gas/
[binutils.git] / gas / config / tc-i386.c
blobaa88394786bb4881032e97964ec71e7357bf402a
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.
6 This file is part of GAS, the GNU Assembler.
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.
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.
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. */
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. */
30 #include "as.h"
31 #include "safe-ctype.h"
32 #include "subsegs.h"
33 #include "dwarf2dbg.h"
34 #include "dw2gencfi.h"
35 #include "opcode/i386.h"
36 #include "elf/x86-64.h"
38 #ifndef REGISTER_WARNINGS
39 #define REGISTER_WARNINGS 1
40 #endif
42 #ifndef INFER_ADDR_PREFIX
43 #define INFER_ADDR_PREFIX 1
44 #endif
46 #ifndef SCALE1_WHEN_NO_INDEX
47 /* Specifying a scale factor besides 1 when there is no index is
48 futile. eg. `mov (%ebx,2),%al' does exactly the same as
49 `mov (%ebx),%al'. To slavishly follow what the programmer
50 specified, set SCALE1_WHEN_NO_INDEX to 0. */
51 #define SCALE1_WHEN_NO_INDEX 1
52 #endif
54 #ifndef DEFAULT_ARCH
55 #define DEFAULT_ARCH "i386"
56 #endif
58 #ifndef INLINE
59 #if __GNUC__ >= 2
60 #define INLINE __inline__
61 #else
62 #define INLINE
63 #endif
64 #endif
66 static void set_code_flag (int);
67 static void set_16bit_gcc_code_flag (int);
68 static void set_intel_syntax (int);
69 static void set_cpu_arch (int);
70 #ifdef TE_PE
71 static void pe_directive_secrel (int);
72 #endif
73 static void signed_cons (int);
74 static char *output_invalid (int c);
75 static int i386_operand (char *);
76 static int i386_intel_operand (char *, int);
77 static const reg_entry *parse_register (char *, char **);
78 static char *parse_insn (char *, char *);
79 static char *parse_operands (char *, const char *);
80 static void swap_operands (void);
81 static void swap_2_operands (int, int);
82 static void optimize_imm (void);
83 static void optimize_disp (void);
84 static int match_template (void);
85 static int check_string (void);
86 static int process_suffix (void);
87 static int check_byte_reg (void);
88 static int check_long_reg (void);
89 static int check_qword_reg (void);
90 static int check_word_reg (void);
91 static int finalize_imm (void);
92 static int process_operands (void);
93 static const seg_entry *build_modrm_byte (void);
94 static void output_insn (void);
95 static void output_imm (fragS *, offsetT);
96 static void output_disp (fragS *, offsetT);
97 #ifndef I386COFF
98 static void s_bss (int);
99 #endif
100 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
101 static void handle_large_common (int small ATTRIBUTE_UNUSED);
102 #endif
104 static const char *default_arch = DEFAULT_ARCH;
106 /* 'md_assemble ()' gathers together information and puts it into a
107 i386_insn. */
109 union i386_op
111 expressionS *disps;
112 expressionS *imms;
113 const reg_entry *regs;
116 struct _i386_insn
118 /* TM holds the template for the insn were currently assembling. */
119 template tm;
121 /* SUFFIX holds the instruction mnemonic suffix if given.
122 (e.g. 'l' for 'movl') */
123 char suffix;
125 /* OPERANDS gives the number of given operands. */
126 unsigned int operands;
128 /* REG_OPERANDS, DISP_OPERANDS, MEM_OPERANDS, IMM_OPERANDS give the number
129 of given register, displacement, memory operands and immediate
130 operands. */
131 unsigned int reg_operands, disp_operands, mem_operands, imm_operands;
133 /* TYPES [i] is the type (see above #defines) which tells us how to
134 use OP[i] for the corresponding operand. */
135 unsigned int types[MAX_OPERANDS];
137 /* Displacement expression, immediate expression, or register for each
138 operand. */
139 union i386_op op[MAX_OPERANDS];
141 /* Flags for operands. */
142 unsigned int flags[MAX_OPERANDS];
143 #define Operand_PCrel 1
145 /* Relocation type for operand */
146 enum bfd_reloc_code_real reloc[MAX_OPERANDS];
148 /* BASE_REG, INDEX_REG, and LOG2_SCALE_FACTOR are used to encode
149 the base index byte below. */
150 const reg_entry *base_reg;
151 const reg_entry *index_reg;
152 unsigned int log2_scale_factor;
154 /* SEG gives the seg_entries of this insn. They are zero unless
155 explicit segment overrides are given. */
156 const seg_entry *seg[2];
158 /* PREFIX holds all the given prefix opcodes (usually null).
159 PREFIXES is the number of prefix opcodes. */
160 unsigned int prefixes;
161 unsigned char prefix[MAX_PREFIXES];
163 /* RM and SIB are the modrm byte and the sib byte where the
164 addressing modes of this insn are encoded. */
166 modrm_byte rm;
167 rex_byte rex;
168 sib_byte sib;
171 typedef struct _i386_insn i386_insn;
173 /* List of chars besides those in app.c:symbol_chars that can start an
174 operand. Used to prevent the scrubber eating vital white-space. */
175 const char extra_symbol_chars[] = "*%-(["
176 #ifdef LEX_AT
178 #endif
179 #ifdef LEX_QM
181 #endif
184 #if (defined (TE_I386AIX) \
185 || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) \
186 && !defined (TE_GNU) \
187 && !defined (TE_LINUX) \
188 && !defined (TE_NETWARE) \
189 && !defined (TE_FreeBSD) \
190 && !defined (TE_NetBSD)))
191 /* This array holds the chars that always start a comment. If the
192 pre-processor is disabled, these aren't very useful. The option
193 --divide will remove '/' from this list. */
194 const char *i386_comment_chars = "#/";
195 #define SVR4_COMMENT_CHARS 1
196 #define PREFIX_SEPARATOR '\\'
198 #else
199 const char *i386_comment_chars = "#";
200 #define PREFIX_SEPARATOR '/'
201 #endif
203 /* This array holds the chars that only start a comment at the beginning of
204 a line. If the line seems to have the form '# 123 filename'
205 .line and .file directives will appear in the pre-processed output.
206 Note that input_file.c hand checks for '#' at the beginning of the
207 first line of the input file. This is because the compiler outputs
208 #NO_APP at the beginning of its output.
209 Also note that comments started like this one will always work if
210 '/' isn't otherwise defined. */
211 const char line_comment_chars[] = "#/";
213 const char line_separator_chars[] = ";";
215 /* Chars that can be used to separate mant from exp in floating point
216 nums. */
217 const char EXP_CHARS[] = "eE";
219 /* Chars that mean this number is a floating point constant
220 As in 0f12.456
221 or 0d1.2345e12. */
222 const char FLT_CHARS[] = "fFdDxX";
224 /* Tables for lexical analysis. */
225 static char mnemonic_chars[256];
226 static char register_chars[256];
227 static char operand_chars[256];
228 static char identifier_chars[256];
229 static char digit_chars[256];
231 /* Lexical macros. */
232 #define is_mnemonic_char(x) (mnemonic_chars[(unsigned char) x])
233 #define is_operand_char(x) (operand_chars[(unsigned char) x])
234 #define is_register_char(x) (register_chars[(unsigned char) x])
235 #define is_space_char(x) ((x) == ' ')
236 #define is_identifier_char(x) (identifier_chars[(unsigned char) x])
237 #define is_digit_char(x) (digit_chars[(unsigned char) x])
239 /* All non-digit non-letter characters that may occur in an operand. */
240 static char operand_special_chars[] = "%$-+(,)*._~/<>|&^!:[@]";
242 /* md_assemble() always leaves the strings it's passed unaltered. To
243 effect this we maintain a stack of saved characters that we've smashed
244 with '\0's (indicating end of strings for various sub-fields of the
245 assembler instruction). */
246 static char save_stack[32];
247 static char *save_stack_p;
248 #define END_STRING_AND_SAVE(s) \
249 do { *save_stack_p++ = *(s); *(s) = '\0'; } while (0)
250 #define RESTORE_END_STRING(s) \
251 do { *(s) = *--save_stack_p; } while (0)
253 /* The instruction we're assembling. */
254 static i386_insn i;
256 /* Possible templates for current insn. */
257 static const templates *current_templates;
259 /* Per instruction expressionS buffers: max displacements & immediates. */
260 static expressionS disp_expressions[MAX_MEMORY_OPERANDS];
261 static expressionS im_expressions[MAX_IMMEDIATE_OPERANDS];
263 /* Current operand we are working on. */
264 static int this_operand;
266 /* We support four different modes. FLAG_CODE variable is used to distinguish
267 these. */
269 enum flag_code {
270 CODE_32BIT,
271 CODE_16BIT,
272 CODE_64BIT };
273 #define NUM_FLAG_CODE ((int) CODE_64BIT + 1)
275 static enum flag_code flag_code;
276 static unsigned int object_64bit;
277 static int use_rela_relocations = 0;
279 /* The names used to print error messages. */
280 static const char *flag_code_names[] =
282 "32",
283 "16",
284 "64"
287 /* 1 for intel syntax,
288 0 if att syntax. */
289 static int intel_syntax = 0;
291 /* 1 if register prefix % not required. */
292 static int allow_naked_reg = 0;
294 /* Register prefix used for error message. */
295 static const char *register_prefix = "%";
297 /* Used in 16 bit gcc mode to add an l suffix to call, ret, enter,
298 leave, push, and pop instructions so that gcc has the same stack
299 frame as in 32 bit mode. */
300 static char stackop_size = '\0';
302 /* Non-zero to optimize code alignment. */
303 int optimize_align_code = 1;
305 /* Non-zero to quieten some warnings. */
306 static int quiet_warnings = 0;
308 /* CPU name. */
309 static const char *cpu_arch_name = NULL;
310 static const char *cpu_sub_arch_name = NULL;
312 /* CPU feature flags. */
313 static unsigned int cpu_arch_flags = CpuUnknownFlags | CpuNo64;
315 /* If we have selected a cpu we are generating instructions for. */
316 static int cpu_arch_tune_set = 0;
318 /* Cpu we are generating instructions for. */
319 static enum processor_type cpu_arch_tune = PROCESSOR_UNKNOWN;
321 /* CPU feature flags of cpu we are generating instructions for. */
322 static unsigned int cpu_arch_tune_flags = 0;
324 /* CPU instruction set architecture used. */
325 static enum processor_type cpu_arch_isa = PROCESSOR_UNKNOWN;
327 /* CPU feature flags of instruction set architecture used. */
328 static unsigned int cpu_arch_isa_flags = 0;
330 /* If set, conditional jumps are not automatically promoted to handle
331 larger than a byte offset. */
332 static unsigned int no_cond_jump_promotion = 0;
334 /* Pre-defined "_GLOBAL_OFFSET_TABLE_". */
335 static symbolS *GOT_symbol;
337 /* The dwarf2 return column, adjusted for 32 or 64 bit. */
338 unsigned int x86_dwarf2_return_column;
340 /* The dwarf2 data alignment, adjusted for 32 or 64 bit. */
341 int x86_cie_data_alignment;
343 /* Interface to relax_segment.
344 There are 3 major relax states for 386 jump insns because the
345 different types of jumps add different sizes to frags when we're
346 figuring out what sort of jump to choose to reach a given label. */
348 /* Types. */
349 #define UNCOND_JUMP 0
350 #define COND_JUMP 1
351 #define COND_JUMP86 2
353 /* Sizes. */
354 #define CODE16 1
355 #define SMALL 0
356 #define SMALL16 (SMALL | CODE16)
357 #define BIG 2
358 #define BIG16 (BIG | CODE16)
360 #ifndef INLINE
361 #ifdef __GNUC__
362 #define INLINE __inline__
363 #else
364 #define INLINE
365 #endif
366 #endif
368 #define ENCODE_RELAX_STATE(type, size) \
369 ((relax_substateT) (((type) << 2) | (size)))
370 #define TYPE_FROM_RELAX_STATE(s) \
371 ((s) >> 2)
372 #define DISP_SIZE_FROM_RELAX_STATE(s) \
373 ((((s) & 3) == BIG ? 4 : (((s) & 3) == BIG16 ? 2 : 1)))
375 /* This table is used by relax_frag to promote short jumps to long
376 ones where necessary. SMALL (short) jumps may be promoted to BIG
377 (32 bit long) ones, and SMALL16 jumps to BIG16 (16 bit long). We
378 don't allow a short jump in a 32 bit code segment to be promoted to
379 a 16 bit offset jump because it's slower (requires data size
380 prefix), and doesn't work, unless the destination is in the bottom
381 64k of the code segment (The top 16 bits of eip are zeroed). */
383 const relax_typeS md_relax_table[] =
385 /* The fields are:
386 1) most positive reach of this state,
387 2) most negative reach of this state,
388 3) how many bytes this mode will have in the variable part of the frag
389 4) which index into the table to try if we can't fit into this one. */
391 /* UNCOND_JUMP states. */
392 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG)},
393 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16)},
394 /* dword jmp adds 4 bytes to frag:
395 0 extra opcode bytes, 4 displacement bytes. */
396 {0, 0, 4, 0},
397 /* word jmp adds 2 byte2 to frag:
398 0 extra opcode bytes, 2 displacement bytes. */
399 {0, 0, 2, 0},
401 /* COND_JUMP states. */
402 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG)},
403 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG16)},
404 /* dword conditionals adds 5 bytes to frag:
405 1 extra opcode byte, 4 displacement bytes. */
406 {0, 0, 5, 0},
407 /* word conditionals add 3 bytes to frag:
408 1 extra opcode byte, 2 displacement bytes. */
409 {0, 0, 3, 0},
411 /* COND_JUMP86 states. */
412 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG)},
413 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG16)},
414 /* dword conditionals adds 5 bytes to frag:
415 1 extra opcode byte, 4 displacement bytes. */
416 {0, 0, 5, 0},
417 /* word conditionals add 4 bytes to frag:
418 1 displacement byte and a 3 byte long branch insn. */
419 {0, 0, 4, 0}
422 static const arch_entry cpu_arch[] =
424 {"generic32", PROCESSOR_GENERIC32,
425 Cpu186|Cpu286|Cpu386},
426 {"generic64", PROCESSOR_GENERIC64,
427 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
428 |CpuMMX2|CpuSSE|CpuSSE2},
429 {"i8086", PROCESSOR_UNKNOWN,
431 {"i186", PROCESSOR_UNKNOWN,
432 Cpu186},
433 {"i286", PROCESSOR_UNKNOWN,
434 Cpu186|Cpu286},
435 {"i386", PROCESSOR_GENERIC32,
436 Cpu186|Cpu286|Cpu386},
437 {"i486", PROCESSOR_I486,
438 Cpu186|Cpu286|Cpu386|Cpu486},
439 {"i586", PROCESSOR_PENTIUM,
440 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586},
441 {"i686", PROCESSOR_PENTIUMPRO,
442 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686},
443 {"pentium", PROCESSOR_PENTIUM,
444 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586},
445 {"pentiumpro",PROCESSOR_PENTIUMPRO,
446 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686},
447 {"pentiumii", PROCESSOR_PENTIUMPRO,
448 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX},
449 {"pentiumiii",PROCESSOR_PENTIUMPRO,
450 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuMMX2|CpuSSE},
451 {"pentium4", PROCESSOR_PENTIUM4,
452 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
453 |CpuMMX2|CpuSSE|CpuSSE2},
454 {"prescott", PROCESSOR_NOCONA,
455 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
456 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
457 {"nocona", PROCESSOR_NOCONA,
458 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
459 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
460 {"yonah", PROCESSOR_CORE,
461 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
462 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
463 {"core", PROCESSOR_CORE,
464 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
465 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
466 {"merom", PROCESSOR_CORE2,
467 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
468 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSSE3},
469 {"core2", PROCESSOR_CORE2,
470 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
471 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSSE3},
472 {"k6", PROCESSOR_K6,
473 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX},
474 {"k6_2", PROCESSOR_K6,
475 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX|Cpu3dnow},
476 {"athlon", PROCESSOR_ATHLON,
477 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6
478 |CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA},
479 {"sledgehammer", PROCESSOR_K8,
480 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6
481 |CpuSledgehammer|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2},
482 {"opteron", PROCESSOR_K8,
483 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6
484 |CpuSledgehammer|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2},
485 {"k8", PROCESSOR_K8,
486 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6
487 |CpuSledgehammer|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2},
488 {"amdfam10", PROCESSOR_AMDFAM10,
489 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuSledgehammer
490 |CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2|CpuSSE3|CpuSSE4a
491 |CpuABM},
492 {".mmx", PROCESSOR_UNKNOWN,
493 CpuMMX},
494 {".sse", PROCESSOR_UNKNOWN,
495 CpuMMX|CpuMMX2|CpuSSE},
496 {".sse2", PROCESSOR_UNKNOWN,
497 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2},
498 {".sse3", PROCESSOR_UNKNOWN,
499 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
500 {".ssse3", PROCESSOR_UNKNOWN,
501 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSSE3},
502 {".3dnow", PROCESSOR_UNKNOWN,
503 CpuMMX|Cpu3dnow},
504 {".3dnowa", PROCESSOR_UNKNOWN,
505 CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA},
506 {".padlock", PROCESSOR_UNKNOWN,
507 CpuPadLock},
508 {".pacifica", PROCESSOR_UNKNOWN,
509 CpuSVME},
510 {".svme", PROCESSOR_UNKNOWN,
511 CpuSVME},
512 {".sse4a", PROCESSOR_UNKNOWN,
513 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSE4a},
514 {".abm", PROCESSOR_UNKNOWN,
515 CpuABM}
518 const pseudo_typeS md_pseudo_table[] =
520 #if !defined(OBJ_AOUT) && !defined(USE_ALIGN_PTWO)
521 {"align", s_align_bytes, 0},
522 #else
523 {"align", s_align_ptwo, 0},
524 #endif
525 {"arch", set_cpu_arch, 0},
526 #ifndef I386COFF
527 {"bss", s_bss, 0},
528 #endif
529 {"ffloat", float_cons, 'f'},
530 {"dfloat", float_cons, 'd'},
531 {"tfloat", float_cons, 'x'},
532 {"value", cons, 2},
533 {"slong", signed_cons, 4},
534 {"noopt", s_ignore, 0},
535 {"optim", s_ignore, 0},
536 {"code16gcc", set_16bit_gcc_code_flag, CODE_16BIT},
537 {"code16", set_code_flag, CODE_16BIT},
538 {"code32", set_code_flag, CODE_32BIT},
539 {"code64", set_code_flag, CODE_64BIT},
540 {"intel_syntax", set_intel_syntax, 1},
541 {"att_syntax", set_intel_syntax, 0},
542 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
543 {"largecomm", handle_large_common, 0},
544 #else
545 {"file", (void (*) (int)) dwarf2_directive_file, 0},
546 {"loc", dwarf2_directive_loc, 0},
547 {"loc_mark_labels", dwarf2_directive_loc_mark_labels, 0},
548 #endif
549 #ifdef TE_PE
550 {"secrel32", pe_directive_secrel, 0},
551 #endif
552 {0, 0, 0}
555 /* For interface with expression (). */
556 extern char *input_line_pointer;
558 /* Hash table for instruction mnemonic lookup. */
559 static struct hash_control *op_hash;
561 /* Hash table for register lookup. */
562 static struct hash_control *reg_hash;
564 void
565 i386_align_code (fragS *fragP, int count)
567 /* Various efficient no-op patterns for aligning code labels.
568 Note: Don't try to assemble the instructions in the comments.
569 0L and 0w are not legal. */
570 static const char f32_1[] =
571 {0x90}; /* nop */
572 static const char f32_2[] =
573 {0x66,0x90}; /* xchg %ax,%ax */
574 static const char f32_3[] =
575 {0x8d,0x76,0x00}; /* leal 0(%esi),%esi */
576 static const char f32_4[] =
577 {0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
578 static const char f32_5[] =
579 {0x90, /* nop */
580 0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
581 static const char f32_6[] =
582 {0x8d,0xb6,0x00,0x00,0x00,0x00}; /* leal 0L(%esi),%esi */
583 static const char f32_7[] =
584 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
585 static const char f32_8[] =
586 {0x90, /* nop */
587 0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
588 static const char f32_9[] =
589 {0x89,0xf6, /* movl %esi,%esi */
590 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
591 static const char f32_10[] =
592 {0x8d,0x76,0x00, /* leal 0(%esi),%esi */
593 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
594 static const char f32_11[] =
595 {0x8d,0x74,0x26,0x00, /* leal 0(%esi,1),%esi */
596 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
597 static const char f32_12[] =
598 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
599 0x8d,0xbf,0x00,0x00,0x00,0x00}; /* leal 0L(%edi),%edi */
600 static const char f32_13[] =
601 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
602 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
603 static const char f32_14[] =
604 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00, /* leal 0L(%esi,1),%esi */
605 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
606 static const char f32_15[] =
607 {0xeb,0x0d,0x90,0x90,0x90,0x90,0x90, /* jmp .+15; lotsa nops */
608 0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90};
609 static const char f16_3[] =
610 {0x8d,0x74,0x00}; /* lea 0(%esi),%esi */
611 static const char f16_4[] =
612 {0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
613 static const char f16_5[] =
614 {0x90, /* nop */
615 0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
616 static const char f16_6[] =
617 {0x89,0xf6, /* mov %si,%si */
618 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
619 static const char f16_7[] =
620 {0x8d,0x74,0x00, /* lea 0(%si),%si */
621 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
622 static const char f16_8[] =
623 {0x8d,0xb4,0x00,0x00, /* lea 0w(%si),%si */
624 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
625 static const char *const f32_patt[] = {
626 f32_1, f32_2, f32_3, f32_4, f32_5, f32_6, f32_7, f32_8,
627 f32_9, f32_10, f32_11, f32_12, f32_13, f32_14, f32_15
629 static const char *const f16_patt[] = {
630 f32_1, f32_2, f16_3, f16_4, f16_5, f16_6, f16_7, f16_8,
631 f32_15, f32_15, f32_15, f32_15, f32_15, f32_15, f32_15
633 /* nopl (%[re]ax) */
634 static const char alt_3[] =
635 {0x0f,0x1f,0x00};
636 /* nopl 0(%[re]ax) */
637 static const char alt_4[] =
638 {0x0f,0x1f,0x40,0x00};
639 /* nopl 0(%[re]ax,%[re]ax,1) */
640 static const char alt_5[] =
641 {0x0f,0x1f,0x44,0x00,0x00};
642 /* nopw 0(%[re]ax,%[re]ax,1) */
643 static const char alt_6[] =
644 {0x66,0x0f,0x1f,0x44,0x00,0x00};
645 /* nopl 0L(%[re]ax) */
646 static const char alt_7[] =
647 {0x0f,0x1f,0x80,0x00,0x00,0x00,0x00};
648 /* nopl 0L(%[re]ax,%[re]ax,1) */
649 static const char alt_8[] =
650 {0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
651 /* nopw 0L(%[re]ax,%[re]ax,1) */
652 static const char alt_9[] =
653 {0x66,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
654 /* nopw %cs:0L(%[re]ax,%[re]ax,1) */
655 static const char alt_10[] =
656 {0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
657 /* data16
658 nopw %cs:0L(%[re]ax,%[re]ax,1) */
659 static const char alt_long_11[] =
660 {0x66,
661 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
662 /* data16
663 data16
664 nopw %cs:0L(%[re]ax,%[re]ax,1) */
665 static const char alt_long_12[] =
666 {0x66,
667 0x66,
668 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
669 /* data16
670 data16
671 data16
672 nopw %cs:0L(%[re]ax,%[re]ax,1) */
673 static const char alt_long_13[] =
674 {0x66,
675 0x66,
676 0x66,
677 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
678 /* data16
679 data16
680 data16
681 data16
682 nopw %cs:0L(%[re]ax,%[re]ax,1) */
683 static const char alt_long_14[] =
684 {0x66,
685 0x66,
686 0x66,
687 0x66,
688 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
689 /* data16
690 data16
691 data16
692 data16
693 data16
694 nopw %cs:0L(%[re]ax,%[re]ax,1) */
695 static const char alt_long_15[] =
696 {0x66,
697 0x66,
698 0x66,
699 0x66,
700 0x66,
701 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
702 /* nopl 0(%[re]ax,%[re]ax,1)
703 nopw 0(%[re]ax,%[re]ax,1) */
704 static const char alt_short_11[] =
705 {0x0f,0x1f,0x44,0x00,0x00,
706 0x66,0x0f,0x1f,0x44,0x00,0x00};
707 /* nopw 0(%[re]ax,%[re]ax,1)
708 nopw 0(%[re]ax,%[re]ax,1) */
709 static const char alt_short_12[] =
710 {0x66,0x0f,0x1f,0x44,0x00,0x00,
711 0x66,0x0f,0x1f,0x44,0x00,0x00};
712 /* nopw 0(%[re]ax,%[re]ax,1)
713 nopl 0L(%[re]ax) */
714 static const char alt_short_13[] =
715 {0x66,0x0f,0x1f,0x44,0x00,0x00,
716 0x0f,0x1f,0x80,0x00,0x00,0x00,0x00};
717 /* nopl 0L(%[re]ax)
718 nopl 0L(%[re]ax) */
719 static const char alt_short_14[] =
720 {0x0f,0x1f,0x80,0x00,0x00,0x00,0x00,
721 0x0f,0x1f,0x80,0x00,0x00,0x00,0x00};
722 /* nopl 0L(%[re]ax)
723 nopl 0L(%[re]ax,%[re]ax,1) */
724 static const char alt_short_15[] =
725 {0x0f,0x1f,0x80,0x00,0x00,0x00,0x00,
726 0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
727 static const char *const alt_short_patt[] = {
728 f32_1, f32_2, alt_3, alt_4, alt_5, alt_6, alt_7, alt_8,
729 alt_9, alt_10, alt_short_11, alt_short_12, alt_short_13,
730 alt_short_14, alt_short_15
732 static const char *const alt_long_patt[] = {
733 f32_1, f32_2, alt_3, alt_4, alt_5, alt_6, alt_7, alt_8,
734 alt_9, alt_10, alt_long_11, alt_long_12, alt_long_13,
735 alt_long_14, alt_long_15
738 if (count <= 0 || count > 15)
739 return;
741 /* We need to decide which NOP sequence to use for 32bit and
742 64bit. When -mtune= is used:
744 1. For PROCESSOR_I486, PROCESSOR_PENTIUM and PROCESSOR_GENERIC32,
745 f32_patt will be used.
746 2. For PROCESSOR_K8 and PROCESSOR_AMDFAM10 in 64bit, NOPs with
747 0x66 prefix will be used.
748 3. For PROCESSOR_CORE2, alt_long_patt will be used.
749 4. For PROCESSOR_PENTIUMPRO, PROCESSOR_PENTIUM4, PROCESSOR_NOCONA,
750 PROCESSOR_CORE, PROCESSOR_CORE2, PROCESSOR_K6, PROCESSOR_ATHLON
751 and PROCESSOR_GENERIC64, alt_short_patt will be used.
753 When -mtune= isn't used, alt_short_patt will be used if
754 cpu_arch_isa_flags has Cpu686. Otherwise, f32_patt will be used.
