2007-01-04 H.J. Lu <hongjiu.lu@intel.com>
[binutils.git] / gas / config / tc-i386.c
blob67bd39eb5c3bb9e21221151aeb2d606014e0ee5a
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_imm_operands (void);
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 0x66 prefix will be used.
747 3. For PROCESSOR_CORE2, alt_long_patt will be used.
748 4. For PROCESSOR_PENTIUMPRO, PROCESSOR_PENTIUM4, PROCESSOR_NOCONA,
749 PROCESSOR_CORE, PROCESSOR_CORE2, PROCESSOR_K6, PROCESSOR_ATHLON
750 and PROCESSOR_GENERIC64, alt_short_patt will be used.
752 When -mtune= isn't used, alt_short_patt will be used if
753 cpu_arch_isa_flags has Cpu686. Otherwise, f32_patt will be used.
755 When -march= or .arch is used, we can't use anything beyond
756 cpu_arch_isa_flags. */
758 if (flag_code == CODE_16BIT)
760 memcpy (fragP->fr_literal + fragP->fr_fix,
761 f16_patt[count - 1], count);
762 if (count > 8)
763 /* Adjust jump offset. */
764 fragP->fr_literal[fragP->fr_fix + 1] = count - 2;
766 else if (flag_code == CODE_64BIT && cpu_arch_tune == PROCESSOR_K8)
768 int i;
769 int nnops = (count + 3) / 4;
770 int len = count / nnops;
771 int remains = count - nnops * len;
772 int pos = 0;
774 /* The recommended way to pad 64bit code is to use NOPs preceded
775 by maximally four 0x66 prefixes. Balance the size of nops. */
776 for (i = 0; i < remains; i++)
778 memset (fragP->fr_literal + fragP->fr_fix + pos, 0x66, len);
779 fragP->fr_literal[fragP->fr_fix + pos + len] = 0x90;
780 pos += len + 1;
782 for (; i < nnops; i++)
784 memset (fragP->fr_literal + fragP->fr_fix + pos, 0x66, len - 1);
785 fragP->fr_literal[fragP->fr_fix + pos + len - 1] = 0x90;
786 pos += len;
789 else
791 const char *const *patt = NULL;
793 if (cpu_arch_isa == PROCESSOR_UNKNOWN)
795 /* PROCESSOR_UNKNOWN means that all ISAs may be used. */
796 switch (cpu_arch_tune)
798 case PROCESSOR_UNKNOWN:
799 /* We use cpu_arch_isa_flags to check if we SHOULD
800 optimize for Cpu686. */
801 if ((cpu_arch_isa_flags & Cpu686) != 0)
802 patt = alt_short_patt;
803 else
804 patt = f32_patt;
805 break;
806 case PROCESSOR_CORE2:
807 patt = alt_long_patt;
808 break;
809 case PROCESSOR_PENTIUMPRO:
810 case PROCESSOR_PENTIUM4:
811 case PROCESSOR_NOCONA:
812 case PROCESSOR_CORE:
813 case PROCESSOR_K6:
814 case PROCESSOR_ATHLON:
815 case PROCESSOR_K8:
816 case PROCESSOR_GENERIC64:
817 case PROCESSOR_AMDFAM10:
818 patt = alt_short_patt;
819 break;
820 case PROCESSOR_I486:
821 case PROCESSOR_PENTIUM:
822 case PROCESSOR_GENERIC32:
823 patt = f32_patt;
824 break;
827 else
829 switch (cpu_arch_tune)
831 case PROCESSOR_UNKNOWN:
832 /* When cpu_arch_isa is net, cpu_arch_tune shouldn't be
833 PROCESSOR_UNKNOWN. */
834 abort ();
835 break;
837 case PROCESSOR_I486:
838 case PROCESSOR_PENTIUM:
839 case PROCESSOR_PENTIUMPRO:
840 case PROCESSOR_PENTIUM4:
841 case PROCESSOR_NOCONA:
842 case PROCESSOR_CORE:
843 case PROCESSOR_K6:
844 case PROCESSOR_ATHLON:
845 case PROCESSOR_K8:
846 case PROCESSOR_AMDFAM10:
847 case PROCESSOR_GENERIC32:
848 /* We use cpu_arch_isa_flags to check if we CAN optimize
849 for Cpu686. */
850 if ((cpu_arch_isa_flags & Cpu686) != 0)
851 patt = alt_short_patt;
852 else
853 patt = f32_patt;
854 break;
855 case PROCESSOR_CORE2:
856 if ((cpu_arch_isa_flags & Cpu686) != 0)
857 patt = alt_long_patt;
858 else
859 patt = f32_patt;
860 break;
861 case PROCESSOR_GENERIC64:
862 patt = alt_short_patt;
863 break;
867 memcpy (fragP->fr_literal + fragP->fr_fix,
868 patt[count - 1], count);
870 fragP->fr_var = count;
873 static INLINE unsigned int
874 mode_from_disp_size (unsigned int t)
876 return (t & Disp8) ? 1 : (t & (Disp16 | Disp32 | Disp32S)) ? 2 : 0;
879 static INLINE int
880 fits_in_signed_byte (offsetT num)
882 return (num >= -128) && (num <= 127);
885 static INLINE int
886 fits_in_unsigned_byte (offsetT num)
888 return (num & 0xff) == num;
891 static INLINE int
892 fits_in_unsigned_word (offsetT num)
894 return (num & 0xffff) == num;
897 static INLINE int
898 fits_in_signed_word (offsetT num)
900 return (-32768 <= num) && (num <= 32767);
903 static INLINE int
904 fits_in_signed_long (offsetT num ATTRIBUTE_UNUSED)
906 #ifndef BFD64
907 return 1;
908 #else
909 return (!(((offsetT) -1 << 31) & num)
910 || (((offsetT) -1 << 31) & num) == ((offsetT) -1 << 31));
911 #endif
912 } /* fits_in_signed_long() */
914 static INLINE int
915 fits_in_unsigned_long (offsetT num ATTRIBUTE_UNUSED)
917 #ifndef BFD64
918 return 1;
919 #else
920 return (num & (((offsetT) 2 << 31) - 1)) == num;
921 #endif
922 } /* fits_in_unsigned_long() */
924 static unsigned int
925 smallest_imm_type (offsetT num)
927 if (cpu_arch_flags != (Cpu186 | Cpu286 | Cpu386 | Cpu486 | CpuNo64))
929 /* This code is disabled on the 486 because all the Imm1 forms
930 in the opcode table are slower on the i486. They're the
931 versions with the implicitly specified single-position
932 displacement, which has another syntax if you really want to
933 use that form. */
934 if (num == 1)
935 return Imm1 | Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64;
937 return (fits_in_signed_byte (num)
938 ? (Imm8S | Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
939 : fits_in_unsigned_byte (num)
940 ? (Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
941 : (fits_in_signed_word (num) || fits_in_unsigned_word (num))
942 ? (Imm16 | Imm32 | Imm32S | Imm64)
943 : fits_in_signed_long (num)
944 ? (Imm32 | Imm32S | Imm64)
945 : fits_in_unsigned_long (num)
946 ? (Imm32 | Imm64)
947 : Imm64);
950 static offsetT
951 offset_in_range (offsetT val, int size)
953 addressT mask;
955 switch (size)
957 case 1: mask = ((addressT) 1 << 8) - 1; break;
958 case 2: mask = ((addressT) 1 << 16) - 1; break;
959 case 4: mask = ((addressT) 2 << 31) - 1; break;
960 #ifdef BFD64
961 case 8: mask = ((addressT) 2 << 63) - 1; break;
962 #endif
963 default: abort ();
966 /* If BFD64, sign extend val. */
967 if (!use_rela_relocations)
968 if ((val & ~(((addressT) 2 << 31) - 1)) == 0)
969 val = (val ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
971 if ((val & ~mask) != 0 && (val & ~mask) != ~mask)
973 char buf1[40], buf2[40];
975 sprint_value (buf1, val);
976 sprint_value (buf2, val & mask);
977 as_warn (_("%s shortened to %s"), buf1, buf2);
979 return val & mask;
982 /* Returns 0 if attempting to add a prefix where one from the same
983 class already exists, 1 if non rep/repne added, 2 if rep/repne
984 added. */
985 static int
986 add_prefix (unsigned int prefix)
988 int ret = 1;
989 unsigned int q;
991 if (prefix >= REX_OPCODE && prefix < REX_OPCODE + 16
992 && flag_code == CODE_64BIT)
994 if ((i.prefix[REX_PREFIX] & prefix & REX_MODE64)
995 || ((i.prefix[REX_PREFIX] & (REX_EXTX | REX_EXTY | REX_EXTZ))
996 && (prefix & (REX_EXTX | REX_EXTY | REX_EXTZ))))
997 ret = 0;
998 q = REX_PREFIX;
1000 else
1002 switch (prefix)
1004 default:
1005 abort ();
1007 case CS_PREFIX_OPCODE:
1008 case DS_PREFIX_OPCODE:
1009 case ES_PREFIX_OPCODE:
1010 case FS_PREFIX_OPCODE:
1011 case GS_PREFIX_OPCODE:
1012 case SS_PREFIX_OPCODE:
1013 q = SEG_PREFIX;
1014 break;
1016 case REPNE_PREFIX_OPCODE:
1017 case REPE_PREFIX_OPCODE:
1018 ret = 2;
1019 /* fall thru */
1020 case LOCK_PREFIX_OPCODE:
1021 q = LOCKREP_PREFIX;
1022 break;
1024 case FWAIT_OPCODE:
1025 q = WAIT_PREFIX;
1026 break;
1028 case ADDR_PREFIX_OPCODE:
1029 q = ADDR_PREFIX;
1030 break;
1032 case DATA_PREFIX_OPCODE:
1033 q = DATA_PREFIX;
1034 break;
1036 if (i.prefix[q] != 0)
1037 ret = 0;
1040 if (ret)
1042 if (!i.prefix[q])
1043 ++i.prefixes;
1044 i.prefix[q] |= prefix;
1046 else
1047 as_bad (_("same type of prefix used twice"));
1049 return ret;
1052 static void
1053 set_code_flag (int value)
1055 flag_code = value;
1056 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
1057 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
1058 if (value == CODE_64BIT && !(cpu_arch_flags & CpuSledgehammer))
1060 as_bad (_("64bit mode not supported on this CPU."));
1062 if (value == CODE_32BIT && !(cpu_arch_flags & Cpu386))
1064 as_bad (_("32bit mode not supported on this CPU."));
1066 stackop_size = '\0';
1069 static void
1070 set_16bit_gcc_code_flag (int new_code_flag)
1072 flag_code = new_code_flag;
1073 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
1074 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
1075 stackop_size = LONG_MNEM_SUFFIX;
1078 static void
1079 set_intel_syntax (int syntax_flag)
1081 /* Find out if register prefixing is specified. */
1082 int ask_naked_reg = 0;
1084 SKIP_WHITESPACE ();
1085 if (!is_end_of_line[(unsigned char) *input_line_pointer])
1087 char *string = input_line_pointer;
1088 int e = get_symbol_end ();
1090 if (strcmp (string, "prefix") == 0)
1091 ask_naked_reg = 1;
1092 else if (strcmp (string, "noprefix") == 0)
1093 ask_naked_reg = -1;
1094 else
1095 as_bad (_("bad argument to syntax directive."));
1096 *input_line_pointer = e;
1098 demand_empty_rest_of_line ();
1100 intel_syntax = syntax_flag;
1102 if (ask_naked_reg == 0)
1103 allow_naked_reg = (intel_syntax
1104 && (bfd_get_symbol_leading_char (stdoutput) != '\0'));
1105 else
1106 allow_naked_reg = (ask_naked_reg < 0);
1108 if (intel_syntax && allow_naked_reg)
1110 identifier_chars['%'] = '%';
1111 register_prefix = "";
1113 else
1114 identifier_chars['%'] = 0;
1115 identifier_chars['$'] = intel_syntax ? '$' : 0;
1118 static void
1119 set_cpu_arch (int dummy ATTRIBUTE_UNUSED)
1121 SKIP_WHITESPACE ();
1123 if (!is_end_of_line[(unsigned char) *input_line_pointer])
1125 char *string = input_line_pointer;
1126 int e = get_symbol_end ();
1127 unsigned int i;
1129 for (i = 0; i < ARRAY_SIZE (cpu_arch); i++)
1131 if (strcmp (string, cpu_arch[i].name) == 0)
1133 if (*string != '.')
1135 cpu_arch_name = cpu_arch[i].name;
1136 cpu_sub_arch_name = NULL;
1137 cpu_arch_flags = (cpu_arch[i].flags
1138 | (flag_code == CODE_64BIT ? Cpu64 : CpuNo64));
1139 cpu_arch_isa = cpu_arch[i].type;
1140 cpu_arch_isa_flags = cpu_arch[i].flags;
1141 if (!cpu_arch_tune_set)
1143 cpu_arch_tune = cpu_arch_isa;
1144 cpu_arch_tune_flags = cpu_arch_isa_flags;
1146 break;
1148 if ((cpu_arch_flags | cpu_arch[i].flags) != cpu_arch_flags)
1150 cpu_sub_arch_name = cpu_arch[i].name;
1151 cpu_arch_flags |= cpu_arch[i].flags;
1153 *input_line_pointer = e;
1154 demand_empty_rest_of_line ();
1155 return;
1158 if (i >= ARRAY_SIZE (cpu_arch))
1159 as_bad (_("no such architecture: `%s'"), string);
1161 *input_line_pointer = e;
1163 else
1164 as_bad (_("missing cpu architecture"));
1166 no_cond_jump_promotion = 0;
1167 if (*input_line_pointer == ','
1168 && !is_end_of_line[(unsigned char) input_line_pointer[1]])
1170 char *string = ++input_line_pointer;
1171 int e = get_symbol_end ();
1173 if (strcmp (string, "nojumps") == 0)
1174 no_cond_jump_promotion = 1;
1175 else if (strcmp (string, "jumps") == 0)
1177 else
1178 as_bad (_("no such architecture modifier: `%s'"), string);
1180 *input_line_pointer = e;
1183 demand_empty_rest_of_line ();
1186 unsigned long
1187 i386_mach ()
1189 if (!strcmp (default_arch, "x86_64"))
1190 return bfd_mach_x86_64;
1191 else if (!strcmp (default_arch, "i386"))
1192 return bfd_mach_i386_i386;
1193 else
1194 as_fatal (_("Unknown architecture"));
1197 void
1198 md_begin ()
1200 const char *hash_err;
1202 /* Initialize op_hash hash table. */
1203 op_hash = hash_new ();
1206 const template *optab;
1207 templates *core_optab;
1209 /* Setup for loop. */
1210 optab = i386_optab;
1211 core_optab = (templates *) xmalloc (sizeof (templates));
1212 core_optab->start = optab;
1214 while (1)
1216 ++optab;
1217 if (optab->name == NULL
1218 || strcmp (optab->name, (optab - 1)->name) != 0)
1220 /* different name --> ship out current template list;
1221 add to hash table; & begin anew. */
1222 core_optab->end = optab;
1223 hash_err = hash_insert (op_hash,
1224 (optab - 1)->name,
1225 (PTR) core_optab);
1226 if (hash_err)
1228 as_fatal (_("Internal Error: Can't hash %s: %s"),
1229 (optab - 1)->name,
1230 hash_err);
1232 if (optab->name == NULL)
1233 break;
1234 core_optab = (templates *) xmalloc (sizeof (templates));
1235 core_optab->start = optab;
1240 /* Initialize reg_hash hash table. */
1241 reg_hash = hash_new ();
1243 const reg_entry *regtab;
1245 for (regtab = i386_regtab;
1246 regtab < i386_regtab + sizeof (i386_regtab) / sizeof (i386_regtab[0]);
1247 regtab++)
1249 hash_err = hash_insert (reg_hash, regtab->reg_name, (PTR) regtab);
1250 if (hash_err)
1251 as_fatal (_("Internal Error: Can't hash %s: %s"),
1252 regtab->reg_name,
1253 hash_err);
1257 /* Fill in lexical tables: mnemonic_chars, operand_chars. */
1259 int c;
1260 char *p;
1262 for (c = 0; c < 256; c++)
1264 if (ISDIGIT (c))
1266 digit_chars[c] = c;
1267 mnemonic_chars[c] = c;
1268 register_chars[c] = c;
1269 operand_chars[c] = c;
1271 else if (ISLOWER (c))
1273 mnemonic_chars[c] = c;
1274 register_chars[c] = c;
1275 operand_chars[c] = c;
1277 else if (ISUPPER (c))
1279 mnemonic_chars[c] = TOLOWER (c);
1280 register_chars[c] = mnemonic_chars[c];
1281 operand_chars[c] = c;
1284 if (ISALPHA (c) || ISDIGIT (c))
1285 identifier_chars[c] = c;
1286 else if (c >= 128)
1288 identifier_chars[c] = c;
1289 operand_chars[c] = c;
1293 #ifdef LEX_AT
1294 identifier_chars['@'] = '@';
1295 #endif
1296 #ifdef LEX_QM
1297 identifier_chars['?'] = '?';
1298 operand_chars['?'] = '?';
1299 #endif
1300 digit_chars['-'] = '-';
1301 mnemonic_chars['-'] = '-';
1302 identifier_chars['_'] = '_';
1303 identifier_chars['.'] = '.';
1305 for (p = operand_special_chars; *p != '\0'; p++)
1306 operand_chars[(unsigned char) *p] = *p;
1309 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1310 if (IS_ELF)
1312 record_alignment (text_section, 2);
1313 record_alignment (data_section, 2);
1314 record_alignment (bss_section, 2);
1316 #endif
1318 if (flag_code == CODE_64BIT)
1320 x86_dwarf2_return_column = 16;
1321 x86_cie_data_alignment = -8;
1323 else
1325 x86_dwarf2_return_column = 8;
1326 x86_cie_data_alignment = -4;
1330 void
1331 i386_print_statistics (FILE *file)
1333 hash_print_statistics (file, "i386 opcode", op_hash);
1334 hash_print_statistics (file, "i386 register", reg_hash);
1337 #ifdef DEBUG386
1339 /* Debugging routines for md_assemble. */
1340 static void pte (template *);
1341 static void pt (unsigned int);
1342 static void pe (expressionS *);
1343 static void ps (symbolS *);
1345 static void
1346 pi (char *line, i386_insn *x)
1348 unsigned int i;
1350 fprintf (stdout, "%s: template ", line);
1351 pte (&x->tm);
1352 fprintf (stdout, " address: base %s index %s scale %x\n",
1353 x->base_reg ? x->base_reg->reg_name : "none",
1354 x->index_reg ? x->index_reg->reg_name : "none",
1355 x->log2_scale_factor);
1356 fprintf (stdout, " modrm: mode %x reg %x reg/mem %x\n",
1357 x->rm.mode, x->rm.reg, x->rm.regmem);
1358 fprintf (stdout, " sib: base %x index %x scale %x\n",
1359 x->sib.base, x->sib.index, x->sib.scale);
1360 fprintf (stdout, " rex: 64bit %x extX %x extY %x extZ %x\n",
1361 (x->rex & REX_MODE64) != 0,
1362 (x->rex & REX_EXTX) != 0,
1363 (x->rex & REX_EXTY) != 0,
1364 (x->rex & REX_EXTZ) != 0);
1365 for (i = 0; i < x->operands; i++)
1367 fprintf (stdout, " #%d: ", i + 1);
1368 pt (x->types[i]);
1369 fprintf (stdout, "\n");
1370 if (x->types[i]
1371 & (Reg | SReg2 | SReg3 | Control | Debug | Test | RegMMX | RegXMM))
1372 fprintf (stdout, "%s\n", x->op[i].regs->reg_name);
1373 if (x->types[i] & Imm)
1374 pe (x->op[i].imms);
1375 if (x->types[i] & Disp)
1376 pe (x->op[i].disps);
1380 static void
1381 pte (template *t)
1383 unsigned int i;
1384 fprintf (stdout, " %d operands ", t->operands);
1385 fprintf (stdout, "opcode %x ", t->base_opcode);
1386 if (t->extension_opcode != None)
1387 fprintf (stdout, "ext %x ", t->extension_opcode);
1388 if (t->opcode_modifier & D)
1389 fprintf (stdout, "D");
1390 if (t->opcode_modifier & W)
1391 fprintf (stdout, "W");
1392 fprintf (stdout, "\n");
1393 for (i = 0; i < t->operands; i++)
1395 fprintf (stdout, " #%d type ", i + 1);
1396 pt (t->operand_types[i]);
1397 fprintf (stdout, "\n");
1401 static void
1402 pe (expressionS *e)
1404 fprintf (stdout, " operation %d\n", e->X_op);
1405 fprintf (stdout, " add_number %ld (%lx)\n",
1406 (long) e->X_add_number, (long) e->X_add_number);
1407 if (e->X_add_symbol)
1409 fprintf (stdout, " add_symbol ");
1410 ps (e->X_add_symbol);
1411 fprintf (stdout, "\n");
1413 if (e->X_op_symbol)
1415 fprintf (stdout, " op_symbol ");
1416 ps (e->X_op_symbol);
1417 fprintf (stdout, "\n");
1421 static void
1422 ps (symbolS *s)
1424 fprintf (stdout, "%s type %s%s",
1425 S_GET_NAME (s),
1426 S_IS_EXTERNAL (s) ? "EXTERNAL " : "",
1427 segment_name (S_GET_SEGMENT (s)));
1430 static struct type_name
1432 unsigned int mask;
1433 char *tname;
1435 const type_names[] =
1437 { Reg8, "r8" },
1438 { Reg16, "r16" },
1439 { Reg32, "r32" },
1440 { Reg64, "r64" },
1441 { Imm8, "i8" },
1442 { Imm8S, "i8s" },
1443 { Imm16, "i16" },
1444 { Imm32, "i32" },
1445 { Imm32S, "i32s" },
1446 { Imm64, "i64" },
1447 { Imm1, "i1" },
1448 { BaseIndex, "BaseIndex" },
1449 { Disp8, "d8" },
1450 { Disp16, "d16" },
1451 { Disp32, "d32" },
1452 { Disp32S, "d32s" },
1453 { Disp64, "d64" },
1454 { InOutPortReg, "InOutPortReg" },
1455 { ShiftCount, "ShiftCount" },
1456 { Control, "control reg" },
1457 { Test, "test reg" },
1458 { Debug, "debug reg" },
1459 { FloatReg, "FReg" },
1460 { FloatAcc, "FAcc" },
1461 { SReg2, "SReg2" },
1462 { SReg3, "SReg3" },
1463 { Acc, "Acc" },
1464 { JumpAbsolute, "Jump Absolute" },
1465 { RegMMX, "rMMX" },
1466 { RegXMM, "rXMM" },
1467 { EsSeg, "es" },
1468 { 0, "" }
1471 static void
1472 pt (t)
1473 unsigned int t;
1475 const struct type_name *ty;
1477 for (ty = type_names; ty->mask; ty++)
1478 if (t & ty->mask)
1479 fprintf (stdout, "%s, ", ty->tname);
1480 fflush (stdout);
1483 #endif /* DEBUG386 */
1485 static bfd_reloc_code_real_type
1486 reloc (unsigned int size,
1487 int pcrel,
1488 int sign,
1489 bfd_reloc_code_real_type other)
1491 if (other != NO_RELOC)
1493 reloc_howto_type *reloc;
1495 if (size == 8)
1496 switch (other)
1498 case BFD_RELOC_X86_64_GOT32:
1499 return BFD_RELOC_X86_64_GOT64;
1500 break;
1501 case BFD_RELOC_X86_64_PLTOFF64:
1502 return BFD_RELOC_X86_64_PLTOFF64;
1503 break;
1504 case BFD_RELOC_X86_64_GOTPC32:
1505 other = BFD_RELOC_X86_64_GOTPC64;
1506 break;
1507 case BFD_RELOC_X86_64_GOTPCREL:
1508 other = BFD_RELOC_X86_64_GOTPCREL64;
1509 break;
1510 case BFD_RELOC_X86_64_TPOFF32:
1511 other = BFD_RELOC_X86_64_TPOFF64;
1512 break;
1513 case BFD_RELOC_X86_64_DTPOFF32:
1514 other = BFD_RELOC_X86_64_DTPOFF64;
1515 break;
1516 default:
1517 break;
1520 /* Sign-checking 4-byte relocations in 16-/32-bit code is pointless. */
1521 if (size == 4 && flag_code != CODE_64BIT)
1522 sign = -1;
1524 reloc = bfd_reloc_type_lookup (stdoutput, other);
1525 if (!reloc)
1526 as_bad (_("unknown relocation (%u)"), other);
1527 else if (size != bfd_get_reloc_size (reloc))
1528 as_bad (_("%u-byte relocation cannot be applied to %u-byte field"),
1529 bfd_get_reloc_size (reloc),
1530 size);
1531 else if (pcrel && !reloc->pc_relative)
1532 as_bad (_("non-pc-relative relocation for pc-relative field"));
1533 else if ((reloc->complain_on_overflow == complain_overflow_signed
1534 && !sign)
1535 || (reloc->complain_on_overflow == complain_overflow_unsigned
1536 && sign > 0))
1537 as_bad (_("relocated field and relocation type differ in signedness"));
1538 else
1539 return other;
1540 return NO_RELOC;
1543 if (pcrel)
1545 if (!sign)
1546 as_bad (_("there are no unsigned pc-relative relocations"));
1547 switch (size)
1549 case 1: return BFD_RELOC_8_PCREL;
1550 case 2: return BFD_RELOC_16_PCREL;
1551 case 4: return BFD_RELOC_32_PCREL;
1552 case 8: return BFD_RELOC_64_PCREL;
1554 as_bad (_("cannot do %u byte pc-relative relocation"), size);
1556 else
1558 if (sign > 0)
1559 switch (size)
1561 case 4: return BFD_RELOC_X86_64_32S;
1563 else
1564 switch (size)
1566 case 1: return BFD_RELOC_8;
1567 case 2: return BFD_RELOC_16;
1568 case 4: return BFD_RELOC_32;
1569 case 8: return BFD_RELOC_64;
1571 as_bad (_("cannot do %s %u byte relocation"),
1572 sign > 0 ? "signed" : "unsigned", size);
1575 abort ();
1576 return BFD_RELOC_NONE;
1579 /* Here we decide which fixups can be adjusted to make them relative to
1580 the beginning of the section instead of the symbol. Basically we need
1581 to make sure that the dynamic relocations are done correctly, so in
1582 some cases we force the original symbol to be used. */
1585 tc_i386_fix_adjustable (fixS *fixP ATTRIBUTE_UNUSED)
1587 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1588 if (!IS_ELF)
1589 return 1;
1591 /* Don't adjust pc-relative references to merge sections in 64-bit
1592 mode. */
1593 if (use_rela_relocations
1594 && (S_GET_SEGMENT (fixP->fx_addsy)->flags & SEC_MERGE) != 0
1595 && fixP->fx_pcrel)
1596 return 0;
1598 /* The x86_64 GOTPCREL are represented as 32bit PCrel relocations
1599 and changed later by validate_fix. */
1600 if (GOT_symbol && fixP->fx_subsy == GOT_symbol
1601 && fixP->fx_r_type == BFD_RELOC_32_PCREL)
1602 return 0;
1604 /* adjust_reloc_syms doesn't know about the GOT. */
1605 if (fixP->fx_r_type == BFD_RELOC_386_GOTOFF
1606 || fixP->fx_r_type == BFD_RELOC_386_PLT32
1607 || fixP->fx_r_type == BFD_RELOC_386_GOT32
1608 || fixP->fx_r_type == BFD_RELOC_386_TLS_GD
1609 || fixP->fx_r_type == BFD_RELOC_386_TLS_LDM
1610 || fixP->fx_r_type == BFD_RELOC_386_TLS_LDO_32
1611 || fixP->fx_r_type == BFD_RELOC_386_TLS_IE_32
1612 || fixP->fx_r_type == BFD_RELOC_386_TLS_IE
1613 || fixP->fx_r_type == BFD_RELOC_386_TLS_GOTIE
1614 || fixP->fx_r_type == BFD_RELOC_386_TLS_LE_32
1615 || fixP->fx_r_type == BFD_RELOC_386_TLS_LE
1616 || fixP->fx_r_type == BFD_RELOC_386_TLS_GOTDESC
1617 || fixP->fx_r_type == BFD_RELOC_386_TLS_DESC_CALL
1618 || fixP->fx_r_type == BFD_RELOC_X86_64_PLT32
1619 || fixP->fx_r_type == BFD_RELOC_X86_64_GOT32
1620 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPCREL
1621 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSGD
1622 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSLD
1623 || fixP->fx_r_type == BFD_RELOC_X86_64_DTPOFF32
1624 || fixP->fx_r_type == BFD_RELOC_X86_64_DTPOFF64
1625 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTTPOFF
1626 || fixP->fx_r_type == BFD_RELOC_X86_64_TPOFF32
1627 || fixP->fx_r_type == BFD_RELOC_X86_64_TPOFF64
1628 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTOFF64
1629 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPC32_TLSDESC
1630 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSDESC_CALL
1631 || fixP->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1632 || fixP->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1633 return 0;
1634 #endif
1635 return 1;
1638 static int
1639 intel_float_operand (const char *mnemonic)
1641 /* Note that the value returned is meaningful only for opcodes with (memory)
1642 operands, hence the code here is free to improperly handle opcodes that
1643 have no operands (for better performance and smaller code). */
1645 if (mnemonic[0] != 'f')
1646 return 0; /* non-math */
1648 switch (mnemonic[1])
1650 /* fclex, fdecstp, fdisi, femms, feni, fincstp, finit, fsetpm, and
1651 the fs segment override prefix not currently handled because no
1652 call path can make opcodes without operands get here */
1653 case 'i':
1654 return 2 /* integer op */;
1655 case 'l':
1656 if (mnemonic[2] == 'd' && (mnemonic[3] == 'c' || mnemonic[3] == 'e'))
1657 return 3; /* fldcw/fldenv */
1658 break;
1659 case 'n':
1660 if (mnemonic[2] != 'o' /* fnop */)
1661 return 3; /* non-waiting control op */
1662 break;
1663 case 'r':
1664 if (mnemonic[2] == 's')
1665 return 3; /* frstor/frstpm */
1666 break;
1667 case 's':
1668 if (mnemonic[2] == 'a')
1669 return 3; /* fsave */
1670 if (mnemonic[2] == 't')
1672 switch (mnemonic[3])
1674 case 'c': /* fstcw */
1675 case 'd': /* fstdw */
1676 case 'e': /* fstenv */
1677 case 's': /* fsts[gw] */
1678 return 3;
1681 break;
1682 case 'x':
1683 if (mnemonic[2] == 'r' || mnemonic[2] == 's')
1684 return 0; /* fxsave/fxrstor are not really math ops */
1685 break;
1688 return 1;
1691 /* This is the guts of the machine-dependent assembler. LINE points to a
1692 machine dependent instruction. This function is supposed to emit
1693 the frags/bytes it assembles to. */
1695 void
1696 md_assemble (line)
1697 char *line;
1699 int j;
1700 char mnemonic[MAX_MNEM_SIZE];
1702 /* Initialize globals. */
1703 memset (&i, '\0', sizeof (i));
1704 for (j = 0; j < MAX_OPERANDS; j++)
1705 i.reloc[j] = NO_RELOC;
1706 memset (disp_expressions, '\0', sizeof (disp_expressions));
1707 memset (im_expressions, '\0', sizeof (im_expressions));
1708 save_stack_p = save_stack;
1710 /* First parse an instruction mnemonic & call i386_operand for the operands.
