Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame^] | 1 | #ifndef _PARISC_BITOPS_H |
| 2 | #define _PARISC_BITOPS_H |
| 3 | |
| 4 | #include <linux/compiler.h> |
| 5 | #include <asm/system.h> |
| 6 | #include <asm/byteorder.h> |
| 7 | #include <asm/atomic.h> |
| 8 | |
| 9 | /* |
| 10 | * HP-PARISC specific bit operations |
| 11 | * for a detailed description of the functions please refer |
| 12 | * to include/asm-i386/bitops.h or kerneldoc |
| 13 | */ |
| 14 | |
| 15 | #ifdef __LP64__ |
| 16 | # define SHIFT_PER_LONG 6 |
| 17 | #ifndef BITS_PER_LONG |
| 18 | # define BITS_PER_LONG 64 |
| 19 | #endif |
| 20 | #else |
| 21 | # define SHIFT_PER_LONG 5 |
| 22 | #ifndef BITS_PER_LONG |
| 23 | # define BITS_PER_LONG 32 |
| 24 | #endif |
| 25 | #endif |
| 26 | |
| 27 | #define CHOP_SHIFTCOUNT(x) ((x) & (BITS_PER_LONG - 1)) |
| 28 | |
| 29 | |
| 30 | #define smp_mb__before_clear_bit() smp_mb() |
| 31 | #define smp_mb__after_clear_bit() smp_mb() |
| 32 | |
| 33 | static __inline__ void set_bit(int nr, volatile unsigned long * address) |
| 34 | { |
| 35 | unsigned long mask; |
| 36 | unsigned long *addr = (unsigned long *) address; |
| 37 | unsigned long flags; |
| 38 | |
| 39 | addr += (nr >> SHIFT_PER_LONG); |
| 40 | mask = 1L << CHOP_SHIFTCOUNT(nr); |
| 41 | _atomic_spin_lock_irqsave(addr, flags); |
| 42 | *addr |= mask; |
| 43 | _atomic_spin_unlock_irqrestore(addr, flags); |
| 44 | } |
| 45 | |
| 46 | static __inline__ void __set_bit(int nr, volatile unsigned long * address) |
| 47 | { |
| 48 | unsigned long mask; |
| 49 | unsigned long *addr = (unsigned long *) address; |
| 50 | |
| 51 | addr += (nr >> SHIFT_PER_LONG); |
| 52 | mask = 1L << CHOP_SHIFTCOUNT(nr); |
| 53 | *addr |= mask; |
| 54 | } |
| 55 | |
| 56 | static __inline__ void clear_bit(int nr, volatile unsigned long * address) |
| 57 | { |
| 58 | unsigned long mask; |
| 59 | unsigned long *addr = (unsigned long *) address; |
| 60 | unsigned long flags; |
| 61 | |
| 62 | addr += (nr >> SHIFT_PER_LONG); |
| 63 | mask = 1L << CHOP_SHIFTCOUNT(nr); |
| 64 | _atomic_spin_lock_irqsave(addr, flags); |
| 65 | *addr &= ~mask; |
| 66 | _atomic_spin_unlock_irqrestore(addr, flags); |
| 67 | } |
| 68 | |
| 69 | static __inline__ void __clear_bit(unsigned long nr, volatile unsigned long * address) |
| 70 | { |
| 71 | unsigned long mask; |
| 72 | unsigned long *addr = (unsigned long *) address; |
| 73 | |
| 74 | addr += (nr >> SHIFT_PER_LONG); |
| 75 | mask = 1L << CHOP_SHIFTCOUNT(nr); |
| 76 | *addr &= ~mask; |
| 77 | } |
| 78 | |
| 79 | static __inline__ void change_bit(int nr, volatile unsigned long * address) |
| 80 | { |
| 81 | unsigned long mask; |
| 82 | unsigned long *addr = (unsigned long *) address; |
| 83 | unsigned long flags; |
| 84 | |
| 85 | addr += (nr >> SHIFT_PER_LONG); |
| 86 | mask = 1L << CHOP_SHIFTCOUNT(nr); |
| 87 | _atomic_spin_lock_irqsave(addr, flags); |
| 88 | *addr ^= mask; |
| 89 | _atomic_spin_unlock_irqrestore(addr, flags); |
| 90 | } |
| 91 | |
