* tc-i386.h (i386_target_format): Define even for ELFs.
[binutils.git] / gas / config / tc-i386.c
blob0485c6784e48b43e8a534bf76b0340a955520834
1 /* i386.c -- Assemble code for the Intel 80386
2 Copyright (C) 1989, 91, 92, 93, 94, 95, 96, 97, 98, 99, 2000
3 Free Software Foundation, Inc.
5 This file is part of GAS, the GNU Assembler.
7 GAS is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 2, or (at your option)
10 any later version.
12 GAS is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with GAS; see the file COPYING. If not, write to the Free
19 Software Foundation, 59 Temple Place - Suite 330, Boston, MA
20 02111-1307, USA. */
22 /* Intel 80386 machine specific gas.
23 Written by Eliot Dresselhaus (eliot@mgm.mit.edu).
24 x86_64 support by Jan Hubicka (jh@suse.cz)
25 Bugs & suggestions are completely welcome. This is free software.
26 Please help us make it better. */
28 #include <ctype.h>
30 #include "as.h"
31 #include "subsegs.h"
32 #include "dwarf2dbg.h"
33 #include "opcode/i386.h"
35 #ifndef REGISTER_WARNINGS
36 #define REGISTER_WARNINGS 1
37 #endif
39 #ifndef INFER_ADDR_PREFIX
40 #define INFER_ADDR_PREFIX 1
41 #endif
43 #ifndef SCALE1_WHEN_NO_INDEX
44 /* Specifying a scale factor besides 1 when there is no index is
45 futile. eg. `mov (%ebx,2),%al' does exactly the same as
46 `mov (%ebx),%al'. To slavishly follow what the programmer
47 specified, set SCALE1_WHEN_NO_INDEX to 0. */
48 #define SCALE1_WHEN_NO_INDEX 1
49 #endif
51 #define true 1
52 #define false 0
54 static unsigned int mode_from_disp_size PARAMS ((unsigned int));
55 static int fits_in_signed_byte PARAMS ((offsetT));
56 static int fits_in_unsigned_byte PARAMS ((offsetT));
57 static int fits_in_unsigned_word PARAMS ((offsetT));
58 static int fits_in_signed_word PARAMS ((offsetT));
59 static int fits_in_unsigned_long PARAMS ((offsetT));
60 static int fits_in_signed_long PARAMS ((offsetT));
61 static int smallest_imm_type PARAMS ((offsetT));
62 static offsetT offset_in_range PARAMS ((offsetT, int));
63 static int add_prefix PARAMS ((unsigned int));
64 static void set_code_flag PARAMS ((int));
65 static void set_16bit_gcc_code_flag PARAMS ((int));
66 static void set_intel_syntax PARAMS ((int));
67 static void set_cpu_arch PARAMS ((int));
69 #ifdef BFD_ASSEMBLER
70 static bfd_reloc_code_real_type reloc
71 PARAMS ((int, int, int, bfd_reloc_code_real_type));
72 #endif
74 #ifndef DEFAULT_ARCH
75 #define DEFAULT_ARCH "i386"
76 #endif
77 static char *default_arch = DEFAULT_ARCH;
79 /* 'md_assemble ()' gathers together information and puts it into a
80 i386_insn. */
82 union i386_op
84 expressionS *disps;
85 expressionS *imms;
86 const reg_entry *regs;
89 struct _i386_insn
91 /* TM holds the template for the insn were currently assembling. */
92 template tm;
94 /* SUFFIX holds the instruction mnemonic suffix if given.
95 (e.g. 'l' for 'movl') */
96 char suffix;
98 /* OPERANDS gives the number of given operands. */
99 unsigned int operands;
101 /* REG_OPERANDS, DISP_OPERANDS, MEM_OPERANDS, IMM_OPERANDS give the number
102 of given register, displacement, memory operands and immediate
103 operands. */
104 unsigned int reg_operands, disp_operands, mem_operands, imm_operands;
106 /* TYPES [i] is the type (see above #defines) which tells us how to
107 use OP[i] for the corresponding operand. */
108 unsigned int types[MAX_OPERANDS];
110 /* Displacement expression, immediate expression, or register for each
111 operand. */
112 union i386_op op[MAX_OPERANDS];
114 /* Flags for operands. */
115 unsigned int flags[MAX_OPERANDS];
116 #define Operand_PCrel 1
118 /* Relocation type for operand */
119 #ifdef BFD_ASSEMBLER
120 enum bfd_reloc_code_real disp_reloc[MAX_OPERANDS];
121 #else
122 int disp_reloc[MAX_OPERANDS];
123 #endif
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;
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];
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];
140 /* RM and SIB are the modrm byte and the sib byte where the
141 addressing modes of this insn are encoded. */
143 modrm_byte rm;
144 rex_byte rex;
145 sib_byte sib;
148 typedef struct _i386_insn i386_insn;
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
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
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
184 const char line_separator_chars[] = ";";
186 /* Chars that can be used to separate mant from exp in floating point
187 nums. */
188 const char EXP_CHARS[] = "eE";
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";
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];
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])
210 /* All non-digit non-letter charcters that may occur in an operand. */
211 static char operand_special_chars[] = "%$-+(,)*._~/<>|&^!:[@]";
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)
224 /* The instruction we're assembling. */
225 static i386_insn i;
227 /* Possible templates for current insn. */
228 static const templates *current_templates;
230 /* Per instruction expressionS buffers: 2 displacements & 2 immediate max. */
231 static expressionS disp_expressions[2], im_expressions[2];
233 /* Current operand we are working on. */
234 static int this_operand;
236 /* We support four different modes. FLAG_CODE variable is used to distinguish
237 these. */
239 enum flag_code {
240 CODE_32BIT,
241 CODE_16BIT,
242 CODE_64BIT };
244 static enum flag_code flag_code;
245 static int use_rela_relocations = 0;
247 /* The names used to print error messages. */
248 static const char *flag_code_names[] =
250 "32",
251 "16",
252 "64"
255 /* 1 for intel syntax,
256 0 if att syntax. */
257 static int intel_syntax = 0;
259 /* 1 if register prefix % not required. */
260 static int allow_naked_reg = 0;
262 /* Used in 16 bit gcc mode to add an l suffix to call, ret, enter,
263 leave, push, and pop instructions so that gcc has the same stack
264 frame as in 32 bit mode. */
265 static char stackop_size = '\0';
267 /* Non-zero to quieten some warnings. */
268 static int quiet_warnings = 0;
270 /* CPU name. */
271 static const char *cpu_arch_name = NULL;
273 /* CPU feature flags. */
274 static unsigned int cpu_arch_flags = CpuUnknownFlags|CpuNo64;
276 /* Interface to relax_segment.
277 There are 2 relax states for 386 jump insns: one for conditional &
278 one for unconditional jumps. This is because these two types of
279 jumps add different sizes to frags when we're figuring out what
280 sort of jump to choose to reach a given label. */
282 /* Types. */
283 #define COND_JUMP 1
284 #define UNCOND_JUMP 2
285 /* Sizes. */
286 #define CODE16 1
287 #define SMALL 0
288 #define SMALL16 (SMALL|CODE16)
289 #define BIG 2
290 #define BIG16 (BIG|CODE16)
292 #ifndef INLINE
293 #ifdef __GNUC__
294 #define INLINE __inline__
295 #else
296 #define INLINE
297 #endif
298 #endif
300 #define ENCODE_RELAX_STATE(type,size) \
301 ((relax_substateT) ((type<<2) | (size)))
302 #define SIZE_FROM_RELAX_STATE(s) \
303 ( (((s) & 0x3) == BIG ? 4 : (((s) & 0x3) == BIG16 ? 2 : 1)) )
305 /* This table is used by relax_frag to promote short jumps to long
306 ones where necessary. SMALL (short) jumps may be promoted to BIG
307 (32 bit long) ones, and SMALL16 jumps to BIG16 (16 bit long). We
308 don't allow a short jump in a 32 bit code segment to be promoted to
309 a 16 bit offset jump because it's slower (requires data size
310 prefix), and doesn't work, unless the destination is in the bottom
311 64k of the code segment (The top 16 bits of eip are zeroed). */
313 const relax_typeS md_relax_table[] =
315 /* The fields are:
316 1) most positive reach of this state,
317 2) most negative reach of this state,
318 3) how many bytes this mode will add to the size of the current frag
319 4) which index into the table to try if we can't fit into this one. */
320 {1, 1, 0, 0},
321 {1, 1, 0, 0},
322 {1, 1, 0, 0},
323 {1, 1, 0, 0},
325 {127 + 1, -128 + 1, 0, ENCODE_RELAX_STATE (COND_JUMP, BIG)},
326 {127 + 1, -128 + 1, 0, ENCODE_RELAX_STATE (COND_JUMP, BIG16)},
327 /* dword conditionals adds 4 bytes to frag:
328 1 extra opcode byte, 3 extra displacement bytes. */
329 {0, 0, 4, 0},
330 /* word conditionals add 2 bytes to frag:
331 1 extra opcode byte, 1 extra displacement byte. */
332 {0, 0, 2, 0},
334 {127 + 1, -128 + 1, 0, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG)},
335 {127 + 1, -128 + 1, 0, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16)},
336 /* dword jmp adds 3 bytes to frag:
337 0 extra opcode bytes, 3 extra displacement bytes. */
338 {0, 0, 3, 0},
339 /* word jmp adds 1 byte to frag:
340 0 extra opcode bytes, 1 extra displacement byte. */
341 {0, 0, 1, 0}
345 static const arch_entry cpu_arch[] = {
346 {"i8086", Cpu086 },
347 {"i186", Cpu086|Cpu186 },
348 {"i286", Cpu086|Cpu186|Cpu286 },
349 {"i386", Cpu086|Cpu186|Cpu286|Cpu386 },
350 {"i486", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486 },
351 {"i586", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
352 {"i686", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
353 {"pentium", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
354 {"pentiumpro",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
355 {"k6", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX|Cpu3dnow },
356 {"athlon", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuMMX|Cpu3dnow },
357 {"sledgehammer",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuSledgehammer|CpuMMX|Cpu3dnow|CpuSSE },
358 {NULL, 0 }
361 void
362 i386_align_code (fragP, count)
363 fragS *fragP;
364 int count;
366 /* Various efficient no-op patterns for aligning code labels.
367 Note: Don't try to assemble the instructions in the comments.
368 0L and 0w are not legal. */
369 static const char f32_1[] =
370 {0x90}; /* nop */
371 static const char f32_2[] =
372 {0x89,0xf6}; /* movl %esi,%esi */
373 static const char f32_3[] =
374 {0x8d,0x76,0x00}; /* leal 0(%esi),%esi */
375 static const char f32_4[] =
376 {0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
377 static const char f32_5[] =
378 {0x90, /* nop */
379 0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
380 static const char f32_6[] =
381 {0x8d,0xb6,0x00,0x00,0x00,0x00}; /* leal 0L(%esi),%esi */
382 static const char f32_7[] =
383 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
384 static const char f32_8[] =
385 {0x90, /* nop */
386 0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
387 static const char f32_9[] =
388 {0x89,0xf6, /* movl %esi,%esi */
389 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
390 static const char f32_10[] =
391 {0x8d,0x76,0x00, /* leal 0(%esi),%esi */
392 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
393 static const char f32_11[] =
394 {0x8d,0x74,0x26,0x00, /* leal 0(%esi,1),%esi */
395 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
396 static const char f32_12[] =
397 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
398 0x8d,0xbf,0x00,0x00,0x00,0x00}; /* leal 0L(%edi),%edi */
399 static const char f32_13[] =
400 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
401 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
402 static const char f32_14[] =
403 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00, /* leal 0L(%esi,1),%esi */
404 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
405 static const char f32_15[] =
406 {0xeb,0x0d,0x90,0x90,0x90,0x90,0x90, /* jmp .+15; lotsa nops */
407 0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90};
408 static const char f16_3[] =
409 {0x8d,0x74,0x00}; /* lea 0(%esi),%esi */
410 static const char f16_4[] =
411 {0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
412 static const char f16_5[] =
413 {0x90, /* nop */
414 0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
415 static const char f16_6[] =
416 {0x89,0xf6, /* mov %si,%si */
417 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
418 static const char f16_7[] =
419 {0x8d,0x74,0x00, /* lea 0(%si),%si */
420 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
421 static const char f16_8[] =
422 {0x8d,0xb4,0x00,0x00, /* lea 0w(%si),%si */
423 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
424 static const char *const f32_patt[] = {
425 f32_1, f32_2, f32_3, f32_4, f32_5, f32_6, f32_7, f32_8,
426 f32_9, f32_10, f32_11, f32_12, f32_13, f32_14, f32_15
428 static const char *const f16_patt[] = {
429 f32_1, f32_2, f16_3, f16_4, f16_5, f16_6, f16_7, f16_8,
430 f32_15, f32_15, f32_15, f32_15, f32_15, f32_15, f32_15
433 /* ??? We can't use these fillers for x86_64, since they often kills the
434 upper halves. Solve later. */
435 if (flag_code == CODE_64BIT)
436 count = 1;
438 if (count > 0 && count <= 15)
440 if (flag_code == CODE_16BIT)
442 memcpy (fragP->fr_literal + fragP->fr_fix,
443 f16_patt[count - 1], count);
444 if (count > 8)
445 /* Adjust jump offset. */
446 fragP->fr_literal[fragP->fr_fix + 1] = count - 2;
448 else
449 memcpy (fragP->fr_literal + fragP->fr_fix,
450 f32_patt[count - 1], count);
451 fragP->fr_var = count;
455 static char *output_invalid PARAMS ((int c));
456 static int i386_operand PARAMS ((char *operand_string));
457 static int i386_intel_operand PARAMS ((char *operand_string, int got_a_float));
458 static const reg_entry *parse_register PARAMS ((char *reg_string,
459 char **end_op));
461 #ifndef I386COFF
462 static void s_bss PARAMS ((int));
463 #endif
465 symbolS *GOT_symbol; /* Pre-defined "_GLOBAL_OFFSET_TABLE_". */
467 static INLINE unsigned int
468 mode_from_disp_size (t)
469 unsigned int t;
471 return (t & Disp8) ? 1 : (t & (Disp16 | Disp32 | Disp32S)) ? 2 : 0;
474 static INLINE int
475 fits_in_signed_byte (num)
476 offsetT num;
478 return (num >= -128) && (num <= 127);
481 static INLINE int
482 fits_in_unsigned_byte (num)
483 offsetT num;
485 return (num & 0xff) == num;
488 static INLINE int
489 fits_in_unsigned_word (num)
490 offsetT num;
492 return (num & 0xffff) == num;
495 static INLINE int
496 fits_in_signed_word (num)
497 offsetT num;
499 return (-32768 <= num) && (num <= 32767);
501 static INLINE int
502 fits_in_signed_long (num)
503 offsetT num ATTRIBUTE_UNUSED;
505 #ifndef BFD64
506 return 1;
507 #else
508 return (!(((offsetT) -1 << 31) & num)
509 || (((offsetT) -1 << 31) & num) == ((offsetT) -1 << 31));
510 #endif
511 } /* fits_in_signed_long() */
512 static INLINE int
513 fits_in_unsigned_long (num)
514 offsetT num ATTRIBUTE_UNUSED;
516 #ifndef BFD64
517 return 1;
518 #else
519 return (num & (((offsetT) 2 << 31) - 1)) == num;
520 #endif
521 } /* fits_in_unsigned_long() */
523 static int
524 smallest_imm_type (num)
525 offsetT num;
527 if (cpu_arch_flags != (Cpu086 | Cpu186 | Cpu286 | Cpu386 | Cpu486 | CpuNo64)
528 && !(cpu_arch_flags & (CpuUnknown)))
530 /* This code is disabled on the 486 because all the Imm1 forms
531 in the opcode table are slower on the i486. They're the
532 versions with the implicitly specified single-position
533 displacement, which has another syntax if you really want to
534 use that form. */
535 if (num == 1)
536 return Imm1 | Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64;
538 return (fits_in_signed_byte (num)
539 ? (Imm8S | Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
540 : fits_in_unsigned_byte (num)
541 ? (Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
542 : (fits_in_signed_word (num) || fits_in_unsigned_word (num))
543 ? (Imm16 | Imm32 | Imm32S | Imm64)
544 : fits_in_signed_long (num)
545 ? (Imm32 | Imm32S | Imm64)
546 : fits_in_unsigned_long (num)
547 ? (Imm32 | Imm64)
548 : Imm64);
551 static offsetT
552 offset_in_range (val, size)
553 offsetT val;
554 int size;
556 addressT mask;
558 switch (size)
560 case 1: mask = ((addressT) 1 << 8) - 1; break;
561 case 2: mask = ((addressT) 1 << 16) - 1; break;
562 case 4: mask = ((addressT) 2 << 31) - 1; break;
563 #ifdef BFD64
564 case 8: mask = ((addressT) 2 << 63) - 1; break;
565 #endif
566 default: abort ();
569 /* If BFD64, sign extend val. */
570 if (!use_rela_relocations)
571 if ((val & ~(((addressT) 2 << 31) - 1)) == 0)
572 val = (val ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
574 if ((val & ~mask) != 0 && (val & ~mask) != ~mask)
576 char buf1[40], buf2[40];
578 sprint_value (buf1, val);
579 sprint_value (buf2, val & mask);
580 as_warn (_("%s shortened to %s"), buf1, buf2);
582 return val & mask;
585 /* Returns 0 if attempting to add a prefix where one from the same
586 class already exists, 1 if non rep/repne added, 2 if rep/repne
587 added. */
588 static int
589 add_prefix (prefix)
590 unsigned int prefix;
592 int ret = 1;
593 int q;
595 if (prefix >= 0x40 && prefix < 0x50 && flag_code == CODE_64BIT)
596 q = REX_PREFIX;
597 else
598 switch (prefix)
600 default:
601 abort ();
603 case CS_PREFIX_OPCODE:
604 case DS_PREFIX_OPCODE:
605 case ES_PREFIX_OPCODE:
606 case FS_PREFIX_OPCODE:
607 case GS_PREFIX_OPCODE:
608 case SS_PREFIX_OPCODE:
609 q = SEG_PREFIX;
610 break;
612 case REPNE_PREFIX_OPCODE:
613 case REPE_PREFIX_OPCODE:
614 ret = 2;
615 /* fall thru */
616 case LOCK_PREFIX_OPCODE:
617 q = LOCKREP_PREFIX;
618 break;
620 case FWAIT_OPCODE:
621 q = WAIT_PREFIX;
622 break;
624 case ADDR_PREFIX_OPCODE:
625 q = ADDR_PREFIX;
626 break;
628 case DATA_PREFIX_OPCODE:
629 q = DATA_PREFIX;
630 break;
633 if (i.prefix[q])
635 as_bad (_("same type of prefix used twice"));
636 return 0;
639 i.prefixes += 1;
640 i.prefix[q] = prefix;
641 return ret;
644 static void
645 set_code_flag (value)
646 int value;
648 flag_code = value;
649 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
650 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
651 if (value == CODE_64BIT && !(cpu_arch_flags & CpuSledgehammer))
653 as_bad (_("64bit mode not supported on this CPU."));
655 if (value == CODE_32BIT && !(cpu_arch_flags & Cpu386))
657 as_bad (_("32bit mode not supported on this CPU."));
659 stackop_size = '\0';
662 static void
663 set_16bit_gcc_code_flag (new_code_flag)
664 int new_code_flag;
666 flag_code = new_code_flag;
667 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
668 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
669 stackop_size = 'l';
672 static void
673 set_intel_syntax (syntax_flag)
674 int syntax_flag;
676 /* Find out if register prefixing is specified. */
677 int ask_naked_reg = 0;
679 SKIP_WHITESPACE ();
680 if (! is_end_of_line[(unsigned char) *input_line_pointer])
682 char *string = input_line_pointer;
683 int e = get_symbol_end ();
685 if (strcmp (string, "prefix") == 0)
686 ask_naked_reg = 1;
687 else if (strcmp (string, "noprefix") == 0)
