| 1 | /* i386.c -- Assemble code for the Intel 80386 |
| 2 | Copyright 1989, 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, |
| 3 | 2000, 2001, 2002, 2003, 2004, 2005 |
| 4 | Free Software Foundation, Inc. |
| 5 | |
| 6 | This file is part of GAS, the GNU Assembler. |
| 7 | |
| 8 | GAS is free software; you can redistribute it and/or modify |
| 9 | it under the terms of the GNU General Public License as published by |
| 10 | the Free Software Foundation; either version 2, or (at your option) |
| 11 | any later version. |
| 12 | |
| 13 | GAS is distributed in the hope that it will be useful, |
| 14 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 15 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 16 | GNU General Public License for more details. |
| 17 | |
| 18 | You should have received a copy of the GNU General Public License |
| 19 | along with GAS; see the file COPYING. If not, write to the Free |
| 20 | Software Foundation, 51 Franklin Street - Fifth Floor, Boston, MA |
| 21 | 02110-1301, USA. */ |
| 22 | |
| 23 | /* Intel 80386 machine specific gas. |
| 24 | Written by Eliot Dresselhaus (eliot@mgm.mit.edu). |
| 25 | x86_64 support by Jan Hubicka (jh@suse.cz) |
| 26 | VIA PadLock support by Michal Ludvig (mludvig@suse.cz) |
| 27 | Bugs & suggestions are completely welcome. This is free software. |
| 28 | Please help us make it better. */ |
| 29 | |
| 30 | #include "as.h" |
| 31 | #include "safe-ctype.h" |
| 32 | #include "subsegs.h" |
| 33 | #include "dwarf2dbg.h" |
| 34 | #include "dw2gencfi.h" |
| 35 | #include "opcode/i386.h" |
| 36 | #include "elf/x86-64.h" |
| 37 | |
| 38 | #ifndef REGISTER_WARNINGS |
| 39 | #define REGISTER_WARNINGS 1 |
| 40 | #endif |
| 41 | |
| 42 | #ifndef INFER_ADDR_PREFIX |
| 43 | #define INFER_ADDR_PREFIX 1 |
| 44 | #endif |
| 45 | |
| 46 | #ifndef SCALE1_WHEN_NO_INDEX |
| 47 | /* Specifying a scale factor besides 1 when there is no index is |
| 48 | futile. eg. `mov (%ebx,2),%al' does exactly the same as |
| 49 | `mov (%ebx),%al'. To slavishly follow what the programmer |
| 50 | specified, set SCALE1_WHEN_NO_INDEX to 0. */ |
| 51 | #define SCALE1_WHEN_NO_INDEX 1 |
| 52 | #endif |
| 53 | |
| 54 | #ifndef DEFAULT_ARCH |
| 55 | #define DEFAULT_ARCH "i386" |
| 56 | #endif |
| 57 | |
| 58 | #ifndef INLINE |
| 59 | #if __GNUC__ >= 2 |
| 60 | #define INLINE __inline__ |
| 61 | #else |
| 62 | #define INLINE |
| 63 | #endif |
| 64 | #endif |
| 65 | |
| 66 | static INLINE unsigned int mode_from_disp_size PARAMS ((unsigned int)); |
| 67 | static INLINE int fits_in_signed_byte PARAMS ((offsetT)); |
| 68 | static INLINE int fits_in_unsigned_byte PARAMS ((offsetT)); |
| 69 | static INLINE int fits_in_unsigned_word PARAMS ((offsetT)); |
| 70 | static INLINE int fits_in_signed_word PARAMS ((offsetT)); |
| 71 | static INLINE int fits_in_unsigned_long PARAMS ((offsetT)); |
| 72 | static INLINE int fits_in_signed_long PARAMS ((offsetT)); |
| 73 | static int smallest_imm_type PARAMS ((offsetT)); |
| 74 | static offsetT offset_in_range PARAMS ((offsetT, int)); |
| 75 | static int add_prefix PARAMS ((unsigned int)); |
| 76 | static void set_code_flag PARAMS ((int)); |
| 77 | static void set_16bit_gcc_code_flag PARAMS ((int)); |
| 78 | static void set_intel_syntax PARAMS ((int)); |
| 79 | static void set_cpu_arch PARAMS ((int)); |
| 80 | #ifdef TE_PE |
| 81 | static void pe_directive_secrel PARAMS ((int)); |
| 82 | #endif |
| 83 | static char *output_invalid PARAMS ((int c)); |
| 84 | static int i386_operand PARAMS ((char *operand_string)); |
| 85 | static int i386_intel_operand PARAMS ((char *operand_string, int got_a_float)); |
| 86 | static const reg_entry *parse_register PARAMS ((char *reg_string, |
| 87 | char **end_op)); |
| 88 | static char *parse_insn PARAMS ((char *, char *)); |
| 89 | static char *parse_operands PARAMS ((char *, const char *)); |
| 90 | static void swap_operands PARAMS ((void)); |
| 91 | static void optimize_imm PARAMS ((void)); |
| 92 | static void optimize_disp PARAMS ((void)); |
| 93 | static int match_template PARAMS ((void)); |
| 94 | static int check_string PARAMS ((void)); |
| 95 | static int process_suffix PARAMS ((void)); |
| 96 | static int check_byte_reg PARAMS ((void)); |
| 97 | static int check_long_reg PARAMS ((void)); |
| 98 | static int check_qword_reg PARAMS ((void)); |
| 99 | static int check_word_reg PARAMS ((void)); |
| 100 | static int finalize_imm PARAMS ((void)); |
| 101 | static int process_operands PARAMS ((void)); |
| 102 | static const seg_entry *build_modrm_byte PARAMS ((void)); |
| 103 | static void output_insn PARAMS ((void)); |
| 104 | static void output_branch PARAMS ((void)); |
| 105 | static void output_jump PARAMS ((void)); |
| 106 | static void output_interseg_jump PARAMS ((void)); |
| 107 | static void output_imm PARAMS ((fragS *insn_start_frag, |
| 108 | offsetT insn_start_off)); |
| 109 | static void output_disp PARAMS ((fragS *insn_start_frag, |
| 110 | offsetT insn_start_off)); |
| 111 | #ifndef I386COFF |
| 112 | static void s_bss PARAMS ((int)); |
| 113 | #endif |
| 114 | #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) |
| 115 | static void handle_large_common (int small ATTRIBUTE_UNUSED); |
| 116 | #endif |
| 117 | |
| 118 | static const char *default_arch = DEFAULT_ARCH; |
| 119 | |
| 120 | /* 'md_assemble ()' gathers together information and puts it into a |
| 121 | i386_insn. */ |
| 122 | |
| 123 | union i386_op |
| 124 | { |
| 125 | expressionS *disps; |
| 126 | expressionS *imms; |
| 127 | const reg_entry *regs; |
| 128 | }; |
| 129 | |
| 130 | struct _i386_insn |
| 131 | { |
| 132 | /* TM holds the template for the insn were currently assembling. */ |
| 133 | template tm; |
| 134 | |
| 135 | /* SUFFIX holds the instruction mnemonic suffix if given. |
| 136 | (e.g. 'l' for 'movl') */ |
| 137 | char suffix; |
| 138 | |
| 139 | /* OPERANDS gives the number of given operands. */ |
| 140 | unsigned int operands; |
| 141 | |
| 142 | /* REG_OPERANDS, DISP_OPERANDS, MEM_OPERANDS, IMM_OPERANDS give the number |
| 143 | of given register, displacement, memory operands and immediate |
| 144 | operands. */ |
| 145 | unsigned int reg_operands, disp_operands, mem_operands, imm_operands; |
| 146 | |
| 147 | /* TYPES [i] is the type (see above #defines) which tells us how to |
| 148 | use OP[i] for the corresponding operand. */ |
| 149 | unsigned int types[MAX_OPERANDS]; |
| 150 | |
| 151 | /* Displacement expression, immediate expression, or register for each |
| 152 | operand. */ |
| 153 | union i386_op op[MAX_OPERANDS]; |
| 154 | |
| 155 | /* Flags for operands. */ |
| 156 | unsigned int flags[MAX_OPERANDS]; |
| 157 | #define Operand_PCrel 1 |
| 158 | |
| 159 | /* Relocation type for operand */ |
| 160 | enum bfd_reloc_code_real reloc[MAX_OPERANDS]; |
| 161 | |
| 162 | /* BASE_REG, INDEX_REG, and LOG2_SCALE_FACTOR are used to encode |
| 163 | the base index byte below. */ |
| 164 | const reg_entry *base_reg; |
| 165 | const reg_entry *index_reg; |
| 166 | unsigned int log2_scale_factor; |
| 167 | |
| 168 | /* SEG gives the seg_entries of this insn. They are zero unless |
| 169 | explicit segment overrides are given. */ |
| 170 | const seg_entry *seg[2]; |
| 171 | |
| 172 | /* PREFIX holds all the given prefix opcodes (usually null). |
| 173 | PREFIXES is the number of prefix opcodes. */ |
| 174 | unsigned int prefixes; |
| 175 | unsigned char prefix[MAX_PREFIXES]; |
| 176 | |
| 177 | /* RM and SIB are the modrm byte and the sib byte where the |
| 178 | addressing modes of this insn are encoded. */ |
| 179 | |
| 180 | modrm_byte rm; |
| 181 | rex_byte rex; |
| 182 | sib_byte sib; |
| 183 | }; |
| 184 | |
| 185 | typedef struct _i386_insn i386_insn; |
| 186 | |
| 187 | /* List of chars besides those in app.c:symbol_chars that can start an |
| 188 | operand. Used to prevent the scrubber eating vital white-space. */ |
| 189 | const char extra_symbol_chars[] = "*%-([" |
| 190 | #ifdef LEX_AT |
| 191 | "@" |
| 192 | #endif |
| 193 | #ifdef LEX_QM |
| 194 | "?" |
| 195 | #endif |
| 196 | ; |
| 197 | |
| 198 | #if (defined (TE_I386AIX) \ |
| 199 | || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) \ |
| 200 | && !defined (TE_LINUX) \ |
| 201 | && !defined (TE_NETWARE) \ |
| 202 | && !defined (TE_FreeBSD) \ |
| 203 | && !defined (TE_NetBSD))) |
| 204 | /* This array holds the chars that always start a comment. If the |
| 205 | pre-processor is disabled, these aren't very useful. */ |
| 206 | const char comment_chars[] = "#/"; |
| 207 | #define PREFIX_SEPARATOR '\\' |
| 208 | |
| 209 | /* This array holds the chars that only start a comment at the beginning of |
| 210 | a line. If the line seems to have the form '# 123 filename' |
| 211 | .line and .file directives will appear in the pre-processed output. |
| 212 | Note that input_file.c hand checks for '#' at the beginning of the |
| 213 | first line of the input file. This is because the compiler outputs |
| 214 | #NO_APP at the beginning of its output. |
| 215 | Also note that comments started like this one will always work if |
| 216 | '/' isn't otherwise defined. */ |
| 217 | const char line_comment_chars[] = "#"; |
| 218 | |
| 219 | #else |
| 220 | /* Putting '/' here makes it impossible to use the divide operator. |
| 221 | However, we need it for compatibility with SVR4 systems. */ |
| 222 | const char comment_chars[] = "#"; |
| 223 | #define PREFIX_SEPARATOR '/' |
| 224 | |
| 225 | const char line_comment_chars[] = "/#"; |
| 226 | #endif |
| 227 | |
| 228 | const char line_separator_chars[] = ";"; |
| 229 | |
| 230 | /* Chars that can be used to separate mant from exp in floating point |
| 231 | nums. */ |
| 232 | const char EXP_CHARS[] = "eE"; |
| 233 | |
| 234 | /* Chars that mean this number is a floating point constant |
| 235 | As in 0f12.456 |
| 236 | or 0d1.2345e12. */ |
| 237 | const char FLT_CHARS[] = "fFdDxX"; |
| 238 | |
| 239 | /* Tables for lexical analysis. */ |
| 240 | static char mnemonic_chars[256]; |
| 241 | static char register_chars[256]; |
| 242 | static char operand_chars[256]; |
| 243 | static char identifier_chars[256]; |
| 244 | static char digit_chars[256]; |
| 245 | |
| 246 | /* Lexical macros. */ |
| 247 | #define is_mnemonic_char(x) (mnemonic_chars[(unsigned char) x]) |
| 248 | #define is_operand_char(x) (operand_chars[(unsigned char) x]) |
| 249 | #define is_register_char(x) (register_chars[(unsigned char) x]) |
| 250 | #define is_space_char(x) ((x) == ' ') |
| 251 | #define is_identifier_char(x) (identifier_chars[(unsigned char) x]) |
| 252 | #define is_digit_char(x) (digit_chars[(unsigned char) x]) |
| 253 | |
| 254 | /* All non-digit non-letter characters that may occur in an operand. */ |
| 255 | static char operand_special_chars[] = "%$-+(,)*._~/<>|&^!:[@]"; |
| 256 | |
| 257 | /* md_assemble() always leaves the strings it's passed unaltered. To |
| 258 | effect this we maintain a stack of saved characters that we've smashed |
| 259 | with '\0's (indicating end of strings for various sub-fields of the |
| 260 | assembler instruction). */ |
| 261 | static char save_stack[32]; |
| 262 | static char *save_stack_p; |
| 263 | #define END_STRING_AND_SAVE(s) \ |
| 264 | do { *save_stack_p++ = *(s); *(s) = '\0'; } while (0) |
| 265 | #define RESTORE_END_STRING(s) \ |
| 266 | do { *(s) = *--save_stack_p; } while (0) |
| 267 | |
| 268 | /* The instruction we're assembling. */ |
| 269 | static i386_insn i; |
| 270 | |
| 271 | /* Possible templates for current insn. */ |
| 272 | static const templates *current_templates; |
| 273 | |
| 274 | /* Per instruction expressionS buffers: 2 displacements & 2 immediate max. */ |
| 275 | static expressionS disp_expressions[2], im_expressions[2]; |
| 276 | |
| 277 | /* Current operand we are working on. */ |
| 278 | static int this_operand; |
| 279 | |
| 280 | /* We support four different modes. FLAG_CODE variable is used to distinguish |
| 281 | these. */ |
| 282 | |
| 283 | enum flag_code { |
| 284 | CODE_32BIT, |
| 285 | CODE_16BIT, |
| 286 | CODE_64BIT }; |
| 287 | #define NUM_FLAG_CODE ((int) CODE_64BIT + 1) |
| 288 | |
| 289 | static enum flag_code flag_code; |
| 290 | static unsigned int object_64bit; |
| 291 | static int use_rela_relocations = 0; |
| 292 | |
| 293 | /* The names used to print error messages. */ |
| 294 | static const char *flag_code_names[] = |
| 295 | { |
| 296 | "32", |
| 297 | "16", |
| 298 | "64" |
| 299 | }; |
| 300 | |
| 301 | /* 1 for intel syntax, |
| 302 | 0 if att syntax. */ |
| 303 | static int intel_syntax = 0; |
| 304 | |
| 305 | /* 1 if register prefix % not required. */ |
| 306 | static int allow_naked_reg = 0; |
| 307 | |
| 308 | /* Used in 16 bit gcc mode to add an l suffix to call, ret, enter, |
| 309 | leave, push, and pop instructions so that gcc has the same stack |
| 310 | frame as in 32 bit mode. */ |
| 311 | static char stackop_size = '\0'; |
| 312 | |
| 313 | /* Non-zero to optimize code alignment. */ |
| 314 | int optimize_align_code = 1; |
| 315 | |
| 316 | /* Non-zero to quieten some warnings. */ |
| 317 | static int quiet_warnings = 0; |
| 318 | |
| 319 | /* CPU name. */ |
| 320 | static const char *cpu_arch_name = NULL; |
| 321 | static const char *cpu_sub_arch_name = NULL; |
| 322 | |
| 323 | /* CPU feature flags. */ |
| 324 | static unsigned int cpu_arch_flags = CpuUnknownFlags | CpuNo64; |
| 325 | |
| 326 | /* If set, conditional jumps are not automatically promoted to handle |
| 327 | larger than a byte offset. */ |
| 328 | static unsigned int no_cond_jump_promotion = 0; |
| 329 | |
| 330 | /* Pre-defined "_GLOBAL_OFFSET_TABLE_". */ |
| 331 | static symbolS *GOT_symbol; |
| 332 | |
| 333 | /* The dwarf2 return column, adjusted for 32 or 64 bit. */ |
| 334 | unsigned int x86_dwarf2_return_column; |
| 335 | |
| 336 | /* The dwarf2 data alignment, adjusted for 32 or 64 bit. */ |
| 337 | int x86_cie_data_alignment; |
| 338 | |
| 339 | /* Interface to relax_segment. |
| 340 | There are 3 major relax states for 386 jump insns because the |
| 341 | different types of jumps add different sizes to frags when we're |
| 342 | figuring out what sort of jump to choose to reach a given label. */ |
| 343 | |
| 344 | /* Types. */ |
| 345 | #define UNCOND_JUMP 0 |
| 346 | #define COND_JUMP 1 |
| 347 | #define COND_JUMP86 2 |
| 348 | |
| 349 | /* Sizes. */ |
| 350 | #define CODE16 1 |
| 351 | #define SMALL 0 |
| 352 | #define SMALL16 (SMALL | CODE16) |
| 353 | #define BIG 2 |
| 354 | #define BIG16 (BIG | CODE16) |
| 355 | |
| 356 | #ifndef INLINE |
| 357 | #ifdef __GNUC__ |
| 358 | #define INLINE __inline__ |
| 359 | #else |
| 360 | #define INLINE |
| 361 | #endif |
| 362 | #endif |
| 363 | |
| 364 | #define ENCODE_RELAX_STATE(type, size) \ |
| 365 | ((relax_substateT) (((type) << 2) | (size))) |
| 366 | #define TYPE_FROM_RELAX_STATE(s) \ |
| 367 | ((s) >> 2) |
| 368 | #define DISP_SIZE_FROM_RELAX_STATE(s) \ |
| 369 | ((((s) & 3) == BIG ? 4 : (((s) & 3) == BIG16 ? 2 : 1))) |
| 370 | |
| 371 | /* This table is used by relax_frag to promote short jumps to long |
| 372 | ones where necessary. SMALL (short) jumps may be promoted to BIG |
| 373 | (32 bit long) ones, and SMALL16 jumps to BIG16 (16 bit long). We |
| 374 | don't allow a short jump in a 32 bit code segment to be promoted to |
| 375 | a 16 bit offset jump because it's slower (requires data size |
| 376 | prefix), and doesn't work, unless the destination is in the bottom |
| 377 | 64k of the code segment (The top 16 bits of eip are zeroed). */ |
| 378 | |
| 379 | const relax_typeS md_relax_table[] = |
| 380 | { |
| 381 | /* The fields are: |
| 382 | 1) most positive reach of this state, |
| 383 | 2) most negative reach of this state, |
| 384 | 3) how many bytes this mode will have in the variable part of the frag |
| 385 | 4) which index into the table to try if we can't fit into this one. */ |
| 386 | |
| 387 | /* UNCOND_JUMP states. */ |
| 388 | {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG)}, |
| 389 | {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16)}, |
| 390 | /* dword jmp adds 4 bytes to frag: |
| 391 | 0 extra opcode bytes, 4 displacement bytes. */ |
| 392 | {0, 0, 4, 0}, |
| 393 | /* word jmp adds 2 byte2 to frag: |
| 394 | 0 extra opcode bytes, 2 displacement bytes. */ |
| 395 | {0, 0, 2, 0}, |
| 396 | |
| 397 | /* COND_JUMP states. */ |
| 398 | {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG)}, |
| 399 | {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG16)}, |
| 400 | /* dword conditionals adds 5 bytes to frag: |
| 401 | 1 extra opcode byte, 4 displacement bytes. */ |
| 402 | {0, 0, 5, 0}, |
| 403 | /* word conditionals add 3 bytes to frag: |
| 404 | 1 extra opcode byte, 2 displacement bytes. */ |
| 405 | {0, 0, 3, 0}, |
| 406 | |
| 407 | /* COND_JUMP86 states. */ |
| 408 | {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG)}, |
| 409 | {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG16)}, |
| 410 | /* dword conditionals adds 5 bytes to frag: |
| 411 | 1 extra opcode byte, 4 displacement bytes. */ |
| 412 | {0, 0, 5, 0}, |
| 413 | /* word conditionals add 4 bytes to frag: |
| 414 | 1 displacement byte and a 3 byte long branch insn. */ |
| 415 | {0, 0, 4, 0} |
| 416 | }; |
| 417 | |
| 418 | static const arch_entry cpu_arch[] = { |
| 419 | {"i8086", Cpu086 }, |
| 420 | {"i186", Cpu086|Cpu186 }, |
| 421 | {"i286", Cpu086|Cpu186|Cpu286 }, |
| 422 | {"i386", Cpu086|Cpu186|Cpu286|Cpu386 }, |
| 423 | {"i486", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486 }, |
| 424 | {"i586", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586 }, |
| 425 | {"i686", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686 }, |
| 426 | {"pentium", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586 }, |
| 427 | {"pentiumpro",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686 }, |
| 428 | {"pentiumii", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX }, |
| 429 | {"pentiumiii",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuMMX2|CpuSSE }, |
| 430 | {"pentium4", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX|CpuMMX2|CpuSSE|CpuSSE2 }, |
| 431 | {"prescott", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuPNI }, |
| 432 | {"k6", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX }, |
| 433 | {"k6_2", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX|Cpu3dnow }, |
| 434 | {"athlon", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA }, |
| 435 | {"sledgehammer",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuSledgehammer|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2 }, |
| 436 | {"opteron", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuSledgehammer|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2 }, |
| 437 | {".mmx", CpuMMX }, |
| 438 | {".sse", CpuMMX|CpuMMX2|CpuSSE }, |
| 439 | {".sse2", CpuMMX|CpuMMX2|CpuSSE|CpuSSE2 }, |
| 440 | {".sse3", CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3 }, |
| 441 | {".3dnow", CpuMMX|Cpu3dnow }, |
| 442 | {".3dnowa", CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA }, |
| 443 | {".padlock", CpuPadLock }, |
| 444 | {".pacifica", CpuSVME }, |
| 445 | {".svme", CpuSVME }, |
| 446 | {NULL, 0 } |
| 447 | }; |
| 448 | |
| 449 | const pseudo_typeS md_pseudo_table[] = |
| 450 | { |
| 451 | #if !defined(OBJ_AOUT) && !defined(USE_ALIGN_PTWO) |
| 452 | {"align", s_align_bytes, 0}, |
| 453 | #else |
| 454 | {"align", s_align_ptwo, 0}, |
| 455 | #endif |
| 456 | {"arch", set_cpu_arch, 0}, |
| 457 | #ifndef I386COFF |
| 458 | {"bss", s_bss, 0}, |
| 459 | #endif |
| 460 | {"ffloat", float_cons, 'f'}, |
| 461 | {"dfloat", float_cons, 'd'}, |
| 462 | {"tfloat", float_cons, 'x'}, |
| 463 | {"value", cons, 2}, |
| 464 | {"noopt", s_ignore, 0}, |
| 465 | {"optim", s_ignore, 0}, |
| 466 | {"code16gcc", set_16bit_gcc_code_flag, CODE_16BIT}, |
| 467 | {"code16", set_code_flag, CODE_16BIT}, |
| 468 | {"code32", set_code_flag, CODE_32BIT}, |
| 469 | {"code64", set_code_flag, CODE_64BIT}, |
| 470 | {"intel_syntax", set_intel_syntax, 1}, |
| 471 | {"att_syntax", set_intel_syntax, 0}, |
| 472 | {"file", (void (*) PARAMS ((int))) dwarf2_directive_file, 0}, |
| 473 | {"loc", dwarf2_directive_loc, 0}, |
| 474 | #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) |
| 475 | {"largecomm", handle_large_common, 0}, |
| 476 | #endif |
| 477 | #ifdef TE_PE |
| 478 | {"secrel32", pe_directive_secrel, 0}, |
| 479 | #endif |
| 480 | {0, 0, 0} |
| 481 | }; |
| 482 | |
| 483 | /* For interface with expression (). */ |
| 484 | extern char *input_line_pointer; |
| 485 | |
| 486 | /* Hash table for instruction mnemonic lookup. */ |
| 487 | static struct hash_control *op_hash; |
| 488 | |
| 489 | /* Hash table for register lookup. */ |
| 490 | static struct hash_control *reg_hash; |
| 491 | \f |
| 492 | void |
| 493 | i386_align_code (fragP, count) |
| 494 | fragS *fragP; |
| 495 | int count; |
| 496 | { |
| 497 | /* Various efficient no-op patterns for aligning code labels. |
| 498 | Note: Don't try to assemble the instructions in the comments. |
| 499 | 0L and 0w are not legal. */ |
| 500 | static const char f32_1[] = |
| 501 | {0x90}; /* nop */ |
| 502 | static const char f32_2[] = |
| 503 | {0x89,0xf6}; /* movl %esi,%esi */ |
| 504 | static const char f32_3[] = |
| 505 | {0x8d,0x76,0x00}; /* leal 0(%esi),%esi */ |
| 506 | static const char f32_4[] = |
| 507 | {0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */ |
| 508 | static const char f32_5[] = |
| 509 | {0x90, /* nop */ |
| 510 | 0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */ |
| 511 | static const char f32_6[] = |
| 512 | {0x8d,0xb6,0x00,0x00,0x00,0x00}; /* leal 0L(%esi),%esi */ |
| 513 | static const char f32_7[] = |
| 514 | {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */ |
| 515 | static const char f32_8[] = |
| 516 | {0x90, /* nop */ |
| 517 | 0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */ |
| 518 | static const char f32_9[] = |
| 519 | {0x89,0xf6, /* movl %esi,%esi */ |
| 520 | 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */ |
| 521 | static const char f32_10[] = |
| 522 | {0x8d,0x76,0x00, /* leal 0(%esi),%esi */ |
| 523 | 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */ |
| 524 | static const char f32_11[] = |
| 525 | {0x8d,0x74,0x26,0x00, /* leal 0(%esi,1),%esi */ |
| 526 | 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */ |
| 527 | static const char f32_12[] = |
| 528 | {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */ |
| 529 | 0x8d,0xbf,0x00,0x00,0x00,0x00}; /* leal 0L(%edi),%edi */ |
| 530 | static const char f32_13[] = |
| 531 | {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */ |
| 532 | 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */ |
| 533 | static const char f32_14[] = |
| 534 | {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00, /* leal 0L(%esi,1),%esi */ |
| 535 | 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */ |
| 536 | static const char f32_15[] = |
| 537 | {0xeb,0x0d,0x90,0x90,0x90,0x90,0x90, /* jmp .+15; lotsa nops */ |
| 538 | 0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90}; |
| 539 | static const char f16_3[] = |
| 540 | {0x8d,0x74,0x00}; /* lea 0(%esi),%esi */ |
| 541 | static const char f16_4[] = |
| 542 | {0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */ |
| 543 | static const char f16_5[] = |
| 544 | {0x90, /* nop */ |
| 545 | 0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */ |
| 546 | static const char f16_6[] = |
| 547 | {0x89,0xf6, /* mov %si,%si */ |
| 548 | 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */ |
| 549 | static const char f16_7[] = |
| 550 | {0x8d,0x74,0x00, /* lea 0(%si),%si */ |
| 551 | 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */ |
| 552 | static const char f16_8[] = |
| 553 | {0x8d,0xb4,0x00,0x00, /* lea 0w(%si),%si */ |
| 554 | 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */ |
| 555 | static const char *const f32_patt[] = { |
| 556 | f32_1, f32_2, f32_3, f32_4, f32_5, f32_6, f32_7, f32_8, |
| 557 | f32_9, f32_10, f32_11, f32_12, f32_13, f32_14, f32_15 |
| 558 | }; |
| 559 | static const char *const f16_patt[] = { |
| 560 | f32_1, f32_2, f16_3, f16_4, f16_5, f16_6, f16_7, f16_8, |
| 561 | f32_15, f32_15, f32_15, f32_15, f32_15, f32_15, f32_15 |
| 562 | }; |
| 563 | |
| 564 | if (count <= 0 || count > 15) |
| 565 | return; |
| 566 | |
| 567 | /* The recommended way to pad 64bit code is to use NOPs preceded by |
| 568 | maximally four 0x66 prefixes. Balance the size of nops. */ |
| 569 | if (flag_code == CODE_64BIT) |
| 570 | { |
| 571 | int i; |
| 572 | int nnops = (count + 3) / 4; |
| 573 | int len = count / nnops; |
| 574 | int remains = count - nnops * len; |
| 575 | int pos = 0; |
| 576 | |
| 577 | for (i = 0; i < remains; i++) |
| 578 | { |
| 579 | memset (fragP->fr_literal + fragP->fr_fix + pos, 0x66, len); |
| 580 | fragP->fr_literal[fragP->fr_fix + pos + len] = 0x90; |
| 581 | pos += len + 1; |
| 582 | } |
| 583 | for (; i < nnops; i++) |
| 584 | { |
| 585 | memset (fragP->fr_literal + fragP->fr_fix + pos, 0x66, len - 1); |
| 586 | fragP->fr_literal[fragP->fr_fix + pos + len - 1] = 0x90; |
| 587 | pos += len; |
| 588 | } |
| 589 | } |
| 590 | else |
| 591 | if (flag_code == CODE_16BIT) |
| 592 | { |
| 593 | memcpy (fragP->fr_literal + fragP->fr_fix, |
| 594 | f16_patt[count - 1], count); |
| 595 | if (count > 8) |
| 596 | /* Adjust jump offset. */ |
| 597 | fragP->fr_literal[fragP->fr_fix + 1] = count - 2; |
| 598 | } |
| 599 | else |
| 600 | memcpy (fragP->fr_literal + fragP->fr_fix, |
| 601 | f32_patt[count - 1], count); |
| 602 | fragP->fr_var = count; |
| 603 | } |
| 604 | |
| 605 | static INLINE unsigned int |
| 606 | mode_from_disp_size (t) |
| 607 | unsigned int t; |
| 608 | { |
| 609 | return (t & Disp8) ? 1 : (t & (Disp16 | Disp32 | Disp32S)) ? 2 : 0; |
| 610 | } |
| 611 | |
| 612 | static INLINE int |
| 613 | fits_in_signed_byte (num) |
| 614 | offsetT num; |
| 615 | { |
| 616 | return (num >= -128) && (num <= 127); |
| 617 | } |
| 618 | |
| 619 | static INLINE int |
| 620 | fits_in_unsigned_byte (num) |
| 621 | offsetT num; |
| 622 | { |
| 623 | return (num & 0xff) == num; |
| 624 | } |
| 625 | |
| 626 | static INLINE int |
| 627 | fits_in_unsigned_word (num) |
| 628 | offsetT num; |
| 629 | { |
| 630 | return (num & 0xffff) == num; |
| 631 | } |
| 632 | |
| 633 | static INLINE int |
| 634 | fits_in_signed_word (num) |
| 635 | offsetT num; |
| 636 | { |
| 637 | return (-32768 <= num) && (num <= 32767); |
| 638 | } |
| 639 | static INLINE int |
| 640 | fits_in_signed_long (num) |
| 641 | offsetT num ATTRIBUTE_UNUSED; |
| 642 | { |
| 643 | #ifndef BFD64 |
| 644 | return 1; |
| 645 | #else |
| 646 | return (!(((offsetT) -1 << 31) & num) |
| 647 | || (((offsetT) -1 << 31) & num) == ((offsetT) -1 << 31)); |
| 648 | #endif |
| 649 | } /* fits_in_signed_long() */ |
| 650 | static INLINE int |
| 651 | fits_in_unsigned_long (num) |
| 652 | offsetT num ATTRIBUTE_UNUSED; |
| 653 | { |
| 654 | #ifndef BFD64 |
| 655 | return 1; |
| 656 | #else |
| 657 | return (num & (((offsetT) 2 << 31) - 1)) == num; |
| 658 | #endif |
| 659 | } /* fits_in_unsigned_long() */ |
| 660 | |
| 661 | static int |
| 662 | smallest_imm_type (num) |
| 663 | offsetT num; |
| 664 | { |
| 665 | if (cpu_arch_flags != (Cpu086 | Cpu186 | Cpu286 | Cpu386 | Cpu486 | CpuNo64)) |
| 666 | { |
| 667 | /* This code is disabled on the 486 because all the Imm1 forms |
| 668 | in the opcode table are slower on the i486. They're the |
| 669 | versions with the implicitly specified single-position |
| 670 | displacement, which has another syntax if you really want to |
| 671 | use that form. */ |
| 672 | if (num == 1) |
| 673 | return Imm1 | Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64; |
| 674 | } |
| 675 | return (fits_in_signed_byte (num) |
| 676 | ? (Imm8S | Imm8 | Imm16 | Imm32 | Imm32S | Imm64) |
| 677 | : fits_in_unsigned_byte (num) |
| 678 | ? (Imm8 | Imm16 | Imm32 | Imm32S | Imm64) |
| 679 | : (fits_in_signed_word (num) || fits_in_unsigned_word (num)) |
| 680 | ? (Imm16 | Imm32 | Imm32S | Imm64) |
| 681 | : fits_in_signed_long (num) |
| 682 | ? (Imm32 | Imm32S | Imm64) |
| 683 | : fits_in_unsigned_long (num) |
| 684 | ? (Imm32 | Imm64) |
| 685 | : Imm64); |
| 686 | } |
| 687 | |
| 688 | static offsetT |
| 689 | offset_in_range (val, size) |
| 690 | offsetT val; |
| 691 | int size; |
| 692 | { |
| 693 | addressT mask; |
| 694 | |
| 695 | switch (size) |
| 696 | { |
| 697 | case 1: mask = ((addressT) 1 << 8) - 1; break; |
| 698 | case 2: mask = ((addressT) 1 << 16) - 1; break; |
| 699 | case 4: mask = ((addressT) 2 << 31) - 1; break; |
| 700 | #ifdef BFD64 |
| 701 | case 8: mask = ((addressT) 2 << 63) - 1; break; |
| 702 | #endif |
| 703 | default: abort (); |
| 704 | } |
| 705 | |
| 706 | /* If BFD64, sign extend val. */ |
| 707 | if (!use_rela_relocations) |
| 708 | if ((val & ~(((addressT) 2 << 31) - 1)) == 0) |
| 709 | val = (val ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31); |
| 710 | |
| 711 | if ((val & ~mask) != 0 && (val & ~mask) != ~mask) |
| 712 | { |
| 713 | char buf1[40], buf2[40]; |
| 714 | |
| 715 | sprint_value (buf1, val); |
| 716 | sprint_value (buf2, val & mask); |
| 717 | as_warn (_("%s shortened to %s"), buf1, buf2); |
| 718 | } |
| 719 | return val & mask; |
| 720 | } |
| 721 | |
| 722 | /* Returns 0 if attempting to add a prefix where one from the same |
| 723 | class already exists, 1 if non rep/repne added, 2 if rep/repne |
| 724 | added. */ |
| 725 | static int |
| 726 | add_prefix (prefix) |
| 727 | unsigned int prefix; |
| 728 | { |
| 729 | int ret = 1; |
| 730 | int q; |
| 731 | |
| 732 | if (prefix >= REX_OPCODE && prefix < REX_OPCODE + 16 |
| 733 | && flag_code == CODE_64BIT) |
| 734 | q = REX_PREFIX; |
| 735 | else |
| 736 | switch (prefix) |
| 737 | { |
| 738 | default: |
| 739 | abort (); |
| 740 | |
| 741 | case CS_PREFIX_OPCODE: |
| 742 | case DS_PREFIX_OPCODE: |
| 743 | case ES_PREFIX_OPCODE: |
| 744 | case FS_PREFIX_OPCODE: |
| 745 | case GS_PREFIX_OPCODE: |
| 746 | case SS_PREFIX_OPCODE: |
| 747 | q = SEG_PREFIX; |
| 748 | break; |
| 749 | |
| 750 | case REPNE_PREFIX_OPCODE: |
| 751 | case REPE_PREFIX_OPCODE: |
| 752 | ret = 2; |
| 753 | /* fall thru */ |
| 754 | case LOCK_PREFIX_OPCODE: |
| 755 | q = LOCKREP_PREFIX; |
| 756 | break; |
| 757 | |
| 758 | case FWAIT_OPCODE: |
| 759 | q = WAIT_PREFIX; |
| 760 | break; |
| 761 | |
| 762 | case ADDR_PREFIX_OPCODE: |