| 92 | static __inline__ void __change_bit(int nr, volatile unsigned long * address) |
| 93 | { |
| 94 | unsigned long mask; |
| 95 | unsigned long *addr = (unsigned long *) address; |
| 96 | |
| 97 | addr += (nr >> SHIFT_PER_LONG); |
| 98 | mask = 1L << CHOP_SHIFTCOUNT(nr); |
| 99 | *addr ^= mask; |
| 100 | } |
| 101 | |
| 102 | static __inline__ int test_and_set_bit(int nr, volatile unsigned long * address) |
| 103 | { |
| 104 | unsigned long mask; |
| 105 | unsigned long *addr = (unsigned long *) address; |
| 106 | int oldbit; |
| 107 | unsigned long flags; |
| 108 | |
| 109 | addr += (nr >> SHIFT_PER_LONG); |
| 110 | mask = 1L << CHOP_SHIFTCOUNT(nr); |
| 111 | _atomic_spin_lock_irqsave(addr, flags); |
| 112 | oldbit = (*addr & mask) ? 1 : 0; |
| 113 | *addr |= mask; |
| 114 | _atomic_spin_unlock_irqrestore(addr, flags); |
| 115 | |
| 116 | return oldbit; |
| 117 | } |
| 118 | |
| 119 | static __inline__ int __test_and_set_bit(int nr, volatile unsigned long * address) |
| 120 | { |
| 121 | unsigned long mask; |
| 122 | unsigned long *addr = (unsigned long *) address; |
| 123 | int oldbit; |
| 124 | |
| 125 | addr += (nr >> SHIFT_PER_LONG); |
| 126 | mask = 1L << CHOP_SHIFTCOUNT(nr); |
| 127 | oldbit = (*addr & mask) ? 1 : 0; |
| 128 | *addr |= mask; |
| 129 | |
| 130 | return oldbit; |
| 131 | } |
| 132 | |
| 133 | static __inline__ int test_and_clear_bit(int nr, volatile unsigned long * address) |
| 134 | { |
| 135 | unsigned long mask; |
| 136 | unsigned long *addr = (unsigned long *) address; |
| 137 | int oldbit; |
| 138 | unsigned long flags; |
| 139 | |
| 140 | addr += (nr >> SHIFT_PER_LONG); |
| 141 | mask = 1L << CHOP_SHIFTCOUNT(nr); |
| 142 | _atomic_spin_lock_irqsave(addr, flags); |
| 143 | oldbit = (*addr & mask) ? 1 : 0; |
| 144 | *addr &= ~mask; |
| 145 | _atomic_spin_unlock_irqrestore(addr, flags); |
| 146 | |
| 147 | return oldbit; |
| 148 | } |
| 149 | |
| 150 | static __inline__ int __test_and_clear_bit(int nr, volatile unsigned long * address) |
| 151 | { |
| 152 | unsigned long mask; |
| 153 | unsigned long *addr = (unsigned long *) address; |
| 154 | int oldbit; |
| 155 | |
| 156 | addr += (nr >> SHIFT_PER_LONG); |
| 157 | mask = 1L << CHOP_SHIFTCOUNT(nr); |
| 158 | oldbit = (*addr & mask) ? 1 : 0; |
| 159 | *addr &= ~mask; |
| 160 | |
| 161 | return oldbit; |
| 162 | } |
| 163 | |
| 164 | static __inline__ int test_and_change_bit(int nr, volatile unsigned long * address) |
| 165 | { |
| 166 | unsigned long mask; |
| 167 | unsigned long *addr = (unsigned long *) address; |
| 168 | int oldbit; |
| 169 | unsigned long flags; |
| 170 | |
| 171 | addr += (nr >> SHIFT_PER_LONG); |
| 172 | mask = 1L << CHOP_SHIFTCOUNT(nr); |
| 173 | _atomic_spin_lock_irqsave(addr, flags); |
| 174 | oldbit = (*addr & mask) ? 1 : 0; |
| 175 | *addr ^= mask; |
| 176 | _atomic_spin_unlock_irqrestore(addr, flags); |
| 177 | |
| 178 | return oldbit; |
| 179 | } |
| 180 | |
| 181 | static __inline__ int __test_and_change_bit(int nr, volatile unsigned long * address) |
| 182 | { |
