include/asm-powerpc/bitops.h

   1 /*
   2  * PowerPC atomic bit operations.
   3  *
   4  * Merged version by David Gibson <david@gibson.dropbear.id.au>.
   5  * Based on ppc64 versions by: Dave Engebretsen, Todd Inglett, Don
   6  * Reed, Pat McCarthy, Peter Bergner, Anton Blanchard.  They
   7  * originally took it from the ppc32 code.
   8  *
   9  * Within a word, bits are numbered LSB first.  Lot's of places make
  10  * this assumption by directly testing bits with (val & (1<<nr)).
  11  * This can cause confusion for large (> 1 word) bitmaps on a
  12  * big-endian system because, unlike little endian, the number of each
  13  * bit depends on the word size.
  14  *
  15  * The bitop functions are defined to work on unsigned longs, so for a
  16  * ppc64 system the bits end up numbered:
  17  *   |63..............0|127............64|191...........128|255...........196|
  18  * and on ppc32:
  19  *   |31.....0|63....31|95....64|127...96|159..128|191..160|223..192|255..224|
  20  *
  21  * There are a few little-endian macros used mostly for filesystem
  22  * bitmaps, these work on similar bit arrays layouts, but
  23  * byte-oriented:
  24  *   |7...0|15...8|23...16|31...24|39...32|47...40|55...48|63...56|
  25  *
  26  * The main difference is that bit 3-5 (64b) or 3-4 (32b) in the bit
  27  * number field needs to be reversed compared to the big-endian bit
  28  * fields. This can be achieved by XOR with 0x38 (64b) or 0x18 (32b).
  29  *
  30  * This program is free software; you can redistribute it and/or
  31  * modify it under the terms of the GNU General Public License
  32  * as published by the Free Software Foundation; either version
  33  * 2 of the License, or (at your option) any later version.
  34  */
  35
  36 #ifndef _ASM_POWERPC_BITOPS_H
  37 #define _ASM_POWERPC_BITOPS_H
  38
  39 #ifdef __KERNEL__
  40
  41 #include <linux/compiler.h>
  42 #include <asm/atomic.h>
  43 #include <asm/asm-compat.h>
  44 #include <asm/synch.h>
  45
  46 /*
  47  * clear_bit doesn't imply a memory barrier
  48  */
  49 #define smp_mb__before_clear_bit()      smp_mb()
  50 #define smp_mb__after_clear_bit()       smp_mb()
  51
  52 #define BITOP_MASK(nr)          (1UL << ((nr) % BITS_PER_LONG))
  53 #define BITOP_WORD(nr)          ((nr) / BITS_PER_LONG)
  54 #define BITOP_LE_SWIZZLE        ((BITS_PER_LONG-1) & ~0x7)
  55
  56 static __inline__ void set_bit(int nr, volatile unsigned long *addr)
  57 {
  58         unsigned long old;
  59         unsigned long mask = BITOP_MASK(nr);
  60         unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr);
  61
  62         __asm__ __volatile__(
  63 "1:"    PPC_LLARX "%0,0,%3      # set_bit\n"
  64         "or     %0,%0,%2\n"
  65         PPC405_ERR77(0,%3)
  66         PPC_STLCX "%0,0,%3\n"
  67         "bne-   1b"
  68         : "=&r"(old), "=m"(*p)
  69         : "r"(mask), "r"(p), "m"(*p)
  70         : "cc" );
  71 }
  72
  73 static __inline__ void clear_bit(int nr, volatile unsigned long *addr)
  74 {
  75         unsigned long old;
  76         unsigned long mask = BITOP_MASK(nr);
  77         unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr);
  78
  79         __asm__ __volatile__(
  80 "1:"    PPC_LLARX "%0,0,%3      # clear_bit\n"
  81         "andc   %0,%0,%2\n"
  82         PPC405_ERR77(0,%3)
  83         PPC_STLCX "%0,0,%3\n"
  84         "bne-   1b"
  85         : "=&r"(old), "=m"(*p)
  86         : "r"(mask), "r"(p), "m"(*p)
  87         : "cc" );
  88 }
  89
  90 static __inline__ void change_bit(int nr, volatile unsigned long *addr)
  91 {
  92         unsigned long old;
  93         unsigned long mask = BITOP_MASK(nr);
  94         unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr);
  95
  96         __asm__ __volatile__(
  97 "1:"    PPC_LLARX "%0,0,%3      # change_bit\n"
  98         "xor    %0,%0,%2\n"
  99         PPC405_ERR77(0,%3)
 100         PPC_STLCX "%0,0,%3\n"
 101         "bne-   1b"
 102         : "=&r"(old), "=m"(*p)
 103         : "r"(mask), "r"(p), "m"(*p)
 104         : "cc" );
 105 }
 106
 107 static __inline__ int test_and_set_bit(unsigned long nr,
 108                                        volatile unsigned long *addr)
 109 {
 110         unsigned long old, t;
 111         unsigned long mask = BITOP_MASK(nr);
 112         unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr);
 113
 114         __asm__ __volatile__(
 115         EIEIO_ON_SMP
 116 "1:"    PPC_LLARX "%0,0,%3              # test_and_set_bit\n"
 117         "or     %1,%0,%2 \n"
 118         PPC405_ERR77(0,%3)
 119         PPC_STLCX "%1,0,%3 \n"
 120         "bne-   1b"
