* config/tc-i386.c: Formatting.
[deliverable/binutils-gdb.git] / gas / config / tc-i386.c
1 /* 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
4 Free Software Foundation, Inc.
5
6 This file is part of GAS, the GNU Assembler.
7
8 GAS is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 2, or (at your option)
11 any later version.
12
13 GAS is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
17
18 You should have received a copy of the GNU General Public License
19 along with GAS; see the file COPYING. If not, write to the Free
20 Software Foundation, 51 Franklin Street - Fifth Floor, Boston, MA
21 02110-1301, USA. */
22
23 /* Intel 80386 machine specific gas.
24 Written by Eliot Dresselhaus (eliot@mgm.mit.edu).
25 x86_64 support by Jan Hubicka (jh@suse.cz)
26 VIA PadLock support by Michal Ludvig (mludvig@suse.cz)
27 Bugs & suggestions are completely welcome. This is free software.
28 Please help us make it better. */
29
30 #include "as.h"
31 #include "safe-ctype.h"
32 #include "subsegs.h"
33 #include "dwarf2dbg.h"
34 #include "dw2gencfi.h"
35 #include "opcode/i386.h"
36 #include "elf/x86-64.h"
37
38 #ifndef REGISTER_WARNINGS
39 #define REGISTER_WARNINGS 1
40 #endif
41
42 #ifndef INFER_ADDR_PREFIX
43 #define INFER_ADDR_PREFIX 1
44 #endif
45
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
53
54 #ifndef DEFAULT_ARCH
55 #define DEFAULT_ARCH "i386"
56 #endif
57
58 #ifndef INLINE
59 #if __GNUC__ >= 2
60 #define INLINE __inline__
61 #else
62 #define INLINE
63 #endif
64 #endif
65
66 static INLINE unsigned int mode_from_disp_size PARAMS ((unsigned int));
67 static INLINE int fits_in_signed_byte PARAMS ((offsetT));
68 static INLINE int fits_in_unsigned_byte PARAMS ((offsetT));
69 static INLINE int fits_in_unsigned_word PARAMS ((offsetT));
70 static INLINE int fits_in_signed_word PARAMS ((offsetT));
71 static INLINE int fits_in_unsigned_long PARAMS ((offsetT));
72 static INLINE int fits_in_signed_long PARAMS ((offsetT));
73 static int smallest_imm_type PARAMS ((offsetT));
74 static offsetT offset_in_range PARAMS ((offsetT, int));
75 static int add_prefix PARAMS ((unsigned int));
76 static void set_code_flag PARAMS ((int));
77 static void set_16bit_gcc_code_flag PARAMS ((int));
78 static void set_intel_syntax PARAMS ((int));
79 static void set_cpu_arch PARAMS ((int));
80 #ifdef TE_PE
81 static void pe_directive_secrel PARAMS ((int));
82 #endif
83 static void signed_cons PARAMS ((int));
84 static char *output_invalid PARAMS ((int c));
85 static int i386_operand PARAMS ((char *operand_string));
86 static int i386_intel_operand PARAMS ((char *operand_string, int got_a_float));
87 static const reg_entry *parse_register PARAMS ((char *reg_string,
88 char **end_op));
89 static char *parse_insn PARAMS ((char *, char *));
90 static char *parse_operands PARAMS ((char *, const char *));
91 static void swap_operands PARAMS ((void));
92 static void optimize_imm PARAMS ((void));
93 static void optimize_disp PARAMS ((void));
94 static int match_template PARAMS ((void));
95 static int check_string PARAMS ((void));
96 static int process_suffix PARAMS ((void));
97 static int check_byte_reg PARAMS ((void));
98 static int check_long_reg PARAMS ((void));
99 static int check_qword_reg PARAMS ((void));
100 static int check_word_reg PARAMS ((void));
101 static int finalize_imm PARAMS ((void));
102 static int process_operands PARAMS ((void));
103 static const seg_entry *build_modrm_byte PARAMS ((void));
104 static void output_insn PARAMS ((void));
105 static void output_branch PARAMS ((void));
106 static void output_jump PARAMS ((void));
107 static void output_interseg_jump PARAMS ((void));
108 static void output_imm PARAMS ((fragS *insn_start_frag,
109 offsetT insn_start_off));
110 static void output_disp PARAMS ((fragS *insn_start_frag,
111 offsetT insn_start_off));
112 #ifndef I386COFF
113 static void s_bss PARAMS ((int));
114 #endif
115 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
116 static void handle_large_common (int small ATTRIBUTE_UNUSED);
117 #endif
118
119 static const char *default_arch = DEFAULT_ARCH;
120
121 /* 'md_assemble ()' gathers together information and puts it into a
122 i386_insn. */
123
124 union i386_op
125 {
126 expressionS *disps;
127 expressionS *imms;
128 const reg_entry *regs;
129 };
130
131 struct _i386_insn
132 {
133 /* TM holds the template for the insn were currently assembling. */
134 template tm;
135
136 /* SUFFIX holds the instruction mnemonic suffix if given.
137 (e.g. 'l' for 'movl') */
138 char suffix;
139
140 /* OPERANDS gives the number of given operands. */
141 unsigned int operands;
142
143 /* REG_OPERANDS, DISP_OPERANDS, MEM_OPERANDS, IMM_OPERANDS give the number
144 of given register, displacement, memory operands and immediate
145 operands. */
146 unsigned int reg_operands, disp_operands, mem_operands, imm_operands;
147
148 /* TYPES [i] is the type (see above #defines) which tells us how to
149 use OP[i] for the corresponding operand. */
150 unsigned int types[MAX_OPERANDS];
151
152 /* Displacement expression, immediate expression, or register for each
153 operand. */
154 union i386_op op[MAX_OPERANDS];
155
156 /* Flags for operands. */
157 unsigned int flags[MAX_OPERANDS];
158 #define Operand_PCrel 1
159
160 /* Relocation type for operand */
161 enum bfd_reloc_code_real reloc[MAX_OPERANDS];
162
163 /* BASE_REG, INDEX_REG, and LOG2_SCALE_FACTOR are used to encode
164 the base index byte below. */
165 const reg_entry *base_reg;
166 const reg_entry *index_reg;
167 unsigned int log2_scale_factor;
168
169 /* SEG gives the seg_entries of this insn. They are zero unless
170 explicit segment overrides are given. */
171 const seg_entry *seg[2];
172
173 /* PREFIX holds all the given prefix opcodes (usually null).
174 PREFIXES is the number of prefix opcodes. */
175 unsigned int prefixes;
176 unsigned char prefix[MAX_PREFIXES];
177
178 /* RM and SIB are the modrm byte and the sib byte where the
179 addressing modes of this insn are encoded. */
180
181 modrm_byte rm;
182 rex_byte rex;
183 sib_byte sib;
184 };
185
186 typedef struct _i386_insn i386_insn;
187
188 /* List of chars besides those in app.c:symbol_chars that can start an
189 operand. Used to prevent the scrubber eating vital white-space. */
190 const char extra_symbol_chars[] = "*%-(["
191 #ifdef LEX_AT
192 "@"
193 #endif
194 #ifdef LEX_QM
195 "?"
196 #endif
197 ;
198
199 #if (defined (TE_I386AIX) \
200 || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) \
201 && !defined (TE_GNU) \
202 && !defined (TE_LINUX) \
203 && !defined (TE_NETWARE) \
204 && !defined (TE_FreeBSD) \
205 && !defined (TE_NetBSD)))
206 /* This array holds the chars that always start a comment. If the
207 pre-processor is disabled, these aren't very useful. The option
208 --divide will remove '/' from this list. */
209 const char *i386_comment_chars = "#/";
210 #define SVR4_COMMENT_CHARS 1
211 #define PREFIX_SEPARATOR '\\'
212
213 #else
214 const char *i386_comment_chars = "#";
215 #define PREFIX_SEPARATOR '/'
216 #endif
217
218 /* This array holds the chars that only start a comment at the beginning of
219 a line. If the line seems to have the form '# 123 filename'
220 .line and .file directives will appear in the pre-processed output.
221 Note that input_file.c hand checks for '#' at the beginning of the
222 first line of the input file. This is because the compiler outputs
223 #NO_APP at the beginning of its output.
224 Also note that comments started like this one will always work if
225 '/' isn't otherwise defined. */
226 const char line_comment_chars[] = "#/";
227
228 const char line_separator_chars[] = ";";
229
230 /* Chars that can be used to separate mant from exp in floating point
231 nums. */
232 const char EXP_CHARS[] = "eE";
233
234 /* Chars that mean this number is a floating point constant
235 As in 0f12.456
236 or 0d1.2345e12. */
237 const char FLT_CHARS[] = "fFdDxX";
238
239 /* Tables for lexical analysis. */
240 static char mnemonic_chars[256];
241 static char register_chars[256];
242 static char operand_chars[256];
243 static char identifier_chars[256];
244 static char digit_chars[256];
245
246 /* Lexical macros. */
247 #define is_mnemonic_char(x) (mnemonic_chars[(unsigned char) x])
248 #define is_operand_char(x) (operand_chars[(unsigned char) x])
249 #define is_register_char(x) (register_chars[(unsigned char) x])
250 #define is_space_char(x) ((x) == ' ')
251 #define is_identifier_char(x) (identifier_chars[(unsigned char) x])
252 #define is_digit_char(x) (digit_chars[(unsigned char) x])
253
254 /* All non-digit non-letter characters that may occur in an operand. */
255 static char operand_special_chars[] = "%$-+(,)*._~/<>|&^!:[@]";
256
257 /* md_assemble() always leaves the strings it's passed unaltered. To
258 effect this we maintain a stack of saved characters that we've smashed
259 with '\0's (indicating end of strings for various sub-fields of the
260 assembler instruction). */
261 static char save_stack[32];
262 static char *save_stack_p;
263 #define END_STRING_AND_SAVE(s) \
264 do { *save_stack_p++ = *(s); *(s) = '\0'; } while (0)
265 #define RESTORE_END_STRING(s) \
266 do { *(s) = *--save_stack_p; } while (0)
267
268 /* The instruction we're assembling. */
269 static i386_insn i;
270
271 /* Possible templates for current insn. */
272 static const templates *current_templates;
273
274 /* Per instruction expressionS buffers: 2 displacements & 2 immediate max. */
275 static expressionS disp_expressions[2], im_expressions[2];
276
277 /* Current operand we are working on. */
278 static int this_operand;
279
280 /* We support four different modes. FLAG_CODE variable is used to distinguish
281 these. */
282
283 enum flag_code {
284 CODE_32BIT,
285 CODE_16BIT,
286 CODE_64BIT };
287 #define NUM_FLAG_CODE ((int) CODE_64BIT + 1)
288
289 static enum flag_code flag_code;
290 static unsigned int object_64bit;
291 static int use_rela_relocations = 0;
292
293 /* The names used to print error messages. */
294 static const char *flag_code_names[] =
295 {
296 "32",
297 "16",
298 "64"
299 };
300
301 /* 1 for intel syntax,
302 0 if att syntax. */
303 static int intel_syntax = 0;
304
305 /* 1 if register prefix % not required. */
306 static int allow_naked_reg = 0;
307
308 /* Used in 16 bit gcc mode to add an l suffix to call, ret, enter,
309 leave, push, and pop instructions so that gcc has the same stack
310 frame as in 32 bit mode. */
311 static char stackop_size = '\0';
312
313 /* Non-zero to optimize code alignment. */
314 int optimize_align_code = 1;
315
316 /* Non-zero to quieten some warnings. */
317 static int quiet_warnings = 0;
318
319 /* CPU name. */
320 static const char *cpu_arch_name = NULL;
321 static const char *cpu_sub_arch_name = NULL;
322
323 /* CPU feature flags. */
324 static unsigned int cpu_arch_flags = CpuUnknownFlags | CpuNo64;
325
326 /* If set, conditional jumps are not automatically promoted to handle
327 larger than a byte offset. */
328 static unsigned int no_cond_jump_promotion = 0;
329
330 /* Pre-defined "_GLOBAL_OFFSET_TABLE_". */
331 static symbolS *GOT_symbol;
332
333 /* The dwarf2 return column, adjusted for 32 or 64 bit. */
334 unsigned int x86_dwarf2_return_column;
335
336 /* The dwarf2 data alignment, adjusted for 32 or 64 bit. */
337 int x86_cie_data_alignment;
338
339 /* Interface to relax_segment.
340 There are 3 major relax states for 386 jump insns because the
341 different types of jumps add different sizes to frags when we're
342 figuring out what sort of jump to choose to reach a given label. */
343
344 /* Types. */
345 #define UNCOND_JUMP 0
346 #define COND_JUMP 1
347 #define COND_JUMP86 2
348
349 /* Sizes. */
350 #define CODE16 1
351 #define SMALL 0
352 #define SMALL16 (SMALL | CODE16)
353 #define BIG 2
354 #define BIG16 (BIG | CODE16)
355
356 #ifndef INLINE
357 #ifdef __GNUC__
358 #define INLINE __inline__
359 #else
360 #define INLINE
361 #endif
362 #endif
363
364 #define ENCODE_RELAX_STATE(type, size) \
365 ((relax_substateT) (((type) << 2) | (size)))
366 #define TYPE_FROM_RELAX_STATE(s) \
367 ((s) >> 2)
368 #define DISP_SIZE_FROM_RELAX_STATE(s) \
369 ((((s) & 3) == BIG ? 4 : (((s) & 3) == BIG16 ? 2 : 1)))
370
371 /* This table is used by relax_frag to promote short jumps to long
372 ones where necessary. SMALL (short) jumps may be promoted to BIG
373 (32 bit long) ones, and SMALL16 jumps to BIG16 (16 bit long). We
374 don't allow a short jump in a 32 bit code segment to be promoted to
375 a 16 bit offset jump because it's slower (requires data size
376 prefix), and doesn't work, unless the destination is in the bottom
377 64k of the code segment (The top 16 bits of eip are zeroed). */
378
379 const relax_typeS md_relax_table[] =
380 {
381 /* The fields are:
382 1) most positive reach of this state,
383 2) most negative reach of this state,
384 3) how many bytes this mode will have in the variable part of the frag
385 4) which index into the table to try if we can't fit into this one. */
386
387 /* UNCOND_JUMP states. */
388 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG)},
389 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16)},
390 /* dword jmp adds 4 bytes to frag:
391 0 extra opcode bytes, 4 displacement bytes. */
392 {0, 0, 4, 0},
393 /* word jmp adds 2 byte2 to frag:
394 0 extra opcode bytes, 2 displacement bytes. */
395 {0, 0, 2, 0},
396
397 /* COND_JUMP states. */
398 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG)},
399 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG16)},
400 /* dword conditionals adds 5 bytes to frag:
401 1 extra opcode byte, 4 displacement bytes. */
402 {0, 0, 5, 0},
403 /* word conditionals add 3 bytes to frag:
404 1 extra opcode byte, 2 displacement bytes. */
405 {0, 0, 3, 0},
406
407 /* COND_JUMP86 states. */
408 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG)},
409 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG16)},
410 /* dword conditionals adds 5 bytes to frag:
411 1 extra opcode byte, 4 displacement bytes. */
412 {0, 0, 5, 0},
413 /* word conditionals add 4 bytes to frag:
414 1 displacement byte and a 3 byte long branch insn. */
415 {0, 0, 4, 0}
416 };
417
418 static const arch_entry cpu_arch[] = {
419 {"i8086", Cpu086 },
420 {"i186", Cpu086|Cpu186 },
421 {"i286", Cpu086|Cpu186|Cpu286 },
422 {"i386", Cpu086|Cpu186|Cpu286|Cpu386 },
423 {"i486", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486 },
424 {"i586", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586 },
425 {"i686", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686 },
426 {"pentium", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586 },
427 {"pentiumpro",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686 },
428 {"pentiumii", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX },
429 {"pentiumiii",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuMMX2|CpuSSE },
430 {"pentium4", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX|CpuMMX2|CpuSSE|CpuSSE2 },
431 {"prescott", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuPNI },
432 {"k6", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX },
433 {"k6_2", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX|Cpu3dnow },
434 {"athlon", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA },
435 {"sledgehammer",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuSledgehammer|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2 },
436 {"opteron", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuSledgehammer|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2 },
437 {".mmx", CpuMMX },
438 {".sse", CpuMMX|CpuMMX2|CpuSSE },
439 {".sse2", CpuMMX|CpuMMX2|CpuSSE|CpuSSE2 },
440 {".sse3", CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3 },
441 {".3dnow", CpuMMX|Cpu3dnow },
442 {".3dnowa", CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA },
443 {".padlock", CpuPadLock },
444 {".pacifica", CpuSVME },
445 {".svme", CpuSVME },
446 {NULL, 0 }
447 };
448
449 const pseudo_typeS md_pseudo_table[] =
450 {
451 #if !defined(OBJ_AOUT) && !defined(USE_ALIGN_PTWO)
452 {"align", s_align_bytes, 0},
453 #else
454 {"align", s_align_ptwo, 0},
455 #endif
456 {"arch", set_cpu_arch, 0},
457 #ifndef I386COFF
458 {"bss", s_bss, 0},
459 #endif
460 {"ffloat", float_cons, 'f'},
461 {"dfloat", float_cons, 'd'},
462 {"tfloat", float_cons, 'x'},
463 {"value", cons, 2},
464 {"slong", signed_cons, 4},
465 {"noopt", s_ignore, 0},
466 {"optim", s_ignore, 0},
467 {"code16gcc", set_16bit_gcc_code_flag, CODE_16BIT},
468 {"code16", set_code_flag, CODE_16BIT},
469 {"code32", set_code_flag, CODE_32BIT},
470 {"code64", set_code_flag, CODE_64BIT},
471 {"intel_syntax", set_intel_syntax, 1},
472 {"att_syntax", set_intel_syntax, 0},
473 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
474 {"largecomm", handle_large_common, 0},
475 #else
476 {"file", (void (*) PARAMS ((int))) dwarf2_directive_file, 0},
477 {"loc", dwarf2_directive_loc, 0},
478 {"loc_mark_labels", dwarf2_directive_loc_mark_labels, 0},
479 #endif
480 #ifdef TE_PE
481 {"secrel32", pe_directive_secrel, 0},
482 #endif
483 {0, 0, 0}
484 };
485
486 /* For interface with expression (). */
487 extern char *input_line_pointer;
488
489 /* Hash table for instruction mnemonic lookup. */
490 static struct hash_control *op_hash;
491
492 /* Hash table for register lookup. */
493 static struct hash_control *reg_hash;
494 \f
495 void
496 i386_align_code (fragP, count)
497 fragS *fragP;
498 int count;
499 {
500 /* Various efficient no-op patterns for aligning code labels.
501 Note: Don't try to assemble the instructions in the comments.
502 0L and 0w are not legal. */
503 static const char f32_1[] =
504 {0x90}; /* nop */
505 static const char f32_2[] =
506 {0x89,0xf6}; /* movl %esi,%esi */
507 static const char f32_3[] =
508 {0x8d,0x76,0x00}; /* leal 0(%esi),%esi */
509 static const char f32_4[] =
510 {0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
511 static const char f32_5[] =
512 {0x90, /* nop */
513 0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
514 static const char f32_6[] =
515 {0x8d,0xb6,0x00,0x00,0x00,0x00}; /* leal 0L(%esi),%esi */
516 static const char f32_7[] =
517 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
518 static const char f32_8[] =
519 {0x90, /* nop */
520 0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
521 static const char f32_9[] =
522 {0x89,0xf6, /* movl %esi,%esi */
523 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
524 static const char f32_10[] =
525 {0x8d,0x76,0x00, /* leal 0(%esi),%esi */
526 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
527 static const char f32_11[] =
528 {0x8d,0x74,0x26,0x00, /* leal 0(%esi,1),%esi */
529 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
530 static const char f32_12[] =
531 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
532 0x8d,0xbf,0x00,0x00,0x00,0x00}; /* leal 0L(%edi),%edi */
533 static const char f32_13[] =
534 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
535 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
536 static const char f32_14[] =
537 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00, /* leal 0L(%esi,1),%esi */
538 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
539 static const char f32_15[] =
540 {0xeb,0x0d,0x90,0x90,0x90,0x90,0x90, /* jmp .+15; lotsa nops */
541 0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90};
542 static const char f16_3[] =
543 {0x8d,0x74,0x00}; /* lea 0(%esi),%esi */
544 static const char f16_4[] =
545 {0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
546 static const char f16_5[] =
547 {0x90, /* nop */
548 0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
549 static const char f16_6[] =
550 {0x89,0xf6, /* mov %si,%si */
551 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
552 static const char f16_7[] =
553 {0x8d,0x74,0x00, /* lea 0(%si),%si */
554 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
555 static const char f16_8[] =
556 {0x8d,0xb4,0x00,0x00, /* lea 0w(%si),%si */
557 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
558 static const char *const f32_patt[] = {
559 f32_1, f32_2, f32_3, f32_4, f32_5, f32_6, f32_7, f32_8,
560 f32_9, f32_10, f32_11, f32_12, f32_13, f32_14, f32_15
561 };
562 static const char *const f16_patt[] = {
563 f32_1, f32_2, f16_3, f16_4, f16_5, f16_6, f16_7, f16_8,
564 f32_15, f32_15, f32_15, f32_15, f32_15, f32_15, f32_15
565 };
566
567 if (count <= 0 || count > 15)
568 return;
569
570 /* The recommended way to pad 64bit code is to use NOPs preceded by
571 maximally four 0x66 prefixes. Balance the size of nops. */
572 if (flag_code == CODE_64BIT)
573 {
574 int i;
575 int nnops = (count + 3) / 4;
576 int len = count / nnops;
577 int remains = count - nnops * len;
578 int pos = 0;
579
580 for (i = 0; i < remains; i++)
581 {
582 memset (fragP->fr_literal + fragP->fr_fix + pos, 0x66, len);
583 fragP->fr_literal[fragP->fr_fix + pos + len] = 0x90;
584 pos += len + 1;
585 }
586 for (; i < nnops; i++)
587 {
588 memset (fragP->fr_literal + fragP->fr_fix + pos, 0x66, len - 1);
589 fragP->fr_literal[fragP->fr_fix + pos + len - 1] = 0x90;
590 pos += len;
591 }
592 }
593 else
594 if (flag_code == CODE_16BIT)
595 {
596 memcpy (fragP->fr_literal + fragP->fr_fix,
597 f16_patt[count - 1], count);
598 if (count > 8)
599 /* Adjust jump offset. */
600 fragP->fr_literal[fragP->fr_fix + 1] = count - 2;
601 }
602 else
603 memcpy (fragP->fr_literal + fragP->fr_fix,
604 f32_patt[count - 1], count);
605 fragP->fr_var = count;
606 }
607
608 static INLINE unsigned int
609 mode_from_disp_size (t)
610 unsigned int t;
611 {
612 return (t & Disp8) ? 1 : (t & (Disp16 | Disp32 | Disp32S)) ? 2 : 0;
613 }
614
615 static INLINE int
616 fits_in_signed_byte (num)
617 offsetT num;
618 {
619 return (num >= -128) && (num <= 127);
620 }
621
622 static INLINE int
623 fits_in_unsigned_byte (num)
624 offsetT num;
625 {
626 return (num & 0xff) == num;
627 }
628
629 static INLINE int
630 fits_in_unsigned_word (num)
631 offsetT num;
632 {
633 return (num & 0xffff) == num;
634 }
635
636 static INLINE int
637 fits_in_signed_word (num)
638 offsetT num;
639 {
640 return (-32768 <= num) && (num <= 32767);
641 }
642 static INLINE int
643 fits_in_signed_long (num)
644 offsetT num ATTRIBUTE_UNUSED;
645 {
646 #ifndef BFD64
647 return 1;
648 #else
649 return (!(((offsetT) -1 << 31) & num)
650 || (((offsetT) -1 << 31) & num) == ((offsetT) -1 << 31));
651 #endif
652 } /* fits_in_signed_long() */
653 static INLINE int
654 fits_in_unsigned_long (num)
655 offsetT num ATTRIBUTE_UNUSED;
656 {
657 #ifndef BFD64
658 return 1;
659 #else
660 return (num & (((offsetT) 2 << 31) - 1)) == num;
661 #endif
662 } /* fits_in_unsigned_long() */
663
664 static int
665 smallest_imm_type (num)
666 offsetT num;
667 {
668 if (cpu_arch_flags != (Cpu086 | Cpu186 | Cpu286 | Cpu386 | Cpu486 | CpuNo64))
669 {
670 /* This code is disabled on the 486 because all the Imm1 forms
671 in the opcode table are slower on the i486. They're the
672 versions with the implicitly specified single-position
673 displacement, which has another syntax if you really want to
674 use that form. */
675 if (num == 1)
676 return Imm1 | Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64;
677 }
678 return (fits_in_signed_byte (num)
679 ? (Imm8S | Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
680 : fits_in_unsigned_byte (num)
681 ? (Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
682 : (fits_in_signed_word (num) || fits_in_unsigned_word (num))
683 ? (Imm16 | Imm32 | Imm32S | Imm64)
684 : fits_in_signed_long (num)
685 ? (Imm32 | Imm32S | Imm64)
686 : fits_in_unsigned_long (num)
687 ? (Imm32 | Imm64)
688 : Imm64);
689 }
690
691 static offsetT
692 offset_in_range (val, size)
693 offsetT val;
694 int size;
695 {
696 addressT mask;
697
698 switch (size)
699 {
700 case 1: mask = ((addressT) 1 << 8) - 1; break;
701 case 2: mask = ((addressT) 1 << 16) - 1; break;
702 case 4: mask = ((addressT) 2 << 31) - 1; break;
703 #ifdef BFD64
704 case 8: mask = ((addressT) 2 << 63) - 1; break;
705 #endif
706 default: abort ();
707 }
708
709 /* If BFD64, sign extend val. */
710 if (!use_rela_relocations)
711 if ((val & ~(((addressT) 2 << 31) - 1)) == 0)
712 val = (val ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
713
714 if ((val & ~mask) != 0 && (val & ~mask) != ~mask)
715 {
716 char buf1[40], buf2[40];
717
718 sprint_value (buf1, val);
719 sprint_value (buf2, val & mask);
720 as_warn (_("%s shortened to %s"), buf1, buf2);
721 }
722 return val & mask;
723 }
724
725 /* Returns 0 if attempting to add a prefix where one from the same
726 class already exists, 1 if non rep/repne added, 2 if rep/repne
727 added. */
728 static int
729 add_prefix (prefix)
730 unsigned int prefix;
731 {
732 int ret = 1;
733 unsigned int q;
734
735 if (prefix >= REX_OPCODE && prefix < REX_OPCODE + 16
736 && flag_code == CODE_64BIT)
737 {
738 if ((i.prefix[REX_PREFIX] & prefix & REX_MODE64)
739 || ((i.prefix[REX_PREFIX] & (REX_EXTX | REX_EXTY | REX_EXTZ))
740 && (prefix & (REX_EXTX | REX_EXTY | REX_EXTZ))))
741 ret = 0;
742 q = REX_PREFIX;
743 }
744 else
745 {
746 switch (prefix)
747 {
748 default:
749 abort ();
750
751 case CS_PREFIX_OPCODE:
752 case DS_PREFIX_OPCODE:
753 case ES_PREFIX_OPCODE:
754 case FS_PREFIX_OPCODE:
755 case GS_PREFIX_OPCODE:
756 case SS_PREFIX_OPCODE:
757 q = SEG_PREFIX;
758 break;
759
760 case REPNE_PREFIX_OPCODE:
761 case REPE_PREFIX_OPCODE:
762 ret = 2;
763 /* fall thru */
764 case LOCK_PREFIX_OPCODE:
765 q = LOCKREP_PREFIX;
766 break;
767
768 case FWAIT_OPCODE:
769 q = WAIT_PREFIX;
770 break;
771
772 case ADDR_PREFIX_OPCODE:
773 q = ADDR_PREFIX;
774 break;
775
776 case DATA_PREFIX_OPCODE:
777 q = DATA_PREFIX;
778 break;
779 }
780 if (i.prefix[q] != 0)
781 ret = 0;
782 }
783
784 if (ret)
785 {
786 if (!i.prefix[q])
787 ++i.prefixes;
788 i.prefix[q] |= prefix;
789 }
790 else
791 as_bad (_("same type of prefix used twice"));
792
793 return ret;
794 }
795
796 static void
797 set_code_flag (value)
798 int value;
799 {
800 flag_code = value;
801 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
802 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
803 if (value == CODE_64BIT && !(cpu_arch_flags & CpuSledgehammer))
804 {
805 as_bad (_("64bit mode not supported on this CPU."));
806 }
807 if (value == CODE_32BIT && !(cpu_arch_flags & Cpu386))
808 {
809 as_bad (_("32bit mode not supported on this CPU."));
810 }
811 stackop_size = '\0';
812 }
813
814 static void
815 set_16bit_gcc_code_flag (new_code_flag)
816 int new_code_flag;
817 {
818 flag_code = new_code_flag;
819 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
820 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
821 stackop_size = LONG_MNEM_SUFFIX;
822 }
823
824 static void
825 set_intel_syntax (syntax_flag)
826 int syntax_flag;
827 {
828 /* Find out if register prefixing is specified. */
829 int ask_naked_reg = 0;
830
831 SKIP_WHITESPACE ();
832 if (!is_end_of_line[(unsigned char) *input_line_pointer])
833 {
834 char *string = input_line_pointer;
835 int e = get_symbol_end ();
836
837 if (strcmp (string, "prefix") == 0)
838 ask_naked_reg = 1;
839 else if (strcmp (string, "noprefix") == 0)
840 ask_naked_reg = -1;
841 else
842 as_bad (_("bad argument to syntax directive."));
843 *input_line_pointer = e;
844 }
845 demand_empty_rest_of_line ();
846
847 intel_syntax = syntax_flag;
848
849 if (ask_naked_reg == 0)
850 allow_naked_reg = (intel_syntax
851 && (bfd_get_symbol_leading_char (stdoutput) != '\0'));
852 else
853 allow_naked_reg = (ask_naked_reg < 0);
854
855 identifier_chars['%'] = intel_syntax && allow_naked_reg ? '%' : 0;
856 identifier_chars['$'] = intel_syntax ? '$' : 0;
857 }
858
859 static void
860 set_cpu_arch (dummy)
861 int dummy ATTRIBUTE_UNUSED;
862 {
863 SKIP_WHITESPACE ();
864
865 if (!is_end_of_line[(unsigned char) *input_line_pointer])
866 {
867 char *string = input_line_pointer;
868 int e = get_symbol_end ();
869 int i;
870
871 for (i = 0; cpu_arch[i].name; i++)
872 {
873 if (strcmp (string, cpu_arch[i].name) == 0)
874 {
875 if (*string != '.')
