arch/powerpc/include/asm/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 #ifndef _LINUX_BITOPS_H
  42 #error only <linux/bitops.h> can be included directly
  43 #endif
  44
  45 #include <linux/compiler.h>
  46 #include <asm/asm-compat.h>
  47 #include <asm/synch.h>
  48
  49 /*
  50  * clear_bit doesn't imply a memory barrier
  51  */
  52 #define smp_mb__before_clear_bit()      smp_mb()
  53 #define smp_mb__after_clear_bit()       smp_mb()
  54
  55 #define BITOP_MASK(nr)          (1UL << ((nr) % BITS_PER_LONG))
  56 #define BITOP_WORD(nr)          ((nr) / BITS_PER_LONG)
  57 #define BITOP_LE_SWIZZLE        ((BITS_PER_LONG-1) & ~0x7)
  58
  59 /* Macro for generating the ***_bits() functions */
  60 #define DEFINE_BITOP(fn, op, prefix, postfix)   \
  61 static __inline__ void fn(unsigned long mask,   \
  62                 volatile unsigned long *_p)     \
  63 {                                               \
  64         unsigned long old;                      \
  65         unsigned long *p = (unsigned long *)_p; \
  66         __asm__ __volatile__ (                  \
  67         prefix                                  \
  68 "1:"    PPC_LLARX(%0,0,%3,0) "\n"               \
  69         stringify_in_c(op) "%0,%0,%2\n"         \
  70         PPC405_ERR77(0,%3)                      \
  71         PPC_STLCX "%0,0,%3\n"                   \
  72         "bne- 1b\n"                             \
  73         postfix                                 \
  74         : "=&r" (old), "+m" (*p)                \
  75         : "r" (mask), "r" (p)                   \
  76         : "cc", "memory");                      \
  77 }
  78
  79 DEFINE_BITOP(set_bits, or, "", "")
  80 DEFINE_BITOP(clear_bits, andc, "", "")
  81 DEFINE_BITOP(clear_bits_unlock, andc, PPC_RELEASE_BARRIER, "")
  82 DEFINE_BITOP(change_bits, xor, "", "")
  83
  84 static __inline__ void set_bit(int nr, volatile unsigned long *addr)
  85 {
  86         set_bits(BITOP_MASK(nr), addr + BITOP_WORD(nr));
  87 }
  88
  89 static __inline__ void clear_bit(int nr, volatile unsigned long *addr)
  90 {
  91         clear_bits(BITOP_MASK(nr), addr + BITOP_WORD(nr));
  92 }
  93
  94 static __inline__ void clear_bit_unlock(int nr, volatile unsigned long *addr)
  95 {
  96         clear_bits_unlock(BITOP_MASK(nr), addr + BITOP_WORD(nr));
  97 }
  98
  99 static __inline__ void change_bit(int nr, volatile unsigned long *addr)
 100 {
 101         change_bits(BITOP_MASK(nr), addr + BITOP_WORD(nr));
 102 }
 103
 104 /* Like DEFINE_BITOP(), with changes to the arguments to 'op' and the output
 105  * operands. */
 106 #define DEFINE_TESTOP(fn, op, prefix, postfix, eh)      \
 107 static __inline__ unsigned long fn(                     \
 108                 unsigned long mask,                     \
 109                 volatile unsigned long *_p)             \
 110 {                                                       \
 111         unsigned long old, t;                           \
 112         unsigned long *p = (unsigned long *)_p;         \
 113         __asm__ __volatile__ (                          \
 114         prefix                                          \
 115 "1:"    PPC_LLARX(%0,0,%3,eh) "\n"                      \
 116         stringify_in_c(op) "%1,%0,%2\n"                 \
 117         PPC405_ERR77(0,%3)                              \
 118         PPC_STLCX "%1,0,%3\n"                           \
 119         "bne- 1b\n"                                     \
 120         postfix                                         \
 121         : "=&r" (old), "=&r" (t)                        \
 122         : "r" (mask), "r" (p)                           \
 123         : "cc", "memory");                              \
 124         return (old & mask);                            \
 125 }
 126
 127 DEFINE_TESTOP(test_and_set_bits, or, PPC_RELEASE_BARRIER,
 128               PPC_ACQUIRE_BARRIER, 0)
 129 DEFINE_TESTOP(test_and_set_bits_lock, or, "",
 130               PPC_ACQUIRE_BARRIER, 1)
 131 DEFINE_TESTOP(test_and_clear_bits, andc, PPC_RELEASE_BARRIER,
 132               PPC_ACQUIRE_BARRIER, 0)
