* config/tc-i386.c (i386_displacement): Call as_bad for bad GOTOFF
[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
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 <ctype.h>
30
31 #include "as.h"
32 #include "subsegs.h"
33 #include "dwarf2dbg.h"
34 #include "opcode/i386.h"
35
36 #ifndef REGISTER_WARNINGS
37 #define REGISTER_WARNINGS 1
38 #endif
39
40 #ifndef INFER_ADDR_PREFIX
41 #define INFER_ADDR_PREFIX 1
42 #endif
43
44 #ifndef SCALE1_WHEN_NO_INDEX
45 /* Specifying a scale factor besides 1 when there is no index is
46 futile. eg. `mov (%ebx,2),%al' does exactly the same as
47 `mov (%ebx),%al'. To slavishly follow what the programmer
48 specified, set SCALE1_WHEN_NO_INDEX to 0. */
49 #define SCALE1_WHEN_NO_INDEX 1
50 #endif
51
52 #define true 1
53 #define false 0
54
55 static unsigned int mode_from_disp_size PARAMS ((unsigned int));
56 static int fits_in_signed_byte PARAMS ((offsetT));
57 static int fits_in_unsigned_byte PARAMS ((offsetT));
58 static int fits_in_unsigned_word PARAMS ((offsetT));
59 static int fits_in_signed_word PARAMS ((offsetT));
60 static int fits_in_unsigned_long PARAMS ((offsetT));
61 static int fits_in_signed_long PARAMS ((offsetT));
62 static int smallest_imm_type PARAMS ((offsetT));
63 static offsetT offset_in_range PARAMS ((offsetT, int));
64 static int add_prefix PARAMS ((unsigned int));
65 static void set_code_flag PARAMS ((int));
66 static void set_16bit_gcc_code_flag PARAMS ((int));
67 static void set_intel_syntax PARAMS ((int));
68 static void set_cpu_arch PARAMS ((int));
69
70 #ifdef BFD_ASSEMBLER
71 static bfd_reloc_code_real_type reloc
72 PARAMS ((int, int, int, bfd_reloc_code_real_type));
73 #define RELOC_ENUM enum bfd_reloc_code_real
74 #else
75 #define RELOC_ENUM int
76 #endif
77
78 #ifndef DEFAULT_ARCH
79 #define DEFAULT_ARCH "i386"
80 #endif
81 static char *default_arch = DEFAULT_ARCH;
82
83 /* 'md_assemble ()' gathers together information and puts it into a
84 i386_insn. */
85
86 union i386_op
87 {
88 expressionS *disps;
89 expressionS *imms;
90 const reg_entry *regs;
91 };
92
93 struct _i386_insn
94 {
95 /* TM holds the template for the insn were currently assembling. */
96 template tm;
97
98 /* SUFFIX holds the instruction mnemonic suffix if given.
99 (e.g. 'l' for 'movl') */
100 char suffix;
101
102 /* OPERANDS gives the number of given operands. */
103 unsigned int operands;
104
105 /* REG_OPERANDS, DISP_OPERANDS, MEM_OPERANDS, IMM_OPERANDS give the number
106 of given register, displacement, memory operands and immediate
107 operands. */
108 unsigned int reg_operands, disp_operands, mem_operands, imm_operands;
109
110 /* TYPES [i] is the type (see above #defines) which tells us how to
111 use OP[i] for the corresponding operand. */
112 unsigned int types[MAX_OPERANDS];
113
114 /* Displacement expression, immediate expression, or register for each
115 operand. */
116 union i386_op op[MAX_OPERANDS];
117
118 /* Flags for operands. */
119 unsigned int flags[MAX_OPERANDS];
120 #define Operand_PCrel 1
121
122 /* Relocation type for operand */
123 RELOC_ENUM reloc[MAX_OPERANDS];
124
125 /* BASE_REG, INDEX_REG, and LOG2_SCALE_FACTOR are used to encode
126 the base index byte below. */
127 const reg_entry *base_reg;
128 const reg_entry *index_reg;
129 unsigned int log2_scale_factor;
130
131 /* SEG gives the seg_entries of this insn. They are zero unless
132 explicit segment overrides are given. */
133 const seg_entry *seg[2];
134
135 /* PREFIX holds all the given prefix opcodes (usually null).
136 PREFIXES is the number of prefix opcodes. */
137 unsigned int prefixes;
138 unsigned char prefix[MAX_PREFIXES];
139
140 /* RM and SIB are the modrm byte and the sib byte where the
141 addressing modes of this insn are encoded. */
142
143 modrm_byte rm;
144 rex_byte rex;
145 sib_byte sib;
146 };
147
148 typedef struct _i386_insn i386_insn;
149
150 /* List of chars besides those in app.c:symbol_chars that can start an
151 operand. Used to prevent the scrubber eating vital white-space. */
152 #ifdef LEX_AT
153 const char extra_symbol_chars[] = "*%-(@";
154 #else
155 const char extra_symbol_chars[] = "*%-(";
156 #endif
157
158 /* This array holds the chars that always start a comment. If the
159 pre-processor is disabled, these aren't very useful. */
160 #if defined (TE_I386AIX) || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) && ! defined (TE_LINUX) && !defined(TE_FreeBSD))
161 /* Putting '/' here makes it impossible to use the divide operator.
162 However, we need it for compatibility with SVR4 systems. */
163 const char comment_chars[] = "#/";
164 #define PREFIX_SEPARATOR '\\'
165 #else
166 const char comment_chars[] = "#";
167 #define PREFIX_SEPARATOR '/'
168 #endif
169
170 /* This array holds the chars that only start a comment at the beginning of
171 a line. If the line seems to have the form '# 123 filename'
172 .line and .file directives will appear in the pre-processed output.
173 Note that input_file.c hand checks for '#' at the beginning of the
174 first line of the input file. This is because the compiler outputs
175 #NO_APP at the beginning of its output.
176 Also note that comments started like this one will always work if
177 '/' isn't otherwise defined. */
178 #if defined (TE_I386AIX) || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) && ! defined (TE_LINUX) && !defined(TE_FreeBSD))
179 const char line_comment_chars[] = "";
180 #else
181 const char line_comment_chars[] = "/";
182 #endif
183
184 const char line_separator_chars[] = ";";
185
186 /* Chars that can be used to separate mant from exp in floating point
187 nums. */
188 const char EXP_CHARS[] = "eE";
189
190 /* Chars that mean this number is a floating point constant
191 As in 0f12.456
192 or 0d1.2345e12. */
193 const char FLT_CHARS[] = "fFdDxX";
194
195 /* Tables for lexical analysis. */
196 static char mnemonic_chars[256];
197 static char register_chars[256];
198 static char operand_chars[256];
199 static char identifier_chars[256];
200 static char digit_chars[256];
201
202 /* Lexical macros. */
203 #define is_mnemonic_char(x) (mnemonic_chars[(unsigned char) x])
204 #define is_operand_char(x) (operand_chars[(unsigned char) x])
205 #define is_register_char(x) (register_chars[(unsigned char) x])
206 #define is_space_char(x) ((x) == ' ')
207 #define is_identifier_char(x) (identifier_chars[(unsigned char) x])
208 #define is_digit_char(x) (digit_chars[(unsigned char) x])
209
210 /* All non-digit non-letter charcters that may occur in an operand. */
211 static char operand_special_chars[] = "%$-+(,)*._~/<>|&^!:[@]";
212
213 /* md_assemble() always leaves the strings it's passed unaltered. To
214 effect this we maintain a stack of saved characters that we've smashed
215 with '\0's (indicating end of strings for various sub-fields of the
216 assembler instruction). */
217 static char save_stack[32];
218 static char *save_stack_p;
219 #define END_STRING_AND_SAVE(s) \
220 do { *save_stack_p++ = *(s); *(s) = '\0'; } while (0)
221 #define RESTORE_END_STRING(s) \
222 do { *(s) = *--save_stack_p; } while (0)
223
224 /* The instruction we're assembling. */
225 static i386_insn i;
226
227 /* Possible templates for current insn. */
228 static const templates *current_templates;
229
230 /* Per instruction expressionS buffers: 2 displacements & 2 immediate max. */
231 static expressionS disp_expressions[2], im_expressions[2];
232
233 /* Current operand we are working on. */
234 static int this_operand;
235
236 /* We support four different modes. FLAG_CODE variable is used to distinguish
237 these. */
238
239 enum flag_code {
240 CODE_32BIT,
241 CODE_16BIT,
242 CODE_64BIT };
243 #define NUM_FLAG_CODE ((int) CODE_64BIT + 1)
244
245 static enum flag_code flag_code;
246 static int use_rela_relocations = 0;
247
248 /* The names used to print error messages. */
249 static const char *flag_code_names[] =
250 {
251 "32",
252 "16",
253 "64"
254 };
255
256 /* 1 for intel syntax,
257 0 if att syntax. */
258 static int intel_syntax = 0;
259
260 /* 1 if register prefix % not required. */
261 static int allow_naked_reg = 0;
262
263 /* Used in 16 bit gcc mode to add an l suffix to call, ret, enter,
264 leave, push, and pop instructions so that gcc has the same stack
265 frame as in 32 bit mode. */
266 static char stackop_size = '\0';
267
268 /* Non-zero to quieten some warnings. */
269 static int quiet_warnings = 0;
270
271 /* CPU name. */
272 static const char *cpu_arch_name = NULL;
273
274 /* CPU feature flags. */
275 static unsigned int cpu_arch_flags = CpuUnknownFlags|CpuNo64;
276
277 /* If set, conditional jumps are not automatically promoted to handle
278 larger than a byte offset. */
279 static unsigned int no_cond_jump_promotion = 0;
280
281 /* Interface to relax_segment.
282 There are 3 major relax states for 386 jump insns because the
283 different types of jumps add different sizes to frags when we're
284 figuring out what sort of jump to choose to reach a given label. */
285
286 /* Types. */
287 #define UNCOND_JUMP 0
288 #define COND_JUMP 1
289 #define COND_JUMP86 2
290
291 /* Sizes. */
292 #define CODE16 1
293 #define SMALL 0
294 #define SMALL16 (SMALL|CODE16)
295 #define BIG 2
296 #define BIG16 (BIG|CODE16)
297
298 #ifndef INLINE
299 #ifdef __GNUC__
300 #define INLINE __inline__
301 #else
302 #define INLINE
303 #endif
304 #endif
305
306 #define ENCODE_RELAX_STATE(type, size) \
307 ((relax_substateT) (((type) << 2) | (size)))
308 #define TYPE_FROM_RELAX_STATE(s) \
309 ((s) >> 2)
310 #define DISP_SIZE_FROM_RELAX_STATE(s) \
311 ((((s) & 3) == BIG ? 4 : (((s) & 3) == BIG16 ? 2 : 1)))
312
313 /* This table is used by relax_frag to promote short jumps to long
314 ones where necessary. SMALL (short) jumps may be promoted to BIG
315 (32 bit long) ones, and SMALL16 jumps to BIG16 (16 bit long). We
316 don't allow a short jump in a 32 bit code segment to be promoted to
317 a 16 bit offset jump because it's slower (requires data size
318 prefix), and doesn't work, unless the destination is in the bottom
319 64k of the code segment (The top 16 bits of eip are zeroed). */
320
321 const relax_typeS md_relax_table[] =
322 {
323 /* The fields are:
324 1) most positive reach of this state,
325 2) most negative reach of this state,
326 3) how many bytes this mode will have in the variable part of the frag
327 4) which index into the table to try if we can't fit into this one. */
328
329 /* UNCOND_JUMP states. */
330 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG)},
331 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16)},
332 /* dword jmp adds 4 bytes to frag:
333 0 extra opcode bytes, 4 displacement bytes. */
334 {0, 0, 4, 0},
335 /* word jmp adds 2 byte2 to frag:
336 0 extra opcode bytes, 2 displacement bytes. */
337 {0, 0, 2, 0},
338
339 /* COND_JUMP states. */
340 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG)},
341 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG16)},
342 /* dword conditionals adds 5 bytes to frag:
343 1 extra opcode byte, 4 displacement bytes. */
344 {0, 0, 5, 0},
345 /* word conditionals add 3 bytes to frag:
346 1 extra opcode byte, 2 displacement bytes. */
347 {0, 0, 3, 0},
348
349 /* COND_JUMP86 states. */
350 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG)},
351 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG16)},
352 /* dword conditionals adds 5 bytes to frag:
353 1 extra opcode byte, 4 displacement bytes. */
354 {0, 0, 5, 0},
355 /* word conditionals add 4 bytes to frag:
356 1 displacement byte and a 3 byte long branch insn. */
357 {0, 0, 4, 0}
358 };
359
360 static const arch_entry cpu_arch[] = {
361 {"i8086", Cpu086 },
362 {"i186", Cpu086|Cpu186 },
363 {"i286", Cpu086|Cpu186|Cpu286 },
364 {"i386", Cpu086|Cpu186|Cpu286|Cpu386 },
365 {"i486", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486 },
366 {"i586", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
367 {"i686", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
368 {"pentium", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
369 {"pentiumpro",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
370 {"pentium4", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX|CpuSSE|CpuSSE2 },
371 {"k6", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX|Cpu3dnow },
372 {"athlon", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuMMX|Cpu3dnow },
373 {"sledgehammer",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuSledgehammer|CpuMMX|Cpu3dnow|CpuSSE|CpuSSE2 },
374 {NULL, 0 }
375 };
376
377 void
378 i386_align_code (fragP, count)
379 fragS *fragP;
380 int count;
381 {
382 /* Various efficient no-op patterns for aligning code labels.
383 Note: Don't try to assemble the instructions in the comments.
384 0L and 0w are not legal. */
385 static const char f32_1[] =
386 {0x90}; /* nop */
387 static const char f32_2[] =
388 {0x89,0xf6}; /* movl %esi,%esi */
389 static const char f32_3[] =
390 {0x8d,0x76,0x00}; /* leal 0(%esi),%esi */
391 static const char f32_4[] =
392 {0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
393 static const char f32_5[] =
394 {0x90, /* nop */
395 0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
396 static const char f32_6[] =
397 {0x8d,0xb6,0x00,0x00,0x00,0x00}; /* leal 0L(%esi),%esi */
398 static const char f32_7[] =
399 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
400 static const char f32_8[] =
401 {0x90, /* nop */
402 0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
403 static const char f32_9[] =
404 {0x89,0xf6, /* movl %esi,%esi */
405 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
406 static const char f32_10[] =
407 {0x8d,0x76,0x00, /* leal 0(%esi),%esi */
408 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
409 static const char f32_11[] =
410 {0x8d,0x74,0x26,0x00, /* leal 0(%esi,1),%esi */
411 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
412 static const char f32_12[] =
413 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
414 0x8d,0xbf,0x00,0x00,0x00,0x00}; /* leal 0L(%edi),%edi */
415 static const char f32_13[] =
416 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
417 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
418 static const char f32_14[] =
419 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00, /* leal 0L(%esi,1),%esi */
420 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
421 static const char f32_15[] =
422 {0xeb,0x0d,0x90,0x90,0x90,0x90,0x90, /* jmp .+15; lotsa nops */
423 0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90};
424 static const char f16_3[] =
425 {0x8d,0x74,0x00}; /* lea 0(%esi),%esi */
426 static const char f16_4[] =
427 {0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
428 static const char f16_5[] =
429 {0x90, /* nop */
430 0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
431 static const char f16_6[] =
432 {0x89,0xf6, /* mov %si,%si */
433 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
434 static const char f16_7[] =
435 {0x8d,0x74,0x00, /* lea 0(%si),%si */
436 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
437 static const char f16_8[] =
438 {0x8d,0xb4,0x00,0x00, /* lea 0w(%si),%si */
439 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
440 static const char *const f32_patt[] = {
441 f32_1, f32_2, f32_3, f32_4, f32_5, f32_6, f32_7, f32_8,
442 f32_9, f32_10, f32_11, f32_12, f32_13, f32_14, f32_15
443 };
444 static const char *const f16_patt[] = {
445 f32_1, f32_2, f16_3, f16_4, f16_5, f16_6, f16_7, f16_8,
446 f32_15, f32_15, f32_15, f32_15, f32_15, f32_15, f32_15
447 };
448
449 /* ??? We can't use these fillers for x86_64, since they often kills the
450 upper halves. Solve later. */
451 if (flag_code == CODE_64BIT)
452 count = 1;
453
454 if (count > 0 && count <= 15)
455 {
456 if (flag_code == CODE_16BIT)
457 {
458 memcpy (fragP->fr_literal + fragP->fr_fix,
459 f16_patt[count - 1], count);
460 if (count > 8)
461 /* Adjust jump offset. */
462 fragP->fr_literal[fragP->fr_fix + 1] = count - 2;
463 }
464 else
465 memcpy (fragP->fr_literal + fragP->fr_fix,
466 f32_patt[count - 1], count);
467 fragP->fr_var = count;
468 }
469 }
470
471 static char *output_invalid PARAMS ((int c));
472 static int i386_operand PARAMS ((char *operand_string));
473 static int i386_intel_operand PARAMS ((char *operand_string, int got_a_float));
474 static const reg_entry *parse_register PARAMS ((char *reg_string,
475 char **end_op));
476
477 #ifndef I386COFF
478 static void s_bss PARAMS ((int));
479 #endif
480
481 symbolS *GOT_symbol; /* Pre-defined "_GLOBAL_OFFSET_TABLE_". */
482
483 static INLINE unsigned int
484 mode_from_disp_size (t)
485 unsigned int t;
486 {
487 return (t & Disp8) ? 1 : (t & (Disp16 | Disp32 | Disp32S)) ? 2 : 0;
488 }
489
490 static INLINE int
491 fits_in_signed_byte (num)
492 offsetT num;
493 {
494 return (num >= -128) && (num <= 127);
495 }
496
497 static INLINE int
498 fits_in_unsigned_byte (num)
499 offsetT num;
500 {
501 return (num & 0xff) == num;
502 }
503
504 static INLINE int
505 fits_in_unsigned_word (num)
506 offsetT num;
507 {
508 return (num & 0xffff) == num;
509 }
510
511 static INLINE int
512 fits_in_signed_word (num)
513 offsetT num;
514 {
515 return (-32768 <= num) && (num <= 32767);
516 }
517 static INLINE int
518 fits_in_signed_long (num)
519 offsetT num ATTRIBUTE_UNUSED;
520 {
521 #ifndef BFD64
522 return 1;
523 #else
524 return (!(((offsetT) -1 << 31) & num)
525 || (((offsetT) -1 << 31) & num) == ((offsetT) -1 << 31));
526 #endif
527 } /* fits_in_signed_long() */
528 static INLINE int
529 fits_in_unsigned_long (num)
530 offsetT num ATTRIBUTE_UNUSED;
531 {
532 #ifndef BFD64
533 return 1;
534 #else
535 return (num & (((offsetT) 2 << 31) - 1)) == num;
536 #endif
537 } /* fits_in_unsigned_long() */
538
539 static int
540 smallest_imm_type (num)
541 offsetT num;
542 {
543 if (cpu_arch_flags != (Cpu086 | Cpu186 | Cpu286 | Cpu386 | Cpu486 | CpuNo64)
544 && !(cpu_arch_flags & (CpuUnknown)))
545 {
546 /* This code is disabled on the 486 because all the Imm1 forms
547 in the opcode table are slower on the i486. They're the
548 versions with the implicitly specified single-position
549 displacement, which has another syntax if you really want to
550 use that form. */
551 if (num == 1)
552 return Imm1 | Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64;
553 }
554 return (fits_in_signed_byte (num)
555 ? (Imm8S | Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
556 : fits_in_unsigned_byte (num)
557 ? (Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
558 : (fits_in_signed_word (num) || fits_in_unsigned_word (num))
559 ? (Imm16 | Imm32 | Imm32S | Imm64)
560 : fits_in_signed_long (num)
561 ? (Imm32 | Imm32S | Imm64)
562 : fits_in_unsigned_long (num)
563 ? (Imm32 | Imm64)
564 : Imm64);
565 }
566
567 static offsetT
568 offset_in_range (val, size)
569 offsetT val;
570 int size;
571 {
572 addressT mask;
573
574 switch (size)
575 {
576 case 1: mask = ((addressT) 1 << 8) - 1; break;
577 case 2: mask = ((addressT) 1 << 16) - 1; break;
578 case 4: mask = ((addressT) 2 << 31) - 1; break;
579 #ifdef BFD64
580 case 8: mask = ((addressT) 2 << 63) - 1; break;
581 #endif
582 default: abort ();
583 }
584
585 /* If BFD64, sign extend val. */
586 if (!use_rela_relocations)
587 if ((val & ~(((addressT) 2 << 31) - 1)) == 0)
588 val = (val ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
589
590 if ((val & ~mask) != 0 && (val & ~mask) != ~mask)
591 {
592 char buf1[40], buf2[40];
593
594 sprint_value (buf1, val);
595 sprint_value (buf2, val & mask);
596 as_warn (_("%s shortened to %s"), buf1, buf2);
597 }
598 return val & mask;
599 }
600
601 /* Returns 0 if attempting to add a prefix where one from the same
602 class already exists, 1 if non rep/repne added, 2 if rep/repne
603 added. */
604 static int
605 add_prefix (prefix)
606 unsigned int prefix;
607 {
608 int ret = 1;
609 int q;
610
611 if (prefix >= 0x40 && prefix < 0x50 && flag_code == CODE_64BIT)
612 q = REX_PREFIX;
613 else
614 switch (prefix)
615 {
616 default:
617 abort ();
618
619 case CS_PREFIX_OPCODE:
620 case DS_PREFIX_OPCODE:
621 case ES_PREFIX_OPCODE:
622 case FS_PREFIX_OPCODE:
623 case GS_PREFIX_OPCODE:
624 case SS_PREFIX_OPCODE:
625 q = SEG_PREFIX;
626 break;
627
628 case REPNE_PREFIX_OPCODE:
629 case REPE_PREFIX_OPCODE:
630 ret = 2;
631 /* fall thru */
632 case LOCK_PREFIX_OPCODE:
633 q = LOCKREP_PREFIX;
634 break;
635
636 case FWAIT_OPCODE:
637 q = WAIT_PREFIX;
638 break;
639
640 case ADDR_PREFIX_OPCODE:
641 q = ADDR_PREFIX;
642 break;
643
644 case DATA_PREFIX_OPCODE:
645 q = DATA_PREFIX;
646 break;
647 }
648
649 if (i.prefix[q])
650 {
651 as_bad (_("same type of prefix used twice"));
652 return 0;
653 }
654
655 i.prefixes += 1;
656 i.prefix[q] = prefix;
657 return ret;
658 }
659
660 static void
661 set_code_flag (value)
662 int value;
663 {
664 flag_code = value;
665 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
666 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
667 if (value == CODE_64BIT && !(cpu_arch_flags & CpuSledgehammer))
668 {
669 as_bad (_("64bit mode not supported on this CPU."));
670 }
671 if (value == CODE_32BIT && !(cpu_arch_flags & Cpu386))
672 {
673 as_bad (_("32bit mode not supported on this CPU."));
674 }
675 stackop_size = '\0';
676 }
677
678 static void
679 set_16bit_gcc_code_flag (new_code_flag)
680 int new_code_flag;
681 {
682 flag_code = new_code_flag;