688 ask_naked_reg = -1;
689 else
690 as_bad (_("bad argument to syntax directive."));
691 *input_line_pointer = e;
693 demand_empty_rest_of_line ();
695 intel_syntax = syntax_flag;
697 if (ask_naked_reg == 0)
699 #ifdef BFD_ASSEMBLER
700 allow_naked_reg = (intel_syntax
701 && (bfd_get_symbol_leading_char (stdoutput) != '\0'));
702 #else
703 /* Conservative default. */
704 allow_naked_reg = 0;
705 #endif
707 else
708 allow_naked_reg = (ask_naked_reg < 0);
711 static void
712 set_cpu_arch (dummy)
713 int dummy ATTRIBUTE_UNUSED;
715 SKIP_WHITESPACE ();
717 if (! is_end_of_line[(unsigned char) *input_line_pointer])
719 char *string = input_line_pointer;
720 int e = get_symbol_end ();
721 int i;
723 for (i = 0; cpu_arch[i].name; i++)
725 if (strcmp (string, cpu_arch[i].name) == 0)
727 cpu_arch_name = cpu_arch[i].name;
728 cpu_arch_flags = cpu_arch[i].flags | (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
729 break;
732 if (!cpu_arch[i].name)
733 as_bad (_("no such architecture: `%s'"), string);
735 *input_line_pointer = e;
737 else
738 as_bad (_("missing cpu architecture"));
740 demand_empty_rest_of_line ();
743 const pseudo_typeS md_pseudo_table[] =
745 #if !defined(OBJ_AOUT) && !defined(USE_ALIGN_PTWO)
746 {"align", s_align_bytes, 0},
747 #else
748 {"align", s_align_ptwo, 0},
749 #endif
750 {"arch", set_cpu_arch, 0},
751 #ifndef I386COFF
752 {"bss", s_bss, 0},
753 #endif
754 {"ffloat", float_cons, 'f'},
755 {"dfloat", float_cons, 'd'},
756 {"tfloat", float_cons, 'x'},
757 {"value", cons, 2},
758 {"noopt", s_ignore, 0},
759 {"optim", s_ignore, 0},
760 {"code16gcc", set_16bit_gcc_code_flag, CODE_16BIT},
761 {"code16", set_code_flag, CODE_16BIT},
762 {"code32", set_code_flag, CODE_32BIT},
763 {"code64", set_code_flag, CODE_64BIT},
764 {"intel_syntax", set_intel_syntax, 1},
765 {"att_syntax", set_intel_syntax, 0},
766 {"file", dwarf2_directive_file, 0},
767 {"loc", dwarf2_directive_loc, 0},
768 {0, 0, 0}
771 /* For interface with expression (). */
772 extern char *input_line_pointer;
774 /* Hash table for instruction mnemonic lookup. */
775 static struct hash_control *op_hash;
777 /* Hash table for register lookup. */
778 static struct hash_control *reg_hash;
780 void
781 md_begin ()
783 const char *hash_err;
785 /* Initialize op_hash hash table. */
786 op_hash = hash_new ();
789 register const template *optab;
790 register templates *core_optab;
792 /* Setup for loop. */
793 optab = i386_optab;
794 core_optab = (templates *) xmalloc (sizeof (templates));
795 core_optab->start = optab;
797 while (1)
799 ++optab;
800 if (optab->name == NULL
801 || strcmp (optab->name, (optab - 1)->name) != 0)
803 /* different name --> ship out current template list;
804 add to hash table; & begin anew. */
805 core_optab->end = optab;
806 hash_err = hash_insert (op_hash,
807 (optab - 1)->name,
808 (PTR) core_optab);
809 if (hash_err)
811 as_fatal (_("Internal Error: Can't hash %s: %s"),
812 (optab - 1)->name,
813 hash_err);
815 if (optab->name == NULL)
816 break;
817 core_optab = (templates *) xmalloc (sizeof (templates));
818 core_optab->start = optab;
823 /* Initialize reg_hash hash table. */
824 reg_hash = hash_new ();
826 register const reg_entry *regtab;
828 for (regtab = i386_regtab;
829 regtab < i386_regtab + sizeof (i386_regtab) / sizeof (i386_regtab[0]);
830 regtab++)
832 hash_err = hash_insert (reg_hash, regtab->reg_name, (PTR) regtab);
833 if (hash_err)
834 as_fatal (_("Internal Error: Can't hash %s: %s"),
835 regtab->reg_name,
836 hash_err);
840 /* Fill in lexical tables: mnemonic_chars, operand_chars. */
842 register int c;
843 register char *p;
845 for (c = 0; c < 256; c++)
847 if (isdigit (c))
849 digit_chars[c] = c;
850 mnemonic_chars[c] = c;
851 register_chars[c] = c;
852 operand_chars[c] = c;
854 else if (islower (c))
856 mnemonic_chars[c] = c;
857 register_chars[c] = c;
858 operand_chars[c] = c;
860 else if (isupper (c))
862 mnemonic_chars[c] = tolower (c);
863 register_chars[c] = mnemonic_chars[c];
864 operand_chars[c] = c;
867 if (isalpha (c) || isdigit (c))
868 identifier_chars[c] = c;
869 else if (c >= 128)
871 identifier_chars[c] = c;
872 operand_chars[c] = c;
876 #ifdef LEX_AT
877 identifier_chars['@'] = '@';
878 #endif
879 digit_chars['-'] = '-';
880 identifier_chars['_'] = '_';
881 identifier_chars['.'] = '.';
883 for (p = operand_special_chars; *p != '\0'; p++)
884 operand_chars[(unsigned char) *p] = *p;
887 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
888 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
890 record_alignment (text_section, 2);
891 record_alignment (data_section, 2);
892 record_alignment (bss_section, 2);
894 #endif
897 void
898 i386_print_statistics (file)
899 FILE *file;
901 hash_print_statistics (file, "i386 opcode", op_hash);
902 hash_print_statistics (file, "i386 register", reg_hash);
905 #ifdef DEBUG386
907 /* Debugging routines for md_assemble. */
908 static void pi PARAMS ((char *, i386_insn *));
909 static void pte PARAMS ((template *));
910 static void pt PARAMS ((unsigned int));
911 static void pe PARAMS ((expressionS *));
912 static void ps PARAMS ((symbolS *));
914 static void
915 pi (line, x)
916 char *line;
917 i386_insn *x;
919 register template *p;
920 int i;
922 fprintf (stdout, "%s: template ", line);
923 pte (&x->tm);
924 fprintf (stdout, " modrm: mode %x reg %x reg/mem %x",
925 x->rm.mode, x->rm.reg, x->rm.regmem);
926 fprintf (stdout, " base %x index %x scale %x\n",
927 x->bi.base, x->bi.index, x->bi.scale);
928 for (i = 0; i < x->operands; i++)
930 fprintf (stdout, " #%d: ", i + 1);
931 pt (x->types[i]);
932 fprintf (stdout, "\n");
933 if (x->types[i]
934 & (Reg | SReg2 | SReg3 | Control | Debug | Test | RegMMX | RegXMM))
935 fprintf (stdout, "%s\n", x->op[i].regs->reg_name);
936 if (x->types[i] & Imm)
937 pe (x->op[i].imms);
938 if (x->types[i] & Disp)
939 pe (x->op[i].disps);
943 static void
944 pte (t)
945 template *t;
947 int i;
948 fprintf (stdout, " %d operands ", t->operands);
949 fprintf (stdout, "opcode %x ", t->base_opcode);
950 if (t->extension_opcode != None)
951 fprintf (stdout, "ext %x ", t->extension_opcode);
952 if (t->opcode_modifier & D)
953 fprintf (stdout, "D");
954 if (t->opcode_modifier & W)
955 fprintf (stdout, "W");
956 fprintf (stdout, "\n");
957 for (i = 0; i < t->operands; i++)
959 fprintf (stdout, " #%d type ", i + 1);
960 pt (t->operand_types[i]);
961 fprintf (stdout, "\n");
965 static void
966 pe (e)
967 expressionS *e;
969 fprintf (stdout, " operation %d\n", e->X_op);
970 fprintf (stdout, " add_number %ld (%lx)\n",
971 (long) e->X_add_number, (long) e->X_add_number);
972 if (e->X_add_symbol)
974 fprintf (stdout, " add_symbol ");
975 ps (e->X_add_symbol);
976 fprintf (stdout, "\n");
978 if (e->X_op_symbol)
980 fprintf (stdout, " op_symbol ");
981 ps (e->X_op_symbol);
982 fprintf (stdout, "\n");
986 static void
987 ps (s)
988 symbolS *s;
990 fprintf (stdout, "%s type %s%s",
991 S_GET_NAME (s),
992 S_IS_EXTERNAL (s) ? "EXTERNAL " : "",
993 segment_name (S_GET_SEGMENT (s)));
996 struct type_name
998 unsigned int mask;
999 char *tname;
1002 type_names[] =
1004 { Reg8, "r8" },
1005 { Reg16, "r16" },
1006 { Reg32, "r32" },
1007 { Imm8, "i8" },
1008 { Imm8S, "i8s" },
1009 { Imm16, "i16" },
1010 { Imm32, "i32" },
1011 { Imm1, "i1" },
1012 { BaseIndex, "BaseIndex" },
1013 { Disp8, "d8" },
1014 { Disp16, "d16" },
1015 { Disp32, "d32" },
1016 { InOutPortReg, "InOutPortReg" },
1017 { ShiftCount, "ShiftCount" },
1018 { Control, "control reg" },
1019 { Test, "test reg" },
1020 { Debug, "debug reg" },
1021 { FloatReg, "FReg" },
1022 { FloatAcc, "FAcc" },
1023 { SReg2, "SReg2" },
1024 { SReg3, "SReg3" },
1025 { Acc, "Acc" },
1026 { JumpAbsolute, "Jump Absolute" },
1027 { RegMMX, "rMMX" },
1028 { RegXMM, "rXMM" },
1029 { EsSeg, "es" },
1030 { 0, "" }
1033 static void
1034 pt (t)
1035 unsigned int t;
1037 register struct type_name *ty;
1039 if (t == Unknown)
1041 fprintf (stdout, _("Unknown"));
1043 else
1045 for (ty = type_names; ty->mask; ty++)
1046 if (t & ty->mask)
1047 fprintf (stdout, "%s, ", ty->tname);
1049 fflush (stdout);
1052 #endif /* DEBUG386 */
1055 tc_i386_force_relocation (fixp)
1056 struct fix *fixp;
1058 #ifdef BFD_ASSEMBLER
1059 if (fixp->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1060 || fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1061 return 1;
1062 return 0;
1063 #else
1064 /* For COFF. */
1065 return fixp->fx_r_type == 7;
1066 #endif
1069 #ifdef BFD_ASSEMBLER
1071 static bfd_reloc_code_real_type
1072 reloc (size, pcrel, sign, other)
1073 int size;
1074 int pcrel;
1075 int sign;
1076 bfd_reloc_code_real_type other;
1078 if (other != NO_RELOC)
1079 return other;
1081 if (pcrel)
1083 if (!sign)
1084 as_bad(_("There are no unsigned pc-relative relocations"));
1085 switch (size)
1087 case 1: return BFD_RELOC_8_PCREL;
1088 case 2: return BFD_RELOC_16_PCREL;
1089 case 4: return BFD_RELOC_32_PCREL;
1091 as_bad (_("can not do %d byte pc-relative relocation"), size);
1093 else
1095 if (sign)
1096 switch (size)
1098 case 4: return BFD_RELOC_X86_64_32S;
1100 else
1101 switch (size)
1103 case 1: return BFD_RELOC_8;
1104 case 2: return BFD_RELOC_16;
1105 case 4: return BFD_RELOC_32;
1106 case 8: return BFD_RELOC_64;
1108 as_bad (_("can not do %s %d byte relocation"),
1109 sign ? "signed" : "unsigned", size);
1112 abort();
1113 return BFD_RELOC_NONE;
1116 /* Here we decide which fixups can be adjusted to make them relative to
1117 the beginning of the section instead of the symbol. Basically we need
1118 to make sure that the dynamic relocations are done correctly, so in
1119 some cases we force the original symbol to be used. */
1122 tc_i386_fix_adjustable (fixP)
1123 fixS *fixP;
1125 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1126 /* Prevent all adjustments to global symbols, or else dynamic
1127 linking will not work correctly. */
1128 if (S_IS_EXTERNAL (fixP->fx_addsy)
1129 || S_IS_WEAK (fixP->fx_addsy))
1130 return 0;
1131 #endif
1132 /* adjust_reloc_syms doesn't know about the GOT. */
1133 if (fixP->fx_r_type == BFD_RELOC_386_GOTOFF
1134 || fixP->fx_r_type == BFD_RELOC_386_PLT32
1135 || fixP->fx_r_type == BFD_RELOC_386_GOT32
1136 || fixP->fx_r_type == BFD_RELOC_X86_64_PLT32
1137 || fixP->fx_r_type == BFD_RELOC_X86_64_GOT32
1138 || fixP->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1139 || fixP->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1140 return 0;
1141 return 1;
1143 #else
1144 #define reloc(SIZE,PCREL,OTHER) 0
1145 #define BFD_RELOC_16 0
1146 #define BFD_RELOC_32 0
1147 #define BFD_RELOC_16_PCREL 0
1148 #define BFD_RELOC_32_PCREL 0
1149 #define BFD_RELOC_386_PLT32 0
1150 #define BFD_RELOC_386_GOT32 0
1151 #define BFD_RELOC_386_GOTOFF 0
1152 #define BFD_RELOC_X86_64_PLT32 0
1153 #define BFD_RELOC_X86_64_GOT32 0
1154 #endif
1156 static int intel_float_operand PARAMS ((char *mnemonic));
1158 static int
1159 intel_float_operand (mnemonic)
1160 char *mnemonic;
1162 if (mnemonic[0] == 'f' && mnemonic[1] == 'i')
1163 return 2;
1165 if (mnemonic[0] == 'f')
1166 return 1;
1168 return 0;
1171 /* This is the guts of the machine-dependent assembler. LINE points to a
1172 machine dependent instruction. This function is supposed to emit
1173 the frags/bytes it assembles to. */
1175 void
1176 md_assemble (line)
1177 char *line;
1179 /* Points to template once we've found it. */
1180 const template *t;
1182 /* Count the size of the instruction generated. */
1183 int insn_size = 0;
1185 int j;
1187 char mnemonic[MAX_MNEM_SIZE];
1189 /* Initialize globals. */
1190 memset (&i, '\0', sizeof (i));
1191 for (j = 0; j < MAX_OPERANDS; j++)
1192 i.disp_reloc[j] = NO_RELOC;
1193 memset (disp_expressions, '\0', sizeof (disp_expressions));
1194 memset (im_expressions, '\0', sizeof (im_expressions));
1195 save_stack_p = save_stack;
1197 /* First parse an instruction mnemonic & call i386_operand for the operands.
1198 We assume that the scrubber has arranged it so that line[0] is the valid
1199 start of a (possibly prefixed) mnemonic. */
1201 char *l = line;
1202 char *token_start = l;
1203 char *mnem_p;
1205 /* Non-zero if we found a prefix only acceptable with string insns. */
1206 const char *expecting_string_instruction = NULL;
1208 while (1)
1210 mnem_p = mnemonic;
1211 while ((*mnem_p = mnemonic_chars[(unsigned char) *l]) != 0)
1213 mnem_p++;
1214 if (mnem_p >= mnemonic + sizeof (mnemonic))
1216 as_bad (_("no such instruction: `%s'"), token_start);
1217 return;
1219 l++;
1221 if (!is_space_char (*l)
1222 && *l != END_OF_INSN
1223 && *l != PREFIX_SEPARATOR)
1225 as_bad (_("invalid character %s in mnemonic"),
1226 output_invalid (*l));
1227 return;
1229 if (token_start == l)
1231 if (*l == PREFIX_SEPARATOR)
1232 as_bad (_("expecting prefix; got nothing"));
1233 else
1234 as_bad (_("expecting mnemonic; got nothing"));
1235 return;
1238 /* Look up instruction (or prefix) via hash table. */
1239 current_templates = hash_find (op_hash, mnemonic);
1241 if (*l != END_OF_INSN
1242 && (! is_space_char (*l) || l[1] != END_OF_INSN)
1243 && current_templates
1244 && (current_templates->start->opcode_modifier & IsPrefix))
1246 /* If we are in 16-bit mode, do not allow addr16 or data16.
1247 Similarly, in 32-bit mode, do not allow addr32 or data32. */
1248 if ((current_templates->start->opcode_modifier & (Size16 | Size32))
1249 && (((current_templates->start->opcode_modifier & Size32) != 0)
1250 ^ (flag_code == CODE_16BIT)))
1252 as_bad (_("redundant %s prefix"),
1253 current_templates->start->name);
1254 return;
1256 /* Add prefix, checking for repeated prefixes. */
1257 switch (add_prefix (current_templates->start->base_opcode))
1259 case 0:
1260 return;
1261 case 2:
1262 expecting_string_instruction = current_templates->start->name;
1263 break;
1265 /* Skip past PREFIX_SEPARATOR and reset token_start. */
1266 token_start = ++l;
1268 else
1269 break;
1272 if (!current_templates)
1274 /* See if we can get a match by trimming off a suffix. */
1275 switch (mnem_p[-1])
1277 case WORD_MNEM_SUFFIX:
1278 case BYTE_MNEM_SUFFIX:
1279 case QWORD_MNEM_SUFFIX:
1280 i.suffix = mnem_p[-1];
1281 mnem_p[-1] = '\0';
1282 current_templates = hash_find (op_hash, mnemonic);
1283 break;
1284 case SHORT_MNEM_SUFFIX:
1285 case LONG_MNEM_SUFFIX:
1286 if (!intel_syntax)
1288 i.suffix = mnem_p[-1];
1289 mnem_p[-1] = '\0';
1290 current_templates = hash_find (op_hash, mnemonic);
1292 break;
1294 /* Intel Syntax. */
1295 case 'd':
1296 if (intel_syntax)
1298 if (intel_float_operand (mnemonic))
1299 i.suffix = SHORT_MNEM_SUFFIX;
1300 else
1301 i.suffix = LONG_MNEM_SUFFIX;
1302 mnem_p[-1] = '\0';
1303 current_templates = hash_find (op_hash, mnemonic);
1305 break;
1307 if (!current_templates)
1309 as_bad (_("no such instruction: `%s'"), token_start);
1310 return;
1314 /* Check if instruction is supported on specified architecture. */
1315 if (cpu_arch_flags != 0)
1317 if ((current_templates->start->cpu_flags & ~(Cpu64 | CpuNo64))
1318 & ~(cpu_arch_flags & ~(Cpu64 | CpuNo64)))
1320 as_warn (_("`%s' is not supported on `%s'"),
1321 current_templates->start->name, cpu_arch_name);
1323 else if ((Cpu386 & ~cpu_arch_flags) && (flag_code != CODE_16BIT))
1325 as_warn (_("use .code16 to ensure correct addressing mode"));
1329 /* Check for rep/repne without a string instruction. */
1330 if (expecting_string_instruction
1331 && !(current_templates->start->opcode_modifier & IsString))
1333 as_bad (_("expecting string instruction after `%s'"),
1334 expecting_string_instruction);
1335 return;
1338 /* There may be operands to parse. */
1339 if (*l != END_OF_INSN)
1341 /* 1 if operand is pending after ','. */
1342 unsigned int expecting_operand = 0;
1344 /* Non-zero if operand parens not balanced. */
1345 unsigned int paren_not_balanced;
1349 /* Skip optional white space before operand. */
1350 if (is_space_char (*l))
1351 ++l;
1352 if (!is_operand_char (*l) && *l != END_OF_INSN)
1354 as_bad (_("invalid character %s before operand %d"),
1355 output_invalid (*l),
1356 i.operands + 1);
1357 return;
1359 token_start = l; /* after white space */
1360 paren_not_balanced = 0;
1361 while (paren_not_balanced || *l != ',')
1363 if (*l == END_OF_INSN)
1365 if (paren_not_balanced)
1367 if (!intel_syntax)
1368 as_bad (_("unbalanced parenthesis in operand %d."),
1369 i.operands + 1);
1370 else
1371 as_bad (_("unbalanced brackets in operand %d."),
1372 i.operands + 1);
1373 return;
1375 else
1376 break; /* we are done */
1378 else if (!is_operand_char (*l) && !is_space_char (*l))
1380 as_bad (_("invalid character %s in operand %d"),
1381 output_invalid (*l),
1382 i.operands + 1);
1383 return;
1385 if (!intel_syntax)
1387 if (*l == '(')
1388 ++paren_not_balanced;
1389 if (*l == ')')
1390 --paren_not_balanced;
1392 else
1394 if (*l == '[')
1395 ++paren_not_balanced;
1396 if (*l == ']')
1397 --paren_not_balanced;
1399 l++;
1401 if (l != token_start)
1402 { /* Yes, we've read in another operand. */
1403 unsigned int operand_ok;
1404 this_operand = i.operands++;
1405 if (i.operands > MAX_OPERANDS)
1407 as_bad (_("spurious operands; (%d operands/instruction max)"),
1408 MAX_OPERANDS);
1409 return;
1411 /* Now parse operand adding info to 'i' as we go along. */
1412 END_STRING_AND_SAVE (l);
1414 if (intel_syntax)
1415 operand_ok =
1416 i386_intel_operand (token_start,
1417 intel_float_operand (mnemonic));
1418 else
1419 operand_ok = i386_operand (token_start);
1421 RESTORE_END_STRING (l);
1422 if (!operand_ok)
1423 return;
1425 else
1427 if (expecting_operand)
1429 expecting_operand_after_comma:
1430 as_bad (_("expecting operand after ','; got nothing"));
1431 return;
1433 if (*l == ',')
1435 as_bad (_("expecting operand before ','; got nothing"));
1436 return;
1440 /* Now *l must be either ',' or END_OF_INSN. */
1441 if (*l == ',')
1443 if (*++l == END_OF_INSN)
1445 /* Just skip it, if it's \n complain. */
1446 goto expecting_operand_after_comma;
1448 expecting_operand = 1;
1451 while (*l != END_OF_INSN);
1455 /* Now we've parsed the mnemonic into a set of templates, and have the
1456 operands at hand.
1458 Next, we find a template that matches the given insn,
1459 making sure the overlap of the given operands types is consistent
1460 with the template operand types. */
1462 #define MATCH(overlap, given, template) \
1463 ((overlap & ~JumpAbsolute) \
1464 && ((given) & (BaseIndex|JumpAbsolute)) == ((overlap) & (BaseIndex|JumpAbsolute)))
1466 /* If given types r0 and r1 are registers they must be of the same type
1467 unless the expected operand type register overlap is null.
