* tc-pdp11.c: Use VAX float format support for PDP-11 target.
[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
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, 59 Temple Place - Suite 330, Boston, MA
21 02111-1307, 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 Bugs & suggestions are completely welcome. This is free software.
27 Please help us make it better. */
28
29 #include "as.h"
30 #include "safe-ctype.h"
31 #include "subsegs.h"
32 #include "dwarf2dbg.h"
33 #include "opcode/i386.h"
34
35 #ifndef REGISTER_WARNINGS
36 #define REGISTER_WARNINGS 1
37 #endif
38
39 #ifndef INFER_ADDR_PREFIX
40 #define INFER_ADDR_PREFIX 1
41 #endif
42
43 #ifndef SCALE1_WHEN_NO_INDEX
44 /* Specifying a scale factor besides 1 when there is no index is
45 futile. eg. `mov (%ebx,2),%al' does exactly the same as
46 `mov (%ebx),%al'. To slavishly follow what the programmer
47 specified, set SCALE1_WHEN_NO_INDEX to 0. */
48 #define SCALE1_WHEN_NO_INDEX 1
49 #endif
50
51 #ifndef true
52 #define true 1
53 #endif
54 #ifndef false
55 #define false 0
56 #endif
57
58 static unsigned int mode_from_disp_size PARAMS ((unsigned int));
59 static int fits_in_signed_byte PARAMS ((offsetT));
60 static int fits_in_unsigned_byte PARAMS ((offsetT));
61 static int fits_in_unsigned_word PARAMS ((offsetT));
62 static int fits_in_signed_word PARAMS ((offsetT));
63 static int fits_in_unsigned_long PARAMS ((offsetT));
64 static int fits_in_signed_long PARAMS ((offsetT));
65 static int smallest_imm_type PARAMS ((offsetT));
66 static offsetT offset_in_range PARAMS ((offsetT, int));
67 static int add_prefix PARAMS ((unsigned int));
68 static void set_code_flag PARAMS ((int));
69 static void set_16bit_gcc_code_flag PARAMS ((int));
70 static void set_intel_syntax PARAMS ((int));
71 static void set_cpu_arch PARAMS ((int));
72
73 #ifdef BFD_ASSEMBLER
74 static bfd_reloc_code_real_type reloc
75 PARAMS ((int, int, int, bfd_reloc_code_real_type));
76 #define RELOC_ENUM enum bfd_reloc_code_real
77 #else
78 #define RELOC_ENUM int
79 #endif
80
81 #ifndef DEFAULT_ARCH
82 #define DEFAULT_ARCH "i386"
83 #endif
84 static const char *default_arch = DEFAULT_ARCH;
85
86 /* 'md_assemble ()' gathers together information and puts it into a
87 i386_insn. */
88
89 union i386_op
90 {
91 expressionS *disps;
92 expressionS *imms;
93 const reg_entry *regs;
94 };
95
96 struct _i386_insn
97 {
98 /* TM holds the template for the insn were currently assembling. */
99 template tm;
100
101 /* SUFFIX holds the instruction mnemonic suffix if given.
102 (e.g. 'l' for 'movl') */
103 char suffix;
104
105 /* OPERANDS gives the number of given operands. */
106 unsigned int operands;
107
108 /* REG_OPERANDS, DISP_OPERANDS, MEM_OPERANDS, IMM_OPERANDS give the number
109 of given register, displacement, memory operands and immediate
110 operands. */
111 unsigned int reg_operands, disp_operands, mem_operands, imm_operands;
112
113 /* TYPES [i] is the type (see above #defines) which tells us how to
114 use OP[i] for the corresponding operand. */
115 unsigned int types[MAX_OPERANDS];
116
117 /* Displacement expression, immediate expression, or register for each
118 operand. */
119 union i386_op op[MAX_OPERANDS];
120
121 /* Flags for operands. */
122 unsigned int flags[MAX_OPERANDS];
123 #define Operand_PCrel 1
124
125 /* Relocation type for operand */
126 RELOC_ENUM reloc[MAX_OPERANDS];
127
128 /* BASE_REG, INDEX_REG, and LOG2_SCALE_FACTOR are used to encode
129 the base index byte below. */
130 const reg_entry *base_reg;
131 const reg_entry *index_reg;
132 unsigned int log2_scale_factor;
133
134 /* SEG gives the seg_entries of this insn. They are zero unless
135 explicit segment overrides are given. */
136 const seg_entry *seg[2];
137
138 /* PREFIX holds all the given prefix opcodes (usually null).
139 PREFIXES is the number of prefix opcodes. */
140 unsigned int prefixes;
141 unsigned char prefix[MAX_PREFIXES];
142
143 /* RM and SIB are the modrm byte and the sib byte where the
144 addressing modes of this insn are encoded. */
145
146 modrm_byte rm;
147 rex_byte rex;
148 sib_byte sib;
149 };
150
151 typedef struct _i386_insn i386_insn;
152
153 /* List of chars besides those in app.c:symbol_chars that can start an
154 operand. Used to prevent the scrubber eating vital white-space. */
155 #ifdef LEX_AT
156 const char extra_symbol_chars[] = "*%-(@";
157 #else
158 const char extra_symbol_chars[] = "*%-(";
159 #endif
160
161 /* This array holds the chars that always start a comment. If the
162 pre-processor is disabled, these aren't very useful. */
163 #if defined (TE_I386AIX) || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) && ! defined (TE_LINUX) && !defined(TE_FreeBSD) && !defined(TE_NetBSD))
164 /* Putting '/' here makes it impossible to use the divide operator.
165 However, we need it for compatibility with SVR4 systems. */
166 const char comment_chars[] = "#/";
167 #define PREFIX_SEPARATOR '\\'
168 #else
169 const char comment_chars[] = "#";
170 #define PREFIX_SEPARATOR '/'
171 #endif
172
173 /* This array holds the chars that only start a comment at the beginning of
174 a line. If the line seems to have the form '# 123 filename'
175 .line and .file directives will appear in the pre-processed output.
176 Note that input_file.c hand checks for '#' at the beginning of the
177 first line of the input file. This is because the compiler outputs
178 #NO_APP at the beginning of its output.
179 Also note that comments started like this one will always work if
180 '/' isn't otherwise defined. */
181 #if defined (TE_I386AIX) || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) && ! defined (TE_LINUX) && !defined(TE_FreeBSD) && !defined(TE_NetBSD))
182 const char line_comment_chars[] = "";
183 #else
184 const char line_comment_chars[] = "/";
185 #endif
186
187 const char line_separator_chars[] = ";";
188
189 /* Chars that can be used to separate mant from exp in floating point
190 nums. */
191 const char EXP_CHARS[] = "eE";
192
193 /* Chars that mean this number is a floating point constant
194 As in 0f12.456
195 or 0d1.2345e12. */
196 const char FLT_CHARS[] = "fFdDxX";
197
198 /* Tables for lexical analysis. */
199 static char mnemonic_chars[256];
200 static char register_chars[256];
201 static char operand_chars[256];
202 static char identifier_chars[256];
203 static char digit_chars[256];
204
205 /* Lexical macros. */
206 #define is_mnemonic_char(x) (mnemonic_chars[(unsigned char) x])
207 #define is_operand_char(x) (operand_chars[(unsigned char) x])
208 #define is_register_char(x) (register_chars[(unsigned char) x])
209 #define is_space_char(x) ((x) == ' ')
210 #define is_identifier_char(x) (identifier_chars[(unsigned char) x])
211 #define is_digit_char(x) (digit_chars[(unsigned char) x])
212
213 /* All non-digit non-letter charcters that may occur in an operand. */
214 static char operand_special_chars[] = "%$-+(,)*._~/<>|&^!:[@]";
215
216 /* md_assemble() always leaves the strings it's passed unaltered. To
217 effect this we maintain a stack of saved characters that we've smashed
218 with '\0's (indicating end of strings for various sub-fields of the
219 assembler instruction). */
220 static char save_stack[32];
221 static char *save_stack_p;
222 #define END_STRING_AND_SAVE(s) \
223 do { *save_stack_p++ = *(s); *(s) = '\0'; } while (0)
224 #define RESTORE_END_STRING(s) \
225 do { *(s) = *--save_stack_p; } while (0)
226
227 /* The instruction we're assembling. */
228 static i386_insn i;
229
230 /* Possible templates for current insn. */
231 static const templates *current_templates;
232
233 /* Per instruction expressionS buffers: 2 displacements & 2 immediate max. */
234 static expressionS disp_expressions[2], im_expressions[2];
235
236 /* Current operand we are working on. */
237 static int this_operand;
238
239 /* We support four different modes. FLAG_CODE variable is used to distinguish
240 these. */
241
242 enum flag_code {
243 CODE_32BIT,
244 CODE_16BIT,
245 CODE_64BIT };
246 #define NUM_FLAG_CODE ((int) CODE_64BIT + 1)
247
248 static enum flag_code flag_code;
249 static int use_rela_relocations = 0;
250
251 /* The names used to print error messages. */
252 static const char *flag_code_names[] =
253 {
254 "32",
255 "16",
256 "64"
257 };
258
259 /* 1 for intel syntax,
260 0 if att syntax. */
261 static int intel_syntax = 0;
262
263 /* 1 if register prefix % not required. */
264 static int allow_naked_reg = 0;
265
266 /* Used in 16 bit gcc mode to add an l suffix to call, ret, enter,
267 leave, push, and pop instructions so that gcc has the same stack
268 frame as in 32 bit mode. */
269 static char stackop_size = '\0';
270
271 /* Non-zero to quieten some warnings. */
272 static int quiet_warnings = 0;
273
274 /* CPU name. */
275 static const char *cpu_arch_name = NULL;
276
277 /* CPU feature flags. */
278 static unsigned int cpu_arch_flags = CpuUnknownFlags|CpuNo64;
279
280 /* If set, conditional jumps are not automatically promoted to handle
281 larger than a byte offset. */
282 static unsigned int no_cond_jump_promotion = 0;
283
284 /* Interface to relax_segment.
285 There are 3 major relax states for 386 jump insns because the
286 different types of jumps add different sizes to frags when we're
287 figuring out what sort of jump to choose to reach a given label. */
288
289 /* Types. */
290 #define UNCOND_JUMP 0
291 #define COND_JUMP 1
292 #define COND_JUMP86 2
293
294 /* Sizes. */
295 #define CODE16 1
296 #define SMALL 0
297 #define SMALL16 (SMALL|CODE16)
298 #define BIG 2
299 #define BIG16 (BIG|CODE16)
300
301 #ifndef INLINE
302 #ifdef __GNUC__
303 #define INLINE __inline__
304 #else
305 #define INLINE
306 #endif
307 #endif
308
309 #define ENCODE_RELAX_STATE(type, size) \
310 ((relax_substateT) (((type) << 2) | (size)))
311 #define TYPE_FROM_RELAX_STATE(s) \
312 ((s) >> 2)
313 #define DISP_SIZE_FROM_RELAX_STATE(s) \
314 ((((s) & 3) == BIG ? 4 : (((s) & 3) == BIG16 ? 2 : 1)))
315
316 /* This table is used by relax_frag to promote short jumps to long
317 ones where necessary. SMALL (short) jumps may be promoted to BIG
318 (32 bit long) ones, and SMALL16 jumps to BIG16 (16 bit long). We
319 don't allow a short jump in a 32 bit code segment to be promoted to
320 a 16 bit offset jump because it's slower (requires data size
321 prefix), and doesn't work, unless the destination is in the bottom
322 64k of the code segment (The top 16 bits of eip are zeroed). */
323
324 const relax_typeS md_relax_table[] =
325 {
326 /* The fields are:
327 1) most positive reach of this state,
328 2) most negative reach of this state,
329 3) how many bytes this mode will have in the variable part of the frag
330 4) which index into the table to try if we can't fit into this one. */
331
332 /* UNCOND_JUMP states. */
333 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG)},
334 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16)},
335 /* dword jmp adds 4 bytes to frag:
336 0 extra opcode bytes, 4 displacement bytes. */
337 {0, 0, 4, 0},
338 /* word jmp adds 2 byte2 to frag:
339 0 extra opcode bytes, 2 displacement bytes. */
340 {0, 0, 2, 0},
341
342 /* COND_JUMP states. */
343 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG)},
344 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG16)},
345 /* dword conditionals adds 5 bytes to frag:
346 1 extra opcode byte, 4 displacement bytes. */
347 {0, 0, 5, 0},
348 /* word conditionals add 3 bytes to frag:
349 1 extra opcode byte, 2 displacement bytes. */
350 {0, 0, 3, 0},
351
352 /* COND_JUMP86 states. */
353 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG)},
354 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG16)},
355 /* dword conditionals adds 5 bytes to frag:
356 1 extra opcode byte, 4 displacement bytes. */
357 {0, 0, 5, 0},
358 /* word conditionals add 4 bytes to frag:
359 1 displacement byte and a 3 byte long branch insn. */
360 {0, 0, 4, 0}
361 };
362
363 static const arch_entry cpu_arch[] = {
364 {"i8086", Cpu086 },
365 {"i186", Cpu086|Cpu186 },
366 {"i286", Cpu086|Cpu186|Cpu286 },
367 {"i386", Cpu086|Cpu186|Cpu286|Cpu386 },
368 {"i486", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486 },
369 {"i586", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
370 {"i686", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
371 {"pentium", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
372 {"pentiumpro",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
373 {"pentium4", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX|CpuSSE|CpuSSE2 },
374 {"k6", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX|Cpu3dnow },
375 {"athlon", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuMMX|Cpu3dnow },
376 {"sledgehammer",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuSledgehammer|CpuMMX|Cpu3dnow|CpuSSE|CpuSSE2 },
377 {NULL, 0 }
378 };
379
380 void
381 i386_align_code (fragP, count)
382 fragS *fragP;
383 int count;
384 {
385 /* Various efficient no-op patterns for aligning code labels.
386 Note: Don't try to assemble the instructions in the comments.
387 0L and 0w are not legal. */
388 static const char f32_1[] =
389 {0x90}; /* nop */
390 static const char f32_2[] =
391 {0x89,0xf6}; /* movl %esi,%esi */
392 static const char f32_3[] =
393 {0x8d,0x76,0x00}; /* leal 0(%esi),%esi */
394 static const char f32_4[] =
395 {0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
396 static const char f32_5[] =
397 {0x90, /* nop */
398 0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
399 static const char f32_6[] =
400 {0x8d,0xb6,0x00,0x00,0x00,0x00}; /* leal 0L(%esi),%esi */
401 static const char f32_7[] =
402 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
403 static const char f32_8[] =
404 {0x90, /* nop */
405 0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
406 static const char f32_9[] =
407 {0x89,0xf6, /* movl %esi,%esi */
408 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
409 static const char f32_10[] =
410 {0x8d,0x76,0x00, /* leal 0(%esi),%esi */
411 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
412 static const char f32_11[] =
413 {0x8d,0x74,0x26,0x00, /* leal 0(%esi,1),%esi */
414 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
415 static const char f32_12[] =
416 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
417 0x8d,0xbf,0x00,0x00,0x00,0x00}; /* leal 0L(%edi),%edi */
418 static const char f32_13[] =
419 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
420 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
421 static const char f32_14[] =
422 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00, /* leal 0L(%esi,1),%esi */
423 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
424 static const char f32_15[] =
425 {0xeb,0x0d,0x90,0x90,0x90,0x90,0x90, /* jmp .+15; lotsa nops */
426 0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90};
427 static const char f16_3[] =
428 {0x8d,0x74,0x00}; /* lea 0(%esi),%esi */
429 static const char f16_4[] =
430 {0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
431 static const char f16_5[] =
432 {0x90, /* nop */
433 0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
434 static const char f16_6[] =
435 {0x89,0xf6, /* mov %si,%si */
436 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
437 static const char f16_7[] =
438 {0x8d,0x74,0x00, /* lea 0(%si),%si */
439 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
440 static const char f16_8[] =
441 {0x8d,0xb4,0x00,0x00, /* lea 0w(%si),%si */
442 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
443 static const char *const f32_patt[] = {
444 f32_1, f32_2, f32_3, f32_4, f32_5, f32_6, f32_7, f32_8,
445 f32_9, f32_10, f32_11, f32_12, f32_13, f32_14, f32_15
446 };
447 static const char *const f16_patt[] = {
448 f32_1, f32_2, f16_3, f16_4, f16_5, f16_6, f16_7, f16_8,
449 f32_15, f32_15, f32_15, f32_15, f32_15, f32_15, f32_15
450 };
451
452 /* ??? We can't use these fillers for x86_64, since they often kills the
453 upper halves. Solve later. */
454 if (flag_code == CODE_64BIT)
455 count = 1;
456
457 if (count > 0 && count <= 15)
458 {
459 if (flag_code == CODE_16BIT)
460 {
461 memcpy (fragP->fr_literal + fragP->fr_fix,
462 f16_patt[count - 1], count);
463 if (count > 8)
464 /* Adjust jump offset. */
465 fragP->fr_literal[fragP->fr_fix + 1] = count - 2;
466 }
467 else
468 memcpy (fragP->fr_literal + fragP->fr_fix,
469 f32_patt[count - 1], count);
470 fragP->fr_var = count;
471 }
472 }
473
474 static char *output_invalid PARAMS ((int c));
475 static int i386_operand PARAMS ((char *operand_string));
476 static int i386_intel_operand PARAMS ((char *operand_string, int got_a_float));
477 static const reg_entry *parse_register PARAMS ((char *reg_string,
478 char **end_op));
479
480 #ifndef I386COFF
481 static void s_bss PARAMS ((int));
482 #endif
483
484 symbolS *GOT_symbol; /* Pre-defined "_GLOBAL_OFFSET_TABLE_". */
485
486 static INLINE unsigned int
487 mode_from_disp_size (t)
488 unsigned int t;
489 {
490 return (t & Disp8) ? 1 : (t & (Disp16 | Disp32 | Disp32S)) ? 2 : 0;
491 }
492
493 static INLINE int
494 fits_in_signed_byte (num)
495 offsetT num;
496 {
497 return (num >= -128) && (num <= 127);
498 }
499
500 static INLINE int
501 fits_in_unsigned_byte (num)
502 offsetT num;
503 {
504 return (num & 0xff) == num;
505 }
506
507 static INLINE int
508 fits_in_unsigned_word (num)
509 offsetT num;
510 {
511 return (num & 0xffff) == num;
512 }
513
514 static INLINE int
515 fits_in_signed_word (num)
516 offsetT num;
517 {
518 return (-32768 <= num) && (num <= 32767);
519 }
520 static INLINE int
521 fits_in_signed_long (num)
522 offsetT num ATTRIBUTE_UNUSED;
523 {
524 #ifndef BFD64
525 return 1;
526 #else
527 return (!(((offsetT) -1 << 31) & num)
528 || (((offsetT) -1 << 31) & num) == ((offsetT) -1 << 31));
529 #endif
530 } /* fits_in_signed_long() */
531 static INLINE int
532 fits_in_unsigned_long (num)
533 offsetT num ATTRIBUTE_UNUSED;
534 {
535 #ifndef BFD64
536 return 1;
537 #else
538 return (num & (((offsetT) 2 << 31) - 1)) == num;
539 #endif
540 } /* fits_in_unsigned_long() */
541
542 static int
543 smallest_imm_type (num)
544 offsetT num;
545 {
546 if (cpu_arch_flags != (Cpu086 | Cpu186 | Cpu286 | Cpu386 | Cpu486 | CpuNo64))
547 {
548 /* This code is disabled on the 486 because all the Imm1 forms
549 in the opcode table are slower on the i486. They're the
550 versions with the implicitly specified single-position
551 displacement, which has another syntax if you really want to
552 use that form. */
553 if (num == 1)
554 return Imm1 | Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64;
555 }
556 return (fits_in_signed_byte (num)
557 ? (Imm8S | Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
558 : fits_in_unsigned_byte (num)
559 ? (Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
560 : (fits_in_signed_word (num) || fits_in_unsigned_word (num))
561 ? (Imm16 | Imm32 | Imm32S | Imm64)
562 : fits_in_signed_long (num)
563 ? (Imm32 | Imm32S | Imm64)
564 : fits_in_unsigned_long (num)
565 ? (Imm32 | Imm64)
566 : Imm64);
567 }
568
569 static offsetT
570 offset_in_range (val, size)
571 offsetT val;
572 int size;
573 {
574 addressT mask;
575
576 switch (size)
577 {
578 case 1: mask = ((addressT) 1 << 8) - 1; break;
579 case 2: mask = ((addressT) 1 << 16) - 1; break;
580 case 4: mask = ((addressT) 2 << 31) - 1; break;
581 #ifdef BFD64
582 case 8: mask = ((addressT) 2 << 63) - 1; break;
583 #endif
584 default: abort ();
585 }
586
587 /* If BFD64, sign extend val. */
588 if (!use_rela_relocations)
589 if ((val & ~(((addressT) 2 << 31) - 1)) == 0)
590 val = (val ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
591
592 if ((val & ~mask) != 0 && (val & ~mask) != ~mask)
593 {
594 char buf1[40], buf2[40];
595
596 sprint_value (buf1, val);
597 sprint_value (buf2, val & mask);
598 as_warn (_("%s shortened to %s"), buf1, buf2);
599 }
600 return val & mask;
601 }
602
603 /* Returns 0 if attempting to add a prefix where one from the same
604 class already exists, 1 if non rep/repne added, 2 if rep/repne
605 added. */
606 static int
607 add_prefix (prefix)
608 unsigned int prefix;
609 {
610 int ret = 1;
611 int q;
612
613 if (prefix >= 0x40 && prefix < 0x50 && flag_code == CODE_64BIT)
614 q = REX_PREFIX;
615 else
616 switch (prefix)
617 {
618 default:
619 abort ();
620
621 case CS_PREFIX_OPCODE:
622 case DS_PREFIX_OPCODE:
623 case ES_PREFIX_OPCODE:
624 case FS_PREFIX_OPCODE:
625 case GS_PREFIX_OPCODE:
626 case SS_PREFIX_OPCODE:
627 q = SEG_PREFIX;
628 break;
629
630 case REPNE_PREFIX_OPCODE:
631 case REPE_PREFIX_OPCODE:
632 ret = 2;
633 /* fall thru */
634 case LOCK_PREFIX_OPCODE:
635 q = LOCKREP_PREFIX;
636 break;
637
638 case FWAIT_OPCODE:
639 q = WAIT_PREFIX;
640 break;
641
642 case ADDR_PREFIX_OPCODE:
643 q = ADDR_PREFIX;
644 break;
645
646 case DATA_PREFIX_OPCODE:
647 q = DATA_PREFIX;
648 break;
649 }
650
651 if (i.prefix[q])
652 {
653 as_bad (_("same type of prefix used twice"));
654 return 0;
655 }
656
657 i.prefixes += 1;
658 i.prefix[q] = prefix;
659 return ret;
660 }
661
662 static void
663 set_code_flag (value)
664 int value;
665 {
666 flag_code = value;
667 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
668 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
669 if (value == CODE_64BIT && !(cpu_arch_flags & CpuSledgehammer))
670 {
671 as_bad (_("64bit mode not supported on this CPU."));
672 }
673 if (value == CODE_32BIT && !(cpu_arch_flags & Cpu386))
674 {
675 as_bad (_("32bit mode not supported on this CPU."));
676 }
677 stackop_size = '\0';
678 }
679
680 static void
681 set_16bit_gcc_code_flag (new_code_flag)
682 int new_code_flag;
683 {
684 flag_code = new_code_flag;
685 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
686 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
687 stackop_size = 'l';
688 }
689
690 static void
691 set_intel_syntax (syntax_flag)
692 int syntax_flag;
693 {
694 /* Find out if register prefixing is specified. */
695 int ask_naked_reg = 0;
