include/asm-parisc/bitops.h

   1 #ifndef _PARISC_BITOPS_H
   2 #define _PARISC_BITOPS_H
   3
   4 #include <linux/compiler.h>
   5 #include <asm/types.h>          /* for BITS_PER_LONG/SHIFT_PER_LONG */
   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 #define CHOP_SHIFTCOUNT(x) (((unsigned long) (x)) & (BITS_PER_LONG - 1))
  16
  17
  18 #define smp_mb__before_clear_bit()      smp_mb()
  19 #define smp_mb__after_clear_bit()       smp_mb()
  20
  21 /* See http://marc.theaimsgroup.com/?t=108826637900003 for discussion
  22  * on use of volatile and __*_bit() (set/clear/change):
  23  *      *_bit() want use of volatile.
  24  *      __*_bit() are "relaxed" and don't use spinlock or volatile.
  25  */
  26
  27 static __inline__ void set_bit(int nr, volatile unsigned long * addr)
  28 {
  29         unsigned long mask = 1UL << CHOP_SHIFTCOUNT(nr);
  30         unsigned long flags;
  31
  32         addr += (nr >> SHIFT_PER_LONG);
  33         _atomic_spin_lock_irqsave(addr, flags);
  34         *addr |= mask;
  35         _atomic_spin_unlock_irqrestore(addr, flags);
  36 }
  37
  38 static __inline__ void __set_bit(unsigned long nr, volatile unsigned long * addr)
  39 {
  40         unsigned long *m = (unsigned long *) addr + (nr >> SHIFT_PER_LONG);
  41
  42         *m |= 1UL << CHOP_SHIFTCOUNT(nr);
  43 }
  44
  45 static __inline__ void clear_bit(int nr, volatile unsigned long * addr)
  46 {
  47         unsigned long mask = ~(1UL << CHOP_SHIFTCOUNT(nr));
  48         unsigned long flags;
  49
  50         addr += (nr >> SHIFT_PER_LONG);
  51         _atomic_spin_lock_irqsave(addr, flags);
  52         *addr &= mask;
  53         _atomic_spin_unlock_irqrestore(addr, flags);
  54 }
  55
  56 static __inline__ void __clear_bit(unsigned long nr, volatile unsigned long * addr)
  57 {
  58         unsigned long *m = (unsigned long *) addr + (nr >> SHIFT_PER_LONG);
  59
  60         *m &= ~(1UL << CHOP_SHIFTCOUNT(nr));
  61 }
  62
  63 static __inline__ void change_bit(int nr, volatile unsigned long * addr)
  64 {
  65         unsigned long mask = 1UL << CHOP_SHIFTCOUNT(nr);
  66         unsigned long flags;
  67
  68         addr += (nr >> SHIFT_PER_LONG);
  69         _atomic_spin_lock_irqsave(addr, flags);
  70         *addr ^= mask;
  71         _atomic_spin_unlock_irqrestore(addr, flags);
  72 }
  73
  74 static __inline__ void __change_bit(unsigned long nr, volatile unsigned long * addr)
  75 {
  76         unsigned long *m = (unsigned long *) addr + (nr >> SHIFT_PER_LONG);
  77
  78         *m ^= 1UL << CHOP_SHIFTCOUNT(nr);
  79 }
  80
  81 static __inline__ int test_and_set_bit(int nr, volatile unsigned long * addr)
  82 {
  83         unsigned long mask = 1UL << CHOP_SHIFTCOUNT(nr);
  84         unsigned long oldbit;
  85         unsigned long flags;
  86
  87         addr += (nr >> SHIFT_PER_LONG);
  88         _atomic_spin_lock_irqsave(addr, flags);
  89         oldbit = *addr;
  90         *addr = oldbit | mask;
  91         _atomic_spin_unlock_irqrestore(addr, flags);
  92
  93         return (oldbit & mask) ? 1 : 0;
  94 }
  95
  96 static __inline__ int __test_and_set_bit(int nr, volatile unsigned long * address)
  97 {
  98         unsigned long mask = 1UL << CHOP_SHIFTCOUNT(nr);
  99         unsigned long oldbit;