683 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
684 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
685 stackop_size = 'l';
686 }
687
688 static void
689 set_intel_syntax (syntax_flag)
690 int syntax_flag;
691 {
692 /* Find out if register prefixing is specified. */
693 int ask_naked_reg = 0;
694
695 SKIP_WHITESPACE ();
696 if (! is_end_of_line[(unsigned char) *input_line_pointer])
697 {
698 char *string = input_line_pointer;
699 int e = get_symbol_end ();
700
701 if (strcmp (string, "prefix") == 0)
702 ask_naked_reg = 1;
703 else if (strcmp (string, "noprefix") == 0)
704 ask_naked_reg = -1;
705 else
706 as_bad (_("bad argument to syntax directive."));
707 *input_line_pointer = e;
708 }
709 demand_empty_rest_of_line ();
710
711 intel_syntax = syntax_flag;
712
713 if (ask_naked_reg == 0)
714 {
715 #ifdef BFD_ASSEMBLER
716 allow_naked_reg = (intel_syntax
717 && (bfd_get_symbol_leading_char (stdoutput) != '\0'));
718 #else
719 /* Conservative default. */
720 allow_naked_reg = 0;
721 #endif
722 }
723 else
724 allow_naked_reg = (ask_naked_reg < 0);
725 }
726
727 static void
728 set_cpu_arch (dummy)
729 int dummy ATTRIBUTE_UNUSED;
730 {
731 SKIP_WHITESPACE ();
732
733 if (! is_end_of_line[(unsigned char) *input_line_pointer])
734 {
735 char *string = input_line_pointer;
736 int e = get_symbol_end ();
737 int i;
738
739 for (i = 0; cpu_arch[i].name; i++)
740 {
741 if (strcmp (string, cpu_arch[i].name) == 0)
742 {
743 cpu_arch_name = cpu_arch[i].name;
744 cpu_arch_flags = (cpu_arch[i].flags
745 | (flag_code == CODE_64BIT ? Cpu64 : CpuNo64));
746 break;
747 }
748 }
749 if (!cpu_arch[i].name)
750 as_bad (_("no such architecture: `%s'"), string);
751
752 *input_line_pointer = e;
753 }
754 else
755 as_bad (_("missing cpu architecture"));
756
757 no_cond_jump_promotion = 0;
758 if (*input_line_pointer == ','
759 && ! is_end_of_line[(unsigned char) input_line_pointer[1]])
760 {
761 char *string = ++input_line_pointer;
762 int e = get_symbol_end ();
763
764 if (strcmp (string, "nojumps") == 0)
765 no_cond_jump_promotion = 1;
766 else if (strcmp (string, "jumps") == 0)
767 ;
768 else
769 as_bad (_("no such architecture modifier: `%s'"), string);
770
771 *input_line_pointer = e;
772 }
773
774 demand_empty_rest_of_line ();
775 }
776
777 const pseudo_typeS md_pseudo_table[] =
778 {
779 #if !defined(OBJ_AOUT) && !defined(USE_ALIGN_PTWO)
780 {"align", s_align_bytes, 0},
781 #else
782 {"align", s_align_ptwo, 0},
783 #endif
784 {"arch", set_cpu_arch, 0},
785 #ifndef I386COFF
786 {"bss", s_bss, 0},
787 #endif
788 {"ffloat", float_cons, 'f'},
789 {"dfloat", float_cons, 'd'},
790 {"tfloat", float_cons, 'x'},
791 {"value", cons, 2},
792 {"noopt", s_ignore, 0},
793 {"optim", s_ignore, 0},
794 {"code16gcc", set_16bit_gcc_code_flag, CODE_16BIT},
795 {"code16", set_code_flag, CODE_16BIT},
796 {"code32", set_code_flag, CODE_32BIT},
797 {"code64", set_code_flag, CODE_64BIT},
798 {"intel_syntax", set_intel_syntax, 1},
799 {"att_syntax", set_intel_syntax, 0},
800 {"file", dwarf2_directive_file, 0},
801 {"loc", dwarf2_directive_loc, 0},
802 {0, 0, 0}
803 };
804
805 /* For interface with expression (). */
806 extern char *input_line_pointer;
807
808 /* Hash table for instruction mnemonic lookup. */
809 static struct hash_control *op_hash;
810
811 /* Hash table for register lookup. */
812 static struct hash_control *reg_hash;
813 \f
814 #ifdef BFD_ASSEMBLER
815 unsigned long
816 i386_mach ()
817 {
818 if (!strcmp (default_arch, "x86_64"))
819 return bfd_mach_x86_64;
820 else if (!strcmp (default_arch, "i386"))
821 return bfd_mach_i386_i386;
822 else
823 as_fatal (_("Unknown architecture"));
824 }
825 #endif
826 \f
827 void
828 md_begin ()
829 {
830 const char *hash_err;
831
832 /* Initialize op_hash hash table. */
833 op_hash = hash_new ();
834
835 {
836 register const template *optab;
837 register templates *core_optab;
838
839 /* Setup for loop. */
840 optab = i386_optab;
841 core_optab = (templates *) xmalloc (sizeof (templates));
842 core_optab->start = optab;
843
844 while (1)
845 {
846 ++optab;
847 if (optab->name == NULL
848 || strcmp (optab->name, (optab - 1)->name) != 0)
849 {
850 /* different name --> ship out current template list;
851 add to hash table; & begin anew. */
852 core_optab->end = optab;
853 hash_err = hash_insert (op_hash,
854 (optab - 1)->name,
855 (PTR) core_optab);
856 if (hash_err)
857 {
858 as_fatal (_("Internal Error: Can't hash %s: %s"),
859 (optab - 1)->name,
860 hash_err);
861 }
862 if (optab->name == NULL)
863 break;
864 core_optab = (templates *) xmalloc (sizeof (templates));
865 core_optab->start = optab;
866 }
867 }
868 }
869
870 /* Initialize reg_hash hash table. */
871 reg_hash = hash_new ();
872 {
873 register const reg_entry *regtab;
874
875 for (regtab = i386_regtab;
876 regtab < i386_regtab + sizeof (i386_regtab) / sizeof (i386_regtab[0]);
877 regtab++)
878 {
879 hash_err = hash_insert (reg_hash, regtab->reg_name, (PTR) regtab);
880 if (hash_err)
881 as_fatal (_("Internal Error: Can't hash %s: %s"),
882 regtab->reg_name,
883 hash_err);
884 }
885 }
886
887 /* Fill in lexical tables: mnemonic_chars, operand_chars. */
888 {
889 register int c;
890 register char *p;
891
892 for (c = 0; c < 256; c++)
893 {
894 if (isdigit (c))
895 {
896 digit_chars[c] = c;
897 mnemonic_chars[c] = c;
898 register_chars[c] = c;
899 operand_chars[c] = c;
900 }
901 else if (islower (c))
902 {
903 mnemonic_chars[c] = c;
904 register_chars[c] = c;
905 operand_chars[c] = c;
906 }
907 else if (isupper (c))
908 {
909 mnemonic_chars[c] = tolower (c);
910 register_chars[c] = mnemonic_chars[c];
911 operand_chars[c] = c;
912 }
913
914 if (isalpha (c) || isdigit (c))
915 identifier_chars[c] = c;
916 else if (c >= 128)
917 {
918 identifier_chars[c] = c;
919 operand_chars[c] = c;
920 }
921 }
922
923 #ifdef LEX_AT
924 identifier_chars['@'] = '@';
925 #endif
926 digit_chars['-'] = '-';
927 identifier_chars['_'] = '_';
928 identifier_chars['.'] = '.';
929
930 for (p = operand_special_chars; *p != '\0'; p++)
931 operand_chars[(unsigned char) *p] = *p;
932 }
933
934 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
935 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
936 {
937 record_alignment (text_section, 2);
938 record_alignment (data_section, 2);
939 record_alignment (bss_section, 2);
940 }
941 #endif
942 }
943
944 void
945 i386_print_statistics (file)
946 FILE *file;
947 {
948 hash_print_statistics (file, "i386 opcode", op_hash);
949 hash_print_statistics (file, "i386 register", reg_hash);
950 }
951 \f
952 #ifdef DEBUG386
953
954 /* Debugging routines for md_assemble. */
955 static void pi PARAMS ((char *, i386_insn *));
956 static void pte PARAMS ((template *));
957 static void pt PARAMS ((unsigned int));
958 static void pe PARAMS ((expressionS *));
959 static void ps PARAMS ((symbolS *));
960
961 static void
962 pi (line, x)
963 char *line;
964 i386_insn *x;
965 {
966 unsigned int i;
967
968 fprintf (stdout, "%s: template ", line);
969 pte (&x->tm);
970 fprintf (stdout, " address: base %s index %s scale %x\n",
971 x->base_reg ? x->base_reg->reg_name : "none",
972 x->index_reg ? x->index_reg->reg_name : "none",
973 x->log2_scale_factor);
974 fprintf (stdout, " modrm: mode %x reg %x reg/mem %x\n",
975 x->rm.mode, x->rm.reg, x->rm.regmem);
976 fprintf (stdout, " sib: base %x index %x scale %x\n",
977 x->sib.base, x->sib.index, x->sib.scale);
978 fprintf (stdout, " rex: 64bit %x extX %x extY %x extZ %x\n",
979 x->rex.mode64, x->rex.extX, x->rex.extY, x->rex.extZ);
980 for (i = 0; i < x->operands; i++)
981 {
982 fprintf (stdout, " #%d: ", i + 1);
983 pt (x->types[i]);
984 fprintf (stdout, "\n");
985 if (x->types[i]
986 & (Reg | SReg2 | SReg3 | Control | Debug | Test | RegMMX | RegXMM))
987 fprintf (stdout, "%s\n", x->op[i].regs->reg_name);
988 if (x->types[i] & Imm)
989 pe (x->op[i].imms);
990 if (x->types[i] & Disp)
991 pe (x->op[i].disps);
992 }
993 }
994
995 static void
996 pte (t)
997 template *t;
998 {
999 unsigned int i;
1000 fprintf (stdout, " %d operands ", t->operands);
1001 fprintf (stdout, "opcode %x ", t->base_opcode);
1002 if (t->extension_opcode != None)
1003 fprintf (stdout, "ext %x ", t->extension_opcode);
1004 if (t->opcode_modifier & D)
1005 fprintf (stdout, "D");
1006 if (t->opcode_modifier & W)
1007 fprintf (stdout, "W");
1008 fprintf (stdout, "\n");
1009 for (i = 0; i < t->operands; i++)
1010 {
1011 fprintf (stdout, " #%d type ", i + 1);
1012 pt (t->operand_types[i]);
1013 fprintf (stdout, "\n");
1014 }
1015 }
1016
1017 static void
1018 pe (e)
1019 expressionS *e;
1020 {
1021 fprintf (stdout, " operation %d\n", e->X_op);
1022 fprintf (stdout, " add_number %ld (%lx)\n",
1023 (long) e->X_add_number, (long) e->X_add_number);
1024 if (e->X_add_symbol)
1025 {
1026 fprintf (stdout, " add_symbol ");
1027 ps (e->X_add_symbol);
1028 fprintf (stdout, "\n");
1029 }
1030 if (e->X_op_symbol)
1031 {
1032 fprintf (stdout, " op_symbol ");
1033 ps (e->X_op_symbol);
1034 fprintf (stdout, "\n");
1035 }
1036 }
1037
1038 static void
1039 ps (s)
1040 symbolS *s;
1041 {
1042 fprintf (stdout, "%s type %s%s",
1043 S_GET_NAME (s),
1044 S_IS_EXTERNAL (s) ? "EXTERNAL " : "",
1045 segment_name (S_GET_SEGMENT (s)));
1046 }
1047
1048 struct type_name
1049 {
1050 unsigned int mask;
1051 char *tname;
1052 }
1053
1054 type_names[] =
1055 {
1056 { Reg8, "r8" },
1057 { Reg16, "r16" },
1058 { Reg32, "r32" },
1059 { Reg64, "r64" },
1060 { Imm8, "i8" },
1061 { Imm8S, "i8s" },
1062 { Imm16, "i16" },
1063 { Imm32, "i32" },
1064 { Imm32S, "i32s" },
1065 { Imm64, "i64" },
1066 { Imm1, "i1" },
1067 { BaseIndex, "BaseIndex" },
1068 { Disp8, "d8" },
1069 { Disp16, "d16" },
1070 { Disp32, "d32" },
1071 { Disp32S, "d32s" },
1072 { Disp64, "d64" },
1073 { InOutPortReg, "InOutPortReg" },
1074 { ShiftCount, "ShiftCount" },
1075 { Control, "control reg" },
1076 { Test, "test reg" },
1077 { Debug, "debug reg" },
1078 { FloatReg, "FReg" },
1079 { FloatAcc, "FAcc" },
1080 { SReg2, "SReg2" },
1081 { SReg3, "SReg3" },
1082 { Acc, "Acc" },
1083 { JumpAbsolute, "Jump Absolute" },
1084 { RegMMX, "rMMX" },
1085 { RegXMM, "rXMM" },
1086 { EsSeg, "es" },
1087 { 0, "" }
1088 };
1089
1090 static void
1091 pt (t)
1092 unsigned int t;
1093 {
1094 register struct type_name *ty;
1095
1096 for (ty = type_names; ty->mask; ty++)
1097 if (t & ty->mask)
1098 fprintf (stdout, "%s, ", ty->tname);
1099 fflush (stdout);
1100 }
1101
1102 #endif /* DEBUG386 */
1103 \f
1104 int
1105 tc_i386_force_relocation (fixp)
1106 struct fix *fixp;
1107 {
1108 #ifdef BFD_ASSEMBLER
1109 if (fixp->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1110 || fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1111 return 1;
1112 return 0;
1113 #else
1114 /* For COFF. */
1115 return fixp->fx_r_type == 7;
1116 #endif
1117 }
1118
1119 #ifdef BFD_ASSEMBLER
1120
1121 static bfd_reloc_code_real_type
1122 reloc (size, pcrel, sign, other)
1123 int size;
1124 int pcrel;
1125 int sign;
1126 bfd_reloc_code_real_type other;
1127 {
1128 if (other != NO_RELOC)
1129 return other;
1130
1131 if (pcrel)
1132 {
1133 if (!sign)
1134 as_bad (_("There are no unsigned pc-relative relocations"));
1135 switch (size)
1136 {
1137 case 1: return BFD_RELOC_8_PCREL;
1138 case 2: return BFD_RELOC_16_PCREL;
1139 case 4: return BFD_RELOC_32_PCREL;
1140 }
1141 as_bad (_("can not do %d byte pc-relative relocation"), size);
1142 }
1143 else
1144 {
1145 if (sign)
1146 switch (size)
1147 {
1148 case 4: return BFD_RELOC_X86_64_32S;
1149 }
1150 else
1151 switch (size)
1152 {
1153 case 1: return BFD_RELOC_8;
1154 case 2: return BFD_RELOC_16;
1155 case 4: return BFD_RELOC_32;
1156 case 8: return BFD_RELOC_64;
1157 }
1158 as_bad (_("can not do %s %d byte relocation"),
1159 sign ? "signed" : "unsigned", size);
1160 }
1161
1162 abort ();
1163 return BFD_RELOC_NONE;
1164 }
1165
1166 /* Here we decide which fixups can be adjusted to make them relative to
1167 the beginning of the section instead of the symbol. Basically we need
1168 to make sure that the dynamic relocations are done correctly, so in
1169 some cases we force the original symbol to be used. */
1170
1171 int
1172 tc_i386_fix_adjustable (fixP)
1173 fixS *fixP;
1174 {
1175 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1176 /* Prevent all adjustments to global symbols, or else dynamic
1177 linking will not work correctly. */
1178 if (S_IS_EXTERNAL (fixP->fx_addsy)
1179 || S_IS_WEAK (fixP->fx_addsy))
1180 return 0;
1181 #endif
1182 /* adjust_reloc_syms doesn't know about the GOT. */
1183 if (fixP->fx_r_type == BFD_RELOC_386_GOTOFF
1184 || fixP->fx_r_type == BFD_RELOC_386_PLT32
1185 || fixP->fx_r_type == BFD_RELOC_386_GOT32
1186 || fixP->fx_r_type == BFD_RELOC_X86_64_PLT32
1187 || fixP->fx_r_type == BFD_RELOC_X86_64_GOT32
1188 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPCREL
1189 || fixP->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1190 || fixP->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1191 return 0;
1192 return 1;
1193 }
1194 #else
1195 #define reloc(SIZE,PCREL,SIGN,OTHER) 0
1196 #define BFD_RELOC_16 0
1197 #define BFD_RELOC_32 0
1198 #define BFD_RELOC_16_PCREL 0
1199 #define BFD_RELOC_32_PCREL 0
1200 #define BFD_RELOC_386_PLT32 0
1201 #define BFD_RELOC_386_GOT32 0
1202 #define BFD_RELOC_386_GOTOFF 0
1203 #define BFD_RELOC_X86_64_PLT32 0
1204 #define BFD_RELOC_X86_64_GOT32 0
1205 #define BFD_RELOC_X86_64_GOTPCREL 0
1206 #endif
1207
1208 static int intel_float_operand PARAMS ((char *mnemonic));
1209
1210 static int
1211 intel_float_operand (mnemonic)
1212 char *mnemonic;
1213 {
1214 if (mnemonic[0] == 'f' && mnemonic[1] == 'i')
1215 return 2;
1216
1217 if (mnemonic[0] == 'f')
1218 return 1;
1219
1220 return 0;
1221 }
1222
1223 /* This is the guts of the machine-dependent assembler. LINE points to a
1224 machine dependent instruction. This function is supposed to emit
1225 the frags/bytes it assembles to. */
1226
1227 void
1228 md_assemble (line)
1229 char *line;
1230 {
1231 /* Points to template once we've found it. */
1232 const template *t;
1233
1234 /* Count the size of the instruction generated. Does not include
1235 variable part of jump insns before relax. */
1236 int insn_size = 0;
1237
1238 int j;
1239
1240 char mnemonic[MAX_MNEM_SIZE];
1241
1242 /* Initialize globals. */
1243 memset (&i, '\0', sizeof (i));
1244 for (j = 0; j < MAX_OPERANDS; j++)
1245 i.reloc[j] = NO_RELOC;
1246 memset (disp_expressions, '\0', sizeof (disp_expressions));
1247 memset (im_expressions, '\0', sizeof (im_expressions));
1248 save_stack_p = save_stack;
1249
1250 /* First parse an instruction mnemonic & call i386_operand for the operands.
1251 We assume that the scrubber has arranged it so that line[0] is the valid
1252 start of a (possibly prefixed) mnemonic. */
1253 {
1254 char *l = line;
1255 char *token_start = l;
1256 char *mnem_p;
1257
1258 /* Non-zero if we found a prefix only acceptable with string insns. */
1259 const char *expecting_string_instruction = NULL;
1260
1261 while (1)
1262 {
1263 mnem_p = mnemonic;
1264 while ((*mnem_p = mnemonic_chars[(unsigned char) *l]) != 0)
1265 {
1266 mnem_p++;
1267 if (mnem_p >= mnemonic + sizeof (mnemonic))
1268 {
1269 as_bad (_("no such instruction: `%s'"), token_start);
1270 return;
1271 }
1272 l++;
1273 }
1274 if (!is_space_char (*l)
1275 && *l != END_OF_INSN
1276 && *l != PREFIX_SEPARATOR)
1277 {
1278 as_bad (_("invalid character %s in mnemonic"),
1279 output_invalid (*l));
1280 return;
1281 }
1282 if (token_start == l)
1283 {
1284 if (*l == PREFIX_SEPARATOR)
1285 as_bad (_("expecting prefix; got nothing"));
1286 else
1287 as_bad (_("expecting mnemonic; got nothing"));
1288 return;
1289 }
1290
1291 /* Look up instruction (or prefix) via hash table. */
1292 current_templates = hash_find (op_hash, mnemonic);
1293
1294 if (*l != END_OF_INSN
1295 && (! is_space_char (*l) || l[1] != END_OF_INSN)
1296 && current_templates
1297 && (current_templates->start->opcode_modifier & IsPrefix))
1298 {
1299 /* If we are in 16-bit mode, do not allow addr16 or data16.
1300 Similarly, in 32-bit mode, do not allow addr32 or data32. */
1301 if ((current_templates->start->opcode_modifier & (Size16 | Size32))
1302 && (((current_templates->start->opcode_modifier & Size32) != 0)
1303 ^ (flag_code == CODE_16BIT)))
1304 {
1305 as_bad (_("redundant %s prefix"),
1306 current_templates->start->name);
1307 return;
1308 }
1309 /* Add prefix, checking for repeated prefixes. */
1310 switch (add_prefix (current_templates->start->base_opcode))
1311 {
1312 case 0:
1313 return;
1314 case 2:
1315 expecting_string_instruction = current_templates->start->name;
1316 break;
1317 }
1318 /* Skip past PREFIX_SEPARATOR and reset token_start. */
1319 token_start = ++l;
1320 }
1321 else
1322 break;
1323 }
1324
1325 if (!current_templates)
1326 {
1327 /* See if we can get a match by trimming off a suffix. */
1328 switch (mnem_p[-1])
1329 {
1330 case WORD_MNEM_SUFFIX:
1331 case BYTE_MNEM_SUFFIX:
1332 case QWORD_MNEM_SUFFIX:
1333 i.suffix = mnem_p[-1];
1334 mnem_p[-1] = '\0';
1335 current_templates = hash_find (op_hash, mnemonic);
1336 break;
1337 case SHORT_MNEM_SUFFIX:
1338 case LONG_MNEM_SUFFIX:
1339 if (!intel_syntax)
1340 {
1341 i.suffix = mnem_p[-1];
1342 mnem_p[-1] = '\0';
1343 current_templates = hash_find (op_hash, mnemonic);
1344 }
1345 break;
1346
1347 /* Intel Syntax. */
1348 case 'd':
1349 if (intel_syntax)
1350 {
1351 if (intel_float_operand (mnemonic))
1352 i.suffix = SHORT_MNEM_SUFFIX;
1353 else
1354 i.suffix = LONG_MNEM_SUFFIX;
1355 mnem_p[-1] = '\0';
1356 current_templates = hash_find (op_hash, mnemonic);
1357 }
1358 break;
1359 }
1360 if (!current_templates)
1361 {
1362 as_bad (_("no such instruction: `%s'"), token_start);
1363 return;
1364 }
1365 }
1366
1367 /* Check if instruction is supported on specified architecture. */
1368 if (cpu_arch_flags != 0)
1369 {
1370 if ((current_templates->start->cpu_flags & ~(Cpu64 | CpuNo64))
1371 & ~(cpu_arch_flags & ~(Cpu64 | CpuNo64)))
1372 {
1373 as_warn (_("`%s' is not supported on `%s'"),
1374 current_templates->start->name, cpu_arch_name);
1375 }
1376 else if ((Cpu386 & ~cpu_arch_flags) && (flag_code != CODE_16BIT))
1377 {
1378 as_warn (_("use .code16 to ensure correct addressing mode"));
1379 }
1380 }
1381
1382 /* Check for rep/repne without a string instruction. */
1383 if (expecting_string_instruction
1384 && !(current_templates->start->opcode_modifier & IsString))
1385 {
1386 as_bad (_("expecting string instruction after `%s'"),
1387 expecting_string_instruction);
1388 return;
1389 }
1390
1391 /* There may be operands to parse. */
1392 if (*l != END_OF_INSN)
1393 {
1394 /* 1 if operand is pending after ','. */
1395 unsigned int expecting_operand = 0;
1396
1397 /* Non-zero if operand parens not balanced. */
1398 unsigned int paren_not_balanced;
1399
1400 do
1401 {
1402 /* Skip optional white space before operand. */
1403 if (is_space_char (*l))
1404 ++l;
1405 if (!is_operand_char (*l) && *l != END_OF_INSN)
1406 {
1407 as_bad (_("invalid character %s before operand %d"),
1408 output_invalid (*l),
1409 i.operands + 1);
1410 return;
1411 }
1412 token_start = l; /* after white space */
1413 paren_not_balanced = 0;
1414 while (paren_not_balanced || *l != ',')
1415 {
1416 if (*l == END_OF_INSN)
1417 {
1418 if (paren_not_balanced)
1419 {
1420 if (!intel_syntax)
1421 as_bad (_("unbalanced parenthesis in operand %d."),
1422 i.operands + 1);
1423 else
1424 as_bad (_("unbalanced brackets in operand %d."),
1425 i.operands + 1);
1426 return;
1427 }
1428 else
1429 break; /* we are done */
1430 }
1431 else if (!is_operand_char (*l) && !is_space_char (*l))
1432 {
1433 as_bad (_("invalid character %s in operand %d"),
1434 output_invalid (*l),
1435 i.operands + 1);
1436 return;
1437 }
1438 if (!intel_syntax)
1439 {
1440 if (*l == '(')
1441 ++paren_not_balanced;
1442 if (*l == ')')
1443 --paren_not_balanced;
1444 }
1445 else
1446 {
1447 if (*l == '[')
1448 ++paren_not_balanced;
1449 if (*l == ']')
1450 --paren_not_balanced;
1451 }
1452 l++;
1453 }
1454 if (l != token_start)
1455 { /* Yes, we've read in another operand. */
1456 unsigned int operand_ok;
1457 this_operand = i.operands++;
1458 if (i.operands > MAX_OPERANDS)
1459 {
1460 as_bad (_("spurious operands; (%d operands/instruction max)"),
1461 MAX_OPERANDS);
1462 return;
1463 }
1464 /* Now parse operand adding info to 'i' as we go along. */
1465 END_STRING_AND_SAVE (l);
1466
1467 if (intel_syntax)
1468 operand_ok =
1469 i386_intel_operand (token_start,
1470 intel_float_operand (mnemonic));
1471 else
1472 operand_ok = i386_operand (token_start);
1473
1474 RESTORE_END_STRING (l);
1475 if (!operand_ok)
1476 return;
1477 }
1478 else
1479 {
1480 if (expecting_operand)
1481 {
1482 expecting_operand_after_comma:
1483 as_bad (_("expecting operand after ','; got nothing"));
1484 return;
1485 }
1486 if (*l == ',')
1487 {
1488 as_bad (_("expecting operand before ','; got nothing"));
1489 return;
1490 }
1491 }
1492
1493 /* Now *l must be either ',' or END_OF_INSN. */
1494 if (*l == ',')
1495 {
1496 if (*++l == END_OF_INSN)
1497 {
1498 /* Just skip it, if it's \n complain. */
1499 goto expecting_operand_after_comma;
1500 }
1501 expecting_operand = 1;
1502 }
1503 }
1504 while (*l != END_OF_INSN);
1505 }
1506 }
1507
1508 /* Now we've parsed the mnemonic into a set of templates, and have the
1509 operands at hand.
