David Gibson | a0e60b2 | 2005-11-01 17:28:10 +1100 | [diff] [blame] | 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/synch.h> |
| 44 | |
| 45 | /* |
| 46 | * clear_bit doesn't imply a memory barrier |
| 47 | */ |
| 48 | #define smp_mb__before_clear_bit() smp_mb() |
| 49 | #define smp_mb__after_clear_bit() smp_mb() |
| 50 | |
| 51 | #define BITOP_MASK(nr) (1UL << ((nr) % BITS_PER_LONG)) |
| 52 | #define BITOP_WORD(nr) ((nr) / BITS_PER_LONG) |
| 53 | #define BITOP_LE_SWIZZLE ((BITS_PER_LONG-1) & ~0x7) |
| 54 | |
| 55 | #ifdef CONFIG_PPC64 |
| 56 | #define LARXL "ldarx" |
| 57 | #define STCXL "stdcx." |
| 58 | #define CNTLZL "cntlzd" |
| 59 | #else |
| 60 | #define LARXL "lwarx" |
| 61 | #define STCXL "stwcx." |
| 62 | #define CNTLZL "cntlzw" |
| 63 | #endif |
| 64 | |
| 65 | static __inline__ void set_bit(int nr, volatile unsigned long *addr) |
| 66 | { |
| 67 | unsigned long old; |
| 68 | unsigned long mask = BITOP_MASK(nr); |
| 69 | unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr); |
| 70 | |
| 71 | __asm__ __volatile__( |
| 72 | "1:" LARXL " %0,0,%3 # set_bit\n" |
| 73 | "or %0,%0,%2\n" |
| 74 | PPC405_ERR77(0,%3) |
| 75 | STCXL " %0,0,%3\n" |
| 76 | "bne- 1b" |
| 77 | : "=&r"(old), "=m"(*p) |
| 78 | : "r"(mask), "r"(p), "m"(*p) |
| 79 | : "cc" ); |
| 80 | } |
| 81 | |
| 82 | static __inline__ void clear_bit(int nr, volatile unsigned long *addr) |
| 83 | { |
| 84 | unsigned long old; |
| 85 | unsigned long mask = BITOP_MASK(nr); |
| 86 | unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr); |
| 87 | |
| 88 | __asm__ __volatile__( |
| 89 | "1:" LARXL " %0,0,%3 # set_bit\n" |
| 90 | "andc %0,%0,%2\n" |
| 91 | PPC405_ERR77(0,%3) |
| 92 | STCXL " %0,0,%3\n" |
| 93 | "bne- 1b" |
| 94 | : "=&r"(old), "=m"(*p) |
| 95 | : "r"(mask), "r"(p), "m"(*p) |
| 96 | : "cc" ); |
| 97 | } |
| 98 | |
| 99 | static __inline__ void change_bit(int nr, volatile unsigned long *addr) |
| 100 | { |
| 101 | unsigned long old; |
| 102 | unsigned long mask = BITOP_MASK(nr); |
| 103 | unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr); |
| 104 | |
| 105 | __asm__ __volatile__( |
| 106 | "1:" LARXL " %0,0,%3 # set_bit\n" |
| 107 | "xor %0,%0,%2\n" |
| 108 | PPC405_ERR77(0,%3) |
| 109 | STCXL " %0,0,%3\n" |
| 110 | "bne- 1b" |
| 111 | : "=&r"(old), "=m"(*p) |
| 112 | : "r"(mask), "r"(p), "m"(*p) |
| 113 | : "cc" ); |
| 114 | } |
| 115 | |
| 116 | static __inline__ int test_and_set_bit(unsigned long nr, |
| 117 | volatile unsigned long *addr) |
| 118 | { |
| 119 | unsigned long old, t; |
| 120 | unsigned long mask = BITOP_MASK(nr); |
| 121 | unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr); |
| 122 | |
| 123 | __asm__ __volatile__( |
| 124 | EIEIO_ON_SMP |
| 125 | "1:" LARXL " %0,0,%3 # test_and_set_bit\n" |
| 126 | "or %1,%0,%2 \n" |
| 127 | PPC405_ERR77(0,%3) |
| 128 | STCXL " %1,0,%3 \n" |
| 129 | "bne- 1b" |
| 130 | ISYNC_ON_SMP |
| 131 | : "=&r" (old), "=&r" (t) |
| 132 | : "r" (mask), "r" (p) |
| 133 | : "cc", "memory"); |
| 134 | |
| 135 | return (old & mask) != 0; |
| 136 | } |
| 137 | |
| 138 | static __inline__ int test_and_clear_bit(unsigned long nr, |
| 139 | volatile unsigned long *addr) |
| 140 | { |
| 141 | unsigned long old, t; |
| 142 | unsigned long mask = BITOP_MASK(nr); |
| 143 | unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr); |