876 {
877 cpu_arch_name = cpu_arch[i].name;
878 cpu_sub_arch_name = NULL;
879 cpu_arch_flags = (cpu_arch[i].flags
880 | (flag_code == CODE_64BIT ? Cpu64 : CpuNo64));
881 break;
882 }
883 if ((cpu_arch_flags | cpu_arch[i].flags) != cpu_arch_flags)
884 {
885 cpu_sub_arch_name = cpu_arch[i].name;
886 cpu_arch_flags |= cpu_arch[i].flags;
887 }
888 *input_line_pointer = e;
889 demand_empty_rest_of_line ();
890 return;
891 }
892 }
893 if (!cpu_arch[i].name)
894 as_bad (_("no such architecture: `%s'"), string);
895
896 *input_line_pointer = e;
897 }
898 else
899 as_bad (_("missing cpu architecture"));
900
901 no_cond_jump_promotion = 0;
902 if (*input_line_pointer == ','
903 && !is_end_of_line[(unsigned char) input_line_pointer[1]])
904 {
905 char *string = ++input_line_pointer;
906 int e = get_symbol_end ();
907
908 if (strcmp (string, "nojumps") == 0)
909 no_cond_jump_promotion = 1;
910 else if (strcmp (string, "jumps") == 0)
911 ;
912 else
913 as_bad (_("no such architecture modifier: `%s'"), string);
914
915 *input_line_pointer = e;
916 }
917
918 demand_empty_rest_of_line ();
919 }
920
921 unsigned long
922 i386_mach ()
923 {
924 if (!strcmp (default_arch, "x86_64"))
925 return bfd_mach_x86_64;
926 else if (!strcmp (default_arch, "i386"))
927 return bfd_mach_i386_i386;
928 else
929 as_fatal (_("Unknown architecture"));
930 }
931 \f
932 void
933 md_begin ()
934 {
935 const char *hash_err;
936
937 /* Initialize op_hash hash table. */
938 op_hash = hash_new ();
939
940 {
941 const template *optab;
942 templates *core_optab;
943
944 /* Setup for loop. */
945 optab = i386_optab;
946 core_optab = (templates *) xmalloc (sizeof (templates));
947 core_optab->start = optab;
948
949 while (1)
950 {
951 ++optab;
952 if (optab->name == NULL
953 || strcmp (optab->name, (optab - 1)->name) != 0)
954 {
955 /* different name --> ship out current template list;
956 add to hash table; & begin anew. */
957 core_optab->end = optab;
958 hash_err = hash_insert (op_hash,
959 (optab - 1)->name,
960 (PTR) core_optab);
961 if (hash_err)
962 {
963 as_fatal (_("Internal Error: Can't hash %s: %s"),
964 (optab - 1)->name,
965 hash_err);
966 }
967 if (optab->name == NULL)
968 break;
969 core_optab = (templates *) xmalloc (sizeof (templates));
970 core_optab->start = optab;
971 }
972 }
973 }
974
975 /* Initialize reg_hash hash table. */
976 reg_hash = hash_new ();
977 {
978 const reg_entry *regtab;
979
980 for (regtab = i386_regtab;
981 regtab < i386_regtab + sizeof (i386_regtab) / sizeof (i386_regtab[0]);
982 regtab++)
983 {
984 hash_err = hash_insert (reg_hash, regtab->reg_name, (PTR) regtab);
985 if (hash_err)
986 as_fatal (_("Internal Error: Can't hash %s: %s"),
987 regtab->reg_name,
988 hash_err);
989 }
990 }
991
992 /* Fill in lexical tables: mnemonic_chars, operand_chars. */
993 {
994 int c;
995 char *p;
996
997 for (c = 0; c < 256; c++)
998 {
999 if (ISDIGIT (c))
1000 {
1001 digit_chars[c] = c;
1002 mnemonic_chars[c] = c;
1003 register_chars[c] = c;
1004 operand_chars[c] = c;
1005 }
1006 else if (ISLOWER (c))
1007 {
1008 mnemonic_chars[c] = c;
1009 register_chars[c] = c;
1010 operand_chars[c] = c;
1011 }
1012 else if (ISUPPER (c))
1013 {
1014 mnemonic_chars[c] = TOLOWER (c);
1015 register_chars[c] = mnemonic_chars[c];
1016 operand_chars[c] = c;
1017 }
1018
1019 if (ISALPHA (c) || ISDIGIT (c))
1020 identifier_chars[c] = c;
1021 else if (c >= 128)
1022 {
1023 identifier_chars[c] = c;
1024 operand_chars[c] = c;
1025 }
1026 }
1027
1028 #ifdef LEX_AT
1029 identifier_chars['@'] = '@';
1030 #endif
1031 #ifdef LEX_QM
1032 identifier_chars['?'] = '?';
1033 operand_chars['?'] = '?';
1034 #endif
1035 digit_chars['-'] = '-';
1036 mnemonic_chars['-'] = '-';
1037 identifier_chars['_'] = '_';
1038 identifier_chars['.'] = '.';
1039
1040 for (p = operand_special_chars; *p != '\0'; p++)
1041 operand_chars[(unsigned char) *p] = *p;
1042 }
1043
1044 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1045 if (IS_ELF)
1046 {
1047 record_alignment (text_section, 2);
1048 record_alignment (data_section, 2);
1049 record_alignment (bss_section, 2);
1050 }
1051 #endif
1052
1053 if (flag_code == CODE_64BIT)
1054 {
1055 x86_dwarf2_return_column = 16;
1056 x86_cie_data_alignment = -8;
1057 }
1058 else
1059 {
1060 x86_dwarf2_return_column = 8;
1061 x86_cie_data_alignment = -4;
1062 }
1063 }
1064
1065 void
1066 i386_print_statistics (file)
1067 FILE *file;
1068 {
1069 hash_print_statistics (file, "i386 opcode", op_hash);
1070 hash_print_statistics (file, "i386 register", reg_hash);
1071 }
1072 \f
1073 #ifdef DEBUG386
1074
1075 /* Debugging routines for md_assemble. */
1076 static void pi PARAMS ((char *, i386_insn *));
1077 static void pte PARAMS ((template *));
1078 static void pt PARAMS ((unsigned int));
1079 static void pe PARAMS ((expressionS *));
1080 static void ps PARAMS ((symbolS *));
1081
1082 static void
1083 pi (line, x)
1084 char *line;
1085 i386_insn *x;
1086 {
1087 unsigned int i;
1088
1089 fprintf (stdout, "%s: template ", line);
1090 pte (&x->tm);
1091 fprintf (stdout, " address: base %s index %s scale %x\n",
1092 x->base_reg ? x->base_reg->reg_name : "none",
1093 x->index_reg ? x->index_reg->reg_name : "none",
1094 x->log2_scale_factor);
1095 fprintf (stdout, " modrm: mode %x reg %x reg/mem %x\n",
1096 x->rm.mode, x->rm.reg, x->rm.regmem);
1097 fprintf (stdout, " sib: base %x index %x scale %x\n",
1098 x->sib.base, x->sib.index, x->sib.scale);
1099 fprintf (stdout, " rex: 64bit %x extX %x extY %x extZ %x\n",
1100 (x->rex & REX_MODE64) != 0,
1101 (x->rex & REX_EXTX) != 0,
1102 (x->rex & REX_EXTY) != 0,
1103 (x->rex & REX_EXTZ) != 0);
1104 for (i = 0; i < x->operands; i++)
1105 {
1106 fprintf (stdout, " #%d: ", i + 1);
1107 pt (x->types[i]);
1108 fprintf (stdout, "\n");
1109 if (x->types[i]
1110 & (Reg | SReg2 | SReg3 | Control | Debug | Test | RegMMX | RegXMM))
1111 fprintf (stdout, "%s\n", x->op[i].regs->reg_name);
1112 if (x->types[i] & Imm)
1113 pe (x->op[i].imms);
1114 if (x->types[i] & Disp)
1115 pe (x->op[i].disps);
1116 }
1117 }
1118
1119 static void
1120 pte (t)
1121 template *t;
1122 {
1123 unsigned int i;
1124 fprintf (stdout, " %d operands ", t->operands);
1125 fprintf (stdout, "opcode %x ", t->base_opcode);
1126 if (t->extension_opcode != None)
1127 fprintf (stdout, "ext %x ", t->extension_opcode);
1128 if (t->opcode_modifier & D)
1129 fprintf (stdout, "D");
1130 if (t->opcode_modifier & W)
1131 fprintf (stdout, "W");
1132 fprintf (stdout, "\n");
1133 for (i = 0; i < t->operands; i++)
1134 {
1135 fprintf (stdout, " #%d type ", i + 1);
1136 pt (t->operand_types[i]);
1137 fprintf (stdout, "\n");
1138 }
1139 }
1140
1141 static void
1142 pe (e)
1143 expressionS *e;
1144 {
1145 fprintf (stdout, " operation %d\n", e->X_op);
1146 fprintf (stdout, " add_number %ld (%lx)\n",
1147 (long) e->X_add_number, (long) e->X_add_number);
1148 if (e->X_add_symbol)
1149 {
1150 fprintf (stdout, " add_symbol ");
1151 ps (e->X_add_symbol);
1152 fprintf (stdout, "\n");
1153 }
1154 if (e->X_op_symbol)
1155 {
1156 fprintf (stdout, " op_symbol ");
1157 ps (e->X_op_symbol);
1158 fprintf (stdout, "\n");
1159 }
1160 }
1161
1162 static void
1163 ps (s)
1164 symbolS *s;
1165 {
1166 fprintf (stdout, "%s type %s%s",
1167 S_GET_NAME (s),
1168 S_IS_EXTERNAL (s) ? "EXTERNAL " : "",
1169 segment_name (S_GET_SEGMENT (s)));
1170 }
1171
1172 static struct type_name
1173 {
1174 unsigned int mask;
1175 char *tname;
1176 }
1177 const type_names[] =
1178 {
1179 { Reg8, "r8" },
1180 { Reg16, "r16" },
1181 { Reg32, "r32" },
1182 { Reg64, "r64" },
1183 { Imm8, "i8" },
1184 { Imm8S, "i8s" },
1185 { Imm16, "i16" },
1186 { Imm32, "i32" },
1187 { Imm32S, "i32s" },
1188 { Imm64, "i64" },
1189 { Imm1, "i1" },
1190 { BaseIndex, "BaseIndex" },
1191 { Disp8, "d8" },
1192 { Disp16, "d16" },
1193 { Disp32, "d32" },
1194 { Disp32S, "d32s" },
1195 { Disp64, "d64" },
1196 { InOutPortReg, "InOutPortReg" },
1197 { ShiftCount, "ShiftCount" },
1198 { Control, "control reg" },
1199 { Test, "test reg" },
1200 { Debug, "debug reg" },
1201 { FloatReg, "FReg" },
1202 { FloatAcc, "FAcc" },
1203 { SReg2, "SReg2" },
1204 { SReg3, "SReg3" },
1205 { Acc, "Acc" },
1206 { JumpAbsolute, "Jump Absolute" },
1207 { RegMMX, "rMMX" },
1208 { RegXMM, "rXMM" },
1209 { EsSeg, "es" },
1210 { 0, "" }
1211 };
1212
1213 static void
1214 pt (t)
1215 unsigned int t;
1216 {
1217 const struct type_name *ty;
1218
1219 for (ty = type_names; ty->mask; ty++)
1220 if (t & ty->mask)
1221 fprintf (stdout, "%s, ", ty->tname);
1222 fflush (stdout);
1223 }
1224
1225 #endif /* DEBUG386 */
1226 \f
1227 static bfd_reloc_code_real_type
1228 reloc (unsigned int size,
1229 int pcrel,
1230 int sign,
1231 bfd_reloc_code_real_type other)
1232 {
1233 if (other != NO_RELOC)
1234 {
1235 reloc_howto_type *reloc;
1236
1237 if (size == 8)
1238 switch (other)
1239 {
1240 case BFD_RELOC_X86_64_GOT32:
1241 return BFD_RELOC_X86_64_GOT64;
1242 break;
1243 case BFD_RELOC_X86_64_PLTOFF64:
1244 return BFD_RELOC_X86_64_PLTOFF64;
1245 break;
1246 case BFD_RELOC_X86_64_GOTPC32:
1247 other = BFD_RELOC_X86_64_GOTPC64;
1248 break;
1249 case BFD_RELOC_X86_64_GOTPCREL:
1250 other = BFD_RELOC_X86_64_GOTPCREL64;
1251 break;
1252 case BFD_RELOC_X86_64_TPOFF32:
1253 other = BFD_RELOC_X86_64_TPOFF64;
1254 break;
1255 case BFD_RELOC_X86_64_DTPOFF32:
1256 other = BFD_RELOC_X86_64_DTPOFF64;
1257 break;
1258 default:
1259 break;
1260 }
1261
1262 /* Sign-checking 4-byte relocations in 16-/32-bit code is pointless. */
1263 if (size == 4 && flag_code != CODE_64BIT)
1264 sign = -1;
1265
1266 reloc = bfd_reloc_type_lookup (stdoutput, other);
1267 if (!reloc)
1268 as_bad (_("unknown relocation (%u)"), other);
1269 else if (size != bfd_get_reloc_size (reloc))
1270 as_bad (_("%u-byte relocation cannot be applied to %u-byte field"),
1271 bfd_get_reloc_size (reloc),
1272 size);
1273 else if (pcrel && !reloc->pc_relative)
1274 as_bad (_("non-pc-relative relocation for pc-relative field"));
1275 else if ((reloc->complain_on_overflow == complain_overflow_signed
1276 && !sign)
1277 || (reloc->complain_on_overflow == complain_overflow_unsigned
1278 && sign > 0))
1279 as_bad (_("relocated field and relocation type differ in signedness"));
1280 else
1281 return other;
1282 return NO_RELOC;
1283 }
1284
1285 if (pcrel)
1286 {
1287 if (!sign)
1288 as_bad (_("there are no unsigned pc-relative relocations"));
1289 switch (size)
1290 {
1291 case 1: return BFD_RELOC_8_PCREL;
1292 case 2: return BFD_RELOC_16_PCREL;
1293 case 4: return BFD_RELOC_32_PCREL;
1294 case 8: return BFD_RELOC_64_PCREL;
1295 }
1296 as_bad (_("cannot do %u byte pc-relative relocation"), size);
1297 }
1298 else
1299 {
1300 if (sign > 0)
1301 switch (size)
1302 {
1303 case 4: return BFD_RELOC_X86_64_32S;
1304 }
1305 else
1306 switch (size)
1307 {
1308 case 1: return BFD_RELOC_8;
1309 case 2: return BFD_RELOC_16;
1310 case 4: return BFD_RELOC_32;
1311 case 8: return BFD_RELOC_64;
1312 }
1313 as_bad (_("cannot do %s %u byte relocation"),
1314 sign > 0 ? "signed" : "unsigned", size);
1315 }
1316
1317 abort ();
1318 return BFD_RELOC_NONE;
1319 }
1320
1321 /* Here we decide which fixups can be adjusted to make them relative to
1322 the beginning of the section instead of the symbol. Basically we need
1323 to make sure that the dynamic relocations are done correctly, so in
1324 some cases we force the original symbol to be used. */
1325
1326 int
1327 tc_i386_fix_adjustable (fixP)
1328 fixS *fixP ATTRIBUTE_UNUSED;
1329 {
1330 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1331 if (!IS_ELF)
1332 return 1;
1333
1334 /* Don't adjust pc-relative references to merge sections in 64-bit
1335 mode. */
1336 if (use_rela_relocations
1337 && (S_GET_SEGMENT (fixP->fx_addsy)->flags & SEC_MERGE) != 0
1338 && fixP->fx_pcrel)
1339 return 0;
1340
1341 /* The x86_64 GOTPCREL are represented as 32bit PCrel relocations
1342 and changed later by validate_fix. */
1343 if (GOT_symbol && fixP->fx_subsy == GOT_symbol
1344 && fixP->fx_r_type == BFD_RELOC_32_PCREL)
1345 return 0;
1346
1347 /* adjust_reloc_syms doesn't know about the GOT. */
1348 if (fixP->fx_r_type == BFD_RELOC_386_GOTOFF
1349 || fixP->fx_r_type == BFD_RELOC_386_PLT32
1350 || fixP->fx_r_type == BFD_RELOC_386_GOT32
1351 || fixP->fx_r_type == BFD_RELOC_386_TLS_GD
1352 || fixP->fx_r_type == BFD_RELOC_386_TLS_LDM
1353 || fixP->fx_r_type == BFD_RELOC_386_TLS_LDO_32
1354 || fixP->fx_r_type == BFD_RELOC_386_TLS_IE_32
1355 || fixP->fx_r_type == BFD_RELOC_386_TLS_IE
1356 || fixP->fx_r_type == BFD_RELOC_386_TLS_GOTIE
1357 || fixP->fx_r_type == BFD_RELOC_386_TLS_LE_32
1358 || fixP->fx_r_type == BFD_RELOC_386_TLS_LE
1359 || fixP->fx_r_type == BFD_RELOC_386_TLS_GOTDESC
1360 || fixP->fx_r_type == BFD_RELOC_386_TLS_DESC_CALL
1361 || fixP->fx_r_type == BFD_RELOC_X86_64_PLT32
1362 || fixP->fx_r_type == BFD_RELOC_X86_64_GOT32
1363 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPCREL
1364 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSGD
1365 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSLD
1366 || fixP->fx_r_type == BFD_RELOC_X86_64_DTPOFF32
1367 || fixP->fx_r_type == BFD_RELOC_X86_64_DTPOFF64
1368 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTTPOFF
1369 || fixP->fx_r_type == BFD_RELOC_X86_64_TPOFF32
1370 || fixP->fx_r_type == BFD_RELOC_X86_64_TPOFF64
1371 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTOFF64
1372 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPC32_TLSDESC
1373 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSDESC_CALL
1374 || fixP->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1375 || fixP->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1376 return 0;
1377 #endif
1378 return 1;
1379 }
1380
1381 static int intel_float_operand PARAMS ((const char *mnemonic));
1382
1383 static int
1384 intel_float_operand (mnemonic)
1385 const char *mnemonic;
1386 {
1387 /* Note that the value returned is meaningful only for opcodes with (memory)
1388 operands, hence the code here is free to improperly handle opcodes that
1389 have no operands (for better performance and smaller code). */
1390
1391 if (mnemonic[0] != 'f')
1392 return 0; /* non-math */
1393
1394 switch (mnemonic[1])
1395 {
1396 /* fclex, fdecstp, fdisi, femms, feni, fincstp, finit, fsetpm, and
1397 the fs segment override prefix not currently handled because no
1398 call path can make opcodes without operands get here */
1399 case 'i':
1400 return 2 /* integer op */;
1401 case 'l':
1402 if (mnemonic[2] == 'd' && (mnemonic[3] == 'c' || mnemonic[3] == 'e'))
1403 return 3; /* fldcw/fldenv */
1404 break;
1405 case 'n':
1406 if (mnemonic[2] != 'o' /* fnop */)
1407 return 3; /* non-waiting control op */
1408 break;
1409 case 'r':
1410 if (mnemonic[2] == 's')
1411 return 3; /* frstor/frstpm */
1412 break;
1413 case 's':
1414 if (mnemonic[2] == 'a')
1415 return 3; /* fsave */
1416 if (mnemonic[2] == 't')
1417 {
1418 switch (mnemonic[3])
1419 {
1420 case 'c': /* fstcw */
1421 case 'd': /* fstdw */
1422 case 'e': /* fstenv */
1423 case 's': /* fsts[gw] */
1424 return 3;
1425 }
1426 }
1427 break;
1428 case 'x':
1429 if (mnemonic[2] == 'r' || mnemonic[2] == 's')
1430 return 0; /* fxsave/fxrstor are not really math ops */
1431 break;
1432 }
1433
1434 return 1;
1435 }
1436
1437 /* This is the guts of the machine-dependent assembler. LINE points to a
1438 machine dependent instruction. This function is supposed to emit
1439 the frags/bytes it assembles to. */
1440
1441 void
1442 md_assemble (line)
1443 char *line;
1444 {
1445 int j;
1446 char mnemonic[MAX_MNEM_SIZE];
1447
1448 /* Initialize globals. */
1449 memset (&i, '\0', sizeof (i));
1450 for (j = 0; j < MAX_OPERANDS; j++)
1451 i.reloc[j] = NO_RELOC;
1452 memset (disp_expressions, '\0', sizeof (disp_expressions));
1453 memset (im_expressions, '\0', sizeof (im_expressions));
1454 save_stack_p = save_stack;
1455
1456 /* First parse an instruction mnemonic & call i386_operand for the operands.
1457 We assume that the scrubber has arranged it so that line[0] is the valid
1458 start of a (possibly prefixed) mnemonic. */
1459
1460 line = parse_insn (line, mnemonic);
1461 if (line == NULL)
1462 return;
1463
1464 line = parse_operands (line, mnemonic);
1465 if (line == NULL)
1466 return;
1467
1468 /* Now we've parsed the mnemonic into a set of templates, and have the
1469 operands at hand. */
1470
1471 /* All intel opcodes have reversed operands except for "bound" and
1472 "enter". We also don't reverse intersegment "jmp" and "call"
1473 instructions with 2 immediate operands so that the immediate segment
1474 precedes the offset, as it does when in AT&T mode. "enter" and the
1475 intersegment "jmp" and "call" instructions are the only ones that
1476 have two immediate operands. */
1477 if (intel_syntax && i.operands > 1
1478 && (strcmp (mnemonic, "bound") != 0)
1479 && (strcmp (mnemonic, "invlpga") != 0)
1480 && !((i.types[0] & Imm) && (i.types[1] & Imm)))
1481 swap_operands ();
1482
1483 if (i.imm_operands)
1484 optimize_imm ();
1485
1486 /* Don't optimize displacement for movabs since it only takes 64bit
1487 displacement. */
1488 if (i.disp_operands
1489 && (flag_code != CODE_64BIT
1490 || strcmp (mnemonic, "movabs") != 0))
1491 optimize_disp ();
1492
1493 /* Next, we find a template that matches the given insn,
1494 making sure the overlap of the given operands types is consistent
1495 with the template operand types. */
1496
1497 if (!match_template ())
1498 return;
1499
1500 if (intel_syntax)
1501 {
1502 /* Undo SYSV386_COMPAT brokenness when in Intel mode. See i386.h */
1503 if (SYSV386_COMPAT
1504 && (i.tm.base_opcode & 0xfffffde0) == 0xdce0)
1505 i.tm.base_opcode ^= FloatR;
1506
1507 /* Zap movzx and movsx suffix. The suffix may have been set from
1508 "word ptr" or "byte ptr" on the source operand, but we'll use
1509 the suffix later to choose the destination register. */
1510 if ((i.tm.base_opcode & ~9) == 0x0fb6)
1511 {
1512 if (i.reg_operands < 2
1513 && !i.suffix
1514 && (~i.tm.opcode_modifier
1515 & (No_bSuf
1516 | No_wSuf
1517 | No_lSuf
1518 | No_sSuf
1519 | No_xSuf
1520 | No_qSuf)))
1521 as_bad (_("ambiguous operand size for `%s'"), i.tm.name);
1522
1523 i.suffix = 0;
1524 }
1525 }
1526
1527 if (i.tm.opcode_modifier & FWait)
1528 if (!add_prefix (FWAIT_OPCODE))
1529 return;
1530
1531 /* Check string instruction segment overrides. */
1532 if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0)
1533 {
1534 if (!check_string ())
1535 return;
1536 }
1537
1538 if (!process_suffix ())
1539 return;
1540
1541 /* Make still unresolved immediate matches conform to size of immediate
1542 given in i.suffix. */
1543 if (!finalize_imm ())
1544 return;
1545
1546 if (i.types[0] & Imm1)
1547 i.imm_operands = 0; /* kludge for shift insns. */
1548 if (i.types[0] & ImplicitRegister)
1549 i.reg_operands--;
1550 if (i.types[1] & ImplicitRegister)
1551 i.reg_operands--;
1552 if (i.types[2] & ImplicitRegister)
1553 i.reg_operands--;
1554
1555 if (i.tm.opcode_modifier & ImmExt)
1556 {
1557 expressionS *exp;
1558
1559 if ((i.tm.cpu_flags & CpuPNI) && i.operands > 0)
1560 {
1561 /* These Intel Prescott New Instructions have the fixed
1562 operands with an opcode suffix which is coded in the same
1563 place as an 8-bit immediate field would be. Here we check
1564 those operands and remove them afterwards. */
1565 unsigned int x;
1566
1567 for (x = 0; x < i.operands; x++)
1568 if (i.op[x].regs->reg_num != x)
1569 as_bad (_("can't use register '%%%s' as operand %d in '%s'."),
1570 i.op[x].regs->reg_name, x + 1, i.tm.name);
1571 i.operands = 0;
1572 }
1573
1574 /* These AMD 3DNow! and Intel Katmai New Instructions have an
1575 opcode suffix which is coded in the same place as an 8-bit
1576 immediate field would be. Here we fake an 8-bit immediate
1577 operand from the opcode suffix stored in tm.extension_opcode. */
1578
1579 assert (i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS);
1580
1581 exp = &im_expressions[i.imm_operands++];
1582 i.op[i.operands].imms = exp;
1583 i.types[i.operands++] = Imm8;
1584 exp->X_op = O_constant;
1585 exp->X_add_number = i.tm.extension_opcode;
1586 i.tm.extension_opcode = None;
1587 }
1588
1589 /* For insns with operands there are more diddles to do to the opcode. */
1590 if (i.operands)
1591 {
1592 if (!process_operands ())
1593 return;
1594 }
1595 else if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
1596 {
1597 /* UnixWare fsub no args is alias for fsubp, fadd -> faddp, etc. */
1598 as_warn (_("translating to `%sp'"), i.tm.name);
1599 }
1600
1601 /* Handle conversion of 'int $3' --> special int3 insn. */
1602 if (i.tm.base_opcode == INT_OPCODE && i.op[0].imms->X_add_number == 3)
1603 {
1604 i.tm.base_opcode = INT3_OPCODE;
1605 i.imm_operands = 0;
1606 }
1607
1608 if ((i.tm.opcode_modifier & (Jump | JumpByte | JumpDword))
1609 && i.op[0].disps->X_op == O_constant)
1610 {
1611 /* Convert "jmp constant" (and "call constant") to a jump (call) to
1612 the absolute address given by the constant. Since ix86 jumps and
1613 calls are pc relative, we need to generate a reloc. */
1614 i.op[0].disps->X_add_symbol = &abs_symbol;
1615 i.op[0].disps->X_op = O_symbol;
1616 }
1617
1618 if ((i.tm.opcode_modifier & Rex64) != 0)
1619 i.rex |= REX_MODE64;
1620
1621 /* For 8 bit registers we need an empty rex prefix. Also if the
1622 instruction already has a prefix, we need to convert old
1623 registers to new ones. */
1624
1625 if (((i.types[0] & Reg8) != 0
1626 && (i.op[0].regs->reg_flags & RegRex64) != 0)
1627 || ((i.types[1] & Reg8) != 0
1628 && (i.op[1].regs->reg_flags & RegRex64) != 0)
1629 || (((i.types[0] & Reg8) != 0 || (i.types[1] & Reg8) != 0)
1630 && i.rex != 0))
1631 {
1632 int x;
1633
1634 i.rex |= REX_OPCODE;
1635 for (x = 0; x < 2; x++)
1636 {
1637 /* Look for 8 bit operand that uses old registers. */
1638 if ((i.types[x] & Reg8) != 0
1639 && (i.op[x].regs->reg_flags & RegRex64) == 0)
1640 {
1641 /* In case it is "hi" register, give up. */
1642 if (i.op[x].regs->reg_num > 3)
1643 as_bad (_("can't encode register '%%%s' in an instruction requiring REX prefix."),
1644 i.op[x].regs->reg_name);
1645
1646 /* Otherwise it is equivalent to the extended register.
1647 Since the encoding doesn't change this is merely
1648 cosmetic cleanup for debug output. */
1649
1650 i.op[x].regs = i.op[x].regs + 8;
1651 }
1652 }
1653 }
1654
1655 if (i.rex != 0)
1656 add_prefix (REX_OPCODE | i.rex);
1657
1658 /* We are ready to output the insn. */
1659 output_insn ();
1660 }
1661
1662 static char *
1663 parse_insn (line, mnemonic)
1664 char *line;
1665 char *mnemonic;
1666 {
1667 char *l = line;
1668 char *token_start = l;
1669 char *mnem_p;
1670 int supported;
1671 const template *t;
1672
1673 /* Non-zero if we found a prefix only acceptable with string insns. */
1674 const char *expecting_string_instruction = NULL;
1675
1676 while (1)
1677 {
1678 mnem_p = mnemonic;
1679 while ((*mnem_p = mnemonic_chars[(unsigned char) *l]) != 0)
1680 {
1681 mnem_p++;
1682 if (mnem_p >= mnemonic + MAX_MNEM_SIZE)
1683 {
1684 as_bad (_("no such instruction: `%s'"), token_start);
1685 return NULL;
1686 }
1687 l++;
1688 }
1689 if (!is_space_char (*l)
1690 && *l != END_OF_INSN
1691 && (intel_syntax
1692 || (*l != PREFIX_SEPARATOR
1693 && *l != ',')))
1694 {
1695 as_bad (_("invalid character %s in mnemonic"),
1696 output_invalid (*l));
1697 return NULL;
1698 }
1699 if (token_start == l)
1700 {
1701 if (!intel_syntax && *l == PREFIX_SEPARATOR)
1702 as_bad (_("expecting prefix; got nothing"));
1703 else
1704 as_bad (_("expecting mnemonic; got nothing"));
1705 return NULL;
1706 }
1707
1708 /* Look up instruction (or prefix) via hash table. */
1709 current_templates = hash_find (op_hash, mnemonic);
1710
1711 if (*l != END_OF_INSN
1712 && (!is_space_char (*l) || l[1] != END_OF_INSN)
1713 && current_templates
1714 && (current_templates->start->opcode_modifier & IsPrefix))
1715 {
1716 if (current_templates->start->cpu_flags
1717 & (flag_code != CODE_64BIT ? Cpu64 : CpuNo64))
1718 {
1719 as_bad ((flag_code != CODE_64BIT
1720 ? _("`%s' is only supported in 64-bit mode")
1721 : _("`%s' is not supported in 64-bit mode")),
1722 current_templates->start->name);
1723 return NULL;
1724 }
1725 /* If we are in 16-bit mode, do not allow addr16 or data16.
1726 Similarly, in 32-bit mode, do not allow addr32 or data32. */
1727 if ((current_templates->start->opcode_modifier & (Size16 | Size32))
1728 && flag_code != CODE_64BIT
1729 && (((current_templates->start->opcode_modifier & Size32) != 0)
1730 ^ (flag_code == CODE_16BIT)))
1731 {
1732 as_bad (_("redundant %s prefix"),
1733 current_templates->start->name);
1734 return NULL;
1735 }
1736 /* Add prefix, checking for repeated prefixes. */
1737 switch (add_prefix (current_templates->start->base_opcode))
1738 {
1739 case 0:
1740 return NULL;
1741 case 2:
1742 expecting_string_instruction = current_templates->start->name;
1743 break;
1744 }
1745 /* Skip past PREFIX_SEPARATOR and reset token_start. */
1746 token_start = ++l;
1747 }
1748 else
1749 break;
1750 }
1751
1752 if (!current_templates)
1753 {
1754 /* See if we can get a match by trimming off a suffix. */
1755 switch (mnem_p[-1])
1756 {
1757 case WORD_MNEM_SUFFIX:
1758 if (intel_syntax && (intel_float_operand (mnemonic) & 2))
1759 i.suffix = SHORT_MNEM_SUFFIX;
1760 else
1761 case BYTE_MNEM_SUFFIX:
1762 case QWORD_MNEM_SUFFIX:
1763 i.suffix = mnem_p[-1];
1764 mnem_p[-1] = '\0';
1765 current_templates = hash_find (op_hash, mnemonic);
1766 break;
1767 case SHORT_MNEM_SUFFIX:
1768 case LONG_MNEM_SUFFIX:
1769 if (!intel_syntax)
1770 {
1771 i.suffix = mnem_p[-1];
1772 mnem_p[-1] = '\0';
1773 current_templates = hash_find (op_hash, mnemonic);
1774 }
1775 break;
1776
1777 /* Intel Syntax. */
1778 case 'd':
1779 if (intel_syntax)
1780 {
1781 if (intel_float_operand (mnemonic) == 1)
1782 i.suffix = SHORT_MNEM_SUFFIX;
1783 else
1784 i.suffix = LONG_MNEM_SUFFIX;
1785 mnem_p[-1] = '\0';
1786 current_templates = hash_find (op_hash, mnemonic);
1787 }
1788 break;
1789 }
1790 if (!current_templates)
1791 {
1792 as_bad (_("no such instruction: `%s'"), token_start);
1793 return NULL;
1794 }
1795 }
1796
1797 if (current_templates->start->opcode_modifier & (Jump | JumpByte))
1798 {
1799 /* Check for a branch hint. We allow ",pt" and ",pn" for
1800 predict taken and predict not taken respectively.
