)"), other);
1529 else if (size != bfd_get_reloc_size (reloc))
1530 as_bad (_("%u-byte relocation cannot be applied to %u-byte field"),
1531 bfd_get_reloc_size (reloc),
1532 size);
1533 else if (pcrel && !reloc->pc_relative)
1534 as_bad (_("non-pc-relative relocation for pc-relative field"));
1535 else if ((reloc->complain_on_overflow == complain_overflow_signed
1536 && !sign)
1537 || (reloc->complain_on_overflow == complain_overflow_unsigned
1538 && sign > 0))
1539 as_bad (_("relocated field and relocation type differ in signedness"));
1540 else
1541 return other;
1542 return NO_RELOC;
1543 }
1544
1545 if (pcrel)
1546 {
1547 if (!sign)
1548 as_bad (_("there are no unsigned pc-relative relocations"));
1549 switch (size)
1550 {
1551 case 1: return BFD_RELOC_8_PCREL;
1552 case 2: return BFD_RELOC_16_PCREL;
1553 case 4: return BFD_RELOC_32_PCREL;
1554 case 8: return BFD_RELOC_64_PCREL;
1555 }
1556 as_bad (_("cannot do %u byte pc-relative relocation"), size);
1557 }
1558 else
1559 {
1560 if (sign > 0)
1561 switch (size)
1562 {
1563 case 4: return BFD_RELOC_X86_64_32S;
1564 }
1565 else
1566 switch (size)
1567 {
1568 case 1: return BFD_RELOC_8;
1569 case 2: return BFD_RELOC_16;
1570 case 4: return BFD_RELOC_32;
1571 case 8: return BFD_RELOC_64;
1572 }
1573 as_bad (_("cannot do %s %u byte relocation"),
1574 sign > 0 ? "signed" : "unsigned", size);
1575 }
1576
1577 abort ();
1578 return BFD_RELOC_NONE;
1579 }
1580
1581 /* Here we decide which fixups can be adjusted to make them relative to
1582 the beginning of the section instead of the symbol. Basically we need
1583 to make sure that the dynamic relocations are done correctly, so in
1584 some cases we force the original symbol to be used. */
1585
1586 int
1587 tc_i386_fix_adjustable (fixS *fixP ATTRIBUTE_UNUSED)
1588 {
1589 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1590 if (!IS_ELF)
1591 return 1;
1592
1593 /* Don't adjust pc-relative references to merge sections in 64-bit
1594 mode. */
1595 if (use_rela_relocations
1596 && (S_GET_SEGMENT (fixP->fx_addsy)->flags & SEC_MERGE) != 0
1597 && fixP->fx_pcrel)
1598 return 0;
1599
1600 /* The x86_64 GOTPCREL are represented as 32bit PCrel relocations
1601 and changed later by validate_fix. */
1602 if (GOT_symbol && fixP->fx_subsy == GOT_symbol
1603 && fixP->fx_r_type == BFD_RELOC_32_PCREL)
1604 return 0;
1605
1606 /* adjust_reloc_syms doesn't know about the GOT. */
1607 if (fixP->fx_r_type == BFD_RELOC_386_GOTOFF
1608 || fixP->fx_r_type == BFD_RELOC_386_PLT32
1609 || fixP->fx_r_type == BFD_RELOC_386_GOT32
1610 || fixP->fx_r_type == BFD_RELOC_386_TLS_GD
1611 || fixP->fx_r_type == BFD_RELOC_386_TLS_LDM
1612 || fixP->fx_r_type == BFD_RELOC_386_TLS_LDO_32
1613 || fixP->fx_r_type == BFD_RELOC_386_TLS_IE_32
1614 || fixP->fx_r_type == BFD_RELOC_386_TLS_IE
1615 || fixP->fx_r_type == BFD_RELOC_386_TLS_GOTIE
1616 || fixP->fx_r_type == BFD_RELOC_386_TLS_LE_32
1617 || fixP->fx_r_type == BFD_RELOC_386_TLS_LE
1618 || fixP->fx_r_type == BFD_RELOC_386_TLS_GOTDESC
1619 || fixP->fx_r_type == BFD_RELOC_386_TLS_DESC_CALL
1620 || fixP->fx_r_type == BFD_RELOC_X86_64_PLT32
1621 || fixP->fx_r_type == BFD_RELOC_X86_64_GOT32
1622 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPCREL
1623 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSGD
1624 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSLD
1625 || fixP->fx_r_type == BFD_RELOC_X86_64_DTPOFF32
1626 || fixP->fx_r_type == BFD_RELOC_X86_64_DTPOFF64
1627 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTTPOFF
1628 || fixP->fx_r_type == BFD_RELOC_X86_64_TPOFF32
1629 || fixP->fx_r_type == BFD_RELOC_X86_64_TPOFF64
1630 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTOFF64
1631 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPC32_TLSDESC
1632 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSDESC_CALL
1633 || fixP->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1634 || fixP->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1635 return 0;
1636 #endif
1637 return 1;
1638 }
1639
1640 static int
1641 intel_float_operand (const char *mnemonic)
1642 {
1643 /* Note that the value returned is meaningful only for opcodes with (memory)
1644 operands, hence the code here is free to improperly handle opcodes that
1645 have no operands (for better performance and smaller code). */
1646
1647 if (mnemonic[0] != 'f')
1648 return 0; /* non-math */
1649
1650 switch (mnemonic[1])
1651 {
1652 /* fclex, fdecstp, fdisi, femms, feni, fincstp, finit, fsetpm, and
1653 the fs segment override prefix not currently handled because no
1654 call path can make opcodes without operands get here */
1655 case 'i':
1656 return 2 /* integer op */;
1657 case 'l':
1658 if (mnemonic[2] == 'd' && (mnemonic[3] == 'c' || mnemonic[3] == 'e'))
1659 return 3; /* fldcw/fldenv */
1660 break;
1661 case 'n':
1662 if (mnemonic[2] != 'o' /* fnop */)
1663 return 3; /* non-waiting control op */
1664 break;
1665 case 'r':
1666 if (mnemonic[2] == 's')
1667 return 3; /* frstor/frstpm */
1668 break;
1669 case 's':
1670 if (mnemonic[2] == 'a')
1671 return 3; /* fsave */
1672 if (mnemonic[2] == 't')
1673 {
1674 switch (mnemonic[3])
1675 {
1676 case 'c': /* fstcw */
1677 case 'd': /* fstdw */
1678 case 'e': /* fstenv */
1679 case 's': /* fsts[gw] */
1680 return 3;
1681 }
1682 }
1683 break;
1684 case 'x':
1685 if (mnemonic[2] == 'r' || mnemonic[2] == 's')
1686 return 0; /* fxsave/fxrstor are not really math ops */
1687 break;
1688 }
1689
1690 return 1;
1691 }
1692
1693 /* This is the guts of the machine-dependent assembler. LINE points to a
1694 machine dependent instruction. This function is supposed to emit
1695 the frags/bytes it assembles to. */
1696
1697 void
1698 md_assemble (line)
1699 char *line;
1700 {
1701 int j;
1702 char mnemonic[MAX_MNEM_SIZE];
1703
1704 /* Initialize globals. */
1705 memset (&i, '\0', sizeof (i));
1706 for (j = 0; j < MAX_OPERANDS; j++)
1707 i.reloc[j] = NO_RELOC;
1708 memset (disp_expressions, '\0', sizeof (disp_expressions));
1709 memset (im_expressions, '\0', sizeof (im_expressions));
1710 save_stack_p = save_stack;
1711
1712 /* First parse an instruction mnemonic & call i386_operand for the operands.
1713 We assume that the scrubber has arranged it so that line[0] is the valid
1714 start of a (possibly prefixed) mnemonic. */
1715
1716 line = parse_insn (line, mnemonic);
1717 if (line == NULL)
1718 return;
1719
1720 line = parse_operands (line, mnemonic);
1721 if (line == NULL)
1722 return;
1723
1724 /* The order of the immediates should be reversed
1725 for 2 immediates extrq and insertq instructions */
1726 if ((i.imm_operands == 2)
1727 && ((strcmp (mnemonic, "extrq") == 0)
1728 || (strcmp (mnemonic, "insertq") == 0)))
1729 {
1730 swap_2_operands (0, 1);
1731 /* "extrq" and insertq" are the only two instructions whose operands
1732 have to be reversed even though they have two immediate operands.
1733 */
1734 if (intel_syntax)
1735 swap_operands ();
1736 }
1737
1738 /* Now we've parsed the mnemonic into a set of templates, and have the
1739 operands at hand. */
1740
1741 /* All intel opcodes have reversed operands except for "bound" and
1742 "enter". We also don't reverse intersegment "jmp" and "call"
1743 instructions with 2 immediate operands so that the immediate segment
1744 precedes the offset, as it does when in AT&T mode. */
1745 if (intel_syntax
1746 && i.operands > 1
1747 && (strcmp (mnemonic, "bound") != 0)
1748 && (strcmp (mnemonic, "invlpga") != 0)
1749 && !((i.types[0] & Imm) && (i.types[1] & Imm)))
1750 swap_operands ();
1751
1752 if (i.imm_operands)
1753 optimize_imm ();
1754
1755 /* Don't optimize displacement for movabs since it only takes 64bit
1756 displacement. */
1757 if (i.disp_operands
1758 && (flag_code != CODE_64BIT
1759 || strcmp (mnemonic, "movabs") != 0))
1760 optimize_disp ();
1761
1762 /* Next, we find a template that matches the given insn,
1763 making sure the overlap of the given operands types is consistent
1764 with the template operand types. */
1765
1766 if (!match_template ())
1767 return;
1768
1769 if (intel_syntax)
1770 {
1771 /* Undo SYSV386_COMPAT brokenness when in Intel mode. See i386.h */
1772 if (SYSV386_COMPAT
1773 && (i.tm.base_opcode & 0xfffffde0) == 0xdce0)
1774 i.tm.base_opcode ^= Opcode_FloatR;
1775
1776 /* Zap movzx and movsx suffix. The suffix may have been set from
1777 "word ptr" or "byte ptr" on the source operand, but we'll use
1778 the suffix later to choose the destination register. */
1779 if ((i.tm.base_opcode & ~9) == 0x0fb6)
1780 {
1781 if (i.reg_operands < 2
1782 && !i.suffix
1783 && (~i.tm.opcode_modifier
1784 & (No_bSuf
1785 | No_wSuf
1786 | No_lSuf
1787 | No_sSuf
1788 | No_xSuf
1789 | No_qSuf)))
1790 as_bad (_("ambiguous operand size for `%s'"), i.tm.name);
1791
1792 i.suffix = 0;
1793 }
1794 }
1795
1796 if (i.tm.opcode_modifier & FWait)
1797 if (!add_prefix (FWAIT_OPCODE))
1798 return;
1799
1800 /* Check string instruction segment overrides. */
1801 if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0)
1802 {
1803 if (!check_string ())
1804 return;
1805 }
1806
1807 if (!process_suffix ())
1808 return;
1809
1810 /* Make still unresolved immediate matches conform to size of immediate
1811 given in i.suffix. */
1812 if (!finalize_imm ())
1813 return;
1814
1815 if (i.types[0] & Imm1)
1816 i.imm_operands = 0; /* kludge for shift insns. */
1817 if (i.types[0] & ImplicitRegister)
1818 i.reg_operands--;
1819 if (i.types[1] & ImplicitRegister)
1820 i.reg_operands--;
1821 if (i.types[2] & ImplicitRegister)
1822 i.reg_operands--;
1823
1824 if (i.tm.opcode_modifier & ImmExt)
1825 {
1826 expressionS *exp;
1827
1828 if ((i.tm.cpu_flags & CpuSSE3) && i.operands > 0)
1829 {
1830 /* Streaming SIMD extensions 3 Instructions have the fixed
1831 operands with an opcode suffix which is coded in the same
1832 place as an 8-bit immediate field would be. Here we check
1833 those operands and remove them afterwards. */
1834 unsigned int x;
1835
1836 for (x = 0; x < i.operands; x++)
1837 if (i.op[x].regs->reg_num != x)
1838 as_bad (_("can't use register '%s%s' as operand %d in '%s'."),
1839 register_prefix,
1840 i.op[x].regs->reg_name,
1841 x + 1,
1842 i.tm.name);
1843 i.operands = 0;
1844 }
1845
1846 /* These AMD 3DNow! and Intel Katmai New Instructions have an
1847 opcode suffix which is coded in the same place as an 8-bit
1848 immediate field would be. Here we fake an 8-bit immediate
1849 operand from the opcode suffix stored in tm.extension_opcode. */
1850
1851 assert (i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS);
1852
1853 exp = &im_expressions[i.imm_operands++];
1854 i.op[i.operands].imms = exp;
1855 i.types[i.operands++] = Imm8;
1856 exp->X_op = O_constant;
1857 exp->X_add_number = i.tm.extension_opcode;
1858 i.tm.extension_opcode = None;
1859 }
1860
1861 /* For insns with operands there are more diddles to do to the opcode. */
1862 if (i.operands)
1863 {
1864 if (!process_operands ())
1865 return;
1866 }
1867 else if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
1868 {
1869 /* UnixWare fsub no args is alias for fsubp, fadd -> faddp, etc. */
1870 as_warn (_("translating to `%sp'"), i.tm.name);
1871 }
1872
1873 /* Handle conversion of 'int $3' --> special int3 insn. */
1874 if (i.tm.base_opcode == INT_OPCODE && i.op[0].imms->X_add_number == 3)
1875 {
1876 i.tm.base_opcode = INT3_OPCODE;
1877 i.imm_operands = 0;
1878 }
1879
1880 if ((i.tm.opcode_modifier & (Jump | JumpByte | JumpDword))
1881 && i.op[0].disps->X_op == O_constant)
1882 {
1883 /* Convert "jmp constant" (and "call constant") to a jump (call) to
1884 the absolute address given by the constant. Since ix86 jumps and
1885 calls are pc relative, we need to generate a reloc. */
1886 i.op[0].disps->X_add_symbol = &abs_symbol;
1887 i.op[0].disps->X_op = O_symbol;
1888 }
1889
1890 if ((i.tm.opcode_modifier & Rex64) != 0)
1891 i.rex |= REX_W;
1892
1893 /* For 8 bit registers we need an empty rex prefix. Also if the
1894 instruction already has a prefix, we need to convert old
1895 registers to new ones. */
1896
1897 if (((i.types[0] & Reg8) != 0
1898 && (i.op[0].regs->reg_flags & RegRex64) != 0)
1899 || ((i.types[1] & Reg8) != 0
1900 && (i.op[1].regs->reg_flags & RegRex64) != 0)
1901 || (((i.types[0] & Reg8) != 0 || (i.types[1] & Reg8) != 0)
1902 && i.rex != 0))
1903 {
1904 int x;
1905
1906 i.rex |= REX_OPCODE;
1907 for (x = 0; x < 2; x++)
1908 {
1909 /* Look for 8 bit operand that uses old registers. */
1910 if ((i.types[x] & Reg8) != 0
1911 && (i.op[x].regs->reg_flags & RegRex64) == 0)
1912 {
1913 /* In case it is "hi" register, give up. */
1914 if (i.op[x].regs->reg_num > 3)
1915 as_bad (_("can't encode register '%s%s' in an "
1916 "instruction requiring REX prefix."),
1917 register_prefix, i.op[x].regs->reg_name);
1918
1919 /* Otherwise it is equivalent to the extended register.
1920 Since the encoding doesn't change this is merely
1921 cosmetic cleanup for debug output. */
1922
1923 i.op[x].regs = i.op[x].regs + 8;
1924 }
1925 }
1926 }
1927
1928 if (i.rex != 0)
1929 add_prefix (REX_OPCODE | i.rex);
1930
1931 /* We are ready to output the insn. */
1932 output_insn ();
1933 }
1934
1935 static char *
1936 parse_insn (char *line, char *mnemonic)
1937 {
1938 char *l = line;
1939 char *token_start = l;
1940 char *mnem_p;
1941 int supported;
1942 const template *t;
1943
1944 /* Non-zero if we found a prefix only acceptable with string insns. */
1945 const char *expecting_string_instruction = NULL;
1946
1947 while (1)
1948 {
1949 mnem_p = mnemonic;
1950 while ((*mnem_p = mnemonic_chars[(unsigned char) *l]) != 0)
1951 {
1952 mnem_p++;
1953 if (mnem_p >= mnemonic + MAX_MNEM_SIZE)
1954 {
1955 as_bad (_("no such instruction: `%s'"), token_start);
1956 return NULL;
1957 }
1958 l++;
1959 }
1960 if (!is_space_char (*l)
1961 && *l != END_OF_INSN
1962 && (intel_syntax
1963 || (*l != PREFIX_SEPARATOR
1964 && *l != ',')))
1965 {
1966 as_bad (_("invalid character %s in mnemonic"),
1967 output_invalid (*l));
1968 return NULL;
1969 }
1970 if (token_start == l)
1971 {
1972 if (!intel_syntax && *l == PREFIX_SEPARATOR)
1973 as_bad (_("expecting prefix; got nothing"));
1974 else
1975 as_bad (_("expecting mnemonic; got nothing"));
1976 return NULL;
1977 }
1978
1979 /* Look up instruction (or prefix) via hash table. */
1980 current_templates = hash_find (op_hash, mnemonic);
1981
1982 if (*l != END_OF_INSN
1983 && (!is_space_char (*l) || l[1] != END_OF_INSN)
1984 && current_templates
1985 && (current_templates->start->opcode_modifier & IsPrefix))
1986 {
1987 if (current_templates->start->cpu_flags
1988 & (flag_code != CODE_64BIT ? Cpu64 : CpuNo64))
1989 {
1990 as_bad ((flag_code != CODE_64BIT
1991 ? _("`%s' is only supported in 64-bit mode")
1992 : _("`%s' is not supported in 64-bit mode")),
1993 current_templates->start->name);
1994 return NULL;
1995 }
1996 /* If we are in 16-bit mode, do not allow addr16 or data16.
1997 Similarly, in 32-bit mode, do not allow addr32 or data32. */
1998 if ((current_templates->start->opcode_modifier & (Size16 | Size32))
1999 && flag_code != CODE_64BIT
2000 && (((current_templates->start->opcode_modifier & Size32) != 0)
2001 ^ (flag_code == CODE_16BIT)))
2002 {
2003 as_bad (_("redundant %s prefix"),
2004 current_templates->start->name);
2005 return NULL;
2006 }
2007 /* Add prefix, checking for repeated prefixes. */
2008 switch (add_prefix (current_templates->start->base_opcode))
2009 {
2010 case 0:
2011 return NULL;
2012 case 2:
2013 expecting_string_instruction = current_templates->start->name;
2014 break;
2015 }
2016 /* Skip past PREFIX_SEPARATOR and reset token_start. */
2017 token_start = ++l;
2018 }
2019 else
2020 break;
2021 }
2022
2023 if (!current_templates)
2024 {
2025 /* See if we can get a match by trimming off a suffix. */
2026 switch (mnem_p[-1])
2027 {
2028 case WORD_MNEM_SUFFIX:
2029 if (intel_syntax && (intel_float_operand (mnemonic) & 2))
2030 i.suffix = SHORT_MNEM_SUFFIX;
2031 else
2032 case BYTE_MNEM_SUFFIX:
2033 case QWORD_MNEM_SUFFIX:
2034 i.suffix = mnem_p[-1];
2035 mnem_p[-1] = '\0';
2036 current_templates = hash_find (op_hash, mnemonic);
2037 break;
2038 case SHORT_MNEM_SUFFIX:
2039 case LONG_MNEM_SUFFIX:
2040 if (!intel_syntax)
2041 {
2042 i.suffix = mnem_p[-1];
2043 mnem_p[-1] = '\0';
2044 current_templates = hash_find (op_hash, mnemonic);
2045 }
2046 break;
2047
2048 /* Intel Syntax. */
2049 case 'd':
2050 if (intel_syntax)
2051 {
2052 if (intel_float_operand (mnemonic) == 1)
2053 i.suffix = SHORT_MNEM_SUFFIX;
2054 else
2055 i.suffix = LONG_MNEM_SUFFIX;
2056 mnem_p[-1] = '\0';
2057 current_templates = hash_find (op_hash, mnemonic);
2058 }
2059 break;
2060 }
2061 if (!current_templates)
2062 {
2063 as_bad (_("no such instruction: `%s'"), token_start);
2064 return NULL;
2065 }
2066 }
2067
2068 if (current_templates->start->opcode_modifier & (Jump | JumpByte))
2069 {
2070 /* Check for a branch hint. We allow ",pt" and ",pn" for
2071 predict taken and predict not taken respectively.
2072 I'm not sure that branch hints actually do anything on loop
2073 and jcxz insns (JumpByte) for current Pentium4 chips. They
2074 may work in the future and it doesn't hurt to accept them
2075 now. */
2076 if (l[0] == ',' && l[1] == 'p')
2077 {
2078 if (l[2] == 't')
2079 {
2080 if (!add_prefix (DS_PREFIX_OPCODE))
2081 return NULL;
2082 l += 3;
2083 }
2084 else if (l[2] == 'n')
2085 {
2086 if (!add_prefix (CS_PREFIX_OPCODE))
2087 return NULL;
2088 l += 3;
2089 }
2090 }
2091 }
2092 /* Any other comma loses. */
2093 if (*l == ',')
2094 {
2095 as_bad (_("invalid character %s in mnemonic"),
2096 output_invalid (*l));
2097 return NULL;
2098 }
2099
2100 /* Check if instruction is supported on specified architecture. */
2101 supported = 0;
2102 for (t = current_templates->start; t < current_templates->end; ++t)
2103 {
2104 if (!((t->cpu_flags & ~(Cpu64 | CpuNo64))
2105 & ~(cpu_arch_flags & ~(Cpu64 | CpuNo64))))
2106 supported |= 1;
2107 if (!(t->cpu_flags & (flag_code == CODE_64BIT ? CpuNo64 : Cpu64)))
2108 supported |= 2;
2109 }
2110 if (!(supported & 2))
2111 {
2112 as_bad (flag_code == CODE_64BIT
2113 ? _("`%s' is not supported in 64-bit mode")
2114 : _("`%s' is only supported in 64-bit mode"),
2115 current_templates->start->name);
2116 return NULL;
2117 }
2118 if (!(supported & 1))
2119 {
2120 as_warn (_("`%s' is not supported on `%s%s'"),
2121 current_templates->start->name,
2122 cpu_arch_name,
2123 cpu_sub_arch_name ? cpu_sub_arch_name : "");
2124 }
2125 else if ((Cpu386 & ~cpu_arch_flags) && (flag_code != CODE_16BIT))
2126 {
2127 as_warn (_("use .code16 to ensure correct addressing mode"));
2128 }
2129
2130 /* Check for rep/repne without a string instruction. */
2131 if (expecting_string_instruction)
2132 {
2133 static templates override;
2134
2135 for (t = current_templates->start; t < current_templates->end; ++t)
2136 if (t->opcode_modifier & IsString)
2137 break;
2138 if (t >= current_templates->end)
2139 {
2140 as_bad (_("expecting string instruction after `%s'"),
2141 expecting_string_instruction);
2142 return NULL;
2143 }
2144 for (override.start = t; t < current_templates->end; ++t)
2145 if (!(t->opcode_modifier & IsString))
2146 break;
2147 override.end = t;
2148 current_templates = &override;
2149 }
2150
2151 return l;
2152 }
2153
2154 static char *
2155 parse_operands (char *l, const char *mnemonic)
2156 {
2157 char *token_start;
2158
2159 /* 1 if operand is pending after ','. */
2160 unsigned int expecting_operand = 0;
2161
2162 /* Non-zero if operand parens not balanced. */
2163 unsigned int paren_not_balanced;
2164
2165 while (*l != END_OF_INSN)
2166 {
2167 /* Skip optional white space before operand. */
2168 if (is_space_char (*l))
2169 ++l;
2170 if (!is_operand_char (*l) && *l != END_OF_INSN)
2171 {
2172 as_bad (_("invalid character %s before operand %d"),
2173 output_invalid (*l),
2174 i.operands + 1);
2175 return NULL;
2176 }
2177 token_start = l; /* after white space */
2178 paren_not_balanced = 0;
2179 while (paren_not_balanced || *l != ',')
2180 {
2181 if (*l == END_OF_INSN)
2182 {
2183 if (paren_not_balanced)
2184 {
2185 if (!intel_syntax)
2186 as_bad (_("unbalanced parenthesis in operand %d."),
2187 i.operands + 1);
2188 else
2189 as_bad (_("unbalanced brackets in operand %d."),
2190 i.operands + 1);
2191 return NULL;
2192 }
2193 else
2194 break; /* we are done */
2195 }
2196 else if (!is_operand_char (*l) && !is_space_char (*l))
2197 {
2198 as_bad (_("invalid character %s in operand %d"),
2199 output_invalid (*l),
2200 i.operands + 1);
2201 return NULL;
2202 }
2203 if (!intel_syntax)
2204 {
2205 if (*l == '(')
2206 ++paren_not_balanced;
2207 if (*l == ')')
2208 --paren_not_balanced;
2209 }
2210 else
2211 {
2212 if (*l == '[')
2213 ++paren_not_balanced;
2214 if (*l == ']')
2215 --paren_not_balanced;
2216 }
2217 l++;
2218 }
2219 if (l != token_start)
2220 { /* Yes, we've read in another operand. */
2221 unsigned int operand_ok;
2222 this_operand = i.operands++;
2223 if (i.operands > MAX_OPERANDS)
2224 {
2225 as_bad (_("spurious operands; (%d operands/instruction max)"),
2226 MAX_OPERANDS);
2227 return NULL;
2228 }
2229 /* Now parse operand adding info to 'i' as we go along. */
2230 END_STRING_AND_SAVE (l);
2231
2232 if (intel_syntax)
2233 operand_ok =
2234 i386_intel_operand (token_start,
2235 intel_float_operand (mnemonic));
2236 else
2237 operand_ok = i386_operand (token_start);
2238
2239 RESTORE_END_STRING (l);
2240 if (!operand_ok)
2241 return NULL;
2242 }
2243 else
2244 {
2245 if (expecting_operand)
2246 {
2247 expecting_operand_after_comma:
2248 as_bad (_("expecting operand after ','; got nothing"));
2249 return NULL;
2250 }
2251 if (*l == ',')
2252 {
2253 as_bad (_("expecting operand before ','; got nothing"));
2254 return NULL;
2255 }
2256 }
2257
2258 /* Now *l must be either ',' or END_OF_INSN. */
2259 if (*l == ',')
2260 {
2261 if (*++l == END_OF_INSN)
2262 {
2263 /* Just skip it, if it's \n complain. */
2264 goto expecting_operand_after_comma;
2265 }
2266 expecting_operand = 1;
2267 }
2268 }
2269 return l;
2270 }
2271
2272 static void
2273 swap_2_operands (int xchg1, int xchg2)
2274 {
2275 union i386_op temp_op;
2276 unsigned int temp_type;
2277 enum bfd_reloc_code_real temp_reloc;
2278
2279 temp_type = i.types[xchg2];
2280 i.types[xchg2] = i.types[xchg1];
2281 i.types[xchg1] = temp_type;
2282 temp_op = i.op[xchg2];
2283 i.op[xchg2] = i.op[xchg1];
2284 i.op[xchg1] = temp_op;
2285 temp_reloc = i.reloc[xchg2];
2286 i.reloc[xchg2] = i.reloc[xchg1];
2287 i.reloc[xchg1] = temp_reloc;
2288 }
2289
2290 static void
2291 swap_operands (void)
2292 {
2293 switch (i.operands)
2294 {
2295 case 4:
2296 swap_2_operands (1, i.operands - 2);
2297 case 3:
2298 case 2:
2299 swap_2_operands (0, i.operands - 1);
2300 break;
2301 default:
2302 abort ();
2303 }
2304
2305 if (i.mem_operands == 2)
2306 {
2307 const seg_entry *temp_seg;
2308 temp_seg = i.seg[0];
2309 i.seg[0] = i.seg[1];
2310 i.seg[1] = temp_seg;
2311 }
2312 }
2313
2314 /* Try to ensure constant immediates are represented in the smallest
2315 opcode possible. */
2316 static void
2317 optimize_imm (void)
2318 {
2319 char guess_suffix = 0;
2320 int op;
2321
2322 if (i.suffix)
2323 guess_suffix = i.suffix;
2324 else if (i.reg_operands)
2325 {
2326 /* Figure out a suffix from the last register operand specified.