| 144 | |
| 145 | __asm__ __volatile__( |
| 146 | EIEIO_ON_SMP |
| 147 | "1:" LARXL " %0,0,%3 # test_and_clear_bit\n" |
| 148 | "andc %1,%0,%2 \n" |
| 149 | PPC405_ERR77(0,%3) |
| 150 | STCXL " %1,0,%3 \n" |
| 151 | "bne- 1b" |
| 152 | ISYNC_ON_SMP |
| 153 | : "=&r" (old), "=&r" (t) |
| 154 | : "r" (mask), "r" (p) |
| 155 | : "cc", "memory"); |
| 156 | |
| 157 | return (old & mask) != 0; |
| 158 | } |
| 159 | |
| 160 | static __inline__ int test_and_change_bit(unsigned long nr, |
| 161 | volatile unsigned long *addr) |
| 162 | { |
| 163 | unsigned long old, t; |
| 164 | unsigned long mask = BITOP_MASK(nr); |
| 165 | unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr); |
| 166 | |
| 167 | __asm__ __volatile__( |
| 168 | EIEIO_ON_SMP |
| 169 | "1:" LARXL " %0,0,%3 # test_and_change_bit\n" |
| 170 | "xor %1,%0,%2 \n" |
| 171 | PPC405_ERR77(0,%3) |
| 172 | STCXL " %1,0,%3 \n" |
| 173 | "bne- 1b" |
| 174 | ISYNC_ON_SMP |
| 175 | : "=&r" (old), "=&r" (t) |
| 176 | : "r" (mask), "r" (p) |
| 177 | : "cc", "memory"); |
| 178 | |
| 179 | return (old & mask) != 0; |
| 180 | } |
| 181 | |
| 182 | static __inline__ void set_bits(unsigned long mask, unsigned long *addr) |
| 183 | { |
| 184 | unsigned long old; |
| 185 | |
| 186 | __asm__ __volatile__( |
| 187 | "1:" LARXL " %0,0,%3 # set_bit\n" |
| 188 | "or %0,%0,%2\n" |
| 189 | STCXL " %0,0,%3\n" |
| 190 | "bne- 1b" |
| 191 | : "=&r" (old), "=m" (*addr) |
| 192 | : "r" (mask), "r" (addr), "m" (*addr) |
| 193 | : "cc"); |
| 194 | } |
| 195 | |
| 196 | /* Non-atomic versions */ |
| 197 | static __inline__ int test_bit(unsigned long nr, |
| 198 | __const__ volatile unsigned long *addr) |
| 199 | { |
| 200 | return 1UL & (addr[BITOP_WORD(nr)] >> (nr & (BITS_PER_LONG-1))); |
| 201 | } |
| 202 | |
| 203 | static __inline__ void __set_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 __clear_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__ void __change_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 | |
| 227 | *p ^= mask; |
| 228 | } |
| 229 | |
| 230 | static __inline__ int __test_and_set_bit(unsigned long nr, |
| 231 | volatile unsigned long *addr) |
| 232 | { |
| 233 | unsigned long mask = BITOP_MASK(nr); |
| 234 | unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr); |
| 235 | unsigned long old = *p; |
| 236 | |
| 237 | *p = old | mask; |
| 238 | return (old & mask) != 0; |
| 239 | } |
| 240 | |
| 241 | static __inline__ int __test_and_clear_bit(unsigned long nr, |
| 242 | volatile unsigned long *addr) |
| 243 | { |
| 244 | unsigned long mask = BITOP_MASK(nr); |
| 245 | unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr); |
| 246 | unsigned long old = *p; |
| 247 | |
| 248 | *p = old & ~mask; |
| 249 | return (old & mask) != 0; |
| 250 | } |
| 251 | |
| 252 | static __inline__ int __test_and_change_bit(unsigned long nr, |
| 253 | volatile unsigned long *addr) |
| 254 | { |
| 255 | unsigned long mask = BITOP_MASK(nr); |
| 256 | unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr); |
| 257 | unsigned long old = *p; |
| 258 | |
| 259 | *p = old ^ mask; |
| 260 | return (old & mask) != 0; |
| 261 | } |
| 262 | |
| 263 | /* |
| 264 | * Return the zero-based bit position (LE, not IBM bit numbering) of |
| 265 | * the most significant 1-bit in a double word. |
| 266 | */ |
| 267 | static __inline__ int __ilog2(unsigned long x) |