1510
1511 Next, we find a template that matches the given insn,
1512 making sure the overlap of the given operands types is consistent
1513 with the template operand types. */
1514
1515 #define MATCH(overlap, given, template) \
1516 ((overlap & ~JumpAbsolute) \
1517 && ((given) & (BaseIndex|JumpAbsolute)) == ((overlap) & (BaseIndex|JumpAbsolute)))
1518
1519 /* If given types r0 and r1 are registers they must be of the same type
1520 unless the expected operand type register overlap is null.
1521 Note that Acc in a template matches every size of reg. */
1522 #define CONSISTENT_REGISTER_MATCH(m0, g0, t0, m1, g1, t1) \
1523 ( ((g0) & Reg) == 0 || ((g1) & Reg) == 0 || \
1524 ((g0) & Reg) == ((g1) & Reg) || \
1525 ((((m0) & Acc) ? Reg : (t0)) & (((m1) & Acc) ? Reg : (t1)) & Reg) == 0 )
1526
1527 {
1528 register unsigned int overlap0, overlap1;
1529 unsigned int overlap2;
1530 unsigned int found_reverse_match;
1531 int suffix_check;
1532
1533 /* All intel opcodes have reversed operands except for "bound" and
1534 "enter". We also don't reverse intersegment "jmp" and "call"
1535 instructions with 2 immediate operands so that the immediate segment
1536 precedes the offset, as it does when in AT&T mode. "enter" and the
1537 intersegment "jmp" and "call" instructions are the only ones that
1538 have two immediate operands. */
1539 if (intel_syntax && i.operands > 1
1540 && (strcmp (mnemonic, "bound") != 0)
1541 && !((i.types[0] & Imm) && (i.types[1] & Imm)))
1542 {
1543 union i386_op temp_op;
1544 unsigned int temp_type;
1545 RELOC_ENUM temp_reloc;
1546 int xchg1 = 0;
1547 int xchg2 = 0;
1548
1549 if (i.operands == 2)
1550 {
1551 xchg1 = 0;
1552 xchg2 = 1;
1553 }
1554 else if (i.operands == 3)
1555 {
1556 xchg1 = 0;
1557 xchg2 = 2;
1558 }
1559 temp_type = i.types[xchg2];
1560 i.types[xchg2] = i.types[xchg1];
1561 i.types[xchg1] = temp_type;
1562 temp_op = i.op[xchg2];
1563 i.op[xchg2] = i.op[xchg1];
1564 i.op[xchg1] = temp_op;
1565 temp_reloc = i.reloc[xchg2];
1566 i.reloc[xchg2] = i.reloc[xchg1];
1567 i.reloc[xchg1] = temp_reloc;
1568
1569 if (i.mem_operands == 2)
1570 {
1571 const seg_entry *temp_seg;
1572 temp_seg = i.seg[0];
1573 i.seg[0] = i.seg[1];
1574 i.seg[1] = temp_seg;
1575 }
1576 }
1577
1578 if (i.imm_operands)
1579 {
1580 /* Try to ensure constant immediates are represented in the smallest
1581 opcode possible. */
1582 char guess_suffix = 0;
1583 int op;
1584
1585 if (i.suffix)
1586 guess_suffix = i.suffix;
1587 else if (i.reg_operands)
1588 {
1589 /* Figure out a suffix from the last register operand specified.
1590 We can't do this properly yet, ie. excluding InOutPortReg,
1591 but the following works for instructions with immediates.
1592 In any case, we can't set i.suffix yet. */
1593 for (op = i.operands; --op >= 0;)
1594 if (i.types[op] & Reg)
1595 {
1596 if (i.types[op] & Reg8)
1597 guess_suffix = BYTE_MNEM_SUFFIX;
1598 else if (i.types[op] & Reg16)
1599 guess_suffix = WORD_MNEM_SUFFIX;
1600 else if (i.types[op] & Reg32)
1601 guess_suffix = LONG_MNEM_SUFFIX;
1602 else if (i.types[op] & Reg64)
1603 guess_suffix = QWORD_MNEM_SUFFIX;
1604 break;
1605 }
1606 }
1607 else if ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0))
1608 guess_suffix = WORD_MNEM_SUFFIX;
1609
1610 for (op = i.operands; --op >= 0;)
1611 if (i.types[op] & Imm)
1612 {
1613 switch (i.op[op].imms->X_op)
1614 {
1615 case O_constant:
1616 /* If a suffix is given, this operand may be shortened. */
1617 switch (guess_suffix)
1618 {
1619 case LONG_MNEM_SUFFIX:
1620 i.types[op] |= Imm32 | Imm64;
1621 break;
1622 case WORD_MNEM_SUFFIX:
1623 i.types[op] |= Imm16 | Imm32S | Imm32 | Imm64;
1624 break;
1625 case BYTE_MNEM_SUFFIX:
1626 i.types[op] |= Imm16 | Imm8 | Imm8S | Imm32S | Imm32 | Imm64;
1627 break;
1628 }
1629
1630 /* If this operand is at most 16 bits, convert it
1631 to a signed 16 bit number before trying to see
1632 whether it will fit in an even smaller size.
1633 This allows a 16-bit operand such as $0xffe0 to
1634 be recognised as within Imm8S range. */
1635 if ((i.types[op] & Imm16)
1636 && (i.op[op].imms->X_add_number & ~(offsetT) 0xffff) == 0)
1637 {
1638 i.op[op].imms->X_add_number =
1639 (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000);
1640 }
1641 if ((i.types[op] & Imm32)
1642 && (i.op[op].imms->X_add_number & ~(((offsetT) 2 << 31) - 1)) == 0)
1643 {
1644 i.op[op].imms->X_add_number =
1645 (i.op[op].imms->X_add_number ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
1646 }
1647 i.types[op] |= smallest_imm_type (i.op[op].imms->X_add_number);
1648 /* We must avoid matching of Imm32 templates when 64bit only immediate is available. */
1649 if (guess_suffix == QWORD_MNEM_SUFFIX)
1650 i.types[op] &= ~Imm32;
1651 break;
1652 case O_absent:
1653 case O_register:
1654 abort ();
1655 /* Symbols and expressions. */
1656 default:
1657 /* Convert symbolic operand to proper sizes for matching. */
1658 switch (guess_suffix)
1659 {
1660 case QWORD_MNEM_SUFFIX:
1661 i.types[op] = Imm64 | Imm32S;
1662 break;
1663 case LONG_MNEM_SUFFIX:
1664 i.types[op] = Imm32 | Imm64;
1665 break;
1666 case WORD_MNEM_SUFFIX:
1667 i.types[op] = Imm16 | Imm32 | Imm64;
1668 break;
1669 break;
1670 case BYTE_MNEM_SUFFIX:
1671 i.types[op] = Imm8 | Imm8S | Imm16 | Imm32S | Imm32;
1672 break;
1673 break;
1674 }
1675 break;
1676 }
1677 }
1678 }
1679
1680 if (i.disp_operands)
1681 {
1682 /* Try to use the smallest displacement type too. */
1683 int op;
1684
1685 for (op = i.operands; --op >= 0;)
1686 if ((i.types[op] & Disp)
1687 && i.op[op].disps->X_op == O_constant)
1688 {
1689 offsetT disp = i.op[op].disps->X_add_number;
1690
1691 if (i.types[op] & Disp16)
1692 {
1693 /* We know this operand is at most 16 bits, so
1694 convert to a signed 16 bit number before trying
1695 to see whether it will fit in an even smaller
1696 size. */
1697
1698 disp = (((disp & 0xffff) ^ 0x8000) - 0x8000);
1699 }
1700 else if (i.types[op] & Disp32)
1701 {
1702 /* We know this operand is at most 32 bits, so convert to a
1703 signed 32 bit number before trying to see whether it will
1704 fit in an even smaller size. */
1705 disp &= (((offsetT) 2 << 31) - 1);
1706 disp = (disp ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
1707 }
1708 if (flag_code == CODE_64BIT)
1709 {
1710 if (fits_in_signed_long (disp))
1711 i.types[op] |= Disp32S;
1712 if (fits_in_unsigned_long (disp))
1713 i.types[op] |= Disp32;
1714 }
1715 if ((i.types[op] & (Disp32 | Disp32S | Disp16))
1716 && fits_in_signed_byte (disp))
1717 i.types[op] |= Disp8;
1718 }
1719 }
1720
1721 overlap0 = 0;
1722 overlap1 = 0;
1723 overlap2 = 0;
1724 found_reverse_match = 0;
1725 suffix_check = (i.suffix == BYTE_MNEM_SUFFIX
1726 ? No_bSuf
1727 : (i.suffix == WORD_MNEM_SUFFIX
1728 ? No_wSuf
1729 : (i.suffix == SHORT_MNEM_SUFFIX
1730 ? No_sSuf
1731 : (i.suffix == LONG_MNEM_SUFFIX
1732 ? No_lSuf
1733 : (i.suffix == QWORD_MNEM_SUFFIX
1734 ? No_qSuf
1735 : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX ? No_xSuf : 0))))));
1736
1737 for (t = current_templates->start;
1738 t < current_templates->end;
1739 t++)
1740 {
1741 /* Must have right number of operands. */
1742 if (i.operands != t->operands)
1743 continue;
1744
1745 /* Check the suffix, except for some instructions in intel mode. */
1746 if ((t->opcode_modifier & suffix_check)
1747 && !(intel_syntax
1748 && (t->opcode_modifier & IgnoreSize))
1749 && !(intel_syntax
1750 && t->base_opcode == 0xd9
1751 && (t->extension_opcode == 5 /* 0xd9,5 "fldcw" */
1752 || t->extension_opcode == 7))) /* 0xd9,7 "f{n}stcw" */
1753 continue;
1754
1755 /* Do not verify operands when there are none. */
1756 else if (!t->operands)
1757 {
1758 if (t->cpu_flags & ~cpu_arch_flags)
1759 continue;
1760 /* We've found a match; break out of loop. */
1761 break;
1762 }
1763
1764 overlap0 = i.types[0] & t->operand_types[0];
1765 switch (t->operands)
1766 {
1767 case 1:
1768 if (!MATCH (overlap0, i.types[0], t->operand_types[0]))
1769 continue;
1770 break;
1771 case 2:
1772 case 3:
1773 overlap1 = i.types[1] & t->operand_types[1];
1774 if (!MATCH (overlap0, i.types[0], t->operand_types[0])
1775 || !MATCH (overlap1, i.types[1], t->operand_types[1])
1776 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
1777 t->operand_types[0],
1778 overlap1, i.types[1],
1779 t->operand_types[1]))
1780 {
1781 /* Check if other direction is valid ... */
1782 if ((t->opcode_modifier & (D|FloatD)) == 0)
1783 continue;
1784
1785 /* Try reversing direction of operands. */
1786 overlap0 = i.types[0] & t->operand_types[1];
1787 overlap1 = i.types[1] & t->operand_types[0];
1788 if (!MATCH (overlap0, i.types[0], t->operand_types[1])
1789 || !MATCH (overlap1, i.types[1], t->operand_types[0])
1790 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
1791 t->operand_types[1],
1792 overlap1, i.types[1],
1793 t->operand_types[0]))
1794 {
1795 /* Does not match either direction. */
1796 continue;
1797 }
1798 /* found_reverse_match holds which of D or FloatDR
1799 we've found. */
1800 found_reverse_match = t->opcode_modifier & (D|FloatDR);
1801 }
1802 /* Found a forward 2 operand match here. */
1803 else if (t->operands == 3)
1804 {
1805 /* Here we make use of the fact that there are no
1806 reverse match 3 operand instructions, and all 3
1807 operand instructions only need to be checked for
1808 register consistency between operands 2 and 3. */
1809 overlap2 = i.types[2] & t->operand_types[2];
1810 if (!MATCH (overlap2, i.types[2], t->operand_types[2])
1811 || !CONSISTENT_REGISTER_MATCH (overlap1, i.types[1],
1812 t->operand_types[1],
1813 overlap2, i.types[2],
1814 t->operand_types[2]))
1815
1816 continue;
1817 }
1818 /* Found either forward/reverse 2 or 3 operand match here:
1819 slip through to break. */
1820 }
1821 if (t->cpu_flags & ~cpu_arch_flags)
1822 {
1823 found_reverse_match = 0;
1824 continue;
1825 }
1826 /* We've found a match; break out of loop. */
1827 break;
1828 }
1829 if (t == current_templates->end)
1830 {
1831 /* We found no match. */
1832 as_bad (_("suffix or operands invalid for `%s'"),
1833 current_templates->start->name);
1834 return;
1835 }
1836
1837 if (!quiet_warnings)
1838 {
1839 if (!intel_syntax
1840 && ((i.types[0] & JumpAbsolute)
1841 != (t->operand_types[0] & JumpAbsolute)))
1842 {
1843 as_warn (_("indirect %s without `*'"), t->name);
1844 }
1845
1846 if ((t->opcode_modifier & (IsPrefix|IgnoreSize))
1847 == (IsPrefix|IgnoreSize))
1848 {
1849 /* Warn them that a data or address size prefix doesn't
1850 affect assembly of the next line of code. */
1851 as_warn (_("stand-alone `%s' prefix"), t->name);
1852 }
1853 }
1854
1855 /* Copy the template we found. */
1856 i.tm = *t;
1857 if (found_reverse_match)
1858 {
1859 /* If we found a reverse match we must alter the opcode
1860 direction bit. found_reverse_match holds bits to change
1861 (different for int & float insns). */
1862
1863 i.tm.base_opcode ^= found_reverse_match;
1864
1865 i.tm.operand_types[0] = t->operand_types[1];
1866 i.tm.operand_types[1] = t->operand_types[0];
1867 }
1868
1869 /* Undo SYSV386_COMPAT brokenness when in Intel mode. See i386.h */
1870 if (SYSV386_COMPAT
1871 && intel_syntax
1872 && (i.tm.base_opcode & 0xfffffde0) == 0xdce0)
1873 i.tm.base_opcode ^= FloatR;
1874
1875 if (i.tm.opcode_modifier & FWait)
1876 if (! add_prefix (FWAIT_OPCODE))
1877 return;
1878
1879 /* Check string instruction segment overrides. */
1880 if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0)
1881 {
1882 int mem_op = (i.types[0] & AnyMem) ? 0 : 1;
1883 if ((i.tm.operand_types[mem_op] & EsSeg) != 0)
1884 {
1885 if (i.seg[0] != NULL && i.seg[0] != &es)
1886 {
1887 as_bad (_("`%s' operand %d must use `%%es' segment"),
1888 i.tm.name,
1889 mem_op + 1);
1890 return;
1891 }
1892 /* There's only ever one segment override allowed per instruction.
1893 This instruction possibly has a legal segment override on the
1894 second operand, so copy the segment to where non-string
1895 instructions store it, allowing common code. */
1896 i.seg[0] = i.seg[1];
1897 }
1898 else if ((i.tm.operand_types[mem_op + 1] & EsSeg) != 0)
1899 {
1900 if (i.seg[1] != NULL && i.seg[1] != &es)
1901 {
1902 as_bad (_("`%s' operand %d must use `%%es' segment"),
1903 i.tm.name,
1904 mem_op + 2);
1905 return;
1906 }
1907 }
1908 }
1909
1910 if (i.reg_operands && flag_code < CODE_64BIT)
1911 {
1912 int op;
1913 for (op = i.operands; --op >= 0;)
1914 if ((i.types[op] & Reg)
1915 && (i.op[op].regs->reg_flags & (RegRex64|RegRex)))
1916 {
1917 as_bad (_("Extended register `%%%s' available only in 64bit mode."),
1918 i.op[op].regs->reg_name);
1919 return;
1920 }
1921 }
1922
1923 /* If matched instruction specifies an explicit instruction mnemonic
1924 suffix, use it. */
1925 if (i.tm.opcode_modifier & (Size16 | Size32 | Size64))
1926 {
1927 if (i.tm.opcode_modifier & Size16)
1928 i.suffix = WORD_MNEM_SUFFIX;
1929 else if (i.tm.opcode_modifier & Size64)
1930 i.suffix = QWORD_MNEM_SUFFIX;
1931 else
1932 i.suffix = LONG_MNEM_SUFFIX;
1933 }
1934 else if (i.reg_operands)
1935 {
1936 /* If there's no instruction mnemonic suffix we try to invent one
1937 based on register operands. */
1938 if (!i.suffix)
1939 {
1940 /* We take i.suffix from the last register operand specified,
1941 Destination register type is more significant than source
1942 register type. */
1943 int op;
1944 for (op = i.operands; --op >= 0;)
1945 if ((i.types[op] & Reg)
1946 && !(i.tm.operand_types[op] & InOutPortReg))
1947 {
1948 i.suffix = ((i.types[op] & Reg8) ? BYTE_MNEM_SUFFIX :
1949 (i.types[op] & Reg16) ? WORD_MNEM_SUFFIX :
1950 (i.types[op] & Reg64) ? QWORD_MNEM_SUFFIX :
1951 LONG_MNEM_SUFFIX);
1952 break;
1953 }
1954 }
1955 else if (i.suffix == BYTE_MNEM_SUFFIX)
1956 {
1957 int op;
1958 for (op = i.operands; --op >= 0;)
1959 {
1960 /* If this is an eight bit register, it's OK. If it's
1961 the 16 or 32 bit version of an eight bit register,
1962 we will just use the low portion, and that's OK too. */
1963 if (i.types[op] & Reg8)
1964 continue;
1965
1966 /* movzx and movsx should not generate this warning. */
1967 if (intel_syntax
1968 && (i.tm.base_opcode == 0xfb7
1969 || i.tm.base_opcode == 0xfb6
1970 || i.tm.base_opcode == 0x63
1971 || i.tm.base_opcode == 0xfbe
1972 || i.tm.base_opcode == 0xfbf))
1973 continue;
1974
1975 if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4
1976 #if 0
1977 /* Check that the template allows eight bit regs
1978 This kills insns such as `orb $1,%edx', which
1979 maybe should be allowed. */
1980 && (i.tm.operand_types[op] & (Reg8|InOutPortReg))
1981 #endif
1982 )
1983 {
1984 /* Prohibit these changes in the 64bit mode, since
1985 the lowering is more complicated. */
1986 if (flag_code == CODE_64BIT
1987 && (i.tm.operand_types[op] & InOutPortReg) == 0)
1988 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
1989 i.op[op].regs->reg_name,
1990 i.suffix);
1991 #if REGISTER_WARNINGS
1992 if (!quiet_warnings
1993 && (i.tm.operand_types[op] & InOutPortReg) == 0)
1994 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
1995 (i.op[op].regs
1996 + (i.types[op] & Reg16
1997 ? REGNAM_AL - REGNAM_AX
1998 : REGNAM_AL - REGNAM_EAX))->reg_name,
1999 i.op[op].regs->reg_name,
2000 i.suffix);
2001 #endif
2002 continue;
2003 }
2004 /* Any other register is bad. */
2005 if (i.types[op] & (Reg | RegMMX | RegXMM
2006 | SReg2 | SReg3
2007 | Control | Debug | Test
2008 | FloatReg | FloatAcc))
2009 {
2010 as_bad (_("`%%%s' not allowed with `%s%c'"),
2011 i.op[op].regs->reg_name,
2012 i.tm.name,
2013 i.suffix);
2014 return;
2015 }
2016 }
2017 }
2018 else if (i.suffix == LONG_MNEM_SUFFIX)
2019 {
2020 int op;
2021
2022 for (op = i.operands; --op >= 0;)
2023 /* Reject eight bit registers, except where the template
2024 requires them. (eg. movzb) */
2025 if ((i.types[op] & Reg8) != 0
2026 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2027 {
2028 as_bad (_("`%%%s' not allowed with `%s%c'"),
2029 i.op[op].regs->reg_name,
2030 i.tm.name,
2031 i.suffix);
2032 return;
2033 }
2034 /* Warn if the e prefix on a general reg is missing. */
2035 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2036 && (i.types[op] & Reg16) != 0
2037 && (i.tm.operand_types[op] & (Reg32|Acc)) != 0)
2038 {
2039 /* Prohibit these changes in the 64bit mode, since
2040 the lowering is more complicated. */
2041 if (flag_code == CODE_64BIT)
2042 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2043 i.op[op].regs->reg_name,
2044 i.suffix);
2045 #if REGISTER_WARNINGS
2046 else
2047 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2048 (i.op[op].regs + REGNAM_EAX - REGNAM_AX)->reg_name,
2049 i.op[op].regs->reg_name,
2050 i.suffix);
2051 #endif
2052 }
2053 /* Warn if the r prefix on a general reg is missing. */
2054 else if ((i.types[op] & Reg64) != 0
2055 && (i.tm.operand_types[op] & (Reg32|Acc)) != 0)
2056 {
2057 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2058 i.op[op].regs->reg_name,
2059 i.suffix);
2060 }
2061 }
2062 else if (i.suffix == QWORD_MNEM_SUFFIX)
2063 {
2064 int op;
2065
2066 for (op = i.operands; --op >= 0; )
2067 /* Reject eight bit registers, except where the template
2068 requires them. (eg. movzb) */
2069 if ((i.types[op] & Reg8) != 0
2070 && (i.tm.operand_types[op] & (Reg16|Reg32|Acc)) != 0)
2071 {
2072 as_bad (_("`%%%s' not allowed with `%s%c'"),
2073 i.op[op].regs->reg_name,
2074 i.tm.name,
2075 i.suffix);
2076 return;
2077 }
2078 /* Warn if the e prefix on a general reg is missing. */
2079 else if (((i.types[op] & Reg16) != 0
2080 || (i.types[op] & Reg32) != 0)
2081 && (i.tm.operand_types[op] & (Reg32|Acc)) != 0)
2082 {
2083 /* Prohibit these changes in the 64bit mode, since
2084 the lowering is more complicated. */
2085 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2086 i.op[op].regs->reg_name,
2087 i.suffix);
2088 }
2089 }
2090 else if (i.suffix == WORD_MNEM_SUFFIX)
2091 {
2092 int op;
2093 for (op = i.operands; --op >= 0;)
2094 /* Reject eight bit registers, except where the template
2095 requires them. (eg. movzb) */
2096 if ((i.types[op] & Reg8) != 0
2097 && (i.tm.operand_types[op] & (Reg16|Reg32|Acc)) != 0)
2098 {
2099 as_bad (_("`%%%s' not allowed with `%s%c'"),
2100 i.op[op].regs->reg_name,
2101 i.tm.name,
2102 i.suffix);
2103 return;
2104 }
2105 /* Warn if the e prefix on a general reg is present. */
2106 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2107 && (i.types[op] & Reg32) != 0
2108 && (i.tm.operand_types[op] & (Reg16|Acc)) != 0)
2109 {
2110 /* Prohibit these changes in the 64bit mode, since
2111 the lowering is more complicated. */
2112 if (flag_code == CODE_64BIT)
2113 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2114 i.op[op].regs->reg_name,
2115 i.suffix);
2116 else
2117 #if REGISTER_WARNINGS
2118 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2119 (i.op[op].regs + REGNAM_AX - REGNAM_EAX)->reg_name,
2120 i.op[op].regs->reg_name,
2121 i.suffix);
2122 #endif
2123 }
2124 }
2125 else if (intel_syntax && (i.tm.opcode_modifier & IgnoreSize))
2126 /* Do nothing if the instruction is going to ignore the prefix. */
2127 ;
2128 else
2129 abort ();
2130 }
2131 else if ((i.tm.opcode_modifier & DefaultSize) && !i.suffix)
2132 {
2133 i.suffix = stackop_size;
2134 }
2135 /* Make still unresolved immediate matches conform to size of immediate
2136 given in i.suffix. Note: overlap2 cannot be an immediate! */
2137 if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S))
2138 && overlap0 != Imm8 && overlap0 != Imm8S
2139 && overlap0 != Imm16 && overlap0 != Imm32S
2140 && overlap0 != Imm32 && overlap0 != Imm64)
2141 {
2142 if (i.suffix)
2143 {
2144 overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX ? (Imm8 | Imm8S) :
2145 (i.suffix == WORD_MNEM_SUFFIX ? Imm16 :
2146 (i.suffix == QWORD_MNEM_SUFFIX ? Imm64 | Imm32S : Imm32)));
2147 }
2148 else if (overlap0 == (Imm16 | Imm32S | Imm32)
2149 || overlap0 == (Imm16 | Imm32)
2150 || overlap0 == (Imm16 | Imm32S))
2151 {
2152 overlap0 =
2153 ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)) ? Imm16 : Imm32S;
2154 }
2155 if (overlap0 != Imm8 && overlap0 != Imm8S
2156 && overlap0 != Imm16 && overlap0 != Imm32S
2157 && overlap0 != Imm32 && overlap0 != Imm64)
2158 {
2159 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size"));
2160 return;
2161 }
2162 }
2163 if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32))
2164 && overlap1 != Imm8 && overlap1 != Imm8S
2165 && overlap1 != Imm16 && overlap1 != Imm32S
2166 && overlap1 != Imm32 && overlap1 != Imm64)
2167 {
2168 if (i.suffix)
2169 {
2170 overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX ? (Imm8 | Imm8S) :
2171 (i.suffix == WORD_MNEM_SUFFIX ? Imm16 :
2172 (i.suffix == QWORD_MNEM_SUFFIX ? Imm64 | Imm32S : Imm32)));
2173 }
2174 else if (overlap1 == (Imm16 | Imm32 | Imm32S)
2175 || overlap1 == (Imm16 | Imm32)
2176 || overlap1 == (Imm16 | Imm32S))
2177 {
2178 overlap1 =
2179 ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)) ? Imm16 : Imm32S;
2180 }
2181 if (overlap1 != Imm8 && overlap1 != Imm8S
2182 && overlap1 != Imm16 && overlap1 != Imm32S
2183 && overlap1 != Imm32 && overlap1 != Imm64)
2184 {
2185 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size %x %c"),overlap1, i.suffix);
2186 return;
2187 }
2188 }
2189 assert ((overlap2 & Imm) == 0);
2190
2191 i.types[0] = overlap0;
2192 if (overlap0 & ImplicitRegister)
2193 i.reg_operands--;
2194 if (overlap0 & Imm1)
2195 i.imm_operands = 0; /* kludge for shift insns. */
2196
2197 i.types[1] = overlap1;
2198 if (overlap1 & ImplicitRegister)
2199 i.reg_operands--;
2200
2201 i.types[2] = overlap2;
2202 if (overlap2 & ImplicitRegister)
2203 i.reg_operands--;
2204
2205 /* Finalize opcode. First, we change the opcode based on the operand
2206 size given by i.suffix: We need not change things for byte insns. */
2207
2208 if (!i.suffix && (i.tm.opcode_modifier & W))
2209 {
2210 as_bad (_("no instruction mnemonic suffix given and no register operands; can't size instruction"));
2211 return;
2212 }
2213
2214 /* For movzx and movsx, need to check the register type. */
2215 if (intel_syntax
2216 && (i.tm.base_opcode == 0xfb6 || i.tm.base_opcode == 0xfbe))
2217 if (i.suffix && i.suffix == BYTE_MNEM_SUFFIX)
2218 {
2219 unsigned int prefix = DATA_PREFIX_OPCODE;
2220
2221 if ((i.op[1].regs->reg_type & Reg16) != 0)
2222 if (!add_prefix (prefix))
2223 return;
2224 }
2225
2226 if (i.suffix && i.suffix != BYTE_MNEM_SUFFIX)
2227 {
2228 /* It's not a byte, select word/dword operation. */
2229 if (i.tm.opcode_modifier & W)
2230 {
2231 if (i.tm.opcode_modifier & ShortForm)
2232 i.tm.base_opcode |= 8;
2233 else
2234 i.tm.base_opcode |= 1;
2235 }
2236 /* Now select between word & dword operations via the operand
2237 size prefix, except for instructions that will ignore this
2238 prefix anyway. */
2239 if (i.suffix != QWORD_MNEM_SUFFIX
2240 && (i.suffix == LONG_MNEM_SUFFIX) == (flag_code == CODE_16BIT)
2241 && !(i.tm.opcode_modifier & IgnoreSize))
2242 {
2243 unsigned int prefix = DATA_PREFIX_OPCODE;
2244 if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */
2245 prefix = ADDR_PREFIX_OPCODE;
2246
2247 if (! add_prefix (prefix))
2248 return;
2249 }
2250
2251 /* Set mode64 for an operand. */
2252 if (i.suffix == QWORD_MNEM_SUFFIX
2253 && !(i.tm.opcode_modifier & NoRex64))
2254 {
2255 i.rex.mode64 = 1;
2256 if (flag_code < CODE_64BIT)
2257 {
2258 as_bad (_("64bit operations available only in 64bit modes."));
2259 return;
2260 }
2261 }
2262
2263 /* Size floating point instruction. */
2264 if (i.suffix == LONG_MNEM_SUFFIX)
2265 {
2266 if (i.tm.opcode_modifier & FloatMF)
2267 i.tm.base_opcode ^= 4;
2268 }
2269 }
2270
2271 if (i.tm.opcode_modifier & ImmExt)
2272 {
2273 /* These AMD 3DNow! and Intel Katmai New Instructions have an
2274 opcode suffix which is coded in the same place as an 8-bit
2275 immediate field would be. Here we fake an 8-bit immediate
2276 operand from the opcode suffix stored in tm.extension_opcode. */
2277
2278 expressionS *exp;
2279
2280 assert (i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS);
2281
2282 exp = &im_expressions[i.imm_operands++];
2283 i.op[i.operands].imms = exp;
2284 i.types[i.operands++] = Imm8;
2285 exp->X_op = O_constant;
2286 exp->X_add_number = i.tm.extension_opcode;
2287 i.tm.extension_opcode = None;
2288 }
2289
2290 /* For insns with operands there are more diddles to do to the opcode. */
2291 if (i.operands)
2292 {
2293 /* Default segment register this instruction will use
2294 for memory accesses. 0 means unknown.