2327 We can't do this properly yet, ie. excluding InOutPortReg,
2328 but the following works for instructions with immediates.
2329 In any case, we can't set i.suffix yet. */
2330 for (op = i.operands; --op >= 0;)
2331 if (i.types[op] & Reg)
2332 {
2333 if (i.types[op] & Reg8)
2334 guess_suffix = BYTE_MNEM_SUFFIX;
2335 else if (i.types[op] & Reg16)
2336 guess_suffix = WORD_MNEM_SUFFIX;
2337 else if (i.types[op] & Reg32)
2338 guess_suffix = LONG_MNEM_SUFFIX;
2339 else if (i.types[op] & Reg64)
2340 guess_suffix = QWORD_MNEM_SUFFIX;
2341 break;
2342 }
2343 }
2344 else if ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0))
2345 guess_suffix = WORD_MNEM_SUFFIX;
2346
2347 for (op = i.operands; --op >= 0;)
2348 if (i.types[op] & Imm)
2349 {
2350 switch (i.op[op].imms->X_op)
2351 {
2352 case O_constant:
2353 /* If a suffix is given, this operand may be shortened. */
2354 switch (guess_suffix)
2355 {
2356 case LONG_MNEM_SUFFIX:
2357 i.types[op] |= Imm32 | Imm64;
2358 break;
2359 case WORD_MNEM_SUFFIX:
2360 i.types[op] |= Imm16 | Imm32S | Imm32 | Imm64;
2361 break;
2362 case BYTE_MNEM_SUFFIX:
2363 i.types[op] |= Imm16 | Imm8 | Imm8S | Imm32S | Imm32 | Imm64;
2364 break;
2365 }
2366
2367 /* If this operand is at most 16 bits, convert it
2368 to a signed 16 bit number before trying to see
2369 whether it will fit in an even smaller size.
2370 This allows a 16-bit operand such as $0xffe0 to
2371 be recognised as within Imm8S range. */
2372 if ((i.types[op] & Imm16)
2373 && (i.op[op].imms->X_add_number & ~(offsetT) 0xffff) == 0)
2374 {
2375 i.op[op].imms->X_add_number =
2376 (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000);
2377 }
2378 if ((i.types[op] & Imm32)
2379 && ((i.op[op].imms->X_add_number & ~(((offsetT) 2 << 31) - 1))
2380 == 0))
2381 {
2382 i.op[op].imms->X_add_number = ((i.op[op].imms->X_add_number
2383 ^ ((offsetT) 1 << 31))
2384 - ((offsetT) 1 << 31));
2385 }
2386 i.types[op] |= smallest_imm_type (i.op[op].imms->X_add_number);
2387
2388 /* We must avoid matching of Imm32 templates when 64bit
2389 only immediate is available. */
2390 if (guess_suffix == QWORD_MNEM_SUFFIX)
2391 i.types[op] &= ~Imm32;
2392 break;
2393
2394 case O_absent:
2395 case O_register:
2396 abort ();
2397
2398 /* Symbols and expressions. */
2399 default:
2400 /* Convert symbolic operand to proper sizes for matching, but don't
2401 prevent matching a set of insns that only supports sizes other
2402 than those matching the insn suffix. */
2403 {
2404 unsigned int mask, allowed = 0;
2405 const template *t;
2406
2407 for (t = current_templates->start;
2408 t < current_templates->end;
2409 ++t)
2410 allowed |= t->operand_types[op];
2411 switch (guess_suffix)
2412 {
2413 case QWORD_MNEM_SUFFIX:
2414 mask = Imm64 | Imm32S;
2415 break;
2416 case LONG_MNEM_SUFFIX:
2417 mask = Imm32;
2418 break;
2419 case WORD_MNEM_SUFFIX:
2420 mask = Imm16;
2421 break;
2422 case BYTE_MNEM_SUFFIX:
2423 mask = Imm8;
2424 break;
2425 default:
2426 mask = 0;
2427 break;
2428 }
2429 if (mask & allowed)
2430 i.types[op] &= mask;
2431 }
2432 break;
2433 }
2434 }
2435 }
2436
2437 /* Try to use the smallest displacement type too. */
2438 static void
2439 optimize_disp (void)
2440 {
2441 int op;
2442
2443 for (op = i.operands; --op >= 0;)
2444 if (i.types[op] & Disp)
2445 {
2446 if (i.op[op].disps->X_op == O_constant)
2447 {
2448 offsetT disp = i.op[op].disps->X_add_number;
2449
2450 if ((i.types[op] & Disp16)
2451 && (disp & ~(offsetT) 0xffff) == 0)
2452 {
2453 /* If this operand is at most 16 bits, convert
2454 to a signed 16 bit number and don't use 64bit
2455 displacement. */
2456 disp = (((disp & 0xffff) ^ 0x8000) - 0x8000);
2457 i.types[op] &= ~Disp64;
2458 }
2459 if ((i.types[op] & Disp32)
2460 && (disp & ~(((offsetT) 2 << 31) - 1)) == 0)
2461 {
2462 /* If this operand is at most 32 bits, convert
2463 to a signed 32 bit number and don't use 64bit
2464 displacement. */
2465 disp &= (((offsetT) 2 << 31) - 1);
2466 disp = (disp ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
2467 i.types[op] &= ~Disp64;
2468 }
2469 if (!disp && (i.types[op] & BaseIndex))
2470 {
2471 i.types[op] &= ~Disp;
2472 i.op[op].disps = 0;
2473 i.disp_operands--;
2474 }
2475 else if (flag_code == CODE_64BIT)
2476 {
2477 if (fits_in_signed_long (disp))
2478 {
2479 i.types[op] &= ~Disp64;
2480 i.types[op] |= Disp32S;
2481 }
2482 if (fits_in_unsigned_long (disp))
2483 i.types[op] |= Disp32;
2484 }
2485 if ((i.types[op] & (Disp32 | Disp32S | Disp16))
2486 && fits_in_signed_byte (disp))
2487 i.types[op] |= Disp8;
2488 }
2489 else if (i.reloc[op] == BFD_RELOC_386_TLS_DESC_CALL
2490 || i.reloc[op] == BFD_RELOC_X86_64_TLSDESC_CALL)
2491 {
2492 fix_new_exp (frag_now, frag_more (0) - frag_now->fr_literal, 0,
2493 i.op[op].disps, 0, i.reloc[op]);
2494 i.types[op] &= ~Disp;
2495 }
2496 else
2497 /* We only support 64bit displacement on constants. */
2498 i.types[op] &= ~Disp64;
2499 }
2500 }
2501
2502 static int
2503 match_template (void)
2504 {
2505 /* Points to template once we've found it. */
2506 const template *t;
2507 unsigned int overlap0, overlap1, overlap2, overlap3;
2508 unsigned int found_reverse_match;
2509 int suffix_check;
2510 unsigned int operand_types [MAX_OPERANDS];
2511 int addr_prefix_disp;
2512 unsigned int j;
2513
2514 #if MAX_OPERANDS != 4
2515 # error "MAX_OPERANDS must be 4."
2516 #endif
2517
2518 #define MATCH(overlap, given, template) \
2519 ((overlap & ~JumpAbsolute) \
2520 && (((given) & (BaseIndex | JumpAbsolute)) \
2521 == ((overlap) & (BaseIndex | JumpAbsolute))))
2522
2523 /* If given types r0 and r1 are registers they must be of the same type
2524 unless the expected operand type register overlap is null.
2525 Note that Acc in a template matches every size of reg. */
2526 #define CONSISTENT_REGISTER_MATCH(m0, g0, t0, m1, g1, t1) \
2527 (((g0) & Reg) == 0 || ((g1) & Reg) == 0 \
2528 || ((g0) & Reg) == ((g1) & Reg) \
2529 || ((((m0) & Acc) ? Reg : (t0)) & (((m1) & Acc) ? Reg : (t1)) & Reg) == 0 )
2530
2531 overlap0 = 0;
2532 overlap1 = 0;
2533 overlap2 = 0;
2534 overlap3 = 0;
2535 found_reverse_match = 0;
2536 for (j = 0; j < MAX_OPERANDS; j++)
2537 operand_types [j] = 0;
2538 addr_prefix_disp = -1;
2539 suffix_check = (i.suffix == BYTE_MNEM_SUFFIX
2540 ? No_bSuf
2541 : (i.suffix == WORD_MNEM_SUFFIX
2542 ? No_wSuf
2543 : (i.suffix == SHORT_MNEM_SUFFIX
2544 ? No_sSuf
2545 : (i.suffix == LONG_MNEM_SUFFIX
2546 ? No_lSuf
2547 : (i.suffix == QWORD_MNEM_SUFFIX
2548 ? No_qSuf
2549 : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX
2550 ? No_xSuf : 0))))));
2551
2552 for (t = current_templates->start; t < current_templates->end; t++)
2553 {
2554 addr_prefix_disp = -1;
2555
2556 /* Must have right number of operands. */
2557 if (i.operands != t->operands)
2558 continue;
2559
2560 /* Check the suffix, except for some instructions in intel mode. */
2561 if ((t->opcode_modifier & suffix_check)
2562 && !(intel_syntax
2563 && (t->opcode_modifier & IgnoreSize)))
2564 continue;
2565
2566 for (j = 0; j < MAX_OPERANDS; j++)
2567 operand_types [j] = t->operand_types [j];
2568
2569 /* In general, don't allow 64-bit operands in 32-bit mode. */
2570 if (i.suffix == QWORD_MNEM_SUFFIX
2571 && flag_code != CODE_64BIT
2572 && (intel_syntax
2573 ? (!(t->opcode_modifier & IgnoreSize)
2574 && !intel_float_operand (t->name))
2575 : intel_float_operand (t->name) != 2)
2576 && (!(operand_types[0] & (RegMMX | RegXMM))
2577 || !(operand_types[t->operands > 1] & (RegMMX | RegXMM)))
2578 && (t->base_opcode != 0x0fc7
2579 || t->extension_opcode != 1 /* cmpxchg8b */))
2580 continue;
2581
2582 /* Do not verify operands when there are none. */
2583 else if (!t->operands)
2584 {
2585 if (t->cpu_flags & ~cpu_arch_flags)
2586 continue;
2587 /* We've found a match; break out of loop. */
2588 break;
2589 }
2590
2591 /* Address size prefix will turn Disp64/Disp32/Disp16 operand
2592 into Disp32/Disp16/Disp32 operand. */
2593 if (i.prefix[ADDR_PREFIX] != 0)
2594 {
2595 unsigned int DispOn = 0, DispOff = 0;
2596
2597 switch (flag_code)
2598 {
2599 case CODE_16BIT:
2600 DispOn = Disp32;
2601 DispOff = Disp16;
2602 break;
2603 case CODE_32BIT:
2604 DispOn = Disp16;
2605 DispOff = Disp32;
2606 break;
2607 case CODE_64BIT:
2608 DispOn = Disp32;
2609 DispOff = Disp64;
2610 break;
2611 }
2612
2613 for (j = 0; j < MAX_OPERANDS; j++)
2614 {
2615 /* There should be only one Disp operand. */
2616 if ((operand_types[j] & DispOff))
2617 {
2618 addr_prefix_disp = j;
2619 operand_types[j] |= DispOn;
2620 operand_types[j] &= ~DispOff;
2621 break;
2622 }
2623 }
2624 }
2625
2626 overlap0 = i.types[0] & operand_types[0];
2627 switch (t->operands)
2628 {
2629 case 1:
2630 if (!MATCH (overlap0, i.types[0], operand_types[0]))
2631 continue;
2632 break;
2633 case 2:
2634 /* xchg %eax, %eax is a special case. It is an aliase for nop
2635 only in 32bit mode and we can use opcode 0x90. In 64bit
2636 mode, we can't use 0x90 for xchg %eax, %eax since it should
2637 zero-extend %eax to %rax. */
2638 if (flag_code == CODE_64BIT
2639 && t->base_opcode == 0x90
2640 && i.types [0] == (Acc | Reg32)
2641 && i.types [1] == (Acc | Reg32))
2642 continue;
2643 case 3:
2644 case 4:
2645 overlap1 = i.types[1] & operand_types[1];
2646 if (!MATCH (overlap0, i.types[0], operand_types[0])
2647 || !MATCH (overlap1, i.types[1], operand_types[1])
2648 /* monitor in SSE3 is a very special case. The first
2649 register and the second register may have different
2650 sizes. The same applies to crc32 in SSE4.2. */
2651 || !((t->base_opcode == 0x0f01
2652 && t->extension_opcode == 0xc8)
2653 || t->base_opcode == 0xf20f38f1
2654 || CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2655 operand_types[0],
2656 overlap1, i.types[1],
2657 operand_types[1])))
2658 {
2659 /* Check if other direction is valid ... */
2660 if ((t->opcode_modifier & (D | FloatD)) == 0)
2661 continue;
2662
2663 /* Try reversing direction of operands. */
2664 overlap0 = i.types[0] & operand_types[1];
2665 overlap1 = i.types[1] & operand_types[0];
2666 if (!MATCH (overlap0, i.types[0], operand_types[1])
2667 || !MATCH (overlap1, i.types[1], operand_types[0])
2668 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2669 operand_types[1],
2670 overlap1, i.types[1],
2671 operand_types[0]))
2672 {
2673 /* Does not match either direction. */
2674 continue;
2675 }
2676 /* found_reverse_match holds which of D or FloatDR
2677 we've found. */
2678 if ((t->opcode_modifier & D))
2679 found_reverse_match = Opcode_D;
2680 else if ((t->opcode_modifier & FloatD))
2681 found_reverse_match = Opcode_FloatD;
2682 else
2683 found_reverse_match = 0;
2684 if ((t->opcode_modifier & FloatR))
2685 found_reverse_match |= Opcode_FloatR;
2686 }
2687 else
2688 {
2689 /* Found a forward 2 operand match here. */
2690 switch (t->operands)
2691 {
2692 case 4:
2693 overlap3 = i.types[3] & operand_types[3];
2694 case 3:
2695 overlap2 = i.types[2] & operand_types[2];
2696 break;
2697 }
2698
2699 switch (t->operands)
2700 {
2701 case 4:
2702 if (!MATCH (overlap3, i.types[3], operand_types[3])
2703 || !CONSISTENT_REGISTER_MATCH (overlap2,
2704 i.types[2],
2705 operand_types[2],
2706 overlap3,
2707 i.types[3],
2708 operand_types[3]))
2709 continue;
2710 case 3:
2711 /* Here we make use of the fact that there are no
2712 reverse match 3 operand instructions, and all 3
2713 operand instructions only need to be checked for
2714 register consistency between operands 2 and 3. */
2715 if (!MATCH (overlap2, i.types[2], operand_types[2])
2716 || !CONSISTENT_REGISTER_MATCH (overlap1,
2717 i.types[1],
2718 operand_types[1],
2719 overlap2,
2720 i.types[2],
2721 operand_types[2]))
2722 continue;
2723 break;
2724 }
2725 }
2726 /* Found either forward/reverse 2, 3 or 4 operand match here:
2727 slip through to break. */
2728 }
2729 if (t->cpu_flags & ~cpu_arch_flags)
2730 {
2731 found_reverse_match = 0;
2732 continue;
2733 }
2734 /* We've found a match; break out of loop. */
2735 break;
2736 }
2737
2738 if (t == current_templates->end)
2739 {
2740 /* We found no match. */
2741 as_bad (_("suffix or operands invalid for `%s'"),
2742 current_templates->start->name);
2743 return 0;
2744 }
2745
2746 if (!quiet_warnings)
2747 {
2748 if (!intel_syntax
2749 && ((i.types[0] & JumpAbsolute)
2750 != (operand_types[0] & JumpAbsolute)))
2751 {
2752 as_warn (_("indirect %s without `*'"), t->name);
2753 }
2754
2755 if ((t->opcode_modifier & (IsPrefix | IgnoreSize))
2756 == (IsPrefix | IgnoreSize))
2757 {
2758 /* Warn them that a data or address size prefix doesn't
2759 affect assembly of the next line of code. */
2760 as_warn (_("stand-alone `%s' prefix"), t->name);
2761 }
2762 }
2763
2764 /* Copy the template we found. */
2765 i.tm = *t;
2766
2767 if (addr_prefix_disp != -1)
2768 i.tm.operand_types[addr_prefix_disp]
2769 = operand_types[addr_prefix_disp];
2770
2771 if (found_reverse_match)
2772 {
2773 /* If we found a reverse match we must alter the opcode
2774 direction bit. found_reverse_match holds bits to change
2775 (different for int & float insns). */
2776
2777 i.tm.base_opcode ^= found_reverse_match;
2778
2779 i.tm.operand_types[0] = operand_types[1];
2780 i.tm.operand_types[1] = operand_types[0];
2781 }
2782
2783 return 1;
2784 }
2785
2786 static int
2787 check_string (void)
2788 {
2789 int mem_op = (i.types[0] & AnyMem) ? 0 : 1;
2790 if ((i.tm.operand_types[mem_op] & EsSeg) != 0)
2791 {
2792 if (i.seg[0] != NULL && i.seg[0] != &es)
2793 {
2794 as_bad (_("`%s' operand %d must use `%%es' segment"),
2795 i.tm.name,
2796 mem_op + 1);
2797 return 0;
2798 }
2799 /* There's only ever one segment override allowed per instruction.
2800 This instruction possibly has a legal segment override on the
2801 second operand, so copy the segment to where non-string
2802 instructions store it, allowing common code. */
2803 i.seg[0] = i.seg[1];
2804 }
2805 else if ((i.tm.operand_types[mem_op + 1] & EsSeg) != 0)
2806 {
2807 if (i.seg[1] != NULL && i.seg[1] != &es)
2808 {
2809 as_bad (_("`%s' operand %d must use `%%es' segment"),
2810 i.tm.name,
2811 mem_op + 2);
2812 return 0;
2813 }
2814 }
2815 return 1;
2816 }
2817
2818 static int
2819 process_suffix (void)
2820 {
2821 /* If matched instruction specifies an explicit instruction mnemonic
2822 suffix, use it. */
2823 if (i.tm.opcode_modifier & (Size16 | Size32 | Size64))
2824 {
2825 if (i.tm.opcode_modifier & Size16)
2826 i.suffix = WORD_MNEM_SUFFIX;
2827 else if (i.tm.opcode_modifier & Size64)
2828 i.suffix = QWORD_MNEM_SUFFIX;
2829 else
2830 i.suffix = LONG_MNEM_SUFFIX;
2831 }
2832 else if (i.reg_operands)
2833 {
2834 /* If there's no instruction mnemonic suffix we try to invent one
2835 based on register operands. */
2836 if (!i.suffix)
2837 {
2838 /* We take i.suffix from the last register operand specified,
2839 Destination register type is more significant than source
2840 register type. crc32 in SSE4.2 prefers source register
2841 type. */
2842 if (i.tm.base_opcode == 0xf20f38f1)
2843 {
2844 if ((i.types[0] & Reg))
2845 i.suffix = ((i.types[0] & Reg16) ? WORD_MNEM_SUFFIX :
2846 LONG_MNEM_SUFFIX);
2847 }
2848 else if (i.tm.base_opcode == 0xf20f38f0)
2849 {
2850 if ((i.types[0] & Reg8))
2851 i.suffix = BYTE_MNEM_SUFFIX;
2852 }
2853
2854 if (!i.suffix)
2855 {
2856 int op;
2857
2858 if (i.tm.base_opcode == 0xf20f38f1
2859 || i.tm.base_opcode == 0xf20f38f0)
2860 {
2861 /* We have to know the operand size for crc32. */
2862 as_bad (_("ambiguous memory operand size for `%s`"),
2863 i.tm.name);
2864 return 0;
2865 }
2866
2867 for (op = i.operands; --op >= 0;)
2868 if ((i.types[op] & Reg)
2869 && !(i.tm.operand_types[op] & InOutPortReg))
2870 {
2871 i.suffix = ((i.types[op] & Reg8) ? BYTE_MNEM_SUFFIX :
2872 (i.types[op] & Reg16) ? WORD_MNEM_SUFFIX :
2873 (i.types[op] & Reg64) ? QWORD_MNEM_SUFFIX :
2874 LONG_MNEM_SUFFIX);
2875 break;
2876 }
2877 }
2878 }
2879 else if (i.suffix == BYTE_MNEM_SUFFIX)
2880 {
2881 if (!check_byte_reg ())
2882 return 0;
2883 }
2884 else if (i.suffix == LONG_MNEM_SUFFIX)
2885 {
2886 if (!check_long_reg ())
2887 return 0;
2888 }
2889 else if (i.suffix == QWORD_MNEM_SUFFIX)
2890 {
2891 if (!check_qword_reg ())
2892 return 0;
2893 }
2894 else if (i.suffix == WORD_MNEM_SUFFIX)
2895 {
2896 if (!check_word_reg ())
2897 return 0;
2898 }
2899 else if (intel_syntax && (i.tm.opcode_modifier & IgnoreSize))
2900 /* Do nothing if the instruction is going to ignore the prefix. */
2901 ;
2902 else
2903 abort ();
2904 }
2905 else if ((i.tm.opcode_modifier & DefaultSize)
2906 && !i.suffix
2907 /* exclude fldenv/frstor/fsave/fstenv */
2908 && (i.tm.opcode_modifier & No_sSuf))
2909 {
2910 i.suffix = stackop_size;
2911 }
2912 else if (intel_syntax
2913 && !i.suffix
2914 && ((i.tm.operand_types[0] & JumpAbsolute)
2915 || (i.tm.opcode_modifier & (JumpByte|JumpInterSegment))
2916 || (i.tm.base_opcode == 0x0f01 /* [ls][gi]dt */
2917 && i.tm.extension_opcode <= 3)))
2918 {
2919 switch (flag_code)
2920 {
2921 case CODE_64BIT:
2922 if (!(i.tm.opcode_modifier & No_qSuf))
2923 {
2924 i.suffix = QWORD_MNEM_SUFFIX;
2925 break;
2926 }
2927 case CODE_32BIT:
2928 if (!(i.tm.opcode_modifier & No_lSuf))
2929 i.suffix = LONG_MNEM_SUFFIX;
2930 break;
2931 case CODE_16BIT:
2932 if (!(i.tm.opcode_modifier & No_wSuf))
2933 i.suffix = WORD_MNEM_SUFFIX;
2934 break;
2935 }
2936 }
2937
2938 if (!i.suffix)
2939 {
2940 if (!intel_syntax)
2941 {
2942 if (i.tm.opcode_modifier & W)
2943 {
2944 as_bad (_("no instruction mnemonic suffix given and "
2945 "no register operands; can't size instruction"));
2946 return 0;
2947 }
2948 }
2949 else
2950 {
2951 unsigned int suffixes = (~i.tm.opcode_modifier
2952 & (No_bSuf
2953 | No_wSuf
2954 | No_lSuf
2955 | No_sSuf
2956 | No_xSuf
2957 | No_qSuf));
2958
2959 if ((i.tm.opcode_modifier & W)
2960 || ((suffixes & (suffixes - 1))
2961 && !(i.tm.opcode_modifier & (DefaultSize | IgnoreSize))))
2962 {
2963 as_bad (_("ambiguous operand size for `%s'"), i.tm.name);
2964 return 0;
2965 }
2966 }
2967 }
2968
2969 /* Change the opcode based on the operand size given by i.suffix;
2970 We don't need to change things for byte insns. */
2971
2972 if (i.suffix && i.suffix != BYTE_MNEM_SUFFIX)
2973 {
2974 /* It's not a byte, select word/dword operation. */
2975 if (i.tm.opcode_modifier & W)
2976 {
2977 if (i.tm.opcode_modifier & ShortForm)
2978 i.tm.base_opcode |= 8;
2979 else
2980 i.tm.base_opcode |= 1;
2981 }
2982
2983 /* Now select between word & dword operations via the operand
2984 size prefix, except for instructions that will ignore this
2985 prefix anyway. */
2986 if (i.tm.base_opcode == 0x0f01 && i.tm.extension_opcode == 0xc8)
2987 {
2988 /* monitor in SSE3 is a very special case. The default size
2989 of AX is the size of mode. The address size override
2990 prefix will change the size of AX. */
2991 if (i.op->regs[0].reg_type &
2992 (flag_code == CODE_32BIT ? Reg16 : Reg32))
2993 if (!add_prefix (ADDR_PREFIX_OPCODE))
2994 return 0;
2995 }
2996 else if (i.suffix != QWORD_MNEM_SUFFIX
2997 && i.suffix != LONG_DOUBLE_MNEM_SUFFIX
2998 && !(i.tm.opcode_modifier & (IgnoreSize | FloatMF))
2999 && ((i.suffix == LONG_MNEM_SUFFIX) == (flag_code == CODE_16BIT)
3000 || (flag_code == CODE_64BIT
3001 && (i.tm.opcode_modifier & JumpByte))))
3002 {
3003 unsigned int prefix = DATA_PREFIX_OPCODE;
3004
3005 if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */
3006 prefix = ADDR_PREFIX_OPCODE;
3007
3008 if (!add_prefix (prefix))
3009 return 0;
3010 }
3011
3012 /* Set mode64 for an operand. */
3013 if (i.suffix == QWORD_MNEM_SUFFIX
3014 && flag_code == CODE_64BIT
3015 && (i.tm.opcode_modifier & NoRex64) == 0)
3016 {
3017 /* Special case for xchg %rax,%rax. It is NOP and doesn't
3018 need rex64. */
3019 if (i.operands != 2
3020 || i.types [0] != (Acc | Reg64)
3021 || i.types [1] != (Acc | Reg64)
3022 || i.tm.base_opcode != 0x90)
3023 i.rex |= REX_W;
3024 }
3025
3026 /* Size floating point instruction. */
3027 if (i.suffix == LONG_MNEM_SUFFIX)
3028 if (i.tm.opcode_modifier & FloatMF)
3029 i.tm.base_opcode ^= 4;
3030 }
3031
3032 return 1;
3033 }
3034
3035 static int
3036 check_byte_reg (void)
3037 {
3038 int op;
3039
3040 for (op = i.operands; --op >= 0;)
3041 {
3042 /* If this is an eight bit register, it's OK. If it's the 16 or
3043 32 bit version of an eight bit register, we will just use the
3044 low portion, and that's OK too. */
3045 if (i.types[op] & Reg8)
3046 continue;
3047
3048 /* movzx and movsx should not generate this warning. */
3049 if (intel_syntax
3050 && (i.tm.base_opcode == 0xfb7
3051 || i.tm.base_opcode == 0xfb6
3052 || i.tm.base_opcode == 0x63
3053 || i.tm.base_opcode == 0xfbe
3054 || i.tm.base_opcode == 0xfbf))
3055 continue;
3056
3057 /* crc32 doesn't generate this warning. */
3058 if (i.tm.base_opcode == 0xf20f38f0)
3059 continue;
3060
3061 if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4)
3062 {
3063 /* Prohibit these changes in the 64bit mode, since the
3064 lowering is more complicated. */
3065 if (flag_code == CODE_64BIT
3066 && (i.tm.operand_types[op] & InOutPortReg) == 0)
3067 {
3068 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3069 register_prefix, i.op[op].regs->reg_name,
3070 i.suffix);
3071 return 0;
3072 }
3073 #if REGISTER_WARNINGS
3074 if (!quiet_warnings
3075 && (i.tm.operand_types[op] & InOutPortReg) == 0)
3076 as_warn (_("using `%s%s' instead of `%s%s' due to `%c' suffix"),
3077 register_prefix,
3078 (i.op[op].regs + (i.types[op] & Reg16
3079 ? REGNAM_AL - REGNAM_AX
3080 : REGNAM_AL - REGNAM_EAX))->reg_name,
3081 register_prefix,
3082 i.op[op].regs->reg_name,
3083 i.suffix);
3084 #endif
3085 continue;
3086 }
3087 /* Any other register is bad. */