756 When -march= or .arch is used, we can't use anything beyond
757 cpu_arch_isa_flags. */
759 if (flag_code == CODE_16BIT)
761 memcpy (fragP->fr_literal + fragP->fr_fix,
762 f16_patt[count - 1], count);
763 if (count > 8)
764 /* Adjust jump offset. */
765 fragP->fr_literal[fragP->fr_fix + 1] = count - 2;
767 else if (flag_code == CODE_64BIT && cpu_arch_tune == PROCESSOR_K8)
769 int i;
770 int nnops = (count + 3) / 4;
771 int len = count / nnops;
772 int remains = count - nnops * len;
773 int pos = 0;
775 /* The recommended way to pad 64bit code is to use NOPs preceded
776 by maximally four 0x66 prefixes. Balance the size of nops. */
777 for (i = 0; i < remains; i++)
779 memset (fragP->fr_literal + fragP->fr_fix + pos, 0x66, len);
780 fragP->fr_literal[fragP->fr_fix + pos + len] = 0x90;
781 pos += len + 1;
783 for (; i < nnops; i++)
785 memset (fragP->fr_literal + fragP->fr_fix + pos, 0x66, len - 1);
786 fragP->fr_literal[fragP->fr_fix + pos + len - 1] = 0x90;
787 pos += len;
790 else
792 const char *const *patt = NULL;
794 if (cpu_arch_isa == PROCESSOR_UNKNOWN)
796 /* PROCESSOR_UNKNOWN means that all ISAs may be used. */
797 switch (cpu_arch_tune)
799 case PROCESSOR_UNKNOWN:
800 /* We use cpu_arch_isa_flags to check if we SHOULD
801 optimize for Cpu686. */
802 if ((cpu_arch_isa_flags & Cpu686) != 0)
803 patt = alt_short_patt;
804 else
805 patt = f32_patt;
806 break;
807 case PROCESSOR_CORE2:
808 patt = alt_long_patt;
809 break;
810 case PROCESSOR_PENTIUMPRO:
811 case PROCESSOR_PENTIUM4:
812 case PROCESSOR_NOCONA:
813 case PROCESSOR_CORE:
814 case PROCESSOR_K6:
815 case PROCESSOR_ATHLON:
816 case PROCESSOR_K8:
817 case PROCESSOR_GENERIC64:
818 case PROCESSOR_AMDFAM10:
819 patt = alt_short_patt;
820 break;
821 case PROCESSOR_I486:
822 case PROCESSOR_PENTIUM:
823 case PROCESSOR_GENERIC32:
824 patt = f32_patt;
825 break;
828 else
830 switch (cpu_arch_tune)
832 case PROCESSOR_UNKNOWN:
833 /* When cpu_arch_isa is net, cpu_arch_tune shouldn't be
834 PROCESSOR_UNKNOWN. */
835 abort ();
836 break;
838 case PROCESSOR_I486:
839 case PROCESSOR_PENTIUM:
840 case PROCESSOR_PENTIUMPRO:
841 case PROCESSOR_PENTIUM4:
842 case PROCESSOR_NOCONA:
843 case PROCESSOR_CORE:
844 case PROCESSOR_K6:
845 case PROCESSOR_ATHLON:
846 case PROCESSOR_K8:
847 case PROCESSOR_AMDFAM10:
848 case PROCESSOR_GENERIC32:
849 /* We use cpu_arch_isa_flags to check if we CAN optimize
850 for Cpu686. */
851 if ((cpu_arch_isa_flags & Cpu686) != 0)
852 patt = alt_short_patt;
853 else
854 patt = f32_patt;
855 break;
856 case PROCESSOR_CORE2:
857 if ((cpu_arch_isa_flags & Cpu686) != 0)
858 patt = alt_long_patt;
859 else
860 patt = f32_patt;
861 break;
862 case PROCESSOR_GENERIC64:
863 patt = alt_short_patt;
864 break;
868 memcpy (fragP->fr_literal + fragP->fr_fix,
869 patt[count - 1], count);
871 fragP->fr_var = count;
874 static INLINE unsigned int
875 mode_from_disp_size (unsigned int t)
877 return (t & Disp8) ? 1 : (t & (Disp16 | Disp32 | Disp32S)) ? 2 : 0;
880 static INLINE int
881 fits_in_signed_byte (offsetT num)
883 return (num >= -128) && (num <= 127);
886 static INLINE int
887 fits_in_unsigned_byte (offsetT num)
889 return (num & 0xff) == num;
892 static INLINE int
893 fits_in_unsigned_word (offsetT num)
895 return (num & 0xffff) == num;
898 static INLINE int
899 fits_in_signed_word (offsetT num)
901 return (-32768 <= num) && (num <= 32767);
904 static INLINE int
905 fits_in_signed_long (offsetT num ATTRIBUTE_UNUSED)
907 #ifndef BFD64
908 return 1;
909 #else
910 return (!(((offsetT) -1 << 31) & num)
911 || (((offsetT) -1 << 31) & num) == ((offsetT) -1 << 31));
912 #endif
913 } /* fits_in_signed_long() */
915 static INLINE int
916 fits_in_unsigned_long (offsetT num ATTRIBUTE_UNUSED)
918 #ifndef BFD64
919 return 1;
920 #else
921 return (num & (((offsetT) 2 << 31) - 1)) == num;
922 #endif
923 } /* fits_in_unsigned_long() */
925 static unsigned int
926 smallest_imm_type (offsetT num)
928 if (cpu_arch_flags != (Cpu186 | Cpu286 | Cpu386 | Cpu486 | CpuNo64))
930 /* This code is disabled on the 486 because all the Imm1 forms
931 in the opcode table are slower on the i486. They're the
932 versions with the implicitly specified single-position
933 displacement, which has another syntax if you really want to
934 use that form. */
935 if (num == 1)
936 return Imm1 | Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64;
938 return (fits_in_signed_byte (num)
939 ? (Imm8S | Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
940 : fits_in_unsigned_byte (num)
941 ? (Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
942 : (fits_in_signed_word (num) || fits_in_unsigned_word (num))
943 ? (Imm16 | Imm32 | Imm32S | Imm64)
944 : fits_in_signed_long (num)
945 ? (Imm32 | Imm32S | Imm64)
946 : fits_in_unsigned_long (num)
947 ? (Imm32 | Imm64)
948 : Imm64);
951 static offsetT
952 offset_in_range (offsetT val, int size)
954 addressT mask;
956 switch (size)
958 case 1: mask = ((addressT) 1 << 8) - 1; break;
959 case 2: mask = ((addressT) 1 << 16) - 1; break;
960 case 4: mask = ((addressT) 2 << 31) - 1; break;
961 #ifdef BFD64
962 case 8: mask = ((addressT) 2 << 63) - 1; break;
963 #endif
964 default: abort ();
967 /* If BFD64, sign extend val. */
968 if (!use_rela_relocations)
969 if ((val & ~(((addressT) 2 << 31) - 1)) == 0)
970 val = (val ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
972 if ((val & ~mask) != 0 && (val & ~mask) != ~mask)
974 char buf1[40], buf2[40];
976 sprint_value (buf1, val);
977 sprint_value (buf2, val & mask);
978 as_warn (_("%s shortened to %s"), buf1, buf2);
980 return val & mask;
983 /* Returns 0 if attempting to add a prefix where one from the same
984 class already exists, 1 if non rep/repne added, 2 if rep/repne
985 added. */
986 static int
987 add_prefix (unsigned int prefix)
989 int ret = 1;
990 unsigned int q;
992 if (prefix >= REX_OPCODE && prefix < REX_OPCODE + 16
993 && flag_code == CODE_64BIT)
995 if ((i.prefix[REX_PREFIX] & prefix & REX_MODE64)
996 || ((i.prefix[REX_PREFIX] & (REX_EXTX | REX_EXTY | REX_EXTZ))
997 && (prefix & (REX_EXTX | REX_EXTY | REX_EXTZ))))
998 ret = 0;
999 q = REX_PREFIX;
1001 else
1003 switch (prefix)
1005 default:
1006 abort ();
1008 case CS_PREFIX_OPCODE:
1009 case DS_PREFIX_OPCODE:
1010 case ES_PREFIX_OPCODE:
1011 case FS_PREFIX_OPCODE:
1012 case GS_PREFIX_OPCODE:
1013 case SS_PREFIX_OPCODE:
1014 q = SEG_PREFIX;
1015 break;
1017 case REPNE_PREFIX_OPCODE:
1018 case REPE_PREFIX_OPCODE:
1019 ret = 2;
1020 /* fall thru */
1021 case LOCK_PREFIX_OPCODE:
1022 q = LOCKREP_PREFIX;
1023 break;
1025 case FWAIT_OPCODE:
1026 q = WAIT_PREFIX;
1027 break;
1029 case ADDR_PREFIX_OPCODE:
1030 q = ADDR_PREFIX;
1031 break;
1033 case DATA_PREFIX_OPCODE:
1034 q = DATA_PREFIX;
1035 break;
1037 if (i.prefix[q] != 0)
1038 ret = 0;
1041 if (ret)
1043 if (!i.prefix[q])
1044 ++i.prefixes;
1045 i.prefix[q] |= prefix;
1047 else
1048 as_bad (_("same type of prefix used twice"));
1050 return ret;
1053 static void
1054 set_code_flag (int value)
1056 flag_code = value;
1057 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
1058 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
1059 if (value == CODE_64BIT && !(cpu_arch_flags & CpuSledgehammer))
1061 as_bad (_("64bit mode not supported on this CPU."));
1063 if (value == CODE_32BIT && !(cpu_arch_flags & Cpu386))
1065 as_bad (_("32bit mode not supported on this CPU."));
1067 stackop_size = '\0';
1070 static void
1071 set_16bit_gcc_code_flag (int new_code_flag)
1073 flag_code = new_code_flag;
1074 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
1075 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
1076 stackop_size = LONG_MNEM_SUFFIX;
1079 static void
1080 set_intel_syntax (int syntax_flag)
1082 /* Find out if register prefixing is specified. */
1083 int ask_naked_reg = 0;
1085 SKIP_WHITESPACE ();
1086 if (!is_end_of_line[(unsigned char) *input_line_pointer])
1088 char *string = input_line_pointer;
1089 int e = get_symbol_end ();
1091 if (strcmp (string, "prefix") == 0)
1092 ask_naked_reg = 1;
1093 else if (strcmp (string, "noprefix") == 0)
1094 ask_naked_reg = -1;
1095 else
1096 as_bad (_("bad argument to syntax directive."));
1097 *input_line_pointer = e;
1099 demand_empty_rest_of_line ();
1101 intel_syntax = syntax_flag;
1103 if (ask_naked_reg == 0)
1104 allow_naked_reg = (intel_syntax
1105 && (bfd_get_symbol_leading_char (stdoutput) != '\0'));
1106 else
1107 allow_naked_reg = (ask_naked_reg < 0);
1109 identifier_chars['%'] = intel_syntax && allow_naked_reg ? '%' : 0;
1110 identifier_chars['$'] = intel_syntax ? '$' : 0;
1111 register_prefix = allow_naked_reg ? "" : "%";
1114 static void
1115 set_cpu_arch (int dummy ATTRIBUTE_UNUSED)
1117 SKIP_WHITESPACE ();
1119 if (!is_end_of_line[(unsigned char) *input_line_pointer])
1121 char *string = input_line_pointer;
1122 int e = get_symbol_end ();
1123 unsigned int i;
1125 for (i = 0; i < ARRAY_SIZE (cpu_arch); i++)
1127 if (strcmp (string, cpu_arch[i].name) == 0)
1129 if (*string != '.')
1131 cpu_arch_name = cpu_arch[i].name;
1132 cpu_sub_arch_name = NULL;
1133 cpu_arch_flags = (cpu_arch[i].flags
1134 | (flag_code == CODE_64BIT
1135 ? Cpu64 : CpuNo64));
1136 cpu_arch_isa = cpu_arch[i].type;
1137 cpu_arch_isa_flags = cpu_arch[i].flags;
1138 if (!cpu_arch_tune_set)
1140 cpu_arch_tune = cpu_arch_isa;
1141 cpu_arch_tune_flags = cpu_arch_isa_flags;
1143 break;
1145 if ((cpu_arch_flags | cpu_arch[i].flags) != cpu_arch_flags)
1147 cpu_sub_arch_name = cpu_arch[i].name;
1148 cpu_arch_flags |= cpu_arch[i].flags;
1150 *input_line_pointer = e;
1151 demand_empty_rest_of_line ();
1152 return;
1155 if (i >= ARRAY_SIZE (cpu_arch))
1156 as_bad (_("no such architecture: `%s'"), string);
1158 *input_line_pointer = e;
1160 else
1161 as_bad (_("missing cpu architecture"));
1163 no_cond_jump_promotion = 0;
1164 if (*input_line_pointer == ','
1165 && !is_end_of_line[(unsigned char) input_line_pointer[1]])
1167 char *string = ++input_line_pointer;
1168 int e = get_symbol_end ();
1170 if (strcmp (string, "nojumps") == 0)
1171 no_cond_jump_promotion = 1;
1172 else if (strcmp (string, "jumps") == 0)
1174 else
1175 as_bad (_("no such architecture modifier: `%s'"), string);
1177 *input_line_pointer = e;
1180 demand_empty_rest_of_line ();
1183 unsigned long
1184 i386_mach ()
1186 if (!strcmp (default_arch, "x86_64"))
1187 return bfd_mach_x86_64;
1188 else if (!strcmp (default_arch, "i386"))
1189 return bfd_mach_i386_i386;
1190 else
1191 as_fatal (_("Unknown architecture"));
1194 void
1195 md_begin ()
1197 const char *hash_err;
1199 /* Initialize op_hash hash table. */
1200 op_hash = hash_new ();
1203 const template *optab;
1204 templates *core_optab;
1206 /* Setup for loop. */
1207 optab = i386_optab;
1208 core_optab = (templates *) xmalloc (sizeof (templates));
1209 core_optab->start = optab;
1211 while (1)
1213 ++optab;
1214 if (optab->name == NULL
1215 || strcmp (optab->name, (optab - 1)->name) != 0)
1217 /* different name --> ship out current template list;
1218 add to hash table; & begin anew. */
1219 core_optab->end = optab;
1220 hash_err = hash_insert (op_hash,
1221 (optab - 1)->name,
1222 (PTR) core_optab);
1223 if (hash_err)
1225 as_fatal (_("Internal Error: Can't hash %s: %s"),
1226 (optab - 1)->name,
1227 hash_err);
1229 if (optab->name == NULL)
1230 break;
1231 core_optab = (templates *) xmalloc (sizeof (templates));
1232 core_optab->start = optab;
1237 /* Initialize reg_hash hash table. */
1238 reg_hash = hash_new ();
1240 const reg_entry *regtab;
1242 for (regtab = i386_regtab;
1243 regtab < i386_regtab + sizeof (i386_regtab) / sizeof (i386_regtab[0]);
1244 regtab++)
1246 hash_err = hash_insert (reg_hash, regtab->reg_name, (PTR) regtab);
1247 if (hash_err)
1248 as_fatal (_("Internal Error: Can't hash %s: %s"),
1249 regtab->reg_name,
1250 hash_err);
1254 /* Fill in lexical tables: mnemonic_chars, operand_chars. */
1256 int c;
1257 char *p;
1259 for (c = 0; c < 256; c++)
1261 if (ISDIGIT (c))
1263 digit_chars[c] = c;
1264 mnemonic_chars[c] = c;
1265 register_chars[c] = c;
1266 operand_chars[c] = c;
1268 else if (ISLOWER (c))
1270 mnemonic_chars[c] = c;
1271 register_chars[c] = c;
1272 operand_chars[c] = c;
1274 else if (ISUPPER (c))
1276 mnemonic_chars[c] = TOLOWER (c);
1277 register_chars[c] = mnemonic_chars[c];
1278 operand_chars[c] = c;
1281 if (ISALPHA (c) || ISDIGIT (c))
1282 identifier_chars[c] = c;
1283 else if (c >= 128)
1285 identifier_chars[c] = c;
1286 operand_chars[c] = c;
1290 #ifdef LEX_AT
1291 identifier_chars['@'] = '@';
1292 #endif
1293 #ifdef LEX_QM
1294 identifier_chars['?'] = '?';
1295 operand_chars['?'] = '?';
1296 #endif
1297 digit_chars['-'] = '-';
1298 mnemonic_chars['-'] = '-';
1299 identifier_chars['_'] = '_';
1300 identifier_chars['.'] = '.';
1302 for (p = operand_special_chars; *p != '\0'; p++)
1303 operand_chars[(unsigned char) *p] = *p;
1306 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1307 if (IS_ELF)
1309 record_alignment (text_section, 2);
1310 record_alignment (data_section, 2);
1311 record_alignment (bss_section, 2);
1313 #endif
1315 if (flag_code == CODE_64BIT)
1317 x86_dwarf2_return_column = 16;
1318 x86_cie_data_alignment = -8;
1320 else
1322 x86_dwarf2_return_column = 8;
1323 x86_cie_data_alignment = -4;
1327 void
1328 i386_print_statistics (FILE *file)
1330 hash_print_statistics (file, "i386 opcode", op_hash);
1331 hash_print_statistics (file, "i386 register", reg_hash);
1334 #ifdef DEBUG386
1336 /* Debugging routines for md_assemble. */
1337 static void pte (template *);
1338 static void pt (unsigned int);
1339 static void pe (expressionS *);
1340 static void ps (symbolS *);
1342 static void
1343 pi (char *line, i386_insn *x)
1345 unsigned int i;
1347 fprintf (stdout, "%s: template ", line);
1348 pte (&x->tm);
1349 fprintf (stdout, " address: base %s index %s scale %x\n",
1350 x->base_reg ? x->base_reg->reg_name : "none",
1351 x->index_reg ? x->index_reg->reg_name : "none",
1352 x->log2_scale_factor);
1353 fprintf (stdout, " modrm: mode %x reg %x reg/mem %x\n",
1354 x->rm.mode, x->rm.reg, x->rm.regmem);
1355 fprintf (stdout, " sib: base %x index %x scale %x\n",
1356 x->sib.base, x->sib.index, x->sib.scale);
1357 fprintf (stdout, " rex: 64bit %x extX %x extY %x extZ %x\n",
1358 (x->rex & REX_MODE64) != 0,
1359 (x->rex & REX_EXTX) != 0,
1360 (x->rex & REX_EXTY) != 0,
1361 (x->rex & REX_EXTZ) != 0);
1362 for (i = 0; i < x->operands; i++)
1364 fprintf (stdout, " #%d: ", i + 1);
1365 pt (x->types[i]);
1366 fprintf (stdout, "\n");
1367 if (x->types[i]
1368 & (Reg | SReg2 | SReg3 | Control | Debug | Test | RegMMX | RegXMM))
1369 fprintf (stdout, "%s\n", x->op[i].regs->reg_name);
1370 if (x->types[i] & Imm)
1371 pe (x->op[i].imms);
1372 if (x->types[i] & Disp)
1373 pe (x->op[i].disps);
1377 static void
1378 pte (template *t)
1380 unsigned int i;
1381 fprintf (stdout, " %d operands ", t->operands);
1382 fprintf (stdout, "opcode %x ", t->base_opcode);
1383 if (t->extension_opcode != None)
1384 fprintf (stdout, "ext %x ", t->extension_opcode);
1385 if (t->opcode_modifier & D)
1386 fprintf (stdout, "D");
1387 if (t->opcode_modifier & W)
1388 fprintf (stdout, "W");
1389 fprintf (stdout, "\n");
1390 for (i = 0; i < t->operands; i++)
1392 fprintf (stdout, " #%d type ", i + 1);
1393 pt (t->operand_types[i]);
1394 fprintf (stdout, "\n");
1398 static void
1399 pe (expressionS *e)
1401 fprintf (stdout, " operation %d\n", e->X_op);
1402 fprintf (stdout, " add_number %ld (%lx)\n",
1403 (long) e->X_add_number, (long) e->X_add_number);
1404 if (e->X_add_symbol)
1406 fprintf (stdout, " add_symbol ");
1407 ps (e->X_add_symbol);
1408 fprintf (stdout, "\n");
1410 if (e->X_op_symbol)
1412 fprintf (stdout, " op_symbol ");
1413 ps (e->X_op_symbol);
1414 fprintf (stdout, "\n");
1418 static void
1419 ps (symbolS *s)
1421 fprintf (stdout, "%s type %s%s",
1422 S_GET_NAME (s),
1423 S_IS_EXTERNAL (s) ? "EXTERNAL " : "",
1424 segment_name (S_GET_SEGMENT (s)));
1427 static struct type_name
1429 unsigned int mask;
1430 char *tname;
1432 const type_names[] =
1434 { Reg8, "r8" },
1435 { Reg16, "r16" },
1436 { Reg32, "r32" },
1437 { Reg64, "r64" },
1438 { Imm8, "i8" },
1439 { Imm8S, "i8s" },
1440 { Imm16, "i16" },
1441 { Imm32, "i32" },
1442 { Imm32S, "i32s" },
1443 { Imm64, "i64" },
1444 { Imm1, "i1" },
1445 { BaseIndex, "BaseIndex" },
1446 { Disp8, "d8" },
1447 { Disp16, "d16" },
1448 { Disp32, "d32" },
1449 { Disp32S, "d32s" },
1450 { Disp64, "d64" },
1451 { InOutPortReg, "InOutPortReg" },
1452 { ShiftCount, "ShiftCount" },
1453 { Control, "control reg" },
1454 { Test, "test reg" },
1455 { Debug, "debug reg" },
1456 { FloatReg, "FReg" },
1457 { FloatAcc, "FAcc" },
1458 { SReg2, "SReg2" },
1459 { SReg3, "SReg3" },
1460 { Acc, "Acc" },
1461 { JumpAbsolute, "Jump Absolute" },
1462 { RegMMX, "rMMX" },
1463 { RegXMM, "rXMM" },
1464 { EsSeg, "es" },
1465 { 0, "" }
1468 static void
1469 pt (t)
1470 unsigned int t;
1472 const struct type_name *ty;
1474 for (ty = type_names; ty->mask; ty++)
1475 if (t & ty->mask)
1476 fprintf (stdout, "%s, ", ty->tname);
1477 fflush (stdout);
1480 #endif /* DEBUG386 */
1482 static bfd_reloc_code_real_type
1483 reloc (unsigned int size,
1484 int pcrel,
1485 int sign,
1486 bfd_reloc_code_real_type other)
1488 if (other != NO_RELOC)
1490 reloc_howto_type *reloc;
1492 if (size == 8)
1493 switch (other)
1495 case BFD_RELOC_X86_64_GOT32:
1496 return BFD_RELOC_X86_64_GOT64;
1497 break;
1498 case BFD_RELOC_X86_64_PLTOFF64:
1499 return BFD_RELOC_X86_64_PLTOFF64;
1500 break;
1501 case BFD_RELOC_X86_64_GOTPC32:
1502 other = BFD_RELOC_X86_64_GOTPC64;
1503 break;
1504 case BFD_RELOC_X86_64_GOTPCREL:
1505 other = BFD_RELOC_X86_64_GOTPCREL64;
1506 break;
1507 case BFD_RELOC_X86_64_TPOFF32:
1508 other = BFD_RELOC_X86_64_TPOFF64;
1509 break;
1510 case BFD_RELOC_X86_64_DTPOFF32:
1511 other = BFD_RELOC_X86_64_DTPOFF64;
1512 break;
1513 default:
1514 break;
1517 /* Sign-checking 4-byte relocations in 16-/32-bit code is pointless. */
1518 if (size == 4 && flag_code != CODE_64BIT)
1519 sign = -1;
1521 reloc = bfd_reloc_type_lookup (stdoutput, other);
1522 if (!reloc)
1523 as_bad (_("unknown relocation (%u)"), other);
1524 else if (size != bfd_get_reloc_size (reloc))
1525 as_bad (_("%u-byte relocation cannot be applied to %u-byte field"),
1526 bfd_get_reloc_size (reloc),
1527 size);
1528 else if (pcrel && !reloc->pc_relative)
1529 as_bad (_("non-pc-relative relocation for pc-relative field"));
1530 else if ((reloc->complain_on_overflow == complain_overflow_signed
1531 && !sign)
1532 || (reloc->complain_on_overflow == complain_overflow_unsigned
1533 && sign > 0))
1534 as_bad (_("relocated field and relocation type differ in signedness"));
1535 else
1536 return other;
1537 return NO_RELOC;
1540 if (pcrel)
1542 if (!sign)
1543 as_bad (_("there are no unsigned pc-relative relocations"));
1544 switch (size)
1546 case 1: return BFD_RELOC_8_PCREL;
1547 case 2: return BFD_RELOC_16_PCREL;
1548 case 4: return BFD_RELOC_32_PCREL;
1549 case 8: return BFD_RELOC_64_PCREL;
1551 as_bad (_("cannot do %u byte pc-relative relocation"), size);
1553 else
1555 if (sign > 0)
1556 switch (size)
1558 case 4: return BFD_RELOC_X86_64_32S;
1560 else
1561 switch (size)
1563 case 1: return BFD_RELOC_8;
1564 case 2: return BFD_RELOC_16;
1565 case 4: return BFD_RELOC_32;
1566 case 8: return BFD_RELOC_64;
1568 as_bad (_("cannot do %s %u byte relocation"),
1569 sign > 0 ? "signed" : "unsigned", size);
1572 abort ();
1573 return BFD_RELOC_NONE;
1576 /* Here we decide which fixups can be adjusted to make them relative to
1577 the beginning of the section instead of the symbol. Basically we need
1578 to make sure that the dynamic relocations are done correctly, so in
1579 some cases we force the original symbol to be used. */
1582 tc_i386_fix_adjustable (fixS *fixP ATTRIBUTE_UNUSED)
1584 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1585 if (!IS_ELF)
1586 return 1;
1588 /* Don't adjust pc-relative references to merge sections in 64-bit
1589 mode. */
1590 if (use_rela_relocations
1591 && (S_GET_SEGMENT (fixP->fx_addsy)->flags & SEC_MERGE) != 0
1592 && fixP->fx_pcrel)
1593 return 0;
1595 /* The x86_64 GOTPCREL are represented as 32bit PCrel relocations
1596 and changed later by validate_fix. */
1597 if (GOT_symbol && fixP->fx_subsy == GOT_symbol
1598 && fixP->fx_r_type == BFD_RELOC_32_PCREL)
1599 return 0;
1601 /* adjust_reloc_syms doesn't know about the GOT. */
1602 if (fixP->fx_r_type == BFD_RELOC_386_GOTOFF
1603 || fixP->fx_r_type == BFD_RELOC_386_PLT32
1604 || fixP->fx_r_type == BFD_RELOC_386_GOT32
1605 || fixP->fx_r_type == BFD_RELOC_386_TLS_GD
1606 || fixP->fx_r_type == BFD_RELOC_386_TLS_LDM
1607 || fixP->fx_r_type == BFD_RELOC_386_TLS_LDO_32
1608 || fixP->fx_r_type == BFD_RELOC_386_TLS_IE_32
1609 || fixP->fx_r_type == BFD_RELOC_386_TLS_IE
1610 || fixP->fx_r_type == BFD_RELOC_386_TLS_GOTIE
1611 || fixP->fx_r_type == BFD_RELOC_386_TLS_LE_32
1612 || fixP->fx_r_type == BFD_RELOC_386_TLS_LE
1613 || fixP->fx_r_type == BFD_RELOC_386_TLS_GOTDESC
1614 || fixP->fx_r_type == BFD_RELOC_386_TLS_DESC_CALL
1615 || fixP->fx_r_type == BFD_RELOC_X86_64_PLT32
1616 || fixP->fx_r_type == BFD_RELOC_X86_64_GOT32
1617 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPCREL
1618 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSGD
1619 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSLD
1620 || fixP->fx_r_type == BFD_RELOC_X86_64_DTPOFF32
1621 || fixP->fx_r_type == BFD_RELOC_X86_64_DTPOFF64
1622 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTTPOFF
1623 || fixP->fx_r_type == BFD_RELOC_X86_64_TPOFF32
1624 || fixP->fx_r_type == BFD_RELOC_X86_64_TPOFF64
1625 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTOFF64
1626 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPC32_TLSDESC
1627 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSDESC_CALL
1628 || fixP->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1629 || fixP->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1630 return 0;
1631 #endif
1632 return 1;
1635 static int
1636 intel_float_operand (const char *mnemonic)
1638 /* Note that the value returned is meaningful only for opcodes with (memory)
1639 operands, hence the code here is free to improperly handle opcodes that
1640 have no operands (for better performance and smaller code). */
1642 if (mnemonic[0] != 'f')
1643 return 0; /* non-math */
1645 switch (mnemonic[1])
1647 /* fclex, fdecstp, fdisi, femms, feni, fincstp, finit, fsetpm, and
1648 the fs segment override prefix not currently handled because no
1649 call path can make opcodes without operands get here */
1650 case 'i':
1651 return 2 /* integer op */;
1652 case 'l':
1653 if (mnemonic[2] == 'd' && (mnemonic[3] == 'c' || mnemonic[3] == 'e'))
1654 return 3; /* fldcw/fldenv */
1655 break;
1656 case 'n':
1657 if (mnemonic[2] != 'o' /* fnop */)
1658 return 3; /* non-waiting control op */
1659 break;
1660 case 'r':
1661 if (mnemonic[2] == 's')
1662 return 3; /* frstor/frstpm */
1663 break;
1664 case 's':
1665 if (mnemonic[2] == 'a')
1666 return 3; /* fsave */
1667 if (mnemonic[2] == 't')
1669 switch (mnemonic[3])
1671 case 'c': /* fstcw */
1672 case 'd': /* fstdw */
1673 case 'e': /* fstenv */
1674 case 's': /* fsts[gw] */
1675 return 3;
1678 break;
1679 case 'x':
1680 if (mnemonic[2] == 'r' || mnemonic[2] == 's')
1681 return 0; /* fxsave/fxrstor are not really math ops */
1682 break;
1685 return 1;
1688 /* This is the guts of the machine-dependent assembler. LINE points to a
1689 machine dependent instruction. This function is supposed to emit
1690 the frags/bytes it assembles to. */
1692 void
1693 md_assemble (line)
1694 char *line;
1696 int j;
1697 char mnemonic[MAX_MNEM_SIZE];
1699 /* Initialize globals. */
1700 memset (&i, '\0', sizeof (i));
1701 for (j = 0; j < MAX_OPERANDS; j++)
1702 i.reloc[j] = NO_RELOC;
1703 memset (disp_expressions, '\0', sizeof (disp_expressions));
1704 memset (im_expressions, '\0', sizeof (im_expressions));
1705 save_stack_p = save_stack;
1707 /* First parse an instruction mnemonic & call i386_operand for the operands.
1708 We assume that the scrubber has arranged it so that line[0] is the valid
1709 start of a (possibly prefixed) mnemonic. */
1711 line = parse_insn (line, mnemonic);
1712 if (line == NULL)
1713 return;
1715 line = parse_operands (line, mnemonic);
1716 if (line == NULL)
1717 return;
1719 /* The order of the immediates should be reversed
1720 for 2 immediates extrq and insertq instructions */
1721 if ((i.imm_operands == 2)
1722 && ((strcmp (mnemonic, "extrq") == 0)
1723 || (strcmp (mnemonic, "insertq") == 0)))
1725 swap_2_operands (0, 1);
1726 /* "extrq" and insertq" are the only two instructions whose operands
1727 have to be reversed even though they have two immediate operands.
1729 if (intel_syntax)
1730 swap_operands ();
1733 /* Now we've parsed the mnemonic into a set of templates, and have the
1734 operands at hand. */
1736 /* All intel opcodes have reversed operands except for "bound" and
1737 "enter". We also don't reverse intersegment "jmp" and "call"
1738 instructions with 2 immediate operands so that the immediate segment
1739 precedes the offset, as it does when in AT&T mode. */
1740 if (intel_syntax
1741 && i.operands > 1
1742 && (strcmp (mnemonic, "bound") != 0)
1743 && (strcmp (mnemonic, "invlpga") != 0)
1744 && !((i.types[0] & Imm) && (i.types[1] & Imm)))
1745 swap_operands ();
1747 if (i.imm_operands)
1748 optimize_imm ();
1750 /* Don't optimize displacement for movabs since it only takes 64bit
1751 displacement. */
1752 if (i.disp_operands
1753 && (flag_code != CODE_64BIT
1754 || strcmp (mnemonic, "movabs") != 0))
1755 optimize_disp ();
1757 /* Next, we find a template that matches the given insn,
1758 making sure the overlap of the given operands types is consistent
1759 with the template operand types. */
1761 if (!match_template ())
1762 return;
1764 if (intel_syntax)
1766 /* Undo SYSV386_COMPAT brokenness when in Intel mode. See i386.h */
1767 if (SYSV386_COMPAT
1768 && (i.tm.base_opcode & 0xfffffde0) == 0xdce0)
1769 i.tm.base_opcode ^= FloatR;
1771 /* Zap movzx and movsx suffix. The suffix may have been set from
1772 "word ptr" or "byte ptr" on the source operand, but we'll use
1773 the suffix later to choose the destination register. */
1774 if ((i.tm.base_opcode & ~9) == 0x0fb6)
1776 if (i.reg_operands < 2
1777 && !i.suffix
1778 && (~i.tm.opcode_modifier
1779 & (No_bSuf
1780 | No_wSuf
1781 | No_lSuf
1782 | No_sSuf
1783 | No_xSuf
1784 | No_qSuf)))
1785 as_bad (_("ambiguous operand size for `%s'"), i.tm.name);
1787 i.suffix = 0;
1791 if (i.tm.opcode_modifier & FWait)
1792 if (!add_prefix (FWAIT_OPCODE))
1793 return;
1795 /* Check string instruction segment overrides. */
1796 if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0)
1798 if (!check_string ())
1799 return;
1802 if (!process_suffix ())
1803 return;
1805 /* Make still unresolved immediate matches conform to size of immediate
1806 given in i.suffix. */
1807 if (!finalize_imm ())
1808 return;
1810 if (i.types[0] & Imm1)
1811 i.imm_operands = 0; /* kludge for shift insns. */
1812 if (i.types[0] & ImplicitRegister)
1813 i.reg_operands--;
1814 if (i.types[1] & ImplicitRegister)
1815 i.reg_operands--;
1816 if (i.types[2] & ImplicitRegister)
1817 i.reg_operands--;
1819 if (i.tm.opcode_modifier & ImmExt)
1821 expressionS *exp;
1823 if ((i.tm.cpu_flags & CpuSSE3) && i.operands > 0)
1825 /* Streaming SIMD extensions 3 Instructions have the fixed
1826 operands with an opcode suffix which is coded in the same
1827 place as an 8-bit immediate field would be. Here we check
1828 those operands and remove them afterwards. */
1829 unsigned int x;
1831 for (x = 0; x < i.operands; x++)
1832 if (i.op[x].regs->reg_num != x)
1833 as_bad (_("can't use register '%%%s' as operand %d in '%s'."),
1834 i.op[x].regs->reg_name, x + 1, i.tm.name);
1835 i.operands = 0;
1838 /* These AMD 3DNow! and Intel Katmai New Instructions have an
1839 opcode suffix which is coded in the same place as an 8-bit
1840 immediate field would be. Here we fake an 8-bit immediate
1841 operand from the opcode suffix stored in tm.extension_opcode. */
1843 assert (i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS);
1845 exp = &im_expressions[i.imm_operands++];
1846 i.op[i.operands].imms = exp;
1847 i.types[i.operands++] = Imm8;
1848 exp->X_op = O_constant;
1849 exp->X_add_number = i.tm.extension_opcode;
1850 i.tm.extension_opcode = None;
1853 /* For insns with operands there are more diddles to do to the opcode. */
1854 if (i.operands)
1856 if (!process_operands ())
1857 return;
1859 else if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
1861 /* UnixWare fsub no args is alias for fsubp, fadd -> faddp, etc. */
1862 as_warn (_("translating to `%sp'"), i.tm.name);
1865 /* Handle conversion of 'int $3' --> special int3 insn. */
1866 if (i.tm.base_opcode == INT_OPCODE && i.op[0].imms->X_add_number == 3)
1868 i.tm.base_opcode = INT3_OPCODE;
1869 i.imm_operands = 0;
1872 if ((i.tm.opcode_modifier & (Jump | JumpByte | JumpDword))
1873 && i.op[0].disps->X_op == O_constant)
1875 /* Convert "jmp constant" (and "call constant") to a jump (call) to
1876 the absolute address given by the constant. Since ix86 jumps and
1877 calls are pc relative, we need to generate a reloc. */
1878 i.op[0].disps->X_add_symbol = &abs_symbol;
1879 i.op[0].disps->X_op = O_symbol;
1882 if ((i.tm.opcode_modifier & Rex64) != 0)
1883 i.rex |= REX_MODE64;
1885 /* For 8 bit registers we need an empty rex prefix. Also if the
1886 instruction already has a prefix, we need to convert old
1887 registers to new ones. */
1889 if (((i.types[0] & Reg8) != 0
1890 && (i.op[0].regs->reg_flags & RegRex64) != 0)
1891 || ((i.types[1] & Reg8) != 0
1892 && (i.op[1].regs->reg_flags & RegRex64) != 0)
1893 || (((i.types[0] & Reg8) != 0 || (i.types[1] & Reg8) != 0)
1894 && i.rex != 0))
1896 int x;
1898 i.rex |= REX_OPCODE;
1899 for (x = 0; x < 2; x++)
1901 /* Look for 8 bit operand that uses old registers. */
1902 if ((i.types[x] & Reg8) != 0
1903 && (i.op[x].regs->reg_flags & RegRex64) == 0)
1905 /* In case it is "hi" register, give up. */
1906 if (i.op[x].regs->reg_num > 3)
1907 as_bad (_("can't encode register '%%%s' in an "
1908 "instruction requiring REX prefix."),
1909 i.op[x].regs->reg_name);
1911 /* Otherwise it is equivalent to the extended register.