| 183 | unsigned long mask; |
| 184 | unsigned long *addr = (unsigned long *) address; |
| 185 | int oldbit; |
| 186 | |
| 187 | addr += (nr >> SHIFT_PER_LONG); |
| 188 | mask = 1L << CHOP_SHIFTCOUNT(nr); |
| 189 | oldbit = (*addr & mask) ? 1 : 0; |
| 190 | *addr ^= mask; |
| 191 | |
| 192 | return oldbit; |
| 193 | } |
| 194 | |
| 195 | static __inline__ int test_bit(int nr, const volatile unsigned long *address) |
| 196 | { |
| 197 | unsigned long mask; |
| 198 | const unsigned long *addr = (const unsigned long *)address; |
| 199 | |
| 200 | addr += (nr >> SHIFT_PER_LONG); |
| 201 | mask = 1L << CHOP_SHIFTCOUNT(nr); |
| 202 | |
| 203 | return !!(*addr & mask); |
| 204 | } |
| 205 | |
| 206 | #ifdef __KERNEL__ |
| 207 | |
| 208 | /** |
| 209 | * __ffs - find first bit in word. returns 0 to "BITS_PER_LONG-1". |
| 210 | * @word: The word to search |
| 211 | * |
| 212 | * __ffs() return is undefined if no bit is set. |
| 213 | * |
| 214 | * 32-bit fast __ffs by LaMont Jones "lamont At hp com". |
| 215 | * 64-bit enhancement by Grant Grundler "grundler At parisc-linux org". |
| 216 | * (with help from willy/jejb to get the semantics right) |
| 217 | * |
| 218 | * This algorithm avoids branches by making use of nullification. |
| 219 | * One side effect of "extr" instructions is it sets PSW[N] bit. |
| 220 | * How PSW[N] (nullify next insn) gets set is determined by the |
| 221 | * "condition" field (eg "<>" or "TR" below) in the extr* insn. |
| 222 | * Only the 1st and one of either the 2cd or 3rd insn will get executed. |
| 223 | * Each set of 3 insn will get executed in 2 cycles on PA8x00 vs 16 or so |
| 224 | * cycles for each mispredicted branch. |
| 225 | */ |
| 226 | |
| 227 | static __inline__ unsigned long __ffs(unsigned long x) |
| 228 | { |
| 229 | unsigned long ret; |
| 230 | |
| 231 | __asm__( |
| 232 | #if BITS_PER_LONG > 32 |
| 233 | " ldi 63,%1\n" |
| 234 | " extrd,u,*<> %0,63,32,%%r0\n" |
| 235 | " extrd,u,*TR %0,31,32,%0\n" /* move top 32-bits down */ |
| 236 | " addi -32,%1,%1\n" |
| 237 | #else |
| 238 | " ldi 31,%1\n" |
| 239 | #endif |
| 240 | " extru,<> %0,31,16,%%r0\n" |
| 241 | " extru,TR %0,15,16,%0\n" /* xxxx0000 -> 0000xxxx */ |
| 242 | " addi -16,%1,%1\n" |
| 243 | " extru,<> %0,31,8,%%r0\n" |
| 244 | " extru,TR %0,23,8,%0\n" /* 0000xx00 -> 000000xx */ |
| 245 | " addi -8,%1,%1\n" |
| 246 | " extru,<> %0,31,4,%%r0\n" |
| 247 | " extru,TR %0,27,4,%0\n" /* 000000x0 -> 0000000x */ |
| 248 | " addi -4,%1,%1\n" |
| 249 | " extru,<> %0,31,2,%%r0\n" |
| 250 | " extru,TR %0,29,2,%0\n" /* 0000000y, 1100b -> 0011b */ |
| 251 | " addi -2,%1,%1\n" |
| 252 | " extru,= %0,31,1,%%r0\n" /* check last bit */ |
| 253 | " addi -1,%1,%1\n" |
| 254 | : "+r" (x), "=r" (ret) ); |
| 255 | return ret; |
| 256 | } |
| 257 | |
| 258 | /* Undefined if no bit is zero. */ |
| 259 | #define ffz(x) __ffs(~x) |
| 260 | |
| 261 | /* |
| 262 | * ffs: find first bit set. returns 1 to BITS_PER_LONG or 0 (if none set) |
| 263 | * This is defined the same way as the libc and compiler builtin |