696
697 SKIP_WHITESPACE ();
698 if (! is_end_of_line[(unsigned char) *input_line_pointer])
699 {
700 char *string = input_line_pointer;
701 int e = get_symbol_end ();
702
703 if (strcmp (string, "prefix") == 0)
704 ask_naked_reg = 1;
705 else if (strcmp (string, "noprefix") == 0)
706 ask_naked_reg = -1;
707 else
708 as_bad (_("bad argument to syntax directive."));
709 *input_line_pointer = e;
710 }
711 demand_empty_rest_of_line ();
712
713 intel_syntax = syntax_flag;
714
715 if (ask_naked_reg == 0)
716 {
717 #ifdef BFD_ASSEMBLER
718 allow_naked_reg = (intel_syntax
719 && (bfd_get_symbol_leading_char (stdoutput) != '\0'));
720 #else
721 /* Conservative default. */
722 allow_naked_reg = 0;
723 #endif
724 }
725 else
726 allow_naked_reg = (ask_naked_reg < 0);
727 }
728
729 static void
730 set_cpu_arch (dummy)
731 int dummy ATTRIBUTE_UNUSED;
732 {
733 SKIP_WHITESPACE ();
734
735 if (! is_end_of_line[(unsigned char) *input_line_pointer])
736 {
737 char *string = input_line_pointer;
738 int e = get_symbol_end ();
739 int i;
740
741 for (i = 0; cpu_arch[i].name; i++)
742 {
743 if (strcmp (string, cpu_arch[i].name) == 0)
744 {
745 cpu_arch_name = cpu_arch[i].name;
746 cpu_arch_flags = (cpu_arch[i].flags
747 | (flag_code == CODE_64BIT ? Cpu64 : CpuNo64));
748 break;
749 }
750 }
751 if (!cpu_arch[i].name)
752 as_bad (_("no such architecture: `%s'"), string);
753
754 *input_line_pointer = e;
755 }
756 else
757 as_bad (_("missing cpu architecture"));
758
759 no_cond_jump_promotion = 0;
760 if (*input_line_pointer == ','
761 && ! is_end_of_line[(unsigned char) input_line_pointer[1]])
762 {
763 char *string = ++input_line_pointer;
764 int e = get_symbol_end ();
765
766 if (strcmp (string, "nojumps") == 0)
767 no_cond_jump_promotion = 1;
768 else if (strcmp (string, "jumps") == 0)
769 ;
770 else
771 as_bad (_("no such architecture modifier: `%s'"), string);
772
773 *input_line_pointer = e;
774 }
775
776 demand_empty_rest_of_line ();
777 }
778
779 const pseudo_typeS md_pseudo_table[] =
780 {
781 #if !defined(OBJ_AOUT) && !defined(USE_ALIGN_PTWO)
782 {"align", s_align_bytes, 0},
783 #else
784 {"align", s_align_ptwo, 0},
785 #endif
786 {"arch", set_cpu_arch, 0},
787 #ifndef I386COFF
788 {"bss", s_bss, 0},
789 #endif
790 {"ffloat", float_cons, 'f'},
791 {"dfloat", float_cons, 'd'},
792 {"tfloat", float_cons, 'x'},
793 {"value", cons, 2},
794 {"noopt", s_ignore, 0},
795 {"optim", s_ignore, 0},
796 {"code16gcc", set_16bit_gcc_code_flag, CODE_16BIT},
797 {"code16", set_code_flag, CODE_16BIT},
798 {"code32", set_code_flag, CODE_32BIT},
799 {"code64", set_code_flag, CODE_64BIT},
800 {"intel_syntax", set_intel_syntax, 1},
801 {"att_syntax", set_intel_syntax, 0},
802 {"file", dwarf2_directive_file, 0},
803 {"loc", dwarf2_directive_loc, 0},
804 {0, 0, 0}
805 };
806
807 /* For interface with expression (). */
808 extern char *input_line_pointer;
809
810 /* Hash table for instruction mnemonic lookup. */
811 static struct hash_control *op_hash;
812
813 /* Hash table for register lookup. */
814 static struct hash_control *reg_hash;
815 \f
816 #ifdef BFD_ASSEMBLER
817 unsigned long
818 i386_mach ()
819 {
820 if (!strcmp (default_arch, "x86_64"))
821 return bfd_mach_x86_64;
822 else if (!strcmp (default_arch, "i386"))
823 return bfd_mach_i386_i386;
824 else
825 as_fatal (_("Unknown architecture"));
826 }
827 #endif
828 \f
829 void
830 md_begin ()
831 {
832 const char *hash_err;
833
834 /* Initialize op_hash hash table. */
835 op_hash = hash_new ();
836
837 {
838 register const template *optab;
839 register templates *core_optab;
840
841 /* Setup for loop. */
842 optab = i386_optab;
843 core_optab = (templates *) xmalloc (sizeof (templates));
844 core_optab->start = optab;
845
846 while (1)
847 {
848 ++optab;
849 if (optab->name == NULL
850 || strcmp (optab->name, (optab - 1)->name) != 0)
851 {
852 /* different name --> ship out current template list;
853 add to hash table; & begin anew. */
854 core_optab->end = optab;
855 hash_err = hash_insert (op_hash,
856 (optab - 1)->name,
857 (PTR) core_optab);
858 if (hash_err)
859 {
860 as_fatal (_("Internal Error: Can't hash %s: %s"),
861 (optab - 1)->name,
862 hash_err);
863 }
864 if (optab->name == NULL)
865 break;
866 core_optab = (templates *) xmalloc (sizeof (templates));
867 core_optab->start = optab;
868 }
869 }
870 }
871
872 /* Initialize reg_hash hash table. */
873 reg_hash = hash_new ();
874 {
875 register const reg_entry *regtab;
876
877 for (regtab = i386_regtab;
878 regtab < i386_regtab + sizeof (i386_regtab) / sizeof (i386_regtab[0]);
879 regtab++)
880 {
881 hash_err = hash_insert (reg_hash, regtab->reg_name, (PTR) regtab);
882 if (hash_err)
883 as_fatal (_("Internal Error: Can't hash %s: %s"),
884 regtab->reg_name,
885 hash_err);
886 }
887 }
888
889 /* Fill in lexical tables: mnemonic_chars, operand_chars. */
890 {
891 register int c;
892 register char *p;
893
894 for (c = 0; c < 256; c++)
895 {
896 if (ISDIGIT (c))
897 {
898 digit_chars[c] = c;
899 mnemonic_chars[c] = c;
900 register_chars[c] = c;
901 operand_chars[c] = c;
902 }
903 else if (ISLOWER (c))
904 {
905 mnemonic_chars[c] = c;
906 register_chars[c] = c;
907 operand_chars[c] = c;
908 }
909 else if (ISUPPER (c))
910 {
911 mnemonic_chars[c] = TOLOWER (c);
912 register_chars[c] = mnemonic_chars[c];
913 operand_chars[c] = c;
914 }
915
916 if (ISALPHA (c) || ISDIGIT (c))
917 identifier_chars[c] = c;
918 else if (c >= 128)
919 {
920 identifier_chars[c] = c;
921 operand_chars[c] = c;
922 }
923 }
924
925 #ifdef LEX_AT
926 identifier_chars['@'] = '@';
927 #endif
928 digit_chars['-'] = '-';
929 identifier_chars['_'] = '_';
930 identifier_chars['.'] = '.';
931
932 for (p = operand_special_chars; *p != '\0'; p++)
933 operand_chars[(unsigned char) *p] = *p;
934 }
935
936 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
937 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
938 {
939 record_alignment (text_section, 2);
940 record_alignment (data_section, 2);
941 record_alignment (bss_section, 2);
942 }
943 #endif
944 }
945
946 void
947 i386_print_statistics (file)
948 FILE *file;
949 {
950 hash_print_statistics (file, "i386 opcode", op_hash);
951 hash_print_statistics (file, "i386 register", reg_hash);
952 }
953 \f
954 #ifdef DEBUG386
955
956 /* Debugging routines for md_assemble. */
957 static void pi PARAMS ((char *, i386_insn *));
958 static void pte PARAMS ((template *));
959 static void pt PARAMS ((unsigned int));
960 static void pe PARAMS ((expressionS *));
961 static void ps PARAMS ((symbolS *));
962
963 static void
964 pi (line, x)
965 char *line;
966 i386_insn *x;
967 {
968 unsigned int i;
969
970 fprintf (stdout, "%s: template ", line);
971 pte (&x->tm);
972 fprintf (stdout, " address: base %s index %s scale %x\n",
973 x->base_reg ? x->base_reg->reg_name : "none",
974 x->index_reg ? x->index_reg->reg_name : "none",
975 x->log2_scale_factor);
976 fprintf (stdout, " modrm: mode %x reg %x reg/mem %x\n",
977 x->rm.mode, x->rm.reg, x->rm.regmem);
978 fprintf (stdout, " sib: base %x index %x scale %x\n",
979 x->sib.base, x->sib.index, x->sib.scale);
980 fprintf (stdout, " rex: 64bit %x extX %x extY %x extZ %x\n",
981 x->rex.mode64, x->rex.extX, x->rex.extY, x->rex.extZ);
982 for (i = 0; i < x->operands; i++)
983 {
984 fprintf (stdout, " #%d: ", i + 1);
985 pt (x->types[i]);
986 fprintf (stdout, "\n");
987 if (x->types[i]
988 & (Reg | SReg2 | SReg3 | Control | Debug | Test | RegMMX | RegXMM))
989 fprintf (stdout, "%s\n", x->op[i].regs->reg_name);
990 if (x->types[i] & Imm)
991 pe (x->op[i].imms);
992 if (x->types[i] & Disp)
993 pe (x->op[i].disps);
994 }
995 }
996
997 static void
998 pte (t)
999 template *t;
1000 {
1001 unsigned int i;
1002 fprintf (stdout, " %d operands ", t->operands);
1003 fprintf (stdout, "opcode %x ", t->base_opcode);
1004 if (t->extension_opcode != None)
1005 fprintf (stdout, "ext %x ", t->extension_opcode);
1006 if (t->opcode_modifier & D)
1007 fprintf (stdout, "D");
1008 if (t->opcode_modifier & W)
1009 fprintf (stdout, "W");
1010 fprintf (stdout, "\n");
1011 for (i = 0; i < t->operands; i++)
1012 {
1013 fprintf (stdout, " #%d type ", i + 1);
1014 pt (t->operand_types[i]);
1015 fprintf (stdout, "\n");
1016 }
1017 }
1018
1019 static void
1020 pe (e)
1021 expressionS *e;
1022 {
1023 fprintf (stdout, " operation %d\n", e->X_op);
1024 fprintf (stdout, " add_number %ld (%lx)\n",
1025 (long) e->X_add_number, (long) e->X_add_number);
1026 if (e->X_add_symbol)
1027 {
1028 fprintf (stdout, " add_symbol ");
1029 ps (e->X_add_symbol);
1030 fprintf (stdout, "\n");
1031 }
1032 if (e->X_op_symbol)
1033 {
1034 fprintf (stdout, " op_symbol ");
1035 ps (e->X_op_symbol);
1036 fprintf (stdout, "\n");
1037 }
1038 }
1039
1040 static void
1041 ps (s)
1042 symbolS *s;
1043 {
1044 fprintf (stdout, "%s type %s%s",
1045 S_GET_NAME (s),
1046 S_IS_EXTERNAL (s) ? "EXTERNAL " : "",
1047 segment_name (S_GET_SEGMENT (s)));
1048 }
1049
1050 struct type_name
1051 {
1052 unsigned int mask;
1053 char *tname;
1054 }
1055
1056 type_names[] =
1057 {
1058 { Reg8, "r8" },
1059 { Reg16, "r16" },
1060 { Reg32, "r32" },
1061 { Reg64, "r64" },
1062 { Imm8, "i8" },
1063 { Imm8S, "i8s" },
1064 { Imm16, "i16" },
1065 { Imm32, "i32" },
1066 { Imm32S, "i32s" },
1067 { Imm64, "i64" },
1068 { Imm1, "i1" },
1069 { BaseIndex, "BaseIndex" },
1070 { Disp8, "d8" },
1071 { Disp16, "d16" },
1072 { Disp32, "d32" },
1073 { Disp32S, "d32s" },
1074 { Disp64, "d64" },
1075 { InOutPortReg, "InOutPortReg" },
1076 { ShiftCount, "ShiftCount" },
1077 { Control, "control reg" },
1078 { Test, "test reg" },
1079 { Debug, "debug reg" },
1080 { FloatReg, "FReg" },
1081 { FloatAcc, "FAcc" },
1082 { SReg2, "SReg2" },
1083 { SReg3, "SReg3" },
1084 { Acc, "Acc" },
1085 { JumpAbsolute, "Jump Absolute" },
1086 { RegMMX, "rMMX" },
1087 { RegXMM, "rXMM" },
1088 { EsSeg, "es" },
1089 { 0, "" }
1090 };
1091
1092 static void
1093 pt (t)
1094 unsigned int t;
1095 {
1096 register struct type_name *ty;
1097
1098 for (ty = type_names; ty->mask; ty++)
1099 if (t & ty->mask)
1100 fprintf (stdout, "%s, ", ty->tname);
1101 fflush (stdout);
1102 }
1103
1104 #endif /* DEBUG386 */
1105 \f
1106 int
1107 tc_i386_force_relocation (fixp)
1108 struct fix *fixp;
1109 {
1110 #ifdef BFD_ASSEMBLER
1111 if (fixp->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1112 || fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1113 return 1;
1114 return 0;
1115 #else
1116 /* For COFF. */
1117 return fixp->fx_r_type == 7;
1118 #endif
1119 }
1120
1121 #ifdef BFD_ASSEMBLER
1122
1123 static bfd_reloc_code_real_type
1124 reloc (size, pcrel, sign, other)
1125 int size;
1126 int pcrel;
1127 int sign;
1128 bfd_reloc_code_real_type other;
1129 {
1130 if (other != NO_RELOC)
1131 return other;
1132
1133 if (pcrel)
1134 {
1135 if (!sign)
1136 as_bad (_("There are no unsigned pc-relative relocations"));
1137 switch (size)
1138 {
1139 case 1: return BFD_RELOC_8_PCREL;
1140 case 2: return BFD_RELOC_16_PCREL;
1141 case 4: return BFD_RELOC_32_PCREL;
1142 }
1143 as_bad (_("can not do %d byte pc-relative relocation"), size);
1144 }
1145 else
1146 {
1147 if (sign)
1148 switch (size)
1149 {
1150 case 4: return BFD_RELOC_X86_64_32S;
1151 }
1152 else
1153 switch (size)
1154 {
1155 case 1: return BFD_RELOC_8;
1156 case 2: return BFD_RELOC_16;
1157 case 4: return BFD_RELOC_32;
1158 case 8: return BFD_RELOC_64;
1159 }
1160 as_bad (_("can not do %s %d byte relocation"),
1161 sign ? "signed" : "unsigned", size);
1162 }
1163
1164 abort ();
1165 return BFD_RELOC_NONE;
1166 }
1167
1168 /* Here we decide which fixups can be adjusted to make them relative to
1169 the beginning of the section instead of the symbol. Basically we need
1170 to make sure that the dynamic relocations are done correctly, so in
1171 some cases we force the original symbol to be used. */
1172
1173 int
1174 tc_i386_fix_adjustable (fixP)
1175 fixS *fixP;
1176 {
1177 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1178 /* Prevent all adjustments to global symbols, or else dynamic
1179 linking will not work correctly. */
1180 if (S_IS_EXTERNAL (fixP->fx_addsy)
1181 || S_IS_WEAK (fixP->fx_addsy)
1182 /* Don't adjust pc-relative references to merge sections in 64-bit
1183 mode. */
1184 || (use_rela_relocations
1185 && (S_GET_SEGMENT (fixP->fx_addsy)->flags & SEC_MERGE) != 0
1186 && fixP->fx_pcrel))
1187 return 0;
1188 #endif
1189 /* adjust_reloc_syms doesn't know about the GOT. */
1190 if (fixP->fx_r_type == BFD_RELOC_386_GOTOFF
1191 || fixP->fx_r_type == BFD_RELOC_386_PLT32
1192 || fixP->fx_r_type == BFD_RELOC_386_GOT32
1193 || fixP->fx_r_type == BFD_RELOC_X86_64_PLT32
1194 || fixP->fx_r_type == BFD_RELOC_X86_64_GOT32
1195 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPCREL
1196 || fixP->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1197 || fixP->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1198 return 0;
1199 return 1;
1200 }
1201 #else
1202 #define reloc(SIZE,PCREL,SIGN,OTHER) 0
1203 #define BFD_RELOC_16 0
1204 #define BFD_RELOC_32 0
1205 #define BFD_RELOC_16_PCREL 0
1206 #define BFD_RELOC_32_PCREL 0
1207 #define BFD_RELOC_386_PLT32 0
1208 #define BFD_RELOC_386_GOT32 0
1209 #define BFD_RELOC_386_GOTOFF 0
1210 #define BFD_RELOC_X86_64_PLT32 0
1211 #define BFD_RELOC_X86_64_GOT32 0
1212 #define BFD_RELOC_X86_64_GOTPCREL 0
1213 #endif
1214
1215 static int intel_float_operand PARAMS ((char *mnemonic));
1216
1217 static int
1218 intel_float_operand (mnemonic)
1219 char *mnemonic;
1220 {
1221 if (mnemonic[0] == 'f' && mnemonic[1] == 'i')
1222 return 2;
1223
1224 if (mnemonic[0] == 'f')
1225 return 1;
1226
1227 return 0;
1228 }
1229
1230 /* This is the guts of the machine-dependent assembler. LINE points to a
1231 machine dependent instruction. This function is supposed to emit
1232 the frags/bytes it assembles to. */
1233
1234 void
1235 md_assemble (line)
1236 char *line;
1237 {
1238 /* Points to template once we've found it. */
1239 const template *t;
1240
1241 int j;
1242
1243 char mnemonic[MAX_MNEM_SIZE];
1244
1245 /* Initialize globals. */
1246 memset (&i, '\0', sizeof (i));
1247 for (j = 0; j < MAX_OPERANDS; j++)
1248 i.reloc[j] = NO_RELOC;
1249 memset (disp_expressions, '\0', sizeof (disp_expressions));
1250 memset (im_expressions, '\0', sizeof (im_expressions));
1251 save_stack_p = save_stack;
1252
1253 /* First parse an instruction mnemonic & call i386_operand for the operands.
1254 We assume that the scrubber has arranged it so that line[0] is the valid
1255 start of a (possibly prefixed) mnemonic. */
1256 {
1257 char *l = line;
1258 char *token_start = l;
1259 char *mnem_p;
1260
1261 /* Non-zero if we found a prefix only acceptable with string insns. */
1262 const char *expecting_string_instruction = NULL;
1263
1264 while (1)
1265 {
1266 mnem_p = mnemonic;
1267 while ((*mnem_p = mnemonic_chars[(unsigned char) *l]) != 0)
1268 {
1269 mnem_p++;
1270 if (mnem_p >= mnemonic + sizeof (mnemonic))
1271 {
1272 as_bad (_("no such instruction: `%s'"), token_start);
1273 return;
1274 }
1275 l++;
1276 }
1277 if (!is_space_char (*l)
1278 && *l != END_OF_INSN
1279 && *l != PREFIX_SEPARATOR
1280 && *l != ',')
1281 {
1282 as_bad (_("invalid character %s in mnemonic"),
1283 output_invalid (*l));
1284 return;
1285 }
1286 if (token_start == l)
1287 {
1288 if (*l == PREFIX_SEPARATOR)
1289 as_bad (_("expecting prefix; got nothing"));
1290 else
1291 as_bad (_("expecting mnemonic; got nothing"));
1292 return;
1293 }
1294
1295 /* Look up instruction (or prefix) via hash table. */
1296 current_templates = hash_find (op_hash, mnemonic);
1297
1298 if (*l != END_OF_INSN
1299 && (! is_space_char (*l) || l[1] != END_OF_INSN)
1300 && current_templates
1301 && (current_templates->start->opcode_modifier & IsPrefix))
1302 {
1303 /* If we are in 16-bit mode, do not allow addr16 or data16.
1304 Similarly, in 32-bit mode, do not allow addr32 or data32. */
1305 if ((current_templates->start->opcode_modifier & (Size16 | Size32))
1306 && flag_code != CODE_64BIT
1307 && (((current_templates->start->opcode_modifier & Size32) != 0)
1308 ^ (flag_code == CODE_16BIT)))
1309 {
1310 as_bad (_("redundant %s prefix"),
1311 current_templates->start->name);
1312 return;
1313 }
1314 /* Add prefix, checking for repeated prefixes. */
1315 switch (add_prefix (current_templates->start->base_opcode))
1316 {
1317 case 0:
1318 return;
1319 case 2:
1320 expecting_string_instruction = current_templates->start->name;
1321 break;
1322 }
1323 /* Skip past PREFIX_SEPARATOR and reset token_start. */
1324 token_start = ++l;
1325 }
1326 else
1327 break;
1328 }
1329
1330 if (!current_templates)
1331 {
1332 /* See if we can get a match by trimming off a suffix. */
1333 switch (mnem_p[-1])
1334 {
1335 case WORD_MNEM_SUFFIX:
1336 case BYTE_MNEM_SUFFIX:
1337 case QWORD_MNEM_SUFFIX:
1338 i.suffix = mnem_p[-1];
1339 mnem_p[-1] = '\0';
1340 current_templates = hash_find (op_hash, mnemonic);
1341 break;
1342 case SHORT_MNEM_SUFFIX:
1343 case LONG_MNEM_SUFFIX:
1344 if (!intel_syntax)
1345 {
1346 i.suffix = mnem_p[-1];
1347 mnem_p[-1] = '\0';
1348 current_templates = hash_find (op_hash, mnemonic);
1349 }
1350 break;
1351
1352 /* Intel Syntax. */
1353 case 'd':
1354 if (intel_syntax)
1355 {
1356 if (intel_float_operand (mnemonic))
1357 i.suffix = SHORT_MNEM_SUFFIX;
1358 else
1359 i.suffix = LONG_MNEM_SUFFIX;
1360 mnem_p[-1] = '\0';
1361 current_templates = hash_find (op_hash, mnemonic);
1362 }
1363 break;
1364 }
1365 if (!current_templates)
1366 {
1367 as_bad (_("no such instruction: `%s'"), token_start);
1368 return;
1369 }
1370 }
1371
1372 if (current_templates->start->opcode_modifier & (Jump | JumpByte))
1373 {
1374 /* Check for a branch hint. We allow ",pt" and ",pn" for
1375 predict taken and predict not taken respectively.
1376 I'm not sure that branch hints actually do anything on loop
1377 and jcxz insns (JumpByte) for current Pentium4 chips. They
1378 may work in the future and it doesn't hurt to accept them
1379 now. */
1380 if (l[0] == ',' && l[1] == 'p')
1381 {
1382 if (l[2] == 't')
1383 {
1384 if (! add_prefix (DS_PREFIX_OPCODE))
1385 return;
1386 l += 3;
1387 }
1388 else if (l[2] == 'n')
1389 {
1390 if (! add_prefix (CS_PREFIX_OPCODE))
1391 return;
1392 l += 3;
1393 }
1394 }
1395 }
1396 /* Any other comma loses. */
1397 if (*l == ',')
1398 {
1399 as_bad (_("invalid character %s in mnemonic"),
1400 output_invalid (*l));
1401 return;
1402 }
1403
1404 /* Check if instruction is supported on specified architecture. */
1405 if ((current_templates->start->cpu_flags & ~(Cpu64 | CpuNo64))
1406 & ~(cpu_arch_flags & ~(Cpu64 | CpuNo64)))
1407 {
1408 as_warn (_("`%s' is not supported on `%s'"),
1409 current_templates->start->name, cpu_arch_name);
1410 }
1411 else if ((Cpu386 & ~cpu_arch_flags) && (flag_code != CODE_16BIT))
1412 {
1413 as_warn (_("use .code16 to ensure correct addressing mode"));
1414 }
1415
1416 /* Check for rep/repne without a string instruction. */
1417 if (expecting_string_instruction
1418 && !(current_templates->start->opcode_modifier & IsString))
1419 {
1420 as_bad (_("expecting string instruction after `%s'"),
1421 expecting_string_instruction);
1422 return;
1423 }
1424
1425 /* There may be operands to parse. */
1426 if (*l != END_OF_INSN)
1427 {
1428 /* 1 if operand is pending after ','. */
1429 unsigned int expecting_operand = 0;
1430
1431 /* Non-zero if operand parens not balanced. */
1432 unsigned int paren_not_balanced;
1433
1434 do
1435 {
1436 /* Skip optional white space before operand. */
1437 if (is_space_char (*l))
1438 ++l;
1439 if (!is_operand_char (*l) && *l != END_OF_INSN)
1440 {
1441 as_bad (_("invalid character %s before operand %d"),
1442 output_invalid (*l),
1443 i.operands + 1);
1444 return;
1445 }
1446 token_start = l; /* after white space */
1447 paren_not_balanced = 0;
1448 while (paren_not_balanced || *l != ',')
1449 {
1450 if (*l == END_OF_INSN)
1451 {
1452 if (paren_not_balanced)
1453 {
1454 if (!intel_syntax)
1455 as_bad (_("unbalanced parenthesis in operand %d."),
1456 i.operands + 1);
1457 else
1458 as_bad (_("unbalanced brackets in operand %d."),
1459 i.operands + 1);
1460 return;
1461 }
1462 else
1463 break; /* we are done */
1464 }
1465 else if (!is_operand_char (*l) && !is_space_char (*l))
1466 {
1467 as_bad (_("invalid character %s in operand %d"),
1468 output_invalid (*l),
1469 i.operands + 1);
1470 return;
1471 }
1472 if (!intel_syntax)
1473 {
1474 if (*l == '(')
1475 ++paren_not_balanced;
1476 if (*l == ')')
1477 --paren_not_balanced;
1478 }
1479 else
1480 {
1481 if (*l == '[')
1482 ++paren_not_balanced;
1483 if (*l == ']')
1484 --paren_not_balanced;
1485 }
1486 l++;
1487 }
1488 if (l != token_start)
1489 { /* Yes, we've read in another operand. */
1490 unsigned int operand_ok;
1491 this_operand = i.operands++;
1492 if (i.operands > MAX_OPERANDS)
1493 {
1494 as_bad (_("spurious operands; (%d operands/instruction max)"),
1495 MAX_OPERANDS);
1496 return;
1497 }
1498 /* Now parse operand adding info to 'i' as we go along. */
1499 END_STRING_AND_SAVE (l);
1500
1501 if (intel_syntax)
1502 operand_ok =
1503 i386_intel_operand (token_start,
1504 intel_float_operand (mnemonic));
1505 else
1506 operand_ok = i386_operand (token_start);
1507
1508 RESTORE_END_STRING (l);
1509 if (!operand_ok)
1510 return;
1511 }
1512 else
1513 {
1514 if (expecting_operand)
1515 {
1516 expecting_operand_after_comma:
1517 as_bad (_("expecting operand after ','; got nothing"));
1518 return;
1519 }
1520 if (*l == ',')
1521 {
1522 as_bad (_("expecting operand before ','; got nothing"));
1523 return;
1524 }
1525 }
1526
1527 /* Now *l must be either ',' or END_OF_INSN. */
1528 if (*l == ',')
1529 {
1530 if (*++l == END_OF_INSN)
1531 {
1532 /* Just skip it, if it's \n complain. */
1533 goto expecting_operand_after_comma;
1534 }
1535 expecting_operand = 1;
1536 }
1537 }
1538 while (*l != END_OF_INSN);
1539 }
1540 }
1541
1542 /* Now we've parsed the mnemonic into a set of templates, and have the
1543 operands at hand.