1912 Since the encoding doesn't change this is merely
1913 cosmetic cleanup for debug output. */
1915 i.op[x].regs = i.op[x].regs + 8;
1920 if (i.rex != 0)
1921 add_prefix (REX_OPCODE | i.rex);
1923 /* We are ready to output the insn. */
1924 output_insn ();
1927 static char *
1928 parse_insn (char *line, char *mnemonic)
1930 char *l = line;
1931 char *token_start = l;
1932 char *mnem_p;
1933 int supported;
1934 const template *t;
1936 /* Non-zero if we found a prefix only acceptable with string insns. */
1937 const char *expecting_string_instruction = NULL;
1939 while (1)
1941 mnem_p = mnemonic;
1942 while ((*mnem_p = mnemonic_chars[(unsigned char) *l]) != 0)
1944 mnem_p++;
1945 if (mnem_p >= mnemonic + MAX_MNEM_SIZE)
1947 as_bad (_("no such instruction: `%s'"), token_start);
1948 return NULL;
1950 l++;
1952 if (!is_space_char (*l)
1953 && *l != END_OF_INSN
1954 && (intel_syntax
1955 || (*l != PREFIX_SEPARATOR
1956 && *l != ',')))
1958 as_bad (_("invalid character %s in mnemonic"),
1959 output_invalid (*l));
1960 return NULL;
1962 if (token_start == l)
1964 if (!intel_syntax && *l == PREFIX_SEPARATOR)
1965 as_bad (_("expecting prefix; got nothing"));
1966 else
1967 as_bad (_("expecting mnemonic; got nothing"));
1968 return NULL;
1971 /* Look up instruction (or prefix) via hash table. */
1972 current_templates = hash_find (op_hash, mnemonic);
1974 if (*l != END_OF_INSN
1975 && (!is_space_char (*l) || l[1] != END_OF_INSN)
1976 && current_templates
1977 && (current_templates->start->opcode_modifier & IsPrefix))
1979 if (current_templates->start->cpu_flags
1980 & (flag_code != CODE_64BIT ? Cpu64 : CpuNo64))
1982 as_bad ((flag_code != CODE_64BIT
1983 ? _("`%s' is only supported in 64-bit mode")
1984 : _("`%s' is not supported in 64-bit mode")),
1985 current_templates->start->name);
1986 return NULL;
1988 /* If we are in 16-bit mode, do not allow addr16 or data16.
1989 Similarly, in 32-bit mode, do not allow addr32 or data32. */
1990 if ((current_templates->start->opcode_modifier & (Size16 | Size32))
1991 && flag_code != CODE_64BIT
1992 && (((current_templates->start->opcode_modifier & Size32) != 0)
1993 ^ (flag_code == CODE_16BIT)))
1995 as_bad (_("redundant %s prefix"),
1996 current_templates->start->name);
1997 return NULL;
1999 /* Add prefix, checking for repeated prefixes. */
2000 switch (add_prefix (current_templates->start->base_opcode))
2002 case 0:
2003 return NULL;
2004 case 2:
2005 expecting_string_instruction = current_templates->start->name;
2006 break;
2008 /* Skip past PREFIX_SEPARATOR and reset token_start. */
2009 token_start = ++l;
2011 else
2012 break;
2015 if (!current_templates)
2017 /* See if we can get a match by trimming off a suffix. */
2018 switch (mnem_p[-1])
2020 case WORD_MNEM_SUFFIX:
2021 if (intel_syntax && (intel_float_operand (mnemonic) & 2))
2022 i.suffix = SHORT_MNEM_SUFFIX;
2023 else
2024 case BYTE_MNEM_SUFFIX:
2025 case QWORD_MNEM_SUFFIX:
2026 i.suffix = mnem_p[-1];
2027 mnem_p[-1] = '\0';
2028 current_templates = hash_find (op_hash, mnemonic);
2029 break;
2030 case SHORT_MNEM_SUFFIX:
2031 case LONG_MNEM_SUFFIX:
2032 if (!intel_syntax)
2034 i.suffix = mnem_p[-1];
2035 mnem_p[-1] = '\0';
2036 current_templates = hash_find (op_hash, mnemonic);
2038 break;
2040 /* Intel Syntax. */
2041 case 'd':
2042 if (intel_syntax)
2044 if (intel_float_operand (mnemonic) == 1)
2045 i.suffix = SHORT_MNEM_SUFFIX;
2046 else
2047 i.suffix = LONG_MNEM_SUFFIX;
2048 mnem_p[-1] = '\0';
2049 current_templates = hash_find (op_hash, mnemonic);
2051 break;
2053 if (!current_templates)
2055 as_bad (_("no such instruction: `%s'"), token_start);
2056 return NULL;
2060 if (current_templates->start->opcode_modifier & (Jump | JumpByte))
2062 /* Check for a branch hint. We allow ",pt" and ",pn" for
2063 predict taken and predict not taken respectively.
2064 I'm not sure that branch hints actually do anything on loop
2065 and jcxz insns (JumpByte) for current Pentium4 chips. They
2066 may work in the future and it doesn't hurt to accept them
2067 now. */
2068 if (l[0] == ',' && l[1] == 'p')
2070 if (l[2] == 't')
2072 if (!add_prefix (DS_PREFIX_OPCODE))
2073 return NULL;
2074 l += 3;
2076 else if (l[2] == 'n')
2078 if (!add_prefix (CS_PREFIX_OPCODE))
2079 return NULL;
2080 l += 3;
2084 /* Any other comma loses. */
2085 if (*l == ',')
2087 as_bad (_("invalid character %s in mnemonic"),
2088 output_invalid (*l));
2089 return NULL;
2092 /* Check if instruction is supported on specified architecture. */
2093 supported = 0;
2094 for (t = current_templates->start; t < current_templates->end; ++t)
2096 if (!((t->cpu_flags & ~(Cpu64 | CpuNo64))
2097 & ~(cpu_arch_flags & ~(Cpu64 | CpuNo64))))
2098 supported |= 1;
2099 if (!(t->cpu_flags & (flag_code == CODE_64BIT ? CpuNo64 : Cpu64)))
2100 supported |= 2;
2102 if (!(supported & 2))
2104 as_bad (flag_code == CODE_64BIT
2105 ? _("`%s' is not supported in 64-bit mode")
2106 : _("`%s' is only supported in 64-bit mode"),
2107 current_templates->start->name);
2108 return NULL;
2110 if (!(supported & 1))
2112 as_warn (_("`%s' is not supported on `%s%s'"),
2113 current_templates->start->name,
2114 cpu_arch_name,
2115 cpu_sub_arch_name ? cpu_sub_arch_name : "");
2117 else if ((Cpu386 & ~cpu_arch_flags) && (flag_code != CODE_16BIT))
2119 as_warn (_("use .code16 to ensure correct addressing mode"));
2122 /* Check for rep/repne without a string instruction. */
2123 if (expecting_string_instruction)
2125 static templates override;
2127 for (t = current_templates->start; t < current_templates->end; ++t)
2128 if (t->opcode_modifier & IsString)
2129 break;
2130 if (t >= current_templates->end)
2132 as_bad (_("expecting string instruction after `%s'"),
2133 expecting_string_instruction);
2134 return NULL;
2136 for (override.start = t; t < current_templates->end; ++t)
2137 if (!(t->opcode_modifier & IsString))
2138 break;
2139 override.end = t;
2140 current_templates = &override;
2143 return l;
2146 static char *
2147 parse_operands (char *l, const char *mnemonic)
2149 char *token_start;
2151 /* 1 if operand is pending after ','. */
2152 unsigned int expecting_operand = 0;
2154 /* Non-zero if operand parens not balanced. */
2155 unsigned int paren_not_balanced;
2157 while (*l != END_OF_INSN)
2159 /* Skip optional white space before operand. */
2160 if (is_space_char (*l))
2161 ++l;
2162 if (!is_operand_char (*l) && *l != END_OF_INSN)
2164 as_bad (_("invalid character %s before operand %d"),
2165 output_invalid (*l),
2166 i.operands + 1);
2167 return NULL;
2169 token_start = l; /* after white space */
2170 paren_not_balanced = 0;
2171 while (paren_not_balanced || *l != ',')
2173 if (*l == END_OF_INSN)
2175 if (paren_not_balanced)
2177 if (!intel_syntax)
2178 as_bad (_("unbalanced parenthesis in operand %d."),
2179 i.operands + 1);
2180 else
2181 as_bad (_("unbalanced brackets in operand %d."),
2182 i.operands + 1);
2183 return NULL;
2185 else
2186 break; /* we are done */
2188 else if (!is_operand_char (*l) && !is_space_char (*l))
2190 as_bad (_("invalid character %s in operand %d"),
2191 output_invalid (*l),
2192 i.operands + 1);
2193 return NULL;
2195 if (!intel_syntax)
2197 if (*l == '(')
2198 ++paren_not_balanced;
2199 if (*l == ')')
2200 --paren_not_balanced;
2202 else
2204 if (*l == '[')
2205 ++paren_not_balanced;
2206 if (*l == ']')
2207 --paren_not_balanced;
2209 l++;
2211 if (l != token_start)
2212 { /* Yes, we've read in another operand. */
2213 unsigned int operand_ok;
2214 this_operand = i.operands++;
2215 if (i.operands > MAX_OPERANDS)
2217 as_bad (_("spurious operands; (%d operands/instruction max)"),
2218 MAX_OPERANDS);
2219 return NULL;
2221 /* Now parse operand adding info to 'i' as we go along. */
2222 END_STRING_AND_SAVE (l);
2224 if (intel_syntax)
2225 operand_ok =
2226 i386_intel_operand (token_start,
2227 intel_float_operand (mnemonic));
2228 else
2229 operand_ok = i386_operand (token_start);
2231 RESTORE_END_STRING (l);
2232 if (!operand_ok)
2233 return NULL;
2235 else
2237 if (expecting_operand)
2239 expecting_operand_after_comma:
2240 as_bad (_("expecting operand after ','; got nothing"));
2241 return NULL;
2243 if (*l == ',')
2245 as_bad (_("expecting operand before ','; got nothing"));
2246 return NULL;
2250 /* Now *l must be either ',' or END_OF_INSN. */
2251 if (*l == ',')
2253 if (*++l == END_OF_INSN)
2255 /* Just skip it, if it's \n complain. */
2256 goto expecting_operand_after_comma;
2258 expecting_operand = 1;
2261 return l;
2264 static void
2265 swap_2_operands (int xchg1, int xchg2)
2267 union i386_op temp_op;
2268 unsigned int temp_type;
2269 enum bfd_reloc_code_real temp_reloc;
2271 temp_type = i.types[xchg2];
2272 i.types[xchg2] = i.types[xchg1];
2273 i.types[xchg1] = temp_type;
2274 temp_op = i.op[xchg2];
2275 i.op[xchg2] = i.op[xchg1];
2276 i.op[xchg1] = temp_op;
2277 temp_reloc = i.reloc[xchg2];
2278 i.reloc[xchg2] = i.reloc[xchg1];
2279 i.reloc[xchg1] = temp_reloc;
2282 static void
2283 swap_operands (void)
2285 switch (i.operands)
2287 case 4:
2288 swap_2_operands (1, i.operands - 2);
2289 case 3:
2290 case 2:
2291 swap_2_operands (0, i.operands - 1);
2292 break;
2293 default:
2294 abort ();
2297 if (i.mem_operands == 2)
2299 const seg_entry *temp_seg;
2300 temp_seg = i.seg[0];
2301 i.seg[0] = i.seg[1];
2302 i.seg[1] = temp_seg;
2306 /* Try to ensure constant immediates are represented in the smallest
2307 opcode possible. */
2308 static void
2309 optimize_imm (void)
2311 char guess_suffix = 0;
2312 int op;
2314 if (i.suffix)
2315 guess_suffix = i.suffix;
2316 else if (i.reg_operands)
2318 /* Figure out a suffix from the last register operand specified.
2319 We can't do this properly yet, ie. excluding InOutPortReg,
2320 but the following works for instructions with immediates.
2321 In any case, we can't set i.suffix yet. */
2322 for (op = i.operands; --op >= 0;)
2323 if (i.types[op] & Reg)
2325 if (i.types[op] & Reg8)
2326 guess_suffix = BYTE_MNEM_SUFFIX;
2327 else if (i.types[op] & Reg16)
2328 guess_suffix = WORD_MNEM_SUFFIX;
2329 else if (i.types[op] & Reg32)
2330 guess_suffix = LONG_MNEM_SUFFIX;
2331 else if (i.types[op] & Reg64)
2332 guess_suffix = QWORD_MNEM_SUFFIX;
2333 break;
2336 else if ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0))
2337 guess_suffix = WORD_MNEM_SUFFIX;
2339 for (op = i.operands; --op >= 0;)
2340 if (i.types[op] & Imm)
2342 switch (i.op[op].imms->X_op)
2344 case O_constant:
2345 /* If a suffix is given, this operand may be shortened. */
2346 switch (guess_suffix)
2348 case LONG_MNEM_SUFFIX:
2349 i.types[op] |= Imm32 | Imm64;
2350 break;
2351 case WORD_MNEM_SUFFIX:
2352 i.types[op] |= Imm16 | Imm32S | Imm32 | Imm64;
2353 break;
2354 case BYTE_MNEM_SUFFIX:
2355 i.types[op] |= Imm16 | Imm8 | Imm8S | Imm32S | Imm32 | Imm64;
2356 break;
2359 /* If this operand is at most 16 bits, convert it
2360 to a signed 16 bit number before trying to see
2361 whether it will fit in an even smaller size.
2362 This allows a 16-bit operand such as $0xffe0 to
2363 be recognised as within Imm8S range. */
2364 if ((i.types[op] & Imm16)
2365 && (i.op[op].imms->X_add_number & ~(offsetT) 0xffff) == 0)
2367 i.op[op].imms->X_add_number =
2368 (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000);
2370 if ((i.types[op] & Imm32)
2371 && ((i.op[op].imms->X_add_number & ~(((offsetT) 2 << 31) - 1))
2372 == 0))
2374 i.op[op].imms->X_add_number = ((i.op[op].imms->X_add_number
2375 ^ ((offsetT) 1 << 31))
2376 - ((offsetT) 1 << 31));
2378 i.types[op] |= smallest_imm_type (i.op[op].imms->X_add_number);
2380 /* We must avoid matching of Imm32 templates when 64bit
2381 only immediate is available. */
2382 if (guess_suffix == QWORD_MNEM_SUFFIX)
2383 i.types[op] &= ~Imm32;
2384 break;
2386 case O_absent:
2387 case O_register:
2388 abort ();
2390 /* Symbols and expressions. */
2391 default:
2392 /* Convert symbolic operand to proper sizes for matching, but don't
2393 prevent matching a set of insns that only supports sizes other
2394 than those matching the insn suffix. */
2396 unsigned int mask, allowed = 0;
2397 const template *t;
2399 for (t = current_templates->start;
2400 t < current_templates->end;
2401 ++t)
2402 allowed |= t->operand_types[op];
2403 switch (guess_suffix)
2405 case QWORD_MNEM_SUFFIX:
2406 mask = Imm64 | Imm32S;
2407 break;
2408 case LONG_MNEM_SUFFIX:
2409 mask = Imm32;
2410 break;
2411 case WORD_MNEM_SUFFIX:
2412 mask = Imm16;
2413 break;
2414 case BYTE_MNEM_SUFFIX:
2415 mask = Imm8;
2416 break;
2417 default:
2418 mask = 0;
2419 break;
2421 if (mask & allowed)
2422 i.types[op] &= mask;
2424 break;
2429 /* Try to use the smallest displacement type too. */
2430 static void
2431 optimize_disp (void)
2433 int op;
2435 for (op = i.operands; --op >= 0;)
2436 if (i.types[op] & Disp)
2438 if (i.op[op].disps->X_op == O_constant)
2440 offsetT disp = i.op[op].disps->X_add_number;
2442 if ((i.types[op] & Disp16)
2443 && (disp & ~(offsetT) 0xffff) == 0)
2445 /* If this operand is at most 16 bits, convert
2446 to a signed 16 bit number and don't use 64bit
2447 displacement. */
2448 disp = (((disp & 0xffff) ^ 0x8000) - 0x8000);
2449 i.types[op] &= ~Disp64;
2451 if ((i.types[op] & Disp32)
2452 && (disp & ~(((offsetT) 2 << 31) - 1)) == 0)
2454 /* If this operand is at most 32 bits, convert
2455 to a signed 32 bit number and don't use 64bit
2456 displacement. */
2457 disp &= (((offsetT) 2 << 31) - 1);
2458 disp = (disp ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
2459 i.types[op] &= ~Disp64;
2461 if (!disp && (i.types[op] & BaseIndex))
2463 i.types[op] &= ~Disp;
2464 i.op[op].disps = 0;
2465 i.disp_operands--;
2467 else if (flag_code == CODE_64BIT)
2469 if (fits_in_signed_long (disp))
2471 i.types[op] &= ~Disp64;
2472 i.types[op] |= Disp32S;
2474 if (fits_in_unsigned_long (disp))
2475 i.types[op] |= Disp32;
2477 if ((i.types[op] & (Disp32 | Disp32S | Disp16))
2478 && fits_in_signed_byte (disp))
2479 i.types[op] |= Disp8;
2481 else if (i.reloc[op] == BFD_RELOC_386_TLS_DESC_CALL
2482 || i.reloc[op] == BFD_RELOC_X86_64_TLSDESC_CALL)
2484 fix_new_exp (frag_now, frag_more (0) - frag_now->fr_literal, 0,
2485 i.op[op].disps, 0, i.reloc[op]);
2486 i.types[op] &= ~Disp;
2488 else
2489 /* We only support 64bit displacement on constants. */
2490 i.types[op] &= ~Disp64;
2494 static int
2495 match_template (void)
2497 /* Points to template once we've found it. */
2498 const template *t;
2499 unsigned int overlap0, overlap1, overlap2, overlap3;
2500 unsigned int found_reverse_match;
2501 int suffix_check;
2502 unsigned int operand_types [MAX_OPERANDS];
2503 int addr_prefix_disp;
2504 unsigned int j;
2506 #if MAX_OPERANDS != 4
2507 # error "MAX_OPERANDS must be 4."
2508 #endif
2510 #define MATCH(overlap, given, template) \
2511 ((overlap & ~JumpAbsolute) \
2512 && (((given) & (BaseIndex | JumpAbsolute)) \
2513 == ((overlap) & (BaseIndex | JumpAbsolute))))
2515 /* If given types r0 and r1 are registers they must be of the same type
2516 unless the expected operand type register overlap is null.
2517 Note that Acc in a template matches every size of reg. */
2518 #define CONSISTENT_REGISTER_MATCH(m0, g0, t0, m1, g1, t1) \
2519 (((g0) & Reg) == 0 || ((g1) & Reg) == 0 \
2520 || ((g0) & Reg) == ((g1) & Reg) \
2521 || ((((m0) & Acc) ? Reg : (t0)) & (((m1) & Acc) ? Reg : (t1)) & Reg) == 0 )
2523 overlap0 = 0;
2524 overlap1 = 0;
2525 overlap2 = 0;
2526 overlap3 = 0;
2527 found_reverse_match = 0;
2528 for (j = 0; j < MAX_OPERANDS; j++)
2529 operand_types [j] = 0;
2530 addr_prefix_disp = -1;
2531 suffix_check = (i.suffix == BYTE_MNEM_SUFFIX
2532 ? No_bSuf
2533 : (i.suffix == WORD_MNEM_SUFFIX
2534 ? No_wSuf
2535 : (i.suffix == SHORT_MNEM_SUFFIX
2536 ? No_sSuf
2537 : (i.suffix == LONG_MNEM_SUFFIX
2538 ? No_lSuf
2539 : (i.suffix == QWORD_MNEM_SUFFIX
2540 ? No_qSuf
2541 : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX
2542 ? No_xSuf : 0))))));
2544 for (t = current_templates->start; t < current_templates->end; t++)
2546 addr_prefix_disp = -1;
2548 /* Must have right number of operands. */
2549 if (i.operands != t->operands)
2550 continue;
2552 /* Check the suffix, except for some instructions in intel mode. */
2553 if ((t->opcode_modifier & suffix_check)
2554 && !(intel_syntax
2555 && (t->opcode_modifier & IgnoreSize)))
2556 continue;
2558 for (j = 0; j < MAX_OPERANDS; j++)
2559 operand_types [j] = t->operand_types [j];
2561 /* In general, don't allow 64-bit operands in 32-bit mode. */
2562 if (i.suffix == QWORD_MNEM_SUFFIX
2563 && flag_code != CODE_64BIT
2564 && (intel_syntax
2565 ? (!(t->opcode_modifier & IgnoreSize)
2566 && !intel_float_operand (t->name))
2567 : intel_float_operand (t->name) != 2)
2568 && (!(operand_types[0] & (RegMMX | RegXMM))
2569 || !(operand_types[t->operands > 1] & (RegMMX | RegXMM)))
2570 && (t->base_opcode != 0x0fc7
2571 || t->extension_opcode != 1 /* cmpxchg8b */))
2572 continue;
2574 /* Do not verify operands when there are none. */
2575 else if (!t->operands)
2577 if (t->cpu_flags & ~cpu_arch_flags)
2578 continue;
2579 /* We've found a match; break out of loop. */
2580 break;
2583 /* Address size prefix will turn Disp64/Disp32/Disp16 operand
2584 into Disp32/Disp16/Disp32 operand. */
2585 if (i.prefix[ADDR_PREFIX] != 0)
2587 unsigned int DispOn = 0, DispOff = 0;
2589 switch (flag_code)
2591 case CODE_16BIT:
2592 DispOn = Disp32;
2593 DispOff = Disp16;
2594 break;
2595 case CODE_32BIT:
2596 DispOn = Disp16;
2597 DispOff = Disp32;
2598 break;
2599 case CODE_64BIT:
2600 DispOn = Disp32;
2601 DispOff = Disp64;
2602 break;
2605 for (j = 0; j < MAX_OPERANDS; j++)
2607 /* There should be only one Disp operand. */
2608 if ((operand_types[j] & DispOff))
2610 addr_prefix_disp = j;
2611 operand_types[j] |= DispOn;
2612 operand_types[j] &= ~DispOff;
2613 break;
2618 overlap0 = i.types[0] & operand_types[0];
2619 switch (t->operands)
2621 case 1:
2622 if (!MATCH (overlap0, i.types[0], operand_types[0]))
2623 continue;
2624 break;
2625 case 2:
2626 case 3:
2627 case 4:
2628 overlap1 = i.types[1] & operand_types[1];
2629 if (!MATCH (overlap0, i.types[0], operand_types[0])
2630 || !MATCH (overlap1, i.types[1], operand_types[1])
2631 /* monitor in SSE3 is a very special case. The first
2632 register and the second register may have different
2633 sizes. */
2634 || !((t->base_opcode == 0x0f01
2635 && t->extension_opcode == 0xc8)
2636 || CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2637 operand_types[0],
2638 overlap1, i.types[1],
2639 operand_types[1])))
2641 /* Check if other direction is valid ... */
2642 if ((t->opcode_modifier & (D | FloatD)) == 0)
2643 continue;
2645 /* Try reversing direction of operands. */
2646 overlap0 = i.types[0] & operand_types[1];
2647 overlap1 = i.types[1] & operand_types[0];
2648 if (!MATCH (overlap0, i.types[0], operand_types[1])
2649 || !MATCH (overlap1, i.types[1], operand_types[0])
2650 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2651 operand_types[1],
2652 overlap1, i.types[1],
2653 operand_types[0]))
2655 /* Does not match either direction. */
2656 continue;
2658 /* found_reverse_match holds which of D or FloatDR
2659 we've found. */
2660 found_reverse_match = t->opcode_modifier & (D | FloatDR);
2662 else
2664 /* Found a forward 2 operand match here. */
2665 switch (t->operands)
2667 case 4:
2668 overlap3 = i.types[3] & operand_types[3];
2669 case 3:
2670 overlap2 = i.types[2] & operand_types[2];
2671 break;
2674 switch (t->operands)
2676 case 4:
2677 if (!MATCH (overlap3, i.types[3], operand_types[3])
2678 || !CONSISTENT_REGISTER_MATCH (overlap2,
2679 i.types[2],
2680 operand_types[2],
2681 overlap3,
2682 i.types[3],
2683 operand_types[3]))
2684 continue;
2685 case 3:
2686 /* Here we make use of the fact that there are no
2687 reverse match 3 operand instructions, and all 3
2688 operand instructions only need to be checked for
2689 register consistency between operands 2 and 3. */
2690 if (!MATCH (overlap2, i.types[2], operand_types[2])
2691 || !CONSISTENT_REGISTER_MATCH (overlap1,
2692 i.types[1],
2693 operand_types[1],
2694 overlap2,
2695 i.types[2],
2696 operand_types[2]))
2697 continue;
2698 break;
2701 /* Found either forward/reverse 2, 3 or 4 operand match here:
2702 slip through to break. */
2704 if (t->cpu_flags & ~cpu_arch_flags)
2706 found_reverse_match = 0;
2707 continue;
2709 /* We've found a match; break out of loop. */
2710 break;
2713 if (t == current_templates->end)
2715 /* We found no match. */
2716 as_bad (_("suffix or operands invalid for `%s'"),
2717 current_templates->start->name);
2718 return 0;
2721 if (!quiet_warnings)
2723 if (!intel_syntax
2724 && ((i.types[0] & JumpAbsolute)
2725 != (operand_types[0] & JumpAbsolute)))
2727 as_warn (_("indirect %s without `*'"), t->name);
2730 if ((t->opcode_modifier & (IsPrefix | IgnoreSize))
2731 == (IsPrefix | IgnoreSize))
2733 /* Warn them that a data or address size prefix doesn't
2734 affect assembly of the next line of code. */
2735 as_warn (_("stand-alone `%s' prefix"), t->name);
2739 /* Copy the template we found. */
2740 i.tm = *t;
2742 if (addr_prefix_disp != -1)
2743 i.tm.operand_types[addr_prefix_disp]
2744 = operand_types[addr_prefix_disp];
2746 if (found_reverse_match)
2748 /* If we found a reverse match we must alter the opcode
2749 direction bit. found_reverse_match holds bits to change
2750 (different for int & float insns). */
2752 i.tm.base_opcode ^= found_reverse_match;
2754 i.tm.operand_types[0] = operand_types[1];
2755 i.tm.operand_types[1] = operand_types[0];
2758 return 1;
2761 static int
2762 check_string (void)
2764 int mem_op = (i.types[0] & AnyMem) ? 0 : 1;
2765 if ((i.tm.operand_types[mem_op] & EsSeg) != 0)
2767 if (i.seg[0] != NULL && i.seg[0] != &es)
2769 as_bad (_("`%s' operand %d must use `%%es' segment"),
2770 i.tm.name,
2771 mem_op + 1);
2772 return 0;
2774 /* There's only ever one segment override allowed per instruction.
