2007-05-14 Mei Ligang <ligang@sunnorth.com.cn>
[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 2, or (at your option)
11 any later version.
12
13 GAS is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
17
18 You should have received a copy of the GNU General Public License
19 along with GAS; see the file COPYING. If not, write to the Free
20 Software Foundation, 51 Franklin Street - Fifth Floor, Boston, MA
21 02110-1301, USA. */
22
23 /* Intel 80386 machine specific gas.
24 Written by Eliot Dresselhaus (eliot@mgm.mit.edu).
25 x86_64 support by Jan Hubicka (jh@suse.cz)
26 VIA PadLock support by Michal Ludvig (mludvig@suse.cz)
27 Bugs & suggestions are completely welcome. This is free software.
28 Please help us make it better. */
29
30 #include "as.h"
31 #include "safe-ctype.h"
32 #include "subsegs.h"
33 #include "dwarf2dbg.h"
34 #include "dw2gencfi.h"
35 #include "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 static void
4080 output_disp (fragS *insn_start_frag, offsetT insn_start_off)
4081 {
4082 char *p;
4083 unsigned int n;
4084
4085 for (n = 0; n < i.operands; n++)
4086 {
4087 if (i.types[n] & Disp)
4088 {
4089 if (i.op[n].disps->X_op == O_constant)
4090 {
4091 int size;
4092 offsetT val;
4093
4094 size = 4;
4095 if (i.types[n] & (Disp8 | Disp16 | Disp64))
4096 {
4097 size = 2;
4098 if (i.types[n] & Disp8)
4099 size = 1;
4100 if (i.types[n] & Disp64)
4101 size = 8;
4102 }
4103 val = offset_in_range (i.op[n].disps->X_add_number,
4104 size);
4105 p = frag_more (size);
4106 md_number_to_chars (p, val, size);
4107 }
4108 else
4109 {
4110 enum bfd_reloc_code_real reloc_type;
4111 int size = 4;
4112 int sign = 0;
4113 int pcrel = (i.flags[n] & Operand_PCrel) != 0;
4114
4115 /* The PC relative address is computed relative
4116 to the instruction boundary, so in case immediate
4117 fields follows, we need to adjust the value. */
4118 if (pcrel && i.imm_operands)
4119 {
4120 int imm_size = 4;
4121 unsigned int n1;
4122
4123 for (n1 = 0; n1 < i.operands; n1++)
4124 if (i.types[n1] & Imm)
4125 {
4126 if (i.types[n1] & (Imm8 | Imm8S | Imm16 | Imm64))
4127 {
4128 imm_size = 2;
4129 if (i.types[n1] & (Imm8 | Imm8S))
4130 imm_size = 1;
4131 if (i.types[n1] & Imm64)
4132 imm_size = 8;
4133 }
4134 break;
4135 }
4136 /* We should find the immediate. */
4137 if (n1 == i.operands)
4138 abort ();
4139 i.op[n].disps->X_add_number -= imm_size;
4140 }
4141
4142 if (i.types[n] & Disp32S)
4143 sign = 1;
4144
4145 if (i.types[n] & (Disp16 | Disp64))
4146 {
4147 size = 2;
4148 if (i.types[n] & Disp64)
4149 size = 8;
4150 }
4151
4152 p = frag_more (size);
4153 reloc_type = reloc (size, pcrel, sign, i.reloc[n]);
4154 if (GOT_symbol
4155 && GOT_symbol == i.op[n].disps->X_add_symbol
4156 && (((reloc_type == BFD_RELOC_32
4157 || reloc_type == BFD_RELOC_X86_64_32S
4158 || (reloc_type == BFD_RELOC_64
4159 && object_64bit))
4160 && (i.op[n].disps->X_op == O_symbol
4161 || (i.op[n].disps->X_op == O_add
4162 && ((symbol_get_value_expression
4163 (i.op[n].disps->X_op_symbol)->X_op)
4164 == O_subtract))))
4165 || reloc_type == BFD_RELOC_32_PCREL))
4166 {
4167 offsetT add;
4168
4169 if (insn_start_frag == frag_now)
4170 add = (p - frag_now->fr_literal) - insn_start_off;
4171 else
4172 {
4173 fragS *fr;
4174
4175 add = insn_start_frag->fr_fix - insn_start_off;
4176 for (fr = insn_start_frag->fr_next;
4177 fr && fr != frag_now; fr = fr->fr_next)
4178 add += fr->fr_fix;
4179 add += p - frag_now->fr_literal;
4180 }
4181
4182 if (!object_64bit)
4183 {
4184 reloc_type = BFD_RELOC_386_GOTPC;
4185 i.op[n].imms->X_add_number += add;
4186 }
4187 else if (reloc_type == BFD_RELOC_64)
4188 reloc_type = BFD_RELOC_X86_64_GOTPC64;
4189 else
4190 /* Don't do the adjustment for x86-64, as there
4191 the pcrel addressing is relative to the _next_
4192 insn, and that is taken care of in other code. */
4193 reloc_type = BFD_RELOC_X86_64_GOTPC32;
4194 }
4195 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
4196 i.op[n].disps, pcrel, reloc_type);
4197 }
4198 }
4199 }
4200 }
4201
4202 static void
4203 output_imm (fragS *insn_start_frag, offsetT insn_start_off)
4204 {
4205 char *p;
4206 unsigned int n;
4207
4208 for (n = 0; n < i.operands; n++)
4209 {
4210 if (i.types[n] & Imm)
4211 {
4212 if (i.op[n].imms->X_op == O_constant)
4213 {
4214 int size;
4215 offsetT val;
4216
4217 size = 4;
4218 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
4219 {
4220 size = 2;
4221 if (i.types[n] & (Imm8 | Imm8S))
4222 size = 1;
4223 else if (i.types[n] & Imm64)
4224 size = 8;
4225 }
4226 val = offset_in_range (i.op[n].imms->X_add_number,
4227 size);
4228 p = frag_more (size);
4229 md_number_to_chars (p, val, size);
4230 }
4231 else
4232 {
4233 /* Not absolute_section.
4234 Need a 32-bit fixup (don't support 8bit
4235 non-absolute imms). Try to support other
4236 sizes ... */
4237 enum bfd_reloc_code_real reloc_type;
4238 int size = 4;
4239 int sign = 0;
4240
4241 if ((i.types[n] & (Imm32S))
4242 && (i.suffix == QWORD_MNEM_SUFFIX
4243 || (!i.suffix && (i.tm.opcode_modifier & No_lSuf))))
4244 sign = 1;
4245 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
4246 {
4247 size = 2;
4248 if (i.types[n] & (Imm8 | Imm8S))
4249 size = 1;
4250 if (i.types[n] & Imm64)
4251 size = 8;
4252 }
4253
4254 p = frag_more (size);
4255 reloc_type = reloc (size, 0, sign, i.reloc[n]);
4256
4257 /* This is tough to explain. We end up with this one if we
4258 * have operands that look like
4259 * "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal here is to
4260 * obtain the absolute address of the GOT, and it is strongly
4261 * preferable from a performance point of view to avoid using
4262 * a runtime relocation for this. The actual sequence of
4263 * instructions often look something like:
4264 *
4265 * call .L66
4266 * .L66:
4267 * popl %ebx
4268 * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
4269 *
4270 * The call and pop essentially return the absolute address
4271 * of the label .L66 and store it in %ebx. The linker itself
4272 * will ultimately change the first operand of the addl so
4273 * that %ebx points to the GOT, but to keep things simple, the
4274 * .o file must have this operand set so that it generates not
4275 * the absolute address of .L66, but the absolute address of
4276 * itself. This allows the linker itself simply treat a GOTPC
4277 * relocation as asking for a pcrel offset to the GOT to be
4278 * added in, and the addend of the relocation is stored in the
4279 * operand field for the instruction itself.
4280 *
4281 * Our job here is to fix the operand so that it would add
4282 * the correct offset so that %ebx would point to itself. The
4283 * thing that is tricky is that .-.L66 will point to the
4284 * beginning of the instruction, so we need to further modify
4285 * the operand so that it will point to itself. There are
4286 * other cases where you have something like:
4287 *
4288 * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
4289 *
4290 * and here no correction would be required. Internally in
4291 * the assembler we treat operands of this form as not being
4292 * pcrel since the '.' is explicitly mentioned, and I wonder
4293 * whether it would simplify matters to do it this way. Who
4294 * knows. In earlier versions of the PIC patches, the
4295 * pcrel_adjust field was used to store the correction, but
4296 * since the expression is not pcrel, I felt it would be
4297 * confusing to do it this way. */
4298
4299 if ((reloc_type == BFD_RELOC_32
4300 || reloc_type == BFD_RELOC_X86_64_32S
4301 || reloc_type == BFD_RELOC_64)
4302 && GOT_symbol
4303 && GOT_symbol == i.op[n].imms->X_add_symbol
4304 && (i.op[n].imms->X_op == O_symbol
4305 || (i.op[n].imms->X_op == O_add
4306 && ((symbol_get_value_expression
4307 (i.op[n].imms->X_op_symbol)->X_op)
4308 == O_subtract))))
4309 {
4310 offsetT add;
4311
4312 if (insn_start_frag == frag_now)
4313 add = (p - frag_now->fr_literal) - insn_start_off;
4314 else
4315 {
4316 fragS *fr;
4317
4318 add = insn_start_frag->fr_fix - insn_start_off;
4319 for (fr = insn_start_frag->fr_next;
4320 fr && fr != frag_now; fr = fr->fr_next)
4321 add += fr->fr_fix;
4322 add += p - frag_now->fr_literal;
4323 }
4324
4325 if (!object_64bit)
4326 reloc_type = BFD_RELOC_386_GOTPC;
4327 else if (size == 4)
4328 reloc_type = BFD_RELOC_X86_64_GOTPC32;
4329 else if (size == 8)
4330 reloc_type = BFD_RELOC_X86_64_GOTPC64;
4331 i.op[n].imms->X_add_number += add;
4332 }
4333 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
4334 i.op[n].imms, 0, reloc_type);
4335 }
4336 }
4337 }
4338 }
4339 \f
4340 /* x86_cons_fix_new is called via the expression parsing code when a
4341 reloc is needed. We use this hook to get the correct .got reloc. */
4342 static enum bfd_reloc_code_real got_reloc = NO_RELOC;
4343 static int cons_sign = -1;
4344
4345 void
4346 x86_cons_fix_new (fragS *frag, unsigned int off, unsigned int len,
4347 expressionS *exp)
4348 {
4349 enum bfd_reloc_code_real r = reloc (len, 0, cons_sign, got_reloc);
4350
4351 got_reloc = NO_RELOC;
4352
4353 #ifdef TE_PE
4354 if (exp->X_op == O_secrel)
4355 {
4356 exp->X_op = O_symbol;
4357 r = BFD_RELOC_32_SECREL;
4358 }
4359 #endif
4360
4361 fix_new_exp (frag, off, len, exp, 0, r);
4362 }
4363
4364 #if (!defined (OBJ_ELF) && !defined (OBJ_MAYBE_ELF)) || defined (LEX_AT)
4365 # define lex_got(reloc, adjust, types) NULL
4366 #else
4367 /* Parse operands of the form
4368 <symbol>@GOTOFF+<nnn>
4369 and similar .plt or .got references.
4370
4371 If we find one, set up the correct relocation in RELOC and copy the
4372 input string, minus the `@GOTOFF' into a malloc'd buffer for
4373 parsing by the calling routine. Return this buffer, and if ADJUST
4374 is non-null set it to the length of the string we removed from the
4375 input line. Otherwise return NULL. */
4376 static char *
4377 lex_got (enum bfd_reloc_code_real *reloc,
4378 int *adjust,
4379 unsigned int *types)
4380 {
4381 /* Some of the relocations depend on the size of what field is to
4382 be relocated. But in our callers i386_immediate and i386_displacement
4383 we don't yet know the operand size (this will be set by insn
4384 matching). Hence we record the word32 relocation here,
4385 and adjust the reloc according to the real size in reloc(). */
4386 static const struct {
4387 const char *str;
4388 const enum bfd_reloc_code_real rel[2];
4389 const unsigned int types64;
4390 } gotrel[] = {
4391 { "PLTOFF", { 0,
4392 BFD_RELOC_X86_64_PLTOFF64 },
4393 Imm64 },
4394 { "PLT", { BFD_RELOC_386_PLT32,
4395 BFD_RELOC_X86_64_PLT32 },
4396 Imm32 | Imm32S | Disp32 },
4397 { "GOTPLT", { 0,
4398 BFD_RELOC_X86_64_GOTPLT64 },
4399 Imm64 | Disp64 },
4400 { "GOTOFF", { BFD_RELOC_386_GOTOFF,
4401 BFD_RELOC_X86_64_GOTOFF64 },
4402 Imm64 | Disp64 },
4403 { "GOTPCREL", { 0,
4404 BFD_RELOC_X86_64_GOTPCREL },
4405 Imm32 | Imm32S | Disp32 },
4406 { "TLSGD", { BFD_RELOC_386_TLS_GD,
4407 BFD_RELOC_X86_64_TLSGD },
4408 Imm32 | Imm32S | Disp32 },
4409 { "TLSLDM", { BFD_RELOC_386_TLS_LDM,
4410 0 },
4411 0 },
4412 { "TLSLD", { 0,
4413 BFD_RELOC_X86_64_TLSLD },
4414 Imm32 | Imm32S | Disp32 },
4415 { "GOTTPOFF", { BFD_RELOC_386_TLS_IE_32,
4416 BFD_RELOC_X86_64_GOTTPOFF },
4417 Imm32 | Imm32S | Disp32 },
4418 { "TPOFF", { BFD_RELOC_386_TLS_LE_32,
4419 BFD_RELOC_X86_64_TPOFF32 },
4420 Imm32 | Imm32S | Imm64 | Disp32 | Disp64 },
4421 { "NTPOFF", { BFD_RELOC_386_TLS_LE,
4422 0 },
4423 0 },
4424 { "DTPOFF", { BFD_RELOC_386_TLS_LDO_32,
4425 BFD_RELOC_X86_64_DTPOFF32 },
4426 Imm32 | Imm32S | Imm64 | Disp32 | Disp64 },
4427 { "GOTNTPOFF",{ BFD_RELOC_386_TLS_GOTIE,
4428 0 },
4429 0 },
4430 { "INDNTPOFF",{ BFD_RELOC_386_TLS_IE,
4431 0 },
4432 0 },
4433 { "GOT", { BFD_RELOC_386_GOT32,
4434 BFD_RELOC_X86_64_GOT32 },
4435 Imm32 | Imm32S | Disp32 | Imm64 },
4436 { "TLSDESC", { BFD_RELOC_386_TLS_GOTDESC,
4437 BFD_RELOC_X86_64_GOTPC32_TLSDESC },
4438 Imm32 | Imm32S | Disp32 },
4439 { "TLSCALL", { BFD_RELOC_386_TLS_DESC_CALL,
4440 BFD_RELOC_X86_64_TLSDESC_CALL },
4441 Imm32 | Imm32S | Disp32 }
4442 };
4443 char *cp;
4444 unsigned int j;
4445
4446 if (!IS_ELF)
4447 return NULL;
4448
4449 for (cp = input_line_pointer; *cp != '@'; cp++)
4450 if (is_end_of_line[(unsigned char) *cp])
4451 return NULL;
4452
4453 for (j = 0; j < sizeof (gotrel) / sizeof (gotrel[0]); j++)
4454 {
4455 int len;
4456
4457 len = strlen (gotrel[j].str);
4458 if (strncasecmp (cp + 1, gotrel[j].str, len) == 0)
4459 {
4460 if (gotrel[j].rel[object_64bit] != 0)
4461 {
4462 int first, second;
4463 char *tmpbuf, *past_reloc;
4464
4465 *reloc = gotrel[j].rel[object_64bit];
4466 if (adjust)
4467 *adjust = len;
4468
4469 if (types)
4470 {
4471 if (flag_code != CODE_64BIT)
4472 *types = Imm32 | Disp32;
4473 else
4474 *types = gotrel[j].types64;
4475 }
4476
4477 if (GOT_symbol == NULL)
4478 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
4479
4480 /* The length of the first part of our input line. */
4481 first = cp - input_line_pointer;
4482
4483 /* The second part goes from after the reloc token until
4484 (and including) an end_of_line char. Don't use strlen
4485 here as the end_of_line char may not be a NUL. */
4486 past_reloc = cp + 1 + len;
4487 for (cp = past_reloc; !is_end_of_line[(unsigned char) *cp++]; )
4488 ;
4489 second = cp - past_reloc;
4490
4491 /* Allocate and copy string. The trailing NUL shouldn't
4492 be necessary, but be safe. */
4493 tmpbuf = xmalloc (first + second + 2);
4494 memcpy (tmpbuf, input_line_pointer, first);
4495 if (second != 0 && *past_reloc != ' ')