2295 This is only for optimizing out unnecessary segment overrides. */
2296 const seg_entry *default_seg = 0;
2297
2298 /* The imul $imm, %reg instruction is converted into
2299 imul $imm, %reg, %reg, and the clr %reg instruction
2300 is converted into xor %reg, %reg. */
2301 if (i.tm.opcode_modifier & regKludge)
2302 {
2303 unsigned int first_reg_op = (i.types[0] & Reg) ? 0 : 1;
2304 /* Pretend we saw the extra register operand. */
2305 assert (i.op[first_reg_op + 1].regs == 0);
2306 i.op[first_reg_op + 1].regs = i.op[first_reg_op].regs;
2307 i.types[first_reg_op + 1] = i.types[first_reg_op];
2308 i.reg_operands = 2;
2309 }
2310
2311 if (i.tm.opcode_modifier & ShortForm)
2312 {
2313 /* The register or float register operand is in operand 0 or 1. */
2314 unsigned int op = (i.types[0] & (Reg | FloatReg)) ? 0 : 1;
2315 /* Register goes in low 3 bits of opcode. */
2316 i.tm.base_opcode |= i.op[op].regs->reg_num;
2317 if (i.op[op].regs->reg_flags & RegRex)
2318 i.rex.extZ = 1;
2319 if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
2320 {
2321 /* Warn about some common errors, but press on regardless.
2322 The first case can be generated by gcc (<= 2.8.1). */
2323 if (i.operands == 2)
2324 {
2325 /* Reversed arguments on faddp, fsubp, etc. */
2326 as_warn (_("translating to `%s %%%s,%%%s'"), i.tm.name,
2327 i.op[1].regs->reg_name,
2328 i.op[0].regs->reg_name);
2329 }
2330 else
2331 {
2332 /* Extraneous `l' suffix on fp insn. */
2333 as_warn (_("translating to `%s %%%s'"), i.tm.name,
2334 i.op[0].regs->reg_name);
2335 }
2336 }
2337 }
2338 else if (i.tm.opcode_modifier & Modrm)
2339 {
2340 /* The opcode is completed (modulo i.tm.extension_opcode which
2341 must be put into the modrm byte).
2342 Now, we make the modrm & index base bytes based on all the
2343 info we've collected. */
2344
2345 /* i.reg_operands MUST be the number of real register operands;
2346 implicit registers do not count. */
2347 if (i.reg_operands == 2)
2348 {
2349 unsigned int source, dest;
2350 source = ((i.types[0]
2351 & (Reg | RegMMX | RegXMM
2352 | SReg2 | SReg3
2353 | Control | Debug | Test))
2354 ? 0 : 1);
2355 dest = source + 1;
2356
2357 i.rm.mode = 3;
2358 /* One of the register operands will be encoded in the
2359 i.tm.reg field, the other in the combined i.tm.mode
2360 and i.tm.regmem fields. If no form of this
2361 instruction supports a memory destination operand,
2362 then we assume the source operand may sometimes be
2363 a memory operand and so we need to store the
2364 destination in the i.rm.reg field. */
2365 if ((i.tm.operand_types[dest] & AnyMem) == 0)
2366 {
2367 i.rm.reg = i.op[dest].regs->reg_num;
2368 i.rm.regmem = i.op[source].regs->reg_num;
2369 if (i.op[dest].regs->reg_flags & RegRex)
2370 i.rex.extX = 1;
2371 if (i.op[source].regs->reg_flags & RegRex)
2372 i.rex.extZ = 1;
2373 }
2374 else
2375 {
2376 i.rm.reg = i.op[source].regs->reg_num;
2377 i.rm.regmem = i.op[dest].regs->reg_num;
2378 if (i.op[dest].regs->reg_flags & RegRex)
2379 i.rex.extZ = 1;
2380 if (i.op[source].regs->reg_flags & RegRex)
2381 i.rex.extX = 1;
2382 }
2383 }
2384 else
2385 { /* If it's not 2 reg operands... */
2386 if (i.mem_operands)
2387 {
2388 unsigned int fake_zero_displacement = 0;
2389 unsigned int op = ((i.types[0] & AnyMem)
2390 ? 0
2391 : (i.types[1] & AnyMem) ? 1 : 2);
2392
2393 default_seg = &ds;
2394
2395 if (! i.base_reg)
2396 {
2397 i.rm.mode = 0;
2398 if (! i.disp_operands)
2399 fake_zero_displacement = 1;
2400 if (! i.index_reg)
2401 {
2402 /* Operand is just <disp> */
2403 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
2404 {
2405 i.rm.regmem = NO_BASE_REGISTER_16;
2406 i.types[op] &= ~Disp;
2407 i.types[op] |= Disp16;
2408 }
2409 else if (flag_code != CODE_64BIT)
2410 {
2411 i.rm.regmem = NO_BASE_REGISTER;
2412 i.types[op] &= ~Disp;
2413 i.types[op] |= Disp32;
2414 }
2415 else
2416 {
2417 /* 64bit mode overwrites the 32bit
2418 absolute addressing by RIP relative
2419 addressing and absolute addressing
2420 is encoded by one of the redundant
2421 SIB forms. */
2422
2423 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2424 i.sib.base = NO_BASE_REGISTER;
2425 i.sib.index = NO_INDEX_REGISTER;
2426 i.types[op] &= ~Disp;
2427 i.types[op] |= Disp32S;
2428 }
2429 }
2430 else /* ! i.base_reg && i.index_reg */
2431 {
2432 i.sib.index = i.index_reg->reg_num;
2433 i.sib.base = NO_BASE_REGISTER;
2434 i.sib.scale = i.log2_scale_factor;
2435 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2436 i.types[op] &= ~Disp;
2437 if (flag_code != CODE_64BIT)
2438 i.types[op] |= Disp32; /* Must be 32 bit */
2439 else
2440 i.types[op] |= Disp32S;
2441 if (i.index_reg->reg_flags & RegRex)
2442 i.rex.extY = 1;
2443 }
2444 }
2445 /* RIP addressing for 64bit mode. */
2446 else if (i.base_reg->reg_type == BaseIndex)
2447 {
2448 i.rm.regmem = NO_BASE_REGISTER;
2449 i.types[op] &= ~Disp;
2450 i.types[op] |= Disp32S;
2451 i.flags[op] = Operand_PCrel;
2452 }
2453 else if (i.base_reg->reg_type & Reg16)
2454 {
2455 switch (i.base_reg->reg_num)
2456 {
2457 case 3: /* (%bx) */
2458 if (! i.index_reg)
2459 i.rm.regmem = 7;
2460 else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */
2461 i.rm.regmem = i.index_reg->reg_num - 6;
2462 break;
2463 case 5: /* (%bp) */
2464 default_seg = &ss;
2465 if (! i.index_reg)
2466 {
2467 i.rm.regmem = 6;
2468 if ((i.types[op] & Disp) == 0)
2469 {
2470 /* fake (%bp) into 0(%bp) */
2471 i.types[op] |= Disp8;
2472 fake_zero_displacement = 1;
2473 }
2474 }
2475 else /* (%bp,%si) -> 2, or (%bp,%di) -> 3 */
2476 i.rm.regmem = i.index_reg->reg_num - 6 + 2;
2477 break;
2478 default: /* (%si) -> 4 or (%di) -> 5 */
2479 i.rm.regmem = i.base_reg->reg_num - 6 + 4;
2480 }
2481 i.rm.mode = mode_from_disp_size (i.types[op]);
2482 }
2483 else /* i.base_reg and 32/64 bit mode */
2484 {
2485 if (flag_code == CODE_64BIT
2486 && (i.types[op] & Disp))
2487 {
2488 if (i.types[op] & Disp8)
2489 i.types[op] = Disp8 | Disp32S;
2490 else
2491 i.types[op] = Disp32S;
2492 }
2493 i.rm.regmem = i.base_reg->reg_num;
2494 if (i.base_reg->reg_flags & RegRex)
2495 i.rex.extZ = 1;
2496 i.sib.base = i.base_reg->reg_num;
2497 /* x86-64 ignores REX prefix bit here to avoid
2498 decoder complications. */
2499 if ((i.base_reg->reg_num & 7) == EBP_REG_NUM)
2500 {
2501 default_seg = &ss;
2502 if (i.disp_operands == 0)
2503 {
2504 fake_zero_displacement = 1;
2505 i.types[op] |= Disp8;
2506 }
2507 }
2508 else if (i.base_reg->reg_num == ESP_REG_NUM)
2509 {
2510 default_seg = &ss;
2511 }
2512 i.sib.scale = i.log2_scale_factor;
2513 if (! i.index_reg)
2514 {
2515 /* <disp>(%esp) becomes two byte modrm
2516 with no index register. We've already
2517 stored the code for esp in i.rm.regmem
2518 ie. ESCAPE_TO_TWO_BYTE_ADDRESSING. Any
2519 base register besides %esp will not use
2520 the extra modrm byte. */
2521 i.sib.index = NO_INDEX_REGISTER;
2522 #if ! SCALE1_WHEN_NO_INDEX
2523 /* Another case where we force the second
2524 modrm byte. */
2525 if (i.log2_scale_factor)
2526 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2527 #endif
2528 }
2529 else
2530 {
2531 i.sib.index = i.index_reg->reg_num;
2532 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2533 if (i.index_reg->reg_flags & RegRex)
2534 i.rex.extY = 1;
2535 }
2536 i.rm.mode = mode_from_disp_size (i.types[op]);
2537 }
2538
2539 if (fake_zero_displacement)
2540 {
2541 /* Fakes a zero displacement assuming that i.types[op]
2542 holds the correct displacement size. */
2543 expressionS *exp;
2544
2545 assert (i.op[op].disps == 0);
2546 exp = &disp_expressions[i.disp_operands++];
2547 i.op[op].disps = exp;
2548 exp->X_op = O_constant;
2549 exp->X_add_number = 0;
2550 exp->X_add_symbol = (symbolS *) 0;
2551 exp->X_op_symbol = (symbolS *) 0;
2552 }
2553 }
2554
2555 /* Fill in i.rm.reg or i.rm.regmem field with register
2556 operand (if any) based on i.tm.extension_opcode.
2557 Again, we must be careful to make sure that
2558 segment/control/debug/test/MMX registers are coded
2559 into the i.rm.reg field. */
2560 if (i.reg_operands)
2561 {
2562 unsigned int op =
2563 ((i.types[0]
2564 & (Reg | RegMMX | RegXMM
2565 | SReg2 | SReg3
2566 | Control | Debug | Test))
2567 ? 0
2568 : ((i.types[1]
2569 & (Reg | RegMMX | RegXMM
2570 | SReg2 | SReg3
2571 | Control | Debug | Test))
2572 ? 1
2573 : 2));
2574 /* If there is an extension opcode to put here, the
2575 register number must be put into the regmem field. */
2576 if (i.tm.extension_opcode != None)
2577 {
2578 i.rm.regmem = i.op[op].regs->reg_num;
2579 if (i.op[op].regs->reg_flags & RegRex)
2580 i.rex.extZ = 1;
2581 }
2582 else
2583 {
2584 i.rm.reg = i.op[op].regs->reg_num;
2585 if (i.op[op].regs->reg_flags & RegRex)
2586 i.rex.extX = 1;
2587 }
2588
2589 /* Now, if no memory operand has set i.rm.mode = 0, 1, 2
2590 we must set it to 3 to indicate this is a register
2591 operand in the regmem field. */
2592 if (!i.mem_operands)
2593 i.rm.mode = 3;
2594 }
2595
2596 /* Fill in i.rm.reg field with extension opcode (if any). */
2597 if (i.tm.extension_opcode != None)
2598 i.rm.reg = i.tm.extension_opcode;
2599 }
2600 }
2601 else if (i.tm.opcode_modifier & (Seg2ShortForm | Seg3ShortForm))
2602 {
2603 if (i.tm.base_opcode == POP_SEG_SHORT
2604 && i.op[0].regs->reg_num == 1)
2605 {
2606 as_bad (_("you can't `pop %%cs'"));
2607 return;
2608 }
2609 i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
2610 if (i.op[0].regs->reg_flags & RegRex)
2611 i.rex.extZ = 1;
2612 }
2613 else if ((i.tm.base_opcode & ~(D|W)) == MOV_AX_DISP32)
2614 {
2615 default_seg = &ds;
2616 }
2617 else if ((i.tm.opcode_modifier & IsString) != 0)
2618 {
2619 /* For the string instructions that allow a segment override
2620 on one of their operands, the default segment is ds. */
2621 default_seg = &ds;
2622 }
2623
2624 /* If a segment was explicitly specified,
2625 and the specified segment is not the default,
2626 use an opcode prefix to select it.
2627 If we never figured out what the default segment is,
2628 then default_seg will be zero at this point,
2629 and the specified segment prefix will always be used. */
2630 if ((i.seg[0]) && (i.seg[0] != default_seg))
2631 {
2632 if (! add_prefix (i.seg[0]->seg_prefix))
2633 return;
2634 }
2635 }
2636 else if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
2637 {
2638 /* UnixWare fsub no args is alias for fsubp, fadd -> faddp, etc. */
2639 as_warn (_("translating to `%sp'"), i.tm.name);
2640 }
2641 }
2642
2643 /* Handle conversion of 'int $3' --> special int3 insn. */
2644 if (i.tm.base_opcode == INT_OPCODE && i.op[0].imms->X_add_number == 3)
2645 {
2646 i.tm.base_opcode = INT3_OPCODE;
2647 i.imm_operands = 0;
2648 }
2649
2650 if ((i.tm.opcode_modifier & (Jump | JumpByte | JumpDword))
2651 && i.op[0].disps->X_op == O_constant)
2652 {
2653 /* Convert "jmp constant" (and "call constant") to a jump (call) to
2654 the absolute address given by the constant. Since ix86 jumps and
2655 calls are pc relative, we need to generate a reloc. */
2656 i.op[0].disps->X_add_symbol = &abs_symbol;
2657 i.op[0].disps->X_op = O_symbol;
2658 }
2659
2660 if (i.tm.opcode_modifier & Rex64)
2661 i.rex.mode64 = 1;
2662
2663 /* For 8bit registers we would need an empty rex prefix.
2664 Also in the case instruction is already having prefix,
2665 we need to convert old registers to new ones. */
2666
2667 if (((i.types[0] & Reg8) && (i.op[0].regs->reg_flags & RegRex64))
2668 || ((i.types[1] & Reg8) && (i.op[1].regs->reg_flags & RegRex64))
2669 || ((i.rex.mode64 || i.rex.extX || i.rex.extY || i.rex.extZ || i.rex.empty)
2670 && ((i.types[0] & Reg8) || (i.types[1] & Reg8))))
2671 {
2672 int x;
2673 i.rex.empty = 1;
2674 for (x = 0; x < 2; x++)
2675 {
2676 /* Look for 8bit operand that does use old registers. */
2677 if (i.types[x] & Reg8
2678 && !(i.op[x].regs->reg_flags & RegRex64))
2679 {
2680 /* In case it is "hi" register, give up. */
2681 if (i.op[x].regs->reg_num > 3)
2682 as_bad (_("Can't encode registers '%%%s' in the instruction requiring REX prefix.\n"),
2683 i.op[x].regs->reg_name);
2684
2685 /* Otherwise it is equivalent to the extended register.
2686 Since the encoding don't change this is merely cosmetical
2687 cleanup for debug output. */
2688
2689 i.op[x].regs = i.op[x].regs + 8;
2690 }
2691 }
2692 }
2693
2694 if (i.rex.mode64 || i.rex.extX || i.rex.extY || i.rex.extZ || i.rex.empty)
2695 add_prefix (0x40
2696 | (i.rex.mode64 ? 8 : 0)
2697 | (i.rex.extX ? 4 : 0)
2698 | (i.rex.extY ? 2 : 0)
2699 | (i.rex.extZ ? 1 : 0));
2700
2701 /* We are ready to output the insn. */
2702 {
2703 register char *p;
2704
2705 /* Output jumps. */
2706 if (i.tm.opcode_modifier & Jump)
2707 {
2708 int code16;
2709 int prefix;
2710
2711 code16 = 0;
2712 if (flag_code == CODE_16BIT)
2713 code16 = CODE16;
2714
2715 prefix = 0;
2716 if (i.prefix[DATA_PREFIX])
2717 {
2718 prefix = 1;
2719 i.prefixes -= 1;
2720 code16 ^= CODE16;
2721 }
2722 if (i.prefix[REX_PREFIX])
2723 {
2724 prefix++;
2725 i.prefixes--;
2726 }
2727
2728 if (i.prefixes != 0 && !intel_syntax)
2729 as_warn (_("skipping prefixes on this instruction"));
2730
2731 /* It's always a symbol; End frag & setup for relax.
2732 Make sure there is enough room in this frag for the largest
2733 instruction we may generate in md_convert_frag. This is 2
2734 bytes for the opcode and room for the prefix and largest
2735 displacement. */
2736 frag_grow (prefix + 2 + 4);
2737 insn_size += prefix + 1;
2738 /* Prefix and 1 opcode byte go in fr_fix. */
2739 p = frag_more (prefix + 1);
2740 if (i.prefix[DATA_PREFIX])
2741 *p++ = DATA_PREFIX_OPCODE;
2742 if (i.prefix[REX_PREFIX])
2743 *p++ = i.prefix[REX_PREFIX];
2744 *p = i.tm.base_opcode;
2745 /* 1 possible extra opcode + displacement go in var part.