1468 Note that Acc in a template matches every size of reg. */
1469 #define CONSISTENT_REGISTER_MATCH(m0, g0, t0, m1, g1, t1) \
1470 ( ((g0) & Reg) == 0 || ((g1) & Reg) == 0 || \
1471 ((g0) & Reg) == ((g1) & Reg) || \
1472 ((((m0) & Acc) ? Reg : (t0)) & (((m1) & Acc) ? Reg : (t1)) & Reg) == 0 )
1475 register unsigned int overlap0, overlap1;
1476 unsigned int overlap2;
1477 unsigned int found_reverse_match;
1478 int suffix_check;
1480 /* All intel opcodes have reversed operands except for "bound" and
1481 "enter". We also don't reverse intersegment "jmp" and "call"
1482 instructions with 2 immediate operands so that the immediate segment
1483 precedes the offset, as it does when in AT&T mode. "enter" and the
1484 intersegment "jmp" and "call" instructions are the only ones that
1485 have two immediate operands. */
1486 if (intel_syntax && i.operands > 1
1487 && (strcmp (mnemonic, "bound") != 0)
1488 && !((i.types[0] & Imm) && (i.types[1] & Imm)))
1490 union i386_op temp_op;
1491 unsigned int temp_type;
1492 #ifdef BFD_ASSEMBLER
1493 enum bfd_reloc_code_real temp_reloc;
1494 #else
1495 int temp_reloc;
1496 #endif
1497 int xchg1 = 0;
1498 int xchg2 = 0;
1500 if (i.operands == 2)
1502 xchg1 = 0;
1503 xchg2 = 1;
1505 else if (i.operands == 3)
1507 xchg1 = 0;
1508 xchg2 = 2;
1510 temp_type = i.types[xchg2];
1511 i.types[xchg2] = i.types[xchg1];
1512 i.types[xchg1] = temp_type;
1513 temp_op = i.op[xchg2];
1514 i.op[xchg2] = i.op[xchg1];
1515 i.op[xchg1] = temp_op;
1516 temp_reloc = i.disp_reloc[xchg2];
1517 i.disp_reloc[xchg2] = i.disp_reloc[xchg1];
1518 i.disp_reloc[xchg1] = temp_reloc;
1520 if (i.mem_operands == 2)
1522 const seg_entry *temp_seg;
1523 temp_seg = i.seg[0];
1524 i.seg[0] = i.seg[1];
1525 i.seg[1] = temp_seg;
1529 if (i.imm_operands)
1531 /* Try to ensure constant immediates are represented in the smallest
1532 opcode possible. */
1533 char guess_suffix = 0;
1534 int op;
1536 if (i.suffix)
1537 guess_suffix = i.suffix;
1538 else if (i.reg_operands)
1540 /* Figure out a suffix from the last register operand specified.
1541 We can't do this properly yet, ie. excluding InOutPortReg,
1542 but the following works for instructions with immediates.
1543 In any case, we can't set i.suffix yet. */
1544 for (op = i.operands; --op >= 0;)
1545 if (i.types[op] & Reg)
1547 if (i.types[op] & Reg8)
1548 guess_suffix = BYTE_MNEM_SUFFIX;
1549 else if (i.types[op] & Reg16)
1550 guess_suffix = WORD_MNEM_SUFFIX;
1551 else if (i.types[op] & Reg32)
1552 guess_suffix = LONG_MNEM_SUFFIX;
1553 else if (i.types[op] & Reg64)
1554 guess_suffix = QWORD_MNEM_SUFFIX;
1555 break;
1558 else if ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0))
1559 guess_suffix = WORD_MNEM_SUFFIX;
1561 for (op = i.operands; --op >= 0;)
1562 if (i.types[op] & Imm)
1564 switch (i.op[op].imms->X_op)
1566 case O_constant:
1567 /* If a suffix is given, this operand may be shortened. */
1568 switch (guess_suffix)
1570 case LONG_MNEM_SUFFIX:
1571 i.types[op] |= Imm32 | Imm64;
1572 break;
1573 case WORD_MNEM_SUFFIX:
1574 i.types[op] |= Imm16 | Imm32S | Imm32 | Imm64;
1575 break;
1576 case BYTE_MNEM_SUFFIX:
1577 i.types[op] |= Imm16 | Imm8 | Imm8S | Imm32S | Imm32 | Imm64;
1578 break;
1581 /* If this operand is at most 16 bits, convert it to a
1582 signed 16 bit number before trying to see whether it will
1583 fit in an even smaller size. This allows a 16-bit operand
1584 such as $0xffe0 to be recognised as within Imm8S range. */
1585 if ((i.types[op] & Imm16)
1586 && (i.op[op].imms->X_add_number & ~(offsetT)0xffff) == 0)
1588 i.op[op].imms->X_add_number =
1589 (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000);
1591 if ((i.types[op] & Imm32)
1592 && (i.op[op].imms->X_add_number & ~(((offsetT) 2 << 31) - 1)) == 0)
1594 i.op[op].imms->X_add_number =
1595 (i.op[op].imms->X_add_number ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
1597 i.types[op] |= smallest_imm_type (i.op[op].imms->X_add_number);
1598 /* We must avoid matching of Imm32 templates when 64bit only immediate is available. */
1599 if (guess_suffix == QWORD_MNEM_SUFFIX)
1600 i.types[op] &= ~Imm32;
1601 break;
1602 case O_absent:
1603 case O_register:
1604 abort();
1605 /* Symbols and expressions. */
1606 default:
1607 /* Convert symbolic operand to proper sizes for matching. */
1608 switch (guess_suffix)
1610 case QWORD_MNEM_SUFFIX:
1611 i.types[op] = Imm64 | Imm32S;
1612 break;
1613 case LONG_MNEM_SUFFIX:
1614 i.types[op] = Imm32 | Imm64;
1615 break;
1616 case WORD_MNEM_SUFFIX:
1617 i.types[op] = Imm16 | Imm32 | Imm64;
1618 break;
1619 break;
1620 case BYTE_MNEM_SUFFIX:
1621 i.types[op] = Imm8 | Imm8S | Imm16 | Imm32S | Imm32;
1622 break;
1623 break;
1625 break;
1630 if (i.disp_operands)
1632 /* Try to use the smallest displacement type too. */
1633 int op;
1635 for (op = i.operands; --op >= 0;)
1636 if ((i.types[op] & Disp)
1637 && i.op[op].imms->X_op == O_constant)
1639 offsetT disp = i.op[op].disps->X_add_number;
1641 if (i.types[op] & Disp16)
1643 /* We know this operand is at most 16 bits, so
1644 convert to a signed 16 bit number before trying
1645 to see whether it will fit in an even smaller
1646 size. */
1648 disp = (((disp & 0xffff) ^ 0x8000) - 0x8000);
1650 else if (i.types[op] & Disp32)
1652 /* We know this operand is at most 32 bits, so convert to a
1653 signed 32 bit number before trying to see whether it will
1654 fit in an even smaller size. */
1655 disp &= (((offsetT) 2 << 31) - 1);
1656 disp = (disp ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
1658 if (flag_code == CODE_64BIT)
1660 if (fits_in_signed_long (disp))
1661 i.types[op] |= Disp32S;
1662 if (fits_in_unsigned_long (disp))
1663 i.types[op] |= Disp32;
1665 if ((i.types[op] & (Disp32 | Disp32S | Disp16))
1666 && fits_in_signed_byte (disp))
1667 i.types[op] |= Disp8;
1671 overlap0 = 0;
1672 overlap1 = 0;
1673 overlap2 = 0;
1674 found_reverse_match = 0;
1675 suffix_check = (i.suffix == BYTE_MNEM_SUFFIX
1676 ? No_bSuf
1677 : (i.suffix == WORD_MNEM_SUFFIX
1678 ? No_wSuf
1679 : (i.suffix == SHORT_MNEM_SUFFIX
1680 ? No_sSuf
1681 : (i.suffix == LONG_MNEM_SUFFIX
1682 ? No_lSuf
1683 : (i.suffix == QWORD_MNEM_SUFFIX
1684 ? No_qSuf
1685 : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX ? No_xSuf : 0))))));
1687 for (t = current_templates->start;
1688 t < current_templates->end;
1689 t++)
1691 /* Must have right number of operands. */
1692 if (i.operands != t->operands)
1693 continue;
1695 /* Check the suffix, except for some instructions in intel mode. */
1696 if ((t->opcode_modifier & suffix_check)
1697 && !(intel_syntax
1698 && (t->opcode_modifier & IgnoreSize))
1699 && !(intel_syntax
1700 && t->base_opcode == 0xd9
1701 && (t->extension_opcode == 5 /* 0xd9,5 "fldcw" */
1702 || t->extension_opcode == 7))) /* 0xd9,7 "f{n}stcw" */
1703 continue;
1705 else if (!t->operands)
1706 /* 0 operands always matches. */
1707 break;
1709 overlap0 = i.types[0] & t->operand_types[0];
1710 switch (t->operands)
1712 case 1:
1713 if (!MATCH (overlap0, i.types[0], t->operand_types[0]))
1714 continue;
1715 break;
1716 case 2:
1717 case 3:
1718 overlap1 = i.types[1] & t->operand_types[1];
1719 if (!MATCH (overlap0, i.types[0], t->operand_types[0])
1720 || !MATCH (overlap1, i.types[1], t->operand_types[1])
1721 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
1722 t->operand_types[0],
1723 overlap1, i.types[1],
1724 t->operand_types[1]))
1726 /* Check if other direction is valid ... */
1727 if ((t->opcode_modifier & (D|FloatD)) == 0)
1728 continue;
1730 /* Try reversing direction of operands. */
1731 overlap0 = i.types[0] & t->operand_types[1];
1732 overlap1 = i.types[1] & t->operand_types[0];
1733 if (!MATCH (overlap0, i.types[0], t->operand_types[1])
1734 || !MATCH (overlap1, i.types[1], t->operand_types[0])
1735 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
1736 t->operand_types[1],
1737 overlap1, i.types[1],
1738 t->operand_types[0]))
1740 /* Does not match either direction. */
1741 continue;
1743 /* found_reverse_match holds which of D or FloatDR
1744 we've found. */
1745 found_reverse_match = t->opcode_modifier & (D|FloatDR);
1747 /* Found a forward 2 operand match here. */
1748 else if (t->operands == 3)
1750 /* Here we make use of the fact that there are no
1751 reverse match 3 operand instructions, and all 3
1752 operand instructions only need to be checked for
1753 register consistency between operands 2 and 3. */
1754 overlap2 = i.types[2] & t->operand_types[2];
1755 if (!MATCH (overlap2, i.types[2], t->operand_types[2])
1756 || !CONSISTENT_REGISTER_MATCH (overlap1, i.types[1],
1757 t->operand_types[1],
1758 overlap2, i.types[2],
1759 t->operand_types[2]))
1761 continue;
1763 /* Found either forward/reverse 2 or 3 operand match here:
1764 slip through to break. */
1766 if (t->cpu_flags & ~cpu_arch_flags)
1768 found_reverse_match = 0;
1769 continue;
1771 /* We've found a match; break out of loop. */
1772 break;
1774 if (t == current_templates->end)
1776 /* We found no match. */
1777 as_bad (_("suffix or operands invalid for `%s'"),
1778 current_templates->start->name);
1779 return;
1782 if (!quiet_warnings)
1784 if (!intel_syntax
1785 && ((i.types[0] & JumpAbsolute)
1786 != (t->operand_types[0] & JumpAbsolute)))
1788 as_warn (_("indirect %s without `*'"), t->name);
1791 if ((t->opcode_modifier & (IsPrefix|IgnoreSize))
1792 == (IsPrefix|IgnoreSize))
1794 /* Warn them that a data or address size prefix doesn't
1795 affect assembly of the next line of code. */
1796 as_warn (_("stand-alone `%s' prefix"), t->name);
1800 /* Copy the template we found. */
1801 i.tm = *t;
1802 if (found_reverse_match)
1804 /* If we found a reverse match we must alter the opcode
1805 direction bit. found_reverse_match holds bits to change
1806 (different for int & float insns). */
1808 i.tm.base_opcode ^= found_reverse_match;
1810 i.tm.operand_types[0] = t->operand_types[1];
1811 i.tm.operand_types[1] = t->operand_types[0];
1814 /* Undo SYSV386_COMPAT brokenness when in Intel mode. See i386.h */
1815 if (SYSV386_COMPAT
1816 && intel_syntax
1817 && (i.tm.base_opcode & 0xfffffde0) == 0xdce0)
1818 i.tm.base_opcode ^= FloatR;
1820 if (i.tm.opcode_modifier & FWait)
1821 if (! add_prefix (FWAIT_OPCODE))
1822 return;
1824 /* Check string instruction segment overrides. */
1825 if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0)
1827 int mem_op = (i.types[0] & AnyMem) ? 0 : 1;
1828 if ((i.tm.operand_types[mem_op] & EsSeg) != 0)
1830 if (i.seg[0] != NULL && i.seg[0] != &es)
1832 as_bad (_("`%s' operand %d must use `%%es' segment"),
1833 i.tm.name,
1834 mem_op + 1);
1835 return;
1837 /* There's only ever one segment override allowed per instruction.
1838 This instruction possibly has a legal segment override on the
1839 second operand, so copy the segment to where non-string
1840 instructions store it, allowing common code. */
1841 i.seg[0] = i.seg[1];
1843 else if ((i.tm.operand_types[mem_op + 1] & EsSeg) != 0)
1845 if (i.seg[1] != NULL && i.seg[1] != &es)
1847 as_bad (_("`%s' operand %d must use `%%es' segment"),
1848 i.tm.name,
1849 mem_op + 2);
1850 return;
1855 if (i.reg_operands && flag_code < CODE_64BIT)
1857 int op;
1858 for (op = i.operands; --op >= 0; )
1859 if ((i.types[op] & Reg)
1860 && (i.op[op].regs->reg_flags & (RegRex64|RegRex)))
1861 as_bad (_("Extended register `%%%s' available only in 64bit mode."),
1862 i.op[op].regs->reg_name);
1865 /* If matched instruction specifies an explicit instruction mnemonic
1866 suffix, use it. */
1867 if (i.tm.opcode_modifier & (Size16 | Size32 | Size64))
1869 if (i.tm.opcode_modifier & Size16)
1870 i.suffix = WORD_MNEM_SUFFIX;
1871 else if (i.tm.opcode_modifier & Size64)
1872 i.suffix = QWORD_MNEM_SUFFIX;
1873 else
1874 i.suffix = LONG_MNEM_SUFFIX;
1876 else if (i.reg_operands)
1878 /* If there's no instruction mnemonic suffix we try to invent one
1879 based on register operands. */
1880 if (!i.suffix)
1882 /* We take i.suffix from the last register operand specified,
1883 Destination register type is more significant than source
1884 register type. */
1885 int op;
1886 for (op = i.operands; --op >= 0;)
1887 if ((i.types[op] & Reg)
1888 && !(i.tm.operand_types[op] & InOutPortReg))
1890 i.suffix = ((i.types[op] & Reg8) ? BYTE_MNEM_SUFFIX :
1891 (i.types[op] & Reg16) ? WORD_MNEM_SUFFIX :
1892 (i.types[op] & Reg64) ? QWORD_MNEM_SUFFIX :
1893 LONG_MNEM_SUFFIX);
1894 break;
1897 else if (i.suffix == BYTE_MNEM_SUFFIX)
1899 int op;
1900 for (op = i.operands; --op >= 0;)
1902 /* If this is an eight bit register, it's OK. If it's
1903 the 16 or 32 bit version of an eight bit register,
1904 we will just use the low portion, and that's OK too. */
1905 if (i.types[op] & Reg8)
1906 continue;
1908 /* movzx and movsx should not generate this warning. */
1909 if (intel_syntax
1910 && (i.tm.base_opcode == 0xfb7
1911 || i.tm.base_opcode == 0xfb6
1912 || i.tm.base_opcode == 0x63
1913 || i.tm.base_opcode == 0xfbe
1914 || i.tm.base_opcode == 0xfbf))
1915 continue;
1917 if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4
1918 #if 0
1919 /* Check that the template allows eight bit regs
1920 This kills insns such as `orb $1,%edx', which
1921 maybe should be allowed. */
1922 && (i.tm.operand_types[op] & (Reg8|InOutPortReg))
1923 #endif
1926 /* Prohibit these changes in the 64bit mode, since
1927 the lowering is more complicated. */
1928 if (flag_code == CODE_64BIT
1929 && (i.tm.operand_types[op] & InOutPortReg) == 0)
1930 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
1931 i.op[op].regs->reg_name,
1932 i.suffix);
1933 #if REGISTER_WARNINGS
1934 if (!quiet_warnings
1935 && (i.tm.operand_types[op] & InOutPortReg) == 0)
1936 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
1937 (i.op[op].regs - (i.types[op] & Reg16 ? 8 : 16))->reg_name,
1938 i.op[op].regs->reg_name,
1939 i.suffix);
1940 #endif
1941 continue;
1943 /* Any other register is bad. */
1944 if (i.types[op] & (Reg | RegMMX | RegXMM
1945 | SReg2 | SReg3
1946 | Control | Debug | Test
1947 | FloatReg | FloatAcc))
1949 as_bad (_("`%%%s' not allowed with `%s%c'"),
1950 i.op[op].regs->reg_name,
1951 i.tm.name,
1952 i.suffix);
1953 return;
1957 else if (i.suffix == LONG_MNEM_SUFFIX)
1959 int op;
1961 for (op = i.operands; --op >= 0;)
1962 /* Reject eight bit registers, except where the template
1963 requires them. (eg. movzb) */
1964 if ((i.types[op] & Reg8) != 0
1965 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
1967 as_bad (_("`%%%s' not allowed with `%s%c'"),
1968 i.op[op].regs->reg_name,
1969 i.tm.name,
1970 i.suffix);
1971 return;
1973 /* Warn if the e prefix on a general reg is missing. */
1974 else if ((!quiet_warnings || flag_code == CODE_64BIT)
1975 && (i.types[op] & Reg16) != 0
1976 && (i.tm.operand_types[op] & (Reg32|Acc)) != 0)
1978 /* Prohibit these changes in the 64bit mode, since
1979 the lowering is more complicated. */
1980 if (flag_code == CODE_64BIT)
1981 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
1982 i.op[op].regs->reg_name,
1983 i.suffix);
1984 #if REGISTER_WARNINGS
1985 else
1986 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
1987 (i.op[op].regs + 8)->reg_name,
1988 i.op[op].regs->reg_name,
1989 i.suffix);
1990 #endif
1992 /* Warn if the r prefix on a general reg is missing. */
1993 else if ((i.types[op] & Reg64) != 0
1994 && (i.tm.operand_types[op] & (Reg32|Acc)) != 0)
1996 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
1997 i.op[op].regs->reg_name,
1998 i.suffix);
2001 else if (i.suffix == QWORD_MNEM_SUFFIX)
2003 int op;
2004 if (flag_code < CODE_64BIT)
2005 as_bad (_("64bit operations available only in 64bit modes."));
2007 for (op = i.operands; --op >= 0; )
2008 /* Reject eight bit registers, except where the template
2009 requires them. (eg. movzb) */
2010 if ((i.types[op] & Reg8) != 0
2011 && (i.tm.operand_types[op] & (Reg16|Reg32|Acc)) != 0)
2013 as_bad (_("`%%%s' not allowed with `%s%c'"),
2014 i.op[op].regs->reg_name,
2015 i.tm.name,
2016 i.suffix);
2017 return;
2019 /* Warn if the e prefix on a general reg is missing. */
2020 else if (((i.types[op] & Reg16) != 0
2021 || (i.types[op] & Reg32) != 0)
2022 && (i.tm.operand_types[op] & (Reg32|Acc)) != 0)
2024 /* Prohibit these changes in the 64bit mode, since
2025 the lowering is more complicated. */
2026 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2027 i.op[op].regs->reg_name,
2028 i.suffix);
2031 else if (i.suffix == WORD_MNEM_SUFFIX)
2033 int op;
2034 for (op = i.operands; --op >= 0;)
2035 /* Reject eight bit registers, except where the template
2036 requires them. (eg. movzb) */
2037 if ((i.types[op] & Reg8) != 0
2038 && (i.tm.operand_types[op] & (Reg16|Reg32|Acc)) != 0)
2040 as_bad (_("`%%%s' not allowed with `%s%c'"),
2041 i.op[op].regs->reg_name,
2042 i.tm.name,
2043 i.suffix);
2044 return;
2046 /* Warn if the e prefix on a general reg is present. */
2047 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2048 && (i.types[op] & Reg32) != 0
2049 && (i.tm.operand_types[op] & (Reg16|Acc)) != 0)
2051 /* Prohibit these changes in the 64bit mode, since
2052 the lowering is more complicated. */
2053 if (flag_code == CODE_64BIT)
2054 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2055 i.op[op].regs->reg_name,
2056 i.suffix);
2057 else
2058 #if REGISTER_WARNINGS
2059 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2060 (i.op[op].regs - 8)->reg_name,
2061 i.op[op].regs->reg_name,
2062 i.suffix);
2063 #endif
2066 else if (intel_syntax && (i.tm.opcode_modifier & IgnoreSize))
2067 /* Do nothing if the instruction is going to ignore the prefix. */
2069 else
2070 abort ();
2072 else if ((i.tm.opcode_modifier & DefaultSize) && !i.suffix)
2074 i.suffix = stackop_size;
2076 /* Make still unresolved immediate matches conform to size of immediate
2077 given in i.suffix. Note: overlap2 cannot be an immediate! */
2078 if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S))
2079 && overlap0 != Imm8 && overlap0 != Imm8S
2080 && overlap0 != Imm16 && overlap0 != Imm32S
2081 && overlap0 != Imm32 && overlap0 != Imm64)
2083 if (i.suffix)
2085 overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX ? (Imm8 | Imm8S) :
2086 (i.suffix == WORD_MNEM_SUFFIX ? Imm16 :
2087 (i.suffix == QWORD_MNEM_SUFFIX ? Imm64 | Imm32S : Imm32)));
2089 else if (overlap0 == (Imm16 | Imm32S | Imm32)
2090 || overlap0 == (Imm16 | Imm32)
2091 || overlap0 == (Imm16 | Imm32S))
2093 overlap0 =
2094 ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)) ? Imm16 : Imm32S;
2096 if (overlap0 != Imm8 && overlap0 != Imm8S
2097 && overlap0 != Imm16 && overlap0 != Imm32S
2098 && overlap0 != Imm32 && overlap0 != Imm64)
2100 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size"));
2101 return;
2104 if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32))
2105 && overlap1 != Imm8 && overlap1 != Imm8S
2106 && overlap1 != Imm16 && overlap1 != Imm32S
2107 && overlap1 != Imm32 && overlap1 != Imm64)
2109 if (i.suffix)
2111 overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX ? (Imm8 | Imm8S) :
2112 (i.suffix == WORD_MNEM_SUFFIX ? Imm16 :
2113 (i.suffix == QWORD_MNEM_SUFFIX ? Imm64 | Imm32S : Imm32)));
2115 else if (overlap1 == (Imm16 | Imm32 | Imm32S)
2116 || overlap1 == (Imm16 | Imm32)
2117 || overlap1 == (Imm16 | Imm32S))
2119 overlap1 =
2120 ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)) ? Imm16 : Imm32S;
2122 if (overlap1 != Imm8 && overlap1 != Imm8S
2123 && overlap1 != Imm16 && overlap1 != Imm32S
2124 && overlap1 != Imm32 && overlap1 != Imm64)
2126 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size %x %c"),overlap1, i.suffix);
2127 return;
2130 assert ((overlap2 & Imm) == 0);
2132 i.types[0] = overlap0;
2133 if (overlap0 & ImplicitRegister)
2134 i.reg_operands--;
2135 if (overlap0 & Imm1)
2136 i.imm_operands = 0; /* kludge for shift insns. */
2138 i.types[1] = overlap1;
2139 if (overlap1 & ImplicitRegister)
2140 i.reg_operands--;
2142 i.types[2] = overlap2;
2143 if (overlap2 & ImplicitRegister)
2144 i.reg_operands--;
2146 /* Finalize opcode. First, we change the opcode based on the operand
2147 size given by i.suffix: We need not change things for byte insns. */
2149 if (!i.suffix && (i.tm.opcode_modifier & W))
2151 as_bad (_("no instruction mnemonic suffix given and no register operands; can't size instruction"));
2152 return;
2155 /* For movzx and movsx, need to check the register type. */
2156 if (intel_syntax
2157 && (i.tm.base_opcode == 0xfb6 || i.tm.base_opcode == 0xfbe))
2158 if (i.suffix && i.suffix == BYTE_MNEM_SUFFIX)
2160 unsigned int prefix = DATA_PREFIX_OPCODE;
2162 if ((i.op[1].regs->reg_type & Reg16) != 0)
2163 if (!add_prefix (prefix))
2164 return;
2167 if (i.suffix && i.suffix != BYTE_MNEM_SUFFIX)
2169 /* It's not a byte, select word/dword operation. */
2170 if (i.tm.opcode_modifier & W)
2172 if (i.tm.opcode_modifier & ShortForm)
2173 i.tm.base_opcode |= 8;
2174 else
2175 i.tm.base_opcode |= 1;
2177 /* Now select between word & dword operations via the operand
2178 size prefix, except for instructions that will ignore this
2179 prefix anyway. */
2180 if (i.suffix != QWORD_MNEM_SUFFIX
2181 && (i.suffix == LONG_MNEM_SUFFIX) == (flag_code == CODE_16BIT)
2182 && !(i.tm.opcode_modifier & IgnoreSize))
2184 unsigned int prefix = DATA_PREFIX_OPCODE;
2185 if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */
2186 prefix = ADDR_PREFIX_OPCODE;
2188 if (! add_prefix (prefix))
2189 return;
2192 /* Set mode64 for an operand. */
2193 if (i.suffix == QWORD_MNEM_SUFFIX
2194 && !(i.tm.opcode_modifier & NoRex64))
2195 i.rex.mode64 = 1;
2197 /* Size floating point instruction. */
2198 if (i.suffix == LONG_MNEM_SUFFIX)
2200 if (i.tm.opcode_modifier & FloatMF)
2201 i.tm.base_opcode ^= 4;
2205 if (i.tm.opcode_modifier & ImmExt)
2207 /* These AMD 3DNow! and Intel Katmai New Instructions have an
2208 opcode suffix which is coded in the same place as an 8-bit
2209 immediate field would be. Here we fake an 8-bit immediate
2210 operand from the opcode suffix stored in tm.extension_opcode. */
2212 expressionS *exp;
2214 assert (i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS);
2216 exp = &im_expressions[i.imm_operands++];
2217 i.op[i.operands].imms = exp;
2218 i.types[i.operands++] = Imm8;
2219 exp->X_op = O_constant;
2220 exp->X_add_number = i.tm.extension_opcode;
2221 i.tm.extension_opcode = None;
2224 /* For insns with operands there are more diddles to do to the opcode. */
2225 if (i.operands)
2227 /* Default segment register this instruction will use
2228 for memory accesses. 0 means unknown.