4496 /* Replace the relocation token with ' ', so that
4497 errors like foo@GOTOFF1 will be detected. */
4498 tmpbuf[first++] = ' ';
4499 memcpy (tmpbuf + first, past_reloc, second);
4500 tmpbuf[first + second] = '\0';
4501 return tmpbuf;
4502 }
4503
4504 as_bad (_("@%s reloc is not supported with %d-bit output format"),
4505 gotrel[j].str, 1 << (5 + object_64bit));
4506 return NULL;
4507 }
4508 }
4509
4510 /* Might be a symbol version string. Don't as_bad here. */
4511 return NULL;
4512 }
4513
4514 void
4515 x86_cons (expressionS *exp, int size)
4516 {
4517 if (size == 4 || (object_64bit && size == 8))
4518 {
4519 /* Handle @GOTOFF and the like in an expression. */
4520 char *save;
4521 char *gotfree_input_line;
4522 int adjust;
4523
4524 save = input_line_pointer;
4525 gotfree_input_line = lex_got (&got_reloc, &adjust, NULL);
4526 if (gotfree_input_line)
4527 input_line_pointer = gotfree_input_line;
4528
4529 expression (exp);
4530
4531 if (gotfree_input_line)
4532 {
4533 /* expression () has merrily parsed up to the end of line,
4534 or a comma - in the wrong buffer. Transfer how far
4535 input_line_pointer has moved to the right buffer. */
4536 input_line_pointer = (save
4537 + (input_line_pointer - gotfree_input_line)
4538 + adjust);
4539 free (gotfree_input_line);
4540 }
4541 }
4542 else
4543 expression (exp);
4544 }
4545 #endif
4546
4547 static void signed_cons (int size)
4548 {
4549 if (flag_code == CODE_64BIT)
4550 cons_sign = 1;
4551 cons (size);
4552 cons_sign = -1;
4553 }
4554
4555 #ifdef TE_PE
4556 static void
4557 pe_directive_secrel (dummy)
4558 int dummy ATTRIBUTE_UNUSED;
4559 {
4560 expressionS exp;
4561
4562 do
4563 {
4564 expression (&exp);
4565 if (exp.X_op == O_symbol)
4566 exp.X_op = O_secrel;
4567
4568 emit_expr (&exp, 4);
4569 }
4570 while (*input_line_pointer++ == ',');
4571
4572 input_line_pointer--;
4573 demand_empty_rest_of_line ();
4574 }
4575 #endif
4576
4577 static int
4578 i386_immediate (char *imm_start)
4579 {
4580 char *save_input_line_pointer;
4581 char *gotfree_input_line;
4582 segT exp_seg = 0;
4583 expressionS *exp;
4584 unsigned int types = ~0U;
4585
4586 if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
4587 {
4588 as_bad (_("at most %d immediate operands are allowed"),
4589 MAX_IMMEDIATE_OPERANDS);
4590 return 0;
4591 }
4592
4593 exp = &im_expressions[i.imm_operands++];
4594 i.op[this_operand].imms = exp;
4595
4596 if (is_space_char (*imm_start))
4597 ++imm_start;
4598
4599 save_input_line_pointer = input_line_pointer;
4600 input_line_pointer = imm_start;
4601
4602 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL, &types);
4603 if (gotfree_input_line)
4604 input_line_pointer = gotfree_input_line;
4605
4606 exp_seg = expression (exp);
4607
4608 SKIP_WHITESPACE ();
4609 if (*input_line_pointer)
4610 as_bad (_("junk `%s' after expression"), input_line_pointer);
4611
4612 input_line_pointer = save_input_line_pointer;
4613 if (gotfree_input_line)
4614 free (gotfree_input_line);
4615
4616 if (exp->X_op == O_absent || exp->X_op == O_big)
4617 {
4618 /* Missing or bad expr becomes absolute 0. */
4619 as_bad (_("missing or invalid immediate expression `%s' taken as 0"),
4620 imm_start);
4621 exp->X_op = O_constant;
4622 exp->X_add_number = 0;
4623 exp->X_add_symbol = (symbolS *) 0;
4624 exp->X_op_symbol = (symbolS *) 0;
4625 }
4626 else if (exp->X_op == O_constant)
4627 {
4628 /* Size it properly later. */
4629 i.types[this_operand] |= Imm64;
4630 /* If BFD64, sign extend val. */
4631 if (!use_rela_relocations
4632 && (exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0)
4633 exp->X_add_number
4634 = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
4635 }
4636 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4637 else if (OUTPUT_FLAVOR == bfd_target_aout_flavour
4638 && exp_seg != absolute_section
4639 && exp_seg != text_section
4640 && exp_seg != data_section
4641 && exp_seg != bss_section
4642 && exp_seg != undefined_section
4643 && !bfd_is_com_section (exp_seg))
4644 {
4645 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
4646 return 0;
4647 }
4648 #endif
4649 else if (!intel_syntax && exp->X_op == O_register)
4650 {
4651 as_bad (_("illegal immediate register operand %s"), imm_start);
4652 return 0;
4653 }
4654 else
4655 {
4656 /* This is an address. The size of the address will be
4657 determined later, depending on destination register,
4658 suffix, or the default for the section. */
4659 i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64;
4660 i.types[this_operand] &= types;
4661 }
4662
4663 return 1;
4664 }
4665
4666 static char *
4667 i386_scale (char *scale)
4668 {
4669 offsetT val;
4670 char *save = input_line_pointer;
4671
4672 input_line_pointer = scale;
4673 val = get_absolute_expression ();
4674
4675 switch (val)
4676 {
4677 case 1:
4678 i.log2_scale_factor = 0;
4679 break;
4680 case 2:
4681 i.log2_scale_factor = 1;
4682 break;
4683 case 4:
4684 i.log2_scale_factor = 2;
4685 break;
4686 case 8:
4687 i.log2_scale_factor = 3;
4688 break;
4689 default:
4690 {
4691 char sep = *input_line_pointer;
4692
4693 *input_line_pointer = '\0';
4694 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
4695 scale);
4696 *input_line_pointer = sep;
4697 input_line_pointer = save;
4698 return NULL;
4699 }
4700 }
4701 if (i.log2_scale_factor != 0 && i.index_reg == 0)
4702 {
4703 as_warn (_("scale factor of %d without an index register"),
4704 1 << i.log2_scale_factor);
4705 #if SCALE1_WHEN_NO_INDEX
4706 i.log2_scale_factor = 0;
4707 #endif
4708 }
4709 scale = input_line_pointer;
4710 input_line_pointer = save;
4711 return scale;
4712 }
4713
4714 static int
4715 i386_displacement (char *disp_start, char *disp_end)
4716 {
4717 expressionS *exp;
4718 segT exp_seg = 0;
4719 char *save_input_line_pointer;
4720 char *gotfree_input_line;
4721 int bigdisp, override;
4722 unsigned int types = Disp;
4723
4724 if (i.disp_operands == MAX_MEMORY_OPERANDS)
4725 {
4726 as_bad (_("at most %d displacement operands are allowed"),
4727 MAX_MEMORY_OPERANDS);
4728 return 0;
4729 }
4730
4731 if ((i.types[this_operand] & JumpAbsolute)
4732 || !(current_templates->start->opcode_modifier & (Jump | JumpDword)))
4733 {
4734 bigdisp = Disp32;
4735 override = (i.prefix[ADDR_PREFIX] != 0);
4736 }
4737 else
4738 {
4739 /* For PC-relative branches, the width of the displacement
4740 is dependent upon data size, not address size. */
4741 bigdisp = 0;
4742 override = (i.prefix[DATA_PREFIX] != 0);
4743 }
4744 if (flag_code == CODE_64BIT)
4745 {
4746 if (!bigdisp)
4747 bigdisp = ((override || i.suffix == WORD_MNEM_SUFFIX)
4748 ? Disp16
4749 : Disp32S | Disp32);
4750 else if (!override)
4751 bigdisp = Disp64 | Disp32S | Disp32;
4752 }
4753 else
4754 {
4755 if (!bigdisp)
4756 {
4757 if (!override)
4758 override = (i.suffix == (flag_code != CODE_16BIT
4759 ? WORD_MNEM_SUFFIX
4760 : LONG_MNEM_SUFFIX));
4761 bigdisp = Disp32;
4762 }
4763 if ((flag_code == CODE_16BIT) ^ override)
4764 bigdisp = Disp16;
4765 }
4766 i.types[this_operand] |= bigdisp;
4767
4768 exp = &disp_expressions[i.disp_operands];
4769 i.op[this_operand].disps = exp;
4770 i.disp_operands++;
4771 save_input_line_pointer = input_line_pointer;
4772 input_line_pointer = disp_start;
4773 END_STRING_AND_SAVE (disp_end);
4774
4775 #ifndef GCC_ASM_O_HACK
4776 #define GCC_ASM_O_HACK 0
4777 #endif
4778 #if GCC_ASM_O_HACK
4779 END_STRING_AND_SAVE (disp_end + 1);
4780 if ((i.types[this_operand] & BaseIndex) != 0
4781 && displacement_string_end[-1] == '+')
4782 {
4783 /* This hack is to avoid a warning when using the "o"
4784 constraint within gcc asm statements.
4785 For instance:
4786
4787 #define _set_tssldt_desc(n,addr,limit,type) \
4788 __asm__ __volatile__ ( \
4789 "movw %w2,%0\n\t" \
4790 "movw %w1,2+%0\n\t" \
4791 "rorl $16,%1\n\t" \
4792 "movb %b1,4+%0\n\t" \
4793 "movb %4,5+%0\n\t" \
4794 "movb $0,6+%0\n\t" \
4795 "movb %h1,7+%0\n\t" \
4796 "rorl $16,%1" \
4797 : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
4798
4799 This works great except that the output assembler ends
4800 up looking a bit weird if it turns out that there is
4801 no offset. You end up producing code that looks like:
4802
4803 #APP
4804 movw $235,(%eax)
4805 movw %dx,2+(%eax)
4806 rorl $16,%edx
4807 movb %dl,4+(%eax)
4808 movb $137,5+(%eax)
4809 movb $0,6+(%eax)
4810 movb %dh,7+(%eax)
4811 rorl $16,%edx
4812 #NO_APP
4813
4814 So here we provide the missing zero. */
4815
4816 *displacement_string_end = '0';
4817 }
4818 #endif
4819 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL, &types);
4820 if (gotfree_input_line)
4821 input_line_pointer = gotfree_input_line;
4822
4823 exp_seg = expression (exp);
4824
4825 SKIP_WHITESPACE ();
4826 if (*input_line_pointer)
4827 as_bad (_("junk `%s' after expression"), input_line_pointer);
4828 #if GCC_ASM_O_HACK
4829 RESTORE_END_STRING (disp_end + 1);
4830 #endif
4831 RESTORE_END_STRING (disp_end);
4832 input_line_pointer = save_input_line_pointer;
4833 if (gotfree_input_line)
4834 free (gotfree_input_line);
4835
4836 /* We do this to make sure that the section symbol is in
4837 the symbol table. We will ultimately change the relocation
4838 to be relative to the beginning of the section. */
4839 if (i.reloc[this_operand] == BFD_RELOC_386_GOTOFF
4840 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
4841 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTOFF64)
4842 {
4843 if (exp->X_op != O_symbol)
4844 {
4845 as_bad (_("bad expression used with @%s"),
4846 (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
4847 ? "GOTPCREL"
4848 : "GOTOFF"));
4849 return 0;
4850 }
4851
4852 if (S_IS_LOCAL (exp->X_add_symbol)
4853 && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
4854 section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
4855 exp->X_op = O_subtract;
4856 exp->X_op_symbol = GOT_symbol;
4857 if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
4858 i.reloc[this_operand] = BFD_RELOC_32_PCREL;
4859 else if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTOFF64)
4860 i.reloc[this_operand] = BFD_RELOC_64;
4861 else
4862 i.reloc[this_operand] = BFD_RELOC_32;
4863 }
4864
4865 if (exp->X_op == O_absent || exp->X_op == O_big)
4866 {
4867 /* Missing or bad expr becomes absolute 0. */
4868 as_bad (_("missing or invalid displacement expression `%s' taken as 0"),
4869 disp_start);
4870 exp->X_op = O_constant;
4871 exp->X_add_number = 0;
4872 exp->X_add_symbol = (symbolS *) 0;
4873 exp->X_op_symbol = (symbolS *) 0;
4874 }
4875
4876 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4877 if (exp->X_op != O_constant
4878 && OUTPUT_FLAVOR == bfd_target_aout_flavour
4879 && exp_seg != absolute_section
4880 && exp_seg != text_section
4881 && exp_seg != data_section
4882 && exp_seg != bss_section
4883 && exp_seg != undefined_section
4884 && !bfd_is_com_section (exp_seg))
4885 {
4886 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
4887 return 0;
4888 }
4889 #endif
4890
4891 if (!(i.types[this_operand] & ~Disp))
4892 i.types[this_operand] &= types;
4893
4894 return 1;
4895 }
4896
4897 /* Make sure the memory operand we've been dealt is valid.
4898 Return 1 on success, 0 on a failure. */
4899
4900 static int
4901 i386_index_check (const char *operand_string)
4902 {
4903 int ok;
4904 #if INFER_ADDR_PREFIX
4905 int fudged = 0;
4906
4907 tryprefix:
4908 #endif
4909 ok = 1;
4910 if ((current_templates->start->cpu_flags & CpuSVME)
4911 && current_templates->end[-1].operand_types[0] == AnyMem)
4912 {
4913 /* Memory operands of SVME insns are special in that they only allow
4914 rAX as their memory address and ignore any segment override. */
4915 unsigned RegXX;
4916
4917 /* SKINIT is even more restrictive: it always requires EAX. */
4918 if (strcmp (current_templates->start->name, "skinit") == 0)
4919 RegXX = Reg32;
4920 else if (flag_code == CODE_64BIT)
4921 RegXX = i.prefix[ADDR_PREFIX] == 0 ? Reg64 : Reg32;
4922 else
4923 RegXX = ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0)
4924 ? Reg16
4925 : Reg32);
4926 if (!i.base_reg
4927 || !(i.base_reg->reg_type & Acc)
4928 || !(i.base_reg->reg_type & RegXX)
4929 || i.index_reg
4930 || (i.types[0] & Disp))
4931 ok = 0;
4932 }
4933 else if (flag_code == CODE_64BIT)
4934 {
4935 unsigned RegXX = (i.prefix[ADDR_PREFIX] == 0 ? Reg64 : Reg32);
4936
4937 if ((i.base_reg
4938 && ((i.base_reg->reg_type & RegXX) == 0)
4939 && (i.base_reg->reg_type != BaseIndex
4940 || i.index_reg))
4941 || (i.index_reg
4942 && ((i.index_reg->reg_type & (RegXX | BaseIndex))
4943 != (RegXX | BaseIndex))))
4944 ok = 0;
4945 }
4946 else
4947 {
4948 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
4949 {
4950 /* 16bit checks. */
4951 if ((i.base_reg
4952 && ((i.base_reg->reg_type & (Reg16 | BaseIndex | RegRex))
4953 != (Reg16 | BaseIndex)))
4954 || (i.index_reg
4955 && (((i.index_reg->reg_type & (Reg16 | BaseIndex))
4956 != (Reg16 | BaseIndex))
4957 || !(i.base_reg
4958 && i.base_reg->reg_num < 6
4959 && i.index_reg->reg_num >= 6
4960 && i.log2_scale_factor == 0))))
4961 ok = 0;
4962 }
4963 else
4964 {
4965 /* 32bit checks. */
4966 if ((i.base_reg
4967 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
4968 || (i.index_reg
4969 && ((i.index_reg->reg_type & (Reg32 | BaseIndex | RegRex))
4970 != (Reg32 | BaseIndex))))
4971 ok = 0;
4972 }
4973 }
4974 if (!ok)
4975 {
4976 #if INFER_ADDR_PREFIX
4977 if (i.prefix[ADDR_PREFIX] == 0)
4978 {
4979 i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
4980 i.prefixes += 1;