1801 I'm not sure that branch hints actually do anything on loop
1802 and jcxz insns (JumpByte) for current Pentium4 chips. They
1803 may work in the future and it doesn't hurt to accept them
1804 now. */
1805 if (l[0] == ',' && l[1] == 'p')
1806 {
1807 if (l[2] == 't')
1808 {
1809 if (!add_prefix (DS_PREFIX_OPCODE))
1810 return NULL;
1811 l += 3;
1812 }
1813 else if (l[2] == 'n')
1814 {
1815 if (!add_prefix (CS_PREFIX_OPCODE))
1816 return NULL;
1817 l += 3;
1818 }
1819 }
1820 }
1821 /* Any other comma loses. */
1822 if (*l == ',')
1823 {
1824 as_bad (_("invalid character %s in mnemonic"),
1825 output_invalid (*l));
1826 return NULL;
1827 }
1828
1829 /* Check if instruction is supported on specified architecture. */
1830 supported = 0;
1831 for (t = current_templates->start; t < current_templates->end; ++t)
1832 {
1833 if (!((t->cpu_flags & ~(Cpu64 | CpuNo64))
1834 & ~(cpu_arch_flags & ~(Cpu64 | CpuNo64))))
1835 supported |= 1;
1836 if (!(t->cpu_flags & (flag_code == CODE_64BIT ? CpuNo64 : Cpu64)))
1837 supported |= 2;
1838 }
1839 if (!(supported & 2))
1840 {
1841 as_bad (flag_code == CODE_64BIT
1842 ? _("`%s' is not supported in 64-bit mode")
1843 : _("`%s' is only supported in 64-bit mode"),
1844 current_templates->start->name);
1845 return NULL;
1846 }
1847 if (!(supported & 1))
1848 {
1849 as_warn (_("`%s' is not supported on `%s%s'"),
1850 current_templates->start->name,
1851 cpu_arch_name,
1852 cpu_sub_arch_name ? cpu_sub_arch_name : "");
1853 }
1854 else if ((Cpu386 & ~cpu_arch_flags) && (flag_code != CODE_16BIT))
1855 {
1856 as_warn (_("use .code16 to ensure correct addressing mode"));
1857 }
1858
1859 /* Check for rep/repne without a string instruction. */
1860 if (expecting_string_instruction)
1861 {
1862 static templates override;
1863
1864 for (t = current_templates->start; t < current_templates->end; ++t)
1865 if (t->opcode_modifier & IsString)
1866 break;
1867 if (t >= current_templates->end)
1868 {
1869 as_bad (_("expecting string instruction after `%s'"),
1870 expecting_string_instruction);
1871 return NULL;
1872 }
1873 for (override.start = t; t < current_templates->end; ++t)
1874 if (!(t->opcode_modifier & IsString))
1875 break;
1876 override.end = t;
1877 current_templates = &override;
1878 }
1879
1880 return l;
1881 }
1882
1883 static char *
1884 parse_operands (l, mnemonic)
1885 char *l;
1886 const char *mnemonic;
1887 {
1888 char *token_start;
1889
1890 /* 1 if operand is pending after ','. */
1891 unsigned int expecting_operand = 0;
1892
1893 /* Non-zero if operand parens not balanced. */
1894 unsigned int paren_not_balanced;
1895
1896 while (*l != END_OF_INSN)
1897 {
1898 /* Skip optional white space before operand. */
1899 if (is_space_char (*l))
1900 ++l;
1901 if (!is_operand_char (*l) && *l != END_OF_INSN)
1902 {
1903 as_bad (_("invalid character %s before operand %d"),
1904 output_invalid (*l),
1905 i.operands + 1);
1906 return NULL;
1907 }
1908 token_start = l; /* after white space */
1909 paren_not_balanced = 0;
1910 while (paren_not_balanced || *l != ',')
1911 {
1912 if (*l == END_OF_INSN)
1913 {
1914 if (paren_not_balanced)
1915 {
1916 if (!intel_syntax)
1917 as_bad (_("unbalanced parenthesis in operand %d."),
1918 i.operands + 1);
1919 else
1920 as_bad (_("unbalanced brackets in operand %d."),
1921 i.operands + 1);
1922 return NULL;
1923 }
1924 else
1925 break; /* we are done */
1926 }
1927 else if (!is_operand_char (*l) && !is_space_char (*l))
1928 {
1929 as_bad (_("invalid character %s in operand %d"),
1930 output_invalid (*l),
1931 i.operands + 1);
1932 return NULL;
1933 }
1934 if (!intel_syntax)
1935 {
1936 if (*l == '(')
1937 ++paren_not_balanced;
1938 if (*l == ')')
1939 --paren_not_balanced;
1940 }
1941 else
1942 {
1943 if (*l == '[')
1944 ++paren_not_balanced;
1945 if (*l == ']')
1946 --paren_not_balanced;
1947 }
1948 l++;
1949 }
1950 if (l != token_start)
1951 { /* Yes, we've read in another operand. */
1952 unsigned int operand_ok;
1953 this_operand = i.operands++;
1954 if (i.operands > MAX_OPERANDS)
1955 {
1956 as_bad (_("spurious operands; (%d operands/instruction max)"),
1957 MAX_OPERANDS);
1958 return NULL;
1959 }
1960 /* Now parse operand adding info to 'i' as we go along. */
1961 END_STRING_AND_SAVE (l);
1962
1963 if (intel_syntax)
1964 operand_ok =
1965 i386_intel_operand (token_start,
1966 intel_float_operand (mnemonic));
1967 else
1968 operand_ok = i386_operand (token_start);
1969
1970 RESTORE_END_STRING (l);
1971 if (!operand_ok)
1972 return NULL;
1973 }
1974 else
1975 {
1976 if (expecting_operand)
1977 {
1978 expecting_operand_after_comma:
1979 as_bad (_("expecting operand after ','; got nothing"));
1980 return NULL;
1981 }
1982 if (*l == ',')
1983 {
1984 as_bad (_("expecting operand before ','; got nothing"));
1985 return NULL;
1986 }
1987 }
1988
1989 /* Now *l must be either ',' or END_OF_INSN. */
1990 if (*l == ',')
1991 {
1992 if (*++l == END_OF_INSN)
1993 {
1994 /* Just skip it, if it's \n complain. */
1995 goto expecting_operand_after_comma;
1996 }
1997 expecting_operand = 1;
1998 }
1999 }
2000 return l;
2001 }
2002
2003 static void
2004 swap_operands ()
2005 {
2006 union i386_op temp_op;
2007 unsigned int temp_type;
2008 enum bfd_reloc_code_real temp_reloc;
2009 int xchg1 = 0;
2010 int xchg2 = 0;
2011
2012 if (i.operands == 2)
2013 {
2014 xchg1 = 0;
2015 xchg2 = 1;
2016 }
2017 else if (i.operands == 3)
2018 {
2019 xchg1 = 0;
2020 xchg2 = 2;
2021 }
2022 temp_type = i.types[xchg2];
2023 i.types[xchg2] = i.types[xchg1];
2024 i.types[xchg1] = temp_type;
2025 temp_op = i.op[xchg2];
2026 i.op[xchg2] = i.op[xchg1];
2027 i.op[xchg1] = temp_op;
2028 temp_reloc = i.reloc[xchg2];
2029 i.reloc[xchg2] = i.reloc[xchg1];
2030 i.reloc[xchg1] = temp_reloc;
2031
2032 if (i.mem_operands == 2)
2033 {
2034 const seg_entry *temp_seg;
2035 temp_seg = i.seg[0];
2036 i.seg[0] = i.seg[1];
2037 i.seg[1] = temp_seg;
2038 }
2039 }
2040
2041 /* Try to ensure constant immediates are represented in the smallest
2042 opcode possible. */
2043 static void
2044 optimize_imm ()
2045 {
2046 char guess_suffix = 0;
2047 int op;
2048
2049 if (i.suffix)
2050 guess_suffix = i.suffix;
2051 else if (i.reg_operands)
2052 {
2053 /* Figure out a suffix from the last register operand specified.
2054 We can't do this properly yet, ie. excluding InOutPortReg,
2055 but the following works for instructions with immediates.
2056 In any case, we can't set i.suffix yet. */
2057 for (op = i.operands; --op >= 0;)
2058 if (i.types[op] & Reg)
2059 {
2060 if (i.types[op] & Reg8)
2061 guess_suffix = BYTE_MNEM_SUFFIX;
2062 else if (i.types[op] & Reg16)
2063 guess_suffix = WORD_MNEM_SUFFIX;
2064 else if (i.types[op] & Reg32)
2065 guess_suffix = LONG_MNEM_SUFFIX;
2066 else if (i.types[op] & Reg64)
2067 guess_suffix = QWORD_MNEM_SUFFIX;
2068 break;
2069 }
2070 }
2071 else if ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0))
2072 guess_suffix = WORD_MNEM_SUFFIX;
2073
2074 for (op = i.operands; --op >= 0;)
2075 if (i.types[op] & Imm)
2076 {
2077 switch (i.op[op].imms->X_op)
2078 {
2079 case O_constant:
2080 /* If a suffix is given, this operand may be shortened. */
2081 switch (guess_suffix)
2082 {
2083 case LONG_MNEM_SUFFIX:
2084 i.types[op] |= Imm32 | Imm64;
2085 break;
2086 case WORD_MNEM_SUFFIX:
2087 i.types[op] |= Imm16 | Imm32S | Imm32 | Imm64;
2088 break;
2089 case BYTE_MNEM_SUFFIX:
2090 i.types[op] |= Imm16 | Imm8 | Imm8S | Imm32S | Imm32 | Imm64;
2091 break;
2092 }
2093
2094 /* If this operand is at most 16 bits, convert it
2095 to a signed 16 bit number before trying to see
2096 whether it will fit in an even smaller size.
2097 This allows a 16-bit operand such as $0xffe0 to
2098 be recognised as within Imm8S range. */
2099 if ((i.types[op] & Imm16)
2100 && (i.op[op].imms->X_add_number & ~(offsetT) 0xffff) == 0)
2101 {
2102 i.op[op].imms->X_add_number =
2103 (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000);
2104 }
2105 if ((i.types[op] & Imm32)
2106 && ((i.op[op].imms->X_add_number & ~(((offsetT) 2 << 31) - 1))
2107 == 0))
2108 {
2109 i.op[op].imms->X_add_number = ((i.op[op].imms->X_add_number
2110 ^ ((offsetT) 1 << 31))
2111 - ((offsetT) 1 << 31));
2112 }
2113 i.types[op] |= smallest_imm_type (i.op[op].imms->X_add_number);
2114
2115 /* We must avoid matching of Imm32 templates when 64bit
2116 only immediate is available. */
2117 if (guess_suffix == QWORD_MNEM_SUFFIX)
2118 i.types[op] &= ~Imm32;
2119 break;
2120
2121 case O_absent:
2122 case O_register:
2123 abort ();
2124
2125 /* Symbols and expressions. */
2126 default:
2127 /* Convert symbolic operand to proper sizes for matching, but don't
2128 prevent matching a set of insns that only supports sizes other
2129 than those matching the insn suffix. */
2130 {
2131 unsigned int mask, allowed = 0;
2132 const template *t;
2133
2134 for (t = current_templates->start; t < current_templates->end; ++t)
2135 allowed |= t->operand_types[op];
2136 switch (guess_suffix)
2137 {
2138 case QWORD_MNEM_SUFFIX:
2139 mask = Imm64 | Imm32S;
2140 break;
2141 case LONG_MNEM_SUFFIX:
2142 mask = Imm32;
2143 break;
2144 case WORD_MNEM_SUFFIX:
2145 mask = Imm16;
2146 break;
2147 case BYTE_MNEM_SUFFIX:
2148 mask = Imm8;
2149 break;
2150 default:
2151 mask = 0;
2152 break;
2153 }
2154 if (mask & allowed)
2155 i.types[op] &= mask;
2156 }
2157 break;
2158 }
2159 }
2160 }
2161
2162 /* Try to use the smallest displacement type too. */
2163 static void
2164 optimize_disp ()
2165 {
2166 int op;
2167
2168 for (op = i.operands; --op >= 0;)
2169 if (i.types[op] & Disp)
2170 {
2171 if (i.op[op].disps->X_op == O_constant)
2172 {
2173 offsetT disp = i.op[op].disps->X_add_number;
2174
2175 if ((i.types[op] & Disp16)
2176 && (disp & ~(offsetT) 0xffff) == 0)
2177 {
2178 /* If this operand is at most 16 bits, convert
2179 to a signed 16 bit number and don't use 64bit
2180 displacement. */
2181 disp = (((disp & 0xffff) ^ 0x8000) - 0x8000);
2182 i.types[op] &= ~Disp64;
2183 }
2184 if ((i.types[op] & Disp32)
2185 && (disp & ~(((offsetT) 2 << 31) - 1)) == 0)
2186 {
2187 /* If this operand is at most 32 bits, convert
2188 to a signed 32 bit number and don't use 64bit
2189 displacement. */
2190 disp &= (((offsetT) 2 << 31) - 1);
2191 disp = (disp ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
2192 i.types[op] &= ~Disp64;
2193 }
2194 if (!disp && (i.types[op] & BaseIndex))
2195 {
2196 i.types[op] &= ~Disp;
2197 i.op[op].disps = 0;
2198 i.disp_operands--;
2199 }
2200 else if (flag_code == CODE_64BIT)
2201 {
2202 if (fits_in_signed_long (disp))
2203 {
2204 i.types[op] &= ~Disp64;
2205 i.types[op] |= Disp32S;
2206 }
2207 if (fits_in_unsigned_long (disp))
2208 i.types[op] |= Disp32;
2209 }
2210 if ((i.types[op] & (Disp32 | Disp32S | Disp16))
2211 && fits_in_signed_byte (disp))
2212 i.types[op] |= Disp8;
2213 }
2214 else if (i.reloc[op] == BFD_RELOC_386_TLS_DESC_CALL
2215 || i.reloc[op] == BFD_RELOC_X86_64_TLSDESC_CALL)
2216 {
2217 fix_new_exp (frag_now, frag_more (0) - frag_now->fr_literal, 0,
2218 i.op[op].disps, 0, i.reloc[op]);
2219 i.types[op] &= ~Disp;
2220 }
2221 else
2222 /* We only support 64bit displacement on constants. */
2223 i.types[op] &= ~Disp64;
2224 }
2225 }
2226
2227 static int
2228 match_template ()
2229 {
2230 /* Points to template once we've found it. */
2231 const template *t;
2232 unsigned int overlap0, overlap1, overlap2;
2233 unsigned int found_reverse_match;
2234 int suffix_check;
2235
2236 #define MATCH(overlap, given, template) \
2237 ((overlap & ~JumpAbsolute) \
2238 && (((given) & (BaseIndex | JumpAbsolute)) \
2239 == ((overlap) & (BaseIndex | JumpAbsolute))))
2240
2241 /* If given types r0 and r1 are registers they must be of the same type
2242 unless the expected operand type register overlap is null.
2243 Note that Acc in a template matches every size of reg. */
2244 #define CONSISTENT_REGISTER_MATCH(m0, g0, t0, m1, g1, t1) \
2245 (((g0) & Reg) == 0 || ((g1) & Reg) == 0 \
2246 || ((g0) & Reg) == ((g1) & Reg) \
2247 || ((((m0) & Acc) ? Reg : (t0)) & (((m1) & Acc) ? Reg : (t1)) & Reg) == 0 )
2248
2249 overlap0 = 0;
2250 overlap1 = 0;
2251 overlap2 = 0;
2252 found_reverse_match = 0;
2253 suffix_check = (i.suffix == BYTE_MNEM_SUFFIX
2254 ? No_bSuf
2255 : (i.suffix == WORD_MNEM_SUFFIX
2256 ? No_wSuf
2257 : (i.suffix == SHORT_MNEM_SUFFIX
2258 ? No_sSuf
2259 : (i.suffix == LONG_MNEM_SUFFIX
2260 ? No_lSuf
2261 : (i.suffix == QWORD_MNEM_SUFFIX
2262 ? No_qSuf
2263 : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX
2264 ? No_xSuf : 0))))));
2265
2266 for (t = current_templates->start; t < current_templates->end; t++)
2267 {
2268 /* Must have right number of operands. */
2269 if (i.operands != t->operands)
2270 continue;
2271
2272 /* Check the suffix, except for some instructions in intel mode. */
2273 if ((t->opcode_modifier & suffix_check)
2274 && !(intel_syntax
2275 && (t->opcode_modifier & IgnoreSize)))
2276 continue;
2277
2278 /* In general, don't allow 64-bit operands in 32-bit mode. */
2279 if (i.suffix == QWORD_MNEM_SUFFIX
2280 && flag_code != CODE_64BIT
2281 && (intel_syntax
2282 ? (!(t->opcode_modifier & IgnoreSize)
2283 && !intel_float_operand (t->name))
2284 : intel_float_operand (t->name) != 2)
2285 && (!(t->operand_types[0] & (RegMMX | RegXMM))
2286 || !(t->operand_types[t->operands > 1] & (RegMMX | RegXMM)))
2287 && (t->base_opcode != 0x0fc7
2288 || t->extension_opcode != 1 /* cmpxchg8b */))
2289 continue;
2290
2291 /* Do not verify operands when there are none. */
2292 else if (!t->operands)
2293 {
2294 if (t->cpu_flags & ~cpu_arch_flags)
2295 continue;
2296 /* We've found a match; break out of loop. */
2297 break;
2298 }
2299
2300 overlap0 = i.types[0] & t->operand_types[0];
2301 switch (t->operands)
2302 {
2303 case 1:
2304 if (!MATCH (overlap0, i.types[0], t->operand_types[0]))
2305 continue;
2306 break;
2307 case 2:
2308 case 3:
2309 overlap1 = i.types[1] & t->operand_types[1];
2310 if (!MATCH (overlap0, i.types[0], t->operand_types[0])
2311 || !MATCH (overlap1, i.types[1], t->operand_types[1])
2312 /* monitor in SSE3 is a very special case. The first
2313 register and the second register may have differnet
2314 sizes. */
2315 || !((t->base_opcode == 0x0f01
2316 && t->extension_opcode == 0xc8)
2317 || CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2318 t->operand_types[0],
2319 overlap1, i.types[1],
2320 t->operand_types[1])))
2321 {
2322 /* Check if other direction is valid ... */
2323 if ((t->opcode_modifier & (D | FloatD)) == 0)
2324 continue;
2325
2326 /* Try reversing direction of operands. */
2327 overlap0 = i.types[0] & t->operand_types[1];
2328 overlap1 = i.types[1] & t->operand_types[0];
2329 if (!MATCH (overlap0, i.types[0], t->operand_types[1])
2330 || !MATCH (overlap1, i.types[1], t->operand_types[0])
2331 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2332 t->operand_types[1],
2333 overlap1, i.types[1],
2334 t->operand_types[0]))
2335 {
2336 /* Does not match either direction. */
2337 continue;
2338 }
2339 /* found_reverse_match holds which of D or FloatDR
2340 we've found. */
2341 found_reverse_match = t->opcode_modifier & (D | FloatDR);
2342 }
2343 /* Found a forward 2 operand match here. */
2344 else if (t->operands == 3)
2345 {
2346 /* Here we make use of the fact that there are no
2347 reverse match 3 operand instructions, and all 3
2348 operand instructions only need to be checked for
2349 register consistency between operands 2 and 3. */
2350 overlap2 = i.types[2] & t->operand_types[2];
2351 if (!MATCH (overlap2, i.types[2], t->operand_types[2])
2352 || !CONSISTENT_REGISTER_MATCH (overlap1, i.types[1],
2353 t->operand_types[1],
2354 overlap2, i.types[2],
2355 t->operand_types[2]))
2356
2357 continue;
2358 }
2359 /* Found either forward/reverse 2 or 3 operand match here:
2360 slip through to break. */
2361 }
2362 if (t->cpu_flags & ~cpu_arch_flags)
2363 {
2364 found_reverse_match = 0;
2365 continue;
2366 }
2367 /* We've found a match; break out of loop. */
2368 break;
2369 }
2370
2371 if (t == current_templates->end)
2372 {
2373 /* We found no match. */
2374 as_bad (_("suffix or operands invalid for `%s'"),
2375 current_templates->start->name);
2376 return 0;
2377 }
2378
2379 if (!quiet_warnings)
2380 {
2381 if (!intel_syntax
2382 && ((i.types[0] & JumpAbsolute)
2383 != (t->operand_types[0] & JumpAbsolute)))
2384 {
2385 as_warn (_("indirect %s without `*'"), t->name);
2386 }
2387
2388 if ((t->opcode_modifier & (IsPrefix | IgnoreSize))
2389 == (IsPrefix | IgnoreSize))
2390 {
2391 /* Warn them that a data or address size prefix doesn't
2392 affect assembly of the next line of code. */
2393 as_warn (_("stand-alone `%s' prefix"), t->name);
2394 }
2395 }
2396
2397 /* Copy the template we found. */
2398 i.tm = *t;
2399 if (found_reverse_match)
2400 {
2401 /* If we found a reverse match we must alter the opcode
2402 direction bit. found_reverse_match holds bits to change
2403 (different for int & float insns). */
2404
2405 i.tm.base_opcode ^= found_reverse_match;
2406
2407 i.tm.operand_types[0] = t->operand_types[1];
2408 i.tm.operand_types[1] = t->operand_types[0];
2409 }
2410
2411 return 1;
2412 }
2413
2414 static int
2415 check_string ()
2416 {
2417 int mem_op = (i.types[0] & AnyMem) ? 0 : 1;
2418 if ((i.tm.operand_types[mem_op] & EsSeg) != 0)
2419 {
2420 if (i.seg[0] != NULL && i.seg[0] != &es)
2421 {
2422 as_bad (_("`%s' operand %d must use `%%es' segment"),
2423 i.tm.name,
2424 mem_op + 1);
2425 return 0;
2426 }
2427 /* There's only ever one segment override allowed per instruction.
2428 This instruction possibly has a legal segment override on the
2429 second operand, so copy the segment to where non-string
2430 instructions store it, allowing common code. */
2431 i.seg[0] = i.seg[1];
2432 }
2433 else if ((i.tm.operand_types[mem_op + 1] & EsSeg) != 0)
2434 {
2435 if (i.seg[1] != NULL && i.seg[1] != &es)
2436 {
2437 as_bad (_("`%s' operand %d must use `%%es' segment"),
2438 i.tm.name,
2439 mem_op + 2);
2440 return 0;
2441 }
2442 }
2443 return 1;
2444 }
2445
2446 static int
2447 process_suffix (void)
2448 {
2449 /* If matched instruction specifies an explicit instruction mnemonic
2450 suffix, use it. */
2451 if (i.tm.opcode_modifier & (Size16 | Size32 | Size64))
2452 {
2453 if (i.tm.opcode_modifier & Size16)
2454 i.suffix = WORD_MNEM_SUFFIX;
2455 else if (i.tm.opcode_modifier & Size64)
2456 i.suffix = QWORD_MNEM_SUFFIX;
2457 else
2458 i.suffix = LONG_MNEM_SUFFIX;
2459 }
2460 else if (i.reg_operands)
2461 {
2462 /* If there's no instruction mnemonic suffix we try to invent one
2463 based on register operands. */
2464 if (!i.suffix)
2465 {
2466 /* We take i.suffix from the last register operand specified,
2467 Destination register type is more significant than source
2468 register type. */
2469 int op;
2470
2471 for (op = i.operands; --op >= 0;)
2472 if ((i.types[op] & Reg)
2473 && !(i.tm.operand_types[op] & InOutPortReg))
2474 {
2475 i.suffix = ((i.types[op] & Reg8) ? BYTE_MNEM_SUFFIX :
2476 (i.types[op] & Reg16) ? WORD_MNEM_SUFFIX :
2477 (i.types[op] & Reg64) ? QWORD_MNEM_SUFFIX :
2478 LONG_MNEM_SUFFIX);
2479 break;
2480 }
2481 }
2482 else if (i.suffix == BYTE_MNEM_SUFFIX)
2483 {
2484 if (!check_byte_reg ())
2485 return 0;
2486 }
2487 else if (i.suffix == LONG_MNEM_SUFFIX)
2488 {
2489 if (!check_long_reg ())
2490 return 0;
2491 }
2492 else if (i.suffix == QWORD_MNEM_SUFFIX)
2493 {
2494 if (!check_qword_reg ())
2495 return 0;
2496 }
2497 else if (i.suffix == WORD_MNEM_SUFFIX)
2498 {
2499 if (!check_word_reg ())
2500 return 0;
2501 }
2502 else if (intel_syntax && (i.tm.opcode_modifier & IgnoreSize))
2503 /* Do nothing if the instruction is going to ignore the prefix. */
2504 ;
2505 else
2506 abort ();
2507 }
2508 else if ((i.tm.opcode_modifier & DefaultSize)
2509 && !i.suffix
2510 /* exclude fldenv/frstor/fsave/fstenv */
2511 && (i.tm.opcode_modifier & No_sSuf))
2512 {
2513 i.suffix = stackop_size;
2514 }
2515 else if (intel_syntax
2516 && !i.suffix
2517 && ((i.tm.operand_types[0] & JumpAbsolute)
2518 || (i.tm.opcode_modifier & (JumpByte|JumpInterSegment))
2519 || (i.tm.base_opcode == 0x0f01 /* [ls][gi]dt */
2520 && i.tm.extension_opcode <= 3)))
2521 {
2522 switch (flag_code)
2523 {
2524 case CODE_64BIT:
2525 if (!(i.tm.opcode_modifier & No_qSuf))
2526 {
2527 i.suffix = QWORD_MNEM_SUFFIX;
2528 break;
2529 }
2530 case CODE_32BIT:
2531 if (!(i.tm.opcode_modifier & No_lSuf))
2532 i.suffix = LONG_MNEM_SUFFIX;
2533 break;
2534 case CODE_16BIT:
2535 if (!(i.tm.opcode_modifier & No_wSuf))
2536 i.suffix = WORD_MNEM_SUFFIX;
2537 break;
2538 }
2539 }
2540
2541 if (!i.suffix)
2542 {
2543 if (!intel_syntax)
2544 {
2545 if (i.tm.opcode_modifier & W)
2546 {
2547 as_bad (_("no instruction mnemonic suffix given and no register operands; can't size instruction"));
2548 return 0;
2549 }
2550 }
2551 else
2552 {
2553 unsigned int suffixes = (~i.tm.opcode_modifier
2554 & (No_bSuf
2555 | No_wSuf
2556 | No_lSuf
2557 | No_sSuf
2558 | No_xSuf
2559 | No_qSuf));
2560
2561 if ((i.tm.opcode_modifier & W)
2562 || ((suffixes & (suffixes - 1))
2563 && !(i.tm.opcode_modifier & (DefaultSize | IgnoreSize))))
2564 {
2565 as_bad (_("ambiguous operand size for `%s'"), i.tm.name);
2566 return 0;
2567 }
2568 }
2569 }
2570
2571 /* Change the opcode based on the operand size given by i.suffix;
2572 We don't need to change things for byte insns. */
2573
2574 if (i.suffix && i.suffix != BYTE_MNEM_SUFFIX)
2575 {
2576 /* It's not a byte, select word/dword operation. */
2577 if (i.tm.opcode_modifier & W)
2578 {
2579 if (i.tm.opcode_modifier & ShortForm)
2580 i.tm.base_opcode |= 8;
2581 else
2582 i.tm.base_opcode |= 1;
2583 }
2584
2585 /* Now select between word & dword operations via the operand
2586 size prefix, except for instructions that will ignore this
2587 prefix anyway. */
2588 if (i.tm.base_opcode == 0x0f01 && i.tm.extension_opcode == 0xc8)
2589 {
2590 /* monitor in SSE3 is a very special case. The default size
2591 of AX is the size of mode. The address size override
2592 prefix will change the size of AX. */
2593 if (i.op->regs[0].reg_type &
2594 (flag_code == CODE_32BIT ? Reg16 : Reg32))
2595 if (!add_prefix (ADDR_PREFIX_OPCODE))
2596 return 0;
2597 }
2598 else if (i.suffix != QWORD_MNEM_SUFFIX
2599 && i.suffix != LONG_DOUBLE_MNEM_SUFFIX
2600 && !(i.tm.opcode_modifier & (IgnoreSize | FloatMF))
2601 && ((i.suffix == LONG_MNEM_SUFFIX) == (flag_code == CODE_16BIT)
2602 || (flag_code == CODE_64BIT
2603 && (i.tm.opcode_modifier & JumpByte))))
2604 {
2605 unsigned int prefix = DATA_PREFIX_OPCODE;
2606
2607 if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */
2608 prefix = ADDR_PREFIX_OPCODE;
2609
2610 if (!add_prefix (prefix))
2611 return 0;
2612 }
2613
2614 /* Set mode64 for an operand. */
2615 if (i.suffix == QWORD_MNEM_SUFFIX
2616 && flag_code == CODE_64BIT
2617 && (i.tm.opcode_modifier & NoRex64) == 0)
2618 i.rex |= REX_MODE64;
2619
2620 /* Size floating point instruction. */
2621 if (i.suffix == LONG_MNEM_SUFFIX)
2622 if (i.tm.opcode_modifier & FloatMF)
2623 i.tm.base_opcode ^= 4;
2624 }
2625
2626 return 1;
2627 }
2628
2629 static int
2630 check_byte_reg (void)
2631 {
2632 int op;
2633
2634 for (op = i.operands; --op >= 0;)
2635 {
2636 /* If this is an eight bit register, it's OK. If it's the 16 or
2637 32 bit version of an eight bit register, we will just use the
2638 low portion, and that's OK too. */
2639 if (i.types[op] & Reg8)
2640 continue;
2641
2642 /* movzx and movsx should not generate this warning. */
2643 if (intel_syntax
2644 && (i.tm.base_opcode == 0xfb7
2645 || i.tm.base_opcode == 0xfb6
2646 || i.tm.base_opcode == 0x63
2647 || i.tm.base_opcode == 0xfbe
2648 || i.tm.base_opcode == 0xfbf))
2649 continue;
2650
2651 if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4)
2652 {
2653 /* Prohibit these changes in the 64bit mode, since the
2654 lowering is more complicated. */
2655 if (flag_code == CODE_64BIT
2656 && (i.tm.operand_types[op] & InOutPortReg) == 0)
2657 {
2658 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2659 i.op[op].regs->reg_name,
2660 i.suffix);
2661 return 0;
2662 }
2663 #if REGISTER_WARNINGS
2664 if (!quiet_warnings
2665 && (i.tm.operand_types[op] & InOutPortReg) == 0)
2666 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2667 (i.op[op].regs + (i.types[op] & Reg16
2668 ? REGNAM_AL - REGNAM_AX
2669 : REGNAM_AL - REGNAM_EAX))->reg_name,
2670 i.op[op].regs->reg_name,
2671 i.suffix);
2672 #endif
2673 continue;
2674 }
2675 /* Any other register is bad. */
2676 if (i.types[op] & (Reg | RegMMX | RegXMM
2677 | SReg2 | SReg3
2678 | Control | Debug | Test
2679 | FloatReg | FloatAcc))
2680 {
2681 as_bad (_("`%%%s' not allowed with `%s%c'"),
2682 i.op[op].regs->reg_name,
2683 i.tm.name,
2684 i.suffix);
2685 return 0;
2686 }
2687 }
2688 return 1;
2689 }
2690
2691 static int
2692 check_long_reg ()
2693 {
2694 int op;
2695
2696 for (op = i.operands; --op >= 0;)
2697 /* Reject eight bit registers, except where the template requires
2698 them. (eg. movzb) */
2699 if ((i.types[op] & Reg8) != 0
2700 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2701 {
2702 as_bad (_("`%%%s' not allowed with `%s%c'"),
2703 i.op[op].regs->reg_name,
2704 i.tm.name,
2705 i.suffix);
2706 return 0;
2707 }
2708 /* Warn if the e prefix on a general reg is missing. */
2709 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2710 && (i.types[op] & Reg16) != 0
2711 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2712 {
2713 /* Prohibit these changes in the 64bit mode, since the
2714 lowering is more complicated. */
2715 if (flag_code == CODE_64BIT)
2716 {
2717 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2718 i.op[op].regs->reg_name,
2719 i.suffix);
2720 return 0;
2721 }
2722 #if REGISTER_WARNINGS
2723 else
2724 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2725 (i.op[op].regs + REGNAM_EAX - REGNAM_AX)->reg_name,
2726 i.op[op].regs->reg_name,
2727 i.suffix);
2728 #endif
2729 }
2730 /* Warn if the r prefix on a general reg is missing. */
2731 else if ((i.types[op] & Reg64) != 0
2732 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2733 {
2734 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2735 i.op[op].regs->reg_name,
2736 i.suffix);
2737 return 0;
2738 }
2739 return 1;
2740 }
2741
2742 static int
2743 check_qword_reg ()
2744 {
2745 int op;
2746
2747 for (op = i.operands; --op >= 0; )
2748 /* Reject eight bit registers, except where the template requires
2749 them. (eg. movzb) */
2750 if ((i.types[op] & Reg8) != 0
2751 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2752 {
2753 as_bad (_("`%%%s' not allowed with `%s%c'"),
2754 i.op[op].regs->reg_name,
2755 i.tm.name,
2756 i.suffix);
2757 return 0;
2758 }
2759 /* Warn if the e prefix on a general reg is missing. */
2760 else if (((i.types[op] & Reg16) != 0
2761 || (i.types[op] & Reg32) != 0)
2762 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2763 {
2764 /* Prohibit these changes in the 64bit mode, since the
2765 lowering is more complicated. */
2766 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2767 i.op[op].regs->reg_name,
2768 i.suffix);
2769 return 0;
2770 }
2771 return 1;
2772 }
2773
2774 static int
2775 check_word_reg ()
2776 {
2777 int op;
2778 for (op = i.operands; --op >= 0;)
2779 /* Reject eight bit registers, except where the template requires
2780 them. (eg. movzb) */
2781 if ((i.types[op] & Reg8) != 0
2782 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2783 {
2784 as_bad (_("`%%%s' not allowed with `%s%c'"),
2785 i.op[op].regs->reg_name,
2786 i.tm.name,
2787 i.suffix);
2788 return 0;
2789 }
2790 /* Warn if the e prefix on a general reg is present. */
2791 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2792 && (i.types[op] & Reg32) != 0
2793 && (i.tm.operand_types[op] & (Reg16 | Acc)) != 0)
2794 {
2795 /* Prohibit these changes in the 64bit mode, since the
2796 lowering is more complicated. */
2797 if (flag_code == CODE_64BIT)
2798 {
2799 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2800 i.op[op].regs->reg_name,
2801 i.suffix);
2802 return 0;
2803 }
2804 else
2805 #if REGISTER_WARNINGS
2806 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2807 (i.op[op].regs + REGNAM_AX - REGNAM_EAX)->reg_name,
2808 i.op[op].regs->reg_name,
2809 i.suffix);
2810 #endif
2811 }
2812 return 1;
2813 }
2814
2815 static int
2816 finalize_imm ()
2817 {
2818 unsigned int overlap0, overlap1, overlap2;
2819
2820 overlap0 = i.types[0] & i.tm.operand_types[0];
2821 if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64))
2822 && overlap0 != Imm8 && overlap0 != Imm8S
2823 && overlap0 != Imm16 && overlap0 != Imm32S
2824 && overlap0 != Imm32 && overlap0 != Imm64)
2825 {
2826 if (i.suffix)
2827 {
2828 overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX
2829 ? Imm8 | Imm8S
2830 : (i.suffix == WORD_MNEM_SUFFIX
2831 ? Imm16
2832 : (i.suffix == QWORD_MNEM_SUFFIX
2833 ? Imm64 | Imm32S
2834 : Imm32)));
2835 }
2836 else if (overlap0 == (Imm16 | Imm32S | Imm32)
2837 || overlap0 == (Imm16 | Imm32)
2838 || overlap0 == (Imm16 | Imm32S))
2839 {
2840 overlap0 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
2841 ? Imm16 : Imm32S);
2842 }
2843 if (overlap0 != Imm8 && overlap0 != Imm8S
2844 && overlap0 != Imm16 && overlap0 != Imm32S
2845 && overlap0 != Imm32 && overlap0 != Imm64)
2846 {
2847 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size"));
2848 return 0;
2849 }
2850 }
2851 i.types[0] = overlap0;
2852
2853 overlap1 = i.types[1] & i.tm.operand_types[1];
2854 if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32 | Imm64))
2855 && overlap1 != Imm8 && overlap1 != Imm8S
2856 && overlap1 != Imm16 && overlap1 != Imm32S
2857 && overlap1 != Imm32 && overlap1 != Imm64)
2858 {
2859 if (i.suffix)
2860 {
2861 overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX
2862 ? Imm8 | Imm8S
2863 : (i.suffix == WORD_MNEM_SUFFIX
2864 ? Imm16
2865 : (i.suffix == QWORD_MNEM_SUFFIX
2866 ? Imm64 | Imm32S
2867 : Imm32)));
2868 }
2869 else if (overlap1 == (Imm16 | Imm32 | Imm32S)
2870 || overlap1 == (Imm16 | Imm32)
2871 || overlap1 == (Imm16 | Imm32S))
2872 {
2873 overlap1 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
2874 ? Imm16 : Imm32S);
2875 }
2876 if (overlap1 != Imm8 && overlap1 != Imm8S
2877 && overlap1 != Imm16 && overlap1 != Imm32S
2878 && overlap1 != Imm32 && overlap1 != Imm64)
2879 {
2880 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size %x %c"),overlap1, i.suffix);
2881 return 0;
2882 }
2883 }
2884 i.types[1] = overlap1;
2885
2886 overlap2 = i.types[2] & i.tm.operand_types[2];
2887 assert ((overlap2 & Imm) == 0);
2888 i.types[2] = overlap2;
2889
2890 return 1;
2891 }
2892
2893 static int
2894 process_operands ()
2895 {
2896 /* Default segment register this instruction will use for memory
2897 accesses. 0 means unknown. This is only for optimizing out
2898 unnecessary segment overrides. */
2899 const seg_entry *default_seg = 0;
2900
2901 /* The imul $imm, %reg instruction is converted into
2902 imul $imm, %reg, %reg, and the clr %reg instruction
2903 is converted into xor %reg, %reg. */
2904 if (i.tm.opcode_modifier & regKludge)
2905 {
2906 unsigned int first_reg_op = (i.types[0] & Reg) ? 0 : 1;
2907 /* Pretend we saw the extra register operand. */
2908 assert (i.op[first_reg_op + 1].regs == 0);
2909 i.op[first_reg_op + 1].regs = i.op[first_reg_op].regs;
2910 i.types[first_reg_op + 1] = i.types[first_reg_op];
2911 i.reg_operands = 2;
2912 }
2913
2914 if (i.tm.opcode_modifier & ShortForm)
2915 {
2916 /* The register or float register operand is in operand 0 or 1. */
2917 unsigned int op = (i.types[0] & (Reg | FloatReg)) ? 0 : 1;
2918 /* Register goes in low 3 bits of opcode. */
2919 i.tm.base_opcode |= i.op[op].regs->reg_num;
2920 if ((i.op[op].regs->reg_flags & RegRex) != 0)
2921 i.rex |= REX_EXTZ;
2922 if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
2923 {
2924 /* Warn about some common errors, but press on regardless.
