| 1 | #ifndef BSWAP_H |
| 2 | #define BSWAP_H |
| 3 | |
| 4 | #include "qemu/target-info.h" |
| 5 | #include "exec/memop.h" |
| 6 | |
| 7 | #undef bswap16 |
| 8 | #define bswap16(_x) __builtin_bswap16(_x) |
| 9 | #undef bswap32 |
| 10 | #define bswap32(_x) __builtin_bswap32(_x) |
| 11 | #undef bswap64 |
| 12 | #define bswap64(_x) __builtin_bswap64(_x) |
| 13 | |
| 14 | static inline uint32_t bswap24(uint32_t x) |
| 15 | { |
| 16 | return (((x & 0x000000ffU) << 16) | |
| 17 | ((x & 0x0000ff00U) << 0) | |
| 18 | ((x & 0x00ff0000U) >> 16)); |
| 19 | } |
| 20 | |
| 21 | static inline void bswap16s(uint16_t *s) |
| 22 | { |
| 23 | *s = __builtin_bswap16(*s); |
| 24 | } |
| 25 | |
| 26 | static inline void bswap24s(uint32_t *s) |
| 27 | { |
| 28 | *s = bswap24(*s & 0x00ffffffU); |
| 29 | } |
| 30 | |
| 31 | static inline void bswap32s(uint32_t *s) |
| 32 | { |
| 33 | *s = __builtin_bswap32(*s); |
| 34 | } |
| 35 | |
| 36 | static inline void bswap64s(uint64_t *s) |
| 37 | { |
| 38 | *s = __builtin_bswap64(*s); |
| 39 | } |
| 40 | |
| 41 | #if HOST_BIG_ENDIAN |
| 42 | #define be_bswap(v, size) (v) |
| 43 | #define le_bswap(v, size) glue(__builtin_bswap, size)(v) |
| 44 | #define be_bswap24(v) (v) |
| 45 | #define le_bswap24(v) bswap24(v) |
| 46 | #define be_bswaps(v, size) |
| 47 | #define le_bswaps(p, size) \ |
| 48 | do { *p = glue(__builtin_bswap, size)(*p); } while (0) |
| 49 | #else |
| 50 | #define le_bswap(v, size) (v) |
| 51 | #define be_bswap24(v) bswap24(v) |
| 52 | #define le_bswap24(v) (v) |
| 53 | #define be_bswap(v, size) glue(__builtin_bswap, size)(v) |
| 54 | #define le_bswaps(v, size) |
| 55 | #define be_bswaps(p, size) \ |
| 56 | do { *p = glue(__builtin_bswap, size)(*p); } while (0) |
| 57 | #endif |
| 58 | |
| 59 | /** |
| 60 | * Endianness conversion functions between host cpu and specified endianness. |
| 61 | * (We list the complete set of prototypes produced by the macros below |
| 62 | * to assist people who search the headers to find their definitions.) |
| 63 | * |
| 64 | * uint16_t le16_to_cpu(uint16_t v); |
| 65 | * uint32_t le32_to_cpu(uint32_t v); |
| 66 | * uint64_t le64_to_cpu(uint64_t v); |
| 67 | * uint16_t be16_to_cpu(uint16_t v); |
| 68 | * uint32_t be32_to_cpu(uint32_t v); |
| 69 | * uint64_t be64_to_cpu(uint64_t v); |
| 70 | * |
| 71 | * Convert the value @v from the specified format to the native |
| 72 | * endianness of the host CPU by byteswapping if necessary, and |
| 73 | * return the converted value. |
| 74 | * |
| 75 | * uint16_t cpu_to_le16(uint16_t v); |
| 76 | * uint32_t cpu_to_le32(uint32_t v); |
| 77 | * uint64_t cpu_to_le64(uint64_t v); |
| 78 | * uint16_t cpu_to_be16(uint16_t v); |
| 79 | * uint32_t cpu_to_be32(uint32_t v); |
| 80 | * uint64_t cpu_to_be64(uint64_t v); |
| 81 | * |
| 82 | * Convert the value @v from the native endianness of the host CPU to |
| 83 | * the specified format by byteswapping if necessary, and return |
| 84 | * the converted value. |
| 85 | * |
| 86 | * void le16_to_cpus(uint16_t *v); |