4981 /* Change the size of any displacement too. At most one of
4982 Disp16 or Disp32 is set.
4983 FIXME. There doesn't seem to be any real need for separate
4984 Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
4985 Removing them would probably clean up the code quite a lot. */
4986 if (flag_code != CODE_64BIT
4987 && (i.types[this_operand] & (Disp16 | Disp32)))
4988 i.types[this_operand] ^= (Disp16 | Disp32);
4989 fudged = 1;
4990 goto tryprefix;
4991 }
4992 if (fudged)
4993 as_bad (_("`%s' is not a valid base/index expression"),
4994 operand_string);
4995 else
4996 #endif
4997 as_bad (_("`%s' is not a valid %s bit base/index expression"),
4998 operand_string,
4999 flag_code_names[flag_code]);
5000 }
5001 return ok;
5002 }
5003
5004 /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
5005 on error. */
5006
5007 static int
5008 i386_operand (char *operand_string)
5009 {
5010 const reg_entry *r;
5011 char *end_op;
5012 char *op_string = operand_string;
5013
5014 if (is_space_char (*op_string))
5015 ++op_string;
5016
5017 /* We check for an absolute prefix (differentiating,
5018 for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
5019 if (*op_string == ABSOLUTE_PREFIX)
5020 {
5021 ++op_string;
5022 if (is_space_char (*op_string))
5023 ++op_string;
5024 i.types[this_operand] |= JumpAbsolute;
5025 }
5026
5027 /* Check if operand is a register. */
5028 if ((r = parse_register (op_string, &end_op)) != NULL)
5029 {
5030 /* Check for a segment override by searching for ':' after a
5031 segment register. */
5032 op_string = end_op;
5033 if (is_space_char (*op_string))
5034 ++op_string;
5035 if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
5036 {
5037 switch (r->reg_num)
5038 {
5039 case 0:
5040 i.seg[i.mem_operands] = &es;
5041 break;
5042 case 1:
5043 i.seg[i.mem_operands] = &cs;
5044 break;
5045 case 2:
5046 i.seg[i.mem_operands] = &ss;
5047 break;
5048 case 3:
5049 i.seg[i.mem_operands] = &ds;
5050 break;
5051 case 4:
5052 i.seg[i.mem_operands] = &fs;
5053 break;
5054 case 5:
5055 i.seg[i.mem_operands] = &gs;
5056 break;
5057 }
5058
5059 /* Skip the ':' and whitespace. */
5060 ++op_string;
5061 if (is_space_char (*op_string))
5062 ++op_string;
5063
5064 if (!is_digit_char (*op_string)
5065 && !is_identifier_char (*op_string)
5066 && *op_string != '('
5067 && *op_string != ABSOLUTE_PREFIX)
5068 {
5069 as_bad (_("bad memory operand `%s'"), op_string);
5070 return 0;
5071 }
5072 /* Handle case of %es:*foo. */
5073 if (*op_string == ABSOLUTE_PREFIX)
5074 {
5075 ++op_string;
5076 if (is_space_char (*op_string))
5077 ++op_string;
5078 i.types[this_operand] |= JumpAbsolute;
5079 }
5080 goto do_memory_reference;
5081 }
5082 if (*op_string)
5083 {
5084 as_bad (_("junk `%s' after register"), op_string);
5085 return 0;
5086 }
5087 i.types[this_operand] |= r->reg_type & ~BaseIndex;
5088 i.op[this_operand].regs = r;
5089 i.reg_operands++;
5090 }
5091 else if (*op_string == REGISTER_PREFIX)
5092 {
5093 as_bad (_("bad register name `%s'"), op_string);
5094 return 0;
5095 }
5096 else if (*op_string == IMMEDIATE_PREFIX)
5097 {
5098 ++op_string;
5099 if (i.types[this_operand] & JumpAbsolute)
5100 {
5101 as_bad (_("immediate operand illegal with absolute jump"));
5102 return 0;
5103 }
5104 if (!i386_immediate (op_string))
5105 return 0;
5106 }
5107 else if (is_digit_char (*op_string)
5108 || is_identifier_char (*op_string)
5109 || *op_string == '(')
5110 {
5111 /* This is a memory reference of some sort. */
5112 char *base_string;
5113
5114 /* Start and end of displacement string expression (if found). */
5115 char *displacement_string_start;
5116 char *displacement_string_end;
5117
5118 do_memory_reference:
5119 if ((i.mem_operands == 1
5120 && (current_templates->start->opcode_modifier & IsString) == 0)
5121 || i.mem_operands == 2)
5122 {
5123 as_bad (_("too many memory references for `%s'"),
5124 current_templates->start->name);
5125 return 0;
5126 }
5127
5128 /* Check for base index form. We detect the base index form by
5129 looking for an ')' at the end of the operand, searching
5130 for the '(' matching it, and finding a REGISTER_PREFIX or ','
5131 after the '('. */
5132 base_string = op_string + strlen (op_string);
5133
5134 --base_string;
5135 if (is_space_char (*base_string))
5136 --base_string;
5137
5138 /* If we only have a displacement, set-up for it to be parsed later. */
5139 displacement_string_start = op_string;
5140 displacement_string_end = base_string + 1;
5141
5142 if (*base_string == ')')
5143 {
5144 char *temp_string;
5145 unsigned int parens_balanced = 1;
5146 /* We've already checked that the number of left & right ()'s are
5147 equal, so this loop will not be infinite. */
5148 do
5149 {
5150 base_string--;
5151 if (*base_string == ')')
5152 parens_balanced++;
5153 if (*base_string == '(')
5154 parens_balanced--;
5155 }
5156 while (parens_balanced);
5157
5158 temp_string = base_string;
5159
5160 /* Skip past '(' and whitespace. */
5161 ++base_string;
5162 if (is_space_char (*base_string))
5163 ++base_string;
5164
5165 if (*base_string == ','
5166 || ((i.base_reg = parse_register (base_string, &end_op))
5167 != NULL))
5168 {
5169 displacement_string_end = temp_string;
5170
5171 i.types[this_operand] |= BaseIndex;
5172
5173 if (i.base_reg)
5174 {
5175 base_string = end_op;
5176 if (is_space_char (*base_string))
5177 ++base_string;
5178 }
5179
5180 /* There may be an index reg or scale factor here. */
5181 if (*base_string == ',')
5182 {
5183 ++base_string;
5184 if (is_space_char (*base_string))
5185 ++base_string;
5186
5187 if ((i.index_reg = parse_register (base_string, &end_op))
5188 != NULL)
5189 {
5190 base_string = end_op;
5191 if (is_space_char (*base_string))
5192 ++base_string;
5193 if (*base_string == ',')
5194 {
5195 ++base_string;
5196 if (is_space_char (*base_string))
5197 ++base_string;
5198 }
5199 else if (*base_string != ')')
5200 {
5201 as_bad (_("expecting `,' or `)' "
5202 "after index register in `%s'"),
5203 operand_string);
5204 return 0;
5205 }
5206 }
5207 else if (*base_string == REGISTER_PREFIX)
5208 {
5209 as_bad (_("bad register name `%s'"), base_string);
5210 return 0;
5211 }
5212
5213 /* Check for scale factor. */
5214 if (*base_string != ')')
5215 {
5216 char *end_scale = i386_scale (base_string);
5217
5218 if (!end_scale)
5219 return 0;
5220
5221 base_string = end_scale;
5222 if (is_space_char (*base_string))
5223 ++base_string;
5224 if (*base_string != ')')
5225 {
5226 as_bad (_("expecting `)' "
5227 "after scale factor in `%s'"),
5228 operand_string);
5229 return 0;
5230 }
5231 }
5232 else if (!i.index_reg)
5233 {
5234 as_bad (_("expecting index register or scale factor "
5235 "after `,'; got '%c'"),
5236 *base_string);
5237 return 0;
5238 }
5239 }
5240 else if (*base_string != ')')
5241 {
5242 as_bad (_("expecting `,' or `)' "
5243 "after base register in `%s'"),
5244 operand_string);
5245 return 0;
5246 }
5247 }
5248 else if (*base_string == REGISTER_PREFIX)
5249 {
5250 as_bad (_("bad register name `%s'"), base_string);
5251 return 0;
5252 }
5253 }
5254
5255 /* If there's an expression beginning the operand, parse it,
5256 assuming displacement_string_start and
5257 displacement_string_end are meaningful. */
5258 if (displacement_string_start != displacement_string_end)
5259 {
5260 if (!i386_displacement (displacement_string_start,
5261 displacement_string_end))
5262 return 0;
5263 }
5264
5265 /* Special case for (%dx) while doing input/output op. */
5266 if (i.base_reg
5267 && i.base_reg->reg_type == (Reg16 | InOutPortReg)
5268 && i.index_reg == 0
5269 && i.log2_scale_factor == 0
5270 && i.seg[i.mem_operands] == 0
5271 && (i.types[this_operand] & Disp) == 0)
5272 {
5273 i.types[this_operand] = InOutPortReg;
5274 return 1;
5275 }
5276
5277 if (i386_index_check (operand_string) == 0)
5278 return 0;
5279 i.mem_operands++;
5280 }
5281 else
5282 {
5283 /* It's not a memory operand; argh! */
5284 as_bad (_("invalid char %s beginning operand %d `%s'"),
5285 output_invalid (*op_string),
5286 this_operand + 1,
5287 op_string);
5288 return 0;
5289 }
5290 return 1; /* Normal return. */
5291 }
5292 \f
5293 /* md_estimate_size_before_relax()
5294
5295 Called just before relax() for rs_machine_dependent frags. The x86
5296 assembler uses these frags to handle variable size jump
5297 instructions.
5298
5299 Any symbol that is now undefined will not become defined.
5300 Return the correct fr_subtype in the frag.
5301 Return the initial "guess for variable size of frag" to caller.
5302 The guess is actually the growth beyond the fixed part. Whatever
5303 we do to grow the fixed or variable part contributes to our
5304 returned value. */
5305
5306 int
5307 md_estimate_size_before_relax (fragP, segment)
5308 fragS *fragP;
5309 segT segment;
5310 {
5311 /* We've already got fragP->fr_subtype right; all we have to do is
5312 check for un-relaxable symbols. On an ELF system, we can't relax
5313 an externally visible symbol, because it may be overridden by a
5314 shared library. */
5315 if (S_GET_SEGMENT (fragP->fr_symbol) != segment
5316 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5317 || (IS_ELF
5318 && (S_IS_EXTERNAL (fragP->fr_symbol)
5319 || S_IS_WEAK (fragP->fr_symbol)))
5320 #endif
5321 )
5322 {
5323 /* Symbol is undefined in this segment, or we need to keep a
5324 reloc so that weak symbols can be overridden. */
5325 int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
5326 enum bfd_reloc_code_real reloc_type;
5327 unsigned char *opcode;
5328 int old_fr_fix;
5329
5330 if (fragP->fr_var != NO_RELOC)
5331 reloc_type = fragP->fr_var;
5332 else if (size == 2)
5333 reloc_type = BFD_RELOC_16_PCREL;
5334 else
5335 reloc_type = BFD_RELOC_32_PCREL;
5336
5337 old_fr_fix = fragP->fr_fix;
5338 opcode = (unsigned char *) fragP->fr_opcode;
5339
5340 switch (TYPE_FROM_RELAX_STATE (fragP->fr_subtype))
5341 {
5342 case UNCOND_JUMP:
5343 /* Make jmp (0xeb) a (d)word displacement jump. */
5344 opcode[0] = 0xe9;
5345 fragP->fr_fix += size;
5346 fix_new (fragP, old_fr_fix, size,
5347 fragP->fr_symbol,
5348 fragP->fr_offset, 1,
5349 reloc_type);
5350 break;
5351
5352 case COND_JUMP86:
5353 if (size == 2
5354 && (!no_cond_jump_promotion || fragP->fr_var != NO_RELOC))
5355 {
5356 /* Negate the condition, and branch past an
5357 unconditional jump. */
5358 opcode[0] ^= 1;
5359 opcode[1] = 3;
5360 /* Insert an unconditional jump. */
5361 opcode[2] = 0xe9;
5362 /* We added two extra opcode bytes, and have a two byte
5363 offset. */
5364 fragP->fr_fix += 2 + 2;
5365 fix_new (fragP, old_fr_fix + 2, 2,
5366 fragP->fr_symbol,
5367 fragP->fr_offset, 1,
5368 reloc_type);
5369 break;
5370 }
5371 /* Fall through. */
5372
5373 case COND_JUMP:
5374 if (no_cond_jump_promotion && fragP->fr_var == NO_RELOC)
5375 {
5376 fixS *fixP;
5377
5378 fragP->fr_fix += 1;
5379 fixP = fix_new (fragP, old_fr_fix, 1,
5380 fragP->fr_symbol,
5381 fragP->fr_offset, 1,
5382 BFD_RELOC_8_PCREL);
5383 fixP->fx_signed = 1;
5384 break;
5385 }
5386
5387 /* This changes the byte-displacement jump 0x7N
5388 to the (d)word-displacement jump 0x0f,0x8N. */
5389 opcode[1] = opcode[0] + 0x10;
5390 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
5391 /* We've added an opcode byte. */
5392 fragP->fr_fix += 1 + size;
5393 fix_new (fragP, old_fr_fix + 1, size,
5394 fragP->fr_symbol,
5395 fragP->fr_offset, 1,
5396 reloc_type);
5397 break;
5398
5399 default:
5400 BAD_CASE (fragP->fr_subtype);
5401 break;
5402 }
5403 frag_wane (fragP);
5404 return fragP->fr_fix - old_fr_fix;
5405 }
5406
5407 /* Guess size depending on current relax state. Initially the relax
5408 state will correspond to a short jump and we return 1, because
5409 the variable part of the frag (the branch offset) is one byte
5410 long. However, we can relax a section more than once and in that
5411 case we must either set fr_subtype back to the unrelaxed state,
5412 or return the value for the appropriate branch. */
5413 return md_relax_table[fragP->fr_subtype].rlx_length;
5414 }
5415
5416 /* Called after relax() is finished.
5417
5418 In: Address of frag.
5419 fr_type == rs_machine_dependent.
5420 fr_subtype is what the address relaxed to.