2746 Pass reloc in fr_var. */
2747 frag_var (rs_machine_dependent,
2748 1 + 4,
2749 i.reloc[0],
2750 ((unsigned char) *p == JUMP_PC_RELATIVE
2751 ? ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL) | code16
2752 : ((cpu_arch_flags & Cpu386) != 0
2753 ? ENCODE_RELAX_STATE (COND_JUMP, SMALL) | code16
2754 : ENCODE_RELAX_STATE (COND_JUMP86, SMALL) | code16)),
2755 i.op[0].disps->X_add_symbol,
2756 i.op[0].disps->X_add_number,
2757 p);
2758 }
2759 else if (i.tm.opcode_modifier & (JumpByte | JumpDword))
2760 {
2761 int size;
2762
2763 if (i.tm.opcode_modifier & JumpByte)
2764 {
2765 /* This is a loop or jecxz type instruction. */
2766 size = 1;
2767 if (i.prefix[ADDR_PREFIX])
2768 {
2769 insn_size += 1;
2770 FRAG_APPEND_1_CHAR (ADDR_PREFIX_OPCODE);
2771 i.prefixes -= 1;
2772 }
2773 }
2774 else
2775 {
2776 int code16;
2777
2778 code16 = 0;
2779 if (flag_code == CODE_16BIT)
2780 code16 = CODE16;
2781
2782 if (i.prefix[DATA_PREFIX])
2783 {
2784 insn_size += 1;
2785 FRAG_APPEND_1_CHAR (DATA_PREFIX_OPCODE);
2786 i.prefixes -= 1;
2787 code16 ^= CODE16;
2788 }
2789
2790 size = 4;
2791 if (code16)
2792 size = 2;
2793 }
2794
2795 if (i.prefix[REX_PREFIX])
2796 {
2797 FRAG_APPEND_1_CHAR (i.prefix[REX_PREFIX]);
2798 insn_size++;
2799 i.prefixes -= 1;
2800 }
2801
2802 if (i.prefixes != 0 && !intel_syntax)
2803 as_warn (_("skipping prefixes on this instruction"));
2804
2805 if (fits_in_unsigned_byte (i.tm.base_opcode))
2806 {
2807 insn_size += 1 + size;
2808 p = frag_more (1 + size);
2809 }
2810 else
2811 {
2812 /* Opcode can be at most two bytes. */
2813 insn_size += 2 + size;
2814 p = frag_more (2 + size);
2815 *p++ = (i.tm.base_opcode >> 8) & 0xff;
2816 }
2817 *p++ = i.tm.base_opcode & 0xff;
2818
2819 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
2820 i.op[0].disps, 1, reloc (size, 1, 1, i.reloc[0]));
2821 }
2822 else if (i.tm.opcode_modifier & JumpInterSegment)
2823 {
2824 int size;
2825 int prefix;
2826 int code16;
2827
2828 code16 = 0;
2829 if (flag_code == CODE_16BIT)
2830 code16 = CODE16;
2831
2832 prefix = 0;
2833 if (i.prefix[DATA_PREFIX])
2834 {
2835 prefix = 1;
2836 i.prefixes -= 1;
2837 code16 ^= CODE16;
2838 }
2839 if (i.prefix[REX_PREFIX])
2840 {
2841 prefix++;
2842 i.prefixes -= 1;
2843 }
2844
2845 size = 4;
2846 if (code16)
2847 size = 2;
2848
2849 if (i.prefixes != 0 && !intel_syntax)
2850 as_warn (_("skipping prefixes on this instruction"));
2851
2852 /* 1 opcode; 2 segment; offset */
2853 insn_size += prefix + 1 + 2 + size;
2854 p = frag_more (prefix + 1 + 2 + size);
2855
2856 if (i.prefix[DATA_PREFIX])
2857 *p++ = DATA_PREFIX_OPCODE;
2858
2859 if (i.prefix[REX_PREFIX])
2860 *p++ = i.prefix[REX_PREFIX];
2861
2862 *p++ = i.tm.base_opcode;
2863 if (i.op[1].imms->X_op == O_constant)
2864 {
2865 offsetT n = i.op[1].imms->X_add_number;
2866
2867 if (size == 2
2868 && !fits_in_unsigned_word (n)
2869 && !fits_in_signed_word (n))
2870 {
2871 as_bad (_("16-bit jump out of range"));
2872 return;
2873 }
2874 md_number_to_chars (p, n, size);
2875 }
2876 else
2877 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
2878 i.op[1].imms, 0, reloc (size, 0, 0, i.reloc[1]));
2879 if (i.op[0].imms->X_op != O_constant)
2880 as_bad (_("can't handle non absolute segment in `%s'"),
2881 i.tm.name);
2882 md_number_to_chars (p + size, (valueT) i.op[0].imms->X_add_number, 2);
2883 }
2884 else
2885 {
2886 /* Output normal instructions here. */
2887 unsigned char *q;
2888
2889 /* All opcodes on i386 have eighter 1 or 2 bytes. We may use third
2890 byte for the SSE instructions to specify prefix they require. */
2891 if (i.tm.base_opcode & 0xff0000)
2892 add_prefix ((i.tm.base_opcode >> 16) & 0xff);
2893
2894 /* The prefix bytes. */
2895 for (q = i.prefix;
2896 q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
2897 q++)
2898 {
2899 if (*q)
2900 {
2901 insn_size += 1;
2902 p = frag_more (1);
2903 md_number_to_chars (p, (valueT) *q, 1);
2904 }
2905 }
2906
2907 /* Now the opcode; be careful about word order here! */
2908 if (fits_in_unsigned_byte (i.tm.base_opcode))
2909 {
2910 insn_size += 1;
2911 FRAG_APPEND_1_CHAR (i.tm.base_opcode);
2912 }
2913 else
2914 {
2915 insn_size += 2;
2916 p = frag_more (2);
2917 /* Put out high byte first: can't use md_number_to_chars! */
2918 *p++ = (i.tm.base_opcode >> 8) & 0xff;
2919 *p = i.tm.base_opcode & 0xff;
2920 }
2921
2922 /* Now the modrm byte and sib byte (if present). */
2923 if (i.tm.opcode_modifier & Modrm)
2924 {
2925 insn_size += 1;
2926 p = frag_more (1);
2927 md_number_to_chars (p,
2928 (valueT) (i.rm.regmem << 0
2929 | i.rm.reg << 3
2930 | i.rm.mode << 6),
2931 1);
2932 /* If i.rm.regmem == ESP (4)
2933 && i.rm.mode != (Register mode)
2934 && not 16 bit
2935 ==> need second modrm byte. */
2936 if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING
2937 && i.rm.mode != 3
2938 && !(i.base_reg && (i.base_reg->reg_type & Reg16) != 0))
2939 {
2940 insn_size += 1;
2941 p = frag_more (1);
2942 md_number_to_chars (p,
2943 (valueT) (i.sib.base << 0
2944 | i.sib.index << 3
2945 | i.sib.scale << 6),
2946 1);
2947 }
2948 }
2949
2950 if (i.disp_operands)
2951 {
2952 register unsigned int n;
2953
2954 for (n = 0; n < i.operands; n++)
2955 {
2956 if (i.types[n] & Disp)
2957 {
2958 if (i.op[n].disps->X_op == O_constant)
2959 {
2960 int size;
2961 offsetT val;
2962
2963 size = 4;
2964 if (i.types[n] & (Disp8 | Disp16 | Disp64))
2965 {
2966 size = 2;
2967 if (i.types[n] & Disp8)
2968 size = 1;
2969 if (i.types[n] & Disp64)
2970 size = 8;
2971 }
2972 val = offset_in_range (i.op[n].disps->X_add_number,
2973 size);
2974 insn_size += size;
2975 p = frag_more (size);
2976 md_number_to_chars (p, val, size);
2977 }
2978 else
2979 {
2980 int size = 4;
2981 int sign = 0;
2982 int pcrel = (i.flags[n] & Operand_PCrel) != 0;
2983
2984 /* The PC relative address is computed relative
2985 to the instruction boundary, so in case immediate
2986 fields follows, we need to adjust the value. */
2987 if (pcrel && i.imm_operands)
2988 {
2989 int imm_size = 4;
2990 register unsigned int n1;
2991
2992 for (n1 = 0; n1 < i.operands; n1++)
2993 if (i.types[n1] & Imm)
2994 {
2995 if (i.types[n1] & (Imm8 | Imm8S | Imm16 | Imm64))
2996 {
2997 imm_size = 2;
2998 if (i.types[n1] & (Imm8 | Imm8S))
2999 imm_size = 1;
3000 if (i.types[n1] & Imm64)
3001 imm_size = 8;
3002 }
3003 break;
3004 }
3005 /* We should find the immediate. */
3006 if (n1 == i.operands)
3007 abort ();
3008 i.op[n].disps->X_add_number -= imm_size;
3009 }
3010
3011 if (i.types[n] & Disp32S)
3012 sign = 1;
3013
3014 if (i.types[n] & (Disp16 | Disp64))
3015 {
3016 size = 2;
3017 if (i.types[n] & Disp64)
3018 size = 8;
3019 }
3020
3021 insn_size += size;
3022 p = frag_more (size);
3023 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3024 i.op[n].disps, pcrel,
3025 reloc (size, pcrel, sign, i.reloc[n]));
3026 }
3027 }
3028 }
3029 }
3030
3031 /* Output immediate. */
3032 if (i.imm_operands)
3033 {
3034 register unsigned int n;
3035
3036 for (n = 0; n < i.operands; n++)
3037 {
3038 if (i.types[n] & Imm)
3039 {
3040 if (i.op[n].imms->X_op == O_constant)
3041 {
3042 int size;
3043 offsetT val;
3044
3045 size = 4;
3046 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3047 {
3048 size = 2;
3049 if (i.types[n] & (Imm8 | Imm8S))
3050 size = 1;
3051 else if (i.types[n] & Imm64)
3052 size = 8;
3053 }
3054 val = offset_in_range (i.op[n].imms->X_add_number,
3055 size);
3056 insn_size += size;
3057 p = frag_more (size);
3058 md_number_to_chars (p, val, size);
3059 }
3060 else
3061 {
3062 /* Not absolute_section.
3063 Need a 32-bit fixup (don't support 8bit
3064 non-absolute imms). Try to support other
3065 sizes ... */
3066 RELOC_ENUM reloc_type;
3067 int size = 4;
3068 int sign = 0;
3069
3070 if ((i.types[n] & (Imm32S))
3071 && i.suffix == QWORD_MNEM_SUFFIX)
3072 sign = 1;
3073 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3074 {
3075 size = 2;
3076 if (i.types[n] & (Imm8 | Imm8S))
3077 size = 1;
3078 if (i.types[n] & Imm64)
3079 size = 8;
3080 }
3081
3082 insn_size += size;
3083 p = frag_more (size);
3084 reloc_type = reloc (size, 0, sign, i.reloc[n]);
3085 #ifdef BFD_ASSEMBLER
3086 if (reloc_type == BFD_RELOC_32
3087 && GOT_symbol
3088 && GOT_symbol == i.op[n].imms->X_add_symbol
3089 && (i.op[n].imms->X_op == O_symbol
3090 || (i.op[n].imms->X_op == O_add
3091 && ((symbol_get_value_expression
3092 (i.op[n].imms->X_op_symbol)->X_op)
3093 == O_subtract))))
3094 {
3095 /* We don't support dynamic linking on x86-64 yet. */
3096 if (flag_code == CODE_64BIT)
3097 abort ();
3098 reloc_type = BFD_RELOC_386_GOTPC;
3099 i.op[n].imms->X_add_number += 3;
3100 }
3101 #endif
3102 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3103 i.op[n].imms, 0, reloc_type);
3104 }
3105 }
3106 }
3107 }
3108 }
3109
3110 dwarf2_emit_insn (insn_size);
3111
3112 #ifdef DEBUG386
3113 if (flag_debug)
3114 {
3115 pi (line, &i);
3116 }
3117 #endif /* DEBUG386 */
3118 }
3119 }
3120 \f
3121 #ifndef LEX_AT
3122 static char *lex_got PARAMS ((RELOC_ENUM *, int *));
3123
3124 /* Parse operands of the form
3125 <symbol>@GOTOFF+<nnn>
3126 and similar .plt or .got references.
3127
3128 If we find one, set up the correct relocation in RELOC and copy the
3129 input string, minus the `@GOTOFF' into a malloc'd buffer for
3130 parsing by the calling routine. Return this buffer, and if ADJUST
3131 is non-null set it to the length of the string we removed from the
3132 input line. Otherwise return NULL. */
3133 static char *
3134 lex_got (reloc, adjust)
3135 RELOC_ENUM *reloc;
3136 int *adjust;
3137 {
3138 static const char * const mode_name[NUM_FLAG_CODE] = { "32", "16", "64" };
3139 static const struct {
3140 const char *str;
3141 const RELOC_ENUM rel[NUM_FLAG_CODE];
3142 } gotrel[] = {
3143 { "PLT", { BFD_RELOC_386_PLT32, 0, BFD_RELOC_X86_64_PLT32 } },
3144 { "GOTOFF", { BFD_RELOC_386_GOTOFF, 0, 0 } },
3145 { "GOTPCREL", { 0, 0, BFD_RELOC_X86_64_GOTPCREL } },
3146 { "GOT", { BFD_RELOC_386_GOT32, 0, BFD_RELOC_X86_64_GOT32 } }
3147 };
3148 char *cp;
3149 unsigned int j;
3150
3151 for (cp = input_line_pointer; *cp != '@'; cp++)
3152 if (is_end_of_line[(unsigned char) *cp])
3153 return NULL;
3154
3155 for (j = 0; j < sizeof (gotrel) / sizeof (gotrel[0]); j++)
3156 {
3157 int len;
3158
3159 len = strlen (gotrel[j].str);
3160 if (strncmp (cp + 1, gotrel[j].str, len) == 0)
3161 {
3162 if (gotrel[j].rel[(unsigned int) flag_code] != 0)
3163 {
3164 int first;
3165 char *tmpbuf;
3166
3167 *reloc = gotrel[j].rel[(unsigned int) flag_code];
3168
3169 if (GOT_symbol == NULL)
3170 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
3171
3172 /* Replace the relocation token with ' ', so that
3173 errors like foo@GOTOFF1 will be detected. */
3174 first = cp - input_line_pointer;
3175 tmpbuf = xmalloc (strlen (input_line_pointer));
3176 memcpy (tmpbuf, input_line_pointer, first);
3177 tmpbuf[first] = ' ';
3178 strcpy (tmpbuf + first + 1, cp + 1 + len);
3179 if (adjust)
3180 *adjust = len;
3181 return tmpbuf;
3182 }
3183
3184 as_bad (_("@%s reloc is not supported in %s bit mode"),
3185 gotrel[j].str, mode_name[(unsigned int) flag_code]);
3186 return NULL;
3187 }
3188 }
3189
3190 /* Might be a symbol version string. Don't as_bad here. */
3191 return NULL;
3192 }
3193
3194 /* x86_cons_fix_new is called via the expression parsing code when a
3195 reloc is needed. We use this hook to get the correct .got reloc. */
3196 static RELOC_ENUM got_reloc = NO_RELOC;
3197
3198 void
3199 x86_cons_fix_new (frag, off, len, exp)
3200 fragS *frag;
3201 unsigned int off;
3202 unsigned int len;
3203 expressionS *exp;
3204 {
3205 RELOC_ENUM r = reloc (len, 0, 0, got_reloc);
3206 got_reloc = NO_RELOC;
3207 fix_new_exp (frag, off, len, exp, 0, r);
3208 }
3209
3210 void
3211 x86_cons (exp, size)
3212 expressionS *exp;
3213 int size;
3214 {
3215 if (size == 4)
3216 {
3217 /* Handle @GOTOFF and the like in an expression. */
3218 char *save;
3219 char *gotfree_input_line;
3220 int adjust;
3221
3222 save = input_line_pointer;
3223 gotfree_input_line = lex_got (&got_reloc, &adjust);
3224 if (gotfree_input_line)
3225 input_line_pointer = gotfree_input_line;
3226
3227 expression (exp);
3228
3229 if (gotfree_input_line)
3230 {
3231 /* expression () has merrily parsed up to the end of line,
3232 or a comma - in the wrong buffer. Transfer how far
3233 input_line_pointer has moved to the right buffer. */
3234 input_line_pointer = (save
3235 + (input_line_pointer - gotfree_input_line)
3236 + adjust);
3237 free (gotfree_input_line);
3238 }
3239 }
3240 else
3241 expression (exp);
3242 }
3243 #endif
3244
3245 static int i386_immediate PARAMS ((char *));
3246
3247 static int
3248 i386_immediate (imm_start)
3249 char *imm_start;
3250 {
3251 char *save_input_line_pointer;
3252 #ifndef LEX_AT
3253 char *gotfree_input_line;
3254 #endif
3255 segT exp_seg = 0;
3256 expressionS *exp;
3257
3258 if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
3259 {
3260 as_bad (_("only 1 or 2 immediate operands are allowed"));
3261 return 0;
3262 }
3263
3264 exp = &im_expressions[i.imm_operands++];
3265 i.op[this_operand].imms = exp;
3266
3267 if (is_space_char (*imm_start))
3268 ++imm_start;
3269
3270 save_input_line_pointer = input_line_pointer;
3271 input_line_pointer = imm_start;
3272
3273 #ifndef LEX_AT
3274 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
3275 if (gotfree_input_line)
3276 input_line_pointer = gotfree_input_line;
3277 #endif
3278
3279 exp_seg = expression (exp);
3280
3281 SKIP_WHITESPACE ();
3282 if (*input_line_pointer)
3283 as_bad (_("junk `%s' after expression"), input_line_pointer);
3284
3285 input_line_pointer = save_input_line_pointer;
3286 #ifndef LEX_AT
3287 if (gotfree_input_line)
3288 free (gotfree_input_line);
3289 #endif
3290
3291 if (exp->X_op == O_absent || exp->X_op == O_big)
3292 {
3293 /* Missing or bad expr becomes absolute 0. */
3294 as_bad (_("missing or invalid immediate expression `%s' taken as 0"),
3295 imm_start);
3296 exp->X_op = O_constant;
3297 exp->X_add_number = 0;
3298 exp->X_add_symbol = (symbolS *) 0;
3299 exp->X_op_symbol = (symbolS *) 0;
3300 }
3301 else if (exp->X_op == O_constant)
3302 {
3303 /* Size it properly later. */
3304 i.types[this_operand] |= Imm64;
3305 /* If BFD64, sign extend val. */
3306 if (!use_rela_relocations)
3307 if ((exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0)
3308 exp->X_add_number = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
3309 }
3310 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3311 else if (1
3312 #ifdef BFD_ASSEMBLER
3313 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3314 #endif
3315 && exp_seg != text_section
3316 && exp_seg != data_section
3317 && exp_seg != bss_section
3318 && exp_seg != undefined_section
3319 #ifdef BFD_ASSEMBLER
3320 && !bfd_is_com_section (exp_seg)
3321 #endif
3322 )
3323 {
3324 #ifdef BFD_ASSEMBLER
3325 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3326 #else
3327 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
3328 #endif
3329 return 0;
3330 }
3331 #endif
3332 else
3333 {
3334 /* This is an address. The size of the address will be
3335 determined later, depending on destination register,
3336 suffix, or the default for the section. */
3337 i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64;
3338 }
3339
3340 return 1;
3341 }
3342
3343 static char *i386_scale PARAMS ((char *));
3344
3345 static char *
3346 i386_scale (scale)
3347 char *scale;
3348 {
3349 offsetT val;
3350 char *save = input_line_pointer;
3351
3352 input_line_pointer = scale;
3353 val = get_absolute_expression ();
3354
3355 switch (val)
3356 {
3357 case 0:
3358 case 1:
3359 i.log2_scale_factor = 0;
3360 break;
3361 case 2:
3362 i.log2_scale_factor = 1;
3363 break;
3364 case 4:
3365 i.log2_scale_factor = 2;
3366 break;
3367 case 8:
3368 i.log2_scale_factor = 3;
3369 break;
3370 default:
3371 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
3372 scale);
3373 input_line_pointer = save;
3374 return NULL;
3375 }
3376 if (i.log2_scale_factor != 0 && ! i.index_reg)
3377 {
3378 as_warn (_("scale factor of %d without an index register"),
3379 1 << i.log2_scale_factor);
3380 #if SCALE1_WHEN_NO_INDEX
3381 i.log2_scale_factor = 0;
3382 #endif
3383 }
3384 scale = input_line_pointer;
3385 input_line_pointer = save;
3386 return scale;
3387 }
3388
3389 static int i386_displacement PARAMS ((char *, char *));
3390
3391 static int
3392 i386_displacement (disp_start, disp_end)
3393 char *disp_start;
3394 char *disp_end;
3395 {
3396 register expressionS *exp;
3397 segT exp_seg = 0;
3398 char *save_input_line_pointer;
3399 #ifndef LEX_AT
3400 char *gotfree_input_line;
3401 #endif
3402 int bigdisp = Disp32;
3403
3404 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3405 bigdisp = Disp16;
3406 if (flag_code == CODE_64BIT)
3407 bigdisp = Disp64;
3408 i.types[this_operand] |= bigdisp;
3409
3410 exp = &disp_expressions[i.disp_operands];
3411 i.op[this_operand].disps = exp;
3412 i.disp_operands++;
3413 save_input_line_pointer = input_line_pointer;
3414 input_line_pointer = disp_start;
3415 END_STRING_AND_SAVE (disp_end);
3416
3417 #ifndef GCC_ASM_O_HACK
3418 #define GCC_ASM_O_HACK 0
3419 #endif
3420 #if GCC_ASM_O_HACK
3421 END_STRING_AND_SAVE (disp_end + 1);
3422 if ((i.types[this_operand] & BaseIndex) != 0
3423 && displacement_string_end[-1] == '+')
3424 {
3425 /* This hack is to avoid a warning when using the "o"
3426 constraint within gcc asm statements.
3427 For instance:
3428
3429 #define _set_tssldt_desc(n,addr,limit,type) \
3430 __asm__ __volatile__ ( \
3431 "movw %w2,%0\n\t" \
3432 "movw %w1,2+%0\n\t" \
3433 "rorl $16,%1\n\t" \
3434 "movb %b1,4+%0\n\t" \
3435 "movb %4,5+%0\n\t" \
3436 "movb $0,6+%0\n\t" \
3437 "movb %h1,7+%0\n\t" \
3438 "rorl $16,%1" \
3439 : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
3440
3441 This works great except that the output assembler ends
3442 up looking a bit weird if it turns out that there is
3443 no offset. You end up producing code that looks like:
3444
3445 #APP
3446 movw $235,(%eax)
3447 movw %dx,2+(%eax)
3448 rorl $16,%edx
3449 movb %dl,4+(%eax)
3450 movb $137,5+(%eax)
3451 movb $0,6+(%eax)
3452 movb %dh,7+(%eax)
3453 rorl $16,%edx
3454 #NO_APP
3455
3456 So here we provide the missing zero. */
3457
3458 *displacement_string_end = '0';
3459 }
3460 #endif
3461 #ifndef LEX_AT
3462 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
3463 if (gotfree_input_line)
3464 input_line_pointer = gotfree_input_line;
3465 #endif
3466
3467 exp_seg = expression (exp);
3468
3469 SKIP_WHITESPACE ();
3470 if (*input_line_pointer)
3471 as_bad (_("junk `%s' after expression"), input_line_pointer);
3472 #if GCC_ASM_O_HACK
3473 RESTORE_END_STRING (disp_end + 1);
3474 #endif
3475 RESTORE_END_STRING (disp_end);
3476 input_line_pointer = save_input_line_pointer;
3477 #ifndef LEX_AT
3478 if (gotfree_input_line)
3479 free (gotfree_input_line);
3480 #endif
3481
3482 #ifdef BFD_ASSEMBLER
3483 /* We do this to make sure that the section symbol is in
3484 the symbol table. We will ultimately change the relocation
3485 to be relative to the beginning of the section. */
3486 if (i.reloc[this_operand] == BFD_RELOC_386_GOTOFF
3487 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3488 {
3489 if (exp->X_op != O_symbol)
3490 {
3491 as_bad (_("bad expression used with @%s"),
3492 (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
3493 ? "GOTPCREL"
3494 : "GOTOFF"));
3495 return 0;
3496 }
3497
3498 if (S_IS_LOCAL (exp->X_add_symbol)
3499 && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
3500 section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
3501 exp->X_op = O_subtract;
3502 exp->X_op_symbol = GOT_symbol;
3503 if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3504 i.reloc[this_operand] = BFD_RELOC_32_PCREL;
3505 else
3506 i.reloc[this_operand] = BFD_RELOC_32;
3507 }
3508 #endif
3509
3510 if (exp->X_op == O_absent || exp->X_op == O_big)
3511 {
3512 /* Missing or bad expr becomes absolute 0. */
3513 as_bad (_("missing or invalid displacement expression `%s' taken as 0"),
3514 disp_start);
3515 exp->X_op = O_constant;
3516 exp->X_add_number = 0;
3517 exp->X_add_symbol = (symbolS *) 0;
3518 exp->X_op_symbol = (symbolS *) 0;
3519 }
3520
3521 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3522 if (exp->X_op != O_constant
3523 #ifdef BFD_ASSEMBLER
3524 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3525 #endif
3526 && exp_seg != text_section
3527 && exp_seg != data_section
3528 && exp_seg != bss_section
3529 && exp_seg != undefined_section)
3530 {
3531 #ifdef BFD_ASSEMBLER
3532 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3533 #else
3534 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
3535 #endif
3536 return 0;
3537 }
3538 #endif
3539 else if (flag_code == CODE_64BIT)
3540 i.types[this_operand] |= Disp32S | Disp32;
3541 return 1;
3542 }
3543
3544 static int i386_index_check PARAMS ((const char *));
3545
3546 /* Make sure the memory operand we've been dealt is valid.