1544
1545 Next, we find a template that matches the given insn,
1546 making sure the overlap of the given operands types is consistent
1547 with the template operand types. */
1548
1549 #define MATCH(overlap, given, template) \
1550 ((overlap & ~JumpAbsolute) \
1551 && ((given) & (BaseIndex|JumpAbsolute)) == ((overlap) & (BaseIndex|JumpAbsolute)))
1552
1553 /* If given types r0 and r1 are registers they must be of the same type
1554 unless the expected operand type register overlap is null.
1555 Note that Acc in a template matches every size of reg. */
1556 #define CONSISTENT_REGISTER_MATCH(m0, g0, t0, m1, g1, t1) \
1557 ( ((g0) & Reg) == 0 || ((g1) & Reg) == 0 || \
1558 ((g0) & Reg) == ((g1) & Reg) || \
1559 ((((m0) & Acc) ? Reg : (t0)) & (((m1) & Acc) ? Reg : (t1)) & Reg) == 0 )
1560
1561 {
1562 register unsigned int overlap0, overlap1;
1563 unsigned int overlap2;
1564 unsigned int found_reverse_match;
1565 int suffix_check;
1566
1567 /* All intel opcodes have reversed operands except for "bound" and
1568 "enter". We also don't reverse intersegment "jmp" and "call"
1569 instructions with 2 immediate operands so that the immediate segment
1570 precedes the offset, as it does when in AT&T mode. "enter" and the
1571 intersegment "jmp" and "call" instructions are the only ones that
1572 have two immediate operands. */
1573 if (intel_syntax && i.operands > 1
1574 && (strcmp (mnemonic, "bound") != 0)
1575 && !((i.types[0] & Imm) && (i.types[1] & Imm)))
1576 {
1577 union i386_op temp_op;
1578 unsigned int temp_type;
1579 RELOC_ENUM temp_reloc;
1580 int xchg1 = 0;
1581 int xchg2 = 0;
1582
1583 if (i.operands == 2)
1584 {
1585 xchg1 = 0;
1586 xchg2 = 1;
1587 }
1588 else if (i.operands == 3)
1589 {
1590 xchg1 = 0;
1591 xchg2 = 2;
1592 }
1593 temp_type = i.types[xchg2];
1594 i.types[xchg2] = i.types[xchg1];
1595 i.types[xchg1] = temp_type;
1596 temp_op = i.op[xchg2];
1597 i.op[xchg2] = i.op[xchg1];
1598 i.op[xchg1] = temp_op;
1599 temp_reloc = i.reloc[xchg2];
1600 i.reloc[xchg2] = i.reloc[xchg1];
1601 i.reloc[xchg1] = temp_reloc;
1602
1603 if (i.mem_operands == 2)
1604 {
1605 const seg_entry *temp_seg;
1606 temp_seg = i.seg[0];
1607 i.seg[0] = i.seg[1];
1608 i.seg[1] = temp_seg;
1609 }
1610 }
1611
1612 if (i.imm_operands)
1613 {
1614 /* Try to ensure constant immediates are represented in the smallest
1615 opcode possible. */
1616 char guess_suffix = 0;
1617 int op;
1618
1619 if (i.suffix)
1620 guess_suffix = i.suffix;
1621 else if (i.reg_operands)
1622 {
1623 /* Figure out a suffix from the last register operand specified.
1624 We can't do this properly yet, ie. excluding InOutPortReg,
1625 but the following works for instructions with immediates.
1626 In any case, we can't set i.suffix yet. */
1627 for (op = i.operands; --op >= 0;)
1628 if (i.types[op] & Reg)
1629 {
1630 if (i.types[op] & Reg8)
1631 guess_suffix = BYTE_MNEM_SUFFIX;
1632 else if (i.types[op] & Reg16)
1633 guess_suffix = WORD_MNEM_SUFFIX;
1634 else if (i.types[op] & Reg32)
1635 guess_suffix = LONG_MNEM_SUFFIX;
1636 else if (i.types[op] & Reg64)
1637 guess_suffix = QWORD_MNEM_SUFFIX;
1638 break;
1639 }
1640 }
1641 else if ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0))
1642 guess_suffix = WORD_MNEM_SUFFIX;
1643
1644 for (op = i.operands; --op >= 0;)
1645 if (i.types[op] & Imm)
1646 {
1647 switch (i.op[op].imms->X_op)
1648 {
1649 case O_constant:
1650 /* If a suffix is given, this operand may be shortened. */
1651 switch (guess_suffix)
1652 {
1653 case LONG_MNEM_SUFFIX:
1654 i.types[op] |= Imm32 | Imm64;
1655 break;
1656 case WORD_MNEM_SUFFIX:
1657 i.types[op] |= Imm16 | Imm32S | Imm32 | Imm64;
1658 break;
1659 case BYTE_MNEM_SUFFIX:
1660 i.types[op] |= Imm16 | Imm8 | Imm8S | Imm32S | Imm32 | Imm64;
1661 break;
1662 }
1663
1664 /* If this operand is at most 16 bits, convert it
1665 to a signed 16 bit number before trying to see
1666 whether it will fit in an even smaller size.
1667 This allows a 16-bit operand such as $0xffe0 to
1668 be recognised as within Imm8S range. */
1669 if ((i.types[op] & Imm16)
1670 && (i.op[op].imms->X_add_number & ~(offsetT) 0xffff) == 0)
1671 {
1672 i.op[op].imms->X_add_number =
1673 (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000);
1674 }
1675 if ((i.types[op] & Imm32)
1676 && (i.op[op].imms->X_add_number & ~(((offsetT) 2 << 31) - 1)) == 0)
1677 {
1678 i.op[op].imms->X_add_number =
1679 (i.op[op].imms->X_add_number ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
1680 }
1681 i.types[op] |= smallest_imm_type (i.op[op].imms->X_add_number);
1682 /* We must avoid matching of Imm32 templates when 64bit only immediate is available. */
1683 if (guess_suffix == QWORD_MNEM_SUFFIX)
1684 i.types[op] &= ~Imm32;
1685 break;
1686 case O_absent:
1687 case O_register:
1688 abort ();
1689 /* Symbols and expressions. */
1690 default:
1691 /* Convert symbolic operand to proper sizes for matching. */
1692 switch (guess_suffix)
1693 {
1694 case QWORD_MNEM_SUFFIX:
1695 i.types[op] = Imm64 | Imm32S;
1696 break;
1697 case LONG_MNEM_SUFFIX:
1698 i.types[op] = Imm32 | Imm64;
1699 break;
1700 case WORD_MNEM_SUFFIX:
1701 i.types[op] = Imm16 | Imm32 | Imm64;
1702 break;
1703 break;
1704 case BYTE_MNEM_SUFFIX:
1705 i.types[op] = Imm8 | Imm8S | Imm16 | Imm32S | Imm32;
1706 break;
1707 break;
1708 }
1709 break;
1710 }
1711 }
1712 }
1713
1714 if (i.disp_operands)
1715 {
1716 /* Try to use the smallest displacement type too. */
1717 int op;
1718
1719 for (op = i.operands; --op >= 0;)
1720 if ((i.types[op] & Disp)
1721 && i.op[op].disps->X_op == O_constant)
1722 {
1723 offsetT disp = i.op[op].disps->X_add_number;
1724
1725 if (i.types[op] & Disp16)
1726 {
1727 /* We know this operand is at most 16 bits, so
1728 convert to a signed 16 bit number before trying
1729 to see whether it will fit in an even smaller
1730 size. */
1731
1732 disp = (((disp & 0xffff) ^ 0x8000) - 0x8000);
1733 }
1734 else if (i.types[op] & Disp32)
1735 {
1736 /* We know this operand is at most 32 bits, so convert to a
1737 signed 32 bit number before trying to see whether it will
1738 fit in an even smaller size. */
1739 disp &= (((offsetT) 2 << 31) - 1);
1740 disp = (disp ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
1741 }
1742 if (flag_code == CODE_64BIT)
1743 {
1744 if (fits_in_signed_long (disp))
1745 i.types[op] |= Disp32S;
1746 if (fits_in_unsigned_long (disp))
1747 i.types[op] |= Disp32;
1748 }
1749 if ((i.types[op] & (Disp32 | Disp32S | Disp16))
1750 && fits_in_signed_byte (disp))
1751 i.types[op] |= Disp8;
1752 }
1753 }
1754
1755 overlap0 = 0;
1756 overlap1 = 0;
1757 overlap2 = 0;
1758 found_reverse_match = 0;
1759 suffix_check = (i.suffix == BYTE_MNEM_SUFFIX
1760 ? No_bSuf
1761 : (i.suffix == WORD_MNEM_SUFFIX
1762 ? No_wSuf
1763 : (i.suffix == SHORT_MNEM_SUFFIX
1764 ? No_sSuf
1765 : (i.suffix == LONG_MNEM_SUFFIX
1766 ? No_lSuf
1767 : (i.suffix == QWORD_MNEM_SUFFIX
1768 ? No_qSuf
1769 : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX ? No_xSuf : 0))))));
1770
1771 for (t = current_templates->start;
1772 t < current_templates->end;
1773 t++)
1774 {
1775 /* Must have right number of operands. */
1776 if (i.operands != t->operands)
1777 continue;
1778
1779 /* Check the suffix, except for some instructions in intel mode. */
1780 if ((t->opcode_modifier & suffix_check)
1781 && !(intel_syntax
1782 && (t->opcode_modifier & IgnoreSize))
1783 && !(intel_syntax
1784 && t->base_opcode == 0xd9
1785 && (t->extension_opcode == 5 /* 0xd9,5 "fldcw" */
1786 || t->extension_opcode == 7))) /* 0xd9,7 "f{n}stcw" */
1787 continue;
1788
1789 /* Do not verify operands when there are none. */
1790 else if (!t->operands)
1791 {
1792 if (t->cpu_flags & ~cpu_arch_flags)
1793 continue;
1794 /* We've found a match; break out of loop. */
1795 break;
1796 }
1797
1798 overlap0 = i.types[0] & t->operand_types[0];
1799 switch (t->operands)
1800 {
1801 case 1:
1802 if (!MATCH (overlap0, i.types[0], t->operand_types[0]))
1803 continue;
1804 break;
1805 case 2:
1806 case 3:
1807 overlap1 = i.types[1] & t->operand_types[1];
1808 if (!MATCH (overlap0, i.types[0], t->operand_types[0])
1809 || !MATCH (overlap1, i.types[1], t->operand_types[1])
1810 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
1811 t->operand_types[0],
1812 overlap1, i.types[1],
1813 t->operand_types[1]))
1814 {
1815 /* Check if other direction is valid ... */
1816 if ((t->opcode_modifier & (D|FloatD)) == 0)
1817 continue;
1818
1819 /* Try reversing direction of operands. */
1820 overlap0 = i.types[0] & t->operand_types[1];
1821 overlap1 = i.types[1] & t->operand_types[0];
1822 if (!MATCH (overlap0, i.types[0], t->operand_types[1])
1823 || !MATCH (overlap1, i.types[1], t->operand_types[0])
1824 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
1825 t->operand_types[1],
1826 overlap1, i.types[1],
1827 t->operand_types[0]))
1828 {
1829 /* Does not match either direction. */
1830 continue;
1831 }
1832 /* found_reverse_match holds which of D or FloatDR
1833 we've found. */
1834 found_reverse_match = t->opcode_modifier & (D|FloatDR);
1835 }
1836 /* Found a forward 2 operand match here. */
1837 else if (t->operands == 3)
1838 {
1839 /* Here we make use of the fact that there are no
1840 reverse match 3 operand instructions, and all 3
1841 operand instructions only need to be checked for
1842 register consistency between operands 2 and 3. */
1843 overlap2 = i.types[2] & t->operand_types[2];
1844 if (!MATCH (overlap2, i.types[2], t->operand_types[2])
1845 || !CONSISTENT_REGISTER_MATCH (overlap1, i.types[1],
1846 t->operand_types[1],
1847 overlap2, i.types[2],
1848 t->operand_types[2]))
1849
1850 continue;
1851 }
1852 /* Found either forward/reverse 2 or 3 operand match here:
1853 slip through to break. */
1854 }
1855 if (t->cpu_flags & ~cpu_arch_flags)
1856 {
1857 found_reverse_match = 0;
1858 continue;
1859 }
1860 /* We've found a match; break out of loop. */
1861 break;
1862 }
1863 if (t == current_templates->end)
1864 {
1865 /* We found no match. */
1866 as_bad (_("suffix or operands invalid for `%s'"),
1867 current_templates->start->name);
1868 return;
1869 }
1870
1871 if (!quiet_warnings)
1872 {
1873 if (!intel_syntax
1874 && ((i.types[0] & JumpAbsolute)
1875 != (t->operand_types[0] & JumpAbsolute)))
1876 {
1877 as_warn (_("indirect %s without `*'"), t->name);
1878 }
1879
1880 if ((t->opcode_modifier & (IsPrefix|IgnoreSize))
1881 == (IsPrefix|IgnoreSize))
1882 {
1883 /* Warn them that a data or address size prefix doesn't
1884 affect assembly of the next line of code. */
1885 as_warn (_("stand-alone `%s' prefix"), t->name);
1886 }
1887 }
1888
1889 /* Copy the template we found. */
1890 i.tm = *t;
1891 if (found_reverse_match)
1892 {
1893 /* If we found a reverse match we must alter the opcode
1894 direction bit. found_reverse_match holds bits to change
1895 (different for int & float insns). */
1896
1897 i.tm.base_opcode ^= found_reverse_match;
1898
1899 i.tm.operand_types[0] = t->operand_types[1];
1900 i.tm.operand_types[1] = t->operand_types[0];
1901 }
1902
1903 /* Undo SYSV386_COMPAT brokenness when in Intel mode. See i386.h */
1904 if (SYSV386_COMPAT
1905 && intel_syntax
1906 && (i.tm.base_opcode & 0xfffffde0) == 0xdce0)
1907 i.tm.base_opcode ^= FloatR;
1908
1909 if (i.tm.opcode_modifier & FWait)
1910 if (! add_prefix (FWAIT_OPCODE))
1911 return;
1912
1913 /* Check string instruction segment overrides. */
1914 if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0)
1915 {
1916 int mem_op = (i.types[0] & AnyMem) ? 0 : 1;
1917 if ((i.tm.operand_types[mem_op] & EsSeg) != 0)
1918 {
1919 if (i.seg[0] != NULL && i.seg[0] != &es)
1920 {
1921 as_bad (_("`%s' operand %d must use `%%es' segment"),
1922 i.tm.name,
1923 mem_op + 1);
1924 return;
1925 }
1926 /* There's only ever one segment override allowed per instruction.
1927 This instruction possibly has a legal segment override on the
1928 second operand, so copy the segment to where non-string
1929 instructions store it, allowing common code. */
1930 i.seg[0] = i.seg[1];
1931 }
1932 else if ((i.tm.operand_types[mem_op + 1] & EsSeg) != 0)
1933 {
1934 if (i.seg[1] != NULL && i.seg[1] != &es)
1935 {
1936 as_bad (_("`%s' operand %d must use `%%es' segment"),
1937 i.tm.name,
1938 mem_op + 2);
1939 return;
1940 }
1941 }
1942 }
1943
1944 /* If matched instruction specifies an explicit instruction mnemonic
1945 suffix, use it. */
1946 if (i.tm.opcode_modifier & (Size16 | Size32 | Size64))
1947 {
1948 if (i.tm.opcode_modifier & Size16)
1949 i.suffix = WORD_MNEM_SUFFIX;
1950 else if (i.tm.opcode_modifier & Size64)
1951 i.suffix = QWORD_MNEM_SUFFIX;
1952 else
1953 i.suffix = LONG_MNEM_SUFFIX;
1954 }
1955 else if (i.reg_operands)
1956 {
1957 /* If there's no instruction mnemonic suffix we try to invent one
1958 based on register operands. */
1959 if (!i.suffix)
1960 {
1961 /* We take i.suffix from the last register operand specified,
1962 Destination register type is more significant than source
1963 register type. */
1964 int op;
1965 for (op = i.operands; --op >= 0;)
1966 if ((i.types[op] & Reg)
1967 && !(i.tm.operand_types[op] & InOutPortReg))
1968 {
1969 i.suffix = ((i.types[op] & Reg8) ? BYTE_MNEM_SUFFIX :
1970 (i.types[op] & Reg16) ? WORD_MNEM_SUFFIX :
1971 (i.types[op] & Reg64) ? QWORD_MNEM_SUFFIX :
1972 LONG_MNEM_SUFFIX);
1973 break;
1974 }
1975 }
1976 else if (i.suffix == BYTE_MNEM_SUFFIX)
1977 {
1978 int op;
1979 for (op = i.operands; --op >= 0;)
1980 {
1981 /* If this is an eight bit register, it's OK. If it's
1982 the 16 or 32 bit version of an eight bit register,
1983 we will just use the low portion, and that's OK too. */
1984 if (i.types[op] & Reg8)
1985 continue;
1986
1987 /* movzx and movsx should not generate this warning. */
1988 if (intel_syntax
1989 && (i.tm.base_opcode == 0xfb7
1990 || i.tm.base_opcode == 0xfb6
1991 || i.tm.base_opcode == 0x63
1992 || i.tm.base_opcode == 0xfbe
1993 || i.tm.base_opcode == 0xfbf))
1994 continue;
1995
1996 if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4
1997 #if 0
1998 /* Check that the template allows eight bit regs
1999 This kills insns such as `orb $1,%edx', which
2000 maybe should be allowed. */
2001 && (i.tm.operand_types[op] & (Reg8|InOutPortReg))
2002 #endif
2003 )
2004 {
2005 /* Prohibit these changes in the 64bit mode, since
2006 the lowering is more complicated. */
2007 if (flag_code == CODE_64BIT
2008 && (i.tm.operand_types[op] & InOutPortReg) == 0)
2009 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2010 i.op[op].regs->reg_name,
2011 i.suffix);
2012 #if REGISTER_WARNINGS
2013 if (!quiet_warnings
2014 && (i.tm.operand_types[op] & InOutPortReg) == 0)
2015 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2016 (i.op[op].regs
2017 + (i.types[op] & Reg16
2018 ? REGNAM_AL - REGNAM_AX
2019 : REGNAM_AL - REGNAM_EAX))->reg_name,
2020 i.op[op].regs->reg_name,
2021 i.suffix);
2022 #endif
2023 continue;
2024 }
2025 /* Any other register is bad. */
2026 if (i.types[op] & (Reg | RegMMX | RegXMM
2027 | SReg2 | SReg3
2028 | Control | Debug | Test
2029 | FloatReg | FloatAcc))
2030 {
2031 as_bad (_("`%%%s' not allowed with `%s%c'"),
2032 i.op[op].regs->reg_name,
2033 i.tm.name,
2034 i.suffix);
2035 return;
2036 }
2037 }
2038 }
2039 else if (i.suffix == LONG_MNEM_SUFFIX)
2040 {
2041 int op;
2042
2043 for (op = i.operands; --op >= 0;)
2044 /* Reject eight bit registers, except where the template
2045 requires them. (eg. movzb) */
2046 if ((i.types[op] & Reg8) != 0
2047 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2048 {
2049 as_bad (_("`%%%s' not allowed with `%s%c'"),
2050 i.op[op].regs->reg_name,
2051 i.tm.name,
2052 i.suffix);
2053 return;
2054 }
2055 /* Warn if the e prefix on a general reg is missing. */
2056 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2057 && (i.types[op] & Reg16) != 0
2058 && (i.tm.operand_types[op] & (Reg32|Acc)) != 0)
2059 {
2060 /* Prohibit these changes in the 64bit mode, since
2061 the lowering is more complicated. */
2062 if (flag_code == CODE_64BIT)
2063 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2064 i.op[op].regs->reg_name,
2065 i.suffix);
2066 #if REGISTER_WARNINGS
2067 else
2068 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2069 (i.op[op].regs + REGNAM_EAX - REGNAM_AX)->reg_name,
2070 i.op[op].regs->reg_name,
2071 i.suffix);
2072 #endif
2073 }
2074 /* Warn if the r prefix on a general reg is missing. */
2075 else if ((i.types[op] & Reg64) != 0
2076 && (i.tm.operand_types[op] & (Reg32|Acc)) != 0)
2077 {
2078 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2079 i.op[op].regs->reg_name,
2080 i.suffix);
2081 }
2082 }
2083 else if (i.suffix == QWORD_MNEM_SUFFIX)
2084 {
2085 int op;
2086
2087 for (op = i.operands; --op >= 0; )
2088 /* Reject eight bit registers, except where the template
2089 requires them. (eg. movzb) */
2090 if ((i.types[op] & Reg8) != 0
2091 && (i.tm.operand_types[op] & (Reg16|Reg32|Acc)) != 0)
2092 {
2093 as_bad (_("`%%%s' not allowed with `%s%c'"),
2094 i.op[op].regs->reg_name,
2095 i.tm.name,
2096 i.suffix);
2097 return;
2098 }
2099 /* Warn if the e prefix on a general reg is missing. */
2100 else if (((i.types[op] & Reg16) != 0
2101 || (i.types[op] & Reg32) != 0)
2102 && (i.tm.operand_types[op] & (Reg32|Acc)) != 0)
2103 {
2104 /* Prohibit these changes in the 64bit mode, since
2105 the lowering is more complicated. */
2106 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2107 i.op[op].regs->reg_name,
2108 i.suffix);
2109 }
2110 }
2111 else if (i.suffix == WORD_MNEM_SUFFIX)
2112 {
2113 int op;
2114 for (op = i.operands; --op >= 0;)
2115 /* Reject eight bit registers, except where the template
2116 requires them. (eg. movzb) */
2117 if ((i.types[op] & Reg8) != 0
2118 && (i.tm.operand_types[op] & (Reg16|Reg32|Acc)) != 0)
2119 {
2120 as_bad (_("`%%%s' not allowed with `%s%c'"),
2121 i.op[op].regs->reg_name,
2122 i.tm.name,
2123 i.suffix);
2124 return;
2125 }
2126 /* Warn if the e prefix on a general reg is present. */
2127 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2128 && (i.types[op] & Reg32) != 0
2129 && (i.tm.operand_types[op] & (Reg16|Acc)) != 0)
2130 {
2131 /* Prohibit these changes in the 64bit mode, since
2132 the lowering is more complicated. */
2133 if (flag_code == CODE_64BIT)
2134 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2135 i.op[op].regs->reg_name,
2136 i.suffix);
2137 else
2138 #if REGISTER_WARNINGS
2139 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2140 (i.op[op].regs + REGNAM_AX - REGNAM_EAX)->reg_name,
2141 i.op[op].regs->reg_name,
2142 i.suffix);
2143 #endif
2144 }
2145 }
2146 else if (intel_syntax && (i.tm.opcode_modifier & IgnoreSize))
2147 /* Do nothing if the instruction is going to ignore the prefix. */
2148 ;
2149 else
2150 abort ();
2151 }
2152 else if ((i.tm.opcode_modifier & DefaultSize) && !i.suffix)
2153 {
2154 i.suffix = stackop_size;
2155 }
2156 /* Make still unresolved immediate matches conform to size of immediate
2157 given in i.suffix. Note: overlap2 cannot be an immediate! */
2158 if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S))
2159 && overlap0 != Imm8 && overlap0 != Imm8S
2160 && overlap0 != Imm16 && overlap0 != Imm32S
2161 && overlap0 != Imm32 && overlap0 != Imm64)
2162 {
2163 if (i.suffix)
2164 {
2165 overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX ? (Imm8 | Imm8S) :
2166 (i.suffix == WORD_MNEM_SUFFIX ? Imm16 :
2167 (i.suffix == QWORD_MNEM_SUFFIX ? Imm64 | Imm32S : Imm32)));
2168 }
2169 else if (overlap0 == (Imm16 | Imm32S | Imm32)
2170 || overlap0 == (Imm16 | Imm32)
2171 || overlap0 == (Imm16 | Imm32S))
2172 {
2173 overlap0 =
2174 ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)) ? Imm16 : Imm32S;
2175 }
2176 if (overlap0 != Imm8 && overlap0 != Imm8S
2177 && overlap0 != Imm16 && overlap0 != Imm32S
2178 && overlap0 != Imm32 && overlap0 != Imm64)
2179 {
2180 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size"));
2181 return;
2182 }
2183 }
2184 if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32))
2185 && overlap1 != Imm8 && overlap1 != Imm8S
2186 && overlap1 != Imm16 && overlap1 != Imm32S
2187 && overlap1 != Imm32 && overlap1 != Imm64)
2188 {
2189 if (i.suffix)
2190 {
2191 overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX ? (Imm8 | Imm8S) :
2192 (i.suffix == WORD_MNEM_SUFFIX ? Imm16 :
2193 (i.suffix == QWORD_MNEM_SUFFIX ? Imm64 | Imm32S : Imm32)));
2194 }
2195 else if (overlap1 == (Imm16 | Imm32 | Imm32S)
2196 || overlap1 == (Imm16 | Imm32)
2197 || overlap1 == (Imm16 | Imm32S))
2198 {
2199 overlap1 =
2200 ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)) ? Imm16 : Imm32S;
2201 }
2202 if (overlap1 != Imm8 && overlap1 != Imm8S
2203 && overlap1 != Imm16 && overlap1 != Imm32S
2204 && overlap1 != Imm32 && overlap1 != Imm64)
2205 {
2206 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size %x %c"),overlap1, i.suffix);
2207 return;
2208 }
2209 }
2210 assert ((overlap2 & Imm) == 0);
2211
2212 i.types[0] = overlap0;
2213 if (overlap0 & ImplicitRegister)
2214 i.reg_operands--;
2215 if (overlap0 & Imm1)
2216 i.imm_operands = 0; /* kludge for shift insns. */
2217
2218 i.types[1] = overlap1;
2219 if (overlap1 & ImplicitRegister)
2220 i.reg_operands--;
2221
2222 i.types[2] = overlap2;
2223 if (overlap2 & ImplicitRegister)
2224 i.reg_operands--;
2225
2226 /* Finalize opcode. First, we change the opcode based on the operand
2227 size given by i.suffix: We need not change things for byte insns. */
2228
2229 if (!i.suffix && (i.tm.opcode_modifier & W))
2230 {
2231 as_bad (_("no instruction mnemonic suffix given and no register operands; can't size instruction"));
2232 return;
2233 }
2234
2235 /* For movzx and movsx, need to check the register type. */
2236 if (intel_syntax
2237 && (i.tm.base_opcode == 0xfb6 || i.tm.base_opcode == 0xfbe))
2238 if (i.suffix && i.suffix == BYTE_MNEM_SUFFIX)
2239 {
2240 unsigned int prefix = DATA_PREFIX_OPCODE;
2241
2242 if ((i.op[1].regs->reg_type & Reg16) != 0)
2243 if (!add_prefix (prefix))
2244 return;
2245 }
2246
2247 if (i.suffix && i.suffix != BYTE_MNEM_SUFFIX)
2248 {
2249 /* It's not a byte, select word/dword operation. */
2250 if (i.tm.opcode_modifier & W)
2251 {
2252 if (i.tm.opcode_modifier & ShortForm)
2253 i.tm.base_opcode |= 8;
2254 else
2255 i.tm.base_opcode |= 1;
2256 }
2257 /* Now select between word & dword operations via the operand
2258 size prefix, except for instructions that will ignore this
2259 prefix anyway. */
2260 if (i.suffix != QWORD_MNEM_SUFFIX
2261 && (i.suffix == LONG_MNEM_SUFFIX) == (flag_code == CODE_16BIT)
2262 && !(i.tm.opcode_modifier & IgnoreSize))
2263 {
2264 unsigned int prefix = DATA_PREFIX_OPCODE;
2265 if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */
2266 prefix = ADDR_PREFIX_OPCODE;
2267
2268 if (! add_prefix (prefix))
2269 return;
2270 }
2271
2272 if (i.suffix != QWORD_MNEM_SUFFIX && (flag_code == CODE_64BIT)
2273 && !(i.tm.opcode_modifier & IgnoreSize)
2274 && (i.tm.opcode_modifier & JumpByte))
2275 {
2276 if (! add_prefix (ADDR_PREFIX_OPCODE))
2277 return;
2278 }
2279
2280 /* Set mode64 for an operand. */
2281 if (i.suffix == QWORD_MNEM_SUFFIX
2282 && !(i.tm.opcode_modifier & NoRex64))
2283 {
2284 i.rex.mode64 = 1;
2285 if (flag_code < CODE_64BIT)
2286 {
2287 as_bad (_("64bit operations available only in 64bit modes."));
2288 return;
2289 }
2290 }
2291
2292 /* Size floating point instruction. */
2293 if (i.suffix == LONG_MNEM_SUFFIX)
2294 {
2295 if (i.tm.opcode_modifier & FloatMF)
2296 i.tm.base_opcode ^= 4;
2297 }
2298 }
2299
2300 if (i.tm.opcode_modifier & ImmExt)
2301 {
2302 /* These AMD 3DNow! and Intel Katmai New Instructions have an
2303 opcode suffix which is coded in the same place as an 8-bit
2304 immediate field would be. Here we fake an 8-bit immediate
2305 operand from the opcode suffix stored in tm.extension_opcode. */
2306
2307 expressionS *exp;
2308
2309 assert (i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS);
2310
2311 exp = &im_expressions[i.imm_operands++];
2312 i.op[i.operands].imms = exp;
2313 i.types[i.operands++] = Imm8;
2314 exp->X_op = O_constant;
2315 exp->X_add_number = i.tm.extension_opcode;
2316 i.tm.extension_opcode = None;
2317 }
2318
2319 /* For insns with operands there are more diddles to do to the opcode. */
2320 if (i.operands)
2321 {
2322 /* Default segment register this instruction will use
2323 for memory accesses. 0 means unknown.
