include/asm-sparc/bitops.h

   1 /* $Id: bitops.h,v 1.67 2001/11/19 18:36:34 davem Exp $
   2  * bitops.h: Bit string operations on the Sparc.
   3  *
   4  * Copyright 1995 David S. Miller (davem@caip.rutgers.edu)
   5  * Copyright 1996 Eddie C. Dost   (ecd@skynet.be)
   6  * Copyright 2001 Anton Blanchard (anton@samba.org)
   7  */
   8
   9 #ifndef _SPARC_BITOPS_H
  10 #define _SPARC_BITOPS_H
  11
  12 #include <linux/compiler.h>
  13 #include <asm/byteorder.h>
  14
  15 #ifdef __KERNEL__
  16
  17 /*
  18  * Set bit 'nr' in 32-bit quantity at address 'addr' where bit '0'
  19  * is in the highest of the four bytes and bit '31' is the high bit
  20  * within the first byte. Sparc is BIG-Endian. Unless noted otherwise
  21  * all bit-ops return 0 if bit was previously clear and != 0 otherwise.
  22  */
  23 static __inline__ int test_and_set_bit(unsigned long nr, volatile unsigned long *addr)
  24 {
  25         register unsigned long mask asm("g2");
  26         register unsigned long *ADDR asm("g1");
  27
  28         ADDR = ((unsigned long *) addr) + (nr >> 5);
  29         mask = 1 << (nr & 31);
  30
  31         __asm__ __volatile__(
  32         "mov    %%o7, %%g4\n\t"
  33         "call   ___set_bit\n\t"
  34         " add   %%o7, 8, %%o7\n"
  35         : "=&r" (mask)
  36         : "0" (mask), "r" (ADDR)
  37         : "g3", "g4", "g5", "g7", "memory", "cc");
  38
  39         return mask != 0;
  40 }
  41
  42 static __inline__ void set_bit(unsigned long nr, volatile unsigned long *addr)
  43 {
  44         register unsigned long mask asm("g2");
  45         register unsigned long *ADDR asm("g1");
  46
  47         ADDR = ((unsigned long *) addr) + (nr >> 5);
  48         mask = 1 << (nr & 31);
  49
  50         __asm__ __volatile__(
  51         "mov    %%o7, %%g4\n\t"
  52         "call   ___set_bit\n\t"
  53         " add   %%o7, 8, %%o7\n"
  54         : "=&r" (mask)
  55         : "0" (mask), "r" (ADDR)
  56         : "g3", "g4", "g5", "g7", "cc");
  57 }
  58
  59 static __inline__ int test_and_clear_bit(unsigned long nr, volatile unsigned long *addr)
  60 {
  61         register unsigned long mask asm("g2");
  62         register unsigned long *ADDR asm("g1");
  63
  64         ADDR = ((unsigned long *) addr) + (nr >> 5);
  65         mask = 1 << (nr & 31);
  66
  67         __asm__ __volatile__(
  68         "mov    %%o7, %%g4\n\t"
  69         "call   ___clear_bit\n\t"
  70         " add   %%o7, 8, %%o7\n"
  71         : "=&r" (mask)
  72         : "0" (mask), "r" (ADDR)
  73         : "g3", "g4", "g5", "g7", "memory", "cc");
  74
  75         return mask != 0;
  76 }
  77
  78 static __inline__ void clear_bit(unsigned long nr, volatile unsigned long *addr)
  79 {
  80         register unsigned long mask asm("g2");
  81         register unsigned long *ADDR asm("g1");
  82
  83         ADDR = ((unsigned long *) addr) + (nr >> 5);
  84         mask = 1 << (nr & 31);
  85
  86         __asm__ __volatile__(
  87         "mov    %%o7, %%g4\n\t"
  88         "call   ___clear_bit\n\t"
  89         " add   %%o7, 8, %%o7\n"
  90         : "=&r" (mask)
  91         : "0" (mask), "r" (ADDR)
  92         : "g3", "g4", "g5", "g7", "cc");
  93 }
  94
  95 static __inline__ int test_and_change_bit(unsigned long nr, volatile unsigned long *addr)
  96 {
  97         register unsigned long mask asm("g2");
  98         register unsigned long *ADDR asm("g1");
  99
 100         ADDR = ((unsigned long *) addr) + (nr >> 5);
 101         mask = 1 << (nr & 31);
 102
 103         __asm__ __volatile__(
 104         "mov    %%o7, %%g4\n\t"
 105         "call   ___change_bit\n\t"
 106         " add   %%o7, 8, %%o7\n"
 107         : "=&r" (mask)
 108         : "0" (mask), "r" (ADDR)
 109         : "g3", "g4", "g5", "g7", "memory", "cc");
 110
 111         return mask != 0;
 112 }
 113
 114 static __inline__ void change_bit(unsigned long nr, volatile unsigned long *addr)
 115 {
 116         register unsigned long mask asm("g2");