 100         unsigned long *addr = (unsigned long *)address + (nr >> SHIFT_PER_LONG);
 101
 102         oldbit = *addr;
 103         *addr = oldbit | mask;
 104
 105         return (oldbit & mask) ? 1 : 0;
 106 }
 107
 108 static __inline__ int test_and_clear_bit(int nr, volatile unsigned long * addr)
 109 {
 110         unsigned long mask = 1UL << CHOP_SHIFTCOUNT(nr);
 111         unsigned long oldbit;
 112         unsigned long flags;
 113
 114         addr += (nr >> SHIFT_PER_LONG);
 115         _atomic_spin_lock_irqsave(addr, flags);
 116         oldbit = *addr;
 117         *addr = oldbit & ~mask;
 118         _atomic_spin_unlock_irqrestore(addr, flags);
 119
 120         return (oldbit & mask) ? 1 : 0;
 121 }
 122
 123 static __inline__ int __test_and_clear_bit(int nr, volatile unsigned long * address)
 124 {
 125         unsigned long mask = 1UL << CHOP_SHIFTCOUNT(nr);
 126         unsigned long *addr = (unsigned long *)address + (nr >> SHIFT_PER_LONG);
 127         unsigned long oldbit;
 128
 129         oldbit = *addr;
 130         *addr = oldbit & ~mask;
 131
 132         return (oldbit & mask) ? 1 : 0;
 133 }
 134
 135 static __inline__ int test_and_change_bit(int nr, volatile unsigned long * addr)
 136 {
 137         unsigned long mask = 1UL << CHOP_SHIFTCOUNT(nr);
 138         unsigned long oldbit;
 139         unsigned long flags;
 140
 141         addr += (nr >> SHIFT_PER_LONG);
 142         _atomic_spin_lock_irqsave(addr, flags);
 143         oldbit = *addr;
 144         *addr = oldbit ^ mask;
 145         _atomic_spin_unlock_irqrestore(addr, flags);
 146
 147         return (oldbit & mask) ? 1 : 0;
 148 }
 149
 150 static __inline__ int __test_and_change_bit(int nr, volatile unsigned long * address)
 151 {
 152         unsigned long mask = 1UL << CHOP_SHIFTCOUNT(nr);
 153         unsigned long *addr = (unsigned long *)address + (nr >> SHIFT_PER_LONG);
 154         unsigned long oldbit;
 155
 156         oldbit = *addr;
 157         *addr = oldbit ^ mask;
 158
 159         return (oldbit & mask) ? 1 : 0;
 160 }
 161
 162 static __inline__ int test_bit(int nr, const volatile unsigned long *address)
 163 {
 164         unsigned long mask = 1UL << CHOP_SHIFTCOUNT(nr);
 165         const unsigned long *addr = (const unsigned long *)address + (nr >> SHIFT_PER_LONG);
 166
 167         return !!(*addr & mask);
 168 }
 169
 170 #ifdef __KERNEL__
 171
 172 /**
 173  * __ffs - find first bit in word. returns 0 to "BITS_PER_LONG-1".
 174  * @word: The word to search
 175  *
 176  * __ffs() return is undefined if no bit is set.
 177  *
 178  * 32-bit fast __ffs by LaMont Jones "lamont At hp com".
 179  * 64-bit enhancement by Grant Grundler "grundler At parisc-linux org".
 180  * (with help from willy/jejb to get the semantics right)
 181  *
 182  * This algorithm avoids branches by making use of nullification.
 183  * One side effect of "extr" instructions is it sets PSW[N] bit.
 184  * How PSW[N] (nullify next insn) gets set is determined by the
 185  * "condition" field (eg "<>" or "TR" below) in the extr* insn.
 186  * Only the 1st and one of either the 2cd or 3rd insn will get executed.
 187  * Each set of 3 insn will get executed in 2 cycles on PA8x00 vs 16 or so
 188  * cycles for each mispredicted branch.
