Chris Zankel | 9a8fd55 | 2005-06-23 22:01:26 -0700 | [diff] [blame^] | 1 | /* |
| 2 | * include/asm-xtensa/bitops.h |
| 3 | * |
| 4 | * Atomic operations that C can't guarantee us.Useful for resource counting etc. |
| 5 | * |
| 6 | * This file is subject to the terms and conditions of the GNU General Public |
| 7 | * License. See the file "COPYING" in the main directory of this archive |
| 8 | * for more details. |
| 9 | * |
| 10 | * Copyright (C) 2001 - 2005 Tensilica Inc. |
| 11 | */ |
| 12 | |
| 13 | #ifndef _XTENSA_BITOPS_H |
| 14 | #define _XTENSA_BITOPS_H |
| 15 | |
| 16 | #ifdef __KERNEL__ |
| 17 | |
| 18 | #include <asm/processor.h> |
| 19 | #include <asm/byteorder.h> |
| 20 | #include <asm/system.h> |
| 21 | |
| 22 | #ifdef CONFIG_SMP |
| 23 | # error SMP not supported on this architecture |
| 24 | #endif |
| 25 | |
| 26 | static __inline__ void set_bit(int nr, volatile void * addr) |
| 27 | { |
| 28 | unsigned long mask = 1 << (nr & 0x1f); |
| 29 | unsigned long *a = ((unsigned long *)addr) + (nr >> 5); |
| 30 | unsigned long flags; |
| 31 | |
| 32 | local_irq_save(flags); |
| 33 | *a |= mask; |
| 34 | local_irq_restore(flags); |
| 35 | } |
| 36 | |
| 37 | static __inline__ void __set_bit(int nr, volatile unsigned long * addr) |
| 38 | { |
| 39 | unsigned long mask = 1 << (nr & 0x1f); |
| 40 | unsigned long *a = ((unsigned long *)addr) + (nr >> 5); |
| 41 | |
| 42 | *a |= mask; |
| 43 | } |
| 44 | |
| 45 | static __inline__ void clear_bit(int nr, volatile void * addr) |
| 46 | { |
| 47 | unsigned long mask = 1 << (nr & 0x1f); |
| 48 | unsigned long *a = ((unsigned long *)addr) + (nr >> 5); |
| 49 | unsigned long flags; |
| 50 | |
| 51 | local_irq_save(flags); |
| 52 | *a &= ~mask; |
| 53 | local_irq_restore(flags); |
| 54 | } |
| 55 | |
| 56 | static __inline__ void __clear_bit(int nr, volatile unsigned long *addr) |
| 57 | { |
| 58 | unsigned long mask = 1 << (nr & 0x1f); |
| 59 | unsigned long *a = ((unsigned long *)addr) + (nr >> 5); |
| 60 | |
| 61 | *a &= ~mask; |
| 62 | } |
| 63 | |
| 64 | /* |
| 65 | * clear_bit() doesn't provide any barrier for the compiler. |
| 66 | */ |
| 67 | |
| 68 | #define smp_mb__before_clear_bit() barrier() |
| 69 | #define smp_mb__after_clear_bit() barrier() |
| 70 | |
| 71 | static __inline__ void change_bit(int nr, volatile void * addr) |
| 72 | { |
| 73 | unsigned long mask = 1 << (nr & 0x1f); |
| 74 | unsigned long *a = ((unsigned long *)addr) + (nr >> 5); |
| 75 | unsigned long flags; |
| 76 | |
| 77 | local_irq_save(flags); |
| 78 | *a ^= mask; |
| 79 | local_irq_restore(flags); |
| 80 | } |
| 81 | |
| 82 | static __inline__ void __change_bit(int nr, volatile void * addr) |
| 83 | { |
| 84 | unsigned long mask = 1 << (nr & 0x1f); |
| 85 | unsigned long *a = ((unsigned long *)addr) + (nr >> 5); |
| 86 | |
| 87 | *a ^= mask; |
| 88 | } |
| 89 | |