| 763 | q = ADDR_PREFIX; |
| 764 | break; |
| 765 | |
| 766 | case DATA_PREFIX_OPCODE: |
| 767 | q = DATA_PREFIX; |
| 768 | break; |
| 769 | } |
| 770 | |
| 771 | if (i.prefix[q] != 0) |
| 772 | { |
| 773 | as_bad (_("same type of prefix used twice")); |
| 774 | return 0; |
| 775 | } |
| 776 | |
| 777 | i.prefixes += 1; |
| 778 | i.prefix[q] = prefix; |
| 779 | return ret; |
| 780 | } |
| 781 | |
| 782 | static void |
| 783 | set_code_flag (value) |
| 784 | int value; |
| 785 | { |
| 786 | flag_code = value; |
| 787 | cpu_arch_flags &= ~(Cpu64 | CpuNo64); |
| 788 | cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64); |
| 789 | if (value == CODE_64BIT && !(cpu_arch_flags & CpuSledgehammer)) |
| 790 | { |
| 791 | as_bad (_("64bit mode not supported on this CPU.")); |
| 792 | } |
| 793 | if (value == CODE_32BIT && !(cpu_arch_flags & Cpu386)) |
| 794 | { |
| 795 | as_bad (_("32bit mode not supported on this CPU.")); |
| 796 | } |
| 797 | stackop_size = '\0'; |
| 798 | } |
| 799 | |
| 800 | static void |
| 801 | set_16bit_gcc_code_flag (new_code_flag) |
| 802 | int new_code_flag; |
| 803 | { |
| 804 | flag_code = new_code_flag; |
| 805 | cpu_arch_flags &= ~(Cpu64 | CpuNo64); |
| 806 | cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64); |
| 807 | stackop_size = LONG_MNEM_SUFFIX; |
| 808 | } |
| 809 | |
| 810 | static void |
| 811 | set_intel_syntax (syntax_flag) |
| 812 | int syntax_flag; |
| 813 | { |
| 814 | /* Find out if register prefixing is specified. */ |
| 815 | int ask_naked_reg = 0; |
| 816 | |
| 817 | SKIP_WHITESPACE (); |
| 818 | if (!is_end_of_line[(unsigned char) *input_line_pointer]) |
| 819 | { |
| 820 | char *string = input_line_pointer; |
| 821 | int e = get_symbol_end (); |
| 822 | |
| 823 | if (strcmp (string, "prefix") == 0) |
| 824 | ask_naked_reg = 1; |
| 825 | else if (strcmp (string, "noprefix") == 0) |
| 826 | ask_naked_reg = -1; |
| 827 | else |
| 828 | as_bad (_("bad argument to syntax directive.")); |
| 829 | *input_line_pointer = e; |
| 830 | } |
| 831 | demand_empty_rest_of_line (); |
| 832 | |
| 833 | intel_syntax = syntax_flag; |
| 834 | |
| 835 | if (ask_naked_reg == 0) |
| 836 | allow_naked_reg = (intel_syntax |
| 837 | && (bfd_get_symbol_leading_char (stdoutput) != '\0')); |
| 838 | else |
| 839 | allow_naked_reg = (ask_naked_reg < 0); |
| 840 | |
| 841 | identifier_chars['%'] = intel_syntax && allow_naked_reg ? '%' : 0; |
| 842 | identifier_chars['$'] = intel_syntax ? '$' : 0; |
| 843 | } |
| 844 | |
| 845 | static void |
| 846 | set_cpu_arch (dummy) |
| 847 | int dummy ATTRIBUTE_UNUSED; |
| 848 | { |
| 849 | SKIP_WHITESPACE (); |
| 850 | |
| 851 | if (!is_end_of_line[(unsigned char) *input_line_pointer]) |
| 852 | { |
| 853 | char *string = input_line_pointer; |
| 854 | int e = get_symbol_end (); |
| 855 | int i; |
| 856 | |
| 857 | for (i = 0; cpu_arch[i].name; i++) |
| 858 | { |
| 859 | if (strcmp (string, cpu_arch[i].name) == 0) |
| 860 | { |
| 861 | if (*string != '.') |
| 862 | { |
| 863 | cpu_arch_name = cpu_arch[i].name; |
| 864 | cpu_sub_arch_name = NULL; |
| 865 | cpu_arch_flags = (cpu_arch[i].flags |
| 866 | | (flag_code == CODE_64BIT ? Cpu64 : CpuNo64)); |
| 867 | break; |
| 868 | } |
| 869 | if ((cpu_arch_flags | cpu_arch[i].flags) != cpu_arch_flags) |
| 870 | { |
| 871 | cpu_sub_arch_name = cpu_arch[i].name; |
| 872 | cpu_arch_flags |= cpu_arch[i].flags; |
| 873 | } |
| 874 | *input_line_pointer = e; |
| 875 | demand_empty_rest_of_line (); |
| 876 | return; |
| 877 | } |
| 878 | } |
| 879 | if (!cpu_arch[i].name) |
| 880 | as_bad (_("no such architecture: `%s'"), string); |
| 881 | |
| 882 | *input_line_pointer = e; |
| 883 | } |
| 884 | else |
| 885 | as_bad (_("missing cpu architecture")); |
| 886 | |
| 887 | no_cond_jump_promotion = 0; |
| 888 | if (*input_line_pointer == ',' |
| 889 | && !is_end_of_line[(unsigned char) input_line_pointer[1]]) |
| 890 | { |
| 891 | char *string = ++input_line_pointer; |
| 892 | int e = get_symbol_end (); |
| 893 | |
| 894 | if (strcmp (string, "nojumps") == 0) |
| 895 | no_cond_jump_promotion = 1; |
| 896 | else if (strcmp (string, "jumps") == 0) |
| 897 | ; |
| 898 | else |
| 899 | as_bad (_("no such architecture modifier: `%s'"), string); |
| 900 | |
| 901 | *input_line_pointer = e; |
| 902 | } |
| 903 | |
| 904 | demand_empty_rest_of_line (); |
| 905 | } |
| 906 | |
| 907 | unsigned long |
| 908 | i386_mach () |
| 909 | { |
| 910 | if (!strcmp (default_arch, "x86_64")) |
| 911 | return bfd_mach_x86_64; |
| 912 | else if (!strcmp (default_arch, "i386")) |
| 913 | return bfd_mach_i386_i386; |
| 914 | else |
| 915 | as_fatal (_("Unknown architecture")); |
| 916 | } |
| 917 | \f |
| 918 | void |
| 919 | md_begin () |
| 920 | { |
| 921 | const char *hash_err; |
| 922 | |
| 923 | /* Initialize op_hash hash table. */ |
| 924 | op_hash = hash_new (); |
| 925 | |
| 926 | { |
| 927 | const template *optab; |
| 928 | templates *core_optab; |
| 929 | |
| 930 | /* Setup for loop. */ |
| 931 | optab = i386_optab; |
| 932 | core_optab = (templates *) xmalloc (sizeof (templates)); |
| 933 | core_optab->start = optab; |
| 934 | |
| 935 | while (1) |
| 936 | { |
| 937 | ++optab; |
| 938 | if (optab->name == NULL |
| 939 | || strcmp (optab->name, (optab - 1)->name) != 0) |
| 940 | { |
| 941 | /* different name --> ship out current template list; |
| 942 | add to hash table; & begin anew. */ |
| 943 | core_optab->end = optab; |
| 944 | hash_err = hash_insert (op_hash, |
| 945 | (optab - 1)->name, |
| 946 | (PTR) core_optab); |
| 947 | if (hash_err) |
| 948 | { |
| 949 | as_fatal (_("Internal Error: Can't hash %s: %s"), |
| 950 | (optab - 1)->name, |
| 951 | hash_err); |
| 952 | } |
| 953 | if (optab->name == NULL) |
| 954 | break; |
| 955 | core_optab = (templates *) xmalloc (sizeof (templates)); |
| 956 | core_optab->start = optab; |
| 957 | } |
| 958 | } |
| 959 | } |
| 960 | |
| 961 | /* Initialize reg_hash hash table. */ |
| 962 | reg_hash = hash_new (); |
| 963 | { |
| 964 | const reg_entry *regtab; |
| 965 | |
| 966 | for (regtab = i386_regtab; |
| 967 | regtab < i386_regtab + sizeof (i386_regtab) / sizeof (i386_regtab[0]); |
| 968 | regtab++) |
| 969 | { |
| 970 | hash_err = hash_insert (reg_hash, regtab->reg_name, (PTR) regtab); |
| 971 | if (hash_err) |
| 972 | as_fatal (_("Internal Error: Can't hash %s: %s"), |
| 973 | regtab->reg_name, |
| 974 | hash_err); |
| 975 | } |
| 976 | } |
| 977 | |
| 978 | /* Fill in lexical tables: mnemonic_chars, operand_chars. */ |
| 979 | { |
| 980 | int c; |
| 981 | char *p; |
| 982 | |
| 983 | for (c = 0; c < 256; c++) |
| 984 | { |
| 985 | if (ISDIGIT (c)) |
| 986 | { |
| 987 | digit_chars[c] = c; |
| 988 | mnemonic_chars[c] = c; |
| 989 | register_chars[c] = c; |
| 990 | operand_chars[c] = c; |
| 991 | } |
| 992 | else if (ISLOWER (c)) |
| 993 | { |
| 994 | mnemonic_chars[c] = c; |
| 995 | register_chars[c] = c; |
| 996 | operand_chars[c] = c; |
| 997 | } |
| 998 | else if (ISUPPER (c)) |
| 999 | { |
| 1000 | mnemonic_chars[c] = TOLOWER (c); |
| 1001 | register_chars[c] = mnemonic_chars[c]; |
| 1002 | operand_chars[c] = c; |
| 1003 | } |
| 1004 | |
| 1005 | if (ISALPHA (c) || ISDIGIT (c)) |
| 1006 | identifier_chars[c] = c; |
| 1007 | else if (c >= 128) |
| 1008 | { |
| 1009 | identifier_chars[c] = c; |
| 1010 | operand_chars[c] = c; |
| 1011 | } |
| 1012 | } |
| 1013 | |
| 1014 | #ifdef LEX_AT |
| 1015 | identifier_chars['@'] = '@'; |
| 1016 | #endif |
| 1017 | #ifdef LEX_QM |
| 1018 | identifier_chars['?'] = '?'; |
| 1019 | operand_chars['?'] = '?'; |
| 1020 | #endif |
| 1021 | digit_chars['-'] = '-'; |
| 1022 | mnemonic_chars['-'] = '-'; |
| 1023 | identifier_chars['_'] = '_'; |
| 1024 | identifier_chars['.'] = '.'; |
| 1025 | |
| 1026 | for (p = operand_special_chars; *p != '\0'; p++) |
| 1027 | operand_chars[(unsigned char) *p] = *p; |
| 1028 | } |
| 1029 | |
| 1030 | #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) |
| 1031 | if (IS_ELF) |
| 1032 | { |
| 1033 | record_alignment (text_section, 2); |
| 1034 | record_alignment (data_section, 2); |
| 1035 | record_alignment (bss_section, 2); |
| 1036 | } |
| 1037 | #endif |
| 1038 | |
| 1039 | if (flag_code == CODE_64BIT) |
| 1040 | { |
| 1041 | x86_dwarf2_return_column = 16; |
| 1042 | x86_cie_data_alignment = -8; |
| 1043 | } |
| 1044 | else |
| 1045 | { |
| 1046 | x86_dwarf2_return_column = 8; |
| 1047 | x86_cie_data_alignment = -4; |
| 1048 | } |
| 1049 | } |
| 1050 | |
| 1051 | void |
| 1052 | i386_print_statistics (file) |
| 1053 | FILE *file; |
| 1054 | { |
| 1055 | hash_print_statistics (file, "i386 opcode", op_hash); |
| 1056 | hash_print_statistics (file, "i386 register", reg_hash); |
| 1057 | } |
| 1058 | \f |
| 1059 | #ifdef DEBUG386 |
| 1060 | |
| 1061 | /* Debugging routines for md_assemble. */ |
| 1062 | static void pi PARAMS ((char *, i386_insn *)); |
| 1063 | static void pte PARAMS ((template *)); |
| 1064 | static void pt PARAMS ((unsigned int)); |
| 1065 | static void pe PARAMS ((expressionS *)); |
| 1066 | static void ps PARAMS ((symbolS *)); |
| 1067 | |
| 1068 | static void |
| 1069 | pi (line, x) |
| 1070 | char *line; |
| 1071 | i386_insn *x; |
| 1072 | { |
| 1073 | unsigned int i; |
| 1074 | |
| 1075 | fprintf (stdout, "%s: template ", line); |
| 1076 | pte (&x->tm); |
| 1077 | fprintf (stdout, " address: base %s index %s scale %x\n", |
| 1078 | x->base_reg ? x->base_reg->reg_name : "none", |
| 1079 | x->index_reg ? x->index_reg->reg_name : "none", |
| 1080 | x->log2_scale_factor); |
| 1081 | fprintf (stdout, " modrm: mode %x reg %x reg/mem %x\n", |
| 1082 | x->rm.mode, x->rm.reg, x->rm.regmem); |
| 1083 | fprintf (stdout, " sib: base %x index %x scale %x\n", |
| 1084 | x->sib.base, x->sib.index, x->sib.scale); |
| 1085 | fprintf (stdout, " rex: 64bit %x extX %x extY %x extZ %x\n", |
| 1086 | (x->rex & REX_MODE64) != 0, |
| 1087 | (x->rex & REX_EXTX) != 0, |
| 1088 | (x->rex & REX_EXTY) != 0, |
| 1089 | (x->rex & REX_EXTZ) != 0); |
| 1090 | for (i = 0; i < x->operands; i++) |
| 1091 | { |
| 1092 | fprintf (stdout, " #%d: ", i + 1); |
| 1093 | pt (x->types[i]); |
| 1094 | fprintf (stdout, "\n"); |
| 1095 | if (x->types[i] |
| 1096 | & (Reg | SReg2 | SReg3 | Control | Debug | Test | RegMMX | RegXMM)) |
| 1097 | fprintf (stdout, "%s\n", x->op[i].regs->reg_name); |
| 1098 | if (x->types[i] & Imm) |
| 1099 | pe (x->op[i].imms); |
| 1100 | if (x->types[i] & Disp) |
| 1101 | pe (x->op[i].disps); |
| 1102 | } |
| 1103 | } |
| 1104 | |
| 1105 | static void |
| 1106 | pte (t) |
| 1107 | template *t; |
| 1108 | { |
| 1109 | unsigned int i; |
| 1110 | fprintf (stdout, " %d operands ", t->operands); |
| 1111 | fprintf (stdout, "opcode %x ", t->base_opcode); |
| 1112 | if (t->extension_opcode != None) |
| 1113 | fprintf (stdout, "ext %x ", t->extension_opcode); |
| 1114 | if (t->opcode_modifier & D) |
| 1115 | fprintf (stdout, "D"); |
| 1116 | if (t->opcode_modifier & W) |
| 1117 | fprintf (stdout, "W"); |
| 1118 | fprintf (stdout, "\n"); |
| 1119 | for (i = 0; i < t->operands; i++) |
| 1120 | { |
| 1121 | fprintf (stdout, " #%d type ", i + 1); |
| 1122 | pt (t->operand_types[i]); |
| 1123 | fprintf (stdout, "\n"); |
| 1124 | } |
| 1125 | } |
| 1126 | |
| 1127 | static void |
| 1128 | pe (e) |
| 1129 | expressionS *e; |
| 1130 | { |
| 1131 | fprintf (stdout, " operation %d\n", e->X_op); |
| 1132 | fprintf (stdout, " add_number %ld (%lx)\n", |
| 1133 | (long) e->X_add_number, (long) e->X_add_number); |
| 1134 | if (e->X_add_symbol) |
| 1135 | { |
| 1136 | fprintf (stdout, " add_symbol "); |
| 1137 | ps (e->X_add_symbol); |
| 1138 | fprintf (stdout, "\n"); |
| 1139 | } |
| 1140 | if (e->X_op_symbol) |
| 1141 | { |
| 1142 | fprintf (stdout, " op_symbol "); |
| 1143 | ps (e->X_op_symbol); |
| 1144 | fprintf (stdout, "\n"); |
| 1145 | } |
| 1146 | } |
| 1147 | |
| 1148 | static void |
| 1149 | ps (s) |
| 1150 | symbolS *s; |
| 1151 | { |
| 1152 | fprintf (stdout, "%s type %s%s", |
| 1153 | S_GET_NAME (s), |
| 1154 | S_IS_EXTERNAL (s) ? "EXTERNAL " : "", |
| 1155 | segment_name (S_GET_SEGMENT (s))); |
| 1156 | } |
| 1157 | |
| 1158 | struct type_name |
| 1159 | { |
| 1160 | unsigned int mask; |
| 1161 | char *tname; |
| 1162 | } |
| 1163 | |
| 1164 | static const type_names[] = |
| 1165 | { |
| 1166 | { Reg8, "r8" }, |
| 1167 | { Reg16, "r16" }, |
| 1168 | { Reg32, "r32" }, |
| 1169 | { Reg64, "r64" }, |
| 1170 | { Imm8, "i8" }, |
| 1171 | { Imm8S, "i8s" }, |
| 1172 | { Imm16, "i16" }, |
| 1173 | { Imm32, "i32" }, |
| 1174 | { Imm32S, "i32s" }, |
| 1175 | { Imm64, "i64" }, |
| 1176 | { Imm1, "i1" }, |
| 1177 | { BaseIndex, "BaseIndex" }, |
| 1178 | { Disp8, "d8" }, |
| 1179 | { Disp16, "d16" }, |
| 1180 | { Disp32, "d32" }, |
| 1181 | { Disp32S, "d32s" }, |
| 1182 | { Disp64, "d64" }, |
| 1183 | { InOutPortReg, "InOutPortReg" }, |
| 1184 | { ShiftCount, "ShiftCount" }, |
| 1185 | { Control, "control reg" }, |
| 1186 | { Test, "test reg" }, |
| 1187 | { Debug, "debug reg" }, |
| 1188 | { FloatReg, "FReg" }, |
| 1189 | { FloatAcc, "FAcc" }, |
| 1190 | { SReg2, "SReg2" }, |
| 1191 | { SReg3, "SReg3" }, |
| 1192 | { Acc, "Acc" }, |
| 1193 | { JumpAbsolute, "Jump Absolute" }, |
| 1194 | { RegMMX, "rMMX" }, |
| 1195 | { RegXMM, "rXMM" }, |
| 1196 | { EsSeg, "es" }, |
| 1197 | { 0, "" } |
| 1198 | }; |
| 1199 | |
| 1200 | static void |
| 1201 | pt (t) |
| 1202 | unsigned int t; |
| 1203 | { |
| 1204 | const struct type_name *ty; |
| 1205 | |
| 1206 | for (ty = type_names; ty->mask; ty++) |
| 1207 | if (t & ty->mask) |
| 1208 | fprintf (stdout, "%s, ", ty->tname); |
| 1209 | fflush (stdout); |
| 1210 | } |
| 1211 | |
| 1212 | #endif /* DEBUG386 */ |
| 1213 | \f |
| 1214 | static bfd_reloc_code_real_type |
| 1215 | reloc (unsigned int size, |
| 1216 | int pcrel, |
| 1217 | int sign, |
| 1218 | bfd_reloc_code_real_type other) |
| 1219 | { |
| 1220 | if (other != NO_RELOC) |
| 1221 | { |
| 1222 | reloc_howto_type *reloc; |
| 1223 | |
| 1224 | if (size == 8) |
| 1225 | switch (other) |
| 1226 | { |
| 1227 | case BFD_RELOC_X86_64_TPOFF32: |
| 1228 | other = BFD_RELOC_X86_64_TPOFF64; |
| 1229 | break; |
| 1230 | case BFD_RELOC_X86_64_DTPOFF32: |
| 1231 | other = BFD_RELOC_X86_64_DTPOFF64; |
| 1232 | break; |
| 1233 | default: |
| 1234 | break; |
| 1235 | } |
| 1236 | reloc = bfd_reloc_type_lookup (stdoutput, other); |
| 1237 | if (!reloc) |
| 1238 | as_bad (_("unknown relocation (%u)"), other); |
| 1239 | else if (size != bfd_get_reloc_size (reloc)) |
| 1240 | as_bad (_("%u-byte relocation cannot be applied to %u-byte field"), |
| 1241 | bfd_get_reloc_size (reloc), |
| 1242 | size); |
| 1243 | else if (pcrel && !reloc->pc_relative) |
| 1244 | as_bad (_("non-pc-relative relocation for pc-relative field")); |
| 1245 | else if ((reloc->complain_on_overflow == complain_overflow_signed |
| 1246 | && !sign) |
| 1247 | || (reloc->complain_on_overflow == complain_overflow_unsigned |
| 1248 | && sign > 0)) |
| 1249 | as_bad (_("relocated field and relocation type differ in signedness")); |
| 1250 | else |
| 1251 | return other; |
| 1252 | return NO_RELOC; |
| 1253 | } |
| 1254 | |
| 1255 | if (pcrel) |
| 1256 | { |
| 1257 | if (!sign) |
| 1258 | as_bad (_("there are no unsigned pc-relative relocations")); |
| 1259 | switch (size) |
| 1260 | { |
| 1261 | case 1: return BFD_RELOC_8_PCREL; |
| 1262 | case 2: return BFD_RELOC_16_PCREL; |
| 1263 | case 4: return BFD_RELOC_32_PCREL; |
| 1264 | case 8: return BFD_RELOC_64_PCREL; |
| 1265 | } |
| 1266 | as_bad (_("cannot do %u byte pc-relative relocation"), size); |
| 1267 | } |
| 1268 | else |
| 1269 | { |
| 1270 | if (sign > 0) |
| 1271 | switch (size) |
| 1272 | { |
| 1273 | case 4: return BFD_RELOC_X86_64_32S; |
| 1274 | } |
| 1275 | else |
| 1276 | switch (size) |
| 1277 | { |
| 1278 | case 1: return BFD_RELOC_8; |
| 1279 | case 2: return BFD_RELOC_16; |
| 1280 | case 4: return BFD_RELOC_32; |
| 1281 | case 8: return BFD_RELOC_64; |
| 1282 | } |
| 1283 | as_bad (_("cannot do %s %u byte relocation"), |
| 1284 | sign > 0 ? "signed" : "unsigned", size); |
| 1285 | } |
| 1286 | |
| 1287 | abort (); |
| 1288 | return BFD_RELOC_NONE; |
| 1289 | } |
| 1290 | |
| 1291 | /* Here we decide which fixups can be adjusted to make them relative to |
| 1292 | the beginning of the section instead of the symbol. Basically we need |
| 1293 | to make sure that the dynamic relocations are done correctly, so in |
| 1294 | some cases we force the original symbol to be used. */ |
| 1295 | |
| 1296 | int |
| 1297 | tc_i386_fix_adjustable (fixP) |
| 1298 | fixS *fixP ATTRIBUTE_UNUSED; |
| 1299 | { |
| 1300 | #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) |
| 1301 | if (!IS_ELF) |
| 1302 | return 1; |
| 1303 | |
| 1304 | /* Don't adjust pc-relative references to merge sections in 64-bit |
| 1305 | mode. */ |
| 1306 | if (use_rela_relocations |
| 1307 | && (S_GET_SEGMENT (fixP->fx_addsy)->flags & SEC_MERGE) != 0 |
| 1308 | && fixP->fx_pcrel) |
| 1309 | return 0; |
| 1310 | |
| 1311 | /* The x86_64 GOTPCREL are represented as 32bit PCrel relocations |
| 1312 | and changed later by validate_fix. */ |
| 1313 | if (GOT_symbol && fixP->fx_subsy == GOT_symbol |
| 1314 | && fixP->fx_r_type == BFD_RELOC_32_PCREL) |
| 1315 | return 0; |
| 1316 | |
| 1317 | /* adjust_reloc_syms doesn't know about the GOT. */ |
| 1318 | if (fixP->fx_r_type == BFD_RELOC_386_GOTOFF |
| 1319 | || fixP->fx_r_type == BFD_RELOC_386_PLT32 |
| 1320 | || fixP->fx_r_type == BFD_RELOC_386_GOT32 |
| 1321 | || fixP->fx_r_type == BFD_RELOC_386_TLS_GD |
| 1322 | || fixP->fx_r_type == BFD_RELOC_386_TLS_LDM |
| 1323 | || fixP->fx_r_type == BFD_RELOC_386_TLS_LDO_32 |
| 1324 | || fixP->fx_r_type == BFD_RELOC_386_TLS_IE_32 |
| 1325 | || fixP->fx_r_type == BFD_RELOC_386_TLS_IE |
| 1326 | || fixP->fx_r_type == BFD_RELOC_386_TLS_GOTIE |
| 1327 | || fixP->fx_r_type == BFD_RELOC_386_TLS_LE_32 |
| 1328 | || fixP->fx_r_type == BFD_RELOC_386_TLS_LE |
| 1329 | || fixP->fx_r_type == BFD_RELOC_X86_64_PLT32 |
| 1330 | || fixP->fx_r_type == BFD_RELOC_X86_64_GOT32 |
| 1331 | || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPCREL |
| 1332 | || fixP->fx_r_type == BFD_RELOC_X86_64_TLSGD |
| 1333 | || fixP->fx_r_type == BFD_RELOC_X86_64_TLSLD |
| 1334 | || fixP->fx_r_type == BFD_RELOC_X86_64_DTPOFF32 |
| 1335 | || fixP->fx_r_type == BFD_RELOC_X86_64_DTPOFF64 |
| 1336 | || fixP->fx_r_type == BFD_RELOC_X86_64_GOTTPOFF |
| 1337 | || fixP->fx_r_type == BFD_RELOC_X86_64_TPOFF32 |
| 1338 | || fixP->fx_r_type == BFD_RELOC_X86_64_TPOFF64 |
| 1339 | || fixP->fx_r_type == BFD_RELOC_X86_64_GOTOFF64 |
| 1340 | || fixP->fx_r_type == BFD_RELOC_VTABLE_INHERIT |
| 1341 | || fixP->fx_r_type == BFD_RELOC_VTABLE_ENTRY) |
| 1342 | return 0; |
| 1343 | #endif |
| 1344 | return 1; |
| 1345 | } |
| 1346 | |
| 1347 | static int intel_float_operand PARAMS ((const char *mnemonic)); |
| 1348 | |
| 1349 | static int |
| 1350 | intel_float_operand (mnemonic) |
| 1351 | const char *mnemonic; |
| 1352 | { |
| 1353 | /* Note that the value returned is meaningful only for opcodes with (memory) |
| 1354 | operands, hence the code here is free to improperly handle opcodes that |
| 1355 | have no operands (for better performance and smaller code). */ |
| 1356 | |
| 1357 | if (mnemonic[0] != 'f') |
| 1358 | return 0; /* non-math */ |
| 1359 | |
| 1360 | switch (mnemonic[1]) |
| 1361 | { |
| 1362 | /* fclex, fdecstp, fdisi, femms, feni, fincstp, finit, fsetpm, and |
| 1363 | the fs segment override prefix not currently handled because no |
| 1364 | call path can make opcodes without operands get here */ |
| 1365 | case 'i': |
| 1366 | return 2 /* integer op */; |
| 1367 | case 'l': |
| 1368 | if (mnemonic[2] == 'd' && (mnemonic[3] == 'c' || mnemonic[3] == 'e')) |
| 1369 | return 3; /* fldcw/fldenv */ |
| 1370 | break; |
| 1371 | case 'n': |
| 1372 | if (mnemonic[2] != 'o' /* fnop */) |
| 1373 | return 3; /* non-waiting control op */ |
| 1374 | break; |
| 1375 | case 'r': |
| 1376 | if (mnemonic[2] == 's') |
| 1377 | return 3; /* frstor/frstpm */ |
| 1378 | break; |
| 1379 | case 's': |
| 1380 | if (mnemonic[2] == 'a') |
| 1381 | return 3; /* fsave */ |
| 1382 | if (mnemonic[2] == 't') |
| 1383 | { |
| 1384 | switch (mnemonic[3]) |
| 1385 | { |
| 1386 | case 'c': /* fstcw */ |
| 1387 | case 'd': /* fstdw */ |
| 1388 | case 'e': /* fstenv */ |
| 1389 | case 's': /* fsts[gw] */ |
| 1390 | return 3; |
| 1391 | } |
| 1392 | } |
| 1393 | break; |
| 1394 | case 'x': |
| 1395 | if (mnemonic[2] == 'r' || mnemonic[2] == 's') |
| 1396 | return 0; /* fxsave/fxrstor are not really math ops */ |
| 1397 | break; |
| 1398 | } |
| 1399 | |
| 1400 | return 1; |
| 1401 | } |
| 1402 | |
| 1403 | /* This is the guts of the machine-dependent assembler. LINE points to a |
| 1404 | machine dependent instruction. This function is supposed to emit |
| 1405 | the frags/bytes it assembles to. */ |
| 1406 | |
| 1407 | void |
| 1408 | md_assemble (line) |
| 1409 | char *line; |
| 1410 | { |
| 1411 | int j; |
| 1412 | char mnemonic[MAX_MNEM_SIZE]; |
| 1413 | |
| 1414 | /* Initialize globals. */ |
| 1415 | memset (&i, '\0', sizeof (i)); |
| 1416 | for (j = 0; j < MAX_OPERANDS; j++) |
| 1417 | i.reloc[j] = NO_RELOC; |
| 1418 | memset (disp_expressions, '\0', sizeof (disp_expressions)); |
| 1419 | memset (im_expressions, '\0', sizeof (im_expressions)); |
| 1420 | save_stack_p = save_stack; |
| 1421 | |
| 1422 | /* First parse an instruction mnemonic & call i386_operand for the operands. |
| 1423 | We assume that the scrubber has arranged it so that line[0] is the valid |
| 1424 | start of a (possibly prefixed) mnemonic. */ |
| 1425 | |
| 1426 | line = parse_insn (line, mnemonic); |
| 1427 | if (line == NULL) |
| 1428 | return; |
| 1429 | |
| 1430 | line = parse_operands (line, mnemonic); |
| 1431 | if (line == NULL) |
| 1432 | return; |
| 1433 | |
| 1434 | /* Now we've parsed the mnemonic into a set of templates, and have the |
| 1435 | operands at hand. */ |
| 1436 | |
| 1437 | /* All intel opcodes have reversed operands except for "bound" and |
| 1438 | "enter". We also don't reverse intersegment "jmp" and "call" |
| 1439 | instructions with 2 immediate operands so that the immediate segment |
| 1440 | precedes the offset, as it does when in AT&T mode. "enter" and the |
| 1441 | intersegment "jmp" and "call" instructions are the only ones that |
| 1442 | have two immediate operands. */ |
| 1443 | if (intel_syntax && i.operands > 1 |
| 1444 | && (strcmp (mnemonic, "bound") != 0) |
| 1445 | && (strcmp (mnemonic, "invlpga") != 0) |
| 1446 | && !((i.types[0] & Imm) && (i.types[1] & Imm))) |
| 1447 | swap_operands (); |
| 1448 | |
| 1449 | if (i.imm_operands) |
| 1450 | optimize_imm (); |
| 1451 | |
| 1452 | /* Don't optimize displacement for movabs since it only takes 64bit |
| 1453 | displacement. */ |
| 1454 | if (i.disp_operands |
| 1455 | && (flag_code != CODE_64BIT |
| 1456 | || strcmp (mnemonic, "movabs") != 0)) |
| 1457 | optimize_disp (); |
| 1458 | |
| 1459 | /* Next, we find a template that matches the given insn, |
| 1460 | making sure the overlap of the given operands types is consistent |
| 1461 | with the template operand types. */ |
| 1462 | |
| 1463 | if (!match_template ()) |
| 1464 | return; |
| 1465 | |
| 1466 | if (intel_syntax) |
| 1467 | { |
| 1468 | /* Undo SYSV386_COMPAT brokenness when in Intel mode. See i386.h */ |
| 1469 | if (SYSV386_COMPAT |
| 1470 | && (i.tm.base_opcode & 0xfffffde0) == 0xdce0) |
| 1471 | i.tm.base_opcode ^= FloatR; |
| 1472 | |
| 1473 | /* Zap movzx and movsx suffix. The suffix may have been set from |
| 1474 | "word ptr" or "byte ptr" on the source operand, but we'll use |
| 1475 | the suffix later to choose the destination register. */ |
| 1476 | if ((i.tm.base_opcode & ~9) == 0x0fb6) |
| 1477 | { |
| 1478 | if (i.reg_operands < 2 |
| 1479 | && !i.suffix |
| 1480 | && (~i.tm.opcode_modifier |
| 1481 | & (No_bSuf |
| 1482 | | No_wSuf |
| 1483 | | No_lSuf |
| 1484 | | No_sSuf |
| 1485 | | No_xSuf |
| 1486 | | No_qSuf))) |
| 1487 | as_bad (_("ambiguous operand size for `%s'"), i.tm.name); |
| 1488 | |
| 1489 | i.suffix = 0; |
| 1490 | } |
| 1491 | } |
| 1492 | |
| 1493 | if (i.tm.opcode_modifier & FWait) |
| 1494 | if (!add_prefix (FWAIT_OPCODE)) |
| 1495 | return; |
| 1496 | |
| 1497 | /* Check string instruction segment overrides. */ |
| 1498 | if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0) |
| 1499 | { |
| 1500 | if (!check_string ()) |
| 1501 | return; |
| 1502 | } |
| 1503 | |
| 1504 | if (!process_suffix ()) |
| 1505 | return; |
| 1506 | |
| 1507 | /* Make still unresolved immediate matches conform to size of immediate |
| 1508 | given in i.suffix. */ |
| 1509 | if (!finalize_imm ()) |
| 1510 | return; |
| 1511 | |
| 1512 | if (i.types[0] & Imm1) |
| 1513 | i.imm_operands = 0; /* kludge for shift insns. */ |
| 1514 | if (i.types[0] & ImplicitRegister) |
| 1515 | i.reg_operands--; |
| 1516 | if (i.types[1] & ImplicitRegister) |
| 1517 | i.reg_operands--; |
| 1518 | if (i.types[2] & ImplicitRegister) |
| 1519 | i.reg_operands--; |
| 1520 | |
| 1521 | if (i.tm.opcode_modifier & ImmExt) |
| 1522 | { |
| 1523 | expressionS *exp; |
| 1524 | |
| 1525 | if ((i.tm.cpu_flags & CpuPNI) && i.operands > 0) |
| 1526 | { |
| 1527 | /* These Intel Prescott New Instructions have the fixed |
| 1528 | operands with an opcode suffix which is coded in the same |
| 1529 | place as an 8-bit immediate field would be. Here we check |
| 1530 | those operands and remove them afterwards. */ |
| 1531 | unsigned int x; |
| 1532 | |
| 1533 | for (x = 0; x < i.operands; x++) |
| 1534 | if (i.op[x].regs->reg_num != x) |
| 1535 | as_bad (_("can't use register '%%%s' as operand %d in '%s'."), |
| 1536 | i.op[x].regs->reg_name, x + 1, i.tm.name); |
| 1537 | i.operands = 0; |
| 1538 | } |
| 1539 | |
| 1540 | /* These AMD 3DNow! and Intel Katmai New Instructions have an |
| 1541 | opcode suffix which is coded in the same place as an 8-bit |
| 1542 | immediate field would be. Here we fake an 8-bit immediate |
| 1543 | operand from the opcode suffix stored in tm.extension_opcode. */ |
| 1544 | |
| 1545 | assert (i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS); |
| 1546 | |
| 1547 | exp = &im_expressions[i.imm_operands++]; |
| 1548 | i.op[i.operands].imms = exp; |
| 1549 | i.types[i.operands++] = Imm8; |
| 1550 | exp->X_op = O_constant; |
| 1551 | exp->X_add_number = i.tm.extension_opcode; |
| 1552 | i.tm.extension_opcode = None; |
| 1553 | } |
| 1554 | |
| 1555 | /* For insns with operands there are more diddles to do to the opcode. */ |
| 1556 | if (i.operands) |
| 1557 | { |
| 1558 | if (!process_operands ()) |
| 1559 | return; |
| 1560 | } |
| 1561 | else if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0) |
| 1562 | { |
| 1563 | /* UnixWare fsub no args is alias for fsubp, fadd -> faddp, etc. */ |
| 1564 | as_warn (_("translating to `%sp'"), i.tm.name); |
| 1565 | } |
| 1566 | |
| 1567 | /* Handle conversion of 'int $3' --> special int3 insn. */ |
| 1568 | if (i.tm.base_opcode == INT_OPCODE && i.op[0].imms->X_add_number == 3) |
| 1569 | { |
| 1570 | i.tm.base_opcode = INT3_OPCODE; |
| 1571 | i.imm_operands = 0; |
| 1572 | } |
| 1573 | |
| 1574 | if ((i.tm.opcode_modifier & (Jump | JumpByte | JumpDword)) |
| 1575 | && i.op[0].disps->X_op == O_constant) |
| 1576 | { |
| 1577 | /* Convert "jmp constant" (and "call constant") to a jump (call) to |
| 1578 | the absolute address given by the constant. Since ix86 jumps and |
| 1579 | calls are pc relative, we need to generate a reloc. */ |
| 1580 | i.op[0].disps->X_add_symbol = &abs_symbol; |
| 1581 | i.op[0].disps->X_op = O_symbol; |
| 1582 | } |
| 1583 | |
| 1584 | if ((i.tm.opcode_modifier & Rex64) != 0) |
| 1585 | i.rex |= REX_MODE64; |
| 1586 | |
| 1587 | /* For 8 bit registers we need an empty rex prefix. Also if the |
| 1588 | instruction already has a prefix, we need to convert old |
| 1589 | registers to new ones. */ |
| 1590 | |
| 1591 | if (((i.types[0] & Reg8) != 0 |
| 1592 | && (i.op[0].regs->reg_flags & RegRex64) != 0) |
| 1593 | || ((i.types[1] & Reg8) != 0 |
| 1594 | && (i.op[1].regs->reg_flags & RegRex64) != 0) |
| 1595 | || (((i.types[0] & Reg8) != 0 || (i.types[1] & Reg8) != 0) |
| 1596 | && i.rex != 0)) |
| 1597 | { |
| 1598 | int x; |
| 1599 | |
| 1600 | i.rex |= REX_OPCODE; |
| 1601 | for (x = 0; x < 2; x++) |
| 1602 | { |
| 1603 | /* Look for 8 bit operand that uses old registers. */ |
| 1604 | if ((i.types[x] & Reg8) != 0 |
| 1605 | && (i.op[x].regs->reg_flags & RegRex64) == 0) |
| 1606 | { |
| 1607 | /* In case it is "hi" register, give up. */ |
| 1608 | if (i.op[x].regs->reg_num > 3) |
| 1609 | as_bad (_("can't encode register '%%%s' in an instruction requiring REX prefix."), |
| 1610 | i.op[x].regs->reg_name); |
| 1611 | |
| 1612 | /* Otherwise it is equivalent to the extended register. |
| 1613 | Since the encoding doesn't change this is merely |
| 1614 | cosmetic cleanup for debug output. */ |
| 1615 | |
| 1616 | i.op[x].regs = i.op[x].regs + 8; |
| 1617 | } |
| 1618 | } |
| 1619 | } |
| 1620 | |
| 1621 | if (i.rex != 0) |
| 1622 | add_prefix (REX_OPCODE | i.rex); |
| 1623 | |
| 1624 | /* We are ready to output the insn. */ |
| 1625 | output_insn (); |
| 1626 | } |
| 1627 | |
| 1628 | static char * |
| 1629 | parse_insn (line, mnemonic) |
| 1630 | char *line; |
| 1631 | char *mnemonic; |
| 1632 | { |
| 1633 | char *l = line; |
| 1634 | char *token_start = l; |
| 1635 | char *mnem_p; |
| 1636 | int supported; |
| 1637 | const template *t; |
| 1638 | |
| 1639 | /* Non-zero if we found a prefix only acceptable with string insns. */ |
| 1640 | const char *expecting_string_instruction = NULL; |
| 1641 | |
| 1642 | while (1) |
| 1643 | { |
| 1644 | mnem_p = mnemonic; |
| 1645 | while ((*mnem_p = mnemonic_chars[(unsigned char) *l]) != 0) |
| 1646 | { |
| 1647 | mnem_p++; |
| 1648 | if (mnem_p >= mnemonic + MAX_MNEM_SIZE) |
| 1649 | { |
| 1650 | as_bad (_("no such instruction: `%s'"), token_start); |
| 1651 | return NULL; |
| 1652 | } |
| 1653 | l++; |
| 1654 | } |
| 1655 | if (!is_space_char (*l) |
| 1656 | && *l != END_OF_INSN |
| 1657 | && (intel_syntax |
| 1658 | || (*l != PREFIX_SEPARATOR |
| 1659 | && *l != ','))) |
| 1660 | { |
| 1661 | as_bad (_("invalid character %s in mnemonic"), |
| 1662 | output_invalid (*l)); |
| 1663 | return NULL; |
| 1664 | } |
| 1665 | if (token_start == l) |
| 1666 | { |
| 1667 | if (!intel_syntax && *l == PREFIX_SEPARATOR) |