2229 This is only for optimizing out unnecessary segment overrides. */
2230 const seg_entry *default_seg = 0;
2232 /* The imul $imm, %reg instruction is converted into
2233 imul $imm, %reg, %reg, and the clr %reg instruction
2234 is converted into xor %reg, %reg. */
2235 if (i.tm.opcode_modifier & regKludge)
2237 unsigned int first_reg_op = (i.types[0] & Reg) ? 0 : 1;
2238 /* Pretend we saw the extra register operand. */
2239 assert (i.op[first_reg_op + 1].regs == 0);
2240 i.op[first_reg_op + 1].regs = i.op[first_reg_op].regs;
2241 i.types[first_reg_op + 1] = i.types[first_reg_op];
2242 i.reg_operands = 2;
2245 if (i.tm.opcode_modifier & ShortForm)
2247 /* The register or float register operand is in operand 0 or 1. */
2248 unsigned int op = (i.types[0] & (Reg | FloatReg)) ? 0 : 1;
2249 /* Register goes in low 3 bits of opcode. */
2250 i.tm.base_opcode |= i.op[op].regs->reg_num;
2251 if (i.op[op].regs->reg_flags & RegRex)
2252 i.rex.extZ=1;
2253 if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
2255 /* Warn about some common errors, but press on regardless.
2256 The first case can be generated by gcc (<= 2.8.1). */
2257 if (i.operands == 2)
2259 /* Reversed arguments on faddp, fsubp, etc. */
2260 as_warn (_("translating to `%s %%%s,%%%s'"), i.tm.name,
2261 i.op[1].regs->reg_name,
2262 i.op[0].regs->reg_name);
2264 else
2266 /* Extraneous `l' suffix on fp insn. */
2267 as_warn (_("translating to `%s %%%s'"), i.tm.name,
2268 i.op[0].regs->reg_name);
2272 else if (i.tm.opcode_modifier & Modrm)
2274 /* The opcode is completed (modulo i.tm.extension_opcode which
2275 must be put into the modrm byte).
2276 Now, we make the modrm & index base bytes based on all the
2277 info we've collected. */
2279 /* i.reg_operands MUST be the number of real register operands;
2280 implicit registers do not count. */
2281 if (i.reg_operands == 2)
2283 unsigned int source, dest;
2284 source = ((i.types[0]
2285 & (Reg | RegMMX | RegXMM
2286 | SReg2 | SReg3
2287 | Control | Debug | Test))
2288 ? 0 : 1);
2289 dest = source + 1;
2291 i.rm.mode = 3;
2292 /* One of the register operands will be encoded in the
2293 i.tm.reg field, the other in the combined i.tm.mode
2294 and i.tm.regmem fields. If no form of this
2295 instruction supports a memory destination operand,
2296 then we assume the source operand may sometimes be
2297 a memory operand and so we need to store the
2298 destination in the i.rm.reg field. */
2299 if ((i.tm.operand_types[dest] & AnyMem) == 0)
2301 i.rm.reg = i.op[dest].regs->reg_num;
2302 i.rm.regmem = i.op[source].regs->reg_num;
2303 if (i.op[dest].regs->reg_flags & RegRex)
2304 i.rex.extX=1;
2305 if (i.op[source].regs->reg_flags & RegRex)
2306 i.rex.extZ=1;
2308 else
2310 i.rm.reg = i.op[source].regs->reg_num;
2311 i.rm.regmem = i.op[dest].regs->reg_num;
2312 if (i.op[dest].regs->reg_flags & RegRex)
2313 i.rex.extZ=1;
2314 if (i.op[source].regs->reg_flags & RegRex)
2315 i.rex.extX=1;
2318 else
2319 { /* If it's not 2 reg operands... */
2320 if (i.mem_operands)
2322 unsigned int fake_zero_displacement = 0;
2323 unsigned int op = ((i.types[0] & AnyMem)
2325 : (i.types[1] & AnyMem) ? 1 : 2);
2327 default_seg = &ds;
2329 if (! i.base_reg)
2331 i.rm.mode = 0;
2332 if (! i.disp_operands)
2333 fake_zero_displacement = 1;
2334 if (! i.index_reg)
2336 /* Operand is just <disp> */
2337 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
2339 i.rm.regmem = NO_BASE_REGISTER_16;
2340 i.types[op] &= ~Disp;
2341 i.types[op] |= Disp16;
2343 else if (flag_code != CODE_64BIT)
2345 i.rm.regmem = NO_BASE_REGISTER;
2346 i.types[op] &= ~Disp;
2347 i.types[op] |= Disp32;
2349 else
2351 /* 64bit mode overwrites the 32bit absolute addressing
2352 by RIP relative addressing and absolute addressing
2353 is encoded by one of the redundant SIB forms. */
2355 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2356 i.sib.base = NO_BASE_REGISTER;
2357 i.sib.index = NO_INDEX_REGISTER;
2358 i.types[op] &= ~Disp;
2359 i.types[op] |= Disp32S;
2362 else /* ! i.base_reg && i.index_reg */
2364 i.sib.index = i.index_reg->reg_num;
2365 i.sib.base = NO_BASE_REGISTER;
2366 i.sib.scale = i.log2_scale_factor;
2367 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2368 i.types[op] &= ~Disp;
2369 if (flag_code != CODE_64BIT)
2370 i.types[op] |= Disp32; /* Must be 32 bit */
2371 else
2372 i.types[op] |= Disp32S;
2373 if (i.index_reg->reg_flags & RegRex)
2374 i.rex.extY=1;
2377 /* RIP addressing for 64bit mode. */
2378 else if (i.base_reg->reg_type == BaseIndex)
2380 i.rm.regmem = NO_BASE_REGISTER;
2381 i.types[op] &= ~Disp;
2382 i.types[op] |= Disp32S;
2383 i.flags[op] = Operand_PCrel;
2385 else if (i.base_reg->reg_type & Reg16)
2387 switch (i.base_reg->reg_num)
2389 case 3: /* (%bx) */
2390 if (! i.index_reg)
2391 i.rm.regmem = 7;
2392 else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */
2393 i.rm.regmem = i.index_reg->reg_num - 6;
2394 break;
2395 case 5: /* (%bp) */
2396 default_seg = &ss;
2397 if (! i.index_reg)
2399 i.rm.regmem = 6;
2400 if ((i.types[op] & Disp) == 0)
2402 /* fake (%bp) into 0(%bp) */
2403 i.types[op] |= Disp8;
2404 fake_zero_displacement = 1;
2407 else /* (%bp,%si) -> 2, or (%bp,%di) -> 3 */
2408 i.rm.regmem = i.index_reg->reg_num - 6 + 2;
2409 break;
2410 default: /* (%si) -> 4 or (%di) -> 5 */
2411 i.rm.regmem = i.base_reg->reg_num - 6 + 4;
2413 i.rm.mode = mode_from_disp_size (i.types[op]);
2415 else /* i.base_reg and 32/64 bit mode */
2417 if (flag_code == CODE_64BIT
2418 && (i.types[op] & Disp))
2420 if (i.types[op] & Disp8)
2421 i.types[op] = Disp8 | Disp32S;
2422 else
2423 i.types[op] = Disp32S;
2425 i.rm.regmem = i.base_reg->reg_num;
2426 if (i.base_reg->reg_flags & RegRex)
2427 i.rex.extZ=1;
2428 i.sib.base = i.base_reg->reg_num;
2429 /* x86-64 ignores REX prefix bit here to avoid
2430 decoder complications. */
2431 if ((i.base_reg->reg_num & 7) == EBP_REG_NUM)
2433 default_seg = &ss;
2434 if (i.disp_operands == 0)
2436 fake_zero_displacement = 1;
2437 i.types[op] |= Disp8;
2440 else if (i.base_reg->reg_num == ESP_REG_NUM)
2442 default_seg = &ss;
2444 i.sib.scale = i.log2_scale_factor;
2445 if (! i.index_reg)
2447 /* <disp>(%esp) becomes two byte modrm
2448 with no index register. We've already
2449 stored the code for esp in i.rm.regmem
2450 ie. ESCAPE_TO_TWO_BYTE_ADDRESSING. Any
2451 base register besides %esp will not use
2452 the extra modrm byte. */
2453 i.sib.index = NO_INDEX_REGISTER;
2454 #if ! SCALE1_WHEN_NO_INDEX
2455 /* Another case where we force the second
2456 modrm byte. */
2457 if (i.log2_scale_factor)
2458 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2459 #endif
2461 else
2463 i.sib.index = i.index_reg->reg_num;
2464 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2465 if (i.index_reg->reg_flags & RegRex)
2466 i.rex.extY=1;
2468 i.rm.mode = mode_from_disp_size (i.types[op]);
2471 if (fake_zero_displacement)
2473 /* Fakes a zero displacement assuming that i.types[op]
2474 holds the correct displacement size. */
2475 expressionS *exp;
2477 assert (i.op[op].disps == 0);
2478 exp = &disp_expressions[i.disp_operands++];
2479 i.op[op].disps = exp;
2480 exp->X_op = O_constant;
2481 exp->X_add_number = 0;
2482 exp->X_add_symbol = (symbolS *) 0;
2483 exp->X_op_symbol = (symbolS *) 0;
2487 /* Fill in i.rm.reg or i.rm.regmem field with register
2488 operand (if any) based on i.tm.extension_opcode.
2489 Again, we must be careful to make sure that
2490 segment/control/debug/test/MMX registers are coded
2491 into the i.rm.reg field. */
2492 if (i.reg_operands)
2494 unsigned int op =
2495 ((i.types[0]
2496 & (Reg | RegMMX | RegXMM
2497 | SReg2 | SReg3
2498 | Control | Debug | Test))
2500 : ((i.types[1]
2501 & (Reg | RegMMX | RegXMM
2502 | SReg2 | SReg3
2503 | Control | Debug | Test))
2505 : 2));
2506 /* If there is an extension opcode to put here, the
2507 register number must be put into the regmem field. */
2508 if (i.tm.extension_opcode != None)
2510 i.rm.regmem = i.op[op].regs->reg_num;
2511 if (i.op[op].regs->reg_flags & RegRex)
2512 i.rex.extZ=1;
2514 else
2516 i.rm.reg = i.op[op].regs->reg_num;
2517 if (i.op[op].regs->reg_flags & RegRex)
2518 i.rex.extX=1;
2521 /* Now, if no memory operand has set i.rm.mode = 0, 1, 2
2522 we must set it to 3 to indicate this is a register
2523 operand in the regmem field. */
2524 if (!i.mem_operands)
2525 i.rm.mode = 3;
2528 /* Fill in i.rm.reg field with extension opcode (if any). */
2529 if (i.tm.extension_opcode != None)
2530 i.rm.reg = i.tm.extension_opcode;
2533 else if (i.tm.opcode_modifier & (Seg2ShortForm | Seg3ShortForm))
2535 if (i.tm.base_opcode == POP_SEG_SHORT
2536 && i.op[0].regs->reg_num == 1)
2538 as_bad (_("you can't `pop %%cs'"));
2539 return;
2541 i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
2542 if (i.op[0].regs->reg_flags & RegRex)
2543 i.rex.extZ = 1;
2545 else if ((i.tm.base_opcode & ~(D|W)) == MOV_AX_DISP32)
2547 default_seg = &ds;
2549 else if ((i.tm.opcode_modifier & IsString) != 0)
2551 /* For the string instructions that allow a segment override
2552 on one of their operands, the default segment is ds. */
2553 default_seg = &ds;
2556 /* If a segment was explicitly specified,
2557 and the specified segment is not the default,
2558 use an opcode prefix to select it.
2559 If we never figured out what the default segment is,
2560 then default_seg will be zero at this point,
2561 and the specified segment prefix will always be used. */
2562 if ((i.seg[0]) && (i.seg[0] != default_seg))
2564 if (! add_prefix (i.seg[0]->seg_prefix))
2565 return;
2568 else if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
2570 /* UnixWare fsub no args is alias for fsubp, fadd -> faddp, etc. */
2571 as_warn (_("translating to `%sp'"), i.tm.name);
2575 /* Handle conversion of 'int $3' --> special int3 insn. */
2576 if (i.tm.base_opcode == INT_OPCODE && i.op[0].imms->X_add_number == 3)
2578 i.tm.base_opcode = INT3_OPCODE;
2579 i.imm_operands = 0;
2582 if ((i.tm.opcode_modifier & (Jump | JumpByte | JumpDword))
2583 && i.op[0].disps->X_op == O_constant)
2585 /* Convert "jmp constant" (and "call constant") to a jump (call) to
2586 the absolute address given by the constant. Since ix86 jumps and
2587 calls are pc relative, we need to generate a reloc. */
2588 i.op[0].disps->X_add_symbol = &abs_symbol;
2589 i.op[0].disps->X_op = O_symbol;
2592 if (i.tm.opcode_modifier & Rex64)
2593 i.rex.mode64 = 1;
2595 /* For 8bit registers we would need an empty rex prefix.
2596 Also in the case instruction is already having prefix,
2597 we need to convert old registers to new ones. */
2599 if (((i.types[0] & Reg8) && (i.op[0].regs->reg_flags & RegRex64))
2600 || ((i.types[1] & Reg8) && (i.op[1].regs->reg_flags & RegRex64))
2601 || ((i.rex.mode64 || i.rex.extX || i.rex.extY || i.rex.extZ || i.rex.empty)
2602 && ((i.types[0] & Reg8) || (i.types[1] & Reg8))))
2604 int x;
2605 i.rex.empty=1;
2606 for (x = 0; x < 2; x++)
2608 /* Look for 8bit operand that does use old registers. */
2609 if (i.types[x] & Reg8
2610 && !(i.op[x].regs->reg_flags & RegRex64))
2612 /* In case it is "hi" register, give up. */
2613 if (i.op[x].regs->reg_num > 3)
2614 as_bad (_("Can't encode registers '%%%s' in the instruction requiring REX prefix.\n"),
2615 i.op[x].regs->reg_name);
2617 /* Otherwise it is equivalent to the extended register.
2618 Since the encoding don't change this is merely cosmetical
2619 cleanup for debug output. */
2621 i.op[x].regs = i.op[x].regs + 8;
2626 if (i.rex.mode64 || i.rex.extX || i.rex.extY || i.rex.extZ || i.rex.empty)
2627 add_prefix (0x40
2628 | (i.rex.mode64 ? 8 : 0)
2629 | (i.rex.extX ? 4 : 0)
2630 | (i.rex.extY ? 2 : 0)
2631 | (i.rex.extZ ? 1 : 0));
2633 /* We are ready to output the insn. */
2635 register char *p;
2637 /* Output jumps. */
2638 if (i.tm.opcode_modifier & Jump)
2640 int size;
2641 int code16;
2642 int prefix;
2644 code16 = 0;
2645 if (flag_code == CODE_16BIT)
2646 code16 = CODE16;
2648 prefix = 0;
2649 if (i.prefix[DATA_PREFIX])
2651 prefix = 1;
2652 i.prefixes -= 1;
2653 code16 ^= CODE16;
2655 if (i.prefix[REX_PREFIX])
2657 prefix++;
2658 i.prefixes --;
2661 size = 4;
2662 if (code16)
2663 size = 2;
2665 if (i.prefixes != 0 && !intel_syntax)
2666 as_warn (_("skipping prefixes on this instruction"));
2668 /* It's always a symbol; End frag & setup for relax.
2669 Make sure there is enough room in this frag for the largest
2670 instruction we may generate in md_convert_frag. This is 2
2671 bytes for the opcode and room for the prefix and largest
2672 displacement. */
2673 frag_grow (prefix + 2 + size);
2674 insn_size += prefix + 1;
2675 /* Prefix and 1 opcode byte go in fr_fix. */
2676 p = frag_more (prefix + 1);
2677 if (i.prefix[DATA_PREFIX])
2678 *p++ = DATA_PREFIX_OPCODE;
2679 if (i.prefix[REX_PREFIX])
2680 *p++ = i.prefix[REX_PREFIX];
2681 *p = i.tm.base_opcode;
2682 /* 1 possible extra opcode + displacement go in var part.