2925 The first case can be generated by gcc (<= 2.8.1). */
2926 if (i.operands == 2)
2927 {
2928 /* Reversed arguments on faddp, fsubp, etc. */
2929 as_warn (_("translating to `%s %%%s,%%%s'"), i.tm.name,
2930 i.op[1].regs->reg_name,
2931 i.op[0].regs->reg_name);
2932 }
2933 else
2934 {
2935 /* Extraneous `l' suffix on fp insn. */
2936 as_warn (_("translating to `%s %%%s'"), i.tm.name,
2937 i.op[0].regs->reg_name);
2938 }
2939 }
2940 }
2941 else if (i.tm.opcode_modifier & Modrm)
2942 {
2943 /* The opcode is completed (modulo i.tm.extension_opcode which
2944 must be put into the modrm byte). Now, we make the modrm and
2945 index base bytes based on all the info we've collected. */
2946
2947 default_seg = build_modrm_byte ();
2948 }
2949 else if (i.tm.opcode_modifier & (Seg2ShortForm | Seg3ShortForm))
2950 {
2951 if (i.tm.base_opcode == POP_SEG_SHORT
2952 && i.op[0].regs->reg_num == 1)
2953 {
2954 as_bad (_("you can't `pop %%cs'"));
2955 return 0;
2956 }
2957 i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
2958 if ((i.op[0].regs->reg_flags & RegRex) != 0)
2959 i.rex |= REX_EXTZ;
2960 }
2961 else if ((i.tm.base_opcode & ~(D | W)) == MOV_AX_DISP32)
2962 {
2963 default_seg = &ds;
2964 }
2965 else if ((i.tm.opcode_modifier & IsString) != 0)
2966 {
2967 /* For the string instructions that allow a segment override
2968 on one of their operands, the default segment is ds. */
2969 default_seg = &ds;
2970 }
2971
2972 if ((i.tm.base_opcode == 0x8d /* lea */
2973 || (i.tm.cpu_flags & CpuSVME))
2974 && i.seg[0] && !quiet_warnings)
2975 as_warn (_("segment override on `%s' is ineffectual"), i.tm.name);
2976
2977 /* If a segment was explicitly specified, and the specified segment
2978 is not the default, use an opcode prefix to select it. If we
2979 never figured out what the default segment is, then default_seg
2980 will be zero at this point, and the specified segment prefix will
2981 always be used. */
2982 if ((i.seg[0]) && (i.seg[0] != default_seg))
2983 {
2984 if (!add_prefix (i.seg[0]->seg_prefix))
2985 return 0;
2986 }
2987 return 1;
2988 }
2989
2990 static const seg_entry *
2991 build_modrm_byte ()
2992 {
2993 const seg_entry *default_seg = 0;
2994
2995 /* i.reg_operands MUST be the number of real register operands;
2996 implicit registers do not count. */
2997 if (i.reg_operands == 2)
2998 {
2999 unsigned int source, dest;
3000 source = ((i.types[0]
3001 & (Reg | RegMMX | RegXMM
3002 | SReg2 | SReg3
3003 | Control | Debug | Test))
3004 ? 0 : 1);
3005 dest = source + 1;
3006
3007 i.rm.mode = 3;
3008 /* One of the register operands will be encoded in the i.tm.reg
3009 field, the other in the combined i.tm.mode and i.tm.regmem
3010 fields. If no form of this instruction supports a memory
3011 destination operand, then we assume the source operand may
3012 sometimes be a memory operand and so we need to store the
3013 destination in the i.rm.reg field. */
3014 if ((i.tm.operand_types[dest] & AnyMem) == 0)
3015 {
3016 i.rm.reg = i.op[dest].regs->reg_num;
3017 i.rm.regmem = i.op[source].regs->reg_num;
3018 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
3019 i.rex |= REX_EXTX;
3020 if ((i.op[source].regs->reg_flags & RegRex) != 0)
3021 i.rex |= REX_EXTZ;
3022 }
3023 else
3024 {
3025 i.rm.reg = i.op[source].regs->reg_num;
3026 i.rm.regmem = i.op[dest].regs->reg_num;
3027 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
3028 i.rex |= REX_EXTZ;
3029 if ((i.op[source].regs->reg_flags & RegRex) != 0)
3030 i.rex |= REX_EXTX;
3031 }
3032 if (flag_code != CODE_64BIT && (i.rex & (REX_EXTX | REX_EXTZ)))
3033 {
3034 if (!((i.types[0] | i.types[1]) & Control))
3035 abort ();
3036 i.rex &= ~(REX_EXTX | REX_EXTZ);
3037 add_prefix (LOCK_PREFIX_OPCODE);
3038 }
3039 }
3040 else
3041 { /* If it's not 2 reg operands... */
3042 if (i.mem_operands)
3043 {
3044 unsigned int fake_zero_displacement = 0;
3045 unsigned int op = ((i.types[0] & AnyMem)
3046 ? 0
3047 : (i.types[1] & AnyMem) ? 1 : 2);
3048
3049 default_seg = &ds;
3050
3051 if (i.base_reg == 0)
3052 {
3053 i.rm.mode = 0;
3054 if (!i.disp_operands)
3055 fake_zero_displacement = 1;
3056 if (i.index_reg == 0)
3057 {
3058 /* Operand is just <disp> */
3059 if (flag_code == CODE_64BIT)
3060 {
3061 /* 64bit mode overwrites the 32bit absolute
3062 addressing by RIP relative addressing and
3063 absolute addressing is encoded by one of the
3064 redundant SIB forms. */
3065 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3066 i.sib.base = NO_BASE_REGISTER;
3067 i.sib.index = NO_INDEX_REGISTER;
3068 i.types[op] = ((i.prefix[ADDR_PREFIX] == 0) ? Disp32S : Disp32);
3069 }
3070 else if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3071 {
3072 i.rm.regmem = NO_BASE_REGISTER_16;
3073 i.types[op] = Disp16;
3074 }
3075 else
3076 {
3077 i.rm.regmem = NO_BASE_REGISTER;
3078 i.types[op] = Disp32;
3079 }
3080 }
3081 else /* !i.base_reg && i.index_reg */
3082 {
3083 i.sib.index = i.index_reg->reg_num;
3084 i.sib.base = NO_BASE_REGISTER;
3085 i.sib.scale = i.log2_scale_factor;
3086 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3087 i.types[op] &= ~Disp;
3088 if (flag_code != CODE_64BIT)
3089 i.types[op] |= Disp32; /* Must be 32 bit */
3090 else
3091 i.types[op] |= Disp32S;
3092 if ((i.index_reg->reg_flags & RegRex) != 0)
3093 i.rex |= REX_EXTY;
3094 }
3095 }
3096 /* RIP addressing for 64bit mode. */
3097 else if (i.base_reg->reg_type == BaseIndex)
3098 {
3099 i.rm.regmem = NO_BASE_REGISTER;
3100 i.types[op] &= ~ Disp;
3101 i.types[op] |= Disp32S;
3102 i.flags[op] = Operand_PCrel;
3103 if (! i.disp_operands)
3104 fake_zero_displacement = 1;
3105 }
3106 else if (i.base_reg->reg_type & Reg16)
3107 {
3108 switch (i.base_reg->reg_num)
3109 {
3110 case 3: /* (%bx) */
3111 if (i.index_reg == 0)
3112 i.rm.regmem = 7;
3113 else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */
3114 i.rm.regmem = i.index_reg->reg_num - 6;
3115 break;
3116 case 5: /* (%bp) */
3117 default_seg = &ss;
3118 if (i.index_reg == 0)
3119 {
3120 i.rm.regmem = 6;
3121 if ((i.types[op] & Disp) == 0)
3122 {
3123 /* fake (%bp) into 0(%bp) */
3124 i.types[op] |= Disp8;
3125 fake_zero_displacement = 1;
3126 }
3127 }
3128 else /* (%bp,%si) -> 2, or (%bp,%di) -> 3 */
3129 i.rm.regmem = i.index_reg->reg_num - 6 + 2;
3130 break;
3131 default: /* (%si) -> 4 or (%di) -> 5 */
3132 i.rm.regmem = i.base_reg->reg_num - 6 + 4;
3133 }
3134 i.rm.mode = mode_from_disp_size (i.types[op]);
3135 }
3136 else /* i.base_reg and 32/64 bit mode */
3137 {
3138 if (flag_code == CODE_64BIT
3139 && (i.types[op] & Disp))
3140 i.types[op] = (i.types[op] & Disp8) | (i.prefix[ADDR_PREFIX] == 0 ? Disp32S : Disp32);
3141
3142 i.rm.regmem = i.base_reg->reg_num;
3143 if ((i.base_reg->reg_flags & RegRex) != 0)
3144 i.rex |= REX_EXTZ;
3145 i.sib.base = i.base_reg->reg_num;
3146 /* x86-64 ignores REX prefix bit here to avoid decoder
3147 complications. */
3148 if ((i.base_reg->reg_num & 7) == EBP_REG_NUM)
3149 {
3150 default_seg = &ss;
3151 if (i.disp_operands == 0)
3152 {
3153 fake_zero_displacement = 1;
3154 i.types[op] |= Disp8;
3155 }
3156 }
3157 else if (i.base_reg->reg_num == ESP_REG_NUM)
3158 {
3159 default_seg = &ss;
3160 }
3161 i.sib.scale = i.log2_scale_factor;
3162 if (i.index_reg == 0)
3163 {
3164 /* <disp>(%esp) becomes two byte modrm with no index
3165 register. We've already stored the code for esp
3166 in i.rm.regmem ie. ESCAPE_TO_TWO_BYTE_ADDRESSING.
3167 Any base register besides %esp will not use the
3168 extra modrm byte. */
3169 i.sib.index = NO_INDEX_REGISTER;
3170 #if !SCALE1_WHEN_NO_INDEX
3171 /* Another case where we force the second modrm byte. */
3172 if (i.log2_scale_factor)
3173 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3174 #endif
3175 }
3176 else
3177 {
3178 i.sib.index = i.index_reg->reg_num;
3179 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3180 if ((i.index_reg->reg_flags & RegRex) != 0)
3181 i.rex |= REX_EXTY;
3182 }
3183
3184 if (i.disp_operands
3185 && (i.reloc[op] == BFD_RELOC_386_TLS_DESC_CALL
3186 || i.reloc[op] == BFD_RELOC_X86_64_TLSDESC_CALL))
3187 i.rm.mode = 0;
3188 else
3189 i.rm.mode = mode_from_disp_size (i.types[op]);
3190 }
3191
3192 if (fake_zero_displacement)
3193 {
3194 /* Fakes a zero displacement assuming that i.types[op]
3195 holds the correct displacement size. */
3196 expressionS *exp;
3197
3198 assert (i.op[op].disps == 0);
3199 exp = &disp_expressions[i.disp_operands++];
3200 i.op[op].disps = exp;
3201 exp->X_op = O_constant;
3202 exp->X_add_number = 0;
3203 exp->X_add_symbol = (symbolS *) 0;
3204 exp->X_op_symbol = (symbolS *) 0;
3205 }
3206 }
3207
3208 /* Fill in i.rm.reg or i.rm.regmem field with register operand
3209 (if any) based on i.tm.extension_opcode. Again, we must be
3210 careful to make sure that segment/control/debug/test/MMX
3211 registers are coded into the i.rm.reg field. */
3212 if (i.reg_operands)
3213 {
3214 unsigned int op =
3215 ((i.types[0]
3216 & (Reg | RegMMX | RegXMM
3217 | SReg2 | SReg3
3218 | Control | Debug | Test))
3219 ? 0
3220 : ((i.types[1]
3221 & (Reg | RegMMX | RegXMM
3222 | SReg2 | SReg3
3223 | Control | Debug | Test))
3224 ? 1
3225 : 2));
3226 /* If there is an extension opcode to put here, the register
3227 number must be put into the regmem field. */
3228 if (i.tm.extension_opcode != None)
3229 {
3230 i.rm.regmem = i.op[op].regs->reg_num;
3231 if ((i.op[op].regs->reg_flags & RegRex) != 0)
3232 i.rex |= REX_EXTZ;
3233 }
3234 else
3235 {
3236 i.rm.reg = i.op[op].regs->reg_num;
3237 if ((i.op[op].regs->reg_flags & RegRex) != 0)
3238 i.rex |= REX_EXTX;
3239 }
3240
3241 /* Now, if no memory operand has set i.rm.mode = 0, 1, 2 we
3242 must set it to 3 to indicate this is a register operand
3243 in the regmem field. */
3244 if (!i.mem_operands)
3245 i.rm.mode = 3;
3246 }
3247
3248 /* Fill in i.rm.reg field with extension opcode (if any). */
3249 if (i.tm.extension_opcode != None)
3250 i.rm.reg = i.tm.extension_opcode;
3251 }
3252 return default_seg;
3253 }
3254
3255 static void
3256 output_branch ()
3257 {
3258 char *p;
3259 int code16;
3260 int prefix;
3261 relax_substateT subtype;
3262 symbolS *sym;
3263 offsetT off;
3264
3265 code16 = 0;
3266 if (flag_code == CODE_16BIT)
3267 code16 = CODE16;
3268
3269 prefix = 0;
3270 if (i.prefix[DATA_PREFIX] != 0)
3271 {
3272 prefix = 1;
3273 i.prefixes -= 1;
3274 code16 ^= CODE16;
3275 }
3276 /* Pentium4 branch hints. */
3277 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
3278 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
3279 {
3280 prefix++;
3281 i.prefixes--;
3282 }
3283 if (i.prefix[REX_PREFIX] != 0)
3284 {
3285 prefix++;
3286 i.prefixes--;
3287 }
3288
3289 if (i.prefixes != 0 && !intel_syntax)
3290 as_warn (_("skipping prefixes on this instruction"));
3291
3292 /* It's always a symbol; End frag & setup for relax.
3293 Make sure there is enough room in this frag for the largest
3294 instruction we may generate in md_convert_frag. This is 2
3295 bytes for the opcode and room for the prefix and largest
3296 displacement. */
3297 frag_grow (prefix + 2 + 4);
3298 /* Prefix and 1 opcode byte go in fr_fix. */
3299 p = frag_more (prefix + 1);
3300 if (i.prefix[DATA_PREFIX] != 0)
3301 *p++ = DATA_PREFIX_OPCODE;
3302 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE
3303 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE)
3304 *p++ = i.prefix[SEG_PREFIX];
3305 if (i.prefix[REX_PREFIX] != 0)
3306 *p++ = i.prefix[REX_PREFIX];
3307 *p = i.tm.base_opcode;
3308
3309 if ((unsigned char) *p == JUMP_PC_RELATIVE)
3310 subtype = ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL);
3311 else if ((cpu_arch_flags & Cpu386) != 0)
3312 subtype = ENCODE_RELAX_STATE (COND_JUMP, SMALL);
3313 else
3314 subtype = ENCODE_RELAX_STATE (COND_JUMP86, SMALL);
3315 subtype |= code16;
3316
3317 sym = i.op[0].disps->X_add_symbol;
3318 off = i.op[0].disps->X_add_number;
3319
3320 if (i.op[0].disps->X_op != O_constant
3321 && i.op[0].disps->X_op != O_symbol)
3322 {
3323 /* Handle complex expressions. */
3324 sym = make_expr_symbol (i.op[0].disps);
3325 off = 0;
3326 }
3327
3328 /* 1 possible extra opcode + 4 byte displacement go in var part.
3329 Pass reloc in fr_var. */
3330 frag_var (rs_machine_dependent, 5, i.reloc[0], subtype, sym, off, p);
3331 }
3332
3333 static void
3334 output_jump ()
3335 {
3336 char *p;
3337 int size;
3338 fixS *fixP;
3339
3340 if (i.tm.opcode_modifier & JumpByte)
3341 {
3342 /* This is a loop or jecxz type instruction. */
3343 size = 1;
3344 if (i.prefix[ADDR_PREFIX] != 0)
3345 {
3346 FRAG_APPEND_1_CHAR (ADDR_PREFIX_OPCODE);
3347 i.prefixes -= 1;
3348 }
3349 /* Pentium4 branch hints. */
3350 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
3351 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
3352 {
3353 FRAG_APPEND_1_CHAR (i.prefix[SEG_PREFIX]);
3354 i.prefixes--;
3355 }
3356 }
3357 else
3358 {
3359 int code16;
3360
3361 code16 = 0;
3362 if (flag_code == CODE_16BIT)
3363 code16 = CODE16;
3364
3365 if (i.prefix[DATA_PREFIX] != 0)
3366 {
3367 FRAG_APPEND_1_CHAR (DATA_PREFIX_OPCODE);
3368 i.prefixes -= 1;
3369 code16 ^= CODE16;
3370 }
3371
3372 size = 4;
3373 if (code16)
3374 size = 2;
3375 }
3376
3377 if (i.prefix[REX_PREFIX] != 0)
3378 {
3379 FRAG_APPEND_1_CHAR (i.prefix[REX_PREFIX]);
3380 i.prefixes -= 1;
3381 }
3382
3383 if (i.prefixes != 0 && !intel_syntax)
3384 as_warn (_("skipping prefixes on this instruction"));
3385
3386 p = frag_more (1 + size);
3387 *p++ = i.tm.base_opcode;
3388
3389 fixP = fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3390 i.op[0].disps, 1, reloc (size, 1, 1, i.reloc[0]));
3391
3392 /* All jumps handled here are signed, but don't use a signed limit
3393 check for 32 and 16 bit jumps as we want to allow wrap around at
3394 4G and 64k respectively. */
3395 if (size == 1)
3396 fixP->fx_signed = 1;
3397 }
3398
3399 static void
3400 output_interseg_jump ()
3401 {
3402 char *p;
3403 int size;
3404 int prefix;
3405 int code16;
3406
3407 code16 = 0;
3408 if (flag_code == CODE_16BIT)
3409 code16 = CODE16;
3410
3411 prefix = 0;
3412 if (i.prefix[DATA_PREFIX] != 0)
3413 {
3414 prefix = 1;
3415 i.prefixes -= 1;
3416 code16 ^= CODE16;
3417 }
3418 if (i.prefix[REX_PREFIX] != 0)
3419 {
3420 prefix++;
3421 i.prefixes -= 1;
3422 }
3423
3424 size = 4;
3425 if (code16)
3426 size = 2;
3427
3428 if (i.prefixes != 0 && !intel_syntax)
3429 as_warn (_("skipping prefixes on this instruction"));
3430
3431 /* 1 opcode; 2 segment; offset */
3432 p = frag_more (prefix + 1 + 2 + size);
3433
3434 if (i.prefix[DATA_PREFIX] != 0)
3435 *p++ = DATA_PREFIX_OPCODE;
3436
3437 if (i.prefix[REX_PREFIX] != 0)
3438 *p++ = i.prefix[REX_PREFIX];
3439
3440 *p++ = i.tm.base_opcode;
3441 if (i.op[1].imms->X_op == O_constant)
3442 {
3443 offsetT n = i.op[1].imms->X_add_number;
3444
3445 if (size == 2
3446 && !fits_in_unsigned_word (n)
3447 && !fits_in_signed_word (n))
3448 {
3449 as_bad (_("16-bit jump out of range"));
3450 return;
3451 }
3452 md_number_to_chars (p, n, size);
3453 }
3454 else
3455 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3456 i.op[1].imms, 0, reloc (size, 0, 0, i.reloc[1]));
3457 if (i.op[0].imms->X_op != O_constant)
3458 as_bad (_("can't handle non absolute segment in `%s'"),
3459 i.tm.name);
3460 md_number_to_chars (p + size, (valueT) i.op[0].imms->X_add_number, 2);
3461 }
3462
3463 static void
3464 output_insn ()
3465 {
3466 fragS *insn_start_frag;
3467 offsetT insn_start_off;
3468
3469 /* Tie dwarf2 debug info to the address at the start of the insn.