| 90 | static __inline__ int test_and_set_bit(int nr, volatile void * addr) |
| 91 | { |
| 92 | unsigned long retval; |
| 93 | unsigned long mask = 1 << (nr & 0x1f); |
| 94 | unsigned long *a = ((unsigned long *)addr) + (nr >> 5); |
| 95 | unsigned long flags; |
| 96 | |
| 97 | local_irq_save(flags); |
| 98 | retval = (mask & *a) != 0; |
| 99 | *a |= mask; |
| 100 | local_irq_restore(flags); |
| 101 | |
| 102 | return retval; |
| 103 | } |
| 104 | |
| 105 | static __inline__ int __test_and_set_bit(int nr, volatile void * addr) |
| 106 | { |
| 107 | unsigned long retval; |
| 108 | unsigned long mask = 1 << (nr & 0x1f); |
| 109 | unsigned long *a = ((unsigned long *)addr) + (nr >> 5); |
| 110 | |
| 111 | retval = (mask & *a) != 0; |
| 112 | *a |= mask; |
| 113 | |
| 114 | return retval; |
| 115 | } |
| 116 | |
| 117 | static __inline__ int test_and_clear_bit(int nr, volatile void * addr) |
| 118 | { |
| 119 | unsigned long retval; |
| 120 | unsigned long mask = 1 << (nr & 0x1f); |
| 121 | unsigned long *a = ((unsigned long *)addr) + (nr >> 5); |
| 122 | unsigned long flags; |
| 123 | |
| 124 | local_irq_save(flags); |
| 125 | retval = (mask & *a) != 0; |
| 126 | *a &= ~mask; |
| 127 | local_irq_restore(flags); |
| 128 | |
| 129 | return retval; |
| 130 | } |
| 131 | |
| 132 | static __inline__ int __test_and_clear_bit(int nr, volatile void * addr) |
| 133 | { |
| 134 | unsigned long mask = 1 << (nr & 0x1f); |
| 135 | unsigned long *a = ((unsigned long *)addr) + (nr >> 5); |
| 136 | unsigned long old = *a; |
| 137 | |
| 138 | *a = old & ~mask; |
| 139 | return (old & mask) != 0; |
| 140 | } |
| 141 | |
| 142 | static __inline__ int test_and_change_bit(int nr, volatile void * addr) |
| 143 | { |
| 144 | unsigned long retval; |
| 145 | unsigned long mask = 1 << (nr & 0x1f); |
| 146 | unsigned long *a = ((unsigned long *)addr) + (nr >> 5); |
| 147 | unsigned long flags; |
| 148 | |
| 149 | local_irq_save(flags); |
| 150 | |
| 151 | retval = (mask & *a) != 0; |
| 152 | *a ^= mask; |
| 153 | local_irq_restore(flags); |
| 154 | |
| 155 | return retval; |
| 156 | } |
| 157 | |
| 158 | /* |
| 159 | * non-atomic version; can be reordered |
| 160 | */ |
| 161 | |
| 162 | static __inline__ int __test_and_change_bit(int nr, volatile void *addr) |
| 163 | { |
| 164 | unsigned long mask = 1 << (nr & 0x1f); |
| 165 | unsigned long *a = ((unsigned long *)addr) + (nr >> 5); |
| 166 | unsigned long old = *a; |
| 167 | |
| 168 | *a = old ^ mask; |
| 169 | return (old & mask) != 0; |
| 170 | } |
| 171 | |
| 172 | static __inline__ int test_bit(int nr, const volatile void *addr) |
| 173 | { |
| 174 | return 1UL & (((const volatile unsigned int *)addr)[nr>>5] >> (nr&31)); |
| 175 | } |
| 176 | |
| 177 | #if XCHAL_HAVE_NSAU |
| 178 | |
| 179 | static __inline__ int __cntlz (unsigned long x) |
| 180 | { |
| 181 | int lz; |