 117         register unsigned long *ADDR asm("g1");
 118
 119         ADDR = ((unsigned long *) addr) + (nr >> 5);
 120         mask = 1 << (nr & 31);
 121
 122         __asm__ __volatile__(
 123         "mov    %%o7, %%g4\n\t"
 124         "call   ___change_bit\n\t"
 125         " add   %%o7, 8, %%o7\n"
 126         : "=&r" (mask)
 127         : "0" (mask), "r" (ADDR)
 128         : "g3", "g4", "g5", "g7", "cc");
 129 }
 130
 131 /*
 132  * non-atomic versions
 133  */
 134 static __inline__ void __set_bit(int nr, volatile unsigned long *addr)
 135 {
 136         unsigned long mask = 1UL << (nr & 0x1f);
 137         unsigned long *p = ((unsigned long *)addr) + (nr >> 5);
 138
 139         *p |= mask;
 140 }
 141
 142 static __inline__ void __clear_bit(int nr, volatile unsigned long *addr)
 143 {
 144         unsigned long mask = 1UL << (nr & 0x1f);
 145         unsigned long *p = ((unsigned long *)addr) + (nr >> 5);
 146
 147         *p &= ~mask;
 148 }
 149
 150 static __inline__ void __change_bit(int nr, volatile unsigned long *addr)
 151 {
 152         unsigned long mask = 1UL << (nr & 0x1f);
 153         unsigned long *p = ((unsigned long *)addr) + (nr >> 5);
 154
 155         *p ^= mask;
 156 }
 157
 158 static __inline__ int __test_and_set_bit(int nr, volatile unsigned long *addr)
 159 {
 160         unsigned long mask = 1UL << (nr & 0x1f);
 161         unsigned long *p = ((unsigned long *)addr) + (nr >> 5);
 162         unsigned long old = *p;
 163
 164         *p = old | mask;
 165         return (old & mask) != 0;
 166 }
 167
 168 static __inline__ int __test_and_clear_bit(int nr, volatile unsigned long *addr)
 169 {
 170         unsigned long mask = 1UL << (nr & 0x1f);
 171         unsigned long *p = ((unsigned long *)addr) + (nr >> 5);
 172         unsigned long old = *p;
 173
 174         *p = old & ~mask;
 175         return (old & mask) != 0;
 176 }
 177
 178 static __inline__ int __test_and_change_bit(int nr, volatile unsigned long *addr)
 179 {
 180         unsigned long mask = 1UL << (nr & 0x1f);
 181         unsigned long *p = ((unsigned long *)addr) + (nr >> 5);
 182         unsigned long old = *p;
 183
 184         *p = old ^ mask;
 185         return (old & mask) != 0;
 186 }
 187
 188 #define smp_mb__before_clear_bit()      do { } while(0)
 189 #define smp_mb__after_clear_bit()       do { } while(0)
 190
 191 /* The following routine need not be atomic. */
 192 static __inline__ int test_bit(int nr, __const__ volatile unsigned long *addr)
 193 {
 194         return (1UL & (((unsigned long *)addr)[nr >> 5] >> (nr & 31))) != 0UL;
 195 }
 196
 197 /* The easy/cheese version for now. */
 198 static __inline__ unsigned long ffz(unsigned long word)
 199 {
 200         unsigned long result = 0;
 201
 202         while(word & 1) {
 203                 result++;
 204                 word >>= 1;
 205         }
 206         return result;
 207 }
 208
 209 /**
 210  * __ffs - find first bit in word.
 211  * @word: The word to search
 212  *
 213  * Undefined if no bit exists, so code should check against 0 first.
 214  */
 215 static __inline__ int __ffs(unsigned long word)
 216 {
 217         int num = 0;
 218
 219         if ((word & 0xffff) == 0) {
 220                 num += 16;
 221                 word >>= 16;
 222         }
 223         if ((word & 0xff) == 0) {
 224                 num += 8;
 225                 word >>= 8;
 226         }
 227         if ((word & 0xf) == 0) {
 228                 num += 4;
 229                 word >>= 4;
 230         }
 231         if ((word & 0x3) == 0) {
 232                 num += 2;
 233                 word >>= 2;
 234         }
 235         if ((word & 0x1) == 0)
 236                 num += 1;
 237         return num;
 238 }
 239
 240 /*
 241  * Every architecture must define this function. It's the fastest
 242  * way of searching a 140-bit bitmap where the first 100 bits are
 243  * unlikely to be set. It's guaranteed that at least one of the 140
 244  * bits is cleared.