3088 if (i.types[op] & (Reg | RegMMX | RegXMM
3089 | SReg2 | SReg3
3090 | Control | Debug | Test
3091 | FloatReg | FloatAcc))
3092 {
3093 as_bad (_("`%s%s' not allowed with `%s%c'"),
3094 register_prefix,
3095 i.op[op].regs->reg_name,
3096 i.tm.name,
3097 i.suffix);
3098 return 0;
3099 }
3100 }
3101 return 1;
3102 }
3103
3104 static int
3105 check_long_reg (void)
3106 {
3107 int op;
3108
3109 for (op = i.operands; --op >= 0;)
3110 /* Reject eight bit registers, except where the template requires
3111 them. (eg. movzb) */
3112 if ((i.types[op] & Reg8) != 0
3113 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
3114 {
3115 as_bad (_("`%s%s' not allowed with `%s%c'"),
3116 register_prefix,
3117 i.op[op].regs->reg_name,
3118 i.tm.name,
3119 i.suffix);
3120 return 0;
3121 }
3122 /* Warn if the e prefix on a general reg is missing. */
3123 else if ((!quiet_warnings || flag_code == CODE_64BIT)
3124 && (i.types[op] & Reg16) != 0
3125 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
3126 {
3127 /* Prohibit these changes in the 64bit mode, since the
3128 lowering is more complicated. */
3129 if (flag_code == CODE_64BIT)
3130 {
3131 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3132 register_prefix, i.op[op].regs->reg_name,
3133 i.suffix);
3134 return 0;
3135 }
3136 #if REGISTER_WARNINGS
3137 else
3138 as_warn (_("using `%s%s' instead of `%s%s' due to `%c' suffix"),
3139 register_prefix,
3140 (i.op[op].regs + REGNAM_EAX - REGNAM_AX)->reg_name,
3141 register_prefix,
3142 i.op[op].regs->reg_name,
3143 i.suffix);
3144 #endif
3145 }
3146 /* Warn if the r prefix on a general reg is missing. */
3147 else if ((i.types[op] & Reg64) != 0
3148 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
3149 {
3150 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3151 register_prefix, i.op[op].regs->reg_name,
3152 i.suffix);
3153 return 0;
3154 }
3155 return 1;
3156 }
3157
3158 static int
3159 check_qword_reg (void)
3160 {
3161 int op;
3162
3163 for (op = i.operands; --op >= 0; )
3164 /* Reject eight bit registers, except where the template requires
3165 them. (eg. movzb) */
3166 if ((i.types[op] & Reg8) != 0
3167 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
3168 {
3169 as_bad (_("`%s%s' not allowed with `%s%c'"),
3170 register_prefix,
3171 i.op[op].regs->reg_name,
3172 i.tm.name,
3173 i.suffix);
3174 return 0;
3175 }
3176 /* Warn if the e prefix on a general reg is missing. */
3177 else if (((i.types[op] & Reg16) != 0
3178 || (i.types[op] & Reg32) != 0)
3179 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
3180 {
3181 /* Prohibit these changes in the 64bit mode, since the
3182 lowering is more complicated. */
3183 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3184 register_prefix, i.op[op].regs->reg_name,
3185 i.suffix);
3186 return 0;
3187 }
3188 return 1;
3189 }
3190
3191 static int
3192 check_word_reg (void)
3193 {
3194 int op;
3195 for (op = i.operands; --op >= 0;)
3196 /* Reject eight bit registers, except where the template requires
3197 them. (eg. movzb) */
3198 if ((i.types[op] & Reg8) != 0
3199 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
3200 {
3201 as_bad (_("`%s%s' not allowed with `%s%c'"),
3202 register_prefix,
3203 i.op[op].regs->reg_name,
3204 i.tm.name,
3205 i.suffix);
3206 return 0;
3207 }
3208 /* Warn if the e prefix on a general reg is present. */
3209 else if ((!quiet_warnings || flag_code == CODE_64BIT)
3210 && (i.types[op] & Reg32) != 0
3211 && (i.tm.operand_types[op] & (Reg16 | Acc)) != 0)
3212 {
3213 /* Prohibit these changes in the 64bit mode, since the
3214 lowering is more complicated. */
3215 if (flag_code == CODE_64BIT)
3216 {
3217 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3218 register_prefix, i.op[op].regs->reg_name,
3219 i.suffix);
3220 return 0;
3221 }
3222 else
3223 #if REGISTER_WARNINGS
3224 as_warn (_("using `%s%s' instead of `%s%s' due to `%c' suffix"),
3225 register_prefix,
3226 (i.op[op].regs + REGNAM_AX - REGNAM_EAX)->reg_name,
3227 register_prefix,
3228 i.op[op].regs->reg_name,
3229 i.suffix);
3230 #endif
3231 }
3232 return 1;
3233 }
3234
3235 static int
3236 finalize_imm (void)
3237 {
3238 unsigned int overlap0, overlap1, overlap2;
3239
3240 overlap0 = i.types[0] & i.tm.operand_types[0];
3241 if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64))
3242 && overlap0 != Imm8 && overlap0 != Imm8S
3243 && overlap0 != Imm16 && overlap0 != Imm32S
3244 && overlap0 != Imm32 && overlap0 != Imm64)
3245 {
3246 if (i.suffix)
3247 {
3248 overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX
3249 ? Imm8 | Imm8S
3250 : (i.suffix == WORD_MNEM_SUFFIX
3251 ? Imm16
3252 : (i.suffix == QWORD_MNEM_SUFFIX
3253 ? Imm64 | Imm32S
3254 : Imm32)));
3255 }
3256 else if (overlap0 == (Imm16 | Imm32S | Imm32)
3257 || overlap0 == (Imm16 | Imm32)
3258 || overlap0 == (Imm16 | Imm32S))
3259 {
3260 overlap0 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
3261 ? Imm16 : Imm32S);
3262 }
3263 if (overlap0 != Imm8 && overlap0 != Imm8S
3264 && overlap0 != Imm16 && overlap0 != Imm32S
3265 && overlap0 != Imm32 && overlap0 != Imm64)
3266 {
3267 as_bad (_("no instruction mnemonic suffix given; "
3268 "can't determine immediate size"));
3269 return 0;
3270 }
3271 }
3272 i.types[0] = overlap0;
3273
3274 overlap1 = i.types[1] & i.tm.operand_types[1];
3275 if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32 | Imm64))
3276 && overlap1 != Imm8 && overlap1 != Imm8S
3277 && overlap1 != Imm16 && overlap1 != Imm32S
3278 && overlap1 != Imm32 && overlap1 != Imm64)
3279 {
3280 if (i.suffix)
3281 {
3282 overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX
3283 ? Imm8 | Imm8S
3284 : (i.suffix == WORD_MNEM_SUFFIX
3285 ? Imm16
3286 : (i.suffix == QWORD_MNEM_SUFFIX
3287 ? Imm64 | Imm32S
3288 : Imm32)));
3289 }
3290 else if (overlap1 == (Imm16 | Imm32 | Imm32S)
3291 || overlap1 == (Imm16 | Imm32)
3292 || overlap1 == (Imm16 | Imm32S))
3293 {
3294 overlap1 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
3295 ? Imm16 : Imm32S);
3296 }
3297 if (overlap1 != Imm8 && overlap1 != Imm8S
3298 && overlap1 != Imm16 && overlap1 != Imm32S
3299 && overlap1 != Imm32 && overlap1 != Imm64)
3300 {
3301 as_bad (_("no instruction mnemonic suffix given; "
3302 "can't determine immediate size %x %c"),
3303 overlap1, i.suffix);
3304 return 0;
3305 }
3306 }
3307 i.types[1] = overlap1;
3308
3309 overlap2 = i.types[2] & i.tm.operand_types[2];
3310 assert ((overlap2 & Imm) == 0);
3311 i.types[2] = overlap2;
3312
3313 return 1;
3314 }
3315
3316 static int
3317 process_operands (void)
3318 {
3319 /* Default segment register this instruction will use for memory
3320 accesses. 0 means unknown. This is only for optimizing out
3321 unnecessary segment overrides. */
3322 const seg_entry *default_seg = 0;
3323
3324 /* The imul $imm, %reg instruction is converted into
3325 imul $imm, %reg, %reg, and the clr %reg instruction
3326 is converted into xor %reg, %reg. */
3327 if (i.tm.opcode_modifier & RegKludge)
3328 {
3329 if ((i.tm.cpu_flags & CpuSSE4_1))
3330 {
3331 /* The first operand in instruction blendvpd, blendvps and
3332 pblendvb in SSE4.1 is implicit and must be xmm0. */
3333 assert (i.operands == 3
3334 && i.reg_operands >= 2
3335 && i.types[0] == RegXMM);
3336 if (i.op[0].regs->reg_num != 0)
3337 {
3338 if (intel_syntax)
3339 as_bad (_("the last operand of `%s' must be `%sxmm0'"),
3340 i.tm.name, register_prefix);
3341 else
3342 as_bad (_("the first operand of `%s' must be `%sxmm0'"),
3343 i.tm.name, register_prefix);
3344 return 0;
3345 }
3346 i.op[0] = i.op[1];
3347 i.op[1] = i.op[2];
3348 i.types[0] = i.types[1];
3349 i.types[1] = i.types[2];
3350 i.operands--;
3351 i.reg_operands--;
3352
3353 /* We need to adjust fields in i.tm since they are used by
3354 build_modrm_byte. */
3355 i.tm.operand_types [0] = i.tm.operand_types [1];
3356 i.tm.operand_types [1] = i.tm.operand_types [2];
3357 i.tm.operands--;
3358 }
3359 else
3360 {
3361 unsigned int first_reg_op = (i.types[0] & Reg) ? 0 : 1;
3362 /* Pretend we saw the extra register operand. */
3363 assert (i.reg_operands == 1
3364 && i.op[first_reg_op + 1].regs == 0);
3365 i.op[first_reg_op + 1].regs = i.op[first_reg_op].regs;
3366 i.types[first_reg_op + 1] = i.types[first_reg_op];
3367 i.operands++;
3368 i.reg_operands++;
3369 }
3370 }
3371
3372 if (i.tm.opcode_modifier & ShortForm)
3373 {
3374 if (i.types[0] & (SReg2 | SReg3))
3375 {
3376 if (i.tm.base_opcode == POP_SEG_SHORT
3377 && i.op[0].regs->reg_num == 1)
3378 {
3379 as_bad (_("you can't `pop %%cs'"));
3380 return 0;
3381 }
3382 i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
3383 if ((i.op[0].regs->reg_flags & RegRex) != 0)
3384 i.rex |= REX_B;
3385 }
3386 else
3387 {
3388 /* The register or float register operand is in operand 0 or 1. */
3389 unsigned int op = (i.types[0] & (Reg | FloatReg)) ? 0 : 1;
3390 /* Register goes in low 3 bits of opcode. */
3391 i.tm.base_opcode |= i.op[op].regs->reg_num;
3392 if ((i.op[op].regs->reg_flags & RegRex) != 0)
3393 i.rex |= REX_B;
3394 if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
3395 {
3396 /* Warn about some common errors, but press on regardless.
3397 The first case can be generated by gcc (<= 2.8.1). */
3398 if (i.operands == 2)
3399 {
3400 /* Reversed arguments on faddp, fsubp, etc. */
3401 as_warn (_("translating to `%s %s%s,%s%s'"), i.tm.name,
3402 register_prefix, i.op[1].regs->reg_name,
3403 register_prefix, i.op[0].regs->reg_name);
3404 }
3405 else
3406 {
3407 /* Extraneous `l' suffix on fp insn. */
3408 as_warn (_("translating to `%s %s%s'"), i.tm.name,
3409 register_prefix, i.op[0].regs->reg_name);
3410 }
3411 }
3412 }
3413 }
3414 else if (i.tm.opcode_modifier & Modrm)
3415 {
3416 /* The opcode is completed (modulo i.tm.extension_opcode which
3417 must be put into the modrm byte). Now, we make the modrm and
3418 index base bytes based on all the info we've collected. */
3419
3420 default_seg = build_modrm_byte ();
3421 }
3422 else if ((i.tm.base_opcode & ~0x3) == MOV_AX_DISP32)
3423 {
3424 default_seg = &ds;
3425 }
3426 else if ((i.tm.opcode_modifier & IsString) != 0)
3427 {
3428 /* For the string instructions that allow a segment override
3429 on one of their operands, the default segment is ds. */
3430 default_seg = &ds;
3431 }
3432
3433 if ((i.tm.base_opcode == 0x8d /* lea */
3434 || (i.tm.cpu_flags & CpuSVME))
3435 && i.seg[0] && !quiet_warnings)
3436 as_warn (_("segment override on `%s' is ineffectual"), i.tm.name);
3437
3438 /* If a segment was explicitly specified, and the specified segment
3439 is not the default, use an opcode prefix to select it. If we
3440 never figured out what the default segment is, then default_seg
3441 will be zero at this point, and the specified segment prefix will
3442 always be used. */
3443 if ((i.seg[0]) && (i.seg[0] != default_seg))
3444 {
3445 if (!add_prefix (i.seg[0]->seg_prefix))
3446 return 0;
3447 }
3448 return 1;
3449 }
3450
3451 static const seg_entry *
3452 build_modrm_byte (void)
3453 {
3454 const seg_entry *default_seg = 0;
3455
3456 /* i.reg_operands MUST be the number of real register operands;
3457 implicit registers do not count. */
3458 if (i.reg_operands == 2)
3459 {
3460 unsigned int source, dest;
3461
3462 switch (i.operands)
3463 {
3464 case 2:
3465 source = 0;
3466 break;
3467 case 3:
3468 /* When there are 3 operands, one of them may be immediate,
3469 which may be the first or the last operand. Otherwise,
3470 the first operand must be shift count register (cl). */
3471 assert (i.imm_operands == 1
3472 || (i.imm_operands == 0
3473 && (i.types[0] & ShiftCount)));
3474 source = (i.types[0] & (Imm | ShiftCount)) ? 1 : 0;
3475 break;
3476 case 4:
3477 /* When there are 4 operands, the first two must be immediate
3478 operands. The source operand will be the 3rd one. */
3479 assert (i.imm_operands == 2
3480 && (i.types[0] & Imm)
3481 && (i.types[1] & Imm));
3482 source = 2;
3483 break;
3484 default:
3485 abort ();
3486 }
3487
3488 dest = source + 1;
3489
3490 i.rm.mode = 3;
3491 /* One of the register operands will be encoded in the i.tm.reg
3492 field, the other in the combined i.tm.mode and i.tm.regmem
3493 fields. If no form of this instruction supports a memory
3494 destination operand, then we assume the source operand may
3495 sometimes be a memory operand and so we need to store the
3496 destination in the i.rm.reg field. */
3497 if ((i.tm.operand_types[dest] & (AnyMem | RegMem)) == 0)
3498 {
3499 i.rm.reg = i.op[dest].regs->reg_num;
3500 i.rm.regmem = i.op[source].regs->reg_num;
3501 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
3502 i.rex |= REX_R;
3503 if ((i.op[source].regs->reg_flags & RegRex) != 0)
3504 i.rex |= REX_B;
3505 }
3506 else
3507 {
3508 i.rm.reg = i.op[source].regs->reg_num;
3509 i.rm.regmem = i.op[dest].regs->reg_num;
3510 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
3511 i.rex |= REX_B;
3512 if ((i.op[source].regs->reg_flags & RegRex) != 0)
3513 i.rex |= REX_R;
3514 }
3515 if (flag_code != CODE_64BIT && (i.rex & (REX_R | REX_B)))
3516 {
3517 if (!((i.types[0] | i.types[1]) & Control))
3518 abort ();
3519 i.rex &= ~(REX_R | REX_B);
3520 add_prefix (LOCK_PREFIX_OPCODE);
3521 }
3522 }
3523 else
3524 { /* If it's not 2 reg operands... */
3525 if (i.mem_operands)
3526 {
3527 unsigned int fake_zero_displacement = 0;
3528 unsigned int op;
3529
3530 for (op = 0; op < i.operands; op++)
3531 if ((i.types[op] & AnyMem))
3532 break;
3533 assert (op < i.operands);
3534
3535 default_seg = &ds;
3536
3537 if (i.base_reg == 0)
3538 {
3539 i.rm.mode = 0;
3540 if (!i.disp_operands)
3541 fake_zero_displacement = 1;
3542 if (i.index_reg == 0)
3543 {
3544 /* Operand is just <disp> */
3545 if (flag_code == CODE_64BIT)
3546 {
3547 /* 64bit mode overwrites the 32bit absolute
3548 addressing by RIP relative addressing and
3549 absolute addressing is encoded by one of the
3550 redundant SIB forms. */
3551 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3552 i.sib.base = NO_BASE_REGISTER;
3553 i.sib.index = NO_INDEX_REGISTER;
3554 i.types[op] = ((i.prefix[ADDR_PREFIX] == 0)
3555 ? Disp32S : Disp32);
3556 }
3557 else if ((flag_code == CODE_16BIT)
3558 ^ (i.prefix[ADDR_PREFIX] != 0))
3559 {
3560 i.rm.regmem = NO_BASE_REGISTER_16;
3561 i.types[op] = Disp16;
3562 }
3563 else
3564 {
3565 i.rm.regmem = NO_BASE_REGISTER;
3566 i.types[op] = Disp32;
3567 }
3568 }
3569 else /* !i.base_reg && i.index_reg */
3570 {
3571 i.sib.index = i.index_reg->reg_num;
3572 i.sib.base = NO_BASE_REGISTER;
3573 i.sib.scale = i.log2_scale_factor;
3574 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3575 i.types[op] &= ~Disp;
3576 if (flag_code != CODE_64BIT)
3577 i.types[op] |= Disp32; /* Must be 32 bit */
3578 else
3579 i.types[op] |= Disp32S;
3580 if ((i.index_reg->reg_flags & RegRex) != 0)
3581 i.rex |= REX_X;
3582 }
3583 }
3584 /* RIP addressing for 64bit mode. */
3585 else if (i.base_reg->reg_type == BaseIndex)
3586 {
3587 i.rm.regmem = NO_BASE_REGISTER;
3588 i.types[op] &= ~ Disp;
3589 i.types[op] |= Disp32S;
3590 i.flags[op] |= Operand_PCrel;
3591 if (! i.disp_operands)
3592 fake_zero_displacement = 1;
3593 }
3594 else if (i.base_reg->reg_type & Reg16)
3595 {
3596 switch (i.base_reg->reg_num)
3597 {
3598 case 3: /* (%bx) */
3599 if (i.index_reg == 0)
3600 i.rm.regmem = 7;
3601 else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */
3602 i.rm.regmem = i.index_reg->reg_num - 6;
3603 break;
3604 case 5: /* (%bp) */
3605 default_seg = &ss;
3606 if (i.index_reg == 0)
3607 {
3608 i.rm.regmem = 6;
3609 if ((i.types[op] & Disp) == 0)
3610 {
3611 /* fake (%bp) into 0(%bp) */
3612 i.types[op] |= Disp8;
3613 fake_zero_displacement = 1;
3614 }
3615 }
3616 else /* (%bp,%si) -> 2, or (%bp,%di) -> 3 */
3617 i.rm.regmem = i.index_reg->reg_num - 6 + 2;
3618 break;
3619 default: /* (%si) -> 4 or (%di) -> 5 */
3620 i.rm.regmem = i.base_reg->reg_num - 6 + 4;
3621 }
3622 i.rm.mode = mode_from_disp_size (i.types[op]);
3623 }
3624 else /* i.base_reg and 32/64 bit mode */
3625 {
3626 if (flag_code == CODE_64BIT
3627 && (i.types[op] & Disp))
3628 i.types[op] = ((i.types[op] & Disp8)
3629 | (i.prefix[ADDR_PREFIX] == 0
3630 ? Disp32S : Disp32));
3631
3632 i.rm.regmem = i.base_reg->reg_num;
3633 if ((i.base_reg->reg_flags & RegRex) != 0)
3634 i.rex |= REX_B;
3635 i.sib.base = i.base_reg->reg_num;
3636 /* x86-64 ignores REX prefix bit here to avoid decoder
3637 complications. */
3638 if ((i.base_reg->reg_num & 7) == EBP_REG_NUM)
3639 {
3640 default_seg = &ss;
3641 if (i.disp_operands == 0)
3642 {
3643 fake_zero_displacement = 1;
3644 i.types[op] |= Disp8;
3645 }
3646 }
3647 else if (i.base_reg->reg_num == ESP_REG_NUM)
3648 {
3649 default_seg = &ss;
3650 }
3651 i.sib.scale = i.log2_scale_factor;
3652 if (i.index_reg == 0)
3653 {
3654 /* <disp>(%esp) becomes two byte modrm with no index
3655 register. We've already stored the code for esp
3656 in i.rm.regmem ie. ESCAPE_TO_TWO_BYTE_ADDRESSING.
3657 Any base register besides %esp will not use the
3658 extra modrm byte. */
3659 i.sib.index = NO_INDEX_REGISTER;
3660 #if !SCALE1_WHEN_NO_INDEX
3661 /* Another case where we force the second modrm byte. */
3662 if (i.log2_scale_factor)
3663 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3664 #endif
3665 }
3666 else
3667 {
3668 i.sib.index = i.index_reg->reg_num;
3669 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3670 if ((i.index_reg->reg_flags & RegRex) != 0)
3671 i.rex |= REX_X;
3672 }
3673
3674 if (i.disp_operands
3675 && (i.reloc[op] == BFD_RELOC_386_TLS_DESC_CALL
3676 || i.reloc[op] == BFD_RELOC_X86_64_TLSDESC_CALL))
3677 i.rm.mode = 0;
3678 else
3679 i.rm.mode = mode_from_disp_size (i.types[op]);
3680 }
3681
3682 if (fake_zero_displacement)
3683 {
3684 /* Fakes a zero displacement assuming that i.types[op]
3685 holds the correct displacement size. */
3686 expressionS *exp;
3687
3688 assert (i.op[op].disps == 0);
3689 exp = &disp_expressions[i.disp_operands++];
3690 i.op[op].disps = exp;
3691 exp->X_op = O_constant;
3692 exp->X_add_number = 0;
3693 exp->X_add_symbol = (symbolS *) 0;
3694 exp->X_op_symbol = (symbolS *) 0;
3695 }
3696 }
3697
3698 /* Fill in i.rm.reg or i.rm.regmem field with register operand
3699 (if any) based on i.tm.extension_opcode. Again, we must be
3700 careful to make sure that segment/control/debug/test/MMX
3701 registers are coded into the i.rm.reg field. */
3702 if (i.reg_operands)
3703 {
3704 unsigned int op;
3705
3706 for (op = 0; op < i.operands; op++)
3707 if ((i.types[op] & (Reg | RegMMX | RegXMM
3708 | SReg2 | SReg3
3709 | Control | Debug | Test)))
3710 break;
3711 assert (op < i.operands);
3712
3713 /* If there is an extension opcode to put here, the register
3714 number must be put into the regmem field. */
3715 if (i.tm.extension_opcode != None)
3716 {
3717 i.rm.regmem = i.op[op].regs->reg_num;
3718 if ((i.op[op].regs->reg_flags & RegRex) != 0)
3719 i.rex |= REX_B;
3720 }
3721 else
3722 {
3723 i.rm.reg = i.op[op].regs->reg_num;
3724 if ((i.op[op].regs->reg_flags & RegRex) != 0)
3725 i.rex |= REX_R;
3726 }
3727
3728 /* Now, if no memory operand has set i.rm.mode = 0, 1, 2 we
3729 must set it to 3 to indicate this is a register operand
3730 in the regmem field. */
3731 if (!i.mem_operands)
3732 i.rm.mode = 3;
3733 }
3734
3735 /* Fill in i.rm.reg field with extension opcode (if any). */
3736 if (i.tm.extension_opcode != None)
3737 i.rm.reg = i.tm.extension_opcode;
3738 }
3739 return default_seg;
3740 }
3741
3742 static void
3743 output_branch (void)
3744 {
3745 char *p;
3746 int code16;
3747 int prefix;
3748 relax_substateT subtype;
3749 symbolS *sym;
3750 offsetT off;
3751
3752 code16 = 0;
3753 if (flag_code == CODE_16BIT)
3754 code16 = CODE16;
3755
3756 prefix = 0;
3757 if (i.prefix[DATA_PREFIX] != 0)
3758 {
3759 prefix = 1;
3760 i.prefixes -= 1;
3761 code16 ^= CODE16;
3762 }
3763 /* Pentium4 branch hints. */
3764 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
3765 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
3766 {
3767 prefix++;
3768 i.prefixes--;
3769 }
3770 if (i.prefix[REX_PREFIX] != 0)
3771 {
3772 prefix++;
3773 i.prefixes--;
3774 }
3775
3776 if (i.prefixes != 0 && !intel_syntax)
3777 as_warn (_("skipping prefixes on this instruction"));
3778
3779 /* It's always a symbol; End frag & setup for relax.
3780 Make sure there is enough room in this frag for the largest
3781 instruction we may generate in md_convert_frag. This is 2
3782 bytes for the opcode and room for the prefix and largest
3783 displacement. */
3784 frag_grow (prefix + 2 + 4);
3785 /* Prefix and 1 opcode byte go in fr_fix. */
3786 p = frag_more (prefix + 1);
3787 if (i.prefix[DATA_PREFIX] != 0)
3788 *p++ = DATA_PREFIX_OPCODE;
3789 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE
3790 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE)
3791 *p++ = i.prefix[SEG_PREFIX];
3792 if (i.prefix[REX_PREFIX] != 0)
3793 *p++ = i.prefix[REX_PREFIX];
3794 *p = i.tm.base_opcode;
3795
3796 if ((unsigned char) *p == JUMP_PC_RELATIVE)
3797 subtype = ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL);
3798 else if ((cpu_arch_flags & Cpu386) != 0)
3799 subtype = ENCODE_RELAX_STATE (COND_JUMP, SMALL);
3800 else
3801 subtype = ENCODE_RELAX_STATE (COND_JUMP86, SMALL);
3802 subtype |= code16;
3803
3804 sym = i.op[0].disps->X_add_symbol;
3805 off = i.op[0].disps->X_add_number;
3806
3807 if (i.op[0].disps->X_op != O_constant
3808 && i.op[0].disps->X_op != O_symbol)
3809 {
3810 /* Handle complex expressions. */
3811 sym = make_expr_symbol (i.op[0].disps);
3812 off = 0;
3813 }
3814
3815 /* 1 possible extra opcode + 4 byte displacement go in var part.