5421
5422 Out: Any fixSs and constants are set up.
5423 Caller will turn frag into a ".space 0". */
5424
5425 void
5426 md_convert_frag (abfd, sec, fragP)
5427 bfd *abfd ATTRIBUTE_UNUSED;
5428 segT sec ATTRIBUTE_UNUSED;
5429 fragS *fragP;
5430 {
5431 unsigned char *opcode;
5432 unsigned char *where_to_put_displacement = NULL;
5433 offsetT target_address;
5434 offsetT opcode_address;
5435 unsigned int extension = 0;
5436 offsetT displacement_from_opcode_start;
5437
5438 opcode = (unsigned char *) fragP->fr_opcode;
5439
5440 /* Address we want to reach in file space. */
5441 target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
5442
5443 /* Address opcode resides at in file space. */
5444 opcode_address = fragP->fr_address + fragP->fr_fix;
5445
5446 /* Displacement from opcode start to fill into instruction. */
5447 displacement_from_opcode_start = target_address - opcode_address;
5448
5449 if ((fragP->fr_subtype & BIG) == 0)
5450 {
5451 /* Don't have to change opcode. */
5452 extension = 1; /* 1 opcode + 1 displacement */
5453 where_to_put_displacement = &opcode[1];
5454 }
5455 else
5456 {
5457 if (no_cond_jump_promotion
5458 && TYPE_FROM_RELAX_STATE (fragP->fr_subtype) != UNCOND_JUMP)
5459 as_warn_where (fragP->fr_file, fragP->fr_line,
5460 _("long jump required"));
5461
5462 switch (fragP->fr_subtype)
5463 {
5464 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
5465 extension = 4; /* 1 opcode + 4 displacement */
5466 opcode[0] = 0xe9;
5467 where_to_put_displacement = &opcode[1];
5468 break;
5469
5470 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
5471 extension = 2; /* 1 opcode + 2 displacement */
5472 opcode[0] = 0xe9;
5473 where_to_put_displacement = &opcode[1];
5474 break;
5475
5476 case ENCODE_RELAX_STATE (COND_JUMP, BIG):
5477 case ENCODE_RELAX_STATE (COND_JUMP86, BIG):
5478 extension = 5; /* 2 opcode + 4 displacement */
5479 opcode[1] = opcode[0] + 0x10;
5480 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
5481 where_to_put_displacement = &opcode[2];
5482 break;
5483
5484 case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
5485 extension = 3; /* 2 opcode + 2 displacement */
5486 opcode[1] = opcode[0] + 0x10;
5487 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
5488 where_to_put_displacement = &opcode[2];
5489 break;
5490
5491 case ENCODE_RELAX_STATE (COND_JUMP86, BIG16):
5492 extension = 4;
5493 opcode[0] ^= 1;
5494 opcode[1] = 3;
5495 opcode[2] = 0xe9;
5496 where_to_put_displacement = &opcode[3];
5497 break;
5498
5499 default:
5500 BAD_CASE (fragP->fr_subtype);
5501 break;
5502 }
5503 }
5504
5505 /* If size if less then four we are sure that the operand fits,
5506 but if it's 4, then it could be that the displacement is larger
5507 then -/+ 2GB. */
5508 if (DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype) == 4
5509 && object_64bit
5510 && ((addressT) (displacement_from_opcode_start - extension
5511 + ((addressT) 1 << 31))
5512 > (((addressT) 2 << 31) - 1)))
5513 {
5514 as_bad_where (fragP->fr_file, fragP->fr_line,
5515 _("jump target out of range"));
5516 /* Make us emit 0. */
5517 displacement_from_opcode_start = extension;
5518 }
5519 /* Now put displacement after opcode. */
5520 md_number_to_chars ((char *) where_to_put_displacement,
5521 (valueT) (displacement_from_opcode_start - extension),
5522 DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
5523 fragP->fr_fix += extension;
5524 }
5525 \f
5526 /* Size of byte displacement jmp. */
5527 int md_short_jump_size = 2;
5528
5529 /* Size of dword displacement jmp. */
5530 int md_long_jump_size = 5;
5531
5532 void
5533 md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
5534 char *ptr;
5535 addressT from_addr, to_addr;
5536 fragS *frag ATTRIBUTE_UNUSED;
5537 symbolS *to_symbol ATTRIBUTE_UNUSED;
5538 {
5539 offsetT offset;
5540
5541 offset = to_addr - (from_addr + 2);
5542 /* Opcode for byte-disp jump. */
5543 md_number_to_chars (ptr, (valueT) 0xeb, 1);
5544 md_number_to_chars (ptr + 1, (valueT) offset, 1);
5545 }
5546
5547 void
5548 md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
5549 char *ptr;
5550 addressT from_addr, to_addr;
5551 fragS *frag ATTRIBUTE_UNUSED;
5552 symbolS *to_symbol ATTRIBUTE_UNUSED;
5553 {
5554 offsetT offset;
5555
5556 offset = to_addr - (from_addr + 5);
5557 md_number_to_chars (ptr, (valueT) 0xe9, 1);
5558 md_number_to_chars (ptr + 1, (valueT) offset, 4);
5559 }
5560 \f
5561 /* Apply a fixup (fixS) to segment data, once it has been determined
5562 by our caller that we have all the info we need to fix it up.
5563
5564 On the 386, immediates, displacements, and data pointers are all in
5565 the same (little-endian) format, so we don't need to care about which
5566 we are handling. */
5567
5568 void
5569 md_apply_fix (fixP, valP, seg)
5570 /* The fix we're to put in. */
5571 fixS *fixP;
5572 /* Pointer to the value of the bits. */
5573 valueT *valP;
5574 /* Segment fix is from. */
5575 segT seg ATTRIBUTE_UNUSED;
5576 {
5577 char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
5578 valueT value = *valP;
5579
5580 #if !defined (TE_Mach)
5581 if (fixP->fx_pcrel)
5582 {
5583 switch (fixP->fx_r_type)
5584 {
5585 default:
5586 break;
5587
5588 case BFD_RELOC_64:
5589 fixP->fx_r_type = BFD_RELOC_64_PCREL;
5590 break;
5591 case BFD_RELOC_32:
5592 case BFD_RELOC_X86_64_32S:
5593 fixP->fx_r_type = BFD_RELOC_32_PCREL;
5594 break;
5595 case BFD_RELOC_16:
5596 fixP->fx_r_type = BFD_RELOC_16_PCREL;
5597 break;
5598 case BFD_RELOC_8:
5599 fixP->fx_r_type = BFD_RELOC_8_PCREL;
5600 break;
5601 }
5602 }
5603
5604 if (fixP->fx_addsy != NULL
5605 && (fixP->fx_r_type == BFD_RELOC_32_PCREL
5606 || fixP->fx_r_type == BFD_RELOC_64_PCREL
5607 || fixP->fx_r_type == BFD_RELOC_16_PCREL
5608 || fixP->fx_r_type == BFD_RELOC_8_PCREL)
5609 && !use_rela_relocations)
5610 {
5611 /* This is a hack. There should be a better way to handle this.
5612 This covers for the fact that bfd_install_relocation will
5613 subtract the current location (for partial_inplace, PC relative
5614 relocations); see more below. */
5615 #ifndef OBJ_AOUT
5616 if (IS_ELF
5617 #ifdef TE_PE
5618 || OUTPUT_FLAVOR == bfd_target_coff_flavour
5619 #endif
5620 )
5621 value += fixP->fx_where + fixP->fx_frag->fr_address;
5622 #endif
5623 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5624 if (IS_ELF)
5625 {
5626 segT sym_seg = S_GET_SEGMENT (fixP->fx_addsy);
5627
5628 if ((sym_seg == seg
5629 || (symbol_section_p (fixP->fx_addsy)
5630 && sym_seg != absolute_section))
5631 && !generic_force_reloc (fixP))
5632 {
5633 /* Yes, we add the values in twice. This is because
5634 bfd_install_relocation subtracts them out again. I think
5635 bfd_install_relocation is broken, but I don't dare change
5636 it. FIXME. */
5637 value += fixP->fx_where + fixP->fx_frag->fr_address;
5638 }
5639 }
5640 #endif
5641 #if defined (OBJ_COFF) && defined (TE_PE)
5642 /* For some reason, the PE format does not store a
5643 section address offset for a PC relative symbol. */
5644 if (S_GET_SEGMENT (fixP->fx_addsy) != seg
5645 || S_IS_WEAK (fixP->fx_addsy))
5646 value += md_pcrel_from (fixP);
5647 #endif
5648 }
5649
5650 /* Fix a few things - the dynamic linker expects certain values here,
5651 and we must not disappoint it. */
5652 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5653 if (IS_ELF && fixP->fx_addsy)
5654 switch (fixP->fx_r_type)
5655 {
5656 case BFD_RELOC_386_PLT32:
5657 case BFD_RELOC_X86_64_PLT32:
5658 /* Make the jump instruction point to the address of the operand. At
5659 runtime we merely add the offset to the actual PLT entry. */
5660 value = -4;
5661 break;
5662
5663 case BFD_RELOC_386_TLS_GD:
5664 case BFD_RELOC_386_TLS_LDM:
5665 case BFD_RELOC_386_TLS_IE_32:
5666 case BFD_RELOC_386_TLS_IE:
5667 case BFD_RELOC_386_TLS_GOTIE:
5668 case BFD_RELOC_386_TLS_GOTDESC:
5669 case BFD_RELOC_X86_64_TLSGD:
5670 case BFD_RELOC_X86_64_TLSLD:
5671 case BFD_RELOC_X86_64_GOTTPOFF:
5672 case BFD_RELOC_X86_64_GOTPC32_TLSDESC:
5673 value = 0; /* Fully resolved at runtime. No addend. */
5674 /* Fallthrough */
5675 case BFD_RELOC_386_TLS_LE:
5676 case BFD_RELOC_386_TLS_LDO_32:
5677 case BFD_RELOC_386_TLS_LE_32:
5678 case BFD_RELOC_X86_64_DTPOFF32:
5679 case BFD_RELOC_X86_64_DTPOFF64:
5680 case BFD_RELOC_X86_64_TPOFF32:
5681 case BFD_RELOC_X86_64_TPOFF64:
5682 S_SET_THREAD_LOCAL (fixP->fx_addsy);
5683 break;
5684
5685 case BFD_RELOC_386_TLS_DESC_CALL:
5686 case BFD_RELOC_X86_64_TLSDESC_CALL:
5687 value = 0; /* Fully resolved at runtime. No addend. */
5688 S_SET_THREAD_LOCAL (fixP->fx_addsy);
5689 fixP->fx_done = 0;
5690 return;
5691
5692 case BFD_RELOC_386_GOT32:
5693 case BFD_RELOC_X86_64_GOT32:
5694 value = 0; /* Fully resolved at runtime. No addend. */
5695 break;
5696
5697 case BFD_RELOC_VTABLE_INHERIT:
5698 case BFD_RELOC_VTABLE_ENTRY:
5699 fixP->fx_done = 0;
5700 return;
5701
5702 default:
5703 break;
5704 }
5705 #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
5706 *valP = value;
5707 #endif /* !defined (TE_Mach) */
5708
5709 /* Are we finished with this relocation now? */
5710 if (fixP->fx_addsy == NULL)
5711 fixP->fx_done = 1;
5712 else if (use_rela_relocations)
5713 {
5714 fixP->fx_no_overflow = 1;
5715 /* Remember value for tc_gen_reloc. */
5716 fixP->fx_addnumber = value;
5717 value = 0;
5718 }
5719
5720 md_number_to_chars (p, value, fixP->fx_size);
5721 }
5722 \f
5723 #define MAX_LITTLENUMS 6
5724
5725 /* Turn the string pointed to by litP into a floating point constant
5726 of type TYPE, and emit the appropriate bytes. The number of
5727 LITTLENUMS emitted is stored in *SIZEP. An error message is
5728 returned, or NULL on OK. */
5729
5730 char *
5731 md_atof (type, litP, sizeP)
5732 int type;
5733 char *litP;
5734 int *sizeP;
5735 {
5736 int prec;
5737 LITTLENUM_TYPE words[MAX_LITTLENUMS];
5738 LITTLENUM_TYPE *wordP;
5739 char *t;
5740
5741 switch (type)
5742 {
5743 case 'f':
5744 case 'F':
5745 prec = 2;
5746 break;
5747
5748 case 'd':
5749 case 'D':
5750 prec = 4;
5751 break;
5752
5753 case 'x':
5754 case 'X':
5755 prec = 5;
5756 break;
5757
5758 default:
5759 *sizeP = 0;
5760 return _("Bad call to md_atof ()");
5761 }
5762 t = atof_ieee (input_line_pointer, type, words);
5763 if (t)
5764 input_line_pointer = t;
5765
5766 *sizeP = prec * sizeof (LITTLENUM_TYPE);
5767 /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
5768 the bigendian 386. */
5769 for (wordP = words + prec - 1; prec--;)
5770 {
5771 md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
5772 litP += sizeof (LITTLENUM_TYPE);
5773 }
5774 return 0;
5775 }
5776 \f
5777 static char output_invalid_buf[sizeof (unsigned char) * 2 + 6];
5778
5779 static char *
5780 output_invalid (int c)
5781 {
5782 if (ISPRINT (c))
5783 snprintf (output_invalid_buf, sizeof (output_invalid_buf),
5784 "'%c'", c);
5785 else
5786 snprintf (output_invalid_buf, sizeof (output_invalid_buf),
5787 "(0x%x)", (unsigned char) c);
5788 return output_invalid_buf;
5789 }
5790
5791 /* REG_STRING starts *before* REGISTER_PREFIX. */
5792
5793 static const reg_entry *
5794 parse_real_register (char *reg_string, char **end_op)
5795 {
5796 char *s = reg_string;
5797 char *p;
5798 char reg_name_given[MAX_REG_NAME_SIZE + 1];
5799 const reg_entry *r;
5800
5801 /* Skip possible REGISTER_PREFIX and possible whitespace. */
5802 if (*s == REGISTER_PREFIX)
5803 ++s;
5804
5805 if (is_space_char (*s))
5806 ++s;
5807
5808 p = reg_name_given;
5809 while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
5810 {
5811 if (p >= reg_name_given + MAX_REG_NAME_SIZE)
5812 return (const reg_entry *) NULL;
5813 s++;
5814 }
5815
5816 /* For naked regs, make sure that we are not dealing with an identifier.
5817 This prevents confusing an identifier like `eax_var' with register
5818 `eax'. */
5819 if (allow_naked_reg && identifier_chars[(unsigned char) *s])
5820 return (const reg_entry *) NULL;
5821
5822 *end_op = s;
5823
5824 r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
5825
5826 /* Handle floating point regs, allowing spaces in the (i) part. */
5827 if (r == i386_regtab /* %st is first entry of table */)
5828 {
5829 if (is_space_char (*s))
5830 ++s;
5831 if (*s == '(')
5832 {
5833 ++s;
5834 if (is_space_char (*s))
5835 ++s;
5836 if (*s >= '0' && *s <= '7')
5837 {
5838 int fpr = *s - '0';
5839 ++s;
5840 if (is_space_char (*s))
5841 ++s;
5842 if (*s == ')')
5843 {
5844 *end_op = s + 1;
5845 r = hash_find (reg_hash, "st(0)");
5846 know (r);
5847 return r + fpr;
5848 }
5849 }
5850 /* We have "%st(" then garbage. */
5851 return (const reg_entry *) NULL;
5852 }
5853 }
5854
5855 if (r != NULL
5856 && ((r->reg_flags & (RegRex64 | RegRex)) | (r->reg_type & Reg64)) != 0
5857 && (r->reg_type != Control || !(cpu_arch_flags & CpuSledgehammer))
5858 && flag_code != CODE_64BIT)
5859 return (const reg_entry *) NULL;
5860
5861 return r;
5862 }
5863
5864 /* REG_STRING starts *before* REGISTER_PREFIX. */
5865
5866 static const reg_entry *
5867 parse_register (char *reg_string, char **end_op)
5868 {
5869 const reg_entry *r;
5870
5871 if (*reg_string == REGISTER_PREFIX || allow_naked_reg)
5872 r = parse_real_register (reg_string, end_op);
5873 else
5874 r = NULL;
5875 if (!r)
5876 {
5877 char *save = input_line_pointer;
5878 char c;
5879 symbolS *symbolP;
5880
5881 input_line_pointer = reg_string;
5882 c = get_symbol_end ();
5883 symbolP = symbol_find (reg_string);
5884 if (symbolP && S_GET_SEGMENT (symbolP) == reg_section)
5885 {
5886 const expressionS *e = symbol_get_value_expression (symbolP);
5887
5888 know (e->X_op == O_register);
5889 know (e->X_add_number >= 0
5890 && (valueT) e->X_add_number < i386_regtab_size);
5891 r = i386_regtab + e->X_add_number;
5892 *end_op = input_line_pointer;
5893 }
5894 *input_line_pointer = c;
5895 input_line_pointer = save;
5896 }
5897 return r;
5898 }
5899
5900 int
5901 i386_parse_name (char *name, expressionS *e, char *nextcharP)
5902 {
5903 const reg_entry *r;
5904 char *end = input_line_pointer;
5905
5906 *end = *nextcharP;
5907 r = parse_register (name, &input_line_pointer);
5908 if (r && end <= input_line_pointer)
5909 {
5910 *nextcharP = *input_line_pointer;
5911 *input_line_pointer = 0;
5912 e->X_op = O_register;
5913 e->X_add_number = r - i386_regtab;
5914 return 1;
5915 }
5916 input_line_pointer = end;
5917 *end = 0;
5918 return 0;
5919 }
5920
5921 void
5922 md_operand (expressionS *e)
5923 {
5924 if (*input_line_pointer == REGISTER_PREFIX)
5925 {
5926 char *end;
5927 const reg_entry *r = parse_real_register (input_line_pointer, &end);
5928
5929 if (r)
5930 {
5931 e->X_op = O_register;
5932 e->X_add_number = r - i386_regtab;
5933 input_line_pointer = end;
5934 }
5935 }
5936 }
5937
5938 \f
5939 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5940 const char *md_shortopts = "kVQ:sqn";
5941 #else
5942 const char *md_shortopts = "qn";
5943 #endif
5944
5945 #define OPTION_32 (OPTION_MD_BASE + 0)
5946 #define OPTION_64 (OPTION_MD_BASE + 1)
5947 #define OPTION_DIVIDE (OPTION_MD_BASE + 2)
5948 #define OPTION_MARCH (OPTION_MD_BASE + 3)
5949 #define OPTION_MTUNE (OPTION_MD_BASE + 4)
5950
5951 struct option md_longopts[] =
5952 {
5953 {"32", no_argument, NULL, OPTION_32},
5954 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) || defined(TE_PEP)
5955 {"64", no_argument, NULL, OPTION_64},
5956 #endif
5957 {"divide", no_argument, NULL, OPTION_DIVIDE},
5958 {"march", required_argument, NULL, OPTION_MARCH},
5959 {"mtune", required_argument, NULL, OPTION_MTUNE},
5960 {NULL, no_argument, NULL, 0}
5961 };
5962 size_t md_longopts_size = sizeof (md_longopts);
5963
5964 int
5965 md_parse_option (int c, char *arg)
5966 {
5967 unsigned int i;
5968
5969 switch (c)
5970 {
5971 case 'n':
5972 optimize_align_code = 0;
5973 break;
5974
5975 case 'q':
5976 quiet_warnings = 1;
5977 break;
5978
5979 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5980 /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
5981 should be emitted or not. FIXME: Not implemented. */
5982 case 'Q':
5983 break;
5984
5985 /* -V: SVR4 argument to print version ID. */
5986 case 'V':
5987 print_version_id ();
5988 break;
5989
5990 /* -k: Ignore for FreeBSD compatibility. */
5991 case 'k':
5992 break;
5993
5994 case 's':
5995 /* -s: On i386 Solaris, this tells the native assembler to use
5996 .stab instead of .stab.excl. We always use .stab anyhow. */
5997 break;
5998 #endif
5999 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) || defined(TE_PEP)
6000 case OPTION_64:
6001 {
6002 const char **list, **l;
6003
6004 list = bfd_target_list ();
6005 for (l = list; *l != NULL; l++)
6006 if (CONST_STRNEQ (*l, "elf64-x86-64")
6007 || strcmp (*l, "coff-x86-64") == 0
6008 || strcmp (*l, "pe-x86-64") == 0
6009 || strcmp (*l, "pei-x86-64") == 0)
6010 {
6011 default_arch = "x86_64";
6012 break;
6013 }
6014 if (*l == NULL)
6015 as_fatal (_("No compiled in support for x86_64"));
6016 free (list);
6017 }
6018 break;
6019 #endif
6020
6021 case OPTION_32:
6022 default_arch = "i386";
6023 break;
6024
6025 case OPTION_DIVIDE:
6026 #ifdef SVR4_COMMENT_CHARS
6027 {
6028 char *n, *t;
6029 const char *s;
6030
6031 n = (char *) xmalloc (strlen (i386_comment_chars) + 1);
6032 t = n;
6033 for (s = i386_comment_chars; *s != '\0'; s++)
6034 if (*s != '/')
6035 *t++ = *s;
6036 *t = '\0';
6037 i386_comment_chars = n;
6038 }
6039 #endif
6040 break;
6041
6042 case OPTION_MARCH:
6043 if (*arg == '.')