 189  */
 190
 191 static __inline__ unsigned long __ffs(unsigned long x)
 192 {
 193         unsigned long ret;
 194
 195         __asm__(
 196 #ifdef __LP64__
 197                 " ldi       63,%1\n"
 198                 " extrd,u,*<>  %0,63,32,%%r0\n"
 199                 " extrd,u,*TR  %0,31,32,%0\n"   /* move top 32-bits down */
 200                 " addi    -32,%1,%1\n"
 201 #else
 202                 " ldi       31,%1\n"
 203 #endif
 204                 " extru,<>  %0,31,16,%%r0\n"
 205                 " extru,TR  %0,15,16,%0\n"      /* xxxx0000 -> 0000xxxx */
 206                 " addi    -16,%1,%1\n"
 207                 " extru,<>  %0,31,8,%%r0\n"
 208                 " extru,TR  %0,23,8,%0\n"       /* 0000xx00 -> 000000xx */
 209                 " addi    -8,%1,%1\n"
 210                 " extru,<>  %0,31,4,%%r0\n"
 211                 " extru,TR  %0,27,4,%0\n"       /* 000000x0 -> 0000000x */
 212                 " addi    -4,%1,%1\n"
 213                 " extru,<>  %0,31,2,%%r0\n"
 214                 " extru,TR  %0,29,2,%0\n"       /* 0000000y, 1100b -> 0011b */
 215                 " addi    -2,%1,%1\n"
 216                 " extru,=  %0,31,1,%%r0\n"      /* check last bit */
 217                 " addi    -1,%1,%1\n"
 218                         : "+r" (x), "=r" (ret) );
 219         return ret;
 220 }
 221
 222 /* Undefined if no bit is zero. */
 223 #define ffz(x)  __ffs(~x)
 224
 225 /*
 226  * ffs: find first bit set. returns 1 to BITS_PER_LONG or 0 (if none set)
 227  * This is defined the same way as the libc and compiler builtin
 228  * ffs routines, therefore differs in spirit from the above ffz (man ffs).
 229  */
 230 static __inline__ int ffs(int x)
 231 {
 232         return x ? (__ffs((unsigned long)x) + 1) : 0;
 233 }
 234
 235 /*
 236  * fls: find last (most significant) bit set.
 237  * fls(0) = 0, fls(1) = 1, fls(0x80000000) = 32.
 238  */
 239
 240 static __inline__ int fls(int x)
 241 {
 242         int ret;
 243         if (!x)
 244                 return 0;
 245
 246         __asm__(
 247         "       ldi             1,%1\n"
 248         "       extru,<>        %0,15,16,%%r0\n"
 249         "       zdep,TR         %0,15,16,%0\n"          /* xxxx0000 */
 250         "       addi            16,%1,%1\n"
 251         "       extru,<>        %0,7,8,%%r0\n"
 252         "       zdep,TR         %0,23,24,%0\n"          /* xx000000 */
 253         "       addi            8,%1,%1\n"
 254         "       extru,<>        %0,3,4,%%r0\n"
 255         "       zdep,TR         %0,27,28,%0\n"          /* x0000000 */
 256         "       addi            4,%1,%1\n"
 257         "       extru,<>        %0,1,2,%%r0\n"
 258         "       zdep,TR         %0,29,30,%0\n"          /* y0000000 (y&3 = 0) */
 259         "       addi            2,%1,%1\n"
 260         "       extru,=         %0,0,1,%%r0\n"
 261         "       addi            1,%1,%1\n"              /* if y & 8, add 1 */
 262                 : "+r" (x), "=r" (ret) );
 263
 264         return ret;
 265 }
 266
 267 /*
 268  * hweightN: returns the hamming weight (i.e. the number
 269  * of bits set) of a N-bit word
 270  */
 271 #define hweight64(x) generic_hweight64(x)
 272 #define hweight32(x) generic_hweight32(x)
 273 #define hweight16(x) generic_hweight16(x)
 274 #define hweight8(x) generic_hweight8(x)
 275
 276 /*
 277  * Every architecture must define this function. It's the fastest
 278  * way of searching a 140-bit bitmap where the first 100 bits are
 279  * unlikely to be set. It's guaranteed that at least one of the 140
 280  * bits is cleared.