1711 We assume that the scrubber has arranged it so that line[0] is the valid
1712 start of a (possibly prefixed) mnemonic. */
1714 line = parse_insn (line, mnemonic);
1715 if (line == NULL)
1716 return;
1718 line = parse_operands (line, mnemonic);
1719 if (line == NULL)
1720 return;
1722 /* The order of the immediates should be reversed
1723 for 2 immediates extrq and insertq instructions */
1724 if ((i.imm_operands == 2) &&
1725 ((strcmp (mnemonic, "extrq") == 0)
1726 || (strcmp (mnemonic, "insertq") == 0)))
1728 swap_imm_operands ();
1729 /* "extrq" and insertq" are the only two instructions whose operands
1730 have to be reversed even though they have two immediate operands.
1732 if (intel_syntax)
1733 swap_operands ();
1736 /* Now we've parsed the mnemonic into a set of templates, and have the
1737 operands at hand. */
1739 /* All intel opcodes have reversed operands except for "bound" and
1740 "enter". We also don't reverse intersegment "jmp" and "call"
1741 instructions with 2 immediate operands so that the immediate segment
1742 precedes the offset, as it does when in AT&T mode. */
1743 if (intel_syntax && i.operands > 1
1744 && (strcmp (mnemonic, "bound") != 0)
1745 && (strcmp (mnemonic, "invlpga") != 0)
1746 && !((i.types[0] & Imm) && (i.types[1] & Imm)))
1747 swap_operands ();
1749 if (i.imm_operands)
1750 optimize_imm ();
1752 /* Don't optimize displacement for movabs since it only takes 64bit
1753 displacement. */
1754 if (i.disp_operands
1755 && (flag_code != CODE_64BIT
1756 || strcmp (mnemonic, "movabs") != 0))
1757 optimize_disp ();
1759 /* Next, we find a template that matches the given insn,
1760 making sure the overlap of the given operands types is consistent
1761 with the template operand types. */
1763 if (!match_template ())
1764 return;
1766 if (intel_syntax)
1768 /* Undo SYSV386_COMPAT brokenness when in Intel mode. See i386.h */
1769 if (SYSV386_COMPAT
1770 && (i.tm.base_opcode & 0xfffffde0) == 0xdce0)
1771 i.tm.base_opcode ^= FloatR;
1773 /* Zap movzx and movsx suffix. The suffix may have been set from
1774 "word ptr" or "byte ptr" on the source operand, but we'll use
1775 the suffix later to choose the destination register. */
1776 if ((i.tm.base_opcode & ~9) == 0x0fb6)
1778 if (i.reg_operands < 2
1779 && !i.suffix
1780 && (~i.tm.opcode_modifier
1781 & (No_bSuf
1782 | No_wSuf
1783 | No_lSuf
1784 | No_sSuf
1785 | No_xSuf
1786 | No_qSuf)))
1787 as_bad (_("ambiguous operand size for `%s'"), i.tm.name);
1789 i.suffix = 0;
1793 if (i.tm.opcode_modifier & FWait)
1794 if (!add_prefix (FWAIT_OPCODE))
1795 return;
1797 /* Check string instruction segment overrides. */
1798 if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0)
1800 if (!check_string ())
1801 return;
1804 if (!process_suffix ())
1805 return;
1807 /* Make still unresolved immediate matches conform to size of immediate
1808 given in i.suffix. */
1809 if (!finalize_imm ())
1810 return;
1812 if (i.types[0] & Imm1)
1813 i.imm_operands = 0; /* kludge for shift insns. */
1814 if (i.types[0] & ImplicitRegister)
1815 i.reg_operands--;
1816 if (i.types[1] & ImplicitRegister)
1817 i.reg_operands--;
1818 if (i.types[2] & ImplicitRegister)
1819 i.reg_operands--;
1821 if (i.tm.opcode_modifier & ImmExt)
1823 expressionS *exp;
1825 if ((i.tm.cpu_flags & CpuSSE3) && i.operands > 0)
1827 /* Streaming SIMD extensions 3 Instructions have the fixed
1828 operands with an opcode suffix which is coded in the same
1829 place as an 8-bit immediate field would be. Here we check
1830 those operands and remove them afterwards. */
1831 unsigned int x;
1833 for (x = 0; x < i.operands; x++)
1834 if (i.op[x].regs->reg_num != x)
1835 as_bad (_("can't use register '%%%s' as operand %d in '%s'."),
1836 i.op[x].regs->reg_name, x + 1, i.tm.name);
1837 i.operands = 0;
1840 /* These AMD 3DNow! and Intel Katmai New Instructions have an
1841 opcode suffix which is coded in the same place as an 8-bit
1842 immediate field would be. Here we fake an 8-bit immediate
1843 operand from the opcode suffix stored in tm.extension_opcode. */
1845 assert (i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS);
1847 exp = &im_expressions[i.imm_operands++];
1848 i.op[i.operands].imms = exp;
1849 i.types[i.operands++] = Imm8;
1850 exp->X_op = O_constant;
1851 exp->X_add_number = i.tm.extension_opcode;
1852 i.tm.extension_opcode = None;
1855 /* For insns with operands there are more diddles to do to the opcode. */
1856 if (i.operands)
1858 if (!process_operands ())
1859 return;
1861 else if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
1863 /* UnixWare fsub no args is alias for fsubp, fadd -> faddp, etc. */
1864 as_warn (_("translating to `%sp'"), i.tm.name);
1867 /* Handle conversion of 'int $3' --> special int3 insn. */
1868 if (i.tm.base_opcode == INT_OPCODE && i.op[0].imms->X_add_number == 3)
1870 i.tm.base_opcode = INT3_OPCODE;
1871 i.imm_operands = 0;
1874 if ((i.tm.opcode_modifier & (Jump | JumpByte | JumpDword))
1875 && i.op[0].disps->X_op == O_constant)
1877 /* Convert "jmp constant" (and "call constant") to a jump (call) to
1878 the absolute address given by the constant. Since ix86 jumps and
1879 calls are pc relative, we need to generate a reloc. */
1880 i.op[0].disps->X_add_symbol = &abs_symbol;
1881 i.op[0].disps->X_op = O_symbol;
1884 if ((i.tm.opcode_modifier & Rex64) != 0)
1885 i.rex |= REX_MODE64;
1887 /* For 8 bit registers we need an empty rex prefix. Also if the
1888 instruction already has a prefix, we need to convert old
1889 registers to new ones. */
1891 if (((i.types[0] & Reg8) != 0
1892 && (i.op[0].regs->reg_flags & RegRex64) != 0)
1893 || ((i.types[1] & Reg8) != 0
1894 && (i.op[1].regs->reg_flags & RegRex64) != 0)
1895 || (((i.types[0] & Reg8) != 0 || (i.types[1] & Reg8) != 0)
1896 && i.rex != 0))
1898 int x;
1900 i.rex |= REX_OPCODE;
1901 for (x = 0; x < 2; x++)
1903 /* Look for 8 bit operand that uses old registers. */
1904 if ((i.types[x] & Reg8) != 0
1905 && (i.op[x].regs->reg_flags & RegRex64) == 0)
1907 /* In case it is "hi" register, give up. */
1908 if (i.op[x].regs->reg_num > 3)
1909 as_bad (_("can't encode register '%%%s' in an instruction requiring REX prefix."),
1910 i.op[x].regs->reg_name);
1912 /* Otherwise it is equivalent to the extended register.
1913 Since the encoding doesn't change this is merely
1914 cosmetic cleanup for debug output. */
1916 i.op[x].regs = i.op[x].regs + 8;
1921 if (i.rex != 0)
1922 add_prefix (REX_OPCODE | i.rex);
1924 /* We are ready to output the insn. */
1925 output_insn ();
1928 static char *
1929 parse_insn (char *line, char *mnemonic)
1931 char *l = line;
1932 char *token_start = l;
1933 char *mnem_p;
1934 int supported;
1935 const template *t;
1937 /* Non-zero if we found a prefix only acceptable with string insns. */
1938 const char *expecting_string_instruction = NULL;
1940 while (1)
1942 mnem_p = mnemonic;
1943 while ((*mnem_p = mnemonic_chars[(unsigned char) *l]) != 0)
1945 mnem_p++;
1946 if (mnem_p >= mnemonic + MAX_MNEM_SIZE)
1948 as_bad (_("no such instruction: `%s'"), token_start);
1949 return NULL;
1951 l++;
1953 if (!is_space_char (*l)
1954 && *l != END_OF_INSN
1955 && (intel_syntax
1956 || (*l != PREFIX_SEPARATOR
1957 && *l != ',')))
1959 as_bad (_("invalid character %s in mnemonic"),
1960 output_invalid (*l));
1961 return NULL;
1963 if (token_start == l)
1965 if (!intel_syntax && *l == PREFIX_SEPARATOR)
1966 as_bad (_("expecting prefix; got nothing"));
1967 else
1968 as_bad (_("expecting mnemonic; got nothing"));
1969 return NULL;
1972 /* Look up instruction (or prefix) via hash table. */
1973 current_templates = hash_find (op_hash, mnemonic);
1975 if (*l != END_OF_INSN
1976 && (!is_space_char (*l) || l[1] != END_OF_INSN)
1977 && current_templates
1978 && (current_templates->start->opcode_modifier & IsPrefix))
1980 if (current_templates->start->cpu_flags
1981 & (flag_code != CODE_64BIT ? Cpu64 : CpuNo64))
1983 as_bad ((flag_code != CODE_64BIT
1984 ? _("`%s' is only supported in 64-bit mode")
1985 : _("`%s' is not supported in 64-bit mode")),
1986 current_templates->start->name);
1987 return NULL;
1989 /* If we are in 16-bit mode, do not allow addr16 or data16.
1990 Similarly, in 32-bit mode, do not allow addr32 or data32. */
1991 if ((current_templates->start->opcode_modifier & (Size16 | Size32))
1992 && flag_code != CODE_64BIT
1993 && (((current_templates->start->opcode_modifier & Size32) != 0)
1994 ^ (flag_code == CODE_16BIT)))
1996 as_bad (_("redundant %s prefix"),
1997 current_templates->start->name);
1998 return NULL;
2000 /* Add prefix, checking for repeated prefixes. */
2001 switch (add_prefix (current_templates->start->base_opcode))
2003 case 0:
2004 return NULL;
2005 case 2:
2006 expecting_string_instruction = current_templates->start->name;
2007 break;
2009 /* Skip past PREFIX_SEPARATOR and reset token_start. */
2010 token_start = ++l;
2012 else
2013 break;
2016 if (!current_templates)
2018 /* See if we can get a match by trimming off a suffix. */
2019 switch (mnem_p[-1])
2021 case WORD_MNEM_SUFFIX:
2022 if (intel_syntax && (intel_float_operand (mnemonic) & 2))
2023 i.suffix = SHORT_MNEM_SUFFIX;
2024 else
2025 case BYTE_MNEM_SUFFIX:
2026 case QWORD_MNEM_SUFFIX:
2027 i.suffix = mnem_p[-1];
2028 mnem_p[-1] = '\0';
2029 current_templates = hash_find (op_hash, mnemonic);
2030 break;
2031 case SHORT_MNEM_SUFFIX:
2032 case LONG_MNEM_SUFFIX:
2033 if (!intel_syntax)
2035 i.suffix = mnem_p[-1];
2036 mnem_p[-1] = '\0';
2037 current_templates = hash_find (op_hash, mnemonic);
2039 break;
2041 /* Intel Syntax. */
2042 case 'd':
2043 if (intel_syntax)
2045 if (intel_float_operand (mnemonic) == 1)
2046 i.suffix = SHORT_MNEM_SUFFIX;
2047 else
2048 i.suffix = LONG_MNEM_SUFFIX;
2049 mnem_p[-1] = '\0';
2050 current_templates = hash_find (op_hash, mnemonic);
2052 break;
2054 if (!current_templates)
2056 as_bad (_("no such instruction: `%s'"), token_start);
2057 return NULL;
2061 if (current_templates->start->opcode_modifier & (Jump | JumpByte))
2063 /* Check for a branch hint. We allow ",pt" and ",pn" for
2064 predict taken and predict not taken respectively.
2065 I'm not sure that branch hints actually do anything on loop
2066 and jcxz insns (JumpByte) for current Pentium4 chips. They
2067 may work in the future and it doesn't hurt to accept them
2068 now. */
2069 if (l[0] == ',' && l[1] == 'p')
2071 if (l[2] == 't')
2073 if (!add_prefix (DS_PREFIX_OPCODE))
2074 return NULL;
2075 l += 3;
2077 else if (l[2] == 'n')
2079 if (!add_prefix (CS_PREFIX_OPCODE))
2080 return NULL;
2081 l += 3;
2085 /* Any other comma loses. */
2086 if (*l == ',')
2088 as_bad (_("invalid character %s in mnemonic"),
2089 output_invalid (*l));
2090 return NULL;
2093 /* Check if instruction is supported on specified architecture. */
2094 supported = 0;
2095 for (t = current_templates->start; t < current_templates->end; ++t)
2097 if (!((t->cpu_flags & ~(Cpu64 | CpuNo64))
2098 & ~(cpu_arch_flags & ~(Cpu64 | CpuNo64))))
2099 supported |= 1;
2100 if (!(t->cpu_flags & (flag_code == CODE_64BIT ? CpuNo64 : Cpu64)))
2101 supported |= 2;
2103 if (!(supported & 2))
2105 as_bad (flag_code == CODE_64BIT
2106 ? _("`%s' is not supported in 64-bit mode")
2107 : _("`%s' is only supported in 64-bit mode"),
2108 current_templates->start->name);
2109 return NULL;
2111 if (!(supported & 1))
2113 as_warn (_("`%s' is not supported on `%s%s'"),
2114 current_templates->start->name,
2115 cpu_arch_name,
2116 cpu_sub_arch_name ? cpu_sub_arch_name : "");
2118 else if ((Cpu386 & ~cpu_arch_flags) && (flag_code != CODE_16BIT))
2120 as_warn (_("use .code16 to ensure correct addressing mode"));
2123 /* Check for rep/repne without a string instruction. */
2124 if (expecting_string_instruction)
2126 static templates override;
2128 for (t = current_templates->start; t < current_templates->end; ++t)
2129 if (t->opcode_modifier & IsString)
2130 break;
2131 if (t >= current_templates->end)
2133 as_bad (_("expecting string instruction after `%s'"),
2134 expecting_string_instruction);
2135 return NULL;
2137 for (override.start = t; t < current_templates->end; ++t)
2138 if (!(t->opcode_modifier & IsString))
2139 break;
2140 override.end = t;
2141 current_templates = &override;
2144 return l;
2147 static char *
2148 parse_operands (char *l, const char *mnemonic)
2150 char *token_start;
2152 /* 1 if operand is pending after ','. */
2153 unsigned int expecting_operand = 0;
2155 /* Non-zero if operand parens not balanced. */
2156 unsigned int paren_not_balanced;
2158 while (*l != END_OF_INSN)
2160 /* Skip optional white space before operand. */
2161 if (is_space_char (*l))
2162 ++l;
2163 if (!is_operand_char (*l) && *l != END_OF_INSN)
2165 as_bad (_("invalid character %s before operand %d"),
2166 output_invalid (*l),
2167 i.operands + 1);
2168 return NULL;
2170 token_start = l; /* after white space */
2171 paren_not_balanced = 0;
2172 while (paren_not_balanced || *l != ',')
2174 if (*l == END_OF_INSN)
2176 if (paren_not_balanced)
2178 if (!intel_syntax)
2179 as_bad (_("unbalanced parenthesis in operand %d."),
2180 i.operands + 1);
2181 else
2182 as_bad (_("unbalanced brackets in operand %d."),
2183 i.operands + 1);
2184 return NULL;
2186 else
2187 break; /* we are done */
2189 else if (!is_operand_char (*l) && !is_space_char (*l))
2191 as_bad (_("invalid character %s in operand %d"),
2192 output_invalid (*l),
2193 i.operands + 1);
2194 return NULL;
2196 if (!intel_syntax)
2198 if (*l == '(')
2199 ++paren_not_balanced;
2200 if (*l == ')')
2201 --paren_not_balanced;
2203 else
2205 if (*l == '[')
2206 ++paren_not_balanced;
2207 if (*l == ']')
2208 --paren_not_balanced;
2210 l++;
2212 if (l != token_start)
2213 { /* Yes, we've read in another operand. */
2214 unsigned int operand_ok;
2215 this_operand = i.operands++;
2216 if (i.operands > MAX_OPERANDS)
2218 as_bad (_("spurious operands; (%d operands/instruction max)"),
2219 MAX_OPERANDS);
2220 return NULL;
2222 /* Now parse operand adding info to 'i' as we go along. */
2223 END_STRING_AND_SAVE (l);
2225 if (intel_syntax)
2226 operand_ok =
2227 i386_intel_operand (token_start,
2228 intel_float_operand (mnemonic));
2229 else
2230 operand_ok = i386_operand (token_start);
2232 RESTORE_END_STRING (l);
2233 if (!operand_ok)
2234 return NULL;
2236 else
2238 if (expecting_operand)
2240 expecting_operand_after_comma:
2241 as_bad (_("expecting operand after ','; got nothing"));
2242 return NULL;
2244 if (*l == ',')
2246 as_bad (_("expecting operand before ','; got nothing"));
2247 return NULL;
2251 /* Now *l must be either ',' or END_OF_INSN. */
2252 if (*l == ',')
2254 if (*++l == END_OF_INSN)
2256 /* Just skip it, if it's \n complain. */
2257 goto expecting_operand_after_comma;
2259 expecting_operand = 1;
2262 return l;
2265 static void
2266 swap_imm_operands (void)
2268 union i386_op temp_op;
2269 unsigned int temp_type;
2270 enum bfd_reloc_code_real temp_reloc;
2271 int xchg1 = 0;
2272 int xchg2 = 1;
2274 temp_type = i.types[xchg2];
2275 i.types[xchg2] = i.types[xchg1];
2276 i.types[xchg1] = temp_type;
2277 temp_op = i.op[xchg2];
2278 i.op[xchg2] = i.op[xchg1];
2279 i.op[xchg1] = temp_op;
2280 temp_reloc = i.reloc[xchg2];
2281 i.reloc[xchg2] = i.reloc[xchg1];
2282 i.reloc[xchg1] = temp_reloc;
2286 static void
2287 swap_operands (void)
2289 union i386_op temp_op;
2290 unsigned int temp_type;
2291 enum bfd_reloc_code_real temp_reloc;
2292 int xchg1, xchg2;
2294 switch (i.operands)
2296 case 4:
2297 xchg1 = 1;
2298 xchg2 = i.operands - 2;
2299 temp_type = i.types[xchg2];
2300 i.types[xchg2] = i.types[xchg1];
2301 i.types[xchg1] = temp_type;
2302 temp_op = i.op[xchg2];
2303 i.op[xchg2] = i.op[xchg1];
2304 i.op[xchg1] = temp_op;
2305 temp_reloc = i.reloc[xchg2];
2306 i.reloc[xchg2] = i.reloc[xchg1];
2307 i.reloc[xchg1] = temp_reloc;
2308 case 3:
2309 case 2:
2310 xchg1 = 0;
2311 xchg2 = i.operands - 1;
2312 temp_type = i.types[xchg2];
2313 i.types[xchg2] = i.types[xchg1];
2314 i.types[xchg1] = temp_type;
2315 temp_op = i.op[xchg2];
2316 i.op[xchg2] = i.op[xchg1];
2317 i.op[xchg1] = temp_op;
2318 temp_reloc = i.reloc[xchg2];
2319 i.reloc[xchg2] = i.reloc[xchg1];
2320 i.reloc[xchg1] = temp_reloc;
2321 break;
2322 default:
2323 abort ();
2326 if (i.mem_operands == 2)
2328 const seg_entry *temp_seg;
2329 temp_seg = i.seg[0];
2330 i.seg[0] = i.seg[1];
2331 i.seg[1] = temp_seg;
2335 /* Try to ensure constant immediates are represented in the smallest
2336 opcode possible. */
2337 static void
2338 optimize_imm (void)
2340 char guess_suffix = 0;
2341 int op;
2343 if (i.suffix)
2344 guess_suffix = i.suffix;
2345 else if (i.reg_operands)
2347 /* Figure out a suffix from the last register operand specified.
2348 We can't do this properly yet, ie. excluding InOutPortReg,
2349 but the following works for instructions with immediates.
2350 In any case, we can't set i.suffix yet. */
2351 for (op = i.operands; --op >= 0;)
2352 if (i.types[op] & Reg)
2354 if (i.types[op] & Reg8)
2355 guess_suffix = BYTE_MNEM_SUFFIX;
2356 else if (i.types[op] & Reg16)
2357 guess_suffix = WORD_MNEM_SUFFIX;
2358 else if (i.types[op] & Reg32)
2359 guess_suffix = LONG_MNEM_SUFFIX;
2360 else if (i.types[op] & Reg64)
2361 guess_suffix = QWORD_MNEM_SUFFIX;
2362 break;
2365 else if ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0))
2366 guess_suffix = WORD_MNEM_SUFFIX;
2368 for (op = i.operands; --op >= 0;)
2369 if (i.types[op] & Imm)
2371 switch (i.op[op].imms->X_op)
2373 case O_constant:
2374 /* If a suffix is given, this operand may be shortened. */
2375 switch (guess_suffix)
2377 case LONG_MNEM_SUFFIX:
2378 i.types[op] |= Imm32 | Imm64;
2379 break;
2380 case WORD_MNEM_SUFFIX:
2381 i.types[op] |= Imm16 | Imm32S | Imm32 | Imm64;
2382 break;
2383 case BYTE_MNEM_SUFFIX:
2384 i.types[op] |= Imm16 | Imm8 | Imm8S | Imm32S | Imm32 | Imm64;
2385 break;
2388 /* If this operand is at most 16 bits, convert it
2389 to a signed 16 bit number before trying to see
2390 whether it will fit in an even smaller size.
2391 This allows a 16-bit operand such as $0xffe0 to
2392 be recognised as within Imm8S range. */
2393 if ((i.types[op] & Imm16)
2394 && (i.op[op].imms->X_add_number & ~(offsetT) 0xffff) == 0)
2396 i.op[op].imms->X_add_number =
2397 (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000);
2399 if ((i.types[op] & Imm32)
2400 && ((i.op[op].imms->X_add_number & ~(((offsetT) 2 << 31) - 1))
2401 == 0))
2403 i.op[op].imms->X_add_number = ((i.op[op].imms->X_add_number
2404 ^ ((offsetT) 1 << 31))
2405 - ((offsetT) 1 << 31));
2407 i.types[op] |= smallest_imm_type (i.op[op].imms->X_add_number);
2409 /* We must avoid matching of Imm32 templates when 64bit
2410 only immediate is available. */
2411 if (guess_suffix == QWORD_MNEM_SUFFIX)
2412 i.types[op] &= ~Imm32;
2413 break;
2415 case O_absent:
2416 case O_register:
2417 abort ();
2419 /* Symbols and expressions. */
2420 default:
2421 /* Convert symbolic operand to proper sizes for matching, but don't
2422 prevent matching a set of insns that only supports sizes other
2423 than those matching the insn suffix. */
2425 unsigned int mask, allowed = 0;
2426 const template *t;
2428 for (t = current_templates->start; t < current_templates->end; ++t)
2429 allowed |= t->operand_types[op];
2430 switch (guess_suffix)
2432 case QWORD_MNEM_SUFFIX:
2433 mask = Imm64 | Imm32S;
2434 break;
2435 case LONG_MNEM_SUFFIX:
2436 mask = Imm32;
2437 break;
2438 case WORD_MNEM_SUFFIX:
2439 mask = Imm16;
2440 break;
2441 case BYTE_MNEM_SUFFIX:
2442 mask = Imm8;
2443 break;
2444 default:
2445 mask = 0;
2446 break;
2448 if (mask & allowed)
2449 i.types[op] &= mask;
2451 break;
2456 /* Try to use the smallest displacement type too. */
2457 static void
2458 optimize_disp (void)
2460 int op;
2462 for (op = i.operands; --op >= 0;)
2463 if (i.types[op] & Disp)
2465 if (i.op[op].disps->X_op == O_constant)
2467 offsetT disp = i.op[op].disps->X_add_number;
2469 if ((i.types[op] & Disp16)
2470 && (disp & ~(offsetT) 0xffff) == 0)
2472 /* If this operand is at most 16 bits, convert
2473 to a signed 16 bit number and don't use 64bit
2474 displacement. */
2475 disp = (((disp & 0xffff) ^ 0x8000) - 0x8000);
2476 i.types[op] &= ~Disp64;
2478 if ((i.types[op] & Disp32)
2479 && (disp & ~(((offsetT) 2 << 31) - 1)) == 0)
2481 /* If this operand is at most 32 bits, convert
2482 to a signed 32 bit number and don't use 64bit
2483 displacement. */
2484 disp &= (((offsetT) 2 << 31) - 1);
2485 disp = (disp ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
2486 i.types[op] &= ~Disp64;
2488 if (!disp && (i.types[op] & BaseIndex))
2490 i.types[op] &= ~Disp;
2491 i.op[op].disps = 0;
2492 i.disp_operands--;
2494 else if (flag_code == CODE_64BIT)
2496 if (fits_in_signed_long (disp))
2498 i.types[op] &= ~Disp64;
2499 i.types[op] |= Disp32S;
2501 if (fits_in_unsigned_long (disp))
2502 i.types[op] |= Disp32;
2504 if ((i.types[op] & (Disp32 | Disp32S | Disp16))
2505 && fits_in_signed_byte (disp))
2506 i.types[op] |= Disp8;
2508 else if (i.reloc[op] == BFD_RELOC_386_TLS_DESC_CALL
2509 || i.reloc[op] == BFD_RELOC_X86_64_TLSDESC_CALL)
2511 fix_new_exp (frag_now, frag_more (0) - frag_now->fr_literal, 0,
2512 i.op[op].disps, 0, i.reloc[op]);
2513 i.types[op] &= ~Disp;
2515 else
2516 /* We only support 64bit displacement on constants. */
2517 i.types[op] &= ~Disp64;
2521 static int
2522 match_template (void)
2524 /* Points to template once we've found it. */
2525 const template *t;
2526 unsigned int overlap0, overlap1, overlap2, overlap3;
2527 unsigned int found_reverse_match;
2528 int suffix_check;
2529 unsigned int operand_types [MAX_OPERANDS];
2530 int addr_prefix_disp;
2531 unsigned int j;
2533 #if MAX_OPERANDS != 4
2534 # error "MAX_OPERANDS must be 4."