| 1668 | as_bad (_("expecting prefix; got nothing")); |
| 1669 | else |
| 1670 | as_bad (_("expecting mnemonic; got nothing")); |
| 1671 | return NULL; |
| 1672 | } |
| 1673 | |
| 1674 | /* Look up instruction (or prefix) via hash table. */ |
| 1675 | current_templates = hash_find (op_hash, mnemonic); |
| 1676 | |
| 1677 | if (*l != END_OF_INSN |
| 1678 | && (!is_space_char (*l) || l[1] != END_OF_INSN) |
| 1679 | && current_templates |
| 1680 | && (current_templates->start->opcode_modifier & IsPrefix)) |
| 1681 | { |
| 1682 | if (current_templates->start->cpu_flags |
| 1683 | & (flag_code != CODE_64BIT ? Cpu64 : CpuNo64)) |
| 1684 | { |
| 1685 | as_bad ((flag_code != CODE_64BIT |
| 1686 | ? _("`%s' is only supported in 64-bit mode") |
| 1687 | : _("`%s' is not supported in 64-bit mode")), |
| 1688 | current_templates->start->name); |
| 1689 | return NULL; |
| 1690 | } |
| 1691 | /* If we are in 16-bit mode, do not allow addr16 or data16. |
| 1692 | Similarly, in 32-bit mode, do not allow addr32 or data32. */ |
| 1693 | if ((current_templates->start->opcode_modifier & (Size16 | Size32)) |
| 1694 | && flag_code != CODE_64BIT |
| 1695 | && (((current_templates->start->opcode_modifier & Size32) != 0) |
| 1696 | ^ (flag_code == CODE_16BIT))) |
| 1697 | { |
| 1698 | as_bad (_("redundant %s prefix"), |
| 1699 | current_templates->start->name); |
| 1700 | return NULL; |
| 1701 | } |
| 1702 | /* Add prefix, checking for repeated prefixes. */ |
| 1703 | switch (add_prefix (current_templates->start->base_opcode)) |
| 1704 | { |
| 1705 | case 0: |
| 1706 | return NULL; |
| 1707 | case 2: |
| 1708 | expecting_string_instruction = current_templates->start->name; |
| 1709 | break; |
| 1710 | } |
| 1711 | /* Skip past PREFIX_SEPARATOR and reset token_start. */ |
| 1712 | token_start = ++l; |
| 1713 | } |
| 1714 | else |
| 1715 | break; |
| 1716 | } |
| 1717 | |
| 1718 | if (!current_templates) |
| 1719 | { |
| 1720 | /* See if we can get a match by trimming off a suffix. */ |
| 1721 | switch (mnem_p[-1]) |
| 1722 | { |
| 1723 | case WORD_MNEM_SUFFIX: |
| 1724 | if (intel_syntax && (intel_float_operand (mnemonic) & 2)) |
| 1725 | i.suffix = SHORT_MNEM_SUFFIX; |
| 1726 | else |
| 1727 | case BYTE_MNEM_SUFFIX: |
| 1728 | case QWORD_MNEM_SUFFIX: |
| 1729 | i.suffix = mnem_p[-1]; |
| 1730 | mnem_p[-1] = '\0'; |
| 1731 | current_templates = hash_find (op_hash, mnemonic); |
| 1732 | break; |
| 1733 | case SHORT_MNEM_SUFFIX: |
| 1734 | case LONG_MNEM_SUFFIX: |
| 1735 | if (!intel_syntax) |
| 1736 | { |
| 1737 | i.suffix = mnem_p[-1]; |
| 1738 | mnem_p[-1] = '\0'; |
| 1739 | current_templates = hash_find (op_hash, mnemonic); |
| 1740 | } |
| 1741 | break; |
| 1742 | |
| 1743 | /* Intel Syntax. */ |
| 1744 | case 'd': |
| 1745 | if (intel_syntax) |
| 1746 | { |
| 1747 | if (intel_float_operand (mnemonic) == 1) |
| 1748 | i.suffix = SHORT_MNEM_SUFFIX; |
| 1749 | else |
| 1750 | i.suffix = LONG_MNEM_SUFFIX; |
| 1751 | mnem_p[-1] = '\0'; |
| 1752 | current_templates = hash_find (op_hash, mnemonic); |
| 1753 | } |
| 1754 | break; |
| 1755 | } |
| 1756 | if (!current_templates) |
| 1757 | { |
| 1758 | as_bad (_("no such instruction: `%s'"), token_start); |
| 1759 | return NULL; |
| 1760 | } |
| 1761 | } |
| 1762 | |
| 1763 | if (current_templates->start->opcode_modifier & (Jump | JumpByte)) |
| 1764 | { |
| 1765 | /* Check for a branch hint. We allow ",pt" and ",pn" for |
| 1766 | predict taken and predict not taken respectively. |
| 1767 | I'm not sure that branch hints actually do anything on loop |
| 1768 | and jcxz insns (JumpByte) for current Pentium4 chips. They |
| 1769 | may work in the future and it doesn't hurt to accept them |
| 1770 | now. */ |
| 1771 | if (l[0] == ',' && l[1] == 'p') |
| 1772 | { |
| 1773 | if (l[2] == 't') |
| 1774 | { |
| 1775 | if (!add_prefix (DS_PREFIX_OPCODE)) |
| 1776 | return NULL; |
| 1777 | l += 3; |
| 1778 | } |
| 1779 | else if (l[2] == 'n') |
| 1780 | { |
| 1781 | if (!add_prefix (CS_PREFIX_OPCODE)) |
| 1782 | return NULL; |
| 1783 | l += 3; |
| 1784 | } |
| 1785 | } |
| 1786 | } |
| 1787 | /* Any other comma loses. */ |
| 1788 | if (*l == ',') |
| 1789 | { |
| 1790 | as_bad (_("invalid character %s in mnemonic"), |
| 1791 | output_invalid (*l)); |
| 1792 | return NULL; |
| 1793 | } |
| 1794 | |
| 1795 | /* Check if instruction is supported on specified architecture. */ |
| 1796 | supported = 0; |
| 1797 | for (t = current_templates->start; t < current_templates->end; ++t) |
| 1798 | { |
| 1799 | if (!((t->cpu_flags & ~(Cpu64 | CpuNo64)) |
| 1800 | & ~(cpu_arch_flags & ~(Cpu64 | CpuNo64)))) |
| 1801 | supported |= 1; |
| 1802 | if (!(t->cpu_flags & (flag_code == CODE_64BIT ? CpuNo64 : Cpu64))) |
| 1803 | supported |= 2; |
| 1804 | } |
| 1805 | if (!(supported & 2)) |
| 1806 | { |
| 1807 | as_bad (flag_code == CODE_64BIT |
| 1808 | ? _("`%s' is not supported in 64-bit mode") |
| 1809 | : _("`%s' is only supported in 64-bit mode"), |
| 1810 | current_templates->start->name); |
| 1811 | return NULL; |
| 1812 | } |
| 1813 | if (!(supported & 1)) |
| 1814 | { |
| 1815 | as_warn (_("`%s' is not supported on `%s%s'"), |
| 1816 | current_templates->start->name, |
| 1817 | cpu_arch_name, |
| 1818 | cpu_sub_arch_name ? cpu_sub_arch_name : ""); |
| 1819 | } |
| 1820 | else if ((Cpu386 & ~cpu_arch_flags) && (flag_code != CODE_16BIT)) |
| 1821 | { |
| 1822 | as_warn (_("use .code16 to ensure correct addressing mode")); |
| 1823 | } |
| 1824 | |
| 1825 | /* Check for rep/repne without a string instruction. */ |
| 1826 | if (expecting_string_instruction) |
| 1827 | { |
| 1828 | static templates override; |
| 1829 | |
| 1830 | for (t = current_templates->start; t < current_templates->end; ++t) |
| 1831 | if (t->opcode_modifier & IsString) |
| 1832 | break; |
| 1833 | if (t >= current_templates->end) |
| 1834 | { |
| 1835 | as_bad (_("expecting string instruction after `%s'"), |
| 1836 | expecting_string_instruction); |
| 1837 | return NULL; |
| 1838 | } |
| 1839 | for (override.start = t; t < current_templates->end; ++t) |
| 1840 | if (!(t->opcode_modifier & IsString)) |
| 1841 | break; |
| 1842 | override.end = t; |
| 1843 | current_templates = &override; |
| 1844 | } |
| 1845 | |
| 1846 | return l; |
| 1847 | } |
| 1848 | |
| 1849 | static char * |
| 1850 | parse_operands (l, mnemonic) |
| 1851 | char *l; |
| 1852 | const char *mnemonic; |
| 1853 | { |
| 1854 | char *token_start; |
| 1855 | |
| 1856 | /* 1 if operand is pending after ','. */ |
| 1857 | unsigned int expecting_operand = 0; |
| 1858 | |
| 1859 | /* Non-zero if operand parens not balanced. */ |
| 1860 | unsigned int paren_not_balanced; |
| 1861 | |
| 1862 | while (*l != END_OF_INSN) |
| 1863 | { |
| 1864 | /* Skip optional white space before operand. */ |
| 1865 | if (is_space_char (*l)) |
| 1866 | ++l; |
| 1867 | if (!is_operand_char (*l) && *l != END_OF_INSN) |
| 1868 | { |
| 1869 | as_bad (_("invalid character %s before operand %d"), |
| 1870 | output_invalid (*l), |
| 1871 | i.operands + 1); |
| 1872 | return NULL; |
| 1873 | } |
| 1874 | token_start = l; /* after white space */ |
| 1875 | paren_not_balanced = 0; |
| 1876 | while (paren_not_balanced || *l != ',') |
| 1877 | { |
| 1878 | if (*l == END_OF_INSN) |
| 1879 | { |
| 1880 | if (paren_not_balanced) |
| 1881 | { |
| 1882 | if (!intel_syntax) |
| 1883 | as_bad (_("unbalanced parenthesis in operand %d."), |
| 1884 | i.operands + 1); |
| 1885 | else |
| 1886 | as_bad (_("unbalanced brackets in operand %d."), |
| 1887 | i.operands + 1); |
| 1888 | return NULL; |
| 1889 | } |
| 1890 | else |
| 1891 | break; /* we are done */ |
| 1892 | } |
| 1893 | else if (!is_operand_char (*l) && !is_space_char (*l)) |
| 1894 | { |
| 1895 | as_bad (_("invalid character %s in operand %d"), |
| 1896 | output_invalid (*l), |
| 1897 | i.operands + 1); |
| 1898 | return NULL; |
| 1899 | } |
| 1900 | if (!intel_syntax) |
| 1901 | { |
| 1902 | if (*l == '(') |
| 1903 | ++paren_not_balanced; |
| 1904 | if (*l == ')') |
| 1905 | --paren_not_balanced; |
| 1906 | } |
| 1907 | else |
| 1908 | { |
| 1909 | if (*l == '[') |
| 1910 | ++paren_not_balanced; |
| 1911 | if (*l == ']') |
| 1912 | --paren_not_balanced; |
| 1913 | } |
| 1914 | l++; |
| 1915 | } |
| 1916 | if (l != token_start) |
| 1917 | { /* Yes, we've read in another operand. */ |
| 1918 | unsigned int operand_ok; |
| 1919 | this_operand = i.operands++; |
| 1920 | if (i.operands > MAX_OPERANDS) |
| 1921 | { |
| 1922 | as_bad (_("spurious operands; (%d operands/instruction max)"), |
| 1923 | MAX_OPERANDS); |
| 1924 | return NULL; |
| 1925 | } |
| 1926 | /* Now parse operand adding info to 'i' as we go along. */ |
| 1927 | END_STRING_AND_SAVE (l); |
| 1928 | |
| 1929 | if (intel_syntax) |
| 1930 | operand_ok = |
| 1931 | i386_intel_operand (token_start, |
| 1932 | intel_float_operand (mnemonic)); |
| 1933 | else |
| 1934 | operand_ok = i386_operand (token_start); |
| 1935 | |
| 1936 | RESTORE_END_STRING (l); |
| 1937 | if (!operand_ok) |
| 1938 | return NULL; |
| 1939 | } |
| 1940 | else |
| 1941 | { |
| 1942 | if (expecting_operand) |
| 1943 | { |
| 1944 | expecting_operand_after_comma: |
| 1945 | as_bad (_("expecting operand after ','; got nothing")); |
| 1946 | return NULL; |
| 1947 | } |
| 1948 | if (*l == ',') |
| 1949 | { |
| 1950 | as_bad (_("expecting operand before ','; got nothing")); |
| 1951 | return NULL; |
| 1952 | } |
| 1953 | } |
| 1954 | |
| 1955 | /* Now *l must be either ',' or END_OF_INSN. */ |
| 1956 | if (*l == ',') |
| 1957 | { |
| 1958 | if (*++l == END_OF_INSN) |
| 1959 | { |
| 1960 | /* Just skip it, if it's \n complain. */ |
| 1961 | goto expecting_operand_after_comma; |
| 1962 | } |
| 1963 | expecting_operand = 1; |
| 1964 | } |
| 1965 | } |
| 1966 | return l; |
| 1967 | } |
| 1968 | |
| 1969 | static void |
| 1970 | swap_operands () |
| 1971 | { |
| 1972 | union i386_op temp_op; |
| 1973 | unsigned int temp_type; |
| 1974 | enum bfd_reloc_code_real temp_reloc; |
| 1975 | int xchg1 = 0; |
| 1976 | int xchg2 = 0; |
| 1977 | |
| 1978 | if (i.operands == 2) |
| 1979 | { |
| 1980 | xchg1 = 0; |
| 1981 | xchg2 = 1; |
| 1982 | } |
| 1983 | else if (i.operands == 3) |
| 1984 | { |
| 1985 | xchg1 = 0; |
| 1986 | xchg2 = 2; |
| 1987 | } |
| 1988 | temp_type = i.types[xchg2]; |
| 1989 | i.types[xchg2] = i.types[xchg1]; |
| 1990 | i.types[xchg1] = temp_type; |
| 1991 | temp_op = i.op[xchg2]; |
| 1992 | i.op[xchg2] = i.op[xchg1]; |
| 1993 | i.op[xchg1] = temp_op; |
| 1994 | temp_reloc = i.reloc[xchg2]; |
| 1995 | i.reloc[xchg2] = i.reloc[xchg1]; |
| 1996 | i.reloc[xchg1] = temp_reloc; |
| 1997 | |
| 1998 | if (i.mem_operands == 2) |
| 1999 | { |
| 2000 | const seg_entry *temp_seg; |
| 2001 | temp_seg = i.seg[0]; |
| 2002 | i.seg[0] = i.seg[1]; |
| 2003 | i.seg[1] = temp_seg; |
| 2004 | } |
| 2005 | } |
| 2006 | |
| 2007 | /* Try to ensure constant immediates are represented in the smallest |
| 2008 | opcode possible. */ |
| 2009 | static void |
| 2010 | optimize_imm () |
| 2011 | { |
| 2012 | char guess_suffix = 0; |
| 2013 | int op; |
| 2014 | |
| 2015 | if (i.suffix) |
| 2016 | guess_suffix = i.suffix; |
| 2017 | else if (i.reg_operands) |
| 2018 | { |
| 2019 | /* Figure out a suffix from the last register operand specified. |
| 2020 | We can't do this properly yet, ie. excluding InOutPortReg, |
| 2021 | but the following works for instructions with immediates. |
| 2022 | In any case, we can't set i.suffix yet. */ |
| 2023 | for (op = i.operands; --op >= 0;) |
| 2024 | if (i.types[op] & Reg) |
| 2025 | { |
| 2026 | if (i.types[op] & Reg8) |
| 2027 | guess_suffix = BYTE_MNEM_SUFFIX; |
| 2028 | else if (i.types[op] & Reg16) |
| 2029 | guess_suffix = WORD_MNEM_SUFFIX; |
| 2030 | else if (i.types[op] & Reg32) |
| 2031 | guess_suffix = LONG_MNEM_SUFFIX; |
| 2032 | else if (i.types[op] & Reg64) |
| 2033 | guess_suffix = QWORD_MNEM_SUFFIX; |
| 2034 | break; |
| 2035 | } |
| 2036 | } |
| 2037 | else if ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)) |
| 2038 | guess_suffix = WORD_MNEM_SUFFIX; |
| 2039 | |
| 2040 | for (op = i.operands; --op >= 0;) |
| 2041 | if (i.types[op] & Imm) |
| 2042 | { |
| 2043 | switch (i.op[op].imms->X_op) |
| 2044 | { |
| 2045 | case O_constant: |
| 2046 | /* If a suffix is given, this operand may be shortened. */ |
| 2047 | switch (guess_suffix) |
| 2048 | { |
| 2049 | case LONG_MNEM_SUFFIX: |
| 2050 | i.types[op] |= Imm32 | Imm64; |
| 2051 | break; |
| 2052 | case WORD_MNEM_SUFFIX: |
| 2053 | i.types[op] |= Imm16 | Imm32S | Imm32 | Imm64; |
| 2054 | break; |
| 2055 | case BYTE_MNEM_SUFFIX: |
| 2056 | i.types[op] |= Imm16 | Imm8 | Imm8S | Imm32S | Imm32 | Imm64; |
| 2057 | break; |
| 2058 | } |
| 2059 | |
| 2060 | /* If this operand is at most 16 bits, convert it |
| 2061 | to a signed 16 bit number before trying to see |
| 2062 | whether it will fit in an even smaller size. |
| 2063 | This allows a 16-bit operand such as $0xffe0 to |
| 2064 | be recognised as within Imm8S range. */ |
| 2065 | if ((i.types[op] & Imm16) |
| 2066 | && (i.op[op].imms->X_add_number & ~(offsetT) 0xffff) == 0) |
| 2067 | { |
| 2068 | i.op[op].imms->X_add_number = |
| 2069 | (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000); |
| 2070 | } |
| 2071 | if ((i.types[op] & Imm32) |
| 2072 | && ((i.op[op].imms->X_add_number & ~(((offsetT) 2 << 31) - 1)) |
| 2073 | == 0)) |
| 2074 | { |
| 2075 | i.op[op].imms->X_add_number = ((i.op[op].imms->X_add_number |
| 2076 | ^ ((offsetT) 1 << 31)) |
| 2077 | - ((offsetT) 1 << 31)); |
| 2078 | } |
| 2079 | i.types[op] |= smallest_imm_type (i.op[op].imms->X_add_number); |
| 2080 | |
| 2081 | /* We must avoid matching of Imm32 templates when 64bit |
| 2082 | only immediate is available. */ |
| 2083 | if (guess_suffix == QWORD_MNEM_SUFFIX) |
| 2084 | i.types[op] &= ~Imm32; |
| 2085 | break; |
| 2086 | |
| 2087 | case O_absent: |
| 2088 | case O_register: |
| 2089 | abort (); |
| 2090 | |
| 2091 | /* Symbols and expressions. */ |
| 2092 | default: |
| 2093 | /* Convert symbolic operand to proper sizes for matching, but don't |
| 2094 | prevent matching a set of insns that only supports sizes other |
| 2095 | than those matching the insn suffix. */ |
| 2096 | { |
| 2097 | unsigned int mask, allowed = 0; |
| 2098 | const template *t; |
| 2099 | |
| 2100 | for (t = current_templates->start; t < current_templates->end; ++t) |
| 2101 | allowed |= t->operand_types[op]; |
| 2102 | switch (guess_suffix) |
| 2103 | { |
| 2104 | case QWORD_MNEM_SUFFIX: |
| 2105 | mask = Imm64 | Imm32S; |
| 2106 | break; |
| 2107 | case LONG_MNEM_SUFFIX: |
| 2108 | mask = Imm32; |
| 2109 | break; |
| 2110 | case WORD_MNEM_SUFFIX: |
| 2111 | mask = Imm16; |
| 2112 | break; |
| 2113 | case BYTE_MNEM_SUFFIX: |
| 2114 | mask = Imm8; |
| 2115 | break; |
| 2116 | default: |
| 2117 | mask = 0; |
| 2118 | break; |
| 2119 | } |
| 2120 | if (mask & allowed) |
| 2121 | i.types[op] &= mask; |
| 2122 | } |
| 2123 | break; |
| 2124 | } |
| 2125 | } |
| 2126 | } |
| 2127 | |
| 2128 | /* Try to use the smallest displacement type too. */ |
| 2129 | static void |
| 2130 | optimize_disp () |
| 2131 | { |
| 2132 | int op; |
| 2133 | |
| 2134 | for (op = i.operands; --op >= 0;) |
| 2135 | if (i.types[op] & Disp) |
| 2136 | { |
| 2137 | if (i.op[op].disps->X_op == O_constant) |
| 2138 | { |
| 2139 | offsetT disp = i.op[op].disps->X_add_number; |
| 2140 | |
| 2141 | if ((i.types[op] & Disp16) |
| 2142 | && (disp & ~(offsetT) 0xffff) == 0) |
| 2143 | { |
| 2144 | /* If this operand is at most 16 bits, convert |
| 2145 | to a signed 16 bit number and don't use 64bit |
| 2146 | displacement. */ |
| 2147 | disp = (((disp & 0xffff) ^ 0x8000) - 0x8000); |
| 2148 | i.types[op] &= ~Disp64; |
| 2149 | } |
| 2150 | if ((i.types[op] & Disp32) |
| 2151 | && (disp & ~(((offsetT) 2 << 31) - 1)) == 0) |
| 2152 | { |
| 2153 | /* If this operand is at most 32 bits, convert |
| 2154 | to a signed 32 bit number and don't use 64bit |
| 2155 | displacement. */ |
| 2156 | disp &= (((offsetT) 2 << 31) - 1); |
| 2157 | disp = (disp ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31); |
| 2158 | i.types[op] &= ~Disp64; |
| 2159 | } |
| 2160 | if (!disp && (i.types[op] & BaseIndex)) |
| 2161 | { |
| 2162 | i.types[op] &= ~Disp; |
| 2163 | i.op[op].disps = 0; |
| 2164 | i.disp_operands--; |
| 2165 | } |
| 2166 | else if (flag_code == CODE_64BIT) |
| 2167 | { |
| 2168 | if (fits_in_signed_long (disp)) |
| 2169 | { |
| 2170 | i.types[op] &= ~Disp64; |
| 2171 | i.types[op] |= Disp32S; |
| 2172 | } |
| 2173 | if (fits_in_unsigned_long (disp)) |
| 2174 | i.types[op] |= Disp32; |
| 2175 | } |
| 2176 | if ((i.types[op] & (Disp32 | Disp32S | Disp16)) |
| 2177 | && fits_in_signed_byte (disp)) |
| 2178 | i.types[op] |= Disp8; |
| 2179 | } |
| 2180 | else |
| 2181 | /* We only support 64bit displacement on constants. */ |
| 2182 | i.types[op] &= ~Disp64; |
| 2183 | } |
| 2184 | } |
| 2185 | |
| 2186 | static int |
| 2187 | match_template () |
| 2188 | { |
| 2189 | /* Points to template once we've found it. */ |
| 2190 | const template *t; |
| 2191 | unsigned int overlap0, overlap1, overlap2; |
| 2192 | unsigned int found_reverse_match; |
| 2193 | int suffix_check; |
| 2194 | |
| 2195 | #define MATCH(overlap, given, template) \ |
| 2196 | ((overlap & ~JumpAbsolute) \ |
| 2197 | && (((given) & (BaseIndex | JumpAbsolute)) \ |
| 2198 | == ((overlap) & (BaseIndex | JumpAbsolute)))) |
| 2199 | |
| 2200 | /* If given types r0 and r1 are registers they must be of the same type |
| 2201 | unless the expected operand type register overlap is null. |
| 2202 | Note that Acc in a template matches every size of reg. */ |
| 2203 | #define CONSISTENT_REGISTER_MATCH(m0, g0, t0, m1, g1, t1) \ |
| 2204 | (((g0) & Reg) == 0 || ((g1) & Reg) == 0 \ |
| 2205 | || ((g0) & Reg) == ((g1) & Reg) \ |
| 2206 | || ((((m0) & Acc) ? Reg : (t0)) & (((m1) & Acc) ? Reg : (t1)) & Reg) == 0 ) |
| 2207 | |
| 2208 | overlap0 = 0; |
| 2209 | overlap1 = 0; |
| 2210 | overlap2 = 0; |
| 2211 | found_reverse_match = 0; |
| 2212 | suffix_check = (i.suffix == BYTE_MNEM_SUFFIX |
| 2213 | ? No_bSuf |
| 2214 | : (i.suffix == WORD_MNEM_SUFFIX |
| 2215 | ? No_wSuf |
| 2216 | : (i.suffix == SHORT_MNEM_SUFFIX |
| 2217 | ? No_sSuf |
| 2218 | : (i.suffix == LONG_MNEM_SUFFIX |
| 2219 | ? No_lSuf |
| 2220 | : (i.suffix == QWORD_MNEM_SUFFIX |
| 2221 | ? No_qSuf |
| 2222 | : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX |
| 2223 | ? No_xSuf : 0)))))); |
| 2224 | |
| 2225 | t = current_templates->start; |
| 2226 | if (i.suffix == QWORD_MNEM_SUFFIX |
| 2227 | && flag_code != CODE_64BIT |
| 2228 | && (intel_syntax |
| 2229 | ? !(t->opcode_modifier & IgnoreSize) |
| 2230 | && !intel_float_operand (t->name) |
| 2231 | : intel_float_operand (t->name) != 2) |
| 2232 | && (!(t->operand_types[0] & (RegMMX | RegXMM)) |
| 2233 | || !(t->operand_types[t->operands > 1] & (RegMMX | RegXMM))) |
| 2234 | && (t->base_opcode != 0x0fc7 |
| 2235 | || t->extension_opcode != 1 /* cmpxchg8b */)) |
| 2236 | t = current_templates->end; |
| 2237 | for (; t < current_templates->end; t++) |
| 2238 | { |
| 2239 | /* Must have right number of operands. */ |
| 2240 | if (i.operands != t->operands) |
| 2241 | continue; |
| 2242 | |
| 2243 | /* Check the suffix, except for some instructions in intel mode. */ |
| 2244 | if ((t->opcode_modifier & suffix_check) |
| 2245 | && !(intel_syntax |
| 2246 | && (t->opcode_modifier & IgnoreSize))) |
| 2247 | continue; |
| 2248 | |
| 2249 | /* Do not verify operands when there are none. */ |
| 2250 | else if (!t->operands) |
| 2251 | { |
| 2252 | if (t->cpu_flags & ~cpu_arch_flags) |
| 2253 | continue; |
| 2254 | /* We've found a match; break out of loop. */ |
| 2255 | break; |
| 2256 | } |
| 2257 | |
| 2258 | overlap0 = i.types[0] & t->operand_types[0]; |
| 2259 | switch (t->operands) |
| 2260 | { |
| 2261 | case 1: |
| 2262 | if (!MATCH (overlap0, i.types[0], t->operand_types[0])) |
| 2263 | continue; |
| 2264 | break; |
| 2265 | case 2: |
| 2266 | case 3: |
| 2267 | overlap1 = i.types[1] & t->operand_types[1]; |
| 2268 | if (!MATCH (overlap0, i.types[0], t->operand_types[0]) |
| 2269 | || !MATCH (overlap1, i.types[1], t->operand_types[1]) |
| 2270 | || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0], |
| 2271 | t->operand_types[0], |
| 2272 | overlap1, i.types[1], |
| 2273 | t->operand_types[1])) |
| 2274 | { |
| 2275 | /* Check if other direction is valid ... */ |
| 2276 | if ((t->opcode_modifier & (D | FloatD)) == 0) |
| 2277 | continue; |
| 2278 | |
| 2279 | /* Try reversing direction of operands. */ |
| 2280 | overlap0 = i.types[0] & t->operand_types[1]; |
| 2281 | overlap1 = i.types[1] & t->operand_types[0]; |
| 2282 | if (!MATCH (overlap0, i.types[0], t->operand_types[1]) |
| 2283 | || !MATCH (overlap1, i.types[1], t->operand_types[0]) |
| 2284 | || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0], |
| 2285 | t->operand_types[1], |
| 2286 | overlap1, i.types[1], |
| 2287 | t->operand_types[0])) |
| 2288 | { |
| 2289 | /* Does not match either direction. */ |
| 2290 | continue; |
| 2291 | } |
| 2292 | /* found_reverse_match holds which of D or FloatDR |
| 2293 | we've found. */ |
| 2294 | found_reverse_match = t->opcode_modifier & (D | FloatDR); |
| 2295 | } |
| 2296 | /* Found a forward 2 operand match here. */ |
| 2297 | else if (t->operands == 3) |
| 2298 | { |
| 2299 | /* Here we make use of the fact that there are no |
| 2300 | reverse match 3 operand instructions, and all 3 |
| 2301 | operand instructions only need to be checked for |
| 2302 | register consistency between operands 2 and 3. */ |
| 2303 | overlap2 = i.types[2] & t->operand_types[2]; |
| 2304 | if (!MATCH (overlap2, i.types[2], t->operand_types[2]) |
| 2305 | || !CONSISTENT_REGISTER_MATCH (overlap1, i.types[1], |
| 2306 | t->operand_types[1], |
| 2307 | overlap2, i.types[2], |
| 2308 | t->operand_types[2])) |
| 2309 | |
| 2310 | continue; |
| 2311 | } |
| 2312 | /* Found either forward/reverse 2 or 3 operand match here: |
| 2313 | slip through to break. */ |
| 2314 | } |
| 2315 | if (t->cpu_flags & ~cpu_arch_flags) |
| 2316 | { |
| 2317 | found_reverse_match = 0; |
| 2318 | continue; |
| 2319 | } |
| 2320 | /* We've found a match; break out of loop. */ |
| 2321 | break; |
| 2322 | } |
| 2323 | |
| 2324 | if (t == current_templates->end) |
| 2325 | { |
| 2326 | /* We found no match. */ |
| 2327 | as_bad (_("suffix or operands invalid for `%s'"), |
| 2328 | current_templates->start->name); |
| 2329 | return 0; |
| 2330 | } |
| 2331 | |
| 2332 | if (!quiet_warnings) |
| 2333 | { |
| 2334 | if (!intel_syntax |
| 2335 | && ((i.types[0] & JumpAbsolute) |
| 2336 | != (t->operand_types[0] & JumpAbsolute))) |
| 2337 | { |
| 2338 | as_warn (_("indirect %s without `*'"), t->name); |
| 2339 | } |
| 2340 | |
| 2341 | if ((t->opcode_modifier & (IsPrefix | IgnoreSize)) |
| 2342 | == (IsPrefix | IgnoreSize)) |
| 2343 | { |
| 2344 | /* Warn them that a data or address size prefix doesn't |
| 2345 | affect assembly of the next line of code. */ |
| 2346 | as_warn (_("stand-alone `%s' prefix"), t->name); |
| 2347 | } |
| 2348 | } |
| 2349 | |
| 2350 | /* Copy the template we found. */ |
| 2351 | i.tm = *t; |
| 2352 | if (found_reverse_match) |
| 2353 | { |
| 2354 | /* If we found a reverse match we must alter the opcode |
| 2355 | direction bit. found_reverse_match holds bits to change |
| 2356 | (different for int & float insns). */ |
| 2357 | |
| 2358 | i.tm.base_opcode ^= found_reverse_match; |
| 2359 | |
| 2360 | i.tm.operand_types[0] = t->operand_types[1]; |
| 2361 | i.tm.operand_types[1] = t->operand_types[0]; |
| 2362 | } |
| 2363 | |
| 2364 | return 1; |
| 2365 | } |
| 2366 | |
| 2367 | static int |
| 2368 | check_string () |
| 2369 | { |
| 2370 | int mem_op = (i.types[0] & AnyMem) ? 0 : 1; |
| 2371 | if ((i.tm.operand_types[mem_op] & EsSeg) != 0) |
| 2372 | { |
| 2373 | if (i.seg[0] != NULL && i.seg[0] != &es) |
| 2374 | { |
| 2375 | as_bad (_("`%s' operand %d must use `%%es' segment"), |
| 2376 | i.tm.name, |
| 2377 | mem_op + 1); |
| 2378 | return 0; |
| 2379 | } |
| 2380 | /* There's only ever one segment override allowed per instruction. |
| 2381 | This instruction possibly has a legal segment override on the |
| 2382 | second operand, so copy the segment to where non-string |
| 2383 | instructions store it, allowing common code. */ |
| 2384 | i.seg[0] = i.seg[1]; |
| 2385 | } |
| 2386 | else if ((i.tm.operand_types[mem_op + 1] & EsSeg) != 0) |
| 2387 | { |
| 2388 | if (i.seg[1] != NULL && i.seg[1] != &es) |
| 2389 | { |
| 2390 | as_bad (_("`%s' operand %d must use `%%es' segment"), |
| 2391 | i.tm.name, |
| 2392 | mem_op + 2); |
| 2393 | return 0; |
| 2394 | } |
| 2395 | } |
| 2396 | return 1; |
| 2397 | } |
| 2398 | |
| 2399 | static int |
| 2400 | process_suffix (void) |
| 2401 | { |
| 2402 | /* If matched instruction specifies an explicit instruction mnemonic |
| 2403 | suffix, use it. */ |
| 2404 | if (i.tm.opcode_modifier & (Size16 | Size32 | Size64)) |
| 2405 | { |
| 2406 | if (i.tm.opcode_modifier & Size16) |
| 2407 | i.suffix = WORD_MNEM_SUFFIX; |
| 2408 | else if (i.tm.opcode_modifier & Size64) |
| 2409 | i.suffix = QWORD_MNEM_SUFFIX; |
| 2410 | else |
| 2411 | i.suffix = LONG_MNEM_SUFFIX; |
| 2412 | } |
| 2413 | else if (i.reg_operands) |
| 2414 | { |
| 2415 | /* If there's no instruction mnemonic suffix we try to invent one |
| 2416 | based on register operands. */ |
| 2417 | if (!i.suffix) |
| 2418 | { |
| 2419 | /* We take i.suffix from the last register operand specified, |
| 2420 | Destination register type is more significant than source |
| 2421 | register type. */ |
| 2422 | int op; |
| 2423 | |
| 2424 | for (op = i.operands; --op >= 0;) |
| 2425 | if ((i.types[op] & Reg) |
| 2426 | && !(i.tm.operand_types[op] & InOutPortReg)) |
| 2427 | { |
| 2428 | i.suffix = ((i.types[op] & Reg8) ? BYTE_MNEM_SUFFIX : |
| 2429 | (i.types[op] & Reg16) ? WORD_MNEM_SUFFIX : |
| 2430 | (i.types[op] & Reg64) ? QWORD_MNEM_SUFFIX : |
| 2431 | LONG_MNEM_SUFFIX); |
| 2432 | break; |
| 2433 | } |
| 2434 | } |
| 2435 | else if (i.suffix == BYTE_MNEM_SUFFIX) |
| 2436 | { |
| 2437 | if (!check_byte_reg ()) |
| 2438 | return 0; |
| 2439 | } |
| 2440 | else if (i.suffix == LONG_MNEM_SUFFIX) |
| 2441 | { |
| 2442 | if (!check_long_reg ()) |
| 2443 | return 0; |
| 2444 | } |
| 2445 | else if (i.suffix == QWORD_MNEM_SUFFIX) |
| 2446 | { |
| 2447 | if (!check_qword_reg ()) |
| 2448 | return 0; |
| 2449 | } |
| 2450 | else if (i.suffix == WORD_MNEM_SUFFIX) |
| 2451 | { |
| 2452 | if (!check_word_reg ()) |
| 2453 | return 0; |
| 2454 | } |
| 2455 | else if (intel_syntax && (i.tm.opcode_modifier & IgnoreSize)) |
| 2456 | /* Do nothing if the instruction is going to ignore the prefix. */ |
| 2457 | ; |
| 2458 | else |
| 2459 | abort (); |
| 2460 | } |
| 2461 | else if ((i.tm.opcode_modifier & DefaultSize) |
| 2462 | && !i.suffix |
| 2463 | /* exclude fldenv/frstor/fsave/fstenv */ |
| 2464 | && (i.tm.opcode_modifier & No_sSuf)) |
| 2465 | { |
| 2466 | i.suffix = stackop_size; |
| 2467 | } |
| 2468 | else if (intel_syntax |
| 2469 | && !i.suffix |
| 2470 | && ((i.tm.operand_types[0] & JumpAbsolute) |
| 2471 | || (i.tm.opcode_modifier & (JumpByte|JumpInterSegment)) |
| 2472 | || (i.tm.base_opcode == 0x0f01 /* [ls][gi]dt */ |
| 2473 | && i.tm.extension_opcode <= 3))) |
| 2474 | { |
| 2475 | switch (flag_code) |
| 2476 | { |
| 2477 | case CODE_64BIT: |
| 2478 | if (!(i.tm.opcode_modifier & No_qSuf)) |
| 2479 | { |
| 2480 | i.suffix = QWORD_MNEM_SUFFIX; |
| 2481 | break; |
| 2482 | } |
| 2483 | case CODE_32BIT: |
| 2484 | if (!(i.tm.opcode_modifier & No_lSuf)) |
| 2485 | i.suffix = LONG_MNEM_SUFFIX; |
| 2486 | break; |
| 2487 | case CODE_16BIT: |
| 2488 | if (!(i.tm.opcode_modifier & No_wSuf)) |
| 2489 | i.suffix = WORD_MNEM_SUFFIX; |
| 2490 | break; |
| 2491 | } |
| 2492 | } |
| 2493 | |
| 2494 | if (!i.suffix) |
| 2495 | { |
| 2496 | if (!intel_syntax) |
| 2497 | { |
| 2498 | if (i.tm.opcode_modifier & W) |
| 2499 | { |
| 2500 | as_bad (_("no instruction mnemonic suffix given and no register operands; can't size instruction")); |
| 2501 | return 0; |
| 2502 | } |
| 2503 | } |
| 2504 | else |
| 2505 | { |
| 2506 | unsigned int suffixes = ~i.tm.opcode_modifier |
| 2507 | & (No_bSuf |
| 2508 | | No_wSuf |
| 2509 | | No_lSuf |
| 2510 | | No_sSuf |
| 2511 | | No_xSuf |
| 2512 | | No_qSuf); |
| 2513 | |
| 2514 | if ((i.tm.opcode_modifier & W) |
| 2515 | || ((suffixes & (suffixes - 1)) |
| 2516 | && !(i.tm.opcode_modifier & (DefaultSize | IgnoreSize)))) |
| 2517 | { |
| 2518 | as_bad (_("ambiguous operand size for `%s'"), i.tm.name); |
| 2519 | return 0; |
| 2520 | } |
| 2521 | } |
| 2522 | } |
| 2523 | |
| 2524 | /* Change the opcode based on the operand size given by i.suffix; |
| 2525 | We don't need to change things for byte insns. */ |
| 2526 | |
| 2527 | if (i.suffix && i.suffix != BYTE_MNEM_SUFFIX) |
| 2528 | { |
| 2529 | /* It's not a byte, select word/dword operation. */ |
| 2530 | if (i.tm.opcode_modifier & W) |
| 2531 | { |
| 2532 | if (i.tm.opcode_modifier & ShortForm) |
| 2533 | i.tm.base_opcode |= 8; |
| 2534 | else |
| 2535 | i.tm.base_opcode |= 1; |
| 2536 | } |
| 2537 | |
| 2538 | /* Now select between word & dword operations via the operand |
| 2539 | size prefix, except for instructions that will ignore this |
| 2540 | prefix anyway. */ |
| 2541 | if (i.suffix != QWORD_MNEM_SUFFIX |
| 2542 | && i.suffix != LONG_DOUBLE_MNEM_SUFFIX |
| 2543 | && !(i.tm.opcode_modifier & (IgnoreSize | FloatMF)) |
| 2544 | && ((i.suffix == LONG_MNEM_SUFFIX) == (flag_code == CODE_16BIT) |
| 2545 | || (flag_code == CODE_64BIT |
| 2546 | && (i.tm.opcode_modifier & JumpByte)))) |
| 2547 | { |
| 2548 | unsigned int prefix = DATA_PREFIX_OPCODE; |
| 2549 | |
| 2550 | if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */ |
| 2551 | prefix = ADDR_PREFIX_OPCODE; |
| 2552 | |
| 2553 | if (!add_prefix (prefix)) |
| 2554 | return 0; |
| 2555 | } |
| 2556 | |
| 2557 | /* Set mode64 for an operand. */ |
| 2558 | if (i.suffix == QWORD_MNEM_SUFFIX |
| 2559 | && flag_code == CODE_64BIT |
| 2560 | && (i.tm.opcode_modifier & NoRex64) == 0) |
| 2561 | i.rex |= REX_MODE64; |
| 2562 | |
| 2563 | /* Size floating point instruction. */ |
| 2564 | if (i.suffix == LONG_MNEM_SUFFIX) |
| 2565 | if (i.tm.opcode_modifier & FloatMF) |
| 2566 | i.tm.base_opcode ^= 4; |
| 2567 | } |
| 2568 | |
| 2569 | return 1; |
| 2570 | } |
| 2571 | |