| 264 | * ffs routines, therefore differs in spirit from the above ffz (man ffs). |
| 265 | */ |
| 266 | static __inline__ int ffs(int x) |
| 267 | { |
| 268 | return x ? (__ffs((unsigned long)x) + 1) : 0; |
| 269 | } |
| 270 | |
| 271 | /* |
| 272 | * fls: find last (most significant) bit set. |
| 273 | * fls(0) = 0, fls(1) = 1, fls(0x80000000) = 32. |
| 274 | */ |
| 275 | |
| 276 | static __inline__ int fls(int x) |
| 277 | { |
| 278 | int ret; |
| 279 | if (!x) |
| 280 | return 0; |
| 281 | |
| 282 | __asm__( |
| 283 | " ldi 1,%1\n" |
| 284 | " extru,<> %0,15,16,%%r0\n" |
| 285 | " zdep,TR %0,15,16,%0\n" /* xxxx0000 */ |
| 286 | " addi 16,%1,%1\n" |
| 287 | " extru,<> %0,7,8,%%r0\n" |
| 288 | " zdep,TR %0,23,24,%0\n" /* xx000000 */ |
| 289 | " addi 8,%1,%1\n" |
| 290 | " extru,<> %0,3,4,%%r0\n" |
| 291 | " zdep,TR %0,27,28,%0\n" /* x0000000 */ |
| 292 | " addi 4,%1,%1\n" |
| 293 | " extru,<> %0,1,2,%%r0\n" |
| 294 | " zdep,TR %0,29,30,%0\n" /* y0000000 (y&3 = 0) */ |
| 295 | " addi 2,%1,%1\n" |
| 296 | " extru,= %0,0,1,%%r0\n" |
| 297 | " addi 1,%1,%1\n" /* if y & 8, add 1 */ |
| 298 | : "+r" (x), "=r" (ret) ); |
| 299 | |
| 300 | return ret; |
| 301 | } |
| 302 | |
| 303 | /* |
| 304 | * hweightN: returns the hamming weight (i.e. the number |
| 305 | * of bits set) of a N-bit word |
| 306 | */ |
| 307 | #define hweight64(x) \ |
| 308 | ({ \ |
| 309 | unsigned long __x = (x); \ |
| 310 | unsigned int __w; \ |
| 311 | __w = generic_hweight32((unsigned int) __x); \ |
| 312 | __w += generic_hweight32((unsigned int) (__x>>32)); \ |
| 313 | __w; \ |
| 314 | }) |
| 315 | #define hweight32(x) generic_hweight32(x) |
| 316 | #define hweight16(x) generic_hweight16(x) |
| 317 | #define hweight8(x) generic_hweight8(x) |
| 318 | |
| 319 | /* |
| 320 | * Every architecture must define this function. It's the fastest |
| 321 | * way of searching a 140-bit bitmap where the first 100 bits are |
| 322 | * unlikely to be set. It's guaranteed that at least one of the 140 |
| 323 | * bits is cleared. |
| 324 | */ |
| 325 | static inline int sched_find_first_bit(const unsigned long *b) |
| 326 | { |
| 327 | #ifndef __LP64__ |
| 328 | if (unlikely(b[0])) |
| 329 | return __ffs(b[0]); |
| 330 | if (unlikely(b[1])) |
| 331 | return __ffs(b[1]) + 32; |
| 332 | if (unlikely(b[2])) |
| 333 | return __ffs(b[2]) + 64; |
| 334 | if (b[3]) |
| 335 | return __ffs(b[3]) + 96; |
| 336 | return __ffs(b[4]) + 128; |
| 337 | #else |
| 338 | if (unlikely(b[0])) |
| 339 | return __ffs(b[0]); |
| 340 | if (unlikely(((unsigned int)b[1]))) |
| 341 | return __ffs(b[1]) + 64; |
| 342 | if (b[1] >> 32) |
| 343 | return __ffs(b[1] >> 32) + 96; |
| 344 | return __ffs(b[2]) + 128; |
| 345 | #endif |
| 346 | } |
| 347 | |
| 348 | #endif /* __KERNEL__ */ |
| 349 | |
| 350 | /* |
| 351 | * This implementation of find_{first,next}_zero_bit was stolen from |
| 352 | * Linus' asm-alpha/bitops.h. |
| 353 | */ |
| 354 | #define find_first_zero_bit(addr, size) \ |