 281  */
 282 static inline int sched_find_first_bit(const unsigned long *b)
 283 {
 284 #ifdef __LP64__
 285         if (unlikely(b[0]))
 286                 return __ffs(b[0]);
 287         if (unlikely(b[1]))
 288                 return __ffs(b[1]) + 64;
 289         return __ffs(b[2]) + 128;
 290 #else
 291         if (unlikely(b[0]))
 292                 return __ffs(b[0]);
 293         if (unlikely(b[1]))
 294                 return __ffs(b[1]) + 32;
 295         if (unlikely(b[2]))
 296                 return __ffs(b[2]) + 64;
 297         if (b[3])
 298                 return __ffs(b[3]) + 96;
 299         return __ffs(b[4]) + 128;
 300 #endif
 301 }
 302
 303 #endif /* __KERNEL__ */
 304
 305 /*
 306  * This implementation of find_{first,next}_zero_bit was stolen from
 307  * Linus' asm-alpha/bitops.h.
 308  */
 309 #define find_first_zero_bit(addr, size) \
 310         find_next_zero_bit((addr), (size), 0)
 311
 312 static __inline__ unsigned long find_next_zero_bit(const void * addr, unsigned long size, unsigned long offset)
 313 {
 314         const unsigned long * p = ((unsigned long *) addr) + (offset >> SHIFT_PER_LONG);
 315         unsigned long result = offset & ~(BITS_PER_LONG-1);
 316         unsigned long tmp;
 317
 318         if (offset >= size)
 319                 return size;
 320         size -= result;
 321         offset &= (BITS_PER_LONG-1);
 322         if (offset) {
 323                 tmp = *(p++);
 324                 tmp |= ~0UL >> (BITS_PER_LONG-offset);
 325                 if (size < BITS_PER_LONG)
 326                         goto found_first;
 327                 if (~tmp)
 328                         goto found_middle;
 329                 size -= BITS_PER_LONG;
 330                 result += BITS_PER_LONG;
 331         }
 332         while (size & ~(BITS_PER_LONG -1)) {
 333                 if (~(tmp = *(p++)))
 334                         goto found_middle;
 335                 result += BITS_PER_LONG;
 336                 size -= BITS_PER_LONG;
 337         }
 338         if (!size)
 339                 return result;
 340         tmp = *p;
 341 found_first:
 342         tmp |= ~0UL << size;
 343 found_middle:
 344         return result + ffz(tmp);
 345 }
 346
 347 static __inline__ unsigned long find_next_bit(const unsigned long *addr, unsigned long size, unsigned long offset)
 348 {
 349         const unsigned long *p = addr + (offset >> SHIFT_PER_LONG);
 350         unsigned long result = offset & ~(BITS_PER_LONG-1);
 351         unsigned long tmp;
 352
 353         if (offset >= size)
 354                 return size;
 355         size -= result;
 356         offset &= (BITS_PER_LONG-1);
 357         if (offset) {
 358                 tmp = *(p++);
 359                 tmp &= (~0UL << offset);
 360                 if (size < BITS_PER_LONG)
 361                         goto found_first;
 362                 if (tmp)
 363                         goto found_middle;
 364                 size -= BITS_PER_LONG;
 365                 result += BITS_PER_LONG;
 366         }
 367         while (size & ~(BITS_PER_LONG-1)) {
 368                 if ((tmp = *(p++)))
 369                         goto found_middle;
 370                 result += BITS_PER_LONG;
 371                 size -= BITS_PER_LONG;
 372         }
 373         if (!size)
 374                 return result;
 375         tmp = *p;
 376
 377 found_first:
 378         tmp &= (~0UL >> (BITS_PER_LONG - size));
 379         if (tmp == 0UL)        /* Are any bits set? */
 380                 return result + size; /* Nope. */
 381 found_middle:
 382         return result + __ffs(tmp);
 383 }
 384
 385 /**
 386  * find_first_bit - find the first set bit in a memory region
 387  * @addr: The address to start the search at
 388  * @size: The maximum size to search
 389  *
 390  * Returns the bit-number of the first set bit, not the number of the byte
 391  * containing a bit.
 392  */
 393 #define find_first_bit(addr, size) \
 394         find_next_bit((addr), (size), 0)
 395
 396 #define _EXT2_HAVE_ASM_BITOPS_
 397
 398 #ifdef __KERNEL__
 399 /*
 400  * test_and_{set,clear}_bit guarantee atomicity without
 401  * disabling interrupts.