3816 Pass reloc in fr_var. */
3817 frag_var (rs_machine_dependent, 5, i.reloc[0], subtype, sym, off, p);
3818 }
3819
3820 static void
3821 output_jump (void)
3822 {
3823 char *p;
3824 int size;
3825 fixS *fixP;
3826
3827 if (i.tm.opcode_modifier & JumpByte)
3828 {
3829 /* This is a loop or jecxz type instruction. */
3830 size = 1;
3831 if (i.prefix[ADDR_PREFIX] != 0)
3832 {
3833 FRAG_APPEND_1_CHAR (ADDR_PREFIX_OPCODE);
3834 i.prefixes -= 1;
3835 }
3836 /* Pentium4 branch hints. */
3837 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
3838 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
3839 {
3840 FRAG_APPEND_1_CHAR (i.prefix[SEG_PREFIX]);
3841 i.prefixes--;
3842 }
3843 }
3844 else
3845 {
3846 int code16;
3847
3848 code16 = 0;
3849 if (flag_code == CODE_16BIT)
3850 code16 = CODE16;
3851
3852 if (i.prefix[DATA_PREFIX] != 0)
3853 {
3854 FRAG_APPEND_1_CHAR (DATA_PREFIX_OPCODE);
3855 i.prefixes -= 1;
3856 code16 ^= CODE16;
3857 }
3858
3859 size = 4;
3860 if (code16)
3861 size = 2;
3862 }
3863
3864 if (i.prefix[REX_PREFIX] != 0)
3865 {
3866 FRAG_APPEND_1_CHAR (i.prefix[REX_PREFIX]);
3867 i.prefixes -= 1;
3868 }
3869
3870 if (i.prefixes != 0 && !intel_syntax)
3871 as_warn (_("skipping prefixes on this instruction"));
3872
3873 p = frag_more (1 + size);
3874 *p++ = i.tm.base_opcode;
3875
3876 fixP = fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3877 i.op[0].disps, 1, reloc (size, 1, 1, i.reloc[0]));
3878
3879 /* All jumps handled here are signed, but don't use a signed limit
3880 check for 32 and 16 bit jumps as we want to allow wrap around at
3881 4G and 64k respectively. */
3882 if (size == 1)
3883 fixP->fx_signed = 1;
3884 }
3885
3886 static void
3887 output_interseg_jump (void)
3888 {
3889 char *p;
3890 int size;
3891 int prefix;
3892 int code16;
3893
3894 code16 = 0;
3895 if (flag_code == CODE_16BIT)
3896 code16 = CODE16;
3897
3898 prefix = 0;
3899 if (i.prefix[DATA_PREFIX] != 0)
3900 {
3901 prefix = 1;
3902 i.prefixes -= 1;
3903 code16 ^= CODE16;
3904 }
3905 if (i.prefix[REX_PREFIX] != 0)
3906 {
3907 prefix++;
3908 i.prefixes -= 1;
3909 }
3910
3911 size = 4;
3912 if (code16)
3913 size = 2;
3914
3915 if (i.prefixes != 0 && !intel_syntax)
3916 as_warn (_("skipping prefixes on this instruction"));
3917
3918 /* 1 opcode; 2 segment; offset */
3919 p = frag_more (prefix + 1 + 2 + size);
3920
3921 if (i.prefix[DATA_PREFIX] != 0)
3922 *p++ = DATA_PREFIX_OPCODE;
3923
3924 if (i.prefix[REX_PREFIX] != 0)
3925 *p++ = i.prefix[REX_PREFIX];
3926
3927 *p++ = i.tm.base_opcode;
3928 if (i.op[1].imms->X_op == O_constant)
3929 {
3930 offsetT n = i.op[1].imms->X_add_number;
3931
3932 if (size == 2
3933 && !fits_in_unsigned_word (n)
3934 && !fits_in_signed_word (n))
3935 {
3936 as_bad (_("16-bit jump out of range"));
3937 return;
3938 }
3939 md_number_to_chars (p, n, size);
3940 }
3941 else
3942 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3943 i.op[1].imms, 0, reloc (size, 0, 0, i.reloc[1]));
3944 if (i.op[0].imms->X_op != O_constant)
3945 as_bad (_("can't handle non absolute segment in `%s'"),
3946 i.tm.name);
3947 md_number_to_chars (p + size, (valueT) i.op[0].imms->X_add_number, 2);
3948 }
3949
3950 static void
3951 output_insn (void)
3952 {
3953 fragS *insn_start_frag;
3954 offsetT insn_start_off;
3955
3956 /* Tie dwarf2 debug info to the address at the start of the insn.
3957 We can't do this after the insn has been output as the current
3958 frag may have been closed off. eg. by frag_var. */
3959 dwarf2_emit_insn (0);
3960
3961 insn_start_frag = frag_now;
3962 insn_start_off = frag_now_fix ();
3963
3964 /* Output jumps. */
3965 if (i.tm.opcode_modifier & Jump)
3966 output_branch ();
3967 else if (i.tm.opcode_modifier & (JumpByte | JumpDword))
3968 output_jump ();
3969 else if (i.tm.opcode_modifier & JumpInterSegment)
3970 output_interseg_jump ();
3971 else
3972 {
3973 /* Output normal instructions here. */
3974 char *p;
3975 unsigned char *q;
3976 unsigned int prefix;
3977
3978 /* All opcodes on i386 have either 1 or 2 bytes. SSSE3 and
3979 SSE4 instructions have 3 bytes. We may use one more higher
3980 byte to specify a prefix the instruction requires. Exclude
3981 instructions which are in both SSE4 and ABM. */
3982 if ((i.tm.cpu_flags & (CpuSSSE3 | CpuSSE4)) != 0
3983 && (i.tm.cpu_flags & CpuABM) == 0)
3984 {
3985 if (i.tm.base_opcode & 0xff000000)
3986 {
3987 prefix = (i.tm.base_opcode >> 24) & 0xff;
3988 goto check_prefix;
3989 }
3990 }
3991 else if ((i.tm.base_opcode & 0xff0000) != 0)
3992 {
3993 prefix = (i.tm.base_opcode >> 16) & 0xff;
3994 if ((i.tm.cpu_flags & CpuPadLock) != 0)
3995 {
3996 check_prefix:
3997 if (prefix != REPE_PREFIX_OPCODE
3998 || i.prefix[LOCKREP_PREFIX] != REPE_PREFIX_OPCODE)
3999 add_prefix (prefix);
4000 }
4001 else
4002 add_prefix (prefix);
4003 }
4004
4005 /* The prefix bytes. */
4006 for (q = i.prefix;
4007 q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
4008 q++)
4009 {
4010 if (*q)
4011 {
4012 p = frag_more (1);
4013 md_number_to_chars (p, (valueT) *q, 1);
4014 }
4015 }
4016
4017 /* Now the opcode; be careful about word order here! */
4018 if (fits_in_unsigned_byte (i.tm.base_opcode))
4019 {
4020 FRAG_APPEND_1_CHAR (i.tm.base_opcode);
4021 }
4022 else
4023 {
4024 if ((i.tm.cpu_flags & (CpuSSSE3 | CpuSSE4)) != 0
4025 && (i.tm.cpu_flags & CpuABM) == 0)
4026 {
4027 p = frag_more (3);
4028 *p++ = (i.tm.base_opcode >> 16) & 0xff;
4029 }
4030 else
4031 p = frag_more (2);
4032
4033 /* Put out high byte first: can't use md_number_to_chars! */
4034 *p++ = (i.tm.base_opcode >> 8) & 0xff;
4035 *p = i.tm.base_opcode & 0xff;
4036 }
4037
4038 /* Now the modrm byte and sib byte (if present). */
4039 if (i.tm.opcode_modifier & Modrm)
4040 {
4041 p = frag_more (1);
4042 md_number_to_chars (p,
4043 (valueT) (i.rm.regmem << 0
4044 | i.rm.reg << 3
4045 | i.rm.mode << 6),
4046 1);
4047 /* If i.rm.regmem == ESP (4)
4048 && i.rm.mode != (Register mode)
4049 && not 16 bit
4050 ==> need second modrm byte. */
4051 if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING
4052 && i.rm.mode != 3
4053 && !(i.base_reg && (i.base_reg->reg_type & Reg16) != 0))
4054 {
4055 p = frag_more (1);
4056 md_number_to_chars (p,
4057 (valueT) (i.sib.base << 0
4058 | i.sib.index << 3
4059 | i.sib.scale << 6),
4060 1);
4061 }
4062 }
4063
4064 if (i.disp_operands)
4065 output_disp (insn_start_frag, insn_start_off);
4066
4067 if (i.imm_operands)
4068 output_imm (insn_start_frag, insn_start_off);
4069 }
4070
4071 #ifdef DEBUG386
4072 if (flag_debug)
4073 {
4074 pi ("" /*line*/, &i);
4075 }
4076 #endif /* DEBUG386 */
4077 }
4078
4079 /* Return the size of the displacement operand N. */
4080
4081 static int
4082 disp_size (unsigned int n)
4083 {
4084 int size = 4;
4085 if (i.types[n] & (Disp8 | Disp16 | Disp64))
4086 {
4087 size = 2;
4088 if (i.types[n] & Disp8)
4089 size = 1;
4090 if (i.types[n] & Disp64)
4091 size = 8;
4092 }
4093 return size;
4094 }
4095
4096 /* Return the size of the immediate operand N. */
4097
4098 static int
4099 imm_size (unsigned int n)
4100 {
4101 int size = 4;
4102 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
4103 {
4104 size = 2;
4105 if (i.types[n] & (Imm8 | Imm8S))
4106 size = 1;
4107 if (i.types[n] & Imm64)
4108 size = 8;
4109 }
4110 return size;
4111 }
4112
4113 static void
4114 output_disp (fragS *insn_start_frag, offsetT insn_start_off)
4115 {
4116 char *p;
4117 unsigned int n;
4118
4119 for (n = 0; n < i.operands; n++)
4120 {
4121 if (i.types[n] & Disp)
4122 {
4123 if (i.op[n].disps->X_op == O_constant)
4124 {
4125 int size = disp_size (n);
4126 offsetT val;
4127
4128 val = offset_in_range (i.op[n].disps->X_add_number,
4129 size);
4130 p = frag_more (size);
4131 md_number_to_chars (p, val, size);
4132 }
4133 else
4134 {
4135 enum bfd_reloc_code_real reloc_type;
4136 int size = disp_size (n);
4137 int sign = (i.types[n] & Disp32S) != 0;
4138 int pcrel = (i.flags[n] & Operand_PCrel) != 0;
4139
4140 /* We can't have 8 bit displacement here. */
4141 assert ((i.types[n] & Disp8) == 0);
4142
4143 /* The PC relative address is computed relative
4144 to the instruction boundary, so in case immediate
4145 fields follows, we need to adjust the value. */
4146 if (pcrel && i.imm_operands)
4147 {
4148 unsigned int n1;
4149 int sz = 0;
4150
4151 for (n1 = 0; n1 < i.operands; n1++)
4152 if (i.types[n1] & Imm)
4153 {
4154 /* Only one immediate is allowed for PC
4155 relative address. */
4156 assert (sz == 0);
4157 sz = imm_size (n1);
4158 i.op[n].disps->X_add_number -= sz;
4159 }
4160 /* We should find the immediate. */
4161 assert (sz != 0);
4162 }
4163
4164 p = frag_more (size);
4165 reloc_type = reloc (size, pcrel, sign, i.reloc[n]);
4166 if (GOT_symbol
4167 && GOT_symbol == i.op[n].disps->X_add_symbol
4168 && (((reloc_type == BFD_RELOC_32
4169 || reloc_type == BFD_RELOC_X86_64_32S
4170 || (reloc_type == BFD_RELOC_64
4171 && object_64bit))
4172 && (i.op[n].disps->X_op == O_symbol
4173 || (i.op[n].disps->X_op == O_add
4174 && ((symbol_get_value_expression
4175 (i.op[n].disps->X_op_symbol)->X_op)
4176 == O_subtract))))
4177 || reloc_type == BFD_RELOC_32_PCREL))
4178 {
4179 offsetT add;
4180
4181 if (insn_start_frag == frag_now)
4182 add = (p - frag_now->fr_literal) - insn_start_off;
4183 else
4184 {
4185 fragS *fr;
4186
4187 add = insn_start_frag->fr_fix - insn_start_off;
4188 for (fr = insn_start_frag->fr_next;
4189 fr && fr != frag_now; fr = fr->fr_next)
4190 add += fr->fr_fix;
4191 add += p - frag_now->fr_literal;
4192 }
4193
4194 if (!object_64bit)
4195 {
4196 reloc_type = BFD_RELOC_386_GOTPC;
4197 i.op[n].imms->X_add_number += add;
4198 }
4199 else if (reloc_type == BFD_RELOC_64)
4200 reloc_type = BFD_RELOC_X86_64_GOTPC64;
4201 else
4202 /* Don't do the adjustment for x86-64, as there
4203 the pcrel addressing is relative to the _next_
4204 insn, and that is taken care of in other code. */
4205 reloc_type = BFD_RELOC_X86_64_GOTPC32;
4206 }
4207 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
4208 i.op[n].disps, pcrel, reloc_type);
4209 }
4210 }
4211 }
4212 }
4213
4214 static void
4215 output_imm (fragS *insn_start_frag, offsetT insn_start_off)
4216 {
4217 char *p;
4218 unsigned int n;
4219
4220 for (n = 0; n < i.operands; n++)
4221 {
4222 if (i.types[n] & Imm)
4223 {
4224 if (i.op[n].imms->X_op == O_constant)
4225 {
4226 int size = imm_size (n);
4227 offsetT val;
4228
4229 val = offset_in_range (i.op[n].imms->X_add_number,
4230 size);
4231 p = frag_more (size);
4232 md_number_to_chars (p, val, size);
4233 }
4234 else
4235 {
4236 /* Not absolute_section.
4237 Need a 32-bit fixup (don't support 8bit
4238 non-absolute imms). Try to support other
4239 sizes ... */
4240 enum bfd_reloc_code_real reloc_type;
4241 int size = imm_size (n);
4242 int sign;
4243
4244 if ((i.types[n] & (Imm32S))
4245 && (i.suffix == QWORD_MNEM_SUFFIX
4246 || (!i.suffix && (i.tm.opcode_modifier & No_lSuf))))
4247 sign = 1;
4248 else
4249 sign = 0;
4250
4251 p = frag_more (size);
4252 reloc_type = reloc (size, 0, sign, i.reloc[n]);
4253
4254 /* This is tough to explain. We end up with this one if we
4255 * have operands that look like
4256 * "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal here is to
4257 * obtain the absolute address of the GOT, and it is strongly
4258 * preferable from a performance point of view to avoid using
4259 * a runtime relocation for this. The actual sequence of
4260 * instructions often look something like:
4261 *
4262 * call .L66
4263 * .L66:
4264 * popl %ebx
4265 * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
4266 *
4267 * The call and pop essentially return the absolute address
4268 * of the label .L66 and store it in %ebx. The linker itself
4269 * will ultimately change the first operand of the addl so
4270 * that %ebx points to the GOT, but to keep things simple, the
4271 * .o file must have this operand set so that it generates not
4272 * the absolute address of .L66, but the absolute address of
4273 * itself. This allows the linker itself simply treat a GOTPC
4274 * relocation as asking for a pcrel offset to the GOT to be
4275 * added in, and the addend of the relocation is stored in the
4276 * operand field for the instruction itself.
4277 *
4278 * Our job here is to fix the operand so that it would add
4279 * the correct offset so that %ebx would point to itself. The
4280 * thing that is tricky is that .-.L66 will point to the
4281 * beginning of the instruction, so we need to further modify
4282 * the operand so that it will point to itself. There are
4283 * other cases where you have something like:
4284 *
4285 * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
4286 *
4287 * and here no correction would be required. Internally in
4288 * the assembler we treat operands of this form as not being
4289 * pcrel since the '.' is explicitly mentioned, and I wonder
4290 * whether it would simplify matters to do it this way. Who
4291 * knows. In earlier versions of the PIC patches, the
4292 * pcrel_adjust field was used to store the correction, but
4293 * since the expression is not pcrel, I felt it would be
4294 * confusing to do it this way. */
4295
4296 if ((reloc_type == BFD_RELOC_32
4297 || reloc_type == BFD_RELOC_X86_64_32S
4298 || reloc_type == BFD_RELOC_64)
4299 && GOT_symbol
4300 && GOT_symbol == i.op[n].imms->X_add_symbol
4301 && (i.op[n].imms->X_op == O_symbol
4302 || (i.op[n].imms->X_op == O_add
4303 && ((symbol_get_value_expression
4304 (i.op[n].imms->X_op_symbol)->X_op)
4305 == O_subtract))))
4306 {
4307 offsetT add;
4308
4309 if (insn_start_frag == frag_now)
4310 add = (p - frag_now->fr_literal) - insn_start_off;
4311 else
4312 {
4313 fragS *fr;
4314
4315 add = insn_start_frag->fr_fix - insn_start_off;
4316 for (fr = insn_start_frag->fr_next;
4317 fr && fr != frag_now; fr = fr->fr_next)
4318 add += fr->fr_fix;
4319 add += p - frag_now->fr_literal;
4320 }
4321
4322 if (!object_64bit)
4323 reloc_type = BFD_RELOC_386_GOTPC;
4324 else if (size == 4)
4325 reloc_type = BFD_RELOC_X86_64_GOTPC32;
4326 else if (size == 8)
4327 reloc_type = BFD_RELOC_X86_64_GOTPC64;
4328 i.op[n].imms->X_add_number += add;
4329 }
4330 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
4331 i.op[n].imms, 0, reloc_type);
4332 }
4333 }
4334 }
4335 }
4336 \f
4337 /* x86_cons_fix_new is called via the expression parsing code when a
4338 reloc is needed. We use this hook to get the correct .got reloc. */
4339 static enum bfd_reloc_code_real got_reloc = NO_RELOC;
4340 static int cons_sign = -1;
4341
4342 void
4343 x86_cons_fix_new (fragS *frag, unsigned int off, unsigned int len,
4344 expressionS *exp)
4345 {
4346 enum bfd_reloc_code_real r = reloc (len, 0, cons_sign, got_reloc);
4347
4348 got_reloc = NO_RELOC;
4349
4350 #ifdef TE_PE
4351 if (exp->X_op == O_secrel)
4352 {
4353 exp->X_op = O_symbol;
4354 r = BFD_RELOC_32_SECREL;
4355 }
4356 #endif
4357
4358 fix_new_exp (frag, off, len, exp, 0, r);
4359 }
4360
4361 #if (!defined (OBJ_ELF) && !defined (OBJ_MAYBE_ELF)) || defined (LEX_AT)
4362 # define lex_got(reloc, adjust, types) NULL
4363 #else
4364 /* Parse operands of the form
4365 <symbol>@GOTOFF+<nnn>
4366 and similar .plt or .got references.
4367
4368 If we find one, set up the correct relocation in RELOC and copy the
4369 input string, minus the `@GOTOFF' into a malloc'd buffer for
4370 parsing by the calling routine. Return this buffer, and if ADJUST
4371 is non-null set it to the length of the string we removed from the
4372 input line. Otherwise return NULL. */
4373 static char *
4374 lex_got (enum bfd_reloc_code_real *reloc,
4375 int *adjust,
4376 unsigned int *types)
4377 {
4378 /* Some of the relocations depend on the size of what field is to
4379 be relocated. But in our callers i386_immediate and i386_displacement
4380 we don't yet know the operand size (this will be set by insn
4381 matching). Hence we record the word32 relocation here,
4382 and adjust the reloc according to the real size in reloc(). */
4383 static const struct {
4384 const char *str;
4385 const enum bfd_reloc_code_real rel[2];
4386 const unsigned int types64;
4387 } gotrel[] = {
4388 { "PLTOFF", { 0,
4389 BFD_RELOC_X86_64_PLTOFF64 },
4390 Imm64 },
4391 { "PLT", { BFD_RELOC_386_PLT32,
4392 BFD_RELOC_X86_64_PLT32 },
4393 Imm32 | Imm32S | Disp32 },
4394 { "GOTPLT", { 0,
4395 BFD_RELOC_X86_64_GOTPLT64 },
4396 Imm64 | Disp64 },
4397 { "GOTOFF", { BFD_RELOC_386_GOTOFF,
4398 BFD_RELOC_X86_64_GOTOFF64 },
4399 Imm64 | Disp64 },
4400 { "GOTPCREL", { 0,
4401 BFD_RELOC_X86_64_GOTPCREL },
4402 Imm32 | Imm32S | Disp32 },
4403 { "TLSGD", { BFD_RELOC_386_TLS_GD,
4404 BFD_RELOC_X86_64_TLSGD },
4405 Imm32 | Imm32S | Disp32 },
4406 { "TLSLDM", { BFD_RELOC_386_TLS_LDM,
4407 0 },
4408 0 },
4409 { "TLSLD", { 0,
4410 BFD_RELOC_X86_64_TLSLD },
4411 Imm32 | Imm32S | Disp32 },
4412 { "GOTTPOFF", { BFD_RELOC_386_TLS_IE_32,
4413 BFD_RELOC_X86_64_GOTTPOFF },
4414 Imm32 | Imm32S | Disp32 },
4415 { "TPOFF", { BFD_RELOC_386_TLS_LE_32,
4416 BFD_RELOC_X86_64_TPOFF32 },
4417 Imm32 | Imm32S | Imm64 | Disp32 | Disp64 },
4418 { "NTPOFF", { BFD_RELOC_386_TLS_LE,
4419 0 },
4420 0 },
4421 { "DTPOFF", { BFD_RELOC_386_TLS_LDO_32,
4422 BFD_RELOC_X86_64_DTPOFF32 },
4423 Imm32 | Imm32S | Imm64 | Disp32 | Disp64 },
4424 { "GOTNTPOFF",{ BFD_RELOC_386_TLS_GOTIE,
4425 0 },
4426 0 },
4427 { "INDNTPOFF",{ BFD_RELOC_386_TLS_IE,
4428 0 },
4429 0 },
4430 { "GOT", { BFD_RELOC_386_GOT32,
4431 BFD_RELOC_X86_64_GOT32 },
4432 Imm32 | Imm32S | Disp32 | Imm64 },
4433 { "TLSDESC", { BFD_RELOC_386_TLS_GOTDESC,
4434 BFD_RELOC_X86_64_GOTPC32_TLSDESC },
4435 Imm32 | Imm32S | Disp32 },
4436 { "TLSCALL", { BFD_RELOC_386_TLS_DESC_CALL,
4437 BFD_RELOC_X86_64_TLSDESC_CALL },
4438 Imm32 | Imm32S | Disp32 }
4439 };
4440 char *cp;
4441 unsigned int j;
4442
4443 if (!IS_ELF)
4444 return NULL;
4445
4446 for (cp = input_line_pointer; *cp != '@'; cp++)
4447 if (is_end_of_line[(unsigned char) *cp])
4448 return NULL;
4449
4450 for (j = 0; j < sizeof (gotrel) / sizeof (gotrel[0]); j++)
4451 {
4452 int len;
4453
4454 len = strlen (gotrel[j].str);
4455 if (strncasecmp (cp + 1, gotrel[j].str, len) == 0)
4456 {
4457 if (gotrel[j].rel[object_64bit] != 0)
4458 {
4459 int first, second;
4460 char *tmpbuf, *past_reloc;
4461
4462 *reloc = gotrel[j].rel[object_64bit];
4463 if (adjust)
4464 *adjust = len;
4465
4466 if (types)
4467 {
4468 if (flag_code != CODE_64BIT)
4469 *types = Imm32 | Disp32;
4470 else
4471 *types = gotrel[j].types64;
4472 }
4473
4474 if (GOT_symbol == NULL)
4475 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
4476
4477 /* The length of the first part of our input line. */
4478 first = cp - input_line_pointer;
4479
4480 /* The second part goes from after the reloc token until
4481 (and including) an end_of_line char. Don't use strlen
4482 here as the end_of_line char may not be a NUL. */
4483 past_reloc = cp + 1 + len;
4484 for (cp = past_reloc; !is_end_of_line[(unsigned char) *cp++]; )
4485 ;
4486 second = cp - past_reloc;
4487
4488 /* Allocate and copy string. The trailing NUL shouldn't
4489 be necessary, but be safe. */
4490 tmpbuf = xmalloc (first + second + 2);
4491 memcpy (tmpbuf, input_line_pointer, first);
4492 if (second != 0 && *past_reloc != ' ')
4493 /* Replace the relocation token with ' ', so that
4494 errors like foo@GOTOFF1 will be detected. */
4495 tmpbuf[first++] = ' ';
4496 memcpy (tmpbuf + first, past_reloc, second);
4497 tmpbuf[first + second] = '\0';
4498 return tmpbuf;
4499 }
4500
4501 as_bad (_("@%s reloc is not supported with %d-bit output format"),
4502 gotrel[j].str, 1 << (5 + object_64bit));
4503 return NULL;
4504 }
4505 }
4506
4507 /* Might be a symbol version string. Don't as_bad here. */
4508 return NULL;
4509 }
4510
4511 void
4512 x86_cons (expressionS *exp, int size)
4513 {
4514 if (size == 4 || (object_64bit && size == 8))
4515 {
4516 /* Handle @GOTOFF and the like in an expression. */
4517 char *save;
4518 char *gotfree_input_line;
4519 int adjust;
4520
4521 save = input_line_pointer;
4522 gotfree_input_line = lex_got (&got_reloc, &adjust, NULL);
4523 if (gotfree_input_line)
4524 input_line_pointer = gotfree_input_line;
4525
4526 expression (exp);
4527
4528 if (gotfree_input_line)
4529 {
4530 /* expression () has merrily parsed up to the end of line,
4531 or a comma - in the wrong buffer. Transfer how far
4532 input_line_pointer has moved to the right buffer. */
4533 input_line_pointer = (save
4534 + (input_line_pointer - gotfree_input_line)
4535 + adjust);
4536 free (gotfree_input_line);
4537 }
4538 }
4539 else
4540 expression (exp);
4541 }
4542 #endif
4543
4544 static void signed_cons (int size)
4545 {
4546 if (flag_code == CODE_64BIT)
4547 cons_sign = 1;
4548 cons (size);
4549 cons_sign = -1;
4550 }
4551
4552 #ifdef TE_PE
4553 static void
4554 pe_directive_secrel (dummy)
4555 int dummy ATTRIBUTE_UNUSED;
4556 {
4557 expressionS exp;
4558
4559 do
4560 {
4561 expression (&exp);
4562 if (exp.X_op == O_symbol)
4563 exp.X_op = O_secrel;
4564
4565 emit_expr (&exp, 4);
4566 }
4567 while (*input_line_pointer++ == ',');
4568
4569 input_line_pointer--;
4570 demand_empty_rest_of_line ();
4571 }
4572 #endif
4573
4574 static int
4575 i386_immediate (char *imm_start)
4576 {
4577 char *save_input_line_pointer;
4578 char *gotfree_input_line;
4579 segT exp_seg = 0;
4580 expressionS *exp;
4581 unsigned int types = ~0U;
4582
4583 if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
4584 {
4585 as_bad (_("at most %d immediate operands are allowed"),
4586 MAX_IMMEDIATE_OPERANDS);
4587 return 0;
4588 }
4589
4590 exp = &im_expressions[i.imm_operands++];
4591 i.op[this_operand].imms = exp;
4592
4593 if (is_space_char (*imm_start))
4594 ++imm_start;
4595
4596 save_input_line_pointer = input_line_pointer;
4597 input_line_pointer = imm_start;
4598
4599 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL, &types);
4600 if (gotfree_input_line)
4601 input_line_pointer = gotfree_input_line;
4602
4603 exp_seg = expression (exp);
4604
4605 SKIP_WHITESPACE ();
4606 if (*input_line_pointer)
4607 as_bad (_("junk `%s' after expression"), input_line_pointer);
4608
4609 input_line_pointer = save_input_line_pointer;
4610 if (gotfree_input_line)
4611 free (gotfree_input_line);
4612
4613 if (exp->X_op == O_absent || exp->X_op == O_big)
4614 {
4615 /* Missing or bad expr becomes absolute 0. */
4616 as_bad (_("missing or invalid immediate expression `%s' taken as 0"),
4617 imm_start);
4618 exp->X_op = O_constant;
4619 exp->X_add_number = 0;
4620 exp->X_add_symbol = (symbolS *) 0;
4621 exp->X_op_symbol = (symbolS *) 0;
4622 }
4623 else if (exp->X_op == O_constant)
4624 {
4625 /* Size it properly later. */
4626 i.types[this_operand] |= Imm64;
4627 /* If BFD64, sign extend val. */
4628 if (!use_rela_relocations
4629 && (exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0)
4630 exp->X_add_number
4631 = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
4632 }
4633 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4634 else if (OUTPUT_FLAVOR == bfd_target_aout_flavour
4635 && exp_seg != absolute_section
4636 && exp_seg != text_section
4637 && exp_seg != data_section
4638 && exp_seg != bss_section
4639 && exp_seg != undefined_section
4640 && !bfd_is_com_section (exp_seg))
4641 {
4642 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
4643 return 0;
4644 }
4645 #endif
4646 else if (!intel_syntax && exp->X_op == O_register)
4647 {
4648 as_bad (_("illegal immediate register operand %s"), imm_start);
4649 return 0;
4650 }
4651 else
4652 {
4653 /* This is an address. The size of the address will be
4654 determined later, depending on destination register,
4655 suffix, or the default for the section. */
4656 i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64;
4657 i.types[this_operand] &= types;
4658 }
4659
4660 return 1;
4661 }
4662
4663 static char *
4664 i386_scale (char *scale)
4665 {
4666 offsetT val;
4667 char *save = input_line_pointer;
4668
4669 input_line_pointer = scale;
4670 val = get_absolute_expression ();
4671
4672 switch (val)
4673 {
4674 case 1:
4675 i.log2_scale_factor = 0;
4676 break;
4677 case 2:
4678 i.log2_scale_factor = 1;
4679 break;
4680 case 4:
4681 i.log2_scale_factor = 2;
4682 break;
4683 case 8:
4684 i.log2_scale_factor = 3;
4685 break;
4686 default:
4687 {
4688 char sep = *input_line_pointer;
4689
4690 *input_line_pointer = '\0';
4691 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
4692 scale);
4693 *input_line_pointer = sep;
4694 input_line_pointer = save;
4695 return NULL;
4696 }
4697 }
4698 if (i.log2_scale_factor != 0 && i.index_reg == 0)
4699 {
4700 as_warn (_("scale factor of %d without an index register"),
4701 1 << i.log2_scale_factor);
4702 #if SCALE1_WHEN_NO_INDEX
4703 i.log2_scale_factor = 0;
4704 #endif
4705 }
4706 scale = input_line_pointer;
4707 input_line_pointer = save;
4708 return scale;
4709 }
4710
4711 static int
4712 i386_displacement (char *disp_start, char *disp_end)
4713 {
4714 expressionS *exp;
4715 segT exp_seg = 0;
4716 char *save_input_line_pointer;
4717 char *gotfree_input_line;
4718 int bigdisp, override;
4719 unsigned int types = Disp;
4720
4721 if (i.disp_operands == MAX_MEMORY_OPERANDS)
4722 {
4723 as_bad (_("at most %d displacement operands are allowed"),
4724 MAX_MEMORY_OPERANDS);
4725 return 0;
4726 }
4727
4728 if ((i.types[this_operand] & JumpAbsolute)
4729 || !(current_templates->start->opcode_modifier & (Jump | JumpDword)))
4730 {
4731 bigdisp = Disp32;
4732 override = (i.prefix[ADDR_PREFIX] != 0);
4733 }
4734 else
4735 {
4736 /* For PC-relative branches, the width of the displacement
4737 is dependent upon data size, not address size. */
4738 bigdisp = 0;
4739 override = (i.prefix[DATA_PREFIX] != 0);
4740 }
4741 if (flag_code == CODE_64BIT)
4742 {
4743 if (!bigdisp)
4744 bigdisp = ((override || i.suffix == WORD_MNEM_SUFFIX)
4745 ? Disp16
4746 : Disp32S | Disp32);
4747 else if (!override)
4748 bigdisp = Disp64 | Disp32S | Disp32;
4749 }
4750 else
4751 {
4752 if (!bigdisp)
4753 {
4754 if (!override)
4755 override = (i.suffix == (flag_code != CODE_16BIT
4756 ? WORD_MNEM_SUFFIX
4757 : LONG_MNEM_SUFFIX));
4758 bigdisp = Disp32;
4759 }
4760 if ((flag_code == CODE_16BIT) ^ override)
4761 bigdisp = Disp16;
4762 }
4763 i.types[this_operand] |= bigdisp;
4764
4765 exp = &disp_expressions[i.disp_operands];
4766 i.op[this_operand].disps = exp;
4767 i.disp_operands++;
4768 save_input_line_pointer = input_line_pointer;
4769 input_line_pointer = disp_start;
4770 END_STRING_AND_SAVE (disp_end);
4771
4772 #ifndef GCC_ASM_O_HACK
4773 #define GCC_ASM_O_HACK 0
4774 #endif
4775 #if GCC_ASM_O_HACK
4776 END_STRING_AND_SAVE (disp_end + 1);
4777 if ((i.types[this_operand] & BaseIndex) != 0
4778 && displacement_string_end[-1] == '+')
4779 {
4780 /* This hack is to avoid a warning when using the "o"
4781 constraint within gcc asm statements.