3547 Return 1 on success, 0 on a failure. */
3548
3549 static int
3550 i386_index_check (operand_string)
3551 const char *operand_string;
3552 {
3553 int ok;
3554 #if INFER_ADDR_PREFIX
3555 int fudged = 0;
3556
3557 tryprefix:
3558 #endif
3559 ok = 1;
3560 if (flag_code == CODE_64BIT)
3561 {
3562 /* 64bit checks. */
3563 if ((i.base_reg
3564 && ((i.base_reg->reg_type & Reg64) == 0)
3565 && (i.base_reg->reg_type != BaseIndex
3566 || i.index_reg))
3567 || (i.index_reg
3568 && ((i.index_reg->reg_type & (Reg64|BaseIndex))
3569 != (Reg64|BaseIndex))))
3570 ok = 0;
3571 }
3572 else
3573 {
3574 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3575 {
3576 /* 16bit checks. */
3577 if ((i.base_reg
3578 && ((i.base_reg->reg_type & (Reg16|BaseIndex|RegRex))
3579 != (Reg16|BaseIndex)))
3580 || (i.index_reg
3581 && (((i.index_reg->reg_type & (Reg16|BaseIndex))
3582 != (Reg16|BaseIndex))
3583 || ! (i.base_reg
3584 && i.base_reg->reg_num < 6
3585 && i.index_reg->reg_num >= 6
3586 && i.log2_scale_factor == 0))))
3587 ok = 0;
3588 }
3589 else
3590 {
3591 /* 32bit checks. */
3592 if ((i.base_reg
3593 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
3594 || (i.index_reg
3595 && ((i.index_reg->reg_type & (Reg32|BaseIndex|RegRex))
3596 != (Reg32|BaseIndex))))
3597 ok = 0;
3598 }
3599 }
3600 if (!ok)
3601 {
3602 #if INFER_ADDR_PREFIX
3603 if (flag_code != CODE_64BIT
3604 && i.prefix[ADDR_PREFIX] == 0 && stackop_size != '\0')
3605 {
3606 i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
3607 i.prefixes += 1;
3608 /* Change the size of any displacement too. At most one of
3609 Disp16 or Disp32 is set.
3610 FIXME. There doesn't seem to be any real need for separate
3611 Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
3612 Removing them would probably clean up the code quite a lot. */
3613 if (i.types[this_operand] & (Disp16|Disp32))
3614 i.types[this_operand] ^= (Disp16|Disp32);
3615 fudged = 1;
3616 goto tryprefix;
3617 }
3618 if (fudged)
3619 as_bad (_("`%s' is not a valid base/index expression"),
3620 operand_string);
3621 else
3622 #endif
3623 as_bad (_("`%s' is not a valid %s bit base/index expression"),
3624 operand_string,
3625 flag_code_names[flag_code]);
3626 return 0;
3627 }
3628 return 1;
3629 }
3630
3631 /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
3632 on error. */
3633
3634 static int
3635 i386_operand (operand_string)
3636 char *operand_string;
3637 {
3638 const reg_entry *r;
3639 char *end_op;
3640 char *op_string = operand_string;
3641
3642 if (is_space_char (*op_string))
3643 ++op_string;
3644
3645 /* We check for an absolute prefix (differentiating,
3646 for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
3647 if (*op_string == ABSOLUTE_PREFIX)
3648 {
3649 ++op_string;
3650 if (is_space_char (*op_string))
3651 ++op_string;
3652 i.types[this_operand] |= JumpAbsolute;
3653 }
3654
3655 /* Check if operand is a register. */
3656 if ((*op_string == REGISTER_PREFIX || allow_naked_reg)
3657 && (r = parse_register (op_string, &end_op)) != NULL)
3658 {
3659 /* Check for a segment override by searching for ':' after a
3660 segment register. */
3661 op_string = end_op;
3662 if (is_space_char (*op_string))
3663 ++op_string;
3664 if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
3665 {
3666 switch (r->reg_num)
3667 {
3668 case 0:
3669 i.seg[i.mem_operands] = &es;
3670 break;
3671 case 1:
3672 i.seg[i.mem_operands] = &cs;
3673 break;
3674 case 2:
3675 i.seg[i.mem_operands] = &ss;
3676 break;
3677 case 3:
3678 i.seg[i.mem_operands] = &ds;
3679 break;
3680 case 4:
3681 i.seg[i.mem_operands] = &fs;
3682 break;
3683 case 5:
3684 i.seg[i.mem_operands] = &gs;
3685 break;
3686 }
3687
3688 /* Skip the ':' and whitespace. */
3689 ++op_string;
3690 if (is_space_char (*op_string))
3691 ++op_string;
3692
3693 if (!is_digit_char (*op_string)
3694 && !is_identifier_char (*op_string)
3695 && *op_string != '('
3696 && *op_string != ABSOLUTE_PREFIX)
3697 {
3698 as_bad (_("bad memory operand `%s'"), op_string);
3699 return 0;
3700 }
3701 /* Handle case of %es:*foo. */
3702 if (*op_string == ABSOLUTE_PREFIX)
3703 {
3704 ++op_string;
3705 if (is_space_char (*op_string))
3706 ++op_string;
3707 i.types[this_operand] |= JumpAbsolute;
3708 }
3709 goto do_memory_reference;
3710 }
3711 if (*op_string)
3712 {
3713 as_bad (_("junk `%s' after register"), op_string);
3714 return 0;
3715 }
3716 i.types[this_operand] |= r->reg_type & ~BaseIndex;
3717 i.op[this_operand].regs = r;
3718 i.reg_operands++;
3719 }
3720 else if (*op_string == REGISTER_PREFIX)
3721 {
3722 as_bad (_("bad register name `%s'"), op_string);
3723 return 0;
3724 }
3725 else if (*op_string == IMMEDIATE_PREFIX)
3726 {
3727 ++op_string;
3728 if (i.types[this_operand] & JumpAbsolute)
3729 {
3730 as_bad (_("immediate operand illegal with absolute jump"));
3731 return 0;
3732 }
3733 if (!i386_immediate (op_string))
3734 return 0;
3735 }
3736 else if (is_digit_char (*op_string)
3737 || is_identifier_char (*op_string)
3738 || *op_string == '(')
3739 {
3740 /* This is a memory reference of some sort. */
3741 char *base_string;
3742
3743 /* Start and end of displacement string expression (if found). */
3744 char *displacement_string_start;
3745 char *displacement_string_end;
3746
3747 do_memory_reference:
3748 if ((i.mem_operands == 1
3749 && (current_templates->start->opcode_modifier & IsString) == 0)
3750 || i.mem_operands == 2)
3751 {
3752 as_bad (_("too many memory references for `%s'"),
3753 current_templates->start->name);
3754 return 0;
3755 }
3756
3757 /* Check for base index form. We detect the base index form by
3758 looking for an ')' at the end of the operand, searching
3759 for the '(' matching it, and finding a REGISTER_PREFIX or ','
3760 after the '('. */
3761 base_string = op_string + strlen (op_string);
3762
3763 --base_string;
3764 if (is_space_char (*base_string))
3765 --base_string;
3766
3767 /* If we only have a displacement, set-up for it to be parsed later. */
3768 displacement_string_start = op_string;
3769 displacement_string_end = base_string + 1;
3770
3771 if (*base_string == ')')
3772 {
3773 char *temp_string;
3774 unsigned int parens_balanced = 1;
3775 /* We've already checked that the number of left & right ()'s are
3776 equal, so this loop will not be infinite. */
3777 do
3778 {
3779 base_string--;
3780 if (*base_string == ')')
3781 parens_balanced++;
3782 if (*base_string == '(')
3783 parens_balanced--;
3784 }
3785 while (parens_balanced);
3786
3787 temp_string = base_string;
3788
3789 /* Skip past '(' and whitespace. */
3790 ++base_string;
3791 if (is_space_char (*base_string))
3792 ++base_string;
3793
3794 if (*base_string == ','
3795 || ((*base_string == REGISTER_PREFIX || allow_naked_reg)
3796 && (i.base_reg = parse_register (base_string, &end_op)) != NULL))
3797 {
3798 displacement_string_end = temp_string;
3799
3800 i.types[this_operand] |= BaseIndex;
3801
3802 if (i.base_reg)
3803 {
3804 base_string = end_op;
3805 if (is_space_char (*base_string))
3806 ++base_string;
3807 }
3808
3809 /* There may be an index reg or scale factor here. */
3810 if (*base_string == ',')
3811 {
3812 ++base_string;
3813 if (is_space_char (*base_string))
3814 ++base_string;
3815
3816 if ((*base_string == REGISTER_PREFIX || allow_naked_reg)
3817 && (i.index_reg = parse_register (base_string, &end_op)) != NULL)
3818 {
3819 base_string = end_op;
3820 if (is_space_char (*base_string))
3821 ++base_string;
3822 if (*base_string == ',')
3823 {
3824 ++base_string;
3825 if (is_space_char (*base_string))
3826 ++base_string;
3827 }
3828 else if (*base_string != ')')
3829 {
3830 as_bad (_("expecting `,' or `)' after index register in `%s'"),
3831 operand_string);
3832 return 0;
3833 }
3834 }
3835 else if (*base_string == REGISTER_PREFIX)
3836 {
3837 as_bad (_("bad register name `%s'"), base_string);
3838 return 0;
3839 }
3840
3841 /* Check for scale factor. */
3842 if (*base_string != ')')
3843 {
3844 char *end_scale = i386_scale (base_string);
3845
3846 if (!end_scale)
3847 return 0;
3848
3849 base_string = end_scale;
3850 if (is_space_char (*base_string))
3851 ++base_string;
3852 if (*base_string != ')')
3853 {
3854 as_bad (_("expecting `)' after scale factor in `%s'"),
3855 operand_string);
3856 return 0;
3857 }
3858 }
3859 else if (!i.index_reg)
3860 {
3861 as_bad (_("expecting index register or scale factor after `,'; got '%c'"),
3862 *base_string);
3863 return 0;
3864 }
3865 }
3866 else if (*base_string != ')')
3867 {
3868 as_bad (_("expecting `,' or `)' after base register in `%s'"),
3869 operand_string);
3870 return 0;
3871 }
3872 }
3873 else if (*base_string == REGISTER_PREFIX)
3874 {
3875 as_bad (_("bad register name `%s'"), base_string);
3876 return 0;
3877 }
3878 }
3879
3880 /* If there's an expression beginning the operand, parse it,
3881 assuming displacement_string_start and
3882 displacement_string_end are meaningful. */
3883 if (displacement_string_start != displacement_string_end)
3884 {
3885 if (!i386_displacement (displacement_string_start,
3886 displacement_string_end))
3887 return 0;
3888 }
3889
3890 /* Special case for (%dx) while doing input/output op. */
3891 if (i.base_reg
3892 && i.base_reg->reg_type == (Reg16 | InOutPortReg)
3893 && i.index_reg == 0
3894 && i.log2_scale_factor == 0
3895 && i.seg[i.mem_operands] == 0
3896 && (i.types[this_operand] & Disp) == 0)
3897 {
3898 i.types[this_operand] = InOutPortReg;
3899 return 1;
3900 }
3901
3902 if (i386_index_check (operand_string) == 0)
3903 return 0;
3904 i.mem_operands++;
3905 }
3906 else
3907 {
3908 /* It's not a memory operand; argh! */
3909 as_bad (_("invalid char %s beginning operand %d `%s'"),
3910 output_invalid (*op_string),
3911 this_operand + 1,
3912 op_string);
3913 return 0;
3914 }
3915 return 1; /* Normal return. */
3916 }
3917 \f
3918 /* md_estimate_size_before_relax()
3919
3920 Called just before relax() for rs_machine_dependent frags. The x86
3921 assembler uses these frags to handle variable size jump
3922 instructions.
3923
3924 Any symbol that is now undefined will not become defined.
3925 Return the correct fr_subtype in the frag.
3926 Return the initial "guess for variable size of frag" to caller.
3927 The guess is actually the growth beyond the fixed part. Whatever
3928 we do to grow the fixed or variable part contributes to our
3929 returned value. */
3930
3931 int
3932 md_estimate_size_before_relax (fragP, segment)
3933 register fragS *fragP;
3934 register segT segment;
3935 {
3936 /* We've already got fragP->fr_subtype right; all we have to do is
3937 check for un-relaxable symbols. On an ELF system, we can't relax
3938 an externally visible symbol, because it may be overridden by a
3939 shared library. */
3940 if (S_GET_SEGMENT (fragP->fr_symbol) != segment
3941 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
3942 || S_IS_EXTERNAL (fragP->fr_symbol)
3943 || S_IS_WEAK (fragP->fr_symbol)
3944 #endif
3945 )
3946 {
3947 /* Symbol is undefined in this segment, or we need to keep a
3948 reloc so that weak symbols can be overridden. */
3949 int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
3950 RELOC_ENUM reloc_type;
3951 unsigned char *opcode;
3952 int old_fr_fix;
3953
3954 if (fragP->fr_var != NO_RELOC)
3955 reloc_type = fragP->fr_var;
3956 else if (size == 2)
3957 reloc_type = BFD_RELOC_16_PCREL;
3958 else
3959 reloc_type = BFD_RELOC_32_PCREL;
3960
3961 old_fr_fix = fragP->fr_fix;
3962 opcode = (unsigned char *) fragP->fr_opcode;
3963
3964 switch (TYPE_FROM_RELAX_STATE (fragP->fr_subtype))
3965 {
3966 case UNCOND_JUMP:
3967 /* Make jmp (0xeb) a (d)word displacement jump. */
3968 opcode[0] = 0xe9;
3969 fragP->fr_fix += size;
3970 fix_new (fragP, old_fr_fix, size,
3971 fragP->fr_symbol,
3972 fragP->fr_offset, 1,
3973 reloc_type);
3974 break;
3975
3976 case COND_JUMP86:
3977 if (no_cond_jump_promotion)
3978 goto relax_guess;
3979
3980 if (size == 2)
3981 {
3982 /* Negate the condition, and branch past an
3983 unconditional jump. */
3984 opcode[0] ^= 1;
3985 opcode[1] = 3;
3986 /* Insert an unconditional jump. */
3987 opcode[2] = 0xe9;
3988 /* We added two extra opcode bytes, and have a two byte
3989 offset. */
3990 fragP->fr_fix += 2 + 2;
3991 fix_new (fragP, old_fr_fix + 2, 2,
3992 fragP->fr_symbol,
3993 fragP->fr_offset, 1,
3994 reloc_type);
3995 break;
3996 }
3997 /* Fall through. */
3998
3999 case COND_JUMP:
4000 if (no_cond_jump_promotion)
4001 goto relax_guess;
4002
4003 /* This changes the byte-displacement jump 0x7N
4004 to the (d)word-displacement jump 0x0f,0x8N. */
4005 opcode[1] = opcode[0] + 0x10;
4006 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4007 /* We've added an opcode byte. */
4008 fragP->fr_fix += 1 + size;
4009 fix_new (fragP, old_fr_fix + 1, size,
4010 fragP->fr_symbol,
4011 fragP->fr_offset, 1,
4012 reloc_type);
4013 break;
4014
4015 default:
4016 BAD_CASE (fragP->fr_subtype);
4017 break;
4018 }
4019 frag_wane (fragP);
4020 return fragP->fr_fix - old_fr_fix;
4021 }
4022
4023 relax_guess:
4024 /* Guess size depending on current relax state. Initially the relax
4025 state will correspond to a short jump and we return 1, because
4026 the variable part of the frag (the branch offset) is one byte
4027 long. However, we can relax a section more than once and in that
4028 case we must either set fr_subtype back to the unrelaxed state,
4029 or return the value for the appropriate branch. */
4030 return md_relax_table[fragP->fr_subtype].rlx_length;
4031 }
4032
4033 /* Called after relax() is finished.
4034
4035 In: Address of frag.
4036 fr_type == rs_machine_dependent.
4037 fr_subtype is what the address relaxed to.
4038
4039 Out: Any fixSs and constants are set up.
4040 Caller will turn frag into a ".space 0". */
4041
4042 #ifndef BFD_ASSEMBLER
4043 void
4044 md_convert_frag (headers, sec, fragP)
4045 object_headers *headers ATTRIBUTE_UNUSED;
4046 segT sec ATTRIBUTE_UNUSED;
4047 register fragS *fragP;
4048 #else
4049 void
4050 md_convert_frag (abfd, sec, fragP)
4051 bfd *abfd ATTRIBUTE_UNUSED;
4052 segT sec ATTRIBUTE_UNUSED;
4053 register fragS *fragP;
4054 #endif
4055 {
4056 register unsigned char *opcode;
4057 unsigned char *where_to_put_displacement = NULL;
4058 offsetT target_address;
4059 offsetT opcode_address;
4060 unsigned int extension = 0;
4061 offsetT displacement_from_opcode_start;
4062
4063 opcode = (unsigned char *) fragP->fr_opcode;
4064
4065 /* Address we want to reach in file space. */
4066 target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
4067 #ifdef BFD_ASSEMBLER
4068 /* Not needed otherwise? */
4069 target_address += symbol_get_frag (fragP->fr_symbol)->fr_address;
4070 #endif
4071
4072 /* Address opcode resides at in file space. */
4073 opcode_address = fragP->fr_address + fragP->fr_fix;
4074
4075 /* Displacement from opcode start to fill into instruction. */
4076 displacement_from_opcode_start = target_address - opcode_address;
4077
4078 if ((fragP->fr_subtype & BIG) == 0)
4079 {
4080 /* Don't have to change opcode. */
4081 extension = 1; /* 1 opcode + 1 displacement */
4082 where_to_put_displacement = &opcode[1];
4083 }
4084 else
4085 {
4086 if (no_cond_jump_promotion
4087 && TYPE_FROM_RELAX_STATE (fragP->fr_subtype) != UNCOND_JUMP)
4088 as_warn_where (fragP->fr_file, fragP->fr_line, _("long jump required"));
4089
4090 switch (fragP->fr_subtype)
4091 {
4092 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
4093 extension = 4; /* 1 opcode + 4 displacement */
4094 opcode[0] = 0xe9;
4095 where_to_put_displacement = &opcode[1];
4096 break;
4097
4098 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
4099 extension = 2; /* 1 opcode + 2 displacement */
4100 opcode[0] = 0xe9;
4101 where_to_put_displacement = &opcode[1];
4102 break;
4103
4104 case ENCODE_RELAX_STATE (COND_JUMP, BIG):
4105 case ENCODE_RELAX_STATE (COND_JUMP86, BIG):
4106 extension = 5; /* 2 opcode + 4 displacement */
4107 opcode[1] = opcode[0] + 0x10;
4108 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4109 where_to_put_displacement = &opcode[2];
4110 break;
4111
4112 case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
4113 extension = 3; /* 2 opcode + 2 displacement */
4114 opcode[1] = opcode[0] + 0x10;
4115 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4116 where_to_put_displacement = &opcode[2];
4117 break;
4118
4119 case ENCODE_RELAX_STATE (COND_JUMP86, BIG16):
4120 extension = 4;
4121 opcode[0] ^= 1;
4122 opcode[1] = 3;
4123 opcode[2] = 0xe9;
4124 where_to_put_displacement = &opcode[3];
4125 break;
4126
4127 default:
4128 BAD_CASE (fragP->fr_subtype);
4129 break;
4130 }
4131 }
4132
4133 /* Now put displacement after opcode. */
4134 md_number_to_chars ((char *) where_to_put_displacement,
4135 (valueT) (displacement_from_opcode_start - extension),
4136 DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
4137 fragP->fr_fix += extension;
4138 }
4139 \f
4140 /* Size of byte displacement jmp. */
4141 int md_short_jump_size = 2;
4142
4143 /* Size of dword displacement jmp. */
4144 int md_long_jump_size = 5;
4145
4146 /* Size of relocation record. */
4147 const int md_reloc_size = 8;
4148
4149 void
4150 md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
4151 char *ptr;
4152 addressT from_addr, to_addr;
4153 fragS *frag ATTRIBUTE_UNUSED;
4154 symbolS *to_symbol ATTRIBUTE_UNUSED;
4155 {
4156 offsetT offset;
4157
4158 offset = to_addr - (from_addr + 2);
4159 /* Opcode for byte-disp jump. */
4160 md_number_to_chars (ptr, (valueT) 0xeb, 1);
4161 md_number_to_chars (ptr + 1, (valueT) offset, 1);
4162 }
4163
4164 void
4165 md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
4166 char *ptr;
4167 addressT from_addr, to_addr;
4168 fragS *frag ATTRIBUTE_UNUSED;
4169 symbolS *to_symbol ATTRIBUTE_UNUSED;
4170 {
4171 offsetT offset;
4172
4173 offset = to_addr - (from_addr + 5);
4174 md_number_to_chars (ptr, (valueT) 0xe9, 1);
4175 md_number_to_chars (ptr + 1, (valueT) offset, 4);
4176 }
4177 \f
4178 /* Apply a fixup (fixS) to segment data, once it has been determined
4179 by our caller that we have all the info we need to fix it up.
4180
4181 On the 386, immediates, displacements, and data pointers are all in
4182 the same (little-endian) format, so we don't need to care about which
4183 we are handling. */
4184
4185 int
4186 md_apply_fix3 (fixP, valp, seg)
4187 /* The fix we're to put in. */
4188 fixS *fixP;
4189
4190 /* Pointer to the value of the bits. */
4191 valueT *valp;
4192
4193 /* Segment fix is from. */
4194 segT seg ATTRIBUTE_UNUSED;
4195 {
4196 register char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
4197 valueT value = *valp;
4198
4199 #if defined (BFD_ASSEMBLER) && !defined (TE_Mach)
4200 if (fixP->fx_pcrel)
4201 {
4202 switch (fixP->fx_r_type)
4203 {
4204 default:
4205 break;
4206
4207 case BFD_RELOC_32:
4208 fixP->fx_r_type = BFD_RELOC_32_PCREL;
4209 break;
4210 case BFD_RELOC_16:
4211 fixP->fx_r_type = BFD_RELOC_16_PCREL;
4212 break;
4213 case BFD_RELOC_8:
4214 fixP->fx_r_type = BFD_RELOC_8_PCREL;
4215 break;
4216 }
4217 }
4218
4219 /* This is a hack. There should be a better way to handle this.
4220 This covers for the fact that bfd_install_relocation will
4221 subtract the current location (for partial_inplace, PC relative
4222 relocations); see more below. */
4223 if ((fixP->fx_r_type == BFD_RELOC_32_PCREL
4224 || fixP->fx_r_type == BFD_RELOC_16_PCREL
4225 || fixP->fx_r_type == BFD_RELOC_8_PCREL)
4226 && fixP->fx_addsy && !use_rela_relocations)
4227 {
4228 #ifndef OBJ_AOUT
4229 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4230 #ifdef TE_PE
4231 || OUTPUT_FLAVOR == bfd_target_coff_flavour
4232 #endif
4233 )
4234 value += fixP->fx_where + fixP->fx_frag->fr_address;
4235 #endif
4236 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4237 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
4238 {
4239 segT fseg = S_GET_SEGMENT (fixP->fx_addsy);
4240
4241 if ((fseg == seg
4242 || (symbol_section_p (fixP->fx_addsy)
4243 && fseg != absolute_section))
4244 && ! S_IS_EXTERNAL (fixP->fx_addsy)
4245 && ! S_IS_WEAK (fixP->fx_addsy)
4246 && S_IS_DEFINED (fixP->fx_addsy)
4247 && ! S_IS_COMMON (fixP->fx_addsy))
4248 {
4249 /* Yes, we add the values in twice. This is because
4250 bfd_perform_relocation subtracts them out again. I think
4251 bfd_perform_relocation is broken, but I don't dare change
4252 it. FIXME. */
4253 value += fixP->fx_where + fixP->fx_frag->fr_address;
4254 }
4255 }
4256 #endif
4257 #if defined (OBJ_COFF) && defined (TE_PE)
4258 /* For some reason, the PE format does not store a section
4259 address offset for a PC relative symbol. */
4260 if (S_GET_SEGMENT (fixP->fx_addsy) != seg)
4261 value += md_pcrel_from (fixP);
4262 #endif
4263 }
4264
4265 /* Fix a few things - the dynamic linker expects certain values here,
4266 and we must not dissappoint it. */
4267 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4268 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4269 && fixP->fx_addsy)
4270 switch (fixP->fx_r_type)
4271 {
4272 case BFD_RELOC_386_PLT32:
4273 case BFD_RELOC_X86_64_PLT32:
4274 /* Make the jump instruction point to the address of the operand. At
4275 runtime we merely add the offset to the actual PLT entry. */
4276 value = -4;
4277 break;
4278 case BFD_RELOC_386_GOTPC:
4279
4280 /* This is tough to explain. We end up with this one if we have
4281 * operands that look like "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal
4282 * here is to obtain the absolute address of the GOT, and it is strongly
4283 * preferable from a performance point of view to avoid using a runtime
4284 * relocation for this. The actual sequence of instructions often look
4285 * something like:
4286 *
4287 * call .L66
4288 * .L66:
4289 * popl %ebx
4290 * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
4291 *
4292 * The call and pop essentially return the absolute address of
4293 * the label .L66 and store it in %ebx. The linker itself will
4294 * ultimately change the first operand of the addl so that %ebx points to
4295 * the GOT, but to keep things simple, the .o file must have this operand
4296 * set so that it generates not the absolute address of .L66, but the
4297 * absolute address of itself. This allows the linker itself simply
4298 * treat a GOTPC relocation as asking for a pcrel offset to the GOT to be
4299 * added in, and the addend of the relocation is stored in the operand
4300 * field for the instruction itself.