6044 as_fatal (_("Invalid -march= option: `%s'"), arg);
6045 for (i = 0; i < ARRAY_SIZE (cpu_arch); i++)
6046 {
6047 if (strcmp (arg, cpu_arch [i].name) == 0)
6048 {
6049 cpu_arch_isa = cpu_arch[i].type;
6050 cpu_arch_isa_flags = cpu_arch[i].flags;
6051 if (!cpu_arch_tune_set)
6052 {
6053 cpu_arch_tune = cpu_arch_isa;
6054 cpu_arch_tune_flags = cpu_arch_isa_flags;
6055 }
6056 break;
6057 }
6058 }
6059 if (i >= ARRAY_SIZE (cpu_arch))
6060 as_fatal (_("Invalid -march= option: `%s'"), arg);
6061 break;
6062
6063 case OPTION_MTUNE:
6064 if (*arg == '.')
6065 as_fatal (_("Invalid -mtune= option: `%s'"), arg);
6066 for (i = 0; i < ARRAY_SIZE (cpu_arch); i++)
6067 {
6068 if (strcmp (arg, cpu_arch [i].name) == 0)
6069 {
6070 cpu_arch_tune_set = 1;
6071 cpu_arch_tune = cpu_arch [i].type;
6072 cpu_arch_tune_flags = cpu_arch[i].flags;
6073 break;
6074 }
6075 }
6076 if (i >= ARRAY_SIZE (cpu_arch))
6077 as_fatal (_("Invalid -mtune= option: `%s'"), arg);
6078 break;
6079
6080 default:
6081 return 0;
6082 }
6083 return 1;
6084 }
6085
6086 void
6087 md_show_usage (stream)
6088 FILE *stream;
6089 {
6090 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
6091 fprintf (stream, _("\
6092 -Q ignored\n\
6093 -V print assembler version number\n\
6094 -k ignored\n"));
6095 #endif
6096 fprintf (stream, _("\
6097 -n Do not optimize code alignment\n\
6098 -q quieten some warnings\n"));
6099 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
6100 fprintf (stream, _("\
6101 -s ignored\n"));
6102 #endif
6103 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) || defined(TE_PEP)
6104 fprintf (stream, _("\
6105 --32/--64 generate 32bit/64bit code\n"));
6106 #endif
6107 #ifdef SVR4_COMMENT_CHARS
6108 fprintf (stream, _("\
6109 --divide do not treat `/' as a comment character\n"));
6110 #else
6111 fprintf (stream, _("\
6112 --divide ignored\n"));
6113 #endif
6114 fprintf (stream, _("\
6115 -march=CPU/-mtune=CPU generate code/optimize for CPU, where CPU is one of:\n\
6116 i386, i486, pentium, pentiumpro, pentium4, nocona,\n\
6117 core, core2, k6, athlon, k8, generic32, generic64\n"));
6118
6119 }
6120
6121 #if defined(TE_PEP)
6122 const char *
6123 x86_64_target_format (void)
6124 {
6125 if (strcmp (default_arch, "x86_64") == 0)
6126 {
6127 set_code_flag (CODE_64BIT);
6128 return COFF_TARGET_FORMAT;
6129 }
6130 else if (strcmp (default_arch, "i386") == 0)
6131 {
6132 set_code_flag (CODE_32BIT);
6133 return "coff-i386";
6134 }
6135
6136 as_fatal (_("Unknown architecture"));
6137 return NULL;
6138 }
6139 #endif
6140
6141 #if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
6142 || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
6143
6144 /* Pick the target format to use. */
6145
6146 const char *
6147 i386_target_format (void)
6148 {
6149 if (!strcmp (default_arch, "x86_64"))
6150 {
6151 set_code_flag (CODE_64BIT);
6152 if (cpu_arch_isa_flags == 0)
6153 cpu_arch_isa_flags = Cpu186|Cpu286|Cpu386|Cpu486
6154 |Cpu586|Cpu686|CpuP4|CpuMMX|CpuMMX2
6155 |CpuSSE|CpuSSE2;
6156 if (cpu_arch_tune_flags == 0)
6157 cpu_arch_tune_flags = Cpu186|Cpu286|Cpu386|Cpu486
6158 |Cpu586|Cpu686|CpuP4|CpuMMX|CpuMMX2
6159 |CpuSSE|CpuSSE2;
6160 }
6161 else if (!strcmp (default_arch, "i386"))
6162 {
6163 set_code_flag (CODE_32BIT);
6164 if (cpu_arch_isa_flags == 0)
6165 cpu_arch_isa_flags = Cpu186|Cpu286|Cpu386;
6166 if (cpu_arch_tune_flags == 0)
6167 cpu_arch_tune_flags = Cpu186|Cpu286|Cpu386;
6168 }
6169 else
6170 as_fatal (_("Unknown architecture"));
6171 switch (OUTPUT_FLAVOR)
6172 {
6173 #ifdef OBJ_MAYBE_AOUT
6174 case bfd_target_aout_flavour:
6175 return AOUT_TARGET_FORMAT;
6176 #endif
6177 #ifdef OBJ_MAYBE_COFF
6178 case bfd_target_coff_flavour:
6179 return "coff-i386";
6180 #endif
6181 #if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF)
6182 case bfd_target_elf_flavour:
6183 {
6184 if (flag_code == CODE_64BIT)
6185 {
6186 object_64bit = 1;
6187 use_rela_relocations = 1;
6188 }
6189 return flag_code == CODE_64BIT ? ELF_TARGET_FORMAT64 : ELF_TARGET_FORMAT;
6190 }
6191 #endif
6192 default:
6193 abort ();
6194 return NULL;
6195 }
6196 }
6197
6198 #endif /* OBJ_MAYBE_ more than one */
6199
6200 #if (defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
6201 void
6202 i386_elf_emit_arch_note (void)
6203 {
6204 if (IS_ELF && cpu_arch_name != NULL)
6205 {
6206 char *p;
6207 asection *seg = now_seg;
6208 subsegT subseg = now_subseg;
6209 Elf_Internal_Note i_note;
6210 Elf_External_Note e_note;
6211 asection *note_secp;
6212 int len;
6213
6214 /* Create the .note section. */
6215 note_secp = subseg_new (".note", 0);
6216 bfd_set_section_flags (stdoutput,
6217 note_secp,
6218 SEC_HAS_CONTENTS | SEC_READONLY);
6219
6220 /* Process the arch string. */
6221 len = strlen (cpu_arch_name);
6222
6223 i_note.namesz = len + 1;
6224 i_note.descsz = 0;
6225 i_note.type = NT_ARCH;
6226 p = frag_more (sizeof (e_note.namesz));
6227 md_number_to_chars (p, (valueT) i_note.namesz, sizeof (e_note.namesz));
6228 p = frag_more (sizeof (e_note.descsz));
6229 md_number_to_chars (p, (valueT) i_note.descsz, sizeof (e_note.descsz));
6230 p = frag_more (sizeof (e_note.type));
6231 md_number_to_chars (p, (valueT) i_note.type, sizeof (e_note.type));
6232 p = frag_more (len + 1);
6233 strcpy (p, cpu_arch_name);
6234
6235 frag_align (2, 0, 0);
6236
6237 subseg_set (seg, subseg);
6238 }
6239 }
6240 #endif
6241 \f
6242 symbolS *
6243 md_undefined_symbol (name)
6244 char *name;
6245 {
6246 if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
6247 && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
6248 && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
6249 && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
6250 {
6251 if (!GOT_symbol)
6252 {
6253 if (symbol_find (name))
6254 as_bad (_("GOT already in symbol table"));
6255 GOT_symbol = symbol_new (name, undefined_section,
6256 (valueT) 0, &zero_address_frag);
6257 };
6258 return GOT_symbol;
6259 }
6260 return 0;
6261 }
6262
6263 /* Round up a section size to the appropriate boundary. */
6264
6265 valueT
6266 md_section_align (segment, size)
6267 segT segment ATTRIBUTE_UNUSED;
6268 valueT size;
6269 {
6270 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
6271 if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
6272 {
6273 /* For a.out, force the section size to be aligned. If we don't do
6274 this, BFD will align it for us, but it will not write out the
6275 final bytes of the section. This may be a bug in BFD, but it is
6276 easier to fix it here since that is how the other a.out targets
6277 work. */
6278 int align;
6279
6280 align = bfd_get_section_alignment (stdoutput, segment);
6281 size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
6282 }
6283 #endif
6284
6285 return size;
6286 }
6287
6288 /* On the i386, PC-relative offsets are relative to the start of the
6289 next instruction. That is, the address of the offset, plus its
6290 size, since the offset is always the last part of the insn. */
6291
6292 long
6293 md_pcrel_from (fixS *fixP)
6294 {
6295 return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
6296 }
6297
6298 #ifndef I386COFF
6299
6300 static void
6301 s_bss (int ignore ATTRIBUTE_UNUSED)
6302 {
6303 int temp;
6304
6305 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
6306 if (IS_ELF)
6307 obj_elf_section_change_hook ();
6308 #endif
6309 temp = get_absolute_expression ();
6310 subseg_set (bss_section, (subsegT) temp);
6311 demand_empty_rest_of_line ();
6312 }
6313
6314 #endif
6315
6316 void
6317 i386_validate_fix (fixS *fixp)
6318 {
6319 if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
6320 {
6321 if (fixp->fx_r_type == BFD_RELOC_32_PCREL)
6322 {
6323 if (!object_64bit)
6324 abort ();
6325 fixp->fx_r_type = BFD_RELOC_X86_64_GOTPCREL;
6326 }
6327 else
6328 {
6329 if (!object_64bit)
6330 fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
6331 else
6332 fixp->fx_r_type = BFD_RELOC_X86_64_GOTOFF64;
6333 }
6334 fixp->fx_subsy = 0;
6335 }
6336 }
6337
6338 arelent *
6339 tc_gen_reloc (section, fixp)
6340 asection *section ATTRIBUTE_UNUSED;
6341 fixS *fixp;
6342 {
6343 arelent *rel;
6344 bfd_reloc_code_real_type code;
6345
6346 switch (fixp->fx_r_type)
6347 {
6348 case BFD_RELOC_X86_64_PLT32:
6349 case BFD_RELOC_X86_64_GOT32:
6350 case BFD_RELOC_X86_64_GOTPCREL:
6351 case BFD_RELOC_386_PLT32:
6352 case BFD_RELOC_386_GOT32:
6353 case BFD_RELOC_386_GOTOFF:
6354 case BFD_RELOC_386_GOTPC:
6355 case BFD_RELOC_386_TLS_GD:
6356 case BFD_RELOC_386_TLS_LDM:
6357 case BFD_RELOC_386_TLS_LDO_32:
6358 case BFD_RELOC_386_TLS_IE_32:
6359 case BFD_RELOC_386_TLS_IE:
6360 case BFD_RELOC_386_TLS_GOTIE:
6361 case BFD_RELOC_386_TLS_LE_32:
6362 case BFD_RELOC_386_TLS_LE:
6363 case BFD_RELOC_386_TLS_GOTDESC:
6364 case BFD_RELOC_386_TLS_DESC_CALL:
6365 case BFD_RELOC_X86_64_TLSGD:
6366 case BFD_RELOC_X86_64_TLSLD:
6367 case BFD_RELOC_X86_64_DTPOFF32:
6368 case BFD_RELOC_X86_64_DTPOFF64:
6369 case BFD_RELOC_X86_64_GOTTPOFF:
6370 case BFD_RELOC_X86_64_TPOFF32:
6371 case BFD_RELOC_X86_64_TPOFF64:
6372 case BFD_RELOC_X86_64_GOTOFF64:
6373 case BFD_RELOC_X86_64_GOTPC32:
6374 case BFD_RELOC_X86_64_GOT64:
6375 case BFD_RELOC_X86_64_GOTPCREL64:
6376 case BFD_RELOC_X86_64_GOTPC64:
6377 case BFD_RELOC_X86_64_GOTPLT64:
6378 case BFD_RELOC_X86_64_PLTOFF64:
6379 case BFD_RELOC_X86_64_GOTPC32_TLSDESC:
6380 case BFD_RELOC_X86_64_TLSDESC_CALL:
6381 case BFD_RELOC_RVA:
6382 case BFD_RELOC_VTABLE_ENTRY:
6383 case BFD_RELOC_VTABLE_INHERIT:
6384 #ifdef TE_PE
6385 case BFD_RELOC_32_SECREL:
6386 #endif
6387 code = fixp->fx_r_type;
6388 break;
6389 case BFD_RELOC_X86_64_32S:
6390 if (!fixp->fx_pcrel)
6391 {
6392 /* Don't turn BFD_RELOC_X86_64_32S into BFD_RELOC_32. */
6393 code = fixp->fx_r_type;
6394 break;
6395 }
6396 default:
6397 if (fixp->fx_pcrel)
6398 {
6399 switch (fixp->fx_size)
6400 {
6401 default:
6402 as_bad_where (fixp->fx_file, fixp->fx_line,
6403 _("can not do %d byte pc-relative relocation"),
6404 fixp->fx_size);
6405 code = BFD_RELOC_32_PCREL;
6406 break;
6407 case 1: code = BFD_RELOC_8_PCREL; break;
6408 case 2: code = BFD_RELOC_16_PCREL; break;
6409 case 4: code = BFD_RELOC_32_PCREL; break;
6410 #ifdef BFD64
6411 case 8: code = BFD_RELOC_64_PCREL; break;
6412 #endif
6413 }
6414 }
6415 else
6416 {
6417 switch (fixp->fx_size)
6418 {
6419 default:
6420 as_bad_where (fixp->fx_file, fixp->fx_line,
6421 _("can not do %d byte relocation"),
6422 fixp->fx_size);
6423 code = BFD_RELOC_32;
6424 break;
6425 case 1: code = BFD_RELOC_8; break;
6426 case 2: code = BFD_RELOC_16; break;
6427 case 4: code = BFD_RELOC_32; break;
6428 #ifdef BFD64
6429 case 8: code = BFD_RELOC_64; break;
6430 #endif
6431 }
6432 }
6433 break;
6434 }
6435
6436 if ((code == BFD_RELOC_32
6437 || code == BFD_RELOC_32_PCREL
6438 || code == BFD_RELOC_X86_64_32S)
6439 && GOT_symbol
6440 && fixp->fx_addsy == GOT_symbol)
6441 {
6442 if (!object_64bit)
6443 code = BFD_RELOC_386_GOTPC;
6444 else
6445 code = BFD_RELOC_X86_64_GOTPC32;
6446 }
6447 if ((code == BFD_RELOC_64 || code == BFD_RELOC_64_PCREL)
6448 && GOT_symbol
6449 && fixp->fx_addsy == GOT_symbol)
6450 {
6451 code = BFD_RELOC_X86_64_GOTPC64;
6452 }
6453
6454 rel = (arelent *) xmalloc (sizeof (arelent));
6455 rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
6456 *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
6457
6458 rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
6459
6460 if (!use_rela_relocations)
6461 {
6462 /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
6463 vtable entry to be used in the relocation's section offset. */
6464 if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
6465 rel->address = fixp->fx_offset;
6466
6467 rel->addend = 0;
6468 }
6469 /* Use the rela in 64bit mode. */
6470 else
6471 {
6472 if (!fixp->fx_pcrel)
6473 rel->addend = fixp->fx_offset;
6474 else
6475 switch (code)
6476 {
6477 case BFD_RELOC_X86_64_PLT32:
6478 case BFD_RELOC_X86_64_GOT32:
6479 case BFD_RELOC_X86_64_GOTPCREL:
6480 case BFD_RELOC_X86_64_TLSGD:
6481 case BFD_RELOC_X86_64_TLSLD:
6482 case BFD_RELOC_X86_64_GOTTPOFF:
6483 case BFD_RELOC_X86_64_GOTPC32_TLSDESC:
6484 case BFD_RELOC_X86_64_TLSDESC_CALL:
6485 rel->addend = fixp->fx_offset - fixp->fx_size;
6486 break;
6487 default:
6488 rel->addend = (section->vma
6489 - fixp->fx_size
6490 + fixp->fx_addnumber
6491 + md_pcrel_from (fixp));
6492 break;
6493 }
6494 }
6495
6496 rel->howto = bfd_reloc_type_lookup (stdoutput, code);
6497 if (rel->howto == NULL)
6498 {
6499 as_bad_where (fixp->fx_file, fixp->fx_line,
6500 _("cannot represent relocation type %s"),
6501 bfd_get_reloc_code_name (code));
6502 /* Set howto to a garbage value so that we can keep going. */
6503 rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
6504 assert (rel->howto != NULL);
6505 }
6506
6507 return rel;