| 2572 | static int |
| 2573 | check_byte_reg (void) |
| 2574 | { |
| 2575 | int op; |
| 2576 | |
| 2577 | for (op = i.operands; --op >= 0;) |
| 2578 | { |
| 2579 | /* If this is an eight bit register, it's OK. If it's the 16 or |
| 2580 | 32 bit version of an eight bit register, we will just use the |
| 2581 | low portion, and that's OK too. */ |
| 2582 | if (i.types[op] & Reg8) |
| 2583 | continue; |
| 2584 | |
| 2585 | /* movzx and movsx should not generate this warning. */ |
| 2586 | if (intel_syntax |
| 2587 | && (i.tm.base_opcode == 0xfb7 |
| 2588 | || i.tm.base_opcode == 0xfb6 |
| 2589 | || i.tm.base_opcode == 0x63 |
| 2590 | || i.tm.base_opcode == 0xfbe |
| 2591 | || i.tm.base_opcode == 0xfbf)) |
| 2592 | continue; |
| 2593 | |
| 2594 | if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4) |
| 2595 | { |
| 2596 | /* Prohibit these changes in the 64bit mode, since the |
| 2597 | lowering is more complicated. */ |
| 2598 | if (flag_code == CODE_64BIT |
| 2599 | && (i.tm.operand_types[op] & InOutPortReg) == 0) |
| 2600 | { |
| 2601 | as_bad (_("Incorrect register `%%%s' used with `%c' suffix"), |
| 2602 | i.op[op].regs->reg_name, |
| 2603 | i.suffix); |
| 2604 | return 0; |
| 2605 | } |
| 2606 | #if REGISTER_WARNINGS |
| 2607 | if (!quiet_warnings |
| 2608 | && (i.tm.operand_types[op] & InOutPortReg) == 0) |
| 2609 | as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"), |
| 2610 | (i.op[op].regs + (i.types[op] & Reg16 |
| 2611 | ? REGNAM_AL - REGNAM_AX |
| 2612 | : REGNAM_AL - REGNAM_EAX))->reg_name, |
| 2613 | i.op[op].regs->reg_name, |
| 2614 | i.suffix); |
| 2615 | #endif |
| 2616 | continue; |
| 2617 | } |
| 2618 | /* Any other register is bad. */ |
| 2619 | if (i.types[op] & (Reg | RegMMX | RegXMM |
| 2620 | | SReg2 | SReg3 |
| 2621 | | Control | Debug | Test |
| 2622 | | FloatReg | FloatAcc)) |
| 2623 | { |
| 2624 | as_bad (_("`%%%s' not allowed with `%s%c'"), |
| 2625 | i.op[op].regs->reg_name, |
| 2626 | i.tm.name, |
| 2627 | i.suffix); |
| 2628 | return 0; |
| 2629 | } |
| 2630 | } |
| 2631 | return 1; |
| 2632 | } |
| 2633 | |
| 2634 | static int |
| 2635 | check_long_reg () |
| 2636 | { |
| 2637 | int op; |
| 2638 | |
| 2639 | for (op = i.operands; --op >= 0;) |
| 2640 | /* Reject eight bit registers, except where the template requires |
| 2641 | them. (eg. movzb) */ |
| 2642 | if ((i.types[op] & Reg8) != 0 |
| 2643 | && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0) |
| 2644 | { |
| 2645 | as_bad (_("`%%%s' not allowed with `%s%c'"), |
| 2646 | i.op[op].regs->reg_name, |
| 2647 | i.tm.name, |
| 2648 | i.suffix); |
| 2649 | return 0; |
| 2650 | } |
| 2651 | /* Warn if the e prefix on a general reg is missing. */ |
| 2652 | else if ((!quiet_warnings || flag_code == CODE_64BIT) |
| 2653 | && (i.types[op] & Reg16) != 0 |
| 2654 | && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0) |
| 2655 | { |
| 2656 | /* Prohibit these changes in the 64bit mode, since the |
| 2657 | lowering is more complicated. */ |
| 2658 | if (flag_code == CODE_64BIT) |
| 2659 | { |
| 2660 | as_bad (_("Incorrect register `%%%s' used with `%c' suffix"), |
| 2661 | i.op[op].regs->reg_name, |
| 2662 | i.suffix); |
| 2663 | return 0; |
| 2664 | } |
| 2665 | #if REGISTER_WARNINGS |
| 2666 | else |
| 2667 | as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"), |
| 2668 | (i.op[op].regs + REGNAM_EAX - REGNAM_AX)->reg_name, |
| 2669 | i.op[op].regs->reg_name, |
| 2670 | i.suffix); |
| 2671 | #endif |
| 2672 | } |
| 2673 | /* Warn if the r prefix on a general reg is missing. */ |
| 2674 | else if ((i.types[op] & Reg64) != 0 |
| 2675 | && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0) |
| 2676 | { |
| 2677 | as_bad (_("Incorrect register `%%%s' used with `%c' suffix"), |
| 2678 | i.op[op].regs->reg_name, |
| 2679 | i.suffix); |
| 2680 | return 0; |
| 2681 | } |
| 2682 | return 1; |
| 2683 | } |
| 2684 | |
| 2685 | static int |
| 2686 | check_qword_reg () |
| 2687 | { |
| 2688 | int op; |
| 2689 | |
| 2690 | for (op = i.operands; --op >= 0; ) |
| 2691 | /* Reject eight bit registers, except where the template requires |
| 2692 | them. (eg. movzb) */ |
| 2693 | if ((i.types[op] & Reg8) != 0 |
| 2694 | && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0) |
| 2695 | { |
| 2696 | as_bad (_("`%%%s' not allowed with `%s%c'"), |
| 2697 | i.op[op].regs->reg_name, |
| 2698 | i.tm.name, |
| 2699 | i.suffix); |
| 2700 | return 0; |
| 2701 | } |
| 2702 | /* Warn if the e prefix on a general reg is missing. */ |
| 2703 | else if (((i.types[op] & Reg16) != 0 |
| 2704 | || (i.types[op] & Reg32) != 0) |
| 2705 | && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0) |
| 2706 | { |
| 2707 | /* Prohibit these changes in the 64bit mode, since the |
| 2708 | lowering is more complicated. */ |
| 2709 | as_bad (_("Incorrect register `%%%s' used with `%c' suffix"), |
| 2710 | i.op[op].regs->reg_name, |
| 2711 | i.suffix); |
| 2712 | return 0; |
| 2713 | } |
| 2714 | return 1; |
| 2715 | } |
| 2716 | |
| 2717 | static int |
| 2718 | check_word_reg () |
| 2719 | { |
| 2720 | int op; |
| 2721 | for (op = i.operands; --op >= 0;) |
| 2722 | /* Reject eight bit registers, except where the template requires |
| 2723 | them. (eg. movzb) */ |
| 2724 | if ((i.types[op] & Reg8) != 0 |
| 2725 | && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0) |
| 2726 | { |
| 2727 | as_bad (_("`%%%s' not allowed with `%s%c'"), |
| 2728 | i.op[op].regs->reg_name, |
| 2729 | i.tm.name, |
| 2730 | i.suffix); |
| 2731 | return 0; |
| 2732 | } |
| 2733 | /* Warn if the e prefix on a general reg is present. */ |
| 2734 | else if ((!quiet_warnings || flag_code == CODE_64BIT) |
| 2735 | && (i.types[op] & Reg32) != 0 |
| 2736 | && (i.tm.operand_types[op] & (Reg16 | Acc)) != 0) |
| 2737 | { |
| 2738 | /* Prohibit these changes in the 64bit mode, since the |
| 2739 | lowering is more complicated. */ |
| 2740 | if (flag_code == CODE_64BIT) |
| 2741 | { |
| 2742 | as_bad (_("Incorrect register `%%%s' used with `%c' suffix"), |
| 2743 | i.op[op].regs->reg_name, |
| 2744 | i.suffix); |
| 2745 | return 0; |
| 2746 | } |
| 2747 | else |
| 2748 | #if REGISTER_WARNINGS |
| 2749 | as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"), |
| 2750 | (i.op[op].regs + REGNAM_AX - REGNAM_EAX)->reg_name, |
| 2751 | i.op[op].regs->reg_name, |
| 2752 | i.suffix); |
| 2753 | #endif |
| 2754 | } |
| 2755 | return 1; |
| 2756 | } |
| 2757 | |
| 2758 | static int |
| 2759 | finalize_imm () |
| 2760 | { |
| 2761 | unsigned int overlap0, overlap1, overlap2; |
| 2762 | |
| 2763 | overlap0 = i.types[0] & i.tm.operand_types[0]; |
| 2764 | if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64)) |
| 2765 | && overlap0 != Imm8 && overlap0 != Imm8S |
| 2766 | && overlap0 != Imm16 && overlap0 != Imm32S |
| 2767 | && overlap0 != Imm32 && overlap0 != Imm64) |
| 2768 | { |
| 2769 | if (i.suffix) |
| 2770 | { |
| 2771 | overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX |
| 2772 | ? Imm8 | Imm8S |
| 2773 | : (i.suffix == WORD_MNEM_SUFFIX |
| 2774 | ? Imm16 |
| 2775 | : (i.suffix == QWORD_MNEM_SUFFIX |
| 2776 | ? Imm64 | Imm32S |
| 2777 | : Imm32))); |
| 2778 | } |
| 2779 | else if (overlap0 == (Imm16 | Imm32S | Imm32) |
| 2780 | || overlap0 == (Imm16 | Imm32) |
| 2781 | || overlap0 == (Imm16 | Imm32S)) |
| 2782 | { |
| 2783 | overlap0 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0) |
| 2784 | ? Imm16 : Imm32S); |
| 2785 | } |
| 2786 | if (overlap0 != Imm8 && overlap0 != Imm8S |
| 2787 | && overlap0 != Imm16 && overlap0 != Imm32S |
| 2788 | && overlap0 != Imm32 && overlap0 != Imm64) |
| 2789 | { |
| 2790 | as_bad (_("no instruction mnemonic suffix given; can't determine immediate size")); |
| 2791 | return 0; |
| 2792 | } |
| 2793 | } |
| 2794 | i.types[0] = overlap0; |
| 2795 | |
| 2796 | overlap1 = i.types[1] & i.tm.operand_types[1]; |
| 2797 | if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32 | Imm64)) |
| 2798 | && overlap1 != Imm8 && overlap1 != Imm8S |
| 2799 | && overlap1 != Imm16 && overlap1 != Imm32S |
| 2800 | && overlap1 != Imm32 && overlap1 != Imm64) |
| 2801 | { |
| 2802 | if (i.suffix) |
| 2803 | { |
| 2804 | overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX |
| 2805 | ? Imm8 | Imm8S |
| 2806 | : (i.suffix == WORD_MNEM_SUFFIX |
| 2807 | ? Imm16 |
| 2808 | : (i.suffix == QWORD_MNEM_SUFFIX |
| 2809 | ? Imm64 | Imm32S |
| 2810 | : Imm32))); |
| 2811 | } |
| 2812 | else if (overlap1 == (Imm16 | Imm32 | Imm32S) |
| 2813 | || overlap1 == (Imm16 | Imm32) |
| 2814 | || overlap1 == (Imm16 | Imm32S)) |
| 2815 | { |
| 2816 | overlap1 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0) |
| 2817 | ? Imm16 : Imm32S); |
| 2818 | } |
| 2819 | if (overlap1 != Imm8 && overlap1 != Imm8S |
| 2820 | && overlap1 != Imm16 && overlap1 != Imm32S |
| 2821 | && overlap1 != Imm32 && overlap1 != Imm64) |
| 2822 | { |
| 2823 | as_bad (_("no instruction mnemonic suffix given; can't determine immediate size %x %c"),overlap1, i.suffix); |
| 2824 | return 0; |
| 2825 | } |
| 2826 | } |
| 2827 | i.types[1] = overlap1; |
| 2828 | |
| 2829 | overlap2 = i.types[2] & i.tm.operand_types[2]; |
| 2830 | assert ((overlap2 & Imm) == 0); |
| 2831 | i.types[2] = overlap2; |
| 2832 | |
| 2833 | return 1; |
| 2834 | } |
| 2835 | |
| 2836 | static int |
| 2837 | process_operands () |
| 2838 | { |
| 2839 | /* Default segment register this instruction will use for memory |
| 2840 | accesses. 0 means unknown. This is only for optimizing out |
| 2841 | unnecessary segment overrides. */ |
| 2842 | const seg_entry *default_seg = 0; |
| 2843 | |
| 2844 | /* The imul $imm, %reg instruction is converted into |
| 2845 | imul $imm, %reg, %reg, and the clr %reg instruction |
| 2846 | is converted into xor %reg, %reg. */ |
| 2847 | if (i.tm.opcode_modifier & regKludge) |
| 2848 | { |
| 2849 | unsigned int first_reg_op = (i.types[0] & Reg) ? 0 : 1; |
| 2850 | /* Pretend we saw the extra register operand. */ |
| 2851 | assert (i.op[first_reg_op + 1].regs == 0); |
| 2852 | i.op[first_reg_op + 1].regs = i.op[first_reg_op].regs; |
| 2853 | i.types[first_reg_op + 1] = i.types[first_reg_op]; |
| 2854 | i.reg_operands = 2; |
| 2855 | } |
| 2856 | |
| 2857 | if (i.tm.opcode_modifier & ShortForm) |
| 2858 | { |
| 2859 | /* The register or float register operand is in operand 0 or 1. */ |
| 2860 | unsigned int op = (i.types[0] & (Reg | FloatReg)) ? 0 : 1; |
| 2861 | /* Register goes in low 3 bits of opcode. */ |
| 2862 | i.tm.base_opcode |= i.op[op].regs->reg_num; |
| 2863 | if ((i.op[op].regs->reg_flags & RegRex) != 0) |
| 2864 | i.rex |= REX_EXTZ; |
| 2865 | if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0) |
| 2866 | { |
| 2867 | /* Warn about some common errors, but press on regardless. |
| 2868 | The first case can be generated by gcc (<= 2.8.1). */ |
| 2869 | if (i.operands == 2) |
| 2870 | { |
| 2871 | /* Reversed arguments on faddp, fsubp, etc. */ |
| 2872 | as_warn (_("translating to `%s %%%s,%%%s'"), i.tm.name, |
| 2873 | i.op[1].regs->reg_name, |
| 2874 | i.op[0].regs->reg_name); |
| 2875 | } |
| 2876 | else |
| 2877 | { |
| 2878 | /* Extraneous `l' suffix on fp insn. */ |
| 2879 | as_warn (_("translating to `%s %%%s'"), i.tm.name, |
| 2880 | i.op[0].regs->reg_name); |
| 2881 | } |
| 2882 | } |
| 2883 | } |
| 2884 | else if (i.tm.opcode_modifier & Modrm) |
| 2885 | { |
| 2886 | /* The opcode is completed (modulo i.tm.extension_opcode which |
| 2887 | must be put into the modrm byte). Now, we make the modrm and |
| 2888 | index base bytes based on all the info we've collected. */ |
| 2889 | |
| 2890 | default_seg = build_modrm_byte (); |
| 2891 | } |
| 2892 | else if (i.tm.opcode_modifier & (Seg2ShortForm | Seg3ShortForm)) |
| 2893 | { |
| 2894 | if (i.tm.base_opcode == POP_SEG_SHORT |
| 2895 | && i.op[0].regs->reg_num == 1) |
| 2896 | { |
| 2897 | as_bad (_("you can't `pop %%cs'")); |
| 2898 | return 0; |
| 2899 | } |
| 2900 | i.tm.base_opcode |= (i.op[0].regs->reg_num << 3); |
| 2901 | if ((i.op[0].regs->reg_flags & RegRex) != 0) |
| 2902 | i.rex |= REX_EXTZ; |
| 2903 | } |
| 2904 | else if ((i.tm.base_opcode & ~(D | W)) == MOV_AX_DISP32) |
| 2905 | { |
| 2906 | default_seg = &ds; |
| 2907 | } |
| 2908 | else if ((i.tm.opcode_modifier & IsString) != 0) |
| 2909 | { |
| 2910 | /* For the string instructions that allow a segment override |
| 2911 | on one of their operands, the default segment is ds. */ |
| 2912 | default_seg = &ds; |
| 2913 | } |
| 2914 | |
| 2915 | if ((i.tm.base_opcode == 0x8d /* lea */ |
| 2916 | || (i.tm.cpu_flags & CpuSVME)) |
| 2917 | && i.seg[0] && !quiet_warnings) |
| 2918 | as_warn (_("segment override on `%s' is ineffectual"), i.tm.name); |
| 2919 | |
| 2920 | /* If a segment was explicitly specified, and the specified segment |
| 2921 | is not the default, use an opcode prefix to select it. If we |
| 2922 | never figured out what the default segment is, then default_seg |
| 2923 | will be zero at this point, and the specified segment prefix will |
| 2924 | always be used. */ |
| 2925 | if ((i.seg[0]) && (i.seg[0] != default_seg)) |
| 2926 | { |
| 2927 | if (!add_prefix (i.seg[0]->seg_prefix)) |
| 2928 | return 0; |
| 2929 | } |
| 2930 | return 1; |
| 2931 | } |
| 2932 | |
| 2933 | static const seg_entry * |
| 2934 | build_modrm_byte () |
| 2935 | { |
| 2936 | const seg_entry *default_seg = 0; |
| 2937 | |
| 2938 | /* i.reg_operands MUST be the number of real register operands; |
| 2939 | implicit registers do not count. */ |
| 2940 | if (i.reg_operands == 2) |
| 2941 | { |
| 2942 | unsigned int source, dest; |
| 2943 | source = ((i.types[0] |
| 2944 | & (Reg | RegMMX | RegXMM |
| 2945 | | SReg2 | SReg3 |
| 2946 | | Control | Debug | Test)) |
| 2947 | ? 0 : 1); |
| 2948 | dest = source + 1; |
| 2949 | |
| 2950 | i.rm.mode = 3; |
| 2951 | /* One of the register operands will be encoded in the i.tm.reg |
| 2952 | field, the other in the combined i.tm.mode and i.tm.regmem |
| 2953 | fields. If no form of this instruction supports a memory |
| 2954 | destination operand, then we assume the source operand may |
| 2955 | sometimes be a memory operand and so we need to store the |
| 2956 | destination in the i.rm.reg field. */ |
| 2957 | if ((i.tm.operand_types[dest] & AnyMem) == 0) |
| 2958 | { |
| 2959 | i.rm.reg = i.op[dest].regs->reg_num; |
| 2960 | i.rm.regmem = i.op[source].regs->reg_num; |
| 2961 | if ((i.op[dest].regs->reg_flags & RegRex) != 0) |
| 2962 | i.rex |= REX_EXTX; |
| 2963 | if ((i.op[source].regs->reg_flags & RegRex) != 0) |
| 2964 | i.rex |= REX_EXTZ; |
| 2965 | } |
| 2966 | else |
| 2967 | { |
| 2968 | i.rm.reg = i.op[source].regs->reg_num; |
| 2969 | i.rm.regmem = i.op[dest].regs->reg_num; |
| 2970 | if ((i.op[dest].regs->reg_flags & RegRex) != 0) |
| 2971 | i.rex |= REX_EXTZ; |
| 2972 | if ((i.op[source].regs->reg_flags & RegRex) != 0) |
| 2973 | i.rex |= REX_EXTX; |
| 2974 | } |
| 2975 | if (flag_code != CODE_64BIT && (i.rex & (REX_EXTX | REX_EXTZ))) |
| 2976 | { |
| 2977 | if (!((i.types[0] | i.types[1]) & Control)) |
| 2978 | abort (); |
| 2979 | i.rex &= ~(REX_EXTX | REX_EXTZ); |
| 2980 | add_prefix (LOCK_PREFIX_OPCODE); |
| 2981 | } |
| 2982 | } |
| 2983 | else |
| 2984 | { /* If it's not 2 reg operands... */ |
| 2985 | if (i.mem_operands) |
| 2986 | { |
| 2987 | unsigned int fake_zero_displacement = 0; |
| 2988 | unsigned int op = ((i.types[0] & AnyMem) |
| 2989 | ? 0 |
| 2990 | : (i.types[1] & AnyMem) ? 1 : 2); |
| 2991 | |
| 2992 | default_seg = &ds; |
| 2993 | |
| 2994 | if (i.base_reg == 0) |
| 2995 | { |
| 2996 | i.rm.mode = 0; |
| 2997 | if (!i.disp_operands) |
| 2998 | fake_zero_displacement = 1; |
| 2999 | if (i.index_reg == 0) |
| 3000 | { |
| 3001 | /* Operand is just <disp> */ |
| 3002 | if (flag_code == CODE_64BIT) |
| 3003 | { |
| 3004 | /* 64bit mode overwrites the 32bit absolute |
| 3005 | addressing by RIP relative addressing and |
| 3006 | absolute addressing is encoded by one of the |
| 3007 | redundant SIB forms. */ |
| 3008 | i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING; |
| 3009 | i.sib.base = NO_BASE_REGISTER; |
| 3010 | i.sib.index = NO_INDEX_REGISTER; |
| 3011 | i.types[op] = ((i.prefix[ADDR_PREFIX] == 0) ? Disp32S : Disp32); |
| 3012 | } |
| 3013 | else if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0)) |
| 3014 | { |
| 3015 | i.rm.regmem = NO_BASE_REGISTER_16; |
| 3016 | i.types[op] = Disp16; |
| 3017 | } |
| 3018 | else |
| 3019 | { |
| 3020 | i.rm.regmem = NO_BASE_REGISTER; |
| 3021 | i.types[op] = Disp32; |
| 3022 | } |
| 3023 | } |
| 3024 | else /* !i.base_reg && i.index_reg */ |
| 3025 | { |
| 3026 | i.sib.index = i.index_reg->reg_num; |
| 3027 | i.sib.base = NO_BASE_REGISTER; |
| 3028 | i.sib.scale = i.log2_scale_factor; |
| 3029 | i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING; |
| 3030 | i.types[op] &= ~Disp; |
| 3031 | if (flag_code != CODE_64BIT) |
| 3032 | i.types[op] |= Disp32; /* Must be 32 bit */ |
| 3033 | else |
| 3034 | i.types[op] |= Disp32S; |
| 3035 | if ((i.index_reg->reg_flags & RegRex) != 0) |
| 3036 | i.rex |= REX_EXTY; |
| 3037 | } |
| 3038 | } |
| 3039 | /* RIP addressing for 64bit mode. */ |
| 3040 | else if (i.base_reg->reg_type == BaseIndex) |
| 3041 | { |
| 3042 | i.rm.regmem = NO_BASE_REGISTER; |
| 3043 | i.types[op] &= ~ Disp; |
| 3044 | i.types[op] |= Disp32S; |
| 3045 | i.flags[op] = Operand_PCrel; |
| 3046 | if (! i.disp_operands) |
| 3047 | fake_zero_displacement = 1; |
| 3048 | } |
| 3049 | else if (i.base_reg->reg_type & Reg16) |
| 3050 | { |
| 3051 | switch (i.base_reg->reg_num) |
| 3052 | { |
| 3053 | case 3: /* (%bx) */ |
| 3054 | if (i.index_reg == 0) |
| 3055 | i.rm.regmem = 7; |
| 3056 | else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */ |
| 3057 | i.rm.regmem = i.index_reg->reg_num - 6; |
| 3058 | break; |
| 3059 | case 5: /* (%bp) */ |
| 3060 | default_seg = &ss; |
| 3061 | if (i.index_reg == 0) |
| 3062 | { |
| 3063 | i.rm.regmem = 6; |
| 3064 | if ((i.types[op] & Disp) == 0) |
| 3065 | { |
| 3066 | /* fake (%bp) into 0(%bp) */ |
| 3067 | i.types[op] |= Disp8; |
| 3068 | fake_zero_displacement = 1; |
| 3069 | } |
| 3070 | } |
| 3071 | else /* (%bp,%si) -> 2, or (%bp,%di) -> 3 */ |
| 3072 | i.rm.regmem = i.index_reg->reg_num - 6 + 2; |
| 3073 | break; |
| 3074 | default: /* (%si) -> 4 or (%di) -> 5 */ |
| 3075 | i.rm.regmem = i.base_reg->reg_num - 6 + 4; |
| 3076 | } |
| 3077 | i.rm.mode = mode_from_disp_size (i.types[op]); |
| 3078 | } |
| 3079 | else /* i.base_reg and 32/64 bit mode */ |
| 3080 | { |
| 3081 | if (flag_code == CODE_64BIT |
| 3082 | && (i.types[op] & Disp)) |
| 3083 | i.types[op] = (i.types[op] & Disp8) | (i.prefix[ADDR_PREFIX] == 0 ? Disp32S : Disp32); |
| 3084 | |
| 3085 | i.rm.regmem = i.base_reg->reg_num; |
| 3086 | if ((i.base_reg->reg_flags & RegRex) != 0) |
| 3087 | i.rex |= REX_EXTZ; |
| 3088 | i.sib.base = i.base_reg->reg_num; |
| 3089 | /* x86-64 ignores REX prefix bit here to avoid decoder |
| 3090 | complications. */ |
| 3091 | if ((i.base_reg->reg_num & 7) == EBP_REG_NUM) |
| 3092 | { |
| 3093 | default_seg = &ss; |
| 3094 | if (i.disp_operands == 0) |
| 3095 | { |
| 3096 | fake_zero_displacement = 1; |
| 3097 | i.types[op] |= Disp8; |
| 3098 | } |
| 3099 | } |
| 3100 | else if (i.base_reg->reg_num == ESP_REG_NUM) |
| 3101 | { |
| 3102 | default_seg = &ss; |
| 3103 | } |
| 3104 | i.sib.scale = i.log2_scale_factor; |
| 3105 | if (i.index_reg == 0) |
| 3106 | { |
| 3107 | /* <disp>(%esp) becomes two byte modrm with no index |
| 3108 | register. We've already stored the code for esp |
| 3109 | in i.rm.regmem ie. ESCAPE_TO_TWO_BYTE_ADDRESSING. |
| 3110 | Any base register besides %esp will not use the |
| 3111 | extra modrm byte. */ |
| 3112 | i.sib.index = NO_INDEX_REGISTER; |
| 3113 | #if !SCALE1_WHEN_NO_INDEX |
| 3114 | /* Another case where we force the second modrm byte. */ |
| 3115 | if (i.log2_scale_factor) |
| 3116 | i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING; |
| 3117 | #endif |
| 3118 | } |
| 3119 | else |
| 3120 | { |
| 3121 | i.sib.index = i.index_reg->reg_num; |
| 3122 | i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING; |
| 3123 | if ((i.index_reg->reg_flags & RegRex) != 0) |
| 3124 | i.rex |= REX_EXTY; |
| 3125 | } |
| 3126 | i.rm.mode = mode_from_disp_size (i.types[op]); |
| 3127 | } |
| 3128 | |
| 3129 | if (fake_zero_displacement) |
| 3130 | { |
| 3131 | /* Fakes a zero displacement assuming that i.types[op] |
| 3132 | holds the correct displacement size. */ |
| 3133 | expressionS *exp; |
| 3134 | |
| 3135 | assert (i.op[op].disps == 0); |
| 3136 | exp = &disp_expressions[i.disp_operands++]; |
| 3137 | i.op[op].disps = exp; |
| 3138 | exp->X_op = O_constant; |
| 3139 | exp->X_add_number = 0; |
| 3140 | exp->X_add_symbol = (symbolS *) 0; |
| 3141 | exp->X_op_symbol = (symbolS *) 0; |
| 3142 | } |
| 3143 | } |
| 3144 | |
| 3145 | /* Fill in i.rm.reg or i.rm.regmem field with register operand |
| 3146 | (if any) based on i.tm.extension_opcode. Again, we must be |
| 3147 | careful to make sure that segment/control/debug/test/MMX |
| 3148 | registers are coded into the i.rm.reg field. */ |
| 3149 | if (i.reg_operands) |
| 3150 | { |
| 3151 | unsigned int op = |
| 3152 | ((i.types[0] |
| 3153 | & (Reg | RegMMX | RegXMM |
| 3154 | | SReg2 | SReg3 |
| 3155 | | Control | Debug | Test)) |
| 3156 | ? 0 |
| 3157 | : ((i.types[1] |
| 3158 | & (Reg | RegMMX | RegXMM |
| 3159 | | SReg2 | SReg3 |
| 3160 | | Control | Debug | Test)) |
| 3161 | ? 1 |
| 3162 | : 2)); |
| 3163 | /* If there is an extension opcode to put here, the register |
| 3164 | number must be put into the regmem field. */ |
| 3165 | if (i.tm.extension_opcode != None) |
| 3166 | { |
| 3167 | i.rm.regmem = i.op[op].regs->reg_num; |
| 3168 | if ((i.op[op].regs->reg_flags & RegRex) != 0) |
| 3169 | i.rex |= REX_EXTZ; |
| 3170 | } |
| 3171 | else |
| 3172 | { |
| 3173 | i.rm.reg = i.op[op].regs->reg_num; |
| 3174 | if ((i.op[op].regs->reg_flags & RegRex) != 0) |
| 3175 | i.rex |= REX_EXTX; |
| 3176 | } |
| 3177 | |
| 3178 | /* Now, if no memory operand has set i.rm.mode = 0, 1, 2 we |
| 3179 | must set it to 3 to indicate this is a register operand |
| 3180 | in the regmem field. */ |
| 3181 | if (!i.mem_operands) |
| 3182 | i.rm.mode = 3; |
| 3183 | } |
| 3184 | |
| 3185 | /* Fill in i.rm.reg field with extension opcode (if any). */ |
| 3186 | if (i.tm.extension_opcode != None) |
| 3187 | i.rm.reg = i.tm.extension_opcode; |
| 3188 | } |
| 3189 | return default_seg; |
| 3190 | } |
| 3191 | |
| 3192 | static void |
| 3193 | output_branch () |
| 3194 | { |
| 3195 | char *p; |
| 3196 | int code16; |
| 3197 | int prefix; |
| 3198 | relax_substateT subtype; |
| 3199 | symbolS *sym; |
| 3200 | offsetT off; |
| 3201 | |
| 3202 | code16 = 0; |
| 3203 | if (flag_code == CODE_16BIT) |
| 3204 | code16 = CODE16; |
| 3205 | |
| 3206 | prefix = 0; |
| 3207 | if (i.prefix[DATA_PREFIX] != 0) |
| 3208 | { |
| 3209 | prefix = 1; |
| 3210 | i.prefixes -= 1; |
| 3211 | code16 ^= CODE16; |
| 3212 | } |
| 3213 | /* Pentium4 branch hints. */ |
| 3214 | if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */ |
| 3215 | || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */) |
| 3216 | { |
| 3217 | prefix++; |
| 3218 | i.prefixes--; |
| 3219 | } |
| 3220 | if (i.prefix[REX_PREFIX] != 0) |
| 3221 | { |
| 3222 | prefix++; |
| 3223 | i.prefixes--; |
| 3224 | } |
| 3225 | |
| 3226 | if (i.prefixes != 0 && !intel_syntax) |
| 3227 | as_warn (_("skipping prefixes on this instruction")); |
| 3228 | |
| 3229 | /* It's always a symbol; End frag & setup for relax. |
| 3230 | Make sure there is enough room in this frag for the largest |
| 3231 | instruction we may generate in md_convert_frag. This is 2 |
| 3232 | bytes for the opcode and room for the prefix and largest |
| 3233 | displacement. */ |
| 3234 | frag_grow (prefix + 2 + 4); |
| 3235 | /* Prefix and 1 opcode byte go in fr_fix. */ |
| 3236 | p = frag_more (prefix + 1); |
| 3237 | if (i.prefix[DATA_PREFIX] != 0) |
| 3238 | *p++ = DATA_PREFIX_OPCODE; |
| 3239 | if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE |
| 3240 | || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE) |
| 3241 | *p++ = i.prefix[SEG_PREFIX]; |
| 3242 | if (i.prefix[REX_PREFIX] != 0) |
| 3243 | *p++ = i.prefix[REX_PREFIX]; |
| 3244 | *p = i.tm.base_opcode; |
| 3245 | |
| 3246 | if ((unsigned char) *p == JUMP_PC_RELATIVE) |
| 3247 | subtype = ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL); |
| 3248 | else if ((cpu_arch_flags & Cpu386) != 0) |
| 3249 | subtype = ENCODE_RELAX_STATE (COND_JUMP, SMALL); |
| 3250 | else |
| 3251 | subtype = ENCODE_RELAX_STATE (COND_JUMP86, SMALL); |
| 3252 | subtype |= code16; |
| 3253 | |
| 3254 | sym = i.op[0].disps->X_add_symbol; |
| 3255 | off = i.op[0].disps->X_add_number; |
| 3256 | |
| 3257 | if (i.op[0].disps->X_op != O_constant |
| 3258 | && i.op[0].disps->X_op != O_symbol) |
| 3259 | { |
| 3260 | /* Handle complex expressions. */ |
| 3261 | sym = make_expr_symbol (i.op[0].disps); |
| 3262 | off = 0; |
| 3263 | } |
| 3264 | |
| 3265 | /* 1 possible extra opcode + 4 byte displacement go in var part. |
| 3266 | Pass reloc in fr_var. */ |
| 3267 | frag_var (rs_machine_dependent, 5, i.reloc[0], subtype, sym, off, p); |
| 3268 | } |
| 3269 | |
| 3270 | static void |
| 3271 | output_jump () |
| 3272 | { |
| 3273 | char *p; |
| 3274 | int size; |
| 3275 | fixS *fixP; |
| 3276 | |
| 3277 | if (i.tm.opcode_modifier & JumpByte) |
| 3278 | { |
| 3279 | /* This is a loop or jecxz type instruction. */ |
| 3280 | size = 1; |
| 3281 | if (i.prefix[ADDR_PREFIX] != 0) |
| 3282 | { |
| 3283 | FRAG_APPEND_1_CHAR (ADDR_PREFIX_OPCODE); |
| 3284 | i.prefixes -= 1; |
| 3285 | } |
| 3286 | /* Pentium4 branch hints. */ |
| 3287 | if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */ |
| 3288 | || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */) |
| 3289 | { |
| 3290 | FRAG_APPEND_1_CHAR (i.prefix[SEG_PREFIX]); |
| 3291 | i.prefixes--; |
| 3292 | } |
| 3293 | } |
| 3294 | else |
| 3295 | { |
| 3296 | int code16; |
| 3297 | |
| 3298 | code16 = 0; |
| 3299 | if (flag_code == CODE_16BIT) |
| 3300 | code16 = CODE16; |
| 3301 | |
| 3302 | if (i.prefix[DATA_PREFIX] != 0) |
| 3303 | { |
| 3304 | FRAG_APPEND_1_CHAR (DATA_PREFIX_OPCODE); |
| 3305 | i.prefixes -= 1; |
| 3306 | code16 ^= CODE16; |
| 3307 | } |
| 3308 | |
| 3309 | size = 4; |
| 3310 | if (code16) |
| 3311 | size = 2; |
| 3312 | } |
| 3313 | |
| 3314 | if (i.prefix[REX_PREFIX] != 0) |
| 3315 | { |
| 3316 | FRAG_APPEND_1_CHAR (i.prefix[REX_PREFIX]); |
| 3317 | i.prefixes -= 1; |
| 3318 | } |
| 3319 | |
| 3320 | if (i.prefixes != 0 && !intel_syntax) |
| 3321 | as_warn (_("skipping prefixes on this instruction")); |
| 3322 | |
| 3323 | p = frag_more (1 + size); |
| 3324 | *p++ = i.tm.base_opcode; |
| 3325 | |
| 3326 | fixP = fix_new_exp (frag_now, p - frag_now->fr_literal, size, |
| 3327 | i.op[0].disps, 1, reloc (size, 1, 1, i.reloc[0])); |
| 3328 | |
| 3329 | /* All jumps handled here are signed, but don't use a signed limit |
| 3330 | check for 32 and 16 bit jumps as we want to allow wrap around at |
| 3331 | 4G and 64k respectively. */ |
| 3332 | if (size == 1) |
| 3333 | fixP->fx_signed = 1; |
| 3334 | } |
| 3335 | |
| 3336 | static void |
| 3337 | output_interseg_jump () |
| 3338 | { |
| 3339 | char *p; |
| 3340 | int size; |
| 3341 | int prefix; |
| 3342 | int code16; |
| 3343 | |
| 3344 | code16 = 0; |
| 3345 | if (flag_code == CODE_16BIT) |
| 3346 | code16 = CODE16; |
| 3347 | |
| 3348 | prefix = 0; |
| 3349 | if (i.prefix[DATA_PREFIX] != 0) |
| 3350 | { |
| 3351 | prefix = 1; |
| 3352 | i.prefixes -= 1; |
| 3353 | code16 ^= CODE16; |
| 3354 | } |
| 3355 | if (i.prefix[REX_PREFIX] != 0) |
| 3356 | { |
| 3357 | prefix++; |
| 3358 | i.prefixes -= 1; |
| 3359 | } |
| 3360 | |
| 3361 | size = 4; |
| 3362 | if (code16) |
| 3363 | size = 2; |
| 3364 | |
| 3365 | if (i.prefixes != 0 && !intel_syntax) |
| 3366 | as_warn (_("skipping prefixes on this instruction")); |
| 3367 | |
| 3368 | /* 1 opcode; 2 segment; offset */ |
| 3369 | p = frag_more (prefix + 1 + 2 + size); |
| 3370 | |
| 3371 | if (i.prefix[DATA_PREFIX] != 0) |
| 3372 | *p++ = DATA_PREFIX_OPCODE; |
| 3373 | |
| 3374 | if (i.prefix[REX_PREFIX] != 0) |
| 3375 | *p++ = i.prefix[REX_PREFIX]; |
| 3376 | |
| 3377 | *p++ = i.tm.base_opcode; |
| 3378 | if (i.op[1].imms->X_op == O_constant) |
| 3379 | { |
| 3380 | offsetT n = i.op[1].imms->X_add_number; |
| 3381 | |
| 3382 | if (size == 2 |
| 3383 | && !fits_in_unsigned_word (n) |
| 3384 | && !fits_in_signed_word (n)) |
| 3385 | { |
| 3386 | as_bad (_("16-bit jump out of range")); |
| 3387 | return; |
| 3388 | } |
| 3389 | md_number_to_chars (p, n, size); |
| 3390 | } |
| 3391 | else |
| 3392 | fix_new_exp (frag_now, p - frag_now->fr_literal, size, |
| 3393 | i.op[1].imms, 0, reloc (size, 0, 0, i.reloc[1])); |
| 3394 | if (i.op[0].imms->X_op != O_constant) |
| 3395 | as_bad (_("can't handle non absolute segment in `%s'"), |
| 3396 | i.tm.name); |
| 3397 | md_number_to_chars (p + size, (valueT) i.op[0].imms->X_add_number, 2); |
| 3398 | } |
| 3399 | |
| 3400 | static void |
| 3401 | output_insn () |
| 3402 | { |
| 3403 | fragS *insn_start_frag; |
| 3404 | offsetT insn_start_off; |
| 3405 | |
| 3406 | /* Tie dwarf2 debug info to the address at the start of the insn. |
| 3407 | We can't do this after the insn has been output as the current |
| 3408 | frag may have been closed off. eg. by frag_var. */ |
| 3409 | dwarf2_emit_insn (0); |
| 3410 | |
| 3411 | insn_start_frag = frag_now; |
| 3412 | insn_start_off = frag_now_fix (); |
| 3413 | |
| 3414 | /* Output jumps. */ |
| 3415 | if (i.tm.opcode_modifier & Jump) |
| 3416 | output_branch (); |
| 3417 | else if (i.tm.opcode_modifier & (JumpByte | JumpDword)) |
| 3418 | output_jump (); |
| 3419 | else if (i.tm.opcode_modifier & JumpInterSegment) |
| 3420 | output_interseg_jump (); |
| 3421 | else |
| 3422 | { |
| 3423 | /* Output normal instructions here. */ |
| 3424 | char *p; |
| 3425 | unsigned char *q; |
| 3426 | |
| 3427 | /* All opcodes on i386 have either 1 or 2 bytes. We may use one |
| 3428 | more higher byte to specify a prefix the instruction |
| 3429 | requires. */ |
| 3430 | if ((i.tm.base_opcode & 0xff0000) != 0) |
| 3431 | { |
| 3432 | if ((i.tm.cpu_flags & CpuPadLock) != 0) |
| 3433 | { |
| 3434 | unsigned int prefix; |
| 3435 | prefix = (i.tm.base_opcode >> 16) & 0xff; |
| 3436 | |