| 355 | find_next_zero_bit((addr), (size), 0) |
| 356 | |
| 357 | static __inline__ unsigned long find_next_zero_bit(const void * addr, unsigned long size, unsigned long offset) |
| 358 | { |
| 359 | const unsigned long * p = ((unsigned long *) addr) + (offset >> SHIFT_PER_LONG); |
| 360 | unsigned long result = offset & ~(BITS_PER_LONG-1); |
| 361 | unsigned long tmp; |
| 362 | |
| 363 | if (offset >= size) |
| 364 | return size; |
| 365 | size -= result; |
| 366 | offset &= (BITS_PER_LONG-1); |
| 367 | if (offset) { |
| 368 | tmp = *(p++); |
| 369 | tmp |= ~0UL >> (BITS_PER_LONG-offset); |
| 370 | if (size < BITS_PER_LONG) |
| 371 | goto found_first; |
| 372 | if (~tmp) |
| 373 | goto found_middle; |
| 374 | size -= BITS_PER_LONG; |
| 375 | result += BITS_PER_LONG; |
| 376 | } |
| 377 | while (size & ~(BITS_PER_LONG -1)) { |
| 378 | if (~(tmp = *(p++))) |
| 379 | goto found_middle; |
| 380 | result += BITS_PER_LONG; |
| 381 | size -= BITS_PER_LONG; |
| 382 | } |
| 383 | if (!size) |
| 384 | return result; |
| 385 | tmp = *p; |
| 386 | found_first: |
| 387 | tmp |= ~0UL << size; |
| 388 | found_middle: |
| 389 | return result + ffz(tmp); |
| 390 | } |
| 391 | |
| 392 | static __inline__ unsigned long find_next_bit(const unsigned long *addr, unsigned long size, unsigned long offset) |
| 393 | { |
| 394 | const unsigned long *p = addr + (offset >> 6); |
| 395 | unsigned long result = offset & ~(BITS_PER_LONG-1); |
| 396 | unsigned long tmp; |
| 397 | |
| 398 | if (offset >= size) |
| 399 | return size; |
| 400 | size -= result; |
| 401 | offset &= (BITS_PER_LONG-1); |
| 402 | if (offset) { |
| 403 | tmp = *(p++); |
| 404 | tmp &= (~0UL << offset); |
| 405 | if (size < BITS_PER_LONG) |
| 406 | goto found_first; |
| 407 | if (tmp) |
| 408 | goto found_middle; |
| 409 | size -= BITS_PER_LONG; |
| 410 | result += BITS_PER_LONG; |
| 411 | } |
| 412 | while (size & ~(BITS_PER_LONG-1)) { |
| 413 | if ((tmp = *(p++))) |
| 414 | goto found_middle; |
| 415 | result += BITS_PER_LONG; |
| 416 | size -= BITS_PER_LONG; |
| 417 | } |
| 418 | if (!size) |
| 419 | return result; |
| 420 | tmp = *p; |
| 421 | |
| 422 | found_first: |
| 423 | tmp &= (~0UL >> (BITS_PER_LONG - size)); |
| 424 | if (tmp == 0UL) /* Are any bits set? */ |
| 425 | return result + size; /* Nope. */ |
| 426 | found_middle: |
| 427 | return result + __ffs(tmp); |
| 428 | } |
| 429 | |
| 430 | /** |
| 431 | * find_first_bit - find the first set bit in a memory region |
| 432 | * @addr: The address to start the search at |
| 433 | * @size: The maximum size to search |
| 434 | * |
| 435 | * Returns the bit-number of the first set bit, not the number of the byte |
| 436 | * containing a bit. |
| 437 | */ |
| 438 | #define find_first_bit(addr, size) \ |
| 439 | find_next_bit((addr), (size), 0) |
| 440 | |
| 441 | #define _EXT2_HAVE_ASM_BITOPS_ |
| 442 | |
| 443 | #ifdef __KERNEL__ |
| 444 | /* |
| 445 | * test_and_{set,clear}_bit guarantee atomicity without |
| 446 | * disabling interrupts. |
| 447 | */ |