6508 }
6509
6510 \f
6511 /* Parse operands using Intel syntax. This implements a recursive descent
6512 parser based on the BNF grammar published in Appendix B of the MASM 6.1
6513 Programmer's Guide.
6514
6515 FIXME: We do not recognize the full operand grammar defined in the MASM
6516 documentation. In particular, all the structure/union and
6517 high-level macro operands are missing.
6518
6519 Uppercase words are terminals, lower case words are non-terminals.
6520 Objects surrounded by double brackets '[[' ']]' are optional. Vertical
6521 bars '|' denote choices. Most grammar productions are implemented in
6522 functions called 'intel_<production>'.
6523
6524 Initial production is 'expr'.
6525
6526 addOp + | -
6527
6528 alpha [a-zA-Z]
6529
6530 binOp & | AND | \| | OR | ^ | XOR
6531
6532 byteRegister AL | AH | BL | BH | CL | CH | DL | DH
6533
6534 constant digits [[ radixOverride ]]
6535
6536 dataType BYTE | WORD | DWORD | FWORD | QWORD | TBYTE | OWORD | XMMWORD
6537
6538 digits decdigit
6539 | digits decdigit
6540 | digits hexdigit
6541
6542 decdigit [0-9]
6543
6544 e04 e04 addOp e05
6545 | e05
6546
6547 e05 e05 binOp e06
6548 | e06
6549
6550 e06 e06 mulOp e09
6551 | e09
6552
6553 e09 OFFSET e10
6554 | SHORT e10
6555 | + e10
6556 | - e10
6557 | ~ e10
6558 | NOT e10
6559 | e09 PTR e10
6560 | e09 : e10
6561 | e10
6562
6563 e10 e10 [ expr ]
6564 | e11
6565
6566 e11 ( expr )
6567 | [ expr ]
6568 | constant
6569 | dataType
6570 | id
6571 | $
6572 | register
6573
6574 => expr expr cmpOp e04
6575 | e04
6576
6577 gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
6578 | BP | EBP | SP | ESP | DI | EDI | SI | ESI
6579
6580 hexdigit a | b | c | d | e | f
6581 | A | B | C | D | E | F
6582
6583 id alpha
6584 | id alpha
6585 | id decdigit
6586
6587 mulOp * | / | % | MOD | << | SHL | >> | SHR
6588
6589 quote " | '
6590
6591 register specialRegister
6592 | gpRegister
6593 | byteRegister
6594
6595 segmentRegister CS | DS | ES | FS | GS | SS
6596
6597 specialRegister CR0 | CR2 | CR3 | CR4
6598 | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
6599 | TR3 | TR4 | TR5 | TR6 | TR7
6600
6601 We simplify the grammar in obvious places (e.g., register parsing is
6602 done by calling parse_register) and eliminate immediate left recursion
6603 to implement a recursive-descent parser.
6604
6605 expr e04 expr'
6606
6607 expr' cmpOp e04 expr'
6608 | Empty
6609
6610 e04 e05 e04'
6611
6612 e04' addOp e05 e04'
6613 | Empty
6614
6615 e05 e06 e05'
6616
6617 e05' binOp e06 e05'
6618 | Empty
6619
6620 e06 e09 e06'
6621
6622 e06' mulOp e09 e06'
6623 | Empty
6624
6625 e09 OFFSET e10 e09'
6626 | SHORT e10'
6627 | + e10'
6628 | - e10'
6629 | ~ e10'
6630 | NOT e10'
6631 | e10 e09'
6632
6633 e09' PTR e10 e09'
6634 | : e10 e09'
6635 | Empty
6636
6637 e10 e11 e10'
6638
6639 e10' [ expr ] e10'
6640 | Empty
6641
6642 e11 ( expr )
6643 | [ expr ]
6644 | BYTE
6645 | WORD
6646 | DWORD
6647 | FWORD
6648 | QWORD
6649 | TBYTE
6650 | OWORD
6651 | XMMWORD
6652 | .
6653 | $
6654 | register
6655 | id
6656 | constant */
6657
6658 /* Parsing structure for the intel syntax parser. Used to implement the
6659 semantic actions for the operand grammar. */
6660 struct intel_parser_s
6661 {
6662 char *op_string; /* The string being parsed. */
6663 int got_a_float; /* Whether the operand is a float. */
6664 int op_modifier; /* Operand modifier. */
6665 int is_mem; /* 1 if operand is memory reference. */
6666 int in_offset; /* >=1 if parsing operand of offset. */
6667 int in_bracket; /* >=1 if parsing operand in brackets. */
6668 const reg_entry *reg; /* Last register reference found. */
6669 char *disp; /* Displacement string being built. */
6670 char *next_operand; /* Resume point when splitting operands. */
6671 };
6672
6673 static struct intel_parser_s intel_parser;
6674
6675 /* Token structure for parsing intel syntax. */
6676 struct intel_token
6677 {
6678 int code; /* Token code. */
6679 const reg_entry *reg; /* Register entry for register tokens. */
6680 char *str; /* String representation. */
6681 };
6682
6683 static struct intel_token cur_token, prev_token;
6684
6685 /* Token codes for the intel parser. Since T_SHORT is already used
6686 by COFF, undefine it first to prevent a warning. */
6687 #define T_NIL -1
6688 #define T_CONST 1
6689 #define T_REG 2
6690 #define T_BYTE 3
6691 #define T_WORD 4
6692 #define T_DWORD 5
6693 #define T_FWORD 6
6694 #define T_QWORD 7
6695 #define T_TBYTE 8
6696 #define T_XMMWORD 9
6697 #undef T_SHORT
6698 #define T_SHORT 10
6699 #define T_OFFSET 11
6700 #define T_PTR 12
6701 #define T_ID 13
6702 #define T_SHL 14
6703 #define T_SHR 15
6704
6705 /* Prototypes for intel parser functions. */
6706 static int intel_match_token (int);
6707 static void intel_putback_token (void);
6708 static void intel_get_token (void);
6709 static int intel_expr (void);
6710 static int intel_e04 (void);
6711 static int intel_e05 (void);
6712 static int intel_e06 (void);
6713 static int intel_e09 (void);
6714 static int intel_e10 (void);
6715 static int intel_e11 (void);
6716
6717 static int
6718 i386_intel_operand (char *operand_string, int got_a_float)
6719 {
6720 int ret;
6721 char *p;
6722
6723 p = intel_parser.op_string = xstrdup (operand_string);
6724 intel_parser.disp = (char *) xmalloc (strlen (operand_string) + 1);
6725
6726 for (;;)
6727 {
6728 /* Initialize token holders. */
6729 cur_token.code = prev_token.code = T_NIL;
6730 cur_token.reg = prev_token.reg = NULL;
6731 cur_token.str = prev_token.str = NULL;
6732
6733 /* Initialize parser structure. */
6734 intel_parser.got_a_float = got_a_float;
6735 intel_parser.op_modifier = 0;
6736 intel_parser.is_mem = 0;
6737 intel_parser.in_offset = 0;
6738 intel_parser.in_bracket = 0;
6739 intel_parser.reg = NULL;
6740 intel_parser.disp[0] = '\0';
6741 intel_parser.next_operand = NULL;
6742
6743 /* Read the first token and start the parser. */
6744 intel_get_token ();
6745 ret = intel_expr ();
6746
6747 if (!ret)
6748 break;
6749
6750 if (cur_token.code != T_NIL)
6751 {
6752 as_bad (_("invalid operand for '%s' ('%s' unexpected)"),
6753 current_templates->start->name, cur_token.str);
6754 ret = 0;
6755 }
6756 /* If we found a memory reference, hand it over to i386_displacement
6757 to fill in the rest of the operand fields. */
6758 else if (intel_parser.is_mem)
6759 {
6760 if ((i.mem_operands == 1
6761 && (current_templates->start->opcode_modifier & IsString) == 0)
6762 || i.mem_operands == 2)
6763 {
6764 as_bad (_("too many memory references for '%s'"),
6765 current_templates->start->name);
6766 ret = 0;
6767 }
6768 else
6769 {
6770 char *s = intel_parser.disp;
6771 i.mem_operands++;
6772
6773 if (!quiet_warnings && intel_parser.is_mem < 0)
6774 /* See the comments in intel_bracket_expr. */
6775 as_warn (_("Treating `%s' as memory reference"), operand_string);
6776
6777 /* Add the displacement expression. */
6778 if (*s != '\0')
6779 ret = i386_displacement (s, s + strlen (s));
6780 if (ret)
6781 {
6782 /* Swap base and index in 16-bit memory operands like
6783 [si+bx]. Since i386_index_check is also used in AT&T
6784 mode we have to do that here. */
6785 if (i.base_reg
6786 && i.index_reg
6787 && (i.base_reg->reg_type & Reg16)
6788 && (i.index_reg->reg_type & Reg16)
6789 && i.base_reg->reg_num >= 6
6790 && i.index_reg->reg_num < 6)
6791 {
6792 const reg_entry *base = i.index_reg;
6793
6794 i.index_reg = i.base_reg;
6795 i.base_reg = base;
6796 }
6797 ret = i386_index_check (operand_string);
6798 }
6799 }
6800 }
6801
6802 /* Constant and OFFSET expressions are handled by i386_immediate. */
6803 else if ((intel_parser.op_modifier & (1 << T_OFFSET))
6804 || intel_parser.reg == NULL)
6805 ret = i386_immediate (intel_parser.disp);
6806
6807 if (intel_parser.next_operand && this_operand >= MAX_OPERANDS - 1)
6808 ret = 0;
6809 if (!ret || !intel_parser.next_operand)
6810 break;
6811 intel_parser.op_string = intel_parser.next_operand;
6812 this_operand = i.operands++;
6813 }
6814
6815 free (p);
6816 free (intel_parser.disp);
6817
6818 return ret;
6819 }
6820
6821 #define NUM_ADDRESS_REGS (!!i.base_reg + !!i.index_reg)
6822
6823 /* expr e04 expr'
6824
6825 expr' cmpOp e04 expr'
6826 | Empty */
6827 static int
6828 intel_expr (void)
6829 {
6830 /* XXX Implement the comparison operators. */
6831 return intel_e04 ();
6832 }
6833
6834 /* e04 e05 e04'
6835
6836 e04' addOp e05 e04'
6837 | Empty */
6838 static int
6839 intel_e04 (void)
6840 {
6841 int nregs = -1;
6842
6843 for (;;)
6844 {
6845 if (!intel_e05())
6846 return 0;
6847
6848 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
6849 i.base_reg = i386_regtab + REGNAM_AL; /* al is invalid as base */
6850
6851 if (cur_token.code == '+')
6852 nregs = -1;
6853 else if (cur_token.code == '-')
6854 nregs = NUM_ADDRESS_REGS;
6855 else
6856 return 1;
6857
6858 strcat (intel_parser.disp, cur_token.str);
6859 intel_match_token (cur_token.code);
6860 }
6861 }
6862
6863 /* e05 e06 e05'
6864
6865 e05' binOp e06 e05'
6866 | Empty */
6867 static int
6868 intel_e05 (void)
6869 {
6870 int nregs = ~NUM_ADDRESS_REGS;
6871
6872 for (;;)
6873 {
6874 if (!intel_e06())
6875 return 0;
6876
6877 if (cur_token.code == '&'
6878 || cur_token.code == '|'
6879 || cur_token.code == '^')
6880 {
6881 char str[2];
6882
6883 str[0] = cur_token.code;
6884 str[1] = 0;
6885 strcat (intel_parser.disp, str);
6886 }
6887 else
6888 break;
6889
6890 intel_match_token (cur_token.code);
6891
6892 if (nregs < 0)
6893 nregs = ~nregs;
6894 }
6895 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
6896 i.base_reg = i386_regtab + REGNAM_AL + 1; /* cl is invalid as base */
6897 return 1;
6898 }
6899
6900 /* e06 e09 e06'
6901
6902 e06' mulOp e09 e06'
6903 | Empty */
6904 static int
6905 intel_e06 (void)
6906 {
6907 int nregs = ~NUM_ADDRESS_REGS;
6908
6909 for (;;)
6910 {
6911 if (!intel_e09())
6912 return 0;
6913
6914 if (cur_token.code == '*'
6915 || cur_token.code == '/'
6916 || cur_token.code == '%')
6917 {
6918 char str[2];
6919
6920 str[0] = cur_token.code;
6921 str[1] = 0;
6922 strcat (intel_parser.disp, str);
6923 }
6924 else if (cur_token.code == T_SHL)
6925 strcat (intel_parser.disp, "<<");
6926 else if (cur_token.code == T_SHR)
6927 strcat (intel_parser.disp, ">>");
6928 else
6929 break;
6930
6931 intel_match_token (cur_token.code);
6932
6933 if (nregs < 0)
6934 nregs = ~nregs;
6935 }
6936 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
6937 i.base_reg = i386_regtab + REGNAM_AL + 2; /* dl is invalid as base */
6938 return 1;
6939 }
6940
6941 /* e09 OFFSET e09
6942 | SHORT e09
6943 | + e09
6944 | - e09
6945 | ~ e09
6946 | NOT e09
6947 | e10 e09'
6948
6949 e09' PTR e10 e09'
6950 | : e10 e09'
6951 | Empty */
6952 static int
6953 intel_e09 (void)
6954 {
6955 int nregs = ~NUM_ADDRESS_REGS;
6956 int in_offset = 0;
6957
6958 for (;;)
6959 {
6960 /* Don't consume constants here. */
6961 if (cur_token.code == '+' || cur_token.code == '-')
6962 {
6963 /* Need to look one token ahead - if the next token
6964 is a constant, the current token is its sign. */
6965 int next_code;
6966
6967 intel_match_token (cur_token.code);
6968 next_code = cur_token.code;
6969 intel_putback_token ();
6970 if (next_code == T_CONST)
6971 break;
6972 }
6973
6974 /* e09 OFFSET e09 */
6975 if (cur_token.code == T_OFFSET)
6976 {
6977 if (!in_offset++)
6978 ++intel_parser.in_offset;
6979 }
6980
6981 /* e09 SHORT e09 */
6982 else if (cur_token.code == T_SHORT)
6983 intel_parser.op_modifier |= 1 << T_SHORT;
6984
6985 /* e09 + e09 */
6986 else if (cur_token.code == '+')
6987 strcat (intel_parser.disp, "+");
6988
6989 /* e09 - e09
6990 | ~ e09
6991 | NOT e09 */
6992 else if (cur_token.code == '-' || cur_token.code == '~')
6993 {
6994 char str[2];
6995
6996 if (nregs < 0)
6997 nregs = ~nregs;
6998 str[0] = cur_token.code;
6999 str[1] = 0;
7000 strcat (intel_parser.disp, str);
7001 }
7002
7003 /* e09 e10 e09' */
7004 else
7005 break;
7006
7007 intel_match_token (cur_token.code);
7008 }
7009
7010 for (;;)
7011 {
7012 if (!intel_e10 ())
7013 return 0;
7014
7015 /* e09' PTR e10 e09' */
7016 if (cur_token.code == T_PTR)
7017 {
7018 char suffix;
7019
7020 if (prev_token.code == T_BYTE)
7021 suffix = BYTE_MNEM_SUFFIX;
7022
7023 else if (prev_token.code == T_WORD)
7024 {
7025 if (current_templates->start->name[0] == 'l'
7026 && current_templates->start->name[2] == 's'
7027 && current_templates->start->name[3] == 0)
7028 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
7029 else if (intel_parser.got_a_float == 2) /* "fi..." */
7030 suffix = SHORT_MNEM_SUFFIX;
7031 else
7032 suffix = WORD_MNEM_SUFFIX;
7033 }
7034
7035 else if (prev_token.code == T_DWORD)
7036 {
7037 if (current_templates->start->name[0] == 'l'
7038 && current_templates->start->name[2] == 's'
7039 && current_templates->start->name[3] == 0)
7040 suffix = WORD_MNEM_SUFFIX;
7041 else if (flag_code == CODE_16BIT