4301 *
4302 * Our job here is to fix the operand so that it would add the correct
4303 * offset so that %ebx would point to itself. The thing that is tricky is
4304 * that .-.L66 will point to the beginning of the instruction, so we need
4305 * to further modify the operand so that it will point to itself.
4306 * There are other cases where you have something like:
4307 *
4308 * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
4309 *
4310 * and here no correction would be required. Internally in the assembler
4311 * we treat operands of this form as not being pcrel since the '.' is
4312 * explicitly mentioned, and I wonder whether it would simplify matters
4313 * to do it this way. Who knows. In earlier versions of the PIC patches,
4314 * the pcrel_adjust field was used to store the correction, but since the
4315 * expression is not pcrel, I felt it would be confusing to do it this
4316 * way. */
4317
4318 value -= 1;
4319 break;
4320 case BFD_RELOC_386_GOT32:
4321 case BFD_RELOC_X86_64_GOT32:
4322 value = 0; /* Fully resolved at runtime. No addend. */
4323 break;
4324 case BFD_RELOC_386_GOTOFF:
4325 case BFD_RELOC_X86_64_GOTPCREL:
4326 break;
4327
4328 case BFD_RELOC_VTABLE_INHERIT:
4329 case BFD_RELOC_VTABLE_ENTRY:
4330 fixP->fx_done = 0;
4331 return 1;
4332
4333 default:
4334 break;
4335 }
4336 #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
4337 *valp = value;
4338 #endif /* defined (BFD_ASSEMBLER) && !defined (TE_Mach) */
4339
4340 #ifndef BFD_ASSEMBLER
4341 md_number_to_chars (p, value, fixP->fx_size);
4342 #else
4343 /* Are we finished with this relocation now? */
4344 if (fixP->fx_addsy == 0 && fixP->fx_pcrel == 0)
4345 fixP->fx_done = 1;
4346 else if (use_rela_relocations)
4347 {
4348 fixP->fx_no_overflow = 1;
4349 value = 0;
4350 }
4351 md_number_to_chars (p, value, fixP->fx_size);
4352 #endif
4353
4354 return 1;
4355 }
4356 \f
4357 #define MAX_LITTLENUMS 6
4358
4359 /* Turn the string pointed to by litP into a floating point constant
4360 of type TYPE, and emit the appropriate bytes. The number of
4361 LITTLENUMS emitted is stored in *SIZEP. An error message is
4362 returned, or NULL on OK. */
4363
4364 char *
4365 md_atof (type, litP, sizeP)
4366 int type;
4367 char *litP;
4368 int *sizeP;
4369 {
4370 int prec;
4371 LITTLENUM_TYPE words[MAX_LITTLENUMS];
4372 LITTLENUM_TYPE *wordP;
4373 char *t;
4374
4375 switch (type)
4376 {
4377 case 'f':
4378 case 'F':
4379 prec = 2;
4380 break;
4381
4382 case 'd':
4383 case 'D':
4384 prec = 4;
4385 break;
4386
4387 case 'x':
4388 case 'X':
4389 prec = 5;
4390 break;
4391
4392 default:
4393 *sizeP = 0;
4394 return _("Bad call to md_atof ()");
4395 }
4396 t = atof_ieee (input_line_pointer, type, words);
4397 if (t)
4398 input_line_pointer = t;
4399
4400 *sizeP = prec * sizeof (LITTLENUM_TYPE);
4401 /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
4402 the bigendian 386. */
4403 for (wordP = words + prec - 1; prec--;)
4404 {
4405 md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
4406 litP += sizeof (LITTLENUM_TYPE);
4407 }
4408 return 0;
4409 }
4410 \f
4411 char output_invalid_buf[8];
4412
4413 static char *
4414 output_invalid (c)
4415 int c;
4416 {
4417 if (isprint (c))
4418 sprintf (output_invalid_buf, "'%c'", c);
4419 else
4420 sprintf (output_invalid_buf, "(0x%x)", (unsigned) c);
4421 return output_invalid_buf;
4422 }
4423
4424 /* REG_STRING starts *before* REGISTER_PREFIX. */
4425
4426 static const reg_entry *
4427 parse_register (reg_string, end_op)
4428 char *reg_string;
4429 char **end_op;
4430 {
4431 char *s = reg_string;
4432 char *p;
4433 char reg_name_given[MAX_REG_NAME_SIZE + 1];
4434 const reg_entry *r;
4435
4436 /* Skip possible REGISTER_PREFIX and possible whitespace. */
4437 if (*s == REGISTER_PREFIX)
4438 ++s;
4439
4440 if (is_space_char (*s))
4441 ++s;
4442
4443 p = reg_name_given;
4444 while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
4445 {
4446 if (p >= reg_name_given + MAX_REG_NAME_SIZE)
4447 return (const reg_entry *) NULL;
4448 s++;
4449 }
4450
4451 /* For naked regs, make sure that we are not dealing with an identifier.
4452 This prevents confusing an identifier like `eax_var' with register
4453 `eax'. */
4454 if (allow_naked_reg && identifier_chars[(unsigned char) *s])
4455 return (const reg_entry *) NULL;
4456
4457 *end_op = s;
4458
4459 r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
4460
4461 /* Handle floating point regs, allowing spaces in the (i) part. */
4462 if (r == i386_regtab /* %st is first entry of table */)
4463 {
4464 if (is_space_char (*s))
4465 ++s;
4466 if (*s == '(')
4467 {
4468 ++s;
4469 if (is_space_char (*s))
4470 ++s;
4471 if (*s >= '0' && *s <= '7')
4472 {
4473 r = &i386_float_regtab[*s - '0'];
4474 ++s;
4475 if (is_space_char (*s))
4476 ++s;
4477 if (*s == ')')
4478 {
4479 *end_op = s + 1;
4480 return r;
4481 }
4482 }
4483 /* We have "%st(" then garbage. */
4484 return (const reg_entry *) NULL;
4485 }
4486 }
4487
4488 return r;
4489 }
4490 \f
4491 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4492 const char *md_shortopts = "kVQ:sq";
4493 #else
4494 const char *md_shortopts = "q";
4495 #endif
4496
4497 struct option md_longopts[] = {
4498 #define OPTION_32 (OPTION_MD_BASE + 0)
4499 {"32", no_argument, NULL, OPTION_32},
4500 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4501 #define OPTION_64 (OPTION_MD_BASE + 1)
4502 {"64", no_argument, NULL, OPTION_64},
4503 #endif
4504 {NULL, no_argument, NULL, 0}
4505 };
4506 size_t md_longopts_size = sizeof (md_longopts);
4507
4508 int
4509 md_parse_option (c, arg)
4510 int c;
4511 char *arg ATTRIBUTE_UNUSED;
4512 {
4513 switch (c)
4514 {
4515 case 'q':
4516 quiet_warnings = 1;
4517 break;
4518
4519 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4520 /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
4521 should be emitted or not. FIXME: Not implemented. */
4522 case 'Q':
4523 break;
4524
4525 /* -V: SVR4 argument to print version ID. */
4526 case 'V':
4527 print_version_id ();
4528 break;
4529
4530 /* -k: Ignore for FreeBSD compatibility. */
4531 case 'k':
4532 break;
4533
4534 case 's':
4535 /* -s: On i386 Solaris, this tells the native assembler to use
4536 .stab instead of .stab.excl. We always use .stab anyhow. */
4537 break;
4538
4539 case OPTION_64:
4540 {
4541 const char **list, **l;
4542
4543 list = bfd_target_list ();
4544 for (l = list; *l != NULL; l++)
4545 if (strcmp (*l, "elf64-x86-64") == 0)
4546 {
4547 default_arch = "x86_64";
4548 break;
4549 }
4550 if (*l == NULL)
4551 as_fatal (_("No compiled in support for x86_64"));
4552 free (list);
4553 }
4554 break;
4555 #endif
4556
4557 case OPTION_32:
4558 default_arch = "i386";
4559 break;
4560
4561 default:
4562 return 0;
4563 }
4564 return 1;
4565 }
4566
4567 void
4568 md_show_usage (stream)
4569 FILE *stream;
4570 {
4571 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4572 fprintf (stream, _("\
4573 -Q ignored\n\
4574 -V print assembler version number\n\
4575 -k ignored\n\
4576 -q quieten some warnings\n\
4577 -s ignored\n"));
4578 #else
4579 fprintf (stream, _("\
4580 -q quieten some warnings\n"));
4581 #endif
4582 }
4583
4584 #ifdef BFD_ASSEMBLER
4585 #if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
4586 || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
4587
4588 /* Pick the target format to use. */
4589
4590 const char *
4591 i386_target_format ()
4592 {
4593 if (!strcmp (default_arch, "x86_64"))
4594 set_code_flag (CODE_64BIT);
4595 else if (!strcmp (default_arch, "i386"))
4596 set_code_flag (CODE_32BIT);
4597 else
4598 as_fatal (_("Unknown architecture"));
4599 switch (OUTPUT_FLAVOR)
4600 {
4601 #ifdef OBJ_MAYBE_AOUT
4602 case bfd_target_aout_flavour:
4603 return AOUT_TARGET_FORMAT;
4604 #endif
4605 #ifdef OBJ_MAYBE_COFF
4606 case bfd_target_coff_flavour:
4607 return "coff-i386";
4608 #endif
4609 #if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF)
4610 case bfd_target_elf_flavour:
4611 {
4612 if (flag_code == CODE_64BIT)
4613 use_rela_relocations = 1;
4614 return flag_code == CODE_64BIT ? "elf64-x86-64" : "elf32-i386";
4615 }
4616 #endif
4617 default:
4618 abort ();
4619 return NULL;
4620 }
4621 }
4622
4623 #endif /* OBJ_MAYBE_ more than one */
4624 #endif /* BFD_ASSEMBLER */
4625 \f
4626 symbolS *
4627 md_undefined_symbol (name)
4628 char *name;
4629 {
4630 if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
4631 && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
4632 && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
4633 && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
4634 {
4635 if (!GOT_symbol)
4636 {
4637 if (symbol_find (name))
4638 as_bad (_("GOT already in symbol table"));
4639 GOT_symbol = symbol_new (name, undefined_section,
4640 (valueT) 0, &zero_address_frag);
4641 };
4642 return GOT_symbol;
4643 }
4644 return 0;
4645 }
4646
4647 /* Round up a section size to the appropriate boundary. */
4648
4649 valueT
4650 md_section_align (segment, size)
4651 segT segment ATTRIBUTE_UNUSED;
4652 valueT size;
4653 {
4654 #ifdef BFD_ASSEMBLER
4655 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4656 if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
4657 {
4658 /* For a.out, force the section size to be aligned. If we don't do
4659 this, BFD will align it for us, but it will not write out the
4660 final bytes of the section. This may be a bug in BFD, but it is
4661 easier to fix it here since that is how the other a.out targets
4662 work. */
4663 int align;
4664
4665 align = bfd_get_section_alignment (stdoutput, segment);
4666 size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
4667 }
4668 #endif
4669 #endif
4670
4671 return size;
4672 }
4673
4674 /* On the i386, PC-relative offsets are relative to the start of the
4675 next instruction. That is, the address of the offset, plus its
4676 size, since the offset is always the last part of the insn. */
4677
4678 long
4679 md_pcrel_from (fixP)
4680 fixS *fixP;
4681 {
4682 return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
4683 }
4684
4685 #ifndef I386COFF
4686
4687 static void
4688 s_bss (ignore)
4689 int ignore ATTRIBUTE_UNUSED;
4690 {
4691 register int temp;
4692
4693 temp = get_absolute_expression ();
4694 subseg_set (bss_section, (subsegT) temp);
4695 demand_empty_rest_of_line ();
4696 }
4697
4698 #endif
4699
4700 #ifdef BFD_ASSEMBLER
4701
4702 void
4703 i386_validate_fix (fixp)
4704 fixS *fixp;
4705 {
4706 if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
4707 {
4708 /* GOTOFF relocation are nonsense in 64bit mode. */
4709 if (fixp->fx_r_type == BFD_RELOC_32_PCREL)
4710 {
4711 if (flag_code != CODE_64BIT)
4712 abort ();
4713 fixp->fx_r_type = BFD_RELOC_X86_64_GOTPCREL;
4714 }
4715 else
4716 {
4717 if (flag_code == CODE_64BIT)
4718 abort ();
4719 fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
4720 }
4721 fixp->fx_subsy = 0;
4722 }
4723 }
4724
4725 arelent *
4726 tc_gen_reloc (section, fixp)
4727 asection *section ATTRIBUTE_UNUSED;
4728 fixS *fixp;
4729 {
4730 arelent *rel;
4731 bfd_reloc_code_real_type code;
4732
4733 switch (fixp->fx_r_type)
4734 {
4735 case BFD_RELOC_X86_64_PLT32:
4736 case BFD_RELOC_X86_64_GOT32:
4737 case BFD_RELOC_X86_64_GOTPCREL:
4738 case BFD_RELOC_386_PLT32:
4739 case BFD_RELOC_386_GOT32:
4740 case BFD_RELOC_386_GOTOFF:
4741 case BFD_RELOC_386_GOTPC:
4742 case BFD_RELOC_X86_64_32S:
4743 case BFD_RELOC_RVA:
4744 case BFD_RELOC_VTABLE_ENTRY:
4745 case BFD_RELOC_VTABLE_INHERIT:
4746 code = fixp->fx_r_type;
4747 break;
4748 default:
4749 if (fixp->fx_pcrel)
4750 {
4751 switch (fixp->fx_size)
4752 {
4753 default:
4754 as_bad (_("can not do %d byte pc-relative relocation"),
4755 fixp->fx_size);
4756 code = BFD_RELOC_32_PCREL;
4757 break;
4758 case 1: code = BFD_RELOC_8_PCREL; break;
4759 case 2: code = BFD_RELOC_16_PCREL; break;
4760 case 4: code = BFD_RELOC_32_PCREL; break;
4761 }
4762 }
4763 else
4764 {
4765 switch (fixp->fx_size)
4766 {
4767 default:
4768 as_bad (_("can not do %d byte relocation"), fixp->fx_size);
4769 code = BFD_RELOC_32;
4770 break;
4771 case 1: code = BFD_RELOC_8; break;
4772 case 2: code = BFD_RELOC_16; break;
4773 case 4: code = BFD_RELOC_32; break;
4774 case 8: code = BFD_RELOC_64; break;
4775 }
4776 }
4777 break;
4778 }
4779
4780 if (code == BFD_RELOC_32
4781 && GOT_symbol
4782 && fixp->fx_addsy == GOT_symbol)
4783 {
4784 /* We don't support GOTPC on 64bit targets. */
4785 if (flag_code == CODE_64BIT)
4786 abort ();
4787 code = BFD_RELOC_386_GOTPC;
4788 }
4789
4790 rel = (arelent *) xmalloc (sizeof (arelent));
4791 rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
4792 *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
4793
4794 rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
4795 if (!use_rela_relocations)
4796 {
4797 /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
4798 vtable entry to be used in the relocation's section offset. */
4799 if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
4800 rel->address = fixp->fx_offset;
4801
4802 if (fixp->fx_pcrel)
4803 rel->addend = fixp->fx_addnumber;
4804 else
4805 rel->addend = 0;
4806 }
4807 /* Use the rela in 64bit mode. */
4808 else
4809 {
4810 rel->addend = fixp->fx_offset;
4811 if (fixp->fx_pcrel)
4812 rel->addend -= fixp->fx_size;
4813 }
4814
4815 rel->howto = bfd_reloc_type_lookup (stdoutput, code);
4816 if (rel->howto == NULL)
4817 {
4818 as_bad_where (fixp->fx_file, fixp->fx_line,
4819 _("cannot represent relocation type %s"),
4820 bfd_get_reloc_code_name (code));
4821 /* Set howto to a garbage value so that we can keep going. */
4822 rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
4823 assert (rel->howto != NULL);
4824 }
4825
4826 return rel;
4827 }
4828
4829 #else /* ! BFD_ASSEMBLER */
4830
4831 #if (defined(OBJ_AOUT) | defined(OBJ_BOUT))
4832 void
4833 tc_aout_fix_to_chars (where, fixP, segment_address_in_file)
4834 char *where;
4835 fixS *fixP;
4836 relax_addressT segment_address_in_file;
4837 {
4838 /* In: length of relocation (or of address) in chars: 1, 2 or 4.
4839 Out: GNU LD relocation length code: 0, 1, or 2. */
4840
4841 static const unsigned char nbytes_r_length[] = { 42, 0, 1, 42, 2 };
4842 long r_symbolnum;
4843
4844 know (fixP->fx_addsy != NULL);
4845
4846 md_number_to_chars (where,
4847 (valueT) (fixP->fx_frag->fr_address
4848 + fixP->fx_where - segment_address_in_file),
4849 4);
4850
4851 r_symbolnum = (S_IS_DEFINED (fixP->fx_addsy)
4852 ? S_GET_TYPE (fixP->fx_addsy)
4853 : fixP->fx_addsy->sy_number);
4854
4855 where[6] = (r_symbolnum >> 16) & 0x0ff;
4856 where[5] = (r_symbolnum >> 8) & 0x0ff;
4857 where[4] = r_symbolnum & 0x0ff;
4858 where[7] = ((((!S_IS_DEFINED (fixP->fx_addsy)) << 3) & 0x08)
4859 | ((nbytes_r_length[fixP->fx_size] << 1) & 0x06)
4860 | (((fixP->fx_pcrel << 0) & 0x01) & 0x0f));
4861 }
4862
4863 #endif /* OBJ_AOUT or OBJ_BOUT. */
4864
4865 #if defined (I386COFF)
4866
4867 short
4868 tc_coff_fix2rtype (fixP)
4869 fixS *fixP;
4870 {
4871 if (fixP->fx_r_type == R_IMAGEBASE)
4872 return R_IMAGEBASE;
4873
4874 return (fixP->fx_pcrel ?
4875 (fixP->fx_size == 1 ? R_PCRBYTE :
4876 fixP->fx_size == 2 ? R_PCRWORD :
4877 R_PCRLONG) :
4878 (fixP->fx_size == 1 ? R_RELBYTE :
4879 fixP->fx_size == 2 ? R_RELWORD :
4880 R_DIR32));
4881 }
4882
4883 int
4884 tc_coff_sizemachdep (frag)
4885 fragS *frag;
4886 {
4887 if (frag->fr_next)
4888 return (frag->fr_next->fr_address - frag->fr_address);
4889 else
4890 return 0;
4891 }
4892
4893 #endif /* I386COFF */
4894
4895 #endif /* ! BFD_ASSEMBLER */
4896 \f
4897 /* Parse operands using Intel syntax. This implements a recursive descent
4898 parser based on the BNF grammar published in Appendix B of the MASM 6.1
4899 Programmer's Guide.
4900
4901 FIXME: We do not recognize the full operand grammar defined in the MASM
4902 documentation. In particular, all the structure/union and
4903 high-level macro operands are missing.
4904
4905 Uppercase words are terminals, lower case words are non-terminals.
4906 Objects surrounded by double brackets '[[' ']]' are optional. Vertical
4907 bars '|' denote choices. Most grammar productions are implemented in
4908 functions called 'intel_<production>'.
4909
4910 Initial production is 'expr'.
4911
4912 addOp + | -
4913
4914 alpha [a-zA-Z]
4915
4916 byteRegister AL | AH | BL | BH | CL | CH | DL | DH
4917
4918 constant digits [[ radixOverride ]]
4919
4920 dataType BYTE | WORD | DWORD | QWORD | XWORD
4921
4922 digits decdigit
4923 | digits decdigit
4924 | digits hexdigit
4925
4926 decdigit [0-9]
4927
4928 e05 e05 addOp e06
4929 | e06
4930
4931 e06 e06 mulOp e09
4932 | e09
4933
4934 e09 OFFSET e10
4935 | e09 PTR e10
4936 | e09 : e10
4937 | e10
4938
4939 e10 e10 [ expr ]
4940 | e11
4941
4942 e11 ( expr )
4943 | [ expr ]
4944 | constant
4945 | dataType
4946 | id
4947 | $
4948 | register
4949
4950 => expr SHORT e05
4951 | e05
4952
4953 gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
4954 | BP | EBP | SP | ESP | DI | EDI | SI | ESI
4955
4956 hexdigit a | b | c | d | e | f
4957 | A | B | C | D | E | F
4958
4959 id alpha
4960 | id alpha
4961 | id decdigit
4962
4963 mulOp * | / | MOD
4964
4965 quote " | '
4966
4967 register specialRegister
4968 | gpRegister
4969 | byteRegister
4970
4971 segmentRegister CS | DS | ES | FS | GS | SS
4972
4973 specialRegister CR0 | CR2 | CR3
4974 | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
4975 | TR3 | TR4 | TR5 | TR6 | TR7
4976
4977 We simplify the grammar in obvious places (e.g., register parsing is
4978 done by calling parse_register) and eliminate immediate left recursion
4979 to implement a recursive-descent parser.
4980
4981 expr SHORT e05
4982 | e05
4983
4984 e05 e06 e05'
4985
4986 e05' addOp e06 e05'
4987 | Empty
4988
4989 e06 e09 e06'
4990
4991 e06' mulOp e09 e06'
4992 | Empty
4993
4994 e09 OFFSET e10 e09'
4995 | e10 e09'
4996
4997 e09' PTR e10 e09'
4998 | : e10 e09'
4999 | Empty
5000
5001 e10 e11 e10'
5002
5003 e10' [ expr ] e10'
5004 | Empty
5005
5006 e11 ( expr )
5007 | [ expr ]
5008 | BYTE
5009 | WORD
5010 | DWORD
5011 | QWORD
5012 | XWORD
5013 | .