| 448 | #ifdef __LP64__ |
| 449 | #define ext2_set_bit(nr, addr) __test_and_set_bit((nr) ^ 0x38, (unsigned long *)addr) |
| 450 | #define ext2_set_bit_atomic(l,nr,addr) test_and_set_bit((nr) ^ 0x38, (unsigned long *)addr) |
| 451 | #define ext2_clear_bit(nr, addr) __test_and_clear_bit((nr) ^ 0x38, (unsigned long *)addr) |
| 452 | #define ext2_clear_bit_atomic(l,nr,addr) test_and_clear_bit((nr) ^ 0x38, (unsigned long *)addr) |
| 453 | #else |
| 454 | #define ext2_set_bit(nr, addr) __test_and_set_bit((nr) ^ 0x18, (unsigned long *)addr) |
| 455 | #define ext2_set_bit_atomic(l,nr,addr) test_and_set_bit((nr) ^ 0x18, (unsigned long *)addr) |
| 456 | #define ext2_clear_bit(nr, addr) __test_and_clear_bit((nr) ^ 0x18, (unsigned long *)addr) |
| 457 | #define ext2_clear_bit_atomic(l,nr,addr) test_and_clear_bit((nr) ^ 0x18, (unsigned long *)addr) |
| 458 | #endif |
| 459 | |
| 460 | #endif /* __KERNEL__ */ |
| 461 | |
| 462 | static __inline__ int ext2_test_bit(int nr, __const__ void * addr) |
| 463 | { |
| 464 | __const__ unsigned char *ADDR = (__const__ unsigned char *) addr; |
| 465 | |
| 466 | return (ADDR[nr >> 3] >> (nr & 7)) & 1; |
| 467 | } |
| 468 | |
| 469 | /* |
| 470 | * This implementation of ext2_find_{first,next}_zero_bit was stolen from |
| 471 | * Linus' asm-alpha/bitops.h and modified for a big-endian machine. |
| 472 | */ |
| 473 | |
| 474 | #define ext2_find_first_zero_bit(addr, size) \ |
| 475 | ext2_find_next_zero_bit((addr), (size), 0) |
| 476 | |
| 477 | extern __inline__ unsigned long ext2_find_next_zero_bit(void *addr, |
| 478 | unsigned long size, unsigned long offset) |
| 479 | { |
| 480 | unsigned int *p = ((unsigned int *) addr) + (offset >> 5); |
| 481 | unsigned int result = offset & ~31UL; |
| 482 | unsigned int tmp; |
| 483 | |
| 484 | if (offset >= size) |
| 485 | return size; |
| 486 | size -= result; |
| 487 | offset &= 31UL; |
| 488 | if (offset) { |
| 489 | tmp = cpu_to_le32p(p++); |
| 490 | tmp |= ~0UL >> (32-offset); |
| 491 | if (size < 32) |
| 492 | goto found_first; |
| 493 | if (tmp != ~0U) |
| 494 | goto found_middle; |
| 495 | size -= 32; |
| 496 | result += 32; |
| 497 | } |
| 498 | while (size >= 32) { |
| 499 | if ((tmp = cpu_to_le32p(p++)) != ~0U) |
| 500 | goto found_middle; |
| 501 | result += 32; |
| 502 | size -= 32; |
| 503 | } |
| 504 | if (!size) |
| 505 | return result; |
| 506 | tmp = cpu_to_le32p(p); |
| 507 | found_first: |
| 508 | tmp |= ~0U << size; |
| 509 | found_middle: |
| 510 | return result + ffz(tmp); |
| 511 | } |
| 512 | |
| 513 | /* Bitmap functions for the minix filesystem. */ |
| 514 | #define minix_test_and_set_bit(nr,addr) ext2_set_bit(nr,addr) |
| 515 | #define minix_set_bit(nr,addr) ((void)ext2_set_bit(nr,addr)) |
| 516 | #define minix_test_and_clear_bit(nr,addr) ext2_clear_bit(nr,addr) |
| 517 | #define minix_test_bit(nr,addr) ext2_test_bit(nr,addr) |
| 518 | #define minix_find_first_zero_bit(addr,size) ext2_find_first_zero_bit(addr,size) |
| 519 | |
| 520 | #endif /* _PARISC_BITOPS_H */ |