 402  */
 403
 404 /* '3' is bits per byte */
 405 #define LE_BYTE_ADDR ((sizeof(unsigned long) - 1) << 3)
 406
 407 #define ext2_test_bit(nr, addr) \
 408                         test_bit((nr)   ^ LE_BYTE_ADDR, (unsigned long *)addr)
 409 #define ext2_set_bit(nr, addr)  \
 410                 __test_and_set_bit((nr) ^ LE_BYTE_ADDR, (unsigned long *)addr)
 411 #define ext2_clear_bit(nr, addr) \
 412                 __test_and_clear_bit((nr) ^ LE_BYTE_ADDR, (unsigned long *)addr)
 413
 414 #define ext2_set_bit_atomic(l,nr,addr) \
 415                 test_and_set_bit((nr)   ^ LE_BYTE_ADDR, (unsigned long *)addr)
 416 #define ext2_clear_bit_atomic(l,nr,addr) \
 417                 test_and_clear_bit( (nr) ^ LE_BYTE_ADDR, (unsigned long *)addr)
 418
 419 #endif  /* __KERNEL__ */
 420
 421
 422 #define ext2_find_first_zero_bit(addr, size) \
 423         ext2_find_next_zero_bit((addr), (size), 0)
 424
 425 /* include/linux/byteorder does not support "unsigned long" type */
 426 static inline unsigned long ext2_swabp(unsigned long * x)
 427 {
 428 #ifdef __LP64__
 429         return (unsigned long) __swab64p((u64 *) x);
 430 #else
 431         return (unsigned long) __swab32p((u32 *) x);
 432 #endif
 433 }
 434
 435 /* include/linux/byteorder doesn't support "unsigned long" type */
 436 static inline unsigned long ext2_swab(unsigned long y)
 437 {
 438 #ifdef __LP64__
 439         return (unsigned long) __swab64((u64) y);
 440 #else
 441         return (unsigned long) __swab32((u32) y);
 442 #endif
 443 }
 444
 445 static __inline__ unsigned long ext2_find_next_zero_bit(void *addr, unsigned long size, unsigned long offset)
 446 {
 447         unsigned long *p = (unsigned long *) addr + (offset >> SHIFT_PER_LONG);
 448         unsigned long result = offset & ~(BITS_PER_LONG - 1);
 449         unsigned long tmp;
 450
 451         if (offset >= size)
 452                 return size;
 453         size -= result;
 454         offset &= (BITS_PER_LONG - 1UL);
 455         if (offset) {
 456                 tmp = ext2_swabp(p++);
 457                 tmp |= (~0UL >> (BITS_PER_LONG - offset));
 458                 if (size < BITS_PER_LONG)
 459                         goto found_first;
 460                 if (~tmp)
 461                         goto found_middle;
 462                 size -= BITS_PER_LONG;
 463                 result += BITS_PER_LONG;
 464         }
 465
 466         while (size & ~(BITS_PER_LONG - 1)) {
 467                 if (~(tmp = *(p++)))
 468                         goto found_middle_swap;
 469                 result += BITS_PER_LONG;
 470                 size -= BITS_PER_LONG;
 471         }
 472         if (!size)
 473                 return result;
 474         tmp = ext2_swabp(p);
 475 found_first:
 476         tmp |= ~0UL << size;
 477         if (tmp == ~0UL)        /* Are any bits zero? */
 478                 return result + size; /* Nope. Skip ffz */
 479 found_middle:
 480         return result + ffz(tmp);
 481
 482 found_middle_swap:
 483         return result + ffz(ext2_swab(tmp));
 484 }
 485
 486
 487 /* Bitmap functions for the minix filesystem.  */
 488 #define minix_test_and_set_bit(nr,addr) ext2_set_bit(nr,addr)
 489 #define minix_set_bit(nr,addr) ((void)ext2_set_bit(nr,addr))
 490 #define minix_test_and_clear_bit(nr,addr) ext2_clear_bit(nr,addr)
 491 #define minix_test_bit(nr,addr) ext2_test_bit(nr,addr)
 492 #define minix_find_first_zero_bit(addr,size) ext2_find_first_zero_bit(addr,size)
 493
 494 #endif /* _PARISC_BITOPS_H */