2535 #endif
2537 #define MATCH(overlap, given, template) \
2538 ((overlap & ~JumpAbsolute) \
2539 && (((given) & (BaseIndex | JumpAbsolute)) \
2540 == ((overlap) & (BaseIndex | JumpAbsolute))))
2542 /* If given types r0 and r1 are registers they must be of the same type
2543 unless the expected operand type register overlap is null.
2544 Note that Acc in a template matches every size of reg. */
2545 #define CONSISTENT_REGISTER_MATCH(m0, g0, t0, m1, g1, t1) \
2546 (((g0) & Reg) == 0 || ((g1) & Reg) == 0 \
2547 || ((g0) & Reg) == ((g1) & Reg) \
2548 || ((((m0) & Acc) ? Reg : (t0)) & (((m1) & Acc) ? Reg : (t1)) & Reg) == 0 )
2550 overlap0 = 0;
2551 overlap1 = 0;
2552 overlap2 = 0;
2553 overlap3 = 0;
2554 found_reverse_match = 0;
2555 for (j = 0; j < MAX_OPERANDS; j++)
2556 operand_types [j] = 0;
2557 addr_prefix_disp = -1;
2558 suffix_check = (i.suffix == BYTE_MNEM_SUFFIX
2559 ? No_bSuf
2560 : (i.suffix == WORD_MNEM_SUFFIX
2561 ? No_wSuf
2562 : (i.suffix == SHORT_MNEM_SUFFIX
2563 ? No_sSuf
2564 : (i.suffix == LONG_MNEM_SUFFIX
2565 ? No_lSuf
2566 : (i.suffix == QWORD_MNEM_SUFFIX
2567 ? No_qSuf
2568 : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX
2569 ? No_xSuf : 0))))));
2571 for (t = current_templates->start; t < current_templates->end; t++)
2573 addr_prefix_disp = -1;
2575 /* Must have right number of operands. */
2576 if (i.operands != t->operands)
2577 continue;
2579 /* Check the suffix, except for some instructions in intel mode. */
2580 if ((t->opcode_modifier & suffix_check)
2581 && !(intel_syntax
2582 && (t->opcode_modifier & IgnoreSize)))
2583 continue;
2585 for (j = 0; j < MAX_OPERANDS; j++)
2586 operand_types [j] = t->operand_types [j];
2588 /* In general, don't allow 64-bit operands in 32-bit mode. */
2589 if (i.suffix == QWORD_MNEM_SUFFIX
2590 && flag_code != CODE_64BIT
2591 && (intel_syntax
2592 ? (!(t->opcode_modifier & IgnoreSize)
2593 && !intel_float_operand (t->name))
2594 : intel_float_operand (t->name) != 2)
2595 && (!(operand_types[0] & (RegMMX | RegXMM))
2596 || !(operand_types[t->operands > 1] & (RegMMX | RegXMM)))
2597 && (t->base_opcode != 0x0fc7
2598 || t->extension_opcode != 1 /* cmpxchg8b */))
2599 continue;
2601 /* Do not verify operands when there are none. */
2602 else if (!t->operands)
2604 if (t->cpu_flags & ~cpu_arch_flags)
2605 continue;
2606 /* We've found a match; break out of loop. */
2607 break;
2610 /* Address size prefix will turn Disp64/Disp32/Disp16 operand
2611 into Disp32/Disp16/Disp32 operand. */
2612 if (i.prefix[ADDR_PREFIX] != 0)
2614 unsigned int DispOn = 0, DispOff = 0;
2616 switch (flag_code)
2618 case CODE_16BIT:
2619 DispOn = Disp32;
2620 DispOff = Disp16;
2621 break;
2622 case CODE_32BIT:
2623 DispOn = Disp16;
2624 DispOff = Disp32;
2625 break;
2626 case CODE_64BIT:
2627 DispOn = Disp32;
2628 DispOff = Disp64;
2629 break;
2632 for (j = 0; j < MAX_OPERANDS; j++)
2634 /* There should be only one Disp operand. */
2635 if ((operand_types[j] & DispOff))
2637 addr_prefix_disp = j;
2638 operand_types[j] |= DispOn;
2639 operand_types[j] &= ~DispOff;
2640 break;
2645 overlap0 = i.types[0] & operand_types[0];
2646 switch (t->operands)
2648 case 1:
2649 if (!MATCH (overlap0, i.types[0], operand_types[0]))
2650 continue;
2651 break;
2652 case 2:
2653 case 3:
2654 case 4:
2655 overlap1 = i.types[1] & operand_types[1];
2656 if (!MATCH (overlap0, i.types[0], operand_types[0])
2657 || !MATCH (overlap1, i.types[1], operand_types[1])
2658 /* monitor in SSE3 is a very special case. The first
2659 register and the second register may have different
2660 sizes. */
2661 || !((t->base_opcode == 0x0f01
2662 && t->extension_opcode == 0xc8)
2663 || CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2664 operand_types[0],
2665 overlap1, i.types[1],
2666 operand_types[1])))
2668 /* Check if other direction is valid ... */
2669 if ((t->opcode_modifier & (D | FloatD)) == 0)
2670 continue;
2672 /* Try reversing direction of operands. */
2673 overlap0 = i.types[0] & operand_types[1];
2674 overlap1 = i.types[1] & operand_types[0];
2675 if (!MATCH (overlap0, i.types[0], operand_types[1])
2676 || !MATCH (overlap1, i.types[1], operand_types[0])
2677 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2678 operand_types[1],
2679 overlap1, i.types[1],
2680 operand_types[0]))
2682 /* Does not match either direction. */
2683 continue;
2685 /* found_reverse_match holds which of D or FloatDR
2686 we've found. */
2687 found_reverse_match = t->opcode_modifier & (D | FloatDR);
2689 else
2691 /* Found a forward 2 operand match here. */
2692 switch (t->operands)
2694 case 4:
2695 overlap3 = i.types[3] & operand_types[3];
2696 case 3:
2697 overlap2 = i.types[2] & operand_types[2];
2698 break;
2701 switch (t->operands)
2703 case 4:
2704 if (!MATCH (overlap3, i.types[3], operand_types[3])
2705 || !CONSISTENT_REGISTER_MATCH (overlap2,
2706 i.types[2],
2707 operand_types[2],
2708 overlap3,
2709 i.types[3],
2710 operand_types[3]))
2711 continue;
2712 case 3:
2713 /* Here we make use of the fact that there are no
2714 reverse match 3 operand instructions, and all 3
2715 operand instructions only need to be checked for
2716 register consistency between operands 2 and 3. */
2717 if (!MATCH (overlap2, i.types[2], operand_types[2])
2718 || !CONSISTENT_REGISTER_MATCH (overlap1,
2719 i.types[1],
2720 operand_types[1],
2721 overlap2,
2722 i.types[2],
2723 operand_types[2]))
2724 continue;
2725 break;
2728 /* Found either forward/reverse 2, 3 or 4 operand match here:
2729 slip through to break. */
2731 if (t->cpu_flags & ~cpu_arch_flags)
2733 found_reverse_match = 0;
2734 continue;
2736 /* We've found a match; break out of loop. */
2737 break;
2740 if (t == current_templates->end)
2742 /* We found no match. */
2743 as_bad (_("suffix or operands invalid for `%s'"),
2744 current_templates->start->name);
2745 return 0;
2748 if (!quiet_warnings)
2750 if (!intel_syntax
2751 && ((i.types[0] & JumpAbsolute)
2752 != (operand_types[0] & JumpAbsolute)))
2754 as_warn (_("indirect %s without `*'"), t->name);
2757 if ((t->opcode_modifier & (IsPrefix | IgnoreSize))
2758 == (IsPrefix | IgnoreSize))
2760 /* Warn them that a data or address size prefix doesn't
2761 affect assembly of the next line of code. */
2762 as_warn (_("stand-alone `%s' prefix"), t->name);
2766 /* Copy the template we found. */
2767 i.tm = *t;
2769 if (addr_prefix_disp != -1)
2770 i.tm.operand_types[addr_prefix_disp]
2771 = operand_types[addr_prefix_disp];
2773 if (found_reverse_match)
2775 /* If we found a reverse match we must alter the opcode
2776 direction bit. found_reverse_match holds bits to change
2777 (different for int & float insns). */
2779 i.tm.base_opcode ^= found_reverse_match;
2781 i.tm.operand_types[0] = operand_types[1];
2782 i.tm.operand_types[1] = operand_types[0];
2785 return 1;
2788 static int
2789 check_string (void)
2791 int mem_op = (i.types[0] & AnyMem) ? 0 : 1;
2792 if ((i.tm.operand_types[mem_op] & EsSeg) != 0)
2794 if (i.seg[0] != NULL && i.seg[0] != &es)
2796 as_bad (_("`%s' operand %d must use `%%es' segment"),
2797 i.tm.name,
2798 mem_op + 1);
2799 return 0;
2801 /* There's only ever one segment override allowed per instruction.
2802 This instruction possibly has a legal segment override on the
2803 second operand, so copy the segment to where non-string
2804 instructions store it, allowing common code. */
2805 i.seg[0] = i.seg[1];
2807 else if ((i.tm.operand_types[mem_op + 1] & EsSeg) != 0)
2809 if (i.seg[1] != NULL && i.seg[1] != &es)
2811 as_bad (_("`%s' operand %d must use `%%es' segment"),
2812 i.tm.name,
2813 mem_op + 2);
2814 return 0;
2817 return 1;
2820 static int
2821 process_suffix (void)
2823 /* If matched instruction specifies an explicit instruction mnemonic
2824 suffix, use it. */
2825 if (i.tm.opcode_modifier & (Size16 | Size32 | Size64))
2827 if (i.tm.opcode_modifier & Size16)
2828 i.suffix = WORD_MNEM_SUFFIX;
2829 else if (i.tm.opcode_modifier & Size64)
2830 i.suffix = QWORD_MNEM_SUFFIX;
2831 else
2832 i.suffix = LONG_MNEM_SUFFIX;
2834 else if (i.reg_operands)
2836 /* If there's no instruction mnemonic suffix we try to invent one
2837 based on register operands. */
2838 if (!i.suffix)
2840 /* We take i.suffix from the last register operand specified,
2841 Destination register type is more significant than source
2842 register type. */
2843 int op;
2845 for (op = i.operands; --op >= 0;)
2846 if ((i.types[op] & Reg)
2847 && !(i.tm.operand_types[op] & InOutPortReg))
2849 i.suffix = ((i.types[op] & Reg8) ? BYTE_MNEM_SUFFIX :
2850 (i.types[op] & Reg16) ? WORD_MNEM_SUFFIX :
2851 (i.types[op] & Reg64) ? QWORD_MNEM_SUFFIX :
2852 LONG_MNEM_SUFFIX);
2853 break;
2856 else if (i.suffix == BYTE_MNEM_SUFFIX)
2858 if (!check_byte_reg ())
2859 return 0;
2861 else if (i.suffix == LONG_MNEM_SUFFIX)
2863 if (!check_long_reg ())
2864 return 0;
2866 else if (i.suffix == QWORD_MNEM_SUFFIX)
2868 if (!check_qword_reg ())
2869 return 0;
2871 else if (i.suffix == WORD_MNEM_SUFFIX)
2873 if (!check_word_reg ())
2874 return 0;
2876 else if (intel_syntax && (i.tm.opcode_modifier & IgnoreSize))
2877 /* Do nothing if the instruction is going to ignore the prefix. */
2879 else
2880 abort ();
2882 else if ((i.tm.opcode_modifier & DefaultSize)
2883 && !i.suffix
2884 /* exclude fldenv/frstor/fsave/fstenv */
2885 && (i.tm.opcode_modifier & No_sSuf))
2887 i.suffix = stackop_size;
2889 else if (intel_syntax
2890 && !i.suffix
2891 && ((i.tm.operand_types[0] & JumpAbsolute)
2892 || (i.tm.opcode_modifier & (JumpByte|JumpInterSegment))
2893 || (i.tm.base_opcode == 0x0f01 /* [ls][gi]dt */
2894 && i.tm.extension_opcode <= 3)))
2896 switch (flag_code)
2898 case CODE_64BIT:
2899 if (!(i.tm.opcode_modifier & No_qSuf))
2901 i.suffix = QWORD_MNEM_SUFFIX;
2902 break;
2904 case CODE_32BIT:
2905 if (!(i.tm.opcode_modifier & No_lSuf))
2906 i.suffix = LONG_MNEM_SUFFIX;
2907 break;
2908 case CODE_16BIT:
2909 if (!(i.tm.opcode_modifier & No_wSuf))
2910 i.suffix = WORD_MNEM_SUFFIX;
2911 break;
2915 if (!i.suffix)
2917 if (!intel_syntax)
2919 if (i.tm.opcode_modifier & W)
2921 as_bad (_("no instruction mnemonic suffix given and no register operands; can't size instruction"));
2922 return 0;
2925 else
2927 unsigned int suffixes = (~i.tm.opcode_modifier
2928 & (No_bSuf
2929 | No_wSuf
2930 | No_lSuf
2931 | No_sSuf
2932 | No_xSuf
2933 | No_qSuf));
2935 if ((i.tm.opcode_modifier & W)
2936 || ((suffixes & (suffixes - 1))
2937 && !(i.tm.opcode_modifier & (DefaultSize | IgnoreSize))))
2939 as_bad (_("ambiguous operand size for `%s'"), i.tm.name);
2940 return 0;
2945 /* Change the opcode based on the operand size given by i.suffix;
2946 We don't need to change things for byte insns. */
2948 if (i.suffix && i.suffix != BYTE_MNEM_SUFFIX)
2950 /* It's not a byte, select word/dword operation. */
2951 if (i.tm.opcode_modifier & W)
2953 if (i.tm.opcode_modifier & ShortForm)
2954 i.tm.base_opcode |= 8;
2955 else
2956 i.tm.base_opcode |= 1;
2959 /* Now select between word & dword operations via the operand
2960 size prefix, except for instructions that will ignore this
2961 prefix anyway. */
2962 if (i.tm.base_opcode == 0x0f01 && i.tm.extension_opcode == 0xc8)
2964 /* monitor in SSE3 is a very special case. The default size
2965 of AX is the size of mode. The address size override
2966 prefix will change the size of AX. */
2967 if (i.op->regs[0].reg_type &
2968 (flag_code == CODE_32BIT ? Reg16 : Reg32))
2969 if (!add_prefix (ADDR_PREFIX_OPCODE))
2970 return 0;
2972 else if (i.suffix != QWORD_MNEM_SUFFIX
2973 && i.suffix != LONG_DOUBLE_MNEM_SUFFIX
2974 && !(i.tm.opcode_modifier & (IgnoreSize | FloatMF))
2975 && ((i.suffix == LONG_MNEM_SUFFIX) == (flag_code == CODE_16BIT)
2976 || (flag_code == CODE_64BIT
2977 && (i.tm.opcode_modifier & JumpByte))))
2979 unsigned int prefix = DATA_PREFIX_OPCODE;
2981 if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */
2982 prefix = ADDR_PREFIX_OPCODE;
2984 if (!add_prefix (prefix))
2985 return 0;
2988 /* Set mode64 for an operand. */
2989 if (i.suffix == QWORD_MNEM_SUFFIX
2990 && flag_code == CODE_64BIT
2991 && (i.tm.opcode_modifier & NoRex64) == 0)
2993 /* Special case for xchg %rax,%rax. It is NOP and doesn't
2994 need rex64. */
2995 if (i.operands != 2
2996 || i.types [0] != (Acc | Reg64)
2997 || i.types [1] != (Acc | Reg64)
2998 || strcmp (i.tm.name, "xchg") != 0)
2999 i.rex |= REX_MODE64;
3002 /* Size floating point instruction. */
3003 if (i.suffix == LONG_MNEM_SUFFIX)
3004 if (i.tm.opcode_modifier & FloatMF)
3005 i.tm.base_opcode ^= 4;
3008 return 1;
3011 static int
3012 check_byte_reg (void)
3014 int op;
3016 for (op = i.operands; --op >= 0;)
3018 /* If this is an eight bit register, it's OK. If it's the 16 or
3019 32 bit version of an eight bit register, we will just use the
3020 low portion, and that's OK too. */
3021 if (i.types[op] & Reg8)
3022 continue;
3024 /* movzx and movsx should not generate this warning. */
3025 if (intel_syntax
3026 && (i.tm.base_opcode == 0xfb7
3027 || i.tm.base_opcode == 0xfb6
3028 || i.tm.base_opcode == 0x63
3029 || i.tm.base_opcode == 0xfbe
3030 || i.tm.base_opcode == 0xfbf))
3031 continue;
3033 if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4)
3035 /* Prohibit these changes in the 64bit mode, since the
3036 lowering is more complicated. */
3037 if (flag_code == CODE_64BIT
3038 && (i.tm.operand_types[op] & InOutPortReg) == 0)
3040 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3041 register_prefix, i.op[op].regs->reg_name,
3042 i.suffix);
3043 return 0;
3045 #if REGISTER_WARNINGS
3046 if (!quiet_warnings
3047 && (i.tm.operand_types[op] & InOutPortReg) == 0)
3048 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
3049 (i.op[op].regs + (i.types[op] & Reg16
3050 ? REGNAM_AL - REGNAM_AX
3051 : REGNAM_AL - REGNAM_EAX))->reg_name,
3052 i.op[op].regs->reg_name,
3053 i.suffix);
3054 #endif
3055 continue;
3057 /* Any other register is bad. */
3058 if (i.types[op] & (Reg | RegMMX | RegXMM
3059 | SReg2 | SReg3
3060 | Control | Debug | Test
3061 | FloatReg | FloatAcc))
3063 as_bad (_("`%%%s' not allowed with `%s%c'"),
3064 i.op[op].regs->reg_name,
3065 i.tm.name,
3066 i.suffix);
3067 return 0;
3070 return 1;
3073 static int
3074 check_long_reg (void)
3076 int op;
3078 for (op = i.operands; --op >= 0;)
3079 /* Reject eight bit registers, except where the template requires
3080 them. (eg. movzb) */
3081 if ((i.types[op] & Reg8) != 0
3082 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
3084 as_bad (_("`%%%s' not allowed with `%s%c'"),
3085 i.op[op].regs->reg_name,
3086 i.tm.name,
3087 i.suffix);
3088 return 0;
3090 /* Warn if the e prefix on a general reg is missing. */
3091 else if ((!quiet_warnings || flag_code == CODE_64BIT)
3092 && (i.types[op] & Reg16) != 0
3093 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
3095 /* Prohibit these changes in the 64bit mode, since the
3096 lowering is more complicated. */
3097 if (flag_code == CODE_64BIT)
3099 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3100 register_prefix, i.op[op].regs->reg_name,
3101 i.suffix);
3102 return 0;
3104 #if REGISTER_WARNINGS
3105 else
3106 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
3107 (i.op[op].regs + REGNAM_EAX - REGNAM_AX)->reg_name,
3108 i.op[op].regs->reg_name,
3109 i.suffix);
3110 #endif
3112 /* Warn if the r prefix on a general reg is missing. */
3113 else if ((i.types[op] & Reg64) != 0
3114 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
3116 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3117 register_prefix, i.op[op].regs->reg_name,
3118 i.suffix);
3119 return 0;
3121 return 1;
3124 static int
3125 check_qword_reg (void)
3127 int op;
3129 for (op = i.operands; --op >= 0; )
3130 /* Reject eight bit registers, except where the template requires
3131 them. (eg. movzb) */
3132 if ((i.types[op] & Reg8) != 0
3133 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
3135 as_bad (_("`%%%s' not allowed with `%s%c'"),
3136 i.op[op].regs->reg_name,
3137 i.tm.name,
3138 i.suffix);
3139 return 0;
3141 /* Warn if the e prefix on a general reg is missing. */
3142 else if (((i.types[op] & Reg16) != 0
3143 || (i.types[op] & Reg32) != 0)
3144 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
3146 /* Prohibit these changes in the 64bit mode, since the
3147 lowering is more complicated. */
3148 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3149 register_prefix, i.op[op].regs->reg_name,
3150 i.suffix);
3151 return 0;
3153 return 1;
3156 static int
3157 check_word_reg (void)
3159 int op;
3160 for (op = i.operands; --op >= 0;)
3161 /* Reject eight bit registers, except where the template requires
3162 them. (eg. movzb) */
3163 if ((i.types[op] & Reg8) != 0
3164 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
3166 as_bad (_("`%%%s' not allowed with `%s%c'"),
3167 i.op[op].regs->reg_name,
3168 i.tm.name,
3169 i.suffix);
3170 return 0;
3172 /* Warn if the e prefix on a general reg is present. */
3173 else if ((!quiet_warnings || flag_code == CODE_64BIT)
3174 && (i.types[op] & Reg32) != 0
3175 && (i.tm.operand_types[op] & (Reg16 | Acc)) != 0)
3177 /* Prohibit these changes in the 64bit mode, since the
3178 lowering is more complicated. */
3179 if (flag_code == CODE_64BIT)
3181 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3182 register_prefix, i.op[op].regs->reg_name,
3183 i.suffix);
3184 return 0;
3186 else
3187 #if REGISTER_WARNINGS
3188 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
3189 (i.op[op].regs + REGNAM_AX - REGNAM_EAX)->reg_name,
3190 i.op[op].regs->reg_name,
3191 i.suffix);
3192 #endif
3194 return 1;
3197 static int
3198 finalize_imm (void)
3200 unsigned int overlap0, overlap1, overlap2;
3202 overlap0 = i.types[0] & i.tm.operand_types[0];
3203 if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64))
3204 && overlap0 != Imm8 && overlap0 != Imm8S
3205 && overlap0 != Imm16 && overlap0 != Imm32S
3206 && overlap0 != Imm32 && overlap0 != Imm64)
3208 if (i.suffix)
3210 overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX
3211 ? Imm8 | Imm8S
3212 : (i.suffix == WORD_MNEM_SUFFIX
3213 ? Imm16
3214 : (i.suffix == QWORD_MNEM_SUFFIX
3215 ? Imm64 | Imm32S
3216 : Imm32)));
3218 else if (overlap0 == (Imm16 | Imm32S | Imm32)
3219 || overlap0 == (Imm16 | Imm32)
3220 || overlap0 == (Imm16 | Imm32S))
3222 overlap0 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
3223 ? Imm16 : Imm32S);
3225 if (overlap0 != Imm8 && overlap0 != Imm8S
3226 && overlap0 != Imm16 && overlap0 != Imm32S
3227 && overlap0 != Imm32 && overlap0 != Imm64)
3229 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size"));
3230 return 0;
3233 i.types[0] = overlap0;
3235 overlap1 = i.types[1] & i.tm.operand_types[1];
3236 if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32 | Imm64))
3237 && overlap1 != Imm8 && overlap1 != Imm8S
3238 && overlap1 != Imm16 && overlap1 != Imm32S
3239 && overlap1 != Imm32 && overlap1 != Imm64)
3241 if (i.suffix)
3243 overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX
3244 ? Imm8 | Imm8S
3245 : (i.suffix == WORD_MNEM_SUFFIX
3246 ? Imm16
3247 : (i.suffix == QWORD_MNEM_SUFFIX
3248 ? Imm64 | Imm32S
3249 : Imm32)));
3251 else if (overlap1 == (Imm16 | Imm32 | Imm32S)
3252 || overlap1 == (Imm16 | Imm32)
3253 || overlap1 == (Imm16 | Imm32S))
3255 overlap1 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
3256 ? Imm16 : Imm32S);
3258 if (overlap1 != Imm8 && overlap1 != Imm8S
3259 && overlap1 != Imm16 && overlap1 != Imm32S
3260 && overlap1 != Imm32 && overlap1 != Imm64)
3262 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size %x %c"),overlap1, i.suffix);
3263 return 0;
3266 i.types[1] = overlap1;
3268 overlap2 = i.types[2] & i.tm.operand_types[2];
3269 assert ((overlap2 & Imm) == 0);
3270 i.types[2] = overlap2;
3272 return 1;
3275 static int
3276 process_operands (void)
3278 /* Default segment register this instruction will use for memory
3279 accesses. 0 means unknown. This is only for optimizing out
3280 unnecessary segment overrides. */
3281 const seg_entry *default_seg = 0;
3283 /* The imul $imm, %reg instruction is converted into
3284 imul $imm, %reg, %reg, and the clr %reg instruction
3285 is converted into xor %reg, %reg. */
3286 if (i.tm.opcode_modifier & regKludge)
3288 unsigned int first_reg_op = (i.types[0] & Reg) ? 0 : 1;
3289 /* Pretend we saw the extra register operand. */
3290 assert (i.reg_operands == 1
3291 && i.op[first_reg_op + 1].regs == 0);
3292 i.op[first_reg_op + 1].regs = i.op[first_reg_op].regs;
3293 i.types[first_reg_op + 1] = i.types[first_reg_op];
3294 i.operands++;
3295 i.reg_operands++;
3298 if (i.tm.opcode_modifier & ShortForm)
3300 /* The register or float register operand is in operand 0 or 1. */
3301 unsigned int op = (i.types[0] & (Reg | FloatReg)) ? 0 : 1;
3302 /* Register goes in low 3 bits of opcode. */
3303 i.tm.base_opcode |= i.op[op].regs->reg_num;
3304 if ((i.op[op].regs->reg_flags & RegRex) != 0)
3305 i.rex |= REX_EXTZ;
3306 if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
3308 /* Warn about some common errors, but press on regardless.