5014 | $
5015 | register
5016 | id
5017 | constant */
5018
5019 /* Parsing structure for the intel syntax parser. Used to implement the
5020 semantic actions for the operand grammar. */
5021 struct intel_parser_s
5022 {
5023 char *op_string; /* The string being parsed. */
5024 int got_a_float; /* Whether the operand is a float. */
5025 int op_modifier; /* Operand modifier. */
5026 int is_mem; /* 1 if operand is memory reference. */
5027 const reg_entry *reg; /* Last register reference found. */
5028 char *disp; /* Displacement string being built. */
5029 };
5030
5031 static struct intel_parser_s intel_parser;
5032
5033 /* Token structure for parsing intel syntax. */
5034 struct intel_token
5035 {
5036 int code; /* Token code. */
5037 const reg_entry *reg; /* Register entry for register tokens. */
5038 char *str; /* String representation. */
5039 };
5040
5041 static struct intel_token cur_token, prev_token;
5042
5043 /* Token codes for the intel parser. Since T_SHORT is already used
5044 by COFF, undefine it first to prevent a warning. */
5045 #define T_NIL -1
5046 #define T_CONST 1
5047 #define T_REG 2
5048 #define T_BYTE 3
5049 #define T_WORD 4
5050 #define T_DWORD 5
5051 #define T_QWORD 6
5052 #define T_XWORD 7
5053 #undef T_SHORT
5054 #define T_SHORT 8
5055 #define T_OFFSET 9
5056 #define T_PTR 10
5057 #define T_ID 11
5058
5059 /* Prototypes for intel parser functions. */
5060 static int intel_match_token PARAMS ((int code));
5061 static void intel_get_token PARAMS ((void));
5062 static void intel_putback_token PARAMS ((void));
5063 static int intel_expr PARAMS ((void));
5064 static int intel_e05 PARAMS ((void));
5065 static int intel_e05_1 PARAMS ((void));
5066 static int intel_e06 PARAMS ((void));
5067 static int intel_e06_1 PARAMS ((void));
5068 static int intel_e09 PARAMS ((void));
5069 static int intel_e09_1 PARAMS ((void));
5070 static int intel_e10 PARAMS ((void));
5071 static int intel_e10_1 PARAMS ((void));
5072 static int intel_e11 PARAMS ((void));
5073
5074 static int
5075 i386_intel_operand (operand_string, got_a_float)
5076 char *operand_string;
5077 int got_a_float;
5078 {
5079 int ret;
5080 char *p;
5081
5082 /* Initialize token holders. */
5083 cur_token.code = prev_token.code = T_NIL;
5084 cur_token.reg = prev_token.reg = NULL;
5085 cur_token.str = prev_token.str = NULL;
5086
5087 /* Initialize parser structure. */
5088 p = intel_parser.op_string = (char *) malloc (strlen (operand_string) + 1);
5089 if (p == NULL)
5090 abort ();
5091 strcpy (intel_parser.op_string, operand_string);
5092 intel_parser.got_a_float = got_a_float;
5093 intel_parser.op_modifier = -1;
5094 intel_parser.is_mem = 0;
5095 intel_parser.reg = NULL;
5096 intel_parser.disp = (char *) malloc (strlen (operand_string) + 1);
5097 if (intel_parser.disp == NULL)
5098 abort ();
5099 intel_parser.disp[0] = '\0';
5100
5101 /* Read the first token and start the parser. */
5102 intel_get_token ();
5103 ret = intel_expr ();
5104
5105 if (ret)
5106 {
5107 /* If we found a memory reference, hand it over to i386_displacement
5108 to fill in the rest of the operand fields. */
5109 if (intel_parser.is_mem)
5110 {
5111 if ((i.mem_operands == 1
5112 && (current_templates->start->opcode_modifier & IsString) == 0)
5113 || i.mem_operands == 2)
5114 {
5115 as_bad (_("too many memory references for '%s'"),
5116 current_templates->start->name);
5117 ret = 0;
5118 }
5119 else
5120 {
5121 char *s = intel_parser.disp;
5122 i.mem_operands++;
5123
5124 /* Add the displacement expression. */
5125 if (*s != '\0')
5126 ret = i386_displacement (s, s + strlen (s))
5127 && i386_index_check (s);
5128 }
5129 }
5130
5131 /* Constant and OFFSET expressions are handled by i386_immediate. */
5132 else if (intel_parser.op_modifier == OFFSET_FLAT
5133 || intel_parser.reg == NULL)
5134 ret = i386_immediate (intel_parser.disp);
5135 }
5136
5137 free (p);
5138 free (intel_parser.disp);
5139
5140 return ret;
5141 }
5142
5143 /* expr SHORT e05
5144 | e05 */
5145 static int
5146 intel_expr ()
5147 {
5148 /* expr SHORT e05 */
5149 if (cur_token.code == T_SHORT)
5150 {
5151 intel_parser.op_modifier = SHORT;
5152 intel_match_token (T_SHORT);
5153
5154 return (intel_e05 ());
5155 }
5156
5157 /* expr e05 */
5158 else
5159 return intel_e05 ();
5160 }
5161
5162 /* e05 e06 e05'
5163
5164 e05' addOp e06 e05'
5165 | Empty */
5166 static int
5167 intel_e05 ()
5168 {
5169 return (intel_e06 () && intel_e05_1 ());
5170 }
5171
5172 static int
5173 intel_e05_1 ()
5174 {
5175 /* e05' addOp e06 e05' */
5176 if (cur_token.code == '+' || cur_token.code == '-')
5177 {
5178 strcat (intel_parser.disp, cur_token.str);
5179 intel_match_token (cur_token.code);
5180
5181 return (intel_e06 () && intel_e05_1 ());
5182 }
5183
5184 /* e05' Empty */
5185 else
5186 return 1;
5187 }
5188
5189 /* e06 e09 e06'
5190
5191 e06' mulOp e09 e06'
5192 | Empty */
5193 static int
5194 intel_e06 ()
5195 {
5196 return (intel_e09 () && intel_e06_1 ());
5197 }
5198
5199 static int
5200 intel_e06_1 ()
5201 {
5202 /* e06' mulOp e09 e06' */
5203 if (cur_token.code == '*' || cur_token.code == '/')
5204 {
5205 strcat (intel_parser.disp, cur_token.str);
5206 intel_match_token (cur_token.code);
5207
5208 return (intel_e09 () && intel_e06_1 ());
5209 }
5210
5211 /* e06' Empty */
5212 else
5213 return 1;
5214 }
5215
5216 /* e09 OFFSET e10 e09'
5217 | e10 e09'
5218
5219 e09' PTR e10 e09'
5220 | : e10 e09'
5221 | Empty */
5222 static int
5223 intel_e09 ()
5224 {
5225 /* e09 OFFSET e10 e09' */
5226 if (cur_token.code == T_OFFSET)
5227 {
5228 intel_parser.is_mem = 0;
5229 intel_parser.op_modifier = OFFSET_FLAT;
5230 intel_match_token (T_OFFSET);
5231
5232 return (intel_e10 () && intel_e09_1 ());
5233 }
5234
5235 /* e09 e10 e09' */
5236 else
5237 return (intel_e10 () && intel_e09_1 ());
5238 }
5239
5240 static int
5241 intel_e09_1 ()
5242 {
5243 /* e09' PTR e10 e09' */
5244 if (cur_token.code == T_PTR)
5245 {
5246 if (prev_token.code == T_BYTE)
5247 i.suffix = BYTE_MNEM_SUFFIX;
5248
5249 else if (prev_token.code == T_WORD)
5250 {
5251 if (intel_parser.got_a_float == 2) /* "fi..." */
5252 i.suffix = SHORT_MNEM_SUFFIX;
5253 else
5254 i.suffix = WORD_MNEM_SUFFIX;
5255 }
5256
5257 else if (prev_token.code == T_DWORD)
5258 {
5259 if (intel_parser.got_a_float == 1) /* "f..." */
5260 i.suffix = SHORT_MNEM_SUFFIX;
5261 else
5262 i.suffix = LONG_MNEM_SUFFIX;
5263 }
5264
5265 else if (prev_token.code == T_QWORD)
5266 {
5267 if (intel_parser.got_a_float == 1) /* "f..." */
5268 i.suffix = LONG_MNEM_SUFFIX;
5269 else
5270 i.suffix = QWORD_MNEM_SUFFIX;
5271 }
5272
5273 else if (prev_token.code == T_XWORD)
5274 i.suffix = LONG_DOUBLE_MNEM_SUFFIX;
5275
5276 else
5277 {
5278 as_bad (_("Unknown operand modifier `%s'\n"), prev_token.str);
5279 return 0;
5280 }
5281
5282 intel_match_token (T_PTR);
5283
5284 return (intel_e10 () && intel_e09_1 ());
5285 }
5286
5287 /* e09 : e10 e09' */
5288 else if (cur_token.code == ':')
5289 {
5290 /* Mark as a memory operand only if it's not already known to be an
5291 offset expression. */
5292 if (intel_parser.op_modifier != OFFSET_FLAT)
5293 intel_parser.is_mem = 1;
5294
5295 return (intel_match_token (':') && intel_e10 () && intel_e09_1 ());
5296 }
5297
5298 /* e09' Empty */
5299 else
5300 return 1;
5301 }
5302
5303 /* e10 e11 e10'
5304
5305 e10' [ expr ] e10'
5306 | Empty */
5307 static int
5308 intel_e10 ()
5309 {
5310 return (intel_e11 () && intel_e10_1 ());
5311 }
5312
5313 static int
5314 intel_e10_1 ()
5315 {
5316 /* e10' [ expr ] e10' */
5317 if (cur_token.code == '[')
5318 {
5319 intel_match_token ('[');
5320
5321 /* Mark as a memory operand only if it's not already known to be an
5322 offset expression. If it's an offset expression, we need to keep
5323 the brace in. */
5324 if (intel_parser.op_modifier != OFFSET_FLAT)
5325 intel_parser.is_mem = 1;
5326 else
5327 strcat (intel_parser.disp, "[");
5328
5329 /* Add a '+' to the displacement string if necessary. */
5330 if (*intel_parser.disp != '\0'
5331 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5332 strcat (intel_parser.disp, "+");
5333
5334 if (intel_expr () && intel_match_token (']'))
5335 {
5336 /* Preserve brackets when the operand is an offset expression. */
5337 if (intel_parser.op_modifier == OFFSET_FLAT)
5338 strcat (intel_parser.disp, "]");
5339
5340 return intel_e10_1 ();
5341 }
5342 else
5343 return 0;
5344 }
5345
5346 /* e10' Empty */
5347 else
5348 return 1;
5349 }
5350
5351 /* e11 ( expr )
5352 | [ expr ]
5353 | BYTE
5354 | WORD
5355 | DWORD
5356 | QWORD
5357 | XWORD
5358 | $
5359 | .
5360 | register
5361 | id
5362 | constant */
5363 static int
5364 intel_e11 ()
5365 {
5366 /* e11 ( expr ) */
5367 if (cur_token.code == '(')
5368 {
5369 intel_match_token ('(');
5370 strcat (intel_parser.disp, "(");
5371
5372 if (intel_expr () && intel_match_token (')'))
5373 {
5374 strcat (intel_parser.disp, ")");
5375 return 1;
5376 }
5377 else
5378 return 0;
5379 }
5380
5381 /* e11 [ expr ] */
5382 else if (cur_token.code == '[')
5383 {
5384 intel_match_token ('[');
5385
5386 /* Mark as a memory operand only if it's not already known to be an
5387 offset expression. If it's an offset expression, we need to keep
5388 the brace in. */
5389 if (intel_parser.op_modifier != OFFSET_FLAT)
5390 intel_parser.is_mem = 1;
5391 else
5392 strcat (intel_parser.disp, "[");
5393
5394 /* Operands for jump/call inside brackets denote absolute addresses. */
5395 if (current_templates->start->opcode_modifier & Jump
5396 || current_templates->start->opcode_modifier & JumpDword
5397 || current_templates->start->opcode_modifier & JumpByte
5398 || current_templates->start->opcode_modifier & JumpInterSegment)
5399 i.types[this_operand] |= JumpAbsolute;
5400
5401 /* Add a '+' to the displacement string if necessary. */
5402 if (*intel_parser.disp != '\0'
5403 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5404 strcat (intel_parser.disp, "+");
5405
5406 if (intel_expr () && intel_match_token (']'))
5407 {
5408 /* Preserve brackets when the operand is an offset expression. */
5409 if (intel_parser.op_modifier == OFFSET_FLAT)
5410 strcat (intel_parser.disp, "]");
5411
5412 return 1;
5413 }
5414 else
5415 return 0;
5416 }
5417
5418 /* e11 BYTE
5419 | WORD
5420 | DWORD
5421 | QWORD
5422 | XWORD */
5423 else if (cur_token.code == T_BYTE
5424 || cur_token.code == T_WORD
5425 || cur_token.code == T_DWORD
5426 || cur_token.code == T_QWORD
5427 || cur_token.code == T_XWORD)
5428 {
5429 intel_match_token (cur_token.code);
5430
5431 return 1;
5432 }
5433
5434 /* e11 $
5435 | . */
5436 else if (cur_token.code == '$' || cur_token.code == '.')
5437 {
5438 strcat (intel_parser.disp, cur_token.str);
5439 intel_match_token (cur_token.code);
5440
5441 /* Mark as a memory operand only if it's not already known to be an
5442 offset expression. */
5443 if (intel_parser.op_modifier != OFFSET_FLAT)
5444 intel_parser.is_mem = 1;
5445
5446 return 1;
5447 }
5448
5449 /* e11 register */
5450 else if (cur_token.code == T_REG)
5451 {
5452 const reg_entry *reg = intel_parser.reg = cur_token.reg;
5453
5454 intel_match_token (T_REG);
5455
5456 /* Check for segment change. */
5457 if (cur_token.code == ':')
5458 {
5459 if (reg->reg_type & (SReg2 | SReg3))
5460 {
5461 switch (reg->reg_num)
5462 {
5463 case 0:
5464 i.seg[i.mem_operands] = &es;
5465 break;
5466 case 1:
5467 i.seg[i.mem_operands] = &cs;
5468 break;
5469 case 2:
5470 i.seg[i.mem_operands] = &ss;
5471 break;
5472 case 3:
5473 i.seg[i.mem_operands] = &ds;
5474 break;
5475 case 4:
5476 i.seg[i.mem_operands] = &fs;
5477 break;
5478 case 5:
5479 i.seg[i.mem_operands] = &gs;
5480 break;
5481 }
5482 }
5483 else
5484 {
5485 as_bad (_("`%s' is not a valid segment register"), reg->reg_name);
5486 return 0;
5487 }
5488 }
5489
5490 /* Not a segment register. Check for register scaling. */
5491 else if (cur_token.code == '*')
5492 {
5493 if (!intel_parser.is_mem)
5494 {
5495 as_bad (_("Register scaling only allowed in memory operands."));
5496 return 0;
5497 }
5498
5499 /* What follows must be a valid scale. */
5500 if (intel_match_token ('*')
5501 && strchr ("01248", *cur_token.str))
5502 {
5503 i.index_reg = reg;
5504 i.types[this_operand] |= BaseIndex;
5505
5506 /* Set the scale after setting the register (otherwise,
5507 i386_scale will complain) */
5508 i386_scale (cur_token.str);
5509 intel_match_token (T_CONST);
5510 }
5511 else
5512 {
5513 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
5514 cur_token.str);
5515 return 0;
5516 }
5517 }
5518
5519 /* No scaling. If this is a memory operand, the register is either a
5520 base register (first occurrence) or an index register (second
5521 occurrence). */
5522 else if (intel_parser.is_mem && !(reg->reg_type & (SReg2 | SReg3)))
5523 {
5524 if (i.base_reg && i.index_reg)
5525 {
5526 as_bad (_("Too many register references in memory operand.\n"));
5527 return 0;
5528 }
5529
5530 if (i.base_reg == NULL)
5531 i.base_reg = reg;
5532 else
5533 i.index_reg = reg;
5534
5535 i.types[this_operand] |= BaseIndex;
5536 }
5537
5538 /* Offset modifier. Add the register to the displacement string to be
5539 parsed as an immediate expression after we're done. */
5540 else if (intel_parser.op_modifier == OFFSET_FLAT)
5541 strcat (intel_parser.disp, reg->reg_name);
5542
5543 /* It's neither base nor index nor offset. */
5544 else
5545 {
5546 i.types[this_operand] |= reg->reg_type & ~BaseIndex;
5547 i.op[this_operand].regs = reg;
5548 i.reg_operands++;
5549 }
5550
5551 /* Since registers are not part of the displacement string (except
5552 when we're parsing offset operands), we may need to remove any
5553 preceding '+' from the displacement string. */
5554 if (*intel_parser.disp != '\0'
5555 && intel_parser.op_modifier != OFFSET_FLAT)
5556 {
5557 char *s = intel_parser.disp;
5558 s += strlen (s) - 1;
5559 if (*s == '+')
5560 *s = '\0';
5561 }
5562
5563 return 1;
5564 }
5565
5566 /* e11 id */
5567 else if (cur_token.code == T_ID)
5568 {
5569 /* Add the identifier to the displacement string. */
5570 strcat (intel_parser.disp, cur_token.str);
5571 intel_match_token (T_ID);
5572
5573 /* The identifier represents a memory reference only if it's not
5574 preceded by an offset modifier. */
5575 if (intel_parser.op_modifier != OFFSET_FLAT)
5576 intel_parser.is_mem = 1;
5577
5578 return 1;
5579 }
5580
5581 /* e11 constant */
5582 else if (cur_token.code == T_CONST
5583 || cur_token.code == '-'
5584 || cur_token.code == '+')
5585 {
5586 char *save_str;
5587
5588 /* Allow constants that start with `+' or `-'. */
5589 if (cur_token.code == '-' || cur_token.code == '+')
5590 {
5591 strcat (intel_parser.disp, cur_token.str);
5592 intel_match_token (cur_token.code);
5593 if (cur_token.code != T_CONST)
5594 {
5595 as_bad (_("Syntax error. Expecting a constant. Got `%s'.\n"),
5596 cur_token.str);
5597 return 0;
5598 }
5599 }
5600
5601 save_str = (char *) malloc (strlen (cur_token.str) + 1);
5602 if (save_str == NULL)
5603 abort ();
5604 strcpy (save_str, cur_token.str);
5605
5606 /* Get the next token to check for register scaling. */
5607 intel_match_token (cur_token.code);
5608
5609 /* Check if this constant is a scaling factor for an index register. */
5610 if (cur_token.code == '*')
5611 {
5612 if (intel_match_token ('*') && cur_token.code == T_REG)
5613 {
5614 if (!intel_parser.is_mem)
5615 {
5616 as_bad (_("Register scaling only allowed in memory operands."));
5617 return 0;
5618 }
5619
5620 /* The constant is followed by `* reg', so it must be
5621 a valid scale. */
5622 if (strchr ("01248", *save_str))
5623 {
5624 i.index_reg = cur_token.reg;
5625 i.types[this_operand] |= BaseIndex;
5626
5627 /* Set the scale after setting the register (otherwise,
5628 i386_scale will complain) */
5629 i386_scale (save_str);
5630 intel_match_token (T_REG);
5631
5632 /* Since registers are not part of the displacement
5633 string, we may need to remove any preceding '+' from
5634 the displacement string. */
5635 if (*intel_parser.disp != '\0')
5636 {
5637 char *s = intel_parser.disp;
5638 s += strlen (s) - 1;
5639 if (*s == '+')
5640 *s = '\0';
5641 }
5642
5643 free (save_str);
5644
5645 return 1;
5646 }
5647 else
5648 return 0;
5649 }
5650
5651 /* The constant was not used for register scaling. Since we have
5652 already consumed the token following `*' we now need to put it
5653 back in the stream. */
5654 else
5655 intel_putback_token ();
5656 }
5657
5658 /* Add the constant to the displacement string. */
5659 strcat (intel_parser.disp, save_str);
5660 free (save_str);
5661
5662 return 1;
5663 }
5664
5665 as_bad (_("Unrecognized token '%s'"), cur_token.str);
5666 return 0;
5667 }
5668
5669 /* Match the given token against cur_token. If they match, read the next
5670 token from the operand string. */
5671 static int
5672 intel_match_token (code)
5673 int code;
5674 {
5675 if (cur_token.code == code)
5676 {
5677 intel_get_token ();
5678 return 1;
5679 }
5680 else
5681 {
5682 as_bad (_("Unexpected token `%s'\n"), cur_token.str);
5683 return 0;
5684 }
5685 }
5686
5687 /* Read a new token from intel_parser.op_string and store it in cur_token. */
5688 static void
5689 intel_get_token ()
5690 {
5691 char *end_op;
5692 const reg_entry *reg;
5693 struct intel_token new_token;
5694
5695 new_token.code = T_NIL;
5696 new_token.reg = NULL;
5697 new_token.str = NULL;
5698
5699 /* Free the memory allocated to the previous token and move
5700 cur_token to prev_token. */
5701 if (prev_token.str)
5702 free (prev_token.str);
5703
5704 prev_token = cur_token;
5705
5706 /* Skip whitespace. */
5707 while (is_space_char (*intel_parser.op_string))
5708 intel_parser.op_string++;
5709
5710 /* Return an empty token if we find nothing else on the line. */
5711 if (*intel_parser.op_string == '\0')
5712 {
5713 cur_token = new_token;
5714 return;
5715 }
5716
5717 /* The new token cannot be larger than the remainder of the operand
5718 string. */
5719 new_token.str = (char *) malloc (strlen (intel_parser.op_string) + 1);
5720 if (new_token.str == NULL)
5721 abort ();
5722 new_token.str[0] = '\0';
5723
5724 if (strchr ("0123456789", *intel_parser.op_string))
5725 {
5726 char *p = new_token.str;
5727 char *q = intel_parser.op_string;
5728 new_token.code = T_CONST;
5729
5730 /* Allow any kind of identifier char to encompass floating point and
5731 hexadecimal numbers. */
5732 while (is_identifier_char (*q))
5733 *p++ = *q++;
5734 *p = '\0';
5735
5736 /* Recognize special symbol names [0-9][bf]. */
5737 if (strlen (intel_parser.op_string) == 2
5738 && (intel_parser.op_string[1] == 'b'
5739 || intel_parser.op_string[1] == 'f'))
5740 new_token.code = T_ID;
5741 }
5742
5743 else if (strchr ("+-/*:[]()", *intel_parser.op_string))
5744 {
5745 new_token.code = *intel_parser.op_string;
5746 new_token.str[0] = *intel_parser.op_string;
5747 new_token.str[1] = '\0';
5748 }
5749
5750 else if ((*intel_parser.op_string == REGISTER_PREFIX || allow_naked_reg)
5751 && ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL))
5752 {
5753 new_token.code = T_REG;
5754 new_token.reg = reg;
5755
5756 if (*intel_parser.op_string == REGISTER_PREFIX)
5757 {
5758 new_token.str[0] = REGISTER_PREFIX;
5759 new_token.str[1] = '\0';
5760 }
5761
5762 strcat (new_token.str, reg->reg_name);
5763 }
5764
5765 else if (is_identifier_char (*intel_parser.op_string))
5766 {
5767 char *p = new_token.str;
5768 char *q = intel_parser.op_string;
5769
5770 /* A '.' or '$' followed by an identifier char is an identifier.
5771 Otherwise, it's operator '.' followed by an expression. */
5772 if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
5773 {
5774 new_token.code = *q;
5775 new_token.str[0] = *q;
5776 new_token.str[1] = '\0';
5777 }
5778 else
5779 {
5780 while (is_identifier_char (*q) || *q == '@')
5781 *p++ = *q++;
5782 *p = '\0';
5783
5784 if (strcasecmp (new_token.str, "BYTE") == 0)
5785 new_token.code = T_BYTE;
5786
5787 else if (strcasecmp (new_token.str, "WORD") == 0)
5788 new_token.code = T_WORD;
5789
5790 else if (strcasecmp (new_token.str, "DWORD") == 0)
5791 new_token.code = T_DWORD;
5792
5793 else if (strcasecmp (new_token.str, "QWORD") == 0)
5794 new_token.code = T_QWORD;
5795
5796 else if (strcasecmp (new_token.str, "XWORD") == 0)
5797 new_token.code = T_XWORD;
5798
5799 else if (strcasecmp (new_token.str, "PTR") == 0)
5800 new_token.code = T_PTR;
5801
5802 else if (strcasecmp (new_token.str, "SHORT") == 0)
5803 new_token.code = T_SHORT;
5804
5805 else if (strcasecmp (new_token.str, "OFFSET") == 0)
5806 {
5807 new_token.code = T_OFFSET;
5808
5809 /* ??? This is not mentioned in the MASM grammar but gcc
5810 makes use of it with -mintel-syntax. OFFSET may be
5811 followed by FLAT: */
5812 if (strncasecmp (q, " FLAT:", 6) == 0)
5813 strcat (new_token.str, " FLAT:");
5814 }
5815
5816 /* ??? This is not mentioned in the MASM grammar. */
5817 else if (strcasecmp (new_token.str, "FLAT") == 0)
5818 new_token.code = T_OFFSET;
5819
5820 else
5821 new_token.code = T_ID;
5822 }
5823 }
5824
5825 else
5826 as_bad (_("Unrecognized token `%s'\n"), intel_parser.op_string);
5827
5828 intel_parser.op_string += strlen (new_token.str);
5829 cur_token = new_token;
5830 }
5831
5832 /* Put cur_token back into the token stream and make cur_token point to
5833 prev_token. */
5834 static void
5835 intel_putback_token ()
5836 {
5837 intel_parser.op_string -= strlen (cur_token.str);
5838 free (cur_token.str);
5839 cur_token = prev_token;
5840
5841 /* Forget prev_token. */
5842 prev_token.code = T_NIL;
5843 prev_token.reg = NULL;
5844 prev_token.str = NULL;
5845 }
This page took 0.17683 seconds and 5 git commands to generate.