2324 This is only for optimizing out unnecessary segment overrides. */
2325 const seg_entry *default_seg = 0;
2326
2327 /* The imul $imm, %reg instruction is converted into
2328 imul $imm, %reg, %reg, and the clr %reg instruction
2329 is converted into xor %reg, %reg. */
2330 if (i.tm.opcode_modifier & regKludge)
2331 {
2332 unsigned int first_reg_op = (i.types[0] & Reg) ? 0 : 1;
2333 /* Pretend we saw the extra register operand. */
2334 assert (i.op[first_reg_op + 1].regs == 0);
2335 i.op[first_reg_op + 1].regs = i.op[first_reg_op].regs;
2336 i.types[first_reg_op + 1] = i.types[first_reg_op];
2337 i.reg_operands = 2;
2338 }
2339
2340 if (i.tm.opcode_modifier & ShortForm)
2341 {
2342 /* The register or float register operand is in operand 0 or 1. */
2343 unsigned int op = (i.types[0] & (Reg | FloatReg)) ? 0 : 1;
2344 /* Register goes in low 3 bits of opcode. */
2345 i.tm.base_opcode |= i.op[op].regs->reg_num;
2346 if (i.op[op].regs->reg_flags & RegRex)
2347 i.rex.extZ = 1;
2348 if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
2349 {
2350 /* Warn about some common errors, but press on regardless.
2351 The first case can be generated by gcc (<= 2.8.1). */
2352 if (i.operands == 2)
2353 {
2354 /* Reversed arguments on faddp, fsubp, etc. */
2355 as_warn (_("translating to `%s %%%s,%%%s'"), i.tm.name,
2356 i.op[1].regs->reg_name,
2357 i.op[0].regs->reg_name);
2358 }
2359 else
2360 {
2361 /* Extraneous `l' suffix on fp insn. */
2362 as_warn (_("translating to `%s %%%s'"), i.tm.name,
2363 i.op[0].regs->reg_name);
2364 }
2365 }
2366 }
2367 else if (i.tm.opcode_modifier & Modrm)
2368 {
2369 /* The opcode is completed (modulo i.tm.extension_opcode which
2370 must be put into the modrm byte).
2371 Now, we make the modrm & index base bytes based on all the
2372 info we've collected. */
2373
2374 /* i.reg_operands MUST be the number of real register operands;
2375 implicit registers do not count. */
2376 if (i.reg_operands == 2)
2377 {
2378 unsigned int source, dest;
2379 source = ((i.types[0]
2380 & (Reg | RegMMX | RegXMM
2381 | SReg2 | SReg3
2382 | Control | Debug | Test))
2383 ? 0 : 1);
2384 dest = source + 1;
2385
2386 i.rm.mode = 3;
2387 /* One of the register operands will be encoded in the
2388 i.tm.reg field, the other in the combined i.tm.mode
2389 and i.tm.regmem fields. If no form of this
2390 instruction supports a memory destination operand,
2391 then we assume the source operand may sometimes be
2392 a memory operand and so we need to store the
2393 destination in the i.rm.reg field. */
2394 if ((i.tm.operand_types[dest] & AnyMem) == 0)
2395 {
2396 i.rm.reg = i.op[dest].regs->reg_num;
2397 i.rm.regmem = i.op[source].regs->reg_num;
2398 if (i.op[dest].regs->reg_flags & RegRex)
2399 i.rex.extX = 1;
2400 if (i.op[source].regs->reg_flags & RegRex)
2401 i.rex.extZ = 1;
2402 }
2403 else
2404 {
2405 i.rm.reg = i.op[source].regs->reg_num;
2406 i.rm.regmem = i.op[dest].regs->reg_num;
2407 if (i.op[dest].regs->reg_flags & RegRex)
2408 i.rex.extZ = 1;
2409 if (i.op[source].regs->reg_flags & RegRex)
2410 i.rex.extX = 1;
2411 }
2412 }
2413 else
2414 { /* If it's not 2 reg operands... */
2415 if (i.mem_operands)
2416 {
2417 unsigned int fake_zero_displacement = 0;
2418 unsigned int op = ((i.types[0] & AnyMem)
2419 ? 0
2420 : (i.types[1] & AnyMem) ? 1 : 2);
2421
2422 default_seg = &ds;
2423
2424 if (! i.base_reg)
2425 {
2426 i.rm.mode = 0;
2427 if (! i.disp_operands)
2428 fake_zero_displacement = 1;
2429 if (! i.index_reg)
2430 {
2431 /* Operand is just <disp> */
2432 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0)
2433 && (flag_code != CODE_64BIT))
2434 {
2435 i.rm.regmem = NO_BASE_REGISTER_16;
2436 i.types[op] &= ~Disp;
2437 i.types[op] |= Disp16;
2438 }
2439 else if (flag_code != CODE_64BIT
2440 || (i.prefix[ADDR_PREFIX] != 0))
2441 {
2442 i.rm.regmem = NO_BASE_REGISTER;
2443 i.types[op] &= ~Disp;
2444 i.types[op] |= Disp32;
2445 }
2446 else
2447 {
2448 /* 64bit mode overwrites the 32bit
2449 absolute addressing by RIP relative
2450 addressing and absolute addressing
2451 is encoded by one of the redundant
2452 SIB forms. */
2453
2454 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2455 i.sib.base = NO_BASE_REGISTER;
2456 i.sib.index = NO_INDEX_REGISTER;
2457 i.types[op] &= ~Disp;
2458 i.types[op] |= Disp32S;
2459 }
2460 }
2461 else /* ! i.base_reg && i.index_reg */
2462 {
2463 i.sib.index = i.index_reg->reg_num;
2464 i.sib.base = NO_BASE_REGISTER;
2465 i.sib.scale = i.log2_scale_factor;
2466 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2467 i.types[op] &= ~Disp;
2468 if (flag_code != CODE_64BIT)
2469 i.types[op] |= Disp32; /* Must be 32 bit */
2470 else
2471 i.types[op] |= Disp32S;
2472 if (i.index_reg->reg_flags & RegRex)
2473 i.rex.extY = 1;
2474 }
2475 }
2476 /* RIP addressing for 64bit mode. */
2477 else if (i.base_reg->reg_type == BaseIndex)
2478 {
2479 i.rm.regmem = NO_BASE_REGISTER;
2480 i.types[op] &= ~Disp;
2481 i.types[op] |= Disp32S;
2482 i.flags[op] = Operand_PCrel;
2483 }
2484 else if (i.base_reg->reg_type & Reg16)
2485 {
2486 switch (i.base_reg->reg_num)
2487 {
2488 case 3: /* (%bx) */
2489 if (! i.index_reg)
2490 i.rm.regmem = 7;
2491 else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */
2492 i.rm.regmem = i.index_reg->reg_num - 6;
2493 break;
2494 case 5: /* (%bp) */
2495 default_seg = &ss;
2496 if (! i.index_reg)
2497 {
2498 i.rm.regmem = 6;
2499 if ((i.types[op] & Disp) == 0)
2500 {
2501 /* fake (%bp) into 0(%bp) */
2502 i.types[op] |= Disp8;
2503 fake_zero_displacement = 1;
2504 }
2505 }
2506 else /* (%bp,%si) -> 2, or (%bp,%di) -> 3 */
2507 i.rm.regmem = i.index_reg->reg_num - 6 + 2;
2508 break;
2509 default: /* (%si) -> 4 or (%di) -> 5 */
2510 i.rm.regmem = i.base_reg->reg_num - 6 + 4;
2511 }
2512 i.rm.mode = mode_from_disp_size (i.types[op]);
2513 }
2514 else /* i.base_reg and 32/64 bit mode */
2515 {
2516 if (flag_code == CODE_64BIT
2517 && (i.types[op] & Disp))
2518 {
2519 if (i.types[op] & Disp8)
2520 i.types[op] = Disp8 | Disp32S;
2521 else
2522 i.types[op] = Disp32S;
2523 }
2524 i.rm.regmem = i.base_reg->reg_num;
2525 if (i.base_reg->reg_flags & RegRex)
2526 i.rex.extZ = 1;
2527 i.sib.base = i.base_reg->reg_num;
2528 /* x86-64 ignores REX prefix bit here to avoid
2529 decoder complications. */
2530 if ((i.base_reg->reg_num & 7) == EBP_REG_NUM)
2531 {
2532 default_seg = &ss;
2533 if (i.disp_operands == 0)
2534 {
2535 fake_zero_displacement = 1;
2536 i.types[op] |= Disp8;
2537 }
2538 }
2539 else if (i.base_reg->reg_num == ESP_REG_NUM)
2540 {
2541 default_seg = &ss;
2542 }
2543 i.sib.scale = i.log2_scale_factor;
2544 if (! i.index_reg)
2545 {
2546 /* <disp>(%esp) becomes two byte modrm
2547 with no index register. We've already
2548 stored the code for esp in i.rm.regmem
2549 ie. ESCAPE_TO_TWO_BYTE_ADDRESSING. Any
2550 base register besides %esp will not use
2551 the extra modrm byte. */
2552 i.sib.index = NO_INDEX_REGISTER;
2553 #if ! SCALE1_WHEN_NO_INDEX
2554 /* Another case where we force the second
2555 modrm byte. */
2556 if (i.log2_scale_factor)
2557 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2558 #endif
2559 }
2560 else
2561 {
2562 i.sib.index = i.index_reg->reg_num;
2563 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2564 if (i.index_reg->reg_flags & RegRex)
2565 i.rex.extY = 1;
2566 }
2567 i.rm.mode = mode_from_disp_size (i.types[op]);
2568 }
2569
2570 if (fake_zero_displacement)
2571 {
2572 /* Fakes a zero displacement assuming that i.types[op]
2573 holds the correct displacement size. */
2574 expressionS *exp;
2575
2576 assert (i.op[op].disps == 0);
2577 exp = &disp_expressions[i.disp_operands++];
2578 i.op[op].disps = exp;
2579 exp->X_op = O_constant;
2580 exp->X_add_number = 0;
2581 exp->X_add_symbol = (symbolS *) 0;
2582 exp->X_op_symbol = (symbolS *) 0;
2583 }
2584 }
2585
2586 /* Fill in i.rm.reg or i.rm.regmem field with register
2587 operand (if any) based on i.tm.extension_opcode.
2588 Again, we must be careful to make sure that
2589 segment/control/debug/test/MMX registers are coded
2590 into the i.rm.reg field. */
2591 if (i.reg_operands)
2592 {
2593 unsigned int op =
2594 ((i.types[0]
2595 & (Reg | RegMMX | RegXMM
2596 | SReg2 | SReg3
2597 | Control | Debug | Test))
2598 ? 0
2599 : ((i.types[1]
2600 & (Reg | RegMMX | RegXMM
2601 | SReg2 | SReg3
2602 | Control | Debug | Test))
2603 ? 1
2604 : 2));
2605 /* If there is an extension opcode to put here, the
2606 register number must be put into the regmem field. */
2607 if (i.tm.extension_opcode != None)
2608 {
2609 i.rm.regmem = i.op[op].regs->reg_num;
2610 if (i.op[op].regs->reg_flags & RegRex)
2611 i.rex.extZ = 1;
2612 }
2613 else
2614 {
2615 i.rm.reg = i.op[op].regs->reg_num;
2616 if (i.op[op].regs->reg_flags & RegRex)
2617 i.rex.extX = 1;
2618 }
2619
2620 /* Now, if no memory operand has set i.rm.mode = 0, 1, 2
2621 we must set it to 3 to indicate this is a register
2622 operand in the regmem field. */
2623 if (!i.mem_operands)
2624 i.rm.mode = 3;
2625 }
2626
2627 /* Fill in i.rm.reg field with extension opcode (if any). */
2628 if (i.tm.extension_opcode != None)
2629 i.rm.reg = i.tm.extension_opcode;
2630 }
2631 }
2632 else if (i.tm.opcode_modifier & (Seg2ShortForm | Seg3ShortForm))
2633 {
2634 if (i.tm.base_opcode == POP_SEG_SHORT
2635 && i.op[0].regs->reg_num == 1)
2636 {
2637 as_bad (_("you can't `pop %%cs'"));
2638 return;
2639 }
2640 i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
2641 if (i.op[0].regs->reg_flags & RegRex)
2642 i.rex.extZ = 1;
2643 }
2644 else if ((i.tm.base_opcode & ~(D|W)) == MOV_AX_DISP32)
2645 {
2646 default_seg = &ds;
2647 }
2648 else if ((i.tm.opcode_modifier & IsString) != 0)
2649 {
2650 /* For the string instructions that allow a segment override
2651 on one of their operands, the default segment is ds. */
2652 default_seg = &ds;
2653 }
2654
2655 /* If a segment was explicitly specified,
2656 and the specified segment is not the default,
2657 use an opcode prefix to select it.
2658 If we never figured out what the default segment is,
2659 then default_seg will be zero at this point,
2660 and the specified segment prefix will always be used. */
2661 if ((i.seg[0]) && (i.seg[0] != default_seg))
2662 {
2663 if (! add_prefix (i.seg[0]->seg_prefix))
2664 return;
2665 }
2666 }
2667 else if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
2668 {
2669 /* UnixWare fsub no args is alias for fsubp, fadd -> faddp, etc. */
2670 as_warn (_("translating to `%sp'"), i.tm.name);
2671 }
2672 }
2673
2674 /* Handle conversion of 'int $3' --> special int3 insn. */
2675 if (i.tm.base_opcode == INT_OPCODE && i.op[0].imms->X_add_number == 3)
2676 {
2677 i.tm.base_opcode = INT3_OPCODE;
2678 i.imm_operands = 0;
2679 }
2680
2681 if ((i.tm.opcode_modifier & (Jump | JumpByte | JumpDword))
2682 && i.op[0].disps->X_op == O_constant)
2683 {
2684 /* Convert "jmp constant" (and "call constant") to a jump (call) to
2685 the absolute address given by the constant. Since ix86 jumps and
2686 calls are pc relative, we need to generate a reloc. */
2687 i.op[0].disps->X_add_symbol = &abs_symbol;
2688 i.op[0].disps->X_op = O_symbol;
2689 }
2690
2691 if (i.tm.opcode_modifier & Rex64)
2692 i.rex.mode64 = 1;
2693
2694 /* For 8bit registers we would need an empty rex prefix.
2695 Also in the case instruction is already having prefix,
2696 we need to convert old registers to new ones. */
2697
2698 if (((i.types[0] & Reg8) && (i.op[0].regs->reg_flags & RegRex64))
2699 || ((i.types[1] & Reg8) && (i.op[1].regs->reg_flags & RegRex64))
2700 || ((i.rex.mode64 || i.rex.extX || i.rex.extY || i.rex.extZ || i.rex.empty)
2701 && ((i.types[0] & Reg8) || (i.types[1] & Reg8))))
2702 {
2703 int x;
2704 i.rex.empty = 1;
2705 for (x = 0; x < 2; x++)
2706 {
2707 /* Look for 8bit operand that does use old registers. */
2708 if (i.types[x] & Reg8
2709 && !(i.op[x].regs->reg_flags & RegRex64))
2710 {
2711 /* In case it is "hi" register, give up. */
2712 if (i.op[x].regs->reg_num > 3)
2713 as_bad (_("Can't encode registers '%%%s' in the instruction requiring REX prefix.\n"),
2714 i.op[x].regs->reg_name);
2715
2716 /* Otherwise it is equivalent to the extended register.
2717 Since the encoding don't change this is merely cosmetical
2718 cleanup for debug output. */
2719
2720 i.op[x].regs = i.op[x].regs + 8;
2721 }
2722 }
2723 }
2724
2725 if (i.rex.mode64 || i.rex.extX || i.rex.extY || i.rex.extZ || i.rex.empty)
2726 add_prefix (0x40
2727 | (i.rex.mode64 ? 8 : 0)
2728 | (i.rex.extX ? 4 : 0)
2729 | (i.rex.extY ? 2 : 0)
2730 | (i.rex.extZ ? 1 : 0));
2731
2732 /* We are ready to output the insn. */
2733 {
2734 register char *p;
2735
2736 /* Tie dwarf2 debug info to the address at the start of the insn.
2737 We can't do this after the insn has been output as the current
2738 frag may have been closed off. eg. by frag_var. */
2739 dwarf2_emit_insn (0);
2740
2741 /* Output jumps. */
2742 if (i.tm.opcode_modifier & Jump)
2743 {
2744 int code16;
2745 int prefix;
2746 relax_substateT subtype;
2747 symbolS *sym;
2748 offsetT off;
2749
2750 code16 = 0;
2751 if (flag_code == CODE_16BIT)
2752 code16 = CODE16;
2753
2754 prefix = 0;
2755 if (i.prefix[DATA_PREFIX])
2756 {
2757 prefix = 1;
2758 i.prefixes -= 1;
2759 code16 ^= CODE16;
2760 }
2761 /* Pentium4 branch hints. */
2762 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
2763 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
2764 {
2765 prefix++;
2766 i.prefixes--;
2767 }
2768 if (i.prefix[REX_PREFIX])
2769 {
2770 prefix++;
2771 i.prefixes--;
2772 }
2773
2774 if (i.prefixes != 0 && !intel_syntax)
2775 as_warn (_("skipping prefixes on this instruction"));
2776
2777 /* It's always a symbol; End frag & setup for relax.
2778 Make sure there is enough room in this frag for the largest
2779 instruction we may generate in md_convert_frag. This is 2
2780 bytes for the opcode and room for the prefix and largest
2781 displacement. */
2782 frag_grow (prefix + 2 + 4);
2783 /* Prefix and 1 opcode byte go in fr_fix. */
2784 p = frag_more (prefix + 1);
2785 if (i.prefix[DATA_PREFIX])
2786 *p++ = DATA_PREFIX_OPCODE;
2787 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE
2788 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE)
2789 *p++ = i.prefix[SEG_PREFIX];
2790 if (i.prefix[REX_PREFIX])
2791 *p++ = i.prefix[REX_PREFIX];
2792 *p = i.tm.base_opcode;
2793
2794 if ((unsigned char) *p == JUMP_PC_RELATIVE)
2795 subtype = ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL);
2796 else if ((cpu_arch_flags & Cpu386) != 0)
2797 subtype = ENCODE_RELAX_STATE (COND_JUMP, SMALL);
2798 else
2799 subtype = ENCODE_RELAX_STATE (COND_JUMP86, SMALL);
2800 subtype |= code16;
2801
2802 sym = i.op[0].disps->X_add_symbol;
2803 off = i.op[0].disps->X_add_number;
2804
2805 if (i.op[0].disps->X_op != O_constant
2806 && i.op[0].disps->X_op != O_symbol)
2807 {
2808 /* Handle complex expressions. */
2809 sym = make_expr_symbol (i.op[0].disps);
2810 off = 0;
2811 }
2812
2813 /* 1 possible extra opcode + 4 byte displacement go in var part.
