From 2.11 branch 2001-03-30 Richard Henderson <rth@redhat.com>
[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 int j;
1235
1236 char mnemonic[MAX_MNEM_SIZE];
1237
1238 /* Initialize globals. */
1239 memset (&i, '\0', sizeof (i));
1240 for (j = 0; j < MAX_OPERANDS; j++)
1241 i.reloc[j] = NO_RELOC;
1242 memset (disp_expressions, '\0', sizeof (disp_expressions));
1243 memset (im_expressions, '\0', sizeof (im_expressions));
1244 save_stack_p = save_stack;
1245
1246 /* First parse an instruction mnemonic & call i386_operand for the operands.
1247 We assume that the scrubber has arranged it so that line[0] is the valid
1248 start of a (possibly prefixed) mnemonic. */
1249 {
1250 char *l = line;
1251 char *token_start = l;
1252 char *mnem_p;
1253
1254 /* Non-zero if we found a prefix only acceptable with string insns. */
1255 const char *expecting_string_instruction = NULL;
1256
1257 while (1)
1258 {
1259 mnem_p = mnemonic;
1260 while ((*mnem_p = mnemonic_chars[(unsigned char) *l]) != 0)
1261 {
1262 mnem_p++;
1263 if (mnem_p >= mnemonic + sizeof (mnemonic))
1264 {
1265 as_bad (_("no such instruction: `%s'"), token_start);
1266 return;
1267 }
1268 l++;
1269 }
1270 if (!is_space_char (*l)
1271 && *l != END_OF_INSN
1272 && *l != PREFIX_SEPARATOR)
1273 {
1274 as_bad (_("invalid character %s in mnemonic"),
1275 output_invalid (*l));
1276 return;
1277 }
1278 if (token_start == l)
1279 {
1280 if (*l == PREFIX_SEPARATOR)
1281 as_bad (_("expecting prefix; got nothing"));
1282 else
1283 as_bad (_("expecting mnemonic; got nothing"));
1284 return;
1285 }
1286
1287 /* Look up instruction (or prefix) via hash table. */
1288 current_templates = hash_find (op_hash, mnemonic);
1289
1290 if (*l != END_OF_INSN
1291 && (! is_space_char (*l) || l[1] != END_OF_INSN)
1292 && current_templates
1293 && (current_templates->start->opcode_modifier & IsPrefix))
1294 {
1295 /* If we are in 16-bit mode, do not allow addr16 or data16.
1296 Similarly, in 32-bit mode, do not allow addr32 or data32. */
1297 if ((current_templates->start->opcode_modifier & (Size16 | Size32))
1298 && (((current_templates->start->opcode_modifier & Size32) != 0)
1299 ^ (flag_code == CODE_16BIT)))
1300 {
1301 as_bad (_("redundant %s prefix"),
1302 current_templates->start->name);
1303 return;
1304 }
1305 /* Add prefix, checking for repeated prefixes. */
1306 switch (add_prefix (current_templates->start->base_opcode))
1307 {
1308 case 0:
1309 return;
1310 case 2:
1311 expecting_string_instruction = current_templates->start->name;
1312 break;
1313 }
1314 /* Skip past PREFIX_SEPARATOR and reset token_start. */
1315 token_start = ++l;
1316 }
1317 else
1318 break;
1319 }
1320
1321 if (!current_templates)
1322 {
1323 /* See if we can get a match by trimming off a suffix. */
1324 switch (mnem_p[-1])
1325 {
1326 case WORD_MNEM_SUFFIX:
1327 case BYTE_MNEM_SUFFIX:
1328 case QWORD_MNEM_SUFFIX:
1329 i.suffix = mnem_p[-1];
1330 mnem_p[-1] = '\0';
1331 current_templates = hash_find (op_hash, mnemonic);
1332 break;
1333 case SHORT_MNEM_SUFFIX:
1334 case LONG_MNEM_SUFFIX:
1335 if (!intel_syntax)
1336 {
1337 i.suffix = mnem_p[-1];
1338 mnem_p[-1] = '\0';
1339 current_templates = hash_find (op_hash, mnemonic);
1340 }
1341 break;
1342
1343 /* Intel Syntax. */
1344 case 'd':
1345 if (intel_syntax)
1346 {
1347 if (intel_float_operand (mnemonic))
1348 i.suffix = SHORT_MNEM_SUFFIX;
1349 else
1350 i.suffix = LONG_MNEM_SUFFIX;
1351 mnem_p[-1] = '\0';
1352 current_templates = hash_find (op_hash, mnemonic);
1353 }
1354 break;
1355 }
1356 if (!current_templates)
1357 {
1358 as_bad (_("no such instruction: `%s'"), token_start);
1359 return;
1360 }
1361 }
1362
1363 /* Check if instruction is supported on specified architecture. */
1364 if (cpu_arch_flags != 0)
1365 {
1366 if ((current_templates->start->cpu_flags & ~(Cpu64 | CpuNo64))
1367 & ~(cpu_arch_flags & ~(Cpu64 | CpuNo64)))
1368 {
1369 as_warn (_("`%s' is not supported on `%s'"),
1370 current_templates->start->name, cpu_arch_name);
1371 }
1372 else if ((Cpu386 & ~cpu_arch_flags) && (flag_code != CODE_16BIT))
1373 {
1374 as_warn (_("use .code16 to ensure correct addressing mode"));
1375 }
1376 }
1377
1378 /* Check for rep/repne without a string instruction. */
1379 if (expecting_string_instruction
1380 && !(current_templates->start->opcode_modifier & IsString))
1381 {
1382 as_bad (_("expecting string instruction after `%s'"),
1383 expecting_string_instruction);
1384 return;
1385 }
1386
1387 /* There may be operands to parse. */
1388 if (*l != END_OF_INSN)
1389 {
1390 /* 1 if operand is pending after ','. */
1391 unsigned int expecting_operand = 0;
1392
1393 /* Non-zero if operand parens not balanced. */
1394 unsigned int paren_not_balanced;
1395
1396 do
1397 {
1398 /* Skip optional white space before operand. */
1399 if (is_space_char (*l))
1400 ++l;
1401 if (!is_operand_char (*l) && *l != END_OF_INSN)
1402 {
1403 as_bad (_("invalid character %s before operand %d"),
1404 output_invalid (*l),
1405 i.operands + 1);
1406 return;
1407 }
1408 token_start = l; /* after white space */
1409 paren_not_balanced = 0;
1410 while (paren_not_balanced || *l != ',')
1411 {
1412 if (*l == END_OF_INSN)
1413 {
1414 if (paren_not_balanced)
1415 {
1416 if (!intel_syntax)
1417 as_bad (_("unbalanced parenthesis in operand %d."),
1418 i.operands + 1);
1419 else
1420 as_bad (_("unbalanced brackets in operand %d."),
1421 i.operands + 1);
1422 return;
1423 }
1424 else
1425 break; /* we are done */
1426 }
1427 else if (!is_operand_char (*l) && !is_space_char (*l))
1428 {
1429 as_bad (_("invalid character %s in operand %d"),
1430 output_invalid (*l),
1431 i.operands + 1);
1432 return;
1433 }
1434 if (!intel_syntax)
1435 {
1436 if (*l == '(')
1437 ++paren_not_balanced;
1438 if (*l == ')')
1439 --paren_not_balanced;
1440 }
1441 else
1442 {
1443 if (*l == '[')
1444 ++paren_not_balanced;
1445 if (*l == ']')
1446 --paren_not_balanced;
1447 }
1448 l++;
1449 }
1450 if (l != token_start)
1451 { /* Yes, we've read in another operand. */
1452 unsigned int operand_ok;
1453 this_operand = i.operands++;
1454 if (i.operands > MAX_OPERANDS)
1455 {
1456 as_bad (_("spurious operands; (%d operands/instruction max)"),
1457 MAX_OPERANDS);
1458 return;
1459 }
1460 /* Now parse operand adding info to 'i' as we go along. */
1461 END_STRING_AND_SAVE (l);
1462
1463 if (intel_syntax)
1464 operand_ok =
1465 i386_intel_operand (token_start,
1466 intel_float_operand (mnemonic));
1467 else
1468 operand_ok = i386_operand (token_start);
1469
1470 RESTORE_END_STRING (l);
1471 if (!operand_ok)
1472 return;
1473 }
1474 else
1475 {
1476 if (expecting_operand)
1477 {
1478 expecting_operand_after_comma:
1479 as_bad (_("expecting operand after ','; got nothing"));
1480 return;
1481 }
1482 if (*l == ',')
1483 {
1484 as_bad (_("expecting operand before ','; got nothing"));
1485 return;
1486 }
1487 }
1488
1489 /* Now *l must be either ',' or END_OF_INSN. */
1490 if (*l == ',')
1491 {
1492 if (*++l == END_OF_INSN)
1493 {
1494 /* Just skip it, if it's \n complain. */
1495 goto expecting_operand_after_comma;
1496 }
1497 expecting_operand = 1;
1498 }
1499 }
1500 while (*l != END_OF_INSN);
1501 }
1502 }
1503
1504 /* Now we've parsed the mnemonic into a set of templates, and have the
1505 operands at hand.
1506
1507 Next, we find a template that matches the given insn,
1508 making sure the overlap of the given operands types is consistent
1509 with the template operand types. */
1510
1511 #define MATCH(overlap, given, template) \
1512 ((overlap & ~JumpAbsolute) \
1513 && ((given) & (BaseIndex|JumpAbsolute)) == ((overlap) & (BaseIndex|JumpAbsolute)))
1514
1515 /* If given types r0 and r1 are registers they must be of the same type
1516 unless the expected operand type register overlap is null.
1517 Note that Acc in a template matches every size of reg. */
1518 #define CONSISTENT_REGISTER_MATCH(m0, g0, t0, m1, g1, t1) \
1519 ( ((g0) & Reg) == 0 || ((g1) & Reg) == 0 || \
1520 ((g0) & Reg) == ((g1) & Reg) || \
1521 ((((m0) & Acc) ? Reg : (t0)) & (((m1) & Acc) ? Reg : (t1)) & Reg) == 0 )
1522
1523 {
1524 register unsigned int overlap0, overlap1;
1525 unsigned int overlap2;
1526 unsigned int found_reverse_match;
1527 int suffix_check;
1528
1529 /* All intel opcodes have reversed operands except for "bound" and
1530 "enter". We also don't reverse intersegment "jmp" and "call"
1531 instructions with 2 immediate operands so that the immediate segment
1532 precedes the offset, as it does when in AT&T mode. "enter" and the
1533 intersegment "jmp" and "call" instructions are the only ones that
1534 have two immediate operands. */
1535 if (intel_syntax && i.operands > 1
1536 && (strcmp (mnemonic, "bound") != 0)
1537 && !((i.types[0] & Imm) && (i.types[1] & Imm)))
1538 {
1539 union i386_op temp_op;
1540 unsigned int temp_type;
1541 RELOC_ENUM temp_reloc;
1542 int xchg1 = 0;
1543 int xchg2 = 0;
1544
1545 if (i.operands == 2)
1546 {
1547 xchg1 = 0;
1548 xchg2 = 1;
1549 }
1550 else if (i.operands == 3)
1551 {
1552 xchg1 = 0;
1553 xchg2 = 2;
1554 }
1555 temp_type = i.types[xchg2];
1556 i.types[xchg2] = i.types[xchg1];
1557 i.types[xchg1] = temp_type;
1558 temp_op = i.op[xchg2];
1559 i.op[xchg2] = i.op[xchg1];
1560 i.op[xchg1] = temp_op;
1561 temp_reloc = i.reloc[xchg2];
1562 i.reloc[xchg2] = i.reloc[xchg1];
1563 i.reloc[xchg1] = temp_reloc;
1564
1565 if (i.mem_operands == 2)
1566 {
1567 const seg_entry *temp_seg;
1568 temp_seg = i.seg[0];
1569 i.seg[0] = i.seg[1];
1570 i.seg[1] = temp_seg;
1571 }
1572 }
1573
1574 if (i.imm_operands)
1575 {
1576 /* Try to ensure constant immediates are represented in the smallest
1577 opcode possible. */
1578 char guess_suffix = 0;
1579 int op;
1580
1581 if (i.suffix)
1582 guess_suffix = i.suffix;
1583 else if (i.reg_operands)
1584 {
1585 /* Figure out a suffix from the last register operand specified.
1586 We can't do this properly yet, ie. excluding InOutPortReg,
1587 but the following works for instructions with immediates.
1588 In any case, we can't set i.suffix yet. */
1589 for (op = i.operands; --op >= 0;)
1590 if (i.types[op] & Reg)
1591 {
1592 if (i.types[op] & Reg8)
1593 guess_suffix = BYTE_MNEM_SUFFIX;
1594 else if (i.types[op] & Reg16)
1595 guess_suffix = WORD_MNEM_SUFFIX;
1596 else if (i.types[op] & Reg32)
1597 guess_suffix = LONG_MNEM_SUFFIX;
1598 else if (i.types[op] & Reg64)
1599 guess_suffix = QWORD_MNEM_SUFFIX;
1600 break;
1601 }
1602 }
1603 else if ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0))
1604 guess_suffix = WORD_MNEM_SUFFIX;
1605
1606 for (op = i.operands; --op >= 0;)
1607 if (i.types[op] & Imm)
1608 {
1609 switch (i.op[op].imms->X_op)
1610 {
1611 case O_constant:
1612 /* If a suffix is given, this operand may be shortened. */
1613 switch (guess_suffix)
1614 {
1615 case LONG_MNEM_SUFFIX:
1616 i.types[op] |= Imm32 | Imm64;
1617 break;
1618 case WORD_MNEM_SUFFIX:
1619 i.types[op] |= Imm16 | Imm32S | Imm32 | Imm64;
1620 break;
1621 case BYTE_MNEM_SUFFIX:
1622 i.types[op] |= Imm16 | Imm8 | Imm8S | Imm32S | Imm32 | Imm64;
1623 break;
1624 }
1625
1626 /* If this operand is at most 16 bits, convert it
1627 to a signed 16 bit number before trying to see
1628 whether it will fit in an even smaller size.
1629 This allows a 16-bit operand such as $0xffe0 to
1630 be recognised as within Imm8S range. */
1631 if ((i.types[op] & Imm16)
1632 && (i.op[op].imms->X_add_number & ~(offsetT) 0xffff) == 0)
1633 {
1634 i.op[op].imms->X_add_number =
1635 (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000);
1636 }
1637 if ((i.types[op] & Imm32)
1638 && (i.op[op].imms->X_add_number & ~(((offsetT) 2 << 31) - 1)) == 0)
1639 {
1640 i.op[op].imms->X_add_number =
1641 (i.op[op].imms->X_add_number ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
1642 }
1643 i.types[op] |= smallest_imm_type (i.op[op].imms->X_add_number);
1644 /* We must avoid matching of Imm32 templates when 64bit only immediate is available. */
1645 if (guess_suffix == QWORD_MNEM_SUFFIX)
1646 i.types[op] &= ~Imm32;
1647 break;
1648 case O_absent:
1649 case O_register:
1650 abort ();
1651 /* Symbols and expressions. */
1652 default:
1653 /* Convert symbolic operand to proper sizes for matching. */
1654 switch (guess_suffix)
1655 {
1656 case QWORD_MNEM_SUFFIX:
1657 i.types[op] = Imm64 | Imm32S;
1658 break;
1659 case LONG_MNEM_SUFFIX:
1660 i.types[op] = Imm32 | Imm64;
1661 break;
1662 case WORD_MNEM_SUFFIX:
1663 i.types[op] = Imm16 | Imm32 | Imm64;
1664 break;
1665 break;
1666 case BYTE_MNEM_SUFFIX:
1667 i.types[op] = Imm8 | Imm8S | Imm16 | Imm32S | Imm32;
1668 break;
1669 break;
1670 }
1671 break;
1672 }
1673 }
1674 }
1675
1676 if (i.disp_operands)
1677 {
1678 /* Try to use the smallest displacement type too. */
1679 int op;
1680
1681 for (op = i.operands; --op >= 0;)
1682 if ((i.types[op] & Disp)
1683 && i.op[op].disps->X_op == O_constant)
1684 {
1685 offsetT disp = i.op[op].disps->X_add_number;
1686
1687 if (i.types[op] & Disp16)
1688 {
1689 /* We know this operand is at most 16 bits, so
1690 convert to a signed 16 bit number before trying
1691 to see whether it will fit in an even smaller
1692 size. */
1693
1694 disp = (((disp & 0xffff) ^ 0x8000) - 0x8000);
1695 }
1696 else if (i.types[op] & Disp32)
1697 {
1698 /* We know this operand is at most 32 bits, so convert to a
1699 signed 32 bit number before trying to see whether it will
1700 fit in an even smaller size. */
1701 disp &= (((offsetT) 2 << 31) - 1);
1702 disp = (disp ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
1703 }
1704 if (flag_code == CODE_64BIT)
1705 {
1706 if (fits_in_signed_long (disp))
1707 i.types[op] |= Disp32S;
1708 if (fits_in_unsigned_long (disp))
1709 i.types[op] |= Disp32;
1710 }
1711 if ((i.types[op] & (Disp32 | Disp32S | Disp16))
1712 && fits_in_signed_byte (disp))
1713 i.types[op] |= Disp8;
1714 }
1715 }
1716
1717 overlap0 = 0;
1718 overlap1 = 0;
1719 overlap2 = 0;
1720 found_reverse_match = 0;
1721 suffix_check = (i.suffix == BYTE_MNEM_SUFFIX
1722 ? No_bSuf
1723 : (i.suffix == WORD_MNEM_SUFFIX
1724 ? No_wSuf
1725 : (i.suffix == SHORT_MNEM_SUFFIX
1726 ? No_sSuf
1727 : (i.suffix == LONG_MNEM_SUFFIX
1728 ? No_lSuf
1729 : (i.suffix == QWORD_MNEM_SUFFIX
1730 ? No_qSuf
1731 : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX ? No_xSuf : 0))))));
1732
1733 for (t = current_templates->start;
1734 t < current_templates->end;
1735 t++)
1736 {
1737 /* Must have right number of operands. */
1738 if (i.operands != t->operands)
1739 continue;
1740
1741 /* Check the suffix, except for some instructions in intel mode. */
1742 if ((t->opcode_modifier & suffix_check)
1743 && !(intel_syntax
1744 && (t->opcode_modifier & IgnoreSize))
1745 && !(intel_syntax
1746 && t->base_opcode == 0xd9
1747 && (t->extension_opcode == 5 /* 0xd9,5 "fldcw" */
1748 || t->extension_opcode == 7))) /* 0xd9,7 "f{n}stcw" */
1749 continue;
1750
1751 /* Do not verify operands when there are none. */
1752 else if (!t->operands)
1753 {
1754 if (t->cpu_flags & ~cpu_arch_flags)
1755 continue;
1756 /* We've found a match; break out of loop. */
1757 break;
1758 }
1759
1760 overlap0 = i.types[0] & t->operand_types[0];
1761 switch (t->operands)
1762 {
1763 case 1:
1764 if (!MATCH (overlap0, i.types[0], t->operand_types[0]))
1765 continue;
1766 break;
1767 case 2:
1768 case 3:
1769 overlap1 = i.types[1] & t->operand_types[1];
1770 if (!MATCH (overlap0, i.types[0], t->operand_types[0])
1771 || !MATCH (overlap1, i.types[1], t->operand_types[1])
1772 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
1773 t->operand_types[0],
1774 overlap1, i.types[1],
1775 t->operand_types[1]))
1776 {
1777 /* Check if other direction is valid ... */
1778 if ((t->opcode_modifier & (D|FloatD)) == 0)
1779 continue;
1780
1781 /* Try reversing direction of operands. */
1782 overlap0 = i.types[0] & t->operand_types[1];
1783 overlap1 = i.types[1] & t->operand_types[0];
1784 if (!MATCH (overlap0, i.types[0], t->operand_types[1])
1785 || !MATCH (overlap1, i.types[1], t->operand_types[0])
1786 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
1787 t->operand_types[1],
1788 overlap1, i.types[1],
1789 t->operand_types[0]))
1790 {
1791 /* Does not match either direction. */
1792 continue;
1793 }
1794 /* found_reverse_match holds which of D or FloatDR
1795 we've found. */
1796 found_reverse_match = t->opcode_modifier & (D|FloatDR);
1797 }
1798 /* Found a forward 2 operand match here. */
1799 else if (t->operands == 3)
1800 {
1801 /* Here we make use of the fact that there are no
1802 reverse match 3 operand instructions, and all 3
1803 operand instructions only need to be checked for
1804 register consistency between operands 2 and 3. */
1805 overlap2 = i.types[2] & t->operand_types[2];
1806 if (!MATCH (overlap2, i.types[2], t->operand_types[2])
1807 || !CONSISTENT_REGISTER_MATCH (overlap1, i.types[1],
1808 t->operand_types[1],
1809 overlap2, i.types[2],
1810 t->operand_types[2]))
1811
1812 continue;
1813 }
1814 /* Found either forward/reverse 2 or 3 operand match here:
1815 slip through to break. */
1816 }
1817 if (t->cpu_flags & ~cpu_arch_flags)
1818 {
1819 found_reverse_match = 0;
1820 continue;
1821 }
1822 /* We've found a match; break out of loop. */
1823 break;
1824 }
1825 if (t == current_templates->end)
1826 {
1827 /* We found no match. */
1828 as_bad (_("suffix or operands invalid for `%s'"),
1829 current_templates->start->name);
1830 return;
1831 }
1832
1833 if (!quiet_warnings)
1834 {
1835 if (!intel_syntax
1836 && ((i.types[0] & JumpAbsolute)
1837 != (t->operand_types[0] & JumpAbsolute)))
1838 {
1839 as_warn (_("indirect %s without `*'"), t->name);
1840 }
1841
1842 if ((t->opcode_modifier & (IsPrefix|IgnoreSize))
1843 == (IsPrefix|IgnoreSize))
1844 {
1845 /* Warn them that a data or address size prefix doesn't
1846 affect assembly of the next line of code. */
1847 as_warn (_("stand-alone `%s' prefix"), t->name);
1848 }
1849 }
1850
1851 /* Copy the template we found. */
1852 i.tm = *t;
1853 if (found_reverse_match)
1854 {
1855 /* If we found a reverse match we must alter the opcode
1856 direction bit. found_reverse_match holds bits to change
1857 (different for int & float insns). */
1858
1859 i.tm.base_opcode ^= found_reverse_match;
1860
1861 i.tm.operand_types[0] = t->operand_types[1];
1862 i.tm.operand_types[1] = t->operand_types[0];
1863 }
1864
1865 /* Undo SYSV386_COMPAT brokenness when in Intel mode. See i386.h */
1866 if (SYSV386_COMPAT
1867 && intel_syntax
1868 && (i.tm.base_opcode & 0xfffffde0) == 0xdce0)
1869 i.tm.base_opcode ^= FloatR;
1870
1871 if (i.tm.opcode_modifier & FWait)
1872 if (! add_prefix (FWAIT_OPCODE))
1873 return;
1874
1875 /* Check string instruction segment overrides. */
1876 if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0)
1877 {
1878 int mem_op = (i.types[0] & AnyMem) ? 0 : 1;
1879 if ((i.tm.operand_types[mem_op] & EsSeg) != 0)
1880 {
1881 if (i.seg[0] != NULL && i.seg[0] != &es)
1882 {
1883 as_bad (_("`%s' operand %d must use `%%es' segment"),
1884 i.tm.name,
1885 mem_op + 1);
1886 return;
1887 }
1888 /* There's only ever one segment override allowed per instruction.