 121         ISYNC_ON_SMP
 122         : "=&r" (old), "=&r" (t)
 123         : "r" (mask), "r" (p)
 124         : "cc", "memory");
 125
 126         return (old & mask) != 0;
 127 }
 128
 129 static __inline__ int test_and_clear_bit(unsigned long nr,
 130                                          volatile unsigned long *addr)
 131 {
 132         unsigned long old, t;
 133         unsigned long mask = BITOP_MASK(nr);
 134         unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr);
 135
 136         __asm__ __volatile__(
 137         EIEIO_ON_SMP
 138 "1:"    PPC_LLARX "%0,0,%3              # test_and_clear_bit\n"
 139         "andc   %1,%0,%2 \n"
 140         PPC405_ERR77(0,%3)
 141         PPC_STLCX "%1,0,%3 \n"
 142         "bne-   1b"
 143         ISYNC_ON_SMP
 144         : "=&r" (old), "=&r" (t)
 145         : "r" (mask), "r" (p)
 146         : "cc", "memory");
 147
 148         return (old & mask) != 0;
 149 }
 150
 151 static __inline__ int test_and_change_bit(unsigned long nr,
 152                                           volatile unsigned long *addr)
 153 {
 154         unsigned long old, t;
 155         unsigned long mask = BITOP_MASK(nr);
 156         unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr);
 157
 158         __asm__ __volatile__(
 159         EIEIO_ON_SMP
 160 "1:"    PPC_LLARX "%0,0,%3              # test_and_change_bit\n"
 161         "xor    %1,%0,%2 \n"
 162         PPC405_ERR77(0,%3)
 163         PPC_STLCX "%1,0,%3 \n"
 164         "bne-   1b"
 165         ISYNC_ON_SMP
 166         : "=&r" (old), "=&r" (t)
 167         : "r" (mask), "r" (p)
 168         : "cc", "memory");
 169
 170         return (old & mask) != 0;
 171 }
 172
 173 static __inline__ void set_bits(unsigned long mask, unsigned long *addr)
 174 {
 175         unsigned long old;
 176
 177         __asm__ __volatile__(
 178 "1:"    PPC_LLARX "%0,0,%3         # set_bits\n"
 179         "or     %0,%0,%2\n"
 180         PPC_STLCX "%0,0,%3\n"
 181         "bne-   1b"
 182         : "=&r" (old), "=m" (*addr)
 183         : "r" (mask), "r" (addr), "m" (*addr)
 184         : "cc");
 185 }
 186
 187 /* Non-atomic versions */
 188 static __inline__ int test_bit(unsigned long nr,
 189                                __const__ volatile unsigned long *addr)
 190 {
 191         return 1UL & (addr[BITOP_WORD(nr)] >> (nr & (BITS_PER_LONG-1)));
 192 }
 193
 194 static __inline__ void __set_bit(unsigned long nr,
 195                                  volatile unsigned long *addr)
 196 {
 197         unsigned long mask = BITOP_MASK(nr);
 198         unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr);
 199
 200         *p  |= mask;
 201 }
 202
 203 static __inline__ void __clear_bit(unsigned long nr,
 204                                    volatile unsigned long *addr)
 205 {
 206         unsigned long mask = BITOP_MASK(nr);
 207         unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr);
 208
 209         *p &= ~mask;
 210 }
 211
 212 static __inline__ void __change_bit(unsigned long nr,
 213                                     volatile unsigned long *addr)
 214 {
 215         unsigned long mask = BITOP_MASK(nr);
 216         unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr);
 217
 218         *p ^= mask;
 219 }
 220
 221 static __inline__ int __test_and_set_bit(unsigned long nr,
 222                                          volatile unsigned long *addr)
 223 {
 224         unsigned long mask = BITOP_MASK(nr);
 225         unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr);
 226         unsigned long old = *p;
 227
 228         *p = old | mask;
 229         return (old & mask) != 0;
 230 }
 231
 232 static __inline__ int __test_and_clear_bit(unsigned long nr,
 233                                            volatile unsigned long *addr)
 234 {
 235         unsigned long mask = BITOP_MASK(nr);
 236         unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr);
 237         unsigned long old = *p;
 238
 239         *p = old & ~mask;
 240         return (old & mask) != 0;
 241 }
 242
 243 static __inline__ int __test_and_change_bit(unsigned long nr,
 244                                             volatile unsigned long *addr)
 245 {
 246         unsigned long mask = BITOP_MASK(nr);
 247         unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr);
 248         unsigned long old = *p;
 249
 250         *p = old ^ mask;
 251         return (old & mask) != 0;
 252 }
 253
 254 /*
 255  * Return the zero-based bit position (LE, not IBM bit numbering) of
 256  * the most significant 1-bit in a double word.