| 87 | * void le32_to_cpus(uint32_t *v); |
| 88 | * void le64_to_cpus(uint64_t *v); |
| 89 | * void be16_to_cpus(uint16_t *v); |
| 90 | * void be32_to_cpus(uint32_t *v); |
| 91 | * void be64_to_cpus(uint64_t *v); |
| 92 | * |
| 93 | * Do an in-place conversion of the value pointed to by @v from the |
| 94 | * specified format to the native endianness of the host CPU. |
| 95 | * |
| 96 | * void cpu_to_le16s(uint16_t *v); |
| 97 | * void cpu_to_le32s(uint32_t *v); |
| 98 | * void cpu_to_le64s(uint64_t *v); |
| 99 | * void cpu_to_be16s(uint16_t *v); |
| 100 | * void cpu_to_be32s(uint32_t *v); |
| 101 | * void cpu_to_be64s(uint64_t *v); |
| 102 | * |
| 103 | * Do an in-place conversion of the value pointed to by @v from the |
| 104 | * native endianness of the host CPU to the specified format. |
| 105 | * |
| 106 | * Both X_to_cpu() and cpu_to_X() perform the same operation; you |
| 107 | * should use whichever one is better documenting of the function your |
| 108 | * code is performing. |
| 109 | * |
| 110 | * Do not use these functions for conversion of values which are in guest |
| 111 | * memory, since the data may not be sufficiently aligned for the host CPU's |
| 112 | * load and store instructions. Instead you should use the ld*_p() and |
| 113 | * st*_p() functions, which perform loads and stores of data of any |
| 114 | * required size and endianness and handle possible misalignment. |
| 115 | */ |
| 116 | |
| 117 | #define CPU_CONVERT(endian, size, type)\ |
| 118 | static inline type endian ## size ## _to_cpu(type v)\ |
| 119 | {\ |
| 120 | return glue(endian, _bswap)(v, size);\ |
| 121 | }\ |
| 122 | \ |
| 123 | static inline type cpu_to_ ## endian ## size(type v)\ |
| 124 | {\ |
| 125 | return glue(endian, _bswap)(v, size);\ |
| 126 | }\ |
| 127 | \ |
| 128 | static inline void endian ## size ## _to_cpus(type *p)\ |
| 129 | {\ |
| 130 | glue(endian, _bswaps)(p, size);\ |
| 131 | }\ |
| 132 | \ |
| 133 | static inline void cpu_to_ ## endian ## size ## s(type *p)\ |
| 134 | {\ |
| 135 | glue(endian, _bswaps)(p, size);\ |
| 136 | } |
| 137 | |
| 138 | CPU_CONVERT(be, 16, uint16_t) |
| 139 | CPU_CONVERT(be, 32, uint32_t) |
| 140 | CPU_CONVERT(be, 64, uint64_t) |
| 141 | |
| 142 | CPU_CONVERT(le, 16, uint16_t) |
| 143 | CPU_CONVERT(le, 32, uint32_t) |
| 144 | CPU_CONVERT(le, 64, uint64_t) |
| 145 | |
| 146 | #undef CPU_CONVERT |
| 147 | |
| 148 | /* |
| 149 | * Same as cpu_to_le{16,32,64}, except that gcc will figure the result is |
| 150 | * a compile-time constant if you pass in a constant. So this can be |
| 151 | * used to initialize static variables. |
| 152 | */ |
| 153 | #if HOST_BIG_ENDIAN |
| 154 | # define const_le64(_x) \ |
| 155 | ((((_x) & 0x00000000000000ffULL) << 56) | \ |
| 156 | (((_x) & 0x000000000000ff00ULL) << 40) | \ |
| 157 | (((_x) & 0x0000000000ff0000ULL) << 24) | \ |
| 158 | (((_x) & 0x00000000ff000000ULL) << 8) | \ |
| 159 | (((_x) & 0x000000ff00000000ULL) >> 8) | \ |
| 160 | (((_x) & 0x0000ff0000000000ULL) >> 24) | \ |