1889 This instruction possibly has a legal segment override on the
1890 second operand, so copy the segment to where non-string
1891 instructions store it, allowing common code. */
1892 i.seg[0] = i.seg[1];
1893 }
1894 else if ((i.tm.operand_types[mem_op + 1] & EsSeg) != 0)
1895 {
1896 if (i.seg[1] != NULL && i.seg[1] != &es)
1897 {
1898 as_bad (_("`%s' operand %d must use `%%es' segment"),
1899 i.tm.name,
1900 mem_op + 2);
1901 return;
1902 }
1903 }
1904 }
1905
1906 if (i.reg_operands && flag_code < CODE_64BIT)
1907 {
1908 int op;
1909 for (op = i.operands; --op >= 0;)
1910 if ((i.types[op] & Reg)
1911 && (i.op[op].regs->reg_flags & (RegRex64|RegRex)))
1912 {
1913 as_bad (_("Extended register `%%%s' available only in 64bit mode."),
1914 i.op[op].regs->reg_name);
1915 return;
1916 }
1917 }
1918
1919 /* If matched instruction specifies an explicit instruction mnemonic
1920 suffix, use it. */
1921 if (i.tm.opcode_modifier & (Size16 | Size32 | Size64))
1922 {
1923 if (i.tm.opcode_modifier & Size16)
1924 i.suffix = WORD_MNEM_SUFFIX;
1925 else if (i.tm.opcode_modifier & Size64)
1926 i.suffix = QWORD_MNEM_SUFFIX;
1927 else
1928 i.suffix = LONG_MNEM_SUFFIX;
1929 }
1930 else if (i.reg_operands)
1931 {
1932 /* If there's no instruction mnemonic suffix we try to invent one
1933 based on register operands. */
1934 if (!i.suffix)
1935 {
1936 /* We take i.suffix from the last register operand specified,
1937 Destination register type is more significant than source
1938 register type. */
1939 int op;
1940 for (op = i.operands; --op >= 0;)
1941 if ((i.types[op] & Reg)
1942 && !(i.tm.operand_types[op] & InOutPortReg))
1943 {
1944 i.suffix = ((i.types[op] & Reg8) ? BYTE_MNEM_SUFFIX :
1945 (i.types[op] & Reg16) ? WORD_MNEM_SUFFIX :
1946 (i.types[op] & Reg64) ? QWORD_MNEM_SUFFIX :
1947 LONG_MNEM_SUFFIX);
1948 break;
1949 }
1950 }
1951 else if (i.suffix == BYTE_MNEM_SUFFIX)
1952 {
1953 int op;
1954 for (op = i.operands; --op >= 0;)
1955 {
1956 /* If this is an eight bit register, it's OK. If it's
1957 the 16 or 32 bit version of an eight bit register,
1958 we will just use the low portion, and that's OK too. */
1959 if (i.types[op] & Reg8)
1960 continue;
1961
1962 /* movzx and movsx should not generate this warning. */
1963 if (intel_syntax
1964 && (i.tm.base_opcode == 0xfb7
1965 || i.tm.base_opcode == 0xfb6
1966 || i.tm.base_opcode == 0x63
1967 || i.tm.base_opcode == 0xfbe
1968 || i.tm.base_opcode == 0xfbf))
1969 continue;
1970
1971 if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4
1972 #if 0
1973 /* Check that the template allows eight bit regs
1974 This kills insns such as `orb $1,%edx', which
1975 maybe should be allowed. */
1976 && (i.tm.operand_types[op] & (Reg8|InOutPortReg))
1977 #endif
1978 )
1979 {
1980 /* Prohibit these changes in the 64bit mode, since
1981 the lowering is more complicated. */
1982 if (flag_code == CODE_64BIT
1983 && (i.tm.operand_types[op] & InOutPortReg) == 0)
1984 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
1985 i.op[op].regs->reg_name,
1986 i.suffix);
1987 #if REGISTER_WARNINGS
1988 if (!quiet_warnings
1989 && (i.tm.operand_types[op] & InOutPortReg) == 0)
1990 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
1991 (i.op[op].regs
1992 + (i.types[op] & Reg16
1993 ? REGNAM_AL - REGNAM_AX
1994 : REGNAM_AL - REGNAM_EAX))->reg_name,
1995 i.op[op].regs->reg_name,
1996 i.suffix);
1997 #endif
1998 continue;
1999 }
2000 /* Any other register is bad. */
2001 if (i.types[op] & (Reg | RegMMX | RegXMM
2002 | SReg2 | SReg3
2003 | Control | Debug | Test
2004 | FloatReg | FloatAcc))
2005 {
2006 as_bad (_("`%%%s' not allowed with `%s%c'"),
2007 i.op[op].regs->reg_name,
2008 i.tm.name,
2009 i.suffix);
2010 return;
2011 }
2012 }
2013 }
2014 else if (i.suffix == LONG_MNEM_SUFFIX)
2015 {
2016 int op;
2017
2018 for (op = i.operands; --op >= 0;)
2019 /* Reject eight bit registers, except where the template
2020 requires them. (eg. movzb) */
2021 if ((i.types[op] & Reg8) != 0
2022 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2023 {
2024 as_bad (_("`%%%s' not allowed with `%s%c'"),
2025 i.op[op].regs->reg_name,
2026 i.tm.name,
2027 i.suffix);
2028 return;
2029 }
2030 /* Warn if the e prefix on a general reg is missing. */
2031 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2032 && (i.types[op] & Reg16) != 0
2033 && (i.tm.operand_types[op] & (Reg32|Acc)) != 0)
2034 {
2035 /* Prohibit these changes in the 64bit mode, since
2036 the lowering is more complicated. */
2037 if (flag_code == CODE_64BIT)
2038 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2039 i.op[op].regs->reg_name,
2040 i.suffix);
2041 #if REGISTER_WARNINGS
2042 else
2043 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2044 (i.op[op].regs + REGNAM_EAX - REGNAM_AX)->reg_name,
2045 i.op[op].regs->reg_name,
2046 i.suffix);
2047 #endif
2048 }
2049 /* Warn if the r prefix on a general reg is missing. */
2050 else if ((i.types[op] & Reg64) != 0
2051 && (i.tm.operand_types[op] & (Reg32|Acc)) != 0)
2052 {
2053 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2054 i.op[op].regs->reg_name,
2055 i.suffix);
2056 }
2057 }
2058 else if (i.suffix == QWORD_MNEM_SUFFIX)
2059 {
2060 int op;
2061
2062 for (op = i.operands; --op >= 0; )
2063 /* Reject eight bit registers, except where the template
2064 requires them. (eg. movzb) */
2065 if ((i.types[op] & Reg8) != 0
2066 && (i.tm.operand_types[op] & (Reg16|Reg32|Acc)) != 0)
2067 {
2068 as_bad (_("`%%%s' not allowed with `%s%c'"),
2069 i.op[op].regs->reg_name,
2070 i.tm.name,
2071 i.suffix);
2072 return;
2073 }
2074 /* Warn if the e prefix on a general reg is missing. */
2075 else if (((i.types[op] & Reg16) != 0
2076 || (i.types[op] & Reg32) != 0)
2077 && (i.tm.operand_types[op] & (Reg32|Acc)) != 0)
2078 {
2079 /* Prohibit these changes in the 64bit mode, since
2080 the lowering is more complicated. */
2081 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2082 i.op[op].regs->reg_name,
2083 i.suffix);
2084 }
2085 }
2086 else if (i.suffix == WORD_MNEM_SUFFIX)
2087 {
2088 int op;
2089 for (op = i.operands; --op >= 0;)
2090 /* Reject eight bit registers, except where the template
2091 requires them. (eg. movzb) */
2092 if ((i.types[op] & Reg8) != 0
2093 && (i.tm.operand_types[op] & (Reg16|Reg32|Acc)) != 0)
2094 {
2095 as_bad (_("`%%%s' not allowed with `%s%c'"),
2096 i.op[op].regs->reg_name,
2097 i.tm.name,
2098 i.suffix);
2099 return;
2100 }
2101 /* Warn if the e prefix on a general reg is present. */
2102 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2103 && (i.types[op] & Reg32) != 0
2104 && (i.tm.operand_types[op] & (Reg16|Acc)) != 0)
2105 {
2106 /* Prohibit these changes in the 64bit mode, since
2107 the lowering is more complicated. */
2108 if (flag_code == CODE_64BIT)
2109 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2110 i.op[op].regs->reg_name,
2111 i.suffix);
2112 else
2113 #if REGISTER_WARNINGS
2114 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2115 (i.op[op].regs + REGNAM_AX - REGNAM_EAX)->reg_name,
2116 i.op[op].regs->reg_name,
2117 i.suffix);
2118 #endif
2119 }
2120 }
2121 else if (intel_syntax && (i.tm.opcode_modifier & IgnoreSize))
2122 /* Do nothing if the instruction is going to ignore the prefix. */
2123 ;
2124 else
2125 abort ();
2126 }
2127 else if ((i.tm.opcode_modifier & DefaultSize) && !i.suffix)
2128 {
2129 i.suffix = stackop_size;
2130 }
2131 /* Make still unresolved immediate matches conform to size of immediate
2132 given in i.suffix. Note: overlap2 cannot be an immediate! */
2133 if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S))
2134 && overlap0 != Imm8 && overlap0 != Imm8S
2135 && overlap0 != Imm16 && overlap0 != Imm32S
2136 && overlap0 != Imm32 && overlap0 != Imm64)
2137 {
2138 if (i.suffix)
2139 {
2140 overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX ? (Imm8 | Imm8S) :
2141 (i.suffix == WORD_MNEM_SUFFIX ? Imm16 :
2142 (i.suffix == QWORD_MNEM_SUFFIX ? Imm64 | Imm32S : Imm32)));
2143 }
2144 else if (overlap0 == (Imm16 | Imm32S | Imm32)
2145 || overlap0 == (Imm16 | Imm32)
2146 || overlap0 == (Imm16 | Imm32S))
2147 {
2148 overlap0 =
2149 ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)) ? Imm16 : Imm32S;
2150 }
2151 if (overlap0 != Imm8 && overlap0 != Imm8S
2152 && overlap0 != Imm16 && overlap0 != Imm32S
2153 && overlap0 != Imm32 && overlap0 != Imm64)
2154 {
2155 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size"));
2156 return;
2157 }
2158 }
2159 if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32))
2160 && overlap1 != Imm8 && overlap1 != Imm8S
2161 && overlap1 != Imm16 && overlap1 != Imm32S
2162 && overlap1 != Imm32 && overlap1 != Imm64)
2163 {
2164 if (i.suffix)
2165 {
2166 overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX ? (Imm8 | Imm8S) :
2167 (i.suffix == WORD_MNEM_SUFFIX ? Imm16 :
2168 (i.suffix == QWORD_MNEM_SUFFIX ? Imm64 | Imm32S : Imm32)));
2169 }
2170 else if (overlap1 == (Imm16 | Imm32 | Imm32S)
2171 || overlap1 == (Imm16 | Imm32)
2172 || overlap1 == (Imm16 | Imm32S))
2173 {
2174 overlap1 =
2175 ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)) ? Imm16 : Imm32S;
2176 }
2177 if (overlap1 != Imm8 && overlap1 != Imm8S
2178 && overlap1 != Imm16 && overlap1 != Imm32S
2179 && overlap1 != Imm32 && overlap1 != Imm64)
2180 {
2181 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size %x %c"),overlap1, i.suffix);
2182 return;
2183 }
2184 }
2185 assert ((overlap2 & Imm) == 0);
2186
2187 i.types[0] = overlap0;
2188 if (overlap0 & ImplicitRegister)
2189 i.reg_operands--;
2190 if (overlap0 & Imm1)
2191 i.imm_operands = 0; /* kludge for shift insns. */
2192
2193 i.types[1] = overlap1;
2194 if (overlap1 & ImplicitRegister)
2195 i.reg_operands--;
2196
2197 i.types[2] = overlap2;
2198 if (overlap2 & ImplicitRegister)
2199 i.reg_operands--;
2200
2201 /* Finalize opcode. First, we change the opcode based on the operand
2202 size given by i.suffix: We need not change things for byte insns. */
2203
2204 if (!i.suffix && (i.tm.opcode_modifier & W))
2205 {
2206 as_bad (_("no instruction mnemonic suffix given and no register operands; can't size instruction"));
2207 return;
2208 }
2209
2210 /* For movzx and movsx, need to check the register type. */
2211 if (intel_syntax
2212 && (i.tm.base_opcode == 0xfb6 || i.tm.base_opcode == 0xfbe))
2213 if (i.suffix && i.suffix == BYTE_MNEM_SUFFIX)
2214 {
2215 unsigned int prefix = DATA_PREFIX_OPCODE;
2216
2217 if ((i.op[1].regs->reg_type & Reg16) != 0)
2218 if (!add_prefix (prefix))
2219 return;
2220 }
2221
2222 if (i.suffix && i.suffix != BYTE_MNEM_SUFFIX)
2223 {
2224 /* It's not a byte, select word/dword operation. */
2225 if (i.tm.opcode_modifier & W)
2226 {
2227 if (i.tm.opcode_modifier & ShortForm)
2228 i.tm.base_opcode |= 8;
2229 else
2230 i.tm.base_opcode |= 1;
2231 }
2232 /* Now select between word & dword operations via the operand
2233 size prefix, except for instructions that will ignore this
2234 prefix anyway. */
2235 if (i.suffix != QWORD_MNEM_SUFFIX
2236 && (i.suffix == LONG_MNEM_SUFFIX) == (flag_code == CODE_16BIT)
2237 && !(i.tm.opcode_modifier & IgnoreSize))
2238 {
2239 unsigned int prefix = DATA_PREFIX_OPCODE;
2240 if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */
2241 prefix = ADDR_PREFIX_OPCODE;
2242
2243 if (! add_prefix (prefix))
2244 return;
2245 }
2246
2247 /* Set mode64 for an operand. */
2248 if (i.suffix == QWORD_MNEM_SUFFIX
2249 && !(i.tm.opcode_modifier & NoRex64))
2250 {
2251 i.rex.mode64 = 1;
2252 if (flag_code < CODE_64BIT)
2253 {
2254 as_bad (_("64bit operations available only in 64bit modes."));
2255 return;
2256 }
2257 }
2258
2259 /* Size floating point instruction. */
2260 if (i.suffix == LONG_MNEM_SUFFIX)
2261 {
2262 if (i.tm.opcode_modifier & FloatMF)
2263 i.tm.base_opcode ^= 4;
2264 }
2265 }
2266
2267 if (i.tm.opcode_modifier & ImmExt)
2268 {
2269 /* These AMD 3DNow! and Intel Katmai New Instructions have an
2270 opcode suffix which is coded in the same place as an 8-bit
2271 immediate field would be. Here we fake an 8-bit immediate
2272 operand from the opcode suffix stored in tm.extension_opcode. */
2273
2274 expressionS *exp;
2275
2276 assert (i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS);
2277
2278 exp = &im_expressions[i.imm_operands++];
2279 i.op[i.operands].imms = exp;
2280 i.types[i.operands++] = Imm8;
2281 exp->X_op = O_constant;
2282 exp->X_add_number = i.tm.extension_opcode;
2283 i.tm.extension_opcode = None;
2284 }
2285
2286 /* For insns with operands there are more diddles to do to the opcode. */
2287 if (i.operands)
2288 {
2289 /* Default segment register this instruction will use
2290 for memory accesses. 0 means unknown.
