bitops.h 9.3 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332
  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. #ifndef _ASM_POWERPC_BITOPS_H
  36. #define _ASM_POWERPC_BITOPS_H
  37. #ifdef __KERNEL__
  38. #include <linux/compiler.h>
  39. #include <asm/atomic.h>
  40. #include <asm/asm-compat.h>
  41. #include <asm/synch.h>
  42. /*
  43. * clear_bit doesn't imply a memory barrier
  44. */
  45. #define smp_mb__before_clear_bit() smp_mb()
  46. #define smp_mb__after_clear_bit() smp_mb()
  47. #define BITOP_MASK(nr) (1UL << ((nr) % BITS_PER_LONG))
  48. #define BITOP_WORD(nr) ((nr) / BITS_PER_LONG)
  49. #define BITOP_LE_SWIZZLE ((BITS_PER_LONG-1) & ~0x7)
  50. static __inline__ void set_bit(int nr, volatile unsigned long *addr)
  51. {
  52. unsigned long old;
  53. unsigned long mask = BITOP_MASK(nr);
  54. unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr);
  55. __asm__ __volatile__(
  56. "1:" PPC_LLARX "%0,0,%3 # set_bit\n"
  57. "or %0,%0,%2\n"
  58. PPC405_ERR77(0,%3)
  59. PPC_STLCX "%0,0,%3\n"
  60. "bne- 1b"
  61. : "=&r"(old), "=m"(*p)
  62. : "r"(mask), "r"(p), "m"(*p)
  63. : "cc" );
  64. }
  65. static __inline__ void clear_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. __asm__ __volatile__(
  71. "1:" PPC_LLARX "%0,0,%3 # clear_bit\n"
  72. "andc %0,%0,%2\n"
  73. PPC405_ERR77(0,%3)
  74. PPC_STLCX "%0,0,%3\n"
  75. "bne- 1b"
  76. : "=&r"(old), "=m"(*p)
  77. : "r"(mask), "r"(p), "m"(*p)
  78. : "cc" );
  79. }
  80. static __inline__ void change_bit(int nr, volatile unsigned long *addr)
  81. {
  82. unsigned long old;
  83. unsigned long mask = BITOP_MASK(nr);
  84. unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr);
  85. __asm__ __volatile__(
  86. "1:" PPC_LLARX "%0,0,%3 # change_bit\n"
  87. "xor %0,%0,%2\n"
  88. PPC405_ERR77(0,%3)
  89. PPC_STLCX "%0,0,%3\n"
  90. "bne- 1b"
  91. : "=&r"(old), "=m"(*p)
  92. : "r"(mask), "r"(p), "m"(*p)
  93. : "cc" );
  94. }
  95. static __inline__ int test_and_set_bit(unsigned long nr,
  96. volatile unsigned long *addr)
  97. {
  98. unsigned long old, t;
  99. unsigned long mask = BITOP_MASK(nr);
  100. unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr);
  101. __asm__ __volatile__(
  102. LWSYNC_ON_SMP
  103. "1:" PPC_LLARX "%0,0,%3 # test_and_set_bit\n"
  104. "or %1,%0,%2 \n"
  105. PPC405_ERR77(0,%3)
  106. PPC_STLCX "%1,0,%3 \n"
  107. "bne- 1b"
  108. ISYNC_ON_SMP
  109. : "=&r" (old), "=&r" (t)
  110. : "r" (mask), "r" (p)
  111. : "cc", "memory");
  112. return (old & mask) != 0;
  113. }
  114. static __inline__ int test_and_clear_bit(unsigned long nr,
  115. volatile unsigned long *addr)
  116. {
  117. unsigned long old, t;
  118. unsigned long mask = BITOP_MASK(nr);
  119. unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr);
  120. __asm__ __volatile__(
  121. LWSYNC_ON_SMP
  122. "1:" PPC_LLARX "%0,0,%3 # test_and_clear_bit\n"