| 161 | (((_x) & 0x00ff000000000000ULL) >> 40) | \ |
| 162 | (((_x) & 0xff00000000000000ULL) >> 56)) |
| 163 | # define const_le32(_x) \ |
| 164 | ((((_x) & 0x000000ffU) << 24) | \ |
| 165 | (((_x) & 0x0000ff00U) << 8) | \ |
| 166 | (((_x) & 0x00ff0000U) >> 8) | \ |
| 167 | (((_x) & 0xff000000U) >> 24)) |
| 168 | # define const_le16(_x) \ |
| 169 | ((((_x) & 0x00ff) << 8) | \ |
| 170 | (((_x) & 0xff00) >> 8)) |
| 171 | #else |
| 172 | # define const_le64(_x) (_x) |
| 173 | # define const_le32(_x) (_x) |
| 174 | # define const_le16(_x) (_x) |
| 175 | #endif |
| 176 | |
| 177 | /* unaligned/endian-independent pointer access */ |
| 178 | |
| 179 | /* |
| 180 | * the generic syntax is: |
| 181 | * |
| 182 | * load: ld{type}{sign}{size}_{endian}_p(ptr) |
| 183 | * |
| 184 | * store: st{type}{size}_{endian}_p(ptr, val) |
| 185 | * |
| 186 | * Note there are small differences with the softmmu access API! |
| 187 | * |
| 188 | * type is: |
| 189 | * (empty): integer access |
| 190 | * f : float access |
| 191 | * |
| 192 | * sign is: |
| 193 | * (empty): for 32 or 64 bit sizes (including floats and doubles) |
| 194 | * u : unsigned |
| 195 | * s : signed |
| 196 | * |
| 197 | * size is: |
| 198 | * b: 8 bits |
| 199 | * w: 16 bits |
| 200 | * 24: 24 bits |
| 201 | * l: 32 bits |
| 202 | * q: 64 bits |
| 203 | * |
| 204 | * endian is: |
| 205 | * he : host endian |
| 206 | * be : big endian |
| 207 | * le : little endian |
| 208 | * te : target endian |
| 209 | * (except for byte accesses, which have no endian infix). |
| 210 | * |
| 211 | * In all cases these functions take a host pointer. |
| 212 | * For accessors that take a guest address rather than a |
| 213 | * host address, see the cpu_{ld,st}_* accessors defined in |
| 214 | * cpu_ldst.h. |
| 215 | * |
| 216 | * For cases where the size to be used is not fixed at compile time, |
| 217 | * there are |
| 218 | * stn_{endian}_p(ptr, sz, val) |
| 219 | * which stores @val to @ptr as an @endian-order number @sz bytes in size |
| 220 | * and |
| 221 | * ldn_{endian}_p(ptr, sz) |
| 222 | * which loads @sz bytes from @ptr as an unsigned @endian-order number |
| 223 | * and returns it in a uint64_t. |
| 224 | */ |
| 225 | |
| 226 | static inline int ldub_p(const void *ptr) |
| 227 | { |
| 228 | return *(uint8_t *)ptr; |
| 229 | } |
| 230 | |
| 231 | static inline int ldsb_p(const void *ptr) |
| 232 | { |
| 233 | return *(int8_t *)ptr; |
| 234 | } |
| 235 | |
| 236 | static inline void stb_p(void *ptr, uint8_t v) |
| 237 | { |
| 238 | *(uint8_t *)ptr = v; |
| 239 | } |
| 240 | |
| 241 | /* |
| 242 | * Any compiler worth its salt will turn these memcpy into native unaligned |
| 243 | * operations. Thus we don't need to play games with packed attributes, or |
| 244 | * inline byte-by-byte stores. |
| 245 | * Some compilation environments (eg some fortify-source implementations) |
| 246 | * may intercept memcpy() in a way that defeats the compiler optimization, |