2291 This is only for optimizing out unnecessary segment overrides. */
2292 const seg_entry *default_seg = 0;
2293
2294 /* The imul $imm, %reg instruction is converted into
2295 imul $imm, %reg, %reg, and the clr %reg instruction
2296 is converted into xor %reg, %reg. */
2297 if (i.tm.opcode_modifier & regKludge)
2298 {
2299 unsigned int first_reg_op = (i.types[0] & Reg) ? 0 : 1;
2300 /* Pretend we saw the extra register operand. */
2301 assert (i.op[first_reg_op + 1].regs == 0);
2302 i.op[first_reg_op + 1].regs = i.op[first_reg_op].regs;
2303 i.types[first_reg_op + 1] = i.types[first_reg_op];
2304 i.reg_operands = 2;
2305 }
2306
2307 if (i.tm.opcode_modifier & ShortForm)
2308 {
2309 /* The register or float register operand is in operand 0 or 1. */
2310 unsigned int op = (i.types[0] & (Reg | FloatReg)) ? 0 : 1;
2311 /* Register goes in low 3 bits of opcode. */
2312 i.tm.base_opcode |= i.op[op].regs->reg_num;
2313 if (i.op[op].regs->reg_flags & RegRex)
2314 i.rex.extZ = 1;
2315 if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
2316 {
2317 /* Warn about some common errors, but press on regardless.
2318 The first case can be generated by gcc (<= 2.8.1). */
2319 if (i.operands == 2)
2320 {
2321 /* Reversed arguments on faddp, fsubp, etc. */
2322 as_warn (_("translating to `%s %%%s,%%%s'"), i.tm.name,
2323 i.op[1].regs->reg_name,
2324 i.op[0].regs->reg_name);
2325 }
2326 else
2327 {
2328 /* Extraneous `l' suffix on fp insn. */
2329 as_warn (_("translating to `%s %%%s'"), i.tm.name,
2330 i.op[0].regs->reg_name);
2331 }
2332 }
2333 }
2334 else if (i.tm.opcode_modifier & Modrm)
2335 {
2336 /* The opcode is completed (modulo i.tm.extension_opcode which
2337 must be put into the modrm byte).
2338 Now, we make the modrm & index base bytes based on all the
2339 info we've collected. */
2340
2341 /* i.reg_operands MUST be the number of real register operands;
2342 implicit registers do not count. */
2343 if (i.reg_operands == 2)
2344 {
2345 unsigned int source, dest;
2346 source = ((i.types[0]
2347 & (Reg | RegMMX | RegXMM
2348 | SReg2 | SReg3
2349 | Control | Debug | Test))
2350 ? 0 : 1);
2351 dest = source + 1;
2352
2353 i.rm.mode = 3;
2354 /* One of the register operands will be encoded in the
2355 i.tm.reg field, the other in the combined i.tm.mode
2356 and i.tm.regmem fields. If no form of this
2357 instruction supports a memory destination operand,
2358 then we assume the source operand may sometimes be
2359 a memory operand and so we need to store the
2360 destination in the i.rm.reg field. */
2361 if ((i.tm.operand_types[dest] & AnyMem) == 0)
2362 {
2363 i.rm.reg = i.op[dest].regs->reg_num;
2364 i.rm.regmem = i.op[source].regs->reg_num;
2365 if (i.op[dest].regs->reg_flags & RegRex)
2366 i.rex.extX = 1;
2367 if (i.op[source].regs->reg_flags & RegRex)
2368 i.rex.extZ = 1;
2369 }
2370 else
2371 {
2372 i.rm.reg = i.op[source].regs->reg_num;
2373 i.rm.regmem = i.op[dest].regs->reg_num;
2374 if (i.op[dest].regs->reg_flags & RegRex)
2375 i.rex.extZ = 1;
2376 if (i.op[source].regs->reg_flags & RegRex)
2377 i.rex.extX = 1;
2378 }
2379 }
2380 else
2381 { /* If it's not 2 reg operands... */
2382 if (i.mem_operands)
2383 {
2384 unsigned int fake_zero_displacement = 0;
2385 unsigned int op = ((i.types[0] & AnyMem)
2386 ? 0
2387 : (i.types[1] & AnyMem) ? 1 : 2);
2388
2389 default_seg = &ds;
2390
2391 if (! i.base_reg)
2392 {
2393 i.rm.mode = 0;
2394 if (! i.disp_operands)
2395 fake_zero_displacement = 1;
2396 if (! i.index_reg)
2397 {
2398 /* Operand is just <disp> */
2399 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
2400 {
2401 i.rm.regmem = NO_BASE_REGISTER_16;
2402 i.types[op] &= ~Disp;
2403 i.types[op] |= Disp16;
2404 }
2405 else if (flag_code != CODE_64BIT)
2406 {
2407 i.rm.regmem = NO_BASE_REGISTER;
2408 i.types[op] &= ~Disp;
2409 i.types[op] |= Disp32;
2410 }
2411 else
2412 {
2413 /* 64bit mode overwrites the 32bit
2414 absolute addressing by RIP relative
2415 addressing and absolute addressing
2416 is encoded by one of the redundant
2417 SIB forms. */
2418
2419 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2420 i.sib.base = NO_BASE_REGISTER;
2421 i.sib.index = NO_INDEX_REGISTER;
2422 i.types[op] &= ~Disp;
2423 i.types[op] |= Disp32S;
2424 }
2425 }
2426 else /* ! i.base_reg && i.index_reg */
2427 {
2428 i.sib.index = i.index_reg->reg_num;
2429 i.sib.base = NO_BASE_REGISTER;
2430 i.sib.scale = i.log2_scale_factor;
2431 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2432 i.types[op] &= ~Disp;
2433 if (flag_code != CODE_64BIT)
2434 i.types[op] |= Disp32; /* Must be 32 bit */
2435 else
2436 i.types[op] |= Disp32S;
2437 if (i.index_reg->reg_flags & RegRex)
2438 i.rex.extY = 1;
2439 }
2440 }
2441 /* RIP addressing for 64bit mode. */
2442 else if (i.base_reg->reg_type == BaseIndex)
2443 {
2444 i.rm.regmem = NO_BASE_REGISTER;
2445 i.types[op] &= ~Disp;
2446 i.types[op] |= Disp32S;
2447 i.flags[op] = Operand_PCrel;
2448 }
2449 else if (i.base_reg->reg_type & Reg16)
2450 {
2451 switch (i.base_reg->reg_num)
2452 {
2453 case 3: /* (%bx) */
2454 if (! i.index_reg)
2455 i.rm.regmem = 7;
2456 else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */
2457 i.rm.regmem = i.index_reg->reg_num - 6;
2458 break;
2459 case 5: /* (%bp) */
2460 default_seg = &ss;
2461 if (! i.index_reg)
2462 {
2463 i.rm.regmem = 6;
2464 if ((i.types[op] & Disp) == 0)
2465 {
2466 /* fake (%bp) into 0(%bp) */
2467 i.types[op] |= Disp8;
2468 fake_zero_displacement = 1;
2469 }
2470 }
2471 else /* (%bp,%si) -> 2, or (%bp,%di) -> 3 */
2472 i.rm.regmem = i.index_reg->reg_num - 6 + 2;
2473 break;
2474 default: /* (%si) -> 4 or (%di) -> 5 */
2475 i.rm.regmem = i.base_reg->reg_num - 6 + 4;
2476 }
2477 i.rm.mode = mode_from_disp_size (i.types[op]);
2478 }
2479 else /* i.base_reg and 32/64 bit mode */
2480 {
2481 if (flag_code == CODE_64BIT
2482 && (i.types[op] & Disp))
2483 {
2484 if (i.types[op] & Disp8)
2485 i.types[op] = Disp8 | Disp32S;
2486 else
2487 i.types[op] = Disp32S;
2488 }
2489 i.rm.regmem = i.base_reg->reg_num;
2490 if (i.base_reg->reg_flags & RegRex)
2491 i.rex.extZ = 1;
2492 i.sib.base = i.base_reg->reg_num;
2493 /* x86-64 ignores REX prefix bit here to avoid
2494 decoder complications. */
2495 if ((i.base_reg->reg_num & 7) == EBP_REG_NUM)
2496 {
2497 default_seg = &ss;
2498 if (i.disp_operands == 0)
2499 {
2500 fake_zero_displacement = 1;
2501 i.types[op] |= Disp8;
2502 }
2503 }
2504 else if (i.base_reg->reg_num == ESP_REG_NUM)
2505 {
2506 default_seg = &ss;
2507 }
2508 i.sib.scale = i.log2_scale_factor;
2509 if (! i.index_reg)
2510 {
2511 /* <disp>(%esp) becomes two byte modrm
2512 with no index register. We've already
2513 stored the code for esp in i.rm.regmem
2514 ie. ESCAPE_TO_TWO_BYTE_ADDRESSING. Any
2515 base register besides %esp will not use
2516 the extra modrm byte. */
2517 i.sib.index = NO_INDEX_REGISTER;
2518 #if ! SCALE1_WHEN_NO_INDEX
2519 /* Another case where we force the second
2520 modrm byte. */
2521 if (i.log2_scale_factor)
2522 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2523 #endif
2524 }
2525 else
2526 {
2527 i.sib.index = i.index_reg->reg_num;
2528 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2529 if (i.index_reg->reg_flags & RegRex)
2530 i.rex.extY = 1;
2531 }
2532 i.rm.mode = mode_from_disp_size (i.types[op]);
2533 }
2534
2535 if (fake_zero_displacement)
2536 {
2537 /* Fakes a zero displacement assuming that i.types[op]
2538 holds the correct displacement size. */
2539 expressionS *exp;
2540
2541 assert (i.op[op].disps == 0);
2542 exp = &disp_expressions[i.disp_operands++];
2543 i.op[op].disps = exp;
2544 exp->X_op = O_constant;
2545 exp->X_add_number = 0;
2546 exp->X_add_symbol = (symbolS *) 0;
2547 exp->X_op_symbol = (symbolS *) 0;
2548 }
2549 }
2550
2551 /* Fill in i.rm.reg or i.rm.regmem field with register
2552 operand (if any) based on i.tm.extension_opcode.
2553 Again, we must be careful to make sure that
2554 segment/control/debug/test/MMX registers are coded
2555 into the i.rm.reg field. */
2556 if (i.reg_operands)
2557 {
2558 unsigned int op =
2559 ((i.types[0]
2560 & (Reg | RegMMX | RegXMM
2561 | SReg2 | SReg3
2562 | Control | Debug | Test))
2563 ? 0
2564 : ((i.types[1]
2565 & (Reg | RegMMX | RegXMM
2566 | SReg2 | SReg3
2567 | Control | Debug | Test))
2568 ? 1
2569 : 2));
2570 /* If there is an extension opcode to put here, the
2571 register number must be put into the regmem field. */
2572 if (i.tm.extension_opcode != None)
2573 {
2574 i.rm.regmem = i.op[op].regs->reg_num;
2575 if (i.op[op].regs->reg_flags & RegRex)
2576 i.rex.extZ = 1;
2577 }
2578 else
2579 {
2580 i.rm.reg = i.op[op].regs->reg_num;
2581 if (i.op[op].regs->reg_flags & RegRex)
2582 i.rex.extX = 1;
2583 }
2584
2585 /* Now, if no memory operand has set i.rm.mode = 0, 1, 2
2586 we must set it to 3 to indicate this is a register
2587 operand in the regmem field. */
2588 if (!i.mem_operands)
2589 i.rm.mode = 3;
2590 }
2591
2592 /* Fill in i.rm.reg field with extension opcode (if any). */
2593 if (i.tm.extension_opcode != None)
2594 i.rm.reg = i.tm.extension_opcode;
2595 }
2596 }
2597 else if (i.tm.opcode_modifier & (Seg2ShortForm | Seg3ShortForm))
2598 {
2599 if (i.tm.base_opcode == POP_SEG_SHORT
2600 && i.op[0].regs->reg_num == 1)
2601 {
2602 as_bad (_("you can't `pop %%cs'"));
2603 return;
2604 }
2605 i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
2606 if (i.op[0].regs->reg_flags & RegRex)
2607 i.rex.extZ = 1;
2608 }
2609 else if ((i.tm.base_opcode & ~(D|W)) == MOV_AX_DISP32)
2610 {
2611 default_seg = &ds;
2612 }
2613 else if ((i.tm.opcode_modifier & IsString) != 0)
2614 {
2615 /* For the string instructions that allow a segment override
2616 on one of their operands, the default segment is ds. */
2617 default_seg = &ds;
2618 }
2619
2620 /* If a segment was explicitly specified,
2621 and the specified segment is not the default,
2622 use an opcode prefix to select it.
2623 If we never figured out what the default segment is,
2624 then default_seg will be zero at this point,
2625 and the specified segment prefix will always be used. */
2626 if ((i.seg[0]) && (i.seg[0] != default_seg))
2627 {
2628 if (! add_prefix (i.seg[0]->seg_prefix))
2629 return;
2630 }
2631 }
2632 else if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
2633 {
2634 /* UnixWare fsub no args is alias for fsubp, fadd -> faddp, etc. */
2635 as_warn (_("translating to `%sp'"), i.tm.name);
2636 }
2637 }
2638
2639 /* Handle conversion of 'int $3' --> special int3 insn. */
2640 if (i.tm.base_opcode == INT_OPCODE && i.op[0].imms->X_add_number == 3)
2641 {
2642 i.tm.base_opcode = INT3_OPCODE;
2643 i.imm_operands = 0;
2644 }
2645
2646 if ((i.tm.opcode_modifier & (Jump | JumpByte | JumpDword))
2647 && i.op[0].disps->X_op == O_constant)
2648 {
2649 /* Convert "jmp constant" (and "call constant") to a jump (call) to
2650 the absolute address given by the constant. Since ix86 jumps and
2651 calls are pc relative, we need to generate a reloc. */
2652 i.op[0].disps->X_add_symbol = &abs_symbol;
2653 i.op[0].disps->X_op = O_symbol;
2654 }
2655
2656 if (i.tm.opcode_modifier & Rex64)
2657 i.rex.mode64 = 1;
2658
2659 /* For 8bit registers we would need an empty rex prefix.
2660 Also in the case instruction is already having prefix,
2661 we need to convert old registers to new ones. */
2662
2663 if (((i.types[0] & Reg8) && (i.op[0].regs->reg_flags & RegRex64))
2664 || ((i.types[1] & Reg8) && (i.op[1].regs->reg_flags & RegRex64))
2665 || ((i.rex.mode64 || i.rex.extX || i.rex.extY || i.rex.extZ || i.rex.empty)
2666 && ((i.types[0] & Reg8) || (i.types[1] & Reg8))))
2667 {
2668 int x;
2669 i.rex.empty = 1;
2670 for (x = 0; x < 2; x++)
2671 {
2672 /* Look for 8bit operand that does use old registers. */
2673 if (i.types[x] & Reg8
2674 && !(i.op[x].regs->reg_flags & RegRex64))
2675 {
2676 /* In case it is "hi" register, give up. */
2677 if (i.op[x].regs->reg_num > 3)
2678 as_bad (_("Can't encode registers '%%%s' in the instruction requiring REX prefix.\n"),
2679 i.op[x].regs->reg_name);
2680
2681 /* Otherwise it is equivalent to the extended register.
2682 Since the encoding don't change this is merely cosmetical
2683 cleanup for debug output. */
2684
2685 i.op[x].regs = i.op[x].regs + 8;
2686 }
2687 }
2688 }
2689
2690 if (i.rex.mode64 || i.rex.extX || i.rex.extY || i.rex.extZ || i.rex.empty)
2691 add_prefix (0x40
2692 | (i.rex.mode64 ? 8 : 0)
2693 | (i.rex.extX ? 4 : 0)
2694 | (i.rex.extY ? 2 : 0)
2695 | (i.rex.extZ ? 1 : 0));
2696
2697 /* We are ready to output the insn. */
2698 {
2699 register char *p;
2700
2701 /* Tie dwarf2 debug info to the address at the start of the insn.
2702 We can't do this after the insn has been output as the current
2703 frag may have been closed off. eg. by frag_var. */
2704 dwarf2_emit_insn (0);
2705
2706 /* Output jumps. */
2707 if (i.tm.opcode_modifier & Jump)
2708 {
2709 int code16;
2710 int prefix;
2711
2712 code16 = 0;
2713 if (flag_code == CODE_16BIT)
2714 code16 = CODE16;
2715
2716 prefix = 0;
2717 if (i.prefix[DATA_PREFIX])
2718 {
2719 prefix = 1;
2720 i.prefixes -= 1;
2721 code16 ^= CODE16;
2722 }
2723 if (i.prefix[REX_PREFIX])
2724 {
2725 prefix++;
2726 i.prefixes--;
2727 }
2728
2729 if (i.prefixes != 0 && !intel_syntax)
2730 as_warn (_("skipping prefixes on this instruction"));
2731
2732 /* It's always a symbol; End frag & setup for relax.