  123. "andc %1,%0,%2 \n"
  124. PPC405_ERR77(0,%3)
  125. PPC_STLCX "%1,0,%3 \n"
  126. "bne- 1b"
  127. ISYNC_ON_SMP
  128. : "=&r" (old), "=&r" (t)
  129. : "r" (mask), "r" (p)
  130. : "cc", "memory");
  131. return (old & mask) != 0;
  132. }
  133. static __inline__ int test_and_change_bit(unsigned long nr,
  134. volatile unsigned long *addr)
  135. {
  136. unsigned long old, t;
  137. unsigned long mask = BITOP_MASK(nr);
  138. unsigned long *p = ((unsigned long *)addr) + BITOP_WORD(nr);
  139. __asm__ __volatile__(
  140. LWSYNC_ON_SMP
  141. "1:" PPC_LLARX "%0,0,%3 # test_and_change_bit\n"
  142. "xor %1,%0,%2 \n"
  143. PPC405_ERR77(0,%3)
  144. PPC_STLCX "%1,0,%3 \n"
  145. "bne- 1b"
  146. ISYNC_ON_SMP
  147. : "=&r" (old), "=&r" (t)
  148. : "r" (mask), "r" (p)
  149. : "cc", "memory");
  150. return (old & mask) != 0;
  151. }
  152. static __inline__ void set_bits(unsigned long mask, unsigned long *addr)
  153. {
  154. unsigned long old;
  155. __asm__ __volatile__(
  156. "1:" PPC_LLARX "%0,0,%3 # set_bits\n"
  157. "or %0,%0,%2\n"
  158. PPC_STLCX "%0,0,%3\n"
  159. "bne- 1b"
  160. : "=&r" (old), "=m" (*addr)
  161. : "r" (mask), "r" (addr), "m" (*addr)
  162. : "cc");
  163. }
  164. #include <asm-generic/bitops/non-atomic.h>
  165. /*
  166. * Return the zero-based bit position (LE, not IBM bit numbering) of
  167. * the most significant 1-bit in a double word.
  168. */
  169. static __inline__ int __ilog2(unsigned long x)
  170. {
  171. int lz;
  172. asm (PPC_CNTLZL "%0,%1" : "=r" (lz) : "r" (x));
  173. return BITS_PER_LONG - 1 - lz;
  174. }
  175. /*
  176. * Determines the bit position of the least significant 0 bit in the
  177. * specified double word. The returned bit position will be
  178. * zero-based, starting from the right side (63/31 - 0).
  179. */
  180. static __inline__ unsigned long ffz(unsigned long x)
  181. {
  182. /* no zero exists anywhere in the 8 byte area. */
  183. if ((x = ~x) == 0)
  184. return BITS_PER_LONG;
  185. /*
  186. * Calculate the bit position of the least signficant '1' bit in x
  187. * (since x has been changed this will actually be the least signficant
  188. * '0' bit in * the original x). Note: (x & -x) gives us a mask that
  189. * is the least significant * (RIGHT-most) 1-bit of the value in x.
  190. */
  191. return __ilog2(x & -x);
  192. }
  193. static __inline__ int __ffs(unsigned long x)
  194. {
  195. return __ilog2(x & -x);
  196. }
  197. /*
  198. * ffs: find first bit set. This is defined the same way as
  199. * the libc and compiler builtin ffs routines, therefore
  200. * differs in spirit from the above ffz (man ffs).
  201. */
  202. static __inline__ int ffs(int x)
  203. {
  204. unsigned long i = (unsigned long)x;
  205. return __ilog2(i & -i) + 1;
  206. }
  207. /*
  208. * fls: find last (most-significant) bit set.
  209. * Note fls(0) = 0, fls(1) = 1, fls(0x80000000) = 32.