| 3437 | if (prefix != REPE_PREFIX_OPCODE |
| 3438 | || i.prefix[LOCKREP_PREFIX] != REPE_PREFIX_OPCODE) |
| 3439 | add_prefix (prefix); |
| 3440 | } |
| 3441 | else |
| 3442 | add_prefix ((i.tm.base_opcode >> 16) & 0xff); |
| 3443 | } |
| 3444 | |
| 3445 | /* The prefix bytes. */ |
| 3446 | for (q = i.prefix; |
| 3447 | q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]); |
| 3448 | q++) |
| 3449 | { |
| 3450 | if (*q) |
| 3451 | { |
| 3452 | p = frag_more (1); |
| 3453 | md_number_to_chars (p, (valueT) *q, 1); |
| 3454 | } |
| 3455 | } |
| 3456 | |
| 3457 | /* Now the opcode; be careful about word order here! */ |
| 3458 | if (fits_in_unsigned_byte (i.tm.base_opcode)) |
| 3459 | { |
| 3460 | FRAG_APPEND_1_CHAR (i.tm.base_opcode); |
| 3461 | } |
| 3462 | else |
| 3463 | { |
| 3464 | p = frag_more (2); |
| 3465 | |
| 3466 | /* Put out high byte first: can't use md_number_to_chars! */ |
| 3467 | *p++ = (i.tm.base_opcode >> 8) & 0xff; |
| 3468 | *p = i.tm.base_opcode & 0xff; |
| 3469 | } |
| 3470 | |
| 3471 | /* Now the modrm byte and sib byte (if present). */ |
| 3472 | if (i.tm.opcode_modifier & Modrm) |
| 3473 | { |
| 3474 | p = frag_more (1); |
| 3475 | md_number_to_chars (p, |
| 3476 | (valueT) (i.rm.regmem << 0 |
| 3477 | | i.rm.reg << 3 |
| 3478 | | i.rm.mode << 6), |
| 3479 | 1); |
| 3480 | /* If i.rm.regmem == ESP (4) |
| 3481 | && i.rm.mode != (Register mode) |
| 3482 | && not 16 bit |
| 3483 | ==> need second modrm byte. */ |
| 3484 | if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING |
| 3485 | && i.rm.mode != 3 |
| 3486 | && !(i.base_reg && (i.base_reg->reg_type & Reg16) != 0)) |
| 3487 | { |
| 3488 | p = frag_more (1); |
| 3489 | md_number_to_chars (p, |
| 3490 | (valueT) (i.sib.base << 0 |
| 3491 | | i.sib.index << 3 |
| 3492 | | i.sib.scale << 6), |
| 3493 | 1); |
| 3494 | } |
| 3495 | } |
| 3496 | |
| 3497 | if (i.disp_operands) |
| 3498 | output_disp (insn_start_frag, insn_start_off); |
| 3499 | |
| 3500 | if (i.imm_operands) |
| 3501 | output_imm (insn_start_frag, insn_start_off); |
| 3502 | } |
| 3503 | |
| 3504 | #ifdef DEBUG386 |
| 3505 | if (flag_debug) |
| 3506 | { |
| 3507 | pi (line, &i); |
| 3508 | } |
| 3509 | #endif /* DEBUG386 */ |
| 3510 | } |
| 3511 | |
| 3512 | static void |
| 3513 | output_disp (insn_start_frag, insn_start_off) |
| 3514 | fragS *insn_start_frag; |
| 3515 | offsetT insn_start_off; |
| 3516 | { |
| 3517 | char *p; |
| 3518 | unsigned int n; |
| 3519 | |
| 3520 | for (n = 0; n < i.operands; n++) |
| 3521 | { |
| 3522 | if (i.types[n] & Disp) |
| 3523 | { |
| 3524 | if (i.op[n].disps->X_op == O_constant) |
| 3525 | { |
| 3526 | int size; |
| 3527 | offsetT val; |
| 3528 | |
| 3529 | size = 4; |
| 3530 | if (i.types[n] & (Disp8 | Disp16 | Disp64)) |
| 3531 | { |
| 3532 | size = 2; |
| 3533 | if (i.types[n] & Disp8) |
| 3534 | size = 1; |
| 3535 | if (i.types[n] & Disp64) |
| 3536 | size = 8; |
| 3537 | } |
| 3538 | val = offset_in_range (i.op[n].disps->X_add_number, |
| 3539 | size); |
| 3540 | p = frag_more (size); |
| 3541 | md_number_to_chars (p, val, size); |
| 3542 | } |
| 3543 | else |
| 3544 | { |
| 3545 | enum bfd_reloc_code_real reloc_type; |
| 3546 | int size = 4; |
| 3547 | int sign = 0; |
| 3548 | int pcrel = (i.flags[n] & Operand_PCrel) != 0; |
| 3549 | |
| 3550 | /* The PC relative address is computed relative |
| 3551 | to the instruction boundary, so in case immediate |
| 3552 | fields follows, we need to adjust the value. */ |
| 3553 | if (pcrel && i.imm_operands) |
| 3554 | { |
| 3555 | int imm_size = 4; |
| 3556 | unsigned int n1; |
| 3557 | |
| 3558 | for (n1 = 0; n1 < i.operands; n1++) |
| 3559 | if (i.types[n1] & Imm) |
| 3560 | { |
| 3561 | if (i.types[n1] & (Imm8 | Imm8S | Imm16 | Imm64)) |
| 3562 | { |
| 3563 | imm_size = 2; |
| 3564 | if (i.types[n1] & (Imm8 | Imm8S)) |
| 3565 | imm_size = 1; |
| 3566 | if (i.types[n1] & Imm64) |
| 3567 | imm_size = 8; |
| 3568 | } |
| 3569 | break; |
| 3570 | } |
| 3571 | /* We should find the immediate. */ |
| 3572 | if (n1 == i.operands) |
| 3573 | abort (); |
| 3574 | i.op[n].disps->X_add_number -= imm_size; |
| 3575 | } |
| 3576 | |
| 3577 | if (i.types[n] & Disp32S) |
| 3578 | sign = 1; |
| 3579 | |
| 3580 | if (i.types[n] & (Disp16 | Disp64)) |
| 3581 | { |
| 3582 | size = 2; |
| 3583 | if (i.types[n] & Disp64) |
| 3584 | size = 8; |
| 3585 | } |
| 3586 | |
| 3587 | p = frag_more (size); |
| 3588 | reloc_type = reloc (size, pcrel, sign, i.reloc[n]); |
| 3589 | if (GOT_symbol |
| 3590 | && GOT_symbol == i.op[n].disps->X_add_symbol |
| 3591 | && (((reloc_type == BFD_RELOC_32 |
| 3592 | || reloc_type == BFD_RELOC_X86_64_32S) |
| 3593 | && (i.op[n].disps->X_op == O_symbol |
| 3594 | || (i.op[n].disps->X_op == O_add |
| 3595 | && ((symbol_get_value_expression |
| 3596 | (i.op[n].disps->X_op_symbol)->X_op) |
| 3597 | == O_subtract)))) |
| 3598 | || reloc_type == BFD_RELOC_32_PCREL)) |
| 3599 | { |
| 3600 | offsetT add; |
| 3601 | |
| 3602 | if (insn_start_frag == frag_now) |
| 3603 | add = (p - frag_now->fr_literal) - insn_start_off; |
| 3604 | else |
| 3605 | { |
| 3606 | fragS *fr; |
| 3607 | |
| 3608 | add = insn_start_frag->fr_fix - insn_start_off; |
| 3609 | for (fr = insn_start_frag->fr_next; |
| 3610 | fr && fr != frag_now; fr = fr->fr_next) |
| 3611 | add += fr->fr_fix; |
| 3612 | add += p - frag_now->fr_literal; |
| 3613 | } |
| 3614 | |
| 3615 | if (!object_64bit) |
| 3616 | reloc_type = BFD_RELOC_386_GOTPC; |
| 3617 | else |
| 3618 | reloc_type = BFD_RELOC_X86_64_GOTPC32; |
| 3619 | i.op[n].disps->X_add_number += add; |
| 3620 | } |
| 3621 | fix_new_exp (frag_now, p - frag_now->fr_literal, size, |
| 3622 | i.op[n].disps, pcrel, reloc_type); |
| 3623 | } |
| 3624 | } |
| 3625 | } |
| 3626 | } |
| 3627 | |
| 3628 | static void |
| 3629 | output_imm (insn_start_frag, insn_start_off) |
| 3630 | fragS *insn_start_frag; |
| 3631 | offsetT insn_start_off; |
| 3632 | { |
| 3633 | char *p; |
| 3634 | unsigned int n; |
| 3635 | |
| 3636 | for (n = 0; n < i.operands; n++) |
| 3637 | { |
| 3638 | if (i.types[n] & Imm) |
| 3639 | { |
| 3640 | if (i.op[n].imms->X_op == O_constant) |
| 3641 | { |
| 3642 | int size; |
| 3643 | offsetT val; |
| 3644 | |
| 3645 | size = 4; |
| 3646 | if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64)) |
| 3647 | { |
| 3648 | size = 2; |
| 3649 | if (i.types[n] & (Imm8 | Imm8S)) |
| 3650 | size = 1; |
| 3651 | else if (i.types[n] & Imm64) |
| 3652 | size = 8; |
| 3653 | } |
| 3654 | val = offset_in_range (i.op[n].imms->X_add_number, |
| 3655 | size); |
| 3656 | p = frag_more (size); |
| 3657 | md_number_to_chars (p, val, size); |
| 3658 | } |
| 3659 | else |
| 3660 | { |
| 3661 | /* Not absolute_section. |
| 3662 | Need a 32-bit fixup (don't support 8bit |
| 3663 | non-absolute imms). Try to support other |
| 3664 | sizes ... */ |
| 3665 | enum bfd_reloc_code_real reloc_type; |
| 3666 | int size = 4; |
| 3667 | int sign = 0; |
| 3668 | |
| 3669 | if ((i.types[n] & (Imm32S)) |
| 3670 | && (i.suffix == QWORD_MNEM_SUFFIX |
| 3671 | || (!i.suffix && (i.tm.opcode_modifier & No_lSuf)))) |
| 3672 | sign = 1; |
| 3673 | if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64)) |
| 3674 | { |
| 3675 | size = 2; |
| 3676 | if (i.types[n] & (Imm8 | Imm8S)) |
| 3677 | size = 1; |
| 3678 | if (i.types[n] & Imm64) |
| 3679 | size = 8; |
| 3680 | } |
| 3681 | |
| 3682 | p = frag_more (size); |
| 3683 | reloc_type = reloc (size, 0, sign, i.reloc[n]); |
| 3684 | |
| 3685 | /* This is tough to explain. We end up with this one if we |
| 3686 | * have operands that look like |
| 3687 | * "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal here is to |
| 3688 | * obtain the absolute address of the GOT, and it is strongly |
| 3689 | * preferable from a performance point of view to avoid using |
| 3690 | * a runtime relocation for this. The actual sequence of |
| 3691 | * instructions often look something like: |
| 3692 | * |
| 3693 | * call .L66 |
| 3694 | * .L66: |
| 3695 | * popl %ebx |
| 3696 | * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx |
| 3697 | * |
| 3698 | * The call and pop essentially return the absolute address |
| 3699 | * of the label .L66 and store it in %ebx. The linker itself |
| 3700 | * will ultimately change the first operand of the addl so |
| 3701 | * that %ebx points to the GOT, but to keep things simple, the |
| 3702 | * .o file must have this operand set so that it generates not |
| 3703 | * the absolute address of .L66, but the absolute address of |
| 3704 | * itself. This allows the linker itself simply treat a GOTPC |
| 3705 | * relocation as asking for a pcrel offset to the GOT to be |
| 3706 | * added in, and the addend of the relocation is stored in the |
| 3707 | * operand field for the instruction itself. |
| 3708 | * |
| 3709 | * Our job here is to fix the operand so that it would add |
| 3710 | * the correct offset so that %ebx would point to itself. The |
| 3711 | * thing that is tricky is that .-.L66 will point to the |
| 3712 | * beginning of the instruction, so we need to further modify |
| 3713 | * the operand so that it will point to itself. There are |
| 3714 | * other cases where you have something like: |
| 3715 | * |
| 3716 | * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66] |
| 3717 | * |
| 3718 | * and here no correction would be required. Internally in |
| 3719 | * the assembler we treat operands of this form as not being |
| 3720 | * pcrel since the '.' is explicitly mentioned, and I wonder |
| 3721 | * whether it would simplify matters to do it this way. Who |
| 3722 | * knows. In earlier versions of the PIC patches, the |
| 3723 | * pcrel_adjust field was used to store the correction, but |
| 3724 | * since the expression is not pcrel, I felt it would be |
| 3725 | * confusing to do it this way. */ |
| 3726 | |
| 3727 | if ((reloc_type == BFD_RELOC_32 |
| 3728 | || reloc_type == BFD_RELOC_X86_64_32S) |
| 3729 | && GOT_symbol |
| 3730 | && GOT_symbol == i.op[n].imms->X_add_symbol |
| 3731 | && (i.op[n].imms->X_op == O_symbol |
| 3732 | || (i.op[n].imms->X_op == O_add |
| 3733 | && ((symbol_get_value_expression |
| 3734 | (i.op[n].imms->X_op_symbol)->X_op) |
| 3735 | == O_subtract)))) |
| 3736 | { |
| 3737 | offsetT add; |
| 3738 | |
| 3739 | if (insn_start_frag == frag_now) |
| 3740 | add = (p - frag_now->fr_literal) - insn_start_off; |
| 3741 | else |
| 3742 | { |
| 3743 | fragS *fr; |
| 3744 | |
| 3745 | add = insn_start_frag->fr_fix - insn_start_off; |
| 3746 | for (fr = insn_start_frag->fr_next; |
| 3747 | fr && fr != frag_now; fr = fr->fr_next) |
| 3748 | add += fr->fr_fix; |
| 3749 | add += p - frag_now->fr_literal; |
| 3750 | } |
| 3751 | |
| 3752 | if (!object_64bit) |
| 3753 | reloc_type = BFD_RELOC_386_GOTPC; |
| 3754 | else |
| 3755 | reloc_type = BFD_RELOC_X86_64_GOTPC32; |
| 3756 | i.op[n].imms->X_add_number += add; |
| 3757 | } |
| 3758 | fix_new_exp (frag_now, p - frag_now->fr_literal, size, |
| 3759 | i.op[n].imms, 0, reloc_type); |
| 3760 | } |
| 3761 | } |
| 3762 | } |
| 3763 | } |
| 3764 | \f |
| 3765 | #if (!defined (OBJ_ELF) && !defined (OBJ_MAYBE_ELF)) || defined (LEX_AT) |
| 3766 | # define lex_got(reloc, adjust, types) NULL |
| 3767 | #else |
| 3768 | /* Parse operands of the form |
| 3769 | <symbol>@GOTOFF+<nnn> |
| 3770 | and similar .plt or .got references. |
| 3771 | |
| 3772 | If we find one, set up the correct relocation in RELOC and copy the |
| 3773 | input string, minus the `@GOTOFF' into a malloc'd buffer for |
| 3774 | parsing by the calling routine. Return this buffer, and if ADJUST |
| 3775 | is non-null set it to the length of the string we removed from the |
| 3776 | input line. Otherwise return NULL. */ |
| 3777 | static char * |
| 3778 | lex_got (enum bfd_reloc_code_real *reloc, |
| 3779 | int *adjust, |
| 3780 | unsigned int *types) |
| 3781 | { |
| 3782 | static const struct { |
| 3783 | const char *str; |
| 3784 | const enum bfd_reloc_code_real rel[2]; |
| 3785 | const unsigned int types64; |
| 3786 | } gotrel[] = { |
| 3787 | { "PLT", { BFD_RELOC_386_PLT32, BFD_RELOC_X86_64_PLT32 }, Imm32|Imm32S|Disp32 }, |
| 3788 | { "GOTOFF", { BFD_RELOC_386_GOTOFF, BFD_RELOC_X86_64_GOTOFF64 }, Imm64|Disp64 }, |
| 3789 | { "GOTPCREL", { 0, BFD_RELOC_X86_64_GOTPCREL }, Imm32|Imm32S|Disp32 }, |
| 3790 | { "TLSGD", { BFD_RELOC_386_TLS_GD, BFD_RELOC_X86_64_TLSGD }, Imm32|Imm32S|Disp32 }, |
| 3791 | { "TLSLDM", { BFD_RELOC_386_TLS_LDM, 0 }, 0 }, |
| 3792 | { "TLSLD", { 0, BFD_RELOC_X86_64_TLSLD }, Imm32|Imm32S|Disp32 }, |
| 3793 | { "GOTTPOFF", { BFD_RELOC_386_TLS_IE_32, BFD_RELOC_X86_64_GOTTPOFF }, Imm32|Imm32S|Disp32 }, |
| 3794 | { "TPOFF", { BFD_RELOC_386_TLS_LE_32, BFD_RELOC_X86_64_TPOFF32 }, Imm32|Imm32S|Imm64|Disp32|Disp64 }, |
| 3795 | { "NTPOFF", { BFD_RELOC_386_TLS_LE, 0 }, 0 }, |
| 3796 | { "DTPOFF", { BFD_RELOC_386_TLS_LDO_32, BFD_RELOC_X86_64_DTPOFF32 }, Imm32|Imm32S|Imm64|Disp32|Disp64 }, |
| 3797 | { "GOTNTPOFF",{ BFD_RELOC_386_TLS_GOTIE, 0 }, 0 }, |
| 3798 | { "INDNTPOFF",{ BFD_RELOC_386_TLS_IE, 0 }, 0 }, |
| 3799 | { "GOT", { BFD_RELOC_386_GOT32, BFD_RELOC_X86_64_GOT32 }, Imm32|Imm32S|Disp32 } |
| 3800 | }; |
| 3801 | char *cp; |
| 3802 | unsigned int j; |
| 3803 | |
| 3804 | if (!IS_ELF) |
| 3805 | return NULL; |
| 3806 | |
| 3807 | for (cp = input_line_pointer; *cp != '@'; cp++) |
| 3808 | if (is_end_of_line[(unsigned char) *cp]) |
| 3809 | return NULL; |
| 3810 | |
| 3811 | for (j = 0; j < sizeof (gotrel) / sizeof (gotrel[0]); j++) |
| 3812 | { |
| 3813 | int len; |
| 3814 | |
| 3815 | len = strlen (gotrel[j].str); |
| 3816 | if (strncasecmp (cp + 1, gotrel[j].str, len) == 0) |
| 3817 | { |
| 3818 | if (gotrel[j].rel[object_64bit] != 0) |
| 3819 | { |
| 3820 | int first, second; |
| 3821 | char *tmpbuf, *past_reloc; |
| 3822 | |
| 3823 | *reloc = gotrel[j].rel[object_64bit]; |
| 3824 | if (adjust) |
| 3825 | *adjust = len; |
| 3826 | |
| 3827 | if (types) |
| 3828 | { |
| 3829 | if (flag_code != CODE_64BIT) |
| 3830 | *types = Imm32|Disp32; |
| 3831 | else |
| 3832 | *types = gotrel[j].types64; |
| 3833 | } |
| 3834 | |
| 3835 | if (GOT_symbol == NULL) |
| 3836 | GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME); |
| 3837 | |
| 3838 | /* Replace the relocation token with ' ', so that |
| 3839 | errors like foo@GOTOFF1 will be detected. */ |
| 3840 | |
| 3841 | /* The length of the first part of our input line. */ |
| 3842 | first = cp - input_line_pointer; |
| 3843 | |
| 3844 | /* The second part goes from after the reloc token until |
| 3845 | (and including) an end_of_line char. Don't use strlen |
| 3846 | here as the end_of_line char may not be a NUL. */ |
| 3847 | past_reloc = cp + 1 + len; |
| 3848 | for (cp = past_reloc; !is_end_of_line[(unsigned char) *cp++]; ) |
| 3849 | ; |
| 3850 | second = cp - past_reloc; |
| 3851 | |
| 3852 | /* Allocate and copy string. The trailing NUL shouldn't |
| 3853 | be necessary, but be safe. */ |
| 3854 | tmpbuf = xmalloc (first + second + 2); |
| 3855 | memcpy (tmpbuf, input_line_pointer, first); |
| 3856 | tmpbuf[first] = ' '; |
| 3857 | memcpy (tmpbuf + first + 1, past_reloc, second); |
| 3858 | tmpbuf[first + second + 1] = '\0'; |
| 3859 | return tmpbuf; |
| 3860 | } |
| 3861 | |
| 3862 | as_bad (_("@%s reloc is not supported with %d-bit output format"), |
| 3863 | gotrel[j].str, 1 << (5 + object_64bit)); |
| 3864 | return NULL; |
| 3865 | } |
| 3866 | } |
| 3867 | |
| 3868 | /* Might be a symbol version string. Don't as_bad here. */ |
| 3869 | return NULL; |
| 3870 | } |
| 3871 | |
| 3872 | /* x86_cons_fix_new is called via the expression parsing code when a |
| 3873 | reloc is needed. We use this hook to get the correct .got reloc. */ |
| 3874 | static enum bfd_reloc_code_real got_reloc = NO_RELOC; |
| 3875 | |
| 3876 | void |
| 3877 | x86_cons_fix_new (frag, off, len, exp) |
| 3878 | fragS *frag; |
| 3879 | unsigned int off; |
| 3880 | unsigned int len; |
| 3881 | expressionS *exp; |
| 3882 | { |
| 3883 | enum bfd_reloc_code_real r = reloc (len, 0, -1, got_reloc); |
| 3884 | got_reloc = NO_RELOC; |
| 3885 | fix_new_exp (frag, off, len, exp, 0, r); |
| 3886 | } |
| 3887 | |
| 3888 | void |
| 3889 | x86_cons (exp, size) |
| 3890 | expressionS *exp; |
| 3891 | int size; |
| 3892 | { |
| 3893 | if (size == 4 || (object_64bit && size == 8)) |
| 3894 | { |
| 3895 | /* Handle @GOTOFF and the like in an expression. */ |
| 3896 | char *save; |
| 3897 | char *gotfree_input_line; |
| 3898 | int adjust; |
| 3899 | |
| 3900 | save = input_line_pointer; |
| 3901 | gotfree_input_line = lex_got (&got_reloc, &adjust, NULL); |
| 3902 | if (gotfree_input_line) |
| 3903 | input_line_pointer = gotfree_input_line; |
| 3904 | |
| 3905 | expression (exp); |
| 3906 | |
| 3907 | if (gotfree_input_line) |
| 3908 | { |
| 3909 | /* expression () has merrily parsed up to the end of line, |
| 3910 | or a comma - in the wrong buffer. Transfer how far |
| 3911 | input_line_pointer has moved to the right buffer. */ |
| 3912 | input_line_pointer = (save |
| 3913 | + (input_line_pointer - gotfree_input_line) |
| 3914 | + adjust); |
| 3915 | free (gotfree_input_line); |
| 3916 | } |
| 3917 | } |
| 3918 | else |
| 3919 | expression (exp); |
| 3920 | } |
| 3921 | #endif |
| 3922 | |
| 3923 | #ifdef TE_PE |
| 3924 | |
| 3925 | void |
| 3926 | x86_pe_cons_fix_new (frag, off, len, exp) |
| 3927 | fragS *frag; |
| 3928 | unsigned int off; |
| 3929 | unsigned int len; |
| 3930 | expressionS *exp; |
| 3931 | { |
| 3932 | enum bfd_reloc_code_real r = reloc (len, 0, -1, NO_RELOC); |
| 3933 | |
| 3934 | if (exp->X_op == O_secrel) |
| 3935 | { |
| 3936 | exp->X_op = O_symbol; |
| 3937 | r = BFD_RELOC_32_SECREL; |
| 3938 | } |
| 3939 | |
| 3940 | fix_new_exp (frag, off, len, exp, 0, r); |
| 3941 | } |
| 3942 | |
| 3943 | static void |
| 3944 | pe_directive_secrel (dummy) |
| 3945 | int dummy ATTRIBUTE_UNUSED; |
| 3946 | { |
| 3947 | expressionS exp; |
| 3948 | |
| 3949 | do |
| 3950 | { |
| 3951 | expression (&exp); |
| 3952 | if (exp.X_op == O_symbol) |
| 3953 | exp.X_op = O_secrel; |
| 3954 | |
| 3955 | emit_expr (&exp, 4); |
| 3956 | } |
| 3957 | while (*input_line_pointer++ == ','); |
| 3958 | |
| 3959 | input_line_pointer--; |
| 3960 | demand_empty_rest_of_line (); |
| 3961 | } |
| 3962 | |
| 3963 | #endif |
| 3964 | |
| 3965 | static int i386_immediate PARAMS ((char *)); |
| 3966 | |
| 3967 | static int |
| 3968 | i386_immediate (imm_start) |
| 3969 | char *imm_start; |
| 3970 | { |
| 3971 | char *save_input_line_pointer; |
| 3972 | char *gotfree_input_line; |
| 3973 | segT exp_seg = 0; |
| 3974 | expressionS *exp; |
| 3975 | unsigned int types = ~0U; |
| 3976 | |
| 3977 | if (i.imm_operands == MAX_IMMEDIATE_OPERANDS) |
| 3978 | { |
| 3979 | as_bad (_("only 1 or 2 immediate operands are allowed")); |
| 3980 | return 0; |
| 3981 | } |
| 3982 | |
| 3983 | exp = &im_expressions[i.imm_operands++]; |
| 3984 | i.op[this_operand].imms = exp; |
| 3985 | |
| 3986 | if (is_space_char (*imm_start)) |
| 3987 | ++imm_start; |
| 3988 | |
| 3989 | save_input_line_pointer = input_line_pointer; |
| 3990 | input_line_pointer = imm_start; |
| 3991 | |
| 3992 | gotfree_input_line = lex_got (&i.reloc[this_operand], NULL, &types); |
| 3993 | if (gotfree_input_line) |
| 3994 | input_line_pointer = gotfree_input_line; |
| 3995 | |
| 3996 | exp_seg = expression (exp); |
| 3997 | |
| 3998 | SKIP_WHITESPACE (); |
| 3999 | if (*input_line_pointer) |
| 4000 | as_bad (_("junk `%s' after expression"), input_line_pointer); |
| 4001 | |
| 4002 | input_line_pointer = save_input_line_pointer; |
| 4003 | if (gotfree_input_line) |
| 4004 | free (gotfree_input_line); |
| 4005 | |
| 4006 | if (exp->X_op == O_absent || exp->X_op == O_big) |
| 4007 | { |
| 4008 | /* Missing or bad expr becomes absolute 0. */ |
| 4009 | as_bad (_("missing or invalid immediate expression `%s' taken as 0"), |
| 4010 | imm_start); |
| 4011 | exp->X_op = O_constant; |
| 4012 | exp->X_add_number = 0; |
| 4013 | exp->X_add_symbol = (symbolS *) 0; |
| 4014 | exp->X_op_symbol = (symbolS *) 0; |
| 4015 | } |
| 4016 | else if (exp->X_op == O_constant) |
| 4017 | { |
| 4018 | /* Size it properly later. */ |
| 4019 | i.types[this_operand] |= Imm64; |
| 4020 | /* If BFD64, sign extend val. */ |
| 4021 | if (!use_rela_relocations) |
| 4022 | if ((exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0) |
| 4023 | exp->X_add_number = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31); |
| 4024 | } |
| 4025 | #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT)) |
| 4026 | else if (OUTPUT_FLAVOR == bfd_target_aout_flavour |
| 4027 | && exp_seg != absolute_section |
| 4028 | && exp_seg != text_section |
| 4029 | && exp_seg != data_section |
| 4030 | && exp_seg != bss_section |
| 4031 | && exp_seg != undefined_section |
| 4032 | && !bfd_is_com_section (exp_seg)) |
| 4033 | { |
| 4034 | as_bad (_("unimplemented segment %s in operand"), exp_seg->name); |
| 4035 | return 0; |
| 4036 | } |
| 4037 | #endif |
| 4038 | else |
| 4039 | { |
| 4040 | /* This is an address. The size of the address will be |
| 4041 | determined later, depending on destination register, |
| 4042 | suffix, or the default for the section. */ |
| 4043 | i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64; |
| 4044 | i.types[this_operand] &= types; |
| 4045 | } |
| 4046 | |
| 4047 | return 1; |
| 4048 | } |
| 4049 | |
| 4050 | static char *i386_scale PARAMS ((char *)); |
| 4051 | |
| 4052 | static char * |
| 4053 | i386_scale (scale) |
| 4054 | char *scale; |
| 4055 | { |
| 4056 | offsetT val; |
| 4057 | char *save = input_line_pointer; |
| 4058 | |
| 4059 | input_line_pointer = scale; |
| 4060 | val = get_absolute_expression (); |
| 4061 | |
| 4062 | switch (val) |
| 4063 | { |
| 4064 | case 1: |
| 4065 | i.log2_scale_factor = 0; |
| 4066 | break; |
| 4067 | case 2: |
| 4068 | i.log2_scale_factor = 1; |
| 4069 | break; |
| 4070 | case 4: |
| 4071 | i.log2_scale_factor = 2; |
| 4072 | break; |
| 4073 | case 8: |
| 4074 | i.log2_scale_factor = 3; |
| 4075 | break; |
| 4076 | default: |
| 4077 | { |
| 4078 | char sep = *input_line_pointer; |
| 4079 | |
| 4080 | *input_line_pointer = '\0'; |
| 4081 | as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"), |
| 4082 | scale); |
| 4083 | *input_line_pointer = sep; |
| 4084 | input_line_pointer = save; |
| 4085 | return NULL; |
| 4086 | } |
| 4087 | } |
| 4088 | if (i.log2_scale_factor != 0 && i.index_reg == 0) |
| 4089 | { |
| 4090 | as_warn (_("scale factor of %d without an index register"), |
| 4091 | 1 << i.log2_scale_factor); |
| 4092 | #if SCALE1_WHEN_NO_INDEX |
| 4093 | i.log2_scale_factor = 0; |
| 4094 | #endif |
| 4095 | } |
| 4096 | scale = input_line_pointer; |
| 4097 | input_line_pointer = save; |
| 4098 | return scale; |
| 4099 | } |
| 4100 | |
| 4101 | static int i386_displacement PARAMS ((char *, char *)); |
| 4102 | |
| 4103 | static int |
| 4104 | i386_displacement (disp_start, disp_end) |
| 4105 | char *disp_start; |
| 4106 | char *disp_end; |
| 4107 | { |
| 4108 | expressionS *exp; |
| 4109 | segT exp_seg = 0; |
| 4110 | char *save_input_line_pointer; |
| 4111 | char *gotfree_input_line; |
| 4112 | int bigdisp = Disp32; |
| 4113 | unsigned int types = Disp; |
| 4114 | |
| 4115 | if (flag_code == CODE_64BIT) |
| 4116 | { |
| 4117 | if (i.prefix[ADDR_PREFIX] == 0) |
| 4118 | bigdisp = Disp64 | Disp32S | Disp32; |
| 4119 | } |
| 4120 | else if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0)) |
| 4121 | bigdisp = Disp16; |
| 4122 | i.types[this_operand] |= bigdisp; |
| 4123 | |
| 4124 | exp = &disp_expressions[i.disp_operands]; |
| 4125 | i.op[this_operand].disps = exp; |
| 4126 | i.disp_operands++; |
| 4127 | save_input_line_pointer = input_line_pointer; |
| 4128 | input_line_pointer = disp_start; |
| 4129 | END_STRING_AND_SAVE (disp_end); |
| 4130 | |
| 4131 | #ifndef GCC_ASM_O_HACK |
| 4132 | #define GCC_ASM_O_HACK 0 |
| 4133 | #endif |
| 4134 | #if GCC_ASM_O_HACK |
| 4135 | END_STRING_AND_SAVE (disp_end + 1); |
| 4136 | if ((i.types[this_operand] & BaseIndex) != 0 |
| 4137 | && displacement_string_end[-1] == '+') |
| 4138 | { |
| 4139 | /* This hack is to avoid a warning when using the "o" |
| 4140 | constraint within gcc asm statements. |
| 4141 | For instance: |
| 4142 | |
| 4143 | #define _set_tssldt_desc(n,addr,limit,type) \ |
| 4144 | __asm__ __volatile__ ( \ |
| 4145 | "movw %w2,%0\n\t" \ |
| 4146 | "movw %w1,2+%0\n\t" \ |
| 4147 | "rorl $16,%1\n\t" \ |
| 4148 | "movb %b1,4+%0\n\t" \ |
| 4149 | "movb %4,5+%0\n\t" \ |
| 4150 | "movb $0,6+%0\n\t" \ |
| 4151 | "movb %h1,7+%0\n\t" \ |
| 4152 | "rorl $16,%1" \ |
| 4153 | : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type)) |
| 4154 | |
| 4155 | This works great except that the output assembler ends |
| 4156 | up looking a bit weird if it turns out that there is |
| 4157 | no offset. You end up producing code that looks like: |
| 4158 | |
| 4159 | #APP |
| 4160 | movw $235,(%eax) |
| 4161 | movw %dx,2+(%eax) |
| 4162 | rorl $16,%edx |
| 4163 | movb %dl,4+(%eax) |
| 4164 | movb $137,5+(%eax) |
| 4165 | movb $0,6+(%eax) |
| 4166 | movb %dh,7+(%eax) |
| 4167 | rorl $16,%edx |
| 4168 | #NO_APP |
| 4169 | |
| 4170 | So here we provide the missing zero. */ |
| 4171 | |
| 4172 | *displacement_string_end = '0'; |
| 4173 | } |
| 4174 | #endif |
| 4175 | gotfree_input_line = lex_got (&i.reloc[this_operand], NULL, &types); |
| 4176 | if (gotfree_input_line) |
| 4177 | input_line_pointer = gotfree_input_line; |
| 4178 | |
| 4179 | exp_seg = expression (exp); |
| 4180 | |
| 4181 | SKIP_WHITESPACE (); |
| 4182 | if (*input_line_pointer) |
| 4183 | as_bad (_("junk `%s' after expression"), input_line_pointer); |
| 4184 | #if GCC_ASM_O_HACK |
| 4185 | RESTORE_END_STRING (disp_end + 1); |
| 4186 | #endif |
| 4187 | RESTORE_END_STRING (disp_end); |
| 4188 | input_line_pointer = save_input_line_pointer; |
| 4189 | if (gotfree_input_line) |
| 4190 | free (gotfree_input_line); |
| 4191 | |
| 4192 | /* We do this to make sure that the section symbol is in |
| 4193 | the symbol table. We will ultimately change the relocation |
| 4194 | to be relative to the beginning of the section. */ |
| 4195 | if (i.reloc[this_operand] == BFD_RELOC_386_GOTOFF |
| 4196 | || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL |
| 4197 | || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTOFF64) |
| 4198 | { |
| 4199 | if (exp->X_op != O_symbol) |
| 4200 | { |
| 4201 | as_bad (_("bad expression used with @%s"), |
| 4202 | (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL |
| 4203 | ? "GOTPCREL" |
| 4204 | : "GOTOFF")); |
| 4205 | return 0; |
| 4206 | } |
| 4207 | |
| 4208 | if (S_IS_LOCAL (exp->X_add_symbol) |
| 4209 | && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section) |
| 4210 | section_symbol (S_GET_SEGMENT (exp->X_add_symbol)); |
| 4211 | exp->X_op = O_subtract; |
| 4212 | exp->X_op_symbol = GOT_symbol; |
| 4213 | if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL) |
| 4214 | i.reloc[this_operand] = BFD_RELOC_32_PCREL; |
| 4215 | else if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTOFF64) |
| 4216 | i.reloc[this_operand] = BFD_RELOC_64; |
| 4217 | else |
| 4218 | i.reloc[this_operand] = BFD_RELOC_32; |
| 4219 | } |
| 4220 | |
| 4221 | if (exp->X_op == O_absent || exp->X_op == O_big) |
| 4222 | { |
| 4223 | /* Missing or bad expr becomes absolute 0. */ |
| 4224 | as_bad (_("missing or invalid displacement expression `%s' taken as 0"), |
| 4225 | disp_start); |
| 4226 | exp->X_op = O_constant; |
| 4227 | exp->X_add_number = 0; |
| 4228 | exp->X_add_symbol = (symbolS *) 0; |
| 4229 | exp->X_op_symbol = (symbolS *) 0; |
| 4230 | } |
| 4231 | |
| 4232 | #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT)) |
| 4233 | if (exp->X_op != O_constant |
| 4234 | && OUTPUT_FLAVOR == bfd_target_aout_flavour |
| 4235 | && exp_seg != absolute_section |
| 4236 | && exp_seg != text_section |
| 4237 | && exp_seg != data_section |
| 4238 | && exp_seg != bss_section |
| 4239 | && exp_seg != undefined_section |
| 4240 | && !bfd_is_com_section (exp_seg)) |
| 4241 | { |
| 4242 | as_bad (_("unimplemented segment %s in operand"), exp_seg->name); |
| 4243 | return 0; |
| 4244 | } |
| 4245 | #endif |
| 4246 | |
| 4247 | if (!(i.types[this_operand] & ~Disp)) |
| 4248 | i.types[this_operand] &= types; |
| 4249 | |
| 4250 | return 1; |
| 4251 | } |
| 4252 | |
| 4253 | static int i386_index_check PARAMS ((const char *)); |
| 4254 | |
| 4255 | /* Make sure the memory operand we've been dealt is valid. |
| 4256 | Return 1 on success, 0 on a failure. */ |
| 4257 | |
| 4258 | static int |
| 4259 | i386_index_check (operand_string) |
| 4260 | const char *operand_string; |
| 4261 | { |
| 4262 | int ok; |
| 4263 | #if INFER_ADDR_PREFIX |
| 4264 | int fudged = 0; |
| 4265 | |
| 4266 | tryprefix: |
| 4267 | #endif |
| 4268 | ok = 1; |
| 4269 | if ((current_templates->start->cpu_flags & CpuSVME) |
| 4270 | && current_templates->end[-1].operand_types[0] == AnyMem) |
| 4271 | { |
| 4272 | /* Memory operands of SVME insns are special in that they only allow |
| 4273 | rAX as their memory address and ignore any segment override. */ |
| 4274 | unsigned RegXX; |
| 4275 | |
| 4276 | /* SKINIT is even more restrictive: it always requires EAX. */ |
| 4277 | if (strcmp (current_templates->start->name, "skinit") == 0) |
| 4278 | RegXX = Reg32; |
| 4279 | else if (flag_code == CODE_64BIT) |
| 4280 | RegXX = i.prefix[ADDR_PREFIX] == 0 ? Reg64 : Reg32; |
| 4281 | else |
| 4282 | RegXX = (flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0) |
| 4283 | ? Reg16 |
| 4284 | : Reg32; |
| 4285 | if (!i.base_reg |
| 4286 | || !(i.base_reg->reg_type & Acc) |
| 4287 | || !(i.base_reg->reg_type & RegXX) |
| 4288 | || i.index_reg |
| 4289 | || (i.types[0] & Disp)) |
| 4290 | ok = 0; |
| 4291 | } |
| 4292 | else if (flag_code == CODE_64BIT) |
| 4293 | { |
| 4294 | unsigned RegXX = (i.prefix[ADDR_PREFIX] == 0 ? Reg64 : Reg32); |
| 4295 | |
| 4296 | if ((i.base_reg |
| 4297 | && ((i.base_reg->reg_type & RegXX) == 0) |
| 4298 | && (i.base_reg->reg_type != BaseIndex |
| 4299 | || i.index_reg)) |
| 4300 | || (i.index_reg |
| 4301 | && ((i.index_reg->reg_type & (RegXX | BaseIndex)) |
| 4302 | != (RegXX | BaseIndex)))) |
| 4303 | ok = 0; |
| 4304 | } |
| 4305 | else |
| 4306 | { |
| 4307 | if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0)) |
| 4308 | { |
| 4309 | /* 16bit checks. */ |
| 4310 | if ((i.base_reg |