2775 This instruction possibly has a legal segment override on the
2776 second operand, so copy the segment to where non-string
2777 instructions store it, allowing common code. */
2778 i.seg[0] = i.seg[1];
2780 else if ((i.tm.operand_types[mem_op + 1] & EsSeg) != 0)
2782 if (i.seg[1] != NULL && i.seg[1] != &es)
2784 as_bad (_("`%s' operand %d must use `%%es' segment"),
2785 i.tm.name,
2786 mem_op + 2);
2787 return 0;
2790 return 1;
2793 static int
2794 process_suffix (void)
2796 /* If matched instruction specifies an explicit instruction mnemonic
2797 suffix, use it. */
2798 if (i.tm.opcode_modifier & (Size16 | Size32 | Size64))
2800 if (i.tm.opcode_modifier & Size16)
2801 i.suffix = WORD_MNEM_SUFFIX;
2802 else if (i.tm.opcode_modifier & Size64)
2803 i.suffix = QWORD_MNEM_SUFFIX;
2804 else
2805 i.suffix = LONG_MNEM_SUFFIX;
2807 else if (i.reg_operands)
2809 /* If there's no instruction mnemonic suffix we try to invent one
2810 based on register operands. */
2811 if (!i.suffix)
2813 /* We take i.suffix from the last register operand specified,
2814 Destination register type is more significant than source
2815 register type. */
2816 int op;
2818 for (op = i.operands; --op >= 0;)
2819 if ((i.types[op] & Reg)
2820 && !(i.tm.operand_types[op] & InOutPortReg))
2822 i.suffix = ((i.types[op] & Reg8) ? BYTE_MNEM_SUFFIX :
2823 (i.types[op] & Reg16) ? WORD_MNEM_SUFFIX :
2824 (i.types[op] & Reg64) ? QWORD_MNEM_SUFFIX :
2825 LONG_MNEM_SUFFIX);
2826 break;
2829 else if (i.suffix == BYTE_MNEM_SUFFIX)
2831 if (!check_byte_reg ())
2832 return 0;
2834 else if (i.suffix == LONG_MNEM_SUFFIX)
2836 if (!check_long_reg ())
2837 return 0;
2839 else if (i.suffix == QWORD_MNEM_SUFFIX)
2841 if (!check_qword_reg ())
2842 return 0;
2844 else if (i.suffix == WORD_MNEM_SUFFIX)
2846 if (!check_word_reg ())
2847 return 0;
2849 else if (intel_syntax && (i.tm.opcode_modifier & IgnoreSize))
2850 /* Do nothing if the instruction is going to ignore the prefix. */
2852 else
2853 abort ();
2855 else if ((i.tm.opcode_modifier & DefaultSize)
2856 && !i.suffix
2857 /* exclude fldenv/frstor/fsave/fstenv */
2858 && (i.tm.opcode_modifier & No_sSuf))
2860 i.suffix = stackop_size;
2862 else if (intel_syntax
2863 && !i.suffix
2864 && ((i.tm.operand_types[0] & JumpAbsolute)
2865 || (i.tm.opcode_modifier & (JumpByte|JumpInterSegment))
2866 || (i.tm.base_opcode == 0x0f01 /* [ls][gi]dt */
2867 && i.tm.extension_opcode <= 3)))
2869 switch (flag_code)
2871 case CODE_64BIT:
2872 if (!(i.tm.opcode_modifier & No_qSuf))
2874 i.suffix = QWORD_MNEM_SUFFIX;
2875 break;
2877 case CODE_32BIT:
2878 if (!(i.tm.opcode_modifier & No_lSuf))
2879 i.suffix = LONG_MNEM_SUFFIX;
2880 break;
2881 case CODE_16BIT:
2882 if (!(i.tm.opcode_modifier & No_wSuf))
2883 i.suffix = WORD_MNEM_SUFFIX;
2884 break;
2888 if (!i.suffix)
2890 if (!intel_syntax)
2892 if (i.tm.opcode_modifier & W)
2894 as_bad (_("no instruction mnemonic suffix given and "
2895 "no register operands; can't size instruction"));
2896 return 0;
2899 else
2901 unsigned int suffixes = (~i.tm.opcode_modifier
2902 & (No_bSuf
2903 | No_wSuf
2904 | No_lSuf
2905 | No_sSuf
2906 | No_xSuf
2907 | No_qSuf));
2909 if ((i.tm.opcode_modifier & W)
2910 || ((suffixes & (suffixes - 1))
2911 && !(i.tm.opcode_modifier & (DefaultSize | IgnoreSize))))
2913 as_bad (_("ambiguous operand size for `%s'"), i.tm.name);
2914 return 0;
2919 /* Change the opcode based on the operand size given by i.suffix;
2920 We don't need to change things for byte insns. */
2922 if (i.suffix && i.suffix != BYTE_MNEM_SUFFIX)
2924 /* It's not a byte, select word/dword operation. */
2925 if (i.tm.opcode_modifier & W)
2927 if (i.tm.opcode_modifier & ShortForm)
2928 i.tm.base_opcode |= 8;
2929 else
2930 i.tm.base_opcode |= 1;
2933 /* Now select between word & dword operations via the operand
2934 size prefix, except for instructions that will ignore this
2935 prefix anyway. */
2936 if (i.tm.base_opcode == 0x0f01 && i.tm.extension_opcode == 0xc8)
2938 /* monitor in SSE3 is a very special case. The default size
2939 of AX is the size of mode. The address size override
2940 prefix will change the size of AX. */
2941 if (i.op->regs[0].reg_type &
2942 (flag_code == CODE_32BIT ? Reg16 : Reg32))
2943 if (!add_prefix (ADDR_PREFIX_OPCODE))
2944 return 0;
2946 else if (i.suffix != QWORD_MNEM_SUFFIX
2947 && i.suffix != LONG_DOUBLE_MNEM_SUFFIX
2948 && !(i.tm.opcode_modifier & (IgnoreSize | FloatMF))
2949 && ((i.suffix == LONG_MNEM_SUFFIX) == (flag_code == CODE_16BIT)
2950 || (flag_code == CODE_64BIT
2951 && (i.tm.opcode_modifier & JumpByte))))
2953 unsigned int prefix = DATA_PREFIX_OPCODE;
2955 if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */
2956 prefix = ADDR_PREFIX_OPCODE;
2958 if (!add_prefix (prefix))
2959 return 0;
2962 /* Set mode64 for an operand. */
2963 if (i.suffix == QWORD_MNEM_SUFFIX
2964 && flag_code == CODE_64BIT
2965 && (i.tm.opcode_modifier & NoRex64) == 0)
2967 /* Special case for xchg %rax,%rax. It is NOP and doesn't
2968 need rex64. */
2969 if (i.operands != 2
2970 || i.types [0] != (Acc | Reg64)
2971 || i.types [1] != (Acc | Reg64)
2972 || strcmp (i.tm.name, "xchg") != 0)
2973 i.rex |= REX_MODE64;
2976 /* Size floating point instruction. */
2977 if (i.suffix == LONG_MNEM_SUFFIX)
2978 if (i.tm.opcode_modifier & FloatMF)
2979 i.tm.base_opcode ^= 4;
2982 return 1;
2985 static int
2986 check_byte_reg (void)
2988 int op;
2990 for (op = i.operands; --op >= 0;)
2992 /* If this is an eight bit register, it's OK. If it's the 16 or
2993 32 bit version of an eight bit register, we will just use the
2994 low portion, and that's OK too. */
2995 if (i.types[op] & Reg8)
2996 continue;
2998 /* movzx and movsx should not generate this warning. */
2999 if (intel_syntax
3000 && (i.tm.base_opcode == 0xfb7
3001 || i.tm.base_opcode == 0xfb6
3002 || i.tm.base_opcode == 0x63
3003 || i.tm.base_opcode == 0xfbe
3004 || i.tm.base_opcode == 0xfbf))
3005 continue;
3007 if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4)
3009 /* Prohibit these changes in the 64bit mode, since the
3010 lowering is more complicated. */
3011 if (flag_code == CODE_64BIT
3012 && (i.tm.operand_types[op] & InOutPortReg) == 0)
3014 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3015 register_prefix, i.op[op].regs->reg_name,
3016 i.suffix);
3017 return 0;
3019 #if REGISTER_WARNINGS
3020 if (!quiet_warnings
3021 && (i.tm.operand_types[op] & InOutPortReg) == 0)
3022 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
3023 (i.op[op].regs + (i.types[op] & Reg16
3024 ? REGNAM_AL - REGNAM_AX
3025 : REGNAM_AL - REGNAM_EAX))->reg_name,
3026 i.op[op].regs->reg_name,
3027 i.suffix);
3028 #endif
3029 continue;
3031 /* Any other register is bad. */
3032 if (i.types[op] & (Reg | RegMMX | RegXMM
3033 | SReg2 | SReg3
3034 | Control | Debug | Test
3035 | FloatReg | FloatAcc))
3037 as_bad (_("`%%%s' not allowed with `%s%c'"),
3038 i.op[op].regs->reg_name,
3039 i.tm.name,
3040 i.suffix);
3041 return 0;
3044 return 1;
3047 static int
3048 check_long_reg (void)
3050 int op;
3052 for (op = i.operands; --op >= 0;)
3053 /* Reject eight bit registers, except where the template requires
3054 them. (eg. movzb) */
3055 if ((i.types[op] & Reg8) != 0
3056 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
3058 as_bad (_("`%%%s' not allowed with `%s%c'"),
3059 i.op[op].regs->reg_name,
3060 i.tm.name,
3061 i.suffix);
3062 return 0;
3064 /* Warn if the e prefix on a general reg is missing. */
3065 else if ((!quiet_warnings || flag_code == CODE_64BIT)
3066 && (i.types[op] & Reg16) != 0
3067 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
3069 /* Prohibit these changes in the 64bit mode, since the
3070 lowering is more complicated. */
3071 if (flag_code == CODE_64BIT)
3073 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3074 register_prefix, i.op[op].regs->reg_name,
3075 i.suffix);
3076 return 0;
3078 #if REGISTER_WARNINGS
3079 else
3080 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
3081 (i.op[op].regs + REGNAM_EAX - REGNAM_AX)->reg_name,
3082 i.op[op].regs->reg_name,
3083 i.suffix);
3084 #endif
3086 /* Warn if the r prefix on a general reg is missing. */
3087 else if ((i.types[op] & Reg64) != 0
3088 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
3090 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3091 register_prefix, i.op[op].regs->reg_name,
3092 i.suffix);
3093 return 0;
3095 return 1;
3098 static int
3099 check_qword_reg (void)
3101 int op;
3103 for (op = i.operands; --op >= 0; )
3104 /* Reject eight bit registers, except where the template requires
3105 them. (eg. movzb) */
3106 if ((i.types[op] & Reg8) != 0
3107 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
3109 as_bad (_("`%%%s' not allowed with `%s%c'"),
3110 i.op[op].regs->reg_name,
3111 i.tm.name,
3112 i.suffix);
3113 return 0;
3115 /* Warn if the e prefix on a general reg is missing. */
3116 else if (((i.types[op] & Reg16) != 0
3117 || (i.types[op] & Reg32) != 0)
3118 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
3120 /* Prohibit these changes in the 64bit mode, since the
3121 lowering is more complicated. */
3122 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3123 register_prefix, i.op[op].regs->reg_name,
3124 i.suffix);
3125 return 0;
3127 return 1;
3130 static int
3131 check_word_reg (void)
3133 int op;
3134 for (op = i.operands; --op >= 0;)
3135 /* Reject eight bit registers, except where the template requires
3136 them. (eg. movzb) */
3137 if ((i.types[op] & Reg8) != 0
3138 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
3140 as_bad (_("`%%%s' not allowed with `%s%c'"),
3141 i.op[op].regs->reg_name,
3142 i.tm.name,
3143 i.suffix);
3144 return 0;
3146 /* Warn if the e prefix on a general reg is present. */
3147 else if ((!quiet_warnings || flag_code == CODE_64BIT)
3148 && (i.types[op] & Reg32) != 0
3149 && (i.tm.operand_types[op] & (Reg16 | Acc)) != 0)
3151 /* Prohibit these changes in the 64bit mode, since the
3152 lowering is more complicated. */
3153 if (flag_code == CODE_64BIT)
3155 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3156 register_prefix, i.op[op].regs->reg_name,
3157 i.suffix);
3158 return 0;
3160 else
3161 #if REGISTER_WARNINGS
3162 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
3163 (i.op[op].regs + REGNAM_AX - REGNAM_EAX)->reg_name,
3164 i.op[op].regs->reg_name,
3165 i.suffix);
3166 #endif
3168 return 1;
3171 static int
3172 finalize_imm (void)
3174 unsigned int overlap0, overlap1, overlap2;
3176 overlap0 = i.types[0] & i.tm.operand_types[0];
3177 if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64))
3178 && overlap0 != Imm8 && overlap0 != Imm8S
3179 && overlap0 != Imm16 && overlap0 != Imm32S
3180 && overlap0 != Imm32 && overlap0 != Imm64)
3182 if (i.suffix)
3184 overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX
3185 ? Imm8 | Imm8S
3186 : (i.suffix == WORD_MNEM_SUFFIX
3187 ? Imm16
3188 : (i.suffix == QWORD_MNEM_SUFFIX
3189 ? Imm64 | Imm32S
3190 : Imm32)));
3192 else if (overlap0 == (Imm16 | Imm32S | Imm32)
3193 || overlap0 == (Imm16 | Imm32)
3194 || overlap0 == (Imm16 | Imm32S))
3196 overlap0 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
3197 ? Imm16 : Imm32S);
3199 if (overlap0 != Imm8 && overlap0 != Imm8S
3200 && overlap0 != Imm16 && overlap0 != Imm32S
3201 && overlap0 != Imm32 && overlap0 != Imm64)
3203 as_bad (_("no instruction mnemonic suffix given; "
3204 "can't determine immediate size"));
3205 return 0;
3208 i.types[0] = overlap0;
3210 overlap1 = i.types[1] & i.tm.operand_types[1];
3211 if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32 | Imm64))
3212 && overlap1 != Imm8 && overlap1 != Imm8S
3213 && overlap1 != Imm16 && overlap1 != Imm32S
3214 && overlap1 != Imm32 && overlap1 != Imm64)
3216 if (i.suffix)
3218 overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX
3219 ? Imm8 | Imm8S
3220 : (i.suffix == WORD_MNEM_SUFFIX
3221 ? Imm16
3222 : (i.suffix == QWORD_MNEM_SUFFIX
3223 ? Imm64 | Imm32S
3224 : Imm32)));
3226 else if (overlap1 == (Imm16 | Imm32 | Imm32S)
3227 || overlap1 == (Imm16 | Imm32)
3228 || overlap1 == (Imm16 | Imm32S))
3230 overlap1 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
3231 ? Imm16 : Imm32S);
3233 if (overlap1 != Imm8 && overlap1 != Imm8S
3234 && overlap1 != Imm16 && overlap1 != Imm32S
3235 && overlap1 != Imm32 && overlap1 != Imm64)
3237 as_bad (_("no instruction mnemonic suffix given; "
3238 "can't determine immediate size %x %c"),
3239 overlap1, i.suffix);
3240 return 0;
3243 i.types[1] = overlap1;
3245 overlap2 = i.types[2] & i.tm.operand_types[2];
3246 assert ((overlap2 & Imm) == 0);
3247 i.types[2] = overlap2;
3249 return 1;
3252 static int
3253 process_operands (void)
3255 /* Default segment register this instruction will use for memory
3256 accesses. 0 means unknown. This is only for optimizing out
3257 unnecessary segment overrides. */
3258 const seg_entry *default_seg = 0;
3260 /* The imul $imm, %reg instruction is converted into
3261 imul $imm, %reg, %reg, and the clr %reg instruction
3262 is converted into xor %reg, %reg. */
3263 if (i.tm.opcode_modifier & regKludge)
3265 unsigned int first_reg_op = (i.types[0] & Reg) ? 0 : 1;
3266 /* Pretend we saw the extra register operand. */
3267 assert (i.reg_operands == 1
3268 && i.op[first_reg_op + 1].regs == 0);
3269 i.op[first_reg_op + 1].regs = i.op[first_reg_op].regs;
3270 i.types[first_reg_op + 1] = i.types[first_reg_op];
3271 i.operands++;
3272 i.reg_operands++;
3275 if (i.tm.opcode_modifier & ShortForm)
3277 if (i.types[0] & (SReg2 | SReg3))
3279 if (i.tm.base_opcode == POP_SEG_SHORT
3280 && i.op[0].regs->reg_num == 1)
3282 as_bad (_("you can't `pop %%cs'"));
3283 return 0;
3285 i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
3286 if ((i.op[0].regs->reg_flags & RegRex) != 0)
3287 i.rex |= REX_EXTZ;
3289 else
3291 /* The register or float register operand is in operand 0 or 1. */
3292 unsigned int op = (i.types[0] & (Reg | FloatReg)) ? 0 : 1;
3293 /* Register goes in low 3 bits of opcode. */
3294 i.tm.base_opcode |= i.op[op].regs->reg_num;
3295 if ((i.op[op].regs->reg_flags & RegRex) != 0)
3296 i.rex |= REX_EXTZ;
3297 if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
3299 /* Warn about some common errors, but press on regardless.
3300 The first case can be generated by gcc (<= 2.8.1). */
3301 if (i.operands == 2)
3303 /* Reversed arguments on faddp, fsubp, etc. */
3304 as_warn (_("translating to `%s %%%s,%%%s'"), i.tm.name,
3305 i.op[1].regs->reg_name,
3306 i.op[0].regs->reg_name);
3308 else
3310 /* Extraneous `l' suffix on fp insn. */
3311 as_warn (_("translating to `%s %%%s'"), i.tm.name,
3312 i.op[0].regs->reg_name);
3317 else if (i.tm.opcode_modifier & Modrm)
3319 /* The opcode is completed (modulo i.tm.extension_opcode which
3320 must be put into the modrm byte). Now, we make the modrm and
3321 index base bytes based on all the info we've collected. */
3323 default_seg = build_modrm_byte ();
3325 else if ((i.tm.base_opcode & ~(D | W)) == MOV_AX_DISP32)
3327 default_seg = &ds;
3329 else if ((i.tm.opcode_modifier & IsString) != 0)
3331 /* For the string instructions that allow a segment override
3332 on one of their operands, the default segment is ds. */
3333 default_seg = &ds;
3336 if ((i.tm.base_opcode == 0x8d /* lea */
3337 || (i.tm.cpu_flags & CpuSVME))
3338 && i.seg[0] && !quiet_warnings)
3339 as_warn (_("segment override on `%s' is ineffectual"), i.tm.name);
3341 /* If a segment was explicitly specified, and the specified segment
3342 is not the default, use an opcode prefix to select it. If we
3343 never figured out what the default segment is, then default_seg
3344 will be zero at this point, and the specified segment prefix will
3345 always be used. */
3346 if ((i.seg[0]) && (i.seg[0] != default_seg))
3348 if (!add_prefix (i.seg[0]->seg_prefix))
3349 return 0;
3351 return 1;
3354 static const seg_entry *
3355 build_modrm_byte (void)
3357 const seg_entry *default_seg = 0;
3359 /* i.reg_operands MUST be the number of real register operands;
3360 implicit registers do not count. */
3361 if (i.reg_operands == 2)
3363 unsigned int source, dest;
3365 switch (i.operands)
3367 case 2:
3368 source = 0;
3369 break;
3370 case 3:
3371 /* When there are 3 operands, one of them may be immediate,
3372 which may be the first or the last operand. Otherwise,
3373 the first operand must be shift count register (cl). */
3374 assert (i.imm_operands == 1
3375 || (i.imm_operands == 0
3376 && (i.types[0] & ShiftCount)));
3377 source = (i.types[0] & (Imm | ShiftCount)) ? 1 : 0;
3378 break;
3379 case 4:
3380 /* When there are 4 operands, the first two must be immediate
3381 operands. The source operand will be the 3rd one. */
3382 assert (i.imm_operands == 2
3383 && (i.types[0] & Imm)
3384 && (i.types[1] & Imm));
3385 source = 2;
3386 break;
3387 default:
3388 abort ();
3391 dest = source + 1;
3393 i.rm.mode = 3;
3394 /* One of the register operands will be encoded in the i.tm.reg
3395 field, the other in the combined i.tm.mode and i.tm.regmem
3396 fields. If no form of this instruction supports a memory
3397 destination operand, then we assume the source operand may
3398 sometimes be a memory operand and so we need to store the
3399 destination in the i.rm.reg field. */
3400 if ((i.tm.operand_types[dest] & AnyMem) == 0)
3402 i.rm.reg = i.op[dest].regs->reg_num;
3403 i.rm.regmem = i.op[source].regs->reg_num;
3404 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
3405 i.rex |= REX_EXTX;
3406 if ((i.op[source].regs->reg_flags & RegRex) != 0)
3407 i.rex |= REX_EXTZ;
3409 else
3411 i.rm.reg = i.op[source].regs->reg_num;
3412 i.rm.regmem = i.op[dest].regs->reg_num;
3413 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
3414 i.rex |= REX_EXTZ;
3415 if ((i.op[source].regs->reg_flags & RegRex) != 0)
3416 i.rex |= REX_EXTX;
3418 if (flag_code != CODE_64BIT && (i.rex & (REX_EXTX | REX_EXTZ)))
3420 if (!((i.types[0] | i.types[1]) & Control))
3421 abort ();
3422 i.rex &= ~(REX_EXTX | REX_EXTZ);
3423 add_prefix (LOCK_PREFIX_OPCODE);
3426 else
3427 { /* If it's not 2 reg operands... */
3428 if (i.mem_operands)
3430 unsigned int fake_zero_displacement = 0;
3431 unsigned int op;
3433 for (op = 0; op < i.operands; op++)
3434 if ((i.types[op] & AnyMem))
3435 break;
3436 assert (op < i.operands);
3438 default_seg = &ds;
3440 if (i.base_reg == 0)
3442 i.rm.mode = 0;
3443 if (!i.disp_operands)
3444 fake_zero_displacement = 1;
3445 if (i.index_reg == 0)
3447 /* Operand is just <disp> */
3448 if (flag_code == CODE_64BIT)
3450 /* 64bit mode overwrites the 32bit absolute
3451 addressing by RIP relative addressing and
3452 absolute addressing is encoded by one of the
3453 redundant SIB forms. */
3454 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3455 i.sib.base = NO_BASE_REGISTER;
3456 i.sib.index = NO_INDEX_REGISTER;
3457 i.types[op] = ((i.prefix[ADDR_PREFIX] == 0)
3458 ? Disp32S : Disp32);
3460 else if ((flag_code == CODE_16BIT)
3461 ^ (i.prefix[ADDR_PREFIX] != 0))
3463 i.rm.regmem = NO_BASE_REGISTER_16;
3464 i.types[op] = Disp16;
3466 else
3468 i.rm.regmem = NO_BASE_REGISTER;
3469 i.types[op] = Disp32;
3472 else /* !i.base_reg && i.index_reg */
3474 i.sib.index = i.index_reg->reg_num;
3475 i.sib.base = NO_BASE_REGISTER;
3476 i.sib.scale = i.log2_scale_factor;
3477 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3478 i.types[op] &= ~Disp;
3479 if (flag_code != CODE_64BIT)
3480 i.types[op] |= Disp32; /* Must be 32 bit */
3481 else
3482 i.types[op] |= Disp32S;
3483 if ((i.index_reg->reg_flags & RegRex) != 0)
3484 i.rex |= REX_EXTY;
3487 /* RIP addressing for 64bit mode. */
3488 else if (i.base_reg->reg_type == BaseIndex)
3490 i.rm.regmem = NO_BASE_REGISTER;
3491 i.types[op] &= ~ Disp;
3492 i.types[op] |= Disp32S;
3493 i.flags[op] |= Operand_PCrel;
3494 if (! i.disp_operands)
3495 fake_zero_displacement = 1;
3497 else if (i.base_reg->reg_type & Reg16)
3499 switch (i.base_reg->reg_num)
3501 case 3: /* (%bx) */
3502 if (i.index_reg == 0)
3503 i.rm.regmem = 7;
3504 else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */
3505 i.rm.regmem = i.index_reg->reg_num - 6;
3506 break;
3507 case 5: /* (%bp) */
3508 default_seg = &ss;
3509 if (i.index_reg == 0)
3511 i.rm.regmem = 6;
3512 if ((i.types[op] & Disp) == 0)
3514 /* fake (%bp) into 0(%bp) */
3515 i.types[op] |= Disp8;
3516 fake_zero_displacement = 1;
3519 else /* (%bp,%si) -> 2, or (%bp,%di) -> 3 */
3520 i.rm.regmem = i.index_reg->reg_num - 6 + 2;
3521 break;
3522 default: /* (%si) -> 4 or (%di) -> 5 */
3523 i.rm.regmem = i.base_reg->reg_num - 6 + 4;
3525 i.rm.mode = mode_from_disp_size (i.types[op]);
3527 else /* i.base_reg and 32/64 bit mode */
3529 if (flag_code == CODE_64BIT
3530 && (i.types[op] & Disp))
3531 i.types[op] = ((i.types[op] & Disp8)
3532 | (i.prefix[ADDR_PREFIX] == 0
3533 ? Disp32S : Disp32));
3535 i.rm.regmem = i.base_reg->reg_num;
3536 if ((i.base_reg->reg_flags & RegRex) != 0)
3537 i.rex |= REX_EXTZ;
3538 i.sib.base = i.base_reg->reg_num;
3539 /* x86-64 ignores REX prefix bit here to avoid decoder
3540 complications. */
3541 if ((i.base_reg->reg_num & 7) == EBP_REG_NUM)
3543 default_seg = &ss;
3544 if (i.disp_operands == 0)
3546 fake_zero_displacement = 1;
3547 i.types[op] |= Disp8;
3550 else if (i.base_reg->reg_num == ESP_REG_NUM)
3552 default_seg = &ss;
3554 i.sib.scale = i.log2_scale_factor;
3555 if (i.index_reg == 0)
3557 /* <disp>(%esp) becomes two byte modrm with no index
3558 register. We've already stored the code for esp
3559 in i.rm.regmem ie. ESCAPE_TO_TWO_BYTE_ADDRESSING.
3560 Any base register besides %esp will not use the
3561 extra modrm byte. */
3562 i.sib.index = NO_INDEX_REGISTER;
3563 #if !SCALE1_WHEN_NO_INDEX
3564 /* Another case where we force the second modrm byte. */
3565 if (i.log2_scale_factor)
3566 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3567 #endif
3569 else
3571 i.sib.index = i.index_reg->reg_num;
3572 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3573 if ((i.index_reg->reg_flags & RegRex) != 0)
3574 i.rex |= REX_EXTY;
3577 if (i.disp_operands
3578 && (i.reloc[op] == BFD_RELOC_386_TLS_DESC_CALL
3579 || i.reloc[op] == BFD_RELOC_X86_64_TLSDESC_CALL))
3580 i.rm.mode = 0;
3581 else
3582 i.rm.mode = mode_from_disp_size (i.types[op]);
3585 if (fake_zero_displacement)
3587 /* Fakes a zero displacement assuming that i.types[op]
3588 holds the correct displacement size. */
3589 expressionS *exp;
3591 assert (i.op[op].disps == 0);
3592 exp = &disp_expressions[i.disp_operands++];
3593 i.op[op].disps = exp;
3594 exp->X_op = O_constant;
3595 exp->X_add_number = 0;
3596 exp->X_add_symbol = (symbolS *) 0;
3597 exp->X_op_symbol = (symbolS *) 0;
3601 /* Fill in i.rm.reg or i.rm.regmem field with register operand
3602 (if any) based on i.tm.extension_opcode. Again, we must be
3603 careful to make sure that segment/control/debug/test/MMX
3604 registers are coded into the i.rm.reg field. */
3605 if (i.reg_operands)
3607 unsigned int op;
3609 for (op = 0; op < i.operands; op++)
3610 if ((i.types[op] & (Reg | RegMMX | RegXMM
3611 | SReg2 | SReg3
3612 | Control | Debug | Test)))
3613 break;
3614 assert (op < i.operands);
3616 /* If there is an extension opcode to put here, the register
3617 number must be put into the regmem field. */
3618 if (i.tm.extension_opcode != None)
3620 i.rm.regmem = i.op[op].regs->reg_num;
3621 if ((i.op[op].regs->reg_flags & RegRex) != 0)
3622 i.rex |= REX_EXTZ;
3624 else
3626 i.rm.reg = i.op[op].regs->reg_num;
3627 if ((i.op[op].regs->reg_flags & RegRex) != 0)
3628 i.rex |= REX_EXTX;
3631 /* Now, if no memory operand has set i.rm.mode = 0, 1, 2 we
3632 must set it to 3 to indicate this is a register operand
3633 in the regmem field. */
3634 if (!i.mem_operands)
3635 i.rm.mode = 3;
3638 /* Fill in i.rm.reg field with extension opcode (if any). */
3639 if (i.tm.extension_opcode != None)
3640 i.rm.reg = i.tm.extension_opcode;
3642 return default_seg;
3645 static void
3646 output_branch (void)
3648 char *p;
3649 int code16;
3650 int prefix;
3651 relax_substateT subtype;
3652 symbolS *sym;
3653 offsetT off;
3655 code16 = 0;
3656 if (flag_code == CODE_16BIT)
3657 code16 = CODE16;
3659 prefix = 0;
3660 if (i.prefix[DATA_PREFIX] != 0)
3662 prefix = 1;
3663 i.prefixes -= 1;
3664 code16 ^= CODE16;
3666 /* Pentium4 branch hints. */
3667 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
3668 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
3670 prefix++;
3671 i.prefixes--;
3673 if (i.prefix[REX_PREFIX] != 0)
3675 prefix++;
3676 i.prefixes--;
3679 if (i.prefixes != 0 && !intel_syntax)
3680 as_warn (_("skipping prefixes on this instruction"));
3682 /* It's always a symbol; End frag & setup for relax.
3683 Make sure there is enough room in this frag for the largest
3684 instruction we may generate in md_convert_frag. This is 2
3685 bytes for the opcode and room for the prefix and largest
3686 displacement. */
3687 frag_grow (prefix + 2 + 4);
3688 /* Prefix and 1 opcode byte go in fr_fix. */
3689 p = frag_more (prefix + 1);
3690 if (i.prefix[DATA_PREFIX] != 0)
3691 *p++ = DATA_PREFIX_OPCODE;
3692 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE
3693 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE)
3694 *p++ = i.prefix[SEG_PREFIX];
3695 if (i.prefix[REX_PREFIX] != 0)
3696 *p++ = i.prefix[REX_PREFIX];
3697 *p = i.tm.base_opcode;
3699 if ((unsigned char) *p == JUMP_PC_RELATIVE)
3700 subtype = ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL);
3701 else if ((cpu_arch_flags & Cpu386) != 0)
3702 subtype = ENCODE_RELAX_STATE (COND_JUMP, SMALL);
3703 else
3704 subtype = ENCODE_RELAX_STATE (COND_JUMP86, SMALL);
3705 subtype |= code16;
3707 sym = i.op[0].disps->X_add_symbol;
3708 off = i.op[0].disps->X_add_number;
3710 if (i.op[0].disps->X_op != O_constant
3711 && i.op[0].disps->X_op != O_symbol)
3713 /* Handle complex expressions. */
3714 sym = make_expr_symbol (i.op[0].disps);
3715 off = 0;
3718 /* 1 possible extra opcode + 4 byte displacement go in var part.