 257  */
 258 static __inline__ int __ilog2(unsigned long x)
 259 {
 260         int lz;
 261
 262         asm (PPC_CNTLZL "%0,%1" : "=r" (lz) : "r" (x));
 263         return BITS_PER_LONG - 1 - lz;
 264 }
 265
 266 /*
 267  * Determines the bit position of the least significant 0 bit in the
 268  * specified double word. The returned bit position will be
 269  * zero-based, starting from the right side (63/31 - 0).
 270  */
 271 static __inline__ unsigned long ffz(unsigned long x)
 272 {
 273         /* no zero exists anywhere in the 8 byte area. */
 274         if ((x = ~x) == 0)
 275                 return BITS_PER_LONG;
 276
 277         /*
 278          * Calculate the bit position of the least signficant '1' bit in x
 279          * (since x has been changed this will actually be the least signficant
 280          * '0' bit in * the original x).  Note: (x & -x) gives us a mask that
 281          * is the least significant * (RIGHT-most) 1-bit of the value in x.
 282          */
 283         return __ilog2(x & -x);
 284 }
 285
 286 static __inline__ int __ffs(unsigned long x)
 287 {
 288         return __ilog2(x & -x);
 289 }
 290
 291 /*
 292  * ffs: find first bit set. This is defined the same way as
 293  * the libc and compiler builtin ffs routines, therefore
 294  * differs in spirit from the above ffz (man ffs).
 295  */
 296 static __inline__ int ffs(int x)
 297 {
 298         unsigned long i = (unsigned long)x;
 299         return __ilog2(i & -i) + 1;
 300 }
 301
 302 /*
 303  * fls: find last (most-significant) bit set.
 304  * Note fls(0) = 0, fls(1) = 1, fls(0x80000000) = 32.
 305  */
 306 static __inline__ int fls(unsigned int x)
 307 {
 308         int lz;
 309
 310         asm ("cntlzw %0,%1" : "=r" (lz) : "r" (x));
 311         return 32 - lz;
 312 }
 313
 314 /*
 315  * hweightN: returns the hamming weight (i.e. the number
 316  * of bits set) of a N-bit word
 317  */
 318 #define hweight64(x) generic_hweight64(x)
 319 #define hweight32(x) generic_hweight32(x)
 320 #define hweight16(x) generic_hweight16(x)
 321 #define hweight8(x) generic_hweight8(x)
 322
 323 #define find_first_zero_bit(addr, size) find_next_zero_bit((addr), (size), 0)
 324 unsigned long find_next_zero_bit(const unsigned long *addr,
 325                                  unsigned long size, unsigned long offset);
 326 /**
 327  * find_first_bit - find the first set bit in a memory region
 328  * @addr: The address to start the search at
 329  * @size: The maximum size to search
 330  *
 331  * Returns the bit-number of the first set bit, not the number of the byte
 332  * containing a bit.