2814 Pass reloc in fr_var. */
2815 frag_var (rs_machine_dependent, 5, i.reloc[0], subtype, sym, off, p);
2816 }
2817 else if (i.tm.opcode_modifier & (JumpByte | JumpDword))
2818 {
2819 int size;
2820
2821 if (i.tm.opcode_modifier & JumpByte)
2822 {
2823 /* This is a loop or jecxz type instruction. */
2824 size = 1;
2825 if (i.prefix[ADDR_PREFIX])
2826 {
2827 FRAG_APPEND_1_CHAR (ADDR_PREFIX_OPCODE);
2828 i.prefixes -= 1;
2829 }
2830 /* Pentium4 branch hints. */
2831 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
2832 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
2833 {
2834 FRAG_APPEND_1_CHAR (i.prefix[SEG_PREFIX]);
2835 i.prefixes--;
2836 }
2837 }
2838 else
2839 {
2840 int code16;
2841
2842 code16 = 0;
2843 if (flag_code == CODE_16BIT)
2844 code16 = CODE16;
2845
2846 if (i.prefix[DATA_PREFIX])
2847 {
2848 FRAG_APPEND_1_CHAR (DATA_PREFIX_OPCODE);
2849 i.prefixes -= 1;
2850 code16 ^= CODE16;
2851 }
2852
2853 size = 4;
2854 if (code16)
2855 size = 2;
2856 }
2857
2858 if (i.prefix[REX_PREFIX])
2859 {
2860 FRAG_APPEND_1_CHAR (i.prefix[REX_PREFIX]);
2861 i.prefixes -= 1;
2862 }
2863
2864 if (i.prefixes != 0 && !intel_syntax)
2865 as_warn (_("skipping prefixes on this instruction"));
2866
2867 p = frag_more (1 + size);
2868 *p++ = i.tm.base_opcode;
2869
2870 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
2871 i.op[0].disps, 1, reloc (size, 1, 1, i.reloc[0]));
2872 }
2873 else if (i.tm.opcode_modifier & JumpInterSegment)
2874 {
2875 int size;
2876 int prefix;
2877 int code16;
2878
2879 code16 = 0;
2880 if (flag_code == CODE_16BIT)
2881 code16 = CODE16;
2882
2883 prefix = 0;
2884 if (i.prefix[DATA_PREFIX])
2885 {
2886 prefix = 1;
2887 i.prefixes -= 1;
2888 code16 ^= CODE16;
2889 }
2890 if (i.prefix[REX_PREFIX])
2891 {
2892 prefix++;
2893 i.prefixes -= 1;
2894 }
2895
2896 size = 4;
2897 if (code16)
2898 size = 2;
2899
2900 if (i.prefixes != 0 && !intel_syntax)
2901 as_warn (_("skipping prefixes on this instruction"));
2902
2903 /* 1 opcode; 2 segment; offset */
2904 p = frag_more (prefix + 1 + 2 + size);
2905
2906 if (i.prefix[DATA_PREFIX])
2907 *p++ = DATA_PREFIX_OPCODE;
2908
2909 if (i.prefix[REX_PREFIX])
2910 *p++ = i.prefix[REX_PREFIX];
2911
2912 *p++ = i.tm.base_opcode;
2913 if (i.op[1].imms->X_op == O_constant)
2914 {
2915 offsetT n = i.op[1].imms->X_add_number;
2916
2917 if (size == 2
2918 && !fits_in_unsigned_word (n)
2919 && !fits_in_signed_word (n))
2920 {
2921 as_bad (_("16-bit jump out of range"));
2922 return;
2923 }
2924 md_number_to_chars (p, n, size);
2925 }
2926 else
2927 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
2928 i.op[1].imms, 0, reloc (size, 0, 0, i.reloc[1]));
2929 if (i.op[0].imms->X_op != O_constant)
2930 as_bad (_("can't handle non absolute segment in `%s'"),
2931 i.tm.name);
2932 md_number_to_chars (p + size, (valueT) i.op[0].imms->X_add_number, 2);
2933 }
2934 else
2935 {
2936 /* Output normal instructions here. */
2937 unsigned char *q;
2938
2939 /* All opcodes on i386 have eighter 1 or 2 bytes. We may use third
2940 byte for the SSE instructions to specify prefix they require. */
2941 if (i.tm.base_opcode & 0xff0000)
2942 add_prefix ((i.tm.base_opcode >> 16) & 0xff);
2943
2944 /* The prefix bytes. */
2945 for (q = i.prefix;
2946 q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
2947 q++)
2948 {
2949 if (*q)
2950 {
2951 p = frag_more (1);
2952 md_number_to_chars (p, (valueT) *q, 1);
2953 }
2954 }
2955
2956 /* Now the opcode; be careful about word order here! */
2957 if (fits_in_unsigned_byte (i.tm.base_opcode))
2958 {
2959 FRAG_APPEND_1_CHAR (i.tm.base_opcode);
2960 }
2961 else
2962 {
2963 p = frag_more (2);
2964 /* Put out high byte first: can't use md_number_to_chars! */
2965 *p++ = (i.tm.base_opcode >> 8) & 0xff;
2966 *p = i.tm.base_opcode & 0xff;
2967 }
2968
2969 /* Now the modrm byte and sib byte (if present). */
2970 if (i.tm.opcode_modifier & Modrm)
2971 {
2972 p = frag_more (1);
2973 md_number_to_chars (p,
2974 (valueT) (i.rm.regmem << 0
2975 | i.rm.reg << 3
2976 | i.rm.mode << 6),
2977 1);
2978 /* If i.rm.regmem == ESP (4)
2979 && i.rm.mode != (Register mode)
2980 && not 16 bit
2981 ==> need second modrm byte. */
2982 if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING
2983 && i.rm.mode != 3
2984 && !(i.base_reg && (i.base_reg->reg_type & Reg16) != 0))
2985 {
2986 p = frag_more (1);
2987 md_number_to_chars (p,
2988 (valueT) (i.sib.base << 0
2989 | i.sib.index << 3
2990 | i.sib.scale << 6),
2991 1);
2992 }
2993 }
2994
2995 if (i.disp_operands)
2996 {
2997 register unsigned int n;
2998
2999 for (n = 0; n < i.operands; n++)
3000 {
3001 if (i.types[n] & Disp)
3002 {
3003 if (i.op[n].disps->X_op == O_constant)
3004 {
3005 int size;
3006 offsetT val;
3007
3008 size = 4;
3009 if (i.types[n] & (Disp8 | Disp16 | Disp64))
3010 {
3011 size = 2;
3012 if (i.types[n] & Disp8)
3013 size = 1;
3014 if (i.types[n] & Disp64)
3015 size = 8;
3016 }
3017 val = offset_in_range (i.op[n].disps->X_add_number,
3018 size);
3019 p = frag_more (size);
3020 md_number_to_chars (p, val, size);
3021 }
3022 else
3023 {
3024 int size = 4;
3025 int sign = 0;
3026 int pcrel = (i.flags[n] & Operand_PCrel) != 0;
3027
3028 /* The PC relative address is computed relative
3029 to the instruction boundary, so in case immediate
3030 fields follows, we need to adjust the value. */
3031 if (pcrel && i.imm_operands)
3032 {
3033 int imm_size = 4;
3034 register unsigned int n1;
3035
3036 for (n1 = 0; n1 < i.operands; n1++)
3037 if (i.types[n1] & Imm)
3038 {
3039 if (i.types[n1] & (Imm8 | Imm8S | Imm16 | Imm64))
3040 {
3041 imm_size = 2;
3042 if (i.types[n1] & (Imm8 | Imm8S))
3043 imm_size = 1;
3044 if (i.types[n1] & Imm64)
3045 imm_size = 8;
3046 }
3047 break;
3048 }
3049 /* We should find the immediate. */
3050 if (n1 == i.operands)
3051 abort ();
3052 i.op[n].disps->X_add_number -= imm_size;
3053 }
3054
3055 if (i.types[n] & Disp32S)
3056 sign = 1;
3057
3058 if (i.types[n] & (Disp16 | Disp64))
3059 {
3060 size = 2;
3061 if (i.types[n] & Disp64)
3062 size = 8;
3063 }
3064
3065 p = frag_more (size);
3066 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3067 i.op[n].disps, pcrel,
3068 reloc (size, pcrel, sign, i.reloc[n]));
3069 }
3070 }
3071 }
3072 }
3073
3074 /* Output immediate. */
3075 if (i.imm_operands)
3076 {
3077 register unsigned int n;
3078
3079 for (n = 0; n < i.operands; n++)
3080 {
3081 if (i.types[n] & Imm)
3082 {
3083 if (i.op[n].imms->X_op == O_constant)
3084 {
3085 int size;
3086 offsetT val;
3087
3088 size = 4;
3089 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3090 {
3091 size = 2;
3092 if (i.types[n] & (Imm8 | Imm8S))
3093 size = 1;
3094 else if (i.types[n] & Imm64)
3095 size = 8;
3096 }
3097 val = offset_in_range (i.op[n].imms->X_add_number,
3098 size);
3099 p = frag_more (size);
3100 md_number_to_chars (p, val, size);
3101 }
3102 else
3103 {
3104 /* Not absolute_section.
3105 Need a 32-bit fixup (don't support 8bit
3106 non-absolute imms). Try to support other
3107 sizes ... */
3108 RELOC_ENUM reloc_type;
3109 int size = 4;
3110 int sign = 0;
3111
3112 if ((i.types[n] & (Imm32S))
3113 && i.suffix == QWORD_MNEM_SUFFIX)
3114 sign = 1;
3115 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3116 {
3117 size = 2;
3118 if (i.types[n] & (Imm8 | Imm8S))
3119 size = 1;
3120 if (i.types[n] & Imm64)
3121 size = 8;
3122 }
3123
3124 p = frag_more (size);
3125 reloc_type = reloc (size, 0, sign, i.reloc[n]);
3126 #ifdef BFD_ASSEMBLER
3127 if (reloc_type == BFD_RELOC_32
3128 && GOT_symbol
3129 && GOT_symbol == i.op[n].imms->X_add_symbol
3130 && (i.op[n].imms->X_op == O_symbol
3131 || (i.op[n].imms->X_op == O_add
3132 && ((symbol_get_value_expression
3133 (i.op[n].imms->X_op_symbol)->X_op)
3134 == O_subtract))))
3135 {
3136 /* We don't support dynamic linking on x86-64 yet. */
3137 if (flag_code == CODE_64BIT)
3138 abort ();
3139 reloc_type = BFD_RELOC_386_GOTPC;
3140 i.op[n].imms->X_add_number += 3;
3141 }
3142 #endif
3143 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3144 i.op[n].imms, 0, reloc_type);
3145 }
3146 }
3147 }
3148 }
3149 }
3150
3151 #ifdef DEBUG386
3152 if (flag_debug)
3153 {
3154 pi (line, &i);
3155 }
3156 #endif /* DEBUG386 */
3157 }
3158 }
3159 \f
3160 #ifndef LEX_AT
3161 static char *lex_got PARAMS ((RELOC_ENUM *, int *));
3162
3163 /* Parse operands of the form
3164 <symbol>@GOTOFF+<nnn>
3165 and similar .plt or .got references.
3166
3167 If we find one, set up the correct relocation in RELOC and copy the
3168 input string, minus the `@GOTOFF' into a malloc'd buffer for
3169 parsing by the calling routine. Return this buffer, and if ADJUST
3170 is non-null set it to the length of the string we removed from the
3171 input line. Otherwise return NULL. */
3172 static char *
3173 lex_got (reloc, adjust)
3174 RELOC_ENUM *reloc;
3175 int *adjust;
3176 {
3177 static const char * const mode_name[NUM_FLAG_CODE] = { "32", "16", "64" };
3178 static const struct {
3179 const char *str;
3180 const RELOC_ENUM rel[NUM_FLAG_CODE];
3181 } gotrel[] = {
3182 { "PLT", { BFD_RELOC_386_PLT32, 0, BFD_RELOC_X86_64_PLT32 } },
3183 { "GOTOFF", { BFD_RELOC_386_GOTOFF, 0, 0 } },
3184 { "GOTPCREL", { 0, 0, BFD_RELOC_X86_64_GOTPCREL } },
3185 { "GOT", { BFD_RELOC_386_GOT32, 0, BFD_RELOC_X86_64_GOT32 } }
3186 };
3187 char *cp;
3188 unsigned int j;
3189
3190 for (cp = input_line_pointer; *cp != '@'; cp++)
3191 if (is_end_of_line[(unsigned char) *cp])
3192 return NULL;
3193
3194 for (j = 0; j < sizeof (gotrel) / sizeof (gotrel[0]); j++)
3195 {
3196 int len;
3197
3198 len = strlen (gotrel[j].str);
3199 if (strncasecmp (cp + 1, gotrel[j].str, len) == 0)
3200 {
3201 if (gotrel[j].rel[(unsigned int) flag_code] != 0)
3202 {
3203 int first, second;
3204 char *tmpbuf, *past_reloc;
3205
3206 *reloc = gotrel[j].rel[(unsigned int) flag_code];
3207 if (adjust)
3208 *adjust = len;
3209
3210 if (GOT_symbol == NULL)
3211 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
3212
3213 /* Replace the relocation token with ' ', so that
3214 errors like foo@GOTOFF1 will be detected. */
3215
3216 /* The length of the first part of our input line. */
3217 first = cp - input_line_pointer;
3218
3219 /* The second part goes from after the reloc token until
3220 (and including) an end_of_line char. Don't use strlen
3221 here as the end_of_line char may not be a NUL. */
3222 past_reloc = cp + 1 + len;
3223 for (cp = past_reloc; !is_end_of_line[(unsigned char) *cp++]; )
3224 ;
3225 second = cp - past_reloc;
3226
3227 /* Allocate and copy string. The trailing NUL shouldn't
3228 be necessary, but be safe. */
3229 tmpbuf = xmalloc (first + second + 2);
3230 memcpy (tmpbuf, input_line_pointer, first);
3231 tmpbuf[first] = ' ';
3232 memcpy (tmpbuf + first + 1, past_reloc, second);
3233 tmpbuf[first + second + 1] = '\0';
3234 return tmpbuf;
3235 }
3236
3237 as_bad (_("@%s reloc is not supported in %s bit mode"),
3238 gotrel[j].str, mode_name[(unsigned int) flag_code]);
3239 return NULL;
3240 }
3241 }
3242
3243 /* Might be a symbol version string. Don't as_bad here. */
3244 return NULL;
3245 }
3246
3247 /* x86_cons_fix_new is called via the expression parsing code when a
3248 reloc is needed. We use this hook to get the correct .got reloc. */
3249 static RELOC_ENUM got_reloc = NO_RELOC;
3250
3251 void
3252 x86_cons_fix_new (frag, off, len, exp)
3253 fragS *frag;
3254 unsigned int off;
3255 unsigned int len;
3256 expressionS *exp;
3257 {
3258 RELOC_ENUM r = reloc (len, 0, 0, got_reloc);
3259 got_reloc = NO_RELOC;
3260 fix_new_exp (frag, off, len, exp, 0, r);
3261 }
3262
3263 void
3264 x86_cons (exp, size)
3265 expressionS *exp;
3266 int size;
3267 {
3268 if (size == 4)
3269 {
3270 /* Handle @GOTOFF and the like in an expression. */
3271 char *save;
3272 char *gotfree_input_line;
3273 int adjust;
3274
3275 save = input_line_pointer;
3276 gotfree_input_line = lex_got (&got_reloc, &adjust);
3277 if (gotfree_input_line)
3278 input_line_pointer = gotfree_input_line;
3279
3280 expression (exp);
3281
3282 if (gotfree_input_line)
3283 {
3284 /* expression () has merrily parsed up to the end of line,
3285 or a comma - in the wrong buffer. Transfer how far
3286 input_line_pointer has moved to the right buffer. */
3287 input_line_pointer = (save
3288 + (input_line_pointer - gotfree_input_line)
3289 + adjust);
3290 free (gotfree_input_line);
3291 }
3292 }
3293 else
3294 expression (exp);
3295 }
3296 #endif
3297
3298 static int i386_immediate PARAMS ((char *));
3299
3300 static int
3301 i386_immediate (imm_start)
3302 char *imm_start;
3303 {
3304 char *save_input_line_pointer;
3305 #ifndef LEX_AT
3306 char *gotfree_input_line;
3307 #endif
3308 segT exp_seg = 0;
3309 expressionS *exp;
3310
3311 if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
3312 {
3313 as_bad (_("only 1 or 2 immediate operands are allowed"));
3314 return 0;
3315 }
3316
3317 exp = &im_expressions[i.imm_operands++];
3318 i.op[this_operand].imms = exp;
3319
3320 if (is_space_char (*imm_start))
3321 ++imm_start;
3322
3323 save_input_line_pointer = input_line_pointer;
3324 input_line_pointer = imm_start;
3325
3326 #ifndef LEX_AT
3327 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
3328 if (gotfree_input_line)
3329 input_line_pointer = gotfree_input_line;
3330 #endif
3331
3332 exp_seg = expression (exp);
3333
3334 SKIP_WHITESPACE ();
3335 if (*input_line_pointer)
3336 as_bad (_("junk `%s' after expression"), input_line_pointer);
3337
3338 input_line_pointer = save_input_line_pointer;
3339 #ifndef LEX_AT
3340 if (gotfree_input_line)
3341 free (gotfree_input_line);
3342 #endif
3343
3344 if (exp->X_op == O_absent || exp->X_op == O_big)
3345 {
3346 /* Missing or bad expr becomes absolute 0. */
3347 as_bad (_("missing or invalid immediate expression `%s' taken as 0"),
3348 imm_start);
3349 exp->X_op = O_constant;
3350 exp->X_add_number = 0;
3351 exp->X_add_symbol = (symbolS *) 0;
3352 exp->X_op_symbol = (symbolS *) 0;
3353 }
3354 else if (exp->X_op == O_constant)
3355 {
3356 /* Size it properly later. */
3357 i.types[this_operand] |= Imm64;
3358 /* If BFD64, sign extend val. */
3359 if (!use_rela_relocations)
3360 if ((exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0)
3361 exp->X_add_number = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
3362 }
3363 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3364 else if (1
3365 #ifdef BFD_ASSEMBLER
3366 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3367 #endif
3368 && exp_seg != text_section
3369 && exp_seg != data_section
3370 && exp_seg != bss_section
3371 && exp_seg != undefined_section
3372 #ifdef BFD_ASSEMBLER
3373 && !bfd_is_com_section (exp_seg)
3374 #endif
3375 )
3376 {
3377 #ifdef BFD_ASSEMBLER
3378 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3379 #else
3380 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
3381 #endif
3382 return 0;
3383 }
3384 #endif
3385 else
3386 {
3387 /* This is an address. The size of the address will be
3388 determined later, depending on destination register,
3389 suffix, or the default for the section. */
3390 i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64;
3391 }
3392
3393 return 1;
3394 }
3395
3396 static char *i386_scale PARAMS ((char *));
3397
3398 static char *
3399 i386_scale (scale)
3400 char *scale;
3401 {
3402 offsetT val;
3403 char *save = input_line_pointer;
3404
3405 input_line_pointer = scale;
3406 val = get_absolute_expression ();
3407
3408 switch (val)
3409 {
3410 case 0:
3411 case 1:
3412 i.log2_scale_factor = 0;
3413 break;
3414 case 2:
3415 i.log2_scale_factor = 1;
3416 break;
3417 case 4:
3418 i.log2_scale_factor = 2;
3419 break;
3420 case 8:
3421 i.log2_scale_factor = 3;
3422 break;
3423 default:
3424 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
3425 scale);
3426 input_line_pointer = save;
3427 return NULL;
3428 }
3429 if (i.log2_scale_factor != 0 && ! i.index_reg)
3430 {
3431 as_warn (_("scale factor of %d without an index register"),
3432 1 << i.log2_scale_factor);
3433 #if SCALE1_WHEN_NO_INDEX
3434 i.log2_scale_factor = 0;
3435 #endif
3436 }
3437 scale = input_line_pointer;
3438 input_line_pointer = save;
3439 return scale;
3440 }
3441
3442 static int i386_displacement PARAMS ((char *, char *));
3443
3444 static int
3445 i386_displacement (disp_start, disp_end)
3446 char *disp_start;
3447 char *disp_end;
3448 {
3449 register expressionS *exp;
3450 segT exp_seg = 0;
3451 char *save_input_line_pointer;
3452 #ifndef LEX_AT
3453 char *gotfree_input_line;
3454 #endif
3455 int bigdisp = Disp32;
3456
3457 if (flag_code == CODE_64BIT)
3458 {
3459 if (!i.prefix[ADDR_PREFIX])
3460 bigdisp = Disp64;
3461 }
3462 else if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3463 bigdisp = Disp16;
3464 i.types[this_operand] |= bigdisp;
3465
3466 exp = &disp_expressions[i.disp_operands];
3467 i.op[this_operand].disps = exp;
3468 i.disp_operands++;
3469 save_input_line_pointer = input_line_pointer;
3470 input_line_pointer = disp_start;
3471 END_STRING_AND_SAVE (disp_end);
3472
3473 #ifndef GCC_ASM_O_HACK
3474 #define GCC_ASM_O_HACK 0
3475 #endif
3476 #if GCC_ASM_O_HACK
3477 END_STRING_AND_SAVE (disp_end + 1);
3478 if ((i.types[this_operand] & BaseIndex) != 0
3479 && displacement_string_end[-1] == '+')
3480 {
3481 /* This hack is to avoid a warning when using the "o"
3482 constraint within gcc asm statements.
3483 For instance:
3484
3485 #define _set_tssldt_desc(n,addr,limit,type) \
3486 __asm__ __volatile__ ( \
3487 "movw %w2,%0\n\t" \
3488 "movw %w1,2+%0\n\t" \
3489 "rorl $16,%1\n\t" \
3490 "movb %b1,4+%0\n\t" \
3491 "movb %4,5+%0\n\t" \
3492 "movb $0,6+%0\n\t" \
3493 "movb %h1,7+%0\n\t" \
3494 "rorl $16,%1" \
3495 : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
3496
3497 This works great except that the output assembler ends
3498 up looking a bit weird if it turns out that there is
3499 no offset. You end up producing code that looks like:
3500
3501 #APP
3502 movw $235,(%eax)
3503 movw %dx,2+(%eax)
3504 rorl $16,%edx
3505 movb %dl,4+(%eax)
3506 movb $137,5+(%eax)
3507 movb $0,6+(%eax)
3508 movb %dh,7+(%eax)
3509 rorl $16,%edx
3510 #NO_APP
3511
3512 So here we provide the missing zero. */
3513
3514 *displacement_string_end = '0';
3515 }
3516 #endif
3517 #ifndef LEX_AT
3518 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
3519 if (gotfree_input_line)
3520 input_line_pointer = gotfree_input_line;
3521 #endif
3522
3523 exp_seg = expression (exp);
3524
3525 SKIP_WHITESPACE ();
3526 if (*input_line_pointer)
3527 as_bad (_("junk `%s' after expression"), input_line_pointer);
3528 #if GCC_ASM_O_HACK
3529 RESTORE_END_STRING (disp_end + 1);
3530 #endif
3531 RESTORE_END_STRING (disp_end);
3532 input_line_pointer = save_input_line_pointer;
3533 #ifndef LEX_AT
3534 if (gotfree_input_line)
3535 free (gotfree_input_line);
3536 #endif
3537
3538 #ifdef BFD_ASSEMBLER
3539 /* We do this to make sure that the section symbol is in
3540 the symbol table. We will ultimately change the relocation
3541 to be relative to the beginning of the section. */
3542 if (i.reloc[this_operand] == BFD_RELOC_386_GOTOFF
3543 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3544 {
3545 if (exp->X_op != O_symbol)
3546 {
3547 as_bad (_("bad expression used with @%s"),
3548 (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
3549 ? "GOTPCREL"
3550 : "GOTOFF"));
3551 return 0;
3552 }
3553
3554 if (S_IS_LOCAL (exp->X_add_symbol)
3555 && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
3556 section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
3557 exp->X_op = O_subtract;
3558 exp->X_op_symbol = GOT_symbol;
3559 if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3560 i.reloc[this_operand] = BFD_RELOC_32_PCREL;
3561 else
3562 i.reloc[this_operand] = BFD_RELOC_32;
3563 }
3564 #endif
3565
3566 if (exp->X_op == O_absent || exp->X_op == O_big)
3567 {
3568 /* Missing or bad expr becomes absolute 0. */
3569 as_bad (_("missing or invalid displacement expression `%s' taken as 0"),
3570 disp_start);
3571 exp->X_op = O_constant;
3572 exp->X_add_number = 0;
3573 exp->X_add_symbol = (symbolS *) 0;
3574 exp->X_op_symbol = (symbolS *) 0;
3575 }
3576
3577 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3578 if (exp->X_op != O_constant
3579 #ifdef BFD_ASSEMBLER
3580 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3581 #endif
3582 && exp_seg != text_section
3583 && exp_seg != data_section
3584 && exp_seg != bss_section
3585 && exp_seg != undefined_section)
3586 {
3587 #ifdef BFD_ASSEMBLER
3588 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3589 #else
3590 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
3591 #endif
3592 return 0;
3593 }
3594 #endif
3595 else if (flag_code == CODE_64BIT)
3596 i.types[this_operand] |= Disp32S | Disp32;
3597 return 1;
3598 }
3599
3600 static int i386_index_check PARAMS ((const char *));
3601
3602 /* Make sure the memory operand we've been dealt is valid.