| 268 | { |
| 269 | int lz; |
| 270 | |
| 271 | asm (CNTLZL " %0,%1" : "=r" (lz) : "r" (x)); |
| 272 | return BITS_PER_LONG - 1 - lz; |
| 273 | } |
| 274 | |
| 275 | /* |
| 276 | * Determines the bit position of the least significant 0 bit in the |
| 277 | * specified double word. The returned bit position will be |
| 278 | * zero-based, starting from the right side (63/31 - 0). |
| 279 | */ |
| 280 | static __inline__ unsigned long ffz(unsigned long x) |
| 281 | { |
| 282 | /* no zero exists anywhere in the 8 byte area. */ |
| 283 | if ((x = ~x) == 0) |
| 284 | return BITS_PER_LONG; |
| 285 | |
| 286 | /* |
| 287 | * Calculate the bit position of the least signficant '1' bit in x |
| 288 | * (since x has been changed this will actually be the least signficant |
| 289 | * '0' bit in * the original x). Note: (x & -x) gives us a mask that |
| 290 | * is the least significant * (RIGHT-most) 1-bit of the value in x. |
| 291 | */ |
| 292 | return __ilog2(x & -x); |
| 293 | } |
| 294 | |
| 295 | static __inline__ int __ffs(unsigned long x) |
| 296 | { |
| 297 | return __ilog2(x & -x); |
| 298 | } |
| 299 | |
| 300 | /* |
| 301 | * ffs: find first bit set. This is defined the same way as |
| 302 | * the libc and compiler builtin ffs routines, therefore |
| 303 | * differs in spirit from the above ffz (man ffs). |
| 304 | */ |
| 305 | static __inline__ int ffs(int x) |
| 306 | { |
| 307 | unsigned long i = (unsigned long)x; |
| 308 | return __ilog2(i & -i) + 1; |
| 309 | } |
| 310 | |
| 311 | /* |
| 312 | * fls: find last (most-significant) bit set. |
| 313 | * Note fls(0) = 0, fls(1) = 1, fls(0x80000000) = 32. |
| 314 | */ |
| 315 | static __inline__ int fls(unsigned int x) |
| 316 | { |
| 317 | int lz; |
| 318 | |
| 319 | asm ("cntlzw %0,%1" : "=r" (lz) : "r" (x)); |
| 320 | return 32 - lz; |
| 321 | } |
| 322 | |
| 323 | /* |
| 324 | * hweightN: returns the hamming weight (i.e. the number |
| 325 | * of bits set) of a N-bit word |
| 326 | */ |
| 327 | #define hweight64(x) generic_hweight64(x) |
| 328 | #define hweight32(x) generic_hweight32(x) |
| 329 | #define hweight16(x) generic_hweight16(x) |
| 330 | #define hweight8(x) generic_hweight8(x) |
| 331 | |
| 332 | #define find_first_zero_bit(addr, size) find_next_zero_bit((addr), (size), 0) |
| 333 | unsigned long find_next_zero_bit(const unsigned long *addr, |
| 334 | unsigned long size, unsigned long offset); |
| 335 | /** |
| 336 | * find_first_bit - find the first set bit in a memory region |
| 337 | * @addr: The address to start the search at |
| 338 | * @size: The maximum size to search |
| 339 | * |
| 340 | * Returns the bit-number of the first set bit, not the number of the byte |
| 341 | * containing a bit. |
| 342 | */ |
| 343 | #define find_first_bit(addr, size) find_next_bit((addr), (size), 0) |
| 344 | unsigned long find_next_bit(const unsigned long *addr, |
| 345 | unsigned long size, unsigned long offset); |
| 346 | |
| 347 | /* Little-endian versions */ |
| 348 | |
| 349 | static __inline__ int test_le_bit(unsigned long nr, |
| 350 | __const__ unsigned long *addr) |
| 351 | { |
| 352 | __const__ unsigned char *tmp = (__const__ unsigned char *) addr; |
| 353 | return (tmp[nr >> 3] >> (nr & 7)) & 1; |
| 354 | } |
| 355 | |
| 356 | #define __set_le_bit(nr, addr) \ |