 333  */
 334 #define find_first_bit(addr, size) find_next_bit((addr), (size), 0)
 335 unsigned long find_next_bit(const unsigned long *addr,
 336                             unsigned long size, unsigned long offset);
 337
 338 /* Little-endian versions */
 339
 340 static __inline__ int test_le_bit(unsigned long nr,
 341                                   __const__ unsigned long *addr)
 342 {
 343         __const__ unsigned char *tmp = (__const__ unsigned char *) addr;
 344         return (tmp[nr >> 3] >> (nr & 7)) & 1;
 345 }
 346
 347 #define __set_le_bit(nr, addr) \
 348         __set_bit((nr) ^ BITOP_LE_SWIZZLE, (addr))
 349 #define __clear_le_bit(nr, addr) \
 350         __clear_bit((nr) ^ BITOP_LE_SWIZZLE, (addr))
 351
 352 #define test_and_set_le_bit(nr, addr) \
 353         test_and_set_bit((nr) ^ BITOP_LE_SWIZZLE, (addr))
 354 #define test_and_clear_le_bit(nr, addr) \
 355         test_and_clear_bit((nr) ^ BITOP_LE_SWIZZLE, (addr))
 356
 357 #define __test_and_set_le_bit(nr, addr) \
 358         __test_and_set_bit((nr) ^ BITOP_LE_SWIZZLE, (addr))
 359 #define __test_and_clear_le_bit(nr, addr) \
 360         __test_and_clear_bit((nr) ^ BITOP_LE_SWIZZLE, (addr))
 361
 362 #define find_first_zero_le_bit(addr, size) find_next_zero_le_bit((addr), (size), 0)
 363 unsigned long find_next_zero_le_bit(const unsigned long *addr,
 364                                     unsigned long size, unsigned long offset);
 365
 366 /* Bitmap functions for the ext2 filesystem */
 367
 368 #define ext2_set_bit(nr,addr) \
 369         __test_and_set_le_bit((nr), (unsigned long*)addr)
 370 #define ext2_clear_bit(nr, addr) \
 371         __test_and_clear_le_bit((nr), (unsigned long*)addr)
 372
 373 #define ext2_set_bit_atomic(lock, nr, addr) \
 374         test_and_set_le_bit((nr), (unsigned long*)addr)
 375 #define ext2_clear_bit_atomic(lock, nr, addr) \
 376         test_and_clear_le_bit((nr), (unsigned long*)addr)
 377
 378 #define ext2_test_bit(nr, addr)      test_le_bit((nr),(unsigned long*)addr)
 379
 380 #define ext2_find_first_zero_bit(addr, size) \
 381         find_first_zero_le_bit((unsigned long*)addr, size)
 382 #define ext2_find_next_zero_bit(addr, size, off) \
 383         find_next_zero_le_bit((unsigned long*)addr, size, off)
 384
 385 /* Bitmap functions for the minix filesystem.  */
 386
 387 #define minix_test_and_set_bit(nr,addr) \
 388         __test_and_set_le_bit(nr, (unsigned long *)addr)
 389 #define minix_set_bit(nr,addr) \
 390         __set_le_bit(nr, (unsigned long *)addr)
 391 #define minix_test_and_clear_bit(nr,addr) \
 392         __test_and_clear_le_bit(nr, (unsigned long *)addr)
 393 #define minix_test_bit(nr,addr) \
 394         test_le_bit(nr, (unsigned long *)addr)
 395
 396 #define minix_find_first_zero_bit(addr,size) \
 397         find_first_zero_le_bit((unsigned long *)addr, size)
 398
 399 /*
 400  * Every architecture must define this function. It's the fastest
 401  * way of searching a 140-bit bitmap where the first 100 bits are
 402  * unlikely to be set. It's guaranteed that at least one of the 140
 403  * bits is cleared.
 404  */
 405 static inline int sched_find_first_bit(const unsigned long *b)
 406 {
 407 #ifdef CONFIG_PPC64
 408         if (unlikely(b[0]))
 409                 return __ffs(b[0]);
 410         if (unlikely(b[1]))
 411                 return __ffs(b[1]) + 64;
 412         return __ffs(b[2]) + 128;
 413 #else
 414         if (unlikely(b[0]))
 415                 return __ffs(b[0]);
 416         if (unlikely(b[1]))
 417                 return __ffs(b[1]) + 32;
 418         if (unlikely(b[2]))
 419                 return __ffs(b[2]) + 64;
 420         if (b[3])
 421                 return __ffs(b[3]) + 96;
 422         return __ffs(b[4]) + 128;
 423 #endif
 424 }
 425
 426 #endif /* __KERNEL__ */
 427
 428 #endif /* _ASM_POWERPC_BITOPS_H */