7042 && (current_templates->start->opcode_modifier
7043 & (Jump | JumpDword)))
7044 suffix = LONG_DOUBLE_MNEM_SUFFIX;
7045 else if (intel_parser.got_a_float == 1) /* "f..." */
7046 suffix = SHORT_MNEM_SUFFIX;
7047 else
7048 suffix = LONG_MNEM_SUFFIX;
7049 }
7050
7051 else if (prev_token.code == T_FWORD)
7052 {
7053 if (current_templates->start->name[0] == 'l'
7054 && current_templates->start->name[2] == 's'
7055 && current_templates->start->name[3] == 0)
7056 suffix = LONG_MNEM_SUFFIX;
7057 else if (!intel_parser.got_a_float)
7058 {
7059 if (flag_code == CODE_16BIT)
7060 add_prefix (DATA_PREFIX_OPCODE);
7061 suffix = LONG_DOUBLE_MNEM_SUFFIX;
7062 }
7063 else
7064 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
7065 }
7066
7067 else if (prev_token.code == T_QWORD)
7068 {
7069 if (intel_parser.got_a_float == 1) /* "f..." */
7070 suffix = LONG_MNEM_SUFFIX;
7071 else
7072 suffix = QWORD_MNEM_SUFFIX;
7073 }
7074
7075 else if (prev_token.code == T_TBYTE)
7076 {
7077 if (intel_parser.got_a_float == 1)
7078 suffix = LONG_DOUBLE_MNEM_SUFFIX;
7079 else
7080 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
7081 }
7082
7083 else if (prev_token.code == T_XMMWORD)
7084 {
7085 /* XXX ignored for now, but accepted since gcc uses it */
7086 suffix = 0;
7087 }
7088
7089 else
7090 {
7091 as_bad (_("Unknown operand modifier `%s'"), prev_token.str);
7092 return 0;
7093 }
7094
7095 /* Operands for jump/call using 'ptr' notation denote absolute
7096 addresses. */
7097 if (current_templates->start->opcode_modifier & (Jump | JumpDword))
7098 i.types[this_operand] |= JumpAbsolute;
7099
7100 if (current_templates->start->base_opcode == 0x8d /* lea */)
7101 ;
7102 else if (!i.suffix)
7103 i.suffix = suffix;
7104 else if (i.suffix != suffix)
7105 {
7106 as_bad (_("Conflicting operand modifiers"));
7107 return 0;
7108 }
7109
7110 }
7111
7112 /* e09' : e10 e09' */
7113 else if (cur_token.code == ':')
7114 {
7115 if (prev_token.code != T_REG)
7116 {
7117 /* While {call,jmp} SSSS:OOOO is MASM syntax only when SSSS is a
7118 segment/group identifier (which we don't have), using comma
7119 as the operand separator there is even less consistent, since
7120 there all branches only have a single operand. */
7121 if (this_operand != 0
7122 || intel_parser.in_offset
7123 || intel_parser.in_bracket
7124 || (!(current_templates->start->opcode_modifier
7125 & (Jump|JumpDword|JumpInterSegment))
7126 && !(current_templates->start->operand_types[0]
7127 & JumpAbsolute)))
7128 return intel_match_token (T_NIL);
7129 /* Remember the start of the 2nd operand and terminate 1st
7130 operand here.
7131 XXX This isn't right, yet (when SSSS:OOOO is right operand of
7132 another expression), but it gets at least the simplest case
7133 (a plain number or symbol on the left side) right. */
7134 intel_parser.next_operand = intel_parser.op_string;
7135 *--intel_parser.op_string = '\0';
7136 return intel_match_token (':');
7137 }
7138 }
7139
7140 /* e09' Empty */
7141 else
7142 break;
7143
7144 intel_match_token (cur_token.code);
7145
7146 }
7147
7148 if (in_offset)
7149 {
7150 --intel_parser.in_offset;
7151 if (nregs < 0)
7152 nregs = ~nregs;
7153 if (NUM_ADDRESS_REGS > nregs)
7154 {
7155 as_bad (_("Invalid operand to `OFFSET'"));
7156 return 0;
7157 }
7158 intel_parser.op_modifier |= 1 << T_OFFSET;
7159 }
7160
7161 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
7162 i.base_reg = i386_regtab + REGNAM_AL + 3; /* bl is invalid as base */
7163 return 1;
7164 }
7165
7166 static int
7167 intel_bracket_expr (void)
7168 {
7169 int was_offset = intel_parser.op_modifier & (1 << T_OFFSET);
7170 const char *start = intel_parser.op_string;
7171 int len;
7172
7173 if (i.op[this_operand].regs)
7174 return intel_match_token (T_NIL);
7175
7176 intel_match_token ('[');
7177
7178 /* Mark as a memory operand only if it's not already known to be an
7179 offset expression. If it's an offset expression, we need to keep
7180 the brace in. */
7181 if (!intel_parser.in_offset)
7182 {
7183 ++intel_parser.in_bracket;
7184
7185 /* Operands for jump/call inside brackets denote absolute addresses. */
7186 if (current_templates->start->opcode_modifier & (Jump | JumpDword))
7187 i.types[this_operand] |= JumpAbsolute;
7188
7189 /* Unfortunately gas always diverged from MASM in a respect that can't
7190 be easily fixed without risking to break code sequences likely to be
7191 encountered (the testsuite even check for this): MASM doesn't consider
7192 an expression inside brackets unconditionally as a memory reference.
7193 When that is e.g. a constant, an offset expression, or the sum of the
7194 two, this is still taken as a constant load. gas, however, always
7195 treated these as memory references. As a compromise, we'll try to make
7196 offset expressions inside brackets work the MASM way (since that's
7197 less likely to be found in real world code), but make constants alone
7198 continue to work the traditional gas way. In either case, issue a
7199 warning. */
7200 intel_parser.op_modifier &= ~was_offset;
7201 }
7202 else
7203 strcat (intel_parser.disp, "[");
7204
7205 /* Add a '+' to the displacement string if necessary. */
7206 if (*intel_parser.disp != '\0'
7207 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
7208 strcat (intel_parser.disp, "+");
7209
7210 if (intel_expr ()
7211 && (len = intel_parser.op_string - start - 1,
7212 intel_match_token (']')))
7213 {
7214 /* Preserve brackets when the operand is an offset expression. */
7215 if (intel_parser.in_offset)
7216 strcat (intel_parser.disp, "]");
7217 else
7218 {
7219 --intel_parser.in_bracket;
7220 if (i.base_reg || i.index_reg)
7221 intel_parser.is_mem = 1;
7222 if (!intel_parser.is_mem)
7223 {
7224 if (!(intel_parser.op_modifier & (1 << T_OFFSET)))
7225 /* Defer the warning until all of the operand was parsed. */
7226 intel_parser.is_mem = -1;
7227 else if (!quiet_warnings)
7228 as_warn (_("`[%.*s]' taken to mean just `%.*s'"),
7229 len, start, len, start);
7230 }
7231 }
7232 intel_parser.op_modifier |= was_offset;
7233
7234 return 1;
7235 }
7236 return 0;
7237 }
7238
7239 /* e10 e11 e10'
7240
7241 e10' [ expr ] e10'
7242 | Empty */
7243 static int
7244 intel_e10 (void)
7245 {
7246 if (!intel_e11 ())
7247 return 0;
7248
7249 while (cur_token.code == '[')
7250 {
7251 if (!intel_bracket_expr ())
7252 return 0;
7253 }
7254
7255 return 1;
7256 }
7257
7258 /* e11 ( expr )
7259 | [ expr ]
7260 | BYTE
7261 | WORD
7262 | DWORD
7263 | FWORD
7264 | QWORD
7265 | TBYTE
7266 | OWORD
7267 | XMMWORD
7268 | $
7269 | .
7270 | register
7271 | id
7272 | constant */
7273 static int
7274 intel_e11 (void)
7275 {
7276 switch (cur_token.code)
7277 {
7278 /* e11 ( expr ) */
7279 case '(':
7280 intel_match_token ('(');
7281 strcat (intel_parser.disp, "(");
7282
7283 if (intel_expr () && intel_match_token (')'))
7284 {
7285 strcat (intel_parser.disp, ")");
7286 return 1;
7287 }
7288 return 0;
7289
7290 /* e11 [ expr ] */
7291 case '[':
7292 return intel_bracket_expr ();
7293
7294 /* e11 $
7295 | . */
7296 case '.':
7297 strcat (intel_parser.disp, cur_token.str);
7298 intel_match_token (cur_token.code);
7299
7300 /* Mark as a memory operand only if it's not already known to be an
7301 offset expression. */
7302 if (!intel_parser.in_offset)
7303 intel_parser.is_mem = 1;
7304
7305 return 1;
7306
7307 /* e11 register */
7308 case T_REG:
7309 {
7310 const reg_entry *reg = intel_parser.reg = cur_token.reg;
7311
7312 intel_match_token (T_REG);
7313
7314 /* Check for segment change. */
7315 if (cur_token.code == ':')
7316 {
7317 if (!(reg->reg_type & (SReg2 | SReg3)))
7318 {
7319 as_bad (_("`%s' is not a valid segment register"),
7320 reg->reg_name);
7321 return 0;
7322 }
7323 else if (i.seg[i.mem_operands])
7324 as_warn (_("Extra segment override ignored"));
7325 else
7326 {
7327 if (!intel_parser.in_offset)
7328 intel_parser.is_mem = 1;
7329 switch (reg->reg_num)
7330 {
7331 case 0:
7332 i.seg[i.mem_operands] = &es;
7333 break;
7334 case 1:
7335 i.seg[i.mem_operands] = &cs;
7336 break;
7337 case 2:
7338 i.seg[i.mem_operands] = &ss;
7339 break;
7340 case 3:
7341 i.seg[i.mem_operands] = &ds;
7342 break;
7343 case 4:
7344 i.seg[i.mem_operands] = &fs;
7345 break;
7346 case 5:
7347 i.seg[i.mem_operands] = &gs;
7348 break;
7349 }
7350 }
7351 }
7352
7353 /* Not a segment register. Check for register scaling. */
7354 else if (cur_token.code == '*')
7355 {
7356 if (!intel_parser.in_bracket)
7357 {
7358 as_bad (_("Register scaling only allowed in memory operands"));
7359 return 0;
7360 }
7361
7362 if (reg->reg_type & Reg16) /* Disallow things like [si*1]. */
7363 reg = i386_regtab + REGNAM_AX + 4; /* sp is invalid as index */
7364 else if (i.index_reg)
7365 reg = i386_regtab + REGNAM_EAX + 4; /* esp is invalid as index */
7366
7367 /* What follows must be a valid scale. */
7368 intel_match_token ('*');
7369 i.index_reg = reg;
7370 i.types[this_operand] |= BaseIndex;
7371
7372 /* Set the scale after setting the register (otherwise,
7373 i386_scale will complain) */
7374 if (cur_token.code == '+' || cur_token.code == '-')
7375 {
7376 char *str, sign = cur_token.code;
7377 intel_match_token (cur_token.code);
7378 if (cur_token.code != T_CONST)
7379 {
7380 as_bad (_("Syntax error: Expecting a constant, got `%s'"),
7381 cur_token.str);
7382 return 0;
7383 }
7384 str = (char *) xmalloc (strlen (cur_token.str) + 2);
7385 strcpy (str + 1, cur_token.str);
7386 *str = sign;
7387 if (!i386_scale (str))
7388 return 0;
7389 free (str);
7390 }
7391 else if (!i386_scale (cur_token.str))
7392 return 0;
7393 intel_match_token (cur_token.code);
7394 }
7395
7396 /* No scaling. If this is a memory operand, the register is either a
7397 base register (first occurrence) or an index register (second
7398 occurrence). */
7399 else if (intel_parser.in_bracket)
7400 {
7401
7402 if (!i.base_reg)
7403 i.base_reg = reg;
7404 else if (!i.index_reg)
7405 i.index_reg = reg;
7406 else
7407 {
7408 as_bad (_("Too many register references in memory operand"));
7409 return 0;
7410 }
7411
7412 i.types[this_operand] |= BaseIndex;
7413 }
7414
7415 /* It's neither base nor index. */
7416 else if (!intel_parser.in_offset && !intel_parser.is_mem)
7417 {
7418 i.types[this_operand] |= reg->reg_type & ~BaseIndex;
7419 i.op[this_operand].regs = reg;
7420 i.reg_operands++;
7421 }
7422 else
7423 {
7424 as_bad (_("Invalid use of register"));
7425 return 0;
7426 }
7427
7428 /* Since registers are not part of the displacement string (except
7429 when we're parsing offset operands), we may need to remove any
7430 preceding '+' from the displacement string. */
7431 if (*intel_parser.disp != '\0'
7432 && !intel_parser.in_offset)
7433 {
7434 char *s = intel_parser.disp;
7435 s += strlen (s) - 1;
7436 if (*s == '+')
7437 *s = '\0';
7438 }
7439
7440 return 1;
7441 }
7442
7443 /* e11 BYTE
7444 | WORD
7445 | DWORD
7446 | FWORD
7447 | QWORD
7448 | TBYTE
7449 | OWORD
7450 | XMMWORD */
7451 case T_BYTE:
7452 case T_WORD:
7453 case T_DWORD:
7454 case T_FWORD:
7455 case T_QWORD:
7456 case T_TBYTE:
7457 case T_XMMWORD:
7458 intel_match_token (cur_token.code);
7459
7460 if (cur_token.code == T_PTR)
7461 return 1;
7462
7463 /* It must have been an identifier. */
7464 intel_putback_token ();
7465 cur_token.code = T_ID;
7466 /* FALLTHRU */
7467
7468 /* e11 id
7469 | constant */
7470 case T_ID:
7471 if (!intel_parser.in_offset && intel_parser.is_mem <= 0)
7472 {
7473 symbolS *symbolP;
7474
7475 /* The identifier represents a memory reference only if it's not
7476 preceded by an offset modifier and if it's not an equate. */
7477 symbolP = symbol_find(cur_token.str);
7478 if (!symbolP || S_GET_SEGMENT(symbolP) != absolute_section)
7479 intel_parser.is_mem = 1;
7480 }
7481 /* FALLTHRU */
7482
7483 case T_CONST:
7484 case '-':
7485 case '+':
7486 {
7487 char *save_str, sign = 0;
7488
7489 /* Allow constants that start with `+' or `-'. */
7490 if (cur_token.code == '-' || cur_token.code == '+')
7491 {
7492 sign = cur_token.code;
7493 intel_match_token (cur_token.code);
7494 if (cur_token.code != T_CONST)
7495 {
7496 as_bad (_("Syntax error: Expecting a constant, got `%s'"),
7497 cur_token.str);
7498 return 0;
7499 }
7500 }
7501
7502 save_str = (char *) xmalloc (strlen (cur_token.str) + 2);
7503 strcpy (save_str + !!sign, cur_token.str);
7504 if (sign)
7505 *save_str = sign;
7506
7507 /* Get the next token to check for register scaling. */
7508 intel_match_token (cur_token.code);
7509
7510 /* Check if this constant is a scaling factor for an
7511 index register. */
7512 if (cur_token.code == '*')
7513 {
7514 if (intel_match_token ('*') && cur_token.code == T_REG)
7515 {
7516 const reg_entry *reg = cur_token.reg;
7517
7518 if (!intel_parser.in_bracket)
7519 {
7520 as_bad (_("Register scaling only allowed "
7521 "in memory operands"));
7522 return 0;
7523 }
7524
7525 /* Disallow things like [1*si].