 245  */
 246 static __inline__ int sched_find_first_bit(unsigned long *b)
 247 {
 248
 249         if (unlikely(b[0]))
 250                 return __ffs(b[0]);
 251         if (unlikely(b[1]))
 252                 return __ffs(b[1]) + 32;
 253         if (unlikely(b[2]))
 254                 return __ffs(b[2]) + 64;
 255         if (b[3])
 256                 return __ffs(b[3]) + 96;
 257         return __ffs(b[4]) + 128;
 258 }
 259
 260 /*
 261  * ffs: find first bit set. This is defined the same way as
 262  * the libc and compiler builtin ffs routines, therefore
 263  * differs in spirit from the above ffz (man ffs).
 264  */
 265 static __inline__ int ffs(int x)
 266 {
 267         if (!x)
 268                 return 0;
 269         return __ffs((unsigned long)x) + 1;
 270 }
 271
 272 /*
 273  * fls: find last (most-significant) bit set.
 274  * Note fls(0) = 0, fls(1) = 1, fls(0x80000000) = 32.
 275  */
 276 #define fls(x) generic_fls(x)
 277
 278 /*
 279  * hweightN: returns the hamming weight (i.e. the number
 280  * of bits set) of a N-bit word
 281  */
 282 #define hweight32(x) generic_hweight32(x)
 283 #define hweight16(x) generic_hweight16(x)
 284 #define hweight8(x) generic_hweight8(x)
 285
 286 /*
 287  * find_next_zero_bit() finds the first zero bit in a bit string of length
 288  * 'size' bits, starting the search at bit 'offset'. This is largely based
 289  * on Linus's ALPHA routines, which are pretty portable BTW.
 290  */
 291 static __inline__ unsigned long find_next_zero_bit(unsigned long *addr,
 292     unsigned long size, unsigned long offset)
 293 {
 294         unsigned long *p = addr + (offset >> 5);
 295         unsigned long result = offset & ~31UL;
 296         unsigned long tmp;
 297
 298         if (offset >= size)
 299                 return size;
 300         size -= result;
 301         offset &= 31UL;
 302         if (offset) {
 303                 tmp = *(p++);
 304                 tmp |= ~0UL >> (32-offset);
 305                 if (size < 32)
 306                         goto found_first;
 307                 if (~tmp)
 308                         goto found_middle;
 309                 size -= 32;
 310                 result += 32;
 311         }
 312         while (size & ~31UL) {
 313                 if (~(tmp = *(p++)))
 314                         goto found_middle;
 315                 result += 32;
 316                 size -= 32;
 317         }
 318         if (!size)
 319                 return result;
 320         tmp = *p;
 321
 322 found_first:
 323         tmp |= ~0UL << size;
 324         if (tmp == ~0UL)        /* Are any bits zero? */
 325                 return result + size; /* Nope. */
 326 found_middle:
 327         return result + ffz(tmp);
 328 }
 329
 330 /*
 331  * Linus sez that gcc can optimize the following correctly, we'll see if this
 332  * holds on the Sparc as it does for the ALPHA.
 333  */
 334 #define find_first_zero_bit(addr, size) \
 335         find_next_zero_bit((addr), (size), 0)
 336
 337 /**
 338  * find_next_bit - find the first set bit in a memory region
 339  * @addr: The address to base the search on
 340  * @offset: The bitnumber to start searching at
 341  * @size: The maximum size to search
 342  *
 343  * Scheduler induced bitop, do not use.