3470 We can't do this after the insn has been output as the current
3471 frag may have been closed off. eg. by frag_var. */
3472 dwarf2_emit_insn (0);
3473
3474 insn_start_frag = frag_now;
3475 insn_start_off = frag_now_fix ();
3476
3477 /* Output jumps. */
3478 if (i.tm.opcode_modifier & Jump)
3479 output_branch ();
3480 else if (i.tm.opcode_modifier & (JumpByte | JumpDword))
3481 output_jump ();
3482 else if (i.tm.opcode_modifier & JumpInterSegment)
3483 output_interseg_jump ();
3484 else
3485 {
3486 /* Output normal instructions here. */
3487 char *p;
3488 unsigned char *q;
3489 unsigned int prefix;
3490
3491 /* All opcodes on i386 have either 1 or 2 bytes. Merom New
3492 Instructions have 3 bytes. We may use one more higher byte
3493 to specify a prefix the instruction requires. */
3494 if ((i.tm.cpu_flags & CpuMNI) != 0)
3495 {
3496 if (i.tm.base_opcode & 0xff000000)
3497 {
3498 prefix = (i.tm.base_opcode >> 24) & 0xff;
3499 goto check_prefix;
3500 }
3501 }
3502 else if ((i.tm.base_opcode & 0xff0000) != 0)
3503 {
3504 prefix = (i.tm.base_opcode >> 16) & 0xff;
3505 if ((i.tm.cpu_flags & CpuPadLock) != 0)
3506 {
3507 check_prefix:
3508 if (prefix != REPE_PREFIX_OPCODE
3509 || i.prefix[LOCKREP_PREFIX] != REPE_PREFIX_OPCODE)
3510 add_prefix (prefix);
3511 }
3512 else
3513 add_prefix (prefix);
3514 }
3515
3516 /* The prefix bytes. */
3517 for (q = i.prefix;
3518 q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
3519 q++)
3520 {
3521 if (*q)
3522 {
3523 p = frag_more (1);
3524 md_number_to_chars (p, (valueT) *q, 1);
3525 }
3526 }
3527
3528 /* Now the opcode; be careful about word order here! */
3529 if (fits_in_unsigned_byte (i.tm.base_opcode))
3530 {
3531 FRAG_APPEND_1_CHAR (i.tm.base_opcode);
3532 }
3533 else
3534 {
3535 if ((i.tm.cpu_flags & CpuMNI) != 0)
3536 {
3537 p = frag_more (3);
3538 *p++ = (i.tm.base_opcode >> 16) & 0xff;
3539 }
3540 else
3541 p = frag_more (2);
3542
3543 /* Put out high byte first: can't use md_number_to_chars! */
3544 *p++ = (i.tm.base_opcode >> 8) & 0xff;
3545 *p = i.tm.base_opcode & 0xff;
3546 }
3547
3548 /* Now the modrm byte and sib byte (if present). */
3549 if (i.tm.opcode_modifier & Modrm)
3550 {
3551 p = frag_more (1);
3552 md_number_to_chars (p,
3553 (valueT) (i.rm.regmem << 0
3554 | i.rm.reg << 3
3555 | i.rm.mode << 6),
3556 1);
3557 /* If i.rm.regmem == ESP (4)
3558 && i.rm.mode != (Register mode)
3559 && not 16 bit
3560 ==> need second modrm byte. */
3561 if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING
3562 && i.rm.mode != 3
3563 && !(i.base_reg && (i.base_reg->reg_type & Reg16) != 0))
3564 {
3565 p = frag_more (1);
3566 md_number_to_chars (p,
3567 (valueT) (i.sib.base << 0
3568 | i.sib.index << 3
3569 | i.sib.scale << 6),
3570 1);
3571 }
3572 }
3573
3574 if (i.disp_operands)
3575 output_disp (insn_start_frag, insn_start_off);
3576
3577 if (i.imm_operands)
3578 output_imm (insn_start_frag, insn_start_off);
3579 }
3580
3581 #ifdef DEBUG386
3582 if (flag_debug)
3583 {
3584 pi ("" /*line*/, &i);
3585 }
3586 #endif /* DEBUG386 */
3587 }
3588
3589 static void
3590 output_disp (fragS *insn_start_frag, offsetT insn_start_off)
3591 {
3592 char *p;
3593 unsigned int n;
3594
3595 for (n = 0; n < i.operands; n++)
3596 {
3597 if (i.types[n] & Disp)
3598 {
3599 if (i.op[n].disps->X_op == O_constant)
3600 {
3601 int size;
3602 offsetT val;
3603
3604 size = 4;
3605 if (i.types[n] & (Disp8 | Disp16 | Disp64))
3606 {
3607 size = 2;
3608 if (i.types[n] & Disp8)
3609 size = 1;
3610 if (i.types[n] & Disp64)
3611 size = 8;
3612 }
3613 val = offset_in_range (i.op[n].disps->X_add_number,
3614 size);
3615 p = frag_more (size);
3616 md_number_to_chars (p, val, size);
3617 }
3618 else
3619 {
3620 enum bfd_reloc_code_real reloc_type;
3621 int size = 4;
3622 int sign = 0;
3623 int pcrel = (i.flags[n] & Operand_PCrel) != 0;
3624
3625 /* The PC relative address is computed relative
3626 to the instruction boundary, so in case immediate
3627 fields follows, we need to adjust the value. */
3628 if (pcrel && i.imm_operands)
3629 {
3630 int imm_size = 4;
3631 unsigned int n1;
3632
3633 for (n1 = 0; n1 < i.operands; n1++)
3634 if (i.types[n1] & Imm)
3635 {
3636 if (i.types[n1] & (Imm8 | Imm8S | Imm16 | Imm64))
3637 {
3638 imm_size = 2;
3639 if (i.types[n1] & (Imm8 | Imm8S))
3640 imm_size = 1;
3641 if (i.types[n1] & Imm64)
3642 imm_size = 8;
3643 }
3644 break;
3645 }
3646 /* We should find the immediate. */
3647 if (n1 == i.operands)
3648 abort ();
3649 i.op[n].disps->X_add_number -= imm_size;
3650 }
3651
3652 if (i.types[n] & Disp32S)
3653 sign = 1;
3654
3655 if (i.types[n] & (Disp16 | Disp64))
3656 {
3657 size = 2;
3658 if (i.types[n] & Disp64)
3659 size = 8;
3660 }
3661
3662 p = frag_more (size);
3663 reloc_type = reloc (size, pcrel, sign, i.reloc[n]);
3664 if (GOT_symbol
3665 && GOT_symbol == i.op[n].disps->X_add_symbol
3666 && (((reloc_type == BFD_RELOC_32
3667 || reloc_type == BFD_RELOC_X86_64_32S
3668 || (reloc_type == BFD_RELOC_64
3669 && object_64bit))
3670 && (i.op[n].disps->X_op == O_symbol
3671 || (i.op[n].disps->X_op == O_add
3672 && ((symbol_get_value_expression
3673 (i.op[n].disps->X_op_symbol)->X_op)
3674 == O_subtract))))
3675 || reloc_type == BFD_RELOC_32_PCREL))
3676 {
3677 offsetT add;
3678
3679 if (insn_start_frag == frag_now)
3680 add = (p - frag_now->fr_literal) - insn_start_off;
3681 else
3682 {
3683 fragS *fr;
3684
3685 add = insn_start_frag->fr_fix - insn_start_off;
3686 for (fr = insn_start_frag->fr_next;
3687 fr && fr != frag_now; fr = fr->fr_next)
3688 add += fr->fr_fix;
3689 add += p - frag_now->fr_literal;
3690 }
3691
3692 if (!object_64bit)
3693 {
3694 reloc_type = BFD_RELOC_386_GOTPC;
3695 i.op[n].imms->X_add_number += add;
3696 }
3697 else if (reloc_type == BFD_RELOC_64)
3698 reloc_type = BFD_RELOC_X86_64_GOTPC64;
3699 else
3700 /* Don't do the adjustment for x86-64, as there
3701 the pcrel addressing is relative to the _next_
3702 insn, and that is taken care of in other code. */
3703 reloc_type = BFD_RELOC_X86_64_GOTPC32;
3704 }
3705 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3706 i.op[n].disps, pcrel, reloc_type);
3707 }
3708 }
3709 }
3710 }
3711
3712 static void
3713 output_imm (fragS *insn_start_frag, offsetT insn_start_off)
3714 {
3715 char *p;
3716 unsigned int n;
3717
3718 for (n = 0; n < i.operands; n++)
3719 {
3720 if (i.types[n] & Imm)
3721 {
3722 if (i.op[n].imms->X_op == O_constant)
3723 {
3724 int size;
3725 offsetT val;
3726
3727 size = 4;
3728 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3729 {
3730 size = 2;
3731 if (i.types[n] & (Imm8 | Imm8S))
3732 size = 1;
3733 else if (i.types[n] & Imm64)
3734 size = 8;
3735 }
3736 val = offset_in_range (i.op[n].imms->X_add_number,
3737 size);
3738 p = frag_more (size);
3739 md_number_to_chars (p, val, size);
3740 }
3741 else
3742 {
3743 /* Not absolute_section.
3744 Need a 32-bit fixup (don't support 8bit
3745 non-absolute imms). Try to support other
3746 sizes ... */
3747 enum bfd_reloc_code_real reloc_type;
3748 int size = 4;
3749 int sign = 0;
3750
3751 if ((i.types[n] & (Imm32S))
3752 && (i.suffix == QWORD_MNEM_SUFFIX
3753 || (!i.suffix && (i.tm.opcode_modifier & No_lSuf))))
3754 sign = 1;
3755 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3756 {
3757 size = 2;
3758 if (i.types[n] & (Imm8 | Imm8S))
3759 size = 1;
3760 if (i.types[n] & Imm64)
3761 size = 8;
3762 }
3763
3764 p = frag_more (size);
3765 reloc_type = reloc (size, 0, sign, i.reloc[n]);
3766
3767 /* This is tough to explain. We end up with this one if we
3768 * have operands that look like
3769 * "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal here is to
3770 * obtain the absolute address of the GOT, and it is strongly
3771 * preferable from a performance point of view to avoid using
3772 * a runtime relocation for this. The actual sequence of
3773 * instructions often look something like:
3774 *
3775 * call .L66
3776 * .L66:
3777 * popl %ebx
3778 * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
3779 *
3780 * The call and pop essentially return the absolute address
3781 * of the label .L66 and store it in %ebx. The linker itself
3782 * will ultimately change the first operand of the addl so
3783 * that %ebx points to the GOT, but to keep things simple, the
3784 * .o file must have this operand set so that it generates not
3785 * the absolute address of .L66, but the absolute address of
3786 * itself. This allows the linker itself simply treat a GOTPC
3787 * relocation as asking for a pcrel offset to the GOT to be
3788 * added in, and the addend of the relocation is stored in the
3789 * operand field for the instruction itself.
3790 *
3791 * Our job here is to fix the operand so that it would add
3792 * the correct offset so that %ebx would point to itself. The
3793 * thing that is tricky is that .-.L66 will point to the
3794 * beginning of the instruction, so we need to further modify
3795 * the operand so that it will point to itself. There are
3796 * other cases where you have something like:
3797 *
3798 * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
3799 *
3800 * and here no correction would be required. Internally in
3801 * the assembler we treat operands of this form as not being
3802 * pcrel since the '.' is explicitly mentioned, and I wonder
3803 * whether it would simplify matters to do it this way. Who
3804 * knows. In earlier versions of the PIC patches, the
3805 * pcrel_adjust field was used to store the correction, but
3806 * since the expression is not pcrel, I felt it would be
3807 * confusing to do it this way. */
3808
3809 if ((reloc_type == BFD_RELOC_32
3810 || reloc_type == BFD_RELOC_X86_64_32S
3811 || reloc_type == BFD_RELOC_64)
3812 && GOT_symbol
3813 && GOT_symbol == i.op[n].imms->X_add_symbol
3814 && (i.op[n].imms->X_op == O_symbol
3815 || (i.op[n].imms->X_op == O_add
3816 && ((symbol_get_value_expression
3817 (i.op[n].imms->X_op_symbol)->X_op)
3818 == O_subtract))))
3819 {
3820 offsetT add;
3821
3822 if (insn_start_frag == frag_now)
3823 add = (p - frag_now->fr_literal) - insn_start_off;
3824 else
3825 {
3826 fragS *fr;
3827
3828 add = insn_start_frag->fr_fix - insn_start_off;
3829 for (fr = insn_start_frag->fr_next;
3830 fr && fr != frag_now; fr = fr->fr_next)
3831 add += fr->fr_fix;
3832 add += p - frag_now->fr_literal;
3833 }
3834
3835 if (!object_64bit)
3836 reloc_type = BFD_RELOC_386_GOTPC;
3837 else if (size == 4)
3838 reloc_type = BFD_RELOC_X86_64_GOTPC32;
3839 else if (size == 8)
3840 reloc_type = BFD_RELOC_X86_64_GOTPC64;
3841 i.op[n].imms->X_add_number += add;
3842 }
3843 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3844 i.op[n].imms, 0, reloc_type);
3845 }
3846 }
3847 }
3848 }
3849 \f
3850 /* x86_cons_fix_new is called via the expression parsing code when a
3851 reloc is needed. We use this hook to get the correct .got reloc. */
3852 static enum bfd_reloc_code_real got_reloc = NO_RELOC;
3853 static int cons_sign = -1;
3854
3855 void
3856 x86_cons_fix_new (fragS *frag,
3857 unsigned int off,
3858 unsigned int len,
3859 expressionS *exp)
3860 {
3861 enum bfd_reloc_code_real r = reloc (len, 0, cons_sign, got_reloc);
3862
3863 got_reloc = NO_RELOC;
3864
3865 #ifdef TE_PE
3866 if (exp->X_op == O_secrel)
3867 {
3868 exp->X_op = O_symbol;
3869 r = BFD_RELOC_32_SECREL;
3870 }
3871 #endif
3872
3873 fix_new_exp (frag, off, len, exp, 0, r);
3874 }
3875
3876 #if (!defined (OBJ_ELF) && !defined (OBJ_MAYBE_ELF)) || defined (LEX_AT)
3877 # define lex_got(reloc, adjust, types) NULL
3878 #else
3879 /* Parse operands of the form
3880 <symbol>@GOTOFF+<nnn>
3881 and similar .plt or .got references.
3882
3883 If we find one, set up the correct relocation in RELOC and copy the
3884 input string, minus the `@GOTOFF' into a malloc'd buffer for
3885 parsing by the calling routine. Return this buffer, and if ADJUST
3886 is non-null set it to the length of the string we removed from the
3887 input line. Otherwise return NULL. */
3888 static char *
3889 lex_got (enum bfd_reloc_code_real *reloc,
3890 int *adjust,
3891 unsigned int *types)
3892 {
3893 /* Some of the relocations depend on the size of what field is to
3894 be relocated. But in our callers i386_immediate and i386_displacement
3895 we don't yet know the operand size (this will be set by insn
3896 matching). Hence we record the word32 relocation here,
3897 and adjust the reloc according to the real size in reloc(). */
3898 static const struct {
3899 const char *str;
3900 const enum bfd_reloc_code_real rel[2];
3901 const unsigned int types64;
3902 } gotrel[] = {
3903 { "PLTOFF", { 0, BFD_RELOC_X86_64_PLTOFF64 }, Imm64 },
3904 { "PLT", { BFD_RELOC_386_PLT32, BFD_RELOC_X86_64_PLT32 }, Imm32|Imm32S|Disp32 },
3905 { "GOTPLT", { 0, BFD_RELOC_X86_64_GOTPLT64 }, Imm64|Disp64 },
3906 { "GOTOFF", { BFD_RELOC_386_GOTOFF, BFD_RELOC_X86_64_GOTOFF64 }, Imm64|Disp64 },
3907 { "GOTPCREL", { 0, BFD_RELOC_X86_64_GOTPCREL }, Imm32|Imm32S|Disp32 },
3908 { "TLSGD", { BFD_RELOC_386_TLS_GD, BFD_RELOC_X86_64_TLSGD }, Imm32|Imm32S|Disp32 },
3909 { "TLSLDM", { BFD_RELOC_386_TLS_LDM, 0 }, 0 },
3910 { "TLSLD", { 0, BFD_RELOC_X86_64_TLSLD }, Imm32|Imm32S|Disp32 },
3911 { "GOTTPOFF", { BFD_RELOC_386_TLS_IE_32, BFD_RELOC_X86_64_GOTTPOFF }, Imm32|Imm32S|Disp32 },
3912 { "TPOFF", { BFD_RELOC_386_TLS_LE_32, BFD_RELOC_X86_64_TPOFF32 }, Imm32|Imm32S|Imm64|Disp32|Disp64 },
3913 { "NTPOFF", { BFD_RELOC_386_TLS_LE, 0 }, 0 },
3914 { "DTPOFF", { BFD_RELOC_386_TLS_LDO_32, BFD_RELOC_X86_64_DTPOFF32 }, Imm32|Imm32S|Imm64|Disp32|Disp64 },
3915 { "GOTNTPOFF",{ BFD_RELOC_386_TLS_GOTIE, 0 }, 0 },
3916 { "INDNTPOFF",{ BFD_RELOC_386_TLS_IE, 0 }, 0 },
3917 { "GOT", { BFD_RELOC_386_GOT32, BFD_RELOC_X86_64_GOT32 }, Imm32|Imm32S|Disp32|Imm64 },
3918 { "TLSDESC", { BFD_RELOC_386_TLS_GOTDESC, BFD_RELOC_X86_64_GOTPC32_TLSDESC }, Imm32|Imm32S|Disp32 },
3919 { "TLSCALL", { BFD_RELOC_386_TLS_DESC_CALL, BFD_RELOC_X86_64_TLSDESC_CALL }, Imm32|Imm32S|Disp32 }
3920 };
3921 char *cp;
3922 unsigned int j;
3923
3924 if (!IS_ELF)
3925 return NULL;
3926
3927 for (cp = input_line_pointer; *cp != '@'; cp++)
3928 if (is_end_of_line[(unsigned char) *cp])
3929 return NULL;
3930
3931 for (j = 0; j < sizeof (gotrel) / sizeof (gotrel[0]); j++)
3932 {
3933 int len;
3934
3935 len = strlen (gotrel[j].str);
3936 if (strncasecmp (cp + 1, gotrel[j].str, len) == 0)
3937 {
3938 if (gotrel[j].rel[object_64bit] != 0)
3939 {
3940 int first, second;
3941 char *tmpbuf, *past_reloc;
3942
3943 *reloc = gotrel[j].rel[object_64bit];
3944 if (adjust)
3945 *adjust = len;
3946
3947 if (types)
3948 {
3949 if (flag_code != CODE_64BIT)
3950 *types = Imm32|Disp32;
3951 else
3952 *types = gotrel[j].types64;
3953 }
3954
3955 if (GOT_symbol == NULL)
3956 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
3957
3958 /* Replace the relocation token with ' ', so that
3959 errors like foo@GOTOFF1 will be detected. */
3960
3961 /* The length of the first part of our input line. */
3962 first = cp - input_line_pointer;
3963
3964 /* The second part goes from after the reloc token until
3965 (and including) an end_of_line char. Don't use strlen
3966 here as the end_of_line char may not be a NUL. */
3967 past_reloc = cp + 1 + len;
3968 for (cp = past_reloc; !is_end_of_line[(unsigned char) *cp++]; )
3969 ;
3970 second = cp - past_reloc;
3971
3972 /* Allocate and copy string. The trailing NUL shouldn't
3973 be necessary, but be safe. */
3974 tmpbuf = xmalloc (first + second + 2);
3975 memcpy (tmpbuf, input_line_pointer, first);
3976 tmpbuf[first] = ' ';
3977 memcpy (tmpbuf + first + 1, past_reloc, second);
3978 tmpbuf[first + second + 1] = '\0';
3979 return tmpbuf;
3980 }
3981
3982 as_bad (_("@%s reloc is not supported with %d-bit output format"),
3983 gotrel[j].str, 1 << (5 + object_64bit));
3984 return NULL;
3985 }
3986 }
3987
3988 /* Might be a symbol version string. Don't as_bad here. */
3989 return NULL;
3990 }
3991
3992 void
3993 x86_cons (exp, size)
3994 expressionS *exp;
3995 int size;
3996 {
3997 if (size == 4 || (object_64bit && size == 8))
3998 {
3999 /* Handle @GOTOFF and the like in an expression. */
4000 char *save;
4001 char *gotfree_input_line;
4002 int adjust;
4003
4004 save = input_line_pointer;
4005 gotfree_input_line = lex_got (&got_reloc, &adjust, NULL);
4006 if (gotfree_input_line)
4007 input_line_pointer = gotfree_input_line;
4008
4009 expression (exp);
4010
4011 if (gotfree_input_line)
4012 {
4013 /* expression () has merrily parsed up to the end of line,
4014 or a comma - in the wrong buffer. Transfer how far
4015 input_line_pointer has moved to the right buffer. */
4016 input_line_pointer = (save
4017 + (input_line_pointer - gotfree_input_line)
4018 + adjust);
4019 free (gotfree_input_line);
4020 }
4021 }
4022 else
4023 expression (exp);
4024 }
4025 #endif
4026
4027 static void signed_cons (int size)
4028 {
4029 if (flag_code == CODE_64BIT)
4030 cons_sign = 1;
4031 cons (size);
4032 cons_sign = -1;
4033 }
4034
4035 #ifdef TE_PE
4036 static void
4037 pe_directive_secrel (dummy)
4038 int dummy ATTRIBUTE_UNUSED;
4039 {
4040 expressionS exp;
4041
4042 do
4043 {
4044 expression (&exp);
4045 if (exp.X_op == O_symbol)
4046 exp.X_op = O_secrel;
4047
4048 emit_expr (&exp, 4);
4049 }
4050 while (*input_line_pointer++ == ',');
4051
4052 input_line_pointer--;
4053 demand_empty_rest_of_line ();
4054 }
4055 #endif
4056
4057 static int i386_immediate PARAMS ((char *));
4058
4059 static int
4060 i386_immediate (imm_start)
4061 char *imm_start;
4062 {
4063 char *save_input_line_pointer;
4064 char *gotfree_input_line;
4065 segT exp_seg = 0;
4066 expressionS *exp;
4067 unsigned int types = ~0U;
4068
4069 if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
4070 {
4071 as_bad (_("only 1 or 2 immediate operands are allowed"));
4072 return 0;
4073 }
4074
4075 exp = &im_expressions[i.imm_operands++];
4076 i.op[this_operand].imms = exp;
4077
4078 if (is_space_char (*imm_start))
4079 ++imm_start;
4080
4081 save_input_line_pointer = input_line_pointer;
4082 input_line_pointer = imm_start;
4083
4084 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL, &types);
4085 if (gotfree_input_line)
4086 input_line_pointer = gotfree_input_line;
4087
4088 exp_seg = expression (exp);
4089
4090 SKIP_WHITESPACE ();
4091 if (*input_line_pointer)
4092 as_bad (_("junk `%s' after expression"), input_line_pointer);
4093
4094 input_line_pointer = save_input_line_pointer;
4095 if (gotfree_input_line)
4096 free (gotfree_input_line);
4097
4098 if (exp->X_op == O_absent || exp->X_op == O_big)
4099 {
4100 /* Missing or bad expr becomes absolute 0. */
4101 as_bad (_("missing or invalid immediate expression `%s' taken as 0"),
4102 imm_start);
4103 exp->X_op = O_constant;
4104 exp->X_add_number = 0;
4105 exp->X_add_symbol = (symbolS *) 0;
4106 exp->X_op_symbol = (symbolS *) 0;
4107 }
4108 else if (exp->X_op == O_constant)
4109 {
4110 /* Size it properly later. */
4111 i.types[this_operand] |= Imm64;
4112 /* If BFD64, sign extend val. */
4113 if (!use_rela_relocations)
4114 if ((exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0)
4115 exp->X_add_number = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
4116 }
4117 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4118 else if (OUTPUT_FLAVOR == bfd_target_aout_flavour
4119 && exp_seg != absolute_section
4120 && exp_seg != text_section
4121 && exp_seg != data_section
4122 && exp_seg != bss_section
4123 && exp_seg != undefined_section
4124 && !bfd_is_com_section (exp_seg))
4125 {
4126 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
4127 return 0;
4128 }
4129 #endif
4130 else
4131 {
4132 /* This is an address. The size of the address will be
4133 determined later, depending on destination register,
4134 suffix, or the default for the section. */
4135 i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64;
4136 i.types[this_operand] &= types;
4137 }
4138
4139 return 1;
4140 }
4141
4142 static char *i386_scale PARAMS ((char *));
4143
4144 static char *
4145 i386_scale (scale)
4146 char *scale;
4147 {
4148 offsetT val;
4149 char *save = input_line_pointer;
4150
4151 input_line_pointer = scale;
4152 val = get_absolute_expression ();
4153
4154 switch (val)
4155 {
4156 case 1:
4157 i.log2_scale_factor = 0;
4158 break;
4159 case 2:
4160 i.log2_scale_factor = 1;
4161 break;
4162 case 4:
4163 i.log2_scale_factor = 2;
4164 break;
4165 case 8:
4166 i.log2_scale_factor = 3;
4167 break;
4168 default:
4169 {
4170 char sep = *input_line_pointer;
4171
4172 *input_line_pointer = '\0';
4173 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
4174 scale);
4175 *input_line_pointer = sep;
4176 input_line_pointer = save;
4177 return NULL;
4178 }
4179 }
4180 if (i.log2_scale_factor != 0 && i.index_reg == 0)
4181 {
4182 as_warn (_("scale factor of %d without an index register"),
4183 1 << i.log2_scale_factor);
4184 #if SCALE1_WHEN_NO_INDEX
4185 i.log2_scale_factor = 0;
4186 #endif
4187 }
4188 scale = input_line_pointer;
4189 input_line_pointer = save;
4190 return scale;
4191 }
4192
4193 static int i386_displacement PARAMS ((char *, char *));
4194
4195 static int
4196 i386_displacement (disp_start, disp_end)
4197 char *disp_start;
4198 char *disp_end;
4199 {
4200 expressionS *exp;
4201 segT exp_seg = 0;
4202 char *save_input_line_pointer;
4203 char *gotfree_input_line;
4204 int bigdisp, override;
4205 unsigned int types = Disp;
4206
4207 if ((i.types[this_operand] & JumpAbsolute)
4208 || !(current_templates->start->opcode_modifier & (Jump | JumpDword)))
4209 {
4210 bigdisp = Disp32;
4211 override = (i.prefix[ADDR_PREFIX] != 0);
4212 }
4213 else
4214 {
4215 /* For PC-relative branches, the width of the displacement
4216 is dependent upon data size, not address size. */
4217 bigdisp = 0;
4218 override = (i.prefix[DATA_PREFIX] != 0);
4219 }
4220 if (flag_code == CODE_64BIT)
4221 {
4222 if (!bigdisp)
4223 bigdisp = ((override || i.suffix == WORD_MNEM_SUFFIX)
4224 ? Disp16
4225 : Disp32S | Disp32);
4226 else if (!override)
4227 bigdisp = Disp64 | Disp32S | Disp32;
4228 }
4229 else
4230 {
4231 if (!bigdisp)
4232 {
4233 if (!override)
4234 override = (i.suffix == (flag_code != CODE_16BIT
4235 ? WORD_MNEM_SUFFIX
4236 : LONG_MNEM_SUFFIX));
4237 bigdisp = Disp32;
4238 }
4239 if ((flag_code == CODE_16BIT) ^ override)
4240 bigdisp = Disp16;
4241 }
4242 i.types[this_operand] |= bigdisp;
4243
4244 exp = &disp_expressions[i.disp_operands];
4245 i.op[this_operand].disps = exp;
4246 i.disp_operands++;
4247 save_input_line_pointer = input_line_pointer;
4248 input_line_pointer = disp_start;
4249 END_STRING_AND_SAVE (disp_end);
4250
4251 #ifndef GCC_ASM_O_HACK
4252 #define GCC_ASM_O_HACK 0
4253 #endif
4254 #if GCC_ASM_O_HACK
4255 END_STRING_AND_SAVE (disp_end + 1);
4256 if ((i.types[this_operand] & BaseIndex) != 0
4257 && displacement_string_end[-1] == '+')
4258 {
4259 /* This hack is to avoid a warning when using the "o"
4260 constraint within gcc asm statements.
4261 For instance:
4262
4263 #define _set_tssldt_desc(n,addr,limit,type) \
4264 __asm__ __volatile__ ( \
4265 "movw %w2,%0\n\t" \
4266 "movw %w1,2+%0\n\t" \
4267 "rorl $16,%1\n\t" \
4268 "movb %b1,4+%0\n\t" \
4269 "movb %4,5+%0\n\t" \
4270 "movb $0,6+%0\n\t" \
4271 "movb %h1,7+%0\n\t" \
4272 "rorl $16,%1" \
4273 : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
4274
4275 This works great except that the output assembler ends
4276 up looking a bit weird if it turns out that there is
4277 no offset. You end up producing code that looks like:
4278
4279 #APP
4280 movw $235,(%eax)
4281 movw %dx,2+(%eax)
4282 rorl $16,%edx
4283 movb %dl,4+(%eax)
4284 movb $137,5+(%eax)
4285 movb $0,6+(%eax)
4286 movb %dh,7+(%eax)
4287 rorl $16,%edx
4288 #NO_APP
4289
4290 So here we provide the missing zero. */
4291
4292 *displacement_string_end = '0';
4293 }
4294 #endif
4295 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL, &types);
4296 if (gotfree_input_line)
4297 input_line_pointer = gotfree_input_line;
4298
4299 exp_seg = expression (exp);
4300
4301 SKIP_WHITESPACE ();
4302 if (*input_line_pointer)
4303 as_bad (_("junk `%s' after expression"), input_line_pointer);
4304 #if GCC_ASM_O_HACK
4305 RESTORE_END_STRING (disp_end + 1);
4306 #endif
4307 RESTORE_END_STRING (disp_end);
4308 input_line_pointer = save_input_line_pointer;
4309 if (gotfree_input_line)
4310 free (gotfree_input_line);
4311
4312 /* We do this to make sure that the section symbol is in
4313 the symbol table. We will ultimately change the relocation
4314 to be relative to the beginning of the section. */
4315 if (i.reloc[this_operand] == BFD_RELOC_386_GOTOFF
4316 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
4317 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTOFF64)
4318 {
4319 if (exp->X_op != O_symbol)
4320 {
4321 as_bad (_("bad expression used with @%s"),
4322 (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
4323 ? "GOTPCREL"
4324 : "GOTOFF"));
4325 return 0;
4326 }
4327
4328 if (S_IS_LOCAL (exp->X_add_symbol)
4329 && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
4330 section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
4331 exp->X_op = O_subtract;
4332 exp->X_op_symbol = GOT_symbol;
4333 if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
4334 i.reloc[this_operand] = BFD_RELOC_32_PCREL;
4335 else if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTOFF64)
4336 i.reloc[this_operand] = BFD_RELOC_64;
4337 else
4338 i.reloc[this_operand] = BFD_RELOC_32;
4339 }
4340
4341 if (exp->X_op == O_absent || exp->X_op == O_big)
4342 {
4343 /* Missing or bad expr becomes absolute 0. */
4344 as_bad (_("missing or invalid displacement expression `%s' taken as 0"),
4345 disp_start);
4346 exp->X_op = O_constant;
4347 exp->X_add_number = 0;
4348 exp->X_add_symbol = (symbolS *) 0;
4349 exp->X_op_symbol = (symbolS *) 0;
4350 }
4351
4352 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4353 if (exp->X_op != O_constant
4354 && OUTPUT_FLAVOR == bfd_target_aout_flavour
4355 && exp_seg != absolute_section
4356 && exp_seg != text_section
4357 && exp_seg != data_section
4358 && exp_seg != bss_section
4359 && exp_seg != undefined_section
4360 && !bfd_is_com_section (exp_seg))
4361 {
4362 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
4363 return 0;
4364 }
4365 #endif
4366
4367 if (!(i.types[this_operand] & ~Disp))
4368 i.types[this_operand] &= types;
4369
4370 return 1;
4371 }
4372
4373 static int i386_index_check PARAMS ((const char *));
4374
4375 /* Make sure the memory operand we've been dealt is valid.