 133 DEFINE_TESTOP(test_and_change_bits, xor, PPC_RELEASE_BARRIER,
 134               PPC_ACQUIRE_BARRIER, 0)
 135
 136 static __inline__ int test_and_set_bit(unsigned long nr,
 137                                        volatile unsigned long *addr)
 138 {
 139         return test_and_set_bits(BITOP_MASK(nr), addr + BITOP_WORD(nr)) != 0;
 140 }
 141
 142 static __inline__ int test_and_set_bit_lock(unsigned long nr,
 143                                        volatile unsigned long *addr)
 144 {
 145         return test_and_set_bits_lock(BITOP_MASK(nr),
 146                                 addr + BITOP_WORD(nr)) != 0;
 147 }
 148
 149 static __inline__ int test_and_clear_bit(unsigned long nr,
 150                                          volatile unsigned long *addr)
 151 {
 152         return test_and_clear_bits(BITOP_MASK(nr), addr + BITOP_WORD(nr)) != 0;
 153 }
 154
 155 static __inline__ int test_and_change_bit(unsigned long nr,
 156                                           volatile unsigned long *addr)
 157 {
 158         return test_and_change_bits(BITOP_MASK(nr), addr + BITOP_WORD(nr)) != 0;
 159 }
 160
 161 #include <asm-generic/bitops/non-atomic.h>
 162
 163 static __inline__ void __clear_bit_unlock(int nr, volatile unsigned long *addr)
 164 {
 165         __asm__ __volatile__(PPC_RELEASE_BARRIER "" ::: "memory");
 166         __clear_bit(nr, addr);
 167 }
 168
 169 /*
 170  * Return the zero-based bit position (LE, not IBM bit numbering) of
 171  * the most significant 1-bit in a double word.
 172  */
 173 static __inline__ __attribute__((const))
 174 int __ilog2(unsigned long x)
 175 {
 176         int lz;
 177
 178         asm (PPC_CNTLZL "%0,%1" : "=r" (lz) : "r" (x));
 179         return BITS_PER_LONG - 1 - lz;
 180 }
 181
 182 static inline __attribute__((const))
 183 int __ilog2_u32(u32 n)
 184 {
 185         int bit;
 186         asm ("cntlzw %0,%1" : "=r" (bit) : "r" (n));
 187         return 31 - bit;
 188 }
 189
 190 #ifdef __powerpc64__
 191 static inline __attribute__((const))
 192 int __ilog2_u64(u64 n)
 193 {
 194         int bit;
 195         asm ("cntlzd %0,%1" : "=r" (bit) : "r" (n));
 196         return 63 - bit;
 197 }
 198 #endif
 199
 200 /*
 201  * Determines the bit position of the least significant 0 bit in the
 202  * specified double word. The returned bit position will be
 203  * zero-based, starting from the right side (63/31 - 0).
 204  */
 205 static __inline__ unsigned long ffz(unsigned long x)
 206 {
 207         /* no zero exists anywhere in the 8 byte area. */
 208         if ((x = ~x) == 0)
 209                 return BITS_PER_LONG;
 210
 211         /*
 212          * Calculate the bit position of the least signficant '1' bit in x
 213          * (since x has been changed this will actually be the least signficant
 214          * '0' bit in * the original x).  Note: (x & -x) gives us a mask that
 215          * is the least significant * (RIGHT-most) 1-bit of the value in x.
 216          */
 217         return __ilog2(x & -x);
 218 }
 219
 220 static __inline__ int __ffs(unsigned long x)
 221 {
 222         return __ilog2(x & -x);
 223 }
 224
 225 /*
 226  * ffs: find first bit set. This is defined the same way as
 227  * the libc and compiler builtin ffs routines, therefore
 228  * differs in spirit from the above ffz (man ffs).
 229  */
 230 static __inline__ int ffs(int x)
 231 {
 232         unsigned long i = (unsigned long)x;
 233         return __ilog2(i & -i) + 1;
 234 }
 235
 236 /*
 237  * fls: find last (most-significant) bit set.
 238  * Note fls(0) = 0, fls(1) = 1, fls(0x80000000) = 32.
 239  */
 240 static __inline__ int fls(unsigned int x)
 241 {
 242         int lz;
 243
 244         asm ("cntlzw %0,%1" : "=r" (lz) : "r" (x));
 245         return 32 - lz;
 246 }
 247
 248 static __inline__ unsigned long __fls(unsigned long x)
 249 {
 250         return __ilog2(x);
 251 }
 252
 253 /*
 254  * 64-bit can do this using one cntlzd (count leading zeroes doubleword)
 255  * instruction; for 32-bit we use the generic version, which does two
 256  * 32-bit fls calls.