3719 Pass reloc in fr_var. */
3720 frag_var (rs_machine_dependent, 5, i.reloc[0], subtype, sym, off, p);
3723 static void
3724 output_jump (void)
3726 char *p;
3727 int size;
3728 fixS *fixP;
3730 if (i.tm.opcode_modifier & JumpByte)
3732 /* This is a loop or jecxz type instruction. */
3733 size = 1;
3734 if (i.prefix[ADDR_PREFIX] != 0)
3736 FRAG_APPEND_1_CHAR (ADDR_PREFIX_OPCODE);
3737 i.prefixes -= 1;
3739 /* Pentium4 branch hints. */
3740 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
3741 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
3743 FRAG_APPEND_1_CHAR (i.prefix[SEG_PREFIX]);
3744 i.prefixes--;
3747 else
3749 int code16;
3751 code16 = 0;
3752 if (flag_code == CODE_16BIT)
3753 code16 = CODE16;
3755 if (i.prefix[DATA_PREFIX] != 0)
3757 FRAG_APPEND_1_CHAR (DATA_PREFIX_OPCODE);
3758 i.prefixes -= 1;
3759 code16 ^= CODE16;
3762 size = 4;
3763 if (code16)
3764 size = 2;
3767 if (i.prefix[REX_PREFIX] != 0)
3769 FRAG_APPEND_1_CHAR (i.prefix[REX_PREFIX]);
3770 i.prefixes -= 1;
3773 if (i.prefixes != 0 && !intel_syntax)
3774 as_warn (_("skipping prefixes on this instruction"));
3776 p = frag_more (1 + size);
3777 *p++ = i.tm.base_opcode;
3779 fixP = fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3780 i.op[0].disps, 1, reloc (size, 1, 1, i.reloc[0]));
3782 /* All jumps handled here are signed, but don't use a signed limit
3783 check for 32 and 16 bit jumps as we want to allow wrap around at
3784 4G and 64k respectively. */
3785 if (size == 1)
3786 fixP->fx_signed = 1;
3789 static void
3790 output_interseg_jump (void)
3792 char *p;
3793 int size;
3794 int prefix;
3795 int code16;
3797 code16 = 0;
3798 if (flag_code == CODE_16BIT)
3799 code16 = CODE16;
3801 prefix = 0;
3802 if (i.prefix[DATA_PREFIX] != 0)
3804 prefix = 1;
3805 i.prefixes -= 1;
3806 code16 ^= CODE16;
3808 if (i.prefix[REX_PREFIX] != 0)
3810 prefix++;
3811 i.prefixes -= 1;
3814 size = 4;
3815 if (code16)
3816 size = 2;
3818 if (i.prefixes != 0 && !intel_syntax)
3819 as_warn (_("skipping prefixes on this instruction"));
3821 /* 1 opcode; 2 segment; offset */
3822 p = frag_more (prefix + 1 + 2 + size);
3824 if (i.prefix[DATA_PREFIX] != 0)
3825 *p++ = DATA_PREFIX_OPCODE;
3827 if (i.prefix[REX_PREFIX] != 0)
3828 *p++ = i.prefix[REX_PREFIX];
3830 *p++ = i.tm.base_opcode;
3831 if (i.op[1].imms->X_op == O_constant)
3833 offsetT n = i.op[1].imms->X_add_number;
3835 if (size == 2
3836 && !fits_in_unsigned_word (n)
3837 && !fits_in_signed_word (n))
3839 as_bad (_("16-bit jump out of range"));
3840 return;
3842 md_number_to_chars (p, n, size);
3844 else
3845 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3846 i.op[1].imms, 0, reloc (size, 0, 0, i.reloc[1]));
3847 if (i.op[0].imms->X_op != O_constant)
3848 as_bad (_("can't handle non absolute segment in `%s'"),
3849 i.tm.name);
3850 md_number_to_chars (p + size, (valueT) i.op[0].imms->X_add_number, 2);
3853 static void
3854 output_insn (void)
3856 fragS *insn_start_frag;
3857 offsetT insn_start_off;
3859 /* Tie dwarf2 debug info to the address at the start of the insn.
3860 We can't do this after the insn has been output as the current
3861 frag may have been closed off. eg. by frag_var. */
3862 dwarf2_emit_insn (0);
3864 insn_start_frag = frag_now;
3865 insn_start_off = frag_now_fix ();
3867 /* Output jumps. */
3868 if (i.tm.opcode_modifier & Jump)
3869 output_branch ();
3870 else if (i.tm.opcode_modifier & (JumpByte | JumpDword))
3871 output_jump ();
3872 else if (i.tm.opcode_modifier & JumpInterSegment)
3873 output_interseg_jump ();
3874 else
3876 /* Output normal instructions here. */
3877 char *p;
3878 unsigned char *q;
3879 unsigned int prefix;
3881 /* All opcodes on i386 have either 1 or 2 bytes. Supplemental
3882 Streaming SIMD extensions 3 Instructions have 3 bytes. We may
3883 use one more higher byte to specify a prefix the instruction
3884 requires. */
3885 if ((i.tm.cpu_flags & CpuSSSE3) != 0)
3887 if (i.tm.base_opcode & 0xff000000)
3889 prefix = (i.tm.base_opcode >> 24) & 0xff;
3890 goto check_prefix;
3893 else if ((i.tm.base_opcode & 0xff0000) != 0)
3895 prefix = (i.tm.base_opcode >> 16) & 0xff;
3896 if ((i.tm.cpu_flags & CpuPadLock) != 0)
3898 check_prefix:
3899 if (prefix != REPE_PREFIX_OPCODE
3900 || i.prefix[LOCKREP_PREFIX] != REPE_PREFIX_OPCODE)
3901 add_prefix (prefix);
3903 else
3904 add_prefix (prefix);
3907 /* The prefix bytes. */
3908 for (q = i.prefix;
3909 q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
3910 q++)
3912 if (*q)
3914 p = frag_more (1);
3915 md_number_to_chars (p, (valueT) *q, 1);
3919 /* Now the opcode; be careful about word order here! */
3920 if (fits_in_unsigned_byte (i.tm.base_opcode))
3922 FRAG_APPEND_1_CHAR (i.tm.base_opcode);
3924 else
3926 if ((i.tm.cpu_flags & CpuSSSE3) != 0)
3928 p = frag_more (3);
3929 *p++ = (i.tm.base_opcode >> 16) & 0xff;
3931 else
3932 p = frag_more (2);
3934 /* Put out high byte first: can't use md_number_to_chars! */
3935 *p++ = (i.tm.base_opcode >> 8) & 0xff;
3936 *p = i.tm.base_opcode & 0xff;
3939 /* Now the modrm byte and sib byte (if present). */
3940 if (i.tm.opcode_modifier & Modrm)
3942 p = frag_more (1);
3943 md_number_to_chars (p,
3944 (valueT) (i.rm.regmem << 0
3945 | i.rm.reg << 3
3946 | i.rm.mode << 6),
3948 /* If i.rm.regmem == ESP (4)
3949 && i.rm.mode != (Register mode)
3950 && not 16 bit
3951 ==> need second modrm byte. */
3952 if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING
3953 && i.rm.mode != 3
3954 && !(i.base_reg && (i.base_reg->reg_type & Reg16) != 0))
3956 p = frag_more (1);
3957 md_number_to_chars (p,
3958 (valueT) (i.sib.base << 0
3959 | i.sib.index << 3
3960 | i.sib.scale << 6),
3965 if (i.disp_operands)
3966 output_disp (insn_start_frag, insn_start_off);
3968 if (i.imm_operands)
3969 output_imm (insn_start_frag, insn_start_off);
3972 #ifdef DEBUG386
3973 if (flag_debug)
3975 pi ("" /*line*/, &i);
3977 #endif /* DEBUG386 */
3980 static void
3981 output_disp (fragS *insn_start_frag, offsetT insn_start_off)
3983 char *p;
3984 unsigned int n;
3986 for (n = 0; n < i.operands; n++)
3988 if (i.types[n] & Disp)
3990 if (i.op[n].disps->X_op == O_constant)
3992 int size;
3993 offsetT val;
3995 size = 4;
3996 if (i.types[n] & (Disp8 | Disp16 | Disp64))
3998 size = 2;
3999 if (i.types[n] & Disp8)
4000 size = 1;
4001 if (i.types[n] & Disp64)
4002 size = 8;
4004 val = offset_in_range (i.op[n].disps->X_add_number,
4005 size);
4006 p = frag_more (size);
4007 md_number_to_chars (p, val, size);
4009 else
4011 enum bfd_reloc_code_real reloc_type;
4012 int size = 4;
4013 int sign = 0;
4014 int pcrel = (i.flags[n] & Operand_PCrel) != 0;
4016 /* The PC relative address is computed relative
4017 to the instruction boundary, so in case immediate
4018 fields follows, we need to adjust the value. */
4019 if (pcrel && i.imm_operands)
4021 int imm_size = 4;
4022 unsigned int n1;
4024 for (n1 = 0; n1 < i.operands; n1++)
4025 if (i.types[n1] & Imm)
4027 if (i.types[n1] & (Imm8 | Imm8S | Imm16 | Imm64))
4029 imm_size = 2;
4030 if (i.types[n1] & (Imm8 | Imm8S))
4031 imm_size = 1;
4032 if (i.types[n1] & Imm64)
4033 imm_size = 8;
4035 break;
4037 /* We should find the immediate. */
4038 if (n1 == i.operands)
4039 abort ();
4040 i.op[n].disps->X_add_number -= imm_size;
4043 if (i.types[n] & Disp32S)
4044 sign = 1;
4046 if (i.types[n] & (Disp16 | Disp64))
4048 size = 2;
4049 if (i.types[n] & Disp64)
4050 size = 8;
4053 p = frag_more (size);
4054 reloc_type = reloc (size, pcrel, sign, i.reloc[n]);
4055 if (GOT_symbol
4056 && GOT_symbol == i.op[n].disps->X_add_symbol
4057 && (((reloc_type == BFD_RELOC_32
4058 || reloc_type == BFD_RELOC_X86_64_32S
4059 || (reloc_type == BFD_RELOC_64
4060 && object_64bit))
4061 && (i.op[n].disps->X_op == O_symbol
4062 || (i.op[n].disps->X_op == O_add
4063 && ((symbol_get_value_expression
4064 (i.op[n].disps->X_op_symbol)->X_op)
4065 == O_subtract))))
4066 || reloc_type == BFD_RELOC_32_PCREL))
4068 offsetT add;
4070 if (insn_start_frag == frag_now)
4071 add = (p - frag_now->fr_literal) - insn_start_off;
4072 else
4074 fragS *fr;
4076 add = insn_start_frag->fr_fix - insn_start_off;
4077 for (fr = insn_start_frag->fr_next;
4078 fr && fr != frag_now; fr = fr->fr_next)
4079 add += fr->fr_fix;
4080 add += p - frag_now->fr_literal;
4083 if (!object_64bit)
4085 reloc_type = BFD_RELOC_386_GOTPC;
4086 i.op[n].imms->X_add_number += add;
4088 else if (reloc_type == BFD_RELOC_64)
4089 reloc_type = BFD_RELOC_X86_64_GOTPC64;
4090 else
4091 /* Don't do the adjustment for x86-64, as there
4092 the pcrel addressing is relative to the _next_
4093 insn, and that is taken care of in other code. */
4094 reloc_type = BFD_RELOC_X86_64_GOTPC32;
4096 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
4097 i.op[n].disps, pcrel, reloc_type);
4103 static void
4104 output_imm (fragS *insn_start_frag, offsetT insn_start_off)
4106 char *p;
4107 unsigned int n;
4109 for (n = 0; n < i.operands; n++)
4111 if (i.types[n] & Imm)
4113 if (i.op[n].imms->X_op == O_constant)
4115 int size;
4116 offsetT val;
4118 size = 4;
4119 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
4121 size = 2;
4122 if (i.types[n] & (Imm8 | Imm8S))
4123 size = 1;
4124 else if (i.types[n] & Imm64)
4125 size = 8;
4127 val = offset_in_range (i.op[n].imms->X_add_number,
4128 size);
4129 p = frag_more (size);
4130 md_number_to_chars (p, val, size);
4132 else
4134 /* Not absolute_section.
4135 Need a 32-bit fixup (don't support 8bit
4136 non-absolute imms). Try to support other
4137 sizes ... */
4138 enum bfd_reloc_code_real reloc_type;
4139 int size = 4;
4140 int sign = 0;
4142 if ((i.types[n] & (Imm32S))
4143 && (i.suffix == QWORD_MNEM_SUFFIX
4144 || (!i.suffix && (i.tm.opcode_modifier & No_lSuf))))
4145 sign = 1;
4146 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
4148 size = 2;
4149 if (i.types[n] & (Imm8 | Imm8S))
4150 size = 1;
4151 if (i.types[n] & Imm64)
4152 size = 8;
4155 p = frag_more (size);
4156 reloc_type = reloc (size, 0, sign, i.reloc[n]);
4158 /* This is tough to explain. We end up with this one if we
4159 * have operands that look like
4160 * "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal here is to
4161 * obtain the absolute address of the GOT, and it is strongly
4162 * preferable from a performance point of view to avoid using
4163 * a runtime relocation for this. The actual sequence of
4164 * instructions often look something like:
4166 * call .L66
4167 * .L66:
4168 * popl %ebx
4169 * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
4171 * The call and pop essentially return the absolute address
4172 * of the label .L66 and store it in %ebx. The linker itself
4173 * will ultimately change the first operand of the addl so
4174 * that %ebx points to the GOT, but to keep things simple, the
4175 * .o file must have this operand set so that it generates not
4176 * the absolute address of .L66, but the absolute address of
4177 * itself. This allows the linker itself simply treat a GOTPC
4178 * relocation as asking for a pcrel offset to the GOT to be
4179 * added in, and the addend of the relocation is stored in the
4180 * operand field for the instruction itself.
4182 * Our job here is to fix the operand so that it would add
4183 * the correct offset so that %ebx would point to itself. The
4184 * thing that is tricky is that .-.L66 will point to the
4185 * beginning of the instruction, so we need to further modify
4186 * the operand so that it will point to itself. There are
4187 * other cases where you have something like:
4189 * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
4191 * and here no correction would be required. Internally in
4192 * the assembler we treat operands of this form as not being
4193 * pcrel since the '.' is explicitly mentioned, and I wonder
4194 * whether it would simplify matters to do it this way. Who
4195 * knows. In earlier versions of the PIC patches, the
4196 * pcrel_adjust field was used to store the correction, but
4197 * since the expression is not pcrel, I felt it would be
4198 * confusing to do it this way. */
4200 if ((reloc_type == BFD_RELOC_32
4201 || reloc_type == BFD_RELOC_X86_64_32S
4202 || reloc_type == BFD_RELOC_64)
4203 && GOT_symbol
4204 && GOT_symbol == i.op[n].imms->X_add_symbol
4205 && (i.op[n].imms->X_op == O_symbol
4206 || (i.op[n].imms->X_op == O_add
4207 && ((symbol_get_value_expression
4208 (i.op[n].imms->X_op_symbol)->X_op)
4209 == O_subtract))))
4211 offsetT add;
4213 if (insn_start_frag == frag_now)
4214 add = (p - frag_now->fr_literal) - insn_start_off;
4215 else
4217 fragS *fr;
4219 add = insn_start_frag->fr_fix - insn_start_off;
4220 for (fr = insn_start_frag->fr_next;
4221 fr && fr != frag_now; fr = fr->fr_next)
4222 add += fr->fr_fix;
4223 add += p - frag_now->fr_literal;
4226 if (!object_64bit)
4227 reloc_type = BFD_RELOC_386_GOTPC;
4228 else if (size == 4)
4229 reloc_type = BFD_RELOC_X86_64_GOTPC32;
4230 else if (size == 8)
4231 reloc_type = BFD_RELOC_X86_64_GOTPC64;
4232 i.op[n].imms->X_add_number += add;
4234 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
4235 i.op[n].imms, 0, reloc_type);
4241 /* x86_cons_fix_new is called via the expression parsing code when a
4242 reloc is needed. We use this hook to get the correct .got reloc. */
4243 static enum bfd_reloc_code_real got_reloc = NO_RELOC;
4244 static int cons_sign = -1;
4246 void
4247 x86_cons_fix_new (fragS *frag, unsigned int off, unsigned int len,
4248 expressionS *exp)
4250 enum bfd_reloc_code_real r = reloc (len, 0, cons_sign, got_reloc);
4252 got_reloc = NO_RELOC;
4254 #ifdef TE_PE
4255 if (exp->X_op == O_secrel)
4257 exp->X_op = O_symbol;
4258 r = BFD_RELOC_32_SECREL;
4260 #endif
4262 fix_new_exp (frag, off, len, exp, 0, r);
4265 #if (!defined (OBJ_ELF) && !defined (OBJ_MAYBE_ELF)) || defined (LEX_AT)
4266 # define lex_got(reloc, adjust, types) NULL
4267 #else
4268 /* Parse operands of the form
4269 <symbol>@GOTOFF+<nnn>
4270 and similar .plt or .got references.
4272 If we find one, set up the correct relocation in RELOC and copy the
4273 input string, minus the `@GOTOFF' into a malloc'd buffer for
4274 parsing by the calling routine. Return this buffer, and if ADJUST
4275 is non-null set it to the length of the string we removed from the
4276 input line. Otherwise return NULL. */
4277 static char *
4278 lex_got (enum bfd_reloc_code_real *reloc,
4279 int *adjust,
4280 unsigned int *types)
4282 /* Some of the relocations depend on the size of what field is to
4283 be relocated. But in our callers i386_immediate and i386_displacement
4284 we don't yet know the operand size (this will be set by insn
4285 matching). Hence we record the word32 relocation here,
4286 and adjust the reloc according to the real size in reloc(). */
4287 static const struct {
4288 const char *str;
4289 const enum bfd_reloc_code_real rel[2];
4290 const unsigned int types64;
4291 } gotrel[] = {
4292 { "PLTOFF", { 0,
4293 BFD_RELOC_X86_64_PLTOFF64 },
4294 Imm64 },
4295 { "PLT", { BFD_RELOC_386_PLT32,
4296 BFD_RELOC_X86_64_PLT32 },
4297 Imm32 | Imm32S | Disp32 },
4298 { "GOTPLT", { 0,
4299 BFD_RELOC_X86_64_GOTPLT64 },
4300 Imm64 | Disp64 },
4301 { "GOTOFF", { BFD_RELOC_386_GOTOFF,
4302 BFD_RELOC_X86_64_GOTOFF64 },
4303 Imm64 | Disp64 },
4304 { "GOTPCREL", { 0,
4305 BFD_RELOC_X86_64_GOTPCREL },
4306 Imm32 | Imm32S | Disp32 },
4307 { "TLSGD", { BFD_RELOC_386_TLS_GD,
4308 BFD_RELOC_X86_64_TLSGD },
4309 Imm32 | Imm32S | Disp32 },
4310 { "TLSLDM", { BFD_RELOC_386_TLS_LDM,
4311 0 },
4312 0 },
4313 { "TLSLD", { 0,
4314 BFD_RELOC_X86_64_TLSLD },
4315 Imm32 | Imm32S | Disp32 },
4316 { "GOTTPOFF", { BFD_RELOC_386_TLS_IE_32,
4317 BFD_RELOC_X86_64_GOTTPOFF },
4318 Imm32 | Imm32S | Disp32 },
4319 { "TPOFF", { BFD_RELOC_386_TLS_LE_32,
4320 BFD_RELOC_X86_64_TPOFF32 },
4321 Imm32 | Imm32S | Imm64 | Disp32 | Disp64 },
4322 { "NTPOFF", { BFD_RELOC_386_TLS_LE,
4323 0 },
4324 0 },
4325 { "DTPOFF", { BFD_RELOC_386_TLS_LDO_32,
4326 BFD_RELOC_X86_64_DTPOFF32 },
4327 Imm32 | Imm32S | Imm64 | Disp32 | Disp64 },
4328 { "GOTNTPOFF",{ BFD_RELOC_386_TLS_GOTIE,
4329 0 },
4330 0 },
4331 { "INDNTPOFF",{ BFD_RELOC_386_TLS_IE,
4332 0 },
4333 0 },
4334 { "GOT", { BFD_RELOC_386_GOT32,
4335 BFD_RELOC_X86_64_GOT32 },
4336 Imm32 | Imm32S | Disp32 | Imm64 },
4337 { "TLSDESC", { BFD_RELOC_386_TLS_GOTDESC,
4338 BFD_RELOC_X86_64_GOTPC32_TLSDESC },
4339 Imm32 | Imm32S | Disp32 },
4340 { "TLSCALL", { BFD_RELOC_386_TLS_DESC_CALL,
4341 BFD_RELOC_X86_64_TLSDESC_CALL },
4342 Imm32 | Imm32S | Disp32 }
4344 char *cp;
4345 unsigned int j;
4347 if (!IS_ELF)
4348 return NULL;
4350 for (cp = input_line_pointer; *cp != '@'; cp++)
4351 if (is_end_of_line[(unsigned char) *cp])
4352 return NULL;
4354 for (j = 0; j < sizeof (gotrel) / sizeof (gotrel[0]); j++)
4356 int len;
4358 len = strlen (gotrel[j].str);
4359 if (strncasecmp (cp + 1, gotrel[j].str, len) == 0)
4361 if (gotrel[j].rel[object_64bit] != 0)
4363 int first, second;
4364 char *tmpbuf, *past_reloc;
4366 *reloc = gotrel[j].rel[object_64bit];
4367 if (adjust)
4368 *adjust = len;
4370 if (types)
4372 if (flag_code != CODE_64BIT)
4373 *types = Imm32 | Disp32;
4374 else
4375 *types = gotrel[j].types64;
4378 if (GOT_symbol == NULL)
4379 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
4381 /* Replace the relocation token with ' ', so that
4382 errors like foo@GOTOFF1 will be detected. */
4384 /* The length of the first part of our input line. */
4385 first = cp - input_line_pointer;
4387 /* The second part goes from after the reloc token until
4388 (and including) an end_of_line char. Don't use strlen
4389 here as the end_of_line char may not be a NUL. */
4390 past_reloc = cp + 1 + len;
4391 for (cp = past_reloc; !is_end_of_line[(unsigned char) *cp++]; )
4393 second = cp - past_reloc;
4395 /* Allocate and copy string. The trailing NUL shouldn't
4396 be necessary, but be safe. */
4397 tmpbuf = xmalloc (first + second + 2);
4398 memcpy (tmpbuf, input_line_pointer, first);
4399 tmpbuf[first] = ' ';
4400 memcpy (tmpbuf + first + 1, past_reloc, second);
4401 tmpbuf[first + second + 1] = '\0';
4402 return tmpbuf;
4405 as_bad (_("@%s reloc is not supported with %d-bit output format"),
4406 gotrel[j].str, 1 << (5 + object_64bit));
4407 return NULL;
4411 /* Might be a symbol version string. Don't as_bad here. */
4412 return NULL;
4415 void
4416 x86_cons (expressionS *exp, int size)
4418 if (size == 4 || (object_64bit && size == 8))
4420 /* Handle @GOTOFF and the like in an expression. */
4421 char *save;
4422 char *gotfree_input_line;
4423 int adjust;
4425 save = input_line_pointer;
4426 gotfree_input_line = lex_got (&got_reloc, &adjust, NULL);
4427 if (gotfree_input_line)
4428 input_line_pointer = gotfree_input_line;
4430 expression (exp);
4432 if (gotfree_input_line)
4434 /* expression () has merrily parsed up to the end of line,
4435 or a comma - in the wrong buffer. Transfer how far
4436 input_line_pointer has moved to the right buffer. */
4437 input_line_pointer = (save
4438 + (input_line_pointer - gotfree_input_line)
4439 + adjust);
4440 free (gotfree_input_line);
4443 else
4444 expression (exp);
4446 #endif
4448 static void signed_cons (int size)
4450 if (flag_code == CODE_64BIT)
4451 cons_sign = 1;
4452 cons (size);
4453 cons_sign = -1;
4456 #ifdef TE_PE
4457 static void
4458 pe_directive_secrel (dummy)
4459 int dummy ATTRIBUTE_UNUSED;
4461 expressionS exp;
4465 expression (&exp);
4466 if (exp.X_op == O_symbol)
4467 exp.X_op = O_secrel;
4469 emit_expr (&exp, 4);
4471 while (*input_line_pointer++ == ',');
4473 input_line_pointer--;
4474 demand_empty_rest_of_line ();
4476 #endif
4478 static int
4479 i386_immediate (char *imm_start)
4481 char *save_input_line_pointer;
4482 char *gotfree_input_line;
4483 segT exp_seg = 0;
4484 expressionS *exp;
4485 unsigned int types = ~0U;
4487 if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
4489 as_bad (_("at most %d immediate operands are allowed"),
4490 MAX_IMMEDIATE_OPERANDS);
4491 return 0;
4494 exp = &im_expressions[i.imm_operands++];
4495 i.op[this_operand].imms = exp;
4497 if (is_space_char (*imm_start))
4498 ++imm_start;
4500 save_input_line_pointer = input_line_pointer;
4501 input_line_pointer = imm_start;
4503 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL, &types);
4504 if (gotfree_input_line)
4505 input_line_pointer = gotfree_input_line;
4507 exp_seg = expression (exp);
4509 SKIP_WHITESPACE ();
4510 if (*input_line_pointer)
4511 as_bad (_("junk `%s' after expression"), input_line_pointer);
4513 input_line_pointer = save_input_line_pointer;
4514 if (gotfree_input_line)
4515 free (gotfree_input_line);
4517 if (exp->X_op == O_absent || exp->X_op == O_big)
4519 /* Missing or bad expr becomes absolute 0. */
4520 as_bad (_("missing or invalid immediate expression `%s' taken as 0"),
4521 imm_start);
4522 exp->X_op = O_constant;
4523 exp->X_add_number = 0;
4524 exp->X_add_symbol = (symbolS *) 0;
4525 exp->X_op_symbol = (symbolS *) 0;
4527 else if (exp->X_op == O_constant)
4529 /* Size it properly later. */
4530 i.types[this_operand] |= Imm64;
4531 /* If BFD64, sign extend val. */
4532 if (!use_rela_relocations
4533 && (exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0)
4534 exp->X_add_number
4535 = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
4537 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4538 else if (OUTPUT_FLAVOR == bfd_target_aout_flavour
4539 && exp_seg != absolute_section
4540 && exp_seg != text_section
4541 && exp_seg != data_section
4542 && exp_seg != bss_section
4543 && exp_seg != undefined_section
4544 && !bfd_is_com_section (exp_seg))
4546 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
4547 return 0;
4549 #endif
4550 else if (!intel_syntax && exp->X_op == O_register)
4552 as_bad (_("illegal immediate register operand %s"), imm_start);
4553 return 0;
4555 else
4557 /* This is an address. The size of the address will be
4558 determined later, depending on destination register,
4559 suffix, or the default for the section. */
4560 i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64;
4561 i.types[this_operand] &= types;
4564 return 1;
4567 static char *
4568 i386_scale (char *scale)
4570 offsetT val;
4571 char *save = input_line_pointer;
4573 input_line_pointer = scale;
4574 val = get_absolute_expression ();
4576 switch (val)
4578 case 1:
4579 i.log2_scale_factor = 0;
4580 break;
4581 case 2:
4582 i.log2_scale_factor = 1;
4583 break;
4584 case 4:
4585 i.log2_scale_factor = 2;
4586 break;
4587 case 8:
4588 i.log2_scale_factor = 3;
4589 break;
4590 default:
4592 char sep = *input_line_pointer;
4594 *input_line_pointer = '\0';
4595 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
4596 scale);
4597 *input_line_pointer = sep;
4598 input_line_pointer = save;
4599 return NULL;
4602 if (i.log2_scale_factor != 0 && i.index_reg == 0)
4604 as_warn (_("scale factor of %d without an index register"),
4605 1 << i.log2_scale_factor);
4606 #if SCALE1_WHEN_NO_INDEX
4607 i.log2_scale_factor = 0;
4608 #endif
4610 scale = input_line_pointer;
4611 input_line_pointer = save;
4612 return scale;
4615 static int
4616 i386_displacement (char *disp_start, char *disp_end)
4618 expressionS *exp;
4619 segT exp_seg = 0;
4620 char *save_input_line_pointer;
4621 char *gotfree_input_line;
4622 int bigdisp, override;
4623 unsigned int types = Disp;
4625 if (i.disp_operands == MAX_MEMORY_OPERANDS)
4627 as_bad (_("at most %d displacement operands are allowed"),
4628 MAX_MEMORY_OPERANDS);
4629 return 0;
4632 if ((i.types[this_operand] & JumpAbsolute)
4633 || !(current_templates->start->opcode_modifier & (Jump | JumpDword)))
4635 bigdisp = Disp32;
4636 override = (i.prefix[ADDR_PREFIX] != 0);
4638 else
4640 /* For PC-relative branches, the width of the displacement
4641 is dependent upon data size, not address size. */
4642 bigdisp = 0;
4643 override = (i.prefix[DATA_PREFIX] != 0);
4645 if (flag_code == CODE_64BIT)
4647 if (!bigdisp)
4648 bigdisp = ((override || i.suffix == WORD_MNEM_SUFFIX)
4649 ? Disp16
4650 : Disp32S | Disp32);
4651 else if (!override)
4652 bigdisp = Disp64 | Disp32S | Disp32;
4654 else
4656 if (!bigdisp)
4658 if (!override)
4659 override = (i.suffix == (flag_code != CODE_16BIT
4660 ? WORD_MNEM_SUFFIX
4661 : LONG_MNEM_SUFFIX));
4662 bigdisp = Disp32;
4664 if ((flag_code == CODE_16BIT) ^ override)
4665 bigdisp = Disp16;
4667 i.types[this_operand] |= bigdisp;
4669 exp = &disp_expressions[i.disp_operands];
4670 i.op[this_operand].disps = exp;
4671 i.disp_operands++;
4672 save_input_line_pointer = input_line_pointer;
4673 input_line_pointer = disp_start;
4674 END_STRING_AND_SAVE (disp_end);
4676 #ifndef GCC_ASM_O_HACK
4677 #define GCC_ASM_O_HACK 0
4678 #endif
4679 #if GCC_ASM_O_HACK
4680 END_STRING_AND_SAVE (disp_end + 1);
4681 if ((i.types[this_operand] & BaseIndex) != 0
4682 && displacement_string_end[-1] == '+')
4684 /* This hack is to avoid a warning when using the "o"
4685 constraint within gcc asm statements.
4686 For instance:
4688 #define _set_tssldt_desc(n,addr,limit,type) \
4689 __asm__ __volatile__ ( \
4690 "movw %w2,%0\n\t" \
4691 "movw %w1,2+%0\n\t" \
4692 "rorl $16,%1\n\t" \
4693 "movb %b1,4+%0\n\t" \
4694 "movb %4,5+%0\n\t" \
4695 "movb $0,6+%0\n\t" \
4696 "movb %h1,7+%0\n\t" \
4697 "rorl $16,%1" \
4698 : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
4700 This works great except that the output assembler ends
4701 up looking a bit weird if it turns out that there is
4702 no offset. You end up producing code that looks like:
4704 #APP
4705 movw $235,(%eax)
4706 movw %dx,2+(%eax)
4707 rorl $16,%edx
4708 movb %dl,4+(%eax)
4709 movb $137,5+(%eax)
4710 movb $0,6+(%eax)
4711 movb %dh,7+(%eax)
4712 rorl $16,%edx
4713 #NO_APP
4715 So here we provide the missing zero. */
4717 *displacement_string_end = '0';
4719 #endif
4720 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL, &types);
4721 if (gotfree_input_line)
4722 input_line_pointer = gotfree_input_line;
4724 exp_seg = expression (exp);
4726 SKIP_WHITESPACE ();
4727 if (*input_line_pointer)
4728 as_bad (_("junk `%s' after expression"), input_line_pointer);
4729 #if GCC_ASM_O_HACK
4730 RESTORE_END_STRING (disp_end + 1);
4731 #endif
4732 RESTORE_END_STRING (disp_end);
4733 input_line_pointer = save_input_line_pointer;
4734 if (gotfree_input_line)
4735 free (gotfree_input_line);
4737 /* We do this to make sure that the section symbol is in
4738 the symbol table. We will ultimately change the relocation
4739 to be relative to the beginning of the section. */
4740 if (i.reloc[this_operand] == BFD_RELOC_386_GOTOFF
4741 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
4742 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTOFF64)
4744 if (exp->X_op != O_symbol)
4746 as_bad (_("bad expression used with @%s"),
4747 (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
4748 ? "GOTPCREL"
4749 : "GOTOFF"));
4750 return 0;
4753 if (S_IS_LOCAL (exp->X_add_symbol)
4754 && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
4755 section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
4756 exp->X_op = O_subtract;
4757 exp->X_op_symbol = GOT_symbol;
4758 if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
4759 i.reloc[this_operand] = BFD_RELOC_32_PCREL;
4760 else if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTOFF64)
4761 i.reloc[this_operand] = BFD_RELOC_64;
4762 else
4763 i.reloc[this_operand] = BFD_RELOC_32;
4766 if (exp->X_op == O_absent || exp->X_op == O_big)
4768 /* Missing or bad expr becomes absolute 0. */
4769 as_bad (_("missing or invalid displacement expression `%s' taken as 0"),
4770 disp_start);
4771 exp->X_op = O_constant;
4772 exp->X_add_number = 0;
4773 exp->X_add_symbol = (symbolS *) 0;
4774 exp->X_op_symbol = (symbolS *) 0;
4777 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4778 if (exp->X_op != O_constant
4779 && OUTPUT_FLAVOR == bfd_target_aout_flavour
4780 && exp_seg != absolute_section
4781 && exp_seg != text_section
4782 && exp_seg != data_section
4783 && exp_seg != bss_section
4784 && exp_seg != undefined_section
4785 && !bfd_is_com_section (exp_seg))
4787 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
4788 return 0;
4790 #endif
4792 if (!(i.types[this_operand] & ~Disp))
4793 i.types[this_operand] &= types;
4795 return 1;
4798 /* Make sure the memory operand we've been dealt is valid.