| 182 | asm ("nsau %0, %1" : "=r" (lz) : "r" (x)); |
| 183 | return 31 - lz; |
| 184 | } |
| 185 | |
| 186 | #else |
| 187 | |
| 188 | static __inline__ int __cntlz (unsigned long x) |
| 189 | { |
| 190 | unsigned long sum, x1, x2, x4, x8, x16; |
| 191 | x1 = x & 0xAAAAAAAA; |
| 192 | x2 = x & 0xCCCCCCCC; |
| 193 | x4 = x & 0xF0F0F0F0; |
| 194 | x8 = x & 0xFF00FF00; |
| 195 | x16 = x & 0xFFFF0000; |
| 196 | sum = x2 ? 2 : 0; |
| 197 | sum += (x16 != 0) * 16; |
| 198 | sum += (x8 != 0) * 8; |
| 199 | sum += (x4 != 0) * 4; |
| 200 | sum += (x1 != 0); |
| 201 | |
| 202 | return sum; |
| 203 | } |
| 204 | |
| 205 | #endif |
| 206 | |
| 207 | /* |
| 208 | * ffz: Find first zero in word. Undefined if no zero exists. |
| 209 | * bit 0 is the LSB of addr; bit 32 is the LSB of (addr+1). |
| 210 | */ |
| 211 | |
| 212 | static __inline__ int ffz(unsigned long x) |
| 213 | { |
| 214 | if ((x = ~x) == 0) |
| 215 | return 32; |
| 216 | return __cntlz(x & -x); |
| 217 | } |
| 218 | |
| 219 | /* |
| 220 | * __ffs: Find first bit set in word. Return 0 for bit 0 |
| 221 | */ |
| 222 | |
| 223 | static __inline__ int __ffs(unsigned long x) |
| 224 | { |
| 225 | return __cntlz(x & -x); |
| 226 | } |
| 227 | |
| 228 | /* |
| 229 | * ffs: Find first bit set in word. This is defined the same way as |
| 230 | * the libc and compiler builtin ffs routines, therefore |
| 231 | * differs in spirit from the above ffz (man ffs). |
| 232 | */ |
| 233 | |
| 234 | static __inline__ int ffs(unsigned long x) |
| 235 | { |
| 236 | return __cntlz(x & -x) + 1; |
| 237 | } |
| 238 | |
| 239 | /* |
| 240 | * fls: Find last (most-significant) bit set in word. |
| 241 | * Note fls(0) = 0, fls(1) = 1, fls(0x80000000) = 32. |
| 242 | */ |
| 243 | |
| 244 | static __inline__ int fls (unsigned int x) |
| 245 | { |
| 246 | return __cntlz(x); |
| 247 | } |
| 248 | |
| 249 | static __inline__ int |
| 250 | find_next_bit(const unsigned long *addr, int size, int offset) |
| 251 | { |
| 252 | const unsigned long *p = addr + (offset >> 5); |
| 253 | unsigned long result = offset & ~31UL; |
| 254 | unsigned long tmp; |
| 255 | |
| 256 | if (offset >= size) |
| 257 | return size; |
| 258 | size -= result; |
| 259 | offset &= 31UL; |
| 260 | if (offset) { |
| 261 | tmp = *p++; |
| 262 | tmp &= ~0UL << offset; |
| 263 | if (size < 32) |
| 264 | goto found_first; |
| 265 | if (tmp) |
| 266 | goto found_middle; |
| 267 | size -= 32; |
| 268 | result += 32; |
| 269 | } |
| 270 | while (size >= 32) { |
| 271 | if ((tmp = *p++) != 0) |
| 272 | goto found_middle; |
| 273 | result += 32; |
| 274 | size -= 32; |
| 275 | } |
| 276 | if (!size) |
| 277 | return result; |
| 278 | tmp = *p; |
| 279 | |
| 280 | found_first: |
| 281 | tmp &= ~0UL >> (32 - size); |
| 282 | if (tmp == 0UL) /* Are any bits set? */ |
| 283 | return result + size; /* Nope. */ |
| 284 | found_middle: |