| 247 | * though, so we use __builtin_memcpy() to give ourselves the best chance |
| 248 | * of good performance. |
| 249 | */ |
| 250 | |
| 251 | static inline int lduw_he_p(const void *ptr) |
| 252 | { |
| 253 | uint16_t r; |
| 254 | __builtin_memcpy(&r, ptr, sizeof(r)); |
| 255 | return r; |
| 256 | } |
| 257 | |
| 258 | static inline int ldsw_he_p(const void *ptr) |
| 259 | { |
| 260 | int16_t r; |
| 261 | __builtin_memcpy(&r, ptr, sizeof(r)); |
| 262 | return r; |
| 263 | } |
| 264 | |
| 265 | static inline void stw_he_p(void *ptr, uint16_t v) |
| 266 | { |
| 267 | __builtin_memcpy(ptr, &v, sizeof(v)); |
| 268 | } |
| 269 | |
| 270 | static inline void st24_he_p(void *ptr, uint32_t v) |
| 271 | { |
| 272 | __builtin_memcpy(ptr, &v, 3); |
| 273 | } |
| 274 | |
| 275 | static inline int ldl_he_p(const void *ptr) |
| 276 | { |
| 277 | int32_t r; |
| 278 | __builtin_memcpy(&r, ptr, sizeof(r)); |
| 279 | return r; |
| 280 | } |
| 281 | |
| 282 | static inline void stl_he_p(void *ptr, uint32_t v) |
| 283 | { |
| 284 | __builtin_memcpy(ptr, &v, sizeof(v)); |
| 285 | } |
| 286 | |
| 287 | static inline uint64_t ldq_he_p(const void *ptr) |
| 288 | { |
| 289 | uint64_t r; |
| 290 | __builtin_memcpy(&r, ptr, sizeof(r)); |
| 291 | return r; |
| 292 | } |
| 293 | |
| 294 | static inline void stq_he_p(void *ptr, uint64_t v) |
| 295 | { |
| 296 | __builtin_memcpy(ptr, &v, sizeof(v)); |
| 297 | } |
| 298 | |
| 299 | static inline int lduw_le_p(const void *ptr) |
| 300 | { |
| 301 | return (uint16_t)le_bswap(lduw_he_p(ptr), 16); |
| 302 | } |
| 303 | |
| 304 | static inline int ldsw_le_p(const void *ptr) |
| 305 | { |
| 306 | return (int16_t)le_bswap(lduw_he_p(ptr), 16); |
| 307 | } |
| 308 | |
| 309 | static inline int ldl_le_p(const void *ptr) |
| 310 | { |
| 311 | return le_bswap(ldl_he_p(ptr), 32); |
| 312 | } |
| 313 | |
| 314 | static inline uint64_t ldq_le_p(const void *ptr) |
| 315 | { |
| 316 | return le_bswap(ldq_he_p(ptr), 64); |
| 317 | } |
| 318 | |
| 319 | static inline void stw_le_p(void *ptr, uint16_t v) |
| 320 | { |
| 321 | stw_he_p(ptr, le_bswap(v, 16)); |
| 322 | } |
| 323 | |
| 324 | static inline void st24_le_p(void *ptr, uint32_t v) |
| 325 | { |
| 326 | st24_he_p(ptr, le_bswap24(v)); |
| 327 | } |
| 328 | |
| 329 | static inline void stl_le_p(void *ptr, uint32_t v) |
| 330 | { |
| 331 | stl_he_p(ptr, le_bswap(v, 32)); |
| 332 | } |
| 333 | |
| 334 | static inline void stq_le_p(void *ptr, uint64_t v) |
| 335 | { |
| 336 | stq_he_p(ptr, le_bswap(v, 64)); |
| 337 | } |
| 338 | |
| 339 | static inline int lduw_be_p(const void *ptr) |
| 340 | { |
| 341 | return (uint16_t)be_bswap(lduw_he_p(ptr), 16); |
| 342 | } |
| 343 | |
| 344 | static inline int ldsw_be_p(const void *ptr) |
| 345 | { |
| 346 | return (int16_t)be_bswap(lduw_he_p(ptr), 16); |
| 347 | } |
| 348 | |
| 349 | static inline int ldl_be_p(const void *ptr) |
| 350 | { |
| 351 | return be_bswap(ldl_he_p(ptr), 32); |
| 352 | } |
| 353 | |
| 354 | static inline uint64_t ldq_be_p(const void *ptr) |
| 355 | { |