3309 The first case can be generated by gcc (<= 2.8.1). */
3310 if (i.operands == 2)
3312 /* Reversed arguments on faddp, fsubp, etc. */
3313 as_warn (_("translating to `%s %%%s,%%%s'"), i.tm.name,
3314 i.op[1].regs->reg_name,
3315 i.op[0].regs->reg_name);
3317 else
3319 /* Extraneous `l' suffix on fp insn. */
3320 as_warn (_("translating to `%s %%%s'"), i.tm.name,
3321 i.op[0].regs->reg_name);
3325 else if (i.tm.opcode_modifier & Modrm)
3327 /* The opcode is completed (modulo i.tm.extension_opcode which
3328 must be put into the modrm byte). Now, we make the modrm and
3329 index base bytes based on all the info we've collected. */
3331 default_seg = build_modrm_byte ();
3333 else if (i.tm.opcode_modifier & (Seg2ShortForm | Seg3ShortForm))
3335 if (i.tm.base_opcode == POP_SEG_SHORT
3336 && i.op[0].regs->reg_num == 1)
3338 as_bad (_("you can't `pop %%cs'"));
3339 return 0;
3341 i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
3342 if ((i.op[0].regs->reg_flags & RegRex) != 0)
3343 i.rex |= REX_EXTZ;
3345 else if ((i.tm.base_opcode & ~(D | W)) == MOV_AX_DISP32)
3347 default_seg = &ds;
3349 else if ((i.tm.opcode_modifier & IsString) != 0)
3351 /* For the string instructions that allow a segment override
3352 on one of their operands, the default segment is ds. */
3353 default_seg = &ds;
3356 if ((i.tm.base_opcode == 0x8d /* lea */
3357 || (i.tm.cpu_flags & CpuSVME))
3358 && i.seg[0] && !quiet_warnings)
3359 as_warn (_("segment override on `%s' is ineffectual"), i.tm.name);
3361 /* If a segment was explicitly specified, and the specified segment
3362 is not the default, use an opcode prefix to select it. If we
3363 never figured out what the default segment is, then default_seg
3364 will be zero at this point, and the specified segment prefix will
3365 always be used. */
3366 if ((i.seg[0]) && (i.seg[0] != default_seg))
3368 if (!add_prefix (i.seg[0]->seg_prefix))
3369 return 0;
3371 return 1;
3374 static const seg_entry *
3375 build_modrm_byte (void)
3377 const seg_entry *default_seg = 0;
3379 /* i.reg_operands MUST be the number of real register operands;
3380 implicit registers do not count. */
3381 if (i.reg_operands == 2)
3383 unsigned int source, dest;
3385 switch (i.operands)
3387 case 2:
3388 source = 0;
3389 break;
3390 case 3:
3391 /* When there are 3 operands, one of them may be immediate,
3392 which may be the first or the last operand. Otherwise,
3393 the first operand must be shift count register (cl). */
3394 assert (i.imm_operands == 1
3395 || (i.imm_operands == 0
3396 && (i.types[0] & ShiftCount)));
3397 source = (i.types[0] & (Imm | ShiftCount)) ? 1 : 0;
3398 break;
3399 case 4:
3400 /* When there are 4 operands, the first two must be immediate
3401 operands. The source operand will be the 3rd one. */
3402 assert (i.imm_operands == 2
3403 && (i.types[0] & Imm)
3404 && (i.types[1] & Imm));
3405 source = 2;
3406 break;
3407 default:
3408 abort ();
3411 dest = source + 1;
3413 i.rm.mode = 3;
3414 /* One of the register operands will be encoded in the i.tm.reg
3415 field, the other in the combined i.tm.mode and i.tm.regmem
3416 fields. If no form of this instruction supports a memory
3417 destination operand, then we assume the source operand may
3418 sometimes be a memory operand and so we need to store the
3419 destination in the i.rm.reg field. */
3420 if ((i.tm.operand_types[dest] & AnyMem) == 0)
3422 i.rm.reg = i.op[dest].regs->reg_num;
3423 i.rm.regmem = i.op[source].regs->reg_num;
3424 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
3425 i.rex |= REX_EXTX;
3426 if ((i.op[source].regs->reg_flags & RegRex) != 0)
3427 i.rex |= REX_EXTZ;
3429 else
3431 i.rm.reg = i.op[source].regs->reg_num;
3432 i.rm.regmem = i.op[dest].regs->reg_num;
3433 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
3434 i.rex |= REX_EXTZ;
3435 if ((i.op[source].regs->reg_flags & RegRex) != 0)
3436 i.rex |= REX_EXTX;
3438 if (flag_code != CODE_64BIT && (i.rex & (REX_EXTX | REX_EXTZ)))
3440 if (!((i.types[0] | i.types[1]) & Control))
3441 abort ();
3442 i.rex &= ~(REX_EXTX | REX_EXTZ);
3443 add_prefix (LOCK_PREFIX_OPCODE);
3446 else
3447 { /* If it's not 2 reg operands... */
3448 if (i.mem_operands)
3450 unsigned int fake_zero_displacement = 0;
3451 unsigned int op = ((i.types[0] & AnyMem)
3453 : (i.types[1] & AnyMem) ? 1 : 2);
3455 default_seg = &ds;
3457 if (i.base_reg == 0)
3459 i.rm.mode = 0;
3460 if (!i.disp_operands)
3461 fake_zero_displacement = 1;
3462 if (i.index_reg == 0)
3464 /* Operand is just <disp> */
3465 if (flag_code == CODE_64BIT)
3467 /* 64bit mode overwrites the 32bit absolute
3468 addressing by RIP relative addressing and
3469 absolute addressing is encoded by one of the
3470 redundant SIB forms. */
3471 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3472 i.sib.base = NO_BASE_REGISTER;
3473 i.sib.index = NO_INDEX_REGISTER;
3474 i.types[op] = ((i.prefix[ADDR_PREFIX] == 0)
3475 ? Disp32S : Disp32);
3477 else if ((flag_code == CODE_16BIT)
3478 ^ (i.prefix[ADDR_PREFIX] != 0))
3480 i.rm.regmem = NO_BASE_REGISTER_16;
3481 i.types[op] = Disp16;
3483 else
3485 i.rm.regmem = NO_BASE_REGISTER;
3486 i.types[op] = Disp32;
3489 else /* !i.base_reg && i.index_reg */
3491 i.sib.index = i.index_reg->reg_num;
3492 i.sib.base = NO_BASE_REGISTER;
3493 i.sib.scale = i.log2_scale_factor;
3494 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3495 i.types[op] &= ~Disp;
3496 if (flag_code != CODE_64BIT)
3497 i.types[op] |= Disp32; /* Must be 32 bit */
3498 else
3499 i.types[op] |= Disp32S;
3500 if ((i.index_reg->reg_flags & RegRex) != 0)
3501 i.rex |= REX_EXTY;
3504 /* RIP addressing for 64bit mode. */
3505 else if (i.base_reg->reg_type == BaseIndex)
3507 i.rm.regmem = NO_BASE_REGISTER;
3508 i.types[op] &= ~ Disp;
3509 i.types[op] |= Disp32S;
3510 i.flags[op] |= Operand_PCrel;
3511 if (! i.disp_operands)
3512 fake_zero_displacement = 1;
3514 else if (i.base_reg->reg_type & Reg16)
3516 switch (i.base_reg->reg_num)
3518 case 3: /* (%bx) */
3519 if (i.index_reg == 0)
3520 i.rm.regmem = 7;
3521 else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */
3522 i.rm.regmem = i.index_reg->reg_num - 6;
3523 break;
3524 case 5: /* (%bp) */
3525 default_seg = &ss;
3526 if (i.index_reg == 0)
3528 i.rm.regmem = 6;
3529 if ((i.types[op] & Disp) == 0)
3531 /* fake (%bp) into 0(%bp) */
3532 i.types[op] |= Disp8;
3533 fake_zero_displacement = 1;
3536 else /* (%bp,%si) -> 2, or (%bp,%di) -> 3 */
3537 i.rm.regmem = i.index_reg->reg_num - 6 + 2;
3538 break;
3539 default: /* (%si) -> 4 or (%di) -> 5 */
3540 i.rm.regmem = i.base_reg->reg_num - 6 + 4;
3542 i.rm.mode = mode_from_disp_size (i.types[op]);
3544 else /* i.base_reg and 32/64 bit mode */
3546 if (flag_code == CODE_64BIT
3547 && (i.types[op] & Disp))
3548 i.types[op] = ((i.types[op] & Disp8)
3549 | (i.prefix[ADDR_PREFIX] == 0
3550 ? Disp32S : Disp32));
3552 i.rm.regmem = i.base_reg->reg_num;
3553 if ((i.base_reg->reg_flags & RegRex) != 0)
3554 i.rex |= REX_EXTZ;
3555 i.sib.base = i.base_reg->reg_num;
3556 /* x86-64 ignores REX prefix bit here to avoid decoder
3557 complications. */
3558 if ((i.base_reg->reg_num & 7) == EBP_REG_NUM)
3560 default_seg = &ss;
3561 if (i.disp_operands == 0)
3563 fake_zero_displacement = 1;
3564 i.types[op] |= Disp8;
3567 else if (i.base_reg->reg_num == ESP_REG_NUM)
3569 default_seg = &ss;
3571 i.sib.scale = i.log2_scale_factor;
3572 if (i.index_reg == 0)
3574 /* <disp>(%esp) becomes two byte modrm with no index
3575 register. We've already stored the code for esp
3576 in i.rm.regmem ie. ESCAPE_TO_TWO_BYTE_ADDRESSING.
3577 Any base register besides %esp will not use the
3578 extra modrm byte. */
3579 i.sib.index = NO_INDEX_REGISTER;
3580 #if !SCALE1_WHEN_NO_INDEX
3581 /* Another case where we force the second modrm byte. */
3582 if (i.log2_scale_factor)
3583 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3584 #endif
3586 else
3588 i.sib.index = i.index_reg->reg_num;
3589 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3590 if ((i.index_reg->reg_flags & RegRex) != 0)
3591 i.rex |= REX_EXTY;
3594 if (i.disp_operands
3595 && (i.reloc[op] == BFD_RELOC_386_TLS_DESC_CALL
3596 || i.reloc[op] == BFD_RELOC_X86_64_TLSDESC_CALL))
3597 i.rm.mode = 0;
3598 else
3599 i.rm.mode = mode_from_disp_size (i.types[op]);
3602 if (fake_zero_displacement)
3604 /* Fakes a zero displacement assuming that i.types[op]
3605 holds the correct displacement size. */
3606 expressionS *exp;
3608 assert (i.op[op].disps == 0);
3609 exp = &disp_expressions[i.disp_operands++];
3610 i.op[op].disps = exp;
3611 exp->X_op = O_constant;
3612 exp->X_add_number = 0;
3613 exp->X_add_symbol = (symbolS *) 0;
3614 exp->X_op_symbol = (symbolS *) 0;
3618 /* Fill in i.rm.reg or i.rm.regmem field with register operand
3619 (if any) based on i.tm.extension_opcode. Again, we must be
3620 careful to make sure that segment/control/debug/test/MMX
3621 registers are coded into the i.rm.reg field. */
3622 if (i.reg_operands)
3624 unsigned int op =
3625 ((i.types[0]
3626 & (Reg | RegMMX | RegXMM
3627 | SReg2 | SReg3
3628 | Control | Debug | Test))
3630 : ((i.types[1]
3631 & (Reg | RegMMX | RegXMM
3632 | SReg2 | SReg3
3633 | Control | Debug | Test))
3635 : 2));
3636 /* If there is an extension opcode to put here, the register
3637 number must be put into the regmem field. */
3638 if (i.tm.extension_opcode != None)
3640 i.rm.regmem = i.op[op].regs->reg_num;
3641 if ((i.op[op].regs->reg_flags & RegRex) != 0)
3642 i.rex |= REX_EXTZ;
3644 else
3646 i.rm.reg = i.op[op].regs->reg_num;
3647 if ((i.op[op].regs->reg_flags & RegRex) != 0)
3648 i.rex |= REX_EXTX;
3651 /* Now, if no memory operand has set i.rm.mode = 0, 1, 2 we
3652 must set it to 3 to indicate this is a register operand
3653 in the regmem field. */
3654 if (!i.mem_operands)
3655 i.rm.mode = 3;
3658 /* Fill in i.rm.reg field with extension opcode (if any). */
3659 if (i.tm.extension_opcode != None)
3660 i.rm.reg = i.tm.extension_opcode;
3662 return default_seg;
3665 static void
3666 output_branch (void)
3668 char *p;
3669 int code16;
3670 int prefix;
3671 relax_substateT subtype;
3672 symbolS *sym;
3673 offsetT off;
3675 code16 = 0;
3676 if (flag_code == CODE_16BIT)
3677 code16 = CODE16;
3679 prefix = 0;
3680 if (i.prefix[DATA_PREFIX] != 0)
3682 prefix = 1;
3683 i.prefixes -= 1;
3684 code16 ^= CODE16;
3686 /* Pentium4 branch hints. */
3687 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
3688 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
3690 prefix++;
3691 i.prefixes--;
3693 if (i.prefix[REX_PREFIX] != 0)
3695 prefix++;
3696 i.prefixes--;
3699 if (i.prefixes != 0 && !intel_syntax)
3700 as_warn (_("skipping prefixes on this instruction"));
3702 /* It's always a symbol; End frag & setup for relax.
3703 Make sure there is enough room in this frag for the largest
3704 instruction we may generate in md_convert_frag. This is 2
3705 bytes for the opcode and room for the prefix and largest
3706 displacement. */
3707 frag_grow (prefix + 2 + 4);
3708 /* Prefix and 1 opcode byte go in fr_fix. */
3709 p = frag_more (prefix + 1);
3710 if (i.prefix[DATA_PREFIX] != 0)
3711 *p++ = DATA_PREFIX_OPCODE;
3712 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE
3713 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE)
3714 *p++ = i.prefix[SEG_PREFIX];
3715 if (i.prefix[REX_PREFIX] != 0)
3716 *p++ = i.prefix[REX_PREFIX];
3717 *p = i.tm.base_opcode;
3719 if ((unsigned char) *p == JUMP_PC_RELATIVE)
3720 subtype = ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL);
3721 else if ((cpu_arch_flags & Cpu386) != 0)
3722 subtype = ENCODE_RELAX_STATE (COND_JUMP, SMALL);
3723 else
3724 subtype = ENCODE_RELAX_STATE (COND_JUMP86, SMALL);
3725 subtype |= code16;
3727 sym = i.op[0].disps->X_add_symbol;
3728 off = i.op[0].disps->X_add_number;
3730 if (i.op[0].disps->X_op != O_constant
3731 && i.op[0].disps->X_op != O_symbol)
3733 /* Handle complex expressions. */
3734 sym = make_expr_symbol (i.op[0].disps);
3735 off = 0;
3738 /* 1 possible extra opcode + 4 byte displacement go in var part.
3739 Pass reloc in fr_var. */
3740 frag_var (rs_machine_dependent, 5, i.reloc[0], subtype, sym, off, p);
3743 static void
3744 output_jump (void)
3746 char *p;
3747 int size;
3748 fixS *fixP;
3750 if (i.tm.opcode_modifier & JumpByte)
3752 /* This is a loop or jecxz type instruction. */
3753 size = 1;
3754 if (i.prefix[ADDR_PREFIX] != 0)
3756 FRAG_APPEND_1_CHAR (ADDR_PREFIX_OPCODE);
3757 i.prefixes -= 1;
3759 /* Pentium4 branch hints. */
3760 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
3761 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
3763 FRAG_APPEND_1_CHAR (i.prefix[SEG_PREFIX]);
3764 i.prefixes--;
3767 else
3769 int code16;
3771 code16 = 0;
3772 if (flag_code == CODE_16BIT)
3773 code16 = CODE16;
3775 if (i.prefix[DATA_PREFIX] != 0)
3777 FRAG_APPEND_1_CHAR (DATA_PREFIX_OPCODE);
3778 i.prefixes -= 1;
3779 code16 ^= CODE16;
3782 size = 4;
3783 if (code16)
3784 size = 2;
3787 if (i.prefix[REX_PREFIX] != 0)
3789 FRAG_APPEND_1_CHAR (i.prefix[REX_PREFIX]);
3790 i.prefixes -= 1;
3793 if (i.prefixes != 0 && !intel_syntax)
3794 as_warn (_("skipping prefixes on this instruction"));
3796 p = frag_more (1 + size);
3797 *p++ = i.tm.base_opcode;
3799 fixP = fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3800 i.op[0].disps, 1, reloc (size, 1, 1, i.reloc[0]));
3802 /* All jumps handled here are signed, but don't use a signed limit
3803 check for 32 and 16 bit jumps as we want to allow wrap around at
3804 4G and 64k respectively. */
3805 if (size == 1)
3806 fixP->fx_signed = 1;
3809 static void
3810 output_interseg_jump (void)
3812 char *p;
3813 int size;
3814 int prefix;
3815 int code16;
3817 code16 = 0;
3818 if (flag_code == CODE_16BIT)
3819 code16 = CODE16;
3821 prefix = 0;
3822 if (i.prefix[DATA_PREFIX] != 0)
3824 prefix = 1;
3825 i.prefixes -= 1;
3826 code16 ^= CODE16;
3828 if (i.prefix[REX_PREFIX] != 0)
3830 prefix++;
3831 i.prefixes -= 1;
3834 size = 4;
3835 if (code16)
3836 size = 2;
3838 if (i.prefixes != 0 && !intel_syntax)
3839 as_warn (_("skipping prefixes on this instruction"));
3841 /* 1 opcode; 2 segment; offset */
3842 p = frag_more (prefix + 1 + 2 + size);
3844 if (i.prefix[DATA_PREFIX] != 0)
3845 *p++ = DATA_PREFIX_OPCODE;
3847 if (i.prefix[REX_PREFIX] != 0)
3848 *p++ = i.prefix[REX_PREFIX];
3850 *p++ = i.tm.base_opcode;
3851 if (i.op[1].imms->X_op == O_constant)
3853 offsetT n = i.op[1].imms->X_add_number;
3855 if (size == 2
3856 && !fits_in_unsigned_word (n)
3857 && !fits_in_signed_word (n))
3859 as_bad (_("16-bit jump out of range"));
3860 return;
3862 md_number_to_chars (p, n, size);
3864 else
3865 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3866 i.op[1].imms, 0, reloc (size, 0, 0, i.reloc[1]));
3867 if (i.op[0].imms->X_op != O_constant)
3868 as_bad (_("can't handle non absolute segment in `%s'"),
3869 i.tm.name);
3870 md_number_to_chars (p + size, (valueT) i.op[0].imms->X_add_number, 2);
3873 static void
3874 output_insn (void)
3876 fragS *insn_start_frag;
3877 offsetT insn_start_off;
3879 /* Tie dwarf2 debug info to the address at the start of the insn.
3880 We can't do this after the insn has been output as the current
3881 frag may have been closed off. eg. by frag_var. */
3882 dwarf2_emit_insn (0);
3884 insn_start_frag = frag_now;
3885 insn_start_off = frag_now_fix ();
3887 /* Output jumps. */
3888 if (i.tm.opcode_modifier & Jump)
3889 output_branch ();
3890 else if (i.tm.opcode_modifier & (JumpByte | JumpDword))
3891 output_jump ();
3892 else if (i.tm.opcode_modifier & JumpInterSegment)
3893 output_interseg_jump ();
3894 else
3896 /* Output normal instructions here. */
3897 char *p;
3898 unsigned char *q;
3899 unsigned int prefix;
3901 /* All opcodes on i386 have either 1 or 2 bytes. Supplemental
3902 Streaming SIMD extensions 3 Instructions have 3 bytes. We may
3903 use one more higher byte to specify a prefix the instruction
3904 requires. */
3905 if ((i.tm.cpu_flags & CpuSSSE3) != 0)
3907 if (i.tm.base_opcode & 0xff000000)
3909 prefix = (i.tm.base_opcode >> 24) & 0xff;
3910 goto check_prefix;
3913 else if ((i.tm.base_opcode & 0xff0000) != 0)
3915 prefix = (i.tm.base_opcode >> 16) & 0xff;
3916 if ((i.tm.cpu_flags & CpuPadLock) != 0)
3918 check_prefix:
3919 if (prefix != REPE_PREFIX_OPCODE
3920 || i.prefix[LOCKREP_PREFIX] != REPE_PREFIX_OPCODE)
3921 add_prefix (prefix);
3923 else
3924 add_prefix (prefix);
3927 /* The prefix bytes. */
3928 for (q = i.prefix;
3929 q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
3930 q++)
3932 if (*q)
3934 p = frag_more (1);
3935 md_number_to_chars (p, (valueT) *q, 1);
3939 /* Now the opcode; be careful about word order here! */
3940 if (fits_in_unsigned_byte (i.tm.base_opcode))
3942 FRAG_APPEND_1_CHAR (i.tm.base_opcode);
3944 else
3946 if ((i.tm.cpu_flags & CpuSSSE3) != 0)
3948 p = frag_more (3);
3949 *p++ = (i.tm.base_opcode >> 16) & 0xff;
3951 else
3952 p = frag_more (2);
3954 /* Put out high byte first: can't use md_number_to_chars! */
3955 *p++ = (i.tm.base_opcode >> 8) & 0xff;
3956 *p = i.tm.base_opcode & 0xff;
3959 /* Now the modrm byte and sib byte (if present). */
3960 if (i.tm.opcode_modifier & Modrm)
3962 p = frag_more (1);
3963 md_number_to_chars (p,
3964 (valueT) (i.rm.regmem << 0
3965 | i.rm.reg << 3
3966 | i.rm.mode << 6),
3968 /* If i.rm.regmem == ESP (4)
3969 && i.rm.mode != (Register mode)
3970 && not 16 bit
3971 ==> need second modrm byte. */
3972 if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING
3973 && i.rm.mode != 3
3974 && !(i.base_reg && (i.base_reg->reg_type & Reg16) != 0))
3976 p = frag_more (1);
3977 md_number_to_chars (p,
3978 (valueT) (i.sib.base << 0
3979 | i.sib.index << 3
3980 | i.sib.scale << 6),
3985 if (i.disp_operands)
3986 output_disp (insn_start_frag, insn_start_off);
3988 if (i.imm_operands)
3989 output_imm (insn_start_frag, insn_start_off);
3992 #ifdef DEBUG386
3993 if (flag_debug)
3995 pi ("" /*line*/, &i);
3997 #endif /* DEBUG386 */
4000 static void
4001 output_disp (fragS *insn_start_frag, offsetT insn_start_off)
4003 char *p;
4004 unsigned int n;
4006 for (n = 0; n < i.operands; n++)
4008 if (i.types[n] & Disp)
4010 if (i.op[n].disps->X_op == O_constant)
4012 int size;
4013 offsetT val;
4015 size = 4;
4016 if (i.types[n] & (Disp8 | Disp16 | Disp64))
4018 size = 2;
4019 if (i.types[n] & Disp8)
4020 size = 1;
4021 if (i.types[n] & Disp64)
4022 size = 8;
4024 val = offset_in_range (i.op[n].disps->X_add_number,
4025 size);
4026 p = frag_more (size);
4027 md_number_to_chars (p, val, size);
4029 else
4031 enum bfd_reloc_code_real reloc_type;
4032 int size = 4;
4033 int sign = 0;
4034 int pcrel = (i.flags[n] & Operand_PCrel) != 0;
4036 /* The PC relative address is computed relative
4037 to the instruction boundary, so in case immediate
4038 fields follows, we need to adjust the value. */
4039 if (pcrel && i.imm_operands)
4041 int imm_size = 4;
4042 unsigned int n1;
4044 for (n1 = 0; n1 < i.operands; n1++)
4045 if (i.types[n1] & Imm)
4047 if (i.types[n1] & (Imm8 | Imm8S | Imm16 | Imm64))
4049 imm_size = 2;
4050 if (i.types[n1] & (Imm8 | Imm8S))
4051 imm_size = 1;
4052 if (i.types[n1] & Imm64)
4053 imm_size = 8;
4055 break;
4057 /* We should find the immediate. */
4058 if (n1 == i.operands)
4059 abort ();
4060 i.op[n].disps->X_add_number -= imm_size;
4063 if (i.types[n] & Disp32S)
4064 sign = 1;
4066 if (i.types[n] & (Disp16 | Disp64))
4068 size = 2;
4069 if (i.types[n] & Disp64)
4070 size = 8;
4073 p = frag_more (size);
4074 reloc_type = reloc (size, pcrel, sign, i.reloc[n]);
4075 if (GOT_symbol
4076 && GOT_symbol == i.op[n].disps->X_add_symbol
4077 && (((reloc_type == BFD_RELOC_32
4078 || reloc_type == BFD_RELOC_X86_64_32S
4079 || (reloc_type == BFD_RELOC_64
4080 && object_64bit))
4081 && (i.op[n].disps->X_op == O_symbol
4082 || (i.op[n].disps->X_op == O_add
4083 && ((symbol_get_value_expression
4084 (i.op[n].disps->X_op_symbol)->X_op)
4085 == O_subtract))))
4086 || reloc_type == BFD_RELOC_32_PCREL))
4088 offsetT add;
4090 if (insn_start_frag == frag_now)
4091 add = (p - frag_now->fr_literal) - insn_start_off;
4092 else
4094 fragS *fr;
4096 add = insn_start_frag->fr_fix - insn_start_off;
4097 for (fr = insn_start_frag->fr_next;
4098 fr && fr != frag_now; fr = fr->fr_next)
4099 add += fr->fr_fix;
4100 add += p - frag_now->fr_literal;
4103 if (!object_64bit)
4105 reloc_type = BFD_RELOC_386_GOTPC;
4106 i.op[n].imms->X_add_number += add;
4108 else if (reloc_type == BFD_RELOC_64)
4109 reloc_type = BFD_RELOC_X86_64_GOTPC64;
4110 else
4111 /* Don't do the adjustment for x86-64, as there
4112 the pcrel addressing is relative to the _next_
4113 insn, and that is taken care of in other code. */
4114 reloc_type = BFD_RELOC_X86_64_GOTPC32;
4116 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
4117 i.op[n].disps, pcrel, reloc_type);
4123 static void
4124 output_imm (fragS *insn_start_frag, offsetT insn_start_off)
4126 char *p;
4127 unsigned int n;
4129 for (n = 0; n < i.operands; n++)
4131 if (i.types[n] & Imm)
4133 if (i.op[n].imms->X_op == O_constant)
4135 int size;
4136 offsetT val;
4138 size = 4;
4139 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
4141 size = 2;
4142 if (i.types[n] & (Imm8 | Imm8S))
4143 size = 1;
4144 else if (i.types[n] & Imm64)
4145 size = 8;
4147 val = offset_in_range (i.op[n].imms->X_add_number,
4148 size);
4149 p = frag_more (size);
4150 md_number_to_chars (p, val, size);
4152 else
4154 /* Not absolute_section.
4155 Need a 32-bit fixup (don't support 8bit
4156 non-absolute imms). Try to support other
4157 sizes ... */
4158 enum bfd_reloc_code_real reloc_type;
4159 int size = 4;
4160 int sign = 0;
4162 if ((i.types[n] & (Imm32S))
4163 && (i.suffix == QWORD_MNEM_SUFFIX
4164 || (!i.suffix && (i.tm.opcode_modifier & No_lSuf))))
4165 sign = 1;
4166 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
4168 size = 2;
4169 if (i.types[n] & (Imm8 | Imm8S))
4170 size = 1;
4171 if (i.types[n] & Imm64)
4172 size = 8;
4175 p = frag_more (size);
4176 reloc_type = reloc (size, 0, sign, i.reloc[n]);
4178 /* This is tough to explain. We end up with this one if we
4179 * have operands that look like
4180 * "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal here is to
4181 * obtain the absolute address of the GOT, and it is strongly
4182 * preferable from a performance point of view to avoid using
4183 * a runtime relocation for this. The actual sequence of
4184 * instructions often look something like:
4186 * call .L66
4187 * .L66:
4188 * popl %ebx
4189 * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
4191 * The call and pop essentially return the absolute address
4192 * of the label .L66 and store it in %ebx. The linker itself
4193 * will ultimately change the first operand of the addl so
4194 * that %ebx points to the GOT, but to keep things simple, the
4195 * .o file must have this operand set so that it generates not
4196 * the absolute address of .L66, but the absolute address of
4197 * itself. This allows the linker itself simply treat a GOTPC
4198 * relocation as asking for a pcrel offset to the GOT to be
4199 * added in, and the addend of the relocation is stored in the
4200 * operand field for the instruction itself.