2733 Make sure there is enough room in this frag for the largest
2734 instruction we may generate in md_convert_frag. This is 2
2735 bytes for the opcode and room for the prefix and largest
2736 displacement. */
2737 frag_grow (prefix + 2 + 4);
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 FRAG_APPEND_1_CHAR (ADDR_PREFIX_OPCODE);
2770 i.prefixes -= 1;
2771 }
2772 }
2773 else
2774 {
2775 int code16;
2776
2777 code16 = 0;
2778 if (flag_code == CODE_16BIT)
2779 code16 = CODE16;
2780
2781 if (i.prefix[DATA_PREFIX])
2782 {
2783 FRAG_APPEND_1_CHAR (DATA_PREFIX_OPCODE);
2784 i.prefixes -= 1;
2785 code16 ^= CODE16;
2786 }
2787
2788 size = 4;
2789 if (code16)
2790 size = 2;
2791 }
2792
2793 if (i.prefix[REX_PREFIX])
2794 {
2795 FRAG_APPEND_1_CHAR (i.prefix[REX_PREFIX]);
2796 i.prefixes -= 1;
2797 }
2798
2799 if (i.prefixes != 0 && !intel_syntax)
2800 as_warn (_("skipping prefixes on this instruction"));
2801
2802 if (fits_in_unsigned_byte (i.tm.base_opcode))
2803 {
2804 p = frag_more (1 + size);
2805 }
2806 else
2807 {
2808 /* Opcode can be at most two bytes. */
2809 p = frag_more (2 + size);
2810 *p++ = (i.tm.base_opcode >> 8) & 0xff;
2811 }
2812 *p++ = i.tm.base_opcode & 0xff;
2813
2814 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
2815 i.op[0].disps, 1, reloc (size, 1, 1, i.reloc[0]));
2816 }
2817 else if (i.tm.opcode_modifier & JumpInterSegment)
2818 {
2819 int size;
2820 int prefix;
2821 int code16;
2822
2823 code16 = 0;
2824 if (flag_code == CODE_16BIT)
2825 code16 = CODE16;
2826
2827 prefix = 0;
2828 if (i.prefix[DATA_PREFIX])
2829 {
2830 prefix = 1;
2831 i.prefixes -= 1;
2832 code16 ^= CODE16;
2833 }
2834 if (i.prefix[REX_PREFIX])
2835 {
2836 prefix++;
2837 i.prefixes -= 1;
2838 }
2839
2840 size = 4;
2841 if (code16)
2842 size = 2;
2843
2844 if (i.prefixes != 0 && !intel_syntax)
2845 as_warn (_("skipping prefixes on this instruction"));
2846
2847 /* 1 opcode; 2 segment; offset */
2848 p = frag_more (prefix + 1 + 2 + size);
2849
2850 if (i.prefix[DATA_PREFIX])
2851 *p++ = DATA_PREFIX_OPCODE;
2852
2853 if (i.prefix[REX_PREFIX])
2854 *p++ = i.prefix[REX_PREFIX];
2855
2856 *p++ = i.tm.base_opcode;
2857 if (i.op[1].imms->X_op == O_constant)
2858 {
2859 offsetT n = i.op[1].imms->X_add_number;
2860
2861 if (size == 2
2862 && !fits_in_unsigned_word (n)
2863 && !fits_in_signed_word (n))
2864 {
2865 as_bad (_("16-bit jump out of range"));
2866 return;
2867 }
2868 md_number_to_chars (p, n, size);
2869 }
2870 else
2871 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
2872 i.op[1].imms, 0, reloc (size, 0, 0, i.reloc[1]));
2873 if (i.op[0].imms->X_op != O_constant)
2874 as_bad (_("can't handle non absolute segment in `%s'"),
2875 i.tm.name);
2876 md_number_to_chars (p + size, (valueT) i.op[0].imms->X_add_number, 2);
2877 }
2878 else
2879 {
2880 /* Output normal instructions here. */
2881 unsigned char *q;
2882
2883 /* All opcodes on i386 have eighter 1 or 2 bytes. We may use third
2884 byte for the SSE instructions to specify prefix they require. */
2885 if (i.tm.base_opcode & 0xff0000)
2886 add_prefix ((i.tm.base_opcode >> 16) & 0xff);
2887
2888 /* The prefix bytes. */
2889 for (q = i.prefix;
2890 q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
2891 q++)
2892 {
2893 if (*q)
2894 {
2895 p = frag_more (1);
2896 md_number_to_chars (p, (valueT) *q, 1);
2897 }
2898 }
2899
2900 /* Now the opcode; be careful about word order here! */
2901 if (fits_in_unsigned_byte (i.tm.base_opcode))
2902 {
2903 FRAG_APPEND_1_CHAR (i.tm.base_opcode);
2904 }
2905 else
2906 {
2907 p = frag_more (2);
2908 /* Put out high byte first: can't use md_number_to_chars! */
2909 *p++ = (i.tm.base_opcode >> 8) & 0xff;
2910 *p = i.tm.base_opcode & 0xff;
2911 }
2912
2913 /* Now the modrm byte and sib byte (if present). */
2914 if (i.tm.opcode_modifier & Modrm)
2915 {
2916 p = frag_more (1);
2917 md_number_to_chars (p,
2918 (valueT) (i.rm.regmem << 0
2919 | i.rm.reg << 3
2920 | i.rm.mode << 6),
2921 1);
2922 /* If i.rm.regmem == ESP (4)
2923 && i.rm.mode != (Register mode)
2924 && not 16 bit
2925 ==> need second modrm byte. */
2926 if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING
2927 && i.rm.mode != 3
2928 && !(i.base_reg && (i.base_reg->reg_type & Reg16) != 0))
2929 {
2930 p = frag_more (1);
2931 md_number_to_chars (p,
2932 (valueT) (i.sib.base << 0
2933 | i.sib.index << 3
2934 | i.sib.scale << 6),
2935 1);
2936 }
2937 }
2938
2939 if (i.disp_operands)
2940 {
2941 register unsigned int n;
2942
2943 for (n = 0; n < i.operands; n++)
2944 {
2945 if (i.types[n] & Disp)
2946 {
2947 if (i.op[n].disps->X_op == O_constant)
2948 {
2949 int size;
2950 offsetT val;
2951
2952 size = 4;
2953 if (i.types[n] & (Disp8 | Disp16 | Disp64))
2954 {
2955 size = 2;
2956 if (i.types[n] & Disp8)
2957 size = 1;
2958 if (i.types[n] & Disp64)
2959 size = 8;
2960 }
2961 val = offset_in_range (i.op[n].disps->X_add_number,
2962 size);
2963 p = frag_more (size);
2964 md_number_to_chars (p, val, size);
2965 }
2966 else
2967 {
2968 int size = 4;
2969 int sign = 0;
2970 int pcrel = (i.flags[n] & Operand_PCrel) != 0;
2971
2972 /* The PC relative address is computed relative
2973 to the instruction boundary, so in case immediate
2974 fields follows, we need to adjust the value. */
2975 if (pcrel && i.imm_operands)
2976 {
2977 int imm_size = 4;
2978 register unsigned int n1;
2979
2980 for (n1 = 0; n1 < i.operands; n1++)
2981 if (i.types[n1] & Imm)
2982 {
2983 if (i.types[n1] & (Imm8 | Imm8S | Imm16 | Imm64))
2984 {
2985 imm_size = 2;
2986 if (i.types[n1] & (Imm8 | Imm8S))
2987 imm_size = 1;
2988 if (i.types[n1] & Imm64)
2989 imm_size = 8;
2990 }
2991 break;
2992 }
2993 /* We should find the immediate. */
2994 if (n1 == i.operands)
2995 abort ();
2996 i.op[n].disps->X_add_number -= imm_size;
2997 }
2998
2999 if (i.types[n] & Disp32S)
3000 sign = 1;
3001
3002 if (i.types[n] & (Disp16 | Disp64))
3003 {
3004 size = 2;
3005 if (i.types[n] & Disp64)
3006 size = 8;
3007 }
3008
3009 p = frag_more (size);
3010 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3011 i.op[n].disps, pcrel,
3012 reloc (size, pcrel, sign, i.reloc[n]));
3013 }
3014 }
3015 }
3016 }
3017
3018 /* Output immediate. */
3019 if (i.imm_operands)
3020 {
3021 register unsigned int n;
3022
3023 for (n = 0; n < i.operands; n++)
3024 {
3025 if (i.types[n] & Imm)
3026 {
3027 if (i.op[n].imms->X_op == O_constant)
3028 {
3029 int size;
3030 offsetT val;
3031
3032 size = 4;
3033 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3034 {
3035 size = 2;
3036 if (i.types[n] & (Imm8 | Imm8S))
3037 size = 1;
3038 else if (i.types[n] & Imm64)
3039 size = 8;
3040 }
3041 val = offset_in_range (i.op[n].imms->X_add_number,
3042 size);
3043 p = frag_more (size);
3044 md_number_to_chars (p, val, size);
3045 }
3046 else
3047 {
3048 /* Not absolute_section.
3049 Need a 32-bit fixup (don't support 8bit
3050 non-absolute imms). Try to support other
3051 sizes ... */
3052 RELOC_ENUM reloc_type;
3053 int size = 4;
3054 int sign = 0;
3055
3056 if ((i.types[n] & (Imm32S))
3057 && i.suffix == QWORD_MNEM_SUFFIX)
3058 sign = 1;
3059 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3060 {
3061 size = 2;
3062 if (i.types[n] & (Imm8 | Imm8S))
3063 size = 1;
3064 if (i.types[n] & Imm64)
3065 size = 8;
3066 }
3067
3068 p = frag_more (size);
3069 reloc_type = reloc (size, 0, sign, i.reloc[n]);
3070 #ifdef BFD_ASSEMBLER
3071 if (reloc_type == BFD_RELOC_32
3072 && GOT_symbol
3073 && GOT_symbol == i.op[n].imms->X_add_symbol
3074 && (i.op[n].imms->X_op == O_symbol
3075 || (i.op[n].imms->X_op == O_add
3076 && ((symbol_get_value_expression
3077 (i.op[n].imms->X_op_symbol)->X_op)
3078 == O_subtract))))
3079 {
3080 /* We don't support dynamic linking on x86-64 yet. */
3081 if (flag_code == CODE_64BIT)
3082 abort ();
3083 reloc_type = BFD_RELOC_386_GOTPC;
3084 i.op[n].imms->X_add_number += 3;
3085 }
3086 #endif
3087 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3088 i.op[n].imms, 0, reloc_type);
3089 }
3090 }
3091 }
3092 }
3093 }
3094
3095 #ifdef DEBUG386
3096 if (flag_debug)
3097 {
3098 pi (line, &i);
3099 }
3100 #endif /* DEBUG386 */
3101 }
3102 }
3103 \f
3104 #ifndef LEX_AT
3105 static char *lex_got PARAMS ((RELOC_ENUM *, int *));
3106
3107 /* Parse operands of the form
3108 <symbol>@GOTOFF+<nnn>
3109 and similar .plt or .got references.
3110
3111 If we find one, set up the correct relocation in RELOC and copy the
3112 input string, minus the `@GOTOFF' into a malloc'd buffer for
3113 parsing by the calling routine. Return this buffer, and if ADJUST
3114 is non-null set it to the length of the string we removed from the
3115 input line. Otherwise return NULL. */
3116 static char *
3117 lex_got (reloc, adjust)
3118 RELOC_ENUM *reloc;
3119 int *adjust;
3120 {
3121 static const char * const mode_name[NUM_FLAG_CODE] = { "32", "16", "64" };
3122 static const struct {
3123 const char *str;
3124 const RELOC_ENUM rel[NUM_FLAG_CODE];
3125 } gotrel[] = {
3126 { "PLT", { BFD_RELOC_386_PLT32, 0, BFD_RELOC_X86_64_PLT32 } },
3127 { "GOTOFF", { BFD_RELOC_386_GOTOFF, 0, 0 } },
3128 { "GOTPCREL", { 0, 0, BFD_RELOC_X86_64_GOTPCREL } },
3129 { "GOT", { BFD_RELOC_386_GOT32, 0, BFD_RELOC_X86_64_GOT32 } }
3130 };
3131 char *cp;
3132 unsigned int j;
3133
3134 for (cp = input_line_pointer; *cp != '@'; cp++)
3135 if (is_end_of_line[(unsigned char) *cp])
3136 return NULL;
3137
3138 for (j = 0; j < sizeof (gotrel) / sizeof (gotrel[0]); j++)
3139 {
3140 int len;
3141
3142 len = strlen (gotrel[j].str);
3143 if (strncmp (cp + 1, gotrel[j].str, len) == 0)
3144 {
3145 if (gotrel[j].rel[(unsigned int) flag_code] != 0)
3146 {
3147 int first;
3148 char *tmpbuf;
3149
3150 *reloc = gotrel[j].rel[(unsigned int) flag_code];
3151
3152 if (GOT_symbol == NULL)
3153 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
3154
3155 /* Replace the relocation token with ' ', so that
3156 errors like foo@GOTOFF1 will be detected. */
3157 first = cp - input_line_pointer;
3158 tmpbuf = xmalloc (strlen (input_line_pointer));
3159 memcpy (tmpbuf, input_line_pointer, first);
3160 tmpbuf[first] = ' ';
3161 strcpy (tmpbuf + first + 1, cp + 1 + len);
3162 if (adjust)
3163 *adjust = len;
3164 return tmpbuf;
3165 }
3166
3167 as_bad (_("@%s reloc is not supported in %s bit mode"),
3168 gotrel[j].str, mode_name[(unsigned int) flag_code]);
3169 return NULL;
3170 }
3171 }
3172
3173 /* Might be a symbol version string. Don't as_bad here. */
3174 return NULL;
3175 }
3176
3177 /* x86_cons_fix_new is called via the expression parsing code when a
3178 reloc is needed. We use this hook to get the correct .got reloc. */
3179 static RELOC_ENUM got_reloc = NO_RELOC;
3180
3181 void
3182 x86_cons_fix_new (frag, off, len, exp)
3183 fragS *frag;
3184 unsigned int off;
3185 unsigned int len;
3186 expressionS *exp;
3187 {
3188 RELOC_ENUM r = reloc (len, 0, 0, got_reloc);
3189 got_reloc = NO_RELOC;
3190 fix_new_exp (frag, off, len, exp, 0, r);
3191 }
3192
3193 void
3194 x86_cons (exp, size)
3195 expressionS *exp;
3196 int size;
3197 {
3198 if (size == 4)
3199 {
3200 /* Handle @GOTOFF and the like in an expression. */
3201 char *save;
3202 char *gotfree_input_line;
3203 int adjust;
3204
3205 save = input_line_pointer;
3206 gotfree_input_line = lex_got (&got_reloc, &adjust);
3207 if (gotfree_input_line)
3208 input_line_pointer = gotfree_input_line;
3209
3210 expression (exp);
3211
3212 if (gotfree_input_line)
3213 {
3214 /* expression () has merrily parsed up to the end of line,
3215 or a comma - in the wrong buffer. Transfer how far
3216 input_line_pointer has moved to the right buffer. */
3217 input_line_pointer = (save
3218 + (input_line_pointer - gotfree_input_line)
3219 + adjust);
3220 free (gotfree_input_line);
3221 }
3222 }
3223 else
3224 expression (exp);
3225 }
3226 #endif
3227
3228 static int i386_immediate PARAMS ((char *));
3229
3230 static int
3231 i386_immediate (imm_start)
3232 char *imm_start;
3233 {
3234 char *save_input_line_pointer;
3235 #ifndef LEX_AT
3236 char *gotfree_input_line;
3237 #endif
3238 segT exp_seg = 0;
3239 expressionS *exp;
3240
3241 if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
3242 {
3243 as_bad (_("only 1 or 2 immediate operands are allowed"));
3244 return 0;
3245 }
3246
3247 exp = &im_expressions[i.imm_operands++];
3248 i.op[this_operand].imms = exp;
3249
3250 if (is_space_char (*imm_start))
3251 ++imm_start;
3252
3253 save_input_line_pointer = input_line_pointer;
3254 input_line_pointer = imm_start;
3255
3256 #ifndef LEX_AT
3257 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
3258 if (gotfree_input_line)
3259 input_line_pointer = gotfree_input_line;
3260 #endif
3261
3262 exp_seg = expression (exp);
3263
3264 SKIP_WHITESPACE ();
3265 if (*input_line_pointer)
3266 as_bad (_("junk `%s' after expression"), input_line_pointer);
3267
3268 input_line_pointer = save_input_line_pointer;
3269 #ifndef LEX_AT
3270 if (gotfree_input_line)
3271 free (gotfree_input_line);
3272 #endif
3273
3274 if (exp->X_op == O_absent || exp->X_op == O_big)
3275 {
3276 /* Missing or bad expr becomes absolute 0. */
3277 as_bad (_("missing or invalid immediate expression `%s' taken as 0"),
3278 imm_start);
3279 exp->X_op = O_constant;
3280 exp->X_add_number = 0;
3281 exp->X_add_symbol = (symbolS *) 0;
3282 exp->X_op_symbol = (symbolS *) 0;
3283 }
3284 else if (exp->X_op == O_constant)
3285 {
3286 /* Size it properly later. */
3287 i.types[this_operand] |= Imm64;
3288 /* If BFD64, sign extend val. */
3289 if (!use_rela_relocations)
3290 if ((exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0)
3291 exp->X_add_number = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
3292 }
3293 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3294 else if (1
3295 #ifdef BFD_ASSEMBLER
3296 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3297 #endif
3298 && exp_seg != text_section
3299 && exp_seg != data_section
3300 && exp_seg != bss_section
3301 && exp_seg != undefined_section
3302 #ifdef BFD_ASSEMBLER
3303 && !bfd_is_com_section (exp_seg)
3304 #endif
3305 )
3306 {
3307 #ifdef BFD_ASSEMBLER
3308 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3309 #else
3310 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
3311 #endif
3312 return 0;
3313 }
3314 #endif
3315 else
3316 {
3317 /* This is an address. The size of the address will be
3318 determined later, depending on destination register,
3319 suffix, or the default for the section. */
3320 i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64;
3321 }
3322
3323 return 1;
3324 }
3325
3326 static char *i386_scale PARAMS ((char *));
3327
3328 static char *
3329 i386_scale (scale)
3330 char *scale;
3331 {
3332 offsetT val;
3333 char *save = input_line_pointer;
3334
3335 input_line_pointer = scale;
3336 val = get_absolute_expression ();
3337
3338 switch (val)
3339 {
3340 case 0:
3341 case 1:
3342 i.log2_scale_factor = 0;
3343 break;
3344 case 2:
3345 i.log2_scale_factor = 1;
3346 break;
3347 case 4:
3348 i.log2_scale_factor = 2;
3349 break;
3350 case 8:
3351 i.log2_scale_factor = 3;
3352 break;
3353 default:
3354 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
3355 scale);
3356 input_line_pointer = save;
3357 return NULL;
3358 }
3359 if (i.log2_scale_factor != 0 && ! i.index_reg)
3360 {
3361 as_warn (_("scale factor of %d without an index register"),
3362 1 << i.log2_scale_factor);
3363 #if SCALE1_WHEN_NO_INDEX
3364 i.log2_scale_factor = 0;
3365 #endif
3366 }
3367 scale = input_line_pointer;
3368 input_line_pointer = save;
3369 return scale;
3370 }
3371
3372 static int i386_displacement PARAMS ((char *, char *));
3373
3374 static int
3375 i386_displacement (disp_start, disp_end)
3376 char *disp_start;
3377 char *disp_end;
3378 {
3379 register expressionS *exp;
3380 segT exp_seg = 0;
3381 char *save_input_line_pointer;
3382 #ifndef LEX_AT
3383 char *gotfree_input_line;
3384 #endif
3385 int bigdisp = Disp32;
3386
3387 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3388 bigdisp = Disp16;
3389 if (flag_code == CODE_64BIT)
3390 bigdisp = Disp64;
3391 i.types[this_operand] |= bigdisp;
3392
3393 exp = &disp_expressions[i.disp_operands];
3394 i.op[this_operand].disps = exp;
3395 i.disp_operands++;
3396 save_input_line_pointer = input_line_pointer;
3397 input_line_pointer = disp_start;
3398 END_STRING_AND_SAVE (disp_end);
3399
3400 #ifndef GCC_ASM_O_HACK
3401 #define GCC_ASM_O_HACK 0
3402 #endif
3403 #if GCC_ASM_O_HACK
3404 END_STRING_AND_SAVE (disp_end + 1);
3405 if ((i.types[this_operand] & BaseIndex) != 0
3406 && displacement_string_end[-1] == '+')
3407 {
3408 /* This hack is to avoid a warning when using the "o"
3409 constraint within gcc asm statements.