2683 Pass reloc in fr_var. */
2684 frag_var (rs_machine_dependent,
2685 1 + size,
2686 i.disp_reloc[0],
2687 ((unsigned char) *p == JUMP_PC_RELATIVE
2688 ? ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL) | code16
2689 : ENCODE_RELAX_STATE (COND_JUMP, SMALL) | code16),
2690 i.op[0].disps->X_add_symbol,
2691 i.op[0].disps->X_add_number,
2694 else if (i.tm.opcode_modifier & (JumpByte | JumpDword))
2696 int size;
2698 if (i.tm.opcode_modifier & JumpByte)
2700 /* This is a loop or jecxz type instruction. */
2701 size = 1;
2702 if (i.prefix[ADDR_PREFIX])
2704 insn_size += 1;
2705 FRAG_APPEND_1_CHAR (ADDR_PREFIX_OPCODE);
2706 i.prefixes -= 1;
2709 else
2711 int code16;
2713 code16 = 0;
2714 if (flag_code == CODE_16BIT)
2715 code16 = CODE16;
2717 if (i.prefix[DATA_PREFIX])
2719 insn_size += 1;
2720 FRAG_APPEND_1_CHAR (DATA_PREFIX_OPCODE);
2721 i.prefixes -= 1;
2722 code16 ^= CODE16;
2725 size = 4;
2726 if (code16)
2727 size = 2;
2730 if (i.prefix[REX_PREFIX])
2732 FRAG_APPEND_1_CHAR (i.prefix[REX_PREFIX]);
2733 insn_size++;
2734 i.prefixes -= 1;
2737 if (i.prefixes != 0 && !intel_syntax)
2738 as_warn (_("skipping prefixes on this instruction"));
2740 if (fits_in_unsigned_byte (i.tm.base_opcode))
2742 insn_size += 1 + size;
2743 p = frag_more (1 + size);
2745 else
2747 /* Opcode can be at most two bytes. */
2748 insn_size += 2 + size;
2749 p = frag_more (2 + size);
2750 *p++ = (i.tm.base_opcode >> 8) & 0xff;
2752 *p++ = i.tm.base_opcode & 0xff;
2754 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
2755 i.op[0].disps, 1, reloc (size, 1, 1, i.disp_reloc[0]));
2757 else if (i.tm.opcode_modifier & JumpInterSegment)
2759 int size;
2760 int prefix;
2761 int code16;
2763 code16 = 0;
2764 if (flag_code == CODE_16BIT)
2765 code16 = CODE16;
2767 prefix = 0;
2768 if (i.prefix[DATA_PREFIX])
2770 prefix = 1;
2771 i.prefixes -= 1;
2772 code16 ^= CODE16;
2774 if (i.prefix[REX_PREFIX])
2776 prefix++;
2777 i.prefixes -= 1;
2780 size = 4;
2781 if (code16)
2782 size = 2;
2784 if (i.prefixes != 0 && !intel_syntax)
2785 as_warn (_("skipping prefixes on this instruction"));
2787 /* 1 opcode; 2 segment; offset */
2788 insn_size += prefix + 1 + 2 + size;
2789 p = frag_more (prefix + 1 + 2 + size);
2791 if (i.prefix[DATA_PREFIX])
2792 *p++ = DATA_PREFIX_OPCODE;
2794 if (i.prefix[REX_PREFIX])
2795 *p++ = i.prefix[REX_PREFIX];
2797 *p++ = i.tm.base_opcode;
2798 if (i.op[1].imms->X_op == O_constant)
2800 offsetT n = i.op[1].imms->X_add_number;
2802 if (size == 2
2803 && !fits_in_unsigned_word (n)
2804 && !fits_in_signed_word (n))
2806 as_bad (_("16-bit jump out of range"));
2807 return;
2809 md_number_to_chars (p, n, size);
2811 else
2812 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
2813 i.op[1].imms, 0, reloc (size, 0, 0, i.disp_reloc[0]));
2814 if (i.op[0].imms->X_op != O_constant)
2815 as_bad (_("can't handle non absolute segment in `%s'"),
2816 i.tm.name);
2817 md_number_to_chars (p + size, (valueT) i.op[0].imms->X_add_number, 2);
2819 else
2821 /* Output normal instructions here. */
2822 unsigned char *q;
2824 /* The prefix bytes. */
2825 for (q = i.prefix;
2826 q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
2827 q++)
2829 if (*q)
2831 insn_size += 1;
2832 p = frag_more (1);
2833 md_number_to_chars (p, (valueT) *q, 1);
2837 /* Now the opcode; be careful about word order here! */
2838 if (fits_in_unsigned_byte (i.tm.base_opcode))
2840 insn_size += 1;
2841 FRAG_APPEND_1_CHAR (i.tm.base_opcode);
2843 else if (fits_in_unsigned_word (i.tm.base_opcode))
2845 insn_size += 2;
2846 p = frag_more (2);
2847 /* Put out high byte first: can't use md_number_to_chars! */
2848 *p++ = (i.tm.base_opcode >> 8) & 0xff;
2849 *p = i.tm.base_opcode & 0xff;
2851 else
2852 { /* Opcode is either 3 or 4 bytes. */
2853 if (i.tm.base_opcode & 0xff000000)
2855 insn_size += 4;
2856 p = frag_more (4);
2857 *p++ = (i.tm.base_opcode >> 24) & 0xff;
2859 else
2861 insn_size += 3;
2862 p = frag_more (3);
2864 *p++ = (i.tm.base_opcode >> 16) & 0xff;
2865 *p++ = (i.tm.base_opcode >> 8) & 0xff;
2866 *p = (i.tm.base_opcode) & 0xff;
2869 /* Now the modrm byte and sib byte (if present). */
2870 if (i.tm.opcode_modifier & Modrm)
2872 insn_size += 1;
2873 p = frag_more (1);
2874 md_number_to_chars (p,
2875 (valueT) (i.rm.regmem << 0
2876 | i.rm.reg << 3
2877 | i.rm.mode << 6),
2879 /* If i.rm.regmem == ESP (4)
2880 && i.rm.mode != (Register mode)
2881 && not 16 bit
2882 ==> need second modrm byte. */
2883 if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING
2884 && i.rm.mode != 3
2885 && !(i.base_reg && (i.base_reg->reg_type & Reg16) != 0))
2887 insn_size += 1;
2888 p = frag_more (1);
2889 md_number_to_chars (p,
2890 (valueT) (i.sib.base << 0
2891 | i.sib.index << 3
2892 | i.sib.scale << 6),
2897 if (i.disp_operands)
2899 register unsigned int n;
2901 for (n = 0; n < i.operands; n++)
2903 if (i.types[n] & Disp)
2905 if (i.op[n].disps->X_op == O_constant)
2907 int size;
2908 offsetT val;
2910 size = 4;
2911 if (i.types[n] & (Disp8 | Disp16 | Disp64))
2913 size = 2;
2914 if (i.types[n] & Disp8)
2915 size = 1;
2916 if (i.types[n] & Disp64)
2917 size = 8;
2919 val = offset_in_range (i.op[n].disps->X_add_number,
2920 size);
2921 insn_size += size;
2922 p = frag_more (size);
2923 md_number_to_chars (p, val, size);
2925 else
2927 int size = 4;
2928 int sign = 0;
2929 int pcrel = (i.flags[n] & Operand_PCrel) != 0;
2931 /* The PC relative address is computed relative
2932 to the instruction boundary, so in case immediate
2933 fields follows, we need to adjust the value. */
2934 if (pcrel && i.imm_operands)
2936 int imm_size = 4;
2937 register unsigned int n1;
2939 for (n1 = 0; n1 < i.operands; n1++)
2940 if (i.types[n1] & Imm)
2942 if (i.types[n1] & (Imm8 | Imm8S | Imm16 | Imm64))
2944 imm_size = 2;
2945 if (i.types[n1] & (Imm8 | Imm8S))
2946 imm_size = 1;
2947 if (i.types[n1] & Imm64)
2948 imm_size = 8;
2950 break;
2952 /* We should find the immediate. */
2953 if (n1 == i.operands)
2954 abort();
2955 i.op[n].disps->X_add_number -= imm_size;
2958 if (i.types[n] & Disp32S)
2959 sign = 1;
2961 if (i.types[n] & (Disp16 | Disp64))
2963 size = 2;
2964 if (i.types[n] & Disp64)
2965 size = 8;
2968 insn_size += size;
2969 p = frag_more (size);
2970 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
2971 i.op[n].disps, pcrel,
2972 reloc (size, pcrel, sign, i.disp_reloc[n]));
2978 /* Output immediate. */
2979 if (i.imm_operands)
2981 register unsigned int n;
2983 for (n = 0; n < i.operands; n++)
2985 if (i.types[n] & Imm)
2987 if (i.op[n].imms->X_op == O_constant)
2989 int size;
2990 offsetT val;
2992 size = 4;
2993 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
2995 size = 2;
2996 if (i.types[n] & (Imm8 | Imm8S))
2997 size = 1;
2998 else if (i.types[n] & Imm64)
2999 size = 8;
3001 val = offset_in_range (i.op[n].imms->X_add_number,
3002 size);
3003 insn_size += size;
3004 p = frag_more (size);
3005 md_number_to_chars (p, val, size);
3007 else
3009 /* Not absolute_section.
3010 Need a 32-bit fixup (don't support 8bit
3011 non-absolute imms). Try to support other
3012 sizes ... */
3013 #ifdef BFD_ASSEMBLER
3014 enum bfd_reloc_code_real reloc_type;
3015 #else
3016 int reloc_type;
3017 #endif
3018 int size = 4;
3019 int sign = 0;
3021 if ((i.types[n] & (Imm32S))
3022 && i.suffix == QWORD_MNEM_SUFFIX)
3023 sign = 1;
3024 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3026 size = 2;
3027 if (i.types[n] & (Imm8 | Imm8S))
3028 size = 1;
3029 if (i.types[n] & Imm64)
3030 size = 8;
3033 insn_size += size;
3034 p = frag_more (size);
3035 reloc_type = reloc (size, 0, sign, i.disp_reloc[0]);
3036 #ifdef BFD_ASSEMBLER
3037 if (reloc_type == BFD_RELOC_32
3038 && GOT_symbol
3039 && GOT_symbol == i.op[n].imms->X_add_symbol
3040 && (i.op[n].imms->X_op == O_symbol
3041 || (i.op[n].imms->X_op == O_add
3042 && ((symbol_get_value_expression
3043 (i.op[n].imms->X_op_symbol)->X_op)
3044 == O_subtract))))
3046 /* We don't support dynamic linking on x86-64 yet. */
3047 if (flag_code == CODE_64BIT)
3048 abort();
3049 reloc_type = BFD_RELOC_386_GOTPC;
3050 i.op[n].imms->X_add_number += 3;
3052 #endif
3053 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3054 i.op[n].imms, 0, reloc_type);
3061 dwarf2_emit_insn (insn_size);
3063 #ifdef DEBUG386
3064 if (flag_debug)
3066 pi (line, &i);
3068 #endif /* DEBUG386 */
3072 static int i386_immediate PARAMS ((char *));
3074 static int
3075 i386_immediate (imm_start)
3076 char *imm_start;
3078 char *save_input_line_pointer;
3079 segT exp_seg = 0;
3080 expressionS *exp;
3082 if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
3084 as_bad (_("only 1 or 2 immediate operands are allowed"));
3085 return 0;
3088 exp = &im_expressions[i.imm_operands++];
3089 i.op[this_operand].imms = exp;
3091 if (is_space_char (*imm_start))
3092 ++imm_start;
3094 save_input_line_pointer = input_line_pointer;
3095 input_line_pointer = imm_start;
3097 #ifndef LEX_AT
3099 /* We can have operands of the form
3100 <symbol>@GOTOFF+<nnn>
3101 Take the easy way out here and copy everything
3102 into a temporary buffer... */
3103 register char *cp;
3105 cp = strchr (input_line_pointer, '@');
3106 if (cp != NULL)
3108 char *tmpbuf;
3109 int len = 0;
3110 int first;
3112 /* GOT relocations are not supported in 16 bit mode. */
3113 if (flag_code == CODE_16BIT)
3114 as_bad (_("GOT relocations not supported in 16 bit mode"));
3116 if (GOT_symbol == NULL)
3117 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
3119 if (strncmp (cp + 1, "PLT", 3) == 0)
3121 if (flag_code == CODE_64BIT)
3122 i.disp_reloc[this_operand] = BFD_RELOC_X86_64_PLT32;
3123 else
3124 i.disp_reloc[this_operand] = BFD_RELOC_386_PLT32;
3125 len = 3;
3127 else if (strncmp (cp + 1, "GOTOFF", 6) == 0)
3129 if (flag_code == CODE_64BIT)
3130 as_bad ("GOTOFF relocations are unsupported in 64bit mode.");
3131 i.disp_reloc[this_operand] = BFD_RELOC_386_GOTOFF;
3132 len = 6;
3134 else if (strncmp (cp + 1, "GOT", 3) == 0)
3136 if (flag_code == CODE_64BIT)
3137 i.disp_reloc[this_operand] = BFD_RELOC_X86_64_GOT32;
3138 else
3139 i.disp_reloc[this_operand] = BFD_RELOC_386_GOT32;
3140 len = 3;
3142 else if (strncmp (cp + 1, "GOTPCREL", 3) == 0)
3144 if (flag_code == CODE_64BIT)
3145 i.disp_reloc[this_operand] = BFD_RELOC_X86_64_GOTPCREL;
3146 else
3147 as_bad ("GOTPCREL relocations are supported only in 64bit mode.");
3148 len = 3;
3150 else
3151 as_bad (_("bad reloc specifier in expression"));
3153 /* Replace the relocation token with ' ', so that errors like
3154 foo@GOTOFF1 will be detected. */
3155 first = cp - input_line_pointer;
3156 tmpbuf = (char *) alloca (strlen (input_line_pointer));
3157 memcpy (tmpbuf, input_line_pointer, first);
3158 tmpbuf[first] = ' ';
3159 strcpy (tmpbuf + first + 1, cp + 1 + len);
3160 input_line_pointer = tmpbuf;
3163 #endif
3165 exp_seg = expression (exp);
3167 SKIP_WHITESPACE ();
3168 if (*input_line_pointer)
3169 as_bad (_("ignoring junk `%s' after expression"), input_line_pointer);
3171 input_line_pointer = save_input_line_pointer;
3173 if (exp->X_op == O_absent || exp->X_op == O_big)
3175 /* Missing or bad expr becomes absolute 0. */
3176 as_bad (_("missing or invalid immediate expression `%s' taken as 0"),
3177 imm_start);
3178 exp->X_op = O_constant;
3179 exp->X_add_number = 0;
3180 exp->X_add_symbol = (symbolS *) 0;
3181 exp->X_op_symbol = (symbolS *) 0;
3183 else if (exp->X_op == O_constant)
3185 /* Size it properly later. */
3186 i.types[this_operand] |= Imm64;
3187 /* If BFD64, sign extend val. */
3188 if (!use_rela_relocations)
3189 if ((exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0)
3190 exp->X_add_number = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
3192 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3193 else if (1
3194 #ifdef BFD_ASSEMBLER
3195 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3196 #endif
3197 && exp_seg != text_section
3198 && exp_seg != data_section
3199 && exp_seg != bss_section
3200 && exp_seg != undefined_section
3201 #ifdef BFD_ASSEMBLER
3202 && !bfd_is_com_section (exp_seg)
3203 #endif
3206 #ifdef BFD_ASSEMBLER
3207 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3208 #else
3209 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
3210 #endif
3211 return 0;
3213 #endif
3214 else
3216 /* This is an address. The size of the address will be
3217 determined later, depending on destination register,
3218 suffix, or the default for the section. */
3219 i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64;
3222 return 1;
3225 static int i386_scale PARAMS ((char *));
3227 static int
3228 i386_scale (scale)
3229 char *scale;
3231 if (!isdigit (*scale))
3232 goto bad_scale;
3234 switch (*scale)
3236 case '0':
3237 case '1':
3238 i.log2_scale_factor = 0;
3239 break;
3240 case '2':
3241 i.log2_scale_factor = 1;
3242 break;
3243 case '4':
3244 i.log2_scale_factor = 2;
3245 break;
3246 case '8':
3247 i.log2_scale_factor = 3;
3248 break;
3249 default:
3250 bad_scale:
3251 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
3252 scale);
3253 return 0;
3255 if (i.log2_scale_factor != 0 && ! i.index_reg)
3257 as_warn (_("scale factor of %d without an index register"),
3258 1 << i.log2_scale_factor);
3259 #if SCALE1_WHEN_NO_INDEX
3260 i.log2_scale_factor = 0;
3261 #endif
3263 return 1;
3266 static int i386_displacement PARAMS ((char *, char *));
3268 static int
3269 i386_displacement (disp_start, disp_end)
3270 char *disp_start;
3271 char *disp_end;
3273 register expressionS *exp;
3274 segT exp_seg = 0;
3275 char *save_input_line_pointer;
3276 int bigdisp = Disp32;
3278 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3279 bigdisp = Disp16;
3280 if (flag_code == CODE_64BIT)
3281 bigdisp = Disp64;
3282 i.types[this_operand] |= bigdisp;
3284 exp = &disp_expressions[i.disp_operands];
3285 i.op[this_operand].disps = exp;
3286 i.disp_operands++;
3287 save_input_line_pointer = input_line_pointer;
3288 input_line_pointer = disp_start;
3289 END_STRING_AND_SAVE (disp_end);
3291 #ifndef GCC_ASM_O_HACK
3292 #define GCC_ASM_O_HACK 0
3293 #endif
3294 #if GCC_ASM_O_HACK
3295 END_STRING_AND_SAVE (disp_end + 1);
3296 if ((i.types[this_operand] & BaseIndex) != 0
3297 && displacement_string_end[-1] == '+')
3299 /* This hack is to avoid a warning when using the "o"
3300 constraint within gcc asm statements.
3301 For instance:
3303 #define _set_tssldt_desc(n,addr,limit,type) \
3304 __asm__ __volatile__ ( \
3305 "movw %w2,%0\n\t" \
3306 "movw %w1,2+%0\n\t" \
3307 "rorl $16,%1\n\t" \
3308 "movb %b1,4+%0\n\t" \
3309 "movb %4,5+%0\n\t" \
3310 "movb $0,6+%0\n\t" \
3311 "movb %h1,7+%0\n\t" \
3312 "rorl $16,%1" \
3313 : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
3315 This works great except that the output assembler ends
3316 up looking a bit weird if it turns out that there is
3317 no offset. You end up producing code that looks like:
3319 #APP
3320 movw $235,(%eax)
3321 movw %dx,2+(%eax)
3322 rorl $16,%edx
3323 movb %dl,4+(%eax)
3324 movb $137,5+(%eax)
3325 movb $0,6+(%eax)
3326 movb %dh,7+(%eax)
3327 rorl $16,%edx
3328 #NO_APP
3330 So here we provide the missing zero. */
3332 *displacement_string_end = '0';
3334 #endif
3335 #ifndef LEX_AT
3337 /* We can have operands of the form
3338 <symbol>@GOTOFF+<nnn>
3339 Take the easy way out here and copy everything
3340 into a temporary buffer... */
3341 register char *cp;
3343 cp = strchr (input_line_pointer, '@');
3344 if (cp != NULL)
3346 char *tmpbuf;
3347 int len = 0;
3348 int first;
3350 /* GOT relocations are not supported in 16 bit mode. */
3351 if (flag_code == CODE_16BIT)
3352 as_bad (_("GOT relocations not supported in 16 bit mode"));
3354 if (GOT_symbol == NULL)
3355 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
3357 if (strncmp (cp + 1, "PLT", 3) == 0)
3359 if (flag_code == CODE_64BIT)
3360 i.disp_reloc[this_operand] = BFD_RELOC_X86_64_PLT32;
3361 else
3362 i.disp_reloc[this_operand] = BFD_RELOC_386_PLT32;
3363 len = 3;
3365 else if (strncmp (cp + 1, "GOTOFF", 6) == 0)
3367 if (flag_code == CODE_64BIT)
3368 as_bad ("GOTOFF relocation is not supported in 64bit mode.");
3369 i.disp_reloc[this_operand] = BFD_RELOC_386_GOTOFF;
3370 len = 6;
3372 else if (strncmp (cp + 1, "GOT", 3) == 0)
3374 if (flag_code == CODE_64BIT)
3375 i.disp_reloc[this_operand] = BFD_RELOC_X86_64_GOT32;
3376 else
3377 i.disp_reloc[this_operand] = BFD_RELOC_386_GOT32;
3378 len = 3;
3380 else if (strncmp (cp + 1, "GOTPCREL", 3) == 0)
3382 if (flag_code != CODE_64BIT)
3383 as_bad ("GOTPCREL relocation is supported only in 64bit mode.");
3384 i.disp_reloc[this_operand] = BFD_RELOC_X86_64_GOTPCREL;
3385 len = 3;
3387 else
3388 as_bad (_("bad reloc specifier in expression"));
3390 /* Replace the relocation token with ' ', so that errors like
3391 foo@GOTOFF1 will be detected. */
3392 first = cp - input_line_pointer;
3393 tmpbuf = (char *) alloca (strlen (input_line_pointer));
3394 memcpy (tmpbuf, input_line_pointer, first);
3395 tmpbuf[first] = ' ';
3396 strcpy (tmpbuf + first + 1, cp + 1 + len);
3397 input_line_pointer = tmpbuf;
3400 #endif
3402 exp_seg = expression (exp);
3404 #ifdef BFD_ASSEMBLER
3405 /* We do this to make sure that the section symbol is in
3406 the symbol table. We will ultimately change the relocation
3407 to be relative to the beginning of the section. */
3408 if (i.disp_reloc[this_operand] == BFD_RELOC_386_GOTOFF
3409 || i.disp_reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3411 if (S_IS_LOCAL(exp->X_add_symbol)
3412 && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
3413 section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
3414 assert (exp->X_op == O_symbol);
3415 exp->X_op = O_subtract;
3416 exp->X_op_symbol = GOT_symbol;
3417 i.disp_reloc[this_operand] = BFD_RELOC_32;
3419 #endif
3421 SKIP_WHITESPACE ();
3422 if (*input_line_pointer)
3423 as_bad (_("ignoring junk `%s' after expression"),
3424 input_line_pointer);
3425 #if GCC_ASM_O_HACK
3426 RESTORE_END_STRING (disp_end + 1);
3427 #endif
3428 RESTORE_END_STRING (disp_end);
3429 input_line_pointer = save_input_line_pointer;
3431 if (exp->X_op == O_absent || exp->X_op == O_big)
3433 /* Missing or bad expr becomes absolute 0. */
3434 as_bad (_("missing or invalid displacement expression `%s' taken as 0"),
3435 disp_start);
3436 exp->X_op = O_constant;
3437 exp->X_add_number = 0;
3438 exp->X_add_symbol = (symbolS *) 0;
3439 exp->X_op_symbol = (symbolS *) 0;
3442 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3443 if (exp->X_op != O_constant
3444 #ifdef BFD_ASSEMBLER
3445 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3446 #endif
3447 && exp_seg != text_section
3448 && exp_seg != data_section
3449 && exp_seg != bss_section
3450 && exp_seg != undefined_section)
3452 #ifdef BFD_ASSEMBLER
3453 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3454 #else
3455 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
3456 #endif
3457 return 0;
3459 #endif
3460 else if (flag_code == CODE_64BIT)
3461 i.types[this_operand] |= Disp32S | Disp32;
3462 return 1;
3465 static int i386_index_check PARAMS((const char *));
3467 /* Make sure the memory operand we've been dealt is valid.