3603 Return 1 on success, 0 on a failure. */
3604
3605 static int
3606 i386_index_check (operand_string)
3607 const char *operand_string;
3608 {
3609 int ok;
3610 #if INFER_ADDR_PREFIX
3611 int fudged = 0;
3612
3613 tryprefix:
3614 #endif
3615 ok = 1;
3616 if (flag_code == CODE_64BIT)
3617 {
3618 if (i.prefix[ADDR_PREFIX] == 0)
3619 {
3620 /* 64bit checks. */
3621 if ((i.base_reg
3622 && ((i.base_reg->reg_type & Reg64) == 0)
3623 && (i.base_reg->reg_type != BaseIndex
3624 || i.index_reg))
3625 || (i.index_reg
3626 && ((i.index_reg->reg_type & (Reg64|BaseIndex))
3627 != (Reg64|BaseIndex))))
3628 ok = 0;
3629 }
3630 else
3631 {
3632 /* 32bit checks. */
3633 if ((i.base_reg
3634 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
3635 || (i.index_reg
3636 && ((i.index_reg->reg_type & (Reg32|BaseIndex|RegRex))
3637 != (Reg32|BaseIndex))))
3638 ok = 0;
3639 }
3640 }
3641 else
3642 {
3643 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3644 {
3645 /* 16bit checks. */
3646 if ((i.base_reg
3647 && ((i.base_reg->reg_type & (Reg16|BaseIndex|RegRex))
3648 != (Reg16|BaseIndex)))
3649 || (i.index_reg
3650 && (((i.index_reg->reg_type & (Reg16|BaseIndex))
3651 != (Reg16|BaseIndex))
3652 || ! (i.base_reg
3653 && i.base_reg->reg_num < 6
3654 && i.index_reg->reg_num >= 6
3655 && i.log2_scale_factor == 0))))
3656 ok = 0;
3657 }
3658 else
3659 {
3660 /* 32bit checks. */
3661 if ((i.base_reg
3662 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
3663 || (i.index_reg
3664 && ((i.index_reg->reg_type & (Reg32|BaseIndex|RegRex))
3665 != (Reg32|BaseIndex))))
3666 ok = 0;
3667 }
3668 }
3669 if (!ok)
3670 {
3671 #if INFER_ADDR_PREFIX
3672 if (flag_code != CODE_64BIT
3673 && i.prefix[ADDR_PREFIX] == 0 && stackop_size != '\0')
3674 {
3675 i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
3676 i.prefixes += 1;
3677 /* Change the size of any displacement too. At most one of
3678 Disp16 or Disp32 is set.
3679 FIXME. There doesn't seem to be any real need for separate
3680 Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
3681 Removing them would probably clean up the code quite a lot. */
3682 if (i.types[this_operand] & (Disp16|Disp32))
3683 i.types[this_operand] ^= (Disp16|Disp32);
3684 fudged = 1;
3685 goto tryprefix;
3686 }
3687 if (fudged)
3688 as_bad (_("`%s' is not a valid base/index expression"),
3689 operand_string);
3690 else
3691 #endif
3692 as_bad (_("`%s' is not a valid %s bit base/index expression"),
3693 operand_string,
3694 flag_code_names[flag_code]);
3695 return 0;
3696 }
3697 return 1;
3698 }
3699
3700 /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
3701 on error. */
3702
3703 static int
3704 i386_operand (operand_string)
3705 char *operand_string;
3706 {
3707 const reg_entry *r;
3708 char *end_op;
3709 char *op_string = operand_string;
3710
3711 if (is_space_char (*op_string))
3712 ++op_string;
3713
3714 /* We check for an absolute prefix (differentiating,
3715 for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
3716 if (*op_string == ABSOLUTE_PREFIX)
3717 {
3718 ++op_string;
3719 if (is_space_char (*op_string))
3720 ++op_string;
3721 i.types[this_operand] |= JumpAbsolute;
3722 }
3723
3724 /* Check if operand is a register. */
3725 if ((*op_string == REGISTER_PREFIX || allow_naked_reg)
3726 && (r = parse_register (op_string, &end_op)) != NULL)
3727 {
3728 /* Check for a segment override by searching for ':' after a
3729 segment register. */
3730 op_string = end_op;
3731 if (is_space_char (*op_string))
3732 ++op_string;
3733 if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
3734 {
3735 switch (r->reg_num)
3736 {
3737 case 0:
3738 i.seg[i.mem_operands] = &es;
3739 break;
3740 case 1:
3741 i.seg[i.mem_operands] = &cs;
3742 break;
3743 case 2:
3744 i.seg[i.mem_operands] = &ss;
3745 break;
3746 case 3:
3747 i.seg[i.mem_operands] = &ds;
3748 break;
3749 case 4:
3750 i.seg[i.mem_operands] = &fs;
3751 break;
3752 case 5:
3753 i.seg[i.mem_operands] = &gs;
3754 break;
3755 }
3756
3757 /* Skip the ':' and whitespace. */
3758 ++op_string;
3759 if (is_space_char (*op_string))
3760 ++op_string;
3761
3762 if (!is_digit_char (*op_string)
3763 && !is_identifier_char (*op_string)
3764 && *op_string != '('
3765 && *op_string != ABSOLUTE_PREFIX)
3766 {
3767 as_bad (_("bad memory operand `%s'"), op_string);
3768 return 0;
3769 }
3770 /* Handle case of %es:*foo. */
3771 if (*op_string == ABSOLUTE_PREFIX)
3772 {
3773 ++op_string;
3774 if (is_space_char (*op_string))
3775 ++op_string;
3776 i.types[this_operand] |= JumpAbsolute;
3777 }
3778 goto do_memory_reference;
3779 }
3780 if (*op_string)
3781 {
3782 as_bad (_("junk `%s' after register"), op_string);
3783 return 0;
3784 }
3785 i.types[this_operand] |= r->reg_type & ~BaseIndex;
3786 i.op[this_operand].regs = r;
3787 i.reg_operands++;
3788 }
3789 else if (*op_string == REGISTER_PREFIX)
3790 {
3791 as_bad (_("bad register name `%s'"), op_string);
3792 return 0;
3793 }
3794 else if (*op_string == IMMEDIATE_PREFIX)
3795 {
3796 ++op_string;
3797 if (i.types[this_operand] & JumpAbsolute)
3798 {
3799 as_bad (_("immediate operand illegal with absolute jump"));
3800 return 0;
3801 }
3802 if (!i386_immediate (op_string))
3803 return 0;
3804 }
3805 else if (is_digit_char (*op_string)
3806 || is_identifier_char (*op_string)
3807 || *op_string == '(')
3808 {
3809 /* This is a memory reference of some sort. */
3810 char *base_string;
3811
3812 /* Start and end of displacement string expression (if found). */
3813 char *displacement_string_start;
3814 char *displacement_string_end;
3815
3816 do_memory_reference:
3817 if ((i.mem_operands == 1
3818 && (current_templates->start->opcode_modifier & IsString) == 0)
3819 || i.mem_operands == 2)
3820 {
3821 as_bad (_("too many memory references for `%s'"),
3822 current_templates->start->name);
3823 return 0;
3824 }
3825
3826 /* Check for base index form. We detect the base index form by
3827 looking for an ')' at the end of the operand, searching
3828 for the '(' matching it, and finding a REGISTER_PREFIX or ','
3829 after the '('. */
3830 base_string = op_string + strlen (op_string);
3831
3832 --base_string;
3833 if (is_space_char (*base_string))
3834 --base_string;
3835
3836 /* If we only have a displacement, set-up for it to be parsed later. */
3837 displacement_string_start = op_string;
3838 displacement_string_end = base_string + 1;
3839
3840 if (*base_string == ')')
3841 {
3842 char *temp_string;
3843 unsigned int parens_balanced = 1;
3844 /* We've already checked that the number of left & right ()'s are
3845 equal, so this loop will not be infinite. */
3846 do
3847 {
3848 base_string--;
3849 if (*base_string == ')')
3850 parens_balanced++;
3851 if (*base_string == '(')
3852 parens_balanced--;
3853 }
3854 while (parens_balanced);
3855
3856 temp_string = base_string;
3857
3858 /* Skip past '(' and whitespace. */
3859 ++base_string;
3860 if (is_space_char (*base_string))
3861 ++base_string;
3862
3863 if (*base_string == ','
3864 || ((*base_string == REGISTER_PREFIX || allow_naked_reg)
3865 && (i.base_reg = parse_register (base_string, &end_op)) != NULL))
3866 {
3867 displacement_string_end = temp_string;
3868
3869 i.types[this_operand] |= BaseIndex;
3870
3871 if (i.base_reg)
3872 {
3873 base_string = end_op;
3874 if (is_space_char (*base_string))
3875 ++base_string;
3876 }
3877
3878 /* There may be an index reg or scale factor here. */
3879 if (*base_string == ',')
3880 {
3881 ++base_string;
3882 if (is_space_char (*base_string))
3883 ++base_string;
3884
3885 if ((*base_string == REGISTER_PREFIX || allow_naked_reg)
3886 && (i.index_reg = parse_register (base_string, &end_op)) != NULL)
3887 {
3888 base_string = end_op;
3889 if (is_space_char (*base_string))
3890 ++base_string;
3891 if (*base_string == ',')
3892 {
3893 ++base_string;
3894 if (is_space_char (*base_string))
3895 ++base_string;
3896 }
3897 else if (*base_string != ')')
3898 {
3899 as_bad (_("expecting `,' or `)' after index register in `%s'"),
3900 operand_string);
3901 return 0;
3902 }
3903 }
3904 else if (*base_string == REGISTER_PREFIX)
3905 {
3906 as_bad (_("bad register name `%s'"), base_string);
3907 return 0;
3908 }
3909
3910 /* Check for scale factor. */
3911 if (*base_string != ')')
3912 {
3913 char *end_scale = i386_scale (base_string);
3914
3915 if (!end_scale)
3916 return 0;
3917
3918 base_string = end_scale;
3919 if (is_space_char (*base_string))
3920 ++base_string;
3921 if (*base_string != ')')
3922 {
3923 as_bad (_("expecting `)' after scale factor in `%s'"),
3924 operand_string);
3925 return 0;
3926 }
3927 }
3928 else if (!i.index_reg)
3929 {
3930 as_bad (_("expecting index register or scale factor after `,'; got '%c'"),
3931 *base_string);
3932 return 0;
3933 }
3934 }
3935 else if (*base_string != ')')
3936 {
3937 as_bad (_("expecting `,' or `)' after base register in `%s'"),
3938 operand_string);
3939 return 0;
3940 }
3941 }
3942 else if (*base_string == REGISTER_PREFIX)
3943 {
3944 as_bad (_("bad register name `%s'"), base_string);
3945 return 0;
3946 }
3947 }
3948
3949 /* If there's an expression beginning the operand, parse it,
3950 assuming displacement_string_start and
3951 displacement_string_end are meaningful. */
3952 if (displacement_string_start != displacement_string_end)
3953 {
3954 if (!i386_displacement (displacement_string_start,
3955 displacement_string_end))
3956 return 0;
3957 }
3958
3959 /* Special case for (%dx) while doing input/output op. */
3960 if (i.base_reg
3961 && i.base_reg->reg_type == (Reg16 | InOutPortReg)
3962 && i.index_reg == 0
3963 && i.log2_scale_factor == 0
3964 && i.seg[i.mem_operands] == 0
3965 && (i.types[this_operand] & Disp) == 0)
3966 {
3967 i.types[this_operand] = InOutPortReg;
3968 return 1;
3969 }
3970
3971 if (i386_index_check (operand_string) == 0)
3972 return 0;
3973 i.mem_operands++;
3974 }
3975 else
3976 {
3977 /* It's not a memory operand; argh! */
3978 as_bad (_("invalid char %s beginning operand %d `%s'"),
3979 output_invalid (*op_string),
3980 this_operand + 1,
3981 op_string);
3982 return 0;
3983 }
3984 return 1; /* Normal return. */
3985 }
3986 \f
3987 /* md_estimate_size_before_relax()
3988
3989 Called just before relax() for rs_machine_dependent frags. The x86
3990 assembler uses these frags to handle variable size jump
3991 instructions.
3992
3993 Any symbol that is now undefined will not become defined.
3994 Return the correct fr_subtype in the frag.
3995 Return the initial "guess for variable size of frag" to caller.
3996 The guess is actually the growth beyond the fixed part. Whatever
3997 we do to grow the fixed or variable part contributes to our
3998 returned value. */
3999
4000 int
4001 md_estimate_size_before_relax (fragP, segment)
4002 register fragS *fragP;
4003 register segT segment;
4004 {
4005 /* We've already got fragP->fr_subtype right; all we have to do is
4006 check for un-relaxable symbols. On an ELF system, we can't relax
4007 an externally visible symbol, because it may be overridden by a
4008 shared library. */
4009 if (S_GET_SEGMENT (fragP->fr_symbol) != segment
4010 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4011 || S_IS_EXTERNAL (fragP->fr_symbol)
4012 || S_IS_WEAK (fragP->fr_symbol)
4013 #endif
4014 )
4015 {
4016 /* Symbol is undefined in this segment, or we need to keep a
4017 reloc so that weak symbols can be overridden. */
4018 int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
4019 RELOC_ENUM reloc_type;
4020 unsigned char *opcode;
4021 int old_fr_fix;
4022
4023 if (fragP->fr_var != NO_RELOC)
4024 reloc_type = fragP->fr_var;
4025 else if (size == 2)
4026 reloc_type = BFD_RELOC_16_PCREL;
4027 else
4028 reloc_type = BFD_RELOC_32_PCREL;
4029
4030 old_fr_fix = fragP->fr_fix;
4031 opcode = (unsigned char *) fragP->fr_opcode;
4032
4033 switch (TYPE_FROM_RELAX_STATE (fragP->fr_subtype))
4034 {
4035 case UNCOND_JUMP:
4036 /* Make jmp (0xeb) a (d)word displacement jump. */
4037 opcode[0] = 0xe9;
4038 fragP->fr_fix += size;
4039 fix_new (fragP, old_fr_fix, size,
4040 fragP->fr_symbol,
4041 fragP->fr_offset, 1,
4042 reloc_type);
4043 break;
4044
4045 case COND_JUMP86:
4046 if (no_cond_jump_promotion)
4047 goto relax_guess;
4048
4049 if (size == 2)
4050 {
4051 /* Negate the condition, and branch past an
4052 unconditional jump. */
4053 opcode[0] ^= 1;
4054 opcode[1] = 3;
4055 /* Insert an unconditional jump. */
4056 opcode[2] = 0xe9;
4057 /* We added two extra opcode bytes, and have a two byte
4058 offset. */
4059 fragP->fr_fix += 2 + 2;
4060 fix_new (fragP, old_fr_fix + 2, 2,
4061 fragP->fr_symbol,
4062 fragP->fr_offset, 1,
4063 reloc_type);
4064 break;
4065 }
4066 /* Fall through. */
4067
4068 case COND_JUMP:
4069 if (no_cond_jump_promotion)
4070 goto relax_guess;
4071
4072 /* This changes the byte-displacement jump 0x7N
4073 to the (d)word-displacement jump 0x0f,0x8N. */
4074 opcode[1] = opcode[0] + 0x10;
4075 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4076 /* We've added an opcode byte. */
4077 fragP->fr_fix += 1 + size;
4078 fix_new (fragP, old_fr_fix + 1, size,
4079 fragP->fr_symbol,
4080 fragP->fr_offset, 1,
4081 reloc_type);
4082 break;
4083
4084 default:
4085 BAD_CASE (fragP->fr_subtype);
4086 break;
4087 }
4088 frag_wane (fragP);
4089 return fragP->fr_fix - old_fr_fix;
4090 }
4091
4092 relax_guess:
4093 /* Guess size depending on current relax state. Initially the relax
4094 state will correspond to a short jump and we return 1, because
4095 the variable part of the frag (the branch offset) is one byte
4096 long. However, we can relax a section more than once and in that
4097 case we must either set fr_subtype back to the unrelaxed state,
4098 or return the value for the appropriate branch. */
4099 return md_relax_table[fragP->fr_subtype].rlx_length;
4100 }
4101
4102 /* Called after relax() is finished.
4103
4104 In: Address of frag.
4105 fr_type == rs_machine_dependent.
4106 fr_subtype is what the address relaxed to.
4107
4108 Out: Any fixSs and constants are set up.
4109 Caller will turn frag into a ".space 0". */
4110
4111 #ifndef BFD_ASSEMBLER
4112 void
4113 md_convert_frag (headers, sec, fragP)
4114 object_headers *headers ATTRIBUTE_UNUSED;
4115 segT sec ATTRIBUTE_UNUSED;
4116 register fragS *fragP;
4117 #else
4118 void
4119 md_convert_frag (abfd, sec, fragP)
4120 bfd *abfd ATTRIBUTE_UNUSED;
4121 segT sec ATTRIBUTE_UNUSED;
4122 register fragS *fragP;
4123 #endif
4124 {
4125 register unsigned char *opcode;
4126 unsigned char *where_to_put_displacement = NULL;
4127 offsetT target_address;
4128 offsetT opcode_address;
4129 unsigned int extension = 0;
4130 offsetT displacement_from_opcode_start;
4131
4132 opcode = (unsigned char *) fragP->fr_opcode;
4133
4134 /* Address we want to reach in file space. */
4135 target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
4136
4137 /* Address opcode resides at in file space. */
4138 opcode_address = fragP->fr_address + fragP->fr_fix;
4139
4140 /* Displacement from opcode start to fill into instruction. */
4141 displacement_from_opcode_start = target_address - opcode_address;
4142
4143 if ((fragP->fr_subtype & BIG) == 0)
4144 {
4145 /* Don't have to change opcode. */
4146 extension = 1; /* 1 opcode + 1 displacement */
4147 where_to_put_displacement = &opcode[1];
4148 }
4149 else
4150 {
4151 if (no_cond_jump_promotion
4152 && TYPE_FROM_RELAX_STATE (fragP->fr_subtype) != UNCOND_JUMP)
4153 as_warn_where (fragP->fr_file, fragP->fr_line, _("long jump required"));
4154
4155 switch (fragP->fr_subtype)
4156 {
4157 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
4158 extension = 4; /* 1 opcode + 4 displacement */
4159 opcode[0] = 0xe9;
4160 where_to_put_displacement = &opcode[1];
4161 break;
4162
4163 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
4164 extension = 2; /* 1 opcode + 2 displacement */
4165 opcode[0] = 0xe9;
4166 where_to_put_displacement = &opcode[1];
4167 break;
4168
4169 case ENCODE_RELAX_STATE (COND_JUMP, BIG):
4170 case ENCODE_RELAX_STATE (COND_JUMP86, BIG):
4171 extension = 5; /* 2 opcode + 4 displacement */
4172 opcode[1] = opcode[0] + 0x10;
4173 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4174 where_to_put_displacement = &opcode[2];
4175 break;
4176
4177 case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
4178 extension = 3; /* 2 opcode + 2 displacement */
4179 opcode[1] = opcode[0] + 0x10;
4180 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4181 where_to_put_displacement = &opcode[2];
4182 break;
4183
4184 case ENCODE_RELAX_STATE (COND_JUMP86, BIG16):
4185 extension = 4;
4186 opcode[0] ^= 1;
4187 opcode[1] = 3;
4188 opcode[2] = 0xe9;
4189 where_to_put_displacement = &opcode[3];
4190 break;
4191
4192 default:
4193 BAD_CASE (fragP->fr_subtype);
4194 break;
4195 }
4196 }
4197
4198 /* Now put displacement after opcode. */
4199 md_number_to_chars ((char *) where_to_put_displacement,
4200 (valueT) (displacement_from_opcode_start - extension),
4201 DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
4202 fragP->fr_fix += extension;
4203 }
4204 \f
4205 /* Size of byte displacement jmp. */
4206 int md_short_jump_size = 2;
4207
4208 /* Size of dword displacement jmp. */
4209 int md_long_jump_size = 5;
4210
4211 /* Size of relocation record. */
4212 const int md_reloc_size = 8;
4213
4214 void
4215 md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
4216 char *ptr;
4217 addressT from_addr, to_addr;
4218 fragS *frag ATTRIBUTE_UNUSED;
4219 symbolS *to_symbol ATTRIBUTE_UNUSED;
4220 {
4221 offsetT offset;
4222
4223 offset = to_addr - (from_addr + 2);
4224 /* Opcode for byte-disp jump. */
4225 md_number_to_chars (ptr, (valueT) 0xeb, 1);
4226 md_number_to_chars (ptr + 1, (valueT) offset, 1);
4227 }
4228
4229 void
4230 md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
4231 char *ptr;
4232 addressT from_addr, to_addr;
4233 fragS *frag ATTRIBUTE_UNUSED;
4234 symbolS *to_symbol ATTRIBUTE_UNUSED;
4235 {
4236 offsetT offset;
4237
4238 offset = to_addr - (from_addr + 5);
4239 md_number_to_chars (ptr, (valueT) 0xe9, 1);
4240 md_number_to_chars (ptr + 1, (valueT) offset, 4);
4241 }
4242 \f
4243 /* Apply a fixup (fixS) to segment data, once it has been determined
4244 by our caller that we have all the info we need to fix it up.
4245
4246 On the 386, immediates, displacements, and data pointers are all in
4247 the same (little-endian) format, so we don't need to care about which
4248 we are handling. */
4249
4250 void
4251 md_apply_fix3 (fixP, valP, seg)
4252 /* The fix we're to put in. */
4253 fixS *fixP;
4254 /* Pointer to the value of the bits. */
4255 valueT * valP;
4256 /* Segment fix is from. */
4257 segT seg ATTRIBUTE_UNUSED;
4258 {
4259 char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
4260 valueT value = * valP;
4261
4262 #if defined (BFD_ASSEMBLER) && !defined (TE_Mach)
4263 if (fixP->fx_pcrel)
4264 {
4265 switch (fixP->fx_r_type)
4266 {
4267 default:
4268 break;
4269
4270 case BFD_RELOC_32:
4271 fixP->fx_r_type = BFD_RELOC_32_PCREL;
4272 break;
4273 case BFD_RELOC_16:
4274 fixP->fx_r_type = BFD_RELOC_16_PCREL;
4275 break;
4276 case BFD_RELOC_8:
4277 fixP->fx_r_type = BFD_RELOC_8_PCREL;
4278 break;
4279 }
4280 }
4281
4282 /* This is a hack. There should be a better way to handle this.
4283 This covers for the fact that bfd_install_relocation will
4284 subtract the current location (for partial_inplace, PC relative
4285 relocations); see more below. */
4286 if ((fixP->fx_r_type == BFD_RELOC_32_PCREL
4287 || fixP->fx_r_type == BFD_RELOC_16_PCREL
4288 || fixP->fx_r_type == BFD_RELOC_8_PCREL)
4289 && fixP->fx_addsy && !use_rela_relocations)
4290 {
4291 #ifndef OBJ_AOUT
4292 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4293 #ifdef TE_PE
4294 || OUTPUT_FLAVOR == bfd_target_coff_flavour
4295 #endif
4296 )
4297 value += fixP->fx_where + fixP->fx_frag->fr_address;
4298 #endif
4299 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4300 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
4301 {
4302 segT fseg = S_GET_SEGMENT (fixP->fx_addsy);
4303
4304 if ((fseg == seg
4305 || (symbol_section_p (fixP->fx_addsy)
4306 && fseg != absolute_section))
4307 && ! S_IS_EXTERNAL (fixP->fx_addsy)
4308 && ! S_IS_WEAK (fixP->fx_addsy)
4309 && S_IS_DEFINED (fixP->fx_addsy)
4310 && ! S_IS_COMMON (fixP->fx_addsy))
4311 {
4312 /* Yes, we add the values in twice. This is because
4313 bfd_perform_relocation subtracts them out again. I think
4314 bfd_perform_relocation is broken, but I don't dare change
4315 it. FIXME. */
4316 value += fixP->fx_where + fixP->fx_frag->fr_address;
4317 }
4318 }
4319 #endif
4320 #if defined (OBJ_COFF) && defined (TE_PE)
4321 /* For some reason, the PE format does not store a section
4322 address offset for a PC relative symbol. */
4323 if (S_GET_SEGMENT (fixP->fx_addsy) != seg)
4324 value += md_pcrel_from (fixP);
4325 #endif
4326 }
4327
4328 /* Fix a few things - the dynamic linker expects certain values here,
4329 and we must not dissappoint it. */
4330 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4331 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4332 && fixP->fx_addsy)
4333 switch (fixP->fx_r_type)
4334 {
4335 case BFD_RELOC_386_PLT32:
4336 case BFD_RELOC_X86_64_PLT32:
4337 /* Make the jump instruction point to the address of the operand. At
4338 runtime we merely add the offset to the actual PLT entry. */
4339 value = -4;
4340 break;
4341 case BFD_RELOC_386_GOTPC:
4342
4343 /* This is tough to explain. We end up with this one if we have
4344 * operands that look like "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal
4345 * here is to obtain the absolute address of the GOT, and it is strongly
4346 * preferable from a performance point of view to avoid using a runtime
4347 * relocation for this. The actual sequence of instructions often look
4348 * something like:
4349 *
4350 * call .L66
4351 * .L66:
4352 * popl %ebx
4353 * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
4354 *
4355 * The call and pop essentially return the absolute address of
4356 * the label .L66 and store it in %ebx. The linker itself will
4357 * ultimately change the first operand of the addl so that %ebx points to
4358 * the GOT, but to keep things simple, the .o file must have this operand
4359 * set so that it generates not the absolute address of .L66, but the
4360 * absolute address of itself. This allows the linker itself simply
4361 * treat a GOTPC relocation as asking for a pcrel offset to the GOT to be
4362 * added in, and the addend of the relocation is stored in the operand
4363 * field for the instruction itself.