3410 For instance:
3411
3412 #define _set_tssldt_desc(n,addr,limit,type) \
3413 __asm__ __volatile__ ( \
3414 "movw %w2,%0\n\t" \
3415 "movw %w1,2+%0\n\t" \
3416 "rorl $16,%1\n\t" \
3417 "movb %b1,4+%0\n\t" \
3418 "movb %4,5+%0\n\t" \
3419 "movb $0,6+%0\n\t" \
3420 "movb %h1,7+%0\n\t" \
3421 "rorl $16,%1" \
3422 : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
3423
3424 This works great except that the output assembler ends
3425 up looking a bit weird if it turns out that there is
3426 no offset. You end up producing code that looks like:
3427
3428 #APP
3429 movw $235,(%eax)
3430 movw %dx,2+(%eax)
3431 rorl $16,%edx
3432 movb %dl,4+(%eax)
3433 movb $137,5+(%eax)
3434 movb $0,6+(%eax)
3435 movb %dh,7+(%eax)
3436 rorl $16,%edx
3437 #NO_APP
3438
3439 So here we provide the missing zero. */
3440
3441 *displacement_string_end = '0';
3442 }
3443 #endif
3444 #ifndef LEX_AT
3445 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
3446 if (gotfree_input_line)
3447 input_line_pointer = gotfree_input_line;
3448 #endif
3449
3450 exp_seg = expression (exp);
3451
3452 SKIP_WHITESPACE ();
3453 if (*input_line_pointer)
3454 as_bad (_("junk `%s' after expression"), input_line_pointer);
3455 #if GCC_ASM_O_HACK
3456 RESTORE_END_STRING (disp_end + 1);
3457 #endif
3458 RESTORE_END_STRING (disp_end);
3459 input_line_pointer = save_input_line_pointer;
3460 #ifndef LEX_AT
3461 if (gotfree_input_line)
3462 free (gotfree_input_line);
3463 #endif
3464
3465 #ifdef BFD_ASSEMBLER
3466 /* We do this to make sure that the section symbol is in
3467 the symbol table. We will ultimately change the relocation
3468 to be relative to the beginning of the section. */
3469 if (i.reloc[this_operand] == BFD_RELOC_386_GOTOFF
3470 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3471 {
3472 if (exp->X_op != O_symbol)
3473 {
3474 as_bad (_("bad expression used with @%s"),
3475 (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
3476 ? "GOTPCREL"
3477 : "GOTOFF"));
3478 return 0;
3479 }
3480
3481 if (S_IS_LOCAL (exp->X_add_symbol)
3482 && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
3483 section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
3484 exp->X_op = O_subtract;
3485 exp->X_op_symbol = GOT_symbol;
3486 if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3487 i.reloc[this_operand] = BFD_RELOC_32_PCREL;
3488 else
3489 i.reloc[this_operand] = BFD_RELOC_32;
3490 }
3491 #endif
3492
3493 if (exp->X_op == O_absent || exp->X_op == O_big)
3494 {
3495 /* Missing or bad expr becomes absolute 0. */
3496 as_bad (_("missing or invalid displacement expression `%s' taken as 0"),
3497 disp_start);
3498 exp->X_op = O_constant;
3499 exp->X_add_number = 0;
3500 exp->X_add_symbol = (symbolS *) 0;
3501 exp->X_op_symbol = (symbolS *) 0;
3502 }
3503
3504 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3505 if (exp->X_op != O_constant
3506 #ifdef BFD_ASSEMBLER
3507 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3508 #endif
3509 && exp_seg != text_section
3510 && exp_seg != data_section
3511 && exp_seg != bss_section
3512 && exp_seg != undefined_section)
3513 {
3514 #ifdef BFD_ASSEMBLER
3515 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3516 #else
3517 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
3518 #endif
3519 return 0;
3520 }
3521 #endif
3522 else if (flag_code == CODE_64BIT)
3523 i.types[this_operand] |= Disp32S | Disp32;
3524 return 1;
3525 }
3526
3527 static int i386_index_check PARAMS ((const char *));
3528
3529 /* Make sure the memory operand we've been dealt is valid.
3530 Return 1 on success, 0 on a failure. */
3531
3532 static int
3533 i386_index_check (operand_string)
3534 const char *operand_string;
3535 {
3536 int ok;
3537 #if INFER_ADDR_PREFIX
3538 int fudged = 0;
3539
3540 tryprefix:
3541 #endif
3542 ok = 1;
3543 if (flag_code == CODE_64BIT)
3544 {
3545 /* 64bit checks. */
3546 if ((i.base_reg
3547 && ((i.base_reg->reg_type & Reg64) == 0)
3548 && (i.base_reg->reg_type != BaseIndex
3549 || i.index_reg))
3550 || (i.index_reg
3551 && ((i.index_reg->reg_type & (Reg64|BaseIndex))
3552 != (Reg64|BaseIndex))))
3553 ok = 0;
3554 }
3555 else
3556 {
3557 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3558 {
3559 /* 16bit checks. */
3560 if ((i.base_reg
3561 && ((i.base_reg->reg_type & (Reg16|BaseIndex|RegRex))
3562 != (Reg16|BaseIndex)))
3563 || (i.index_reg
3564 && (((i.index_reg->reg_type & (Reg16|BaseIndex))
3565 != (Reg16|BaseIndex))
3566 || ! (i.base_reg
3567 && i.base_reg->reg_num < 6
3568 && i.index_reg->reg_num >= 6
3569 && i.log2_scale_factor == 0))))
3570 ok = 0;
3571 }
3572 else
3573 {
3574 /* 32bit checks. */
3575 if ((i.base_reg
3576 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
3577 || (i.index_reg
3578 && ((i.index_reg->reg_type & (Reg32|BaseIndex|RegRex))
3579 != (Reg32|BaseIndex))))
3580 ok = 0;
3581 }
3582 }
3583 if (!ok)
3584 {
3585 #if INFER_ADDR_PREFIX
3586 if (flag_code != CODE_64BIT
3587 && i.prefix[ADDR_PREFIX] == 0 && stackop_size != '\0')
3588 {
3589 i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
3590 i.prefixes += 1;
3591 /* Change the size of any displacement too. At most one of
3592 Disp16 or Disp32 is set.
3593 FIXME. There doesn't seem to be any real need for separate
3594 Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
3595 Removing them would probably clean up the code quite a lot. */
3596 if (i.types[this_operand] & (Disp16|Disp32))
3597 i.types[this_operand] ^= (Disp16|Disp32);
3598 fudged = 1;
3599 goto tryprefix;
3600 }
3601 if (fudged)
3602 as_bad (_("`%s' is not a valid base/index expression"),
3603 operand_string);
3604 else
3605 #endif
3606 as_bad (_("`%s' is not a valid %s bit base/index expression"),
3607 operand_string,
3608 flag_code_names[flag_code]);
3609 return 0;
3610 }
3611 return 1;
3612 }
3613
3614 /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
3615 on error. */
3616
3617 static int
3618 i386_operand (operand_string)
3619 char *operand_string;
3620 {
3621 const reg_entry *r;
3622 char *end_op;
3623 char *op_string = operand_string;
3624
3625 if (is_space_char (*op_string))
3626 ++op_string;
3627
3628 /* We check for an absolute prefix (differentiating,
3629 for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
3630 if (*op_string == ABSOLUTE_PREFIX)
3631 {
3632 ++op_string;
3633 if (is_space_char (*op_string))
3634 ++op_string;
3635 i.types[this_operand] |= JumpAbsolute;
3636 }
3637
3638 /* Check if operand is a register. */
3639 if ((*op_string == REGISTER_PREFIX || allow_naked_reg)
3640 && (r = parse_register (op_string, &end_op)) != NULL)
3641 {
3642 /* Check for a segment override by searching for ':' after a
3643 segment register. */
3644 op_string = end_op;
3645 if (is_space_char (*op_string))
3646 ++op_string;
3647 if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
3648 {
3649 switch (r->reg_num)
3650 {
3651 case 0:
3652 i.seg[i.mem_operands] = &es;
3653 break;
3654 case 1:
3655 i.seg[i.mem_operands] = &cs;
3656 break;
3657 case 2:
3658 i.seg[i.mem_operands] = &ss;
3659 break;
3660 case 3:
3661 i.seg[i.mem_operands] = &ds;
3662 break;
3663 case 4:
3664 i.seg[i.mem_operands] = &fs;
3665 break;
3666 case 5:
3667 i.seg[i.mem_operands] = &gs;
3668 break;
3669 }
3670
3671 /* Skip the ':' and whitespace. */
3672 ++op_string;
3673 if (is_space_char (*op_string))
3674 ++op_string;
3675
3676 if (!is_digit_char (*op_string)
3677 && !is_identifier_char (*op_string)
3678 && *op_string != '('
3679 && *op_string != ABSOLUTE_PREFIX)
3680 {
3681 as_bad (_("bad memory operand `%s'"), op_string);
3682 return 0;
3683 }
3684 /* Handle case of %es:*foo. */
3685 if (*op_string == ABSOLUTE_PREFIX)
3686 {
3687 ++op_string;
3688 if (is_space_char (*op_string))
3689 ++op_string;
3690 i.types[this_operand] |= JumpAbsolute;
3691 }
3692 goto do_memory_reference;
3693 }
3694 if (*op_string)
3695 {
3696 as_bad (_("junk `%s' after register"), op_string);
3697 return 0;
3698 }
3699 i.types[this_operand] |= r->reg_type & ~BaseIndex;
3700 i.op[this_operand].regs = r;
3701 i.reg_operands++;
3702 }
3703 else if (*op_string == REGISTER_PREFIX)
3704 {
3705 as_bad (_("bad register name `%s'"), op_string);
3706 return 0;
3707 }
3708 else if (*op_string == IMMEDIATE_PREFIX)
3709 {
3710 ++op_string;
3711 if (i.types[this_operand] & JumpAbsolute)
3712 {
3713 as_bad (_("immediate operand illegal with absolute jump"));
3714 return 0;
3715 }
3716 if (!i386_immediate (op_string))
3717 return 0;
3718 }
3719 else if (is_digit_char (*op_string)
3720 || is_identifier_char (*op_string)
3721 || *op_string == '(')
3722 {
3723 /* This is a memory reference of some sort. */
3724 char *base_string;
3725
3726 /* Start and end of displacement string expression (if found). */
3727 char *displacement_string_start;
3728 char *displacement_string_end;
3729
3730 do_memory_reference:
3731 if ((i.mem_operands == 1
3732 && (current_templates->start->opcode_modifier & IsString) == 0)
3733 || i.mem_operands == 2)
3734 {
3735 as_bad (_("too many memory references for `%s'"),
3736 current_templates->start->name);
3737 return 0;
3738 }
3739
3740 /* Check for base index form. We detect the base index form by
3741 looking for an ')' at the end of the operand, searching
3742 for the '(' matching it, and finding a REGISTER_PREFIX or ','
3743 after the '('. */
3744 base_string = op_string + strlen (op_string);
3745
3746 --base_string;
3747 if (is_space_char (*base_string))
3748 --base_string;
3749
3750 /* If we only have a displacement, set-up for it to be parsed later. */
3751 displacement_string_start = op_string;
3752 displacement_string_end = base_string + 1;
3753
3754 if (*base_string == ')')
3755 {
3756 char *temp_string;
3757 unsigned int parens_balanced = 1;
3758 /* We've already checked that the number of left & right ()'s are
3759 equal, so this loop will not be infinite. */
3760 do
3761 {
3762 base_string--;
3763 if (*base_string == ')')
3764 parens_balanced++;
3765 if (*base_string == '(')
3766 parens_balanced--;
3767 }
3768 while (parens_balanced);
3769
3770 temp_string = base_string;
3771
3772 /* Skip past '(' and whitespace. */
3773 ++base_string;
3774 if (is_space_char (*base_string))
3775 ++base_string;
3776
3777 if (*base_string == ','
3778 || ((*base_string == REGISTER_PREFIX || allow_naked_reg)
3779 && (i.base_reg = parse_register (base_string, &end_op)) != NULL))
3780 {
3781 displacement_string_end = temp_string;
3782
3783 i.types[this_operand] |= BaseIndex;
3784
3785 if (i.base_reg)
3786 {
3787 base_string = end_op;
3788 if (is_space_char (*base_string))
3789 ++base_string;
3790 }
3791
3792 /* There may be an index reg or scale factor here. */
3793 if (*base_string == ',')
3794 {
3795 ++base_string;
3796 if (is_space_char (*base_string))
3797 ++base_string;
3798
3799 if ((*base_string == REGISTER_PREFIX || allow_naked_reg)
3800 && (i.index_reg = parse_register (base_string, &end_op)) != NULL)
3801 {
3802 base_string = end_op;
3803 if (is_space_char (*base_string))
3804 ++base_string;
3805 if (*base_string == ',')
3806 {
3807 ++base_string;
3808 if (is_space_char (*base_string))
3809 ++base_string;
3810 }
3811 else if (*base_string != ')')
3812 {
3813 as_bad (_("expecting `,' or `)' after index register in `%s'"),
3814 operand_string);
3815 return 0;
3816 }
3817 }
3818 else if (*base_string == REGISTER_PREFIX)
3819 {
3820 as_bad (_("bad register name `%s'"), base_string);
3821 return 0;
3822 }
3823
3824 /* Check for scale factor. */
3825 if (*base_string != ')')
3826 {
3827 char *end_scale = i386_scale (base_string);
3828
3829 if (!end_scale)
3830 return 0;
3831
3832 base_string = end_scale;
3833 if (is_space_char (*base_string))
3834 ++base_string;
3835 if (*base_string != ')')
3836 {
3837 as_bad (_("expecting `)' after scale factor in `%s'"),
3838 operand_string);
3839 return 0;
3840 }
3841 }
3842 else if (!i.index_reg)
3843 {
3844 as_bad (_("expecting index register or scale factor after `,'; got '%c'"),
3845 *base_string);
3846 return 0;
3847 }
3848 }
3849 else if (*base_string != ')')
3850 {
3851 as_bad (_("expecting `,' or `)' after base register in `%s'"),
3852 operand_string);
3853 return 0;
3854 }
3855 }
3856 else if (*base_string == REGISTER_PREFIX)
3857 {
3858 as_bad (_("bad register name `%s'"), base_string);
3859 return 0;
3860 }
3861 }
3862
3863 /* If there's an expression beginning the operand, parse it,
3864 assuming displacement_string_start and
3865 displacement_string_end are meaningful. */
3866 if (displacement_string_start != displacement_string_end)
3867 {
3868 if (!i386_displacement (displacement_string_start,
3869 displacement_string_end))
3870 return 0;
3871 }
3872
3873 /* Special case for (%dx) while doing input/output op. */
3874 if (i.base_reg
3875 && i.base_reg->reg_type == (Reg16 | InOutPortReg)
3876 && i.index_reg == 0
3877 && i.log2_scale_factor == 0
3878 && i.seg[i.mem_operands] == 0
3879 && (i.types[this_operand] & Disp) == 0)
3880 {
3881 i.types[this_operand] = InOutPortReg;
3882 return 1;
3883 }
3884
3885 if (i386_index_check (operand_string) == 0)
3886 return 0;
3887 i.mem_operands++;
3888 }
3889 else
3890 {
3891 /* It's not a memory operand; argh! */
3892 as_bad (_("invalid char %s beginning operand %d `%s'"),
3893 output_invalid (*op_string),
3894 this_operand + 1,
3895 op_string);
3896 return 0;
3897 }
3898 return 1; /* Normal return. */
3899 }
3900 \f
3901 /* md_estimate_size_before_relax()
3902
3903 Called just before relax() for rs_machine_dependent frags. The x86
3904 assembler uses these frags to handle variable size jump
3905 instructions.
3906
3907 Any symbol that is now undefined will not become defined.
3908 Return the correct fr_subtype in the frag.
3909 Return the initial "guess for variable size of frag" to caller.
3910 The guess is actually the growth beyond the fixed part. Whatever
3911 we do to grow the fixed or variable part contributes to our
3912 returned value. */
3913
3914 int
3915 md_estimate_size_before_relax (fragP, segment)
3916 register fragS *fragP;
3917 register segT segment;
3918 {
3919 /* We've already got fragP->fr_subtype right; all we have to do is
3920 check for un-relaxable symbols. On an ELF system, we can't relax
3921 an externally visible symbol, because it may be overridden by a
3922 shared library. */
3923 if (S_GET_SEGMENT (fragP->fr_symbol) != segment
3924 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
3925 || S_IS_EXTERNAL (fragP->fr_symbol)
3926 || S_IS_WEAK (fragP->fr_symbol)
3927 #endif
3928 )
3929 {
3930 /* Symbol is undefined in this segment, or we need to keep a
3931 reloc so that weak symbols can be overridden. */
3932 int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
3933 RELOC_ENUM reloc_type;
3934 unsigned char *opcode;
3935 int old_fr_fix;
3936
3937 if (fragP->fr_var != NO_RELOC)
3938 reloc_type = fragP->fr_var;
3939 else if (size == 2)
3940 reloc_type = BFD_RELOC_16_PCREL;
3941 else
3942 reloc_type = BFD_RELOC_32_PCREL;
3943
3944 old_fr_fix = fragP->fr_fix;
3945 opcode = (unsigned char *) fragP->fr_opcode;
3946
3947 switch (TYPE_FROM_RELAX_STATE (fragP->fr_subtype))
3948 {
3949 case UNCOND_JUMP:
3950 /* Make jmp (0xeb) a (d)word displacement jump. */
3951 opcode[0] = 0xe9;
3952 fragP->fr_fix += size;
3953 fix_new (fragP, old_fr_fix, size,
3954 fragP->fr_symbol,
3955 fragP->fr_offset, 1,
3956 reloc_type);
3957 break;
3958
3959 case COND_JUMP86:
3960 if (no_cond_jump_promotion)
3961 goto relax_guess;
3962
3963 if (size == 2)
3964 {
3965 /* Negate the condition, and branch past an
3966 unconditional jump. */
3967 opcode[0] ^= 1;
3968 opcode[1] = 3;
3969 /* Insert an unconditional jump. */
3970 opcode[2] = 0xe9;
3971 /* We added two extra opcode bytes, and have a two byte
3972 offset. */
3973 fragP->fr_fix += 2 + 2;
3974 fix_new (fragP, old_fr_fix + 2, 2,
3975 fragP->fr_symbol,
3976 fragP->fr_offset, 1,
3977 reloc_type);
3978 break;
3979 }
3980 /* Fall through. */
3981
3982 case COND_JUMP:
3983 if (no_cond_jump_promotion)
3984 goto relax_guess;
3985
3986 /* This changes the byte-displacement jump 0x7N
3987 to the (d)word-displacement jump 0x0f,0x8N. */
3988 opcode[1] = opcode[0] + 0x10;
3989 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
3990 /* We've added an opcode byte. */
3991 fragP->fr_fix += 1 + size;
3992 fix_new (fragP, old_fr_fix + 1, size,
3993 fragP->fr_symbol,
3994 fragP->fr_offset, 1,
3995 reloc_type);
3996 break;
3997
3998 default:
3999 BAD_CASE (fragP->fr_subtype);
4000 break;
4001 }
4002 frag_wane (fragP);
4003 return fragP->fr_fix - old_fr_fix;
4004 }
4005
4006 relax_guess:
4007 /* Guess size depending on current relax state. Initially the relax
4008 state will correspond to a short jump and we return 1, because
4009 the variable part of the frag (the branch offset) is one byte
4010 long. However, we can relax a section more than once and in that
4011 case we must either set fr_subtype back to the unrelaxed state,
4012 or return the value for the appropriate branch. */
4013 return md_relax_table[fragP->fr_subtype].rlx_length;
4014 }
4015
4016 /* Called after relax() is finished.