 344  */
 345 static __inline__ int find_next_bit(unsigned long *addr, int size, int offset)
 346 {
 347         unsigned long *p = addr + (offset >> 5);
 348         int num = offset & ~0x1f;
 349         unsigned long word;
 350
 351         word = *p++;
 352         word &= ~((1 << (offset & 0x1f)) - 1);
 353         while (num < size) {
 354                 if (word != 0) {
 355                         return __ffs(word) + num;
 356                 }
 357                 word = *p++;
 358                 num += 0x20;
 359         }
 360         return num;
 361 }
 362
 363 /*
 364  */
 365 static __inline__ int test_le_bit(int nr, __const__ unsigned long * addr)
 366 {
 367         __const__ unsigned char *ADDR = (__const__ unsigned char *) addr;
 368         return (ADDR[nr >> 3] >> (nr & 7)) & 1;
 369 }
 370
 371 /*
 372  * non-atomic versions
 373  */
 374 static __inline__ void __set_le_bit(int nr, unsigned long *addr)
 375 {
 376         unsigned char *ADDR = (unsigned char *)addr;
 377
 378         ADDR += nr >> 3;
 379         *ADDR |= 1 << (nr & 0x07);
 380 }
 381
 382 static __inline__ void __clear_le_bit(int nr, unsigned long *addr)
 383 {
 384         unsigned char *ADDR = (unsigned char *)addr;
 385
 386         ADDR += nr >> 3;
 387         *ADDR &= ~(1 << (nr & 0x07));
 388 }
 389
 390 static __inline__ int __test_and_set_le_bit(int nr, unsigned long *addr)
 391 {
 392         int mask, retval;
 393         unsigned char *ADDR = (unsigned char *)addr;
 394
 395         ADDR += nr >> 3;
 396         mask = 1 << (nr & 0x07);
 397         retval = (mask & *ADDR) != 0;
 398         *ADDR |= mask;
 399         return retval;
 400 }
 401
 402 static __inline__ int __test_and_clear_le_bit(int nr, unsigned long *addr)
 403 {
 404         int mask, retval;
 405         unsigned char *ADDR = (unsigned char *)addr;
 406
 407         ADDR += nr >> 3;
 408         mask = 1 << (nr & 0x07);
 409         retval = (mask & *ADDR) != 0;
 410         *ADDR &= ~mask;
 411         return retval;
 412 }
 413
 414 static __inline__ unsigned long find_next_zero_le_bit(unsigned long *addr,
 415     unsigned long size, unsigned long offset)
 416 {
 417         unsigned long *p = addr + (offset >> 5);
 418         unsigned long result = offset & ~31UL;
 419         unsigned long tmp;
 420
 421         if (offset >= size)
 422                 return size;
 423         size -= result;
 424         offset &= 31UL;
 425         if(offset) {
 426                 tmp = *(p++);
 427                 tmp |= __swab32(~0UL >> (32-offset));
 428                 if(size < 32)
 429                         goto found_first;
 430                 if(~tmp)
 431                         goto found_middle;
 432                 size -= 32;
 433                 result += 32;
 434         }
 435         while(size & ~31UL) {
 436                 if(~(tmp = *(p++)))
 437                         goto found_middle;
 438                 result += 32;
 439                 size -= 32;
 440         }
 441         if(!size)
 442                 return result;
 443         tmp = *p;
 444
 445 found_first:
 446         tmp = __swab32(tmp) | (~0UL << size);
 447         if (tmp == ~0UL)        /* Are any bits zero? */
 448                 return result + size; /* Nope. */
 449         return result + ffz(tmp);
 450
 451 found_middle:
 452         return result + ffz(__swab32(tmp));
 453 }
 454
 455 #define find_first_zero_le_bit(addr, size) \
 456         find_next_zero_le_bit((addr), (size), 0)
 457
 458 #define ext2_set_bit                    __test_and_set_le_bit
 459 #define ext2_clear_bit                  __test_and_clear_le_bit
 460
 461 #define ext2_set_bit_atomic(lock, nr, addr)             \
 462         ({                                              \
 463                 int ret;                                \
 464                 spin_lock(lock);                        \
 465                 ret = ext2_set_bit((nr), (addr));       \
 466                 spin_unlock(lock);                      \
 467                 ret;                                    \
 468         })
 469
 470 #define ext2_clear_bit_atomic(lock, nr, addr)           \
 471         ({                                              \
 472                 int ret;                                \
 473                 spin_lock(lock);                        \
 474                 ret = ext2_clear_bit((nr), (addr));     \
 475                 spin_unlock(lock);                      \
 476                 ret;                                    \
 477         })
 478
 479 #define ext2_test_bit                   test_le_bit
 480 #define ext2_find_first_zero_bit        find_first_zero_le_bit
 481 #define ext2_find_next_zero_bit         find_next_zero_le_bit
 482
 483 /* Bitmap functions for the minix filesystem.  */
 484 #define minix_test_and_set_bit(nr,addr)         test_and_set_bit(nr,addr)
 485 #define minix_set_bit(nr,addr)                  set_bit(nr,addr)
 486 #define minix_test_and_clear_bit(nr,addr)       test_and_clear_bit(nr,addr)
 487 #define minix_test_bit(nr,addr)                 test_bit(nr,addr)
 488 #define minix_find_first_zero_bit(addr,size)    find_first_zero_bit(addr,size)
 489
 490 #endif /* __KERNEL__ */
 491
 492 #endif /* defined(_SPARC_BITOPS_H) */