4782 For instance:
4783
4784 #define _set_tssldt_desc(n,addr,limit,type) \
4785 __asm__ __volatile__ ( \
4786 "movw %w2,%0\n\t" \
4787 "movw %w1,2+%0\n\t" \
4788 "rorl $16,%1\n\t" \
4789 "movb %b1,4+%0\n\t" \
4790 "movb %4,5+%0\n\t" \
4791 "movb $0,6+%0\n\t" \
4792 "movb %h1,7+%0\n\t" \
4793 "rorl $16,%1" \
4794 : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
4795
4796 This works great except that the output assembler ends
4797 up looking a bit weird if it turns out that there is
4798 no offset. You end up producing code that looks like:
4799
4800 #APP
4801 movw $235,(%eax)
4802 movw %dx,2+(%eax)
4803 rorl $16,%edx
4804 movb %dl,4+(%eax)
4805 movb $137,5+(%eax)
4806 movb $0,6+(%eax)
4807 movb %dh,7+(%eax)
4808 rorl $16,%edx
4809 #NO_APP
4810
4811 So here we provide the missing zero. */
4812
4813 *displacement_string_end = '0';
4814 }
4815 #endif
4816 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL, &types);
4817 if (gotfree_input_line)
4818 input_line_pointer = gotfree_input_line;
4819
4820 exp_seg = expression (exp);
4821
4822 SKIP_WHITESPACE ();
4823 if (*input_line_pointer)
4824 as_bad (_("junk `%s' after expression"), input_line_pointer);
4825 #if GCC_ASM_O_HACK
4826 RESTORE_END_STRING (disp_end + 1);
4827 #endif
4828 RESTORE_END_STRING (disp_end);
4829 input_line_pointer = save_input_line_pointer;
4830 if (gotfree_input_line)
4831 free (gotfree_input_line);
4832
4833 /* We do this to make sure that the section symbol is in
4834 the symbol table. We will ultimately change the relocation
4835 to be relative to the beginning of the section. */
4836 if (i.reloc[this_operand] == BFD_RELOC_386_GOTOFF
4837 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
4838 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTOFF64)
4839 {
4840 if (exp->X_op != O_symbol)
4841 {
4842 as_bad (_("bad expression used with @%s"),
4843 (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
4844 ? "GOTPCREL"
4845 : "GOTOFF"));
4846 return 0;
4847 }
4848
4849 if (S_IS_LOCAL (exp->X_add_symbol)
4850 && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
4851 section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
4852 exp->X_op = O_subtract;
4853 exp->X_op_symbol = GOT_symbol;
4854 if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
4855 i.reloc[this_operand] = BFD_RELOC_32_PCREL;
4856 else if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTOFF64)
4857 i.reloc[this_operand] = BFD_RELOC_64;
4858 else
4859 i.reloc[this_operand] = BFD_RELOC_32;
4860 }
4861
4862 if (exp->X_op == O_absent || exp->X_op == O_big)
4863 {
4864 /* Missing or bad expr becomes absolute 0. */
4865 as_bad (_("missing or invalid displacement expression `%s' taken as 0"),
4866 disp_start);
4867 exp->X_op = O_constant;
4868 exp->X_add_number = 0;
4869 exp->X_add_symbol = (symbolS *) 0;
4870 exp->X_op_symbol = (symbolS *) 0;
4871 }
4872
4873 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4874 if (exp->X_op != O_constant
4875 && OUTPUT_FLAVOR == bfd_target_aout_flavour
4876 && exp_seg != absolute_section
4877 && exp_seg != text_section
4878 && exp_seg != data_section
4879 && exp_seg != bss_section
4880 && exp_seg != undefined_section
4881 && !bfd_is_com_section (exp_seg))
4882 {
4883 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
4884 return 0;
4885 }
4886 #endif
4887
4888 if (!(i.types[this_operand] & ~Disp))
4889 i.types[this_operand] &= types;
4890
4891 return 1;
4892 }
4893
4894 /* Make sure the memory operand we've been dealt is valid.
4895 Return 1 on success, 0 on a failure. */
4896
4897 static int
4898 i386_index_check (const char *operand_string)
4899 {
4900 int ok;
4901 #if INFER_ADDR_PREFIX
4902 int fudged = 0;
4903
4904 tryprefix:
4905 #endif
4906 ok = 1;
4907 if ((current_templates->start->cpu_flags & CpuSVME)
4908 && current_templates->end[-1].operand_types[0] == AnyMem)
4909 {
4910 /* Memory operands of SVME insns are special in that they only allow
4911 rAX as their memory address and ignore any segment override. */
4912 unsigned RegXX;
4913
4914 /* SKINIT is even more restrictive: it always requires EAX. */
4915 if (strcmp (current_templates->start->name, "skinit") == 0)
4916 RegXX = Reg32;
4917 else if (flag_code == CODE_64BIT)
4918 RegXX = i.prefix[ADDR_PREFIX] == 0 ? Reg64 : Reg32;
4919 else
4920 RegXX = ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0)
4921 ? Reg16
4922 : Reg32);
4923 if (!i.base_reg
4924 || !(i.base_reg->reg_type & Acc)
4925 || !(i.base_reg->reg_type & RegXX)
4926 || i.index_reg
4927 || (i.types[0] & Disp))
4928 ok = 0;
4929 }
4930 else if (flag_code == CODE_64BIT)
4931 {
4932 unsigned RegXX = (i.prefix[ADDR_PREFIX] == 0 ? Reg64 : Reg32);
4933
4934 if ((i.base_reg
4935 && ((i.base_reg->reg_type & RegXX) == 0)
4936 && (i.base_reg->reg_type != BaseIndex
4937 || i.index_reg))
4938 || (i.index_reg
4939 && ((i.index_reg->reg_type & (RegXX | BaseIndex))
4940 != (RegXX | BaseIndex))))
4941 ok = 0;
4942 }
4943 else
4944 {
4945 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
4946 {
4947 /* 16bit checks. */
4948 if ((i.base_reg
4949 && ((i.base_reg->reg_type & (Reg16 | BaseIndex | RegRex))
4950 != (Reg16 | BaseIndex)))
4951 || (i.index_reg
4952 && (((i.index_reg->reg_type & (Reg16 | BaseIndex))
4953 != (Reg16 | BaseIndex))
4954 || !(i.base_reg
4955 && i.base_reg->reg_num < 6
4956 && i.index_reg->reg_num >= 6
4957 && i.log2_scale_factor == 0))))
4958 ok = 0;
4959 }
4960 else
4961 {
4962 /* 32bit checks. */
4963 if ((i.base_reg
4964 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
4965 || (i.index_reg
4966 && ((i.index_reg->reg_type & (Reg32 | BaseIndex | RegRex))
4967 != (Reg32 | BaseIndex))))
4968 ok = 0;
4969 }
4970 }
4971 if (!ok)
4972 {
4973 #if INFER_ADDR_PREFIX
4974 if (i.prefix[ADDR_PREFIX] == 0)
4975 {
4976 i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
4977 i.prefixes += 1;
4978 /* Change the size of any displacement too. At most one of
4979 Disp16 or Disp32 is set.
4980 FIXME. There doesn't seem to be any real need for separate
4981 Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
4982 Removing them would probably clean up the code quite a lot. */
4983 if (flag_code != CODE_64BIT
4984 && (i.types[this_operand] & (Disp16 | Disp32)))
4985 i.types[this_operand] ^= (Disp16 | Disp32);
4986 fudged = 1;
4987 goto tryprefix;
4988 }
4989 if (fudged)
4990 as_bad (_("`%s' is not a valid base/index expression"),
4991 operand_string);
4992 else
4993 #endif
4994 as_bad (_("`%s' is not a valid %s bit base/index expression"),
4995 operand_string,
4996 flag_code_names[flag_code]);
4997 }
4998 return ok;
4999 }
5000
5001 /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
5002 on error. */
5003
5004 static int
5005 i386_operand (char *operand_string)
5006 {
5007 const reg_entry *r;
5008 char *end_op;
5009 char *op_string = operand_string;
5010
5011 if (is_space_char (*op_string))
5012 ++op_string;
5013
5014 /* We check for an absolute prefix (differentiating,
5015 for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
5016 if (*op_string == ABSOLUTE_PREFIX)
5017 {
5018 ++op_string;
5019 if (is_space_char (*op_string))
5020 ++op_string;
5021 i.types[this_operand] |= JumpAbsolute;
5022 }
5023
5024 /* Check if operand is a register. */
5025 if ((r = parse_register (op_string, &end_op)) != NULL)
5026 {
5027 /* Check for a segment override by searching for ':' after a
5028 segment register. */
5029 op_string = end_op;
5030 if (is_space_char (*op_string))
5031 ++op_string;
5032 if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
5033 {
5034 switch (r->reg_num)
5035 {
5036 case 0:
5037 i.seg[i.mem_operands] = &es;
5038 break;
5039 case 1:
5040 i.seg[i.mem_operands] = &cs;
5041 break;
5042 case 2:
5043 i.seg[i.mem_operands] = &ss;
5044 break;
5045 case 3:
5046 i.seg[i.mem_operands] = &ds;
5047 break;
5048 case 4:
5049 i.seg[i.mem_operands] = &fs;
5050 break;
5051 case 5:
5052 i.seg[i.mem_operands] = &gs;
5053 break;
5054 }
5055
5056 /* Skip the ':' and whitespace. */
5057 ++op_string;
5058 if (is_space_char (*op_string))
5059 ++op_string;
5060
5061 if (!is_digit_char (*op_string)
5062 && !is_identifier_char (*op_string)
5063 && *op_string != '('
5064 && *op_string != ABSOLUTE_PREFIX)
5065 {
5066 as_bad (_("bad memory operand `%s'"), op_string);
5067 return 0;
5068 }
5069 /* Handle case of %es:*foo. */
5070 if (*op_string == ABSOLUTE_PREFIX)
5071 {
5072 ++op_string;
5073 if (is_space_char (*op_string))
5074 ++op_string;
5075 i.types[this_operand] |= JumpAbsolute;
5076 }
5077 goto do_memory_reference;
5078 }
5079 if (*op_string)
5080 {
5081 as_bad (_("junk `%s' after register"), op_string);
5082 return 0;
5083 }
5084 i.types[this_operand] |= r->reg_type & ~BaseIndex;
5085 i.op[this_operand].regs = r;
5086 i.reg_operands++;
5087 }
5088 else if (*op_string == REGISTER_PREFIX)
5089 {
5090 as_bad (_("bad register name `%s'"), op_string);
5091 return 0;
5092 }
5093 else if (*op_string == IMMEDIATE_PREFIX)
5094 {
5095 ++op_string;
5096 if (i.types[this_operand] & JumpAbsolute)
5097 {
5098 as_bad (_("immediate operand illegal with absolute jump"));
5099 return 0;
5100 }
5101 if (!i386_immediate (op_string))
5102 return 0;
5103 }
5104 else if (is_digit_char (*op_string)
5105 || is_identifier_char (*op_string)
5106 || *op_string == '(')
5107 {
5108 /* This is a memory reference of some sort. */
5109 char *base_string;
5110
5111 /* Start and end of displacement string expression (if found). */
5112 char *displacement_string_start;
5113 char *displacement_string_end;
5114
5115 do_memory_reference:
5116 if ((i.mem_operands == 1
5117 && (current_templates->start->opcode_modifier & IsString) == 0)
5118 || i.mem_operands == 2)
5119 {
5120 as_bad (_("too many memory references for `%s'"),
5121 current_templates->start->name);
5122 return 0;
5123 }
5124
5125 /* Check for base index form. We detect the base index form by
5126 looking for an ')' at the end of the operand, searching
5127 for the '(' matching it, and finding a REGISTER_PREFIX or ','
5128 after the '('. */
5129 base_string = op_string + strlen (op_string);
5130
5131 --base_string;
5132 if (is_space_char (*base_string))
5133 --base_string;
5134
5135 /* If we only have a displacement, set-up for it to be parsed later. */
5136 displacement_string_start = op_string;
5137 displacement_string_end = base_string + 1;
5138
5139 if (*base_string == ')')
5140 {
5141 char *temp_string;
5142 unsigned int parens_balanced = 1;
5143 /* We've already checked that the number of left & right ()'s are
5144 equal, so this loop will not be infinite. */
5145 do
5146 {
5147 base_string--;
5148 if (*base_string == ')')
5149 parens_balanced++;
5150 if (*base_string == '(')
5151 parens_balanced--;
5152 }
5153 while (parens_balanced);
5154
5155 temp_string = base_string;
5156
5157 /* Skip past '(' and whitespace. */
5158 ++base_string;
5159 if (is_space_char (*base_string))
5160 ++base_string;
5161
5162 if (*base_string == ','
5163 || ((i.base_reg = parse_register (base_string, &end_op))
5164 != NULL))
5165 {
5166 displacement_string_end = temp_string;
5167
5168 i.types[this_operand] |= BaseIndex;
5169
5170 if (i.base_reg)
5171 {
5172 base_string = end_op;
5173 if (is_space_char (*base_string))
5174 ++base_string;
5175 }
5176
5177 /* There may be an index reg or scale factor here. */
5178 if (*base_string == ',')
5179 {
5180 ++base_string;
5181 if (is_space_char (*base_string))
5182 ++base_string;
5183
5184 if ((i.index_reg = parse_register (base_string, &end_op))
5185 != NULL)
5186 {
5187 base_string = end_op;
5188 if (is_space_char (*base_string))
5189 ++base_string;
5190 if (*base_string == ',')
5191 {
5192 ++base_string;
5193 if (is_space_char (*base_string))
5194 ++base_string;
5195 }
5196 else if (*base_string != ')')
5197 {
5198 as_bad (_("expecting `,' or `)' "
5199 "after index register in `%s'"),
5200 operand_string);
5201 return 0;
5202 }
5203 }
5204 else if (*base_string == REGISTER_PREFIX)
5205 {
5206 as_bad (_("bad register name `%s'"), base_string);
5207 return 0;
5208 }
5209
5210 /* Check for scale factor. */
5211 if (*base_string != ')')
5212 {
5213 char *end_scale = i386_scale (base_string);
5214
5215 if (!end_scale)
5216 return 0;
5217
5218 base_string = end_scale;
5219 if (is_space_char (*base_string))
5220 ++base_string;
5221 if (*base_string != ')')
5222 {
5223 as_bad (_("expecting `)' "
5224 "after scale factor in `%s'"),
5225 operand_string);
5226 return 0;
5227 }
5228 }
5229 else if (!i.index_reg)
5230 {
5231 as_bad (_("expecting index register or scale factor "
5232 "after `,'; got '%c'"),
5233 *base_string);
5234 return 0;
5235 }
5236 }
5237 else if (*base_string != ')')
5238 {
5239 as_bad (_("expecting `,' or `)' "
5240 "after base register in `%s'"),
5241 operand_string);
5242 return 0;
5243 }
5244 }
5245 else if (*base_string == REGISTER_PREFIX)
5246 {
5247 as_bad (_("bad register name `%s'"), base_string);
5248 return 0;
5249 }
5250 }
5251
5252 /* If there's an expression beginning the operand, parse it,
5253 assuming displacement_string_start and
5254 displacement_string_end are meaningful. */
5255 if (displacement_string_start != displacement_string_end)
5256 {
5257 if (!i386_displacement (displacement_string_start,
5258 displacement_string_end))
5259 return 0;
5260 }
5261
5262 /* Special case for (%dx) while doing input/output op. */
5263 if (i.base_reg
5264 && i.base_reg->reg_type == (Reg16 | InOutPortReg)
5265 && i.index_reg == 0
5266 && i.log2_scale_factor == 0
5267 && i.seg[i.mem_operands] == 0
5268 && (i.types[this_operand] & Disp) == 0)
5269 {
5270 i.types[this_operand] = InOutPortReg;
5271 return 1;
5272 }
5273
5274 if (i386_index_check (operand_string) == 0)
5275 return 0;
5276 i.mem_operands++;
5277 }
5278 else
5279 {
5280 /* It's not a memory operand; argh! */
5281 as_bad (_("invalid char %s beginning operand %d `%s'"),
5282 output_invalid (*op_string),
5283 this_operand + 1,
5284 op_string);
5285 return 0;
5286 }
5287 return 1; /* Normal return. */
5288 }
5289 \f
5290 /* md_estimate_size_before_relax()
5291
5292 Called just before relax() for rs_machine_dependent frags. The x86
5293 assembler uses these frags to handle variable size jump
5294 instructions.
5295
5296 Any symbol that is now undefined will not become defined.
5297 Return the correct fr_subtype in the frag.
5298 Return the initial "guess for variable size of frag" to caller.
5299 The guess is actually the growth beyond the fixed part. Whatever
5300 we do to grow the fixed or variable part contributes to our
5301 returned value. */
5302
5303 int
5304 md_estimate_size_before_relax (fragP, segment)
5305 fragS *fragP;
5306 segT segment;
5307 {
5308 /* We've already got fragP->fr_subtype right; all we have to do is
5309 check for un-relaxable symbols. On an ELF system, we can't relax
5310 an externally visible symbol, because it may be overridden by a
5311 shared library. */
5312 if (S_GET_SEGMENT (fragP->fr_symbol) != segment
5313 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5314 || (IS_ELF
5315 && (S_IS_EXTERNAL (fragP->fr_symbol)
5316 || S_IS_WEAK (fragP->fr_symbol)))
5317 #endif
5318 )
5319 {
5320 /* Symbol is undefined in this segment, or we need to keep a
5321 reloc so that weak symbols can be overridden. */
5322 int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
5323 enum bfd_reloc_code_real reloc_type;
5324 unsigned char *opcode;
5325 int old_fr_fix;
5326
5327 if (fragP->fr_var != NO_RELOC)
5328 reloc_type = fragP->fr_var;
5329 else if (size == 2)
5330 reloc_type = BFD_RELOC_16_PCREL;
5331 else
5332 reloc_type = BFD_RELOC_32_PCREL;
5333
5334 old_fr_fix = fragP->fr_fix;
5335 opcode = (unsigned char *) fragP->fr_opcode;
5336
5337 switch (TYPE_FROM_RELAX_STATE (fragP->fr_subtype))
5338 {
5339 case UNCOND_JUMP:
5340 /* Make jmp (0xeb) a (d)word displacement jump. */
5341 opcode[0] = 0xe9;
5342 fragP->fr_fix += size;
5343 fix_new (fragP, old_fr_fix, size,
5344 fragP->fr_symbol,
5345 fragP->fr_offset, 1,
5346 reloc_type);
5347 break;
5348
5349 case COND_JUMP86:
5350 if (size == 2
5351 && (!no_cond_jump_promotion || fragP->fr_var != NO_RELOC))
5352 {
5353 /* Negate the condition, and branch past an
5354 unconditional jump. */
5355 opcode[0] ^= 1;
5356 opcode[1] = 3;
5357 /* Insert an unconditional jump. */
5358 opcode[2] = 0xe9;
5359 /* We added two extra opcode bytes, and have a two byte
5360 offset. */
5361 fragP->fr_fix += 2 + 2;
5362 fix_new (fragP, old_fr_fix + 2, 2,
5363 fragP->fr_symbol,
5364 fragP->fr_offset, 1,
5365 reloc_type);
5366 break;
5367 }
5368 /* Fall through. */
5369
5370 case COND_JUMP:
5371 if (no_cond_jump_promotion && fragP->fr_var == NO_RELOC)
5372 {
5373 fixS *fixP;
5374
5375 fragP->fr_fix += 1;
5376 fixP = fix_new (fragP, old_fr_fix, 1,
5377 fragP->fr_symbol,
5378 fragP->fr_offset, 1,
5379 BFD_RELOC_8_PCREL);
5380 fixP->fx_signed = 1;
5381 break;
5382 }
5383
5384 /* This changes the byte-displacement jump 0x7N
5385 to the (d)word-displacement jump 0x0f,0x8N. */
5386 opcode[1] = opcode[0] + 0x10;
5387 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
5388 /* We've added an opcode byte. */
5389 fragP->fr_fix += 1 + size;
5390 fix_new (fragP, old_fr_fix + 1, size,
5391 fragP->fr_symbol,
5392 fragP->fr_offset, 1,
5393 reloc_type);
5394 break;
5395
5396 default:
5397 BAD_CASE (fragP->fr_subtype);
5398 break;
5399 }
5400 frag_wane (fragP);
5401 return fragP->fr_fix - old_fr_fix;
5402 }
5403
5404 /* Guess size depending on current relax state. Initially the relax
5405 state will correspond to a short jump and we return 1, because
5406 the variable part of the frag (the branch offset) is one byte
5407 long. However, we can relax a section more than once and in that
5408 case we must either set fr_subtype back to the unrelaxed state,
5409 or return the value for the appropriate branch. */
5410 return md_relax_table[fragP->fr_subtype].rlx_length;
5411 }
5412
5413 /* Called after relax() is finished.
5414
5415 In: Address of frag.
5416 fr_type == rs_machine_dependent.
5417 fr_subtype is what the address relaxed to.