4017
4018 In: Address of frag.
4019 fr_type == rs_machine_dependent.
4020 fr_subtype is what the address relaxed to.
4021
4022 Out: Any fixSs and constants are set up.
4023 Caller will turn frag into a ".space 0". */
4024
4025 #ifndef BFD_ASSEMBLER
4026 void
4027 md_convert_frag (headers, sec, fragP)
4028 object_headers *headers ATTRIBUTE_UNUSED;
4029 segT sec ATTRIBUTE_UNUSED;
4030 register fragS *fragP;
4031 #else
4032 void
4033 md_convert_frag (abfd, sec, fragP)
4034 bfd *abfd ATTRIBUTE_UNUSED;
4035 segT sec ATTRIBUTE_UNUSED;
4036 register fragS *fragP;
4037 #endif
4038 {
4039 register unsigned char *opcode;
4040 unsigned char *where_to_put_displacement = NULL;
4041 offsetT target_address;
4042 offsetT opcode_address;
4043 unsigned int extension = 0;
4044 offsetT displacement_from_opcode_start;
4045
4046 opcode = (unsigned char *) fragP->fr_opcode;
4047
4048 /* Address we want to reach in file space. */
4049 target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
4050 #ifdef BFD_ASSEMBLER
4051 /* Not needed otherwise? */
4052 {
4053 /* Local symbols which have already been resolved have a NULL frag. */
4054 fragS *sym_frag = symbol_get_frag (fragP->fr_symbol);
4055 if (sym_frag)
4056 target_address += sym_frag->fr_address;
4057 }
4058 #endif
4059
4060 /* Address opcode resides at in file space. */
4061 opcode_address = fragP->fr_address + fragP->fr_fix;
4062
4063 /* Displacement from opcode start to fill into instruction. */
4064 displacement_from_opcode_start = target_address - opcode_address;
4065
4066 if ((fragP->fr_subtype & BIG) == 0)
4067 {
4068 /* Don't have to change opcode. */
4069 extension = 1; /* 1 opcode + 1 displacement */
4070 where_to_put_displacement = &opcode[1];
4071 }
4072 else
4073 {
4074 if (no_cond_jump_promotion
4075 && TYPE_FROM_RELAX_STATE (fragP->fr_subtype) != UNCOND_JUMP)
4076 as_warn_where (fragP->fr_file, fragP->fr_line, _("long jump required"));
4077
4078 switch (fragP->fr_subtype)
4079 {
4080 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
4081 extension = 4; /* 1 opcode + 4 displacement */
4082 opcode[0] = 0xe9;
4083 where_to_put_displacement = &opcode[1];
4084 break;
4085
4086 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
4087 extension = 2; /* 1 opcode + 2 displacement */
4088 opcode[0] = 0xe9;
4089 where_to_put_displacement = &opcode[1];
4090 break;
4091
4092 case ENCODE_RELAX_STATE (COND_JUMP, BIG):
4093 case ENCODE_RELAX_STATE (COND_JUMP86, BIG):
4094 extension = 5; /* 2 opcode + 4 displacement */
4095 opcode[1] = opcode[0] + 0x10;
4096 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4097 where_to_put_displacement = &opcode[2];
4098 break;
4099
4100 case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
4101 extension = 3; /* 2 opcode + 2 displacement */
4102 opcode[1] = opcode[0] + 0x10;
4103 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4104 where_to_put_displacement = &opcode[2];
4105 break;
4106
4107 case ENCODE_RELAX_STATE (COND_JUMP86, BIG16):
4108 extension = 4;
4109 opcode[0] ^= 1;
4110 opcode[1] = 3;
4111 opcode[2] = 0xe9;
4112 where_to_put_displacement = &opcode[3];
4113 break;
4114
4115 default:
4116 BAD_CASE (fragP->fr_subtype);
4117 break;
4118 }
4119 }
4120
4121 /* Now put displacement after opcode. */
4122 md_number_to_chars ((char *) where_to_put_displacement,
4123 (valueT) (displacement_from_opcode_start - extension),
4124 DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
4125 fragP->fr_fix += extension;
4126 }
4127 \f
4128 /* Size of byte displacement jmp. */
4129 int md_short_jump_size = 2;
4130
4131 /* Size of dword displacement jmp. */
4132 int md_long_jump_size = 5;
4133
4134 /* Size of relocation record. */
4135 const int md_reloc_size = 8;
4136
4137 void
4138 md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
4139 char *ptr;
4140 addressT from_addr, to_addr;
4141 fragS *frag ATTRIBUTE_UNUSED;
4142 symbolS *to_symbol ATTRIBUTE_UNUSED;
4143 {
4144 offsetT offset;
4145
4146 offset = to_addr - (from_addr + 2);
4147 /* Opcode for byte-disp jump. */
4148 md_number_to_chars (ptr, (valueT) 0xeb, 1);
4149 md_number_to_chars (ptr + 1, (valueT) offset, 1);
4150 }
4151
4152 void
4153 md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
4154 char *ptr;
4155 addressT from_addr, to_addr;
4156 fragS *frag ATTRIBUTE_UNUSED;
4157 symbolS *to_symbol ATTRIBUTE_UNUSED;
4158 {
4159 offsetT offset;
4160
4161 offset = to_addr - (from_addr + 5);
4162 md_number_to_chars (ptr, (valueT) 0xe9, 1);
4163 md_number_to_chars (ptr + 1, (valueT) offset, 4);
4164 }
4165 \f
4166 /* Apply a fixup (fixS) to segment data, once it has been determined
4167 by our caller that we have all the info we need to fix it up.
4168
4169 On the 386, immediates, displacements, and data pointers are all in
4170 the same (little-endian) format, so we don't need to care about which
4171 we are handling. */
4172
4173 int
4174 md_apply_fix3 (fixP, valp, seg)
4175 /* The fix we're to put in. */
4176 fixS *fixP;
4177
4178 /* Pointer to the value of the bits. */
4179 valueT *valp;
4180
4181 /* Segment fix is from. */
4182 segT seg ATTRIBUTE_UNUSED;
4183 {
4184 register char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
4185 valueT value = *valp;
4186
4187 #if defined (BFD_ASSEMBLER) && !defined (TE_Mach)
4188 if (fixP->fx_pcrel)
4189 {
4190 switch (fixP->fx_r_type)
4191 {
4192 default:
4193 break;
4194
4195 case BFD_RELOC_32:
4196 fixP->fx_r_type = BFD_RELOC_32_PCREL;
4197 break;
4198 case BFD_RELOC_16:
4199 fixP->fx_r_type = BFD_RELOC_16_PCREL;
4200 break;
4201 case BFD_RELOC_8:
4202 fixP->fx_r_type = BFD_RELOC_8_PCREL;
4203 break;
4204 }
4205 }
4206
4207 /* This is a hack. There should be a better way to handle this.
4208 This covers for the fact that bfd_install_relocation will
4209 subtract the current location (for partial_inplace, PC relative
4210 relocations); see more below. */
4211 if ((fixP->fx_r_type == BFD_RELOC_32_PCREL
4212 || fixP->fx_r_type == BFD_RELOC_16_PCREL
4213 || fixP->fx_r_type == BFD_RELOC_8_PCREL)
4214 && fixP->fx_addsy && !use_rela_relocations)
4215 {
4216 #ifndef OBJ_AOUT
4217 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4218 #ifdef TE_PE
4219 || OUTPUT_FLAVOR == bfd_target_coff_flavour
4220 #endif
4221 )
4222 value += fixP->fx_where + fixP->fx_frag->fr_address;
4223 #endif
4224 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4225 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
4226 {
4227 segT fseg = S_GET_SEGMENT (fixP->fx_addsy);
4228
4229 if ((fseg == seg
4230 || (symbol_section_p (fixP->fx_addsy)
4231 && fseg != absolute_section))
4232 && ! S_IS_EXTERNAL (fixP->fx_addsy)
4233 && ! S_IS_WEAK (fixP->fx_addsy)
4234 && S_IS_DEFINED (fixP->fx_addsy)
4235 && ! S_IS_COMMON (fixP->fx_addsy))
4236 {
4237 /* Yes, we add the values in twice. This is because
4238 bfd_perform_relocation subtracts them out again. I think
4239 bfd_perform_relocation is broken, but I don't dare change
4240 it. FIXME. */
4241 value += fixP->fx_where + fixP->fx_frag->fr_address;
4242 }
4243 }
4244 #endif
4245 #if defined (OBJ_COFF) && defined (TE_PE)
4246 /* For some reason, the PE format does not store a section
4247 address offset for a PC relative symbol. */
4248 if (S_GET_SEGMENT (fixP->fx_addsy) != seg)
4249 value += md_pcrel_from (fixP);
4250 #endif
4251 }
4252
4253 /* Fix a few things - the dynamic linker expects certain values here,
4254 and we must not dissappoint it. */
4255 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4256 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4257 && fixP->fx_addsy)
4258 switch (fixP->fx_r_type)
4259 {
4260 case BFD_RELOC_386_PLT32:
4261 case BFD_RELOC_X86_64_PLT32:
4262 /* Make the jump instruction point to the address of the operand. At
4263 runtime we merely add the offset to the actual PLT entry. */
4264 value = -4;
4265 break;
4266 case BFD_RELOC_386_GOTPC:
4267
4268 /* This is tough to explain. We end up with this one if we have
4269 * operands that look like "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal
4270 * here is to obtain the absolute address of the GOT, and it is strongly
4271 * preferable from a performance point of view to avoid using a runtime
4272 * relocation for this. The actual sequence of instructions often look
4273 * something like:
4274 *
4275 * call .L66
4276 * .L66:
4277 * popl %ebx
4278 * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
4279 *
4280 * The call and pop essentially return the absolute address of
4281 * the label .L66 and store it in %ebx. The linker itself will
4282 * ultimately change the first operand of the addl so that %ebx points to
4283 * the GOT, but to keep things simple, the .o file must have this operand
4284 * set so that it generates not the absolute address of .L66, but the
4285 * absolute address of itself. This allows the linker itself simply
4286 * treat a GOTPC relocation as asking for a pcrel offset to the GOT to be
4287 * added in, and the addend of the relocation is stored in the operand
4288 * field for the instruction itself.
4289 *
4290 * Our job here is to fix the operand so that it would add the correct
4291 * offset so that %ebx would point to itself. The thing that is tricky is
4292 * that .-.L66 will point to the beginning of the instruction, so we need
4293 * to further modify the operand so that it will point to itself.
4294 * There are other cases where you have something like:
4295 *
4296 * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
4297 *
4298 * and here no correction would be required. Internally in the assembler
4299 * we treat operands of this form as not being pcrel since the '.' is
4300 * explicitly mentioned, and I wonder whether it would simplify matters
4301 * to do it this way. Who knows. In earlier versions of the PIC patches,
4302 * the pcrel_adjust field was used to store the correction, but since the
4303 * expression is not pcrel, I felt it would be confusing to do it this
4304 * way. */
4305
4306 value -= 1;
4307 break;
4308 case BFD_RELOC_386_GOT32:
4309 case BFD_RELOC_X86_64_GOT32:
4310 value = 0; /* Fully resolved at runtime. No addend. */
4311 break;
4312 case BFD_RELOC_386_GOTOFF:
4313 case BFD_RELOC_X86_64_GOTPCREL:
4314 break;
4315
4316 case BFD_RELOC_VTABLE_INHERIT:
4317 case BFD_RELOC_VTABLE_ENTRY:
4318 fixP->fx_done = 0;
4319 return 1;
4320
4321 default:
4322 break;
4323 }
4324 #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
4325 *valp = value;
4326 #endif /* defined (BFD_ASSEMBLER) && !defined (TE_Mach) */
4327
4328 #ifndef BFD_ASSEMBLER
4329 md_number_to_chars (p, value, fixP->fx_size);
4330 #else
4331 /* Are we finished with this relocation now? */
4332 if (fixP->fx_addsy == 0 && fixP->fx_pcrel == 0)
4333 fixP->fx_done = 1;
4334 else if (use_rela_relocations)
4335 {
4336 fixP->fx_no_overflow = 1;
4337 value = 0;
4338 }
4339 md_number_to_chars (p, value, fixP->fx_size);
4340 #endif
4341
4342 return 1;
4343 }
4344 \f
4345 #define MAX_LITTLENUMS 6
4346
4347 /* Turn the string pointed to by litP into a floating point constant
4348 of type TYPE, and emit the appropriate bytes. The number of
4349 LITTLENUMS emitted is stored in *SIZEP. An error message is
4350 returned, or NULL on OK. */
4351
4352 char *
4353 md_atof (type, litP, sizeP)
4354 int type;
4355 char *litP;
4356 int *sizeP;
4357 {
4358 int prec;
4359 LITTLENUM_TYPE words[MAX_LITTLENUMS];
4360 LITTLENUM_TYPE *wordP;
4361 char *t;
4362
4363 switch (type)
4364 {
4365 case 'f':
4366 case 'F':
4367 prec = 2;
4368 break;
4369
4370 case 'd':
4371 case 'D':
4372 prec = 4;
4373 break;
4374
4375 case 'x':
4376 case 'X':
4377 prec = 5;
4378 break;
4379
4380 default:
4381 *sizeP = 0;
4382 return _("Bad call to md_atof ()");
4383 }
4384 t = atof_ieee (input_line_pointer, type, words);
4385 if (t)
4386 input_line_pointer = t;
4387
4388 *sizeP = prec * sizeof (LITTLENUM_TYPE);
4389 /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
4390 the bigendian 386. */
4391 for (wordP = words + prec - 1; prec--;)
4392 {
4393 md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
4394 litP += sizeof (LITTLENUM_TYPE);
4395 }
4396 return 0;
4397 }
4398 \f
4399 char output_invalid_buf[8];
4400
4401 static char *
4402 output_invalid (c)
4403 int c;
4404 {
4405 if (isprint (c))
4406 sprintf (output_invalid_buf, "'%c'", c);
4407 else
4408 sprintf (output_invalid_buf, "(0x%x)", (unsigned) c);
4409 return output_invalid_buf;
4410 }
4411
4412 /* REG_STRING starts *before* REGISTER_PREFIX. */
4413
4414 static const reg_entry *
4415 parse_register (reg_string, end_op)
4416 char *reg_string;
4417 char **end_op;
4418 {
4419 char *s = reg_string;
4420 char *p;
4421 char reg_name_given[MAX_REG_NAME_SIZE + 1];
4422 const reg_entry *r;
4423
4424 /* Skip possible REGISTER_PREFIX and possible whitespace. */
4425 if (*s == REGISTER_PREFIX)
4426 ++s;
4427
4428 if (is_space_char (*s))
4429 ++s;
4430
4431 p = reg_name_given;
4432 while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
4433 {
4434 if (p >= reg_name_given + MAX_REG_NAME_SIZE)
4435 return (const reg_entry *) NULL;
4436 s++;
4437 }
4438
4439 /* For naked regs, make sure that we are not dealing with an identifier.