| 4311 | && ((i.base_reg->reg_type & (Reg16 | BaseIndex | RegRex)) |
| 4312 | != (Reg16 | BaseIndex))) |
| 4313 | || (i.index_reg |
| 4314 | && (((i.index_reg->reg_type & (Reg16 | BaseIndex)) |
| 4315 | != (Reg16 | BaseIndex)) |
| 4316 | || !(i.base_reg |
| 4317 | && i.base_reg->reg_num < 6 |
| 4318 | && i.index_reg->reg_num >= 6 |
| 4319 | && i.log2_scale_factor == 0)))) |
| 4320 | ok = 0; |
| 4321 | } |
| 4322 | else |
| 4323 | { |
| 4324 | /* 32bit checks. */ |
| 4325 | if ((i.base_reg |
| 4326 | && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32) |
| 4327 | || (i.index_reg |
| 4328 | && ((i.index_reg->reg_type & (Reg32 | BaseIndex | RegRex)) |
| 4329 | != (Reg32 | BaseIndex)))) |
| 4330 | ok = 0; |
| 4331 | } |
| 4332 | } |
| 4333 | if (!ok) |
| 4334 | { |
| 4335 | #if INFER_ADDR_PREFIX |
| 4336 | if (i.prefix[ADDR_PREFIX] == 0) |
| 4337 | { |
| 4338 | i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE; |
| 4339 | i.prefixes += 1; |
| 4340 | /* Change the size of any displacement too. At most one of |
| 4341 | Disp16 or Disp32 is set. |
| 4342 | FIXME. There doesn't seem to be any real need for separate |
| 4343 | Disp16 and Disp32 flags. The same goes for Imm16 and Imm32. |
| 4344 | Removing them would probably clean up the code quite a lot. */ |
| 4345 | if (flag_code != CODE_64BIT && (i.types[this_operand] & (Disp16 | Disp32))) |
| 4346 | i.types[this_operand] ^= (Disp16 | Disp32); |
| 4347 | fudged = 1; |
| 4348 | goto tryprefix; |
| 4349 | } |
| 4350 | if (fudged) |
| 4351 | as_bad (_("`%s' is not a valid base/index expression"), |
| 4352 | operand_string); |
| 4353 | else |
| 4354 | #endif |
| 4355 | as_bad (_("`%s' is not a valid %s bit base/index expression"), |
| 4356 | operand_string, |
| 4357 | flag_code_names[flag_code]); |
| 4358 | } |
| 4359 | return ok; |
| 4360 | } |
| 4361 | |
| 4362 | /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero |
| 4363 | on error. */ |
| 4364 | |
| 4365 | static int |
| 4366 | i386_operand (operand_string) |
| 4367 | char *operand_string; |
| 4368 | { |
| 4369 | const reg_entry *r; |
| 4370 | char *end_op; |
| 4371 | char *op_string = operand_string; |
| 4372 | |
| 4373 | if (is_space_char (*op_string)) |
| 4374 | ++op_string; |
| 4375 | |
| 4376 | /* We check for an absolute prefix (differentiating, |
| 4377 | for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */ |
| 4378 | if (*op_string == ABSOLUTE_PREFIX) |
| 4379 | { |
| 4380 | ++op_string; |
| 4381 | if (is_space_char (*op_string)) |
| 4382 | ++op_string; |
| 4383 | i.types[this_operand] |= JumpAbsolute; |
| 4384 | } |
| 4385 | |
| 4386 | /* Check if operand is a register. */ |
| 4387 | if ((*op_string == REGISTER_PREFIX || allow_naked_reg) |
| 4388 | && (r = parse_register (op_string, &end_op)) != NULL) |
| 4389 | { |
| 4390 | /* Check for a segment override by searching for ':' after a |
| 4391 | segment register. */ |
| 4392 | op_string = end_op; |
| 4393 | if (is_space_char (*op_string)) |
| 4394 | ++op_string; |
| 4395 | if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3))) |
| 4396 | { |
| 4397 | switch (r->reg_num) |
| 4398 | { |
| 4399 | case 0: |
| 4400 | i.seg[i.mem_operands] = &es; |
| 4401 | break; |
| 4402 | case 1: |
| 4403 | i.seg[i.mem_operands] = &cs; |
| 4404 | break; |
| 4405 | case 2: |
| 4406 | i.seg[i.mem_operands] = &ss; |
| 4407 | break; |
| 4408 | case 3: |
| 4409 | i.seg[i.mem_operands] = &ds; |
| 4410 | break; |
| 4411 | case 4: |
| 4412 | i.seg[i.mem_operands] = &fs; |
| 4413 | break; |
| 4414 | case 5: |
| 4415 | i.seg[i.mem_operands] = &gs; |
| 4416 | break; |
| 4417 | } |
| 4418 | |
| 4419 | /* Skip the ':' and whitespace. */ |
| 4420 | ++op_string; |
| 4421 | if (is_space_char (*op_string)) |
| 4422 | ++op_string; |
| 4423 | |
| 4424 | if (!is_digit_char (*op_string) |
| 4425 | && !is_identifier_char (*op_string) |
| 4426 | && *op_string != '(' |
| 4427 | && *op_string != ABSOLUTE_PREFIX) |
| 4428 | { |
| 4429 | as_bad (_("bad memory operand `%s'"), op_string); |
| 4430 | return 0; |
| 4431 | } |
| 4432 | /* Handle case of %es:*foo. */ |
| 4433 | if (*op_string == ABSOLUTE_PREFIX) |
| 4434 | { |
| 4435 | ++op_string; |
| 4436 | if (is_space_char (*op_string)) |
| 4437 | ++op_string; |
| 4438 | i.types[this_operand] |= JumpAbsolute; |
| 4439 | } |
| 4440 | goto do_memory_reference; |
| 4441 | } |
| 4442 | if (*op_string) |
| 4443 | { |
| 4444 | as_bad (_("junk `%s' after register"), op_string); |
| 4445 | return 0; |
| 4446 | } |
| 4447 | i.types[this_operand] |= r->reg_type & ~BaseIndex; |
| 4448 | i.op[this_operand].regs = r; |
| 4449 | i.reg_operands++; |
| 4450 | } |
| 4451 | else if (*op_string == REGISTER_PREFIX) |
| 4452 | { |
| 4453 | as_bad (_("bad register name `%s'"), op_string); |
| 4454 | return 0; |
| 4455 | } |
| 4456 | else if (*op_string == IMMEDIATE_PREFIX) |
| 4457 | { |
| 4458 | ++op_string; |
| 4459 | if (i.types[this_operand] & JumpAbsolute) |
| 4460 | { |
| 4461 | as_bad (_("immediate operand illegal with absolute jump")); |
| 4462 | return 0; |
| 4463 | } |
| 4464 | if (!i386_immediate (op_string)) |
| 4465 | return 0; |
| 4466 | } |
| 4467 | else if (is_digit_char (*op_string) |
| 4468 | || is_identifier_char (*op_string) |
| 4469 | || *op_string == '(') |
| 4470 | { |
| 4471 | /* This is a memory reference of some sort. */ |
| 4472 | char *base_string; |
| 4473 | |
| 4474 | /* Start and end of displacement string expression (if found). */ |
| 4475 | char *displacement_string_start; |
| 4476 | char *displacement_string_end; |
| 4477 | |
| 4478 | do_memory_reference: |
| 4479 | if ((i.mem_operands == 1 |
| 4480 | && (current_templates->start->opcode_modifier & IsString) == 0) |
| 4481 | || i.mem_operands == 2) |
| 4482 | { |
| 4483 | as_bad (_("too many memory references for `%s'"), |
| 4484 | current_templates->start->name); |
| 4485 | return 0; |
| 4486 | } |
| 4487 | |
| 4488 | /* Check for base index form. We detect the base index form by |
| 4489 | looking for an ')' at the end of the operand, searching |
| 4490 | for the '(' matching it, and finding a REGISTER_PREFIX or ',' |
| 4491 | after the '('. */ |
| 4492 | base_string = op_string + strlen (op_string); |
| 4493 | |
| 4494 | --base_string; |
| 4495 | if (is_space_char (*base_string)) |
| 4496 | --base_string; |
| 4497 | |
| 4498 | /* If we only have a displacement, set-up for it to be parsed later. */ |
| 4499 | displacement_string_start = op_string; |
| 4500 | displacement_string_end = base_string + 1; |
| 4501 | |
| 4502 | if (*base_string == ')') |
| 4503 | { |
| 4504 | char *temp_string; |
| 4505 | unsigned int parens_balanced = 1; |
| 4506 | /* We've already checked that the number of left & right ()'s are |
| 4507 | equal, so this loop will not be infinite. */ |
| 4508 | do |
| 4509 | { |
| 4510 | base_string--; |
| 4511 | if (*base_string == ')') |
| 4512 | parens_balanced++; |
| 4513 | if (*base_string == '(') |
| 4514 | parens_balanced--; |
| 4515 | } |
| 4516 | while (parens_balanced); |
| 4517 | |
| 4518 | temp_string = base_string; |
| 4519 | |
| 4520 | /* Skip past '(' and whitespace. */ |
| 4521 | ++base_string; |
| 4522 | if (is_space_char (*base_string)) |
| 4523 | ++base_string; |
| 4524 | |
| 4525 | if (*base_string == ',' |
| 4526 | || ((*base_string == REGISTER_PREFIX || allow_naked_reg) |
| 4527 | && (i.base_reg = parse_register (base_string, &end_op)) != NULL)) |
| 4528 | { |
| 4529 | displacement_string_end = temp_string; |
| 4530 | |
| 4531 | i.types[this_operand] |= BaseIndex; |
| 4532 | |
| 4533 | if (i.base_reg) |
| 4534 | { |
| 4535 | base_string = end_op; |
| 4536 | if (is_space_char (*base_string)) |
| 4537 | ++base_string; |
| 4538 | } |
| 4539 | |
| 4540 | /* There may be an index reg or scale factor here. */ |
| 4541 | if (*base_string == ',') |
| 4542 | { |
| 4543 | ++base_string; |
| 4544 | if (is_space_char (*base_string)) |
| 4545 | ++base_string; |
| 4546 | |
| 4547 | if ((*base_string == REGISTER_PREFIX || allow_naked_reg) |
| 4548 | && (i.index_reg = parse_register (base_string, &end_op)) != NULL) |
| 4549 | { |
| 4550 | base_string = end_op; |
| 4551 | if (is_space_char (*base_string)) |
| 4552 | ++base_string; |
| 4553 | if (*base_string == ',') |
| 4554 | { |
| 4555 | ++base_string; |
| 4556 | if (is_space_char (*base_string)) |
| 4557 | ++base_string; |
| 4558 | } |
| 4559 | else if (*base_string != ')') |
| 4560 | { |
| 4561 | as_bad (_("expecting `,' or `)' after index register in `%s'"), |
| 4562 | operand_string); |
| 4563 | return 0; |
| 4564 | } |
| 4565 | } |
| 4566 | else if (*base_string == REGISTER_PREFIX) |
| 4567 | { |
| 4568 | as_bad (_("bad register name `%s'"), base_string); |
| 4569 | return 0; |
| 4570 | } |
| 4571 | |
| 4572 | /* Check for scale factor. */ |
| 4573 | if (*base_string != ')') |
| 4574 | { |
| 4575 | char *end_scale = i386_scale (base_string); |
| 4576 | |
| 4577 | if (!end_scale) |
| 4578 | return 0; |
| 4579 | |
| 4580 | base_string = end_scale; |
| 4581 | if (is_space_char (*base_string)) |
| 4582 | ++base_string; |
| 4583 | if (*base_string != ')') |
| 4584 | { |
| 4585 | as_bad (_("expecting `)' after scale factor in `%s'"), |
| 4586 | operand_string); |
| 4587 | return 0; |
| 4588 | } |
| 4589 | } |
| 4590 | else if (!i.index_reg) |
| 4591 | { |
| 4592 | as_bad (_("expecting index register or scale factor after `,'; got '%c'"), |
| 4593 | *base_string); |
| 4594 | return 0; |
| 4595 | } |
| 4596 | } |
| 4597 | else if (*base_string != ')') |
| 4598 | { |
| 4599 | as_bad (_("expecting `,' or `)' after base register in `%s'"), |
| 4600 | operand_string); |
| 4601 | return 0; |
| 4602 | } |
| 4603 | } |
| 4604 | else if (*base_string == REGISTER_PREFIX) |
| 4605 | { |
| 4606 | as_bad (_("bad register name `%s'"), base_string); |
| 4607 | return 0; |
| 4608 | } |
| 4609 | } |
| 4610 | |
| 4611 | /* If there's an expression beginning the operand, parse it, |
| 4612 | assuming displacement_string_start and |
| 4613 | displacement_string_end are meaningful. */ |
| 4614 | if (displacement_string_start != displacement_string_end) |
| 4615 | { |
| 4616 | if (!i386_displacement (displacement_string_start, |
| 4617 | displacement_string_end)) |
| 4618 | return 0; |
| 4619 | } |
| 4620 | |
| 4621 | /* Special case for (%dx) while doing input/output op. */ |
| 4622 | if (i.base_reg |
| 4623 | && i.base_reg->reg_type == (Reg16 | InOutPortReg) |
| 4624 | && i.index_reg == 0 |
| 4625 | && i.log2_scale_factor == 0 |
| 4626 | && i.seg[i.mem_operands] == 0 |
| 4627 | && (i.types[this_operand] & Disp) == 0) |
| 4628 | { |
| 4629 | i.types[this_operand] = InOutPortReg; |
| 4630 | return 1; |
| 4631 | } |
| 4632 | |
| 4633 | if (i386_index_check (operand_string) == 0) |
| 4634 | return 0; |
| 4635 | i.mem_operands++; |
| 4636 | } |
| 4637 | else |
| 4638 | { |
| 4639 | /* It's not a memory operand; argh! */ |
| 4640 | as_bad (_("invalid char %s beginning operand %d `%s'"), |
| 4641 | output_invalid (*op_string), |
| 4642 | this_operand + 1, |
| 4643 | op_string); |
| 4644 | return 0; |
| 4645 | } |
| 4646 | return 1; /* Normal return. */ |
| 4647 | } |
| 4648 | \f |
| 4649 | /* md_estimate_size_before_relax() |
| 4650 | |
| 4651 | Called just before relax() for rs_machine_dependent frags. The x86 |
| 4652 | assembler uses these frags to handle variable size jump |
| 4653 | instructions. |
| 4654 | |
| 4655 | Any symbol that is now undefined will not become defined. |
| 4656 | Return the correct fr_subtype in the frag. |
| 4657 | Return the initial "guess for variable size of frag" to caller. |
| 4658 | The guess is actually the growth beyond the fixed part. Whatever |
| 4659 | we do to grow the fixed or variable part contributes to our |
| 4660 | returned value. */ |
| 4661 | |
| 4662 | int |
| 4663 | md_estimate_size_before_relax (fragP, segment) |
| 4664 | fragS *fragP; |
| 4665 | segT segment; |
| 4666 | { |
| 4667 | /* We've already got fragP->fr_subtype right; all we have to do is |
| 4668 | check for un-relaxable symbols. On an ELF system, we can't relax |
| 4669 | an externally visible symbol, because it may be overridden by a |
| 4670 | shared library. */ |
| 4671 | if (S_GET_SEGMENT (fragP->fr_symbol) != segment |
| 4672 | #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) |
| 4673 | || (IS_ELF |
| 4674 | && (S_IS_EXTERNAL (fragP->fr_symbol) |
| 4675 | || S_IS_WEAK (fragP->fr_symbol))) |
| 4676 | #endif |
| 4677 | ) |
| 4678 | { |
| 4679 | /* Symbol is undefined in this segment, or we need to keep a |
| 4680 | reloc so that weak symbols can be overridden. */ |
| 4681 | int size = (fragP->fr_subtype & CODE16) ? 2 : 4; |
| 4682 | enum bfd_reloc_code_real reloc_type; |
| 4683 | unsigned char *opcode; |
| 4684 | int old_fr_fix; |
| 4685 | |
| 4686 | if (fragP->fr_var != NO_RELOC) |
| 4687 | reloc_type = fragP->fr_var; |
| 4688 | else if (size == 2) |
| 4689 | reloc_type = BFD_RELOC_16_PCREL; |
| 4690 | else |
| 4691 | reloc_type = BFD_RELOC_32_PCREL; |
| 4692 | |
| 4693 | old_fr_fix = fragP->fr_fix; |
| 4694 | opcode = (unsigned char *) fragP->fr_opcode; |
| 4695 | |
| 4696 | switch (TYPE_FROM_RELAX_STATE (fragP->fr_subtype)) |
| 4697 | { |
| 4698 | case UNCOND_JUMP: |
| 4699 | /* Make jmp (0xeb) a (d)word displacement jump. */ |
| 4700 | opcode[0] = 0xe9; |
| 4701 | fragP->fr_fix += size; |
| 4702 | fix_new (fragP, old_fr_fix, size, |
| 4703 | fragP->fr_symbol, |
| 4704 | fragP->fr_offset, 1, |
| 4705 | reloc_type); |
| 4706 | break; |
| 4707 | |
| 4708 | case COND_JUMP86: |
| 4709 | if (size == 2 |
| 4710 | && (!no_cond_jump_promotion || fragP->fr_var != NO_RELOC)) |
| 4711 | { |
| 4712 | /* Negate the condition, and branch past an |
| 4713 | unconditional jump. */ |
| 4714 | opcode[0] ^= 1; |
| 4715 | opcode[1] = 3; |
| 4716 | /* Insert an unconditional jump. */ |
| 4717 | opcode[2] = 0xe9; |
| 4718 | /* We added two extra opcode bytes, and have a two byte |
| 4719 | offset. */ |
| 4720 | fragP->fr_fix += 2 + 2; |
| 4721 | fix_new (fragP, old_fr_fix + 2, 2, |
| 4722 | fragP->fr_symbol, |
| 4723 | fragP->fr_offset, 1, |
| 4724 | reloc_type); |
| 4725 | break; |
| 4726 | } |
| 4727 | /* Fall through. */ |
| 4728 | |
| 4729 | case COND_JUMP: |
| 4730 | if (no_cond_jump_promotion && fragP->fr_var == NO_RELOC) |
| 4731 | { |
| 4732 | fixS *fixP; |
| 4733 | |
| 4734 | fragP->fr_fix += 1; |
| 4735 | fixP = fix_new (fragP, old_fr_fix, 1, |
| 4736 | fragP->fr_symbol, |
| 4737 | fragP->fr_offset, 1, |
| 4738 | BFD_RELOC_8_PCREL); |
| 4739 | fixP->fx_signed = 1; |
| 4740 | break; |
| 4741 | } |
| 4742 | |
| 4743 | /* This changes the byte-displacement jump 0x7N |
| 4744 | to the (d)word-displacement jump 0x0f,0x8N. */ |
| 4745 | opcode[1] = opcode[0] + 0x10; |
| 4746 | opcode[0] = TWO_BYTE_OPCODE_ESCAPE; |
| 4747 | /* We've added an opcode byte. */ |
| 4748 | fragP->fr_fix += 1 + size; |
| 4749 | fix_new (fragP, old_fr_fix + 1, size, |
| 4750 | fragP->fr_symbol, |
| 4751 | fragP->fr_offset, 1, |
| 4752 | reloc_type); |
| 4753 | break; |
| 4754 | |
| 4755 | default: |
| 4756 | BAD_CASE (fragP->fr_subtype); |
| 4757 | break; |
| 4758 | } |
| 4759 | frag_wane (fragP); |
| 4760 | return fragP->fr_fix - old_fr_fix; |
| 4761 | } |
| 4762 | |
| 4763 | /* Guess size depending on current relax state. Initially the relax |
| 4764 | state will correspond to a short jump and we return 1, because |
| 4765 | the variable part of the frag (the branch offset) is one byte |
| 4766 | long. However, we can relax a section more than once and in that |
| 4767 | case we must either set fr_subtype back to the unrelaxed state, |
| 4768 | or return the value for the appropriate branch. */ |
| 4769 | return md_relax_table[fragP->fr_subtype].rlx_length; |
| 4770 | } |
| 4771 | |
| 4772 | /* Called after relax() is finished. |
| 4773 | |
| 4774 | In: Address of frag. |
| 4775 | fr_type == rs_machine_dependent. |
| 4776 | fr_subtype is what the address relaxed to. |
| 4777 | |
| 4778 | Out: Any fixSs and constants are set up. |
| 4779 | Caller will turn frag into a ".space 0". */ |
| 4780 | |
| 4781 | void |
| 4782 | md_convert_frag (abfd, sec, fragP) |
| 4783 | bfd *abfd ATTRIBUTE_UNUSED; |
| 4784 | segT sec ATTRIBUTE_UNUSED; |
| 4785 | fragS *fragP; |
| 4786 | { |
| 4787 | unsigned char *opcode; |
| 4788 | unsigned char *where_to_put_displacement = NULL; |
| 4789 | offsetT target_address; |
| 4790 | offsetT opcode_address; |
| 4791 | unsigned int extension = 0; |
| 4792 | offsetT displacement_from_opcode_start; |
| 4793 | |
| 4794 | opcode = (unsigned char *) fragP->fr_opcode; |
| 4795 | |
| 4796 | /* Address we want to reach in file space. */ |
| 4797 | target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset; |
| 4798 | |
| 4799 | /* Address opcode resides at in file space. */ |
| 4800 | opcode_address = fragP->fr_address + fragP->fr_fix; |
| 4801 | |
| 4802 | /* Displacement from opcode start to fill into instruction. */ |
| 4803 | displacement_from_opcode_start = target_address - opcode_address; |
| 4804 | |
| 4805 | if ((fragP->fr_subtype & BIG) == 0) |
| 4806 | { |
| 4807 | /* Don't have to change opcode. */ |
| 4808 | extension = 1; /* 1 opcode + 1 displacement */ |
| 4809 | where_to_put_displacement = &opcode[1]; |
| 4810 | } |
| 4811 | else |
| 4812 | { |
| 4813 | if (no_cond_jump_promotion |
| 4814 | && TYPE_FROM_RELAX_STATE (fragP->fr_subtype) != UNCOND_JUMP) |
| 4815 | as_warn_where (fragP->fr_file, fragP->fr_line, _("long jump required")); |
| 4816 | |
| 4817 | switch (fragP->fr_subtype) |
| 4818 | { |
| 4819 | case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG): |
| 4820 | extension = 4; /* 1 opcode + 4 displacement */ |
| 4821 | opcode[0] = 0xe9; |
| 4822 | where_to_put_displacement = &opcode[1]; |
| 4823 | break; |
| 4824 | |
| 4825 | case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16): |
| 4826 | extension = 2; /* 1 opcode + 2 displacement */ |
| 4827 | opcode[0] = 0xe9; |
| 4828 | where_to_put_displacement = &opcode[1]; |
| 4829 | break; |
| 4830 | |
| 4831 | case ENCODE_RELAX_STATE (COND_JUMP, BIG): |
| 4832 | case ENCODE_RELAX_STATE (COND_JUMP86, BIG): |
| 4833 | extension = 5; /* 2 opcode + 4 displacement */ |
| 4834 | opcode[1] = opcode[0] + 0x10; |
| 4835 | opcode[0] = TWO_BYTE_OPCODE_ESCAPE; |
| 4836 | where_to_put_displacement = &opcode[2]; |
| 4837 | break; |
| 4838 | |
| 4839 | case ENCODE_RELAX_STATE (COND_JUMP, BIG16): |
| 4840 | extension = 3; /* 2 opcode + 2 displacement */ |
| 4841 | opcode[1] = opcode[0] + 0x10; |
| 4842 | opcode[0] = TWO_BYTE_OPCODE_ESCAPE; |
| 4843 | where_to_put_displacement = &opcode[2]; |
| 4844 | break; |
| 4845 | |
| 4846 | case ENCODE_RELAX_STATE (COND_JUMP86, BIG16): |
| 4847 | extension = 4; |
| 4848 | opcode[0] ^= 1; |
| 4849 | opcode[1] = 3; |
| 4850 | opcode[2] = 0xe9; |
| 4851 | where_to_put_displacement = &opcode[3]; |
| 4852 | break; |
| 4853 | |
| 4854 | default: |
| 4855 | BAD_CASE (fragP->fr_subtype); |
| 4856 | break; |
| 4857 | } |
| 4858 | } |
| 4859 | |
| 4860 | /* Now put displacement after opcode. */ |
| 4861 | md_number_to_chars ((char *) where_to_put_displacement, |
| 4862 | (valueT) (displacement_from_opcode_start - extension), |
| 4863 | DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype)); |
| 4864 | fragP->fr_fix += extension; |
| 4865 | } |
| 4866 | \f |
| 4867 | /* Size of byte displacement jmp. */ |
| 4868 | int md_short_jump_size = 2; |
| 4869 | |
| 4870 | /* Size of dword displacement jmp. */ |
| 4871 | int md_long_jump_size = 5; |
| 4872 | |
| 4873 | void |
| 4874 | md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol) |
| 4875 | char *ptr; |
| 4876 | addressT from_addr, to_addr; |
| 4877 | fragS *frag ATTRIBUTE_UNUSED; |
| 4878 | symbolS *to_symbol ATTRIBUTE_UNUSED; |
| 4879 | { |
| 4880 | offsetT offset; |
| 4881 | |
| 4882 | offset = to_addr - (from_addr + 2); |
| 4883 | /* Opcode for byte-disp jump. */ |
| 4884 | md_number_to_chars (ptr, (valueT) 0xeb, 1); |
| 4885 | md_number_to_chars (ptr + 1, (valueT) offset, 1); |
| 4886 | } |
| 4887 | |
| 4888 | void |
| 4889 | md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol) |
| 4890 | char *ptr; |
| 4891 | addressT from_addr, to_addr; |
| 4892 | fragS *frag ATTRIBUTE_UNUSED; |
| 4893 | symbolS *to_symbol ATTRIBUTE_UNUSED; |
| 4894 | { |
| 4895 | offsetT offset; |
| 4896 | |
| 4897 | offset = to_addr - (from_addr + 5); |
| 4898 | md_number_to_chars (ptr, (valueT) 0xe9, 1); |
| 4899 | md_number_to_chars (ptr + 1, (valueT) offset, 4); |
| 4900 | } |
| 4901 | \f |
| 4902 | /* Apply a fixup (fixS) to segment data, once it has been determined |
| 4903 | by our caller that we have all the info we need to fix it up. |
| 4904 | |
| 4905 | On the 386, immediates, displacements, and data pointers are all in |
| 4906 | the same (little-endian) format, so we don't need to care about which |
| 4907 | we are handling. */ |
| 4908 | |
| 4909 | void |
| 4910 | md_apply_fix (fixP, valP, seg) |
| 4911 | /* The fix we're to put in. */ |
| 4912 | fixS *fixP; |
| 4913 | /* Pointer to the value of the bits. */ |
| 4914 | valueT *valP; |
| 4915 | /* Segment fix is from. */ |
| 4916 | segT seg ATTRIBUTE_UNUSED; |
| 4917 | { |
| 4918 | char *p = fixP->fx_where + fixP->fx_frag->fr_literal; |
| 4919 | valueT value = *valP; |
| 4920 | |
| 4921 | #if !defined (TE_Mach) |
| 4922 | if (fixP->fx_pcrel) |
| 4923 | { |
| 4924 | switch (fixP->fx_r_type) |
| 4925 | { |
| 4926 | default: |
| 4927 | break; |
| 4928 | |
| 4929 | case BFD_RELOC_64: |
| 4930 | fixP->fx_r_type = BFD_RELOC_64_PCREL; |
| 4931 | break; |
| 4932 | case BFD_RELOC_32: |
| 4933 | case BFD_RELOC_X86_64_32S: |
| 4934 | fixP->fx_r_type = BFD_RELOC_32_PCREL; |
| 4935 | break; |
| 4936 | case BFD_RELOC_16: |
| 4937 | fixP->fx_r_type = BFD_RELOC_16_PCREL; |
| 4938 | break; |
| 4939 | case BFD_RELOC_8: |
| 4940 | fixP->fx_r_type = BFD_RELOC_8_PCREL; |
| 4941 | break; |
| 4942 | } |
| 4943 | } |
| 4944 | |
| 4945 | if (fixP->fx_addsy != NULL |
| 4946 | && (fixP->fx_r_type == BFD_RELOC_32_PCREL |
| 4947 | || fixP->fx_r_type == BFD_RELOC_64_PCREL |
| 4948 | || fixP->fx_r_type == BFD_RELOC_16_PCREL |
| 4949 | || fixP->fx_r_type == BFD_RELOC_8_PCREL) |
| 4950 | && !use_rela_relocations) |
| 4951 | { |
| 4952 | /* This is a hack. There should be a better way to handle this. |
| 4953 | This covers for the fact that bfd_install_relocation will |
| 4954 | subtract the current location (for partial_inplace, PC relative |
| 4955 | relocations); see more below. */ |
| 4956 | #ifndef OBJ_AOUT |
| 4957 | if (IS_ELF |
| 4958 | #ifdef TE_PE |
| 4959 | || OUTPUT_FLAVOR == bfd_target_coff_flavour |
| 4960 | #endif |
| 4961 | ) |
| 4962 | value += fixP->fx_where + fixP->fx_frag->fr_address; |
| 4963 | #endif |
| 4964 | #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) |
| 4965 | if (IS_ELF) |
| 4966 | { |
| 4967 | segT sym_seg = S_GET_SEGMENT (fixP->fx_addsy); |
| 4968 | |
| 4969 | if ((sym_seg == seg |
| 4970 | || (symbol_section_p (fixP->fx_addsy) |
| 4971 | && sym_seg != absolute_section)) |
| 4972 | && !generic_force_reloc (fixP)) |
| 4973 | { |
| 4974 | /* Yes, we add the values in twice. This is because |
| 4975 | bfd_install_relocation subtracts them out again. I think |
| 4976 | bfd_install_relocation is broken, but I don't dare change |
| 4977 | it. FIXME. */ |
| 4978 | value += fixP->fx_where + fixP->fx_frag->fr_address; |
| 4979 | } |
| 4980 | } |
| 4981 | #endif |
| 4982 | #if defined (OBJ_COFF) && defined (TE_PE) |
| 4983 | /* For some reason, the PE format does not store a |
| 4984 | section address offset for a PC relative symbol. */ |
| 4985 | if (S_GET_SEGMENT (fixP->fx_addsy) != seg |
| 4986 | || S_IS_WEAK (fixP->fx_addsy)) |
| 4987 | value += md_pcrel_from (fixP); |
| 4988 | #endif |
| 4989 | } |
| 4990 | |
| 4991 | /* Fix a few things - the dynamic linker expects certain values here, |
| 4992 | and we must not disappoint it. */ |
| 4993 | #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) |
| 4994 | if (IS_ELF && fixP->fx_addsy) |
| 4995 | switch (fixP->fx_r_type) |
| 4996 | { |
| 4997 | case BFD_RELOC_386_PLT32: |
| 4998 | case BFD_RELOC_X86_64_PLT32: |
| 4999 | /* Make the jump instruction point to the address of the operand. At |
| 5000 | runtime we merely add the offset to the actual PLT entry. */ |
| 5001 | value = -4; |
| 5002 | break; |
| 5003 | |
| 5004 | case BFD_RELOC_386_TLS_GD: |
| 5005 | case BFD_RELOC_386_TLS_LDM: |
| 5006 | case BFD_RELOC_386_TLS_IE_32: |
| 5007 | case BFD_RELOC_386_TLS_IE: |
| 5008 | case BFD_RELOC_386_TLS_GOTIE: |
| 5009 | case BFD_RELOC_X86_64_TLSGD: |
| 5010 | case BFD_RELOC_X86_64_TLSLD: |
| 5011 | case BFD_RELOC_X86_64_GOTTPOFF: |
| 5012 | value = 0; /* Fully resolved at runtime. No addend. */ |
| 5013 | /* Fallthrough */ |
| 5014 | case BFD_RELOC_386_TLS_LE: |
| 5015 | case BFD_RELOC_386_TLS_LDO_32: |
| 5016 | case BFD_RELOC_386_TLS_LE_32: |
| 5017 | case BFD_RELOC_X86_64_DTPOFF32: |
| 5018 | case BFD_RELOC_X86_64_DTPOFF64: |
| 5019 | case BFD_RELOC_X86_64_TPOFF32: |
| 5020 | case BFD_RELOC_X86_64_TPOFF64: |
| 5021 | S_SET_THREAD_LOCAL (fixP->fx_addsy); |
| 5022 | break; |
| 5023 | |
| 5024 | case BFD_RELOC_386_GOT32: |
| 5025 | case BFD_RELOC_X86_64_GOT32: |
| 5026 | value = 0; /* Fully resolved at runtime. No addend. */ |
| 5027 | break; |
| 5028 | |
| 5029 | case BFD_RELOC_VTABLE_INHERIT: |
| 5030 | case BFD_RELOC_VTABLE_ENTRY: |
| 5031 | fixP->fx_done = 0; |
| 5032 | return; |
| 5033 | |
| 5034 | default: |
| 5035 | break; |
| 5036 | } |
| 5037 | #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */ |
| 5038 | *valP = value; |
| 5039 | #endif /* !defined (TE_Mach) */ |
| 5040 | |
| 5041 | /* Are we finished with this relocation now? */ |
| 5042 | if (fixP->fx_addsy == NULL) |
| 5043 | fixP->fx_done = 1; |
| 5044 | else if (use_rela_relocations) |
| 5045 | { |
| 5046 | fixP->fx_no_overflow = 1; |
| 5047 | /* Remember value for tc_gen_reloc. */ |
| 5048 | fixP->fx_addnumber = value; |
| 5049 | value = 0; |
| 5050 | } |
| 5051 | |
| 5052 | md_number_to_chars (p, value, fixP->fx_size); |
| 5053 | } |
| 5054 | \f |
| 5055 | #define MAX_LITTLENUMS 6 |
| 5056 | |
| 5057 | /* Turn the string pointed to by litP into a floating point constant |
| 5058 | of type TYPE, and emit the appropriate bytes. The number of |
| 5059 | LITTLENUMS emitted is stored in *SIZEP. An error message is |
| 5060 | returned, or NULL on OK. */ |
| 5061 | |
| 5062 | char * |
| 5063 | md_atof (type, litP, sizeP) |
| 5064 | int type; |
| 5065 | char *litP; |
| 5066 | int *sizeP; |
| 5067 | { |
| 5068 | int prec; |
| 5069 | LITTLENUM_TYPE words[MAX_LITTLENUMS]; |
| 5070 | LITTLENUM_TYPE *wordP; |
| 5071 | char *t; |
| 5072 | |
| 5073 | switch (type) |
| 5074 | { |
| 5075 | case 'f': |
| 5076 | case 'F': |
| 5077 | prec = 2; |
| 5078 | break; |
| 5079 | |
| 5080 | case 'd': |
| 5081 | case 'D': |
| 5082 | prec = 4; |
| 5083 | break; |
| 5084 | |
| 5085 | case 'x': |
| 5086 | case 'X': |
| 5087 | prec = 5; |
| 5088 | break; |
| 5089 | |
| 5090 | default: |
| 5091 | *sizeP = 0; |
| 5092 | return _("Bad call to md_atof ()"); |
| 5093 | } |
| 5094 | t = atof_ieee (input_line_pointer, type, words); |
| 5095 | if (t) |
| 5096 | input_line_pointer = t; |
| 5097 | |
| 5098 | *sizeP = prec * sizeof (LITTLENUM_TYPE); |
| 5099 | /* This loops outputs the LITTLENUMs in REVERSE order; in accord with |
| 5100 | the bigendian 386. */ |
| 5101 | for (wordP = words + prec - 1; prec--;) |
| 5102 | { |
| 5103 | md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE)); |
| 5104 | litP += sizeof (LITTLENUM_TYPE); |
| 5105 | } |
| 5106 | return 0; |
| 5107 | } |
| 5108 | \f |
| 5109 | static char output_invalid_buf[8]; |
| 5110 | |
| 5111 | static char * |
| 5112 | output_invalid (c) |
| 5113 | int c; |
| 5114 | { |
| 5115 | if (ISPRINT (c)) |
| 5116 | sprintf (output_invalid_buf, "'%c'", c); |
| 5117 | else |
| 5118 | sprintf (output_invalid_buf, "(0x%x)", (unsigned) c); |
| 5119 | return output_invalid_buf; |
| 5120 | } |
| 5121 | |
| 5122 | /* REG_STRING starts *before* REGISTER_PREFIX. */ |
| 5123 | |
| 5124 | static const reg_entry * |
| 5125 | parse_register (reg_string, end_op) |
| 5126 | char *reg_string; |
| 5127 | char **end_op; |
| 5128 | { |
| 5129 | char *s = reg_string; |
| 5130 | char *p; |
| 5131 | char reg_name_given[MAX_REG_NAME_SIZE + 1]; |
| 5132 | const reg_entry *r; |
| 5133 | |
| 5134 | /* Skip possible REGISTER_PREFIX and possible whitespace. */ |
| 5135 | if (*s == REGISTER_PREFIX) |
| 5136 | ++s; |
| 5137 | |
| 5138 | if (is_space_char (*s)) |
| 5139 | ++s; |
| 5140 | |
| 5141 | p = reg_name_given; |
| 5142 | while ((*p++ = register_chars[(unsigned char) *s]) != '\0') |
| 5143 | { |
| 5144 | if (p >= reg_name_given + MAX_REG_NAME_SIZE) |
| 5145 | return (const reg_entry *) NULL; |
| 5146 | s++; |
| 5147 | } |
| 5148 | |
| 5149 | /* For naked regs, make sure that we are not dealing with an identifier. |
| 5150 | This prevents confusing an identifier like `eax_var' with register |
| 5151 | `eax'. */ |
| 5152 | if (allow_naked_reg && identifier_chars[(unsigned char) *s]) |
| 5153 | return (const reg_entry *) NULL; |
| 5154 | |
| 5155 | *end_op = s; |
| 5156 | |
| 5157 | r = (const reg_entry *) hash_find (reg_hash, reg_name_given); |
| 5158 | |
| 5159 | /* Handle floating point regs, allowing spaces in the (i) part. */ |
| 5160 | if (r == i386_regtab /* %st is first entry of table */) |
| 5161 | { |
| 5162 | if (is_space_char (*s)) |
| 5163 | ++s; |
| 5164 | if (*s == '(') |
| 5165 | { |
| 5166 | ++s; |
| 5167 | if (is_space_char (*s)) |
| 5168 | ++s; |
| 5169 | if (*s >= '0' && *s <= '7') |
| 5170 | { |
| 5171 | r = &i386_float_regtab[*s - '0']; |
| 5172 | ++s; |
| 5173 | if (is_space_char (*s)) |
| 5174 | ++s; |
| 5175 | if (*s == ')') |
| 5176 | { |
| 5177 | *end_op = s + 1; |
| 5178 | return r; |
| 5179 | } |
| 5180 | } |
| 5181 | /* We have "%st(" then garbage. */ |
| 5182 | return (const reg_entry *) NULL; |
| 5183 | } |
| 5184 | } |
| 5185 | |
| 5186 | if (r != NULL |
| 5187 | && ((r->reg_flags & (RegRex64 | RegRex)) | (r->reg_type & Reg64)) != 0 |
| 5188 | && (r->reg_type != Control || !(cpu_arch_flags & CpuSledgehammer)) |
| 5189 | && flag_code != CODE_64BIT) |
| 5190 | return (const reg_entry *) NULL; |
| 5191 | |
| 5192 | return r; |
| 5193 | } |
| 5194 | \f |
| 5195 | #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) |
| 5196 | const char *md_shortopts = "kVQ:sqn"; |
| 5197 | #else |
| 5198 | const char *md_shortopts = "qn"; |
| 5199 | #endif |