4202 * Our job here is to fix the operand so that it would add
4203 * the correct offset so that %ebx would point to itself. The
4204 * thing that is tricky is that .-.L66 will point to the
4205 * beginning of the instruction, so we need to further modify
4206 * the operand so that it will point to itself. There are
4207 * other cases where you have something like:
4209 * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
4211 * and here no correction would be required. Internally in
4212 * the assembler we treat operands of this form as not being
4213 * pcrel since the '.' is explicitly mentioned, and I wonder
4214 * whether it would simplify matters to do it this way. Who
4215 * knows. In earlier versions of the PIC patches, the
4216 * pcrel_adjust field was used to store the correction, but
4217 * since the expression is not pcrel, I felt it would be
4218 * confusing to do it this way. */
4220 if ((reloc_type == BFD_RELOC_32
4221 || reloc_type == BFD_RELOC_X86_64_32S
4222 || reloc_type == BFD_RELOC_64)
4223 && GOT_symbol
4224 && GOT_symbol == i.op[n].imms->X_add_symbol
4225 && (i.op[n].imms->X_op == O_symbol
4226 || (i.op[n].imms->X_op == O_add
4227 && ((symbol_get_value_expression
4228 (i.op[n].imms->X_op_symbol)->X_op)
4229 == O_subtract))))
4231 offsetT add;
4233 if (insn_start_frag == frag_now)
4234 add = (p - frag_now->fr_literal) - insn_start_off;
4235 else
4237 fragS *fr;
4239 add = insn_start_frag->fr_fix - insn_start_off;
4240 for (fr = insn_start_frag->fr_next;
4241 fr && fr != frag_now; fr = fr->fr_next)
4242 add += fr->fr_fix;
4243 add += p - frag_now->fr_literal;
4246 if (!object_64bit)
4247 reloc_type = BFD_RELOC_386_GOTPC;
4248 else if (size == 4)
4249 reloc_type = BFD_RELOC_X86_64_GOTPC32;
4250 else if (size == 8)
4251 reloc_type = BFD_RELOC_X86_64_GOTPC64;
4252 i.op[n].imms->X_add_number += add;
4254 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
4255 i.op[n].imms, 0, reloc_type);
4261 /* x86_cons_fix_new is called via the expression parsing code when a
4262 reloc is needed. We use this hook to get the correct .got reloc. */
4263 static enum bfd_reloc_code_real got_reloc = NO_RELOC;
4264 static int cons_sign = -1;
4266 void
4267 x86_cons_fix_new (fragS *frag, unsigned int off, unsigned int len,
4268 expressionS *exp)
4270 enum bfd_reloc_code_real r = reloc (len, 0, cons_sign, got_reloc);
4272 got_reloc = NO_RELOC;
4274 #ifdef TE_PE
4275 if (exp->X_op == O_secrel)
4277 exp->X_op = O_symbol;
4278 r = BFD_RELOC_32_SECREL;
4280 #endif
4282 fix_new_exp (frag, off, len, exp, 0, r);
4285 #if (!defined (OBJ_ELF) && !defined (OBJ_MAYBE_ELF)) || defined (LEX_AT)
4286 # define lex_got(reloc, adjust, types) NULL
4287 #else
4288 /* Parse operands of the form
4289 <symbol>@GOTOFF+<nnn>
4290 and similar .plt or .got references.
4292 If we find one, set up the correct relocation in RELOC and copy the
4293 input string, minus the `@GOTOFF' into a malloc'd buffer for
4294 parsing by the calling routine. Return this buffer, and if ADJUST
4295 is non-null set it to the length of the string we removed from the
4296 input line. Otherwise return NULL. */
4297 static char *
4298 lex_got (enum bfd_reloc_code_real *reloc,
4299 int *adjust,
4300 unsigned int *types)
4302 /* Some of the relocations depend on the size of what field is to
4303 be relocated. But in our callers i386_immediate and i386_displacement
4304 we don't yet know the operand size (this will be set by insn
4305 matching). Hence we record the word32 relocation here,
4306 and adjust the reloc according to the real size in reloc(). */
4307 static const struct {
4308 const char *str;
4309 const enum bfd_reloc_code_real rel[2];
4310 const unsigned int types64;
4311 } gotrel[] = {
4312 { "PLTOFF", { 0, BFD_RELOC_X86_64_PLTOFF64 }, Imm64 },
4313 { "PLT", { BFD_RELOC_386_PLT32, BFD_RELOC_X86_64_PLT32 }, Imm32|Imm32S|Disp32 },
4314 { "GOTPLT", { 0, BFD_RELOC_X86_64_GOTPLT64 }, Imm64|Disp64 },
4315 { "GOTOFF", { BFD_RELOC_386_GOTOFF, BFD_RELOC_X86_64_GOTOFF64 }, Imm64|Disp64 },
4316 { "GOTPCREL", { 0, BFD_RELOC_X86_64_GOTPCREL }, Imm32|Imm32S|Disp32 },
4317 { "TLSGD", { BFD_RELOC_386_TLS_GD, BFD_RELOC_X86_64_TLSGD }, Imm32|Imm32S|Disp32 },
4318 { "TLSLDM", { BFD_RELOC_386_TLS_LDM, 0 }, 0 },
4319 { "TLSLD", { 0, BFD_RELOC_X86_64_TLSLD }, Imm32|Imm32S|Disp32 },
4320 { "GOTTPOFF", { BFD_RELOC_386_TLS_IE_32, BFD_RELOC_X86_64_GOTTPOFF }, Imm32|Imm32S|Disp32 },
4321 { "TPOFF", { BFD_RELOC_386_TLS_LE_32, BFD_RELOC_X86_64_TPOFF32 }, Imm32|Imm32S|Imm64|Disp32|Disp64 },
4322 { "NTPOFF", { BFD_RELOC_386_TLS_LE, 0 }, 0 },
4323 { "DTPOFF", { BFD_RELOC_386_TLS_LDO_32, BFD_RELOC_X86_64_DTPOFF32 }, Imm32|Imm32S|Imm64|Disp32|Disp64 },
4324 { "GOTNTPOFF",{ BFD_RELOC_386_TLS_GOTIE, 0 }, 0 },
4325 { "INDNTPOFF",{ BFD_RELOC_386_TLS_IE, 0 }, 0 },
4326 { "GOT", { BFD_RELOC_386_GOT32, BFD_RELOC_X86_64_GOT32 }, Imm32|Imm32S|Disp32|Imm64 },
4327 { "TLSDESC", { BFD_RELOC_386_TLS_GOTDESC, BFD_RELOC_X86_64_GOTPC32_TLSDESC }, Imm32|Imm32S|Disp32 },
4328 { "TLSCALL", { BFD_RELOC_386_TLS_DESC_CALL, BFD_RELOC_X86_64_TLSDESC_CALL }, Imm32|Imm32S|Disp32 }
4330 char *cp;
4331 unsigned int j;
4333 if (!IS_ELF)
4334 return NULL;
4336 for (cp = input_line_pointer; *cp != '@'; cp++)
4337 if (is_end_of_line[(unsigned char) *cp])
4338 return NULL;
4340 for (j = 0; j < sizeof (gotrel) / sizeof (gotrel[0]); j++)
4342 int len;
4344 len = strlen (gotrel[j].str);
4345 if (strncasecmp (cp + 1, gotrel[j].str, len) == 0)
4347 if (gotrel[j].rel[object_64bit] != 0)
4349 int first, second;
4350 char *tmpbuf, *past_reloc;
4352 *reloc = gotrel[j].rel[object_64bit];
4353 if (adjust)
4354 *adjust = len;
4356 if (types)
4358 if (flag_code != CODE_64BIT)
4359 *types = Imm32|Disp32;
4360 else
4361 *types = gotrel[j].types64;
4364 if (GOT_symbol == NULL)
4365 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
4367 /* Replace the relocation token with ' ', so that
4368 errors like foo@GOTOFF1 will be detected. */
4370 /* The length of the first part of our input line. */
4371 first = cp - input_line_pointer;
4373 /* The second part goes from after the reloc token until
4374 (and including) an end_of_line char. Don't use strlen
4375 here as the end_of_line char may not be a NUL. */
4376 past_reloc = cp + 1 + len;
4377 for (cp = past_reloc; !is_end_of_line[(unsigned char) *cp++]; )
4379 second = cp - past_reloc;
4381 /* Allocate and copy string. The trailing NUL shouldn't
4382 be necessary, but be safe. */
4383 tmpbuf = xmalloc (first + second + 2);
4384 memcpy (tmpbuf, input_line_pointer, first);
4385 tmpbuf[first] = ' ';
4386 memcpy (tmpbuf + first + 1, past_reloc, second);
4387 tmpbuf[first + second + 1] = '\0';
4388 return tmpbuf;
4391 as_bad (_("@%s reloc is not supported with %d-bit output format"),
4392 gotrel[j].str, 1 << (5 + object_64bit));
4393 return NULL;
4397 /* Might be a symbol version string. Don't as_bad here. */
4398 return NULL;
4401 void
4402 x86_cons (expressionS *exp, int size)
4404 if (size == 4 || (object_64bit && size == 8))
4406 /* Handle @GOTOFF and the like in an expression. */
4407 char *save;
4408 char *gotfree_input_line;
4409 int adjust;
4411 save = input_line_pointer;
4412 gotfree_input_line = lex_got (&got_reloc, &adjust, NULL);
4413 if (gotfree_input_line)
4414 input_line_pointer = gotfree_input_line;
4416 expression (exp);
4418 if (gotfree_input_line)
4420 /* expression () has merrily parsed up to the end of line,
4421 or a comma - in the wrong buffer. Transfer how far
4422 input_line_pointer has moved to the right buffer. */
4423 input_line_pointer = (save
4424 + (input_line_pointer - gotfree_input_line)
4425 + adjust);
4426 free (gotfree_input_line);
4429 else
4430 expression (exp);
4432 #endif
4434 static void signed_cons (int size)
4436 if (flag_code == CODE_64BIT)
4437 cons_sign = 1;
4438 cons (size);
4439 cons_sign = -1;
4442 #ifdef TE_PE
4443 static void
4444 pe_directive_secrel (dummy)
4445 int dummy ATTRIBUTE_UNUSED;
4447 expressionS exp;
4451 expression (&exp);
4452 if (exp.X_op == O_symbol)
4453 exp.X_op = O_secrel;
4455 emit_expr (&exp, 4);
4457 while (*input_line_pointer++ == ',');
4459 input_line_pointer--;
4460 demand_empty_rest_of_line ();
4462 #endif
4464 static int
4465 i386_immediate (char *imm_start)
4467 char *save_input_line_pointer;
4468 char *gotfree_input_line;
4469 segT exp_seg = 0;
4470 expressionS *exp;
4471 unsigned int types = ~0U;
4473 if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
4475 as_bad (_("at most %d immediate operands are allowed"),
4476 MAX_IMMEDIATE_OPERANDS);
4477 return 0;
4480 exp = &im_expressions[i.imm_operands++];
4481 i.op[this_operand].imms = exp;
4483 if (is_space_char (*imm_start))
4484 ++imm_start;
4486 save_input_line_pointer = input_line_pointer;
4487 input_line_pointer = imm_start;
4489 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL, &types);
4490 if (gotfree_input_line)
4491 input_line_pointer = gotfree_input_line;
4493 exp_seg = expression (exp);
4495 SKIP_WHITESPACE ();
4496 if (*input_line_pointer)
4497 as_bad (_("junk `%s' after expression"), input_line_pointer);
4499 input_line_pointer = save_input_line_pointer;
4500 if (gotfree_input_line)
4501 free (gotfree_input_line);
4503 if (exp->X_op == O_absent || exp->X_op == O_big)
4505 /* Missing or bad expr becomes absolute 0. */
4506 as_bad (_("missing or invalid immediate expression `%s' taken as 0"),
4507 imm_start);
4508 exp->X_op = O_constant;
4509 exp->X_add_number = 0;
4510 exp->X_add_symbol = (symbolS *) 0;
4511 exp->X_op_symbol = (symbolS *) 0;
4513 else if (exp->X_op == O_constant)
4515 /* Size it properly later. */
4516 i.types[this_operand] |= Imm64;
4517 /* If BFD64, sign extend val. */
4518 if (!use_rela_relocations)
4519 if ((exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0)
4520 exp->X_add_number = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
4522 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4523 else if (OUTPUT_FLAVOR == bfd_target_aout_flavour
4524 && exp_seg != absolute_section
4525 && exp_seg != text_section
4526 && exp_seg != data_section
4527 && exp_seg != bss_section
4528 && exp_seg != undefined_section
4529 && !bfd_is_com_section (exp_seg))
4531 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
4532 return 0;
4534 #endif
4535 else if (!intel_syntax && exp->X_op == O_register)
4537 as_bad (_("illegal immediate register operand %s"), imm_start);
4538 return 0;
4540 else
4542 /* This is an address. The size of the address will be
4543 determined later, depending on destination register,
4544 suffix, or the default for the section. */
4545 i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64;
4546 i.types[this_operand] &= types;
4549 return 1;
4552 static char *
4553 i386_scale (char *scale)
4555 offsetT val;
4556 char *save = input_line_pointer;
4558 input_line_pointer = scale;
4559 val = get_absolute_expression ();
4561 switch (val)
4563 case 1:
4564 i.log2_scale_factor = 0;
4565 break;
4566 case 2:
4567 i.log2_scale_factor = 1;
4568 break;
4569 case 4:
4570 i.log2_scale_factor = 2;
4571 break;
4572 case 8:
4573 i.log2_scale_factor = 3;
4574 break;
4575 default:
4577 char sep = *input_line_pointer;
4579 *input_line_pointer = '\0';
4580 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
4581 scale);
4582 *input_line_pointer = sep;
4583 input_line_pointer = save;
4584 return NULL;
4587 if (i.log2_scale_factor != 0 && i.index_reg == 0)
4589 as_warn (_("scale factor of %d without an index register"),
4590 1 << i.log2_scale_factor);
4591 #if SCALE1_WHEN_NO_INDEX
4592 i.log2_scale_factor = 0;
4593 #endif
4595 scale = input_line_pointer;
4596 input_line_pointer = save;
4597 return scale;
4600 static int
4601 i386_displacement (char *disp_start, char *disp_end)
4603 expressionS *exp;
4604 segT exp_seg = 0;
4605 char *save_input_line_pointer;
4606 char *gotfree_input_line;
4607 int bigdisp, override;
4608 unsigned int types = Disp;
4610 if (i.disp_operands == MAX_MEMORY_OPERANDS)
4612 as_bad (_("at most %d displacement operands are allowed"),
4613 MAX_MEMORY_OPERANDS);
4614 return 0;
4617 if ((i.types[this_operand] & JumpAbsolute)
4618 || !(current_templates->start->opcode_modifier & (Jump | JumpDword)))
4620 bigdisp = Disp32;
4621 override = (i.prefix[ADDR_PREFIX] != 0);
4623 else
4625 /* For PC-relative branches, the width of the displacement
4626 is dependent upon data size, not address size. */
4627 bigdisp = 0;
4628 override = (i.prefix[DATA_PREFIX] != 0);
4630 if (flag_code == CODE_64BIT)
4632 if (!bigdisp)
4633 bigdisp = ((override || i.suffix == WORD_MNEM_SUFFIX)
4634 ? Disp16
4635 : Disp32S | Disp32);
4636 else if (!override)
4637 bigdisp = Disp64 | Disp32S | Disp32;
4639 else
4641 if (!bigdisp)
4643 if (!override)
4644 override = (i.suffix == (flag_code != CODE_16BIT
4645 ? WORD_MNEM_SUFFIX
4646 : LONG_MNEM_SUFFIX));
4647 bigdisp = Disp32;
4649 if ((flag_code == CODE_16BIT) ^ override)
4650 bigdisp = Disp16;
4652 i.types[this_operand] |= bigdisp;
4654 exp = &disp_expressions[i.disp_operands];
4655 i.op[this_operand].disps = exp;
4656 i.disp_operands++;
4657 save_input_line_pointer = input_line_pointer;
4658 input_line_pointer = disp_start;
4659 END_STRING_AND_SAVE (disp_end);
4661 #ifndef GCC_ASM_O_HACK
4662 #define GCC_ASM_O_HACK 0
4663 #endif
4664 #if GCC_ASM_O_HACK
4665 END_STRING_AND_SAVE (disp_end + 1);
4666 if ((i.types[this_operand] & BaseIndex) != 0
4667 && displacement_string_end[-1] == '+')
4669 /* This hack is to avoid a warning when using the "o"
4670 constraint within gcc asm statements.
4671 For instance:
4673 #define _set_tssldt_desc(n,addr,limit,type) \
4674 __asm__ __volatile__ ( \
4675 "movw %w2,%0\n\t" \
4676 "movw %w1,2+%0\n\t" \
4677 "rorl $16,%1\n\t" \
4678 "movb %b1,4+%0\n\t" \
4679 "movb %4,5+%0\n\t" \
4680 "movb $0,6+%0\n\t" \
4681 "movb %h1,7+%0\n\t" \
4682 "rorl $16,%1" \
4683 : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
4685 This works great except that the output assembler ends
4686 up looking a bit weird if it turns out that there is
4687 no offset. You end up producing code that looks like:
4689 #APP
4690 movw $235,(%eax)
4691 movw %dx,2+(%eax)
4692 rorl $16,%edx
4693 movb %dl,4+(%eax)
4694 movb $137,5+(%eax)
4695 movb $0,6+(%eax)
4696 movb %dh,7+(%eax)
4697 rorl $16,%edx
4698 #NO_APP
4700 So here we provide the missing zero. */
4702 *displacement_string_end = '0';
4704 #endif
4705 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL, &types);
4706 if (gotfree_input_line)
4707 input_line_pointer = gotfree_input_line;
4709 exp_seg = expression (exp);
4711 SKIP_WHITESPACE ();
4712 if (*input_line_pointer)
4713 as_bad (_("junk `%s' after expression"), input_line_pointer);
4714 #if GCC_ASM_O_HACK
4715 RESTORE_END_STRING (disp_end + 1);
4716 #endif
4717 RESTORE_END_STRING (disp_end);
4718 input_line_pointer = save_input_line_pointer;
4719 if (gotfree_input_line)
4720 free (gotfree_input_line);
4722 /* We do this to make sure that the section symbol is in
4723 the symbol table. We will ultimately change the relocation
4724 to be relative to the beginning of the section. */
4725 if (i.reloc[this_operand] == BFD_RELOC_386_GOTOFF
4726 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
4727 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTOFF64)
4729 if (exp->X_op != O_symbol)
4731 as_bad (_("bad expression used with @%s"),
4732 (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
4733 ? "GOTPCREL"
4734 : "GOTOFF"));
4735 return 0;
4738 if (S_IS_LOCAL (exp->X_add_symbol)
4739 && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
4740 section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
4741 exp->X_op = O_subtract;
4742 exp->X_op_symbol = GOT_symbol;
4743 if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
4744 i.reloc[this_operand] = BFD_RELOC_32_PCREL;
4745 else if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTOFF64)
4746 i.reloc[this_operand] = BFD_RELOC_64;
4747 else
4748 i.reloc[this_operand] = BFD_RELOC_32;
4751 if (exp->X_op == O_absent || exp->X_op == O_big)
4753 /* Missing or bad expr becomes absolute 0. */
4754 as_bad (_("missing or invalid displacement expression `%s' taken as 0"),
4755 disp_start);
4756 exp->X_op = O_constant;
4757 exp->X_add_number = 0;
4758 exp->X_add_symbol = (symbolS *) 0;
4759 exp->X_op_symbol = (symbolS *) 0;
4762 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4763 if (exp->X_op != O_constant
4764 && OUTPUT_FLAVOR == bfd_target_aout_flavour
4765 && exp_seg != absolute_section
4766 && exp_seg != text_section
4767 && exp_seg != data_section
4768 && exp_seg != bss_section
4769 && exp_seg != undefined_section
4770 && !bfd_is_com_section (exp_seg))
4772 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
4773 return 0;
4775 #endif
4777 if (!(i.types[this_operand] & ~Disp))
4778 i.types[this_operand] &= types;
4780 return 1;
4783 /* Make sure the memory operand we've been dealt is valid.
4784 Return 1 on success, 0 on a failure. */
4786 static int
4787 i386_index_check (const char *operand_string)
4789 int ok;
4790 #if INFER_ADDR_PREFIX
4791 int fudged = 0;
4793 tryprefix:
4794 #endif
4795 ok = 1;
4796 if ((current_templates->start->cpu_flags & CpuSVME)
4797 && current_templates->end[-1].operand_types[0] == AnyMem)
4799 /* Memory operands of SVME insns are special in that they only allow
4800 rAX as their memory address and ignore any segment override. */
4801 unsigned RegXX;
4803 /* SKINIT is even more restrictive: it always requires EAX. */
4804 if (strcmp (current_templates->start->name, "skinit") == 0)
4805 RegXX = Reg32;
4806 else if (flag_code == CODE_64BIT)
4807 RegXX = i.prefix[ADDR_PREFIX] == 0 ? Reg64 : Reg32;
4808 else
4809 RegXX = ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0)
4810 ? Reg16
4811 : Reg32);
4812 if (!i.base_reg
4813 || !(i.base_reg->reg_type & Acc)
4814 || !(i.base_reg->reg_type & RegXX)
4815 || i.index_reg
4816 || (i.types[0] & Disp))
4817 ok = 0;
4819 else if (flag_code == CODE_64BIT)
4821 unsigned RegXX = (i.prefix[ADDR_PREFIX] == 0 ? Reg64 : Reg32);
4823 if ((i.base_reg
4824 && ((i.base_reg->reg_type & RegXX) == 0)
4825 && (i.base_reg->reg_type != BaseIndex
4826 || i.index_reg))
4827 || (i.index_reg
4828 && ((i.index_reg->reg_type & (RegXX | BaseIndex))
4829 != (RegXX | BaseIndex))))
4830 ok = 0;
4832 else
4834 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
4836 /* 16bit checks. */
4837 if ((i.base_reg
4838 && ((i.base_reg->reg_type & (Reg16 | BaseIndex | RegRex))
4839 != (Reg16 | BaseIndex)))
4840 || (i.index_reg
4841 && (((i.index_reg->reg_type & (Reg16 | BaseIndex))
4842 != (Reg16 | BaseIndex))
4843 || !(i.base_reg
4844 && i.base_reg->reg_num < 6
4845 && i.index_reg->reg_num >= 6
4846 && i.log2_scale_factor == 0))))
4847 ok = 0;
4849 else
4851 /* 32bit checks. */
4852 if ((i.base_reg
4853 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
4854 || (i.index_reg
4855 && ((i.index_reg->reg_type & (Reg32 | BaseIndex | RegRex))
4856 != (Reg32 | BaseIndex))))
4857 ok = 0;
4860 if (!ok)
4862 #if INFER_ADDR_PREFIX
4863 if (i.prefix[ADDR_PREFIX] == 0)
4865 i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
4866 i.prefixes += 1;
4867 /* Change the size of any displacement too. At most one of
4868 Disp16 or Disp32 is set.
4869 FIXME. There doesn't seem to be any real need for separate
4870 Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
4871 Removing them would probably clean up the code quite a lot. */
4872 if (flag_code != CODE_64BIT && (i.types[this_operand] & (Disp16 | Disp32)))
4873 i.types[this_operand] ^= (Disp16 | Disp32);
4874 fudged = 1;
4875 goto tryprefix;
4877 if (fudged)
4878 as_bad (_("`%s' is not a valid base/index expression"),
4879 operand_string);
4880 else
4881 #endif
4882 as_bad (_("`%s' is not a valid %s bit base/index expression"),
4883 operand_string,
4884 flag_code_names[flag_code]);
4886 return ok;
4889 /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
4890 on error. */
4892 static int
4893 i386_operand (char *operand_string)
4895 const reg_entry *r;
4896 char *end_op;
4897 char *op_string = operand_string;
4899 if (is_space_char (*op_string))
4900 ++op_string;
4902 /* We check for an absolute prefix (differentiating,
4903 for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
4904 if (*op_string == ABSOLUTE_PREFIX)
4906 ++op_string;
4907 if (is_space_char (*op_string))
4908 ++op_string;
4909 i.types[this_operand] |= JumpAbsolute;
4912 /* Check if operand is a register. */
4913 if ((r = parse_register (op_string, &end_op)) != NULL)
4915 /* Check for a segment override by searching for ':' after a
4916 segment register. */
4917 op_string = end_op;
4918 if (is_space_char (*op_string))
4919 ++op_string;
4920 if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
4922 switch (r->reg_num)
4924 case 0:
4925 i.seg[i.mem_operands] = &es;
4926 break;
4927 case 1:
4928 i.seg[i.mem_operands] = &cs;
4929 break;
4930 case 2:
4931 i.seg[i.mem_operands] = &ss;
4932 break;
4933 case 3:
4934 i.seg[i.mem_operands] = &ds;
4935 break;
4936 case 4:
4937 i.seg[i.mem_operands] = &fs;
4938 break;
4939 case 5:
4940 i.seg[i.mem_operands] = &gs;
4941 break;
4944 /* Skip the ':' and whitespace. */
4945 ++op_string;
4946 if (is_space_char (*op_string))
4947 ++op_string;
4949 if (!is_digit_char (*op_string)
4950 && !is_identifier_char (*op_string)
4951 && *op_string != '('
4952 && *op_string != ABSOLUTE_PREFIX)
4954 as_bad (_("bad memory operand `%s'"), op_string);
4955 return 0;
4957 /* Handle case of %es:*foo. */
4958 if (*op_string == ABSOLUTE_PREFIX)
4960 ++op_string;
4961 if (is_space_char (*op_string))
4962 ++op_string;
4963 i.types[this_operand] |= JumpAbsolute;
4965 goto do_memory_reference;
4967 if (*op_string)
4969 as_bad (_("junk `%s' after register"), op_string);
4970 return 0;
4972 i.types[this_operand] |= r->reg_type & ~BaseIndex;
4973 i.op[this_operand].regs = r;
4974 i.reg_operands++;
4976 else if (*op_string == REGISTER_PREFIX)
4978 as_bad (_("bad register name `%s'"), op_string);
4979 return 0;
4981 else if (*op_string == IMMEDIATE_PREFIX)
4983 ++op_string;
4984 if (i.types[this_operand] & JumpAbsolute)
4986 as_bad (_("immediate operand illegal with absolute jump"));
4987 return 0;
4989 if (!i386_immediate (op_string))
4990 return 0;
4992 else if (is_digit_char (*op_string)
4993 || is_identifier_char (*op_string)
4994 || *op_string == '(')
4996 /* This is a memory reference of some sort. */
4997 char *base_string;
4999 /* Start and end of displacement string expression (if found). */
5000 char *displacement_string_start;
5001 char *displacement_string_end;
5003 do_memory_reference:
5004 if ((i.mem_operands == 1
5005 && (current_templates->start->opcode_modifier & IsString) == 0)
5006 || i.mem_operands == 2)
5008 as_bad (_("too many memory references for `%s'"),
5009 current_templates->start->name);
5010 return 0;
5013 /* Check for base index form. We detect the base index form by
5014 looking for an ')' at the end of the operand, searching
5015 for the '(' matching it, and finding a REGISTER_PREFIX or ','
5016 after the '('. */
5017 base_string = op_string + strlen (op_string);
5019 --base_string;
5020 if (is_space_char (*base_string))
5021 --base_string;
5023 /* If we only have a displacement, set-up for it to be parsed later. */
5024 displacement_string_start = op_string;
5025 displacement_string_end = base_string + 1;
5027 if (*base_string == ')')
5029 char *temp_string;
5030 unsigned int parens_balanced = 1;
5031 /* We've already checked that the number of left & right ()'s are
5032 equal, so this loop will not be infinite. */
5035 base_string--;
5036 if (*base_string == ')')
5037 parens_balanced++;
5038 if (*base_string == '(')
5039 parens_balanced--;
5041 while (parens_balanced);
5043 temp_string = base_string;
5045 /* Skip past '(' and whitespace. */
5046 ++base_string;
5047 if (is_space_char (*base_string))
5048 ++base_string;
5050 if (*base_string == ','
5051 || ((i.base_reg = parse_register (base_string, &end_op)) != NULL))
5053 displacement_string_end = temp_string;
5055 i.types[this_operand] |= BaseIndex;
5057 if (i.base_reg)
5059 base_string = end_op;
5060 if (is_space_char (*base_string))
5061 ++base_string;
5064 /* There may be an index reg or scale factor here. */
5065 if (*base_string == ',')
5067 ++base_string;
5068 if (is_space_char (*base_string))
5069 ++base_string;
5071 if ((i.index_reg = parse_register (base_string, &end_op)) != NULL)
5073 base_string = end_op;
5074 if (is_space_char (*base_string))
5075 ++base_string;
5076 if (*base_string == ',')
5078 ++base_string;
5079 if (is_space_char (*base_string))
5080 ++base_string;
5082 else if (*base_string != ')')
5084 as_bad (_("expecting `,' or `)' after index register in `%s'"),
5085 operand_string);
5086 return 0;
5089 else if (*base_string == REGISTER_PREFIX)
5091 as_bad (_("bad register name `%s'"), base_string);
5092 return 0;
5095 /* Check for scale factor. */
5096 if (*base_string != ')')
5098 char *end_scale = i386_scale (base_string);
5100 if (!end_scale)
5101 return 0;
5103 base_string = end_scale;
5104 if (is_space_char (*base_string))
5105 ++base_string;
5106 if (*base_string != ')')
5108 as_bad (_("expecting `)' after scale factor in `%s'"),
5109 operand_string);
5110 return 0;
5113 else if (!i.index_reg)
5115 as_bad (_("expecting index register or scale factor after `,'; got '%c'"),
5116 *base_string);
5117 return 0;
5120 else if (*base_string != ')')
5122 as_bad (_("expecting `,' or `)' after base register in `%s'"),
5123 operand_string);
5124 return 0;
5127 else if (*base_string == REGISTER_PREFIX)
5129 as_bad (_("bad register name `%s'"), base_string);
5130 return 0;
5134 /* If there's an expression beginning the operand, parse it,
5135 assuming displacement_string_start and
5136 displacement_string_end are meaningful. */
5137 if (displacement_string_start != displacement_string_end)
5139 if (!i386_displacement (displacement_string_start,
5140 displacement_string_end))
5141 return 0;
5144 /* Special case for (%dx) while doing input/output op. */
5145 if (i.base_reg
5146 && i.base_reg->reg_type == (Reg16 | InOutPortReg)
5147 && i.index_reg == 0
5148 && i.log2_scale_factor == 0
5149 && i.seg[i.mem_operands] == 0
5150 && (i.types[this_operand] & Disp) == 0)
5152 i.types[this_operand] = InOutPortReg;
5153 return 1;
5156 if (i386_index_check (operand_string) == 0)
5157 return 0;
5158 i.mem_operands++;
5160 else
5162 /* It's not a memory operand; argh! */
5163 as_bad (_("invalid char %s beginning operand %d `%s'"),
5164 output_invalid (*op_string),
5165 this_operand + 1,
5166 op_string);
5167 return 0;
5169 return 1; /* Normal return. */
5172 /* md_estimate_size_before_relax()
5174 Called just before relax() for rs_machine_dependent frags. The x86
5175 assembler uses these frags to handle variable size jump
5176 instructions.