4440 This prevents confusing an identifier like `eax_var' with register
4441 `eax'. */
4442 if (allow_naked_reg && identifier_chars[(unsigned char) *s])
4443 return (const reg_entry *) NULL;
4444
4445 *end_op = s;
4446
4447 r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
4448
4449 /* Handle floating point regs, allowing spaces in the (i) part. */
4450 if (r == i386_regtab /* %st is first entry of table */)
4451 {
4452 if (is_space_char (*s))
4453 ++s;
4454 if (*s == '(')
4455 {
4456 ++s;
4457 if (is_space_char (*s))
4458 ++s;
4459 if (*s >= '0' && *s <= '7')
4460 {
4461 r = &i386_float_regtab[*s - '0'];
4462 ++s;
4463 if (is_space_char (*s))
4464 ++s;
4465 if (*s == ')')
4466 {
4467 *end_op = s + 1;
4468 return r;
4469 }
4470 }
4471 /* We have "%st(" then garbage. */
4472 return (const reg_entry *) NULL;
4473 }
4474 }
4475
4476 return r;
4477 }
4478 \f
4479 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4480 const char *md_shortopts = "kVQ:sq";
4481 #else
4482 const char *md_shortopts = "q";
4483 #endif
4484
4485 struct option md_longopts[] = {
4486 #define OPTION_32 (OPTION_MD_BASE + 0)
4487 {"32", no_argument, NULL, OPTION_32},
4488 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4489 #define OPTION_64 (OPTION_MD_BASE + 1)
4490 {"64", no_argument, NULL, OPTION_64},
4491 #endif
4492 {NULL, no_argument, NULL, 0}
4493 };
4494 size_t md_longopts_size = sizeof (md_longopts);
4495
4496 int
4497 md_parse_option (c, arg)
4498 int c;
4499 char *arg ATTRIBUTE_UNUSED;
4500 {
4501 switch (c)
4502 {
4503 case 'q':
4504 quiet_warnings = 1;
4505 break;
4506
4507 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4508 /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
4509 should be emitted or not. FIXME: Not implemented. */
4510 case 'Q':
4511 break;
4512
4513 /* -V: SVR4 argument to print version ID. */
4514 case 'V':
4515 print_version_id ();
4516 break;
4517
4518 /* -k: Ignore for FreeBSD compatibility. */
4519 case 'k':
4520 break;
4521
4522 case 's':
4523 /* -s: On i386 Solaris, this tells the native assembler to use
4524 .stab instead of .stab.excl. We always use .stab anyhow. */
4525 break;
4526
4527 case OPTION_64:
4528 {
4529 const char **list, **l;
4530
4531 list = bfd_target_list ();
4532 for (l = list; *l != NULL; l++)
4533 if (strcmp (*l, "elf64-x86-64") == 0)
4534 {
4535 default_arch = "x86_64";
4536 break;
4537 }
4538 if (*l == NULL)
4539 as_fatal (_("No compiled in support for x86_64"));
4540 free (list);
4541 }
4542 break;
4543 #endif
4544
4545 case OPTION_32:
4546 default_arch = "i386";
4547 break;
4548
4549 default:
4550 return 0;
4551 }
4552 return 1;
4553 }
4554
4555 void
4556 md_show_usage (stream)
4557 FILE *stream;
4558 {
4559 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4560 fprintf (stream, _("\
4561 -Q ignored\n\
4562 -V print assembler version number\n\
4563 -k ignored\n\
4564 -q quieten some warnings\n\
4565 -s ignored\n"));
4566 #else
4567 fprintf (stream, _("\
4568 -q quieten some warnings\n"));
4569 #endif
4570 }
4571
4572 #ifdef BFD_ASSEMBLER
4573 #if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
4574 || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
4575
4576 /* Pick the target format to use. */
4577
4578 const char *
4579 i386_target_format ()
4580 {
4581 if (!strcmp (default_arch, "x86_64"))
4582 set_code_flag (CODE_64BIT);
4583 else if (!strcmp (default_arch, "i386"))
4584 set_code_flag (CODE_32BIT);
4585 else
4586 as_fatal (_("Unknown architecture"));
4587 switch (OUTPUT_FLAVOR)
4588 {
4589 #ifdef OBJ_MAYBE_AOUT
4590 case bfd_target_aout_flavour:
4591 return AOUT_TARGET_FORMAT;
4592 #endif
4593 #ifdef OBJ_MAYBE_COFF
4594 case bfd_target_coff_flavour:
4595 return "coff-i386";
4596 #endif
4597 #if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF)
4598 case bfd_target_elf_flavour:
4599 {
4600 if (flag_code == CODE_64BIT)
4601 use_rela_relocations = 1;
4602 return flag_code == CODE_64BIT ? "elf64-x86-64" : "elf32-i386";
4603 }
4604 #endif
4605 default:
4606 abort ();
4607 return NULL;
4608 }
4609 }
4610
4611 #endif /* OBJ_MAYBE_ more than one */
4612 #endif /* BFD_ASSEMBLER */
4613 \f
4614 symbolS *
4615 md_undefined_symbol (name)
4616 char *name;
4617 {
4618 if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
4619 && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
4620 && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
4621 && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
4622 {
4623 if (!GOT_symbol)
4624 {
4625 if (symbol_find (name))
4626 as_bad (_("GOT already in symbol table"));
4627 GOT_symbol = symbol_new (name, undefined_section,
4628 (valueT) 0, &zero_address_frag);
4629 };
4630 return GOT_symbol;
4631 }
4632 return 0;
4633 }
4634
4635 /* Round up a section size to the appropriate boundary. */
4636
4637 valueT
4638 md_section_align (segment, size)
4639 segT segment ATTRIBUTE_UNUSED;
4640 valueT size;
4641 {
4642 #ifdef BFD_ASSEMBLER
4643 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4644 if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
4645 {
4646 /* For a.out, force the section size to be aligned. If we don't do
4647 this, BFD will align it for us, but it will not write out the
4648 final bytes of the section. This may be a bug in BFD, but it is
4649 easier to fix it here since that is how the other a.out targets
4650 work. */
4651 int align;
4652
4653 align = bfd_get_section_alignment (stdoutput, segment);
4654 size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
4655 }
4656 #endif
4657 #endif
4658
4659 return size;
4660 }
4661
4662 /* On the i386, PC-relative offsets are relative to the start of the
4663 next instruction. That is, the address of the offset, plus its
4664 size, since the offset is always the last part of the insn. */
4665
4666 long
4667 md_pcrel_from (fixP)
4668 fixS *fixP;
4669 {
4670 return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
4671 }
4672
4673 #ifndef I386COFF
4674
4675 static void
4676 s_bss (ignore)
4677 int ignore ATTRIBUTE_UNUSED;
4678 {
4679 register int temp;
4680
4681 temp = get_absolute_expression ();
4682 subseg_set (bss_section, (subsegT) temp);
4683 demand_empty_rest_of_line ();
4684 }
4685
4686 #endif
4687
4688 #ifdef BFD_ASSEMBLER
4689
4690 void
4691 i386_validate_fix (fixp)
4692 fixS *fixp;
4693 {
4694 if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
4695 {
4696 /* GOTOFF relocation are nonsense in 64bit mode. */
4697 if (fixp->fx_r_type == BFD_RELOC_32_PCREL)
4698 {
4699 if (flag_code != CODE_64BIT)
4700 abort ();
4701 fixp->fx_r_type = BFD_RELOC_X86_64_GOTPCREL;
4702 }
4703 else
4704 {
4705 if (flag_code == CODE_64BIT)
4706 abort ();
4707 fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
4708 }
4709 fixp->fx_subsy = 0;
4710 }
4711 }
4712
4713 arelent *
4714 tc_gen_reloc (section, fixp)
4715 asection *section ATTRIBUTE_UNUSED;
4716 fixS *fixp;
4717 {
4718 arelent *rel;
4719 bfd_reloc_code_real_type code;
4720
4721 switch (fixp->fx_r_type)
4722 {
4723 case BFD_RELOC_X86_64_PLT32:
4724 case BFD_RELOC_X86_64_GOT32:
4725 case BFD_RELOC_X86_64_GOTPCREL:
4726 case BFD_RELOC_386_PLT32:
4727 case BFD_RELOC_386_GOT32:
4728 case BFD_RELOC_386_GOTOFF:
4729 case BFD_RELOC_386_GOTPC:
4730 case BFD_RELOC_X86_64_32S:
4731 case BFD_RELOC_RVA:
4732 case BFD_RELOC_VTABLE_ENTRY:
4733 case BFD_RELOC_VTABLE_INHERIT:
4734 code = fixp->fx_r_type;
4735 break;
4736 default:
4737 if (fixp->fx_pcrel)
4738 {
4739 switch (fixp->fx_size)
4740 {
4741 default:
4742 as_bad (_("can not do %d byte pc-relative relocation"),
4743 fixp->fx_size);
4744 code = BFD_RELOC_32_PCREL;
4745 break;
4746 case 1: code = BFD_RELOC_8_PCREL; break;
4747 case 2: code = BFD_RELOC_16_PCREL; break;
4748 case 4: code = BFD_RELOC_32_PCREL; break;
4749 }
4750 }
4751 else
4752 {
4753 switch (fixp->fx_size)
4754 {
4755 default:
4756 as_bad (_("can not do %d byte relocation"), fixp->fx_size);
4757 code = BFD_RELOC_32;
4758 break;
4759 case 1: code = BFD_RELOC_8; break;
4760 case 2: code = BFD_RELOC_16; break;
4761 case 4: code = BFD_RELOC_32; break;
4762 case 8: code = BFD_RELOC_64; break;
4763 }
4764 }
4765 break;
4766 }
4767
4768 if (code == BFD_RELOC_32
4769 && GOT_symbol
4770 && fixp->fx_addsy == GOT_symbol)
4771 {
4772 /* We don't support GOTPC on 64bit targets. */
4773 if (flag_code == CODE_64BIT)
4774 abort ();
4775 code = BFD_RELOC_386_GOTPC;
4776 }
4777
4778 rel = (arelent *) xmalloc (sizeof (arelent));
4779 rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
4780 *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
4781
4782 rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
4783 if (!use_rela_relocations)
4784 {
4785 /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
4786 vtable entry to be used in the relocation's section offset. */
4787 if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
4788 rel->address = fixp->fx_offset;
4789
4790 if (fixp->fx_pcrel)
4791 rel->addend = fixp->fx_addnumber;
4792 else
4793 rel->addend = 0;
4794 }
4795 /* Use the rela in 64bit mode. */
4796 else
4797 {
4798 rel->addend = fixp->fx_offset;
4799 if (fixp->fx_pcrel)
4800 rel->addend -= fixp->fx_size;
4801 }
4802
4803 rel->howto = bfd_reloc_type_lookup (stdoutput, code);
4804 if (rel->howto == NULL)
4805 {
4806 as_bad_where (fixp->fx_file, fixp->fx_line,
4807 _("cannot represent relocation type %s"),
4808 bfd_get_reloc_code_name (code));
4809 /* Set howto to a garbage value so that we can keep going. */
4810 rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
4811 assert (rel->howto != NULL);
4812 }
4813
4814 return rel;
4815 }
4816
4817 #else /* ! BFD_ASSEMBLER */
4818
4819 #if (defined(OBJ_AOUT) | defined(OBJ_BOUT))
4820 void
4821 tc_aout_fix_to_chars (where, fixP, segment_address_in_file)
4822 char *where;
4823 fixS *fixP;
4824 relax_addressT segment_address_in_file;
4825 {
4826 /* In: length of relocation (or of address) in chars: 1, 2 or 4.
4827 Out: GNU LD relocation length code: 0, 1, or 2. */
4828
4829 static const unsigned char nbytes_r_length[] = { 42, 0, 1, 42, 2 };
4830 long r_symbolnum;
4831
4832 know (fixP->fx_addsy != NULL);
4833
4834 md_number_to_chars (where,
4835 (valueT) (fixP->fx_frag->fr_address
4836 + fixP->fx_where - segment_address_in_file),
4837 4);
4838
4839 r_symbolnum = (S_IS_DEFINED (fixP->fx_addsy)
4840 ? S_GET_TYPE (fixP->fx_addsy)
4841 : fixP->fx_addsy->sy_number);
4842
4843 where[6] = (r_symbolnum >> 16) & 0x0ff;
4844 where[5] = (r_symbolnum >> 8) & 0x0ff;
4845 where[4] = r_symbolnum & 0x0ff;
4846 where[7] = ((((!S_IS_DEFINED (fixP->fx_addsy)) << 3) & 0x08)
4847 | ((nbytes_r_length[fixP->fx_size] << 1) & 0x06)
4848 | (((fixP->fx_pcrel << 0) & 0x01) & 0x0f));
4849 }
4850
4851 #endif /* OBJ_AOUT or OBJ_BOUT. */
4852
4853 #if defined (I386COFF)
4854
4855 short
4856 tc_coff_fix2rtype (fixP)
4857 fixS *fixP;
4858 {
4859 if (fixP->fx_r_type == R_IMAGEBASE)
4860 return R_IMAGEBASE;
4861
4862 return (fixP->fx_pcrel ?
4863 (fixP->fx_size == 1 ? R_PCRBYTE :
4864 fixP->fx_size == 2 ? R_PCRWORD :
4865 R_PCRLONG) :
4866 (fixP->fx_size == 1 ? R_RELBYTE :
4867 fixP->fx_size == 2 ? R_RELWORD :
4868 R_DIR32));
4869 }
4870
4871 int
4872 tc_coff_sizemachdep (frag)
4873 fragS *frag;
4874 {
4875 if (frag->fr_next)
4876 return (frag->fr_next->fr_address - frag->fr_address);
4877 else
4878 return 0;
4879 }
4880
4881 #endif /* I386COFF */
4882
4883 #endif /* ! BFD_ASSEMBLER */
4884 \f
4885 /* Parse operands using Intel syntax. This implements a recursive descent
4886 parser based on the BNF grammar published in Appendix B of the MASM 6.1
4887 Programmer's Guide.
4888
4889 FIXME: We do not recognize the full operand grammar defined in the MASM
4890 documentation. In particular, all the structure/union and
4891 high-level macro operands are missing.
4892
4893 Uppercase words are terminals, lower case words are non-terminals.
4894 Objects surrounded by double brackets '[[' ']]' are optional. Vertical
4895 bars '|' denote choices. Most grammar productions are implemented in
4896 functions called 'intel_<production>'.
4897
4898 Initial production is 'expr'.
4899
4900 addOp + | -
4901
4902 alpha [a-zA-Z]
4903
4904 byteRegister AL | AH | BL | BH | CL | CH | DL | DH
4905
4906 constant digits [[ radixOverride ]]
4907
4908 dataType BYTE | WORD | DWORD | QWORD | XWORD
4909
4910 digits decdigit
4911 | digits decdigit
4912 | digits hexdigit
4913
4914 decdigit [0-9]
4915
4916 e05 e05 addOp e06
4917 | e06
4918
4919 e06 e06 mulOp e09
4920 | e09
4921
4922 e09 OFFSET e10
4923 | e09 PTR e10
4924 | e09 : e10
4925 | e10
4926
4927 e10 e10 [ expr ]
4928 | e11
4929
4930 e11 ( expr )
4931 | [ expr ]
4932 | constant
4933 | dataType
4934 | id
4935 | $
4936 | register
4937
4938 => expr SHORT e05
4939 | e05
4940
4941 gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
4942 | BP | EBP | SP | ESP | DI | EDI | SI | ESI
4943
4944 hexdigit a | b | c | d | e | f
4945 | A | B | C | D | E | F
4946
4947 id alpha
4948 | id alpha
4949 | id decdigit
4950
4951 mulOp * | / | MOD
4952
4953 quote " | '
4954
4955 register specialRegister
4956 | gpRegister
4957 | byteRegister
4958
4959 segmentRegister CS | DS | ES | FS | GS | SS
4960
4961 specialRegister CR0 | CR2 | CR3
4962 | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
4963 | TR3 | TR4 | TR5 | TR6 | TR7
4964
4965 We simplify the grammar in obvious places (e.g., register parsing is
4966 done by calling parse_register) and eliminate immediate left recursion
4967 to implement a recursive-descent parser.
4968
4969 expr SHORT e05
4970 | e05
4971
4972 e05 e06 e05'
4973
4974 e05' addOp e06 e05'
4975 | Empty
4976
4977 e06 e09 e06'
4978
4979 e06' mulOp e09 e06'
4980 | Empty
4981
4982 e09 OFFSET e10 e09'
4983 | e10 e09'
4984
4985 e09' PTR e10 e09'
4986 | : e10 e09'
4987 | Empty
4988
4989 e10 e11 e10'
4990
4991 e10' [ expr ] e10'
4992 | Empty
4993
4994 e11 ( expr )
4995 | [ expr ]
4996 | BYTE
4997 | WORD
4998 | DWORD
4999 | QWORD
5000 | XWORD
5001 | .
5002 | $
5003 | register
5004 | id
5005 | constant */
5006
5007 /* Parsing structure for the intel syntax parser. Used to implement the
5008 semantic actions for the operand grammar. */
5009 struct intel_parser_s
5010 {
5011 char *op_string; /* The string being parsed. */
5012 int got_a_float; /* Whether the operand is a float. */
5013 int op_modifier; /* Operand modifier. */
5014 int is_mem; /* 1 if operand is memory reference. */
5015 const reg_entry *reg; /* Last register reference found. */
5016 char *disp; /* Displacement string being built. */
5017 };
5018
5019 static struct intel_parser_s intel_parser;
5020
5021 /* Token structure for parsing intel syntax. */
5022 struct intel_token
5023 {
5024 int code; /* Token code. */
5025 const reg_entry *reg; /* Register entry for register tokens. */
5026 char *str; /* String representation. */
5027 };
5028
5029 static struct intel_token cur_token, prev_token;
5030
5031 /* Token codes for the intel parser. Since T_SHORT is already used
5032 by COFF, undefine it first to prevent a warning. */
5033 #define T_NIL -1
5034 #define T_CONST 1
5035 #define T_REG 2
5036 #define T_BYTE 3
5037 #define T_WORD 4
5038 #define T_DWORD 5
5039 #define T_QWORD 6
5040 #define T_XWORD 7
5041 #undef T_SHORT
5042 #define T_SHORT 8
5043 #define T_OFFSET 9
5044 #define T_PTR 10
5045 #define T_ID 11
5046
5047 /* Prototypes for intel parser functions. */
5048 static int intel_match_token PARAMS ((int code));
5049 static void intel_get_token PARAMS ((void));
5050 static void intel_putback_token PARAMS ((void));
5051 static int intel_expr PARAMS ((void));
5052 static int intel_e05 PARAMS ((void));
5053 static int intel_e05_1 PARAMS ((void));
5054 static int intel_e06 PARAMS ((void));
5055 static int intel_e06_1 PARAMS ((void));
5056 static int intel_e09 PARAMS ((void));
5057 static int intel_e09_1 PARAMS ((void));
5058 static int intel_e10 PARAMS ((void));
5059 static int intel_e10_1 PARAMS ((void));
5060 static int intel_e11 PARAMS ((void));
5061
5062 static int
5063 i386_intel_operand (operand_string, got_a_float)
5064 char *operand_string;
5065 int got_a_float;
5066 {
5067 int ret;
5068 char *p;
5069
5070 /* Initialize token holders. */
5071 cur_token.code = prev_token.code = T_NIL;
5072 cur_token.reg = prev_token.reg = NULL;
5073 cur_token.str = prev_token.str = NULL;
5074
5075 /* Initialize parser structure. */
5076 p = intel_parser.op_string = (char *) malloc (strlen (operand_string) + 1);
5077 if (p == NULL)
5078 abort ();
5079 strcpy (intel_parser.op_string, operand_string);
5080 intel_parser.got_a_float = got_a_float;
5081 intel_parser.op_modifier = -1;
5082 intel_parser.is_mem = 0;
5083 intel_parser.reg = NULL;
5084 intel_parser.disp = (char *) malloc (strlen (operand_string) + 1);
5085 if (intel_parser.disp == NULL)
5086 abort ();
5087 intel_parser.disp[0] = '\0';
5088
5089 /* Read the first token and start the parser. */
5090 intel_get_token ();
5091 ret = intel_expr ();
5092
5093 if (ret)
5094 {
5095 /* If we found a memory reference, hand it over to i386_displacement
5096 to fill in the rest of the operand fields. */
5097 if (intel_parser.is_mem)
5098 {
5099 if ((i.mem_operands == 1
5100 && (current_templates->start->opcode_modifier & IsString) == 0)
5101 || i.mem_operands == 2)
5102 {
5103 as_bad (_("too many memory references for '%s'"),
5104 current_templates->start->name);
5105 ret = 0;
5106 }
5107 else
5108 {
5109 char *s = intel_parser.disp;
5110 i.mem_operands++;
5111
5112 /* Add the displacement expression. */
5113 if (*s != '\0')
5114 ret = i386_displacement (s, s + strlen (s))
5115 && i386_index_check (s);
5116 }
5117 }
5118
5119 /* Constant and OFFSET expressions are handled by i386_immediate. */
5120 else if (intel_parser.op_modifier == OFFSET_FLAT
5121 || intel_parser.reg == NULL)
5122 ret = i386_immediate (intel_parser.disp);
5123 }
5124
5125 free (p);
5126 free (intel_parser.disp);
5127
5128 return ret;
5129 }
5130
5131 /* expr SHORT e05
5132 | e05 */
5133 static int
5134 intel_expr ()
5135 {
5136 /* expr SHORT e05 */
5137 if (cur_token.code == T_SHORT)
5138 {
5139 intel_parser.op_modifier = SHORT;
5140 intel_match_token (T_SHORT);
5141
5142 return (intel_e05 ());
5143 }
5144
5145 /* expr e05 */
5146 else
5147 return intel_e05 ();
5148 }
5149
5150 /* e05 e06 e05'
5151
5152 e05' addOp e06 e05'
5153 | Empty */
5154 static int
5155 intel_e05 ()
5156 {
5157 return (intel_e06 () && intel_e05_1 ());
5158 }
5159
5160 static int
5161 intel_e05_1 ()
5162 {
5163 /* e05' addOp e06 e05' */
5164 if (cur_token.code == '+' || cur_token.code == '-')
5165 {
5166 strcat (intel_parser.disp, cur_token.str);
5167 intel_match_token (cur_token.code);
5168
5169 return (intel_e06 () && intel_e05_1 ());
5170 }
5171
5172 /* e05' Empty */
5173 else
5174 return 1;
5175 }
5176
5177 /* e06 e09 e06'
5178
5179 e06' mulOp e09 e06'
5180 | Empty */
5181 static int
5182 intel_e06 ()
5183 {
5184 return (intel_e09 () && intel_e06_1 ());
5185 }
5186
5187 static int
5188 intel_e06_1 ()
5189 {
5190 /* e06' mulOp e09 e06' */
5191 if (cur_token.code == '*' || cur_token.code == '/')
5192 {
5193 strcat (intel_parser.disp, cur_token.str);
5194 intel_match_token (cur_token.code);
5195
5196 return (intel_e09 () && intel_e06_1 ());
5197 }
5198
5199 /* e06' Empty */
5200 else
5201 return 1;
5202 }
5203
5204 /* e09 OFFSET e10 e09'
5205 | e10 e09'
5206
5207 e09' PTR e10 e09'
5208 | : e10 e09'
5209 | Empty */
5210 static int
5211 intel_e09 ()
5212 {
5213 /* e09 OFFSET e10 e09' */
5214 if (cur_token.code == T_OFFSET)
5215 {
5216 intel_parser.is_mem = 0;
5217 intel_parser.op_modifier = OFFSET_FLAT;
5218 intel_match_token (T_OFFSET);
5219
5220 return (intel_e10 () && intel_e09_1 ());
5221 }
5222
5223 /* e09 e10 e09' */
5224 else
5225 return (intel_e10 () && intel_e09_1 ());
5226 }
5227
5228 static int
5229 intel_e09_1 ()
5230 {
5231 /* e09' PTR e10 e09' */
5232 if (cur_token.code == T_PTR)
5233 {
5234 if (prev_token.code == T_BYTE)
5235 i.suffix = BYTE_MNEM_SUFFIX;
5236
5237 else if (prev_token.code == T_WORD)
5238 {
5239 if (intel_parser.got_a_float == 2) /* "fi..." */
5240 i.suffix = SHORT_MNEM_SUFFIX;
5241 else
5242 i.suffix = WORD_MNEM_SUFFIX;
5243 }
5244
5245 else if (prev_token.code == T_DWORD)
5246 {
5247 if (intel_parser.got_a_float == 1) /* "f..." */
5248 i.suffix = SHORT_MNEM_SUFFIX;
5249 else
5250 i.suffix = LONG_MNEM_SUFFIX;
5251 }
5252
5253 else if (prev_token.code == T_QWORD)
5254 {
5255 if (intel_parser.got_a_float == 1) /* "f..." */
5256 i.suffix = LONG_MNEM_SUFFIX;
5257 else
5258 i.suffix = QWORD_MNEM_SUFFIX;
5259 }
5260
5261 else if (prev_token.code == T_XWORD)
5262 i.suffix = LONG_DOUBLE_MNEM_SUFFIX;
5263
5264 else
5265 {
5266 as_bad (_("Unknown operand modifier `%s'\n"), prev_token.str);
5267 return 0;
5268 }
5269
5270 intel_match_token (T_PTR);
5271
5272 return (intel_e10 () && intel_e09_1 ());
5273 }
5274
5275 /* e09 : e10 e09' */
5276 else if (cur_token.code == ':')
5277 {
5278 /* Mark as a memory operand only if it's not already known to be an
5279 offset expression. */
5280 if (intel_parser.op_modifier != OFFSET_FLAT)
5281 intel_parser.is_mem = 1;
5282
5283 return (intel_match_token (':') && intel_e10 () && intel_e09_1 ());
5284 }
5285
5286 /* e09' Empty */
5287 else
5288 return 1;
5289 }
5290
5291 /* e10 e11 e10'
5292
5293 e10' [ expr ] e10'
5294 | Empty */
5295 static int
5296 intel_e10 ()
5297 {
5298 return (intel_e11 () && intel_e10_1 ());
5299 }
5300
5301 static int
5302 intel_e10_1 ()
5303 {
5304 /* e10' [ expr ] e10' */
5305 if (cur_token.code == '[')
5306 {
5307 intel_match_token ('[');
5308
5309 /* Mark as a memory operand only if it's not already known to be an
5310 offset expression. If it's an offset expression, we need to keep
5311 the brace in. */
5312 if (intel_parser.op_modifier != OFFSET_FLAT)
5313 intel_parser.is_mem = 1;
5314 else
5315 strcat (intel_parser.disp, "[");
5316
5317 /* Add a '+' to the displacement string if necessary. */
5318 if (*intel_parser.disp != '\0'
5319 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5320 strcat (intel_parser.disp, "+");
5321
5322 if (intel_expr () && intel_match_token (']'))
5323 {
5324 /* Preserve brackets when the operand is an offset expression. */
5325 if (intel_parser.op_modifier == OFFSET_FLAT)
5326 strcat (intel_parser.disp, "]");
5327
5328 return intel_e10_1 ();
5329 }
5330 else
5331 return 0;
5332 }
5333
5334 /* e10' Empty */
5335 else
5336 return 1;
5337 }
5338
5339 /* e11 ( expr )
5340 | [ expr ]
5341 | BYTE
5342 | WORD
5343 | DWORD
5344 | QWORD
5345 | XWORD
5346 | $
5347 | .