4376 Return 1 on success, 0 on a failure. */
4377
4378 static int
4379 i386_index_check (operand_string)
4380 const char *operand_string;
4381 {
4382 int ok;
4383 #if INFER_ADDR_PREFIX
4384 int fudged = 0;
4385
4386 tryprefix:
4387 #endif
4388 ok = 1;
4389 if ((current_templates->start->cpu_flags & CpuSVME)
4390 && current_templates->end[-1].operand_types[0] == AnyMem)
4391 {
4392 /* Memory operands of SVME insns are special in that they only allow
4393 rAX as their memory address and ignore any segment override. */
4394 unsigned RegXX;
4395
4396 /* SKINIT is even more restrictive: it always requires EAX. */
4397 if (strcmp (current_templates->start->name, "skinit") == 0)
4398 RegXX = Reg32;
4399 else if (flag_code == CODE_64BIT)
4400 RegXX = i.prefix[ADDR_PREFIX] == 0 ? Reg64 : Reg32;
4401 else
4402 RegXX = ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0)
4403 ? Reg16
4404 : Reg32);
4405 if (!i.base_reg
4406 || !(i.base_reg->reg_type & Acc)
4407 || !(i.base_reg->reg_type & RegXX)
4408 || i.index_reg
4409 || (i.types[0] & Disp))
4410 ok = 0;
4411 }
4412 else if (flag_code == CODE_64BIT)
4413 {
4414 unsigned RegXX = (i.prefix[ADDR_PREFIX] == 0 ? Reg64 : Reg32);
4415
4416 if ((i.base_reg
4417 && ((i.base_reg->reg_type & RegXX) == 0)
4418 && (i.base_reg->reg_type != BaseIndex
4419 || i.index_reg))
4420 || (i.index_reg
4421 && ((i.index_reg->reg_type & (RegXX | BaseIndex))
4422 != (RegXX | BaseIndex))))
4423 ok = 0;
4424 }
4425 else
4426 {
4427 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
4428 {
4429 /* 16bit checks. */
4430 if ((i.base_reg
4431 && ((i.base_reg->reg_type & (Reg16 | BaseIndex | RegRex))
4432 != (Reg16 | BaseIndex)))
4433 || (i.index_reg
4434 && (((i.index_reg->reg_type & (Reg16 | BaseIndex))
4435 != (Reg16 | BaseIndex))
4436 || !(i.base_reg
4437 && i.base_reg->reg_num < 6
4438 && i.index_reg->reg_num >= 6
4439 && i.log2_scale_factor == 0))))
4440 ok = 0;
4441 }
4442 else
4443 {
4444 /* 32bit checks. */
4445 if ((i.base_reg
4446 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
4447 || (i.index_reg
4448 && ((i.index_reg->reg_type & (Reg32 | BaseIndex | RegRex))
4449 != (Reg32 | BaseIndex))))
4450 ok = 0;
4451 }
4452 }
4453 if (!ok)
4454 {
4455 #if INFER_ADDR_PREFIX
4456 if (i.prefix[ADDR_PREFIX] == 0)
4457 {
4458 i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
4459 i.prefixes += 1;
4460 /* Change the size of any displacement too. At most one of
4461 Disp16 or Disp32 is set.
4462 FIXME. There doesn't seem to be any real need for separate
4463 Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
4464 Removing them would probably clean up the code quite a lot. */
4465 if (flag_code != CODE_64BIT && (i.types[this_operand] & (Disp16 | Disp32)))
4466 i.types[this_operand] ^= (Disp16 | Disp32);
4467 fudged = 1;
4468 goto tryprefix;
4469 }
4470 if (fudged)
4471 as_bad (_("`%s' is not a valid base/index expression"),
4472 operand_string);
4473 else
4474 #endif
4475 as_bad (_("`%s' is not a valid %s bit base/index expression"),
4476 operand_string,
4477 flag_code_names[flag_code]);
4478 }
4479 return ok;
4480 }
4481
4482 /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
4483 on error. */
4484
4485 static int
4486 i386_operand (operand_string)
4487 char *operand_string;
4488 {
4489 const reg_entry *r;
4490 char *end_op;
4491 char *op_string = operand_string;
4492
4493 if (is_space_char (*op_string))
4494 ++op_string;
4495
4496 /* We check for an absolute prefix (differentiating,
4497 for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
4498 if (*op_string == ABSOLUTE_PREFIX)
4499 {
4500 ++op_string;
4501 if (is_space_char (*op_string))
4502 ++op_string;
4503 i.types[this_operand] |= JumpAbsolute;
4504 }
4505
4506 /* Check if operand is a register. */
4507 if ((r = parse_register (op_string, &end_op)) != NULL)
4508 {
4509 /* Check for a segment override by searching for ':' after a
4510 segment register. */
4511 op_string = end_op;
4512 if (is_space_char (*op_string))
4513 ++op_string;
4514 if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
4515 {
4516 switch (r->reg_num)
4517 {
4518 case 0:
4519 i.seg[i.mem_operands] = &es;
4520 break;
4521 case 1:
4522 i.seg[i.mem_operands] = &cs;
4523 break;
4524 case 2:
4525 i.seg[i.mem_operands] = &ss;
4526 break;
4527 case 3:
4528 i.seg[i.mem_operands] = &ds;
4529 break;
4530 case 4:
4531 i.seg[i.mem_operands] = &fs;
4532 break;
4533 case 5:
4534 i.seg[i.mem_operands] = &gs;
4535 break;
4536 }
4537
4538 /* Skip the ':' and whitespace. */
4539 ++op_string;
4540 if (is_space_char (*op_string))
4541 ++op_string;
4542
4543 if (!is_digit_char (*op_string)
4544 && !is_identifier_char (*op_string)
4545 && *op_string != '('
4546 && *op_string != ABSOLUTE_PREFIX)
4547 {
4548 as_bad (_("bad memory operand `%s'"), op_string);
4549 return 0;
4550 }
4551 /* Handle case of %es:*foo. */
4552 if (*op_string == ABSOLUTE_PREFIX)
4553 {
4554 ++op_string;
4555 if (is_space_char (*op_string))
4556 ++op_string;
4557 i.types[this_operand] |= JumpAbsolute;
4558 }
4559 goto do_memory_reference;
4560 }
4561 if (*op_string)
4562 {
4563 as_bad (_("junk `%s' after register"), op_string);
4564 return 0;
4565 }
4566 i.types[this_operand] |= r->reg_type & ~BaseIndex;
4567 i.op[this_operand].regs = r;
4568 i.reg_operands++;
4569 }
4570 else if (*op_string == REGISTER_PREFIX)
4571 {
4572 as_bad (_("bad register name `%s'"), op_string);
4573 return 0;
4574 }
4575 else if (*op_string == IMMEDIATE_PREFIX)
4576 {
4577 ++op_string;
4578 if (i.types[this_operand] & JumpAbsolute)
4579 {
4580 as_bad (_("immediate operand illegal with absolute jump"));
4581 return 0;
4582 }
4583 if (!i386_immediate (op_string))
4584 return 0;
4585 }
4586 else if (is_digit_char (*op_string)
4587 || is_identifier_char (*op_string)
4588 || *op_string == '(')
4589 {
4590 /* This is a memory reference of some sort. */
4591 char *base_string;
4592
4593 /* Start and end of displacement string expression (if found). */
4594 char *displacement_string_start;
4595 char *displacement_string_end;
4596
4597 do_memory_reference:
4598 if ((i.mem_operands == 1
4599 && (current_templates->start->opcode_modifier & IsString) == 0)
4600 || i.mem_operands == 2)
4601 {
4602 as_bad (_("too many memory references for `%s'"),
4603 current_templates->start->name);
4604 return 0;
4605 }
4606
4607 /* Check for base index form. We detect the base index form by
4608 looking for an ')' at the end of the operand, searching
4609 for the '(' matching it, and finding a REGISTER_PREFIX or ','
4610 after the '('. */
4611 base_string = op_string + strlen (op_string);
4612
4613 --base_string;
4614 if (is_space_char (*base_string))
4615 --base_string;
4616
4617 /* If we only have a displacement, set-up for it to be parsed later. */
4618 displacement_string_start = op_string;
4619 displacement_string_end = base_string + 1;
4620
4621 if (*base_string == ')')
4622 {
4623 char *temp_string;
4624 unsigned int parens_balanced = 1;
4625 /* We've already checked that the number of left & right ()'s are
4626 equal, so this loop will not be infinite. */
4627 do
4628 {
4629 base_string--;
4630 if (*base_string == ')')
4631 parens_balanced++;
4632 if (*base_string == '(')
4633 parens_balanced--;
4634 }
4635 while (parens_balanced);
4636
4637 temp_string = base_string;
4638
4639 /* Skip past '(' and whitespace. */
4640 ++base_string;
4641 if (is_space_char (*base_string))
4642 ++base_string;
4643
4644 if (*base_string == ','
4645 || ((i.base_reg = parse_register (base_string, &end_op)) != NULL))
4646 {
4647 displacement_string_end = temp_string;
4648
4649 i.types[this_operand] |= BaseIndex;
4650
4651 if (i.base_reg)
4652 {
4653 base_string = end_op;
4654 if (is_space_char (*base_string))
4655 ++base_string;
4656 }
4657
4658 /* There may be an index reg or scale factor here. */
4659 if (*base_string == ',')
4660 {
4661 ++base_string;
4662 if (is_space_char (*base_string))
4663 ++base_string;
4664
4665 if ((i.index_reg = parse_register (base_string, &end_op)) != NULL)
4666 {
4667 base_string = end_op;
4668 if (is_space_char (*base_string))
4669 ++base_string;
4670 if (*base_string == ',')
4671 {
4672 ++base_string;
4673 if (is_space_char (*base_string))
4674 ++base_string;
4675 }
4676 else if (*base_string != ')')
4677 {
4678 as_bad (_("expecting `,' or `)' after index register in `%s'"),
4679 operand_string);
4680 return 0;
4681 }
4682 }
4683 else if (*base_string == REGISTER_PREFIX)
4684 {
4685 as_bad (_("bad register name `%s'"), base_string);
4686 return 0;
4687 }
4688
4689 /* Check for scale factor. */
4690 if (*base_string != ')')
4691 {
4692 char *end_scale = i386_scale (base_string);
4693
4694 if (!end_scale)
4695 return 0;
4696
4697 base_string = end_scale;
4698 if (is_space_char (*base_string))
4699 ++base_string;
4700 if (*base_string != ')')
4701 {
4702 as_bad (_("expecting `)' after scale factor in `%s'"),
4703 operand_string);
4704 return 0;
4705 }
4706 }
4707 else if (!i.index_reg)
4708 {
4709 as_bad (_("expecting index register or scale factor after `,'; got '%c'"),
4710 *base_string);
4711 return 0;
4712 }
4713 }
4714 else if (*base_string != ')')
4715 {
4716 as_bad (_("expecting `,' or `)' after base register in `%s'"),
4717 operand_string);
4718 return 0;
4719 }
4720 }
4721 else if (*base_string == REGISTER_PREFIX)
4722 {
4723 as_bad (_("bad register name `%s'"), base_string);
4724 return 0;
4725 }
4726 }
4727
4728 /* If there's an expression beginning the operand, parse it,
4729 assuming displacement_string_start and
4730 displacement_string_end are meaningful. */
4731 if (displacement_string_start != displacement_string_end)
4732 {
4733 if (!i386_displacement (displacement_string_start,
4734 displacement_string_end))
4735 return 0;
4736 }
4737
4738 /* Special case for (%dx) while doing input/output op. */
4739 if (i.base_reg
4740 && i.base_reg->reg_type == (Reg16 | InOutPortReg)
4741 && i.index_reg == 0
4742 && i.log2_scale_factor == 0
4743 && i.seg[i.mem_operands] == 0
4744 && (i.types[this_operand] & Disp) == 0)
4745 {
4746 i.types[this_operand] = InOutPortReg;
4747 return 1;
4748 }
4749
4750 if (i386_index_check (operand_string) == 0)
4751 return 0;
4752 i.mem_operands++;
4753 }
4754 else
4755 {
4756 /* It's not a memory operand; argh! */
4757 as_bad (_("invalid char %s beginning operand %d `%s'"),
4758 output_invalid (*op_string),
4759 this_operand + 1,
4760 op_string);
4761 return 0;
4762 }
4763 return 1; /* Normal return. */
4764 }
4765 \f
4766 /* md_estimate_size_before_relax()
4767
4768 Called just before relax() for rs_machine_dependent frags. The x86
4769 assembler uses these frags to handle variable size jump
4770 instructions.
4771
4772 Any symbol that is now undefined will not become defined.
4773 Return the correct fr_subtype in the frag.
4774 Return the initial "guess for variable size of frag" to caller.
4775 The guess is actually the growth beyond the fixed part. Whatever
4776 we do to grow the fixed or variable part contributes to our
4777 returned value. */
4778
4779 int
4780 md_estimate_size_before_relax (fragP, segment)
4781 fragS *fragP;
4782 segT segment;
4783 {
4784 /* We've already got fragP->fr_subtype right; all we have to do is
4785 check for un-relaxable symbols. On an ELF system, we can't relax
4786 an externally visible symbol, because it may be overridden by a
4787 shared library. */
4788 if (S_GET_SEGMENT (fragP->fr_symbol) != segment
4789 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4790 || (IS_ELF
4791 && (S_IS_EXTERNAL (fragP->fr_symbol)
4792 || S_IS_WEAK (fragP->fr_symbol)))
4793 #endif
4794 )
4795 {
4796 /* Symbol is undefined in this segment, or we need to keep a
4797 reloc so that weak symbols can be overridden. */
4798 int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
4799 enum bfd_reloc_code_real reloc_type;
4800 unsigned char *opcode;
4801 int old_fr_fix;
4802
4803 if (fragP->fr_var != NO_RELOC)
4804 reloc_type = fragP->fr_var;
4805 else if (size == 2)
4806 reloc_type = BFD_RELOC_16_PCREL;
4807 else
4808 reloc_type = BFD_RELOC_32_PCREL;
4809
4810 old_fr_fix = fragP->fr_fix;
4811 opcode = (unsigned char *) fragP->fr_opcode;
4812
4813 switch (TYPE_FROM_RELAX_STATE (fragP->fr_subtype))
4814 {
4815 case UNCOND_JUMP:
4816 /* Make jmp (0xeb) a (d)word displacement jump. */
4817 opcode[0] = 0xe9;
4818 fragP->fr_fix += size;
4819 fix_new (fragP, old_fr_fix, size,
4820 fragP->fr_symbol,
4821 fragP->fr_offset, 1,
4822 reloc_type);
4823 break;
4824
4825 case COND_JUMP86:
4826 if (size == 2
4827 && (!no_cond_jump_promotion || fragP->fr_var != NO_RELOC))
4828 {
4829 /* Negate the condition, and branch past an
4830 unconditional jump. */
4831 opcode[0] ^= 1;
4832 opcode[1] = 3;
4833 /* Insert an unconditional jump. */
4834 opcode[2] = 0xe9;
4835 /* We added two extra opcode bytes, and have a two byte
4836 offset. */
4837 fragP->fr_fix += 2 + 2;
4838 fix_new (fragP, old_fr_fix + 2, 2,
4839 fragP->fr_symbol,
4840 fragP->fr_offset, 1,
4841 reloc_type);
4842 break;
4843 }
4844 /* Fall through. */
4845
4846 case COND_JUMP:
4847 if (no_cond_jump_promotion && fragP->fr_var == NO_RELOC)
4848 {
4849 fixS *fixP;
4850
4851 fragP->fr_fix += 1;
4852 fixP = fix_new (fragP, old_fr_fix, 1,
4853 fragP->fr_symbol,
4854 fragP->fr_offset, 1,
4855 BFD_RELOC_8_PCREL);
4856 fixP->fx_signed = 1;
4857 break;
4858 }
4859
4860 /* This changes the byte-displacement jump 0x7N
4861 to the (d)word-displacement jump 0x0f,0x8N. */
4862 opcode[1] = opcode[0] + 0x10;
4863 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4864 /* We've added an opcode byte. */
4865 fragP->fr_fix += 1 + size;
4866 fix_new (fragP, old_fr_fix + 1, size,
4867 fragP->fr_symbol,
4868 fragP->fr_offset, 1,
4869 reloc_type);
4870 break;
4871
4872 default:
4873 BAD_CASE (fragP->fr_subtype);
4874 break;
4875 }
4876 frag_wane (fragP);
4877 return fragP->fr_fix - old_fr_fix;
4878 }
4879
4880 /* Guess size depending on current relax state. Initially the relax
4881 state will correspond to a short jump and we return 1, because
4882 the variable part of the frag (the branch offset) is one byte
4883 long. However, we can relax a section more than once and in that
4884 case we must either set fr_subtype back to the unrelaxed state,
4885 or return the value for the appropriate branch. */
4886 return md_relax_table[fragP->fr_subtype].rlx_length;
4887 }
4888
4889 /* Called after relax() is finished.
4890
4891 In: Address of frag.
4892 fr_type == rs_machine_dependent.
4893 fr_subtype is what the address relaxed to.
4894
4895 Out: Any fixSs and constants are set up.
4896 Caller will turn frag into a ".space 0". */
4897
4898 void
4899 md_convert_frag (abfd, sec, fragP)
4900 bfd *abfd ATTRIBUTE_UNUSED;
4901 segT sec ATTRIBUTE_UNUSED;
4902 fragS *fragP;
4903 {
4904 unsigned char *opcode;
4905 unsigned char *where_to_put_displacement = NULL;
4906 offsetT target_address;
4907 offsetT opcode_address;
4908 unsigned int extension = 0;
4909 offsetT displacement_from_opcode_start;
4910
4911 opcode = (unsigned char *) fragP->fr_opcode;
4912
4913 /* Address we want to reach in file space. */
4914 target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
4915
4916 /* Address opcode resides at in file space. */
4917 opcode_address = fragP->fr_address + fragP->fr_fix;
4918
4919 /* Displacement from opcode start to fill into instruction. */
4920 displacement_from_opcode_start = target_address - opcode_address;
4921
4922 if ((fragP->fr_subtype & BIG) == 0)
4923 {
4924 /* Don't have to change opcode. */
4925 extension = 1; /* 1 opcode + 1 displacement */
4926 where_to_put_displacement = &opcode[1];
4927 }
4928 else
4929 {
4930 if (no_cond_jump_promotion
4931 && TYPE_FROM_RELAX_STATE (fragP->fr_subtype) != UNCOND_JUMP)
4932 as_warn_where (fragP->fr_file, fragP->fr_line, _("long jump required"));
4933
4934 switch (fragP->fr_subtype)
4935 {
4936 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
4937 extension = 4; /* 1 opcode + 4 displacement */
4938 opcode[0] = 0xe9;
4939 where_to_put_displacement = &opcode[1];
4940 break;
4941
4942 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
4943 extension = 2; /* 1 opcode + 2 displacement */
4944 opcode[0] = 0xe9;
4945 where_to_put_displacement = &opcode[1];
4946 break;
4947
4948 case ENCODE_RELAX_STATE (COND_JUMP, BIG):
4949 case ENCODE_RELAX_STATE (COND_JUMP86, BIG):
4950 extension = 5; /* 2 opcode + 4 displacement */
4951 opcode[1] = opcode[0] + 0x10;
4952 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4953 where_to_put_displacement = &opcode[2];
4954 break;
4955
4956 case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
4957 extension = 3; /* 2 opcode + 2 displacement */
4958 opcode[1] = opcode[0] + 0x10;
4959 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4960 where_to_put_displacement = &opcode[2];
4961 break;
4962
4963 case ENCODE_RELAX_STATE (COND_JUMP86, BIG16):
4964 extension = 4;
4965 opcode[0] ^= 1;
4966 opcode[1] = 3;
4967 opcode[2] = 0xe9;
4968 where_to_put_displacement = &opcode[3];
4969 break;
4970
4971 default:
4972 BAD_CASE (fragP->fr_subtype);
4973 break;
4974 }
4975 }
4976
4977 /* If size if less then four we are sure that the operand fits,
4978 but if it's 4, then it could be that the displacement is larger
4979 then -/+ 2GB. */
4980 if (DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype) == 4
4981 && object_64bit
4982 && ((addressT) (displacement_from_opcode_start - extension
4983 + ((addressT) 1 << 31))
4984 > (((addressT) 2 << 31) - 1)))
4985 {
4986 as_bad_where (fragP->fr_file, fragP->fr_line,
4987 _("jump target out of range"));
4988 /* Make us emit 0. */
4989 displacement_from_opcode_start = extension;
4990 }
4991 /* Now put displacement after opcode. */
4992 md_number_to_chars ((char *) where_to_put_displacement,
4993 (valueT) (displacement_from_opcode_start - extension),
4994 DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
4995 fragP->fr_fix += extension;
4996 }
4997 \f
4998 /* Size of byte displacement jmp. */
4999 int md_short_jump_size = 2;
5000
5001 /* Size of dword displacement jmp. */
5002 int md_long_jump_size = 5;
5003
5004 void
5005 md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
5006 char *ptr;
5007 addressT from_addr, to_addr;
5008 fragS *frag ATTRIBUTE_UNUSED;
5009 symbolS *to_symbol ATTRIBUTE_UNUSED;
5010 {
5011 offsetT offset;
5012
5013 offset = to_addr - (from_addr + 2);
5014 /* Opcode for byte-disp jump. */
5015 md_number_to_chars (ptr, (valueT) 0xeb, 1);
5016 md_number_to_chars (ptr + 1, (valueT) offset, 1);
5017 }
5018
5019 void
5020 md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
5021 char *ptr;
5022 addressT from_addr, to_addr;
5023 fragS *frag ATTRIBUTE_UNUSED;
5024 symbolS *to_symbol ATTRIBUTE_UNUSED;
5025 {
5026 offsetT offset;
5027
5028 offset = to_addr - (from_addr + 5);
5029 md_number_to_chars (ptr, (valueT) 0xe9, 1);
5030 md_number_to_chars (ptr + 1, (valueT) offset, 4);
5031 }
5032 \f
5033 /* Apply a fixup (fixS) to segment data, once it has been determined
5034 by our caller that we have all the info we need to fix it up.
5035
5036 On the 386, immediates, displacements, and data pointers are all in
5037 the same (little-endian) format, so we don't need to care about which
5038 we are handling. */
5039
5040 void
5041 md_apply_fix (fixP, valP, seg)
5042 /* The fix we're to put in. */
5043 fixS *fixP;
5044 /* Pointer to the value of the bits. */
5045 valueT *valP;
5046 /* Segment fix is from. */
5047 segT seg ATTRIBUTE_UNUSED;
5048 {
5049 char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
5050 valueT value = *valP;
5051
5052 #if !defined (TE_Mach)
5053 if (fixP->fx_pcrel)
5054 {
5055 switch (fixP->fx_r_type)
5056 {
5057 default:
5058 break;
5059
5060 case BFD_RELOC_64:
5061 fixP->fx_r_type = BFD_RELOC_64_PCREL;
5062 break;
5063 case BFD_RELOC_32:
5064 case BFD_RELOC_X86_64_32S:
5065 fixP->fx_r_type = BFD_RELOC_32_PCREL;
5066 break;
5067 case BFD_RELOC_16:
5068 fixP->fx_r_type = BFD_RELOC_16_PCREL;
5069 break;
5070 case BFD_RELOC_8:
5071 fixP->fx_r_type = BFD_RELOC_8_PCREL;
5072 break;
5073 }
5074 }
5075
5076 if (fixP->fx_addsy != NULL
5077 && (fixP->fx_r_type == BFD_RELOC_32_PCREL
5078 || fixP->fx_r_type == BFD_RELOC_64_PCREL
5079 || fixP->fx_r_type == BFD_RELOC_16_PCREL
5080 || fixP->fx_r_type == BFD_RELOC_8_PCREL)
5081 && !use_rela_relocations)
5082 {
5083 /* This is a hack. There should be a better way to handle this.
5084 This covers for the fact that bfd_install_relocation will
5085 subtract the current location (for partial_inplace, PC relative
5086 relocations); see more below. */
5087 #ifndef OBJ_AOUT
5088 if (IS_ELF
5089 #ifdef TE_PE
5090 || OUTPUT_FLAVOR == bfd_target_coff_flavour
5091 #endif
5092 )
5093 value += fixP->fx_where + fixP->fx_frag->fr_address;
5094 #endif
5095 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5096 if (IS_ELF)
5097 {
5098 segT sym_seg = S_GET_SEGMENT (fixP->fx_addsy);
5099
5100 if ((sym_seg == seg
5101 || (symbol_section_p (fixP->fx_addsy)
5102 && sym_seg != absolute_section))
5103 && !generic_force_reloc (fixP))
5104 {
5105 /* Yes, we add the values in twice. This is because
5106 bfd_install_relocation subtracts them out again. I think
5107 bfd_install_relocation is broken, but I don't dare change
5108 it. FIXME. */
5109 value += fixP->fx_where + fixP->fx_frag->fr_address;
5110 }
5111 }
5112 #endif
5113 #if defined (OBJ_COFF) && defined (TE_PE)
5114 /* For some reason, the PE format does not store a
5115 section address offset for a PC relative symbol. */
5116 if (S_GET_SEGMENT (fixP->fx_addsy) != seg
5117 || S_IS_WEAK (fixP->fx_addsy))
5118 value += md_pcrel_from (fixP);
5119 #endif
5120 }
5121
5122 /* Fix a few things - the dynamic linker expects certain values here,
5123 and we must not disappoint it. */
5124 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5125 if (IS_ELF && fixP->fx_addsy)
5126 switch (fixP->fx_r_type)
5127 {
5128 case BFD_RELOC_386_PLT32:
5129 case BFD_RELOC_X86_64_PLT32:
5130 /* Make the jump instruction point to the address of the operand. At
5131 runtime we merely add the offset to the actual PLT entry. */
5132 value = -4;
5133 break;
5134
5135 case BFD_RELOC_386_TLS_GD:
5136 case BFD_RELOC_386_TLS_LDM:
5137 case BFD_RELOC_386_TLS_IE_32:
5138 case BFD_RELOC_386_TLS_IE:
5139 case BFD_RELOC_386_TLS_GOTIE:
5140 case BFD_RELOC_386_TLS_GOTDESC:
5141 case BFD_RELOC_X86_64_TLSGD:
5142 case BFD_RELOC_X86_64_TLSLD:
5143 case BFD_RELOC_X86_64_GOTTPOFF:
5144 case BFD_RELOC_X86_64_GOTPC32_TLSDESC:
5145 value = 0; /* Fully resolved at runtime. No addend. */
5146 /* Fallthrough */
5147 case BFD_RELOC_386_TLS_LE:
5148 case BFD_RELOC_386_TLS_LDO_32:
5149 case BFD_RELOC_386_TLS_LE_32:
5150 case BFD_RELOC_X86_64_DTPOFF32:
5151 case BFD_RELOC_X86_64_DTPOFF64:
5152 case BFD_RELOC_X86_64_TPOFF32:
5153 case BFD_RELOC_X86_64_TPOFF64:
5154 S_SET_THREAD_LOCAL (fixP->fx_addsy);
5155 break;
5156
5157 case BFD_RELOC_386_TLS_DESC_CALL:
5158 case BFD_RELOC_X86_64_TLSDESC_CALL:
5159 value = 0; /* Fully resolved at runtime. No addend. */
5160 S_SET_THREAD_LOCAL (fixP->fx_addsy);
5161 fixP->fx_done = 0;
5162 return;
5163
5164 case BFD_RELOC_386_GOT32:
5165 case BFD_RELOC_X86_64_GOT32:
5166 value = 0; /* Fully resolved at runtime. No addend. */
5167 break;
5168
5169 case BFD_RELOC_VTABLE_INHERIT:
5170 case BFD_RELOC_VTABLE_ENTRY:
5171 fixP->fx_done = 0;
5172 return;
5173
5174 default:
5175 break;
5176 }
5177 #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
5178 *valP = value;
5179 #endif /* !defined (TE_Mach) */
5180
5181 /* Are we finished with this relocation now? */
5182 if (fixP->fx_addsy == NULL)
5183 fixP->fx_done = 1;
5184 else if (use_rela_relocations)
5185 {
5186 fixP->fx_no_overflow = 1;
5187 /* Remember value for tc_gen_reloc. */
5188 fixP->fx_addnumber = value;
5189 value = 0;
5190 }
5191
5192 md_number_to_chars (p, value, fixP->fx_size);
5193 }
5194 \f
5195 #define MAX_LITTLENUMS 6
5196
5197 /* Turn the string pointed to by litP into a floating point constant
5198 of type TYPE, and emit the appropriate bytes. The number of
5199 LITTLENUMS emitted is stored in *SIZEP. An error message is
5200 returned, or NULL on OK. */
5201
5202 char *
5203 md_atof (type, litP, sizeP)
5204 int type;
5205 char *litP;
5206 int *sizeP;
5207 {
5208 int prec;
5209 LITTLENUM_TYPE words[MAX_LITTLENUMS];
5210 LITTLENUM_TYPE *wordP;
5211 char *t;
5212
5213 switch (type)
5214 {
5215 case 'f':
5216 case 'F':
5217 prec = 2;
5218 break;
5219
5220 case 'd':
5221 case 'D':
5222 prec = 4;
5223 break;
5224
5225 case 'x':
5226 case 'X':
5227 prec = 5;
5228 break;
5229
5230 default:
5231 *sizeP = 0;
5232 return _("Bad call to md_atof ()");
5233 }
5234 t = atof_ieee (input_line_pointer, type, words);
5235 if (t)
5236 input_line_pointer = t;
5237
5238 *sizeP = prec * sizeof (LITTLENUM_TYPE);
5239 /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
5240 the bigendian 386. */
5241 for (wordP = words + prec - 1; prec--;)
5242 {
5243 md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
5244 litP += sizeof (LITTLENUM_TYPE);
5245 }
5246 return 0;
5247 }
5248 \f
5249 static char output_invalid_buf[8];
5250
5251 static char *
5252 output_invalid (c)
5253 int c;
5254 {
5255 if (ISPRINT (c))
5256 sprintf (output_invalid_buf, "'%c'", c);
5257 else
5258 sprintf (output_invalid_buf, "(0x%x)", (unsigned) c);
5259 return output_invalid_buf;
5260 }
5261
5262 /* REG_STRING starts *before* REGISTER_PREFIX. */
5263
5264 static const reg_entry *
5265 parse_real_register (char *reg_string, char **end_op)
5266 {
5267 char *s = reg_string;
5268 char *p;
5269 char reg_name_given[MAX_REG_NAME_SIZE + 1];
5270 const reg_entry *r;
5271
5272 /* Skip possible REGISTER_PREFIX and possible whitespace. */
5273 if (*s == REGISTER_PREFIX)
5274 ++s;
5275
5276 if (is_space_char (*s))
5277 ++s;
5278
5279 p = reg_name_given;
5280 while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
5281 {
5282 if (p >= reg_name_given + MAX_REG_NAME_SIZE)
5283 return (const reg_entry *) NULL;
5284 s++;
5285 }
5286
5287 /* For naked regs, make sure that we are not dealing with an identifier.