4799 Return 1 on success, 0 on a failure. */
4801 static int
4802 i386_index_check (const char *operand_string)
4804 int ok;
4805 #if INFER_ADDR_PREFIX
4806 int fudged = 0;
4808 tryprefix:
4809 #endif
4810 ok = 1;
4811 if ((current_templates->start->cpu_flags & CpuSVME)
4812 && current_templates->end[-1].operand_types[0] == AnyMem)
4814 /* Memory operands of SVME insns are special in that they only allow
4815 rAX as their memory address and ignore any segment override. */
4816 unsigned RegXX;
4818 /* SKINIT is even more restrictive: it always requires EAX. */
4819 if (strcmp (current_templates->start->name, "skinit") == 0)
4820 RegXX = Reg32;
4821 else if (flag_code == CODE_64BIT)
4822 RegXX = i.prefix[ADDR_PREFIX] == 0 ? Reg64 : Reg32;
4823 else
4824 RegXX = ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0)
4825 ? Reg16
4826 : Reg32);
4827 if (!i.base_reg
4828 || !(i.base_reg->reg_type & Acc)
4829 || !(i.base_reg->reg_type & RegXX)
4830 || i.index_reg
4831 || (i.types[0] & Disp))
4832 ok = 0;
4834 else if (flag_code == CODE_64BIT)
4836 unsigned RegXX = (i.prefix[ADDR_PREFIX] == 0 ? Reg64 : Reg32);
4838 if ((i.base_reg
4839 && ((i.base_reg->reg_type & RegXX) == 0)
4840 && (i.base_reg->reg_type != BaseIndex
4841 || i.index_reg))
4842 || (i.index_reg
4843 && ((i.index_reg->reg_type & (RegXX | BaseIndex))
4844 != (RegXX | BaseIndex))))
4845 ok = 0;
4847 else
4849 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
4851 /* 16bit checks. */
4852 if ((i.base_reg
4853 && ((i.base_reg->reg_type & (Reg16 | BaseIndex | RegRex))
4854 != (Reg16 | BaseIndex)))
4855 || (i.index_reg
4856 && (((i.index_reg->reg_type & (Reg16 | BaseIndex))
4857 != (Reg16 | BaseIndex))
4858 || !(i.base_reg
4859 && i.base_reg->reg_num < 6
4860 && i.index_reg->reg_num >= 6
4861 && i.log2_scale_factor == 0))))
4862 ok = 0;
4864 else
4866 /* 32bit checks. */
4867 if ((i.base_reg
4868 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
4869 || (i.index_reg
4870 && ((i.index_reg->reg_type & (Reg32 | BaseIndex | RegRex))
4871 != (Reg32 | BaseIndex))))
4872 ok = 0;
4875 if (!ok)
4877 #if INFER_ADDR_PREFIX
4878 if (i.prefix[ADDR_PREFIX] == 0)
4880 i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
4881 i.prefixes += 1;
4882 /* Change the size of any displacement too. At most one of
4883 Disp16 or Disp32 is set.
4884 FIXME. There doesn't seem to be any real need for separate
4885 Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
4886 Removing them would probably clean up the code quite a lot. */
4887 if (flag_code != CODE_64BIT
4888 && (i.types[this_operand] & (Disp16 | Disp32)))
4889 i.types[this_operand] ^= (Disp16 | Disp32);
4890 fudged = 1;
4891 goto tryprefix;
4893 if (fudged)
4894 as_bad (_("`%s' is not a valid base/index expression"),
4895 operand_string);
4896 else
4897 #endif
4898 as_bad (_("`%s' is not a valid %s bit base/index expression"),
4899 operand_string,
4900 flag_code_names[flag_code]);
4902 return ok;
4905 /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
4906 on error. */
4908 static int
4909 i386_operand (char *operand_string)
4911 const reg_entry *r;
4912 char *end_op;
4913 char *op_string = operand_string;
4915 if (is_space_char (*op_string))
4916 ++op_string;
4918 /* We check for an absolute prefix (differentiating,
4919 for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
4920 if (*op_string == ABSOLUTE_PREFIX)
4922 ++op_string;
4923 if (is_space_char (*op_string))
4924 ++op_string;
4925 i.types[this_operand] |= JumpAbsolute;
4928 /* Check if operand is a register. */
4929 if ((r = parse_register (op_string, &end_op)) != NULL)
4931 /* Check for a segment override by searching for ':' after a
4932 segment register. */
4933 op_string = end_op;
4934 if (is_space_char (*op_string))
4935 ++op_string;
4936 if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
4938 switch (r->reg_num)
4940 case 0:
4941 i.seg[i.mem_operands] = &es;
4942 break;
4943 case 1:
4944 i.seg[i.mem_operands] = &cs;
4945 break;
4946 case 2:
4947 i.seg[i.mem_operands] = &ss;
4948 break;
4949 case 3:
4950 i.seg[i.mem_operands] = &ds;
4951 break;
4952 case 4:
4953 i.seg[i.mem_operands] = &fs;
4954 break;
4955 case 5:
4956 i.seg[i.mem_operands] = &gs;
4957 break;
4960 /* Skip the ':' and whitespace. */
4961 ++op_string;
4962 if (is_space_char (*op_string))
4963 ++op_string;
4965 if (!is_digit_char (*op_string)
4966 && !is_identifier_char (*op_string)
4967 && *op_string != '('
4968 && *op_string != ABSOLUTE_PREFIX)
4970 as_bad (_("bad memory operand `%s'"), op_string);
4971 return 0;
4973 /* Handle case of %es:*foo. */
4974 if (*op_string == ABSOLUTE_PREFIX)
4976 ++op_string;
4977 if (is_space_char (*op_string))
4978 ++op_string;
4979 i.types[this_operand] |= JumpAbsolute;
4981 goto do_memory_reference;
4983 if (*op_string)
4985 as_bad (_("junk `%s' after register"), op_string);
4986 return 0;
4988 i.types[this_operand] |= r->reg_type & ~BaseIndex;
4989 i.op[this_operand].regs = r;
4990 i.reg_operands++;
4992 else if (*op_string == REGISTER_PREFIX)
4994 as_bad (_("bad register name `%s'"), op_string);
4995 return 0;
4997 else if (*op_string == IMMEDIATE_PREFIX)
4999 ++op_string;
5000 if (i.types[this_operand] & JumpAbsolute)
5002 as_bad (_("immediate operand illegal with absolute jump"));
5003 return 0;
5005 if (!i386_immediate (op_string))
5006 return 0;
5008 else if (is_digit_char (*op_string)
5009 || is_identifier_char (*op_string)
5010 || *op_string == '(')
5012 /* This is a memory reference of some sort. */
5013 char *base_string;
5015 /* Start and end of displacement string expression (if found). */
5016 char *displacement_string_start;
5017 char *displacement_string_end;
5019 do_memory_reference:
5020 if ((i.mem_operands == 1
5021 && (current_templates->start->opcode_modifier & IsString) == 0)
5022 || i.mem_operands == 2)
5024 as_bad (_("too many memory references for `%s'"),
5025 current_templates->start->name);
5026 return 0;
5029 /* Check for base index form. We detect the base index form by
5030 looking for an ')' at the end of the operand, searching
5031 for the '(' matching it, and finding a REGISTER_PREFIX or ','
5032 after the '('. */
5033 base_string = op_string + strlen (op_string);
5035 --base_string;
5036 if (is_space_char (*base_string))
5037 --base_string;
5039 /* If we only have a displacement, set-up for it to be parsed later. */
5040 displacement_string_start = op_string;
5041 displacement_string_end = base_string + 1;
5043 if (*base_string == ')')
5045 char *temp_string;
5046 unsigned int parens_balanced = 1;
5047 /* We've already checked that the number of left & right ()'s are
5048 equal, so this loop will not be infinite. */
5051 base_string--;
5052 if (*base_string == ')')
5053 parens_balanced++;
5054 if (*base_string == '(')
5055 parens_balanced--;
5057 while (parens_balanced);
5059 temp_string = base_string;
5061 /* Skip past '(' and whitespace. */
5062 ++base_string;
5063 if (is_space_char (*base_string))
5064 ++base_string;
5066 if (*base_string == ','
5067 || ((i.base_reg = parse_register (base_string, &end_op))
5068 != NULL))
5070 displacement_string_end = temp_string;
5072 i.types[this_operand] |= BaseIndex;
5074 if (i.base_reg)
5076 base_string = end_op;
5077 if (is_space_char (*base_string))
5078 ++base_string;
5081 /* There may be an index reg or scale factor here. */
5082 if (*base_string == ',')
5084 ++base_string;
5085 if (is_space_char (*base_string))
5086 ++base_string;
5088 if ((i.index_reg = parse_register (base_string, &end_op))
5089 != NULL)
5091 base_string = end_op;
5092 if (is_space_char (*base_string))
5093 ++base_string;
5094 if (*base_string == ',')
5096 ++base_string;
5097 if (is_space_char (*base_string))
5098 ++base_string;
5100 else if (*base_string != ')')
5102 as_bad (_("expecting `,' or `)' "
5103 "after index register in `%s'"),
5104 operand_string);
5105 return 0;
5108 else if (*base_string == REGISTER_PREFIX)
5110 as_bad (_("bad register name `%s'"), base_string);
5111 return 0;
5114 /* Check for scale factor. */
5115 if (*base_string != ')')
5117 char *end_scale = i386_scale (base_string);
5119 if (!end_scale)
5120 return 0;
5122 base_string = end_scale;
5123 if (is_space_char (*base_string))
5124 ++base_string;
5125 if (*base_string != ')')
5127 as_bad (_("expecting `)' "
5128 "after scale factor in `%s'"),
5129 operand_string);
5130 return 0;
5133 else if (!i.index_reg)
5135 as_bad (_("expecting index register or scale factor "
5136 "after `,'; got '%c'"),
5137 *base_string);
5138 return 0;
5141 else if (*base_string != ')')
5143 as_bad (_("expecting `,' or `)' "
5144 "after base register in `%s'"),
5145 operand_string);
5146 return 0;
5149 else if (*base_string == REGISTER_PREFIX)
5151 as_bad (_("bad register name `%s'"), base_string);
5152 return 0;
5156 /* If there's an expression beginning the operand, parse it,
5157 assuming displacement_string_start and
5158 displacement_string_end are meaningful. */
5159 if (displacement_string_start != displacement_string_end)
5161 if (!i386_displacement (displacement_string_start,
5162 displacement_string_end))
5163 return 0;
5166 /* Special case for (%dx) while doing input/output op. */
5167 if (i.base_reg
5168 && i.base_reg->reg_type == (Reg16 | InOutPortReg)
5169 && i.index_reg == 0
5170 && i.log2_scale_factor == 0
5171 && i.seg[i.mem_operands] == 0
5172 && (i.types[this_operand] & Disp) == 0)
5174 i.types[this_operand] = InOutPortReg;
5175 return 1;
5178 if (i386_index_check (operand_string) == 0)
5179 return 0;
5180 i.mem_operands++;
5182 else
5184 /* It's not a memory operand; argh! */
5185 as_bad (_("invalid char %s beginning operand %d `%s'"),
5186 output_invalid (*op_string),
5187 this_operand + 1,
5188 op_string);
5189 return 0;
5191 return 1; /* Normal return. */
5194 /* md_estimate_size_before_relax()
5196 Called just before relax() for rs_machine_dependent frags. The x86
5197 assembler uses these frags to handle variable size jump
5198 instructions.
5200 Any symbol that is now undefined will not become defined.
5201 Return the correct fr_subtype in the frag.
5202 Return the initial "guess for variable size of frag" to caller.
5203 The guess is actually the growth beyond the fixed part. Whatever
5204 we do to grow the fixed or variable part contributes to our
5205 returned value. */
5208 md_estimate_size_before_relax (fragP, segment)
5209 fragS *fragP;
5210 segT segment;
5212 /* We've already got fragP->fr_subtype right; all we have to do is
5213 check for un-relaxable symbols. On an ELF system, we can't relax
5214 an externally visible symbol, because it may be overridden by a
5215 shared library. */
5216 if (S_GET_SEGMENT (fragP->fr_symbol) != segment
5217 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5218 || (IS_ELF
5219 && (S_IS_EXTERNAL (fragP->fr_symbol)
5220 || S_IS_WEAK (fragP->fr_symbol)))
5221 #endif
5224 /* Symbol is undefined in this segment, or we need to keep a
5225 reloc so that weak symbols can be overridden. */
5226 int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
5227 enum bfd_reloc_code_real reloc_type;
5228 unsigned char *opcode;
5229 int old_fr_fix;
5231 if (fragP->fr_var != NO_RELOC)
5232 reloc_type = fragP->fr_var;
5233 else if (size == 2)
5234 reloc_type = BFD_RELOC_16_PCREL;
5235 else
5236 reloc_type = BFD_RELOC_32_PCREL;
5238 old_fr_fix = fragP->fr_fix;
5239 opcode = (unsigned char *) fragP->fr_opcode;
5241 switch (TYPE_FROM_RELAX_STATE (fragP->fr_subtype))
5243 case UNCOND_JUMP:
5244 /* Make jmp (0xeb) a (d)word displacement jump. */
5245 opcode[0] = 0xe9;
5246 fragP->fr_fix += size;
5247 fix_new (fragP, old_fr_fix, size,
5248 fragP->fr_symbol,
5249 fragP->fr_offset, 1,
5250 reloc_type);
5251 break;
5253 case COND_JUMP86:
5254 if (size == 2
5255 && (!no_cond_jump_promotion || fragP->fr_var != NO_RELOC))
5257 /* Negate the condition, and branch past an
5258 unconditional jump. */
5259 opcode[0] ^= 1;
5260 opcode[1] = 3;
5261 /* Insert an unconditional jump. */
5262 opcode[2] = 0xe9;
5263 /* We added two extra opcode bytes, and have a two byte
5264 offset. */
5265 fragP->fr_fix += 2 + 2;
5266 fix_new (fragP, old_fr_fix + 2, 2,
5267 fragP->fr_symbol,
5268 fragP->fr_offset, 1,
5269 reloc_type);
5270 break;
5272 /* Fall through. */
5274 case COND_JUMP:
5275 if (no_cond_jump_promotion && fragP->fr_var == NO_RELOC)
5277 fixS *fixP;
5279 fragP->fr_fix += 1;
5280 fixP = fix_new (fragP, old_fr_fix, 1,
5281 fragP->fr_symbol,
5282 fragP->fr_offset, 1,
5283 BFD_RELOC_8_PCREL);
5284 fixP->fx_signed = 1;
5285 break;
5288 /* This changes the byte-displacement jump 0x7N
5289 to the (d)word-displacement jump 0x0f,0x8N. */
5290 opcode[1] = opcode[0] + 0x10;
5291 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
5292 /* We've added an opcode byte. */
5293 fragP->fr_fix += 1 + size;
5294 fix_new (fragP, old_fr_fix + 1, size,
5295 fragP->fr_symbol,
5296 fragP->fr_offset, 1,
5297 reloc_type);
5298 break;
5300 default:
5301 BAD_CASE (fragP->fr_subtype);
5302 break;
5304 frag_wane (fragP);
5305 return fragP->fr_fix - old_fr_fix;
5308 /* Guess size depending on current relax state. Initially the relax
5309 state will correspond to a short jump and we return 1, because
5310 the variable part of the frag (the branch offset) is one byte
5311 long. However, we can relax a section more than once and in that
5312 case we must either set fr_subtype back to the unrelaxed state,
5313 or return the value for the appropriate branch. */
5314 return md_relax_table[fragP->fr_subtype].rlx_length;
5317 /* Called after relax() is finished.
5319 In: Address of frag.
5320 fr_type == rs_machine_dependent.
5321 fr_subtype is what the address relaxed to.
5323 Out: Any fixSs and constants are set up.
5324 Caller will turn frag into a ".space 0". */
5326 void
5327 md_convert_frag (abfd, sec, fragP)
5328 bfd *abfd ATTRIBUTE_UNUSED;
5329 segT sec ATTRIBUTE_UNUSED;
5330 fragS *fragP;
5332 unsigned char *opcode;
5333 unsigned char *where_to_put_displacement = NULL;
5334 offsetT target_address;
5335 offsetT opcode_address;
5336 unsigned int extension = 0;
5337 offsetT displacement_from_opcode_start;
5339 opcode = (unsigned char *) fragP->fr_opcode;
5341 /* Address we want to reach in file space. */
5342 target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
5344 /* Address opcode resides at in file space. */
5345 opcode_address = fragP->fr_address + fragP->fr_fix;
5347 /* Displacement from opcode start to fill into instruction. */
5348 displacement_from_opcode_start = target_address - opcode_address;
5350 if ((fragP->fr_subtype & BIG) == 0)
5352 /* Don't have to change opcode. */
5353 extension = 1; /* 1 opcode + 1 displacement */
5354 where_to_put_displacement = &opcode[1];
5356 else
5358 if (no_cond_jump_promotion
5359 && TYPE_FROM_RELAX_STATE (fragP->fr_subtype) != UNCOND_JUMP)
5360 as_warn_where (fragP->fr_file, fragP->fr_line,
5361 _("long jump required"));
5363 switch (fragP->fr_subtype)
5365 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
5366 extension = 4; /* 1 opcode + 4 displacement */
5367 opcode[0] = 0xe9;
5368 where_to_put_displacement = &opcode[1];
5369 break;
5371 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
5372 extension = 2; /* 1 opcode + 2 displacement */
5373 opcode[0] = 0xe9;
5374 where_to_put_displacement = &opcode[1];
5375 break;
5377 case ENCODE_RELAX_STATE (COND_JUMP, BIG):
5378 case ENCODE_RELAX_STATE (COND_JUMP86, BIG):
5379 extension = 5; /* 2 opcode + 4 displacement */
5380 opcode[1] = opcode[0] + 0x10;
5381 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
5382 where_to_put_displacement = &opcode[2];
5383 break;
5385 case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
5386 extension = 3; /* 2 opcode + 2 displacement */
5387 opcode[1] = opcode[0] + 0x10;
5388 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
5389 where_to_put_displacement = &opcode[2];
5390 break;
5392 case ENCODE_RELAX_STATE (COND_JUMP86, BIG16):
5393 extension = 4;
5394 opcode[0] ^= 1;
5395 opcode[1] = 3;
5396 opcode[2] = 0xe9;
5397 where_to_put_displacement = &opcode[3];
5398 break;
5400 default:
5401 BAD_CASE (fragP->fr_subtype);
5402 break;
5406 /* If size if less then four we are sure that the operand fits,
5407 but if it's 4, then it could be that the displacement is larger
5408 then -/+ 2GB. */
5409 if (DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype) == 4
5410 && object_64bit
5411 && ((addressT) (displacement_from_opcode_start - extension
5412 + ((addressT) 1 << 31))
5413 > (((addressT) 2 << 31) - 1)))
5415 as_bad_where (fragP->fr_file, fragP->fr_line,
5416 _("jump target out of range"));
5417 /* Make us emit 0. */
5418 displacement_from_opcode_start = extension;
5420 /* Now put displacement after opcode. */
5421 md_number_to_chars ((char *) where_to_put_displacement,
5422 (valueT) (displacement_from_opcode_start - extension),
5423 DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
5424 fragP->fr_fix += extension;
5427 /* Size of byte displacement jmp. */
5428 int md_short_jump_size = 2;
5430 /* Size of dword displacement jmp. */
5431 int md_long_jump_size = 5;
5433 void
5434 md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
5435 char *ptr;
5436 addressT from_addr, to_addr;
5437 fragS *frag ATTRIBUTE_UNUSED;
5438 symbolS *to_symbol ATTRIBUTE_UNUSED;
5440 offsetT offset;
5442 offset = to_addr - (from_addr + 2);
5443 /* Opcode for byte-disp jump. */
5444 md_number_to_chars (ptr, (valueT) 0xeb, 1);
5445 md_number_to_chars (ptr + 1, (valueT) offset, 1);
5448 void
5449 md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
5450 char *ptr;
5451 addressT from_addr, to_addr;
5452 fragS *frag ATTRIBUTE_UNUSED;
5453 symbolS *to_symbol ATTRIBUTE_UNUSED;
5455 offsetT offset;
5457 offset = to_addr - (from_addr + 5);
5458 md_number_to_chars (ptr, (valueT) 0xe9, 1);
5459 md_number_to_chars (ptr + 1, (valueT) offset, 4);
5462 /* Apply a fixup (fixS) to segment data, once it has been determined
5463 by our caller that we have all the info we need to fix it up.
5465 On the 386, immediates, displacements, and data pointers are all in
5466 the same (little-endian) format, so we don't need to care about which
5467 we are handling. */
5469 void
5470 md_apply_fix (fixP, valP, seg)
5471 /* The fix we're to put in. */
5472 fixS *fixP;
5473 /* Pointer to the value of the bits. */
5474 valueT *valP;
5475 /* Segment fix is from. */
5476 segT seg ATTRIBUTE_UNUSED;
5478 char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
5479 valueT value = *valP;
5481 #if !defined (TE_Mach)
5482 if (fixP->fx_pcrel)
5484 switch (fixP->fx_r_type)
5486 default:
5487 break;
5489 case BFD_RELOC_64:
5490 fixP->fx_r_type = BFD_RELOC_64_PCREL;
5491 break;
5492 case BFD_RELOC_32:
5493 case BFD_RELOC_X86_64_32S:
5494 fixP->fx_r_type = BFD_RELOC_32_PCREL;
5495 break;
5496 case BFD_RELOC_16:
5497 fixP->fx_r_type = BFD_RELOC_16_PCREL;
5498 break;
5499 case BFD_RELOC_8:
5500 fixP->fx_r_type = BFD_RELOC_8_PCREL;
5501 break;
5505 if (fixP->fx_addsy != NULL
5506 && (fixP->fx_r_type == BFD_RELOC_32_PCREL
5507 || fixP->fx_r_type == BFD_RELOC_64_PCREL
5508 || fixP->fx_r_type == BFD_RELOC_16_PCREL
5509 || fixP->fx_r_type == BFD_RELOC_8_PCREL)
5510 && !use_rela_relocations)
5512 /* This is a hack. There should be a better way to handle this.
5513 This covers for the fact that bfd_install_relocation will
5514 subtract the current location (for partial_inplace, PC relative
5515 relocations); see more below. */
5516 #ifndef OBJ_AOUT
5517 if (IS_ELF
5518 #ifdef TE_PE
5519 || OUTPUT_FLAVOR == bfd_target_coff_flavour
5520 #endif
5522 value += fixP->fx_where + fixP->fx_frag->fr_address;
5523 #endif
5524 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5525 if (IS_ELF)
5527 segT sym_seg = S_GET_SEGMENT (fixP->fx_addsy);
5529 if ((sym_seg == seg
5530 || (symbol_section_p (fixP->fx_addsy)
5531 && sym_seg != absolute_section))
5532 && !generic_force_reloc (fixP))
5534 /* Yes, we add the values in twice. This is because
5535 bfd_install_relocation subtracts them out again. I think
5536 bfd_install_relocation is broken, but I don't dare change
5537 it. FIXME. */
5538 value += fixP->fx_where + fixP->fx_frag->fr_address;
5541 #endif
5542 #if defined (OBJ_COFF) && defined (TE_PE)
5543 /* For some reason, the PE format does not store a
5544 section address offset for a PC relative symbol. */
5545 if (S_GET_SEGMENT (fixP->fx_addsy) != seg
5546 || S_IS_WEAK (fixP->fx_addsy))
5547 value += md_pcrel_from (fixP);
5548 #endif
5551 /* Fix a few things - the dynamic linker expects certain values here,
5552 and we must not disappoint it. */
5553 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5554 if (IS_ELF && fixP->fx_addsy)
5555 switch (fixP->fx_r_type)
5557 case BFD_RELOC_386_PLT32:
5558 case BFD_RELOC_X86_64_PLT32:
5559 /* Make the jump instruction point to the address of the operand. At
5560 runtime we merely add the offset to the actual PLT entry. */
5561 value = -4;
5562 break;
5564 case BFD_RELOC_386_TLS_GD:
5565 case BFD_RELOC_386_TLS_LDM:
5566 case BFD_RELOC_386_TLS_IE_32:
5567 case BFD_RELOC_386_TLS_IE:
5568 case BFD_RELOC_386_TLS_GOTIE:
5569 case BFD_RELOC_386_TLS_GOTDESC:
5570 case BFD_RELOC_X86_64_TLSGD:
5571 case BFD_RELOC_X86_64_TLSLD:
5572 case BFD_RELOC_X86_64_GOTTPOFF:
5573 case BFD_RELOC_X86_64_GOTPC32_TLSDESC:
5574 value = 0; /* Fully resolved at runtime. No addend. */
5575 /* Fallthrough */
5576 case BFD_RELOC_386_TLS_LE:
5577 case BFD_RELOC_386_TLS_LDO_32:
5578 case BFD_RELOC_386_TLS_LE_32:
5579 case BFD_RELOC_X86_64_DTPOFF32:
5580 case BFD_RELOC_X86_64_DTPOFF64:
5581 case BFD_RELOC_X86_64_TPOFF32:
5582 case BFD_RELOC_X86_64_TPOFF64:
5583 S_SET_THREAD_LOCAL (fixP->fx_addsy);
5584 break;
5586 case BFD_RELOC_386_TLS_DESC_CALL:
5587 case BFD_RELOC_X86_64_TLSDESC_CALL:
5588 value = 0; /* Fully resolved at runtime. No addend. */
5589 S_SET_THREAD_LOCAL (fixP->fx_addsy);
5590 fixP->fx_done = 0;
5591 return;
5593 case BFD_RELOC_386_GOT32:
5594 case BFD_RELOC_X86_64_GOT32:
5595 value = 0; /* Fully resolved at runtime. No addend. */
5596 break;
5598 case BFD_RELOC_VTABLE_INHERIT:
5599 case BFD_RELOC_VTABLE_ENTRY:
5600 fixP->fx_done = 0;
5601 return;
5603 default:
5604 break;
5606 #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
5607 *valP = value;
5608 #endif /* !defined (TE_Mach) */
5610 /* Are we finished with this relocation now? */
5611 if (fixP->fx_addsy == NULL)
5612 fixP->fx_done = 1;
5613 else if (use_rela_relocations)
5615 fixP->fx_no_overflow = 1;
5616 /* Remember value for tc_gen_reloc. */
5617 fixP->fx_addnumber = value;
5618 value = 0;
5621 md_number_to_chars (p, value, fixP->fx_size);
5624 #define MAX_LITTLENUMS 6
5626 /* Turn the string pointed to by litP into a floating point constant
5627 of type TYPE, and emit the appropriate bytes. The number of
5628 LITTLENUMS emitted is stored in *SIZEP. An error message is
5629 returned, or NULL on OK. */
5631 char *
5632 md_atof (type, litP, sizeP)
5633 int type;
5634 char *litP;
5635 int *sizeP;
5637 int prec;
5638 LITTLENUM_TYPE words[MAX_LITTLENUMS];
5639 LITTLENUM_TYPE *wordP;
5640 char *t;
5642 switch (type)
5644 case 'f':
5645 case 'F':
5646 prec = 2;
5647 break;
5649 case 'd':
5650 case 'D':
5651 prec = 4;
5652 break;
5654 case 'x':
5655 case 'X':
5656 prec = 5;
5657 break;
5659 default:
5660 *sizeP = 0;
5661 return _("Bad call to md_atof ()");
5663 t = atof_ieee (input_line_pointer, type, words);
5664 if (t)
5665 input_line_pointer = t;
5667 *sizeP = prec * sizeof (LITTLENUM_TYPE);
5668 /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
5669 the bigendian 386. */
5670 for (wordP = words + prec - 1; prec--;)
5672 md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
5673 litP += sizeof (LITTLENUM_TYPE);
5675 return 0;
5678 static char output_invalid_buf[sizeof (unsigned char) * 2 + 6];
5680 static char *
5681 output_invalid (int c)
5683 if (ISPRINT (c))
5684 snprintf (output_invalid_buf, sizeof (output_invalid_buf),
5685 "'%c'", c);
5686 else
5687 snprintf (output_invalid_buf, sizeof (output_invalid_buf),
5688 "(0x%x)", (unsigned char) c);
5689 return output_invalid_buf;
5692 /* REG_STRING starts *before* REGISTER_PREFIX. */
5694 static const reg_entry *
5695 parse_real_register (char *reg_string, char **end_op)
5697 char *s = reg_string;
5698 char *p;
5699 char reg_name_given[MAX_REG_NAME_SIZE + 1];
5700 const reg_entry *r;
5702 /* Skip possible REGISTER_PREFIX and possible whitespace. */
5703 if (*s == REGISTER_PREFIX)
5704 ++s;
5706 if (is_space_char (*s))
5707 ++s;
5709 p = reg_name_given;
5710 while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
5712 if (p >= reg_name_given + MAX_REG_NAME_SIZE)
5713 return (const reg_entry *) NULL;
5714 s++;
5717 /* For naked regs, make sure that we are not dealing with an identifier.