3468 Return 1 on success, 0 on a failure. */
3470 static int
3471 i386_index_check (operand_string)
3472 const char *operand_string;
3474 int ok;
3475 #if INFER_ADDR_PREFIX
3476 int fudged = 0;
3478 tryprefix:
3479 #endif
3480 ok = 1;
3481 if (flag_code == CODE_64BIT)
3483 /* 64bit checks. */
3484 if ((i.base_reg
3485 && ((i.base_reg->reg_type & Reg64) == 0)
3486 && (i.base_reg->reg_type != BaseIndex
3487 || i.index_reg))
3488 || (i.index_reg
3489 && ((i.index_reg->reg_type & (Reg64|BaseIndex))
3490 != (Reg64|BaseIndex))))
3491 ok = 0;
3493 else
3495 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3497 /* 16bit checks. */
3498 if ((i.base_reg
3499 && ((i.base_reg->reg_type & (Reg16|BaseIndex|RegRex))
3500 != (Reg16|BaseIndex)))
3501 || (i.index_reg
3502 && (((i.index_reg->reg_type & (Reg16|BaseIndex))
3503 != (Reg16|BaseIndex))
3504 || ! (i.base_reg
3505 && i.base_reg->reg_num < 6
3506 && i.index_reg->reg_num >= 6
3507 && i.log2_scale_factor == 0))))
3508 ok = 0;
3510 else
3512 /* 32bit checks. */
3513 if ((i.base_reg
3514 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
3515 || (i.index_reg
3516 && ((i.index_reg->reg_type & (Reg32|BaseIndex|RegRex))
3517 != (Reg32|BaseIndex))))
3518 ok = 0;
3521 if (!ok)
3523 #if INFER_ADDR_PREFIX
3524 if (flag_code != CODE_64BIT
3525 && i.prefix[ADDR_PREFIX] == 0 && stackop_size != '\0')
3527 i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
3528 i.prefixes += 1;
3529 /* Change the size of any displacement too. At most one of
3530 Disp16 or Disp32 is set.
3531 FIXME. There doesn't seem to be any real need for separate
3532 Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
3533 Removing them would probably clean up the code quite a lot. */
3534 if (i.types[this_operand] & (Disp16|Disp32))
3535 i.types[this_operand] ^= (Disp16|Disp32);
3536 fudged = 1;
3537 goto tryprefix;
3539 if (fudged)
3540 as_bad (_("`%s' is not a valid base/index expression"),
3541 operand_string);
3542 else
3543 #endif
3544 as_bad (_("`%s' is not a valid %s bit base/index expression"),
3545 operand_string,
3546 flag_code_names[flag_code]);
3547 return 0;
3549 return 1;
3552 /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
3553 on error. */
3555 static int
3556 i386_operand (operand_string)
3557 char *operand_string;
3559 const reg_entry *r;
3560 char *end_op;
3561 char *op_string = operand_string;
3563 if (is_space_char (*op_string))
3564 ++op_string;
3566 /* We check for an absolute prefix (differentiating,
3567 for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
3568 if (*op_string == ABSOLUTE_PREFIX)
3570 ++op_string;
3571 if (is_space_char (*op_string))
3572 ++op_string;
3573 i.types[this_operand] |= JumpAbsolute;
3576 /* Check if operand is a register. */
3577 if ((*op_string == REGISTER_PREFIX || allow_naked_reg)
3578 && (r = parse_register (op_string, &end_op)) != NULL)
3580 /* Check for a segment override by searching for ':' after a
3581 segment register. */
3582 op_string = end_op;
3583 if (is_space_char (*op_string))
3584 ++op_string;
3585 if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
3587 switch (r->reg_num)
3589 case 0:
3590 i.seg[i.mem_operands] = &es;
3591 break;
3592 case 1:
3593 i.seg[i.mem_operands] = &cs;
3594 break;
3595 case 2:
3596 i.seg[i.mem_operands] = &ss;
3597 break;
3598 case 3:
3599 i.seg[i.mem_operands] = &ds;
3600 break;
3601 case 4:
3602 i.seg[i.mem_operands] = &fs;
3603 break;
3604 case 5:
3605 i.seg[i.mem_operands] = &gs;
3606 break;
3609 /* Skip the ':' and whitespace. */
3610 ++op_string;
3611 if (is_space_char (*op_string))
3612 ++op_string;
3614 if (!is_digit_char (*op_string)
3615 && !is_identifier_char (*op_string)
3616 && *op_string != '('
3617 && *op_string != ABSOLUTE_PREFIX)
3619 as_bad (_("bad memory operand `%s'"), op_string);
3620 return 0;
3622 /* Handle case of %es:*foo. */
3623 if (*op_string == ABSOLUTE_PREFIX)
3625 ++op_string;
3626 if (is_space_char (*op_string))
3627 ++op_string;
3628 i.types[this_operand] |= JumpAbsolute;
3630 goto do_memory_reference;
3632 if (*op_string)
3634 as_bad (_("junk `%s' after register"), op_string);
3635 return 0;
3637 i.types[this_operand] |= r->reg_type & ~BaseIndex;
3638 i.op[this_operand].regs = r;
3639 i.reg_operands++;
3641 else if (*op_string == REGISTER_PREFIX)
3643 as_bad (_("bad register name `%s'"), op_string);
3644 return 0;
3646 else if (*op_string == IMMEDIATE_PREFIX)
3648 ++op_string;
3649 if (i.types[this_operand] & JumpAbsolute)
3651 as_bad (_("immediate operand illegal with absolute jump"));
3652 return 0;
3654 if (!i386_immediate (op_string))
3655 return 0;
3657 else if (is_digit_char (*op_string)
3658 || is_identifier_char (*op_string)
3659 || *op_string == '(' )
3661 /* This is a memory reference of some sort. */
3662 char *base_string;
3664 /* Start and end of displacement string expression (if found). */
3665 char *displacement_string_start;
3666 char *displacement_string_end;
3668 do_memory_reference:
3669 if ((i.mem_operands == 1
3670 && (current_templates->start->opcode_modifier & IsString) == 0)
3671 || i.mem_operands == 2)
3673 as_bad (_("too many memory references for `%s'"),
3674 current_templates->start->name);
3675 return 0;
3678 /* Check for base index form. We detect the base index form by
3679 looking for an ')' at the end of the operand, searching
3680 for the '(' matching it, and finding a REGISTER_PREFIX or ','
3681 after the '('. */
3682 base_string = op_string + strlen (op_string);
3684 --base_string;
3685 if (is_space_char (*base_string))
3686 --base_string;
3688 /* If we only have a displacement, set-up for it to be parsed later. */
3689 displacement_string_start = op_string;
3690 displacement_string_end = base_string + 1;
3692 if (*base_string == ')')
3694 char *temp_string;
3695 unsigned int parens_balanced = 1;
3696 /* We've already checked that the number of left & right ()'s are
3697 equal, so this loop will not be infinite. */
3700 base_string--;
3701 if (*base_string == ')')
3702 parens_balanced++;
3703 if (*base_string == '(')
3704 parens_balanced--;
3706 while (parens_balanced);
3708 temp_string = base_string;
3710 /* Skip past '(' and whitespace. */
3711 ++base_string;
3712 if (is_space_char (*base_string))
3713 ++base_string;
3715 if (*base_string == ','
3716 || ((*base_string == REGISTER_PREFIX || allow_naked_reg)
3717 && (i.base_reg = parse_register (base_string, &end_op)) != NULL))
3719 displacement_string_end = temp_string;
3721 i.types[this_operand] |= BaseIndex;
3723 if (i.base_reg)
3725 base_string = end_op;
3726 if (is_space_char (*base_string))
3727 ++base_string;
3730 /* There may be an index reg or scale factor here. */
3731 if (*base_string == ',')
3733 ++base_string;
3734 if (is_space_char (*base_string))
3735 ++base_string;
3737 if ((*base_string == REGISTER_PREFIX || allow_naked_reg)
3738 && (i.index_reg = parse_register (base_string, &end_op)) != NULL)
3740 base_string = end_op;
3741 if (is_space_char (*base_string))
3742 ++base_string;
3743 if (*base_string == ',')
3745 ++base_string;
3746 if (is_space_char (*base_string))
3747 ++base_string;
3749 else if (*base_string != ')' )
3751 as_bad (_("expecting `,' or `)' after index register in `%s'"),
3752 operand_string);
3753 return 0;
3756 else if (*base_string == REGISTER_PREFIX)
3758 as_bad (_("bad register name `%s'"), base_string);
3759 return 0;
3762 /* Check for scale factor. */
3763 if (isdigit ((unsigned char) *base_string))
3765 if (!i386_scale (base_string))
3766 return 0;
3768 ++base_string;
3769 if (is_space_char (*base_string))
3770 ++base_string;
3771 if (*base_string != ')')
3773 as_bad (_("expecting `)' after scale factor in `%s'"),
3774 operand_string);
3775 return 0;
3778 else if (!i.index_reg)
3780 as_bad (_("expecting index register or scale factor after `,'; got '%c'"),
3781 *base_string);
3782 return 0;
3785 else if (*base_string != ')')
3787 as_bad (_("expecting `,' or `)' after base register in `%s'"),
3788 operand_string);
3789 return 0;
3792 else if (*base_string == REGISTER_PREFIX)
3794 as_bad (_("bad register name `%s'"), base_string);
3795 return 0;
3799 /* If there's an expression beginning the operand, parse it,
3800 assuming displacement_string_start and
3801 displacement_string_end are meaningful. */
3802 if (displacement_string_start != displacement_string_end)
3804 if (!i386_displacement (displacement_string_start,
3805 displacement_string_end))
3806 return 0;
3809 /* Special case for (%dx) while doing input/output op. */
3810 if (i.base_reg
3811 && i.base_reg->reg_type == (Reg16 | InOutPortReg)
3812 && i.index_reg == 0
3813 && i.log2_scale_factor == 0
3814 && i.seg[i.mem_operands] == 0
3815 && (i.types[this_operand] & Disp) == 0)
3817 i.types[this_operand] = InOutPortReg;
3818 return 1;
3821 if (i386_index_check (operand_string) == 0)
3822 return 0;
3823 i.mem_operands++;
3825 else
3827 /* It's not a memory operand; argh! */
3828 as_bad (_("invalid char %s beginning operand %d `%s'"),
3829 output_invalid (*op_string),
3830 this_operand + 1,
3831 op_string);
3832 return 0;
3834 return 1; /* Normal return. */
3837 /* md_estimate_size_before_relax()
3839 Called just before relax() for rs_machine_dependent frags. The x86
3840 assembler uses these frags to handle variable size jump
3841 instructions.
3843 Any symbol that is now undefined will not become defined.
3844 Return the correct fr_subtype in the frag.
3845 Return the initial "guess for variable size of frag" to caller.
3846 The guess is actually the growth beyond the fixed part. Whatever
3847 we do to grow the fixed or variable part contributes to our
3848 returned value. */
3851 md_estimate_size_before_relax (fragP, segment)
3852 register fragS *fragP;
3853 register segT segment;
3855 /* We've already got fragP->fr_subtype right; all we have to do is
3856 check for un-relaxable symbols. On an ELF system, we can't relax
3857 an externally visible symbol, because it may be overridden by a
3858 shared library. */
3859 if (S_GET_SEGMENT (fragP->fr_symbol) != segment
3860 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
3861 || S_IS_EXTERNAL (fragP->fr_symbol)
3862 || S_IS_WEAK (fragP->fr_symbol)
3863 #endif
3866 /* Symbol is undefined in this segment, or we need to keep a
3867 reloc so that weak symbols can be overridden. */
3868 int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
3869 #ifdef BFD_ASSEMBLER
3870 enum bfd_reloc_code_real reloc_type;
3871 #else
3872 int reloc_type;
3873 #endif
3874 unsigned char *opcode;
3875 int old_fr_fix;
3877 if (fragP->fr_var != NO_RELOC)
3878 reloc_type = fragP->fr_var;
3879 else if (size == 2)
3880 reloc_type = BFD_RELOC_16_PCREL;
3881 else
3882 reloc_type = BFD_RELOC_32_PCREL;
3884 old_fr_fix = fragP->fr_fix;
3885 opcode = (unsigned char *) fragP->fr_opcode;
3887 switch (opcode[0])
3889 case JUMP_PC_RELATIVE:
3890 /* Make jmp (0xeb) a dword displacement jump. */
3891 opcode[0] = 0xe9;
3892 fragP->fr_fix += size;
3893 fix_new (fragP, old_fr_fix, size,
3894 fragP->fr_symbol,
3895 fragP->fr_offset, 1,
3896 reloc_type);
3897 break;
3899 default:
3900 /* This changes the byte-displacement jump 0x7N
3901 to the dword-displacement jump 0x0f,0x8N. */
3902 opcode[1] = opcode[0] + 0x10;
3903 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
3904 /* We've added an opcode byte. */
3905 fragP->fr_fix += 1 + size;
3906 fix_new (fragP, old_fr_fix + 1, size,
3907 fragP->fr_symbol,
3908 fragP->fr_offset, 1,
3909 reloc_type);
3910 break;
3912 frag_wane (fragP);
3913 return fragP->fr_fix - old_fr_fix;
3915 /* Guess a short jump. */
3916 return 1;
3919 /* Called after relax() is finished.
3921 In: Address of frag.
3922 fr_type == rs_machine_dependent.
3923 fr_subtype is what the address relaxed to.
3925 Out: Any fixSs and constants are set up.
3926 Caller will turn frag into a ".space 0". */
3928 #ifndef BFD_ASSEMBLER
3929 void
3930 md_convert_frag (headers, sec, fragP)
3931 object_headers *headers ATTRIBUTE_UNUSED;
3932 segT sec ATTRIBUTE_UNUSED;
3933 register fragS *fragP;
3934 #else
3935 void
3936 md_convert_frag (abfd, sec, fragP)
3937 bfd *abfd ATTRIBUTE_UNUSED;
3938 segT sec ATTRIBUTE_UNUSED;
3939 register fragS *fragP;
3940 #endif
3942 register unsigned char *opcode;
3943 unsigned char *where_to_put_displacement = NULL;
3944 offsetT target_address;
3945 offsetT opcode_address;
3946 unsigned int extension = 0;
3947 offsetT displacement_from_opcode_start;
3949 opcode = (unsigned char *) fragP->fr_opcode;
3951 /* Address we want to reach in file space. */
3952 target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
3953 #ifdef BFD_ASSEMBLER
3954 /* Not needed otherwise? */
3955 target_address += symbol_get_frag (fragP->fr_symbol)->fr_address;
3956 #endif
3958 /* Address opcode resides at in file space. */
3959 opcode_address = fragP->fr_address + fragP->fr_fix;
3961 /* Displacement from opcode start to fill into instruction. */
3962 displacement_from_opcode_start = target_address - opcode_address;
3964 switch (fragP->fr_subtype)
3966 case ENCODE_RELAX_STATE (COND_JUMP, SMALL):
3967 case ENCODE_RELAX_STATE (COND_JUMP, SMALL16):
3968 case ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL):
3969 case ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL16):
3970 /* Don't have to change opcode. */
3971 extension = 1; /* 1 opcode + 1 displacement */
3972 where_to_put_displacement = &opcode[1];
3973 break;
3975 case ENCODE_RELAX_STATE (COND_JUMP, BIG):
3976 extension = 5; /* 2 opcode + 4 displacement */
3977 opcode[1] = opcode[0] + 0x10;
3978 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
3979 where_to_put_displacement = &opcode[2];
3980 break;
3982 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
3983 extension = 4; /* 1 opcode + 4 displacement */
3984 opcode[0] = 0xe9;
3985 where_to_put_displacement = &opcode[1];
3986 break;
3988 case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
3989 extension = 3; /* 2 opcode + 2 displacement */
3990 opcode[1] = opcode[0] + 0x10;
3991 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
3992 where_to_put_displacement = &opcode[2];
3993 break;
3995 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
3996 extension = 2; /* 1 opcode + 2 displacement */
3997 opcode[0] = 0xe9;
3998 where_to_put_displacement = &opcode[1];
3999 break;
4001 default:
4002 BAD_CASE (fragP->fr_subtype);
4003 break;
4005 /* Now put displacement after opcode. */
4006 md_number_to_chars ((char *) where_to_put_displacement,
4007 (valueT) (displacement_from_opcode_start - extension),
4008 SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
4009 fragP->fr_fix += extension;
4012 /* Size of byte displacement jmp. */
4013 int md_short_jump_size = 2;
4015 /* Size of dword displacement jmp. */
4016 int md_long_jump_size = 5;
4018 /* Size of relocation record. */
4019 const int md_reloc_size = 8;
4021 void
4022 md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
4023 char *ptr;
4024 addressT from_addr, to_addr;
4025 fragS *frag ATTRIBUTE_UNUSED;
4026 symbolS *to_symbol ATTRIBUTE_UNUSED;
4028 offsetT offset;
4030 offset = to_addr - (from_addr + 2);
4031 /* Opcode for byte-disp jump. */
4032 md_number_to_chars (ptr, (valueT) 0xeb, 1);
4033 md_number_to_chars (ptr + 1, (valueT) offset, 1);
4036 void
4037 md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
4038 char *ptr;
4039 addressT from_addr, to_addr;
4040 fragS *frag ATTRIBUTE_UNUSED;
4041 symbolS *to_symbol ATTRIBUTE_UNUSED;
4043 offsetT offset;
4045 offset = to_addr - (from_addr + 5);
4046 md_number_to_chars (ptr, (valueT) 0xe9, 1);
4047 md_number_to_chars (ptr + 1, (valueT) offset, 4);
4050 /* Apply a fixup (fixS) to segment data, once it has been determined
4051 by our caller that we have all the info we need to fix it up.
4053 On the 386, immediates, displacements, and data pointers are all in
4054 the same (little-endian) format, so we don't need to care about which
4055 we are handling. */
4058 md_apply_fix3 (fixP, valp, seg)
4059 /* The fix we're to put in. */
4060 fixS *fixP;
4062 /* Pointer to the value of the bits. */
4063 valueT *valp;
4065 /* Segment fix is from. */
4066 segT seg ATTRIBUTE_UNUSED;
4068 register char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
4069 valueT value = *valp;
4071 #if defined (BFD_ASSEMBLER) && !defined (TE_Mach)
4072 if (fixP->fx_pcrel)
4074 switch (fixP->fx_r_type)
4076 default:
4077 break;
4079 case BFD_RELOC_32:
4080 fixP->fx_r_type = BFD_RELOC_32_PCREL;
4081 break;
4082 case BFD_RELOC_16:
4083 fixP->fx_r_type = BFD_RELOC_16_PCREL;
4084 break;
4085 case BFD_RELOC_8:
4086 fixP->fx_r_type = BFD_RELOC_8_PCREL;
4087 break;
4091 /* This is a hack. There should be a better way to handle this.
4092 This covers for the fact that bfd_install_relocation will
4093 subtract the current location (for partial_inplace, PC relative
4094 relocations); see more below. */
4095 if ((fixP->fx_r_type == BFD_RELOC_32_PCREL
4096 || fixP->fx_r_type == BFD_RELOC_16_PCREL
4097 || fixP->fx_r_type == BFD_RELOC_8_PCREL)
4098 && fixP->fx_addsy)
4100 #ifndef OBJ_AOUT
4101 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4102 #ifdef TE_PE
4103 || OUTPUT_FLAVOR == bfd_target_coff_flavour
4104 #endif
4106 value += fixP->fx_where + fixP->fx_frag->fr_address;
4107 #endif
4108 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4109 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
4111 segT fseg = S_GET_SEGMENT (fixP->fx_addsy);
4113 if ((fseg == seg
4114 || (symbol_section_p (fixP->fx_addsy)
4115 && fseg != absolute_section))
4116 && ! S_IS_EXTERNAL (fixP->fx_addsy)
4117 && ! S_IS_WEAK (fixP->fx_addsy)
4118 && S_IS_DEFINED (fixP->fx_addsy)
4119 && ! S_IS_COMMON (fixP->fx_addsy))
4121 /* Yes, we add the values in twice. This is because
4122 bfd_perform_relocation subtracts them out again. I think
4123 bfd_perform_relocation is broken, but I don't dare change
4124 it. FIXME. */
4125 value += fixP->fx_where + fixP->fx_frag->fr_address;
4128 #endif
4129 #if defined (OBJ_COFF) && defined (TE_PE)
4130 /* For some reason, the PE format does not store a section
4131 address offset for a PC relative symbol. */
4132 if (S_GET_SEGMENT (fixP->fx_addsy) != seg)
4133 value += md_pcrel_from (fixP);
4134 #endif
4137 /* Fix a few things - the dynamic linker expects certain values here,
4138 and we must not dissappoint it. */
4139 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4140 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4141 && fixP->fx_addsy)
4142 switch (fixP->fx_r_type)
4144 case BFD_RELOC_386_PLT32:
4145 case BFD_RELOC_X86_64_PLT32:
4146 /* Make the jump instruction point to the address of the operand. At
4147 runtime we merely add the offset to the actual PLT entry. */
4148 value = -4;
4149 break;
4150 case BFD_RELOC_386_GOTPC:
4152 /* This is tough to explain. We end up with this one if we have
4153 * operands that look like "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal
4154 * here is to obtain the absolute address of the GOT, and it is strongly
4155 * preferable from a performance point of view to avoid using a runtime
4156 * relocation for this. The actual sequence of instructions often look
4157 * something like:
4159 * call .L66
4160 * .L66:
4161 * popl %ebx
4162 * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
4164 * The call and pop essentially return the absolute address of
4165 * the label .L66 and store it in %ebx. The linker itself will
4166 * ultimately change the first operand of the addl so that %ebx points to
4167 * the GOT, but to keep things simple, the .o file must have this operand
4168 * set so that it generates not the absolute address of .L66, but the
4169 * absolute address of itself. This allows the linker itself simply
4170 * treat a GOTPC relocation as asking for a pcrel offset to the GOT to be
4171 * added in, and the addend of the relocation is stored in the operand
4172 * field for the instruction itself.