7526 sp and esp are invalid as index. */
7527 if (reg->reg_type & Reg16)
7528 reg = i386_regtab + REGNAM_AX + 4;
7529 else if (i.index_reg)
7530 reg = i386_regtab + REGNAM_EAX + 4;
7531
7532 /* The constant is followed by `* reg', so it must be
7533 a valid scale. */
7534 i.index_reg = reg;
7535 i.types[this_operand] |= BaseIndex;
7536
7537 /* Set the scale after setting the register (otherwise,
7538 i386_scale will complain) */
7539 if (!i386_scale (save_str))
7540 return 0;
7541 intel_match_token (T_REG);
7542
7543 /* Since registers are not part of the displacement
7544 string, we may need to remove any preceding '+' from
7545 the displacement string. */
7546 if (*intel_parser.disp != '\0')
7547 {
7548 char *s = intel_parser.disp;
7549 s += strlen (s) - 1;
7550 if (*s == '+')
7551 *s = '\0';
7552 }
7553
7554 free (save_str);
7555
7556 return 1;
7557 }
7558
7559 /* The constant was not used for register scaling. Since we have
7560 already consumed the token following `*' we now need to put it
7561 back in the stream. */
7562 intel_putback_token ();
7563 }
7564
7565 /* Add the constant to the displacement string. */
7566 strcat (intel_parser.disp, save_str);
7567 free (save_str);
7568
7569 return 1;
7570 }
7571 }
7572
7573 as_bad (_("Unrecognized token '%s'"), cur_token.str);
7574 return 0;
7575 }
7576
7577 /* Match the given token against cur_token. If they match, read the next
7578 token from the operand string. */
7579 static int
7580 intel_match_token (int code)
7581 {
7582 if (cur_token.code == code)
7583 {
7584 intel_get_token ();
7585 return 1;
7586 }
7587 else
7588 {
7589 as_bad (_("Unexpected token `%s'"), cur_token.str);
7590 return 0;
7591 }
7592 }
7593
7594 /* Read a new token from intel_parser.op_string and store it in cur_token. */
7595 static void
7596 intel_get_token (void)
7597 {
7598 char *end_op;
7599 const reg_entry *reg;
7600 struct intel_token new_token;
7601
7602 new_token.code = T_NIL;
7603 new_token.reg = NULL;
7604 new_token.str = NULL;
7605
7606 /* Free the memory allocated to the previous token and move
7607 cur_token to prev_token. */
7608 if (prev_token.str)
7609 free (prev_token.str);
7610
7611 prev_token = cur_token;
7612
7613 /* Skip whitespace. */
7614 while (is_space_char (*intel_parser.op_string))
7615 intel_parser.op_string++;
7616
7617 /* Return an empty token if we find nothing else on the line. */
7618 if (*intel_parser.op_string == '\0')
7619 {
7620 cur_token = new_token;
7621 return;
7622 }
7623
7624 /* The new token cannot be larger than the remainder of the operand
7625 string. */
7626 new_token.str = (char *) xmalloc (strlen (intel_parser.op_string) + 1);
7627 new_token.str[0] = '\0';
7628
7629 if (strchr ("0123456789", *intel_parser.op_string))
7630 {
7631 char *p = new_token.str;
7632 char *q = intel_parser.op_string;
7633 new_token.code = T_CONST;
7634
7635 /* Allow any kind of identifier char to encompass floating point and
7636 hexadecimal numbers. */
7637 while (is_identifier_char (*q))
7638 *p++ = *q++;
7639 *p = '\0';
7640
7641 /* Recognize special symbol names [0-9][bf]. */
7642 if (strlen (intel_parser.op_string) == 2
7643 && (intel_parser.op_string[1] == 'b'
7644 || intel_parser.op_string[1] == 'f'))
7645 new_token.code = T_ID;
7646 }
7647
7648 else if ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL)
7649 {
7650 size_t len = end_op - intel_parser.op_string;
7651
7652 new_token.code = T_REG;
7653 new_token.reg = reg;
7654
7655 memcpy (new_token.str, intel_parser.op_string, len);
7656 new_token.str[len] = '\0';
7657 }
7658
7659 else if (is_identifier_char (*intel_parser.op_string))
7660 {
7661 char *p = new_token.str;
7662 char *q = intel_parser.op_string;
7663
7664 /* A '.' or '$' followed by an identifier char is an identifier.
7665 Otherwise, it's operator '.' followed by an expression. */
7666 if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
7667 {
7668 new_token.code = '.';
7669 new_token.str[0] = '.';
7670 new_token.str[1] = '\0';
7671 }
7672 else
7673 {
7674 while (is_identifier_char (*q) || *q == '@')
7675 *p++ = *q++;
7676 *p = '\0';
7677
7678 if (strcasecmp (new_token.str, "NOT") == 0)
7679 new_token.code = '~';
7680
7681 else if (strcasecmp (new_token.str, "MOD") == 0)
7682 new_token.code = '%';
7683
7684 else if (strcasecmp (new_token.str, "AND") == 0)
7685 new_token.code = '&';
7686
7687 else if (strcasecmp (new_token.str, "OR") == 0)
7688 new_token.code = '|';
7689
7690 else if (strcasecmp (new_token.str, "XOR") == 0)
7691 new_token.code = '^';
7692
7693 else if (strcasecmp (new_token.str, "SHL") == 0)
7694 new_token.code = T_SHL;
7695
7696 else if (strcasecmp (new_token.str, "SHR") == 0)
7697 new_token.code = T_SHR;
7698
7699 else if (strcasecmp (new_token.str, "BYTE") == 0)
7700 new_token.code = T_BYTE;
7701
7702 else if (strcasecmp (new_token.str, "WORD") == 0)
7703 new_token.code = T_WORD;
7704
7705 else if (strcasecmp (new_token.str, "DWORD") == 0)
7706 new_token.code = T_DWORD;
7707
7708 else if (strcasecmp (new_token.str, "FWORD") == 0)
7709 new_token.code = T_FWORD;
7710
7711 else if (strcasecmp (new_token.str, "QWORD") == 0)
7712 new_token.code = T_QWORD;
7713
7714 else if (strcasecmp (new_token.str, "TBYTE") == 0
7715 /* XXX remove (gcc still uses it) */
7716 || strcasecmp (new_token.str, "XWORD") == 0)
7717 new_token.code = T_TBYTE;
7718
7719 else if (strcasecmp (new_token.str, "XMMWORD") == 0
7720 || strcasecmp (new_token.str, "OWORD") == 0)
7721 new_token.code = T_XMMWORD;
7722
7723 else if (strcasecmp (new_token.str, "PTR") == 0)
7724 new_token.code = T_PTR;
7725
7726 else if (strcasecmp (new_token.str, "SHORT") == 0)
7727 new_token.code = T_SHORT;
7728
7729 else if (strcasecmp (new_token.str, "OFFSET") == 0)
7730 {
7731 new_token.code = T_OFFSET;
7732
7733 /* ??? This is not mentioned in the MASM grammar but gcc
7734 makes use of it with -mintel-syntax. OFFSET may be
7735 followed by FLAT: */
7736 if (strncasecmp (q, " FLAT:", 6) == 0)
7737 strcat (new_token.str, " FLAT:");
7738 }
7739
7740 /* ??? This is not mentioned in the MASM grammar. */
7741 else if (strcasecmp (new_token.str, "FLAT") == 0)
7742 {
7743 new_token.code = T_OFFSET;
7744 if (*q == ':')
7745 strcat (new_token.str, ":");
7746 else
7747 as_bad (_("`:' expected"));
7748 }
7749
7750 else
7751 new_token.code = T_ID;
7752 }
7753 }
7754
7755 else if (strchr ("+-/*%|&^:[]()~", *intel_parser.op_string))
7756 {
7757 new_token.code = *intel_parser.op_string;
7758 new_token.str[0] = *intel_parser.op_string;
7759 new_token.str[1] = '\0';
7760 }
7761
7762 else if (strchr ("<>", *intel_parser.op_string)
7763 && *intel_parser.op_string == *(intel_parser.op_string + 1))
7764 {
7765 new_token.code = *intel_parser.op_string == '<' ? T_SHL : T_SHR;
7766 new_token.str[0] = *intel_parser.op_string;
7767 new_token.str[1] = *intel_parser.op_string;
7768 new_token.str[2] = '\0';
7769 }
7770
7771 else
7772 as_bad (_("Unrecognized token `%s'"), intel_parser.op_string);
7773
7774 intel_parser.op_string += strlen (new_token.str);
7775 cur_token = new_token;
7776 }
7777
7778 /* Put cur_token back into the token stream and make cur_token point to
7779 prev_token. */
7780 static void
7781 intel_putback_token (void)
7782 {
7783 if (cur_token.code != T_NIL)
7784 {
7785 intel_parser.op_string -= strlen (cur_token.str);
7786 free (cur_token.str);
7787 }
7788 cur_token = prev_token;
7789
7790 /* Forget prev_token. */
7791 prev_token.code = T_NIL;
7792 prev_token.reg = NULL;
7793 prev_token.str = NULL;
7794 }
7795
7796 int
7797 tc_x86_regname_to_dw2regnum (char *regname)
7798 {
7799 unsigned int regnum;
7800 unsigned int regnames_count;
7801 static const char *const regnames_32[] =
7802 {
7803 "eax", "ecx", "edx", "ebx",
7804 "esp", "ebp", "esi", "edi",
7805 "eip", "eflags", NULL,
7806 "st0", "st1", "st2", "st3",
7807 "st4", "st5", "st6", "st7",
7808 NULL, NULL,
7809 "xmm0", "xmm1", "xmm2", "xmm3",
7810 "xmm4", "xmm5", "xmm6", "xmm7",
7811 "mm0", "mm1", "mm2", "mm3",
7812 "mm4", "mm5", "mm6", "mm7",
7813 "fcw", "fsw", "mxcsr",
7814 "es", "cs", "ss", "ds", "fs", "gs", NULL, NULL,
7815 "tr", "ldtr"
7816 };
7817 static const char *const regnames_64[] =
7818 {
7819 "rax", "rdx", "rcx", "rbx",
7820 "rsi", "rdi", "rbp", "rsp",
7821 "r8", "r9", "r10", "r11",
7822 "r12", "r13", "r14", "r15",
7823 "rip",
7824 "xmm0", "xmm1", "xmm2", "xmm3",
7825 "xmm4", "xmm5", "xmm6", "xmm7",
7826 "xmm8", "xmm9", "xmm10", "xmm11",
7827 "xmm12", "xmm13", "xmm14", "xmm15",
7828 "st0", "st1", "st2", "st3",
7829 "st4", "st5", "st6", "st7",
7830 "mm0", "mm1", "mm2", "mm3",
7831 "mm4", "mm5", "mm6", "mm7",
7832 "rflags",
7833 "es", "cs", "ss", "ds", "fs", "gs", NULL, NULL,
7834 "fs.base", "gs.base", NULL, NULL,
7835 "tr", "ldtr",
7836 "mxcsr", "fcw", "fsw"
7837 };
7838 const char *const *regnames;
7839
7840 if (flag_code == CODE_64BIT)
7841 {
7842 regnames = regnames_64;
7843 regnames_count = ARRAY_SIZE (regnames_64);
7844 }
7845 else
7846 {
7847 regnames = regnames_32;
7848 regnames_count = ARRAY_SIZE (regnames_32);
7849 }
7850
7851 for (regnum = 0; regnum < regnames_count; regnum++)
7852 if (regnames[regnum] != NULL
7853 && strcmp (regname, regnames[regnum]) == 0)
7854 return regnum;
7855
7856 return -1;
7857 }
7858
7859 void
7860 tc_x86_frame_initial_instructions (void)
7861 {
7862 static unsigned int sp_regno;
7863
7864 if (!sp_regno)
7865 sp_regno = tc_x86_regname_to_dw2regnum (flag_code == CODE_64BIT
7866 ? "rsp" : "esp");
7867
7868 cfi_add_CFA_def_cfa (sp_regno, -x86_cie_data_alignment);
7869 cfi_add_CFA_offset (x86_dwarf2_return_column, x86_cie_data_alignment);
7870 }
7871
7872 int
7873 i386_elf_section_type (const char *str, size_t len)
7874 {
7875 if (flag_code == CODE_64BIT
7876 && len == sizeof ("unwind") - 1
7877 && strncmp (str, "unwind", 6) == 0)
7878 return SHT_X86_64_UNWIND;
7879
7880 return -1;
7881 }
7882
7883 #ifdef TE_PE
7884 void
7885 tc_pe_dwarf2_emit_offset (symbolS *symbol, unsigned int size)
7886 {
7887 expressionS expr;
7888
7889 expr.X_op = O_secrel;
7890 expr.X_add_symbol = symbol;
7891 expr.X_add_number = 0;
7892 emit_expr (&expr, size);
7893 }
7894 #endif
7895
7896 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
7897 /* For ELF on x86-64, add support for SHF_X86_64_LARGE. */
7898
7899 int
7900 x86_64_section_letter (int letter, char **ptr_msg)
7901 {
7902 if (flag_code == CODE_64BIT)
7903 {
7904 if (letter == 'l')
7905 return SHF_X86_64_LARGE;
7906
7907 *ptr_msg = _("Bad .section directive: want a,l,w,x,M,S,G,T in string");
7908 }
7909 else
7910 *ptr_msg = _("Bad .section directive: want a,w,x,M,S,G,T in string");
7911 return -1;
7912 }
7913
7914 int
7915 x86_64_section_word (char *str, size_t len)
7916 {
7917 if (len == 5 && flag_code == CODE_64BIT && CONST_STRNEQ (str, "large"))
7918 return SHF_X86_64_LARGE;
7919
7920 return -1;
7921 }
7922
7923 static void
7924 handle_large_common (int small ATTRIBUTE_UNUSED)
7925 {
7926 if (flag_code != CODE_64BIT)
7927 {
7928 s_comm_internal (0, elf_common_parse);
7929 as_warn (_(".largecomm supported only in 64bit mode, producing .comm"));
7930 }
7931 else
7932 {
7933 static segT lbss_section;
7934 asection *saved_com_section_ptr = elf_com_section_ptr;
7935 asection *saved_bss_section = bss_section;
7936
7937 if (lbss_section == NULL)
7938 {
7939 flagword applicable;
7940 segT seg = now_seg;
7941 subsegT subseg = now_subseg;
7942
7943 /* The .lbss section is for local .largecomm symbols. */
7944 lbss_section = subseg_new (".lbss", 0);
7945 applicable = bfd_applicable_section_flags (stdoutput);
7946 bfd_set_section_flags (stdoutput, lbss_section,
7947 applicable & SEC_ALLOC);
7948 seg_info (lbss_section)->bss = 1;
7949
7950 subseg_set (seg, subseg);
7951 }
7952
7953 elf_com_section_ptr = &_bfd_elf_large_com_section;
7954 bss_section = lbss_section;
7955
7956 s_comm_internal (0, elf_common_parse);
7957
7958 elf_com_section_ptr = saved_com_section_ptr;
7959 bss_section = saved_bss_section;
7960 }
7961 }
7962 #endif /* OBJ_ELF || OBJ_MAYBE_ELF */
This page took 0.193698 seconds and 4 git commands to generate.