5348 | register
5349 | id
5350 | constant */
5351 static int
5352 intel_e11 ()
5353 {
5354 /* e11 ( expr ) */
5355 if (cur_token.code == '(')
5356 {
5357 intel_match_token ('(');
5358 strcat (intel_parser.disp, "(");
5359
5360 if (intel_expr () && intel_match_token (')'))
5361 {
5362 strcat (intel_parser.disp, ")");
5363 return 1;
5364 }
5365 else
5366 return 0;
5367 }
5368
5369 /* e11 [ expr ] */
5370 else if (cur_token.code == '[')
5371 {
5372 intel_match_token ('[');
5373
5374 /* Mark as a memory operand only if it's not already known to be an
5375 offset expression. If it's an offset expression, we need to keep
5376 the brace in. */
5377 if (intel_parser.op_modifier != OFFSET_FLAT)
5378 intel_parser.is_mem = 1;
5379 else
5380 strcat (intel_parser.disp, "[");
5381
5382 /* Operands for jump/call inside brackets denote absolute addresses. */
5383 if (current_templates->start->opcode_modifier & Jump
5384 || current_templates->start->opcode_modifier & JumpDword
5385 || current_templates->start->opcode_modifier & JumpByte
5386 || current_templates->start->opcode_modifier & JumpInterSegment)
5387 i.types[this_operand] |= JumpAbsolute;
5388
5389 /* Add a '+' to the displacement string if necessary. */
5390 if (*intel_parser.disp != '\0'
5391 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5392 strcat (intel_parser.disp, "+");
5393
5394 if (intel_expr () && intel_match_token (']'))
5395 {
5396 /* Preserve brackets when the operand is an offset expression. */
5397 if (intel_parser.op_modifier == OFFSET_FLAT)
5398 strcat (intel_parser.disp, "]");
5399
5400 return 1;
5401 }
5402 else
5403 return 0;
5404 }
5405
5406 /* e11 BYTE
5407 | WORD
5408 | DWORD
5409 | QWORD
5410 | XWORD */
5411 else if (cur_token.code == T_BYTE
5412 || cur_token.code == T_WORD
5413 || cur_token.code == T_DWORD
5414 || cur_token.code == T_QWORD
5415 || cur_token.code == T_XWORD)
5416 {
5417 intel_match_token (cur_token.code);
5418
5419 return 1;
5420 }
5421
5422 /* e11 $
5423 | . */
5424 else if (cur_token.code == '$' || cur_token.code == '.')
5425 {
5426 strcat (intel_parser.disp, cur_token.str);
5427 intel_match_token (cur_token.code);
5428
5429 /* Mark as a memory operand only if it's not already known to be an
5430 offset expression. */
5431 if (intel_parser.op_modifier != OFFSET_FLAT)
5432 intel_parser.is_mem = 1;
5433
5434 return 1;
5435 }
5436
5437 /* e11 register */
5438 else if (cur_token.code == T_REG)
5439 {
5440 const reg_entry *reg = intel_parser.reg = cur_token.reg;
5441
5442 intel_match_token (T_REG);
5443
5444 /* Check for segment change. */
5445 if (cur_token.code == ':')
5446 {
5447 if (reg->reg_type & (SReg2 | SReg3))
5448 {
5449 switch (reg->reg_num)
5450 {
5451 case 0:
5452 i.seg[i.mem_operands] = &es;
5453 break;
5454 case 1:
5455 i.seg[i.mem_operands] = &cs;
5456 break;
5457 case 2:
5458 i.seg[i.mem_operands] = &ss;
5459 break;
5460 case 3:
5461 i.seg[i.mem_operands] = &ds;
5462 break;
5463 case 4:
5464 i.seg[i.mem_operands] = &fs;
5465 break;
5466 case 5:
5467 i.seg[i.mem_operands] = &gs;
5468 break;
5469 }
5470 }
5471 else
5472 {
5473 as_bad (_("`%s' is not a valid segment register"), reg->reg_name);
5474 return 0;
5475 }
5476 }
5477
5478 /* Not a segment register. Check for register scaling. */
5479 else if (cur_token.code == '*')
5480 {
5481 if (!intel_parser.is_mem)
5482 {
5483 as_bad (_("Register scaling only allowed in memory operands."));
5484 return 0;
5485 }
5486
5487 /* What follows must be a valid scale. */
5488 if (intel_match_token ('*')
5489 && strchr ("01248", *cur_token.str))
5490 {
5491 i.index_reg = reg;
5492 i.types[this_operand] |= BaseIndex;
5493
5494 /* Set the scale after setting the register (otherwise,
5495 i386_scale will complain) */
5496 i386_scale (cur_token.str);
5497 intel_match_token (T_CONST);
5498 }
5499 else
5500 {
5501 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
5502 cur_token.str);
5503 return 0;
5504 }
5505 }
5506
5507 /* No scaling. If this is a memory operand, the register is either a
5508 base register (first occurrence) or an index register (second
5509 occurrence). */
5510 else if (intel_parser.is_mem && !(reg->reg_type & (SReg2 | SReg3)))
5511 {
5512 if (i.base_reg && i.index_reg)
5513 {
5514 as_bad (_("Too many register references in memory operand.\n"));
5515 return 0;
5516 }
5517
5518 if (i.base_reg == NULL)
5519 i.base_reg = reg;
5520 else
5521 i.index_reg = reg;
5522
5523 i.types[this_operand] |= BaseIndex;
5524 }
5525
5526 /* Offset modifier. Add the register to the displacement string to be
5527 parsed as an immediate expression after we're done. */
5528 else if (intel_parser.op_modifier == OFFSET_FLAT)
5529 strcat (intel_parser.disp, reg->reg_name);
5530
5531 /* It's neither base nor index nor offset. */
5532 else
5533 {
5534 i.types[this_operand] |= reg->reg_type & ~BaseIndex;
5535 i.op[this_operand].regs = reg;
5536 i.reg_operands++;
5537 }
5538
5539 /* Since registers are not part of the displacement string (except
5540 when we're parsing offset operands), we may need to remove any
5541 preceding '+' from the displacement string. */
5542 if (*intel_parser.disp != '\0'
5543 && intel_parser.op_modifier != OFFSET_FLAT)
5544 {
5545 char *s = intel_parser.disp;
5546 s += strlen (s) - 1;
5547 if (*s == '+')
5548 *s = '\0';
5549 }
5550
5551 return 1;
5552 }
5553
5554 /* e11 id */
5555 else if (cur_token.code == T_ID)
5556 {
5557 /* Add the identifier to the displacement string. */
5558 strcat (intel_parser.disp, cur_token.str);
5559 intel_match_token (T_ID);
5560
5561 /* The identifier represents a memory reference only if it's not
5562 preceded by an offset modifier. */
5563 if (intel_parser.op_modifier != OFFSET_FLAT)
5564 intel_parser.is_mem = 1;
5565
5566 return 1;
5567 }
5568
5569 /* e11 constant */
5570 else if (cur_token.code == T_CONST
5571 || cur_token.code == '-'
5572 || cur_token.code == '+')
5573 {
5574 char *save_str;
5575
5576 /* Allow constants that start with `+' or `-'. */
5577 if (cur_token.code == '-' || cur_token.code == '+')
5578 {
5579 strcat (intel_parser.disp, cur_token.str);
5580 intel_match_token (cur_token.code);
5581 if (cur_token.code != T_CONST)
5582 {
5583 as_bad (_("Syntax error. Expecting a constant. Got `%s'.\n"),
5584 cur_token.str);
5585 return 0;
5586 }
5587 }
5588
5589 save_str = (char *) malloc (strlen (cur_token.str) + 1);
5590 if (save_str == NULL)
5591 abort ();
5592 strcpy (save_str, cur_token.str);
5593
5594 /* Get the next token to check for register scaling. */
5595 intel_match_token (cur_token.code);
5596
5597 /* Check if this constant is a scaling factor for an index register. */
5598 if (cur_token.code == '*')
5599 {
5600 if (intel_match_token ('*') && cur_token.code == T_REG)
5601 {
5602 if (!intel_parser.is_mem)
5603 {
5604 as_bad (_("Register scaling only allowed in memory operands."));
5605 return 0;
5606 }
5607
5608 /* The constant is followed by `* reg', so it must be
5609 a valid scale. */
5610 if (strchr ("01248", *save_str))
5611 {
5612 i.index_reg = cur_token.reg;
5613 i.types[this_operand] |= BaseIndex;
5614
5615 /* Set the scale after setting the register (otherwise,
5616 i386_scale will complain) */
5617 i386_scale (save_str);
5618 intel_match_token (T_REG);
5619
5620 /* Since registers are not part of the displacement
5621 string, we may need to remove any preceding '+' from
5622 the displacement string. */
5623 if (*intel_parser.disp != '\0')
5624 {
5625 char *s = intel_parser.disp;
5626 s += strlen (s) - 1;
5627 if (*s == '+')
5628 *s = '\0';
5629 }
5630
5631 free (save_str);
5632
5633 return 1;
5634 }
5635 else
5636 return 0;
5637 }
5638
5639 /* The constant was not used for register scaling. Since we have
5640 already consumed the token following `*' we now need to put it
5641 back in the stream. */
5642 else
5643 intel_putback_token ();
5644 }
5645
5646 /* Add the constant to the displacement string. */
5647 strcat (intel_parser.disp, save_str);
5648 free (save_str);
5649
5650 return 1;
5651 }
5652
5653 as_bad (_("Unrecognized token '%s'"), cur_token.str);
5654 return 0;
5655 }
5656
5657 /* Match the given token against cur_token. If they match, read the next
5658 token from the operand string. */
5659 static int
5660 intel_match_token (code)
5661 int code;
5662 {
5663 if (cur_token.code == code)
5664 {
5665 intel_get_token ();
5666 return 1;
5667 }
5668 else
5669 {
5670 as_bad (_("Unexpected token `%s'\n"), cur_token.str);
5671 return 0;
5672 }
5673 }
5674
5675 /* Read a new token from intel_parser.op_string and store it in cur_token. */
5676 static void
5677 intel_get_token ()
5678 {
5679 char *end_op;
5680 const reg_entry *reg;
5681 struct intel_token new_token;
5682
5683 new_token.code = T_NIL;
5684 new_token.reg = NULL;
5685 new_token.str = NULL;
5686
5687 /* Free the memory allocated to the previous token and move
5688 cur_token to prev_token. */
5689 if (prev_token.str)
5690 free (prev_token.str);
5691
5692 prev_token = cur_token;
5693
5694 /* Skip whitespace. */
5695 while (is_space_char (*intel_parser.op_string))
5696 intel_parser.op_string++;
5697
5698 /* Return an empty token if we find nothing else on the line. */
5699 if (*intel_parser.op_string == '\0')
5700 {
5701 cur_token = new_token;
5702 return;
5703 }
5704
5705 /* The new token cannot be larger than the remainder of the operand
5706 string. */
5707 new_token.str = (char *) malloc (strlen (intel_parser.op_string) + 1);
5708 if (new_token.str == NULL)
5709 abort ();
5710 new_token.str[0] = '\0';
5711
5712 if (strchr ("0123456789", *intel_parser.op_string))
5713 {
5714 char *p = new_token.str;
5715 char *q = intel_parser.op_string;
5716 new_token.code = T_CONST;
5717
5718 /* Allow any kind of identifier char to encompass floating point and
5719 hexadecimal numbers. */
5720 while (is_identifier_char (*q))
5721 *p++ = *q++;
5722 *p = '\0';
5723
5724 /* Recognize special symbol names [0-9][bf]. */
5725 if (strlen (intel_parser.op_string) == 2
5726 && (intel_parser.op_string[1] == 'b'
5727 || intel_parser.op_string[1] == 'f'))
5728 new_token.code = T_ID;
5729 }
5730
5731 else if (strchr ("+-/*:[]()", *intel_parser.op_string))
5732 {
5733 new_token.code = *intel_parser.op_string;
5734 new_token.str[0] = *intel_parser.op_string;
5735 new_token.str[1] = '\0';
5736 }
5737
5738 else if ((*intel_parser.op_string == REGISTER_PREFIX || allow_naked_reg)
5739 && ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL))
5740 {
5741 new_token.code = T_REG;
5742 new_token.reg = reg;
5743
5744 if (*intel_parser.op_string == REGISTER_PREFIX)
5745 {
5746 new_token.str[0] = REGISTER_PREFIX;
5747 new_token.str[1] = '\0';
5748 }
5749
5750 strcat (new_token.str, reg->reg_name);
5751 }
5752
5753 else if (is_identifier_char (*intel_parser.op_string))
5754 {
5755 char *p = new_token.str;
5756 char *q = intel_parser.op_string;
5757
5758 /* A '.' or '$' followed by an identifier char is an identifier.
5759 Otherwise, it's operator '.' followed by an expression. */
5760 if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
5761 {
5762 new_token.code = *q;
5763 new_token.str[0] = *q;
5764 new_token.str[1] = '\0';
5765 }
5766 else
5767 {
5768 while (is_identifier_char (*q) || *q == '@')
5769 *p++ = *q++;
5770 *p = '\0';
5771
5772 if (strcasecmp (new_token.str, "BYTE") == 0)
5773 new_token.code = T_BYTE;
5774
5775 else if (strcasecmp (new_token.str, "WORD") == 0)
5776 new_token.code = T_WORD;
5777
5778 else if (strcasecmp (new_token.str, "DWORD") == 0)
5779 new_token.code = T_DWORD;
5780
5781 else if (strcasecmp (new_token.str, "QWORD") == 0)
5782 new_token.code = T_QWORD;
5783
5784 else if (strcasecmp (new_token.str, "XWORD") == 0)
5785 new_token.code = T_XWORD;
5786
5787 else if (strcasecmp (new_token.str, "PTR") == 0)
5788 new_token.code = T_PTR;
5789
5790 else if (strcasecmp (new_token.str, "SHORT") == 0)
5791 new_token.code = T_SHORT;
5792
5793 else if (strcasecmp (new_token.str, "OFFSET") == 0)
5794 {
5795 new_token.code = T_OFFSET;
5796
5797 /* ??? This is not mentioned in the MASM grammar but gcc
5798 makes use of it with -mintel-syntax. OFFSET may be
5799 followed by FLAT: */
5800 if (strncasecmp (q, " FLAT:", 6) == 0)
5801 strcat (new_token.str, " FLAT:");
5802 }
5803
5804 /* ??? This is not mentioned in the MASM grammar. */
5805 else if (strcasecmp (new_token.str, "FLAT") == 0)
5806 new_token.code = T_OFFSET;
5807
5808 else
5809 new_token.code = T_ID;
5810 }
5811 }
5812
5813 else
5814 as_bad (_("Unrecognized token `%s'\n"), intel_parser.op_string);
5815
5816 intel_parser.op_string += strlen (new_token.str);
5817 cur_token = new_token;
5818 }
5819
5820 /* Put cur_token back into the token stream and make cur_token point to
5821 prev_token. */
5822 static void
5823 intel_putback_token ()
5824 {
5825 intel_parser.op_string -= strlen (cur_token.str);
5826 free (cur_token.str);
5827 cur_token = prev_token;
5828
5829 /* Forget prev_token. */
5830 prev_token.code = T_NIL;
5831 prev_token.reg = NULL;
5832 prev_token.str = NULL;
5833 }
This page took 0.15442 seconds and 5 git commands to generate.