4174 * Our job here is to fix the operand so that it would add the correct
4175 * offset so that %ebx would point to itself. The thing that is tricky is
4176 * that .-.L66 will point to the beginning of the instruction, so we need
4177 * to further modify the operand so that it will point to itself.
4178 * There are other cases where you have something like:
4180 * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
4182 * and here no correction would be required. Internally in the assembler
4183 * we treat operands of this form as not being pcrel since the '.' is
4184 * explicitly mentioned, and I wonder whether it would simplify matters
4185 * to do it this way. Who knows. In earlier versions of the PIC patches,
4186 * the pcrel_adjust field was used to store the correction, but since the
4187 * expression is not pcrel, I felt it would be confusing to do it this
4188 * way. */
4190 value -= 1;
4191 break;
4192 case BFD_RELOC_386_GOT32:
4193 case BFD_RELOC_X86_64_GOT32:
4194 value = 0; /* Fully resolved at runtime. No addend. */
4195 break;
4196 case BFD_RELOC_386_GOTOFF:
4197 case BFD_RELOC_X86_64_GOTPCREL:
4198 break;
4200 case BFD_RELOC_VTABLE_INHERIT:
4201 case BFD_RELOC_VTABLE_ENTRY:
4202 fixP->fx_done = 0;
4203 return 1;
4205 default:
4206 break;
4208 #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
4209 *valp = value;
4210 #endif /* defined (BFD_ASSEMBLER) && !defined (TE_Mach) */
4212 #ifndef BFD_ASSEMBLER
4213 md_number_to_chars (p, value, fixP->fx_size);
4214 #else
4215 /* Are we finished with this relocation now? */
4216 if (fixP->fx_addsy == 0 && fixP->fx_pcrel == 0)
4217 fixP->fx_done = 1;
4218 else if (use_rela_relocations)
4220 fixP->fx_no_overflow = 1;
4221 value = 0;
4223 md_number_to_chars (p, value, fixP->fx_size);
4224 #endif
4226 return 1;
4229 #define MAX_LITTLENUMS 6
4231 /* Turn the string pointed to by litP into a floating point constant
4232 of type TYPE, and emit the appropriate bytes. The number of
4233 LITTLENUMS emitted is stored in *SIZEP. An error message is
4234 returned, or NULL on OK. */
4236 char *
4237 md_atof (type, litP, sizeP)
4238 int type;
4239 char *litP;
4240 int *sizeP;
4242 int prec;
4243 LITTLENUM_TYPE words[MAX_LITTLENUMS];
4244 LITTLENUM_TYPE *wordP;
4245 char *t;
4247 switch (type)
4249 case 'f':
4250 case 'F':
4251 prec = 2;
4252 break;
4254 case 'd':
4255 case 'D':
4256 prec = 4;
4257 break;
4259 case 'x':
4260 case 'X':
4261 prec = 5;
4262 break;
4264 default:
4265 *sizeP = 0;
4266 return _("Bad call to md_atof ()");
4268 t = atof_ieee (input_line_pointer, type, words);
4269 if (t)
4270 input_line_pointer = t;
4272 *sizeP = prec * sizeof (LITTLENUM_TYPE);
4273 /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
4274 the bigendian 386. */
4275 for (wordP = words + prec - 1; prec--;)
4277 md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
4278 litP += sizeof (LITTLENUM_TYPE);
4280 return 0;
4283 char output_invalid_buf[8];
4285 static char *
4286 output_invalid (c)
4287 int c;
4289 if (isprint (c))
4290 sprintf (output_invalid_buf, "'%c'", c);
4291 else
4292 sprintf (output_invalid_buf, "(0x%x)", (unsigned) c);
4293 return output_invalid_buf;
4296 /* REG_STRING starts *before* REGISTER_PREFIX. */
4298 static const reg_entry *
4299 parse_register (reg_string, end_op)
4300 char *reg_string;
4301 char **end_op;
4303 char *s = reg_string;
4304 char *p;
4305 char reg_name_given[MAX_REG_NAME_SIZE + 1];
4306 const reg_entry *r;
4308 /* Skip possible REGISTER_PREFIX and possible whitespace. */
4309 if (*s == REGISTER_PREFIX)
4310 ++s;
4312 if (is_space_char (*s))
4313 ++s;
4315 p = reg_name_given;
4316 while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
4318 if (p >= reg_name_given + MAX_REG_NAME_SIZE)
4319 return (const reg_entry *) NULL;
4320 s++;
4323 /* For naked regs, make sure that we are not dealing with an identifier.
4324 This prevents confusing an identifier like `eax_var' with register
4325 `eax'. */
4326 if (allow_naked_reg && identifier_chars[(unsigned char) *s])
4327 return (const reg_entry *) NULL;
4329 *end_op = s;
4331 r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
4333 /* Handle floating point regs, allowing spaces in the (i) part. */
4334 if (r == i386_regtab /* %st is first entry of table */)
4336 if (is_space_char (*s))
4337 ++s;
4338 if (*s == '(')
4340 ++s;
4341 if (is_space_char (*s))
4342 ++s;
4343 if (*s >= '0' && *s <= '7')
4345 r = &i386_float_regtab[*s - '0'];
4346 ++s;
4347 if (is_space_char (*s))
4348 ++s;
4349 if (*s == ')')
4351 *end_op = s + 1;
4352 return r;
4355 /* We have "%st(" then garbage. */
4356 return (const reg_entry *) NULL;
4360 return r;
4363 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4364 const char *md_shortopts = "kVQ:sq";
4365 #else
4366 const char *md_shortopts = "q";
4367 #endif
4368 struct option md_longopts[] = {
4369 #define OPTION_32 (OPTION_MD_BASE + 0)
4370 {"32", no_argument, NULL, OPTION_32},
4371 #define OPTION_64 (OPTION_MD_BASE + 1)
4372 {"64", no_argument, NULL, OPTION_64},
4373 {NULL, no_argument, NULL, 0}
4375 size_t md_longopts_size = sizeof (md_longopts);
4378 md_parse_option (c, arg)
4379 int c;
4380 char *arg ATTRIBUTE_UNUSED;
4382 switch (c)
4384 case 'q':
4385 quiet_warnings = 1;
4386 break;
4388 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4389 /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
4390 should be emitted or not. FIXME: Not implemented. */
4391 case 'Q':
4392 break;
4394 /* -V: SVR4 argument to print version ID. */
4395 case 'V':
4396 print_version_id ();
4397 break;
4399 /* -k: Ignore for FreeBSD compatibility. */
4400 case 'k':
4401 break;
4403 case 's':
4404 /* -s: On i386 Solaris, this tells the native assembler to use
4405 .stab instead of .stab.excl. We always use .stab anyhow. */
4406 break;
4407 #endif
4408 #ifdef OBJ_ELF
4409 case OPTION_32:
4410 case OPTION_64:
4412 const char **list, **l;
4414 default_arch = c == OPTION_32 ? "i386" : "x86_64";
4415 list = bfd_target_list ();
4416 for (l = list; *l != NULL; l++)
4418 if (c == OPTION_32)
4420 if (strcmp (*l, "elf32-i386") == 0)
4421 break;
4423 else
4425 if (strcmp (*l, "elf64-x86-64") == 0)
4426 break;
4429 if (*l == NULL)
4430 as_fatal (_("No compiled in support for %d bit object file format"),
4431 c == OPTION_32 ? 32 : 64);
4432 free (list);
4434 break;
4435 #endif
4437 default:
4438 return 0;
4440 return 1;
4443 void
4444 md_show_usage (stream)
4445 FILE *stream;
4447 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4448 fprintf (stream, _("\
4449 -Q ignored\n\
4450 -V print assembler version number\n\
4451 -k ignored\n\
4452 -q quieten some warnings\n\
4453 -s ignored\n"));
4454 #else
4455 fprintf (stream, _("\
4456 -q quieten some warnings\n"));
4457 #endif
4460 #ifdef BFD_ASSEMBLER
4461 #if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
4462 || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
4464 /* Pick the target format to use. */
4466 const char *
4467 i386_target_format ()
4469 if (!strcmp (default_arch, "x86_64"))
4470 set_code_flag (CODE_64BIT);
4471 else if (!strcmp (default_arch, "i386"))
4472 set_code_flag (CODE_32BIT);
4473 else
4474 as_fatal (_("Unknown architecture"));
4475 switch (OUTPUT_FLAVOR)
4477 #ifdef OBJ_MAYBE_AOUT
4478 case bfd_target_aout_flavour:
4479 return AOUT_TARGET_FORMAT;
4480 #endif
4481 #ifdef OBJ_MAYBE_COFF
4482 case bfd_target_coff_flavour:
4483 return "coff-i386";
4484 #endif
4485 #if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF)
4486 case bfd_target_elf_flavour:
4488 if (flag_code == CODE_64BIT)
4489 use_rela_relocations = 1;
4490 return flag_code == CODE_64BIT ? "elf64-x86-64" : "elf32-i386";
4492 #endif
4493 default:
4494 abort ();
4495 return NULL;
4499 #endif /* OBJ_MAYBE_ more than one */
4500 #endif /* BFD_ASSEMBLER */
4502 symbolS *
4503 md_undefined_symbol (name)
4504 char *name;
4506 if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
4507 && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
4508 && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
4509 && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
4511 if (!GOT_symbol)
4513 if (symbol_find (name))
4514 as_bad (_("GOT already in symbol table"));
4515 GOT_symbol = symbol_new (name, undefined_section,
4516 (valueT) 0, &zero_address_frag);
4518 return GOT_symbol;
4520 return 0;
4523 /* Round up a section size to the appropriate boundary. */
4525 valueT
4526 md_section_align (segment, size)
4527 segT segment ATTRIBUTE_UNUSED;
4528 valueT size;
4530 #ifdef BFD_ASSEMBLER
4531 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4532 if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
4534 /* For a.out, force the section size to be aligned. If we don't do
4535 this, BFD will align it for us, but it will not write out the
4536 final bytes of the section. This may be a bug in BFD, but it is
4537 easier to fix it here since that is how the other a.out targets
4538 work. */
4539 int align;
4541 align = bfd_get_section_alignment (stdoutput, segment);
4542 size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
4544 #endif
4545 #endif
4547 return size;
4550 /* On the i386, PC-relative offsets are relative to the start of the
4551 next instruction. That is, the address of the offset, plus its
4552 size, since the offset is always the last part of the insn. */
4554 long
4555 md_pcrel_from (fixP)
4556 fixS *fixP;
4558 return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
4561 #ifndef I386COFF
4563 static void
4564 s_bss (ignore)
4565 int ignore ATTRIBUTE_UNUSED;
4567 register int temp;
4569 temp = get_absolute_expression ();
4570 subseg_set (bss_section, (subsegT) temp);
4571 demand_empty_rest_of_line ();
4574 #endif
4576 #ifdef BFD_ASSEMBLER
4578 void
4579 i386_validate_fix (fixp)
4580 fixS *fixp;
4582 if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
4584 /* GOTOFF relocation are nonsense in 64bit mode. */
4585 if (flag_code == CODE_64BIT)
4586 abort();
4587 fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
4588 fixp->fx_subsy = 0;
4592 arelent *
4593 tc_gen_reloc (section, fixp)
4594 asection *section ATTRIBUTE_UNUSED;
4595 fixS *fixp;
4597 arelent *rel;
4598 bfd_reloc_code_real_type code;
4600 switch (fixp->fx_r_type)
4602 case BFD_RELOC_X86_64_PLT32:
4603 case BFD_RELOC_X86_64_GOT32:
4604 case BFD_RELOC_X86_64_GOTPCREL:
4605 case BFD_RELOC_386_PLT32:
4606 case BFD_RELOC_386_GOT32:
4607 case BFD_RELOC_386_GOTOFF:
4608 case BFD_RELOC_386_GOTPC:
4609 case BFD_RELOC_X86_64_32S:
4610 case BFD_RELOC_RVA:
4611 case BFD_RELOC_VTABLE_ENTRY:
4612 case BFD_RELOC_VTABLE_INHERIT:
4613 code = fixp->fx_r_type;
4614 break;
4615 default:
4616 if (fixp->fx_pcrel)
4618 switch (fixp->fx_size)
4620 default:
4621 as_bad (_("can not do %d byte pc-relative relocation"),
4622 fixp->fx_size);
4623 code = BFD_RELOC_32_PCREL;
4624 break;
4625 case 1: code = BFD_RELOC_8_PCREL; break;
4626 case 2: code = BFD_RELOC_16_PCREL; break;
4627 case 4: code = BFD_RELOC_32_PCREL; break;
4630 else
4632 switch (fixp->fx_size)
4634 default:
4635 as_bad (_("can not do %d byte relocation"), fixp->fx_size);
4636 code = BFD_RELOC_32;
4637 break;
4638 case 1: code = BFD_RELOC_8; break;
4639 case 2: code = BFD_RELOC_16; break;
4640 case 4: code = BFD_RELOC_32; break;
4641 case 8: code = BFD_RELOC_64; break;
4644 break;
4647 if (code == BFD_RELOC_32
4648 && GOT_symbol
4649 && fixp->fx_addsy == GOT_symbol)
4651 /* We don't support GOTPC on 64bit targets. */
4652 if (flag_code == CODE_64BIT)
4653 abort();
4654 code = BFD_RELOC_386_GOTPC;
4657 rel = (arelent *) xmalloc (sizeof (arelent));
4658 rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
4659 *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
4661 rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
4662 if (!use_rela_relocations)
4664 /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
4665 vtable entry to be used in the relocation's section offset. */
4666 if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
4667 rel->address = fixp->fx_offset;
4669 if (fixp->fx_pcrel)
4670 rel->addend = fixp->fx_addnumber;
4671 else
4672 rel->addend = 0;
4674 /* Use the rela in 64bit mode. */
4675 else
4677 rel->addend = fixp->fx_offset;
4678 #ifdef OBJ_ELF
4679 /* Ohhh, this is ugly. The problem is that if this is a local global
4680 symbol, the relocation will entirely be performed at link time, not
4681 at assembly time. bfd_perform_reloc doesn't know about this sort
4682 of thing, and as a result we need to fake it out here. */
4683 if ((S_IS_EXTERN (fixp->fx_addsy) || S_IS_WEAK (fixp->fx_addsy))
4684 && !S_IS_COMMON(fixp->fx_addsy))
4685 rel->addend -= symbol_get_bfdsym (fixp->fx_addsy)->value;
4686 #endif
4687 if (fixp->fx_pcrel)
4688 rel->addend -= fixp->fx_size;
4692 rel->howto = bfd_reloc_type_lookup (stdoutput, code);
4693 if (rel->howto == NULL)
4695 as_bad_where (fixp->fx_file, fixp->fx_line,
4696 _("cannot represent relocation type %s"),
4697 bfd_get_reloc_code_name (code));
4698 /* Set howto to a garbage value so that we can keep going. */
4699 rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
4700 assert (rel->howto != NULL);
4703 return rel;
4706 #else /* ! BFD_ASSEMBLER */
4708 #if (defined(OBJ_AOUT) | defined(OBJ_BOUT))
4709 void
4710 tc_aout_fix_to_chars (where, fixP, segment_address_in_file)
4711 char *where;
4712 fixS *fixP;
4713 relax_addressT segment_address_in_file;
4715 /* In: length of relocation (or of address) in chars: 1, 2 or 4.
4716 Out: GNU LD relocation length code: 0, 1, or 2. */
4718 static const unsigned char nbytes_r_length[] = { 42, 0, 1, 42, 2 };
4719 long r_symbolnum;
4721 know (fixP->fx_addsy != NULL);
4723 md_number_to_chars (where,
4724 (valueT) (fixP->fx_frag->fr_address
4725 + fixP->fx_where - segment_address_in_file),
4728 r_symbolnum = (S_IS_DEFINED (fixP->fx_addsy)
4729 ? S_GET_TYPE (fixP->fx_addsy)
4730 : fixP->fx_addsy->sy_number);
4732 where[6] = (r_symbolnum >> 16) & 0x0ff;
4733 where[5] = (r_symbolnum >> 8) & 0x0ff;
4734 where[4] = r_symbolnum & 0x0ff;
4735 where[7] = ((((!S_IS_DEFINED (fixP->fx_addsy)) << 3) & 0x08)
4736 | ((nbytes_r_length[fixP->fx_size] << 1) & 0x06)
4737 | (((fixP->fx_pcrel << 0) & 0x01) & 0x0f));
4740 #endif /* OBJ_AOUT or OBJ_BOUT. */
4742 #if defined (I386COFF)
4744 short
4745 tc_coff_fix2rtype (fixP)
4746 fixS *fixP;
4748 if (fixP->fx_r_type == R_IMAGEBASE)
4749 return R_IMAGEBASE;
4751 return (fixP->fx_pcrel ?
4752 (fixP->fx_size == 1 ? R_PCRBYTE :
4753 fixP->fx_size == 2 ? R_PCRWORD :
4754 R_PCRLONG) :
4755 (fixP->fx_size == 1 ? R_RELBYTE :
4756 fixP->fx_size == 2 ? R_RELWORD :
4757 R_DIR32));
4761 tc_coff_sizemachdep (frag)
4762 fragS *frag;
4764 if (frag->fr_next)
4765 return (frag->fr_next->fr_address - frag->fr_address);
4766 else
4767 return 0;
4770 #endif /* I386COFF */
4772 #endif /* ! BFD_ASSEMBLER */
4774 /* Parse operands using Intel syntax. This implements a recursive descent
4775 parser based on the BNF grammar published in Appendix B of the MASM 6.1
4776 Programmer's Guide.
4778 FIXME: We do not recognize the full operand grammar defined in the MASM
4779 documentation. In particular, all the structure/union and
4780 high-level macro operands are missing.
4782 Uppercase words are terminals, lower case words are non-terminals.
4783 Objects surrounded by double brackets '[[' ']]' are optional. Vertical
4784 bars '|' denote choices. Most grammar productions are implemented in
4785 functions called 'intel_<production>'.
4787 Initial production is 'expr'.
4789 addOp + | -
4791 alpha [a-zA-Z]
4793 byteRegister AL | AH | BL | BH | CL | CH | DL | DH
4795 constant digits [[ radixOverride ]]
4797 dataType BYTE | WORD | DWORD | QWORD | XWORD
4799 digits decdigit
4800 | digits decdigit
4801 | digits hexdigit
4803 decdigit [0-9]
4805 e05 e05 addOp e06
4806 | e06
4808 e06 e06 mulOp e09
4809 | e09
4811 e09 OFFSET e10
4812 | e09 PTR e10
4813 | e09 : e10
4814 | e10
4816 e10 e10 [ expr ]
4817 | e11
4819 e11 ( expr )
4820 | [ expr ]
4821 | constant
4822 | dataType
4823 | id
4825 | register
4827 => expr SHORT e05
4828 | e05
4830 gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
4831 | BP | EBP | SP | ESP | DI | EDI | SI | ESI
4833 hexdigit a | b | c | d | e | f
4834 | A | B | C | D | E | F
4836 id alpha
4837 | id alpha
4838 | id decdigit
4840 mulOp * | / | MOD
4842 quote " | '
4844 register specialRegister
4845 | gpRegister
4846 | byteRegister
4848 segmentRegister CS | DS | ES | FS | GS | SS
4850 specialRegister CR0 | CR2 | CR3
4851 | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
4852 | TR3 | TR4 | TR5 | TR6 | TR7
4854 We simplify the grammar in obvious places (e.g., register parsing is
4855 done by calling parse_register) and eliminate immediate left recursion
4856 to implement a recursive-descent parser.