 257  */
 258 #ifdef __powerpc64__
 259 static __inline__ int fls64(__u64 x)
 260 {
 261         int lz;
 262
 263         asm ("cntlzd %0,%1" : "=r" (lz) : "r" (x));
 264         return 64 - lz;
 265 }
 266 #else
 267 #include <asm-generic/bitops/fls64.h>
 268 #endif /* __powerpc64__ */
 269
 270 #include <asm-generic/bitops/hweight.h>
 271 #include <asm-generic/bitops/find.h>
 272
 273 /* Little-endian versions */
 274
 275 static __inline__ int test_le_bit(unsigned long nr,
 276                                   __const__ unsigned long *addr)
 277 {
 278         __const__ unsigned char *tmp = (__const__ unsigned char *) addr;
 279         return (tmp[nr >> 3] >> (nr & 7)) & 1;
 280 }
 281
 282 #define __set_le_bit(nr, addr) \
 283         __set_bit((nr) ^ BITOP_LE_SWIZZLE, (addr))
 284 #define __clear_le_bit(nr, addr) \
 285         __clear_bit((nr) ^ BITOP_LE_SWIZZLE, (addr))
 286
 287 #define test_and_set_le_bit(nr, addr) \
 288         test_and_set_bit((nr) ^ BITOP_LE_SWIZZLE, (addr))
 289 #define test_and_clear_le_bit(nr, addr) \
 290         test_and_clear_bit((nr) ^ BITOP_LE_SWIZZLE, (addr))
 291
 292 #define __test_and_set_le_bit(nr, addr) \
 293         __test_and_set_bit((nr) ^ BITOP_LE_SWIZZLE, (addr))
 294 #define __test_and_clear_le_bit(nr, addr) \
 295         __test_and_clear_bit((nr) ^ BITOP_LE_SWIZZLE, (addr))
 296
 297 #define find_first_zero_le_bit(addr, size) generic_find_next_zero_le_bit((addr), (size), 0)
 298 unsigned long generic_find_next_zero_le_bit(const unsigned long *addr,
 299                                     unsigned long size, unsigned long offset);
 300
 301 unsigned long generic_find_next_le_bit(const unsigned long *addr,
 302                                     unsigned long size, unsigned long offset);
 303 /* Bitmap functions for the ext2 filesystem */
 304
 305 #define ext2_set_bit(nr,addr) \
 306         __test_and_set_le_bit((nr), (unsigned long*)addr)
 307 #define ext2_clear_bit(nr, addr) \
 308         __test_and_clear_le_bit((nr), (unsigned long*)addr)
 309
 310 #define ext2_set_bit_atomic(lock, nr, addr) \
 311         test_and_set_le_bit((nr), (unsigned long*)addr)
 312 #define ext2_clear_bit_atomic(lock, nr, addr) \
 313         test_and_clear_le_bit((nr), (unsigned long*)addr)
 314
 315 #define ext2_test_bit(nr, addr)      test_le_bit((nr),(unsigned long*)addr)
 316
 317 #define ext2_find_first_zero_bit(addr, size) \
 318         find_first_zero_le_bit((unsigned long*)addr, size)
 319 #define ext2_find_next_zero_bit(addr, size, off) \
 320         generic_find_next_zero_le_bit((unsigned long*)addr, size, off)
 321
 322 #define ext2_find_next_bit(addr, size, off) \
 323         generic_find_next_le_bit((unsigned long *)addr, size, off)
 324 /* Bitmap functions for the minix filesystem.  */
 325
 326 #define minix_test_and_set_bit(nr,addr) \
 327         __test_and_set_le_bit(nr, (unsigned long *)addr)
 328 #define minix_set_bit(nr,addr) \
 329         __set_le_bit(nr, (unsigned long *)addr)
 330 #define minix_test_and_clear_bit(nr,addr) \
 331         __test_and_clear_le_bit(nr, (unsigned long *)addr)
 332 #define minix_test_bit(nr,addr) \
 333         test_le_bit(nr, (unsigned long *)addr)
 334
 335 #define minix_find_first_zero_bit(addr,size) \
 336         find_first_zero_le_bit((unsigned long *)addr, size)
 337
 338 #include <asm-generic/bitops/sched.h>
 339
 340 #endif /* __KERNEL__ */
 341
 342 #endif /* _ASM_POWERPC_BITOPS_H */