4364 *
4365 * Our job here is to fix the operand so that it would add the correct
4366 * offset so that %ebx would point to itself. The thing that is tricky is
4367 * that .-.L66 will point to the beginning of the instruction, so we need
4368 * to further modify the operand so that it will point to itself.
4369 * There are other cases where you have something like:
4370 *
4371 * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
4372 *
4373 * and here no correction would be required. Internally in the assembler
4374 * we treat operands of this form as not being pcrel since the '.' is
4375 * explicitly mentioned, and I wonder whether it would simplify matters
4376 * to do it this way. Who knows. In earlier versions of the PIC patches,
4377 * the pcrel_adjust field was used to store the correction, but since the
4378 * expression is not pcrel, I felt it would be confusing to do it this
4379 * way. */
4380
4381 value -= 1;
4382 break;
4383 case BFD_RELOC_386_GOT32:
4384 case BFD_RELOC_X86_64_GOT32:
4385 value = 0; /* Fully resolved at runtime. No addend. */
4386 break;
4387 case BFD_RELOC_386_GOTOFF:
4388 case BFD_RELOC_X86_64_GOTPCREL:
4389 break;
4390
4391 case BFD_RELOC_VTABLE_INHERIT:
4392 case BFD_RELOC_VTABLE_ENTRY:
4393 fixP->fx_done = 0;
4394 return;
4395
4396 default:
4397 break;
4398 }
4399 #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
4400 * valP = value;
4401 #endif /* defined (BFD_ASSEMBLER) && !defined (TE_Mach) */
4402
4403 /* Are we finished with this relocation now? */
4404 if (fixP->fx_addsy == NULL && fixP->fx_pcrel == 0)
4405 fixP->fx_done = 1;
4406 #ifdef BFD_ASSEMBLER
4407 else if (use_rela_relocations)
4408 {
4409 fixP->fx_no_overflow = 1;
4410 value = 0;
4411 }
4412 #endif
4413 md_number_to_chars (p, value, fixP->fx_size);
4414 }
4415 \f
4416 #define MAX_LITTLENUMS 6
4417
4418 /* Turn the string pointed to by litP into a floating point constant
4419 of type TYPE, and emit the appropriate bytes. The number of
4420 LITTLENUMS emitted is stored in *SIZEP. An error message is
4421 returned, or NULL on OK. */
4422
4423 char *
4424 md_atof (type, litP, sizeP)
4425 int type;
4426 char *litP;
4427 int *sizeP;
4428 {
4429 int prec;
4430 LITTLENUM_TYPE words[MAX_LITTLENUMS];
4431 LITTLENUM_TYPE *wordP;
4432 char *t;
4433
4434 switch (type)
4435 {
4436 case 'f':
4437 case 'F':
4438 prec = 2;
4439 break;
4440
4441 case 'd':
4442 case 'D':
4443 prec = 4;
4444 break;
4445
4446 case 'x':
4447 case 'X':
4448 prec = 5;
4449 break;
4450
4451 default:
4452 *sizeP = 0;
4453 return _("Bad call to md_atof ()");
4454 }
4455 t = atof_ieee (input_line_pointer, type, words);
4456 if (t)
4457 input_line_pointer = t;
4458
4459 *sizeP = prec * sizeof (LITTLENUM_TYPE);
4460 /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
4461 the bigendian 386. */
4462 for (wordP = words + prec - 1; prec--;)
4463 {
4464 md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
4465 litP += sizeof (LITTLENUM_TYPE);
4466 }
4467 return 0;
4468 }
4469 \f
4470 char output_invalid_buf[8];
4471
4472 static char *
4473 output_invalid (c)
4474 int c;
4475 {
4476 if (ISPRINT (c))
4477 sprintf (output_invalid_buf, "'%c'", c);
4478 else
4479 sprintf (output_invalid_buf, "(0x%x)", (unsigned) c);
4480 return output_invalid_buf;
4481 }
4482
4483 /* REG_STRING starts *before* REGISTER_PREFIX. */
4484
4485 static const reg_entry *
4486 parse_register (reg_string, end_op)
4487 char *reg_string;
4488 char **end_op;
4489 {
4490 char *s = reg_string;
4491 char *p;
4492 char reg_name_given[MAX_REG_NAME_SIZE + 1];
4493 const reg_entry *r;
4494
4495 /* Skip possible REGISTER_PREFIX and possible whitespace. */
4496 if (*s == REGISTER_PREFIX)
4497 ++s;
4498
4499 if (is_space_char (*s))
4500 ++s;
4501
4502 p = reg_name_given;
4503 while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
4504 {
4505 if (p >= reg_name_given + MAX_REG_NAME_SIZE)
4506 return (const reg_entry *) NULL;
4507 s++;
4508 }
4509
4510 /* For naked regs, make sure that we are not dealing with an identifier.
4511 This prevents confusing an identifier like `eax_var' with register
4512 `eax'. */
4513 if (allow_naked_reg && identifier_chars[(unsigned char) *s])
4514 return (const reg_entry *) NULL;
4515
4516 *end_op = s;
4517
4518 r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
4519
4520 /* Handle floating point regs, allowing spaces in the (i) part. */
4521 if (r == i386_regtab /* %st is first entry of table */)
4522 {
4523 if (is_space_char (*s))
4524 ++s;
4525 if (*s == '(')
4526 {
4527 ++s;
4528 if (is_space_char (*s))
4529 ++s;
4530 if (*s >= '0' && *s <= '7')
4531 {
4532 r = &i386_float_regtab[*s - '0'];
4533 ++s;
4534 if (is_space_char (*s))
4535 ++s;
4536 if (*s == ')')
4537 {
4538 *end_op = s + 1;
4539 return r;
4540 }
4541 }
4542 /* We have "%st(" then garbage. */
4543 return (const reg_entry *) NULL;
4544 }
4545 }
4546
4547 if (r != NULL
4548 && r->reg_flags & (RegRex64|RegRex)
4549 && flag_code != CODE_64BIT)
4550 {
4551 return (const reg_entry *) NULL;
4552 }
4553
4554 return r;
4555 }
4556 \f
4557 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4558 const char *md_shortopts = "kVQ:sq";
4559 #else
4560 const char *md_shortopts = "q";
4561 #endif
4562
4563 struct option md_longopts[] = {
4564 #define OPTION_32 (OPTION_MD_BASE + 0)
4565 {"32", no_argument, NULL, OPTION_32},
4566 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4567 #define OPTION_64 (OPTION_MD_BASE + 1)
4568 {"64", no_argument, NULL, OPTION_64},
4569 #endif
4570 {NULL, no_argument, NULL, 0}
4571 };
4572 size_t md_longopts_size = sizeof (md_longopts);
4573
4574 int
4575 md_parse_option (c, arg)
4576 int c;
4577 char *arg ATTRIBUTE_UNUSED;
4578 {
4579 switch (c)
4580 {
4581 case 'q':
4582 quiet_warnings = 1;
4583 break;
4584
4585 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4586 /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
4587 should be emitted or not. FIXME: Not implemented. */
4588 case 'Q':
4589 break;
4590
4591 /* -V: SVR4 argument to print version ID. */
4592 case 'V':
4593 print_version_id ();
4594 break;
4595
4596 /* -k: Ignore for FreeBSD compatibility. */
4597 case 'k':
4598 break;
4599
4600 case 's':
4601 /* -s: On i386 Solaris, this tells the native assembler to use
4602 .stab instead of .stab.excl. We always use .stab anyhow. */
4603 break;
4604
4605 case OPTION_64:
4606 {
4607 const char **list, **l;
4608
4609 list = bfd_target_list ();
4610 for (l = list; *l != NULL; l++)
4611 if (strcmp (*l, "elf64-x86-64") == 0)
4612 {
4613 default_arch = "x86_64";
4614 break;
4615 }
4616 if (*l == NULL)
4617 as_fatal (_("No compiled in support for x86_64"));
4618 free (list);
4619 }
4620 break;
4621 #endif
4622
4623 case OPTION_32:
4624 default_arch = "i386";
4625 break;
4626
4627 default:
4628 return 0;
4629 }
4630 return 1;
4631 }
4632
4633 void
4634 md_show_usage (stream)
4635 FILE *stream;
4636 {
4637 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4638 fprintf (stream, _("\
4639 -Q ignored\n\
4640 -V print assembler version number\n\
4641 -k ignored\n\
4642 -q quieten some warnings\n\
4643 -s ignored\n"));
4644 #else
4645 fprintf (stream, _("\
4646 -q quieten some warnings\n"));
4647 #endif
4648 }
4649
4650 #ifdef BFD_ASSEMBLER
4651 #if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
4652 || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
4653
4654 /* Pick the target format to use. */
4655
4656 const char *
4657 i386_target_format ()
4658 {
4659 if (!strcmp (default_arch, "x86_64"))
4660 set_code_flag (CODE_64BIT);
4661 else if (!strcmp (default_arch, "i386"))
4662 set_code_flag (CODE_32BIT);
4663 else
4664 as_fatal (_("Unknown architecture"));
4665 switch (OUTPUT_FLAVOR)
4666 {
4667 #ifdef OBJ_MAYBE_AOUT
4668 case bfd_target_aout_flavour:
4669 return AOUT_TARGET_FORMAT;
4670 #endif
4671 #ifdef OBJ_MAYBE_COFF
4672 case bfd_target_coff_flavour:
4673 return "coff-i386";
4674 #endif
4675 #if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF)
4676 case bfd_target_elf_flavour:
4677 {
4678 if (flag_code == CODE_64BIT)
4679 use_rela_relocations = 1;
4680 return flag_code == CODE_64BIT ? "elf64-x86-64" : "elf32-i386";
4681 }
4682 #endif
4683 default:
4684 abort ();
4685 return NULL;
4686 }
4687 }
4688
4689 #endif /* OBJ_MAYBE_ more than one */
4690
4691 #if (defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
4692 void i386_elf_emit_arch_note ()
4693 {
4694 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4695 && cpu_arch_name != NULL)
4696 {
4697 char *p;
4698 asection *seg = now_seg;
4699 subsegT subseg = now_subseg;
4700 Elf_Internal_Note i_note;
4701 Elf_External_Note e_note;
4702 asection *note_secp;
4703 int len;
4704
4705 /* Create the .note section. */
4706 note_secp = subseg_new (".note", 0);
4707 bfd_set_section_flags (stdoutput,
4708 note_secp,
4709 SEC_HAS_CONTENTS | SEC_READONLY);
4710
4711 /* Process the arch string. */
4712 len = strlen (cpu_arch_name);
4713
4714 i_note.namesz = len + 1;
4715 i_note.descsz = 0;
4716 i_note.type = NT_ARCH;
4717 p = frag_more (sizeof (e_note.namesz));
4718 md_number_to_chars (p, (valueT) i_note.namesz, sizeof (e_note.namesz));
4719 p = frag_more (sizeof (e_note.descsz));
4720 md_number_to_chars (p, (valueT) i_note.descsz, sizeof (e_note.descsz));
4721 p = frag_more (sizeof (e_note.type));
4722 md_number_to_chars (p, (valueT) i_note.type, sizeof (e_note.type));
4723 p = frag_more (len + 1);
4724 strcpy (p, cpu_arch_name);
4725
4726 frag_align (2, 0, 0);
4727
4728 subseg_set (seg, subseg);
4729 }
4730 }
4731 #endif
4732 #endif /* BFD_ASSEMBLER */
4733 \f
4734 symbolS *
4735 md_undefined_symbol (name)
4736 char *name;
4737 {
4738 if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
4739 && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
4740 && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
4741 && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
4742 {
4743 if (!GOT_symbol)
4744 {
4745 if (symbol_find (name))
4746 as_bad (_("GOT already in symbol table"));
4747 GOT_symbol = symbol_new (name, undefined_section,
4748 (valueT) 0, &zero_address_frag);
4749 };
4750 return GOT_symbol;
4751 }
4752 return 0;
4753 }
4754
4755 /* Round up a section size to the appropriate boundary. */
4756
4757 valueT
4758 md_section_align (segment, size)
4759 segT segment ATTRIBUTE_UNUSED;
4760 valueT size;
4761 {
4762 #ifdef BFD_ASSEMBLER
4763 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4764 if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
4765 {
4766 /* For a.out, force the section size to be aligned. If we don't do
4767 this, BFD will align it for us, but it will not write out the
4768 final bytes of the section. This may be a bug in BFD, but it is
4769 easier to fix it here since that is how the other a.out targets
4770 work. */
4771 int align;
4772
4773 align = bfd_get_section_alignment (stdoutput, segment);
4774 size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
4775 }
4776 #endif
4777 #endif
4778
4779 return size;
4780 }
4781
4782 /* On the i386, PC-relative offsets are relative to the start of the
4783 next instruction. That is, the address of the offset, plus its
4784 size, since the offset is always the last part of the insn. */
4785
4786 long
4787 md_pcrel_from (fixP)
4788 fixS *fixP;
4789 {
4790 return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
4791 }
4792
4793 #ifndef I386COFF
4794
4795 static void
4796 s_bss (ignore)
4797 int ignore ATTRIBUTE_UNUSED;
4798 {
4799 register int temp;
4800
4801 temp = get_absolute_expression ();
4802 subseg_set (bss_section, (subsegT) temp);
4803 demand_empty_rest_of_line ();
4804 }
4805
4806 #endif
4807
4808 #ifdef BFD_ASSEMBLER
4809
4810 void
4811 i386_validate_fix (fixp)
4812 fixS *fixp;
4813 {
4814 if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
4815 {
4816 /* GOTOFF relocation are nonsense in 64bit mode. */
4817 if (fixp->fx_r_type == BFD_RELOC_32_PCREL)
4818 {
4819 if (flag_code != CODE_64BIT)
4820 abort ();
4821 fixp->fx_r_type = BFD_RELOC_X86_64_GOTPCREL;
4822 }
4823 else
4824 {
4825 if (flag_code == CODE_64BIT)
4826 abort ();
4827 fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
4828 }
4829 fixp->fx_subsy = 0;
4830 }
4831 }
4832
4833 arelent *
4834 tc_gen_reloc (section, fixp)
4835 asection *section ATTRIBUTE_UNUSED;
4836 fixS *fixp;
4837 {
4838 arelent *rel;
4839 bfd_reloc_code_real_type code;
4840
4841 switch (fixp->fx_r_type)
4842 {
4843 case BFD_RELOC_X86_64_PLT32:
4844 case BFD_RELOC_X86_64_GOT32:
4845 case BFD_RELOC_X86_64_GOTPCREL:
4846 case BFD_RELOC_386_PLT32:
4847 case BFD_RELOC_386_GOT32:
4848 case BFD_RELOC_386_GOTOFF:
4849 case BFD_RELOC_386_GOTPC:
4850 case BFD_RELOC_X86_64_32S:
4851 case BFD_RELOC_RVA:
4852 case BFD_RELOC_VTABLE_ENTRY:
4853 case BFD_RELOC_VTABLE_INHERIT:
4854 code = fixp->fx_r_type;
4855 break;
4856 default:
4857 if (fixp->fx_pcrel)
4858 {
4859 switch (fixp->fx_size)
4860 {
4861 default:
4862 as_bad_where (fixp->fx_file, fixp->fx_line,
4863 _("can not do %d byte pc-relative relocation"),
4864 fixp->fx_size);
4865 code = BFD_RELOC_32_PCREL;
4866 break;
4867 case 1: code = BFD_RELOC_8_PCREL; break;
4868 case 2: code = BFD_RELOC_16_PCREL; break;
4869 case 4: code = BFD_RELOC_32_PCREL; break;
4870 }
4871 }
4872 else
4873 {
4874 switch (fixp->fx_size)
4875 {
4876 default:
4877 as_bad_where (fixp->fx_file, fixp->fx_line,
4878 _("can not do %d byte relocation"),
4879 fixp->fx_size);
4880 code = BFD_RELOC_32;
4881 break;
4882 case 1: code = BFD_RELOC_8; break;
4883 case 2: code = BFD_RELOC_16; break;
4884 case 4: code = BFD_RELOC_32; break;
4885 case 8: code = BFD_RELOC_64; break;
4886 }
4887 }
4888 break;
4889 }
4890
4891 if (code == BFD_RELOC_32
4892 && GOT_symbol
4893 && fixp->fx_addsy == GOT_symbol)
4894 {
4895 /* We don't support GOTPC on 64bit targets. */
4896 if (flag_code == CODE_64BIT)
4897 abort ();
4898 code = BFD_RELOC_386_GOTPC;
4899 }
4900
4901 rel = (arelent *) xmalloc (sizeof (arelent));
4902 rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
4903 *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
4904
4905 rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
4906 if (!use_rela_relocations)
4907 {
4908 /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
4909 vtable entry to be used in the relocation's section offset. */
4910 if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
4911 rel->address = fixp->fx_offset;
4912
4913 if (fixp->fx_pcrel)
4914 rel->addend = fixp->fx_addnumber;
4915 else
4916 rel->addend = 0;
4917 }
4918 /* Use the rela in 64bit mode. */
4919 else
4920 {
4921 rel->addend = fixp->fx_offset;
4922 if (fixp->fx_pcrel)
4923 rel->addend -= fixp->fx_size;
4924 }
4925
4926 rel->howto = bfd_reloc_type_lookup (stdoutput, code);
4927 if (rel->howto == NULL)
4928 {
4929 as_bad_where (fixp->fx_file, fixp->fx_line,
4930 _("cannot represent relocation type %s"),
4931 bfd_get_reloc_code_name (code));
4932 /* Set howto to a garbage value so that we can keep going. */
4933 rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
4934 assert (rel->howto != NULL);
4935 }
4936
4937 return rel;
4938 }
4939
4940 #else /* ! BFD_ASSEMBLER */
4941
4942 #if (defined(OBJ_AOUT) | defined(OBJ_BOUT))
4943 void
4944 tc_aout_fix_to_chars (where, fixP, segment_address_in_file)
4945 char *where;
4946 fixS *fixP;
4947 relax_addressT segment_address_in_file;
4948 {
4949 /* In: length of relocation (or of address) in chars: 1, 2 or 4.
4950 Out: GNU LD relocation length code: 0, 1, or 2. */
4951
4952 static const unsigned char nbytes_r_length[] = { 42, 0, 1, 42, 2 };
4953 long r_symbolnum;
4954
4955 know (fixP->fx_addsy != NULL);
4956
4957 md_number_to_chars (where,
4958 (valueT) (fixP->fx_frag->fr_address
4959 + fixP->fx_where - segment_address_in_file),
4960 4);
4961
4962 r_symbolnum = (S_IS_DEFINED (fixP->fx_addsy)
4963 ? S_GET_TYPE (fixP->fx_addsy)
4964 : fixP->fx_addsy->sy_number);
4965
4966 where[6] = (r_symbolnum >> 16) & 0x0ff;
4967 where[5] = (r_symbolnum >> 8) & 0x0ff;
4968 where[4] = r_symbolnum & 0x0ff;
4969 where[7] = ((((!S_IS_DEFINED (fixP->fx_addsy)) << 3) & 0x08)
4970 | ((nbytes_r_length[fixP->fx_size] << 1) & 0x06)
4971 | (((fixP->fx_pcrel << 0) & 0x01) & 0x0f));
4972 }
4973
4974 #endif /* OBJ_AOUT or OBJ_BOUT. */
4975
4976 #if defined (I386COFF)
4977
4978 short
4979 tc_coff_fix2rtype (fixP)
4980 fixS *fixP;
4981 {
4982 if (fixP->fx_r_type == R_IMAGEBASE)
4983 return R_IMAGEBASE;
4984
4985 return (fixP->fx_pcrel ?
4986 (fixP->fx_size == 1 ? R_PCRBYTE :
4987 fixP->fx_size == 2 ? R_PCRWORD :
4988 R_PCRLONG) :
4989 (fixP->fx_size == 1 ? R_RELBYTE :
4990 fixP->fx_size == 2 ? R_RELWORD :
4991 R_DIR32));
4992 }
4993
4994 int
4995 tc_coff_sizemachdep (frag)
4996 fragS *frag;
4997 {
4998 if (frag->fr_next)
4999 return (frag->fr_next->fr_address - frag->fr_address);
5000 else
5001 return 0;
5002 }
5003
5004 #endif /* I386COFF */
5005
5006 #endif /* ! BFD_ASSEMBLER */
5007 \f
5008 /* Parse operands using Intel syntax. This implements a recursive descent
5009 parser based on the BNF grammar published in Appendix B of the MASM 6.1
5010 Programmer's Guide.
5011
5012 FIXME: We do not recognize the full operand grammar defined in the MASM
5013 documentation. In particular, all the structure/union and
5014 high-level macro operands are missing.
5015
5016 Uppercase words are terminals, lower case words are non-terminals.
5017 Objects surrounded by double brackets '[[' ']]' are optional. Vertical
5018 bars '|' denote choices. Most grammar productions are implemented in
5019 functions called 'intel_<production>'.
5020
5021 Initial production is 'expr'.
5022
5023 addOp + | -
5024
5025 alpha [a-zA-Z]
5026
5027 byteRegister AL | AH | BL | BH | CL | CH | DL | DH
5028
5029 constant digits [[ radixOverride ]]
5030
5031 dataType BYTE | WORD | DWORD | QWORD | XWORD
5032
5033 digits decdigit
5034 | digits decdigit
5035 | digits hexdigit
5036
5037 decdigit [0-9]
5038
5039 e05 e05 addOp e06
5040 | e06
5041
5042 e06 e06 mulOp e09
5043 | e09
5044
5045 e09 OFFSET e10
5046 | e09 PTR e10
5047 | e09 : e10
5048 | e10
5049
5050 e10 e10 [ expr ]
5051 | e11
5052
5053 e11 ( expr )
5054 | [ expr ]
5055 | constant
5056 | dataType
5057 | id
5058 | $
5059 | register
5060
5061 => expr SHORT e05
5062 | e05
5063
5064 gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
5065 | BP | EBP | SP | ESP | DI | EDI | SI | ESI
5066
5067 hexdigit a | b | c | d | e | f
5068 | A | B | C | D | E | F
5069
5070 id alpha
5071 | id alpha
5072 | id decdigit
5073
5074 mulOp * | / | MOD
5075
5076 quote " | '
5077
5078 register specialRegister
5079 | gpRegister
5080 | byteRegister
5081
5082 segmentRegister CS | DS | ES | FS | GS | SS
5083
5084 specialRegister CR0 | CR2 | CR3
5085 | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
5086 | TR3 | TR4 | TR5 | TR6 | TR7
5087
5088 We simplify the grammar in obvious places (e.g., register parsing is
5089 done by calling parse_register) and eliminate immediate left recursion
5090 to implement a recursive-descent parser.
5091
5092 expr SHORT e05
5093 | e05
5094
5095 e05 e06 e05'
5096
5097 e05' addOp e06 e05'
5098 | Empty
5099
5100 e06 e09 e06'
5101
5102 e06' mulOp e09 e06'
5103 | Empty
5104
5105 e09 OFFSET e10 e09'
5106 | e10 e09'
5107
5108 e09' PTR e10 e09'
5109 | : e10 e09'
5110 | Empty
5111
5112 e10 e11 e10'
5113
5114 e10' [ expr ] e10'
5115 | Empty
5116
5117 e11 ( expr )
5118 | [ expr ]
5119 | BYTE
5120 | WORD
5121 | DWORD
5122 | QWORD
5123 | XWORD
5124 | .