5418
5419 Out: Any fixSs and constants are set up.
5420 Caller will turn frag into a ".space 0". */
5421
5422 void
5423 md_convert_frag (abfd, sec, fragP)
5424 bfd *abfd ATTRIBUTE_UNUSED;
5425 segT sec ATTRIBUTE_UNUSED;
5426 fragS *fragP;
5427 {
5428 unsigned char *opcode;
5429 unsigned char *where_to_put_displacement = NULL;
5430 offsetT target_address;
5431 offsetT opcode_address;
5432 unsigned int extension = 0;
5433 offsetT displacement_from_opcode_start;
5434
5435 opcode = (unsigned char *) fragP->fr_opcode;
5436
5437 /* Address we want to reach in file space. */
5438 target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
5439
5440 /* Address opcode resides at in file space. */
5441 opcode_address = fragP->fr_address + fragP->fr_fix;
5442
5443 /* Displacement from opcode start to fill into instruction. */
5444 displacement_from_opcode_start = target_address - opcode_address;
5445
5446 if ((fragP->fr_subtype & BIG) == 0)
5447 {
5448 /* Don't have to change opcode. */
5449 extension = 1; /* 1 opcode + 1 displacement */
5450 where_to_put_displacement = &opcode[1];
5451 }
5452 else
5453 {
5454 if (no_cond_jump_promotion
5455 && TYPE_FROM_RELAX_STATE (fragP->fr_subtype) != UNCOND_JUMP)
5456 as_warn_where (fragP->fr_file, fragP->fr_line,
5457 _("long jump required"));
5458
5459 switch (fragP->fr_subtype)
5460 {
5461 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
5462 extension = 4; /* 1 opcode + 4 displacement */
5463 opcode[0] = 0xe9;
5464 where_to_put_displacement = &opcode[1];
5465 break;
5466
5467 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
5468 extension = 2; /* 1 opcode + 2 displacement */
5469 opcode[0] = 0xe9;
5470 where_to_put_displacement = &opcode[1];
5471 break;
5472
5473 case ENCODE_RELAX_STATE (COND_JUMP, BIG):
5474 case ENCODE_RELAX_STATE (COND_JUMP86, BIG):
5475 extension = 5; /* 2 opcode + 4 displacement */
5476 opcode[1] = opcode[0] + 0x10;
5477 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
5478 where_to_put_displacement = &opcode[2];
5479 break;
5480
5481 case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
5482 extension = 3; /* 2 opcode + 2 displacement */
5483 opcode[1] = opcode[0] + 0x10;
5484 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
5485 where_to_put_displacement = &opcode[2];
5486 break;
5487
5488 case ENCODE_RELAX_STATE (COND_JUMP86, BIG16):
5489 extension = 4;
5490 opcode[0] ^= 1;
5491 opcode[1] = 3;
5492 opcode[2] = 0xe9;
5493 where_to_put_displacement = &opcode[3];
5494 break;
5495
5496 default:
5497 BAD_CASE (fragP->fr_subtype);
5498 break;
5499 }
5500 }
5501
5502 /* If size if less then four we are sure that the operand fits,
5503 but if it's 4, then it could be that the displacement is larger
5504 then -/+ 2GB. */
5505 if (DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype) == 4
5506 && object_64bit
5507 && ((addressT) (displacement_from_opcode_start - extension
5508 + ((addressT) 1 << 31))
5509 > (((addressT) 2 << 31) - 1)))
5510 {
5511 as_bad_where (fragP->fr_file, fragP->fr_line,
5512 _("jump target out of range"));
5513 /* Make us emit 0. */
5514 displacement_from_opcode_start = extension;
5515 }
5516 /* Now put displacement after opcode. */
5517 md_number_to_chars ((char *) where_to_put_displacement,
5518 (valueT) (displacement_from_opcode_start - extension),
5519 DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
5520 fragP->fr_fix += extension;
5521 }
5522 \f
5523 /* Size of byte displacement jmp. */
5524 int md_short_jump_size = 2;
5525
5526 /* Size of dword displacement jmp. */
5527 int md_long_jump_size = 5;
5528
5529 void
5530 md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
5531 char *ptr;
5532 addressT from_addr, to_addr;
5533 fragS *frag ATTRIBUTE_UNUSED;
5534 symbolS *to_symbol ATTRIBUTE_UNUSED;
5535 {
5536 offsetT offset;
5537
5538 offset = to_addr - (from_addr + 2);
5539 /* Opcode for byte-disp jump. */
5540 md_number_to_chars (ptr, (valueT) 0xeb, 1);
5541 md_number_to_chars (ptr + 1, (valueT) offset, 1);
5542 }
5543
5544 void
5545 md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
5546 char *ptr;
5547 addressT from_addr, to_addr;
5548 fragS *frag ATTRIBUTE_UNUSED;
5549 symbolS *to_symbol ATTRIBUTE_UNUSED;
5550 {
5551 offsetT offset;
5552
5553 offset = to_addr - (from_addr + 5);
5554 md_number_to_chars (ptr, (valueT) 0xe9, 1);
5555 md_number_to_chars (ptr + 1, (valueT) offset, 4);
5556 }
5557 \f
5558 /* Apply a fixup (fixS) to segment data, once it has been determined
5559 by our caller that we have all the info we need to fix it up.
5560
5561 On the 386, immediates, displacements, and data pointers are all in
5562 the same (little-endian) format, so we don't need to care about which
5563 we are handling. */
5564
5565 void
5566 md_apply_fix (fixP, valP, seg)
5567 /* The fix we're to put in. */
5568 fixS *fixP;
5569 /* Pointer to the value of the bits. */
5570 valueT *valP;
5571 /* Segment fix is from. */
5572 segT seg ATTRIBUTE_UNUSED;
5573 {
5574 char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
5575 valueT value = *valP;
5576
5577 #if !defined (TE_Mach)
5578 if (fixP->fx_pcrel)
5579 {
5580 switch (fixP->fx_r_type)
5581 {
5582 default:
5583 break;
5584
5585 case BFD_RELOC_64:
5586 fixP->fx_r_type = BFD_RELOC_64_PCREL;
5587 break;
5588 case BFD_RELOC_32:
5589 case BFD_RELOC_X86_64_32S:
5590 fixP->fx_r_type = BFD_RELOC_32_PCREL;
5591 break;
5592 case BFD_RELOC_16:
5593 fixP->fx_r_type = BFD_RELOC_16_PCREL;
5594 break;
5595 case BFD_RELOC_8:
5596 fixP->fx_r_type = BFD_RELOC_8_PCREL;
5597 break;
5598 }
5599 }
5600
5601 if (fixP->fx_addsy != NULL
5602 && (fixP->fx_r_type == BFD_RELOC_32_PCREL
5603 || fixP->fx_r_type == BFD_RELOC_64_PCREL
5604 || fixP->fx_r_type == BFD_RELOC_16_PCREL
5605 || fixP->fx_r_type == BFD_RELOC_8_PCREL)
5606 && !use_rela_relocations)
5607 {
5608 /* This is a hack. There should be a better way to handle this.
5609 This covers for the fact that bfd_install_relocation will
5610 subtract the current location (for partial_inplace, PC relative
5611 relocations); see more below. */
5612 #ifndef OBJ_AOUT
5613 if (IS_ELF
5614 #ifdef TE_PE
5615 || OUTPUT_FLAVOR == bfd_target_coff_flavour
5616 #endif
5617 )
5618 value += fixP->fx_where + fixP->fx_frag->fr_address;
5619 #endif
5620 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5621 if (IS_ELF)
5622 {
5623 segT sym_seg = S_GET_SEGMENT (fixP->fx_addsy);
5624
5625 if ((sym_seg == seg
5626 || (symbol_section_p (fixP->fx_addsy)
5627 && sym_seg != absolute_section))
5628 && !generic_force_reloc (fixP))
5629 {
5630 /* Yes, we add the values in twice. This is because
5631 bfd_install_relocation subtracts them out again. I think
5632 bfd_install_relocation is broken, but I don't dare change
5633 it. FIXME. */
5634 value += fixP->fx_where + fixP->fx_frag->fr_address;
5635 }
5636 }
5637 #endif
5638 #if defined (OBJ_COFF) && defined (TE_PE)
5639 /* For some reason, the PE format does not store a
5640 section address offset for a PC relative symbol. */
5641 if (S_GET_SEGMENT (fixP->fx_addsy) != seg
5642 || S_IS_WEAK (fixP->fx_addsy))
5643 value += md_pcrel_from (fixP);
5644 #endif
5645 }
5646
5647 /* Fix a few things - the dynamic linker expects certain values here,
5648 and we must not disappoint it. */
5649 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5650 if (IS_ELF && fixP->fx_addsy)
5651 switch (fixP->fx_r_type)
5652 {
5653 case BFD_RELOC_386_PLT32:
5654 case BFD_RELOC_X86_64_PLT32:
5655 /* Make the jump instruction point to the address of the operand. At
5656 runtime we merely add the offset to the actual PLT entry. */
5657 value = -4;
5658 break;
5659
5660 case BFD_RELOC_386_TLS_GD:
5661 case BFD_RELOC_386_TLS_LDM:
5662 case BFD_RELOC_386_TLS_IE_32:
5663 case BFD_RELOC_386_TLS_IE:
5664 case BFD_RELOC_386_TLS_GOTIE:
5665 case BFD_RELOC_386_TLS_GOTDESC:
5666 case BFD_RELOC_X86_64_TLSGD:
5667 case BFD_RELOC_X86_64_TLSLD:
5668 case BFD_RELOC_X86_64_GOTTPOFF:
5669 case BFD_RELOC_X86_64_GOTPC32_TLSDESC:
5670 value = 0; /* Fully resolved at runtime. No addend. */
5671 /* Fallthrough */
5672 case BFD_RELOC_386_TLS_LE:
5673 case BFD_RELOC_386_TLS_LDO_32:
5674 case BFD_RELOC_386_TLS_LE_32:
5675 case BFD_RELOC_X86_64_DTPOFF32:
5676 case BFD_RELOC_X86_64_DTPOFF64:
5677 case BFD_RELOC_X86_64_TPOFF32:
5678 case BFD_RELOC_X86_64_TPOFF64:
5679 S_SET_THREAD_LOCAL (fixP->fx_addsy);
5680 break;
5681
5682 case BFD_RELOC_386_TLS_DESC_CALL:
5683 case BFD_RELOC_X86_64_TLSDESC_CALL:
5684 value = 0; /* Fully resolved at runtime. No addend. */
5685 S_SET_THREAD_LOCAL (fixP->fx_addsy);
5686 fixP->fx_done = 0;
5687 return;
5688
5689 case BFD_RELOC_386_GOT32:
5690 case BFD_RELOC_X86_64_GOT32:
5691 value = 0; /* Fully resolved at runtime. No addend. */
5692 break;
5693
5694 case BFD_RELOC_VTABLE_INHERIT:
5695 case BFD_RELOC_VTABLE_ENTRY:
5696 fixP->fx_done = 0;
5697 return;
5698
5699 default:
5700 break;
5701 }
5702 #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
5703 *valP = value;
5704 #endif /* !defined (TE_Mach) */
5705
5706 /* Are we finished with this relocation now? */
5707 if (fixP->fx_addsy == NULL)
5708 fixP->fx_done = 1;
5709 else if (use_rela_relocations)
5710 {
5711 fixP->fx_no_overflow = 1;
5712 /* Remember value for tc_gen_reloc. */
5713 fixP->fx_addnumber = value;
5714 value = 0;
5715 }
5716
5717 md_number_to_chars (p, value, fixP->fx_size);
5718 }
5719 \f
5720 #define MAX_LITTLENUMS 6
5721
5722 /* Turn the string pointed to by litP into a floating point constant
5723 of type TYPE, and emit the appropriate bytes. The number of
5724 LITTLENUMS emitted is stored in *SIZEP. An error message is
5725 returned, or NULL on OK. */
5726
5727 char *
5728 md_atof (type, litP, sizeP)
5729 int type;
5730 char *litP;
5731 int *sizeP;
5732 {
5733 int prec;
5734 LITTLENUM_TYPE words[MAX_LITTLENUMS];
5735 LITTLENUM_TYPE *wordP;
5736 char *t;
5737
5738 switch (type)
5739 {
5740 case 'f':
5741 case 'F':
5742 prec = 2;
5743 break;
5744
5745 case 'd':
5746 case 'D':
5747 prec = 4;
5748 break;
5749
5750 case 'x':
5751 case 'X':
5752 prec = 5;
5753 break;
5754
5755 default:
5756 *sizeP = 0;
5757 return _("Bad call to md_atof ()");
5758 }
5759 t = atof_ieee (input_line_pointer, type, words);
5760 if (t)
5761 input_line_pointer = t;
5762
5763 *sizeP = prec * sizeof (LITTLENUM_TYPE);
5764 /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
5765 the bigendian 386. */
5766 for (wordP = words + prec - 1; prec--;)
5767 {
5768 md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
5769 litP += sizeof (LITTLENUM_TYPE);
5770 }
5771 return 0;
5772 }
5773 \f
5774 static char output_invalid_buf[sizeof (unsigned char) * 2 + 6];
5775
5776 static char *
5777 output_invalid (int c)
5778 {
5779 if (ISPRINT (c))
5780 snprintf (output_invalid_buf, sizeof (output_invalid_buf),
5781 "'%c'", c);
5782 else
5783 snprintf (output_invalid_buf, sizeof (output_invalid_buf),
5784 "(0x%x)", (unsigned char) c);
5785 return output_invalid_buf;
5786 }
5787
5788 /* REG_STRING starts *before* REGISTER_PREFIX. */
5789
5790 static const reg_entry *
5791 parse_real_register (char *reg_string, char **end_op)
5792 {
5793 char *s = reg_string;
5794 char *p;
5795 char reg_name_given[MAX_REG_NAME_SIZE + 1];
5796 const reg_entry *r;
5797
5798 /* Skip possible REGISTER_PREFIX and possible whitespace. */
5799 if (*s == REGISTER_PREFIX)
5800 ++s;
5801
5802 if (is_space_char (*s))
5803 ++s;
5804
5805 p = reg_name_given;
5806 while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
5807 {
5808 if (p >= reg_name_given + MAX_REG_NAME_SIZE)
5809 return (const reg_entry *) NULL;
5810 s++;
5811 }
5812
5813 /* For naked regs, make sure that we are not dealing with an identifier.
5814 This prevents confusing an identifier like `eax_var' with register
5815 `eax'. */
5816 if (allow_naked_reg && identifier_chars[(unsigned char) *s])
5817 return (const reg_entry *) NULL;
5818
5819 *end_op = s;
5820
5821 r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
5822
5823 /* Handle floating point regs, allowing spaces in the (i) part. */
5824 if (r == i386_regtab /* %st is first entry of table */)
5825 {
5826 if (is_space_char (*s))
5827 ++s;
5828 if (*s == '(')
5829 {
5830 ++s;
5831 if (is_space_char (*s))
5832 ++s;
5833 if (*s >= '0' && *s <= '7')
5834 {
5835 int fpr = *s - '0';
5836 ++s;
5837 if (is_space_char (*s))
5838 ++s;
5839 if (*s == ')')
5840 {
5841 *end_op = s + 1;
5842 r = hash_find (reg_hash, "st(0)");
5843 know (r);
5844 return r + fpr;
5845 }
5846 }
5847 /* We have "%st(" then garbage. */
5848 return (const reg_entry *) NULL;
5849 }
5850 }
5851
5852 if (r != NULL
5853 && ((r->reg_flags & (RegRex64 | RegRex)) | (r->reg_type & Reg64)) != 0
5854 && (r->reg_type != Control || !(cpu_arch_flags & CpuSledgehammer))
5855 && flag_code != CODE_64BIT)
5856 return (const reg_entry *) NULL;
5857
5858 return r;
5859 }
5860
5861 /* REG_STRING starts *before* REGISTER_PREFIX. */
5862
5863 static const reg_entry *
5864 parse_register (char *reg_string, char **end_op)
5865 {
5866 const reg_entry *r;
5867
5868 if (*reg_string == REGISTER_PREFIX || allow_naked_reg)
5869 r = parse_real_register (reg_string, end_op);
5870 else
5871 r = NULL;
5872 if (!r)
5873 {
5874 char *save = input_line_pointer;
5875 char c;
5876 symbolS *symbolP;
5877
5878 input_line_pointer = reg_string;
5879 c = get_symbol_end ();
5880 symbolP = symbol_find (reg_string);
5881 if (symbolP && S_GET_SEGMENT (symbolP) == reg_section)
5882 {
5883 const expressionS *e = symbol_get_value_expression (symbolP);
5884
5885 know (e->X_op == O_register);
5886 know (e->X_add_number >= 0
5887 && (valueT) e->X_add_number < i386_regtab_size);
5888 r = i386_regtab + e->X_add_number;
5889 *end_op = input_line_pointer;
5890 }
5891 *input_line_pointer = c;
5892 input_line_pointer = save;
5893 }
5894 return r;
5895 }
5896
5897 int
5898 i386_parse_name (char *name, expressionS *e, char *nextcharP)
5899 {
5900 const reg_entry *r;
5901 char *end = input_line_pointer;
5902
5903 *end = *nextcharP;
5904 r = parse_register (name, &input_line_pointer);
5905 if (r && end <= input_line_pointer)
5906 {
5907 *nextcharP = *input_line_pointer;
5908 *input_line_pointer = 0;
5909 e->X_op = O_register;
5910 e->X_add_number = r - i386_regtab;
5911 return 1;
5912 }
5913 input_line_pointer = end;
5914 *end = 0;
5915 return 0;
5916 }
5917
5918 void
5919 md_operand (expressionS *e)
5920 {
5921 if (*input_line_pointer == REGISTER_PREFIX)
5922 {
5923 char *end;
5924 const reg_entry *r = parse_real_register (input_line_pointer, &end);
5925
5926 if (r)
5927 {
5928 e->X_op = O_register;
5929 e->X_add_number = r - i386_regtab;
5930 input_line_pointer = end;
5931 }
5932 }
5933 }
5934
5935 \f
5936 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5937 const char *md_shortopts = "kVQ:sqn";
5938 #else
5939 const char *md_shortopts = "qn";
5940 #endif
5941
5942 #define OPTION_32 (OPTION_MD_BASE + 0)
5943 #define OPTION_64 (OPTION_MD_BASE + 1)
5944 #define OPTION_DIVIDE (OPTION_MD_BASE + 2)
5945 #define OPTION_MARCH (OPTION_MD_BASE + 3)
5946 #define OPTION_MTUNE (OPTION_MD_BASE + 4)
5947
5948 struct option md_longopts[] =
5949 {
5950 {"32", no_argument, NULL, OPTION_32},
5951 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) || defined(TE_PEP)
5952 {"64", no_argument, NULL, OPTION_64},
5953 #endif
5954 {"divide", no_argument, NULL, OPTION_DIVIDE},
5955 {"march", required_argument, NULL, OPTION_MARCH},
5956 {"mtune", required_argument, NULL, OPTION_MTUNE},
5957 {NULL, no_argument, NULL, 0}
5958 };
5959 size_t md_longopts_size = sizeof (md_longopts);
5960
5961 int
5962 md_parse_option (int c, char *arg)
5963 {
5964 unsigned int i;
5965
5966 switch (c)
5967 {
5968 case 'n':
5969 optimize_align_code = 0;
5970 break;
5971
5972 case 'q':
5973 quiet_warnings = 1;
5974 break;
5975
5976 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5977 /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
5978 should be emitted or not. FIXME: Not implemented. */
5979 case 'Q':
5980 break;
5981
5982 /* -V: SVR4 argument to print version ID. */
5983 case 'V':
5984 print_version_id ();
5985 break;
5986
5987 /* -k: Ignore for FreeBSD compatibility. */
5988 case 'k':
5989 break;
5990
5991 case 's':
5992 /* -s: On i386 Solaris, this tells the native assembler to use
5993 .stab instead of .stab.excl. We always use .stab anyhow. */
5994 break;
5995 #endif
5996 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) || defined(TE_PEP)
5997 case OPTION_64:
5998 {
5999 const char **list, **l;
6000
6001 list = bfd_target_list ();
6002 for (l = list; *l != NULL; l++)
6003 if (CONST_STRNEQ (*l, "elf64-x86-64")
6004 || strcmp (*l, "coff-x86-64") == 0
6005 || strcmp (*l, "pe-x86-64") == 0
6006 || strcmp (*l, "pei-x86-64") == 0)
6007 {
6008 default_arch = "x86_64";
6009 break;
6010 }
6011 if (*l == NULL)
6012 as_fatal (_("No compiled in support for x86_64"));
6013 free (list);
6014 }
6015 break;
6016 #endif
6017
6018 case OPTION_32:
6019 default_arch = "i386";
6020 break;
6021
6022 case OPTION_DIVIDE:
6023 #ifdef SVR4_COMMENT_CHARS
6024 {
6025 char *n, *t;
6026 const char *s;
6027
6028 n = (char *) xmalloc (strlen (i386_comment_chars) + 1);
6029 t = n;
6030 for (s = i386_comment_chars; *s != '\0'; s++)
6031 if (*s != '/')
6032 *t++ = *s;
6033 *t = '\0';
6034 i386_comment_chars = n;
6035 }
6036 #endif
6037 break;
6038
6039 case OPTION_MARCH:
6040 if (*arg == '.')
6041 as_fatal (_("Invalid -march= option: `%s'"), arg);
6042 for (i = 0; i < ARRAY_SIZE (cpu_arch); i++)
6043 {
6044 if (strcmp (arg, cpu_arch [i].name) == 0)
6045 {
6046 cpu_arch_isa = cpu_arch[i].type;
6047 cpu_arch_isa_flags = cpu_arch[i].flags;
6048 if (!cpu_arch_tune_set)
6049 {
6050 cpu_arch_tune = cpu_arch_isa;
6051 cpu_arch_tune_flags = cpu_arch_isa_flags;
6052 }
6053 break;
6054 }
6055 }
6056 if (i >= ARRAY_SIZE (cpu_arch))
6057 as_fatal (_("Invalid -march= option: `%s'"), arg);
6058 break;
6059
6060 case OPTION_MTUNE:
6061 if (*arg == '.')