| 357 | __set_bit((nr) ^ BITOP_LE_SWIZZLE, (addr)) |
| 358 | #define __clear_le_bit(nr, addr) \ |
| 359 | __clear_bit((nr) ^ BITOP_LE_SWIZZLE, (addr)) |
| 360 | |
| 361 | #define test_and_set_le_bit(nr, addr) \ |
| 362 | test_and_set_bit((nr) ^ BITOP_LE_SWIZZLE, (addr)) |
| 363 | #define test_and_clear_le_bit(nr, addr) \ |
| 364 | test_and_clear_bit((nr) ^ BITOP_LE_SWIZZLE, (addr)) |
| 365 | |
| 366 | #define __test_and_set_le_bit(nr, addr) \ |
| 367 | __test_and_set_bit((nr) ^ BITOP_LE_SWIZZLE, (addr)) |
| 368 | #define __test_and_clear_le_bit(nr, addr) \ |
| 369 | __test_and_clear_bit((nr) ^ BITOP_LE_SWIZZLE, (addr)) |
| 370 | |
| 371 | #define find_first_zero_le_bit(addr, size) find_next_zero_le_bit((addr), (size), 0) |
| 372 | unsigned long find_next_zero_le_bit(const unsigned long *addr, |
| 373 | unsigned long size, unsigned long offset); |
| 374 | |
| 375 | /* Bitmap functions for the ext2 filesystem */ |
| 376 | |
| 377 | #define ext2_set_bit(nr,addr) \ |
| 378 | __test_and_set_le_bit((nr), (unsigned long*)addr) |
| 379 | #define ext2_clear_bit(nr, addr) \ |
| 380 | __test_and_clear_le_bit((nr), (unsigned long*)addr) |
| 381 | |
| 382 | #define ext2_set_bit_atomic(lock, nr, addr) \ |
| 383 | test_and_set_le_bit((nr), (unsigned long*)addr) |
| 384 | #define ext2_clear_bit_atomic(lock, nr, addr) \ |
| 385 | test_and_clear_le_bit((nr), (unsigned long*)addr) |
| 386 | |
| 387 | #define ext2_test_bit(nr, addr) test_le_bit((nr),(unsigned long*)addr) |
| 388 | |
| 389 | #define ext2_find_first_zero_bit(addr, size) \ |
| 390 | find_first_zero_le_bit((unsigned long*)addr, size) |
| 391 | #define ext2_find_next_zero_bit(addr, size, off) \ |
| 392 | find_next_zero_le_bit((unsigned long*)addr, size, off) |
| 393 | |
| 394 | /* Bitmap functions for the minix filesystem. */ |
| 395 | |
| 396 | #define minix_test_and_set_bit(nr,addr) \ |
| 397 | __test_and_set_le_bit(nr, (unsigned long *)addr) |
| 398 | #define minix_set_bit(nr,addr) \ |
| 399 | __set_le_bit(nr, (unsigned long *)addr) |
| 400 | #define minix_test_and_clear_bit(nr,addr) \ |
| 401 | __test_and_clear_le_bit(nr, (unsigned long *)addr) |
| 402 | #define minix_test_bit(nr,addr) \ |
| 403 | test_le_bit(nr, (unsigned long *)addr) |
| 404 | |
| 405 | #define minix_find_first_zero_bit(addr,size) \ |
| 406 | find_first_zero_le_bit((unsigned long *)addr, size) |
| 407 | |
| 408 | /* |
| 409 | * Every architecture must define this function. It's the fastest |
| 410 | * way of searching a 140-bit bitmap where the first 100 bits are |
| 411 | * unlikely to be set. It's guaranteed that at least one of the 140 |
| 412 | * bits is cleared. |
| 413 | */ |
| 414 | static inline int sched_find_first_bit(const unsigned long *b) |
| 415 | { |
| 416 | #ifdef CONFIG_PPC64 |
| 417 | if (unlikely(b[0])) |
| 418 | return __ffs(b[0]); |
| 419 | if (unlikely(b[1])) |
| 420 | return __ffs(b[1]) + 64; |
| 421 | return __ffs(b[2]) + 128; |
| 422 | #else |
| 423 | if (unlikely(b[0])) |
| 424 | return __ffs(b[0]); |
| 425 | if (unlikely(b[1])) |
| 426 | return __ffs(b[1]) + 32; |
| 427 | if (unlikely(b[2])) |
| 428 | return __ffs(b[2]) + 64; |
| 429 | if (b[3]) |
| 430 | return __ffs(b[3]) + 96; |
| 431 | return __ffs(b[4]) + 128; |
| 432 | #endif |
| 433 | } |
| 434 | |
| 435 | #endif /* __KERNEL__ */ |
| 436 | |
| 437 | #endif /* _ASM_POWERPC_BITOPS_H */ |