| 5200 | |
| 5201 | struct option md_longopts[] = { |
| 5202 | #define OPTION_32 (OPTION_MD_BASE + 0) |
| 5203 | {"32", no_argument, NULL, OPTION_32}, |
| 5204 | #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) |
| 5205 | #define OPTION_64 (OPTION_MD_BASE + 1) |
| 5206 | {"64", no_argument, NULL, OPTION_64}, |
| 5207 | #endif |
| 5208 | {NULL, no_argument, NULL, 0} |
| 5209 | }; |
| 5210 | size_t md_longopts_size = sizeof (md_longopts); |
| 5211 | |
| 5212 | int |
| 5213 | md_parse_option (c, arg) |
| 5214 | int c; |
| 5215 | char *arg ATTRIBUTE_UNUSED; |
| 5216 | { |
| 5217 | switch (c) |
| 5218 | { |
| 5219 | case 'n': |
| 5220 | optimize_align_code = 0; |
| 5221 | break; |
| 5222 | |
| 5223 | case 'q': |
| 5224 | quiet_warnings = 1; |
| 5225 | break; |
| 5226 | |
| 5227 | #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) |
| 5228 | /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section |
| 5229 | should be emitted or not. FIXME: Not implemented. */ |
| 5230 | case 'Q': |
| 5231 | break; |
| 5232 | |
| 5233 | /* -V: SVR4 argument to print version ID. */ |
| 5234 | case 'V': |
| 5235 | print_version_id (); |
| 5236 | break; |
| 5237 | |
| 5238 | /* -k: Ignore for FreeBSD compatibility. */ |
| 5239 | case 'k': |
| 5240 | break; |
| 5241 | |
| 5242 | case 's': |
| 5243 | /* -s: On i386 Solaris, this tells the native assembler to use |
| 5244 | .stab instead of .stab.excl. We always use .stab anyhow. */ |
| 5245 | break; |
| 5246 | |
| 5247 | case OPTION_64: |
| 5248 | { |
| 5249 | const char **list, **l; |
| 5250 | |
| 5251 | list = bfd_target_list (); |
| 5252 | for (l = list; *l != NULL; l++) |
| 5253 | if (strcmp (*l, "elf64-x86-64") == 0) |
| 5254 | { |
| 5255 | default_arch = "x86_64"; |
| 5256 | break; |
| 5257 | } |
| 5258 | if (*l == NULL) |
| 5259 | as_fatal (_("No compiled in support for x86_64")); |
| 5260 | free (list); |
| 5261 | } |
| 5262 | break; |
| 5263 | #endif |
| 5264 | |
| 5265 | case OPTION_32: |
| 5266 | default_arch = "i386"; |
| 5267 | break; |
| 5268 | |
| 5269 | default: |
| 5270 | return 0; |
| 5271 | } |
| 5272 | return 1; |
| 5273 | } |
| 5274 | |
| 5275 | void |
| 5276 | md_show_usage (stream) |
| 5277 | FILE *stream; |
| 5278 | { |
| 5279 | #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) |
| 5280 | fprintf (stream, _("\ |
| 5281 | -Q ignored\n\ |
| 5282 | -V print assembler version number\n\ |
| 5283 | -k ignored\n\ |
| 5284 | -n Do not optimize code alignment\n\ |
| 5285 | -q quieten some warnings\n\ |
| 5286 | -s ignored\n")); |
| 5287 | #else |
| 5288 | fprintf (stream, _("\ |
| 5289 | -n Do not optimize code alignment\n\ |
| 5290 | -q quieten some warnings\n")); |
| 5291 | #endif |
| 5292 | } |
| 5293 | |
| 5294 | #if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \ |
| 5295 | || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) |
| 5296 | |
| 5297 | /* Pick the target format to use. */ |
| 5298 | |
| 5299 | const char * |
| 5300 | i386_target_format () |
| 5301 | { |
| 5302 | if (!strcmp (default_arch, "x86_64")) |
| 5303 | set_code_flag (CODE_64BIT); |
| 5304 | else if (!strcmp (default_arch, "i386")) |
| 5305 | set_code_flag (CODE_32BIT); |
| 5306 | else |
| 5307 | as_fatal (_("Unknown architecture")); |
| 5308 | switch (OUTPUT_FLAVOR) |
| 5309 | { |
| 5310 | #ifdef OBJ_MAYBE_AOUT |
| 5311 | case bfd_target_aout_flavour: |
| 5312 | return AOUT_TARGET_FORMAT; |
| 5313 | #endif |
| 5314 | #ifdef OBJ_MAYBE_COFF |
| 5315 | case bfd_target_coff_flavour: |
| 5316 | return "coff-i386"; |
| 5317 | #endif |
| 5318 | #if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF) |
| 5319 | case bfd_target_elf_flavour: |
| 5320 | { |
| 5321 | if (flag_code == CODE_64BIT) |
| 5322 | { |
| 5323 | object_64bit = 1; |
| 5324 | use_rela_relocations = 1; |
| 5325 | } |
| 5326 | return flag_code == CODE_64BIT ? "elf64-x86-64" : ELF_TARGET_FORMAT; |
| 5327 | } |
| 5328 | #endif |
| 5329 | default: |
| 5330 | abort (); |
| 5331 | return NULL; |
| 5332 | } |
| 5333 | } |
| 5334 | |
| 5335 | #endif /* OBJ_MAYBE_ more than one */ |
| 5336 | |
| 5337 | #if (defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) |
| 5338 | void i386_elf_emit_arch_note () |
| 5339 | { |
| 5340 | if (IS_ELF && cpu_arch_name != NULL) |
| 5341 | { |
| 5342 | char *p; |
| 5343 | asection *seg = now_seg; |
| 5344 | subsegT subseg = now_subseg; |
| 5345 | Elf_Internal_Note i_note; |
| 5346 | Elf_External_Note e_note; |
| 5347 | asection *note_secp; |
| 5348 | int len; |
| 5349 | |
| 5350 | /* Create the .note section. */ |
| 5351 | note_secp = subseg_new (".note", 0); |
| 5352 | bfd_set_section_flags (stdoutput, |
| 5353 | note_secp, |
| 5354 | SEC_HAS_CONTENTS | SEC_READONLY); |
| 5355 | |
| 5356 | /* Process the arch string. */ |
| 5357 | len = strlen (cpu_arch_name); |
| 5358 | |
| 5359 | i_note.namesz = len + 1; |
| 5360 | i_note.descsz = 0; |
| 5361 | i_note.type = NT_ARCH; |
| 5362 | p = frag_more (sizeof (e_note.namesz)); |
| 5363 | md_number_to_chars (p, (valueT) i_note.namesz, sizeof (e_note.namesz)); |
| 5364 | p = frag_more (sizeof (e_note.descsz)); |
| 5365 | md_number_to_chars (p, (valueT) i_note.descsz, sizeof (e_note.descsz)); |
| 5366 | p = frag_more (sizeof (e_note.type)); |
| 5367 | md_number_to_chars (p, (valueT) i_note.type, sizeof (e_note.type)); |
| 5368 | p = frag_more (len + 1); |
| 5369 | strcpy (p, cpu_arch_name); |
| 5370 | |
| 5371 | frag_align (2, 0, 0); |
| 5372 | |
| 5373 | subseg_set (seg, subseg); |
| 5374 | } |
| 5375 | } |
| 5376 | #endif |
| 5377 | \f |
| 5378 | symbolS * |
| 5379 | md_undefined_symbol (name) |
| 5380 | char *name; |
| 5381 | { |
| 5382 | if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0] |
| 5383 | && name[1] == GLOBAL_OFFSET_TABLE_NAME[1] |
| 5384 | && name[2] == GLOBAL_OFFSET_TABLE_NAME[2] |
| 5385 | && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0) |
| 5386 | { |
| 5387 | if (!GOT_symbol) |
| 5388 | { |
| 5389 | if (symbol_find (name)) |
| 5390 | as_bad (_("GOT already in symbol table")); |
| 5391 | GOT_symbol = symbol_new (name, undefined_section, |
| 5392 | (valueT) 0, &zero_address_frag); |
| 5393 | }; |
| 5394 | return GOT_symbol; |
| 5395 | } |
| 5396 | return 0; |
| 5397 | } |
| 5398 | |
| 5399 | /* Round up a section size to the appropriate boundary. */ |
| 5400 | |
| 5401 | valueT |
| 5402 | md_section_align (segment, size) |
| 5403 | segT segment ATTRIBUTE_UNUSED; |
| 5404 | valueT size; |
| 5405 | { |
| 5406 | #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT)) |
| 5407 | if (OUTPUT_FLAVOR == bfd_target_aout_flavour) |
| 5408 | { |
| 5409 | /* For a.out, force the section size to be aligned. If we don't do |
| 5410 | this, BFD will align it for us, but it will not write out the |
| 5411 | final bytes of the section. This may be a bug in BFD, but it is |
| 5412 | easier to fix it here since that is how the other a.out targets |
| 5413 | work. */ |
| 5414 | int align; |
| 5415 | |
| 5416 | align = bfd_get_section_alignment (stdoutput, segment); |
| 5417 | size = ((size + (1 << align) - 1) & ((valueT) -1 << align)); |
| 5418 | } |
| 5419 | #endif |
| 5420 | |
| 5421 | return size; |
| 5422 | } |
| 5423 | |
| 5424 | /* On the i386, PC-relative offsets are relative to the start of the |
| 5425 | next instruction. That is, the address of the offset, plus its |
| 5426 | size, since the offset is always the last part of the insn. */ |
| 5427 | |
| 5428 | long |
| 5429 | md_pcrel_from (fixP) |
| 5430 | fixS *fixP; |
| 5431 | { |
| 5432 | return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address; |
| 5433 | } |
| 5434 | |
| 5435 | #ifndef I386COFF |
| 5436 | |
| 5437 | static void |
| 5438 | s_bss (ignore) |
| 5439 | int ignore ATTRIBUTE_UNUSED; |
| 5440 | { |
| 5441 | int temp; |
| 5442 | |
| 5443 | #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) |
| 5444 | if (IS_ELF) |
| 5445 | obj_elf_section_change_hook (); |
| 5446 | #endif |
| 5447 | temp = get_absolute_expression (); |
| 5448 | subseg_set (bss_section, (subsegT) temp); |
| 5449 | demand_empty_rest_of_line (); |
| 5450 | } |
| 5451 | |
| 5452 | #endif |
| 5453 | |
| 5454 | void |
| 5455 | i386_validate_fix (fixp) |
| 5456 | fixS *fixp; |
| 5457 | { |
| 5458 | if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol) |
| 5459 | { |
| 5460 | if (fixp->fx_r_type == BFD_RELOC_32_PCREL) |
| 5461 | { |
| 5462 | if (!object_64bit) |
| 5463 | abort (); |
| 5464 | fixp->fx_r_type = BFD_RELOC_X86_64_GOTPCREL; |
| 5465 | } |
| 5466 | else |
| 5467 | { |
| 5468 | if (!object_64bit) |
| 5469 | fixp->fx_r_type = BFD_RELOC_386_GOTOFF; |
| 5470 | else |
| 5471 | fixp->fx_r_type = BFD_RELOC_X86_64_GOTOFF64; |
| 5472 | } |
| 5473 | fixp->fx_subsy = 0; |
| 5474 | } |
| 5475 | } |
| 5476 | |
| 5477 | arelent * |
| 5478 | tc_gen_reloc (section, fixp) |
| 5479 | asection *section ATTRIBUTE_UNUSED; |
| 5480 | fixS *fixp; |
| 5481 | { |
| 5482 | arelent *rel; |
| 5483 | bfd_reloc_code_real_type code; |
| 5484 | |
| 5485 | switch (fixp->fx_r_type) |
| 5486 | { |
| 5487 | case BFD_RELOC_X86_64_PLT32: |
| 5488 | case BFD_RELOC_X86_64_GOT32: |
| 5489 | case BFD_RELOC_X86_64_GOTPCREL: |
| 5490 | case BFD_RELOC_386_PLT32: |
| 5491 | case BFD_RELOC_386_GOT32: |
| 5492 | case BFD_RELOC_386_GOTOFF: |
| 5493 | case BFD_RELOC_386_GOTPC: |
| 5494 | case BFD_RELOC_386_TLS_GD: |
| 5495 | case BFD_RELOC_386_TLS_LDM: |
| 5496 | case BFD_RELOC_386_TLS_LDO_32: |
| 5497 | case BFD_RELOC_386_TLS_IE_32: |
| 5498 | case BFD_RELOC_386_TLS_IE: |
| 5499 | case BFD_RELOC_386_TLS_GOTIE: |
| 5500 | case BFD_RELOC_386_TLS_LE_32: |
| 5501 | case BFD_RELOC_386_TLS_LE: |
| 5502 | case BFD_RELOC_X86_64_TLSGD: |
| 5503 | case BFD_RELOC_X86_64_TLSLD: |
| 5504 | case BFD_RELOC_X86_64_DTPOFF32: |
| 5505 | case BFD_RELOC_X86_64_DTPOFF64: |
| 5506 | case BFD_RELOC_X86_64_GOTTPOFF: |
| 5507 | case BFD_RELOC_X86_64_TPOFF32: |
| 5508 | case BFD_RELOC_X86_64_TPOFF64: |
| 5509 | case BFD_RELOC_X86_64_GOTOFF64: |
| 5510 | case BFD_RELOC_X86_64_GOTPC32: |
| 5511 | case BFD_RELOC_RVA: |
| 5512 | case BFD_RELOC_VTABLE_ENTRY: |
| 5513 | case BFD_RELOC_VTABLE_INHERIT: |
| 5514 | #ifdef TE_PE |
| 5515 | case BFD_RELOC_32_SECREL: |
| 5516 | #endif |
| 5517 | code = fixp->fx_r_type; |
| 5518 | break; |
| 5519 | case BFD_RELOC_X86_64_32S: |
| 5520 | if (!fixp->fx_pcrel) |
| 5521 | { |
| 5522 | /* Don't turn BFD_RELOC_X86_64_32S into BFD_RELOC_32. */ |
| 5523 | code = fixp->fx_r_type; |
| 5524 | break; |
| 5525 | } |
| 5526 | default: |
| 5527 | if (fixp->fx_pcrel) |
| 5528 | { |
| 5529 | switch (fixp->fx_size) |
| 5530 | { |
| 5531 | default: |
| 5532 | as_bad_where (fixp->fx_file, fixp->fx_line, |
| 5533 | _("can not do %d byte pc-relative relocation"), |
| 5534 | fixp->fx_size); |
| 5535 | code = BFD_RELOC_32_PCREL; |
| 5536 | break; |
| 5537 | case 1: code = BFD_RELOC_8_PCREL; break; |
| 5538 | case 2: code = BFD_RELOC_16_PCREL; break; |
| 5539 | case 4: code = BFD_RELOC_32_PCREL; break; |
| 5540 | #ifdef BFD64 |
| 5541 | case 8: code = BFD_RELOC_64_PCREL; break; |
| 5542 | #endif |
| 5543 | } |
| 5544 | } |
| 5545 | else |
| 5546 | { |
| 5547 | switch (fixp->fx_size) |
| 5548 | { |
| 5549 | default: |
| 5550 | as_bad_where (fixp->fx_file, fixp->fx_line, |
| 5551 | _("can not do %d byte relocation"), |
| 5552 | fixp->fx_size); |
| 5553 | code = BFD_RELOC_32; |
| 5554 | break; |
| 5555 | case 1: code = BFD_RELOC_8; break; |
| 5556 | case 2: code = BFD_RELOC_16; break; |
| 5557 | case 4: code = BFD_RELOC_32; break; |
| 5558 | #ifdef BFD64 |
| 5559 | case 8: code = BFD_RELOC_64; break; |
| 5560 | #endif |
| 5561 | } |
| 5562 | } |
| 5563 | break; |
| 5564 | } |
| 5565 | |
| 5566 | if ((code == BFD_RELOC_32 || code == BFD_RELOC_32_PCREL) |
| 5567 | && GOT_symbol |
| 5568 | && fixp->fx_addsy == GOT_symbol) |
| 5569 | { |
| 5570 | if (!object_64bit) |
| 5571 | code = BFD_RELOC_386_GOTPC; |
| 5572 | else |
| 5573 | code = BFD_RELOC_X86_64_GOTPC32; |
| 5574 | } |
| 5575 | |
| 5576 | rel = (arelent *) xmalloc (sizeof (arelent)); |
| 5577 | rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *)); |
| 5578 | *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy); |
| 5579 | |
| 5580 | rel->address = fixp->fx_frag->fr_address + fixp->fx_where; |
| 5581 | |
| 5582 | if (!use_rela_relocations) |
| 5583 | { |
| 5584 | /* HACK: Since i386 ELF uses Rel instead of Rela, encode the |
| 5585 | vtable entry to be used in the relocation's section offset. */ |
| 5586 | if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY) |
| 5587 | rel->address = fixp->fx_offset; |
| 5588 | |
| 5589 | rel->addend = 0; |
| 5590 | } |
| 5591 | /* Use the rela in 64bit mode. */ |
| 5592 | else |
| 5593 | { |
| 5594 | if (!fixp->fx_pcrel) |
| 5595 | rel->addend = fixp->fx_offset; |
| 5596 | else |
| 5597 | switch (code) |
| 5598 | { |
| 5599 | case BFD_RELOC_X86_64_PLT32: |
| 5600 | case BFD_RELOC_X86_64_GOT32: |
| 5601 | case BFD_RELOC_X86_64_GOTPCREL: |
| 5602 | case BFD_RELOC_X86_64_TLSGD: |
| 5603 | case BFD_RELOC_X86_64_TLSLD: |
| 5604 | case BFD_RELOC_X86_64_GOTTPOFF: |
| 5605 | rel->addend = fixp->fx_offset - fixp->fx_size; |
| 5606 | break; |
| 5607 | default: |
| 5608 | rel->addend = (section->vma |
| 5609 | - fixp->fx_size |
| 5610 | + fixp->fx_addnumber |
| 5611 | + md_pcrel_from (fixp)); |
| 5612 | break; |
| 5613 | } |
| 5614 | } |
| 5615 | |
| 5616 | rel->howto = bfd_reloc_type_lookup (stdoutput, code); |
| 5617 | if (rel->howto == NULL) |
| 5618 | { |
| 5619 | as_bad_where (fixp->fx_file, fixp->fx_line, |
| 5620 | _("cannot represent relocation type %s"), |
| 5621 | bfd_get_reloc_code_name (code)); |
| 5622 | /* Set howto to a garbage value so that we can keep going. */ |
| 5623 | rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32); |
| 5624 | assert (rel->howto != NULL); |
| 5625 | } |
| 5626 | |
| 5627 | return rel; |
| 5628 | } |
| 5629 | |
| 5630 | \f |
| 5631 | /* Parse operands using Intel syntax. This implements a recursive descent |
| 5632 | parser based on the BNF grammar published in Appendix B of the MASM 6.1 |
| 5633 | Programmer's Guide. |
| 5634 | |
| 5635 | FIXME: We do not recognize the full operand grammar defined in the MASM |
| 5636 | documentation. In particular, all the structure/union and |
| 5637 | high-level macro operands are missing. |
| 5638 | |
| 5639 | Uppercase words are terminals, lower case words are non-terminals. |
| 5640 | Objects surrounded by double brackets '[[' ']]' are optional. Vertical |
| 5641 | bars '|' denote choices. Most grammar productions are implemented in |
| 5642 | functions called 'intel_<production>'. |
| 5643 | |
| 5644 | Initial production is 'expr'. |
| 5645 | |
| 5646 | addOp + | - |
| 5647 | |
| 5648 | alpha [a-zA-Z] |
| 5649 | |
| 5650 | binOp & | AND | \| | OR | ^ | XOR |
| 5651 | |
| 5652 | byteRegister AL | AH | BL | BH | CL | CH | DL | DH |
| 5653 | |
| 5654 | constant digits [[ radixOverride ]] |
| 5655 | |
| 5656 | dataType BYTE | WORD | DWORD | FWORD | QWORD | TBYTE | OWORD | XMMWORD |
| 5657 | |
| 5658 | digits decdigit |
| 5659 | | digits decdigit |
| 5660 | | digits hexdigit |
| 5661 | |
| 5662 | decdigit [0-9] |
| 5663 | |
| 5664 | e04 e04 addOp e05 |
| 5665 | | e05 |
| 5666 | |
| 5667 | e05 e05 binOp e06 |
| 5668 | | e06 |
| 5669 | |
| 5670 | e06 e06 mulOp e09 |
| 5671 | | e09 |
| 5672 | |
| 5673 | e09 OFFSET e10 |
| 5674 | | SHORT e10 |
| 5675 | | + e10 |
| 5676 | | - e10 |
| 5677 | | ~ e10 |
| 5678 | | NOT e10 |
| 5679 | | e09 PTR e10 |
| 5680 | | e09 : e10 |
| 5681 | | e10 |
| 5682 | |
| 5683 | e10 e10 [ expr ] |
| 5684 | | e11 |
| 5685 | |
| 5686 | e11 ( expr ) |
| 5687 | | [ expr ] |
| 5688 | | constant |
| 5689 | | dataType |
| 5690 | | id |
| 5691 | | $ |
| 5692 | | register |
| 5693 | |
| 5694 | => expr expr cmpOp e04 |
| 5695 | | e04 |
| 5696 | |
| 5697 | gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX |
| 5698 | | BP | EBP | SP | ESP | DI | EDI | SI | ESI |
| 5699 | |
| 5700 | hexdigit a | b | c | d | e | f |
| 5701 | | A | B | C | D | E | F |
| 5702 | |
| 5703 | id alpha |
| 5704 | | id alpha |
| 5705 | | id decdigit |
| 5706 | |
| 5707 | mulOp * | / | % | MOD | << | SHL | >> | SHR |
| 5708 | |
| 5709 | quote " | ' |
| 5710 | |
| 5711 | register specialRegister |
| 5712 | | gpRegister |
| 5713 | | byteRegister |
| 5714 | |
| 5715 | segmentRegister CS | DS | ES | FS | GS | SS |
| 5716 | |
| 5717 | specialRegister CR0 | CR2 | CR3 | CR4 |
| 5718 | | DR0 | DR1 | DR2 | DR3 | DR6 | DR7 |
| 5719 | | TR3 | TR4 | TR5 | TR6 | TR7 |
| 5720 | |
| 5721 | We simplify the grammar in obvious places (e.g., register parsing is |
| 5722 | done by calling parse_register) and eliminate immediate left recursion |
| 5723 | to implement a recursive-descent parser. |
| 5724 | |
| 5725 | expr e04 expr' |
| 5726 | |
| 5727 | expr' cmpOp e04 expr' |
| 5728 | | Empty |
| 5729 | |
| 5730 | e04 e05 e04' |
| 5731 | |
| 5732 | e04' addOp e05 e04' |
| 5733 | | Empty |
| 5734 | |
| 5735 | e05 e06 e05' |
| 5736 | |
| 5737 | e05' binOp e06 e05' |
| 5738 | | Empty |
| 5739 | |
| 5740 | e06 e09 e06' |
| 5741 | |
| 5742 | e06' mulOp e09 e06' |
| 5743 | | Empty |
| 5744 | |
| 5745 | e09 OFFSET e10 e09' |
| 5746 | | SHORT e10' |
| 5747 | | + e10' |
| 5748 | | - e10' |
| 5749 | | ~ e10' |
| 5750 | | NOT e10' |
| 5751 | | e10 e09' |
| 5752 | |
| 5753 | e09' PTR e10 e09' |
| 5754 | | : e10 e09' |
| 5755 | | Empty |
| 5756 | |
| 5757 | e10 e11 e10' |
| 5758 | |
| 5759 | e10' [ expr ] e10' |
| 5760 | | Empty |
| 5761 | |
| 5762 | e11 ( expr ) |
| 5763 | | [ expr ] |
| 5764 | | BYTE |
| 5765 | | WORD |
| 5766 | | DWORD |
| 5767 | | FWORD |
| 5768 | | QWORD |
| 5769 | | TBYTE |
| 5770 | | OWORD |
| 5771 | | XMMWORD |
| 5772 | | . |
| 5773 | | $ |
| 5774 | | register |
| 5775 | | id |
| 5776 | | constant */ |
| 5777 | |
| 5778 | /* Parsing structure for the intel syntax parser. Used to implement the |
| 5779 | semantic actions for the operand grammar. */ |
| 5780 | struct intel_parser_s |
| 5781 | { |
| 5782 | char *op_string; /* The string being parsed. */ |
| 5783 | int got_a_float; /* Whether the operand is a float. */ |
| 5784 | int op_modifier; /* Operand modifier. */ |
| 5785 | int is_mem; /* 1 if operand is memory reference. */ |
| 5786 | int in_offset; /* >=1 if parsing operand of offset. */ |
| 5787 | int in_bracket; /* >=1 if parsing operand in brackets. */ |
| 5788 | const reg_entry *reg; /* Last register reference found. */ |
| 5789 | char *disp; /* Displacement string being built. */ |
| 5790 | char *next_operand; /* Resume point when splitting operands. */ |
| 5791 | }; |
| 5792 | |
| 5793 | static struct intel_parser_s intel_parser; |
| 5794 | |
| 5795 | /* Token structure for parsing intel syntax. */ |
| 5796 | struct intel_token |
| 5797 | { |
| 5798 | int code; /* Token code. */ |
| 5799 | const reg_entry *reg; /* Register entry for register tokens. */ |
| 5800 | char *str; /* String representation. */ |
| 5801 | }; |
| 5802 | |
| 5803 | static struct intel_token cur_token, prev_token; |
| 5804 | |
| 5805 | /* Token codes for the intel parser. Since T_SHORT is already used |
| 5806 | by COFF, undefine it first to prevent a warning. */ |
| 5807 | #define T_NIL -1 |
| 5808 | #define T_CONST 1 |
| 5809 | #define T_REG 2 |
| 5810 | #define T_BYTE 3 |
| 5811 | #define T_WORD 4 |
| 5812 | #define T_DWORD 5 |
| 5813 | #define T_FWORD 6 |
| 5814 | #define T_QWORD 7 |
| 5815 | #define T_TBYTE 8 |
| 5816 | #define T_XMMWORD 9 |
| 5817 | #undef T_SHORT |
| 5818 | #define T_SHORT 10 |
| 5819 | #define T_OFFSET 11 |
| 5820 | #define T_PTR 12 |
| 5821 | #define T_ID 13 |
| 5822 | #define T_SHL 14 |
| 5823 | #define T_SHR 15 |
| 5824 | |
| 5825 | /* Prototypes for intel parser functions. */ |
| 5826 | static int intel_match_token PARAMS ((int code)); |
| 5827 | static void intel_get_token PARAMS ((void)); |
| 5828 | static void intel_putback_token PARAMS ((void)); |
| 5829 | static int intel_expr PARAMS ((void)); |
| 5830 | static int intel_e04 PARAMS ((void)); |
| 5831 | static int intel_e05 PARAMS ((void)); |
| 5832 | static int intel_e06 PARAMS ((void)); |
| 5833 | static int intel_e09 PARAMS ((void)); |
| 5834 | static int intel_bracket_expr PARAMS ((void)); |
| 5835 | static int intel_e10 PARAMS ((void)); |
| 5836 | static int intel_e11 PARAMS ((void)); |
| 5837 | |
| 5838 | static int |
| 5839 | i386_intel_operand (operand_string, got_a_float) |
| 5840 | char *operand_string; |
| 5841 | int got_a_float; |
| 5842 | { |
| 5843 | int ret; |
| 5844 | char *p; |
| 5845 | |
| 5846 | p = intel_parser.op_string = xstrdup (operand_string); |
| 5847 | intel_parser.disp = (char *) xmalloc (strlen (operand_string) + 1); |
| 5848 | |
| 5849 | for (;;) |
| 5850 | { |
| 5851 | /* Initialize token holders. */ |
| 5852 | cur_token.code = prev_token.code = T_NIL; |
| 5853 | cur_token.reg = prev_token.reg = NULL; |
| 5854 | cur_token.str = prev_token.str = NULL; |
| 5855 | |
| 5856 | /* Initialize parser structure. */ |
| 5857 | intel_parser.got_a_float = got_a_float; |
| 5858 | intel_parser.op_modifier = 0; |
| 5859 | intel_parser.is_mem = 0; |
| 5860 | intel_parser.in_offset = 0; |
| 5861 | intel_parser.in_bracket = 0; |
| 5862 | intel_parser.reg = NULL; |
| 5863 | intel_parser.disp[0] = '\0'; |
| 5864 | intel_parser.next_operand = NULL; |
| 5865 | |
| 5866 | /* Read the first token and start the parser. */ |
| 5867 | intel_get_token (); |
| 5868 | ret = intel_expr (); |
| 5869 | |
| 5870 | if (!ret) |
| 5871 | break; |
| 5872 | |
| 5873 | if (cur_token.code != T_NIL) |
| 5874 | { |
| 5875 | as_bad (_("invalid operand for '%s' ('%s' unexpected)"), |
| 5876 | current_templates->start->name, cur_token.str); |
| 5877 | ret = 0; |
| 5878 | } |
| 5879 | /* If we found a memory reference, hand it over to i386_displacement |
| 5880 | to fill in the rest of the operand fields. */ |
| 5881 | else if (intel_parser.is_mem) |
| 5882 | { |
| 5883 | if ((i.mem_operands == 1 |
| 5884 | && (current_templates->start->opcode_modifier & IsString) == 0) |
| 5885 | || i.mem_operands == 2) |
| 5886 | { |
| 5887 | as_bad (_("too many memory references for '%s'"), |
| 5888 | current_templates->start->name); |
| 5889 | ret = 0; |
| 5890 | } |
| 5891 | else |
| 5892 | { |
| 5893 | char *s = intel_parser.disp; |
| 5894 | i.mem_operands++; |
| 5895 | |
| 5896 | if (!quiet_warnings && intel_parser.is_mem < 0) |
| 5897 | /* See the comments in intel_bracket_expr. */ |
| 5898 | as_warn (_("Treating `%s' as memory reference"), operand_string); |
| 5899 | |
| 5900 | /* Add the displacement expression. */ |
| 5901 | if (*s != '\0') |
| 5902 | ret = i386_displacement (s, s + strlen (s)); |
| 5903 | if (ret) |
| 5904 | { |
| 5905 | /* Swap base and index in 16-bit memory operands like |
| 5906 | [si+bx]. Since i386_index_check is also used in AT&T |
| 5907 | mode we have to do that here. */ |
| 5908 | if (i.base_reg |
| 5909 | && i.index_reg |
| 5910 | && (i.base_reg->reg_type & Reg16) |
| 5911 | && (i.index_reg->reg_type & Reg16) |
| 5912 | && i.base_reg->reg_num >= 6 |
| 5913 | && i.index_reg->reg_num < 6) |
| 5914 | { |
| 5915 | const reg_entry *base = i.index_reg; |
| 5916 | |
| 5917 | i.index_reg = i.base_reg; |
| 5918 | i.base_reg = base; |
| 5919 | } |
| 5920 | ret = i386_index_check (operand_string); |
| 5921 | } |
| 5922 | } |
| 5923 | } |
| 5924 | |
| 5925 | /* Constant and OFFSET expressions are handled by i386_immediate. */ |
| 5926 | else if ((intel_parser.op_modifier & (1 << T_OFFSET)) |
| 5927 | || intel_parser.reg == NULL) |
| 5928 | ret = i386_immediate (intel_parser.disp); |
| 5929 | |
| 5930 | if (intel_parser.next_operand && this_operand >= MAX_OPERANDS - 1) |
| 5931 | ret = 0; |
| 5932 | if (!ret || !intel_parser.next_operand) |
| 5933 | break; |
| 5934 | intel_parser.op_string = intel_parser.next_operand; |
| 5935 | this_operand = i.operands++; |
| 5936 | } |
| 5937 | |
| 5938 | free (p); |
| 5939 | free (intel_parser.disp); |
| 5940 | |
| 5941 | return ret; |
| 5942 | } |
| 5943 | |
| 5944 | #define NUM_ADDRESS_REGS (!!i.base_reg + !!i.index_reg) |
| 5945 | |
| 5946 | /* expr e04 expr' |
| 5947 | |
| 5948 | expr' cmpOp e04 expr' |
| 5949 | | Empty */ |
| 5950 | static int |
| 5951 | intel_expr () |
| 5952 | { |
| 5953 | /* XXX Implement the comparison operators. */ |
| 5954 | return intel_e04 (); |
| 5955 | } |
| 5956 | |
| 5957 | /* e04 e05 e04' |
| 5958 | |
| 5959 | e04' addOp e05 e04' |
| 5960 | | Empty */ |
| 5961 | static int |
| 5962 | intel_e04 () |
| 5963 | { |
| 5964 | int nregs = -1; |
| 5965 | |
| 5966 | for (;;) |
| 5967 | { |
| 5968 | if (!intel_e05()) |
| 5969 | return 0; |
| 5970 | |
| 5971 | if (nregs >= 0 && NUM_ADDRESS_REGS > nregs) |
| 5972 | i.base_reg = i386_regtab + REGNAM_AL; /* al is invalid as base */ |
| 5973 | |
| 5974 | if (cur_token.code == '+') |
| 5975 | nregs = -1; |
| 5976 | else if (cur_token.code == '-') |
| 5977 | nregs = NUM_ADDRESS_REGS; |
| 5978 | else |
| 5979 | return 1; |
| 5980 | |
| 5981 | strcat (intel_parser.disp, cur_token.str); |
| 5982 | intel_match_token (cur_token.code); |
| 5983 | } |
| 5984 | } |
| 5985 | |
| 5986 | /* e05 e06 e05' |
| 5987 | |
| 5988 | e05' binOp e06 e05' |
| 5989 | | Empty */ |
| 5990 | static int |
| 5991 | intel_e05 () |
| 5992 | { |
| 5993 | int nregs = ~NUM_ADDRESS_REGS; |
| 5994 | |
| 5995 | for (;;) |
| 5996 | { |
| 5997 | if (!intel_e06()) |
| 5998 | return 0; |
| 5999 | |
| 6000 | if (cur_token.code == '&' || cur_token.code == '|' || cur_token.code == '^') |
| 6001 | { |
| 6002 | char str[2]; |
| 6003 | |
| 6004 | str[0] = cur_token.code; |
| 6005 | str[1] = 0; |
| 6006 | strcat (intel_parser.disp, str); |
| 6007 | } |
| 6008 | else |
| 6009 | break; |
| 6010 | |
| 6011 | intel_match_token (cur_token.code); |
| 6012 | |
| 6013 | if (nregs < 0) |
| 6014 | nregs = ~nregs; |
| 6015 | } |
| 6016 | if (nregs >= 0 && NUM_ADDRESS_REGS > nregs) |
| 6017 | i.base_reg = i386_regtab + REGNAM_AL + 1; /* cl is invalid as base */ |
| 6018 | return 1; |
| 6019 | } |
| 6020 | |
| 6021 | /* e06 e09 e06' |
| 6022 | |
| 6023 | e06' mulOp e09 e06' |
| 6024 | | Empty */ |
| 6025 | static int |
| 6026 | intel_e06 () |
| 6027 | { |
| 6028 | int nregs = ~NUM_ADDRESS_REGS; |
| 6029 | |
| 6030 | for (;;) |
| 6031 | { |
| 6032 | if (!intel_e09()) |
| 6033 | return 0; |
| 6034 | |
| 6035 | if (cur_token.code == '*' || cur_token.code == '/' || cur_token.code == '%') |
| 6036 | { |
| 6037 | char str[2]; |
| 6038 | |
| 6039 | str[0] = cur_token.code; |
| 6040 | str[1] = 0; |
| 6041 | strcat (intel_parser.disp, str); |
| 6042 | } |
| 6043 | else if (cur_token.code == T_SHL) |
| 6044 | strcat (intel_parser.disp, "<<"); |
| 6045 | else if (cur_token.code == T_SHR) |
| 6046 | strcat (intel_parser.disp, ">>"); |
| 6047 | else |
| 6048 | break; |
| 6049 | |
| 6050 | intel_match_token (cur_token.code); |
| 6051 | |
| 6052 | if (nregs < 0) |
| 6053 | nregs = ~nregs; |
| 6054 | } |
| 6055 | if (nregs >= 0 && NUM_ADDRESS_REGS > nregs) |
| 6056 | i.base_reg = i386_regtab + REGNAM_AL + 2; /* dl is invalid as base */ |
| 6057 | return 1; |
| 6058 | } |
| 6059 | |
| 6060 | /* e09 OFFSET e09 |
| 6061 | | SHORT e09 |
| 6062 | | + e09 |
| 6063 | | - e09 |
| 6064 | | ~ e09 |
| 6065 | | NOT e09 |
| 6066 | | e10 e09' |
| 6067 | |
| 6068 | e09' PTR e10 e09' |
| 6069 | | : e10 e09' |
| 6070 | | Empty */ |
| 6071 | static int |
| 6072 | intel_e09 () |
| 6073 | { |
| 6074 | int nregs = ~NUM_ADDRESS_REGS; |
| 6075 | int in_offset = 0; |
| 6076 | |
| 6077 | for (;;) |
| 6078 | { |
| 6079 | /* Don't consume constants here. */ |
| 6080 | if (cur_token.code == '+' || cur_token.code == '-') |
| 6081 | { |
| 6082 | /* Need to look one token ahead - if the next token |
| 6083 | is a constant, the current token is its sign. */ |
| 6084 | int next_code; |
| 6085 | |
| 6086 | intel_match_token (cur_token.code); |
| 6087 | next_code = cur_token.code; |
| 6088 | intel_putback_token (); |
| 6089 | if (next_code == T_CONST) |
| 6090 | break; |
| 6091 | } |
| 6092 | |
| 6093 | /* e09 OFFSET e09 */ |
| 6094 | if (cur_token.code == T_OFFSET) |
| 6095 | { |
| 6096 | if (!in_offset++) |
| 6097 | ++intel_parser.in_offset; |
| 6098 | } |
| 6099 | |
| 6100 | /* e09 SHORT e09 */ |
| 6101 | else if (cur_token.code == T_SHORT) |
| 6102 | intel_parser.op_modifier |= 1 << T_SHORT; |
| 6103 | |
| 6104 | /* e09 + e09 */ |
| 6105 | else if (cur_token.code == '+') |
| 6106 | strcat (intel_parser.disp, "+"); |
| 6107 | |
| 6108 | /* e09 - e09 |
| 6109 | | ~ e09 |
| 6110 | | NOT e09 */ |
| 6111 | else if (cur_token.code == '-' || cur_token.code == '~') |
| 6112 | { |
| 6113 | char str[2]; |
| 6114 | |
| 6115 | if (nregs < 0) |
| 6116 | nregs = ~nregs; |
| 6117 | str[0] = cur_token.code; |
| 6118 | str[1] = 0; |
| 6119 | strcat (intel_parser.disp, str); |
| 6120 | } |
| 6121 | |
| 6122 | /* e09 e10 e09' */ |
| 6123 | else |
| 6124 | break; |
| 6125 | |
| 6126 | intel_match_token (cur_token.code); |
| 6127 | } |
| 6128 | |
| 6129 | for (;;) |
| 6130 | { |
| 6131 | if (!intel_e10 ()) |
| 6132 | return 0; |
| 6133 | |
| 6134 | /* e09' PTR e10 e09' */ |
| 6135 | if (cur_token.code == T_PTR) |
| 6136 | { |
| 6137 | char suffix; |
| 6138 | |
| 6139 | if (prev_token.code == T_BYTE) |
| 6140 | suffix = BYTE_MNEM_SUFFIX; |
| 6141 | |
| 6142 | else if (prev_token.code == T_WORD) |
| 6143 | { |
| 6144 | if (current_templates->start->name[0] == 'l' |
| 6145 | && current_templates->start->name[2] == 's' |
| 6146 | && current_templates->start->name[3] == 0) |
| 6147 | suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */ |
| 6148 | else if (intel_parser.got_a_float == 2) /* "fi..." */ |
| 6149 | suffix = SHORT_MNEM_SUFFIX; |
| 6150 | else |
| 6151 | suffix = WORD_MNEM_SUFFIX; |
| 6152 | } |
| 6153 | |
| 6154 | else if (prev_token.code == T_DWORD) |
| 6155 | { |
| 6156 | if (current_templates->start->name[0] == 'l' |
| 6157 | && current_templates->start->name[2] == 's' |
| 6158 | && current_templates->start->name[3] == 0) |
| 6159 | suffix = WORD_MNEM_SUFFIX; |
| 6160 | else if (flag_code == CODE_16BIT |
| 6161 | && (current_templates->start->opcode_modifier |
| 6162 | & (Jump | JumpDword))) |
| 6163 | suffix = LONG_DOUBLE_MNEM_SUFFIX; |
| 6164 | else if (intel_parser.got_a_float == 1) /* "f..." */ |
| 6165 | suffix = SHORT_MNEM_SUFFIX; |
| 6166 | else |
| 6167 | suffix = LONG_MNEM_SUFFIX; |
| 6168 | } |
| 6169 | |
| 6170 | else if (prev_token.code == T_FWORD) |
| 6171 | { |
| 6172 | if (current_templates->start->name[0] == 'l' |