| 285 | return result + __ffs(tmp); |
| 286 | } |
| 287 | |
| 288 | /** |
| 289 | * find_first_bit - find the first set bit in a memory region |
| 290 | * @addr: The address to start the search at |
| 291 | * @size: The maximum size to search |
| 292 | * |
| 293 | * Returns the bit-number of the first set bit, not the number of the byte |
| 294 | * containing a bit. |
| 295 | */ |
| 296 | |
| 297 | #define find_first_bit(addr, size) \ |
| 298 | find_next_bit((addr), (size), 0) |
| 299 | |
| 300 | static __inline__ int |
| 301 | find_next_zero_bit(const unsigned long *addr, int size, int offset) |
| 302 | { |
| 303 | const unsigned long *p = addr + (offset >> 5); |
| 304 | unsigned long result = offset & ~31UL; |
| 305 | unsigned long tmp; |
| 306 | |
| 307 | if (offset >= size) |
| 308 | return size; |
| 309 | size -= result; |
| 310 | offset &= 31UL; |
| 311 | if (offset) { |
| 312 | tmp = *p++; |
| 313 | tmp |= ~0UL >> (32-offset); |
| 314 | if (size < 32) |
| 315 | goto found_first; |
| 316 | if (~tmp) |
| 317 | goto found_middle; |
| 318 | size -= 32; |
| 319 | result += 32; |
| 320 | } |
| 321 | while (size & ~31UL) { |
| 322 | if (~(tmp = *p++)) |
| 323 | goto found_middle; |
| 324 | result += 32; |
| 325 | size -= 32; |
| 326 | } |
| 327 | if (!size) |
| 328 | return result; |
| 329 | tmp = *p; |
| 330 | |
| 331 | found_first: |
| 332 | tmp |= ~0UL << size; |
| 333 | found_middle: |
| 334 | return result + ffz(tmp); |
| 335 | } |
| 336 | |
| 337 | #define find_first_zero_bit(addr, size) \ |
| 338 | find_next_zero_bit((addr), (size), 0) |
| 339 | |
| 340 | #ifdef __XTENSA_EL__ |
| 341 | # define ext2_set_bit(nr,addr) __test_and_set_bit((nr), (addr)) |
| 342 | # define ext2_set_bit_atomic(lock,nr,addr) test_and_set_bit((nr),(addr)) |
| 343 | # define ext2_clear_bit(nr,addr) __test_and_clear_bit((nr), (addr)) |
| 344 | # define ext2_clear_bit_atomic(lock,nr,addr) test_and_clear_bit((nr),(addr)) |
| 345 | # define ext2_test_bit(nr,addr) test_bit((nr), (addr)) |
| 346 | # define ext2_find_first_zero_bit(addr, size) find_first_zero_bit((addr),(size)) |
| 347 | # define ext2_find_next_zero_bit(addr, size, offset) \ |
| 348 | find_next_zero_bit((addr), (size), (offset)) |
| 349 | #elif defined(__XTENSA_EB__) |
| 350 | # define ext2_set_bit(nr,addr) __test_and_set_bit((nr) ^ 0x18, (addr)) |
| 351 | # define ext2_set_bit_atomic(lock,nr,addr) test_and_set_bit((nr) ^ 0x18, (addr)) |
| 352 | # define ext2_clear_bit(nr,addr) __test_and_clear_bit((nr) ^ 18, (addr)) |
| 353 | # define ext2_clear_bit_atomic(lock,nr,addr) test_and_clear_bit((nr)^0x18,(addr)) |
| 354 | # define ext2_test_bit(nr,addr) test_bit((nr) ^ 0x18, (addr)) |
| 355 | # define ext2_find_first_zero_bit(addr, size) \ |
| 356 | ext2_find_next_zero_bit((addr), (size), 0) |
| 357 | |
| 358 | static __inline__ unsigned long ext2_find_next_zero_bit(void *addr, unsigned long size, unsigned long offset) |