5178 Any symbol that is now undefined will not become defined.
5179 Return the correct fr_subtype in the frag.
5180 Return the initial "guess for variable size of frag" to caller.
5181 The guess is actually the growth beyond the fixed part. Whatever
5182 we do to grow the fixed or variable part contributes to our
5183 returned value. */
5186 md_estimate_size_before_relax (fragP, segment)
5187 fragS *fragP;
5188 segT segment;
5190 /* We've already got fragP->fr_subtype right; all we have to do is
5191 check for un-relaxable symbols. On an ELF system, we can't relax
5192 an externally visible symbol, because it may be overridden by a
5193 shared library. */
5194 if (S_GET_SEGMENT (fragP->fr_symbol) != segment
5195 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5196 || (IS_ELF
5197 && (S_IS_EXTERNAL (fragP->fr_symbol)
5198 || S_IS_WEAK (fragP->fr_symbol)))
5199 #endif
5202 /* Symbol is undefined in this segment, or we need to keep a
5203 reloc so that weak symbols can be overridden. */
5204 int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
5205 enum bfd_reloc_code_real reloc_type;
5206 unsigned char *opcode;
5207 int old_fr_fix;
5209 if (fragP->fr_var != NO_RELOC)
5210 reloc_type = fragP->fr_var;
5211 else if (size == 2)
5212 reloc_type = BFD_RELOC_16_PCREL;
5213 else
5214 reloc_type = BFD_RELOC_32_PCREL;
5216 old_fr_fix = fragP->fr_fix;
5217 opcode = (unsigned char *) fragP->fr_opcode;
5219 switch (TYPE_FROM_RELAX_STATE (fragP->fr_subtype))
5221 case UNCOND_JUMP:
5222 /* Make jmp (0xeb) a (d)word displacement jump. */
5223 opcode[0] = 0xe9;
5224 fragP->fr_fix += size;
5225 fix_new (fragP, old_fr_fix, size,
5226 fragP->fr_symbol,
5227 fragP->fr_offset, 1,
5228 reloc_type);
5229 break;
5231 case COND_JUMP86:
5232 if (size == 2
5233 && (!no_cond_jump_promotion || fragP->fr_var != NO_RELOC))
5235 /* Negate the condition, and branch past an
5236 unconditional jump. */
5237 opcode[0] ^= 1;
5238 opcode[1] = 3;
5239 /* Insert an unconditional jump. */
5240 opcode[2] = 0xe9;
5241 /* We added two extra opcode bytes, and have a two byte
5242 offset. */
5243 fragP->fr_fix += 2 + 2;
5244 fix_new (fragP, old_fr_fix + 2, 2,
5245 fragP->fr_symbol,
5246 fragP->fr_offset, 1,
5247 reloc_type);
5248 break;
5250 /* Fall through. */
5252 case COND_JUMP:
5253 if (no_cond_jump_promotion && fragP->fr_var == NO_RELOC)
5255 fixS *fixP;
5257 fragP->fr_fix += 1;
5258 fixP = fix_new (fragP, old_fr_fix, 1,
5259 fragP->fr_symbol,
5260 fragP->fr_offset, 1,
5261 BFD_RELOC_8_PCREL);
5262 fixP->fx_signed = 1;
5263 break;
5266 /* This changes the byte-displacement jump 0x7N
5267 to the (d)word-displacement jump 0x0f,0x8N. */
5268 opcode[1] = opcode[0] + 0x10;
5269 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
5270 /* We've added an opcode byte. */
5271 fragP->fr_fix += 1 + size;
5272 fix_new (fragP, old_fr_fix + 1, size,
5273 fragP->fr_symbol,
5274 fragP->fr_offset, 1,
5275 reloc_type);
5276 break;
5278 default:
5279 BAD_CASE (fragP->fr_subtype);
5280 break;
5282 frag_wane (fragP);
5283 return fragP->fr_fix - old_fr_fix;
5286 /* Guess size depending on current relax state. Initially the relax
5287 state will correspond to a short jump and we return 1, because
5288 the variable part of the frag (the branch offset) is one byte
5289 long. However, we can relax a section more than once and in that
5290 case we must either set fr_subtype back to the unrelaxed state,
5291 or return the value for the appropriate branch. */
5292 return md_relax_table[fragP->fr_subtype].rlx_length;
5295 /* Called after relax() is finished.
5297 In: Address of frag.
5298 fr_type == rs_machine_dependent.
5299 fr_subtype is what the address relaxed to.
5301 Out: Any fixSs and constants are set up.
5302 Caller will turn frag into a ".space 0". */
5304 void
5305 md_convert_frag (abfd, sec, fragP)
5306 bfd *abfd ATTRIBUTE_UNUSED;
5307 segT sec ATTRIBUTE_UNUSED;
5308 fragS *fragP;
5310 unsigned char *opcode;
5311 unsigned char *where_to_put_displacement = NULL;
5312 offsetT target_address;
5313 offsetT opcode_address;
5314 unsigned int extension = 0;
5315 offsetT displacement_from_opcode_start;
5317 opcode = (unsigned char *) fragP->fr_opcode;
5319 /* Address we want to reach in file space. */
5320 target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
5322 /* Address opcode resides at in file space. */
5323 opcode_address = fragP->fr_address + fragP->fr_fix;
5325 /* Displacement from opcode start to fill into instruction. */
5326 displacement_from_opcode_start = target_address - opcode_address;
5328 if ((fragP->fr_subtype & BIG) == 0)
5330 /* Don't have to change opcode. */
5331 extension = 1; /* 1 opcode + 1 displacement */
5332 where_to_put_displacement = &opcode[1];
5334 else
5336 if (no_cond_jump_promotion
5337 && TYPE_FROM_RELAX_STATE (fragP->fr_subtype) != UNCOND_JUMP)
5338 as_warn_where (fragP->fr_file, fragP->fr_line, _("long jump required"));
5340 switch (fragP->fr_subtype)
5342 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
5343 extension = 4; /* 1 opcode + 4 displacement */
5344 opcode[0] = 0xe9;
5345 where_to_put_displacement = &opcode[1];
5346 break;
5348 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
5349 extension = 2; /* 1 opcode + 2 displacement */
5350 opcode[0] = 0xe9;
5351 where_to_put_displacement = &opcode[1];
5352 break;
5354 case ENCODE_RELAX_STATE (COND_JUMP, BIG):
5355 case ENCODE_RELAX_STATE (COND_JUMP86, BIG):
5356 extension = 5; /* 2 opcode + 4 displacement */
5357 opcode[1] = opcode[0] + 0x10;
5358 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
5359 where_to_put_displacement = &opcode[2];
5360 break;
5362 case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
5363 extension = 3; /* 2 opcode + 2 displacement */
5364 opcode[1] = opcode[0] + 0x10;
5365 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
5366 where_to_put_displacement = &opcode[2];
5367 break;
5369 case ENCODE_RELAX_STATE (COND_JUMP86, BIG16):
5370 extension = 4;
5371 opcode[0] ^= 1;
5372 opcode[1] = 3;
5373 opcode[2] = 0xe9;
5374 where_to_put_displacement = &opcode[3];
5375 break;
5377 default:
5378 BAD_CASE (fragP->fr_subtype);
5379 break;
5383 /* If size if less then four we are sure that the operand fits,
5384 but if it's 4, then it could be that the displacement is larger
5385 then -/+ 2GB. */
5386 if (DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype) == 4
5387 && object_64bit
5388 && ((addressT) (displacement_from_opcode_start - extension
5389 + ((addressT) 1 << 31))
5390 > (((addressT) 2 << 31) - 1)))
5392 as_bad_where (fragP->fr_file, fragP->fr_line,
5393 _("jump target out of range"));
5394 /* Make us emit 0. */
5395 displacement_from_opcode_start = extension;
5397 /* Now put displacement after opcode. */
5398 md_number_to_chars ((char *) where_to_put_displacement,
5399 (valueT) (displacement_from_opcode_start - extension),
5400 DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
5401 fragP->fr_fix += extension;
5404 /* Size of byte displacement jmp. */
5405 int md_short_jump_size = 2;
5407 /* Size of dword displacement jmp. */
5408 int md_long_jump_size = 5;
5410 void
5411 md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
5412 char *ptr;
5413 addressT from_addr, to_addr;
5414 fragS *frag ATTRIBUTE_UNUSED;
5415 symbolS *to_symbol ATTRIBUTE_UNUSED;
5417 offsetT offset;
5419 offset = to_addr - (from_addr + 2);
5420 /* Opcode for byte-disp jump. */
5421 md_number_to_chars (ptr, (valueT) 0xeb, 1);
5422 md_number_to_chars (ptr + 1, (valueT) offset, 1);
5425 void
5426 md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
5427 char *ptr;
5428 addressT from_addr, to_addr;
5429 fragS *frag ATTRIBUTE_UNUSED;
5430 symbolS *to_symbol ATTRIBUTE_UNUSED;
5432 offsetT offset;
5434 offset = to_addr - (from_addr + 5);
5435 md_number_to_chars (ptr, (valueT) 0xe9, 1);
5436 md_number_to_chars (ptr + 1, (valueT) offset, 4);
5439 /* Apply a fixup (fixS) to segment data, once it has been determined
5440 by our caller that we have all the info we need to fix it up.
5442 On the 386, immediates, displacements, and data pointers are all in
5443 the same (little-endian) format, so we don't need to care about which
5444 we are handling. */
5446 void
5447 md_apply_fix (fixP, valP, seg)
5448 /* The fix we're to put in. */
5449 fixS *fixP;
5450 /* Pointer to the value of the bits. */
5451 valueT *valP;
5452 /* Segment fix is from. */
5453 segT seg ATTRIBUTE_UNUSED;
5455 char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
5456 valueT value = *valP;
5458 #if !defined (TE_Mach)
5459 if (fixP->fx_pcrel)
5461 switch (fixP->fx_r_type)
5463 default:
5464 break;
5466 case BFD_RELOC_64:
5467 fixP->fx_r_type = BFD_RELOC_64_PCREL;
5468 break;
5469 case BFD_RELOC_32:
5470 case BFD_RELOC_X86_64_32S:
5471 fixP->fx_r_type = BFD_RELOC_32_PCREL;
5472 break;
5473 case BFD_RELOC_16:
5474 fixP->fx_r_type = BFD_RELOC_16_PCREL;
5475 break;
5476 case BFD_RELOC_8:
5477 fixP->fx_r_type = BFD_RELOC_8_PCREL;
5478 break;
5482 if (fixP->fx_addsy != NULL
5483 && (fixP->fx_r_type == BFD_RELOC_32_PCREL
5484 || fixP->fx_r_type == BFD_RELOC_64_PCREL
5485 || fixP->fx_r_type == BFD_RELOC_16_PCREL
5486 || fixP->fx_r_type == BFD_RELOC_8_PCREL)
5487 && !use_rela_relocations)
5489 /* This is a hack. There should be a better way to handle this.
5490 This covers for the fact that bfd_install_relocation will
5491 subtract the current location (for partial_inplace, PC relative
5492 relocations); see more below. */
5493 #ifndef OBJ_AOUT
5494 if (IS_ELF
5495 #ifdef TE_PE
5496 || OUTPUT_FLAVOR == bfd_target_coff_flavour
5497 #endif
5499 value += fixP->fx_where + fixP->fx_frag->fr_address;
5500 #endif
5501 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5502 if (IS_ELF)
5504 segT sym_seg = S_GET_SEGMENT (fixP->fx_addsy);
5506 if ((sym_seg == seg
5507 || (symbol_section_p (fixP->fx_addsy)
5508 && sym_seg != absolute_section))
5509 && !generic_force_reloc (fixP))
5511 /* Yes, we add the values in twice. This is because
5512 bfd_install_relocation subtracts them out again. I think
5513 bfd_install_relocation is broken, but I don't dare change
5514 it. FIXME. */
5515 value += fixP->fx_where + fixP->fx_frag->fr_address;
5518 #endif
5519 #if defined (OBJ_COFF) && defined (TE_PE)
5520 /* For some reason, the PE format does not store a
5521 section address offset for a PC relative symbol. */
5522 if (S_GET_SEGMENT (fixP->fx_addsy) != seg
5523 || S_IS_WEAK (fixP->fx_addsy))
5524 value += md_pcrel_from (fixP);
5525 #endif
5528 /* Fix a few things - the dynamic linker expects certain values here,
5529 and we must not disappoint it. */
5530 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5531 if (IS_ELF && fixP->fx_addsy)
5532 switch (fixP->fx_r_type)
5534 case BFD_RELOC_386_PLT32:
5535 case BFD_RELOC_X86_64_PLT32:
5536 /* Make the jump instruction point to the address of the operand. At
5537 runtime we merely add the offset to the actual PLT entry. */
5538 value = -4;
5539 break;
5541 case BFD_RELOC_386_TLS_GD:
5542 case BFD_RELOC_386_TLS_LDM:
5543 case BFD_RELOC_386_TLS_IE_32:
5544 case BFD_RELOC_386_TLS_IE:
5545 case BFD_RELOC_386_TLS_GOTIE:
5546 case BFD_RELOC_386_TLS_GOTDESC:
5547 case BFD_RELOC_X86_64_TLSGD:
5548 case BFD_RELOC_X86_64_TLSLD:
5549 case BFD_RELOC_X86_64_GOTTPOFF:
5550 case BFD_RELOC_X86_64_GOTPC32_TLSDESC:
5551 value = 0; /* Fully resolved at runtime. No addend. */
5552 /* Fallthrough */
5553 case BFD_RELOC_386_TLS_LE:
5554 case BFD_RELOC_386_TLS_LDO_32:
5555 case BFD_RELOC_386_TLS_LE_32:
5556 case BFD_RELOC_X86_64_DTPOFF32:
5557 case BFD_RELOC_X86_64_DTPOFF64:
5558 case BFD_RELOC_X86_64_TPOFF32:
5559 case BFD_RELOC_X86_64_TPOFF64:
5560 S_SET_THREAD_LOCAL (fixP->fx_addsy);
5561 break;
5563 case BFD_RELOC_386_TLS_DESC_CALL:
5564 case BFD_RELOC_X86_64_TLSDESC_CALL:
5565 value = 0; /* Fully resolved at runtime. No addend. */
5566 S_SET_THREAD_LOCAL (fixP->fx_addsy);
5567 fixP->fx_done = 0;
5568 return;
5570 case BFD_RELOC_386_GOT32:
5571 case BFD_RELOC_X86_64_GOT32:
5572 value = 0; /* Fully resolved at runtime. No addend. */
5573 break;
5575 case BFD_RELOC_VTABLE_INHERIT:
5576 case BFD_RELOC_VTABLE_ENTRY:
5577 fixP->fx_done = 0;
5578 return;
5580 default:
5581 break;
5583 #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
5584 *valP = value;
5585 #endif /* !defined (TE_Mach) */
5587 /* Are we finished with this relocation now? */
5588 if (fixP->fx_addsy == NULL)
5589 fixP->fx_done = 1;
5590 else if (use_rela_relocations)
5592 fixP->fx_no_overflow = 1;
5593 /* Remember value for tc_gen_reloc. */
5594 fixP->fx_addnumber = value;
5595 value = 0;
5598 md_number_to_chars (p, value, fixP->fx_size);
5601 #define MAX_LITTLENUMS 6
5603 /* Turn the string pointed to by litP into a floating point constant
5604 of type TYPE, and emit the appropriate bytes. The number of
5605 LITTLENUMS emitted is stored in *SIZEP. An error message is
5606 returned, or NULL on OK. */
5608 char *
5609 md_atof (type, litP, sizeP)
5610 int type;
5611 char *litP;
5612 int *sizeP;
5614 int prec;
5615 LITTLENUM_TYPE words[MAX_LITTLENUMS];
5616 LITTLENUM_TYPE *wordP;
5617 char *t;
5619 switch (type)
5621 case 'f':
5622 case 'F':
5623 prec = 2;
5624 break;
5626 case 'd':
5627 case 'D':
5628 prec = 4;
5629 break;
5631 case 'x':
5632 case 'X':
5633 prec = 5;
5634 break;
5636 default:
5637 *sizeP = 0;
5638 return _("Bad call to md_atof ()");
5640 t = atof_ieee (input_line_pointer, type, words);
5641 if (t)
5642 input_line_pointer = t;
5644 *sizeP = prec * sizeof (LITTLENUM_TYPE);
5645 /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
5646 the bigendian 386. */
5647 for (wordP = words + prec - 1; prec--;)
5649 md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
5650 litP += sizeof (LITTLENUM_TYPE);
5652 return 0;
5655 static char output_invalid_buf[sizeof (unsigned char) * 2 + 6];
5657 static char *
5658 output_invalid (int c)
5660 if (ISPRINT (c))
5661 snprintf (output_invalid_buf, sizeof (output_invalid_buf),
5662 "'%c'", c);
5663 else
5664 snprintf (output_invalid_buf, sizeof (output_invalid_buf),
5665 "(0x%x)", (unsigned char) c);
5666 return output_invalid_buf;
5669 /* REG_STRING starts *before* REGISTER_PREFIX. */
5671 static const reg_entry *
5672 parse_real_register (char *reg_string, char **end_op)
5674 char *s = reg_string;
5675 char *p;
5676 char reg_name_given[MAX_REG_NAME_SIZE + 1];
5677 const reg_entry *r;
5679 /* Skip possible REGISTER_PREFIX and possible whitespace. */
5680 if (*s == REGISTER_PREFIX)
5681 ++s;
5683 if (is_space_char (*s))
5684 ++s;
5686 p = reg_name_given;
5687 while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
5689 if (p >= reg_name_given + MAX_REG_NAME_SIZE)
5690 return (const reg_entry *) NULL;
5691 s++;
5694 /* For naked regs, make sure that we are not dealing with an identifier.
5695 This prevents confusing an identifier like `eax_var' with register
5696 `eax'. */
5697 if (allow_naked_reg && identifier_chars[(unsigned char) *s])
5698 return (const reg_entry *) NULL;
5700 *end_op = s;
5702 r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
5704 /* Handle floating point regs, allowing spaces in the (i) part. */
5705 if (r == i386_regtab /* %st is first entry of table */)
5707 if (is_space_char (*s))
5708 ++s;
5709 if (*s == '(')
5711 ++s;
5712 if (is_space_char (*s))
5713 ++s;
5714 if (*s >= '0' && *s <= '7')
5716 r = &i386_float_regtab[*s - '0'];
5717 ++s;
5718 if (is_space_char (*s))
5719 ++s;
5720 if (*s == ')')
5722 *end_op = s + 1;
5723 return r;
5726 /* We have "%st(" then garbage. */
5727 return (const reg_entry *) NULL;
5731 if (r != NULL
5732 && ((r->reg_flags & (RegRex64 | RegRex)) | (r->reg_type & Reg64)) != 0
5733 && (r->reg_type != Control || !(cpu_arch_flags & CpuSledgehammer))
5734 && flag_code != CODE_64BIT)
5735 return (const reg_entry *) NULL;
5737 return r;
5740 /* REG_STRING starts *before* REGISTER_PREFIX. */
5742 static const reg_entry *
5743 parse_register (char *reg_string, char **end_op)
5745 const reg_entry *r;
5747 if (*reg_string == REGISTER_PREFIX || allow_naked_reg)
5748 r = parse_real_register (reg_string, end_op);
5749 else
5750 r = NULL;
5751 if (!r)
5753 char *save = input_line_pointer;
5754 char c;
5755 symbolS *symbolP;
5757 input_line_pointer = reg_string;
5758 c = get_symbol_end ();
5759 symbolP = symbol_find (reg_string);
5760 if (symbolP && S_GET_SEGMENT (symbolP) == reg_section)
5762 const expressionS *e = symbol_get_value_expression (symbolP);
5764 know (e->X_op == O_register);
5765 know (e->X_add_number >= 0 && (valueT) e->X_add_number < ARRAY_SIZE (i386_regtab));
5766 r = i386_regtab + e->X_add_number;
5767 *end_op = input_line_pointer;
5769 *input_line_pointer = c;
5770 input_line_pointer = save;
5772 return r;
5776 i386_parse_name (char *name, expressionS *e, char *nextcharP)
5778 const reg_entry *r;
5779 char *end = input_line_pointer;
5781 *end = *nextcharP;
5782 r = parse_register (name, &input_line_pointer);
5783 if (r && end <= input_line_pointer)
5785 *nextcharP = *input_line_pointer;
5786 *input_line_pointer = 0;
5787 e->X_op = O_register;
5788 e->X_add_number = r - i386_regtab;
5789 return 1;
5791 input_line_pointer = end;
5792 *end = 0;
5793 return 0;
5796 void
5797 md_operand (expressionS *e)
5799 if (*input_line_pointer == REGISTER_PREFIX)
5801 char *end;
5802 const reg_entry *r = parse_real_register (input_line_pointer, &end);
5804 if (r)
5806 e->X_op = O_register;
5807 e->X_add_number = r - i386_regtab;
5808 input_line_pointer = end;
5814 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5815 const char *md_shortopts = "kVQ:sqn";
5816 #else
5817 const char *md_shortopts = "qn";
5818 #endif
5820 #define OPTION_32 (OPTION_MD_BASE + 0)
5821 #define OPTION_64 (OPTION_MD_BASE + 1)
5822 #define OPTION_DIVIDE (OPTION_MD_BASE + 2)
5823 #define OPTION_MARCH (OPTION_MD_BASE + 3)
5824 #define OPTION_MTUNE (OPTION_MD_BASE + 4)
5826 struct option md_longopts[] =
5828 {"32", no_argument, NULL, OPTION_32},
5829 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) || defined(TE_PEP)
5830 {"64", no_argument, NULL, OPTION_64},
5831 #endif
5832 {"divide", no_argument, NULL, OPTION_DIVIDE},
5833 {"march", required_argument, NULL, OPTION_MARCH},
5834 {"mtune", required_argument, NULL, OPTION_MTUNE},
5835 {NULL, no_argument, NULL, 0}
5837 size_t md_longopts_size = sizeof (md_longopts);
5840 md_parse_option (int c, char *arg)
5842 unsigned int i;
5844 switch (c)
5846 case 'n':
5847 optimize_align_code = 0;
5848 break;
5850 case 'q':
5851 quiet_warnings = 1;
5852 break;
5854 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5855 /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
5856 should be emitted or not. FIXME: Not implemented. */
5857 case 'Q':
5858 break;
5860 /* -V: SVR4 argument to print version ID. */
5861 case 'V':
5862 print_version_id ();
5863 break;
5865 /* -k: Ignore for FreeBSD compatibility. */
5866 case 'k':
5867 break;
5869 case 's':
5870 /* -s: On i386 Solaris, this tells the native assembler to use
5871 .stab instead of .stab.excl. We always use .stab anyhow. */
5872 break;
5873 #endif
5874 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) || defined(TE_PEP)
5875 case OPTION_64:
5877 const char **list, **l;
5879 list = bfd_target_list ();
5880 for (l = list; *l != NULL; l++)
5881 if (CONST_STRNEQ (*l, "elf64-x86-64")
5882 || strcmp (*l, "coff-x86-64") == 0
5883 || strcmp (*l, "pe-x86-64") == 0
5884 || strcmp (*l, "pei-x86-64") == 0)
5886 default_arch = "x86_64";
5887 break;
5889 if (*l == NULL)
5890 as_fatal (_("No compiled in support for x86_64"));
5891 free (list);
5893 break;
5894 #endif
5896 case OPTION_32:
5897 default_arch = "i386";
5898 break;
5900 case OPTION_DIVIDE:
5901 #ifdef SVR4_COMMENT_CHARS
5903 char *n, *t;
5904 const char *s;
5906 n = (char *) xmalloc (strlen (i386_comment_chars) + 1);
5907 t = n;
5908 for (s = i386_comment_chars; *s != '\0'; s++)
5909 if (*s != '/')
5910 *t++ = *s;
5911 *t = '\0';
5912 i386_comment_chars = n;
5914 #endif
5915 break;
5917 case OPTION_MARCH:
5918 if (*arg == '.')
5919 as_fatal (_("Invalid -march= option: `%s'"), arg);
5920 for (i = 0; i < ARRAY_SIZE (cpu_arch); i++)
5922 if (strcmp (arg, cpu_arch [i].name) == 0)
5924 cpu_arch_isa = cpu_arch[i].type;
5925 cpu_arch_isa_flags = cpu_arch[i].flags;
5926 if (!cpu_arch_tune_set)
5928 cpu_arch_tune = cpu_arch_isa;
5929 cpu_arch_tune_flags = cpu_arch_isa_flags;
5931 break;
5934 if (i >= ARRAY_SIZE (cpu_arch))
5935 as_fatal (_("Invalid -march= option: `%s'"), arg);
5936 break;
5938 case OPTION_MTUNE:
5939 if (*arg == '.')