5125 | $
5126 | register
5127 | id
5128 | constant */
5129
5130 /* Parsing structure for the intel syntax parser. Used to implement the
5131 semantic actions for the operand grammar. */
5132 struct intel_parser_s
5133 {
5134 char *op_string; /* The string being parsed. */
5135 int got_a_float; /* Whether the operand is a float. */
5136 int op_modifier; /* Operand modifier. */
5137 int is_mem; /* 1 if operand is memory reference. */
5138 const reg_entry *reg; /* Last register reference found. */
5139 char *disp; /* Displacement string being built. */
5140 };
5141
5142 static struct intel_parser_s intel_parser;
5143
5144 /* Token structure for parsing intel syntax. */
5145 struct intel_token
5146 {
5147 int code; /* Token code. */
5148 const reg_entry *reg; /* Register entry for register tokens. */
5149 char *str; /* String representation. */
5150 };
5151
5152 static struct intel_token cur_token, prev_token;
5153
5154 /* Token codes for the intel parser. Since T_SHORT is already used
5155 by COFF, undefine it first to prevent a warning. */
5156 #define T_NIL -1
5157 #define T_CONST 1
5158 #define T_REG 2
5159 #define T_BYTE 3
5160 #define T_WORD 4
5161 #define T_DWORD 5
5162 #define T_QWORD 6
5163 #define T_XWORD 7
5164 #undef T_SHORT
5165 #define T_SHORT 8
5166 #define T_OFFSET 9
5167 #define T_PTR 10
5168 #define T_ID 11
5169
5170 /* Prototypes for intel parser functions. */
5171 static int intel_match_token PARAMS ((int code));
5172 static void intel_get_token PARAMS ((void));
5173 static void intel_putback_token PARAMS ((void));
5174 static int intel_expr PARAMS ((void));
5175 static int intel_e05 PARAMS ((void));
5176 static int intel_e05_1 PARAMS ((void));
5177 static int intel_e06 PARAMS ((void));
5178 static int intel_e06_1 PARAMS ((void));
5179 static int intel_e09 PARAMS ((void));
5180 static int intel_e09_1 PARAMS ((void));
5181 static int intel_e10 PARAMS ((void));
5182 static int intel_e10_1 PARAMS ((void));
5183 static int intel_e11 PARAMS ((void));
5184
5185 static int
5186 i386_intel_operand (operand_string, got_a_float)
5187 char *operand_string;
5188 int got_a_float;
5189 {
5190 int ret;
5191 char *p;
5192
5193 /* Initialize token holders. */
5194 cur_token.code = prev_token.code = T_NIL;
5195 cur_token.reg = prev_token.reg = NULL;
5196 cur_token.str = prev_token.str = NULL;
5197
5198 /* Initialize parser structure. */
5199 p = intel_parser.op_string = (char *) malloc (strlen (operand_string) + 1);
5200 if (p == NULL)
5201 abort ();
5202 strcpy (intel_parser.op_string, operand_string);
5203 intel_parser.got_a_float = got_a_float;
5204 intel_parser.op_modifier = -1;
5205 intel_parser.is_mem = 0;
5206 intel_parser.reg = NULL;
5207 intel_parser.disp = (char *) malloc (strlen (operand_string) + 1);
5208 if (intel_parser.disp == NULL)
5209 abort ();
5210 intel_parser.disp[0] = '\0';
5211
5212 /* Read the first token and start the parser. */
5213 intel_get_token ();
5214 ret = intel_expr ();
5215
5216 if (ret)
5217 {
5218 /* If we found a memory reference, hand it over to i386_displacement
5219 to fill in the rest of the operand fields. */
5220 if (intel_parser.is_mem)
5221 {
5222 if ((i.mem_operands == 1
5223 && (current_templates->start->opcode_modifier & IsString) == 0)
5224 || i.mem_operands == 2)
5225 {
5226 as_bad (_("too many memory references for '%s'"),
5227 current_templates->start->name);
5228 ret = 0;
5229 }
5230 else
5231 {
5232 char *s = intel_parser.disp;
5233 i.mem_operands++;
5234
5235 /* Add the displacement expression. */
5236 if (*s != '\0')
5237 ret = i386_displacement (s, s + strlen (s))
5238 && i386_index_check (s);
5239 }
5240 }
5241
5242 /* Constant and OFFSET expressions are handled by i386_immediate. */
5243 else if (intel_parser.op_modifier == OFFSET_FLAT
5244 || intel_parser.reg == NULL)
5245 ret = i386_immediate (intel_parser.disp);
5246 }
5247
5248 free (p);
5249 free (intel_parser.disp);
5250
5251 return ret;
5252 }
5253
5254 /* expr SHORT e05
5255 | e05 */
5256 static int
5257 intel_expr ()
5258 {
5259 /* expr SHORT e05 */
5260 if (cur_token.code == T_SHORT)
5261 {
5262 intel_parser.op_modifier = SHORT;
5263 intel_match_token (T_SHORT);
5264
5265 return (intel_e05 ());
5266 }
5267
5268 /* expr e05 */
5269 else
5270 return intel_e05 ();
5271 }
5272
5273 /* e05 e06 e05'
5274
5275 e05' addOp e06 e05'
5276 | Empty */
5277 static int
5278 intel_e05 ()
5279 {
5280 return (intel_e06 () && intel_e05_1 ());
5281 }
5282
5283 static int
5284 intel_e05_1 ()
5285 {
5286 /* e05' addOp e06 e05' */
5287 if (cur_token.code == '+' || cur_token.code == '-')
5288 {
5289 strcat (intel_parser.disp, cur_token.str);
5290 intel_match_token (cur_token.code);
5291
5292 return (intel_e06 () && intel_e05_1 ());
5293 }
5294
5295 /* e05' Empty */
5296 else
5297 return 1;
5298 }
5299
5300 /* e06 e09 e06'
5301
5302 e06' mulOp e09 e06'
5303 | Empty */
5304 static int
5305 intel_e06 ()
5306 {
5307 return (intel_e09 () && intel_e06_1 ());
5308 }
5309
5310 static int
5311 intel_e06_1 ()
5312 {
5313 /* e06' mulOp e09 e06' */
5314 if (cur_token.code == '*' || cur_token.code == '/')
5315 {
5316 strcat (intel_parser.disp, cur_token.str);
5317 intel_match_token (cur_token.code);
5318
5319 return (intel_e09 () && intel_e06_1 ());
5320 }
5321
5322 /* e06' Empty */
5323 else
5324 return 1;
5325 }
5326
5327 /* e09 OFFSET e10 e09'
5328 | e10 e09'
5329
5330 e09' PTR e10 e09'
5331 | : e10 e09'
5332 | Empty */
5333 static int
5334 intel_e09 ()
5335 {
5336 /* e09 OFFSET e10 e09' */
5337 if (cur_token.code == T_OFFSET)
5338 {
5339 intel_parser.is_mem = 0;
5340 intel_parser.op_modifier = OFFSET_FLAT;
5341 intel_match_token (T_OFFSET);
5342
5343 return (intel_e10 () && intel_e09_1 ());
5344 }
5345
5346 /* e09 e10 e09' */
5347 else
5348 return (intel_e10 () && intel_e09_1 ());
5349 }
5350
5351 static int
5352 intel_e09_1 ()
5353 {
5354 /* e09' PTR e10 e09' */
5355 if (cur_token.code == T_PTR)
5356 {
5357 if (prev_token.code == T_BYTE)
5358 i.suffix = BYTE_MNEM_SUFFIX;
5359
5360 else if (prev_token.code == T_WORD)
5361 {
5362 if (intel_parser.got_a_float == 2) /* "fi..." */
5363 i.suffix = SHORT_MNEM_SUFFIX;
5364 else
5365 i.suffix = WORD_MNEM_SUFFIX;
5366 }
5367
5368 else if (prev_token.code == T_DWORD)
5369 {
5370 if (intel_parser.got_a_float == 1) /* "f..." */
5371 i.suffix = SHORT_MNEM_SUFFIX;
5372 else
5373 i.suffix = LONG_MNEM_SUFFIX;
5374 }
5375
5376 else if (prev_token.code == T_QWORD)
5377 {
5378 if (intel_parser.got_a_float == 1) /* "f..." */
5379 i.suffix = LONG_MNEM_SUFFIX;
5380 else
5381 i.suffix = QWORD_MNEM_SUFFIX;
5382 }
5383
5384 else if (prev_token.code == T_XWORD)
5385 i.suffix = LONG_DOUBLE_MNEM_SUFFIX;
5386
5387 else
5388 {
5389 as_bad (_("Unknown operand modifier `%s'\n"), prev_token.str);
5390 return 0;
5391 }
5392
5393 intel_match_token (T_PTR);
5394
5395 return (intel_e10 () && intel_e09_1 ());
5396 }
5397
5398 /* e09 : e10 e09' */
5399 else if (cur_token.code == ':')
5400 {
5401 /* Mark as a memory operand only if it's not already known to be an
5402 offset expression. */
5403 if (intel_parser.op_modifier != OFFSET_FLAT)
5404 intel_parser.is_mem = 1;
5405
5406 return (intel_match_token (':') && intel_e10 () && intel_e09_1 ());
5407 }
5408
5409 /* e09' Empty */
5410 else
5411 return 1;
5412 }
5413
5414 /* e10 e11 e10'
5415
5416 e10' [ expr ] e10'
5417 | Empty */
5418 static int
5419 intel_e10 ()
5420 {
5421 return (intel_e11 () && intel_e10_1 ());
5422 }
5423
5424 static int
5425 intel_e10_1 ()
5426 {
5427 /* e10' [ expr ] e10' */
5428 if (cur_token.code == '[')
5429 {
5430 intel_match_token ('[');
5431
5432 /* Mark as a memory operand only if it's not already known to be an
5433 offset expression. If it's an offset expression, we need to keep
5434 the brace in. */
5435 if (intel_parser.op_modifier != OFFSET_FLAT)
5436 intel_parser.is_mem = 1;
5437 else
5438 strcat (intel_parser.disp, "[");
5439
5440 /* Add a '+' to the displacement string if necessary. */
5441 if (*intel_parser.disp != '\0'
5442 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5443 strcat (intel_parser.disp, "+");
5444
5445 if (intel_expr () && intel_match_token (']'))
5446 {
5447 /* Preserve brackets when the operand is an offset expression. */
5448 if (intel_parser.op_modifier == OFFSET_FLAT)
5449 strcat (intel_parser.disp, "]");
5450
5451 return intel_e10_1 ();
5452 }
5453 else
5454 return 0;
5455 }
5456
5457 /* e10' Empty */
5458 else
5459 return 1;
5460 }
5461
5462 /* e11 ( expr )
5463 | [ expr ]
5464 | BYTE
5465 | WORD
5466 | DWORD
5467 | QWORD
5468 | XWORD
5469 | $
5470 | .
5471 | register
5472 | id
5473 | constant */
5474 static int
5475 intel_e11 ()
5476 {
5477 /* e11 ( expr ) */
5478 if (cur_token.code == '(')
5479 {
5480 intel_match_token ('(');
5481 strcat (intel_parser.disp, "(");
5482
5483 if (intel_expr () && intel_match_token (')'))
5484 {
5485 strcat (intel_parser.disp, ")");
5486 return 1;
5487 }
5488 else
5489 return 0;
5490 }
5491
5492 /* e11 [ expr ] */
5493 else if (cur_token.code == '[')
5494 {
5495 intel_match_token ('[');
5496
5497 /* Mark as a memory operand only if it's not already known to be an
5498 offset expression. If it's an offset expression, we need to keep
5499 the brace in. */
5500 if (intel_parser.op_modifier != OFFSET_FLAT)
5501 intel_parser.is_mem = 1;
5502 else
5503 strcat (intel_parser.disp, "[");
5504
5505 /* Operands for jump/call inside brackets denote absolute addresses. */
5506 if (current_templates->start->opcode_modifier & Jump
5507 || current_templates->start->opcode_modifier & JumpDword
5508 || current_templates->start->opcode_modifier & JumpByte
5509 || current_templates->start->opcode_modifier & JumpInterSegment)
5510 i.types[this_operand] |= JumpAbsolute;
5511
5512 /* Add a '+' to the displacement string if necessary. */
5513 if (*intel_parser.disp != '\0'
5514 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5515 strcat (intel_parser.disp, "+");
5516
5517 if (intel_expr () && intel_match_token (']'))
5518 {
5519 /* Preserve brackets when the operand is an offset expression. */
5520 if (intel_parser.op_modifier == OFFSET_FLAT)
5521 strcat (intel_parser.disp, "]");
5522
5523 return 1;
5524 }
5525 else
5526 return 0;
5527 }
5528
5529 /* e11 BYTE
5530 | WORD
5531 | DWORD
5532 | QWORD
5533 | XWORD */
5534 else if (cur_token.code == T_BYTE
5535 || cur_token.code == T_WORD
5536 || cur_token.code == T_DWORD
5537 || cur_token.code == T_QWORD
5538 || cur_token.code == T_XWORD)
5539 {
5540 intel_match_token (cur_token.code);
5541
5542 return 1;
5543 }
5544
5545 /* e11 $
5546 | . */
5547 else if (cur_token.code == '$' || cur_token.code == '.')
5548 {
5549 strcat (intel_parser.disp, cur_token.str);
5550 intel_match_token (cur_token.code);
5551
5552 /* Mark as a memory operand only if it's not already known to be an
5553 offset expression. */
5554 if (intel_parser.op_modifier != OFFSET_FLAT)
5555 intel_parser.is_mem = 1;
5556
5557 return 1;
5558 }
5559
5560 /* e11 register */
5561 else if (cur_token.code == T_REG)
5562 {
5563 const reg_entry *reg = intel_parser.reg = cur_token.reg;
5564
5565 intel_match_token (T_REG);
5566
5567 /* Check for segment change. */
5568 if (cur_token.code == ':')
5569 {
5570 if (reg->reg_type & (SReg2 | SReg3))
5571 {
5572 switch (reg->reg_num)
5573 {
5574 case 0:
5575 i.seg[i.mem_operands] = &es;
5576 break;
5577 case 1:
5578 i.seg[i.mem_operands] = &cs;
5579 break;
5580 case 2:
5581 i.seg[i.mem_operands] = &ss;
5582 break;
5583 case 3:
5584 i.seg[i.mem_operands] = &ds;
5585 break;
5586 case 4:
5587 i.seg[i.mem_operands] = &fs;
5588 break;
5589 case 5:
5590 i.seg[i.mem_operands] = &gs;
5591 break;
5592 }
5593 }
5594 else
5595 {
5596 as_bad (_("`%s' is not a valid segment register"), reg->reg_name);
5597 return 0;
5598 }
5599 }
5600
5601 /* Not a segment register. Check for register scaling. */
5602 else if (cur_token.code == '*')
5603 {
5604 if (!intel_parser.is_mem)
5605 {
5606 as_bad (_("Register scaling only allowed in memory operands."));
5607 return 0;
5608 }
5609
5610 /* What follows must be a valid scale. */
5611 if (intel_match_token ('*')
5612 && strchr ("01248", *cur_token.str))
5613 {
5614 i.index_reg = reg;
5615 i.types[this_operand] |= BaseIndex;
5616
5617 /* Set the scale after setting the register (otherwise,
5618 i386_scale will complain) */
5619 i386_scale (cur_token.str);
5620 intel_match_token (T_CONST);
5621 }
5622 else
5623 {
5624 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
5625 cur_token.str);
5626 return 0;
5627 }
5628 }
5629
5630 /* No scaling. If this is a memory operand, the register is either a
5631 base register (first occurrence) or an index register (second
5632 occurrence). */
5633 else if (intel_parser.is_mem && !(reg->reg_type & (SReg2 | SReg3)))
5634 {
5635 if (i.base_reg && i.index_reg)
5636 {
5637 as_bad (_("Too many register references in memory operand.\n"));
5638 return 0;
5639 }
5640
5641 if (i.base_reg == NULL)
5642 i.base_reg = reg;
5643 else
5644 i.index_reg = reg;
5645
5646 i.types[this_operand] |= BaseIndex;
5647 }
5648
5649 /* Offset modifier. Add the register to the displacement string to be
5650 parsed as an immediate expression after we're done. */
5651 else if (intel_parser.op_modifier == OFFSET_FLAT)
5652 strcat (intel_parser.disp, reg->reg_name);
5653
5654 /* It's neither base nor index nor offset. */
5655 else
5656 {
5657 i.types[this_operand] |= reg->reg_type & ~BaseIndex;
5658 i.op[this_operand].regs = reg;
5659 i.reg_operands++;
5660 }
5661
5662 /* Since registers are not part of the displacement string (except
5663 when we're parsing offset operands), we may need to remove any
5664 preceding '+' from the displacement string. */
5665 if (*intel_parser.disp != '\0'
5666 && intel_parser.op_modifier != OFFSET_FLAT)
5667 {
5668 char *s = intel_parser.disp;
5669 s += strlen (s) - 1;
5670 if (*s == '+')
5671 *s = '\0';
5672 }
5673
5674 return 1;
5675 }
5676
5677 /* e11 id */
5678 else if (cur_token.code == T_ID)
5679 {
5680 /* Add the identifier to the displacement string. */
5681 strcat (intel_parser.disp, cur_token.str);
5682 intel_match_token (T_ID);
5683
5684 /* The identifier represents a memory reference only if it's not
5685 preceded by an offset modifier. */
5686 if (intel_parser.op_modifier != OFFSET_FLAT)
5687 intel_parser.is_mem = 1;
5688
5689 return 1;
5690 }
5691
5692 /* e11 constant */
5693 else if (cur_token.code == T_CONST
5694 || cur_token.code == '-'
5695 || cur_token.code == '+')
5696 {
5697 char *save_str;
5698
5699 /* Allow constants that start with `+' or `-'. */
5700 if (cur_token.code == '-' || cur_token.code == '+')
5701 {
5702 strcat (intel_parser.disp, cur_token.str);
5703 intel_match_token (cur_token.code);
5704 if (cur_token.code != T_CONST)
5705 {
5706 as_bad (_("Syntax error. Expecting a constant. Got `%s'.\n"),
5707 cur_token.str);
5708 return 0;
5709 }
5710 }
5711
5712 save_str = (char *) malloc (strlen (cur_token.str) + 1);
5713 if (save_str == NULL)
5714 abort ();
5715 strcpy (save_str, cur_token.str);
5716
5717 /* Get the next token to check for register scaling. */
5718 intel_match_token (cur_token.code);
5719
5720 /* Check if this constant is a scaling factor for an index register. */
5721 if (cur_token.code == '*')
5722 {
5723 if (intel_match_token ('*') && cur_token.code == T_REG)
5724 {
5725 if (!intel_parser.is_mem)
5726 {
5727 as_bad (_("Register scaling only allowed in memory operands."));
5728 return 0;
5729 }
5730
5731 /* The constant is followed by `* reg', so it must be
5732 a valid scale. */
5733 if (strchr ("01248", *save_str))
5734 {
5735 i.index_reg = cur_token.reg;
5736 i.types[this_operand] |= BaseIndex;
5737
5738 /* Set the scale after setting the register (otherwise,
5739 i386_scale will complain) */
5740 i386_scale (save_str);
5741 intel_match_token (T_REG);
5742
5743 /* Since registers are not part of the displacement
5744 string, we may need to remove any preceding '+' from
5745 the displacement string. */
5746 if (*intel_parser.disp != '\0')
5747 {
5748 char *s = intel_parser.disp;
5749 s += strlen (s) - 1;
5750 if (*s == '+')
5751 *s = '\0';
5752 }
5753
5754 free (save_str);
5755
5756 return 1;
5757 }
5758 else
5759 return 0;
5760 }
5761
5762 /* The constant was not used for register scaling. Since we have
5763 already consumed the token following `*' we now need to put it
5764 back in the stream. */
5765 else
5766 intel_putback_token ();
5767 }
5768
5769 /* Add the constant to the displacement string. */
5770 strcat (intel_parser.disp, save_str);
5771 free (save_str);
5772
5773 return 1;
5774 }
5775
5776 as_bad (_("Unrecognized token '%s'"), cur_token.str);
5777 return 0;
5778 }
5779
5780 /* Match the given token against cur_token. If they match, read the next
5781 token from the operand string. */
5782 static int
5783 intel_match_token (code)
5784 int code;
5785 {
5786 if (cur_token.code == code)
5787 {
5788 intel_get_token ();
5789 return 1;
5790 }
5791 else
5792 {
5793 as_bad (_("Unexpected token `%s'\n"), cur_token.str);
5794 return 0;
5795 }
5796 }
5797
5798 /* Read a new token from intel_parser.op_string and store it in cur_token. */
5799 static void
5800 intel_get_token ()
5801 {
5802 char *end_op;
5803 const reg_entry *reg;
5804 struct intel_token new_token;
5805
5806 new_token.code = T_NIL;
5807 new_token.reg = NULL;
5808 new_token.str = NULL;
5809
5810 /* Free the memory allocated to the previous token and move
5811 cur_token to prev_token. */
5812 if (prev_token.str)
5813 free (prev_token.str);
5814
5815 prev_token = cur_token;
5816
5817 /* Skip whitespace. */
5818 while (is_space_char (*intel_parser.op_string))
5819 intel_parser.op_string++;
5820
5821 /* Return an empty token if we find nothing else on the line. */
5822 if (*intel_parser.op_string == '\0')
5823 {
5824 cur_token = new_token;
5825 return;
5826 }
5827
5828 /* The new token cannot be larger than the remainder of the operand
5829 string. */
5830 new_token.str = (char *) malloc (strlen (intel_parser.op_string) + 1);
5831 if (new_token.str == NULL)
5832 abort ();
5833 new_token.str[0] = '\0';
5834
5835 if (strchr ("0123456789", *intel_parser.op_string))
5836 {
5837 char *p = new_token.str;
5838 char *q = intel_parser.op_string;
5839 new_token.code = T_CONST;
5840
5841 /* Allow any kind of identifier char to encompass floating point and
5842 hexadecimal numbers. */
5843 while (is_identifier_char (*q))
5844 *p++ = *q++;
5845 *p = '\0';
5846
5847 /* Recognize special symbol names [0-9][bf]. */
5848 if (strlen (intel_parser.op_string) == 2
5849 && (intel_parser.op_string[1] == 'b'
5850 || intel_parser.op_string[1] == 'f'))
5851 new_token.code = T_ID;
5852 }
5853
5854 else if (strchr ("+-/*:[]()", *intel_parser.op_string))
5855 {
5856 new_token.code = *intel_parser.op_string;
5857 new_token.str[0] = *intel_parser.op_string;
5858 new_token.str[1] = '\0';
5859 }
5860
5861 else if ((*intel_parser.op_string == REGISTER_PREFIX || allow_naked_reg)
5862 && ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL))
5863 {
5864 new_token.code = T_REG;
5865 new_token.reg = reg;
5866
5867 if (*intel_parser.op_string == REGISTER_PREFIX)
5868 {
5869 new_token.str[0] = REGISTER_PREFIX;
5870 new_token.str[1] = '\0';
5871 }
5872
5873 strcat (new_token.str, reg->reg_name);
5874 }
5875
5876 else if (is_identifier_char (*intel_parser.op_string))
5877 {
5878 char *p = new_token.str;
5879 char *q = intel_parser.op_string;
5880
5881 /* A '.' or '$' followed by an identifier char is an identifier.
5882 Otherwise, it's operator '.' followed by an expression. */
5883 if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
5884 {
5885 new_token.code = *q;
5886 new_token.str[0] = *q;
5887 new_token.str[1] = '\0';
5888 }
5889 else
5890 {
5891 while (is_identifier_char (*q) || *q == '@')
5892 *p++ = *q++;
5893 *p = '\0';
5894
5895 if (strcasecmp (new_token.str, "BYTE") == 0)
5896 new_token.code = T_BYTE;
5897
5898 else if (strcasecmp (new_token.str, "WORD") == 0)
5899 new_token.code = T_WORD;
5900
5901 else if (strcasecmp (new_token.str, "DWORD") == 0)
5902 new_token.code = T_DWORD;
5903
5904 else if (strcasecmp (new_token.str, "QWORD") == 0)
5905 new_token.code = T_QWORD;
5906
5907 else if (strcasecmp (new_token.str, "XWORD") == 0)
5908 new_token.code = T_XWORD;
5909
5910 else if (strcasecmp (new_token.str, "PTR") == 0)
5911 new_token.code = T_PTR;
5912
5913 else if (strcasecmp (new_token.str, "SHORT") == 0)
5914 new_token.code = T_SHORT;
5915
5916 else if (strcasecmp (new_token.str, "OFFSET") == 0)
5917 {
5918 new_token.code = T_OFFSET;
5919
5920 /* ??? This is not mentioned in the MASM grammar but gcc
5921 makes use of it with -mintel-syntax. OFFSET may be
5922 followed by FLAT: */
5923 if (strncasecmp (q, " FLAT:", 6) == 0)
5924 strcat (new_token.str, " FLAT:");
5925 }
5926
5927 /* ??? This is not mentioned in the MASM grammar. */
5928 else if (strcasecmp (new_token.str, "FLAT") == 0)
5929 new_token.code = T_OFFSET;
5930
5931 else
5932 new_token.code = T_ID;
5933 }
5934 }
5935
5936 else
5937 as_bad (_("Unrecognized token `%s'\n"), intel_parser.op_string);
5938
5939 intel_parser.op_string += strlen (new_token.str);
5940 cur_token = new_token;
5941 }
5942
5943 /* Put cur_token back into the token stream and make cur_token point to
5944 prev_token. */
5945 static void
5946 intel_putback_token ()
5947 {
5948 intel_parser.op_string -= strlen (cur_token.str);
5949 free (cur_token.str);
5950 cur_token = prev_token;
5951
5952 /* Forget prev_token. */
5953 prev_token.code = T_NIL;
5954 prev_token.reg = NULL;
5955 prev_token.str = NULL;
5956 }
This page took 0.21643 seconds and 4 git commands to generate.