| 356 | return be_bswap(ldq_he_p(ptr), 64); |
| 357 | } |
| 358 | |
| 359 | static inline void stw_be_p(void *ptr, uint16_t v) |
| 360 | { |
| 361 | stw_he_p(ptr, be_bswap(v, 16)); |
| 362 | } |
| 363 | |
| 364 | static inline void st24_be_p(void *ptr, uint32_t v) |
| 365 | { |
| 366 | st24_he_p(ptr, be_bswap24(v)); |
| 367 | } |
| 368 | |
| 369 | static inline void stl_be_p(void *ptr, uint32_t v) |
| 370 | { |
| 371 | stl_he_p(ptr, be_bswap(v, 32)); |
| 372 | } |
| 373 | |
| 374 | static inline void stq_be_p(void *ptr, uint64_t v) |
| 375 | { |
| 376 | stq_he_p(ptr, be_bswap(v, 64)); |
| 377 | } |
| 378 | |
| 379 | |
| 380 | /** |
| 381 | * ldm_p: Load value from host memory (byteswapping if necessary) |
| 382 | * |
| 383 | * @ptr: the host pointer to be accessed |
| 384 | * @mop: #MemOp mask containing access size and optional byteswapping |
| 385 | * |
| 386 | * Convert the value stored at @ptr in host memory and byteswap if necessary. |
| 387 | * |
| 388 | * Returns: the converted value. |
| 389 | */ |
| 390 | static inline uint64_t ldm_p(const void *ptr, MemOp mop) |
| 391 | { |
| 392 | switch (mop & (MO_SIZE | MO_BSWAP)) { |
| 393 | case MO_8: |
| 394 | return ldub_p(ptr); |
| 395 | case MO_16 | MO_LE: |
| 396 | return lduw_le_p(ptr); |
| 397 | case MO_16 | MO_BE: |
| 398 | return lduw_be_p(ptr); |
| 399 | case MO_32 | MO_LE: |
| 400 | return ldl_le_p(ptr); |
| 401 | case MO_32 | MO_BE: |
| 402 | return ldl_be_p(ptr); |
| 403 | case MO_64 | MO_LE: |
| 404 | return ldq_le_p(ptr); |
| 405 | case MO_64 | MO_BE: |
| 406 | return ldq_be_p(ptr); |
| 407 | default: |
| 408 | g_assert_not_reached(); |
| 409 | } |
| 410 | } |
| 411 | |
| 412 | /** |
| 413 | * stm_p: Store value to host memory (byteswapping if necessary) |
| 414 | * |
| 415 | * @ptr: the host pointer to be accessed |
| 416 | * @mop: #MemOp mask containing access size and optional byteswapping |
| 417 | * @val: the value to store |
| 418 | * |
| 419 | * Convert the value (byteswap if necessary) and store at @ptr in host memory. |
| 420 | */ |
| 421 | static inline void stm_p(void *ptr, MemOp mop, uint64_t val) |
| 422 | { |
| 423 | switch (mop & (MO_SIZE | MO_BSWAP)) { |
| 424 | case MO_8: |
| 425 | stb_p(ptr, val); |
| 426 | break; |
| 427 | case MO_16 | MO_LE: |
| 428 | stw_le_p(ptr, val); |
| 429 | break; |
| 430 | case MO_16 | MO_BE: |
| 431 | stw_be_p(ptr, val); |
| 432 | break; |
| 433 | case MO_32 | MO_LE: |
| 434 | stl_le_p(ptr, val); |
| 435 | break; |
| 436 | case MO_32 | MO_BE: |
| 437 | stl_be_p(ptr, val); |
| 438 | break; |
| 439 | case MO_64 | MO_LE: |
| 440 | stq_le_p(ptr, val); |
| 441 | break; |
| 442 | case MO_64 | MO_BE: |
| 443 | stq_be_p(ptr, val); |
| 444 | break; |
| 445 | default: |
| 446 | g_assert_not_reached(); |
| 447 | } |
| 448 | } |
| 449 | |
| 450 | /* Store v to p as a sz byte value in host order */ |
| 451 | #define DO_STN_LDN_P(END) \ |
| 452 | static inline void stn_## END ## _p(void *ptr, int sz, uint64_t v) \ |
| 453 | { \ |
| 454 | switch (sz) { \ |