| 6173 | && current_templates->start->name[2] == 's' |
| 6174 | && current_templates->start->name[3] == 0) |
| 6175 | suffix = LONG_MNEM_SUFFIX; |
| 6176 | else if (!intel_parser.got_a_float) |
| 6177 | { |
| 6178 | if (flag_code == CODE_16BIT) |
| 6179 | add_prefix (DATA_PREFIX_OPCODE); |
| 6180 | suffix = LONG_DOUBLE_MNEM_SUFFIX; |
| 6181 | } |
| 6182 | else |
| 6183 | suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */ |
| 6184 | } |
| 6185 | |
| 6186 | else if (prev_token.code == T_QWORD) |
| 6187 | { |
| 6188 | if (intel_parser.got_a_float == 1) /* "f..." */ |
| 6189 | suffix = LONG_MNEM_SUFFIX; |
| 6190 | else |
| 6191 | suffix = QWORD_MNEM_SUFFIX; |
| 6192 | } |
| 6193 | |
| 6194 | else if (prev_token.code == T_TBYTE) |
| 6195 | { |
| 6196 | if (intel_parser.got_a_float == 1) |
| 6197 | suffix = LONG_DOUBLE_MNEM_SUFFIX; |
| 6198 | else |
| 6199 | suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */ |
| 6200 | } |
| 6201 | |
| 6202 | else if (prev_token.code == T_XMMWORD) |
| 6203 | { |
| 6204 | /* XXX ignored for now, but accepted since gcc uses it */ |
| 6205 | suffix = 0; |
| 6206 | } |
| 6207 | |
| 6208 | else |
| 6209 | { |
| 6210 | as_bad (_("Unknown operand modifier `%s'"), prev_token.str); |
| 6211 | return 0; |
| 6212 | } |
| 6213 | |
| 6214 | /* Operands for jump/call using 'ptr' notation denote absolute |
| 6215 | addresses. */ |
| 6216 | if (current_templates->start->opcode_modifier & (Jump | JumpDword)) |
| 6217 | i.types[this_operand] |= JumpAbsolute; |
| 6218 | |
| 6219 | if (current_templates->start->base_opcode == 0x8d /* lea */) |
| 6220 | ; |
| 6221 | else if (!i.suffix) |
| 6222 | i.suffix = suffix; |
| 6223 | else if (i.suffix != suffix) |
| 6224 | { |
| 6225 | as_bad (_("Conflicting operand modifiers")); |
| 6226 | return 0; |
| 6227 | } |
| 6228 | |
| 6229 | } |
| 6230 | |
| 6231 | /* e09' : e10 e09' */ |
| 6232 | else if (cur_token.code == ':') |
| 6233 | { |
| 6234 | if (prev_token.code != T_REG) |
| 6235 | { |
| 6236 | /* While {call,jmp} SSSS:OOOO is MASM syntax only when SSSS is a |
| 6237 | segment/group identifier (which we don't have), using comma |
| 6238 | as the operand separator there is even less consistent, since |
| 6239 | there all branches only have a single operand. */ |
| 6240 | if (this_operand != 0 |
| 6241 | || intel_parser.in_offset |
| 6242 | || intel_parser.in_bracket |
| 6243 | || (!(current_templates->start->opcode_modifier |
| 6244 | & (Jump|JumpDword|JumpInterSegment)) |
| 6245 | && !(current_templates->start->operand_types[0] |
| 6246 | & JumpAbsolute))) |
| 6247 | return intel_match_token (T_NIL); |
| 6248 | /* Remember the start of the 2nd operand and terminate 1st |
| 6249 | operand here. |
| 6250 | XXX This isn't right, yet (when SSSS:OOOO is right operand of |
| 6251 | another expression), but it gets at least the simplest case |
| 6252 | (a plain number or symbol on the left side) right. */ |
| 6253 | intel_parser.next_operand = intel_parser.op_string; |
| 6254 | *--intel_parser.op_string = '\0'; |
| 6255 | return intel_match_token (':'); |
| 6256 | } |
| 6257 | } |
| 6258 | |
| 6259 | /* e09' Empty */ |
| 6260 | else |
| 6261 | break; |
| 6262 | |
| 6263 | intel_match_token (cur_token.code); |
| 6264 | |
| 6265 | } |
| 6266 | |
| 6267 | if (in_offset) |
| 6268 | { |
| 6269 | --intel_parser.in_offset; |
| 6270 | if (nregs < 0) |
| 6271 | nregs = ~nregs; |
| 6272 | if (NUM_ADDRESS_REGS > nregs) |
| 6273 | { |
| 6274 | as_bad (_("Invalid operand to `OFFSET'")); |
| 6275 | return 0; |
| 6276 | } |
| 6277 | intel_parser.op_modifier |= 1 << T_OFFSET; |
| 6278 | } |
| 6279 | |
| 6280 | if (nregs >= 0 && NUM_ADDRESS_REGS > nregs) |
| 6281 | i.base_reg = i386_regtab + REGNAM_AL + 3; /* bl is invalid as base */ |
| 6282 | return 1; |
| 6283 | } |
| 6284 | |
| 6285 | static int |
| 6286 | intel_bracket_expr () |
| 6287 | { |
| 6288 | int was_offset = intel_parser.op_modifier & (1 << T_OFFSET); |
| 6289 | const char *start = intel_parser.op_string; |
| 6290 | int len; |
| 6291 | |
| 6292 | if (i.op[this_operand].regs) |
| 6293 | return intel_match_token (T_NIL); |
| 6294 | |
| 6295 | intel_match_token ('['); |
| 6296 | |
| 6297 | /* Mark as a memory operand only if it's not already known to be an |
| 6298 | offset expression. If it's an offset expression, we need to keep |
| 6299 | the brace in. */ |
| 6300 | if (!intel_parser.in_offset) |
| 6301 | { |
| 6302 | ++intel_parser.in_bracket; |
| 6303 | |
| 6304 | /* Operands for jump/call inside brackets denote absolute addresses. */ |
| 6305 | if (current_templates->start->opcode_modifier & (Jump | JumpDword)) |
| 6306 | i.types[this_operand] |= JumpAbsolute; |
| 6307 | |
| 6308 | /* Unfortunately gas always diverged from MASM in a respect that can't |
| 6309 | be easily fixed without risking to break code sequences likely to be |
| 6310 | encountered (the testsuite even check for this): MASM doesn't consider |
| 6311 | an expression inside brackets unconditionally as a memory reference. |
| 6312 | When that is e.g. a constant, an offset expression, or the sum of the |
| 6313 | two, this is still taken as a constant load. gas, however, always |
| 6314 | treated these as memory references. As a compromise, we'll try to make |
| 6315 | offset expressions inside brackets work the MASM way (since that's |
| 6316 | less likely to be found in real world code), but make constants alone |
| 6317 | continue to work the traditional gas way. In either case, issue a |
| 6318 | warning. */ |
| 6319 | intel_parser.op_modifier &= ~was_offset; |
| 6320 | } |
| 6321 | else |
| 6322 | strcat (intel_parser.disp, "["); |
| 6323 | |
| 6324 | /* Add a '+' to the displacement string if necessary. */ |
| 6325 | if (*intel_parser.disp != '\0' |
| 6326 | && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+') |
| 6327 | strcat (intel_parser.disp, "+"); |
| 6328 | |
| 6329 | if (intel_expr () |
| 6330 | && (len = intel_parser.op_string - start - 1, |
| 6331 | intel_match_token (']'))) |
| 6332 | { |
| 6333 | /* Preserve brackets when the operand is an offset expression. */ |
| 6334 | if (intel_parser.in_offset) |
| 6335 | strcat (intel_parser.disp, "]"); |
| 6336 | else |
| 6337 | { |
| 6338 | --intel_parser.in_bracket; |
| 6339 | if (i.base_reg || i.index_reg) |
| 6340 | intel_parser.is_mem = 1; |
| 6341 | if (!intel_parser.is_mem) |
| 6342 | { |
| 6343 | if (!(intel_parser.op_modifier & (1 << T_OFFSET))) |
| 6344 | /* Defer the warning until all of the operand was parsed. */ |
| 6345 | intel_parser.is_mem = -1; |
| 6346 | else if (!quiet_warnings) |
| 6347 | as_warn (_("`[%.*s]' taken to mean just `%.*s'"), len, start, len, start); |
| 6348 | } |
| 6349 | } |
| 6350 | intel_parser.op_modifier |= was_offset; |
| 6351 | |
| 6352 | return 1; |
| 6353 | } |
| 6354 | return 0; |
| 6355 | } |
| 6356 | |
| 6357 | /* e10 e11 e10' |
| 6358 | |
| 6359 | e10' [ expr ] e10' |
| 6360 | | Empty */ |
| 6361 | static int |
| 6362 | intel_e10 () |
| 6363 | { |
| 6364 | if (!intel_e11 ()) |
| 6365 | return 0; |
| 6366 | |
| 6367 | while (cur_token.code == '[') |
| 6368 | { |
| 6369 | if (!intel_bracket_expr ()) |
| 6370 | return 0; |
| 6371 | } |
| 6372 | |
| 6373 | return 1; |
| 6374 | } |
| 6375 | |
| 6376 | /* e11 ( expr ) |
| 6377 | | [ expr ] |
| 6378 | | BYTE |
| 6379 | | WORD |
| 6380 | | DWORD |
| 6381 | | FWORD |
| 6382 | | QWORD |
| 6383 | | TBYTE |
| 6384 | | OWORD |
| 6385 | | XMMWORD |
| 6386 | | $ |
| 6387 | | . |
| 6388 | | register |
| 6389 | | id |
| 6390 | | constant */ |
| 6391 | static int |
| 6392 | intel_e11 () |
| 6393 | { |
| 6394 | switch (cur_token.code) |
| 6395 | { |
| 6396 | /* e11 ( expr ) */ |
| 6397 | case '(': |
| 6398 | intel_match_token ('('); |
| 6399 | strcat (intel_parser.disp, "("); |
| 6400 | |
| 6401 | if (intel_expr () && intel_match_token (')')) |
| 6402 | { |
| 6403 | strcat (intel_parser.disp, ")"); |
| 6404 | return 1; |
| 6405 | } |
| 6406 | return 0; |
| 6407 | |
| 6408 | /* e11 [ expr ] */ |
| 6409 | case '[': |
| 6410 | return intel_bracket_expr (); |
| 6411 | |
| 6412 | /* e11 $ |
| 6413 | | . */ |
| 6414 | case '.': |
| 6415 | strcat (intel_parser.disp, cur_token.str); |
| 6416 | intel_match_token (cur_token.code); |
| 6417 | |
| 6418 | /* Mark as a memory operand only if it's not already known to be an |
| 6419 | offset expression. */ |
| 6420 | if (!intel_parser.in_offset) |
| 6421 | intel_parser.is_mem = 1; |
| 6422 | |
| 6423 | return 1; |
| 6424 | |
| 6425 | /* e11 register */ |
| 6426 | case T_REG: |
| 6427 | { |
| 6428 | const reg_entry *reg = intel_parser.reg = cur_token.reg; |
| 6429 | |
| 6430 | intel_match_token (T_REG); |
| 6431 | |
| 6432 | /* Check for segment change. */ |
| 6433 | if (cur_token.code == ':') |
| 6434 | { |
| 6435 | if (!(reg->reg_type & (SReg2 | SReg3))) |
| 6436 | { |
| 6437 | as_bad (_("`%s' is not a valid segment register"), reg->reg_name); |
| 6438 | return 0; |
| 6439 | } |
| 6440 | else if (i.seg[i.mem_operands]) |
| 6441 | as_warn (_("Extra segment override ignored")); |
| 6442 | else |
| 6443 | { |
| 6444 | if (!intel_parser.in_offset) |
| 6445 | intel_parser.is_mem = 1; |
| 6446 | switch (reg->reg_num) |
| 6447 | { |
| 6448 | case 0: |
| 6449 | i.seg[i.mem_operands] = &es; |
| 6450 | break; |
| 6451 | case 1: |
| 6452 | i.seg[i.mem_operands] = &cs; |
| 6453 | break; |
| 6454 | case 2: |
| 6455 | i.seg[i.mem_operands] = &ss; |
| 6456 | break; |
| 6457 | case 3: |
| 6458 | i.seg[i.mem_operands] = &ds; |
| 6459 | break; |
| 6460 | case 4: |
| 6461 | i.seg[i.mem_operands] = &fs; |
| 6462 | break; |
| 6463 | case 5: |
| 6464 | i.seg[i.mem_operands] = &gs; |
| 6465 | break; |
| 6466 | } |
| 6467 | } |
| 6468 | } |
| 6469 | |
| 6470 | /* Not a segment register. Check for register scaling. */ |
| 6471 | else if (cur_token.code == '*') |
| 6472 | { |
| 6473 | if (!intel_parser.in_bracket) |
| 6474 | { |
| 6475 | as_bad (_("Register scaling only allowed in memory operands")); |
| 6476 | return 0; |
| 6477 | } |
| 6478 | |
| 6479 | if (reg->reg_type & Reg16) /* Disallow things like [si*1]. */ |
| 6480 | reg = i386_regtab + REGNAM_AX + 4; /* sp is invalid as index */ |
| 6481 | else if (i.index_reg) |
| 6482 | reg = i386_regtab + REGNAM_EAX + 4; /* esp is invalid as index */ |
| 6483 | |
| 6484 | /* What follows must be a valid scale. */ |
| 6485 | intel_match_token ('*'); |
| 6486 | i.index_reg = reg; |
| 6487 | i.types[this_operand] |= BaseIndex; |
| 6488 | |
| 6489 | /* Set the scale after setting the register (otherwise, |
| 6490 | i386_scale will complain) */ |
| 6491 | if (cur_token.code == '+' || cur_token.code == '-') |
| 6492 | { |
| 6493 | char *str, sign = cur_token.code; |
| 6494 | intel_match_token (cur_token.code); |
| 6495 | if (cur_token.code != T_CONST) |
| 6496 | { |
| 6497 | as_bad (_("Syntax error: Expecting a constant, got `%s'"), |
| 6498 | cur_token.str); |
| 6499 | return 0; |
| 6500 | } |
| 6501 | str = (char *) xmalloc (strlen (cur_token.str) + 2); |
| 6502 | strcpy (str + 1, cur_token.str); |
| 6503 | *str = sign; |
| 6504 | if (!i386_scale (str)) |
| 6505 | return 0; |
| 6506 | free (str); |
| 6507 | } |
| 6508 | else if (!i386_scale (cur_token.str)) |
| 6509 | return 0; |
| 6510 | intel_match_token (cur_token.code); |
| 6511 | } |
| 6512 | |
| 6513 | /* No scaling. If this is a memory operand, the register is either a |
| 6514 | base register (first occurrence) or an index register (second |
| 6515 | occurrence). */ |
| 6516 | else if (intel_parser.in_bracket && !(reg->reg_type & (SReg2 | SReg3))) |
| 6517 | { |
| 6518 | |
| 6519 | if (!i.base_reg) |
| 6520 | i.base_reg = reg; |
| 6521 | else if (!i.index_reg) |
| 6522 | i.index_reg = reg; |
| 6523 | else |
| 6524 | { |
| 6525 | as_bad (_("Too many register references in memory operand")); |
| 6526 | return 0; |
| 6527 | } |
| 6528 | |
| 6529 | i.types[this_operand] |= BaseIndex; |
| 6530 | } |
| 6531 | |
| 6532 | /* Offset modifier. Add the register to the displacement string to be |
| 6533 | parsed as an immediate expression after we're done. */ |
| 6534 | else if (intel_parser.in_offset) |
| 6535 | { |
| 6536 | as_warn (_("Using register names in OFFSET expressions is deprecated")); |
| 6537 | strcat (intel_parser.disp, reg->reg_name); |
| 6538 | } |
| 6539 | |
| 6540 | /* It's neither base nor index nor offset. */ |
| 6541 | else if (!intel_parser.is_mem) |
| 6542 | { |
| 6543 | i.types[this_operand] |= reg->reg_type & ~BaseIndex; |
| 6544 | i.op[this_operand].regs = reg; |
| 6545 | i.reg_operands++; |
| 6546 | } |
| 6547 | else |
| 6548 | { |
| 6549 | as_bad (_("Invalid use of register")); |
| 6550 | return 0; |
| 6551 | } |
| 6552 | |
| 6553 | /* Since registers are not part of the displacement string (except |
| 6554 | when we're parsing offset operands), we may need to remove any |
| 6555 | preceding '+' from the displacement string. */ |
| 6556 | if (*intel_parser.disp != '\0' |
| 6557 | && !intel_parser.in_offset) |
| 6558 | { |
| 6559 | char *s = intel_parser.disp; |
| 6560 | s += strlen (s) - 1; |
| 6561 | if (*s == '+') |
| 6562 | *s = '\0'; |
| 6563 | } |
| 6564 | |
| 6565 | return 1; |
| 6566 | } |
| 6567 | |
| 6568 | /* e11 BYTE |
| 6569 | | WORD |
| 6570 | | DWORD |
| 6571 | | FWORD |
| 6572 | | QWORD |
| 6573 | | TBYTE |
| 6574 | | OWORD |
| 6575 | | XMMWORD */ |
| 6576 | case T_BYTE: |
| 6577 | case T_WORD: |
| 6578 | case T_DWORD: |
| 6579 | case T_FWORD: |
| 6580 | case T_QWORD: |
| 6581 | case T_TBYTE: |
| 6582 | case T_XMMWORD: |
| 6583 | intel_match_token (cur_token.code); |
| 6584 | |
| 6585 | if (cur_token.code == T_PTR) |
| 6586 | return 1; |
| 6587 | |
| 6588 | /* It must have been an identifier. */ |
| 6589 | intel_putback_token (); |
| 6590 | cur_token.code = T_ID; |
| 6591 | /* FALLTHRU */ |
| 6592 | |
| 6593 | /* e11 id |
| 6594 | | constant */ |
| 6595 | case T_ID: |
| 6596 | if (!intel_parser.in_offset && intel_parser.is_mem <= 0) |
| 6597 | { |
| 6598 | symbolS *symbolP; |
| 6599 | |
| 6600 | /* The identifier represents a memory reference only if it's not |
| 6601 | preceded by an offset modifier and if it's not an equate. */ |
| 6602 | symbolP = symbol_find(cur_token.str); |
| 6603 | if (!symbolP || S_GET_SEGMENT(symbolP) != absolute_section) |
| 6604 | intel_parser.is_mem = 1; |
| 6605 | } |
| 6606 | /* FALLTHRU */ |
| 6607 | |
| 6608 | case T_CONST: |
| 6609 | case '-': |
| 6610 | case '+': |
| 6611 | { |
| 6612 | char *save_str, sign = 0; |
| 6613 | |
| 6614 | /* Allow constants that start with `+' or `-'. */ |
| 6615 | if (cur_token.code == '-' || cur_token.code == '+') |
| 6616 | { |
| 6617 | sign = cur_token.code; |
| 6618 | intel_match_token (cur_token.code); |
| 6619 | if (cur_token.code != T_CONST) |
| 6620 | { |
| 6621 | as_bad (_("Syntax error: Expecting a constant, got `%s'"), |
| 6622 | cur_token.str); |
| 6623 | return 0; |
| 6624 | } |
| 6625 | } |
| 6626 | |
| 6627 | save_str = (char *) xmalloc (strlen (cur_token.str) + 2); |
| 6628 | strcpy (save_str + !!sign, cur_token.str); |
| 6629 | if (sign) |
| 6630 | *save_str = sign; |
| 6631 | |
| 6632 | /* Get the next token to check for register scaling. */ |
| 6633 | intel_match_token (cur_token.code); |
| 6634 | |
| 6635 | /* Check if this constant is a scaling factor for an index register. */ |
| 6636 | if (cur_token.code == '*') |
| 6637 | { |
| 6638 | if (intel_match_token ('*') && cur_token.code == T_REG) |
| 6639 | { |
| 6640 | const reg_entry *reg = cur_token.reg; |
| 6641 | |
| 6642 | if (!intel_parser.in_bracket) |
| 6643 | { |
| 6644 | as_bad (_("Register scaling only allowed in memory operands")); |
| 6645 | return 0; |
| 6646 | } |
| 6647 | |
| 6648 | if (reg->reg_type & Reg16) /* Disallow things like [1*si]. */ |
| 6649 | reg = i386_regtab + REGNAM_AX + 4; /* sp is invalid as index */ |
| 6650 | else if (i.index_reg) |
| 6651 | reg = i386_regtab + REGNAM_EAX + 4; /* esp is invalid as index */ |
| 6652 | |
| 6653 | /* The constant is followed by `* reg', so it must be |
| 6654 | a valid scale. */ |
| 6655 | i.index_reg = reg; |
| 6656 | i.types[this_operand] |= BaseIndex; |
| 6657 | |
| 6658 | /* Set the scale after setting the register (otherwise, |
| 6659 | i386_scale will complain) */ |
| 6660 | if (!i386_scale (save_str)) |
| 6661 | return 0; |
| 6662 | intel_match_token (T_REG); |
| 6663 | |
| 6664 | /* Since registers are not part of the displacement |
| 6665 | string, we may need to remove any preceding '+' from |
| 6666 | the displacement string. */ |
| 6667 | if (*intel_parser.disp != '\0') |
| 6668 | { |
| 6669 | char *s = intel_parser.disp; |
| 6670 | s += strlen (s) - 1; |
| 6671 | if (*s == '+') |
| 6672 | *s = '\0'; |
| 6673 | } |
| 6674 | |
| 6675 | free (save_str); |
| 6676 | |
| 6677 | return 1; |
| 6678 | } |
| 6679 | |
| 6680 | /* The constant was not used for register scaling. Since we have |
| 6681 | already consumed the token following `*' we now need to put it |
| 6682 | back in the stream. */ |
| 6683 | intel_putback_token (); |
| 6684 | } |
| 6685 | |
| 6686 | /* Add the constant to the displacement string. */ |
| 6687 | strcat (intel_parser.disp, save_str); |
| 6688 | free (save_str); |
| 6689 | |
| 6690 | return 1; |
| 6691 | } |
| 6692 | } |
| 6693 | |
| 6694 | as_bad (_("Unrecognized token '%s'"), cur_token.str); |
| 6695 | return 0; |
| 6696 | } |
| 6697 | |
| 6698 | /* Match the given token against cur_token. If they match, read the next |
| 6699 | token from the operand string. */ |
| 6700 | static int |
| 6701 | intel_match_token (code) |
| 6702 | int code; |
| 6703 | { |
| 6704 | if (cur_token.code == code) |
| 6705 | { |
| 6706 | intel_get_token (); |
| 6707 | return 1; |
| 6708 | } |
| 6709 | else |
| 6710 | { |
| 6711 | as_bad (_("Unexpected token `%s'"), cur_token.str); |
| 6712 | return 0; |
| 6713 | } |
| 6714 | } |
| 6715 | |
| 6716 | /* Read a new token from intel_parser.op_string and store it in cur_token. */ |
| 6717 | static void |
| 6718 | intel_get_token () |
| 6719 | { |
| 6720 | char *end_op; |
| 6721 | const reg_entry *reg; |
| 6722 | struct intel_token new_token; |
| 6723 | |
| 6724 | new_token.code = T_NIL; |
| 6725 | new_token.reg = NULL; |
| 6726 | new_token.str = NULL; |
| 6727 | |
| 6728 | /* Free the memory allocated to the previous token and move |
| 6729 | cur_token to prev_token. */ |
| 6730 | if (prev_token.str) |
| 6731 | free (prev_token.str); |
| 6732 | |
| 6733 | prev_token = cur_token; |
| 6734 | |
| 6735 | /* Skip whitespace. */ |
| 6736 | while (is_space_char (*intel_parser.op_string)) |
| 6737 | intel_parser.op_string++; |
| 6738 | |
| 6739 | /* Return an empty token if we find nothing else on the line. */ |
| 6740 | if (*intel_parser.op_string == '\0') |
| 6741 | { |
| 6742 | cur_token = new_token; |
| 6743 | return; |
| 6744 | } |
| 6745 | |
| 6746 | /* The new token cannot be larger than the remainder of the operand |
| 6747 | string. */ |
| 6748 | new_token.str = (char *) xmalloc (strlen (intel_parser.op_string) + 1); |
| 6749 | new_token.str[0] = '\0'; |
| 6750 | |
| 6751 | if (strchr ("0123456789", *intel_parser.op_string)) |
| 6752 | { |
| 6753 | char *p = new_token.str; |
| 6754 | char *q = intel_parser.op_string; |
| 6755 | new_token.code = T_CONST; |
| 6756 | |
| 6757 | /* Allow any kind of identifier char to encompass floating point and |
| 6758 | hexadecimal numbers. */ |
| 6759 | while (is_identifier_char (*q)) |
| 6760 | *p++ = *q++; |
| 6761 | *p = '\0'; |
| 6762 | |
| 6763 | /* Recognize special symbol names [0-9][bf]. */ |
| 6764 | if (strlen (intel_parser.op_string) == 2 |
| 6765 | && (intel_parser.op_string[1] == 'b' |
| 6766 | || intel_parser.op_string[1] == 'f')) |
| 6767 | new_token.code = T_ID; |
| 6768 | } |
| 6769 | |
| 6770 | else if ((*intel_parser.op_string == REGISTER_PREFIX || allow_naked_reg) |
| 6771 | && ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL)) |
| 6772 | { |
| 6773 | new_token.code = T_REG; |
| 6774 | new_token.reg = reg; |
| 6775 | |
| 6776 | if (*intel_parser.op_string == REGISTER_PREFIX) |
| 6777 | { |
| 6778 | new_token.str[0] = REGISTER_PREFIX; |
| 6779 | new_token.str[1] = '\0'; |
| 6780 | } |
| 6781 | |
| 6782 | strcat (new_token.str, reg->reg_name); |
| 6783 | } |
| 6784 | |
| 6785 | else if (is_identifier_char (*intel_parser.op_string)) |
| 6786 | { |
| 6787 | char *p = new_token.str; |
| 6788 | char *q = intel_parser.op_string; |
| 6789 | |
| 6790 | /* A '.' or '$' followed by an identifier char is an identifier. |
| 6791 | Otherwise, it's operator '.' followed by an expression. */ |
| 6792 | if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1))) |
| 6793 | { |
| 6794 | new_token.code = '.'; |
| 6795 | new_token.str[0] = '.'; |
| 6796 | new_token.str[1] = '\0'; |
| 6797 | } |
| 6798 | else |
| 6799 | { |
| 6800 | while (is_identifier_char (*q) || *q == '@') |
| 6801 | *p++ = *q++; |
| 6802 | *p = '\0'; |
| 6803 | |
| 6804 | if (strcasecmp (new_token.str, "NOT") == 0) |
| 6805 | new_token.code = '~'; |
| 6806 | |
| 6807 | else if (strcasecmp (new_token.str, "MOD") == 0) |
| 6808 | new_token.code = '%'; |
| 6809 | |
| 6810 | else if (strcasecmp (new_token.str, "AND") == 0) |
| 6811 | new_token.code = '&'; |
| 6812 | |
| 6813 | else if (strcasecmp (new_token.str, "OR") == 0) |
| 6814 | new_token.code = '|'; |
| 6815 | |
| 6816 | else if (strcasecmp (new_token.str, "XOR") == 0) |
| 6817 | new_token.code = '^'; |
| 6818 | |
| 6819 | else if (strcasecmp (new_token.str, "SHL") == 0) |
| 6820 | new_token.code = T_SHL; |
| 6821 | |
| 6822 | else if (strcasecmp (new_token.str, "SHR") == 0) |
| 6823 | new_token.code = T_SHR; |
| 6824 | |
| 6825 | else if (strcasecmp (new_token.str, "BYTE") == 0) |
| 6826 | new_token.code = T_BYTE; |
| 6827 | |
| 6828 | else if (strcasecmp (new_token.str, "WORD") == 0) |
| 6829 | new_token.code = T_WORD; |
| 6830 | |
| 6831 | else if (strcasecmp (new_token.str, "DWORD") == 0) |
| 6832 | new_token.code = T_DWORD; |
| 6833 | |
| 6834 | else if (strcasecmp (new_token.str, "FWORD") == 0) |
| 6835 | new_token.code = T_FWORD; |
| 6836 | |
| 6837 | else if (strcasecmp (new_token.str, "QWORD") == 0) |
| 6838 | new_token.code = T_QWORD; |
| 6839 | |
| 6840 | else if (strcasecmp (new_token.str, "TBYTE") == 0 |
| 6841 | /* XXX remove (gcc still uses it) */ |
| 6842 | || strcasecmp (new_token.str, "XWORD") == 0) |
| 6843 | new_token.code = T_TBYTE; |
| 6844 | |
| 6845 | else if (strcasecmp (new_token.str, "XMMWORD") == 0 |
| 6846 | || strcasecmp (new_token.str, "OWORD") == 0) |
| 6847 | new_token.code = T_XMMWORD; |
| 6848 | |
| 6849 | else if (strcasecmp (new_token.str, "PTR") == 0) |
| 6850 | new_token.code = T_PTR; |
| 6851 | |
| 6852 | else if (strcasecmp (new_token.str, "SHORT") == 0) |
| 6853 | new_token.code = T_SHORT; |
| 6854 | |
| 6855 | else if (strcasecmp (new_token.str, "OFFSET") == 0) |
| 6856 | { |
| 6857 | new_token.code = T_OFFSET; |
| 6858 | |
| 6859 | /* ??? This is not mentioned in the MASM grammar but gcc |
| 6860 | makes use of it with -mintel-syntax. OFFSET may be |
| 6861 | followed by FLAT: */ |
| 6862 | if (strncasecmp (q, " FLAT:", 6) == 0) |
| 6863 | strcat (new_token.str, " FLAT:"); |
| 6864 | } |
| 6865 | |
| 6866 | /* ??? This is not mentioned in the MASM grammar. */ |
| 6867 | else if (strcasecmp (new_token.str, "FLAT") == 0) |
| 6868 | { |
| 6869 | new_token.code = T_OFFSET; |
| 6870 | if (*q == ':') |
| 6871 | strcat (new_token.str, ":"); |
| 6872 | else |
| 6873 | as_bad (_("`:' expected")); |
| 6874 | } |
| 6875 | |
| 6876 | else |
| 6877 | new_token.code = T_ID; |
| 6878 | } |
| 6879 | } |
| 6880 | |
| 6881 | else if (strchr ("+-/*%|&^:[]()~", *intel_parser.op_string)) |
| 6882 | { |
| 6883 | new_token.code = *intel_parser.op_string; |
| 6884 | new_token.str[0] = *intel_parser.op_string; |
| 6885 | new_token.str[1] = '\0'; |
| 6886 | } |
| 6887 | |
| 6888 | else if (strchr ("<>", *intel_parser.op_string) |
| 6889 | && *intel_parser.op_string == *(intel_parser.op_string + 1)) |
| 6890 | { |
| 6891 | new_token.code = *intel_parser.op_string == '<' ? T_SHL : T_SHR; |
| 6892 | new_token.str[0] = *intel_parser.op_string; |
| 6893 | new_token.str[1] = *intel_parser.op_string; |
| 6894 | new_token.str[2] = '\0'; |
| 6895 | } |
| 6896 | |
| 6897 | else |
| 6898 | as_bad (_("Unrecognized token `%s'"), intel_parser.op_string); |
| 6899 | |
| 6900 | intel_parser.op_string += strlen (new_token.str); |
| 6901 | cur_token = new_token; |
| 6902 | } |
| 6903 | |
| 6904 | /* Put cur_token back into the token stream and make cur_token point to |
| 6905 | prev_token. */ |
| 6906 | static void |
| 6907 | intel_putback_token () |
| 6908 | { |
| 6909 | if (cur_token.code != T_NIL) |
| 6910 | { |
| 6911 | intel_parser.op_string -= strlen (cur_token.str); |
| 6912 | free (cur_token.str); |
| 6913 | } |
| 6914 | cur_token = prev_token; |
| 6915 | |
| 6916 | /* Forget prev_token. */ |
| 6917 | prev_token.code = T_NIL; |
| 6918 | prev_token.reg = NULL; |
| 6919 | prev_token.str = NULL; |
| 6920 | } |
| 6921 | |
| 6922 | int |
| 6923 | tc_x86_regname_to_dw2regnum (const char *regname) |
| 6924 | { |
| 6925 | unsigned int regnum; |
| 6926 | unsigned int regnames_count; |
| 6927 | static const char *const regnames_32[] = |
| 6928 | { |
| 6929 | "eax", "ecx", "edx", "ebx", |
| 6930 | "esp", "ebp", "esi", "edi", |
| 6931 | "eip", "eflags", NULL, |
| 6932 | "st0", "st1", "st2", "st3", |
| 6933 | "st4", "st5", "st6", "st7", |
| 6934 | NULL, NULL, |
| 6935 | "xmm0", "xmm1", "xmm2", "xmm3", |
| 6936 | "xmm4", "xmm5", "xmm6", "xmm7", |
| 6937 | "mm0", "mm1", "mm2", "mm3", |
| 6938 | "mm4", "mm5", "mm6", "mm7", |
| 6939 | "fcw", "fsw", "mxcsr", |
| 6940 | "es", "cs", "ss", "ds", "fs", "gs", NULL, NULL, |
| 6941 | "tr", "ldtr" |
| 6942 | }; |
| 6943 | static const char *const regnames_64[] = |
| 6944 | { |
| 6945 | "rax", "rdx", "rcx", "rbx", |
| 6946 | "rsi", "rdi", "rbp", "rsp", |
| 6947 | "r8", "r9", "r10", "r11", |
| 6948 | "r12", "r13", "r14", "r15", |
| 6949 | "rip", |
| 6950 | "xmm0", "xmm1", "xmm2", "xmm3", |
| 6951 | "xmm4", "xmm5", "xmm6", "xmm7", |
| 6952 | "xmm8", "xmm9", "xmm10", "xmm11", |
| 6953 | "xmm12", "xmm13", "xmm14", "xmm15", |
| 6954 | "st0", "st1", "st2", "st3", |
| 6955 | "st4", "st5", "st6", "st7", |
| 6956 | "mm0", "mm1", "mm2", "mm3", |
| 6957 | "mm4", "mm5", "mm6", "mm7", |
| 6958 | "rflags", |
| 6959 | "es", "cs", "ss", "ds", "fs", "gs", NULL, NULL, |
| 6960 | "fs.base", "gs.base", NULL, NULL, |
| 6961 | "tr", "ldtr", |
| 6962 | "mxcsr", "fcw", "fsw" |
| 6963 | }; |
| 6964 | const char *const *regnames; |
| 6965 | |
| 6966 | if (flag_code == CODE_64BIT) |
| 6967 | { |
| 6968 | regnames = regnames_64; |
| 6969 | regnames_count = ARRAY_SIZE (regnames_64); |
| 6970 | } |
| 6971 | else |
| 6972 | { |
| 6973 | regnames = regnames_32; |
| 6974 | regnames_count = ARRAY_SIZE (regnames_32); |
| 6975 | } |
| 6976 | |
| 6977 | for (regnum = 0; regnum < regnames_count; regnum++) |
| 6978 | if (regnames[regnum] != NULL |
| 6979 | && strcmp (regname, regnames[regnum]) == 0) |
| 6980 | return regnum; |
| 6981 | |
| 6982 | return -1; |
| 6983 | } |
| 6984 | |
| 6985 | void |
| 6986 | tc_x86_frame_initial_instructions (void) |
| 6987 | { |
| 6988 | static unsigned int sp_regno; |
| 6989 | |
| 6990 | if (!sp_regno) |
| 6991 | sp_regno = tc_x86_regname_to_dw2regnum (flag_code == CODE_64BIT |
| 6992 | ? "rsp" : "esp"); |
| 6993 | |
| 6994 | cfi_add_CFA_def_cfa (sp_regno, -x86_cie_data_alignment); |
| 6995 | cfi_add_CFA_offset (x86_dwarf2_return_column, x86_cie_data_alignment); |
| 6996 | } |
| 6997 | |
| 6998 | int |
| 6999 | i386_elf_section_type (const char *str, size_t len) |
| 7000 | { |
| 7001 | if (flag_code == CODE_64BIT |
| 7002 | && len == sizeof ("unwind") - 1 |
| 7003 | && strncmp (str, "unwind", 6) == 0) |
| 7004 | return SHT_X86_64_UNWIND; |
| 7005 | |
| 7006 | return -1; |
| 7007 | } |
| 7008 | |
| 7009 | #ifdef TE_PE |
| 7010 | void |
| 7011 | tc_pe_dwarf2_emit_offset (symbolS *symbol, unsigned int size) |
| 7012 | { |
| 7013 | expressionS expr; |
| 7014 | |
| 7015 | expr.X_op = O_secrel; |
| 7016 | expr.X_add_symbol = symbol; |
| 7017 | expr.X_add_number = 0; |
| 7018 | emit_expr (&expr, size); |
| 7019 | } |
| 7020 | #endif |
| 7021 | |
| 7022 | #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) |
| 7023 | /* For ELF on x86-64, add support for SHF_X86_64_LARGE. */ |
| 7024 | |
| 7025 | int |
| 7026 | x86_64_section_letter (int letter, char **ptr_msg) |
| 7027 | { |
| 7028 | if (flag_code == CODE_64BIT) |
| 7029 | { |
| 7030 | if (letter == 'l') |
| 7031 | return SHF_X86_64_LARGE; |
| 7032 | |
| 7033 | *ptr_msg = _("Bad .section directive: want a,l,w,x,M,S,G,T in string"); |
| 7034 | } |
| 7035 | else |
| 7036 | *ptr_msg = _("Bad .section directive: want a,w,x,M,S,G,T in string"); |
| 7037 | return -1; |
| 7038 | } |
| 7039 | |
| 7040 | int |
| 7041 | x86_64_section_word (char *str, size_t len) |
| 7042 | { |
| 7043 | if (len == 5 && flag_code == CODE_64BIT && strncmp (str, "large", 5) == 0) |
| 7044 | return SHF_X86_64_LARGE; |
| 7045 | |
| 7046 | return -1; |
| 7047 | } |
| 7048 | |
| 7049 | static void |
| 7050 | handle_large_common (int small ATTRIBUTE_UNUSED) |
| 7051 | { |
| 7052 | if (flag_code != CODE_64BIT) |
| 7053 | { |
| 7054 | s_comm_internal (0, elf_common_parse); |
| 7055 | as_warn (_(".largecomm supported only in 64bit mode, producing .comm")); |
| 7056 | } |
| 7057 | else |
| 7058 | { |
| 7059 | static segT lbss_section; |
| 7060 | asection *saved_com_section_ptr = elf_com_section_ptr; |
| 7061 | asection *saved_bss_section = bss_section; |
| 7062 | |
| 7063 | if (lbss_section == NULL) |
| 7064 | { |
| 7065 | flagword applicable; |
| 7066 | segT seg = now_seg; |
| 7067 | subsegT subseg = now_subseg; |
| 7068 | |
| 7069 | /* The .lbss section is for local .largecomm symbols. */ |
| 7070 | lbss_section = subseg_new (".lbss", 0); |
| 7071 | applicable = bfd_applicable_section_flags (stdoutput); |
| 7072 | bfd_set_section_flags (stdoutput, lbss_section, |
| 7073 | applicable & SEC_ALLOC); |
| 7074 | seg_info (lbss_section)->bss = 1; |
| 7075 | |
| 7076 | subseg_set (seg, subseg); |
| 7077 | } |
| 7078 | |
| 7079 | elf_com_section_ptr = &_bfd_elf_large_com_section; |
| 7080 | bss_section = lbss_section; |
| 7081 | |
| 7082 | s_comm_internal (0, elf_common_parse); |
| 7083 | |
| 7084 | elf_com_section_ptr = saved_com_section_ptr; |
| 7085 | bss_section = saved_bss_section; |
| 7086 | } |
| 7087 | } |
| 7088 | #endif /* OBJ_ELF || OBJ_MAYBE_ELF */ |