5718 This prevents confusing an identifier like `eax_var' with register
5719 `eax'. */
5720 if (allow_naked_reg && identifier_chars[(unsigned char) *s])
5721 return (const reg_entry *) NULL;
5723 *end_op = s;
5725 r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
5727 /* Handle floating point regs, allowing spaces in the (i) part. */
5728 if (r == i386_regtab /* %st is first entry of table */)
5730 if (is_space_char (*s))
5731 ++s;
5732 if (*s == '(')
5734 ++s;
5735 if (is_space_char (*s))
5736 ++s;
5737 if (*s >= '0' && *s <= '7')
5739 r = &i386_float_regtab[*s - '0'];
5740 ++s;
5741 if (is_space_char (*s))
5742 ++s;
5743 if (*s == ')')
5745 *end_op = s + 1;
5746 return r;
5749 /* We have "%st(" then garbage. */
5750 return (const reg_entry *) NULL;
5754 if (r != NULL
5755 && ((r->reg_flags & (RegRex64 | RegRex)) | (r->reg_type & Reg64)) != 0
5756 && (r->reg_type != Control || !(cpu_arch_flags & CpuSledgehammer))
5757 && flag_code != CODE_64BIT)
5758 return (const reg_entry *) NULL;
5760 return r;
5763 /* REG_STRING starts *before* REGISTER_PREFIX. */
5765 static const reg_entry *
5766 parse_register (char *reg_string, char **end_op)
5768 const reg_entry *r;
5770 if (*reg_string == REGISTER_PREFIX || allow_naked_reg)
5771 r = parse_real_register (reg_string, end_op);
5772 else
5773 r = NULL;
5774 if (!r)
5776 char *save = input_line_pointer;
5777 char c;
5778 symbolS *symbolP;
5780 input_line_pointer = reg_string;
5781 c = get_symbol_end ();
5782 symbolP = symbol_find (reg_string);
5783 if (symbolP && S_GET_SEGMENT (symbolP) == reg_section)
5785 const expressionS *e = symbol_get_value_expression (symbolP);
5787 know (e->X_op == O_register);
5788 know (e->X_add_number >= 0
5789 && (valueT) e->X_add_number < ARRAY_SIZE (i386_regtab));
5790 r = i386_regtab + e->X_add_number;
5791 *end_op = input_line_pointer;
5793 *input_line_pointer = c;
5794 input_line_pointer = save;
5796 return r;
5800 i386_parse_name (char *name, expressionS *e, char *nextcharP)
5802 const reg_entry *r;
5803 char *end = input_line_pointer;
5805 *end = *nextcharP;
5806 r = parse_register (name, &input_line_pointer);
5807 if (r && end <= input_line_pointer)
5809 *nextcharP = *input_line_pointer;
5810 *input_line_pointer = 0;
5811 e->X_op = O_register;
5812 e->X_add_number = r - i386_regtab;
5813 return 1;
5815 input_line_pointer = end;
5816 *end = 0;
5817 return 0;
5820 void
5821 md_operand (expressionS *e)
5823 if (*input_line_pointer == REGISTER_PREFIX)
5825 char *end;
5826 const reg_entry *r = parse_real_register (input_line_pointer, &end);
5828 if (r)
5830 e->X_op = O_register;
5831 e->X_add_number = r - i386_regtab;
5832 input_line_pointer = end;
5838 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5839 const char *md_shortopts = "kVQ:sqn";
5840 #else
5841 const char *md_shortopts = "qn";
5842 #endif
5844 #define OPTION_32 (OPTION_MD_BASE + 0)
5845 #define OPTION_64 (OPTION_MD_BASE + 1)
5846 #define OPTION_DIVIDE (OPTION_MD_BASE + 2)
5847 #define OPTION_MARCH (OPTION_MD_BASE + 3)
5848 #define OPTION_MTUNE (OPTION_MD_BASE + 4)
5850 struct option md_longopts[] =
5852 {"32", no_argument, NULL, OPTION_32},
5853 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) || defined(TE_PEP)
5854 {"64", no_argument, NULL, OPTION_64},
5855 #endif
5856 {"divide", no_argument, NULL, OPTION_DIVIDE},
5857 {"march", required_argument, NULL, OPTION_MARCH},
5858 {"mtune", required_argument, NULL, OPTION_MTUNE},
5859 {NULL, no_argument, NULL, 0}
5861 size_t md_longopts_size = sizeof (md_longopts);
5864 md_parse_option (int c, char *arg)
5866 unsigned int i;
5868 switch (c)
5870 case 'n':
5871 optimize_align_code = 0;
5872 break;
5874 case 'q':
5875 quiet_warnings = 1;
5876 break;
5878 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5879 /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
5880 should be emitted or not. FIXME: Not implemented. */
5881 case 'Q':
5882 break;
5884 /* -V: SVR4 argument to print version ID. */
5885 case 'V':
5886 print_version_id ();
5887 break;
5889 /* -k: Ignore for FreeBSD compatibility. */
5890 case 'k':
5891 break;
5893 case 's':
5894 /* -s: On i386 Solaris, this tells the native assembler to use
5895 .stab instead of .stab.excl. We always use .stab anyhow. */
5896 break;
5897 #endif
5898 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) || defined(TE_PEP)
5899 case OPTION_64:
5901 const char **list, **l;
5903 list = bfd_target_list ();
5904 for (l = list; *l != NULL; l++)
5905 if (CONST_STRNEQ (*l, "elf64-x86-64")
5906 || strcmp (*l, "coff-x86-64") == 0
5907 || strcmp (*l, "pe-x86-64") == 0
5908 || strcmp (*l, "pei-x86-64") == 0)
5910 default_arch = "x86_64";
5911 break;
5913 if (*l == NULL)
5914 as_fatal (_("No compiled in support for x86_64"));
5915 free (list);
5917 break;
5918 #endif
5920 case OPTION_32:
5921 default_arch = "i386";
5922 break;
5924 case OPTION_DIVIDE:
5925 #ifdef SVR4_COMMENT_CHARS
5927 char *n, *t;
5928 const char *s;
5930 n = (char *) xmalloc (strlen (i386_comment_chars) + 1);
5931 t = n;
5932 for (s = i386_comment_chars; *s != '\0'; s++)
5933 if (*s != '/')
5934 *t++ = *s;
5935 *t = '\0';
5936 i386_comment_chars = n;
5938 #endif
5939 break;
5941 case OPTION_MARCH:
5942 if (*arg == '.')
5943 as_fatal (_("Invalid -march= option: `%s'"), arg);
5944 for (i = 0; i < ARRAY_SIZE (cpu_arch); i++)
5946 if (strcmp (arg, cpu_arch [i].name) == 0)
5948 cpu_arch_isa = cpu_arch[i].type;
5949 cpu_arch_isa_flags = cpu_arch[i].flags;
5950 if (!cpu_arch_tune_set)
5952 cpu_arch_tune = cpu_arch_isa;
5953 cpu_arch_tune_flags = cpu_arch_isa_flags;
5955 break;
5958 if (i >= ARRAY_SIZE (cpu_arch))
5959 as_fatal (_("Invalid -march= option: `%s'"), arg);
5960 break;
5962 case OPTION_MTUNE:
5963 if (*arg == '.')
5964 as_fatal (_("Invalid -mtune= option: `%s'"), arg);
5965 for (i = 0; i < ARRAY_SIZE (cpu_arch); i++)
5967 if (strcmp (arg, cpu_arch [i].name) == 0)
5969 cpu_arch_tune_set = 1;
5970 cpu_arch_tune = cpu_arch [i].type;
5971 cpu_arch_tune_flags = cpu_arch[i].flags;
5972 break;
5975 if (i >= ARRAY_SIZE (cpu_arch))
5976 as_fatal (_("Invalid -mtune= option: `%s'"), arg);
5977 break;
5979 default:
5980 return 0;
5982 return 1;
5985 void
5986 md_show_usage (stream)
5987 FILE *stream;
5989 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5990 fprintf (stream, _("\
5991 -Q ignored\n\
5992 -V print assembler version number\n\
5993 -k ignored\n"));
5994 #endif
5995 fprintf (stream, _("\
5996 -n Do not optimize code alignment\n\
5997 -q quieten some warnings\n"));
5998 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5999 fprintf (stream, _("\
6000 -s ignored\n"));
6001 #endif
6002 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) || defined(TE_PEP)
6003 fprintf (stream, _("\
6004 --32/--64 generate 32bit/64bit code\n"));
6005 #endif
6006 #ifdef SVR4_COMMENT_CHARS
6007 fprintf (stream, _("\
6008 --divide do not treat `/' as a comment character\n"));
6009 #else
6010 fprintf (stream, _("\
6011 --divide ignored\n"));
6012 #endif
6013 fprintf (stream, _("\
6014 -march=CPU/-mtune=CPU generate code/optimize for CPU, where CPU is one of:\n\
6015 i386, i486, pentium, pentiumpro, pentium4, nocona,\n\
6016 core, core2, k6, athlon, k8, generic32, generic64\n"));
6020 #if defined(TE_PEP)
6021 const char *
6022 x86_64_target_format (void)
6024 if (strcmp (default_arch, "x86_64") == 0)
6026 set_code_flag (CODE_64BIT);
6027 return COFF_TARGET_FORMAT;
6029 else if (strcmp (default_arch, "i386") == 0)
6031 set_code_flag (CODE_32BIT);
6032 return "coff-i386";
6035 as_fatal (_("Unknown architecture"));
6036 return NULL;
6038 #endif
6040 #if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
6041 || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
6043 /* Pick the target format to use. */
6045 const char *
6046 i386_target_format (void)
6048 if (!strcmp (default_arch, "x86_64"))
6050 set_code_flag (CODE_64BIT);
6051 if (cpu_arch_isa_flags == 0)
6052 cpu_arch_isa_flags = Cpu186|Cpu286|Cpu386|Cpu486
6053 |Cpu586|Cpu686|CpuP4|CpuMMX|CpuMMX2
6054 |CpuSSE|CpuSSE2;
6055 if (cpu_arch_tune_flags == 0)
6056 cpu_arch_tune_flags = Cpu186|Cpu286|Cpu386|Cpu486
6057 |Cpu586|Cpu686|CpuP4|CpuMMX|CpuMMX2
6058 |CpuSSE|CpuSSE2;
6060 else if (!strcmp (default_arch, "i386"))
6062 set_code_flag (CODE_32BIT);
6063 if (cpu_arch_isa_flags == 0)
6064 cpu_arch_isa_flags = Cpu186|Cpu286|Cpu386;
6065 if (cpu_arch_tune_flags == 0)
6066 cpu_arch_tune_flags = Cpu186|Cpu286|Cpu386;
6068 else
6069 as_fatal (_("Unknown architecture"));
6070 switch (OUTPUT_FLAVOR)
6072 #ifdef OBJ_MAYBE_AOUT
6073 case bfd_target_aout_flavour:
6074 return AOUT_TARGET_FORMAT;
6075 #endif
6076 #ifdef OBJ_MAYBE_COFF
6077 case bfd_target_coff_flavour:
6078 return "coff-i386";
6079 #endif
6080 #if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF)
6081 case bfd_target_elf_flavour:
6083 if (flag_code == CODE_64BIT)
6085 object_64bit = 1;
6086 use_rela_relocations = 1;
6088 return flag_code == CODE_64BIT ? ELF_TARGET_FORMAT64 : ELF_TARGET_FORMAT;
6090 #endif
6091 default:
6092 abort ();
6093 return NULL;
6097 #endif /* OBJ_MAYBE_ more than one */
6099 #if (defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
6100 void
6101 i386_elf_emit_arch_note (void)
6103 if (IS_ELF && cpu_arch_name != NULL)
6105 char *p;
6106 asection *seg = now_seg;
6107 subsegT subseg = now_subseg;
6108 Elf_Internal_Note i_note;
6109 Elf_External_Note e_note;
6110 asection *note_secp;
6111 int len;
6113 /* Create the .note section. */
6114 note_secp = subseg_new (".note", 0);
6115 bfd_set_section_flags (stdoutput,
6116 note_secp,
6117 SEC_HAS_CONTENTS | SEC_READONLY);
6119 /* Process the arch string. */
6120 len = strlen (cpu_arch_name);
6122 i_note.namesz = len + 1;
6123 i_note.descsz = 0;
6124 i_note.type = NT_ARCH;
6125 p = frag_more (sizeof (e_note.namesz));
6126 md_number_to_chars (p, (valueT) i_note.namesz, sizeof (e_note.namesz));
6127 p = frag_more (sizeof (e_note.descsz));
6128 md_number_to_chars (p, (valueT) i_note.descsz, sizeof (e_note.descsz));
6129 p = frag_more (sizeof (e_note.type));
6130 md_number_to_chars (p, (valueT) i_note.type, sizeof (e_note.type));
6131 p = frag_more (len + 1);
6132 strcpy (p, cpu_arch_name);
6134 frag_align (2, 0, 0);
6136 subseg_set (seg, subseg);
6139 #endif
6141 symbolS *
6142 md_undefined_symbol (name)
6143 char *name;
6145 if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
6146 && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
6147 && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
6148 && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
6150 if (!GOT_symbol)
6152 if (symbol_find (name))
6153 as_bad (_("GOT already in symbol table"));
6154 GOT_symbol = symbol_new (name, undefined_section,
6155 (valueT) 0, &zero_address_frag);
6157 return GOT_symbol;
6159 return 0;
6162 /* Round up a section size to the appropriate boundary. */
6164 valueT
6165 md_section_align (segment, size)
6166 segT segment ATTRIBUTE_UNUSED;
6167 valueT size;
6169 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
6170 if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
6172 /* For a.out, force the section size to be aligned. If we don't do
6173 this, BFD will align it for us, but it will not write out the
6174 final bytes of the section. This may be a bug in BFD, but it is
6175 easier to fix it here since that is how the other a.out targets
6176 work. */
6177 int align;
6179 align = bfd_get_section_alignment (stdoutput, segment);
6180 size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
6182 #endif
6184 return size;
6187 /* On the i386, PC-relative offsets are relative to the start of the
6188 next instruction. That is, the address of the offset, plus its
6189 size, since the offset is always the last part of the insn. */
6191 long
6192 md_pcrel_from (fixS *fixP)
6194 return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
6197 #ifndef I386COFF
6199 static void
6200 s_bss (int ignore ATTRIBUTE_UNUSED)
6202 int temp;
6204 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
6205 if (IS_ELF)
6206 obj_elf_section_change_hook ();
6207 #endif
6208 temp = get_absolute_expression ();
6209 subseg_set (bss_section, (subsegT) temp);
6210 demand_empty_rest_of_line ();
6213 #endif
6215 void
6216 i386_validate_fix (fixS *fixp)
6218 if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
6220 if (fixp->fx_r_type == BFD_RELOC_32_PCREL)
6222 if (!object_64bit)
6223 abort ();
6224 fixp->fx_r_type = BFD_RELOC_X86_64_GOTPCREL;
6226 else
6228 if (!object_64bit)
6229 fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
6230 else
6231 fixp->fx_r_type = BFD_RELOC_X86_64_GOTOFF64;
6233 fixp->fx_subsy = 0;
6237 arelent *
6238 tc_gen_reloc (section, fixp)
6239 asection *section ATTRIBUTE_UNUSED;
6240 fixS *fixp;
6242 arelent *rel;
6243 bfd_reloc_code_real_type code;
6245 switch (fixp->fx_r_type)
6247 case BFD_RELOC_X86_64_PLT32:
6248 case BFD_RELOC_X86_64_GOT32:
6249 case BFD_RELOC_X86_64_GOTPCREL:
6250 case BFD_RELOC_386_PLT32:
6251 case BFD_RELOC_386_GOT32:
6252 case BFD_RELOC_386_GOTOFF:
6253 case BFD_RELOC_386_GOTPC:
6254 case BFD_RELOC_386_TLS_GD:
6255 case BFD_RELOC_386_TLS_LDM:
6256 case BFD_RELOC_386_TLS_LDO_32:
6257 case BFD_RELOC_386_TLS_IE_32:
6258 case BFD_RELOC_386_TLS_IE:
6259 case BFD_RELOC_386_TLS_GOTIE:
6260 case BFD_RELOC_386_TLS_LE_32:
6261 case BFD_RELOC_386_TLS_LE:
6262 case BFD_RELOC_386_TLS_GOTDESC:
6263 case BFD_RELOC_386_TLS_DESC_CALL:
6264 case BFD_RELOC_X86_64_TLSGD:
6265 case BFD_RELOC_X86_64_TLSLD:
6266 case BFD_RELOC_X86_64_DTPOFF32:
6267 case BFD_RELOC_X86_64_DTPOFF64:
6268 case BFD_RELOC_X86_64_GOTTPOFF:
6269 case BFD_RELOC_X86_64_TPOFF32:
6270 case BFD_RELOC_X86_64_TPOFF64:
6271 case BFD_RELOC_X86_64_GOTOFF64:
6272 case BFD_RELOC_X86_64_GOTPC32:
6273 case BFD_RELOC_X86_64_GOT64:
6274 case BFD_RELOC_X86_64_GOTPCREL64:
6275 case BFD_RELOC_X86_64_GOTPC64:
6276 case BFD_RELOC_X86_64_GOTPLT64:
6277 case BFD_RELOC_X86_64_PLTOFF64:
6278 case BFD_RELOC_X86_64_GOTPC32_TLSDESC:
6279 case BFD_RELOC_X86_64_TLSDESC_CALL:
6280 case BFD_RELOC_RVA:
6281 case BFD_RELOC_VTABLE_ENTRY:
6282 case BFD_RELOC_VTABLE_INHERIT:
6283 #ifdef TE_PE
6284 case BFD_RELOC_32_SECREL:
6285 #endif
6286 code = fixp->fx_r_type;
6287 break;
6288 case BFD_RELOC_X86_64_32S:
6289 if (!fixp->fx_pcrel)
6291 /* Don't turn BFD_RELOC_X86_64_32S into BFD_RELOC_32. */
6292 code = fixp->fx_r_type;
6293 break;
6295 default:
6296 if (fixp->fx_pcrel)
6298 switch (fixp->fx_size)
6300 default:
6301 as_bad_where (fixp->fx_file, fixp->fx_line,
6302 _("can not do %d byte pc-relative relocation"),
6303 fixp->fx_size);
6304 code = BFD_RELOC_32_PCREL;
6305 break;
6306 case 1: code = BFD_RELOC_8_PCREL; break;
6307 case 2: code = BFD_RELOC_16_PCREL; break;
6308 case 4: code = BFD_RELOC_32_PCREL; break;
6309 #ifdef BFD64
6310 case 8: code = BFD_RELOC_64_PCREL; break;
6311 #endif
6314 else
6316 switch (fixp->fx_size)
6318 default:
6319 as_bad_where (fixp->fx_file, fixp->fx_line,
6320 _("can not do %d byte relocation"),
6321 fixp->fx_size);
6322 code = BFD_RELOC_32;
6323 break;
6324 case 1: code = BFD_RELOC_8; break;
6325 case 2: code = BFD_RELOC_16; break;
6326 case 4: code = BFD_RELOC_32; break;
6327 #ifdef BFD64
6328 case 8: code = BFD_RELOC_64; break;
6329 #endif
6332 break;
6335 if ((code == BFD_RELOC_32
6336 || code == BFD_RELOC_32_PCREL
6337 || code == BFD_RELOC_X86_64_32S)
6338 && GOT_symbol
6339 && fixp->fx_addsy == GOT_symbol)
6341 if (!object_64bit)
6342 code = BFD_RELOC_386_GOTPC;
6343 else
6344 code = BFD_RELOC_X86_64_GOTPC32;
6346 if ((code == BFD_RELOC_64 || code == BFD_RELOC_64_PCREL)
6347 && GOT_symbol
6348 && fixp->fx_addsy == GOT_symbol)
6350 code = BFD_RELOC_X86_64_GOTPC64;
6353 rel = (arelent *) xmalloc (sizeof (arelent));
6354 rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
6355 *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
6357 rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
6359 if (!use_rela_relocations)
6361 /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
6362 vtable entry to be used in the relocation's section offset. */
6363 if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
6364 rel->address = fixp->fx_offset;
6366 rel->addend = 0;
6368 /* Use the rela in 64bit mode. */
6369 else
6371 if (!fixp->fx_pcrel)
6372 rel->addend = fixp->fx_offset;
6373 else
6374 switch (code)
6376 case BFD_RELOC_X86_64_PLT32:
6377 case BFD_RELOC_X86_64_GOT32:
6378 case BFD_RELOC_X86_64_GOTPCREL:
6379 case BFD_RELOC_X86_64_TLSGD:
6380 case BFD_RELOC_X86_64_TLSLD:
6381 case BFD_RELOC_X86_64_GOTTPOFF:
6382 case BFD_RELOC_X86_64_GOTPC32_TLSDESC:
6383 case BFD_RELOC_X86_64_TLSDESC_CALL:
6384 rel->addend = fixp->fx_offset - fixp->fx_size;
6385 break;
6386 default:
6387 rel->addend = (section->vma
6388 - fixp->fx_size
6389 + fixp->fx_addnumber
6390 + md_pcrel_from (fixp));
6391 break;
6395 rel->howto = bfd_reloc_type_lookup (stdoutput, code);
6396 if (rel->howto == NULL)
6398 as_bad_where (fixp->fx_file, fixp->fx_line,
6399 _("cannot represent relocation type %s"),
6400 bfd_get_reloc_code_name (code));
6401 /* Set howto to a garbage value so that we can keep going. */
6402 rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
6403 assert (rel->howto != NULL);
6406 return rel;
6410 /* Parse operands using Intel syntax. This implements a recursive descent
6411 parser based on the BNF grammar published in Appendix B of the MASM 6.1
6412 Programmer's Guide.
6414 FIXME: We do not recognize the full operand grammar defined in the MASM
6415 documentation. In particular, all the structure/union and
6416 high-level macro operands are missing.
6418 Uppercase words are terminals, lower case words are non-terminals.
6419 Objects surrounded by double brackets '[[' ']]' are optional. Vertical
6420 bars '|' denote choices. Most grammar productions are implemented in
6421 functions called 'intel_<production>'.
6423 Initial production is 'expr'.
6425 addOp + | -
6427 alpha [a-zA-Z]
6429 binOp & | AND | \| | OR | ^ | XOR
6431 byteRegister AL | AH | BL | BH | CL | CH | DL | DH
6433 constant digits [[ radixOverride ]]
6435 dataType BYTE | WORD | DWORD | FWORD | QWORD | TBYTE | OWORD | XMMWORD
6437 digits decdigit
6438 | digits decdigit
6439 | digits hexdigit
6441 decdigit [0-9]
6443 e04 e04 addOp e05
6444 | e05
6446 e05 e05 binOp e06
6447 | e06
6449 e06 e06 mulOp e09
6450 | e09
6452 e09 OFFSET e10
6453 | SHORT e10
6454 | + e10
6455 | - e10
6456 | ~ e10
6457 | NOT e10
6458 | e09 PTR e10
6459 | e09 : e10
6460 | e10
6462 e10 e10 [ expr ]
6463 | e11
6465 e11 ( expr )
6466 | [ expr ]
6467 | constant
6468 | dataType
6469 | id
6471 | register
6473 => expr expr cmpOp e04
6474 | e04
6476 gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
6477 | BP | EBP | SP | ESP | DI | EDI | SI | ESI
6479 hexdigit a | b | c | d | e | f
6480 | A | B | C | D | E | F
6482 id alpha
6483 | id alpha
6484 | id decdigit
6486 mulOp * | / | % | MOD | << | SHL | >> | SHR
6488 quote " | '
6490 register specialRegister
6491 | gpRegister
6492 | byteRegister
6494 segmentRegister CS | DS | ES | FS | GS | SS
6496 specialRegister CR0 | CR2 | CR3 | CR4
6497 | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
6498 | TR3 | TR4 | TR5 | TR6 | TR7
6500 We simplify the grammar in obvious places (e.g., register parsing is
6501 done by calling parse_register) and eliminate immediate left recursion
6502 to implement a recursive-descent parser.
6504 expr e04 expr'
6506 expr' cmpOp e04 expr'
6507 | Empty
6509 e04 e05 e04'
6511 e04' addOp e05 e04'
6512 | Empty
6514 e05 e06 e05'
6516 e05' binOp e06 e05'
6517 | Empty
6519 e06 e09 e06'
6521 e06' mulOp e09 e06'
6522 | Empty
6524 e09 OFFSET e10 e09'
6525 | SHORT e10'
6526 | + e10'
6527 | - e10'
6528 | ~ e10'
6529 | NOT e10'
6530 | e10 e09'
6532 e09' PTR e10 e09'
6533 | : e10 e09'
6534 | Empty
6536 e10 e11 e10'
6538 e10' [ expr ] e10'
6539 | Empty
6541 e11 ( expr )
6542 | [ expr ]
6543 | BYTE
6544 | WORD
6545 | DWORD
6546 | FWORD
6547 | QWORD
6548 | TBYTE
6549 | OWORD
6550 | XMMWORD
6553 | register
6554 | id
6555 | constant */
6557 /* Parsing structure for the intel syntax parser. Used to implement the
6558 semantic actions for the operand grammar. */
6559 struct intel_parser_s
6561 char *op_string; /* The string being parsed. */
6562 int got_a_float; /* Whether the operand is a float. */
6563 int op_modifier; /* Operand modifier. */
6564 int is_mem; /* 1 if operand is memory reference. */
6565 int in_offset; /* >=1 if parsing operand of offset. */
6566 int in_bracket; /* >=1 if parsing operand in brackets. */
6567 const reg_entry *reg; /* Last register reference found. */
6568 char *disp; /* Displacement string being built. */
6569 char *next_operand; /* Resume point when splitting operands. */
6572 static struct intel_parser_s intel_parser;
6574 /* Token structure for parsing intel syntax. */
6575 struct intel_token
6577 int code; /* Token code. */
6578 const reg_entry *reg; /* Register entry for register tokens. */
6579 char *str; /* String representation. */
6582 static struct intel_token cur_token, prev_token;
6584 /* Token codes for the intel parser. Since T_SHORT is already used
6585 by COFF, undefine it first to prevent a warning. */
6586 #define T_NIL -1
6587 #define T_CONST 1
6588 #define T_REG 2
6589 #define T_BYTE 3
6590 #define T_WORD 4
6591 #define T_DWORD 5
6592 #define T_FWORD 6
6593 #define T_QWORD 7
6594 #define T_TBYTE 8
6595 #define T_XMMWORD 9
6596 #undef T_SHORT
6597 #define T_SHORT 10
6598 #define T_OFFSET 11
6599 #define T_PTR 12
6600 #define T_ID 13
6601 #define T_SHL 14
6602 #define T_SHR 15
6604 /* Prototypes for intel parser functions. */
6605 static int intel_match_token (int);
6606 static void intel_putback_token (void);
6607 static void intel_get_token (void);
6608 static int intel_expr (void);
6609 static int intel_e04 (void);
6610 static int intel_e05 (void);
6611 static int intel_e06 (void);
6612 static int intel_e09 (void);
6613 static int intel_e10 (void);
6614 static int intel_e11 (void);
6616 static int
6617 i386_intel_operand (char *operand_string, int got_a_float)
6619 int ret;
6620 char *p;
6622 p = intel_parser.op_string = xstrdup (operand_string);
6623 intel_parser.disp = (char *) xmalloc (strlen (operand_string) + 1);
6625 for (;;)
6627 /* Initialize token holders. */
6628 cur_token.code = prev_token.code = T_NIL;
6629 cur_token.reg = prev_token.reg = NULL;
6630 cur_token.str = prev_token.str = NULL;
6632 /* Initialize parser structure. */
6633 intel_parser.got_a_float = got_a_float;
6634 intel_parser.op_modifier = 0;
6635 intel_parser.is_mem = 0;
6636 intel_parser.in_offset = 0;
6637 intel_parser.in_bracket = 0;
6638 intel_parser.reg = NULL;
6639 intel_parser.disp[0] = '\0';
6640 intel_parser.next_operand = NULL;
6642 /* Read the first token and start the parser. */
6643 intel_get_token ();
6644 ret = intel_expr ();
6646 if (!ret)
6647 break;
6649 if (cur_token.code != T_NIL)
6651 as_bad (_("invalid operand for '%s' ('%s' unexpected)"),
6652 current_templates->start->name, cur_token.str);
6653 ret = 0;
6655 /* If we found a memory reference, hand it over to i386_displacement
6656 to fill in the rest of the operand fields. */
6657 else if (intel_parser.is_mem)
6659 if ((i.mem_operands == 1
6660 && (current_templates->start->opcode_modifier & IsString) == 0)
6661 || i.mem_operands == 2)
6663 as_bad (_("too many memory references for '%s'"),
6664 current_templates->start->name);
6665 ret = 0;
6667 else
6669 char *s = intel_parser.disp;
6670 i.mem_operands++;
6672 if (!quiet_warnings && intel_parser.is_mem < 0)
6673 /* See the comments in intel_bracket_expr. */
6674 as_warn (_("Treating `%s' as memory reference"), operand_string);
6676 /* Add the displacement expression. */
6677 if (*s != '\0')
6678 ret = i386_displacement (s, s + strlen (s));
6679 if (ret)
6681 /* Swap base and index in 16-bit memory operands like
6682 [si+bx]. Since i386_index_check is also used in AT&T
6683 mode we have to do that here. */
6684 if (i.base_reg
6685 && i.index_reg
6686 && (i.base_reg->reg_type & Reg16)
6687 && (i.index_reg->reg_type & Reg16)
6688 && i.base_reg->reg_num >= 6
6689 && i.index_reg->reg_num < 6)
6691 const reg_entry *base = i.index_reg;
6693 i.index_reg = i.base_reg;
6694 i.base_reg = base;
6696 ret = i386_index_check (operand_string);
6701 /* Constant and OFFSET expressions are handled by i386_immediate. */
6702 else if ((intel_parser.op_modifier & (1 << T_OFFSET))
6703 || intel_parser.reg == NULL)
6704 ret = i386_immediate (intel_parser.disp);
6706 if (intel_parser.next_operand && this_operand >= MAX_OPERANDS - 1)
6707 ret = 0;
6708 if (!ret || !intel_parser.next_operand)
6709 break;
6710 intel_parser.op_string = intel_parser.next_operand;
6711 this_operand = i.operands++;
6714 free (p);
6715 free (intel_parser.disp);
6717 return ret;
6720 #define NUM_ADDRESS_REGS (!!i.base_reg + !!i.index_reg)
6722 /* expr e04 expr'
6724 expr' cmpOp e04 expr'
6725 | Empty */
6726 static int
6727 intel_expr (void)
6729 /* XXX Implement the comparison operators. */
6730 return intel_e04 ();
6733 /* e04 e05 e04'
6735 e04' addOp e05 e04'
6736 | Empty */
6737 static int
6738 intel_e04 (void)
6740 int nregs = -1;
6742 for (;;)
6744 if (!intel_e05())
6745 return 0;
6747 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
6748 i.base_reg = i386_regtab + REGNAM_AL; /* al is invalid as base */
6750 if (cur_token.code == '+')
6751 nregs = -1;
6752 else if (cur_token.code == '-')
6753 nregs = NUM_ADDRESS_REGS;
6754 else
6755 return 1;
6757 strcat (intel_parser.disp, cur_token.str);
6758 intel_match_token (cur_token.code);
6762 /* e05 e06 e05'
6764 e05' binOp e06 e05'
6765 | Empty */
6766 static int
6767 intel_e05 (void)
6769 int nregs = ~NUM_ADDRESS_REGS;
6771 for (;;)
6773 if (!intel_e06())
6774 return 0;
6776 if (cur_token.code == '&'
6777 || cur_token.code == '|'
6778 || cur_token.code == '^')
6780 char str[2];
6782 str[0] = cur_token.code;
6783 str[1] = 0;
6784 strcat (intel_parser.disp, str);
6786 else
6787 break;
6789 intel_match_token (cur_token.code);
6791 if (nregs < 0)
6792 nregs = ~nregs;
6794 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
6795 i.base_reg = i386_regtab + REGNAM_AL + 1; /* cl is invalid as base */
6796 return 1;
6799 /* e06 e09 e06'
6801 e06' mulOp e09 e06'
6802 | Empty */
6803 static int
6804 intel_e06 (void)
6806 int nregs = ~NUM_ADDRESS_REGS;
6808 for (;;)
6810 if (!intel_e09())
6811 return 0;
6813 if (cur_token.code == '*'
6814 || cur_token.code == '/'
6815 || cur_token.code == '%')
6817 char str[2];
6819 str[0] = cur_token.code;
6820 str[1] = 0;
6821 strcat (intel_parser.disp, str);
6823 else if (cur_token.code == T_SHL)
6824 strcat (intel_parser.disp, "<<");
6825 else if (cur_token.code == T_SHR)
6826 strcat (intel_parser.disp, ">>");
6827 else
6828 break;
6830 intel_match_token (cur_token.code);
6832 if (nregs < 0)
6833 nregs = ~nregs;
6835 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
6836 i.base_reg = i386_regtab + REGNAM_AL + 2; /* dl is invalid as base */
6837 return 1;
6840 /* e09 OFFSET e09
6841 | SHORT e09
6842 | + e09
6843 | - e09
6844 | ~ e09
6845 | NOT e09
6846 | e10 e09'
6848 e09' PTR e10 e09'
6849 | : e10 e09'
6850 | Empty */
6851 static int
6852 intel_e09 (void)
6854 int nregs = ~NUM_ADDRESS_REGS;
6855 int in_offset = 0;
6857 for (;;)
6859 /* Don't consume constants here. */
6860 if (cur_token.code == '+' || cur_token.code == '-')
6862 /* Need to look one token ahead - if the next token
6863 is a constant, the current token is its sign. */
6864 int next_code;
6866 intel_match_token (cur_token.code);
6867 next_code = cur_token.code;
6868 intel_putback_token ();
6869 if (next_code == T_CONST)
6870 break;
6873 /* e09 OFFSET e09 */
6874 if (cur_token.code == T_OFFSET)
6876 if (!in_offset++)
6877 ++intel_parser.in_offset;
6880 /* e09 SHORT e09 */
6881 else if (cur_token.code == T_SHORT)
6882 intel_parser.op_modifier |= 1 << T_SHORT;
6884 /* e09 + e09 */
6885 else if (cur_token.code == '+')
6886 strcat (intel_parser.disp, "+");
6888 /* e09 - e09
6889 | ~ e09
6890 | NOT e09 */
6891 else if (cur_token.code == '-' || cur_token.code == '~')
6893 char str[2];
6895 if (nregs < 0)
6896 nregs = ~nregs;
6897 str[0] = cur_token.code;
6898 str[1] = 0;
6899 strcat (intel_parser.disp, str);
6902 /* e09 e10 e09' */
6903 else
6904 break;
6906 intel_match_token (cur_token.code);
6909 for (;;)
6911 if (!intel_e10 ())
6912 return 0;
6914 /* e09' PTR e10 e09' */
6915 if (cur_token.code == T_PTR)
6917 char suffix;
6919 if (prev_token.code == T_BYTE)
6920 suffix = BYTE_MNEM_SUFFIX;
6922 else if (prev_token.code == T_WORD)
6924 if (current_templates->start->name[0] == 'l'
6925 && current_templates->start->name[2] == 's'
6926 && current_templates->start->name[3] == 0)
6927 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
6928 else if (intel_parser.got_a_float == 2) /* "fi..." */
6929 suffix = SHORT_MNEM_SUFFIX;
6930 else
6931 suffix = WORD_MNEM_SUFFIX;
6934 else if (prev_token.code == T_DWORD)
6936 if (current_templates->start->name[0] == 'l'
6937 && current_templates->start->name[2] == 's'
6938 && current_templates->start->name[3] == 0)
6939 suffix = WORD_MNEM_SUFFIX;
6940 else if (flag_code == CODE_16BIT
6941 && (current_templates->start->opcode_modifier
6942 & (Jump | JumpDword)))
6943 suffix = LONG_DOUBLE_MNEM_SUFFIX;
6944 else if (intel_parser.got_a_float == 1) /* "f..." */
6945 suffix = SHORT_MNEM_SUFFIX;
6946 else
6947 suffix = LONG_MNEM_SUFFIX;
6950 else if (prev_token.code == T_FWORD)
6952 if (current_templates->start->name[0] == 'l'
6953 && current_templates->start->name[2] == 's'
6954 && current_templates->start->name[3] == 0)
6955 suffix = LONG_MNEM_SUFFIX;
6956 else if (!intel_parser.got_a_float)
6958 if (flag_code == CODE_16BIT)
6959 add_prefix (DATA_PREFIX_OPCODE);
6960 suffix = LONG_DOUBLE_MNEM_SUFFIX;
6962 else
6963 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
6966 else if (prev_token.code == T_QWORD)
6968 if (intel_parser.got_a_float == 1) /* "f..." */
6969 suffix = LONG_MNEM_SUFFIX;
6970 else
6971 suffix = QWORD_MNEM_SUFFIX;
6974 else if (prev_token.code == T_TBYTE)
6976 if (intel_parser.got_a_float == 1)
6977 suffix = LONG_DOUBLE_MNEM_SUFFIX;
6978 else
6979 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
6982 else if (prev_token.code == T_XMMWORD)
6984 /* XXX ignored for now, but accepted since gcc uses it */
6985 suffix = 0;
6988 else
6990 as_bad (_("Unknown operand modifier `%s'"), prev_token.str);
6991 return 0;
6994 /* Operands for jump/call using 'ptr' notation denote absolute
6995 addresses. */
6996 if (current_templates->start->opcode_modifier & (Jump | JumpDword))
6997 i.types[this_operand] |= JumpAbsolute;
6999 if (current_templates->start->base_opcode == 0x8d /* lea */)
7001 else if (!i.suffix)
7002 i.suffix = suffix;
7003 else if (i.suffix != suffix)
7005 as_bad (_("Conflicting operand modifiers"));
7006 return 0;
7011 /* e09' : e10 e09' */
7012 else if (cur_token.code == ':')
7014 if (prev_token.code != T_REG)
7016 /* While {call,jmp} SSSS:OOOO is MASM syntax only when SSSS is a
7017 segment/group identifier (which we don't have), using comma
7018 as the operand separator there is even less consistent, since
7019 there all branches only have a single operand. */
7020 if (this_operand != 0
7021 || intel_parser.in_offset
7022 || intel_parser.in_bracket
7023 || (!(current_templates->start->opcode_modifier
7024 & (Jump|JumpDword|JumpInterSegment))
7025 && !(current_templates->start->operand_types[0]
7026 & JumpAbsolute)))
7027 return intel_match_token (T_NIL);
7028 /* Remember the start of the 2nd operand and terminate 1st
7029 operand here.