| 455 | case 1: \ |
| 456 | stb_p(ptr, v); \ |
| 457 | break; \ |
| 458 | case 2: \ |
| 459 | stw_ ## END ## _p(ptr, v); \ |
| 460 | break; \ |
| 461 | case 4: \ |
| 462 | stl_ ## END ## _p(ptr, v); \ |
| 463 | break; \ |
| 464 | case 8: \ |
| 465 | stq_ ## END ## _p(ptr, v); \ |
| 466 | break; \ |
| 467 | default: \ |
| 468 | g_assert_not_reached(); \ |
| 469 | } \ |
| 470 | } \ |
| 471 | static inline uint64_t ldn_## END ## _p(const void *ptr, int sz) \ |
| 472 | { \ |
| 473 | switch (sz) { \ |
| 474 | case 1: \ |
| 475 | return ldub_p(ptr); \ |
| 476 | case 2: \ |
| 477 | return lduw_ ## END ## _p(ptr); \ |
| 478 | case 4: \ |
| 479 | return (uint32_t)ldl_ ## END ## _p(ptr); \ |
| 480 | case 8: \ |
| 481 | return ldq_ ## END ## _p(ptr); \ |
| 482 | default: \ |
| 483 | g_assert_not_reached(); \ |
| 484 | } \ |
| 485 | } |
| 486 | |
| 487 | DO_STN_LDN_P(he) |
| 488 | DO_STN_LDN_P(le) |
| 489 | DO_STN_LDN_P(be) |
| 490 | |
| 491 | #undef DO_STN_LDN_P |
| 492 | |
| 493 | #undef le_bswap |
| 494 | #undef be_bswap |
| 495 | #undef le_bswaps |
| 496 | #undef be_bswaps |
| 497 | |
| 498 | |
| 499 | /* Return ld{word}_{le,be}_p following target endianness. */ |
| 500 | #define LOAD_IMPL(word, args...) \ |
| 501 | do { \ |
| 502 | if (target_big_endian()) { \ |
| 503 | return glue(glue(ld, word), _be_p)(args); \ |
| 504 | } else { \ |
| 505 | return glue(glue(ld, word), _le_p)(args); \ |
| 506 | } \ |
| 507 | } while (0) |
| 508 | |
| 509 | static inline int lduw_p(const void *ptr) |
| 510 | { |
| 511 | LOAD_IMPL(uw, ptr); |
| 512 | } |
| 513 | |
| 514 | static inline int ldsw_p(const void *ptr) |
| 515 | { |
| 516 | LOAD_IMPL(sw, ptr); |
| 517 | } |
| 518 | |
| 519 | static inline int ldl_p(const void *ptr) |
| 520 | { |
| 521 | LOAD_IMPL(l, ptr); |
| 522 | } |
| 523 | |
| 524 | static inline uint64_t ldq_p(const void *ptr) |
| 525 | { |
| 526 | LOAD_IMPL(q, ptr); |
| 527 | } |
| 528 | |
| 529 | static inline uint64_t ldn_p(const void *ptr, int sz) |
| 530 | { |
| 531 | LOAD_IMPL(n, ptr, sz); |
| 532 | } |
| 533 | |
| 534 | #undef LOAD_IMPL |
| 535 | |
| 536 | /* Call st{word}_{le,be}_p following target endianness. */ |
| 537 | #define STORE_IMPL(word, args...) \ |
| 538 | do { \ |
| 539 | if (target_big_endian()) { \ |
| 540 | glue(glue(st, word), _be_p)(args); \ |
| 541 | } else { \ |
| 542 | glue(glue(st, word), _le_p)(args); \ |
| 543 | } \ |
| 544 | } while (0) |
| 545 | |
| 546 | |
| 547 | static inline void stw_p(void *ptr, uint16_t v) |
| 548 | { |
| 549 | STORE_IMPL(w, ptr, v); |
| 550 | } |
| 551 | |
| 552 | static inline void stl_p(void *ptr, uint32_t v) |
| 553 | { |
| 554 | STORE_IMPL(l, ptr, v); |
| 555 | } |
| 556 | |
| 557 | static inline void stq_p(void *ptr, uint64_t v) |
| 558 | { |
| 559 | STORE_IMPL(q, ptr, v); |
| 560 | } |
| 561 | |
| 562 | static inline void stn_p(void *ptr, int sz, uint64_t v) |
| 563 | { |
| 564 | STORE_IMPL(n, ptr, sz, v); |
| 565 | } |
| 566 | |
| 567 | #undef STORE_IMPL |
| 568 | |
| 569 | #endif /* BSWAP_H */ |