5940 as_fatal (_("Invalid -mtune= option: `%s'"), arg);
5941 for (i = 0; i < ARRAY_SIZE (cpu_arch); i++)
5943 if (strcmp (arg, cpu_arch [i].name) == 0)
5945 cpu_arch_tune_set = 1;
5946 cpu_arch_tune = cpu_arch [i].type;
5947 cpu_arch_tune_flags = cpu_arch[i].flags;
5948 break;
5951 if (i >= ARRAY_SIZE (cpu_arch))
5952 as_fatal (_("Invalid -mtune= option: `%s'"), arg);
5953 break;
5955 default:
5956 return 0;
5958 return 1;
5961 void
5962 md_show_usage (stream)
5963 FILE *stream;
5965 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5966 fprintf (stream, _("\
5967 -Q ignored\n\
5968 -V print assembler version number\n\
5969 -k ignored\n"));
5970 #endif
5971 fprintf (stream, _("\
5972 -n Do not optimize code alignment\n\
5973 -q quieten some warnings\n"));
5974 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5975 fprintf (stream, _("\
5976 -s ignored\n"));
5977 #endif
5978 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) || defined(TE_PEP)
5979 fprintf (stream, _("\
5980 --32/--64 generate 32bit/64bit code\n"));
5981 #endif
5982 #ifdef SVR4_COMMENT_CHARS
5983 fprintf (stream, _("\
5984 --divide do not treat `/' as a comment character\n"));
5985 #else
5986 fprintf (stream, _("\
5987 --divide ignored\n"));
5988 #endif
5989 fprintf (stream, _("\
5990 -march=CPU/-mtune=CPU generate code/optimize for CPU, where CPU is one of:\n\
5991 i386, i486, pentium, pentiumpro, pentium4, nocona,\n\
5992 core, core2, k6, athlon, k8, generic32, generic64\n"));
5996 #if defined(TE_PEP)
5997 const char *
5998 x86_64_target_format (void)
6000 if (strcmp (default_arch, "x86_64") == 0)
6002 set_code_flag (CODE_64BIT);
6003 return COFF_TARGET_FORMAT;
6005 else if (strcmp (default_arch, "i386") == 0)
6007 set_code_flag (CODE_32BIT);
6008 return "coff-i386";
6011 as_fatal (_("Unknown architecture"));
6012 return NULL;
6014 #endif
6016 #if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
6017 || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
6019 /* Pick the target format to use. */
6021 const char *
6022 i386_target_format (void)
6024 if (!strcmp (default_arch, "x86_64"))
6026 set_code_flag (CODE_64BIT);
6027 if (cpu_arch_isa_flags == 0)
6028 cpu_arch_isa_flags = Cpu186|Cpu286|Cpu386|Cpu486
6029 |Cpu586|Cpu686|CpuP4|CpuMMX|CpuMMX2
6030 |CpuSSE|CpuSSE2;
6031 if (cpu_arch_tune_flags == 0)
6032 cpu_arch_tune_flags = Cpu186|Cpu286|Cpu386|Cpu486
6033 |Cpu586|Cpu686|CpuP4|CpuMMX|CpuMMX2
6034 |CpuSSE|CpuSSE2;
6036 else if (!strcmp (default_arch, "i386"))
6038 set_code_flag (CODE_32BIT);
6039 if (cpu_arch_isa_flags == 0)
6040 cpu_arch_isa_flags = Cpu186|Cpu286|Cpu386;
6041 if (cpu_arch_tune_flags == 0)
6042 cpu_arch_tune_flags = Cpu186|Cpu286|Cpu386;
6044 else
6045 as_fatal (_("Unknown architecture"));
6046 switch (OUTPUT_FLAVOR)
6048 #ifdef OBJ_MAYBE_AOUT
6049 case bfd_target_aout_flavour:
6050 return AOUT_TARGET_FORMAT;
6051 #endif
6052 #ifdef OBJ_MAYBE_COFF
6053 case bfd_target_coff_flavour:
6054 return "coff-i386";
6055 #endif
6056 #if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF)
6057 case bfd_target_elf_flavour:
6059 if (flag_code == CODE_64BIT)
6061 object_64bit = 1;
6062 use_rela_relocations = 1;
6064 return flag_code == CODE_64BIT ? ELF_TARGET_FORMAT64 : ELF_TARGET_FORMAT;
6066 #endif
6067 default:
6068 abort ();
6069 return NULL;
6073 #endif /* OBJ_MAYBE_ more than one */
6075 #if (defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
6076 void
6077 i386_elf_emit_arch_note (void)
6079 if (IS_ELF && cpu_arch_name != NULL)
6081 char *p;
6082 asection *seg = now_seg;
6083 subsegT subseg = now_subseg;
6084 Elf_Internal_Note i_note;
6085 Elf_External_Note e_note;
6086 asection *note_secp;
6087 int len;
6089 /* Create the .note section. */
6090 note_secp = subseg_new (".note", 0);
6091 bfd_set_section_flags (stdoutput,
6092 note_secp,
6093 SEC_HAS_CONTENTS | SEC_READONLY);
6095 /* Process the arch string. */
6096 len = strlen (cpu_arch_name);
6098 i_note.namesz = len + 1;
6099 i_note.descsz = 0;
6100 i_note.type = NT_ARCH;
6101 p = frag_more (sizeof (e_note.namesz));
6102 md_number_to_chars (p, (valueT) i_note.namesz, sizeof (e_note.namesz));
6103 p = frag_more (sizeof (e_note.descsz));
6104 md_number_to_chars (p, (valueT) i_note.descsz, sizeof (e_note.descsz));
6105 p = frag_more (sizeof (e_note.type));
6106 md_number_to_chars (p, (valueT) i_note.type, sizeof (e_note.type));
6107 p = frag_more (len + 1);
6108 strcpy (p, cpu_arch_name);
6110 frag_align (2, 0, 0);
6112 subseg_set (seg, subseg);
6115 #endif
6117 symbolS *
6118 md_undefined_symbol (name)
6119 char *name;
6121 if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
6122 && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
6123 && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
6124 && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
6126 if (!GOT_symbol)
6128 if (symbol_find (name))
6129 as_bad (_("GOT already in symbol table"));
6130 GOT_symbol = symbol_new (name, undefined_section,
6131 (valueT) 0, &zero_address_frag);
6133 return GOT_symbol;
6135 return 0;
6138 /* Round up a section size to the appropriate boundary. */
6140 valueT
6141 md_section_align (segment, size)
6142 segT segment ATTRIBUTE_UNUSED;
6143 valueT size;
6145 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
6146 if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
6148 /* For a.out, force the section size to be aligned. If we don't do
6149 this, BFD will align it for us, but it will not write out the
6150 final bytes of the section. This may be a bug in BFD, but it is
6151 easier to fix it here since that is how the other a.out targets
6152 work. */
6153 int align;
6155 align = bfd_get_section_alignment (stdoutput, segment);
6156 size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
6158 #endif
6160 return size;
6163 /* On the i386, PC-relative offsets are relative to the start of the
6164 next instruction. That is, the address of the offset, plus its
6165 size, since the offset is always the last part of the insn. */
6167 long
6168 md_pcrel_from (fixS *fixP)
6170 return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
6173 #ifndef I386COFF
6175 static void
6176 s_bss (int ignore ATTRIBUTE_UNUSED)
6178 int temp;
6180 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
6181 if (IS_ELF)
6182 obj_elf_section_change_hook ();
6183 #endif
6184 temp = get_absolute_expression ();
6185 subseg_set (bss_section, (subsegT) temp);
6186 demand_empty_rest_of_line ();
6189 #endif
6191 void
6192 i386_validate_fix (fixS *fixp)
6194 if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
6196 if (fixp->fx_r_type == BFD_RELOC_32_PCREL)
6198 if (!object_64bit)
6199 abort ();
6200 fixp->fx_r_type = BFD_RELOC_X86_64_GOTPCREL;
6202 else
6204 if (!object_64bit)
6205 fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
6206 else
6207 fixp->fx_r_type = BFD_RELOC_X86_64_GOTOFF64;
6209 fixp->fx_subsy = 0;
6213 arelent *
6214 tc_gen_reloc (section, fixp)
6215 asection *section ATTRIBUTE_UNUSED;
6216 fixS *fixp;
6218 arelent *rel;
6219 bfd_reloc_code_real_type code;
6221 switch (fixp->fx_r_type)
6223 case BFD_RELOC_X86_64_PLT32:
6224 case BFD_RELOC_X86_64_GOT32:
6225 case BFD_RELOC_X86_64_GOTPCREL:
6226 case BFD_RELOC_386_PLT32:
6227 case BFD_RELOC_386_GOT32:
6228 case BFD_RELOC_386_GOTOFF:
6229 case BFD_RELOC_386_GOTPC:
6230 case BFD_RELOC_386_TLS_GD:
6231 case BFD_RELOC_386_TLS_LDM:
6232 case BFD_RELOC_386_TLS_LDO_32:
6233 case BFD_RELOC_386_TLS_IE_32:
6234 case BFD_RELOC_386_TLS_IE:
6235 case BFD_RELOC_386_TLS_GOTIE:
6236 case BFD_RELOC_386_TLS_LE_32:
6237 case BFD_RELOC_386_TLS_LE:
6238 case BFD_RELOC_386_TLS_GOTDESC:
6239 case BFD_RELOC_386_TLS_DESC_CALL:
6240 case BFD_RELOC_X86_64_TLSGD:
6241 case BFD_RELOC_X86_64_TLSLD:
6242 case BFD_RELOC_X86_64_DTPOFF32:
6243 case BFD_RELOC_X86_64_DTPOFF64:
6244 case BFD_RELOC_X86_64_GOTTPOFF:
6245 case BFD_RELOC_X86_64_TPOFF32:
6246 case BFD_RELOC_X86_64_TPOFF64:
6247 case BFD_RELOC_X86_64_GOTOFF64:
6248 case BFD_RELOC_X86_64_GOTPC32:
6249 case BFD_RELOC_X86_64_GOT64:
6250 case BFD_RELOC_X86_64_GOTPCREL64:
6251 case BFD_RELOC_X86_64_GOTPC64:
6252 case BFD_RELOC_X86_64_GOTPLT64:
6253 case BFD_RELOC_X86_64_PLTOFF64:
6254 case BFD_RELOC_X86_64_GOTPC32_TLSDESC:
6255 case BFD_RELOC_X86_64_TLSDESC_CALL:
6256 case BFD_RELOC_RVA:
6257 case BFD_RELOC_VTABLE_ENTRY:
6258 case BFD_RELOC_VTABLE_INHERIT:
6259 #ifdef TE_PE
6260 case BFD_RELOC_32_SECREL:
6261 #endif
6262 code = fixp->fx_r_type;
6263 break;
6264 case BFD_RELOC_X86_64_32S:
6265 if (!fixp->fx_pcrel)
6267 /* Don't turn BFD_RELOC_X86_64_32S into BFD_RELOC_32. */
6268 code = fixp->fx_r_type;
6269 break;
6271 default:
6272 if (fixp->fx_pcrel)
6274 switch (fixp->fx_size)
6276 default:
6277 as_bad_where (fixp->fx_file, fixp->fx_line,
6278 _("can not do %d byte pc-relative relocation"),
6279 fixp->fx_size);
6280 code = BFD_RELOC_32_PCREL;
6281 break;
6282 case 1: code = BFD_RELOC_8_PCREL; break;
6283 case 2: code = BFD_RELOC_16_PCREL; break;
6284 case 4: code = BFD_RELOC_32_PCREL; break;
6285 #ifdef BFD64
6286 case 8: code = BFD_RELOC_64_PCREL; break;
6287 #endif
6290 else
6292 switch (fixp->fx_size)
6294 default:
6295 as_bad_where (fixp->fx_file, fixp->fx_line,
6296 _("can not do %d byte relocation"),
6297 fixp->fx_size);
6298 code = BFD_RELOC_32;
6299 break;
6300 case 1: code = BFD_RELOC_8; break;
6301 case 2: code = BFD_RELOC_16; break;
6302 case 4: code = BFD_RELOC_32; break;
6303 #ifdef BFD64
6304 case 8: code = BFD_RELOC_64; break;
6305 #endif
6308 break;
6311 if ((code == BFD_RELOC_32
6312 || code == BFD_RELOC_32_PCREL
6313 || code == BFD_RELOC_X86_64_32S)
6314 && GOT_symbol
6315 && fixp->fx_addsy == GOT_symbol)
6317 if (!object_64bit)
6318 code = BFD_RELOC_386_GOTPC;
6319 else
6320 code = BFD_RELOC_X86_64_GOTPC32;
6322 if ((code == BFD_RELOC_64 || code == BFD_RELOC_64_PCREL)
6323 && GOT_symbol
6324 && fixp->fx_addsy == GOT_symbol)
6326 code = BFD_RELOC_X86_64_GOTPC64;
6329 rel = (arelent *) xmalloc (sizeof (arelent));
6330 rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
6331 *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
6333 rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
6335 if (!use_rela_relocations)
6337 /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
6338 vtable entry to be used in the relocation's section offset. */
6339 if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
6340 rel->address = fixp->fx_offset;
6342 rel->addend = 0;
6344 /* Use the rela in 64bit mode. */
6345 else
6347 if (!fixp->fx_pcrel)
6348 rel->addend = fixp->fx_offset;
6349 else
6350 switch (code)
6352 case BFD_RELOC_X86_64_PLT32:
6353 case BFD_RELOC_X86_64_GOT32:
6354 case BFD_RELOC_X86_64_GOTPCREL:
6355 case BFD_RELOC_X86_64_TLSGD:
6356 case BFD_RELOC_X86_64_TLSLD:
6357 case BFD_RELOC_X86_64_GOTTPOFF:
6358 case BFD_RELOC_X86_64_GOTPC32_TLSDESC:
6359 case BFD_RELOC_X86_64_TLSDESC_CALL:
6360 rel->addend = fixp->fx_offset - fixp->fx_size;
6361 break;
6362 default:
6363 rel->addend = (section->vma
6364 - fixp->fx_size
6365 + fixp->fx_addnumber
6366 + md_pcrel_from (fixp));
6367 break;
6371 rel->howto = bfd_reloc_type_lookup (stdoutput, code);
6372 if (rel->howto == NULL)
6374 as_bad_where (fixp->fx_file, fixp->fx_line,
6375 _("cannot represent relocation type %s"),
6376 bfd_get_reloc_code_name (code));
6377 /* Set howto to a garbage value so that we can keep going. */
6378 rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
6379 assert (rel->howto != NULL);
6382 return rel;
6386 /* Parse operands using Intel syntax. This implements a recursive descent
6387 parser based on the BNF grammar published in Appendix B of the MASM 6.1
6388 Programmer's Guide.
6390 FIXME: We do not recognize the full operand grammar defined in the MASM
6391 documentation. In particular, all the structure/union and
6392 high-level macro operands are missing.
6394 Uppercase words are terminals, lower case words are non-terminals.
6395 Objects surrounded by double brackets '[[' ']]' are optional. Vertical
6396 bars '|' denote choices. Most grammar productions are implemented in
6397 functions called 'intel_<production>'.
6399 Initial production is 'expr'.
6401 addOp + | -
6403 alpha [a-zA-Z]
6405 binOp & | AND | \| | OR | ^ | XOR
6407 byteRegister AL | AH | BL | BH | CL | CH | DL | DH
6409 constant digits [[ radixOverride ]]
6411 dataType BYTE | WORD | DWORD | FWORD | QWORD | TBYTE | OWORD | XMMWORD
6413 digits decdigit
6414 | digits decdigit
6415 | digits hexdigit
6417 decdigit [0-9]
6419 e04 e04 addOp e05
6420 | e05
6422 e05 e05 binOp e06
6423 | e06
6425 e06 e06 mulOp e09
6426 | e09
6428 e09 OFFSET e10
6429 | SHORT e10
6430 | + e10
6431 | - e10
6432 | ~ e10
6433 | NOT e10
6434 | e09 PTR e10
6435 | e09 : e10
6436 | e10
6438 e10 e10 [ expr ]
6439 | e11
6441 e11 ( expr )
6442 | [ expr ]
6443 | constant
6444 | dataType
6445 | id
6447 | register
6449 => expr expr cmpOp e04
6450 | e04
6452 gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
6453 | BP | EBP | SP | ESP | DI | EDI | SI | ESI
6455 hexdigit a | b | c | d | e | f
6456 | A | B | C | D | E | F
6458 id alpha
6459 | id alpha
6460 | id decdigit
6462 mulOp * | / | % | MOD | << | SHL | >> | SHR
6464 quote " | '
6466 register specialRegister
6467 | gpRegister
6468 | byteRegister
6470 segmentRegister CS | DS | ES | FS | GS | SS
6472 specialRegister CR0 | CR2 | CR3 | CR4
6473 | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
6474 | TR3 | TR4 | TR5 | TR6 | TR7
6476 We simplify the grammar in obvious places (e.g., register parsing is
6477 done by calling parse_register) and eliminate immediate left recursion
6478 to implement a recursive-descent parser.
6480 expr e04 expr'
6482 expr' cmpOp e04 expr'
6483 | Empty
6485 e04 e05 e04'
6487 e04' addOp e05 e04'
6488 | Empty
6490 e05 e06 e05'
6492 e05' binOp e06 e05'
6493 | Empty
6495 e06 e09 e06'
6497 e06' mulOp e09 e06'
6498 | Empty
6500 e09 OFFSET e10 e09'
6501 | SHORT e10'
6502 | + e10'
6503 | - e10'
6504 | ~ e10'
6505 | NOT e10'
6506 | e10 e09'
6508 e09' PTR e10 e09'
6509 | : e10 e09'
6510 | Empty
6512 e10 e11 e10'
6514 e10' [ expr ] e10'
6515 | Empty
6517 e11 ( expr )
6518 | [ expr ]
6519 | BYTE
6520 | WORD
6521 | DWORD
6522 | FWORD
6523 | QWORD
6524 | TBYTE
6525 | OWORD
6526 | XMMWORD
6529 | register
6530 | id
6531 | constant */
6533 /* Parsing structure for the intel syntax parser. Used to implement the
6534 semantic actions for the operand grammar. */
6535 struct intel_parser_s
6537 char *op_string; /* The string being parsed. */
6538 int got_a_float; /* Whether the operand is a float. */
6539 int op_modifier; /* Operand modifier. */
6540 int is_mem; /* 1 if operand is memory reference. */
6541 int in_offset; /* >=1 if parsing operand of offset. */
6542 int in_bracket; /* >=1 if parsing operand in brackets. */
6543 const reg_entry *reg; /* Last register reference found. */
6544 char *disp; /* Displacement string being built. */
6545 char *next_operand; /* Resume point when splitting operands. */
6548 static struct intel_parser_s intel_parser;
6550 /* Token structure for parsing intel syntax. */
6551 struct intel_token
6553 int code; /* Token code. */
6554 const reg_entry *reg; /* Register entry for register tokens. */
6555 char *str; /* String representation. */
6558 static struct intel_token cur_token, prev_token;
6560 /* Token codes for the intel parser. Since T_SHORT is already used
6561 by COFF, undefine it first to prevent a warning. */
6562 #define T_NIL -1
6563 #define T_CONST 1
6564 #define T_REG 2
6565 #define T_BYTE 3
6566 #define T_WORD 4
6567 #define T_DWORD 5
6568 #define T_FWORD 6
6569 #define T_QWORD 7
6570 #define T_TBYTE 8
6571 #define T_XMMWORD 9
6572 #undef T_SHORT
6573 #define T_SHORT 10
6574 #define T_OFFSET 11
6575 #define T_PTR 12
6576 #define T_ID 13
6577 #define T_SHL 14
6578 #define T_SHR 15
6580 /* Prototypes for intel parser functions. */
6581 static int intel_match_token (int);
6582 static void intel_putback_token (void);
6583 static void intel_get_token (void);
6584 static int intel_expr (void);
6585 static int intel_e04 (void);
6586 static int intel_e05 (void);
6587 static int intel_e06 (void);
6588 static int intel_e09 (void);
6589 static int intel_e10 (void);
6590 static int intel_e11 (void);
6592 static int
6593 i386_intel_operand (char *operand_string, int got_a_float)
6595 int ret;
6596 char *p;
6598 p = intel_parser.op_string = xstrdup (operand_string);
6599 intel_parser.disp = (char *) xmalloc (strlen (operand_string) + 1);
6601 for (;;)
6603 /* Initialize token holders. */
6604 cur_token.code = prev_token.code = T_NIL;
6605 cur_token.reg = prev_token.reg = NULL;
6606 cur_token.str = prev_token.str = NULL;
6608 /* Initialize parser structure. */
6609 intel_parser.got_a_float = got_a_float;
6610 intel_parser.op_modifier = 0;
6611 intel_parser.is_mem = 0;
6612 intel_parser.in_offset = 0;
6613 intel_parser.in_bracket = 0;
6614 intel_parser.reg = NULL;
6615 intel_parser.disp[0] = '\0';
6616 intel_parser.next_operand = NULL;
6618 /* Read the first token and start the parser. */
6619 intel_get_token ();
6620 ret = intel_expr ();
6622 if (!ret)
6623 break;
6625 if (cur_token.code != T_NIL)
6627 as_bad (_("invalid operand for '%s' ('%s' unexpected)"),
6628 current_templates->start->name, cur_token.str);
6629 ret = 0;
6631 /* If we found a memory reference, hand it over to i386_displacement
6632 to fill in the rest of the operand fields. */
6633 else if (intel_parser.is_mem)
6635 if ((i.mem_operands == 1
6636 && (current_templates->start->opcode_modifier & IsString) == 0)
6637 || i.mem_operands == 2)
6639 as_bad (_("too many memory references for '%s'"),
6640 current_templates->start->name);
6641 ret = 0;
6643 else
6645 char *s = intel_parser.disp;
6646 i.mem_operands++;
6648 if (!quiet_warnings && intel_parser.is_mem < 0)
6649 /* See the comments in intel_bracket_expr. */
6650 as_warn (_("Treating `%s' as memory reference"), operand_string);
6652 /* Add the displacement expression. */
6653 if (*s != '\0')
6654 ret = i386_displacement (s, s + strlen (s));
6655 if (ret)
6657 /* Swap base and index in 16-bit memory operands like
6658 [si+bx]. Since i386_index_check is also used in AT&T
6659 mode we have to do that here. */
6660 if (i.base_reg
6661 && i.index_reg
6662 && (i.base_reg->reg_type & Reg16)
6663 && (i.index_reg->reg_type & Reg16)
6664 && i.base_reg->reg_num >= 6
6665 && i.index_reg->reg_num < 6)
6667 const reg_entry *base = i.index_reg;
6669 i.index_reg = i.base_reg;
6670 i.base_reg = base;
6672 ret = i386_index_check (operand_string);
6677 /* Constant and OFFSET expressions are handled by i386_immediate. */
6678 else if ((intel_parser.op_modifier & (1 << T_OFFSET))
6679 || intel_parser.reg == NULL)
6680 ret = i386_immediate (intel_parser.disp);
6682 if (intel_parser.next_operand && this_operand >= MAX_OPERANDS - 1)
6683 ret = 0;
6684 if (!ret || !intel_parser.next_operand)
6685 break;
6686 intel_parser.op_string = intel_parser.next_operand;
6687 this_operand = i.operands++;
6690 free (p);
6691 free (intel_parser.disp);
6693 return ret;
6696 #define NUM_ADDRESS_REGS (!!i.base_reg + !!i.index_reg)
6698 /* expr e04 expr'
6700 expr' cmpOp e04 expr'
6701 | Empty */
6702 static int
6703 intel_expr (void)
6705 /* XXX Implement the comparison operators. */
6706 return intel_e04 ();
6709 /* e04 e05 e04'
6711 e04' addOp e05 e04'
6712 | Empty */
6713 static int
6714 intel_e04 (void)
6716 int nregs = -1;
6718 for (;;)
6720 if (!intel_e05())
6721 return 0;
6723 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
6724 i.base_reg = i386_regtab + REGNAM_AL; /* al is invalid as base */
6726 if (cur_token.code == '+')
6727 nregs = -1;
6728 else if (cur_token.code == '-')
6729 nregs = NUM_ADDRESS_REGS;
6730 else
6731 return 1;
6733 strcat (intel_parser.disp, cur_token.str);
6734 intel_match_token (cur_token.code);
6738 /* e05 e06 e05'
6740 e05' binOp e06 e05'
6741 | Empty */
6742 static int
6743 intel_e05 (void)
6745 int nregs = ~NUM_ADDRESS_REGS;
6747 for (;;)
6749 if (!intel_e06())
6750 return 0;
6752 if (cur_token.code == '&' || cur_token.code == '|' || cur_token.code == '^')
6754 char str[2];
6756 str[0] = cur_token.code;
6757 str[1] = 0;
6758 strcat (intel_parser.disp, str);
6760 else
6761 break;
6763 intel_match_token (cur_token.code);
6765 if (nregs < 0)
6766 nregs = ~nregs;
6768 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
6769 i.base_reg = i386_regtab + REGNAM_AL + 1; /* cl is invalid as base */
6770 return 1;
6773 /* e06 e09 e06'
6775 e06' mulOp e09 e06'
6776 | Empty */
6777 static int
6778 intel_e06 (void)
6780 int nregs = ~NUM_ADDRESS_REGS;
6782 for (;;)
6784 if (!intel_e09())
6785 return 0;
6787 if (cur_token.code == '*' || cur_token.code == '/' || cur_token.code == '%')
6789 char str[2];
6791 str[0] = cur_token.code;
6792 str[1] = 0;
6793 strcat (intel_parser.disp, str);
6795 else if (cur_token.code == T_SHL)
6796 strcat (intel_parser.disp, "<<");
6797 else if (cur_token.code == T_SHR)
6798 strcat (intel_parser.disp, ">>");
6799 else
6800 break;
6802 intel_match_token (cur_token.code);
6804 if (nregs < 0)
6805 nregs = ~nregs;
6807 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
6808 i.base_reg = i386_regtab + REGNAM_AL + 2; /* dl is invalid as base */
6809 return 1;
6812 /* e09 OFFSET e09
6813 | SHORT e09
6814 | + e09
6815 | - e09
6816 | ~ e09
6817 | NOT e09
6818 | e10 e09'
6820 e09' PTR e10 e09'
6821 | : e10 e09'
6822 | Empty */
6823 static int
6824 intel_e09 (void)
6826 int nregs = ~NUM_ADDRESS_REGS;
6827 int in_offset = 0;
6829 for (;;)
6831 /* Don't consume constants here. */
6832 if (cur_token.code == '+' || cur_token.code == '-')
6834 /* Need to look one token ahead - if the next token
6835 is a constant, the current token is its sign. */
6836 int next_code;
6838 intel_match_token (cur_token.code);
6839 next_code = cur_token.code;
6840 intel_putback_token ();
6841 if (next_code == T_CONST)
6842 break;
6845 /* e09 OFFSET e09 */
6846 if (cur_token.code == T_OFFSET)
6848 if (!in_offset++)
6849 ++intel_parser.in_offset;
6852 /* e09 SHORT e09 */
6853 else if (cur_token.code == T_SHORT)
6854 intel_parser.op_modifier |= 1 << T_SHORT;
6856 /* e09 + e09 */
6857 else if (cur_token.code == '+')
6858 strcat (intel_parser.disp, "+");
6860 /* e09 - e09
6861 | ~ e09
6862 | NOT e09 */
6863 else if (cur_token.code == '-' || cur_token.code == '~')
6865 char str[2];
6867 if (nregs < 0)
6868 nregs = ~nregs;
6869 str[0] = cur_token.code;
6870 str[1] = 0;
6871 strcat (intel_parser.disp, str);
6874 /* e09 e10 e09' */
6875 else
6876 break;
6878 intel_match_token (cur_token.code);
6881 for (;;)
6883 if (!intel_e10 ())
6884 return 0;
6886 /* e09' PTR e10 e09' */
6887 if (cur_token.code == T_PTR)
6889 char suffix;
6891 if (prev_token.code == T_BYTE)
6892 suffix = BYTE_MNEM_SUFFIX;
6894 else if (prev_token.code == T_WORD)
6896 if (current_templates->start->name[0] == 'l'
6897 && current_templates->start->name[2] == 's'
6898 && current_templates->start->name[3] == 0)
6899 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
6900 else if (intel_parser.got_a_float == 2) /* "fi..." */
6901 suffix = SHORT_MNEM_SUFFIX;
6902 else
6903 suffix = WORD_MNEM_SUFFIX;
6906 else if (prev_token.code == T_DWORD)
6908 if (current_templates->start->name[0] == 'l'
6909 && current_templates->start->name[2] == 's'
6910 && current_templates->start->name[3] == 0)
6911 suffix = WORD_MNEM_SUFFIX;
6912 else if (flag_code == CODE_16BIT
6913 && (current_templates->start->opcode_modifier
6914 & (Jump | JumpDword)))
6915 suffix = LONG_DOUBLE_MNEM_SUFFIX;
6916 else if (intel_parser.got_a_float == 1) /* "f..." */
6917 suffix = SHORT_MNEM_SUFFIX;
6918 else
6919 suffix = LONG_MNEM_SUFFIX;
6922 else if (prev_token.code == T_FWORD)
6924 if (current_templates->start->name[0] == 'l'
6925 && current_templates->start->name[2] == 's'
6926 && current_templates->start->name[3] == 0)
6927 suffix = LONG_MNEM_SUFFIX;
6928 else if (!intel_parser.got_a_float)
6930 if (flag_code == CODE_16BIT)
6931 add_prefix (DATA_PREFIX_OPCODE);
6932 suffix = LONG_DOUBLE_MNEM_SUFFIX;
6934 else
6935 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
6938 else if (prev_token.code == T_QWORD)
6940 if (intel_parser.got_a_float == 1) /* "f..." */
6941 suffix = LONG_MNEM_SUFFIX;
6942 else
6943 suffix = QWORD_MNEM_SUFFIX;
6946 else if (prev_token.code == T_TBYTE)
6948 if (intel_parser.got_a_float == 1)
6949 suffix = LONG_DOUBLE_MNEM_SUFFIX;
6950 else
6951 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
6954 else if (prev_token.code == T_XMMWORD)
6956 /* XXX ignored for now, but accepted since gcc uses it */
6957 suffix = 0;
6960 else
6962 as_bad (_("Unknown operand modifier `%s'"), prev_token.str);
6963 return 0;
6966 /* Operands for jump/call using 'ptr' notation denote absolute
6967 addresses. */
6968 if (current_templates->start->opcode_modifier & (Jump | JumpDword))
6969 i.types[this_operand] |= JumpAbsolute;
6971 if (current_templates->start->base_opcode == 0x8d /* lea */)
6973 else if (!i.suffix)
6974 i.suffix = suffix;
6975 else if (i.suffix != suffix)
6977 as_bad (_("Conflicting operand modifiers"));
6978 return 0;
6983 /* e09' : e10 e09' */
6984 else if (cur_token.code == ':')
6986 if (prev_token.code != T_REG)
6988 /* While {call,jmp} SSSS:OOOO is MASM syntax only when SSSS is a
6989 segment/group identifier (which we don't have), using comma
6990 as the operand separator there is even less consistent, since
6991 there all branches only have a single operand. */
6992 if (this_operand != 0
6993 || intel_parser.in_offset
6994 || intel_parser.in_bracket
6995 || (!(current_templates->start->opcode_modifier
6996 & (Jump|JumpDword|JumpInterSegment))
6997 && !(current_templates->start->operand_types[0]
6998 & JumpAbsolute)))
6999 return intel_match_token (T_NIL);
7000 /* Remember the start of the 2nd operand and terminate 1st
7001 operand here.