6062 as_fatal (_("Invalid -mtune= option: `%s'"), arg);
6063 for (i = 0; i < ARRAY_SIZE (cpu_arch); i++)
6064 {
6065 if (strcmp (arg, cpu_arch [i].name) == 0)
6066 {
6067 cpu_arch_tune_set = 1;
6068 cpu_arch_tune = cpu_arch [i].type;
6069 cpu_arch_tune_flags = cpu_arch[i].flags;
6070 break;
6071 }
6072 }
6073 if (i >= ARRAY_SIZE (cpu_arch))
6074 as_fatal (_("Invalid -mtune= option: `%s'"), arg);
6075 break;
6076
6077 default:
6078 return 0;
6079 }
6080 return 1;
6081 }
6082
6083 void
6084 md_show_usage (stream)
6085 FILE *stream;
6086 {
6087 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
6088 fprintf (stream, _("\
6089 -Q ignored\n\
6090 -V print assembler version number\n\
6091 -k ignored\n"));
6092 #endif
6093 fprintf (stream, _("\
6094 -n Do not optimize code alignment\n\
6095 -q quieten some warnings\n"));
6096 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
6097 fprintf (stream, _("\
6098 -s ignored\n"));
6099 #endif
6100 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) || defined(TE_PEP)
6101 fprintf (stream, _("\
6102 --32/--64 generate 32bit/64bit code\n"));
6103 #endif
6104 #ifdef SVR4_COMMENT_CHARS
6105 fprintf (stream, _("\
6106 --divide do not treat `/' as a comment character\n"));
6107 #else
6108 fprintf (stream, _("\
6109 --divide ignored\n"));
6110 #endif
6111 fprintf (stream, _("\
6112 -march=CPU/-mtune=CPU generate code/optimize for CPU, where CPU is one of:\n\
6113 i386, i486, pentium, pentiumpro, pentium4, nocona,\n\
6114 core, core2, k6, athlon, k8, generic32, generic64\n"));
6115
6116 }
6117
6118 #if defined(TE_PEP)
6119 const char *
6120 x86_64_target_format (void)
6121 {
6122 if (strcmp (default_arch, "x86_64") == 0)
6123 {
6124 set_code_flag (CODE_64BIT);
6125 return COFF_TARGET_FORMAT;
6126 }
6127 else if (strcmp (default_arch, "i386") == 0)
6128 {
6129 set_code_flag (CODE_32BIT);
6130 return "coff-i386";
6131 }
6132
6133 as_fatal (_("Unknown architecture"));
6134 return NULL;
6135 }
6136 #endif
6137
6138 #if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
6139 || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
6140
6141 /* Pick the target format to use. */
6142
6143 const char *
6144 i386_target_format (void)
6145 {
6146 if (!strcmp (default_arch, "x86_64"))
6147 {
6148 set_code_flag (CODE_64BIT);
6149 if (cpu_arch_isa_flags == 0)
6150 cpu_arch_isa_flags = Cpu186|Cpu286|Cpu386|Cpu486
6151 |Cpu586|Cpu686|CpuP4|CpuMMX|CpuMMX2
6152 |CpuSSE|CpuSSE2;
6153 if (cpu_arch_tune_flags == 0)
6154 cpu_arch_tune_flags = Cpu186|Cpu286|Cpu386|Cpu486
6155 |Cpu586|Cpu686|CpuP4|CpuMMX|CpuMMX2
6156 |CpuSSE|CpuSSE2;
6157 }
6158 else if (!strcmp (default_arch, "i386"))
6159 {
6160 set_code_flag (CODE_32BIT);
6161 if (cpu_arch_isa_flags == 0)
6162 cpu_arch_isa_flags = Cpu186|Cpu286|Cpu386;
6163 if (cpu_arch_tune_flags == 0)
6164 cpu_arch_tune_flags = Cpu186|Cpu286|Cpu386;
6165 }
6166 else
6167 as_fatal (_("Unknown architecture"));
6168 switch (OUTPUT_FLAVOR)
6169 {
6170 #ifdef OBJ_MAYBE_AOUT
6171 case bfd_target_aout_flavour:
6172 return AOUT_TARGET_FORMAT;
6173 #endif
6174 #ifdef OBJ_MAYBE_COFF
6175 case bfd_target_coff_flavour:
6176 return "coff-i386";
6177 #endif
6178 #if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF)
6179 case bfd_target_elf_flavour:
6180 {
6181 if (flag_code == CODE_64BIT)
6182 {
6183 object_64bit = 1;
6184 use_rela_relocations = 1;
6185 }
6186 return flag_code == CODE_64BIT ? ELF_TARGET_FORMAT64 : ELF_TARGET_FORMAT;
6187 }
6188 #endif
6189 default:
6190 abort ();
6191 return NULL;
6192 }
6193 }
6194
6195 #endif /* OBJ_MAYBE_ more than one */
6196
6197 #if (defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
6198 void
6199 i386_elf_emit_arch_note (void)
6200 {
6201 if (IS_ELF && cpu_arch_name != NULL)
6202 {
6203 char *p;
6204 asection *seg = now_seg;
6205 subsegT subseg = now_subseg;
6206 Elf_Internal_Note i_note;
6207 Elf_External_Note e_note;
6208 asection *note_secp;
6209 int len;
6210
6211 /* Create the .note section. */
6212 note_secp = subseg_new (".note", 0);
6213 bfd_set_section_flags (stdoutput,
6214 note_secp,
6215 SEC_HAS_CONTENTS | SEC_READONLY);
6216
6217 /* Process the arch string. */
6218 len = strlen (cpu_arch_name);
6219
6220 i_note.namesz = len + 1;
6221 i_note.descsz = 0;
6222 i_note.type = NT_ARCH;
6223 p = frag_more (sizeof (e_note.namesz));
6224 md_number_to_chars (p, (valueT) i_note.namesz, sizeof (e_note.namesz));
6225 p = frag_more (sizeof (e_note.descsz));
6226 md_number_to_chars (p, (valueT) i_note.descsz, sizeof (e_note.descsz));
6227 p = frag_more (sizeof (e_note.type));
6228 md_number_to_chars (p, (valueT) i_note.type, sizeof (e_note.type));
6229 p = frag_more (len + 1);
6230 strcpy (p, cpu_arch_name);
6231
6232 frag_align (2, 0, 0);
6233
6234 subseg_set (seg, subseg);
6235 }
6236 }
6237 #endif
6238 \f
6239 symbolS *
6240 md_undefined_symbol (name)
6241 char *name;
6242 {
6243 if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
6244 && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
6245 && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
6246 && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
6247 {
6248 if (!GOT_symbol)
6249 {
6250 if (symbol_find (name))
6251 as_bad (_("GOT already in symbol table"));
6252 GOT_symbol = symbol_new (name, undefined_section,
6253 (valueT) 0, &zero_address_frag);
6254 };
6255 return GOT_symbol;
6256 }
6257 return 0;
6258 }
6259
6260 /* Round up a section size to the appropriate boundary. */
6261
6262 valueT
6263 md_section_align (segment, size)
6264 segT segment ATTRIBUTE_UNUSED;
6265 valueT size;
6266 {
6267 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
6268 if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
6269 {
6270 /* For a.out, force the section size to be aligned. If we don't do
6271 this, BFD will align it for us, but it will not write out the
6272 final bytes of the section. This may be a bug in BFD, but it is
6273 easier to fix it here since that is how the other a.out targets
6274 work. */
6275 int align;
6276
6277 align = bfd_get_section_alignment (stdoutput, segment);
6278 size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
6279 }
6280 #endif
6281
6282 return size;
6283 }
6284
6285 /* On the i386, PC-relative offsets are relative to the start of the
6286 next instruction. That is, the address of the offset, plus its
6287 size, since the offset is always the last part of the insn. */
6288
6289 long
6290 md_pcrel_from (fixS *fixP)
6291 {
6292 return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
6293 }
6294
6295 #ifndef I386COFF
6296
6297 static void
6298 s_bss (int ignore ATTRIBUTE_UNUSED)
6299 {
6300 int temp;
6301
6302 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
6303 if (IS_ELF)
6304 obj_elf_section_change_hook ();
6305 #endif
6306 temp = get_absolute_expression ();
6307 subseg_set (bss_section, (subsegT) temp);
6308 demand_empty_rest_of_line ();
6309 }
6310
6311 #endif
6312
6313 void
6314 i386_validate_fix (fixS *fixp)
6315 {
6316 if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
6317 {
6318 if (fixp->fx_r_type == BFD_RELOC_32_PCREL)
6319 {
6320 if (!object_64bit)
6321 abort ();
6322 fixp->fx_r_type = BFD_RELOC_X86_64_GOTPCREL;
6323 }
6324 else
6325 {
6326 if (!object_64bit)
6327 fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
6328 else
6329 fixp->fx_r_type = BFD_RELOC_X86_64_GOTOFF64;
6330 }
6331 fixp->fx_subsy = 0;
6332 }
6333 }
6334
6335 arelent *
6336 tc_gen_reloc (section, fixp)
6337 asection *section ATTRIBUTE_UNUSED;
6338 fixS *fixp;
6339 {
6340 arelent *rel;
6341 bfd_reloc_code_real_type code;
6342
6343 switch (fixp->fx_r_type)
6344 {
6345 case BFD_RELOC_X86_64_PLT32:
6346 case BFD_RELOC_X86_64_GOT32:
6347 case BFD_RELOC_X86_64_GOTPCREL:
6348 case BFD_RELOC_386_PLT32:
6349 case BFD_RELOC_386_GOT32:
6350 case BFD_RELOC_386_GOTOFF:
6351 case BFD_RELOC_386_GOTPC:
6352 case BFD_RELOC_386_TLS_GD:
6353 case BFD_RELOC_386_TLS_LDM:
6354 case BFD_RELOC_386_TLS_LDO_32:
6355 case BFD_RELOC_386_TLS_IE_32:
6356 case BFD_RELOC_386_TLS_IE:
6357 case BFD_RELOC_386_TLS_GOTIE:
6358 case BFD_RELOC_386_TLS_LE_32:
6359 case BFD_RELOC_386_TLS_LE:
6360 case BFD_RELOC_386_TLS_GOTDESC:
6361 case BFD_RELOC_386_TLS_DESC_CALL:
6362 case BFD_RELOC_X86_64_TLSGD:
6363 case BFD_RELOC_X86_64_TLSLD:
6364 case BFD_RELOC_X86_64_DTPOFF32:
6365 case BFD_RELOC_X86_64_DTPOFF64:
6366 case BFD_RELOC_X86_64_GOTTPOFF:
6367 case BFD_RELOC_X86_64_TPOFF32:
6368 case BFD_RELOC_X86_64_TPOFF64:
6369 case BFD_RELOC_X86_64_GOTOFF64:
6370 case BFD_RELOC_X86_64_GOTPC32:
6371 case BFD_RELOC_X86_64_GOT64:
6372 case BFD_RELOC_X86_64_GOTPCREL64:
6373 case BFD_RELOC_X86_64_GOTPC64:
6374 case BFD_RELOC_X86_64_GOTPLT64:
6375 case BFD_RELOC_X86_64_PLTOFF64:
6376 case BFD_RELOC_X86_64_GOTPC32_TLSDESC:
6377 case BFD_RELOC_X86_64_TLSDESC_CALL:
6378 case BFD_RELOC_RVA:
6379 case BFD_RELOC_VTABLE_ENTRY:
6380 case BFD_RELOC_VTABLE_INHERIT:
6381 #ifdef TE_PE
6382 case BFD_RELOC_32_SECREL:
6383 #endif
6384 code = fixp->fx_r_type;
6385 break;
6386 case BFD_RELOC_X86_64_32S:
6387 if (!fixp->fx_pcrel)
6388 {
6389 /* Don't turn BFD_RELOC_X86_64_32S into BFD_RELOC_32. */
6390 code = fixp->fx_r_type;
6391 break;
6392 }
6393 default:
6394 if (fixp->fx_pcrel)
6395 {
6396 switch (fixp->fx_size)
6397 {
6398 default:
6399 as_bad_where (fixp->fx_file, fixp->fx_line,
6400 _("can not do %d byte pc-relative relocation"),
6401 fixp->fx_size);
6402 code = BFD_RELOC_32_PCREL;
6403 break;
6404 case 1: code = BFD_RELOC_8_PCREL; break;
6405 case 2: code = BFD_RELOC_16_PCREL; break;
6406 case 4: code = BFD_RELOC_32_PCREL; break;
6407 #ifdef BFD64
6408 case 8: code = BFD_RELOC_64_PCREL; break;
6409 #endif
6410 }
6411 }
6412 else
6413 {
6414 switch (fixp->fx_size)
6415 {
6416 default:
6417 as_bad_where (fixp->fx_file, fixp->fx_line,
6418 _("can not do %d byte relocation"),
6419 fixp->fx_size);
6420 code = BFD_RELOC_32;
6421 break;
6422 case 1: code = BFD_RELOC_8; break;
6423 case 2: code = BFD_RELOC_16; break;
6424 case 4: code = BFD_RELOC_32; break;
6425 #ifdef BFD64
6426 case 8: code = BFD_RELOC_64; break;
6427 #endif
6428 }
6429 }
6430 break;
6431 }
6432
6433 if ((code == BFD_RELOC_32
6434 || code == BFD_RELOC_32_PCREL
6435 || code == BFD_RELOC_X86_64_32S)
6436 && GOT_symbol
6437 && fixp->fx_addsy == GOT_symbol)
6438 {
6439 if (!object_64bit)
6440 code = BFD_RELOC_386_GOTPC;
6441 else
6442 code = BFD_RELOC_X86_64_GOTPC32;
6443 }
6444 if ((code == BFD_RELOC_64 || code == BFD_RELOC_64_PCREL)
6445 && GOT_symbol
6446 && fixp->fx_addsy == GOT_symbol)
6447 {
6448 code = BFD_RELOC_X86_64_GOTPC64;
6449 }
6450
6451 rel = (arelent *) xmalloc (sizeof (arelent));
6452 rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
6453 *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
6454
6455 rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
6456
6457 if (!use_rela_relocations)
6458 {
6459 /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
6460 vtable entry to be used in the relocation's section offset. */
6461 if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
6462 rel->address = fixp->fx_offset;
6463
6464 rel->addend = 0;
6465 }
6466 /* Use the rela in 64bit mode. */
6467 else
6468 {
6469 if (!fixp->fx_pcrel)
6470 rel->addend = fixp->fx_offset;
6471 else
6472 switch (code)
6473 {
6474 case BFD_RELOC_X86_64_PLT32:
6475 case BFD_RELOC_X86_64_GOT32:
6476 case BFD_RELOC_X86_64_GOTPCREL:
6477 case BFD_RELOC_X86_64_TLSGD:
6478 case BFD_RELOC_X86_64_TLSLD:
6479 case BFD_RELOC_X86_64_GOTTPOFF:
6480 case BFD_RELOC_X86_64_GOTPC32_TLSDESC:
6481 case BFD_RELOC_X86_64_TLSDESC_CALL:
6482 rel->addend = fixp->fx_offset - fixp->fx_size;
6483 break;
6484 default:
6485 rel->addend = (section->vma
6486 - fixp->fx_size
6487 + fixp->fx_addnumber
6488 + md_pcrel_from (fixp));
6489 break;
6490 }
6491 }
6492
6493 rel->howto = bfd_reloc_type_lookup (stdoutput, code);
6494 if (rel->howto == NULL)
6495 {
6496 as_bad_where (fixp->fx_file, fixp->fx_line,
6497 _("cannot represent relocation type %s"),
6498 bfd_get_reloc_code_name (code));
6499 /* Set howto to a garbage value so that we can keep going. */
6500 rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
6501 assert (rel->howto != NULL);
6502 }
6503
6504 return rel;
6505 }
6506
6507 \f
6508 /* Parse operands using Intel syntax. This implements a recursive descent
6509 parser based on the BNF grammar published in Appendix B of the MASM 6.1
6510 Programmer's Guide.
6511
6512 FIXME: We do not recognize the full operand grammar defined in the MASM
6513 documentation. In particular, all the structure/union and
6514 high-level macro operands are missing.
6515
6516 Uppercase words are terminals, lower case words are non-terminals.
6517 Objects surrounded by double brackets '[[' ']]' are optional. Vertical
6518 bars '|' denote choices. Most grammar productions are implemented in
6519 functions called 'intel_<production>'.
6520
6521 Initial production is 'expr'.
6522
6523 addOp + | -
6524
6525 alpha [a-zA-Z]
6526
6527 binOp & | AND | \| | OR | ^ | XOR
6528
6529 byteRegister AL | AH | BL | BH | CL | CH | DL | DH
6530
6531 constant digits [[ radixOverride ]]
6532
6533 dataType BYTE | WORD | DWORD | FWORD | QWORD | TBYTE | OWORD | XMMWORD
6534
6535 digits decdigit
6536 | digits decdigit
6537 | digits hexdigit
6538
6539 decdigit [0-9]
6540
6541 e04 e04 addOp e05
6542 | e05
6543
6544 e05 e05 binOp e06
6545 | e06
6546
6547 e06 e06 mulOp e09
6548 | e09
6549
6550 e09 OFFSET e10
6551 | SHORT e10
6552 | + e10
6553 | - e10
6554 | ~ e10
6555 | NOT e10
6556 | e09 PTR e10
6557 | e09 : e10
6558 | e10
6559
6560 e10 e10 [ expr ]
6561 | e11
6562
6563 e11 ( expr )
6564 | [ expr ]
6565 | constant
6566 | dataType
6567 | id
6568 | $
6569 | register
6570
6571 => expr expr cmpOp e04
6572 | e04
6573
6574 gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
6575 | BP | EBP | SP | ESP | DI | EDI | SI | ESI
6576
6577 hexdigit a | b | c | d | e | f
6578 | A | B | C | D | E | F
6579
6580 id alpha
6581 | id alpha
6582 | id decdigit
6583
6584 mulOp * | / | % | MOD | << | SHL | >> | SHR
6585
6586 quote " | '
6587
6588 register specialRegister
6589 | gpRegister
6590 | byteRegister
6591
6592 segmentRegister CS | DS | ES | FS | GS | SS
6593
6594 specialRegister CR0 | CR2 | CR3 | CR4
6595 | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
6596 | TR3 | TR4 | TR5 | TR6 | TR7
6597
6598 We simplify the grammar in obvious places (e.g., register parsing is
6599 done by calling parse_register) and eliminate immediate left recursion
6600 to implement a recursive-descent parser.
6601
6602 expr e04 expr'
6603
6604 expr' cmpOp e04 expr'
6605 | Empty
6606
6607 e04 e05 e04'
6608
6609 e04' addOp e05 e04'
6610 | Empty
6611
6612 e05 e06 e05'
6613
6614 e05' binOp e06 e05'
6615 | Empty
6616
6617 e06 e09 e06'
6618
6619 e06' mulOp e09 e06'
6620 | Empty
6621
6622 e09 OFFSET e10 e09'
6623 | SHORT e10'
6624 | + e10'
6625 | - e10'
6626 | ~ e10'
6627 | NOT e10'
6628 | e10 e09'
6629
6630 e09' PTR e10 e09'
6631 | : e10 e09'
6632 | Empty
6633
6634 e10 e11 e10'
6635
6636 e10' [ expr ] e10'
6637 | Empty
6638
6639 e11 ( expr )
6640 | [ expr ]
6641 | BYTE
6642 | WORD
6643 | DWORD
6644 | FWORD
6645 | QWORD
6646 | TBYTE
6647 | OWORD
6648 | XMMWORD
6649 | .
6650 | $
6651 | register
6652 | id
6653 | constant */
6654
6655 /* Parsing structure for the intel syntax parser. Used to implement the
6656 semantic actions for the operand grammar. */
6657 struct intel_parser_s
6658 {
6659 char *op_string; /* The string being parsed. */
6660 int got_a_float; /* Whether the operand is a float. */
6661 int op_modifier; /* Operand modifier. */
6662 int is_mem; /* 1 if operand is memory reference. */
6663 int in_offset; /* >=1 if parsing operand of offset. */
6664 int in_bracket; /* >=1 if parsing operand in brackets. */
6665 const reg_entry *reg; /* Last register reference found. */
6666 char *disp; /* Displacement string being built. */
6667 char *next_operand; /* Resume point when splitting operands. */
6668 };
6669
6670 static struct intel_parser_s intel_parser;
6671
6672 /* Token structure for parsing intel syntax. */
6673 struct intel_token
6674 {
6675 int code; /* Token code. */
6676 const reg_entry *reg; /* Register entry for register tokens. */
6677 char *str; /* String representation. */
6678 };
6679
6680 static struct intel_token cur_token, prev_token;
6681
6682 /* Token codes for the intel parser. Since T_SHORT is already used
6683 by COFF, undefine it first to prevent a warning. */
6684 #define T_NIL -1
6685 #define T_CONST 1
6686 #define T_REG 2
6687 #define T_BYTE 3
6688 #define T_WORD 4
6689 #define T_DWORD 5
6690 #define T_FWORD 6
6691 #define T_QWORD 7
6692 #define T_TBYTE 8
6693 #define T_XMMWORD 9
6694 #undef T_SHORT
6695 #define T_SHORT 10
6696 #define T_OFFSET 11
6697 #define T_PTR 12
6698 #define T_ID 13
6699 #define T_SHL 14
6700 #define T_SHR 15
6701
6702 /* Prototypes for intel parser functions. */
6703 static int intel_match_token (int);
6704 static void intel_putback_token (void);
6705 static void intel_get_token (void);
6706 static int intel_expr (void);
6707 static int intel_e04 (void);
6708 static int intel_e05 (void);
6709 static int intel_e06 (void);
6710 static int intel_e09 (void);
6711 static int intel_e10 (void);
6712 static int intel_e11 (void);
6713
6714 static int
6715 i386_intel_operand (char *operand_string, int got_a_float)
6716 {
6717 int ret;
6718 char *p;
6719
6720 p = intel_parser.op_string = xstrdup (operand_string);
6721 intel_parser.disp = (char *) xmalloc (strlen (operand_string) + 1);
6722
6723 for (;;)
6724 {
6725 /* Initialize token holders. */
6726 cur_token.code = prev_token.code = T_NIL;
6727 cur_token.reg = prev_token.reg = NULL;
6728 cur_token.str = prev_token.str = NULL;
6729
6730 /* Initialize parser structure. */
6731 intel_parser.got_a_float = got_a_float;
6732 intel_parser.op_modifier = 0;
6733 intel_parser.is_mem = 0;
6734 intel_parser.in_offset = 0;
6735 intel_parser.in_bracket = 0;
6736 intel_parser.reg = NULL;
6737 intel_parser.disp[0] = '\0';
6738 intel_parser.next_operand = NULL;
6739
6740 /* Read the first token and start the parser. */
6741 intel_get_token ();
6742 ret = intel_expr ();
6743
6744 if (!ret)
6745 break;
6746
6747 if (cur_token.code != T_NIL)
6748 {
6749 as_bad (_("invalid operand for '%s' ('%s' unexpected)"),
6750 current_templates->start->name, cur_token.str);
6751 ret = 0;
6752 }
6753 /* If we found a memory reference, hand it over to i386_displacement
6754 to fill in the rest of the operand fields. */
6755 else if (intel_parser.is_mem)
6756 {
6757 if ((i.mem_operands == 1
6758 && (current_templates->start->opcode_modifier & IsString) == 0)
6759 || i.mem_operands == 2)
6760 {
6761 as_bad (_("too many memory references for '%s'"),
6762 current_templates->start->name);
6763 ret = 0;
6764 }
6765 else
6766 {
6767 char *s = intel_parser.disp;
6768 i.mem_operands++;
6769
6770 if (!quiet_warnings && intel_parser.is_mem < 0)
6771 /* See the comments in intel_bracket_expr. */
6772 as_warn (_("Treating `%s' as memory reference"), operand_string);
6773
6774 /* Add the displacement expression. */
6775 if (*s != '\0')
6776 ret = i386_displacement (s, s + strlen (s));
6777 if (ret)
6778 {
6779 /* Swap base and index in 16-bit memory operands like
6780 [si+bx]. Since i386_index_check is also used in AT&T
6781 mode we have to do that here. */
6782 if (i.base_reg
6783 && i.index_reg
6784 && (i.base_reg->reg_type & Reg16)
6785 && (i.index_reg->reg_type & Reg16)
6786 && i.base_reg->reg_num >= 6
6787 && i.index_reg->reg_num < 6)
6788 {
6789 const reg_entry *base = i.index_reg;
6790
6791 i.index_reg = i.base_reg;
6792 i.base_reg = base;
6793 }
6794 ret = i386_index_check (operand_string);
6795 }
6796 }
6797 }
6798
6799 /* Constant and OFFSET expressions are handled by i386_immediate. */
6800 else if ((intel_parser.op_modifier & (1 << T_OFFSET))
6801 || intel_parser.reg == NULL)
6802 ret = i386_immediate (intel_parser.disp);
6803
6804 if (intel_parser.next_operand && this_operand >= MAX_OPERANDS - 1)
6805 ret = 0;
6806 if (!ret || !intel_parser.next_operand)
6807 break;
6808 intel_parser.op_string = intel_parser.next_operand;
6809 this_operand = i.operands++;
6810 }
6811
6812 free (p);
6813 free (intel_parser.disp);
6814
6815 return ret;
6816 }
6817
6818 #define NUM_ADDRESS_REGS (!!i.base_reg + !!i.index_reg)
6819
6820 /* expr e04 expr'
6821
6822 expr' cmpOp e04 expr'
6823 | Empty */
6824 static int
6825 intel_expr (void)
6826 {
6827 /* XXX Implement the comparison operators. */
6828 return intel_e04 ();
6829 }
6830
6831 /* e04 e05 e04'
6832
6833 e04' addOp e05 e04'
6834 | Empty */
6835 static int
6836 intel_e04 (void)
6837 {
6838 int nregs = -1;
6839
6840 for (;;)
6841 {
6842 if (!intel_e05())
6843 return 0;
6844
6845 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
6846 i.base_reg = i386_regtab + REGNAM_AL; /* al is invalid as base */
6847
6848 if (cur_token.code == '+')
6849 nregs = -1;
6850 else if (cur_token.code == '-')
6851 nregs = NUM_ADDRESS_REGS;
6852 else
6853 return 1;
6854
6855 strcat (intel_parser.disp, cur_token.str);
6856 intel_match_token (cur_token.code);
6857 }
6858 }
6859
6860 /* e05 e06 e05'
6861
6862 e05' binOp e06 e05'
6863 | Empty */
6864 static int
6865 intel_e05 (void)
6866 {
6867 int nregs = ~NUM_ADDRESS_REGS;
6868
6869 for (;;)
6870 {
6871 if (!intel_e06())
6872 return 0;
6873
6874 if (cur_token.code == '&'
6875 || cur_token.code == '|'
6876 || cur_token.code == '^')
6877 {
6878 char str[2];
6879
6880 str[0] = cur_token.code;
6881 str[1] = 0;
6882 strcat (intel_parser.disp, str);
6883 }
6884 else
6885 break;
6886
6887 intel_match_token (cur_token.code);
6888
6889 if (nregs < 0)
6890 nregs = ~nregs;
6891 }
6892 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
6893 i.base_reg = i386_regtab + REGNAM_AL + 1; /* cl is invalid as base */
6894 return 1;
6895 }
6896
6897 /* e06 e09 e06'
6898
6899 e06' mulOp e09 e06'
6900 | Empty */
6901 static int
6902 intel_e06 (void)
6903 {
6904 int nregs = ~NUM_ADDRESS_REGS;
6905
6906 for (;;)
6907 {
6908 if (!intel_e09())
6909 return 0;
6910
6911 if (cur_token.code == '*'
6912 || cur_token.code == '/'
6913 || cur_token.code == '%')
6914 {
6915 char str[2];
6916
6917 str[0] = cur_token.code;
6918 str[1] = 0;
6919 strcat (intel_parser.disp, str);
6920 }
6921 else if (cur_token.code == T_SHL)
6922 strcat (intel_parser.disp, "<<");
6923 else if (cur_token.code == T_SHR)
6924 strcat (intel_parser.disp, ">>");
6925 else
6926 break;
6927
6928 intel_match_token (cur_token.code);
6929
6930 if (nregs < 0)
6931 nregs = ~nregs;
6932 }
6933 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
6934 i.base_reg = i386_regtab + REGNAM_AL + 2; /* dl is invalid as base */
6935 return 1;
6936 }
6937
6938 /* e09 OFFSET e09
6939 | SHORT e09
6940 | + e09
6941 | - e09
6942 | ~ e09
6943 | NOT e09
6944 | e10 e09'
6945
6946 e09' PTR e10 e09'
6947 | : e10 e09'
6948 | Empty */
6949 static int
6950 intel_e09 (void)
6951 {
6952 int nregs = ~NUM_ADDRESS_REGS;
6953 int in_offset = 0;
6954
6955 for (;;)
6956 {
6957 /* Don't consume constants here. */
6958 if (cur_token.code == '+' || cur_token.code == '-')
6959 {
6960 /* Need to look one token ahead - if the next token
6961 is a constant, the current token is its sign. */
6962 int next_code;
6963
6964 intel_match_token (cur_token.code);
6965 next_code = cur_token.code;
6966 intel_putback_token ();
6967 if (next_code == T_CONST)
6968 break;
6969 }
6970
6971 /* e09 OFFSET e09 */
6972 if (cur_token.code == T_OFFSET)
6973 {
6974 if (!in_offset++)
6975 ++intel_parser.in_offset;
6976 }
6977
6978 /* e09 SHORT e09 */
6979 else if (cur_token.code == T_SHORT)
6980 intel_parser.op_modifier |= 1 << T_SHORT;
6981
6982 /* e09 + e09 */
6983 else if (cur_token.code == '+')
6984 strcat (intel_parser.disp, "+");
6985
6986 /* e09 - e09
6987 | ~ e09
6988 | NOT e09 */
6989 else if (cur_token.code == '-' || cur_token.code == '~')
6990 {
6991 char str[2];
6992
6993 if (nregs < 0)
6994 nregs = ~nregs;
6995 str[0] = cur_token.code;
6996 str[1] = 0;
6997 strcat (intel_parser.disp, str);
6998 }
6999
7000 /* e09 e10 e09' */
7001 else
7002 break;
7003
7004 intel_match_token (cur_token.code);
7005 }
7006
7007 for (;;)
7008 {
7009 if (!intel_e10 ())
7010 return 0;
7011
7012 /* e09' PTR e10 e09' */
7013 if (cur_token.code == T_PTR)
7014 {
7015 char suffix;
7016
7017 if (prev_token.code == T_BYTE)
7018 suffix = BYTE_MNEM_SUFFIX;
7019
7020 else if (prev_token.code == T_WORD)
7021 {
7022 if (current_templates->start->name[0] == 'l'
7023 && current_templates->start->name[2] == 's'
7024 && current_templates->start->name[3] == 0)
7025 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
7026 else if (intel_parser.got_a_float == 2) /* "fi..." */
7027 suffix = SHORT_MNEM_SUFFIX;
7028 else
7029 suffix = WORD_MNEM_SUFFIX;
7030 }
7031
7032 else if (prev_token.code == T_DWORD)
7033 {
7034 if (current_templates->start->name[0] == 'l'
7035 && current_templates->start->name[2] == 's'
7036 && current_templates->start->name[3] == 0)
7037 suffix = WORD_MNEM_SUFFIX;
7038 else if (flag_code == CODE_16BIT
7039 && (current_templates->start->opcode_modifier
7040 & (Jump | JumpDword)))
7041 suffix = LONG_DOUBLE_MNEM_SUFFIX;
7042 else if (intel_parser.got_a_float == 1) /* "f..." */
7043 suffix = SHORT_MNEM_SUFFIX;
7044 else
7045 suffix = LONG_MNEM_SUFFIX;
7046 }
7047
7048 else if (prev_token.code == T_FWORD)
7049 {
7050 if (current_templates->start->name[0] == 'l'
7051 && current_templates->start->name[2] == 's'
7052 && current_templates->start->name[3] == 0)
7053 suffix = LONG_MNEM_SUFFIX;
7054 else if (!intel_parser.got_a_float)
7055 {
7056 if (flag_code == CODE_16BIT)
7057 add_prefix (DATA_PREFIX_OPCODE);
7058 suffix = LONG_DOUBLE_MNEM_SUFFIX;
7059 }
7060 else
7061 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
7062 }
7063
7064 else if (prev_token.code == T_QWORD)
7065 {
7066 if (intel_parser.got_a_float == 1) /* "f..." */
7067 suffix = LONG_MNEM_SUFFIX;
7068 else
7069 suffix = QWORD_MNEM_SUFFIX;
7070 }
7071
7072 else if (prev_token.code == T_TBYTE)
7073 {
7074 if (intel_parser.got_a_float == 1)
7075 suffix = LONG_DOUBLE_MNEM_SUFFIX;
7076 else
7077 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
7078 }
7079
7080 else if (prev_token.code == T_XMMWORD)
7081 {
7082 /* XXX ignored for now, but accepted since gcc uses it */
7083 suffix = 0;
7084 }
7085
7086 else
7087 {
7088 as_bad (_("Unknown operand modifier `%s'"), prev_token.str);
7089 return 0;
7090 }
7091
7092 /* Operands for jump/call using 'ptr' notation denote absolute
7093 addresses. */
7094 if (current_templates->start->opcode_modifier & (Jump | JumpDword))
7095 i.types[this_operand] |= JumpAbsolute;
7096
7097 if (current_templates->start->base_opcode == 0x8d /* lea */)
7098 ;
7099 else if (!i.suffix)
7100 i.suffix = suffix;
7101 else if (i.suffix != suffix)
7102 {
7103 as_bad (_("Conflicting operand modifiers"));
7104 return 0;
7105 }
7106
7107 }
7108
7109 /* e09' : e10 e09' */
7110 else if (cur_token.code == ':')
7111 {
7112 if (prev_token.code != T_REG)
7113 {
7114 /* While {call,jmp} SSSS:OOOO is MASM syntax only when SSSS is a
7115 segment/group identifier (which we don't have), using comma
7116 as the operand separator there is even less consistent, since
7117 there all branches only have a single operand. */
7118 if (this_operand != 0
7119 || intel_parser.in_offset
7120 || intel_parser.in_bracket
7121 || (!(current_templates->start->opcode_modifier
7122 & (Jump|JumpDword|JumpInterSegment))
7123 && !(current_templates->start->operand_types[0]
7124 & JumpAbsolute)))
7125 return intel_match_token (T_NIL);
7126 /* Remember the start of the 2nd operand and terminate 1st
7127 operand here.