| 359 | { |
| 360 | unsigned long *p = ((unsigned long *) addr) + (offset >> 5); |
| 361 | unsigned long result = offset & ~31UL; |
| 362 | unsigned long tmp; |
| 363 | |
| 364 | if (offset >= size) |
| 365 | return size; |
| 366 | size -= result; |
| 367 | offset &= 31UL; |
| 368 | if(offset) { |
| 369 | /* We hold the little endian value in tmp, but then the |
| 370 | * shift is illegal. So we could keep a big endian value |
| 371 | * in tmp, like this: |
| 372 | * |
| 373 | * tmp = __swab32(*(p++)); |
| 374 | * tmp |= ~0UL >> (32-offset); |
| 375 | * |
| 376 | * but this would decrease preformance, so we change the |
| 377 | * shift: |
| 378 | */ |
| 379 | tmp = *(p++); |
| 380 | tmp |= __swab32(~0UL >> (32-offset)); |
| 381 | if(size < 32) |
| 382 | goto found_first; |
| 383 | if(~tmp) |
| 384 | goto found_middle; |
| 385 | size -= 32; |
| 386 | result += 32; |
| 387 | } |
| 388 | while(size & ~31UL) { |
| 389 | if(~(tmp = *(p++))) |
| 390 | goto found_middle; |
| 391 | result += 32; |
| 392 | size -= 32; |
| 393 | } |
| 394 | if(!size) |
| 395 | return result; |
| 396 | tmp = *p; |
| 397 | |
| 398 | found_first: |
| 399 | /* tmp is little endian, so we would have to swab the shift, |
| 400 | * see above. But then we have to swab tmp below for ffz, so |
| 401 | * we might as well do this here. |
| 402 | */ |
| 403 | return result + ffz(__swab32(tmp) | (~0UL << size)); |
| 404 | found_middle: |
| 405 | return result + ffz(__swab32(tmp)); |
| 406 | } |
| 407 | |
| 408 | #else |
| 409 | # error processor byte order undefined! |
| 410 | #endif |
| 411 | |
| 412 | |
| 413 | #define hweight32(x) generic_hweight32(x) |
| 414 | #define hweight16(x) generic_hweight16(x) |
| 415 | #define hweight8(x) generic_hweight8(x) |
| 416 | |
| 417 | /* |
| 418 | * Find the first bit set in a 140-bit bitmap. |
| 419 | * The first 100 bits are unlikely to be set. |
| 420 | */ |
| 421 | |
| 422 | static inline int sched_find_first_bit(const unsigned long *b) |
| 423 | { |
| 424 | if (unlikely(b[0])) |
| 425 | return __ffs(b[0]); |
| 426 | if (unlikely(b[1])) |
| 427 | return __ffs(b[1]) + 32; |
| 428 | if (unlikely(b[2])) |
| 429 | return __ffs(b[2]) + 64; |
| 430 | if (b[3]) |
| 431 | return __ffs(b[3]) + 96; |
| 432 | return __ffs(b[4]) + 128; |
| 433 | } |
| 434 | |
| 435 | |
| 436 | /* Bitmap functions for the minix filesystem. */ |
| 437 | |
| 438 | #define minix_test_and_set_bit(nr,addr) test_and_set_bit(nr,addr) |
| 439 | #define minix_set_bit(nr,addr) set_bit(nr,addr) |
| 440 | #define minix_test_and_clear_bit(nr,addr) test_and_clear_bit(nr,addr) |
| 441 | #define minix_test_bit(nr,addr) test_bit(nr,addr) |
| 442 | #define minix_find_first_zero_bit(addr,size) find_first_zero_bit(addr,size) |
| 443 | |
| 444 | #endif /* __KERNEL__ */ |
| 445 | |
| 446 | #endif /* _XTENSA_BITOPS_H */ |