4858 expr SHORT e05
4859 | e05
4861 e05 e06 e05'
4863 e05' addOp e06 e05'
4864 | Empty
4866 e06 e09 e06'
4868 e06' mulOp e09 e06'
4869 | Empty
4871 e09 OFFSET e10 e09'
4872 | e10 e09'
4874 e09' PTR e10 e09'
4875 | : e10 e09'
4876 | Empty
4878 e10 e11 e10'
4880 e10' [ expr ] e10'
4881 | Empty
4883 e11 ( expr )
4884 | [ expr ]
4885 | BYTE
4886 | WORD
4887 | DWORD
4888 | QWORD
4889 | XWORD
4892 | register
4893 | id
4894 | constant */
4896 /* Parsing structure for the intel syntax parser. Used to implement the
4897 semantic actions for the operand grammar. */
4898 struct intel_parser_s
4900 char *op_string; /* The string being parsed. */
4901 int got_a_float; /* Whether the operand is a float. */
4902 int op_modifier; /* Operand modifier. */
4903 int is_mem; /* 1 if operand is memory reference. */
4904 const reg_entry *reg; /* Last register reference found. */
4905 char *disp; /* Displacement string being built. */
4908 static struct intel_parser_s intel_parser;
4910 /* Token structure for parsing intel syntax. */
4911 struct intel_token
4913 int code; /* Token code. */
4914 const reg_entry *reg; /* Register entry for register tokens. */
4915 char *str; /* String representation. */
4918 static struct intel_token cur_token, prev_token;
4921 /* Token codes for the intel parser. Since T_SHORT is already used
4922 by COFF, undefine it first to prevent a warning. */
4923 #define T_NIL -1
4924 #define T_CONST 1
4925 #define T_REG 2
4926 #define T_BYTE 3
4927 #define T_WORD 4
4928 #define T_DWORD 5
4929 #define T_QWORD 6
4930 #define T_XWORD 7
4931 #undef T_SHORT
4932 #define T_SHORT 8
4933 #define T_OFFSET 9
4934 #define T_PTR 10
4935 #define T_ID 11
4937 /* Prototypes for intel parser functions. */
4938 static int intel_match_token PARAMS ((int code));
4939 static void intel_get_token PARAMS ((void));
4940 static void intel_putback_token PARAMS ((void));
4941 static int intel_expr PARAMS ((void));
4942 static int intel_e05 PARAMS ((void));
4943 static int intel_e05_1 PARAMS ((void));
4944 static int intel_e06 PARAMS ((void));
4945 static int intel_e06_1 PARAMS ((void));
4946 static int intel_e09 PARAMS ((void));
4947 static int intel_e09_1 PARAMS ((void));
4948 static int intel_e10 PARAMS ((void));
4949 static int intel_e10_1 PARAMS ((void));
4950 static int intel_e11 PARAMS ((void));
4952 static int
4953 i386_intel_operand (operand_string, got_a_float)
4954 char *operand_string;
4955 int got_a_float;
4957 int ret;
4958 char *p;
4960 /* Initialize token holders. */
4961 cur_token.code = prev_token.code = T_NIL;
4962 cur_token.reg = prev_token.reg = NULL;
4963 cur_token.str = prev_token.str = NULL;
4965 /* Initialize parser structure. */
4966 p = intel_parser.op_string = (char *)malloc (strlen (operand_string) + 1);
4967 if (p == NULL)
4968 abort ();
4969 strcpy (intel_parser.op_string, operand_string);
4970 intel_parser.got_a_float = got_a_float;
4971 intel_parser.op_modifier = -1;
4972 intel_parser.is_mem = 0;
4973 intel_parser.reg = NULL;
4974 intel_parser.disp = (char *)malloc (strlen (operand_string) + 1);
4975 if (intel_parser.disp == NULL)
4976 abort ();
4977 intel_parser.disp[0] = '\0';
4979 /* Read the first token and start the parser. */
4980 intel_get_token ();
4981 ret = intel_expr ();
4983 if (ret)
4985 /* If we found a memory reference, hand it over to i386_displacement
4986 to fill in the rest of the operand fields. */
4987 if (intel_parser.is_mem)
4989 if ((i.mem_operands == 1
4990 && (current_templates->start->opcode_modifier & IsString) == 0)
4991 || i.mem_operands == 2)
4993 as_bad (_("too many memory references for '%s'"),
4994 current_templates->start->name);
4995 ret = 0;
4997 else
4999 char *s = intel_parser.disp;
5000 i.mem_operands++;
5002 /* Add the displacement expression. */
5003 if (*s != '\0')
5004 ret = i386_displacement (s, s + strlen (s))
5005 && i386_index_check (s);
5009 /* Constant and OFFSET expressions are handled by i386_immediate. */
5010 else if (intel_parser.op_modifier == OFFSET_FLAT
5011 || intel_parser.reg == NULL)
5012 ret = i386_immediate (intel_parser.disp);
5015 free (p);
5016 free (intel_parser.disp);
5018 return ret;
5021 /* expr SHORT e05
5022 | e05 */
5023 static int
5024 intel_expr ()
5026 /* expr SHORT e05 */
5027 if (cur_token.code == T_SHORT)
5029 intel_parser.op_modifier = SHORT;
5030 intel_match_token (T_SHORT);
5032 return (intel_e05 ());
5035 /* expr e05 */
5036 else
5037 return intel_e05 ();
5040 /* e05 e06 e05'
5042 e05' addOp e06 e05'
5043 | Empty */
5044 static int
5045 intel_e05 ()
5047 return (intel_e06 () && intel_e05_1 ());
5050 static int
5051 intel_e05_1 ()
5053 /* e05' addOp e06 e05' */
5054 if (cur_token.code == '+' || cur_token.code == '-')
5056 strcat (intel_parser.disp, cur_token.str);
5057 intel_match_token (cur_token.code);
5059 return (intel_e06 () && intel_e05_1 ());
5062 /* e05' Empty */
5063 else
5064 return 1;
5067 /* e06 e09 e06'
5069 e06' mulOp e09 e06'
5070 | Empty */
5071 static int
5072 intel_e06 ()
5074 return (intel_e09 () && intel_e06_1 ());
5077 static int
5078 intel_e06_1 ()
5080 /* e06' mulOp e09 e06' */
5081 if (cur_token.code == '*' || cur_token.code == '/')
5083 strcat (intel_parser.disp, cur_token.str);
5084 intel_match_token (cur_token.code);
5086 return (intel_e09 () && intel_e06_1 ());
5089 /* e06' Empty */
5090 else
5091 return 1;
5094 /* e09 OFFSET e10 e09'
5095 | e10 e09'
5097 e09' PTR e10 e09'
5098 | : e10 e09'
5099 | Empty */
5100 static int
5101 intel_e09 ()
5103 /* e09 OFFSET e10 e09' */
5104 if (cur_token.code == T_OFFSET)
5106 intel_parser.is_mem = 0;
5107 intel_parser.op_modifier = OFFSET_FLAT;
5108 intel_match_token (T_OFFSET);
5110 return (intel_e10 () && intel_e09_1 ());
5113 /* e09 e10 e09' */
5114 else
5115 return (intel_e10 () && intel_e09_1 ());
5118 static int
5119 intel_e09_1 ()
5121 /* e09' PTR e10 e09' */
5122 if (cur_token.code == T_PTR)
5124 if (prev_token.code == T_BYTE)
5125 i.suffix = BYTE_MNEM_SUFFIX;
5127 else if (prev_token.code == T_WORD)
5129 if (intel_parser.got_a_float == 2) /* "fi..." */
5130 i.suffix = SHORT_MNEM_SUFFIX;
5131 else
5132 i.suffix = WORD_MNEM_SUFFIX;
5135 else if (prev_token.code == T_DWORD)
5137 if (intel_parser.got_a_float == 1) /* "f..." */
5138 i.suffix = SHORT_MNEM_SUFFIX;
5139 else
5140 i.suffix = LONG_MNEM_SUFFIX;
5143 else if (prev_token.code == T_QWORD)
5145 if (intel_parser.got_a_float == 1) /* "f..." */
5146 i.suffix = LONG_MNEM_SUFFIX;
5147 else
5148 i.suffix = QWORD_MNEM_SUFFIX;
5151 else if (prev_token.code == T_XWORD)
5152 i.suffix = LONG_DOUBLE_MNEM_SUFFIX;
5154 else
5156 as_bad (_("Unknown operand modifier `%s'\n"), prev_token.str);
5157 return 0;
5160 intel_match_token (T_PTR);
5162 return (intel_e10 () && intel_e09_1 ());
5165 /* e09 : e10 e09' */
5166 else if (cur_token.code == ':')
5168 /* Mark as a memory operand only if it's not already known to be an
5169 offset expression. */
5170 if (intel_parser.op_modifier != OFFSET_FLAT)
5171 intel_parser.is_mem = 1;
5173 return (intel_match_token (':') && intel_e10 () && intel_e09_1 ());
5176 /* e09' Empty */
5177 else
5178 return 1;
5181 /* e10 e11 e10'
5183 e10' [ expr ] e10'
5184 | Empty */
5185 static int
5186 intel_e10 ()
5188 return (intel_e11 () && intel_e10_1 ());
5191 static int
5192 intel_e10_1 ()
5194 /* e10' [ expr ] e10' */
5195 if (cur_token.code == '[')
5197 intel_match_token ('[');
5199 /* Mark as a memory operand only if it's not already known to be an
5200 offset expression. If it's an offset expression, we need to keep
5201 the brace in. */
5202 if (intel_parser.op_modifier != OFFSET_FLAT)
5203 intel_parser.is_mem = 1;
5204 else
5205 strcat (intel_parser.disp, "[");
5207 /* Add a '+' to the displacement string if necessary. */
5208 if (*intel_parser.disp != '\0'
5209 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5210 strcat (intel_parser.disp, "+");
5212 if (intel_expr () && intel_match_token (']'))
5214 /* Preserve brackets when the operand is an offset expression. */
5215 if (intel_parser.op_modifier == OFFSET_FLAT)
5216 strcat (intel_parser.disp, "]");
5218 return intel_e10_1 ();
5220 else
5221 return 0;
5224 /* e10' Empty */
5225 else
5226 return 1;
5229 /* e11 ( expr )
5230 | [ expr ]
5231 | BYTE
5232 | WORD
5233 | DWORD
5234 | QWORD
5235 | XWORD
5238 | register
5239 | id
5240 | constant */
5241 static int
5242 intel_e11 ()
5244 /* e11 ( expr ) */
5245 if (cur_token.code == '(')
5247 intel_match_token ('(');
5248 strcat (intel_parser.disp, "(");
5250 if (intel_expr () && intel_match_token (')'))
5252 strcat (intel_parser.disp, ")");
5253 return 1;
5255 else
5256 return 0;
5259 /* e11 [ expr ] */
5260 else if (cur_token.code == '[')
5262 intel_match_token ('[');
5264 /* Mark as a memory operand only if it's not already known to be an
5265 offset expression. If it's an offset expression, we need to keep
5266 the brace in. */
5267 if (intel_parser.op_modifier != OFFSET_FLAT)
5268 intel_parser.is_mem = 1;
5269 else
5270 strcat (intel_parser.disp, "[");
5272 /* Operands for jump/call inside brackets denote absolute addresses. */
5273 if (current_templates->start->opcode_modifier & Jump
5274 || current_templates->start->opcode_modifier & JumpDword
5275 || current_templates->start->opcode_modifier & JumpByte
5276 || current_templates->start->opcode_modifier & JumpInterSegment)
5277 i.types[this_operand] |= JumpAbsolute;
5279 /* Add a '+' to the displacement string if necessary. */
5280 if (*intel_parser.disp != '\0'
5281 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5282 strcat (intel_parser.disp, "+");
5284 if (intel_expr () && intel_match_token (']'))
5286 /* Preserve brackets when the operand is an offset expression. */
5287 if (intel_parser.op_modifier == OFFSET_FLAT)
5288 strcat (intel_parser.disp, "]");
5290 return 1;
5292 else
5293 return 0;
5296 /* e11 BYTE
5297 | WORD
5298 | DWORD
5299 | QWORD
5300 | XWORD */
5301 else if (cur_token.code == T_BYTE
5302 || cur_token.code == T_WORD
5303 || cur_token.code == T_DWORD
5304 || cur_token.code == T_QWORD
5305 || cur_token.code == T_XWORD)
5307 intel_match_token (cur_token.code);
5309 return 1;
5312 /* e11 $
5313 | . */
5314 else if (cur_token.code == '$' || cur_token.code == '.')
5316 strcat (intel_parser.disp, cur_token.str);
5317 intel_match_token (cur_token.code);
5319 /* Mark as a memory operand only if it's not already known to be an
5320 offset expression. */
5321 if (intel_parser.op_modifier != OFFSET_FLAT)
5322 intel_parser.is_mem = 1;
5324 return 1;
5327 /* e11 register */
5328 else if (cur_token.code == T_REG)
5330 const reg_entry *reg = intel_parser.reg = cur_token.reg;
5332 intel_match_token (T_REG);
5334 /* Check for segment change. */
5335 if (cur_token.code == ':')
5337 if (reg->reg_type & (SReg2 | SReg3))
5339 switch (reg->reg_num)
5341 case 0:
5342 i.seg[i.mem_operands] = &es;
5343 break;
5344 case 1:
5345 i.seg[i.mem_operands] = &cs;
5346 break;
5347 case 2:
5348 i.seg[i.mem_operands] = &ss;
5349 break;
5350 case 3:
5351 i.seg[i.mem_operands] = &ds;
5352 break;
5353 case 4:
5354 i.seg[i.mem_operands] = &fs;
5355 break;
5356 case 5:
5357 i.seg[i.mem_operands] = &gs;
5358 break;
5361 else
5363 as_bad (_("`%s' is not a valid segment register"), reg->reg_name);
5364 return 0;
5368 /* Not a segment register. Check for register scaling. */
5369 else if (cur_token.code == '*')
5371 if (!intel_parser.is_mem)
5373 as_bad (_("Register scaling only allowed in memory operands."));
5374 return 0;
5377 /* What follows must be a valid scale. */
5378 if (intel_match_token ('*')
5379 && strchr ("01248", *cur_token.str))
5381 i.index_reg = reg;
5382 i.types[this_operand] |= BaseIndex;
5384 /* Set the scale after setting the register (otherwise,
5385 i386_scale will complain) */
5386 i386_scale (cur_token.str);
5387 intel_match_token (T_CONST);
5389 else
5391 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
5392 cur_token.str);
5393 return 0;
5397 /* No scaling. If this is a memory operand, the register is either a
5398 base register (first occurrence) or an index register (second
5399 occurrence). */
5400 else if (intel_parser.is_mem && !(reg->reg_type & (SReg2 | SReg3)))
5402 if (i.base_reg && i.index_reg)
5404 as_bad (_("Too many register references in memory operand.\n"));
5405 return 0;
5408 if (i.base_reg == NULL)
5409 i.base_reg = reg;
5410 else
5411 i.index_reg = reg;
5413 i.types[this_operand] |= BaseIndex;
5416 /* Offset modifier. Add the register to the displacement string to be
5417 parsed as an immediate expression after we're done. */
5418 else if (intel_parser.op_modifier == OFFSET_FLAT)
5419 strcat (intel_parser.disp, reg->reg_name);
5421 /* It's neither base nor index nor offset. */
5422 else
5424 i.types[this_operand] |= reg->reg_type & ~BaseIndex;
5425 i.op[this_operand].regs = reg;
5426 i.reg_operands++;
5429 /* Since registers are not part of the displacement string (except
5430 when we're parsing offset operands), we may need to remove any
5431 preceding '+' from the displacement string. */
5432 if (*intel_parser.disp != '\0'
5433 && intel_parser.op_modifier != OFFSET_FLAT)
5435 char *s = intel_parser.disp;
5436 s += strlen (s) - 1;
5437 if (*s == '+')
5438 *s = '\0';
5441 return 1;
5444 /* e11 id */
5445 else if (cur_token.code == T_ID)
5447 /* Add the identifier to the displacement string. */
5448 strcat (intel_parser.disp, cur_token.str);
5449 intel_match_token (T_ID);
5451 /* The identifier represents a memory reference only if it's not
5452 preceded by an offset modifier. */
5453 if (intel_parser.op_modifier != OFFSET_FLAT)
5454 intel_parser.is_mem = 1;
5456 return 1;
5459 /* e11 constant */
5460 else if (cur_token.code == T_CONST
5461 || cur_token.code == '-'
5462 || cur_token.code == '+')
5464 char *save_str;
5466 /* Allow constants that start with `+' or `-'. */
5467 if (cur_token.code == '-' || cur_token.code == '+')
5469 strcat (intel_parser.disp, cur_token.str);
5470 intel_match_token (cur_token.code);
5471 if (cur_token.code != T_CONST)
5473 as_bad (_("Syntax error. Expecting a constant. Got `%s'.\n"),
5474 cur_token.str);
5475 return 0;
5479 save_str = (char *)malloc (strlen (cur_token.str) + 1);
5480 if (save_str == NULL)
5481 abort ();
5482 strcpy (save_str, cur_token.str);
5484 /* Get the next token to check for register scaling. */
5485 intel_match_token (cur_token.code);
5487 /* Check if this constant is a scaling factor for an index register. */
5488 if (cur_token.code == '*')
5490 if (intel_match_token ('*') && cur_token.code == T_REG)
5492 if (!intel_parser.is_mem)
5494 as_bad (_("Register scaling only allowed in memory operands."));
5495 return 0;
5498 /* The constant is followed by `* reg', so it must be
5499 a valid scale. */
5500 if (strchr ("01248", *save_str))
5502 i.index_reg = cur_token.reg;
5503 i.types[this_operand] |= BaseIndex;
5505 /* Set the scale after setting the register (otherwise,
5506 i386_scale will complain) */
5507 i386_scale (save_str);
5508 intel_match_token (T_REG);
5510 /* Since registers are not part of the displacement
5511 string, we may need to remove any preceding '+' from
5512 the displacement string. */
5513 if (*intel_parser.disp != '\0')
5515 char *s = intel_parser.disp;
5516 s += strlen (s) - 1;
5517 if (*s == '+')
5518 *s = '\0';
5521 free (save_str);
5523 return 1;
5525 else
5526 return 0;
5529 /* The constant was not used for register scaling. Since we have
5530 already consumed the token following `*' we now need to put it
5531 back in the stream. */
5532 else
5533 intel_putback_token ();
5536 /* Add the constant to the displacement string. */
5537 strcat (intel_parser.disp, save_str);
5538 free (save_str);
5540 return 1;
5543 as_bad (_("Unrecognized token '%s'"), cur_token.str);
5544 return 0;
5547 /* Match the given token against cur_token. If they match, read the next
5548 token from the operand string. */
5549 static int
5550 intel_match_token (code)
5551 int code;
5553 if (cur_token.code == code)
5555 intel_get_token ();
5556 return 1;
5558 else
5560 as_bad (_("Unexpected token `%s'\n"), cur_token.str);
5561 return 0;
5565 /* Read a new token from intel_parser.op_string and store it in cur_token. */
5566 static void
5567 intel_get_token ()
5569 char *end_op;
5570 const reg_entry *reg;
5571 struct intel_token new_token;
5573 new_token.code = T_NIL;
5574 new_token.reg = NULL;
5575 new_token.str = NULL;
5577 /* Free the memory allocated to the previous token and move
5578 cur_token to prev_token. */
5579 if (prev_token.str)
5580 free (prev_token.str);
5582 prev_token = cur_token;
5584 /* Skip whitespace. */
5585 while (is_space_char (*intel_parser.op_string))
5586 intel_parser.op_string++;
5588 /* Return an empty token if we find nothing else on the line. */
5589 if (*intel_parser.op_string == '\0')
5591 cur_token = new_token;
5592 return;
5595 /* The new token cannot be larger than the remainder of the operand
5596 string. */
5597 new_token.str = (char *)malloc (strlen (intel_parser.op_string) + 1);
5598 if (new_token.str == NULL)
5599 abort ();
5600 new_token.str[0] = '\0';
5602 if (strchr ("0123456789", *intel_parser.op_string))
5604 char *p = new_token.str;
5605 char *q = intel_parser.op_string;
5606 new_token.code = T_CONST;
5608 /* Allow any kind of identifier char to encompass floating point and
5609 hexadecimal numbers. */
5610 while (is_identifier_char (*q))
5611 *p++ = *q++;
5612 *p = '\0';
5614 /* Recognize special symbol names [0-9][bf]. */
5615 if (strlen (intel_parser.op_string) == 2
5616 && (intel_parser.op_string[1] == 'b'
5617 || intel_parser.op_string[1] == 'f'))
5618 new_token.code = T_ID;
5621 else if (strchr ("+-/*:[]()", *intel_parser.op_string))
5623 new_token.code = *intel_parser.op_string;
5624 new_token.str[0] = *intel_parser.op_string;
5625 new_token.str[1] = '\0';
5628 else if ((*intel_parser.op_string == REGISTER_PREFIX || allow_naked_reg)
5629 && ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL))
5631 new_token.code = T_REG;
5632 new_token.reg = reg;
5634 if (*intel_parser.op_string == REGISTER_PREFIX)
5636 new_token.str[0] = REGISTER_PREFIX;
5637 new_token.str[1] = '\0';
5640 strcat (new_token.str, reg->reg_name);
5643 else if (is_identifier_char (*intel_parser.op_string))
5645 char *p = new_token.str;
5646 char *q = intel_parser.op_string;
5648 /* A '.' or '$' followed by an identifier char is an identifier.
5649 Otherwise, it's operator '.' followed by an expression. */
5650 if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
5652 new_token.code = *q;
5653 new_token.str[0] = *q;
5654 new_token.str[1] = '\0';
5656 else
5658 while (is_identifier_char (*q) || *q == '@')
5659 *p++ = *q++;
5660 *p = '\0';
5662 if (strcasecmp (new_token.str, "BYTE") == 0)
5663 new_token.code = T_BYTE;
5665 else if (strcasecmp (new_token.str, "WORD") == 0)
5666 new_token.code = T_WORD;
5668 else if (strcasecmp (new_token.str, "DWORD") == 0)
5669 new_token.code = T_DWORD;
5671 else if (strcasecmp (new_token.str, "QWORD") == 0)
5672 new_token.code = T_QWORD;
5674 else if (strcasecmp (new_token.str, "XWORD") == 0)
5675 new_token.code = T_XWORD;
5677 else if (strcasecmp (new_token.str, "PTR") == 0)
5678 new_token.code = T_PTR;
5680 else if (strcasecmp (new_token.str, "SHORT") == 0)
5681 new_token.code = T_SHORT;
5683 else if (strcasecmp (new_token.str, "OFFSET") == 0)
5685 new_token.code = T_OFFSET;
5687 /* ??? This is not mentioned in the MASM grammar but gcc
5688 makes use of it with -mintel-syntax. OFFSET may be
5689 followed by FLAT: */
5690 if (strncasecmp (q, " FLAT:", 6) == 0)
5691 strcat (new_token.str, " FLAT:");
5694 /* ??? This is not mentioned in the MASM grammar. */
5695 else if (strcasecmp (new_token.str, "FLAT") == 0)
5696 new_token.code = T_OFFSET;
5698 else
5699 new_token.code = T_ID;
5703 else
5704 as_bad (_("Unrecognized token `%s'\n"), intel_parser.op_string);
5706 intel_parser.op_string += strlen (new_token.str);
5707 cur_token = new_token;
5710 /* Put cur_token back into the token stream and make cur_token point to
5711 prev_token. */
5712 static void
5713 intel_putback_token ()
5715 intel_parser.op_string -= strlen (cur_token.str);
5716 free (cur_token.str);
5717 cur_token = prev_token;
5719 /* Forget prev_token. */
5720 prev_token.code = T_NIL;
5721 prev_token.reg = NULL;
5722 prev_token.str = NULL;