7128 XXX This isn't right, yet (when SSSS:OOOO is right operand of
7129 another expression), but it gets at least the simplest case
7130 (a plain number or symbol on the left side) right. */
7131 intel_parser.next_operand = intel_parser.op_string;
7132 *--intel_parser.op_string = '\0';
7133 return intel_match_token (':');
7134 }
7135 }
7136
7137 /* e09' Empty */
7138 else
7139 break;
7140
7141 intel_match_token (cur_token.code);
7142
7143 }
7144
7145 if (in_offset)
7146 {
7147 --intel_parser.in_offset;
7148 if (nregs < 0)
7149 nregs = ~nregs;
7150 if (NUM_ADDRESS_REGS > nregs)
7151 {
7152 as_bad (_("Invalid operand to `OFFSET'"));
7153 return 0;
7154 }
7155 intel_parser.op_modifier |= 1 << T_OFFSET;
7156 }
7157
7158 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
7159 i.base_reg = i386_regtab + REGNAM_AL + 3; /* bl is invalid as base */
7160 return 1;
7161 }
7162
7163 static int
7164 intel_bracket_expr (void)
7165 {
7166 int was_offset = intel_parser.op_modifier & (1 << T_OFFSET);
7167 const char *start = intel_parser.op_string;
7168 int len;
7169
7170 if (i.op[this_operand].regs)
7171 return intel_match_token (T_NIL);
7172
7173 intel_match_token ('[');
7174
7175 /* Mark as a memory operand only if it's not already known to be an
7176 offset expression. If it's an offset expression, we need to keep
7177 the brace in. */
7178 if (!intel_parser.in_offset)
7179 {
7180 ++intel_parser.in_bracket;
7181
7182 /* Operands for jump/call inside brackets denote absolute addresses. */
7183 if (current_templates->start->opcode_modifier & (Jump | JumpDword))
7184 i.types[this_operand] |= JumpAbsolute;
7185
7186 /* Unfortunately gas always diverged from MASM in a respect that can't
7187 be easily fixed without risking to break code sequences likely to be
7188 encountered (the testsuite even check for this): MASM doesn't consider
7189 an expression inside brackets unconditionally as a memory reference.
7190 When that is e.g. a constant, an offset expression, or the sum of the
7191 two, this is still taken as a constant load. gas, however, always
7192 treated these as memory references. As a compromise, we'll try to make
7193 offset expressions inside brackets work the MASM way (since that's
7194 less likely to be found in real world code), but make constants alone
7195 continue to work the traditional gas way. In either case, issue a
7196 warning. */
7197 intel_parser.op_modifier &= ~was_offset;
7198 }
7199 else
7200 strcat (intel_parser.disp, "[");
7201
7202 /* Add a '+' to the displacement string if necessary. */
7203 if (*intel_parser.disp != '\0'
7204 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
7205 strcat (intel_parser.disp, "+");
7206
7207 if (intel_expr ()
7208 && (len = intel_parser.op_string - start - 1,
7209 intel_match_token (']')))
7210 {
7211 /* Preserve brackets when the operand is an offset expression. */
7212 if (intel_parser.in_offset)
7213 strcat (intel_parser.disp, "]");
7214 else
7215 {
7216 --intel_parser.in_bracket;
7217 if (i.base_reg || i.index_reg)
7218 intel_parser.is_mem = 1;
7219 if (!intel_parser.is_mem)
7220 {
7221 if (!(intel_parser.op_modifier & (1 << T_OFFSET)))
7222 /* Defer the warning until all of the operand was parsed. */
7223 intel_parser.is_mem = -1;
7224 else if (!quiet_warnings)
7225 as_warn (_("`[%.*s]' taken to mean just `%.*s'"),
7226 len, start, len, start);
7227 }
7228 }
7229 intel_parser.op_modifier |= was_offset;
7230
7231 return 1;
7232 }
7233 return 0;
7234 }
7235
7236 /* e10 e11 e10'
7237
7238 e10' [ expr ] e10'
7239 | Empty */
7240 static int
7241 intel_e10 (void)
7242 {
7243 if (!intel_e11 ())
7244 return 0;
7245
7246 while (cur_token.code == '[')
7247 {
7248 if (!intel_bracket_expr ())
7249 return 0;
7250 }
7251
7252 return 1;
7253 }
7254
7255 /* e11 ( expr )
7256 | [ expr ]
7257 | BYTE
7258 | WORD
7259 | DWORD
7260 | FWORD
7261 | QWORD
7262 | TBYTE
7263 | OWORD
7264 | XMMWORD
7265 | $
7266 | .
7267 | register
7268 | id
7269 | constant */
7270 static int
7271 intel_e11 (void)
7272 {
7273 switch (cur_token.code)
7274 {
7275 /* e11 ( expr ) */
7276 case '(':
7277 intel_match_token ('(');
7278 strcat (intel_parser.disp, "(");
7279
7280 if (intel_expr () && intel_match_token (')'))
7281 {
7282 strcat (intel_parser.disp, ")");
7283 return 1;
7284 }
7285 return 0;
7286
7287 /* e11 [ expr ] */
7288 case '[':
7289 return intel_bracket_expr ();
7290
7291 /* e11 $
7292 | . */
7293 case '.':
7294 strcat (intel_parser.disp, cur_token.str);
7295 intel_match_token (cur_token.code);
7296
7297 /* Mark as a memory operand only if it's not already known to be an
7298 offset expression. */
7299 if (!intel_parser.in_offset)
7300 intel_parser.is_mem = 1;
7301
7302 return 1;
7303
7304 /* e11 register */
7305 case T_REG:
7306 {
7307 const reg_entry *reg = intel_parser.reg = cur_token.reg;
7308
7309 intel_match_token (T_REG);
7310
7311 /* Check for segment change. */
7312 if (cur_token.code == ':')
7313 {
7314 if (!(reg->reg_type & (SReg2 | SReg3)))
7315 {
7316 as_bad (_("`%s' is not a valid segment register"),
7317 reg->reg_name);
7318 return 0;
7319 }
7320 else if (i.seg[i.mem_operands])
7321 as_warn (_("Extra segment override ignored"));
7322 else
7323 {
7324 if (!intel_parser.in_offset)
7325 intel_parser.is_mem = 1;
7326 switch (reg->reg_num)
7327 {
7328 case 0:
7329 i.seg[i.mem_operands] = &es;
7330 break;
7331 case 1:
7332 i.seg[i.mem_operands] = &cs;
7333 break;
7334 case 2:
7335 i.seg[i.mem_operands] = &ss;
7336 break;
7337 case 3:
7338 i.seg[i.mem_operands] = &ds;
7339 break;
7340 case 4:
7341 i.seg[i.mem_operands] = &fs;
7342 break;
7343 case 5:
7344 i.seg[i.mem_operands] = &gs;
7345 break;
7346 }
7347 }
7348 }
7349
7350 /* Not a segment register. Check for register scaling. */
7351 else if (cur_token.code == '*')
7352 {
7353 if (!intel_parser.in_bracket)
7354 {
7355 as_bad (_("Register scaling only allowed in memory operands"));
7356 return 0;
7357 }
7358
7359 if (reg->reg_type & Reg16) /* Disallow things like [si*1]. */
7360 reg = i386_regtab + REGNAM_AX + 4; /* sp is invalid as index */
7361 else if (i.index_reg)
7362 reg = i386_regtab + REGNAM_EAX + 4; /* esp is invalid as index */
7363
7364 /* What follows must be a valid scale. */
7365 intel_match_token ('*');
7366 i.index_reg = reg;
7367 i.types[this_operand] |= BaseIndex;
7368
7369 /* Set the scale after setting the register (otherwise,
7370 i386_scale will complain) */
7371 if (cur_token.code == '+' || cur_token.code == '-')
7372 {
7373 char *str, sign = cur_token.code;
7374 intel_match_token (cur_token.code);
7375 if (cur_token.code != T_CONST)
7376 {
7377 as_bad (_("Syntax error: Expecting a constant, got `%s'"),
7378 cur_token.str);
7379 return 0;
7380 }
7381 str = (char *) xmalloc (strlen (cur_token.str) + 2);
7382 strcpy (str + 1, cur_token.str);
7383 *str = sign;
7384 if (!i386_scale (str))
7385 return 0;
7386 free (str);
7387 }
7388 else if (!i386_scale (cur_token.str))
7389 return 0;
7390 intel_match_token (cur_token.code);
7391 }
7392
7393 /* No scaling. If this is a memory operand, the register is either a
7394 base register (first occurrence) or an index register (second
7395 occurrence). */
7396 else if (intel_parser.in_bracket)
7397 {
7398
7399 if (!i.base_reg)
7400 i.base_reg = reg;
7401 else if (!i.index_reg)
7402 i.index_reg = reg;
7403 else
7404 {
7405 as_bad (_("Too many register references in memory operand"));
7406 return 0;
7407 }
7408
7409 i.types[this_operand] |= BaseIndex;
7410 }
7411
7412 /* It's neither base nor index. */
7413 else if (!intel_parser.in_offset && !intel_parser.is_mem)
7414 {
7415 i.types[this_operand] |= reg->reg_type & ~BaseIndex;
7416 i.op[this_operand].regs = reg;
7417 i.reg_operands++;
7418 }
7419 else
7420 {
7421 as_bad (_("Invalid use of register"));
7422 return 0;
7423 }
7424
7425 /* Since registers are not part of the displacement string (except
7426 when we're parsing offset operands), we may need to remove any
7427 preceding '+' from the displacement string. */
7428 if (*intel_parser.disp != '\0'
7429 && !intel_parser.in_offset)
7430 {
7431 char *s = intel_parser.disp;
7432 s += strlen (s) - 1;
7433 if (*s == '+')
7434 *s = '\0';
7435 }
7436
7437 return 1;
7438 }
7439
7440 /* e11 BYTE
7441 | WORD
7442 | DWORD
7443 | FWORD
7444 | QWORD
7445 | TBYTE
7446 | OWORD
7447 | XMMWORD */
7448 case T_BYTE:
7449 case T_WORD:
7450 case T_DWORD:
7451 case T_FWORD:
7452 case T_QWORD:
7453 case T_TBYTE:
7454 case T_XMMWORD:
7455 intel_match_token (cur_token.code);
7456
7457 if (cur_token.code == T_PTR)
7458 return 1;
7459
7460 /* It must have been an identifier. */
7461 intel_putback_token ();
7462 cur_token.code = T_ID;
7463 /* FALLTHRU */
7464
7465 /* e11 id
7466 | constant */
7467 case T_ID:
7468 if (!intel_parser.in_offset && intel_parser.is_mem <= 0)
7469 {
7470 symbolS *symbolP;
7471
7472 /* The identifier represents a memory reference only if it's not
7473 preceded by an offset modifier and if it's not an equate. */
7474 symbolP = symbol_find(cur_token.str);
7475 if (!symbolP || S_GET_SEGMENT(symbolP) != absolute_section)
7476 intel_parser.is_mem = 1;
7477 }
7478 /* FALLTHRU */
7479
7480 case T_CONST:
7481 case '-':
7482 case '+':
7483 {
7484 char *save_str, sign = 0;
7485
7486 /* Allow constants that start with `+' or `-'. */
7487 if (cur_token.code == '-' || cur_token.code == '+')
7488 {
7489 sign = cur_token.code;
7490 intel_match_token (cur_token.code);
7491 if (cur_token.code != T_CONST)
7492 {
7493 as_bad (_("Syntax error: Expecting a constant, got `%s'"),
7494 cur_token.str);
7495 return 0;
7496 }
7497 }
7498
7499 save_str = (char *) xmalloc (strlen (cur_token.str) + 2);
7500 strcpy (save_str + !!sign, cur_token.str);
7501 if (sign)
7502 *save_str = sign;
7503
7504 /* Get the next token to check for register scaling. */
7505 intel_match_token (cur_token.code);
7506
7507 /* Check if this constant is a scaling factor for an
7508 index register. */
7509 if (cur_token.code == '*')
7510 {
7511 if (intel_match_token ('*') && cur_token.code == T_REG)
7512 {
7513 const reg_entry *reg = cur_token.reg;
7514
7515 if (!intel_parser.in_bracket)
7516 {
7517 as_bad (_("Register scaling only allowed "
7518 "in memory operands"));
7519 return 0;
7520 }
7521
7522 /* Disallow things like [1*si].
7523 sp and esp are invalid as index. */
7524 if (reg->reg_type & Reg16)
7525 reg = i386_regtab + REGNAM_AX + 4;
7526 else if (i.index_reg)
7527 reg = i386_regtab + REGNAM_EAX + 4;
7528
7529 /* The constant is followed by `* reg', so it must be
7530 a valid scale. */
7531 i.index_reg = reg;
7532 i.types[this_operand] |= BaseIndex;
7533
7534 /* Set the scale after setting the register (otherwise,
7535 i386_scale will complain) */
7536 if (!i386_scale (save_str))
7537 return 0;
7538 intel_match_token (T_REG);
7539
7540 /* Since registers are not part of the displacement
7541 string, we may need to remove any preceding '+' from
7542 the displacement string. */
7543 if (*intel_parser.disp != '\0')
7544 {
7545 char *s = intel_parser.disp;
7546 s += strlen (s) - 1;
7547 if (*s == '+')
7548 *s = '\0';
7549 }
7550
7551 free (save_str);
7552
7553 return 1;
7554 }
7555
7556 /* The constant was not used for register scaling. Since we have
7557 already consumed the token following `*' we now need to put it
7558 back in the stream. */
7559 intel_putback_token ();
7560 }
7561
7562 /* Add the constant to the displacement string. */
7563 strcat (intel_parser.disp, save_str);
7564 free (save_str);
7565
7566 return 1;
7567 }
7568 }
7569
7570 as_bad (_("Unrecognized token '%s'"), cur_token.str);
7571 return 0;
7572 }
7573
7574 /* Match the given token against cur_token. If they match, read the next
7575 token from the operand string. */
7576 static int
7577 intel_match_token (int code)
7578 {
7579 if (cur_token.code == code)
7580 {
7581 intel_get_token ();
7582 return 1;
7583 }
7584 else
7585 {
7586 as_bad (_("Unexpected token `%s'"), cur_token.str);
7587 return 0;
7588 }
7589 }
7590
7591 /* Read a new token from intel_parser.op_string and store it in cur_token. */
7592 static void
7593 intel_get_token (void)
7594 {
7595 char *end_op;
7596 const reg_entry *reg;
7597 struct intel_token new_token;
7598
7599 new_token.code = T_NIL;
7600 new_token.reg = NULL;
7601 new_token.str = NULL;
7602
7603 /* Free the memory allocated to the previous token and move
7604 cur_token to prev_token. */
7605 if (prev_token.str)
7606 free (prev_token.str);
7607
7608 prev_token = cur_token;
7609
7610 /* Skip whitespace. */
7611 while (is_space_char (*intel_parser.op_string))
7612 intel_parser.op_string++;
7613
7614 /* Return an empty token if we find nothing else on the line. */
7615 if (*intel_parser.op_string == '\0')
7616 {
7617 cur_token = new_token;
7618 return;
7619 }
7620
7621 /* The new token cannot be larger than the remainder of the operand
7622 string. */
7623 new_token.str = (char *) xmalloc (strlen (intel_parser.op_string) + 1);
7624 new_token.str[0] = '\0';
7625
7626 if (strchr ("0123456789", *intel_parser.op_string))
7627 {
7628 char *p = new_token.str;
7629 char *q = intel_parser.op_string;
7630 new_token.code = T_CONST;
7631
7632 /* Allow any kind of identifier char to encompass floating point and
7633 hexadecimal numbers. */
7634 while (is_identifier_char (*q))
7635 *p++ = *q++;
7636 *p = '\0';
7637
7638 /* Recognize special symbol names [0-9][bf]. */
7639 if (strlen (intel_parser.op_string) == 2
7640 && (intel_parser.op_string[1] == 'b'
7641 || intel_parser.op_string[1] == 'f'))
7642 new_token.code = T_ID;
7643 }
7644
7645 else if ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL)
7646 {
7647 size_t len = end_op - intel_parser.op_string;
7648
7649 new_token.code = T_REG;
7650 new_token.reg = reg;
7651
7652 memcpy (new_token.str, intel_parser.op_string, len);
7653 new_token.str[len] = '\0';
7654 }
7655
7656 else if (is_identifier_char (*intel_parser.op_string))
7657 {
7658 char *p = new_token.str;
7659 char *q = intel_parser.op_string;
7660
7661 /* A '.' or '$' followed by an identifier char is an identifier.
7662 Otherwise, it's operator '.' followed by an expression. */
7663 if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
7664 {
7665 new_token.code = '.';
7666 new_token.str[0] = '.';
7667 new_token.str[1] = '\0';
7668 }
7669 else
7670 {
7671 while (is_identifier_char (*q) || *q == '@')
7672 *p++ = *q++;
7673 *p = '\0';
7674
7675 if (strcasecmp (new_token.str, "NOT") == 0)
7676 new_token.code = '~';
7677
7678 else if (strcasecmp (new_token.str, "MOD") == 0)
7679 new_token.code = '%';
7680
7681 else if (strcasecmp (new_token.str, "AND") == 0)
7682 new_token.code = '&';
7683
7684 else if (strcasecmp (new_token.str, "OR") == 0)
7685 new_token.code = '|';
7686
7687 else if (strcasecmp (new_token.str, "XOR") == 0)
7688 new_token.code = '^';
7689
7690 else if (strcasecmp (new_token.str, "SHL") == 0)
7691 new_token.code = T_SHL;
7692
7693 else if (strcasecmp (new_token.str, "SHR") == 0)
7694 new_token.code = T_SHR;
7695
7696 else if (strcasecmp (new_token.str, "BYTE") == 0)
7697 new_token.code = T_BYTE;
7698
7699 else if (strcasecmp (new_token.str, "WORD") == 0)
7700 new_token.code = T_WORD;
7701
7702 else if (strcasecmp (new_token.str, "DWORD") == 0)
7703 new_token.code = T_DWORD;
7704
7705 else if (strcasecmp (new_token.str, "FWORD") == 0)
7706 new_token.code = T_FWORD;
7707
7708 else if (strcasecmp (new_token.str, "QWORD") == 0)
7709 new_token.code = T_QWORD;
7710
7711 else if (strcasecmp (new_token.str, "TBYTE") == 0
7712 /* XXX remove (gcc still uses it) */
7713 || strcasecmp (new_token.str, "XWORD") == 0)
7714 new_token.code = T_TBYTE;
7715
7716 else if (strcasecmp (new_token.str, "XMMWORD") == 0
7717 || strcasecmp (new_token.str, "OWORD") == 0)
7718 new_token.code = T_XMMWORD;
7719
7720 else if (strcasecmp (new_token.str, "PTR") == 0)
7721 new_token.code = T_PTR;
7722
7723 else if (strcasecmp (new_token.str, "SHORT") == 0)
7724 new_token.code = T_SHORT;
7725
7726 else if (strcasecmp (new_token.str, "OFFSET") == 0)
7727 {
7728 new_token.code = T_OFFSET;
7729
7730 /* ??? This is not mentioned in the MASM grammar but gcc
7731 makes use of it with -mintel-syntax. OFFSET may be
7732 followed by FLAT: */
7733 if (strncasecmp (q, " FLAT:", 6) == 0)
7734 strcat (new_token.str, " FLAT:");
7735 }
7736
7737 /* ??? This is not mentioned in the MASM grammar. */
7738 else if (strcasecmp (new_token.str, "FLAT") == 0)
7739 {
7740 new_token.code = T_OFFSET;
7741 if (*q == ':')
7742 strcat (new_token.str, ":");
7743 else
7744 as_bad (_("`:' expected"));
7745 }
7746
7747 else
7748 new_token.code = T_ID;
7749 }
7750 }
7751
7752 else if (strchr ("+-/*%|&^:[]()~", *intel_parser.op_string))
7753 {
7754 new_token.code = *intel_parser.op_string;
7755 new_token.str[0] = *intel_parser.op_string;
7756 new_token.str[1] = '\0';
7757 }
7758
7759 else if (strchr ("<>", *intel_parser.op_string)
7760 && *intel_parser.op_string == *(intel_parser.op_string + 1))
7761 {
7762 new_token.code = *intel_parser.op_string == '<' ? T_SHL : T_SHR;
7763 new_token.str[0] = *intel_parser.op_string;
7764 new_token.str[1] = *intel_parser.op_string;
7765 new_token.str[2] = '\0';
7766 }
7767
7768 else
7769 as_bad (_("Unrecognized token `%s'"), intel_parser.op_string);
7770
7771 intel_parser.op_string += strlen (new_token.str);
7772 cur_token = new_token;
7773 }
7774
7775 /* Put cur_token back into the token stream and make cur_token point to
7776 prev_token. */
7777 static void
7778 intel_putback_token (void)
7779 {
7780 if (cur_token.code != T_NIL)
7781 {
7782 intel_parser.op_string -= strlen (cur_token.str);
7783 free (cur_token.str);
7784 }
7785 cur_token = prev_token;
7786
7787 /* Forget prev_token. */
7788 prev_token.code = T_NIL;
7789 prev_token.reg = NULL;
7790 prev_token.str = NULL;
7791 }
7792
7793 int
7794 tc_x86_regname_to_dw2regnum (char *regname)
7795 {
7796 unsigned int regnum;
7797 unsigned int regnames_count;
7798 static const char *const regnames_32[] =
7799 {
7800 "eax", "ecx", "edx", "ebx",
7801 "esp", "ebp", "esi", "edi",
7802 "eip", "eflags", NULL,
7803 "st0", "st1", "st2", "st3",
7804 "st4", "st5", "st6", "st7",
7805 NULL, NULL,
7806 "xmm0", "xmm1", "xmm2", "xmm3",
7807 "xmm4", "xmm5", "xmm6", "xmm7",
7808 "mm0", "mm1", "mm2", "mm3",
7809 "mm4", "mm5", "mm6", "mm7",
7810 "fcw", "fsw", "mxcsr",
7811 "es", "cs", "ss", "ds", "fs", "gs", NULL, NULL,
7812 "tr", "ldtr"
7813 };
7814 static const char *const regnames_64[] =
7815 {
7816 "rax", "rdx", "rcx", "rbx",
7817 "rsi", "rdi", "rbp", "rsp",
7818 "r8", "r9", "r10", "r11",
7819 "r12", "r13", "r14", "r15",
7820 "rip",
7821 "xmm0", "xmm1", "xmm2", "xmm3",
7822 "xmm4", "xmm5", "xmm6", "xmm7",
7823 "xmm8", "xmm9", "xmm10", "xmm11",
7824 "xmm12", "xmm13", "xmm14", "xmm15",
7825 "st0", "st1", "st2", "st3",
7826 "st4", "st5", "st6", "st7",
7827 "mm0", "mm1", "mm2", "mm3",
7828 "mm4", "mm5", "mm6", "mm7",
7829 "rflags",
7830 "es", "cs", "ss", "ds", "fs", "gs", NULL, NULL,
7831 "fs.base", "gs.base", NULL, NULL,
7832 "tr", "ldtr",
7833 "mxcsr", "fcw", "fsw"
7834 };
7835 const char *const *regnames;
7836
7837 if (flag_code == CODE_64BIT)
7838 {
7839 regnames = regnames_64;
7840 regnames_count = ARRAY_SIZE (regnames_64);
7841 }
7842 else
7843 {
7844 regnames = regnames_32;
7845 regnames_count = ARRAY_SIZE (regnames_32);
7846 }
7847
7848 for (regnum = 0; regnum < regnames_count; regnum++)
7849 if (regnames[regnum] != NULL
7850 && strcmp (regname, regnames[regnum]) == 0)
7851 return regnum;
7852
7853 return -1;
7854 }
7855
7856 void
7857 tc_x86_frame_initial_instructions (void)
7858 {
7859 static unsigned int sp_regno;
7860
7861 if (!sp_regno)
7862 sp_regno = tc_x86_regname_to_dw2regnum (flag_code == CODE_64BIT
7863 ? "rsp" : "esp");
7864
7865 cfi_add_CFA_def_cfa (sp_regno, -x86_cie_data_alignment);
7866 cfi_add_CFA_offset (x86_dwarf2_return_column, x86_cie_data_alignment);
7867 }
7868
7869 int
7870 i386_elf_section_type (const char *str, size_t len)
7871 {
7872 if (flag_code == CODE_64BIT
7873 && len == sizeof ("unwind") - 1
7874 && strncmp (str, "unwind", 6) == 0)
7875 return SHT_X86_64_UNWIND;
7876
7877 return -1;
7878 }
7879
7880 #ifdef TE_PE
7881 void
7882 tc_pe_dwarf2_emit_offset (symbolS *symbol, unsigned int size)
7883 {
7884 expressionS expr;
7885
7886 expr.X_op = O_secrel;
7887 expr.X_add_symbol = symbol;
7888 expr.X_add_number = 0;
7889 emit_expr (&expr, size);
7890 }
7891 #endif
7892
7893 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
7894 /* For ELF on x86-64, add support for SHF_X86_64_LARGE. */
7895
7896 int
7897 x86_64_section_letter (int letter, char **ptr_msg)
7898 {
7899 if (flag_code == CODE_64BIT)
7900 {
7901 if (letter == 'l')
7902 return SHF_X86_64_LARGE;
7903
7904 *ptr_msg = _("Bad .section directive: want a,l,w,x,M,S,G,T in string");
7905 }
7906 else
7907 *ptr_msg = _("Bad .section directive: want a,w,x,M,S,G,T in string");
7908 return -1;
7909 }
7910
7911 int
7912 x86_64_section_word (char *str, size_t len)
7913 {
7914 if (len == 5 && flag_code == CODE_64BIT && CONST_STRNEQ (str, "large"))
7915 return SHF_X86_64_LARGE;
7916
7917 return -1;
7918 }
7919
7920 static void
7921 handle_large_common (int small ATTRIBUTE_UNUSED)
7922 {
7923 if (flag_code != CODE_64BIT)
7924 {
7925 s_comm_internal (0, elf_common_parse);
7926 as_warn (_(".largecomm supported only in 64bit mode, producing .comm"));
7927 }
7928 else
7929 {
7930 static segT lbss_section;
7931 asection *saved_com_section_ptr = elf_com_section_ptr;
7932 asection *saved_bss_section = bss_section;
7933
7934 if (lbss_section == NULL)
7935 {
7936 flagword applicable;
7937 segT seg = now_seg;
7938 subsegT subseg = now_subseg;
7939
7940 /* The .lbss section is for local .largecomm symbols. */
7941 lbss_section = subseg_new (".lbss", 0);
7942 applicable = bfd_applicable_section_flags (stdoutput);
7943 bfd_set_section_flags (stdoutput, lbss_section,
7944 applicable & SEC_ALLOC);
7945 seg_info (lbss_section)->bss = 1;
7946
7947 subseg_set (seg, subseg);
7948 }
7949
7950 elf_com_section_ptr = &_bfd_elf_large_com_section;
7951 bss_section = lbss_section;
7952
7953 s_comm_internal (0, elf_common_parse);
7954
7955 elf_com_section_ptr = saved_com_section_ptr;
7956 bss_section = saved_bss_section;
7957 }
7958 }
7959 #endif /* OBJ_ELF || OBJ_MAYBE_ELF */
This page took 0.296667 seconds and 4 git commands to generate.