  210. */
  211. static __inline__ int fls(unsigned int x)
  212. {
  213. int lz;
  214. asm ("cntlzw %0,%1" : "=r" (lz) : "r" (x));
  215. return 32 - lz;
  216. }
  217. #include <asm-generic/bitops/fls64.h>
  218. #include <asm-generic/bitops/hweight.h>
  219. #define find_first_zero_bit(addr, size) find_next_zero_bit((addr), (size), 0)
  220. unsigned long find_next_zero_bit(const unsigned long *addr,
  221. unsigned long size, unsigned long offset);
  222. /**
  223. * find_first_bit - find the first set bit in a memory region
  224. * @addr: The address to start the search at
  225. * @size: The maximum size to search
  226. *
  227. * Returns the bit-number of the first set bit, not the number of the byte
  228. * containing a bit.
  229. */
  230. #define find_first_bit(addr, size) find_next_bit((addr), (size), 0)
  231. unsigned long find_next_bit(const unsigned long *addr,
  232. unsigned long size, unsigned long offset);
  233. /* Little-endian versions */
  234. static __inline__ int test_le_bit(unsigned long nr,
  235. __const__ unsigned long *addr)
  236. {
  237. __const__ unsigned char *tmp = (__const__ unsigned char *) addr;
  238. return (tmp[nr >> 3] >> (nr & 7)) & 1;
  239. }
  240. #define __set_le_bit(nr, addr) \
  241. __set_bit((nr) ^ BITOP_LE_SWIZZLE, (addr))
  242. #define __clear_le_bit(nr, addr) \
  243. __clear_bit((nr) ^ BITOP_LE_SWIZZLE, (addr))
  244. #define test_and_set_le_bit(nr, addr) \
  245. test_and_set_bit((nr) ^ BITOP_LE_SWIZZLE, (addr))
  246. #define test_and_clear_le_bit(nr, addr) \
  247. test_and_clear_bit((nr) ^ BITOP_LE_SWIZZLE, (addr))
  248. #define __test_and_set_le_bit(nr, addr) \
  249. __test_and_set_bit((nr) ^ BITOP_LE_SWIZZLE, (addr))
  250. #define __test_and_clear_le_bit(nr, addr) \
  251. __test_and_clear_bit((nr) ^ BITOP_LE_SWIZZLE, (addr))
  252. #define find_first_zero_le_bit(addr, size) find_next_zero_le_bit((addr), (size), 0)
  253. unsigned long find_next_zero_le_bit(const unsigned long *addr,
  254. unsigned long size, unsigned long offset);
  255. /* Bitmap functions for the ext2 filesystem */
  256. #define ext2_set_bit(nr,addr) \
  257. __test_and_set_le_bit((nr), (unsigned long*)addr)
  258. #define ext2_clear_bit(nr, addr) \
  259. __test_and_clear_le_bit((nr), (unsigned long*)addr)
  260. #define ext2_set_bit_atomic(lock, nr, addr) \
  261. test_and_set_le_bit((nr), (unsigned long*)addr)
  262. #define ext2_clear_bit_atomic(lock, nr, addr) \
  263. test_and_clear_le_bit((nr), (unsigned long*)addr)
  264. #define ext2_test_bit(nr, addr) test_le_bit((nr),(unsigned long*)addr)
  265. #define ext2_find_first_zero_bit(addr, size) \
  266. find_first_zero_le_bit((unsigned long*)addr, size)
  267. #define ext2_find_next_zero_bit(addr, size, off) \
  268. find_next_zero_le_bit((unsigned long*)addr, size, off)
  269. /* Bitmap functions for the minix filesystem. */
  270. #define minix_test_and_set_bit(nr,addr) \
  271. __test_and_set_le_bit(nr, (unsigned long *)addr)
  272. #define minix_set_bit(nr,addr) \
  273. __set_le_bit(nr, (unsigned long *)addr)
  274. #define minix_test_and_clear_bit(nr,addr) \
  275. __test_and_clear_le_bit(nr, (unsigned long *)addr)
  276. #define minix_test_bit(nr,addr) \
  277. test_le_bit(nr, (unsigned long *)addr)
  278. #define minix_find_first_zero_bit(addr,size) \
  279. find_first_zero_le_bit((unsigned long *)addr, size)
  280. #include <asm-generic/bitops/sched.h>
  281. #endif /* __KERNEL__ */
  282. #endif /* _ASM_POWERPC_BITOPS_H */