5288 This prevents confusing an identifier like `eax_var' with register
5289 `eax'. */
5290 if (allow_naked_reg && identifier_chars[(unsigned char) *s])
5291 return (const reg_entry *) NULL;
5292
5293 *end_op = s;
5294
5295 r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
5296
5297 /* Handle floating point regs, allowing spaces in the (i) part. */
5298 if (r == i386_regtab /* %st is first entry of table */)
5299 {
5300 if (is_space_char (*s))
5301 ++s;
5302 if (*s == '(')
5303 {
5304 ++s;
5305 if (is_space_char (*s))
5306 ++s;
5307 if (*s >= '0' && *s <= '7')
5308 {
5309 r = &i386_float_regtab[*s - '0'];
5310 ++s;
5311 if (is_space_char (*s))
5312 ++s;
5313 if (*s == ')')
5314 {
5315 *end_op = s + 1;
5316 return r;
5317 }
5318 }
5319 /* We have "%st(" then garbage. */
5320 return (const reg_entry *) NULL;
5321 }
5322 }
5323
5324 if (r != NULL
5325 && ((r->reg_flags & (RegRex64 | RegRex)) | (r->reg_type & Reg64)) != 0
5326 && (r->reg_type != Control || !(cpu_arch_flags & CpuSledgehammer))
5327 && flag_code != CODE_64BIT)
5328 return (const reg_entry *) NULL;
5329
5330 return r;
5331 }
5332
5333 /* REG_STRING starts *before* REGISTER_PREFIX. */
5334
5335 static const reg_entry *
5336 parse_register (char *reg_string, char **end_op)
5337 {
5338 const reg_entry *r;
5339
5340 if (*reg_string == REGISTER_PREFIX || allow_naked_reg)
5341 r = parse_real_register (reg_string, end_op);
5342 else
5343 r = NULL;
5344 if (!r)
5345 {
5346 char *save = input_line_pointer;
5347 char c;
5348 symbolS *symbolP;
5349
5350 input_line_pointer = reg_string;
5351 c = get_symbol_end ();
5352 symbolP = symbol_find (reg_string);
5353 if (symbolP && S_GET_SEGMENT (symbolP) == reg_section)
5354 {
5355 const expressionS *e = symbol_get_value_expression (symbolP);
5356
5357 know (e->X_op == O_register);
5358 know (e->X_add_number >= 0 && (valueT) e->X_add_number < ARRAY_SIZE (i386_regtab));
5359 r = i386_regtab + e->X_add_number;
5360 *end_op = input_line_pointer;
5361 }
5362 *input_line_pointer = c;
5363 input_line_pointer = save;
5364 }
5365 return r;
5366 }
5367
5368 int
5369 i386_parse_name (char *name, expressionS *e, char *nextcharP)
5370 {
5371 const reg_entry *r;
5372 char *end = input_line_pointer;
5373
5374 *end = *nextcharP;
5375 r = parse_register (name, &input_line_pointer);
5376 if (r && end <= input_line_pointer)
5377 {
5378 *nextcharP = *input_line_pointer;
5379 *input_line_pointer = 0;
5380 e->X_op = O_register;
5381 e->X_add_number = r - i386_regtab;
5382 return 1;
5383 }
5384 input_line_pointer = end;
5385 *end = 0;
5386 return 0;
5387 }
5388
5389 void
5390 md_operand (expressionS *e)
5391 {
5392 if (*input_line_pointer == REGISTER_PREFIX)
5393 {
5394 char *end;
5395 const reg_entry *r = parse_real_register (input_line_pointer, &end);
5396
5397 if (r)
5398 {
5399 e->X_op = O_register;
5400 e->X_add_number = r - i386_regtab;
5401 input_line_pointer = end;
5402 }
5403 }
5404 }
5405
5406 \f
5407 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5408 const char *md_shortopts = "kVQ:sqn";
5409 #else
5410 const char *md_shortopts = "qn";
5411 #endif
5412
5413 #define OPTION_32 (OPTION_MD_BASE + 0)
5414 #define OPTION_64 (OPTION_MD_BASE + 1)
5415 #define OPTION_DIVIDE (OPTION_MD_BASE + 2)
5416
5417 struct option md_longopts[] = {
5418 {"32", no_argument, NULL, OPTION_32},
5419 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5420 {"64", no_argument, NULL, OPTION_64},
5421 #endif
5422 {"divide", no_argument, NULL, OPTION_DIVIDE},
5423 {NULL, no_argument, NULL, 0}
5424 };
5425 size_t md_longopts_size = sizeof (md_longopts);
5426
5427 int
5428 md_parse_option (c, arg)
5429 int c;
5430 char *arg ATTRIBUTE_UNUSED;
5431 {
5432 switch (c)
5433 {
5434 case 'n':
5435 optimize_align_code = 0;
5436 break;
5437
5438 case 'q':
5439 quiet_warnings = 1;
5440 break;
5441
5442 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5443 /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
5444 should be emitted or not. FIXME: Not implemented. */
5445 case 'Q':
5446 break;
5447
5448 /* -V: SVR4 argument to print version ID. */
5449 case 'V':
5450 print_version_id ();
5451 break;
5452
5453 /* -k: Ignore for FreeBSD compatibility. */
5454 case 'k':
5455 break;
5456
5457 case 's':
5458 /* -s: On i386 Solaris, this tells the native assembler to use
5459 .stab instead of .stab.excl. We always use .stab anyhow. */
5460 break;
5461
5462 case OPTION_64:
5463 {
5464 const char **list, **l;
5465
5466 list = bfd_target_list ();
5467 for (l = list; *l != NULL; l++)
5468 if (strcmp (*l, "elf64-x86-64") == 0)
5469 {
5470 default_arch = "x86_64";
5471 break;
5472 }
5473 if (*l == NULL)
5474 as_fatal (_("No compiled in support for x86_64"));
5475 free (list);
5476 }
5477 break;
5478 #endif
5479
5480 case OPTION_32:
5481 default_arch = "i386";
5482 break;
5483
5484 case OPTION_DIVIDE:
5485 #ifdef SVR4_COMMENT_CHARS
5486 {
5487 char *n, *t;
5488 const char *s;
5489
5490 n = (char *) xmalloc (strlen (i386_comment_chars) + 1);
5491 t = n;
5492 for (s = i386_comment_chars; *s != '\0'; s++)
5493 if (*s != '/')
5494 *t++ = *s;
5495 *t = '\0';
5496 i386_comment_chars = n;
5497 }
5498 #endif
5499 break;
5500
5501 default:
5502 return 0;
5503 }
5504 return 1;
5505 }
5506
5507 void
5508 md_show_usage (stream)
5509 FILE *stream;
5510 {
5511 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5512 fprintf (stream, _("\
5513 -Q ignored\n\
5514 -V print assembler version number\n\
5515 -k ignored\n"));
5516 #endif
5517 fprintf (stream, _("\
5518 -n Do not optimize code alignment\n\
5519 -q quieten some warnings\n"));
5520 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5521 fprintf (stream, _("\
5522 -s ignored\n"));
5523 #endif
5524 #ifdef SVR4_COMMENT_CHARS
5525 fprintf (stream, _("\
5526 --divide do not treat `/' as a comment character\n"));
5527 #else
5528 fprintf (stream, _("\
5529 --divide ignored\n"));
5530 #endif
5531 }
5532
5533 #if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
5534 || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
5535
5536 /* Pick the target format to use. */
5537
5538 const char *
5539 i386_target_format ()
5540 {
5541 if (!strcmp (default_arch, "x86_64"))
5542 set_code_flag (CODE_64BIT);
5543 else if (!strcmp (default_arch, "i386"))
5544 set_code_flag (CODE_32BIT);
5545 else
5546 as_fatal (_("Unknown architecture"));
5547 switch (OUTPUT_FLAVOR)
5548 {
5549 #ifdef OBJ_MAYBE_AOUT
5550 case bfd_target_aout_flavour:
5551 return AOUT_TARGET_FORMAT;
5552 #endif
5553 #ifdef OBJ_MAYBE_COFF
5554 case bfd_target_coff_flavour:
5555 return "coff-i386";
5556 #endif
5557 #if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF)
5558 case bfd_target_elf_flavour:
5559 {
5560 if (flag_code == CODE_64BIT)
5561 {
5562 object_64bit = 1;
5563 use_rela_relocations = 1;
5564 }
5565 return flag_code == CODE_64BIT ? "elf64-x86-64" : ELF_TARGET_FORMAT;
5566 }
5567 #endif
5568 default:
5569 abort ();
5570 return NULL;
5571 }
5572 }
5573
5574 #endif /* OBJ_MAYBE_ more than one */
5575
5576 #if (defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
5577 void i386_elf_emit_arch_note ()
5578 {
5579 if (IS_ELF && cpu_arch_name != NULL)
5580 {
5581 char *p;
5582 asection *seg = now_seg;
5583 subsegT subseg = now_subseg;
5584 Elf_Internal_Note i_note;
5585 Elf_External_Note e_note;
5586 asection *note_secp;
5587 int len;
5588
5589 /* Create the .note section. */
5590 note_secp = subseg_new (".note", 0);
5591 bfd_set_section_flags (stdoutput,
5592 note_secp,
5593 SEC_HAS_CONTENTS | SEC_READONLY);
5594
5595 /* Process the arch string. */
5596 len = strlen (cpu_arch_name);
5597
5598 i_note.namesz = len + 1;
5599 i_note.descsz = 0;
5600 i_note.type = NT_ARCH;
5601 p = frag_more (sizeof (e_note.namesz));
5602 md_number_to_chars (p, (valueT) i_note.namesz, sizeof (e_note.namesz));
5603 p = frag_more (sizeof (e_note.descsz));
5604 md_number_to_chars (p, (valueT) i_note.descsz, sizeof (e_note.descsz));
5605 p = frag_more (sizeof (e_note.type));
5606 md_number_to_chars (p, (valueT) i_note.type, sizeof (e_note.type));
5607 p = frag_more (len + 1);
5608 strcpy (p, cpu_arch_name);
5609
5610 frag_align (2, 0, 0);
5611
5612 subseg_set (seg, subseg);
5613 }
5614 }
5615 #endif
5616 \f
5617 symbolS *
5618 md_undefined_symbol (name)
5619 char *name;
5620 {
5621 if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
5622 && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
5623 && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
5624 && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
5625 {
5626 if (!GOT_symbol)
5627 {
5628 if (symbol_find (name))
5629 as_bad (_("GOT already in symbol table"));
5630 GOT_symbol = symbol_new (name, undefined_section,
5631 (valueT) 0, &zero_address_frag);
5632 };
5633 return GOT_symbol;
5634 }
5635 return 0;
5636 }
5637
5638 /* Round up a section size to the appropriate boundary. */
5639
5640 valueT
5641 md_section_align (segment, size)
5642 segT segment ATTRIBUTE_UNUSED;
5643 valueT size;
5644 {
5645 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
5646 if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
5647 {
5648 /* For a.out, force the section size to be aligned. If we don't do
5649 this, BFD will align it for us, but it will not write out the
5650 final bytes of the section. This may be a bug in BFD, but it is
5651 easier to fix it here since that is how the other a.out targets
5652 work. */
5653 int align;
5654
5655 align = bfd_get_section_alignment (stdoutput, segment);
5656 size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
5657 }
5658 #endif
5659
5660 return size;
5661 }
5662
5663 /* On the i386, PC-relative offsets are relative to the start of the
5664 next instruction. That is, the address of the offset, plus its
5665 size, since the offset is always the last part of the insn. */
5666
5667 long
5668 md_pcrel_from (fixP)
5669 fixS *fixP;
5670 {
5671 return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
5672 }
5673
5674 #ifndef I386COFF
5675
5676 static void
5677 s_bss (ignore)
5678 int ignore ATTRIBUTE_UNUSED;
5679 {
5680 int temp;
5681
5682 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5683 if (IS_ELF)
5684 obj_elf_section_change_hook ();
5685 #endif
5686 temp = get_absolute_expression ();
5687 subseg_set (bss_section, (subsegT) temp);
5688 demand_empty_rest_of_line ();
5689 }
5690
5691 #endif
5692
5693 void
5694 i386_validate_fix (fixp)
5695 fixS *fixp;
5696 {
5697 if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
5698 {
5699 if (fixp->fx_r_type == BFD_RELOC_32_PCREL)
5700 {
5701 if (!object_64bit)
5702 abort ();
5703 fixp->fx_r_type = BFD_RELOC_X86_64_GOTPCREL;
5704 }
5705 else
5706 {
5707 if (!object_64bit)
5708 fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
5709 else
5710 fixp->fx_r_type = BFD_RELOC_X86_64_GOTOFF64;
5711 }
5712 fixp->fx_subsy = 0;
5713 }
5714 }
5715
5716 arelent *
5717 tc_gen_reloc (section, fixp)
5718 asection *section ATTRIBUTE_UNUSED;
5719 fixS *fixp;
5720 {
5721 arelent *rel;
5722 bfd_reloc_code_real_type code;
5723
5724 switch (fixp->fx_r_type)
5725 {
5726 case BFD_RELOC_X86_64_PLT32:
5727 case BFD_RELOC_X86_64_GOT32:
5728 case BFD_RELOC_X86_64_GOTPCREL:
5729 case BFD_RELOC_386_PLT32:
5730 case BFD_RELOC_386_GOT32:
5731 case BFD_RELOC_386_GOTOFF:
5732 case BFD_RELOC_386_GOTPC:
5733 case BFD_RELOC_386_TLS_GD:
5734 case BFD_RELOC_386_TLS_LDM:
5735 case BFD_RELOC_386_TLS_LDO_32:
5736 case BFD_RELOC_386_TLS_IE_32:
5737 case BFD_RELOC_386_TLS_IE:
5738 case BFD_RELOC_386_TLS_GOTIE:
5739 case BFD_RELOC_386_TLS_LE_32:
5740 case BFD_RELOC_386_TLS_LE:
5741 case BFD_RELOC_386_TLS_GOTDESC:
5742 case BFD_RELOC_386_TLS_DESC_CALL:
5743 case BFD_RELOC_X86_64_TLSGD:
5744 case BFD_RELOC_X86_64_TLSLD:
5745 case BFD_RELOC_X86_64_DTPOFF32:
5746 case BFD_RELOC_X86_64_DTPOFF64:
5747 case BFD_RELOC_X86_64_GOTTPOFF:
5748 case BFD_RELOC_X86_64_TPOFF32:
5749 case BFD_RELOC_X86_64_TPOFF64:
5750 case BFD_RELOC_X86_64_GOTOFF64:
5751 case BFD_RELOC_X86_64_GOTPC32:
5752 case BFD_RELOC_X86_64_GOT64:
5753 case BFD_RELOC_X86_64_GOTPCREL64:
5754 case BFD_RELOC_X86_64_GOTPC64:
5755 case BFD_RELOC_X86_64_GOTPLT64:
5756 case BFD_RELOC_X86_64_PLTOFF64:
5757 case BFD_RELOC_X86_64_GOTPC32_TLSDESC:
5758 case BFD_RELOC_X86_64_TLSDESC_CALL:
5759 case BFD_RELOC_RVA:
5760 case BFD_RELOC_VTABLE_ENTRY:
5761 case BFD_RELOC_VTABLE_INHERIT:
5762 #ifdef TE_PE
5763 case BFD_RELOC_32_SECREL:
5764 #endif
5765 code = fixp->fx_r_type;
5766 break;
5767 case BFD_RELOC_X86_64_32S:
5768 if (!fixp->fx_pcrel)
5769 {
5770 /* Don't turn BFD_RELOC_X86_64_32S into BFD_RELOC_32. */
5771 code = fixp->fx_r_type;
5772 break;
5773 }
5774 default:
5775 if (fixp->fx_pcrel)
5776 {
5777 switch (fixp->fx_size)
5778 {
5779 default:
5780 as_bad_where (fixp->fx_file, fixp->fx_line,
5781 _("can not do %d byte pc-relative relocation"),
5782 fixp->fx_size);
5783 code = BFD_RELOC_32_PCREL;
5784 break;
5785 case 1: code = BFD_RELOC_8_PCREL; break;
5786 case 2: code = BFD_RELOC_16_PCREL; break;
5787 case 4: code = BFD_RELOC_32_PCREL; break;
5788 #ifdef BFD64
5789 case 8: code = BFD_RELOC_64_PCREL; break;
5790 #endif
5791 }
5792 }
5793 else
5794 {
5795 switch (fixp->fx_size)
5796 {
5797 default:
5798 as_bad_where (fixp->fx_file, fixp->fx_line,
5799 _("can not do %d byte relocation"),
5800 fixp->fx_size);
5801 code = BFD_RELOC_32;
5802 break;
5803 case 1: code = BFD_RELOC_8; break;
5804 case 2: code = BFD_RELOC_16; break;
5805 case 4: code = BFD_RELOC_32; break;
5806 #ifdef BFD64
5807 case 8: code = BFD_RELOC_64; break;
5808 #endif
5809 }
5810 }
5811 break;
5812 }
5813
5814 if ((code == BFD_RELOC_32
5815 || code == BFD_RELOC_32_PCREL
5816 || code == BFD_RELOC_X86_64_32S)
5817 && GOT_symbol
5818 && fixp->fx_addsy == GOT_symbol)
5819 {
5820 if (!object_64bit)
5821 code = BFD_RELOC_386_GOTPC;
5822 else
5823 code = BFD_RELOC_X86_64_GOTPC32;
5824 }
5825 if ((code == BFD_RELOC_64 || code == BFD_RELOC_64_PCREL)
5826 && GOT_symbol
5827 && fixp->fx_addsy == GOT_symbol)
5828 {
5829 code = BFD_RELOC_X86_64_GOTPC64;
5830 }
5831
5832 rel = (arelent *) xmalloc (sizeof (arelent));
5833 rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
5834 *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
5835
5836 rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
5837
5838 if (!use_rela_relocations)
5839 {
5840 /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
5841 vtable entry to be used in the relocation's section offset. */
5842 if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
5843 rel->address = fixp->fx_offset;
5844
5845 rel->addend = 0;
5846 }
5847 /* Use the rela in 64bit mode. */
5848 else
5849 {
5850 if (!fixp->fx_pcrel)
5851 rel->addend = fixp->fx_offset;
5852 else
5853 switch (code)
5854 {
5855 case BFD_RELOC_X86_64_PLT32:
5856 case BFD_RELOC_X86_64_GOT32:
5857 case BFD_RELOC_X86_64_GOTPCREL:
5858 case BFD_RELOC_X86_64_TLSGD:
5859 case BFD_RELOC_X86_64_TLSLD:
5860 case BFD_RELOC_X86_64_GOTTPOFF:
5861 case BFD_RELOC_X86_64_GOTPC32_TLSDESC:
5862 case BFD_RELOC_X86_64_TLSDESC_CALL:
5863 rel->addend = fixp->fx_offset - fixp->fx_size;
5864 break;
5865 default:
5866 rel->addend = (section->vma
5867 - fixp->fx_size
5868 + fixp->fx_addnumber
5869 + md_pcrel_from (fixp));
5870 break;
5871 }
5872 }
5873
5874 rel->howto = bfd_reloc_type_lookup (stdoutput, code);
5875 if (rel->howto == NULL)
5876 {
5877 as_bad_where (fixp->fx_file, fixp->fx_line,
5878 _("cannot represent relocation type %s"),
5879 bfd_get_reloc_code_name (code));
5880 /* Set howto to a garbage value so that we can keep going. */
5881 rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
5882 assert (rel->howto != NULL);
5883 }
5884
5885 return rel;
5886 }
5887
5888 \f
5889 /* Parse operands using Intel syntax. This implements a recursive descent
5890 parser based on the BNF grammar published in Appendix B of the MASM 6.1
5891 Programmer's Guide.
5892
5893 FIXME: We do not recognize the full operand grammar defined in the MASM
5894 documentation. In particular, all the structure/union and
5895 high-level macro operands are missing.
5896
5897 Uppercase words are terminals, lower case words are non-terminals.
5898 Objects surrounded by double brackets '[[' ']]' are optional. Vertical
5899 bars '|' denote choices. Most grammar productions are implemented in
5900 functions called 'intel_<production>'.
5901
5902 Initial production is 'expr'.
5903
5904 addOp + | -
5905
5906 alpha [a-zA-Z]
5907
5908 binOp & | AND | \| | OR | ^ | XOR
5909
5910 byteRegister AL | AH | BL | BH | CL | CH | DL | DH
5911
5912 constant digits [[ radixOverride ]]
5913
5914 dataType BYTE | WORD | DWORD | FWORD | QWORD | TBYTE | OWORD | XMMWORD
5915
5916 digits decdigit
5917 | digits decdigit
5918 | digits hexdigit
5919
5920 decdigit [0-9]
5921
5922 e04 e04 addOp e05
5923 | e05
5924
5925 e05 e05 binOp e06
5926 | e06
5927
5928 e06 e06 mulOp e09
5929 | e09
5930
5931 e09 OFFSET e10
5932 | SHORT e10
5933 | + e10
5934 | - e10
5935 | ~ e10
5936 | NOT e10
5937 | e09 PTR e10
5938 | e09 : e10
5939 | e10
5940
5941 e10 e10 [ expr ]
5942 | e11
5943
5944 e11 ( expr )
5945 | [ expr ]
5946 | constant
5947 | dataType
5948 | id
5949 | $
5950 | register
5951
5952 => expr expr cmpOp e04
5953 | e04
5954
5955 gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
5956 | BP | EBP | SP | ESP | DI | EDI | SI | ESI
5957
5958 hexdigit a | b | c | d | e | f
5959 | A | B | C | D | E | F
5960
5961 id alpha
5962 | id alpha
5963 | id decdigit
5964
5965 mulOp * | / | % | MOD | << | SHL | >> | SHR
5966
5967 quote " | '
5968
5969 register specialRegister
5970 | gpRegister
5971 | byteRegister
5972
5973 segmentRegister CS | DS | ES | FS | GS | SS
5974
5975 specialRegister CR0 | CR2 | CR3 | CR4
5976 | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
5977 | TR3 | TR4 | TR5 | TR6 | TR7
5978
5979 We simplify the grammar in obvious places (e.g., register parsing is
5980 done by calling parse_register) and eliminate immediate left recursion
5981 to implement a recursive-descent parser.
5982
5983 expr e04 expr'
5984
5985 expr' cmpOp e04 expr'
5986 | Empty
5987
5988 e04 e05 e04'
5989
5990 e04' addOp e05 e04'
5991 | Empty
5992
5993 e05 e06 e05'
5994
5995 e05' binOp e06 e05'
5996 | Empty
5997
5998 e06 e09 e06'
5999
6000 e06' mulOp e09 e06'
6001 | Empty
6002
6003 e09 OFFSET e10 e09'
6004 | SHORT e10'
6005 | + e10'
6006 | - e10'
6007 | ~ e10'
6008 | NOT e10'
6009 | e10 e09'
6010
6011 e09' PTR e10 e09'
6012 | : e10 e09'
6013 | Empty
6014
6015 e10 e11 e10'
6016
6017 e10' [ expr ] e10'
6018 | Empty
6019
6020 e11 ( expr )
6021 | [ expr ]
6022 | BYTE
6023 | WORD
6024 | DWORD
6025 | FWORD
6026 | QWORD
6027 | TBYTE
6028 | OWORD
6029 | XMMWORD
6030 | .
6031 | $
6032 | register
6033 | id
6034 | constant */
6035
6036 /* Parsing structure for the intel syntax parser. Used to implement the
6037 semantic actions for the operand grammar. */
6038 struct intel_parser_s
6039 {
6040 char *op_string; /* The string being parsed. */
6041 int got_a_float; /* Whether the operand is a float. */
6042 int op_modifier; /* Operand modifier. */
6043 int is_mem; /* 1 if operand is memory reference. */
6044 int in_offset; /* >=1 if parsing operand of offset. */
6045 int in_bracket; /* >=1 if parsing operand in brackets. */
6046 const reg_entry *reg; /* Last register reference found. */
6047 char *disp; /* Displacement string being built. */
6048 char *next_operand; /* Resume point when splitting operands. */
6049 };
6050
6051 static struct intel_parser_s intel_parser;
6052
6053 /* Token structure for parsing intel syntax. */
6054 struct intel_token
6055 {
6056 int code; /* Token code. */
6057 const reg_entry *reg; /* Register entry for register tokens. */
6058 char *str; /* String representation. */
6059 };
6060
6061 static struct intel_token cur_token, prev_token;
6062
6063 /* Token codes for the intel parser. Since T_SHORT is already used
6064 by COFF, undefine it first to prevent a warning. */
6065 #define T_NIL -1
6066 #define T_CONST 1
6067 #define T_REG 2
6068 #define T_BYTE 3
6069 #define T_WORD 4
6070 #define T_DWORD 5
6071 #define T_FWORD 6
6072 #define T_QWORD 7
6073 #define T_TBYTE 8
6074 #define T_XMMWORD 9
6075 #undef T_SHORT
6076 #define T_SHORT 10
6077 #define T_OFFSET 11
6078 #define T_PTR 12
6079 #define T_ID 13
6080 #define T_SHL 14
6081 #define T_SHR 15
6082
6083 /* Prototypes for intel parser functions. */
6084 static int intel_match_token PARAMS ((int code));
6085 static void intel_get_token PARAMS ((void));
6086 static void intel_putback_token PARAMS ((void));
6087 static int intel_expr PARAMS ((void));
6088 static int intel_e04 PARAMS ((void));
6089 static int intel_e05 PARAMS ((void));
6090 static int intel_e06 PARAMS ((void));
6091 static int intel_e09 PARAMS ((void));
6092 static int intel_bracket_expr PARAMS ((void));
6093 static int intel_e10 PARAMS ((void));
6094 static int intel_e11 PARAMS ((void));
6095
6096 static int
6097 i386_intel_operand (operand_string, got_a_float)
6098 char *operand_string;
6099 int got_a_float;
6100 {
6101 int ret;
6102 char *p;
6103
6104 p = intel_parser.op_string = xstrdup (operand_string);
6105 intel_parser.disp = (char *) xmalloc (strlen (operand_string) + 1);
6106
6107 for (;;)
6108 {
6109 /* Initialize token holders. */
6110 cur_token.code = prev_token.code = T_NIL;
6111 cur_token.reg = prev_token.reg = NULL;
6112 cur_token.str = prev_token.str = NULL;
6113
6114 /* Initialize parser structure. */
6115 intel_parser.got_a_float = got_a_float;
6116 intel_parser.op_modifier = 0;
6117 intel_parser.is_mem = 0;
6118 intel_parser.in_offset = 0;
6119 intel_parser.in_bracket = 0;
6120 intel_parser.reg = NULL;
6121 intel_parser.disp[0] = '\0';
6122 intel_parser.next_operand = NULL;
6123
6124 /* Read the first token and start the parser. */
6125 intel_get_token ();
6126 ret = intel_expr ();
6127
6128 if (!ret)
6129 break;
6130
6131 if (cur_token.code != T_NIL)
6132 {
6133 as_bad (_("invalid operand for '%s' ('%s' unexpected)"),
6134 current_templates->start->name, cur_token.str);
6135 ret = 0;
6136 }
6137 /* If we found a memory reference, hand it over to i386_displacement
6138 to fill in the rest of the operand fields. */
6139 else if (intel_parser.is_mem)
6140 {
6141 if ((i.mem_operands == 1
6142 && (current_templates->start->opcode_modifier & IsString) == 0)
6143 || i.mem_operands == 2)
6144 {
6145 as_bad (_("too many memory references for '%s'"),
6146 current_templates->start->name);
6147 ret = 0;
6148 }
6149 else
6150 {
6151 char *s = intel_parser.disp;
6152 i.mem_operands++;
6153
6154 if (!quiet_warnings && intel_parser.is_mem < 0)
6155 /* See the comments in intel_bracket_expr. */
6156 as_warn (_("Treating `%s' as memory reference"), operand_string);
6157
6158 /* Add the displacement expression. */
6159 if (*s != '\0')
6160 ret = i386_displacement (s, s + strlen (s));
6161 if (ret)
6162 {
6163 /* Swap base and index in 16-bit memory operands like
6164 [si+bx]. Since i386_index_check is also used in AT&T
6165 mode we have to do that here. */
6166 if (i.base_reg
6167 && i.index_reg
6168 && (i.base_reg->reg_type & Reg16)
6169 && (i.index_reg->reg_type & Reg16)
6170 && i.base_reg->reg_num >= 6
6171 && i.index_reg->reg_num < 6)
6172 {
6173 const reg_entry *base = i.index_reg;
6174
6175 i.index_reg = i.base_reg;
6176 i.base_reg = base;
6177 }
6178 ret = i386_index_check (operand_string);
6179 }
6180 }
6181 }
6182
6183 /* Constant and OFFSET expressions are handled by i386_immediate. */
6184 else if ((intel_parser.op_modifier & (1 << T_OFFSET))
6185 || intel_parser.reg == NULL)
6186 ret = i386_immediate (intel_parser.disp);
6187
6188 if (intel_parser.next_operand && this_operand >= MAX_OPERANDS - 1)
6189 ret = 0;
6190 if (!ret || !intel_parser.next_operand)
6191 break;
6192 intel_parser.op_string = intel_parser.next_operand;
6193 this_operand = i.operands++;
6194 }
6195
6196 free (p);
6197 free (intel_parser.disp);
6198
6199 return ret;
6200 }
6201
6202 #define NUM_ADDRESS_REGS (!!i.base_reg + !!i.index_reg)
6203
6204 /* expr e04 expr'
6205
6206 expr' cmpOp e04 expr'
6207 | Empty */
6208 static int
6209 intel_expr ()
6210 {
6211 /* XXX Implement the comparison operators. */
6212 return intel_e04 ();
6213 }
6214
6215 /* e04 e05 e04'
6216
6217 e04' addOp e05 e04'
6218 | Empty */
6219 static int
6220 intel_e04 ()
6221 {
6222 int nregs = -1;
6223
6224 for (;;)
6225 {
6226 if (!intel_e05())
6227 return 0;
6228
6229 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
6230 i.base_reg = i386_regtab + REGNAM_AL; /* al is invalid as base */
6231
6232 if (cur_token.code == '+')
6233 nregs = -1;
6234 else if (cur_token.code == '-')
6235 nregs = NUM_ADDRESS_REGS;
6236 else
6237 return 1;
6238
6239 strcat (intel_parser.disp, cur_token.str);
6240 intel_match_token (cur_token.code);
6241 }
6242 }
6243
6244 /* e05 e06 e05'
6245
6246 e05' binOp e06 e05'
6247 | Empty */
6248 static int
6249 intel_e05 ()
6250 {
6251 int nregs = ~NUM_ADDRESS_REGS;
6252
6253 for (;;)
6254 {
6255 if (!intel_e06())
6256 return 0;
6257
6258 if (cur_token.code == '&' || cur_token.code == '|' || cur_token.code == '^')
6259 {
6260 char str[2];
6261
6262 str[0] = cur_token.code;
6263 str[1] = 0;
6264 strcat (intel_parser.disp, str);
6265 }
6266 else
6267 break;
6268
6269 intel_match_token (cur_token.code);
6270
6271 if (nregs < 0)
6272 nregs = ~nregs;
6273 }
6274 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
6275 i.base_reg = i386_regtab + REGNAM_AL + 1; /* cl is invalid as base */
6276 return 1;
6277 }
6278
6279 /* e06 e09 e06'
6280
6281 e06' mulOp e09 e06'
6282 | Empty */
6283 static int
6284 intel_e06 ()
6285 {
6286 int nregs = ~NUM_ADDRESS_REGS;
6287
6288 for (;;)
6289 {
6290 if (!intel_e09())
6291 return 0;
6292
6293 if (cur_token.code == '*' || cur_token.code == '/' || cur_token.code == '%')
6294 {
6295 char str[2];
6296
6297 str[0] = cur_token.code;
6298 str[1] = 0;
6299 strcat (intel_parser.disp, str);
6300 }
6301 else if (cur_token.code == T_SHL)
6302 strcat (intel_parser.disp, "<<");
6303 else if (cur_token.code == T_SHR)
6304 strcat (intel_parser.disp, ">>");
6305 else
6306 break;
6307
6308 intel_match_token (cur_token.code);
6309
6310 if (nregs < 0)
6311 nregs = ~nregs;
6312 }
6313 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
6314 i.base_reg = i386_regtab + REGNAM_AL + 2; /* dl is invalid as base */
6315 return 1;
6316 }
6317
6318 /* e09 OFFSET e09
6319 | SHORT e09
6320 | + e09
6321 | - e09
6322 | ~ e09
6323 | NOT e09
6324 | e10 e09'
6325
6326 e09' PTR e10 e09'
6327 | : e10 e09'
6328 | Empty */
6329 static int
6330 intel_e09 ()
6331 {
6332 int nregs = ~NUM_ADDRESS_REGS;
6333 int in_offset = 0;
6334
6335 for (;;)
6336 {
6337 /* Don't consume constants here. */
6338 if (cur_token.code == '+' || cur_token.code == '-')
6339 {
6340 /* Need to look one token ahead - if the next token
6341 is a constant, the current token is its sign. */
6342 int next_code;
6343
6344 intel_match_token (cur_token.code);
6345 next_code = cur_token.code;
6346 intel_putback_token ();
6347 if (next_code == T_CONST)
6348 break;
6349 }
6350
6351 /* e09 OFFSET e09 */
6352 if (cur_token.code == T_OFFSET)
6353 {
6354 if (!in_offset++)
6355 ++intel_parser.in_offset;
6356 }
6357
6358 /* e09 SHORT e09 */
6359 else if (cur_token.code == T_SHORT)
6360 intel_parser.op_modifier |= 1 << T_SHORT;
6361
6362 /* e09 + e09 */
6363 else if (cur_token.code == '+')
6364 strcat (intel_parser.disp, "+");
6365
6366 /* e09 - e09
6367 | ~ e09
6368 | NOT e09 */
6369 else if (cur_token.code == '-' || cur_token.code == '~')
6370 {
6371 char str[2];
6372
6373 if (nregs < 0)
6374 nregs = ~nregs;
6375 str[0] = cur_token.code;
6376 str[1] = 0;
6377 strcat (intel_parser.disp, str);
6378 }
6379
6380 /* e09 e10 e09' */
6381 else
6382 break;
6383
6384 intel_match_token (cur_token.code);
6385 }
6386
6387 for (;;)
6388 {
6389 if (!intel_e10 ())
6390 return 0;
6391
6392 /* e09' PTR e10 e09' */
6393 if (cur_token.code == T_PTR)
6394 {
6395 char suffix;
6396
6397 if (prev_token.code == T_BYTE)
6398 suffix = BYTE_MNEM_SUFFIX;
6399
6400 else if (prev_token.code == T_WORD)
6401 {
6402 if (current_templates->start->name[0] == 'l'
6403 && current_templates->start->name[2] == 's'
6404 && current_templates->start->name[3] == 0)
6405 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
6406 else if (intel_parser.got_a_float == 2) /* "fi..." */
6407 suffix = SHORT_MNEM_SUFFIX;
6408 else
6409 suffix = WORD_MNEM_SUFFIX;
6410 }
6411
6412 else if (prev_token.code == T_DWORD)
6413 {
6414 if (current_templates->start->name[0] == 'l'
6415 && current_templates->start->name[2] == 's'
6416 && current_templates->start->name[3] == 0)
6417 suffix = WORD_MNEM_SUFFIX;
6418 else if (flag_code == CODE_16BIT
6419 && (current_templates->start->opcode_modifier
6420 & (Jump | JumpDword)))
6421 suffix = LONG_DOUBLE_MNEM_SUFFIX;
6422 else if (intel_parser.got_a_float == 1) /* "f..." */
6423 suffix = SHORT_MNEM_SUFFIX;
6424 else
6425 suffix = LONG_MNEM_SUFFIX;
6426 }
6427
6428 else if (prev_token.code == T_FWORD)
6429 {
6430 if (current_templates->start->name[0] == 'l'
6431 && current_templates->start->name[2] == 's'
6432 && current_templates->start->name[3] == 0)
6433 suffix = LONG_MNEM_SUFFIX;
6434 else if (!intel_parser.got_a_float)
6435 {
6436 if (flag_code == CODE_16BIT)
6437 add_prefix (DATA_PREFIX_OPCODE);
6438 suffix = LONG_DOUBLE_MNEM_SUFFIX;
6439 }
6440 else
6441 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
6442 }
6443
6444 else if (prev_token.code == T_QWORD)
6445 {
6446 if (intel_parser.got_a_float == 1) /* "f..." */
6447 suffix = LONG_MNEM_SUFFIX;
6448 else
6449 suffix = QWORD_MNEM_SUFFIX;
6450 }
6451
6452 else if (prev_token.code == T_TBYTE)
6453 {
6454 if (intel_parser.got_a_float == 1)
6455 suffix = LONG_DOUBLE_MNEM_SUFFIX;
6456 else
6457 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
6458 }
6459
6460 else if (prev_token.code == T_XMMWORD)
6461 {
6462 /* XXX ignored for now, but accepted since gcc uses it */
6463 suffix = 0;
6464 }
6465
6466 else
6467 {
6468 as_bad (_("Unknown operand modifier `%s'"), prev_token.str);
6469 return 0;
6470 }
6471
6472 /* Operands for jump/call using 'ptr' notation denote absolute
6473 addresses. */
6474 if (current_templates->start->opcode_modifier & (Jump | JumpDword))
6475 i.types[this_operand] |= JumpAbsolute;
6476
6477 if (current_templates->start->base_opcode == 0x8d /* lea */)
6478 ;
6479 else if (!i.suffix)
6480 i.suffix = suffix;
6481 else if (i.suffix != suffix)
6482 {
6483 as_bad (_("Conflicting operand modifiers"));
6484 return 0;
6485 }
6486
6487 }
6488
6489 /* e09' : e10 e09' */
6490 else if (cur_token.code == ':')
6491 {
6492 if (prev_token.code != T_REG)
6493 {
6494 /* While {call,jmp} SSSS:OOOO is MASM syntax only when SSSS is a
6495 segment/group identifier (which we don't have), using comma
6496 as the operand separator there is even less consistent, since
6497 there all branches only have a single operand. */
6498 if (this_operand != 0
6499 || intel_parser.in_offset
6500 || intel_parser.in_bracket
6501 || (!(current_templates->start->opcode_modifier
6502 & (Jump|JumpDword|JumpInterSegment))
6503 && !(current_templates->start->operand_types[0]
6504 & JumpAbsolute)))
6505 return intel_match_token (T_NIL);
6506 /* Remember the start of the 2nd operand and terminate 1st
6507 operand here.