7002 XXX This isn't right, yet (when SSSS:OOOO is right operand of
7003 another expression), but it gets at least the simplest case
7004 (a plain number or symbol on the left side) right. */
7005 intel_parser.next_operand = intel_parser.op_string;
7006 *--intel_parser.op_string = '\0';
7007 return intel_match_token (':');
7011 /* e09' Empty */
7012 else
7013 break;
7015 intel_match_token (cur_token.code);
7019 if (in_offset)
7021 --intel_parser.in_offset;
7022 if (nregs < 0)
7023 nregs = ~nregs;
7024 if (NUM_ADDRESS_REGS > nregs)
7026 as_bad (_("Invalid operand to `OFFSET'"));
7027 return 0;
7029 intel_parser.op_modifier |= 1 << T_OFFSET;
7032 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
7033 i.base_reg = i386_regtab + REGNAM_AL + 3; /* bl is invalid as base */
7034 return 1;
7037 static int
7038 intel_bracket_expr (void)
7040 int was_offset = intel_parser.op_modifier & (1 << T_OFFSET);
7041 const char *start = intel_parser.op_string;
7042 int len;
7044 if (i.op[this_operand].regs)
7045 return intel_match_token (T_NIL);
7047 intel_match_token ('[');
7049 /* Mark as a memory operand only if it's not already known to be an
7050 offset expression. If it's an offset expression, we need to keep
7051 the brace in. */
7052 if (!intel_parser.in_offset)
7054 ++intel_parser.in_bracket;
7056 /* Operands for jump/call inside brackets denote absolute addresses. */
7057 if (current_templates->start->opcode_modifier & (Jump | JumpDword))
7058 i.types[this_operand] |= JumpAbsolute;
7060 /* Unfortunately gas always diverged from MASM in a respect that can't
7061 be easily fixed without risking to break code sequences likely to be
7062 encountered (the testsuite even check for this): MASM doesn't consider
7063 an expression inside brackets unconditionally as a memory reference.
7064 When that is e.g. a constant, an offset expression, or the sum of the
7065 two, this is still taken as a constant load. gas, however, always
7066 treated these as memory references. As a compromise, we'll try to make
7067 offset expressions inside brackets work the MASM way (since that's
7068 less likely to be found in real world code), but make constants alone
7069 continue to work the traditional gas way. In either case, issue a
7070 warning. */
7071 intel_parser.op_modifier &= ~was_offset;
7073 else
7074 strcat (intel_parser.disp, "[");
7076 /* Add a '+' to the displacement string if necessary. */
7077 if (*intel_parser.disp != '\0'
7078 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
7079 strcat (intel_parser.disp, "+");
7081 if (intel_expr ()
7082 && (len = intel_parser.op_string - start - 1,
7083 intel_match_token (']')))
7085 /* Preserve brackets when the operand is an offset expression. */
7086 if (intel_parser.in_offset)
7087 strcat (intel_parser.disp, "]");
7088 else
7090 --intel_parser.in_bracket;
7091 if (i.base_reg || i.index_reg)
7092 intel_parser.is_mem = 1;
7093 if (!intel_parser.is_mem)
7095 if (!(intel_parser.op_modifier & (1 << T_OFFSET)))
7096 /* Defer the warning until all of the operand was parsed. */
7097 intel_parser.is_mem = -1;
7098 else if (!quiet_warnings)
7099 as_warn (_("`[%.*s]' taken to mean just `%.*s'"), len, start, len, start);
7102 intel_parser.op_modifier |= was_offset;
7104 return 1;
7106 return 0;
7109 /* e10 e11 e10'
7111 e10' [ expr ] e10'
7112 | Empty */
7113 static int
7114 intel_e10 (void)
7116 if (!intel_e11 ())
7117 return 0;
7119 while (cur_token.code == '[')
7121 if (!intel_bracket_expr ())
7122 return 0;
7125 return 1;
7128 /* e11 ( expr )
7129 | [ expr ]
7130 | BYTE
7131 | WORD
7132 | DWORD
7133 | FWORD
7134 | QWORD
7135 | TBYTE
7136 | OWORD
7137 | XMMWORD
7140 | register
7141 | id
7142 | constant */
7143 static int
7144 intel_e11 (void)
7146 switch (cur_token.code)
7148 /* e11 ( expr ) */
7149 case '(':
7150 intel_match_token ('(');
7151 strcat (intel_parser.disp, "(");
7153 if (intel_expr () && intel_match_token (')'))
7155 strcat (intel_parser.disp, ")");
7156 return 1;
7158 return 0;
7160 /* e11 [ expr ] */
7161 case '[':
7162 return intel_bracket_expr ();
7164 /* e11 $
7165 | . */
7166 case '.':
7167 strcat (intel_parser.disp, cur_token.str);
7168 intel_match_token (cur_token.code);
7170 /* Mark as a memory operand only if it's not already known to be an
7171 offset expression. */
7172 if (!intel_parser.in_offset)
7173 intel_parser.is_mem = 1;
7175 return 1;
7177 /* e11 register */
7178 case T_REG:
7180 const reg_entry *reg = intel_parser.reg = cur_token.reg;
7182 intel_match_token (T_REG);
7184 /* Check for segment change. */
7185 if (cur_token.code == ':')
7187 if (!(reg->reg_type & (SReg2 | SReg3)))
7189 as_bad (_("`%s' is not a valid segment register"), reg->reg_name);
7190 return 0;
7192 else if (i.seg[i.mem_operands])
7193 as_warn (_("Extra segment override ignored"));
7194 else
7196 if (!intel_parser.in_offset)
7197 intel_parser.is_mem = 1;
7198 switch (reg->reg_num)
7200 case 0:
7201 i.seg[i.mem_operands] = &es;
7202 break;
7203 case 1:
7204 i.seg[i.mem_operands] = &cs;
7205 break;
7206 case 2:
7207 i.seg[i.mem_operands] = &ss;
7208 break;
7209 case 3:
7210 i.seg[i.mem_operands] = &ds;
7211 break;
7212 case 4:
7213 i.seg[i.mem_operands] = &fs;
7214 break;
7215 case 5:
7216 i.seg[i.mem_operands] = &gs;
7217 break;
7222 /* Not a segment register. Check for register scaling. */
7223 else if (cur_token.code == '*')
7225 if (!intel_parser.in_bracket)
7227 as_bad (_("Register scaling only allowed in memory operands"));
7228 return 0;
7231 if (reg->reg_type & Reg16) /* Disallow things like [si*1]. */
7232 reg = i386_regtab + REGNAM_AX + 4; /* sp is invalid as index */
7233 else if (i.index_reg)
7234 reg = i386_regtab + REGNAM_EAX + 4; /* esp is invalid as index */
7236 /* What follows must be a valid scale. */
7237 intel_match_token ('*');
7238 i.index_reg = reg;
7239 i.types[this_operand] |= BaseIndex;
7241 /* Set the scale after setting the register (otherwise,
7242 i386_scale will complain) */
7243 if (cur_token.code == '+' || cur_token.code == '-')
7245 char *str, sign = cur_token.code;
7246 intel_match_token (cur_token.code);
7247 if (cur_token.code != T_CONST)
7249 as_bad (_("Syntax error: Expecting a constant, got `%s'"),
7250 cur_token.str);
7251 return 0;
7253 str = (char *) xmalloc (strlen (cur_token.str) + 2);
7254 strcpy (str + 1, cur_token.str);
7255 *str = sign;
7256 if (!i386_scale (str))
7257 return 0;
7258 free (str);
7260 else if (!i386_scale (cur_token.str))
7261 return 0;
7262 intel_match_token (cur_token.code);
7265 /* No scaling. If this is a memory operand, the register is either a
7266 base register (first occurrence) or an index register (second
7267 occurrence). */
7268 else if (intel_parser.in_bracket)
7271 if (!i.base_reg)
7272 i.base_reg = reg;
7273 else if (!i.index_reg)
7274 i.index_reg = reg;
7275 else
7277 as_bad (_("Too many register references in memory operand"));
7278 return 0;
7281 i.types[this_operand] |= BaseIndex;
7284 /* It's neither base nor index. */
7285 else if (!intel_parser.in_offset && !intel_parser.is_mem)
7287 i.types[this_operand] |= reg->reg_type & ~BaseIndex;
7288 i.op[this_operand].regs = reg;
7289 i.reg_operands++;
7291 else
7293 as_bad (_("Invalid use of register"));
7294 return 0;
7297 /* Since registers are not part of the displacement string (except
7298 when we're parsing offset operands), we may need to remove any
7299 preceding '+' from the displacement string. */
7300 if (*intel_parser.disp != '\0'
7301 && !intel_parser.in_offset)
7303 char *s = intel_parser.disp;
7304 s += strlen (s) - 1;
7305 if (*s == '+')
7306 *s = '\0';
7309 return 1;
7312 /* e11 BYTE
7313 | WORD
7314 | DWORD
7315 | FWORD
7316 | QWORD
7317 | TBYTE
7318 | OWORD
7319 | XMMWORD */
7320 case T_BYTE:
7321 case T_WORD:
7322 case T_DWORD:
7323 case T_FWORD:
7324 case T_QWORD:
7325 case T_TBYTE:
7326 case T_XMMWORD:
7327 intel_match_token (cur_token.code);
7329 if (cur_token.code == T_PTR)
7330 return 1;
7332 /* It must have been an identifier. */
7333 intel_putback_token ();
7334 cur_token.code = T_ID;
7335 /* FALLTHRU */
7337 /* e11 id
7338 | constant */
7339 case T_ID:
7340 if (!intel_parser.in_offset && intel_parser.is_mem <= 0)
7342 symbolS *symbolP;
7344 /* The identifier represents a memory reference only if it's not
7345 preceded by an offset modifier and if it's not an equate. */
7346 symbolP = symbol_find(cur_token.str);
7347 if (!symbolP || S_GET_SEGMENT(symbolP) != absolute_section)
7348 intel_parser.is_mem = 1;
7350 /* FALLTHRU */
7352 case T_CONST:
7353 case '-':
7354 case '+':
7356 char *save_str, sign = 0;
7358 /* Allow constants that start with `+' or `-'. */
7359 if (cur_token.code == '-' || cur_token.code == '+')
7361 sign = cur_token.code;
7362 intel_match_token (cur_token.code);
7363 if (cur_token.code != T_CONST)
7365 as_bad (_("Syntax error: Expecting a constant, got `%s'"),
7366 cur_token.str);
7367 return 0;
7371 save_str = (char *) xmalloc (strlen (cur_token.str) + 2);
7372 strcpy (save_str + !!sign, cur_token.str);
7373 if (sign)
7374 *save_str = sign;
7376 /* Get the next token to check for register scaling. */
7377 intel_match_token (cur_token.code);
7379 /* Check if this constant is a scaling factor for an index register. */
7380 if (cur_token.code == '*')
7382 if (intel_match_token ('*') && cur_token.code == T_REG)
7384 const reg_entry *reg = cur_token.reg;
7386 if (!intel_parser.in_bracket)
7388 as_bad (_("Register scaling only allowed in memory operands"));
7389 return 0;
7392 if (reg->reg_type & Reg16) /* Disallow things like [1*si]. */
7393 reg = i386_regtab + REGNAM_AX + 4; /* sp is invalid as index */
7394 else if (i.index_reg)
7395 reg = i386_regtab + REGNAM_EAX + 4; /* esp is invalid as index */
7397 /* The constant is followed by `* reg', so it must be
7398 a valid scale. */
7399 i.index_reg = reg;
7400 i.types[this_operand] |= BaseIndex;
7402 /* Set the scale after setting the register (otherwise,
7403 i386_scale will complain) */
7404 if (!i386_scale (save_str))
7405 return 0;
7406 intel_match_token (T_REG);
7408 /* Since registers are not part of the displacement
7409 string, we may need to remove any preceding '+' from
7410 the displacement string. */
7411 if (*intel_parser.disp != '\0')
7413 char *s = intel_parser.disp;
7414 s += strlen (s) - 1;
7415 if (*s == '+')
7416 *s = '\0';
7419 free (save_str);
7421 return 1;
7424 /* The constant was not used for register scaling. Since we have
7425 already consumed the token following `*' we now need to put it
7426 back in the stream. */
7427 intel_putback_token ();
7430 /* Add the constant to the displacement string. */
7431 strcat (intel_parser.disp, save_str);
7432 free (save_str);
7434 return 1;
7438 as_bad (_("Unrecognized token '%s'"), cur_token.str);
7439 return 0;
7442 /* Match the given token against cur_token. If they match, read the next
7443 token from the operand string. */
7444 static int
7445 intel_match_token (int code)
7447 if (cur_token.code == code)
7449 intel_get_token ();
7450 return 1;
7452 else
7454 as_bad (_("Unexpected token `%s'"), cur_token.str);
7455 return 0;
7459 /* Read a new token from intel_parser.op_string and store it in cur_token. */
7460 static void
7461 intel_get_token (void)
7463 char *end_op;
7464 const reg_entry *reg;
7465 struct intel_token new_token;
7467 new_token.code = T_NIL;
7468 new_token.reg = NULL;
7469 new_token.str = NULL;
7471 /* Free the memory allocated to the previous token and move
7472 cur_token to prev_token. */
7473 if (prev_token.str)
7474 free (prev_token.str);
7476 prev_token = cur_token;
7478 /* Skip whitespace. */
7479 while (is_space_char (*intel_parser.op_string))
7480 intel_parser.op_string++;
7482 /* Return an empty token if we find nothing else on the line. */
7483 if (*intel_parser.op_string == '\0')
7485 cur_token = new_token;
7486 return;
7489 /* The new token cannot be larger than the remainder of the operand
7490 string. */
7491 new_token.str = (char *) xmalloc (strlen (intel_parser.op_string) + 1);
7492 new_token.str[0] = '\0';
7494 if (strchr ("0123456789", *intel_parser.op_string))
7496 char *p = new_token.str;
7497 char *q = intel_parser.op_string;
7498 new_token.code = T_CONST;
7500 /* Allow any kind of identifier char to encompass floating point and
7501 hexadecimal numbers. */
7502 while (is_identifier_char (*q))
7503 *p++ = *q++;
7504 *p = '\0';
7506 /* Recognize special symbol names [0-9][bf]. */
7507 if (strlen (intel_parser.op_string) == 2
7508 && (intel_parser.op_string[1] == 'b'
7509 || intel_parser.op_string[1] == 'f'))
7510 new_token.code = T_ID;
7513 else if ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL)
7515 size_t len = end_op - intel_parser.op_string;
7517 new_token.code = T_REG;
7518 new_token.reg = reg;
7520 memcpy (new_token.str, intel_parser.op_string, len);
7521 new_token.str[len] = '\0';
7524 else if (is_identifier_char (*intel_parser.op_string))
7526 char *p = new_token.str;
7527 char *q = intel_parser.op_string;
7529 /* A '.' or '$' followed by an identifier char is an identifier.
7530 Otherwise, it's operator '.' followed by an expression. */
7531 if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
7533 new_token.code = '.';
7534 new_token.str[0] = '.';
7535 new_token.str[1] = '\0';
7537 else
7539 while (is_identifier_char (*q) || *q == '@')
7540 *p++ = *q++;
7541 *p = '\0';
7543 if (strcasecmp (new_token.str, "NOT") == 0)
7544 new_token.code = '~';
7546 else if (strcasecmp (new_token.str, "MOD") == 0)
7547 new_token.code = '%';
7549 else if (strcasecmp (new_token.str, "AND") == 0)
7550 new_token.code = '&';
7552 else if (strcasecmp (new_token.str, "OR") == 0)
7553 new_token.code = '|';
7555 else if (strcasecmp (new_token.str, "XOR") == 0)
7556 new_token.code = '^';
7558 else if (strcasecmp (new_token.str, "SHL") == 0)
7559 new_token.code = T_SHL;
7561 else if (strcasecmp (new_token.str, "SHR") == 0)
7562 new_token.code = T_SHR;
7564 else if (strcasecmp (new_token.str, "BYTE") == 0)
7565 new_token.code = T_BYTE;
7567 else if (strcasecmp (new_token.str, "WORD") == 0)
7568 new_token.code = T_WORD;
7570 else if (strcasecmp (new_token.str, "DWORD") == 0)
7571 new_token.code = T_DWORD;
7573 else if (strcasecmp (new_token.str, "FWORD") == 0)
7574 new_token.code = T_FWORD;
7576 else if (strcasecmp (new_token.str, "QWORD") == 0)
7577 new_token.code = T_QWORD;
7579 else if (strcasecmp (new_token.str, "TBYTE") == 0
7580 /* XXX remove (gcc still uses it) */
7581 || strcasecmp (new_token.str, "XWORD") == 0)
7582 new_token.code = T_TBYTE;
7584 else if (strcasecmp (new_token.str, "XMMWORD") == 0
7585 || strcasecmp (new_token.str, "OWORD") == 0)
7586 new_token.code = T_XMMWORD;
7588 else if (strcasecmp (new_token.str, "PTR") == 0)
7589 new_token.code = T_PTR;
7591 else if (strcasecmp (new_token.str, "SHORT") == 0)
7592 new_token.code = T_SHORT;
7594 else if (strcasecmp (new_token.str, "OFFSET") == 0)
7596 new_token.code = T_OFFSET;
7598 /* ??? This is not mentioned in the MASM grammar but gcc
7599 makes use of it with -mintel-syntax. OFFSET may be
7600 followed by FLAT: */
7601 if (strncasecmp (q, " FLAT:", 6) == 0)
7602 strcat (new_token.str, " FLAT:");
7605 /* ??? This is not mentioned in the MASM grammar. */
7606 else if (strcasecmp (new_token.str, "FLAT") == 0)
7608 new_token.code = T_OFFSET;
7609 if (*q == ':')
7610 strcat (new_token.str, ":");
7611 else
7612 as_bad (_("`:' expected"));
7615 else
7616 new_token.code = T_ID;
7620 else if (strchr ("+-/*%|&^:[]()~", *intel_parser.op_string))
7622 new_token.code = *intel_parser.op_string;
7623 new_token.str[0] = *intel_parser.op_string;
7624 new_token.str[1] = '\0';
7627 else if (strchr ("<>", *intel_parser.op_string)
7628 && *intel_parser.op_string == *(intel_parser.op_string + 1))
7630 new_token.code = *intel_parser.op_string == '<' ? T_SHL : T_SHR;
7631 new_token.str[0] = *intel_parser.op_string;
7632 new_token.str[1] = *intel_parser.op_string;
7633 new_token.str[2] = '\0';
7636 else
7637 as_bad (_("Unrecognized token `%s'"), intel_parser.op_string);
7639 intel_parser.op_string += strlen (new_token.str);
7640 cur_token = new_token;
7643 /* Put cur_token back into the token stream and make cur_token point to
7644 prev_token. */
7645 static void
7646 intel_putback_token (void)
7648 if (cur_token.code != T_NIL)
7650 intel_parser.op_string -= strlen (cur_token.str);
7651 free (cur_token.str);
7653 cur_token = prev_token;
7655 /* Forget prev_token. */
7656 prev_token.code = T_NIL;
7657 prev_token.reg = NULL;
7658 prev_token.str = NULL;
7662 tc_x86_regname_to_dw2regnum (char *regname)
7664 unsigned int regnum;
7665 unsigned int regnames_count;
7666 static const char *const regnames_32[] =
7668 "eax", "ecx", "edx", "ebx",
7669 "esp", "ebp", "esi", "edi",
7670 "eip", "eflags", NULL,
7671 "st0", "st1", "st2", "st3",
7672 "st4", "st5", "st6", "st7",
7673 NULL, NULL,
7674 "xmm0", "xmm1", "xmm2", "xmm3",
7675 "xmm4", "xmm5", "xmm6", "xmm7",
7676 "mm0", "mm1", "mm2", "mm3",
7677 "mm4", "mm5", "mm6", "mm7",
7678 "fcw", "fsw", "mxcsr",
7679 "es", "cs", "ss", "ds", "fs", "gs", NULL, NULL,
7680 "tr", "ldtr"
7682 static const char *const regnames_64[] =
7684 "rax", "rdx", "rcx", "rbx",
7685 "rsi", "rdi", "rbp", "rsp",
7686 "r8", "r9", "r10", "r11",
7687 "r12", "r13", "r14", "r15",
7688 "rip",
7689 "xmm0", "xmm1", "xmm2", "xmm3",
7690 "xmm4", "xmm5", "xmm6", "xmm7",
7691 "xmm8", "xmm9", "xmm10", "xmm11",
7692 "xmm12", "xmm13", "xmm14", "xmm15",
7693 "st0", "st1", "st2", "st3",
7694 "st4", "st5", "st6", "st7",
7695 "mm0", "mm1", "mm2", "mm3",
7696 "mm4", "mm5", "mm6", "mm7",
7697 "rflags",
7698 "es", "cs", "ss", "ds", "fs", "gs", NULL, NULL,
7699 "fs.base", "gs.base", NULL, NULL,
7700 "tr", "ldtr",
7701 "mxcsr", "fcw", "fsw"
7703 const char *const *regnames;
7705 if (flag_code == CODE_64BIT)
7707 regnames = regnames_64;
7708 regnames_count = ARRAY_SIZE (regnames_64);
7710 else
7712 regnames = regnames_32;
7713 regnames_count = ARRAY_SIZE (regnames_32);
7716 for (regnum = 0; regnum < regnames_count; regnum++)
7717 if (regnames[regnum] != NULL
7718 && strcmp (regname, regnames[regnum]) == 0)
7719 return regnum;
7721 return -1;
7724 void
7725 tc_x86_frame_initial_instructions (void)
7727 static unsigned int sp_regno;
7729 if (!sp_regno)
7730 sp_regno = tc_x86_regname_to_dw2regnum (flag_code == CODE_64BIT
7731 ? "rsp" : "esp");
7733 cfi_add_CFA_def_cfa (sp_regno, -x86_cie_data_alignment);
7734 cfi_add_CFA_offset (x86_dwarf2_return_column, x86_cie_data_alignment);
7738 i386_elf_section_type (const char *str, size_t len)
7740 if (flag_code == CODE_64BIT
7741 && len == sizeof ("unwind") - 1
7742 && strncmp (str, "unwind", 6) == 0)
7743 return SHT_X86_64_UNWIND;
7745 return -1;
7748 #ifdef TE_PE
7749 void
7750 tc_pe_dwarf2_emit_offset (symbolS *symbol, unsigned int size)
7752 expressionS expr;
7754 expr.X_op = O_secrel;
7755 expr.X_add_symbol = symbol;
7756 expr.X_add_number = 0;
7757 emit_expr (&expr, size);
7759 #endif
7761 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
7762 /* For ELF on x86-64, add support for SHF_X86_64_LARGE. */
7765 x86_64_section_letter (int letter, char **ptr_msg)
7767 if (flag_code == CODE_64BIT)
7769 if (letter == 'l')
7770 return SHF_X86_64_LARGE;
7772 *ptr_msg = _("Bad .section directive: want a,l,w,x,M,S,G,T in string");
7774 else
7775 *ptr_msg = _("Bad .section directive: want a,w,x,M,S,G,T in string");
7776 return -1;
7780 x86_64_section_word (char *str, size_t len)
7782 if (len == 5 && flag_code == CODE_64BIT && CONST_STRNEQ (str, "large"))
7783 return SHF_X86_64_LARGE;
7785 return -1;
7788 static void
7789 handle_large_common (int small ATTRIBUTE_UNUSED)
7791 if (flag_code != CODE_64BIT)
7793 s_comm_internal (0, elf_common_parse);
7794 as_warn (_(".largecomm supported only in 64bit mode, producing .comm"));
7796 else
7798 static segT lbss_section;
7799 asection *saved_com_section_ptr = elf_com_section_ptr;
7800 asection *saved_bss_section = bss_section;
7802 if (lbss_section == NULL)
7804 flagword applicable;
7805 segT seg = now_seg;
7806 subsegT subseg = now_subseg;
7808 /* The .lbss section is for local .largecomm symbols. */
7809 lbss_section = subseg_new (".lbss", 0);
7810 applicable = bfd_applicable_section_flags (stdoutput);
7811 bfd_set_section_flags (stdoutput, lbss_section,
7812 applicable & SEC_ALLOC);
7813 seg_info (lbss_section)->bss = 1;
7815 subseg_set (seg, subseg);
7818 elf_com_section_ptr = &_bfd_elf_large_com_section;
7819 bss_section = lbss_section;
7821 s_comm_internal (0, elf_common_parse);
7823 elf_com_section_ptr = saved_com_section_ptr;
7824 bss_section = saved_bss_section;
7827 #endif /* OBJ_ELF || OBJ_MAYBE_ELF */