7030 XXX This isn't right, yet (when SSSS:OOOO is right operand of
7031 another expression), but it gets at least the simplest case
7032 (a plain number or symbol on the left side) right. */
7033 intel_parser.next_operand = intel_parser.op_string;
7034 *--intel_parser.op_string = '\0';
7035 return intel_match_token (':');
7039 /* e09' Empty */
7040 else
7041 break;
7043 intel_match_token (cur_token.code);
7047 if (in_offset)
7049 --intel_parser.in_offset;
7050 if (nregs < 0)
7051 nregs = ~nregs;
7052 if (NUM_ADDRESS_REGS > nregs)
7054 as_bad (_("Invalid operand to `OFFSET'"));
7055 return 0;
7057 intel_parser.op_modifier |= 1 << T_OFFSET;
7060 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
7061 i.base_reg = i386_regtab + REGNAM_AL + 3; /* bl is invalid as base */
7062 return 1;
7065 static int
7066 intel_bracket_expr (void)
7068 int was_offset = intel_parser.op_modifier & (1 << T_OFFSET);
7069 const char *start = intel_parser.op_string;
7070 int len;
7072 if (i.op[this_operand].regs)
7073 return intel_match_token (T_NIL);
7075 intel_match_token ('[');
7077 /* Mark as a memory operand only if it's not already known to be an
7078 offset expression. If it's an offset expression, we need to keep
7079 the brace in. */
7080 if (!intel_parser.in_offset)
7082 ++intel_parser.in_bracket;
7084 /* Operands for jump/call inside brackets denote absolute addresses. */
7085 if (current_templates->start->opcode_modifier & (Jump | JumpDword))
7086 i.types[this_operand] |= JumpAbsolute;
7088 /* Unfortunately gas always diverged from MASM in a respect that can't
7089 be easily fixed without risking to break code sequences likely to be
7090 encountered (the testsuite even check for this): MASM doesn't consider
7091 an expression inside brackets unconditionally as a memory reference.
7092 When that is e.g. a constant, an offset expression, or the sum of the
7093 two, this is still taken as a constant load. gas, however, always
7094 treated these as memory references. As a compromise, we'll try to make
7095 offset expressions inside brackets work the MASM way (since that's
7096 less likely to be found in real world code), but make constants alone
7097 continue to work the traditional gas way. In either case, issue a
7098 warning. */
7099 intel_parser.op_modifier &= ~was_offset;
7101 else
7102 strcat (intel_parser.disp, "[");
7104 /* Add a '+' to the displacement string if necessary. */
7105 if (*intel_parser.disp != '\0'
7106 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
7107 strcat (intel_parser.disp, "+");
7109 if (intel_expr ()
7110 && (len = intel_parser.op_string - start - 1,
7111 intel_match_token (']')))
7113 /* Preserve brackets when the operand is an offset expression. */
7114 if (intel_parser.in_offset)
7115 strcat (intel_parser.disp, "]");
7116 else
7118 --intel_parser.in_bracket;
7119 if (i.base_reg || i.index_reg)
7120 intel_parser.is_mem = 1;
7121 if (!intel_parser.is_mem)
7123 if (!(intel_parser.op_modifier & (1 << T_OFFSET)))
7124 /* Defer the warning until all of the operand was parsed. */
7125 intel_parser.is_mem = -1;
7126 else if (!quiet_warnings)
7127 as_warn (_("`[%.*s]' taken to mean just `%.*s'"),
7128 len, start, len, start);
7131 intel_parser.op_modifier |= was_offset;
7133 return 1;
7135 return 0;
7138 /* e10 e11 e10'
7140 e10' [ expr ] e10'
7141 | Empty */
7142 static int
7143 intel_e10 (void)
7145 if (!intel_e11 ())
7146 return 0;
7148 while (cur_token.code == '[')
7150 if (!intel_bracket_expr ())
7151 return 0;
7154 return 1;
7157 /* e11 ( expr )
7158 | [ expr ]
7159 | BYTE
7160 | WORD
7161 | DWORD
7162 | FWORD
7163 | QWORD
7164 | TBYTE
7165 | OWORD
7166 | XMMWORD
7169 | register
7170 | id
7171 | constant */
7172 static int
7173 intel_e11 (void)
7175 switch (cur_token.code)
7177 /* e11 ( expr ) */
7178 case '(':
7179 intel_match_token ('(');
7180 strcat (intel_parser.disp, "(");
7182 if (intel_expr () && intel_match_token (')'))
7184 strcat (intel_parser.disp, ")");
7185 return 1;
7187 return 0;
7189 /* e11 [ expr ] */
7190 case '[':
7191 return intel_bracket_expr ();
7193 /* e11 $
7194 | . */
7195 case '.':
7196 strcat (intel_parser.disp, cur_token.str);
7197 intel_match_token (cur_token.code);
7199 /* Mark as a memory operand only if it's not already known to be an
7200 offset expression. */
7201 if (!intel_parser.in_offset)
7202 intel_parser.is_mem = 1;
7204 return 1;
7206 /* e11 register */
7207 case T_REG:
7209 const reg_entry *reg = intel_parser.reg = cur_token.reg;
7211 intel_match_token (T_REG);
7213 /* Check for segment change. */
7214 if (cur_token.code == ':')
7216 if (!(reg->reg_type & (SReg2 | SReg3)))
7218 as_bad (_("`%s' is not a valid segment register"),
7219 reg->reg_name);
7220 return 0;
7222 else if (i.seg[i.mem_operands])
7223 as_warn (_("Extra segment override ignored"));
7224 else
7226 if (!intel_parser.in_offset)
7227 intel_parser.is_mem = 1;
7228 switch (reg->reg_num)
7230 case 0:
7231 i.seg[i.mem_operands] = &es;
7232 break;
7233 case 1:
7234 i.seg[i.mem_operands] = &cs;
7235 break;
7236 case 2:
7237 i.seg[i.mem_operands] = &ss;
7238 break;
7239 case 3:
7240 i.seg[i.mem_operands] = &ds;
7241 break;
7242 case 4:
7243 i.seg[i.mem_operands] = &fs;
7244 break;
7245 case 5:
7246 i.seg[i.mem_operands] = &gs;
7247 break;
7252 /* Not a segment register. Check for register scaling. */
7253 else if (cur_token.code == '*')
7255 if (!intel_parser.in_bracket)
7257 as_bad (_("Register scaling only allowed in memory operands"));
7258 return 0;
7261 if (reg->reg_type & Reg16) /* Disallow things like [si*1]. */
7262 reg = i386_regtab + REGNAM_AX + 4; /* sp is invalid as index */
7263 else if (i.index_reg)
7264 reg = i386_regtab + REGNAM_EAX + 4; /* esp is invalid as index */
7266 /* What follows must be a valid scale. */
7267 intel_match_token ('*');
7268 i.index_reg = reg;
7269 i.types[this_operand] |= BaseIndex;
7271 /* Set the scale after setting the register (otherwise,
7272 i386_scale will complain) */
7273 if (cur_token.code == '+' || cur_token.code == '-')
7275 char *str, sign = cur_token.code;
7276 intel_match_token (cur_token.code);
7277 if (cur_token.code != T_CONST)
7279 as_bad (_("Syntax error: Expecting a constant, got `%s'"),
7280 cur_token.str);
7281 return 0;
7283 str = (char *) xmalloc (strlen (cur_token.str) + 2);
7284 strcpy (str + 1, cur_token.str);
7285 *str = sign;
7286 if (!i386_scale (str))
7287 return 0;
7288 free (str);
7290 else if (!i386_scale (cur_token.str))
7291 return 0;
7292 intel_match_token (cur_token.code);
7295 /* No scaling. If this is a memory operand, the register is either a
7296 base register (first occurrence) or an index register (second
7297 occurrence). */
7298 else if (intel_parser.in_bracket)
7301 if (!i.base_reg)
7302 i.base_reg = reg;
7303 else if (!i.index_reg)
7304 i.index_reg = reg;
7305 else
7307 as_bad (_("Too many register references in memory operand"));
7308 return 0;
7311 i.types[this_operand] |= BaseIndex;
7314 /* It's neither base nor index. */
7315 else if (!intel_parser.in_offset && !intel_parser.is_mem)
7317 i.types[this_operand] |= reg->reg_type & ~BaseIndex;
7318 i.op[this_operand].regs = reg;
7319 i.reg_operands++;
7321 else
7323 as_bad (_("Invalid use of register"));
7324 return 0;
7327 /* Since registers are not part of the displacement string (except
7328 when we're parsing offset operands), we may need to remove any
7329 preceding '+' from the displacement string. */
7330 if (*intel_parser.disp != '\0'
7331 && !intel_parser.in_offset)
7333 char *s = intel_parser.disp;
7334 s += strlen (s) - 1;
7335 if (*s == '+')
7336 *s = '\0';
7339 return 1;
7342 /* e11 BYTE
7343 | WORD
7344 | DWORD
7345 | FWORD
7346 | QWORD
7347 | TBYTE
7348 | OWORD
7349 | XMMWORD */
7350 case T_BYTE:
7351 case T_WORD:
7352 case T_DWORD:
7353 case T_FWORD:
7354 case T_QWORD:
7355 case T_TBYTE:
7356 case T_XMMWORD:
7357 intel_match_token (cur_token.code);
7359 if (cur_token.code == T_PTR)
7360 return 1;
7362 /* It must have been an identifier. */
7363 intel_putback_token ();
7364 cur_token.code = T_ID;
7365 /* FALLTHRU */
7367 /* e11 id
7368 | constant */
7369 case T_ID:
7370 if (!intel_parser.in_offset && intel_parser.is_mem <= 0)
7372 symbolS *symbolP;
7374 /* The identifier represents a memory reference only if it's not
7375 preceded by an offset modifier and if it's not an equate. */
7376 symbolP = symbol_find(cur_token.str);
7377 if (!symbolP || S_GET_SEGMENT(symbolP) != absolute_section)
7378 intel_parser.is_mem = 1;
7380 /* FALLTHRU */
7382 case T_CONST:
7383 case '-':
7384 case '+':
7386 char *save_str, sign = 0;
7388 /* Allow constants that start with `+' or `-'. */
7389 if (cur_token.code == '-' || cur_token.code == '+')
7391 sign = cur_token.code;
7392 intel_match_token (cur_token.code);
7393 if (cur_token.code != T_CONST)
7395 as_bad (_("Syntax error: Expecting a constant, got `%s'"),
7396 cur_token.str);
7397 return 0;
7401 save_str = (char *) xmalloc (strlen (cur_token.str) + 2);
7402 strcpy (save_str + !!sign, cur_token.str);
7403 if (sign)
7404 *save_str = sign;
7406 /* Get the next token to check for register scaling. */
7407 intel_match_token (cur_token.code);
7409 /* Check if this constant is a scaling factor for an
7410 index register. */
7411 if (cur_token.code == '*')
7413 if (intel_match_token ('*') && cur_token.code == T_REG)
7415 const reg_entry *reg = cur_token.reg;
7417 if (!intel_parser.in_bracket)
7419 as_bad (_("Register scaling only allowed "
7420 "in memory operands"));
7421 return 0;
7424 /* Disallow things like [1*si].
7425 sp and esp are invalid as index. */
7426 if (reg->reg_type & Reg16)
7427 reg = i386_regtab + REGNAM_AX + 4;
7428 else if (i.index_reg)
7429 reg = i386_regtab + REGNAM_EAX + 4;
7431 /* The constant is followed by `* reg', so it must be
7432 a valid scale. */
7433 i.index_reg = reg;
7434 i.types[this_operand] |= BaseIndex;
7436 /* Set the scale after setting the register (otherwise,
7437 i386_scale will complain) */
7438 if (!i386_scale (save_str))
7439 return 0;
7440 intel_match_token (T_REG);
7442 /* Since registers are not part of the displacement
7443 string, we may need to remove any preceding '+' from
7444 the displacement string. */
7445 if (*intel_parser.disp != '\0')
7447 char *s = intel_parser.disp;
7448 s += strlen (s) - 1;
7449 if (*s == '+')
7450 *s = '\0';
7453 free (save_str);
7455 return 1;
7458 /* The constant was not used for register scaling. Since we have
7459 already consumed the token following `*' we now need to put it
7460 back in the stream. */
7461 intel_putback_token ();
7464 /* Add the constant to the displacement string. */
7465 strcat (intel_parser.disp, save_str);
7466 free (save_str);
7468 return 1;
7472 as_bad (_("Unrecognized token '%s'"), cur_token.str);
7473 return 0;
7476 /* Match the given token against cur_token. If they match, read the next
7477 token from the operand string. */
7478 static int
7479 intel_match_token (int code)
7481 if (cur_token.code == code)
7483 intel_get_token ();
7484 return 1;
7486 else
7488 as_bad (_("Unexpected token `%s'"), cur_token.str);
7489 return 0;
7493 /* Read a new token from intel_parser.op_string and store it in cur_token. */
7494 static void
7495 intel_get_token (void)
7497 char *end_op;
7498 const reg_entry *reg;
7499 struct intel_token new_token;
7501 new_token.code = T_NIL;
7502 new_token.reg = NULL;
7503 new_token.str = NULL;
7505 /* Free the memory allocated to the previous token and move
7506 cur_token to prev_token. */
7507 if (prev_token.str)
7508 free (prev_token.str);
7510 prev_token = cur_token;
7512 /* Skip whitespace. */
7513 while (is_space_char (*intel_parser.op_string))
7514 intel_parser.op_string++;
7516 /* Return an empty token if we find nothing else on the line. */
7517 if (*intel_parser.op_string == '\0')
7519 cur_token = new_token;
7520 return;
7523 /* The new token cannot be larger than the remainder of the operand
7524 string. */
7525 new_token.str = (char *) xmalloc (strlen (intel_parser.op_string) + 1);
7526 new_token.str[0] = '\0';
7528 if (strchr ("0123456789", *intel_parser.op_string))
7530 char *p = new_token.str;
7531 char *q = intel_parser.op_string;
7532 new_token.code = T_CONST;
7534 /* Allow any kind of identifier char to encompass floating point and
7535 hexadecimal numbers. */
7536 while (is_identifier_char (*q))
7537 *p++ = *q++;
7538 *p = '\0';
7540 /* Recognize special symbol names [0-9][bf]. */
7541 if (strlen (intel_parser.op_string) == 2
7542 && (intel_parser.op_string[1] == 'b'
7543 || intel_parser.op_string[1] == 'f'))
7544 new_token.code = T_ID;
7547 else if ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL)
7549 size_t len = end_op - intel_parser.op_string;
7551 new_token.code = T_REG;
7552 new_token.reg = reg;
7554 memcpy (new_token.str, intel_parser.op_string, len);
7555 new_token.str[len] = '\0';
7558 else if (is_identifier_char (*intel_parser.op_string))
7560 char *p = new_token.str;
7561 char *q = intel_parser.op_string;
7563 /* A '.' or '$' followed by an identifier char is an identifier.
7564 Otherwise, it's operator '.' followed by an expression. */
7565 if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
7567 new_token.code = '.';
7568 new_token.str[0] = '.';
7569 new_token.str[1] = '\0';
7571 else
7573 while (is_identifier_char (*q) || *q == '@')
7574 *p++ = *q++;
7575 *p = '\0';
7577 if (strcasecmp (new_token.str, "NOT") == 0)
7578 new_token.code = '~';
7580 else if (strcasecmp (new_token.str, "MOD") == 0)
7581 new_token.code = '%';
7583 else if (strcasecmp (new_token.str, "AND") == 0)
7584 new_token.code = '&';
7586 else if (strcasecmp (new_token.str, "OR") == 0)
7587 new_token.code = '|';
7589 else if (strcasecmp (new_token.str, "XOR") == 0)
7590 new_token.code = '^';
7592 else if (strcasecmp (new_token.str, "SHL") == 0)
7593 new_token.code = T_SHL;
7595 else if (strcasecmp (new_token.str, "SHR") == 0)
7596 new_token.code = T_SHR;
7598 else if (strcasecmp (new_token.str, "BYTE") == 0)
7599 new_token.code = T_BYTE;
7601 else if (strcasecmp (new_token.str, "WORD") == 0)
7602 new_token.code = T_WORD;
7604 else if (strcasecmp (new_token.str, "DWORD") == 0)
7605 new_token.code = T_DWORD;
7607 else if (strcasecmp (new_token.str, "FWORD") == 0)
7608 new_token.code = T_FWORD;
7610 else if (strcasecmp (new_token.str, "QWORD") == 0)
7611 new_token.code = T_QWORD;
7613 else if (strcasecmp (new_token.str, "TBYTE") == 0
7614 /* XXX remove (gcc still uses it) */
7615 || strcasecmp (new_token.str, "XWORD") == 0)
7616 new_token.code = T_TBYTE;
7618 else if (strcasecmp (new_token.str, "XMMWORD") == 0
7619 || strcasecmp (new_token.str, "OWORD") == 0)
7620 new_token.code = T_XMMWORD;
7622 else if (strcasecmp (new_token.str, "PTR") == 0)
7623 new_token.code = T_PTR;
7625 else if (strcasecmp (new_token.str, "SHORT") == 0)
7626 new_token.code = T_SHORT;
7628 else if (strcasecmp (new_token.str, "OFFSET") == 0)
7630 new_token.code = T_OFFSET;
7632 /* ??? This is not mentioned in the MASM grammar but gcc
7633 makes use of it with -mintel-syntax. OFFSET may be
7634 followed by FLAT: */
7635 if (strncasecmp (q, " FLAT:", 6) == 0)
7636 strcat (new_token.str, " FLAT:");
7639 /* ??? This is not mentioned in the MASM grammar. */
7640 else if (strcasecmp (new_token.str, "FLAT") == 0)
7642 new_token.code = T_OFFSET;
7643 if (*q == ':')
7644 strcat (new_token.str, ":");
7645 else
7646 as_bad (_("`:' expected"));
7649 else
7650 new_token.code = T_ID;
7654 else if (strchr ("+-/*%|&^:[]()~", *intel_parser.op_string))
7656 new_token.code = *intel_parser.op_string;
7657 new_token.str[0] = *intel_parser.op_string;
7658 new_token.str[1] = '\0';
7661 else if (strchr ("<>", *intel_parser.op_string)
7662 && *intel_parser.op_string == *(intel_parser.op_string + 1))
7664 new_token.code = *intel_parser.op_string == '<' ? T_SHL : T_SHR;
7665 new_token.str[0] = *intel_parser.op_string;
7666 new_token.str[1] = *intel_parser.op_string;
7667 new_token.str[2] = '\0';
7670 else
7671 as_bad (_("Unrecognized token `%s'"), intel_parser.op_string);
7673 intel_parser.op_string += strlen (new_token.str);
7674 cur_token = new_token;
7677 /* Put cur_token back into the token stream and make cur_token point to
7678 prev_token. */
7679 static void
7680 intel_putback_token (void)
7682 if (cur_token.code != T_NIL)
7684 intel_parser.op_string -= strlen (cur_token.str);
7685 free (cur_token.str);
7687 cur_token = prev_token;
7689 /* Forget prev_token. */
7690 prev_token.code = T_NIL;
7691 prev_token.reg = NULL;
7692 prev_token.str = NULL;
7696 tc_x86_regname_to_dw2regnum (char *regname)
7698 unsigned int regnum;
7699 unsigned int regnames_count;
7700 static const char *const regnames_32[] =
7702 "eax", "ecx", "edx", "ebx",
7703 "esp", "ebp", "esi", "edi",
7704 "eip", "eflags", NULL,
7705 "st0", "st1", "st2", "st3",
7706 "st4", "st5", "st6", "st7",
7707 NULL, NULL,
7708 "xmm0", "xmm1", "xmm2", "xmm3",
7709 "xmm4", "xmm5", "xmm6", "xmm7",
7710 "mm0", "mm1", "mm2", "mm3",
7711 "mm4", "mm5", "mm6", "mm7",
7712 "fcw", "fsw", "mxcsr",
7713 "es", "cs", "ss", "ds", "fs", "gs", NULL, NULL,
7714 "tr", "ldtr"
7716 static const char *const regnames_64[] =
7718 "rax", "rdx", "rcx", "rbx",
7719 "rsi", "rdi", "rbp", "rsp",
7720 "r8", "r9", "r10", "r11",
7721 "r12", "r13", "r14", "r15",
7722 "rip",
7723 "xmm0", "xmm1", "xmm2", "xmm3",
7724 "xmm4", "xmm5", "xmm6", "xmm7",
7725 "xmm8", "xmm9", "xmm10", "xmm11",
7726 "xmm12", "xmm13", "xmm14", "xmm15",
7727 "st0", "st1", "st2", "st3",
7728 "st4", "st5", "st6", "st7",
7729 "mm0", "mm1", "mm2", "mm3",
7730 "mm4", "mm5", "mm6", "mm7",
7731 "rflags",
7732 "es", "cs", "ss", "ds", "fs", "gs", NULL, NULL,
7733 "fs.base", "gs.base", NULL, NULL,
7734 "tr", "ldtr",
7735 "mxcsr", "fcw", "fsw"
7737 const char *const *regnames;
7739 if (flag_code == CODE_64BIT)
7741 regnames = regnames_64;
7742 regnames_count = ARRAY_SIZE (regnames_64);
7744 else
7746 regnames = regnames_32;
7747 regnames_count = ARRAY_SIZE (regnames_32);
7750 for (regnum = 0; regnum < regnames_count; regnum++)
7751 if (regnames[regnum] != NULL
7752 && strcmp (regname, regnames[regnum]) == 0)
7753 return regnum;
7755 return -1;
7758 void
7759 tc_x86_frame_initial_instructions (void)
7761 static unsigned int sp_regno;
7763 if (!sp_regno)
7764 sp_regno = tc_x86_regname_to_dw2regnum (flag_code == CODE_64BIT
7765 ? "rsp" : "esp");
7767 cfi_add_CFA_def_cfa (sp_regno, -x86_cie_data_alignment);
7768 cfi_add_CFA_offset (x86_dwarf2_return_column, x86_cie_data_alignment);
7772 i386_elf_section_type (const char *str, size_t len)
7774 if (flag_code == CODE_64BIT
7775 && len == sizeof ("unwind") - 1
7776 && strncmp (str, "unwind", 6) == 0)
7777 return SHT_X86_64_UNWIND;
7779 return -1;
7782 #ifdef TE_PE
7783 void
7784 tc_pe_dwarf2_emit_offset (symbolS *symbol, unsigned int size)
7786 expressionS expr;
7788 expr.X_op = O_secrel;
7789 expr.X_add_symbol = symbol;
7790 expr.X_add_number = 0;
7791 emit_expr (&expr, size);
7793 #endif
7795 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
7796 /* For ELF on x86-64, add support for SHF_X86_64_LARGE. */
7799 x86_64_section_letter (int letter, char **ptr_msg)
7801 if (flag_code == CODE_64BIT)
7803 if (letter == 'l')
7804 return SHF_X86_64_LARGE;
7806 *ptr_msg = _("Bad .section directive: want a,l,w,x,M,S,G,T in string");
7808 else
7809 *ptr_msg = _("Bad .section directive: want a,w,x,M,S,G,T in string");
7810 return -1;
7814 x86_64_section_word (char *str, size_t len)
7816 if (len == 5 && flag_code == CODE_64BIT && CONST_STRNEQ (str, "large"))
7817 return SHF_X86_64_LARGE;
7819 return -1;
7822 static void
7823 handle_large_common (int small ATTRIBUTE_UNUSED)
7825 if (flag_code != CODE_64BIT)
7827 s_comm_internal (0, elf_common_parse);
7828 as_warn (_(".largecomm supported only in 64bit mode, producing .comm"));
7830 else
7832 static segT lbss_section;
7833 asection *saved_com_section_ptr = elf_com_section_ptr;
7834 asection *saved_bss_section = bss_section;
7836 if (lbss_section == NULL)
7838 flagword applicable;
7839 segT seg = now_seg;
7840 subsegT subseg = now_subseg;
7842 /* The .lbss section is for local .largecomm symbols. */
7843 lbss_section = subseg_new (".lbss", 0);
7844 applicable = bfd_applicable_section_flags (stdoutput);
7845 bfd_set_section_flags (stdoutput, lbss_section,
7846 applicable & SEC_ALLOC);
7847 seg_info (lbss_section)->bss = 1;
7849 subseg_set (seg, subseg);
7852 elf_com_section_ptr = &_bfd_elf_large_com_section;
7853 bss_section = lbss_section;
7855 s_comm_internal (0, elf_common_parse);
7857 elf_com_section_ptr = saved_com_section_ptr;
7858 bss_section = saved_bss_section;
7861 #endif /* OBJ_ELF || OBJ_MAYBE_ELF */