6508 XXX This isn't right, yet (when SSSS:OOOO is right operand of
6509 another expression), but it gets at least the simplest case
6510 (a plain number or symbol on the left side) right. */
6511 intel_parser.next_operand = intel_parser.op_string;
6512 *--intel_parser.op_string = '\0';
6513 return intel_match_token (':');
6514 }
6515 }
6516
6517 /* e09' Empty */
6518 else
6519 break;
6520
6521 intel_match_token (cur_token.code);
6522
6523 }
6524
6525 if (in_offset)
6526 {
6527 --intel_parser.in_offset;
6528 if (nregs < 0)
6529 nregs = ~nregs;
6530 if (NUM_ADDRESS_REGS > nregs)
6531 {
6532 as_bad (_("Invalid operand to `OFFSET'"));
6533 return 0;
6534 }
6535 intel_parser.op_modifier |= 1 << T_OFFSET;
6536 }
6537
6538 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
6539 i.base_reg = i386_regtab + REGNAM_AL + 3; /* bl is invalid as base */
6540 return 1;
6541 }
6542
6543 static int
6544 intel_bracket_expr ()
6545 {
6546 int was_offset = intel_parser.op_modifier & (1 << T_OFFSET);
6547 const char *start = intel_parser.op_string;
6548 int len;
6549
6550 if (i.op[this_operand].regs)
6551 return intel_match_token (T_NIL);
6552
6553 intel_match_token ('[');
6554
6555 /* Mark as a memory operand only if it's not already known to be an
6556 offset expression. If it's an offset expression, we need to keep
6557 the brace in. */
6558 if (!intel_parser.in_offset)
6559 {
6560 ++intel_parser.in_bracket;
6561
6562 /* Operands for jump/call inside brackets denote absolute addresses. */
6563 if (current_templates->start->opcode_modifier & (Jump | JumpDword))
6564 i.types[this_operand] |= JumpAbsolute;
6565
6566 /* Unfortunately gas always diverged from MASM in a respect that can't
6567 be easily fixed without risking to break code sequences likely to be
6568 encountered (the testsuite even check for this): MASM doesn't consider
6569 an expression inside brackets unconditionally as a memory reference.
6570 When that is e.g. a constant, an offset expression, or the sum of the
6571 two, this is still taken as a constant load. gas, however, always
6572 treated these as memory references. As a compromise, we'll try to make
6573 offset expressions inside brackets work the MASM way (since that's
6574 less likely to be found in real world code), but make constants alone
6575 continue to work the traditional gas way. In either case, issue a
6576 warning. */
6577 intel_parser.op_modifier &= ~was_offset;
6578 }
6579 else
6580 strcat (intel_parser.disp, "[");
6581
6582 /* Add a '+' to the displacement string if necessary. */
6583 if (*intel_parser.disp != '\0'
6584 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
6585 strcat (intel_parser.disp, "+");
6586
6587 if (intel_expr ()
6588 && (len = intel_parser.op_string - start - 1,
6589 intel_match_token (']')))
6590 {
6591 /* Preserve brackets when the operand is an offset expression. */
6592 if (intel_parser.in_offset)
6593 strcat (intel_parser.disp, "]");
6594 else
6595 {
6596 --intel_parser.in_bracket;
6597 if (i.base_reg || i.index_reg)
6598 intel_parser.is_mem = 1;
6599 if (!intel_parser.is_mem)
6600 {
6601 if (!(intel_parser.op_modifier & (1 << T_OFFSET)))
6602 /* Defer the warning until all of the operand was parsed. */
6603 intel_parser.is_mem = -1;
6604 else if (!quiet_warnings)
6605 as_warn (_("`[%.*s]' taken to mean just `%.*s'"), len, start, len, start);
6606 }
6607 }
6608 intel_parser.op_modifier |= was_offset;
6609
6610 return 1;
6611 }
6612 return 0;
6613 }
6614
6615 /* e10 e11 e10'
6616
6617 e10' [ expr ] e10'
6618 | Empty */
6619 static int
6620 intel_e10 ()
6621 {
6622 if (!intel_e11 ())
6623 return 0;
6624
6625 while (cur_token.code == '[')
6626 {
6627 if (!intel_bracket_expr ())
6628 return 0;
6629 }
6630
6631 return 1;
6632 }
6633
6634 /* e11 ( expr )
6635 | [ expr ]
6636 | BYTE
6637 | WORD
6638 | DWORD
6639 | FWORD
6640 | QWORD
6641 | TBYTE
6642 | OWORD
6643 | XMMWORD
6644 | $
6645 | .
6646 | register
6647 | id
6648 | constant */
6649 static int
6650 intel_e11 ()
6651 {
6652 switch (cur_token.code)
6653 {
6654 /* e11 ( expr ) */
6655 case '(':
6656 intel_match_token ('(');
6657 strcat (intel_parser.disp, "(");
6658
6659 if (intel_expr () && intel_match_token (')'))
6660 {
6661 strcat (intel_parser.disp, ")");
6662 return 1;
6663 }
6664 return 0;
6665
6666 /* e11 [ expr ] */
6667 case '[':
6668 return intel_bracket_expr ();
6669
6670 /* e11 $
6671 | . */
6672 case '.':
6673 strcat (intel_parser.disp, cur_token.str);
6674 intel_match_token (cur_token.code);
6675
6676 /* Mark as a memory operand only if it's not already known to be an
6677 offset expression. */
6678 if (!intel_parser.in_offset)
6679 intel_parser.is_mem = 1;
6680
6681 return 1;
6682
6683 /* e11 register */
6684 case T_REG:
6685 {
6686 const reg_entry *reg = intel_parser.reg = cur_token.reg;
6687
6688 intel_match_token (T_REG);
6689
6690 /* Check for segment change. */
6691 if (cur_token.code == ':')
6692 {
6693 if (!(reg->reg_type & (SReg2 | SReg3)))
6694 {
6695 as_bad (_("`%s' is not a valid segment register"), reg->reg_name);
6696 return 0;
6697 }
6698 else if (i.seg[i.mem_operands])
6699 as_warn (_("Extra segment override ignored"));
6700 else
6701 {
6702 if (!intel_parser.in_offset)
6703 intel_parser.is_mem = 1;
6704 switch (reg->reg_num)
6705 {
6706 case 0:
6707 i.seg[i.mem_operands] = &es;
6708 break;
6709 case 1:
6710 i.seg[i.mem_operands] = &cs;
6711 break;
6712 case 2:
6713 i.seg[i.mem_operands] = &ss;
6714 break;
6715 case 3:
6716 i.seg[i.mem_operands] = &ds;
6717 break;
6718 case 4:
6719 i.seg[i.mem_operands] = &fs;
6720 break;
6721 case 5:
6722 i.seg[i.mem_operands] = &gs;
6723 break;
6724 }
6725 }
6726 }
6727
6728 /* Not a segment register. Check for register scaling. */
6729 else if (cur_token.code == '*')
6730 {
6731 if (!intel_parser.in_bracket)
6732 {
6733 as_bad (_("Register scaling only allowed in memory operands"));
6734 return 0;
6735 }
6736
6737 if (reg->reg_type & Reg16) /* Disallow things like [si*1]. */
6738 reg = i386_regtab + REGNAM_AX + 4; /* sp is invalid as index */
6739 else if (i.index_reg)
6740 reg = i386_regtab + REGNAM_EAX + 4; /* esp is invalid as index */
6741
6742 /* What follows must be a valid scale. */
6743 intel_match_token ('*');
6744 i.index_reg = reg;
6745 i.types[this_operand] |= BaseIndex;
6746
6747 /* Set the scale after setting the register (otherwise,
6748 i386_scale will complain) */
6749 if (cur_token.code == '+' || cur_token.code == '-')
6750 {
6751 char *str, sign = cur_token.code;
6752 intel_match_token (cur_token.code);
6753 if (cur_token.code != T_CONST)
6754 {
6755 as_bad (_("Syntax error: Expecting a constant, got `%s'"),
6756 cur_token.str);
6757 return 0;
6758 }
6759 str = (char *) xmalloc (strlen (cur_token.str) + 2);
6760 strcpy (str + 1, cur_token.str);
6761 *str = sign;
6762 if (!i386_scale (str))
6763 return 0;
6764 free (str);
6765 }
6766 else if (!i386_scale (cur_token.str))
6767 return 0;
6768 intel_match_token (cur_token.code);
6769 }
6770
6771 /* No scaling. If this is a memory operand, the register is either a
6772 base register (first occurrence) or an index register (second
6773 occurrence). */
6774 else if (intel_parser.in_bracket)
6775 {
6776
6777 if (!i.base_reg)
6778 i.base_reg = reg;
6779 else if (!i.index_reg)
6780 i.index_reg = reg;
6781 else
6782 {
6783 as_bad (_("Too many register references in memory operand"));
6784 return 0;
6785 }
6786
6787 i.types[this_operand] |= BaseIndex;
6788 }
6789
6790 /* It's neither base nor index. */
6791 else if (!intel_parser.in_offset && !intel_parser.is_mem)
6792 {
6793 i.types[this_operand] |= reg->reg_type & ~BaseIndex;
6794 i.op[this_operand].regs = reg;
6795 i.reg_operands++;
6796 }
6797 else
6798 {
6799 as_bad (_("Invalid use of register"));
6800 return 0;
6801 }
6802
6803 /* Since registers are not part of the displacement string (except
6804 when we're parsing offset operands), we may need to remove any
6805 preceding '+' from the displacement string. */
6806 if (*intel_parser.disp != '\0'
6807 && !intel_parser.in_offset)
6808 {
6809 char *s = intel_parser.disp;
6810 s += strlen (s) - 1;
6811 if (*s == '+')
6812 *s = '\0';
6813 }
6814
6815 return 1;
6816 }
6817
6818 /* e11 BYTE
6819 | WORD
6820 | DWORD
6821 | FWORD
6822 | QWORD
6823 | TBYTE
6824 | OWORD
6825 | XMMWORD */
6826 case T_BYTE:
6827 case T_WORD:
6828 case T_DWORD:
6829 case T_FWORD:
6830 case T_QWORD:
6831 case T_TBYTE:
6832 case T_XMMWORD:
6833 intel_match_token (cur_token.code);
6834
6835 if (cur_token.code == T_PTR)
6836 return 1;
6837
6838 /* It must have been an identifier. */
6839 intel_putback_token ();
6840 cur_token.code = T_ID;
6841 /* FALLTHRU */
6842
6843 /* e11 id
6844 | constant */
6845 case T_ID:
6846 if (!intel_parser.in_offset && intel_parser.is_mem <= 0)
6847 {
6848 symbolS *symbolP;
6849
6850 /* The identifier represents a memory reference only if it's not
6851 preceded by an offset modifier and if it's not an equate. */
6852 symbolP = symbol_find(cur_token.str);
6853 if (!symbolP || S_GET_SEGMENT(symbolP) != absolute_section)
6854 intel_parser.is_mem = 1;
6855 }
6856 /* FALLTHRU */
6857
6858 case T_CONST:
6859 case '-':
6860 case '+':
6861 {
6862 char *save_str, sign = 0;
6863
6864 /* Allow constants that start with `+' or `-'. */
6865 if (cur_token.code == '-' || cur_token.code == '+')
6866 {
6867 sign = cur_token.code;
6868 intel_match_token (cur_token.code);
6869 if (cur_token.code != T_CONST)
6870 {
6871 as_bad (_("Syntax error: Expecting a constant, got `%s'"),
6872 cur_token.str);
6873 return 0;
6874 }
6875 }
6876
6877 save_str = (char *) xmalloc (strlen (cur_token.str) + 2);
6878 strcpy (save_str + !!sign, cur_token.str);
6879 if (sign)
6880 *save_str = sign;
6881
6882 /* Get the next token to check for register scaling. */
6883 intel_match_token (cur_token.code);
6884
6885 /* Check if this constant is a scaling factor for an index register. */
6886 if (cur_token.code == '*')
6887 {
6888 if (intel_match_token ('*') && cur_token.code == T_REG)
6889 {
6890 const reg_entry *reg = cur_token.reg;
6891
6892 if (!intel_parser.in_bracket)
6893 {
6894 as_bad (_("Register scaling only allowed in memory operands"));
6895 return 0;
6896 }
6897
6898 if (reg->reg_type & Reg16) /* Disallow things like [1*si]. */
6899 reg = i386_regtab + REGNAM_AX + 4; /* sp is invalid as index */
6900 else if (i.index_reg)
6901 reg = i386_regtab + REGNAM_EAX + 4; /* esp is invalid as index */
6902
6903 /* The constant is followed by `* reg', so it must be
6904 a valid scale. */
6905 i.index_reg = reg;
6906 i.types[this_operand] |= BaseIndex;
6907
6908 /* Set the scale after setting the register (otherwise,
6909 i386_scale will complain) */
6910 if (!i386_scale (save_str))
6911 return 0;
6912 intel_match_token (T_REG);
6913
6914 /* Since registers are not part of the displacement
6915 string, we may need to remove any preceding '+' from
6916 the displacement string. */
6917 if (*intel_parser.disp != '\0')
6918 {
6919 char *s = intel_parser.disp;
6920 s += strlen (s) - 1;
6921 if (*s == '+')
6922 *s = '\0';
6923 }
6924
6925 free (save_str);
6926
6927 return 1;
6928 }
6929
6930 /* The constant was not used for register scaling. Since we have
6931 already consumed the token following `*' we now need to put it
6932 back in the stream. */
6933 intel_putback_token ();
6934 }
6935
6936 /* Add the constant to the displacement string. */
6937 strcat (intel_parser.disp, save_str);
6938 free (save_str);
6939
6940 return 1;
6941 }
6942 }
6943
6944 as_bad (_("Unrecognized token '%s'"), cur_token.str);
6945 return 0;
6946 }
6947
6948 /* Match the given token against cur_token. If they match, read the next
6949 token from the operand string. */
6950 static int
6951 intel_match_token (code)
6952 int code;
6953 {
6954 if (cur_token.code == code)
6955 {
6956 intel_get_token ();
6957 return 1;
6958 }
6959 else
6960 {
6961 as_bad (_("Unexpected token `%s'"), cur_token.str);
6962 return 0;
6963 }
6964 }
6965
6966 /* Read a new token from intel_parser.op_string and store it in cur_token. */
6967 static void
6968 intel_get_token ()
6969 {
6970 char *end_op;
6971 const reg_entry *reg;
6972 struct intel_token new_token;
6973
6974 new_token.code = T_NIL;
6975 new_token.reg = NULL;
6976 new_token.str = NULL;
6977
6978 /* Free the memory allocated to the previous token and move
6979 cur_token to prev_token. */
6980 if (prev_token.str)
6981 free (prev_token.str);
6982
6983 prev_token = cur_token;
6984
6985 /* Skip whitespace. */
6986 while (is_space_char (*intel_parser.op_string))
6987 intel_parser.op_string++;
6988
6989 /* Return an empty token if we find nothing else on the line. */
6990 if (*intel_parser.op_string == '\0')
6991 {
6992 cur_token = new_token;
6993 return;
6994 }
6995
6996 /* The new token cannot be larger than the remainder of the operand
6997 string. */
6998 new_token.str = (char *) xmalloc (strlen (intel_parser.op_string) + 1);
6999 new_token.str[0] = '\0';
7000
7001 if (strchr ("0123456789", *intel_parser.op_string))
7002 {
7003 char *p = new_token.str;
7004 char *q = intel_parser.op_string;
7005 new_token.code = T_CONST;
7006
7007 /* Allow any kind of identifier char to encompass floating point and
7008 hexadecimal numbers. */
7009 while (is_identifier_char (*q))
7010 *p++ = *q++;
7011 *p = '\0';
7012
7013 /* Recognize special symbol names [0-9][bf]. */
7014 if (strlen (intel_parser.op_string) == 2
7015 && (intel_parser.op_string[1] == 'b'
7016 || intel_parser.op_string[1] == 'f'))
7017 new_token.code = T_ID;
7018 }
7019
7020 else if ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL)
7021 {
7022 size_t len = end_op - intel_parser.op_string;
7023
7024 new_token.code = T_REG;
7025 new_token.reg = reg;
7026
7027 memcpy (new_token.str, intel_parser.op_string, len);
7028 new_token.str[len] = '\0';
7029 }
7030
7031 else if (is_identifier_char (*intel_parser.op_string))
7032 {
7033 char *p = new_token.str;
7034 char *q = intel_parser.op_string;
7035
7036 /* A '.' or '$' followed by an identifier char is an identifier.
7037 Otherwise, it's operator '.' followed by an expression. */
7038 if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
7039 {
7040 new_token.code = '.';
7041 new_token.str[0] = '.';
7042 new_token.str[1] = '\0';
7043 }
7044 else
7045 {
7046 while (is_identifier_char (*q) || *q == '@')
7047 *p++ = *q++;
7048 *p = '\0';
7049
7050 if (strcasecmp (new_token.str, "NOT") == 0)
7051 new_token.code = '~';
7052
7053 else if (strcasecmp (new_token.str, "MOD") == 0)
7054 new_token.code = '%';
7055
7056 else if (strcasecmp (new_token.str, "AND") == 0)
7057 new_token.code = '&';
7058
7059 else if (strcasecmp (new_token.str, "OR") == 0)
7060 new_token.code = '|';
7061
7062 else if (strcasecmp (new_token.str, "XOR") == 0)
7063 new_token.code = '^';
7064
7065 else if (strcasecmp (new_token.str, "SHL") == 0)
7066 new_token.code = T_SHL;
7067
7068 else if (strcasecmp (new_token.str, "SHR") == 0)
7069 new_token.code = T_SHR;
7070
7071 else if (strcasecmp (new_token.str, "BYTE") == 0)
7072 new_token.code = T_BYTE;
7073
7074 else if (strcasecmp (new_token.str, "WORD") == 0)
7075 new_token.code = T_WORD;
7076
7077 else if (strcasecmp (new_token.str, "DWORD") == 0)
7078 new_token.code = T_DWORD;
7079
7080 else if (strcasecmp (new_token.str, "FWORD") == 0)
7081 new_token.code = T_FWORD;
7082
7083 else if (strcasecmp (new_token.str, "QWORD") == 0)
7084 new_token.code = T_QWORD;
7085
7086 else if (strcasecmp (new_token.str, "TBYTE") == 0
7087 /* XXX remove (gcc still uses it) */
7088 || strcasecmp (new_token.str, "XWORD") == 0)
7089 new_token.code = T_TBYTE;
7090
7091 else if (strcasecmp (new_token.str, "XMMWORD") == 0
7092 || strcasecmp (new_token.str, "OWORD") == 0)
7093 new_token.code = T_XMMWORD;
7094
7095 else if (strcasecmp (new_token.str, "PTR") == 0)
7096 new_token.code = T_PTR;
7097
7098 else if (strcasecmp (new_token.str, "SHORT") == 0)
7099 new_token.code = T_SHORT;
7100
7101 else if (strcasecmp (new_token.str, "OFFSET") == 0)
7102 {
7103 new_token.code = T_OFFSET;
7104
7105 /* ??? This is not mentioned in the MASM grammar but gcc
7106 makes use of it with -mintel-syntax. OFFSET may be
7107 followed by FLAT: */
7108 if (strncasecmp (q, " FLAT:", 6) == 0)
7109 strcat (new_token.str, " FLAT:");
7110 }
7111
7112 /* ??? This is not mentioned in the MASM grammar. */
7113 else if (strcasecmp (new_token.str, "FLAT") == 0)
7114 {
7115 new_token.code = T_OFFSET;
7116 if (*q == ':')
7117 strcat (new_token.str, ":");
7118 else
7119 as_bad (_("`:' expected"));
7120 }
7121
7122 else
7123 new_token.code = T_ID;
7124 }
7125 }
7126
7127 else if (strchr ("+-/*%|&^:[]()~", *intel_parser.op_string))
7128 {
7129 new_token.code = *intel_parser.op_string;
7130 new_token.str[0] = *intel_parser.op_string;
7131 new_token.str[1] = '\0';
7132 }
7133
7134 else if (strchr ("<>", *intel_parser.op_string)
7135 && *intel_parser.op_string == *(intel_parser.op_string + 1))
7136 {
7137 new_token.code = *intel_parser.op_string == '<' ? T_SHL : T_SHR;
7138 new_token.str[0] = *intel_parser.op_string;
7139 new_token.str[1] = *intel_parser.op_string;
7140 new_token.str[2] = '\0';
7141 }
7142
7143 else
7144 as_bad (_("Unrecognized token `%s'"), intel_parser.op_string);
7145
7146 intel_parser.op_string += strlen (new_token.str);
7147 cur_token = new_token;
7148 }
7149
7150 /* Put cur_token back into the token stream and make cur_token point to
7151 prev_token. */
7152 static void
7153 intel_putback_token ()
7154 {
7155 if (cur_token.code != T_NIL)
7156 {
7157 intel_parser.op_string -= strlen (cur_token.str);
7158 free (cur_token.str);
7159 }
7160 cur_token = prev_token;
7161
7162 /* Forget prev_token. */
7163 prev_token.code = T_NIL;
7164 prev_token.reg = NULL;
7165 prev_token.str = NULL;
7166 }
7167
7168 int
7169 tc_x86_regname_to_dw2regnum (const char *regname)
7170 {
7171 unsigned int regnum;
7172 unsigned int regnames_count;
7173 static const char *const regnames_32[] =
7174 {
7175 "eax", "ecx", "edx", "ebx",
7176 "esp", "ebp", "esi", "edi",
7177 "eip", "eflags", NULL,
7178 "st0", "st1", "st2", "st3",
7179 "st4", "st5", "st6", "st7",
7180 NULL, NULL,
7181 "xmm0", "xmm1", "xmm2", "xmm3",
7182 "xmm4", "xmm5", "xmm6", "xmm7",
7183 "mm0", "mm1", "mm2", "mm3",
7184 "mm4", "mm5", "mm6", "mm7",
7185 "fcw", "fsw", "mxcsr",
7186 "es", "cs", "ss", "ds", "fs", "gs", NULL, NULL,
7187 "tr", "ldtr"
7188 };
7189 static const char *const regnames_64[] =
7190 {
7191 "rax", "rdx", "rcx", "rbx",
7192 "rsi", "rdi", "rbp", "rsp",
7193 "r8", "r9", "r10", "r11",
7194 "r12", "r13", "r14", "r15",
7195 "rip",
7196 "xmm0", "xmm1", "xmm2", "xmm3",
7197 "xmm4", "xmm5", "xmm6", "xmm7",
7198 "xmm8", "xmm9", "xmm10", "xmm11",
7199 "xmm12", "xmm13", "xmm14", "xmm15",
7200 "st0", "st1", "st2", "st3",
7201 "st4", "st5", "st6", "st7",
7202 "mm0", "mm1", "mm2", "mm3",
7203 "mm4", "mm5", "mm6", "mm7",
7204 "rflags",
7205 "es", "cs", "ss", "ds", "fs", "gs", NULL, NULL,
7206 "fs.base", "gs.base", NULL, NULL,
7207 "tr", "ldtr",
7208 "mxcsr", "fcw", "fsw"
7209 };
7210 const char *const *regnames;
7211
7212 if (flag_code == CODE_64BIT)
7213 {
7214 regnames = regnames_64;
7215 regnames_count = ARRAY_SIZE (regnames_64);
7216 }
7217 else
7218 {
7219 regnames = regnames_32;
7220 regnames_count = ARRAY_SIZE (regnames_32);
7221 }
7222
7223 for (regnum = 0; regnum < regnames_count; regnum++)
7224 if (regnames[regnum] != NULL
7225 && strcmp (regname, regnames[regnum]) == 0)
7226 return regnum;
7227
7228 return -1;
7229 }
7230
7231 void
7232 tc_x86_frame_initial_instructions (void)
7233 {
7234 static unsigned int sp_regno;
7235
7236 if (!sp_regno)
7237 sp_regno = tc_x86_regname_to_dw2regnum (flag_code == CODE_64BIT
7238 ? "rsp" : "esp");
7239
7240 cfi_add_CFA_def_cfa (sp_regno, -x86_cie_data_alignment);
7241 cfi_add_CFA_offset (x86_dwarf2_return_column, x86_cie_data_alignment);
7242 }
7243
7244 int
7245 i386_elf_section_type (const char *str, size_t len)
7246 {
7247 if (flag_code == CODE_64BIT
7248 && len == sizeof ("unwind") - 1
7249 && strncmp (str, "unwind", 6) == 0)
7250 return SHT_X86_64_UNWIND;
7251
7252 return -1;
7253 }
7254
7255 #ifdef TE_PE
7256 void
7257 tc_pe_dwarf2_emit_offset (symbolS *symbol, unsigned int size)
7258 {
7259 expressionS expr;
7260
7261 expr.X_op = O_secrel;
7262 expr.X_add_symbol = symbol;
7263 expr.X_add_number = 0;
7264 emit_expr (&expr, size);
7265 }
7266 #endif
7267
7268 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
7269 /* For ELF on x86-64, add support for SHF_X86_64_LARGE. */
7270
7271 int
7272 x86_64_section_letter (int letter, char **ptr_msg)
7273 {
7274 if (flag_code == CODE_64BIT)
7275 {
7276 if (letter == 'l')
7277 return SHF_X86_64_LARGE;
7278
7279 *ptr_msg = _("Bad .section directive: want a,l,w,x,M,S,G,T in string");
7280 }
7281 else
7282 *ptr_msg = _("Bad .section directive: want a,w,x,M,S,G,T in string");
7283 return -1;
7284 }
7285
7286 int
7287 x86_64_section_word (char *str, size_t len)
7288 {
7289 if (len == 5 && flag_code == CODE_64BIT && strncmp (str, "large", 5) == 0)
7290 return SHF_X86_64_LARGE;
7291
7292 return -1;
7293 }
7294
7295 static void
7296 handle_large_common (int small ATTRIBUTE_UNUSED)
7297 {
7298 if (flag_code != CODE_64BIT)
7299 {
7300 s_comm_internal (0, elf_common_parse);
7301 as_warn (_(".largecomm supported only in 64bit mode, producing .comm"));
7302 }
7303 else
7304 {
7305 static segT lbss_section;
7306 asection *saved_com_section_ptr = elf_com_section_ptr;
7307 asection *saved_bss_section = bss_section;
7308
7309 if (lbss_section == NULL)
7310 {
7311 flagword applicable;
7312 segT seg = now_seg;
7313 subsegT subseg = now_subseg;
7314
7315 /* The .lbss section is for local .largecomm symbols. */
7316 lbss_section = subseg_new (".lbss", 0);
7317 applicable = bfd_applicable_section_flags (stdoutput);
7318 bfd_set_section_flags (stdoutput, lbss_section,
7319 applicable & SEC_ALLOC);
7320 seg_info (lbss_section)->bss = 1;
7321
7322 subseg_set (seg, subseg);
7323 }
7324
7325 elf_com_section_ptr = &_bfd_elf_large_com_section;
7326 bss_section = lbss_section;
7327
7328 s_comm_internal (0, elf_common_parse);
7329
7330 elf_com_section_ptr = saved_com_section_ptr;
7331 bss_section = saved_bss_section;
7332 }
7333 }
7334 #endif /* OBJ_ELF || OBJ_MAYBE_ELF */
This page took 0.198609 seconds and 4 git commands to generate.