]> bbs.cooldavid.org Git - net-next-2.6.git/blame - include/asm-x86/bitops.h
x86, bitops: make constant-bit set/clear_bit ops faster, adapt, clean up
[net-next-2.6.git] / include / asm-x86 / bitops.h
CommitLineData
1c54d770
JF
1#ifndef _ASM_X86_BITOPS_H
2#define _ASM_X86_BITOPS_H
3
4/*
5 * Copyright 1992, Linus Torvalds.
6 */
7
8#ifndef _LINUX_BITOPS_H
9#error only <linux/bitops.h> can be included directly
10#endif
11
12#include <linux/compiler.h>
13#include <asm/alternative.h>
14
15/*
16 * These have to be done with inline assembly: that way the bit-setting
17 * is guaranteed to be atomic. All bit operations return 0 if the bit
18 * was cleared before the operation and != 0 if it was not.
19 *
20 * bit 0 is the LSB of addr; bit 32 is the LSB of (addr+1).
21 */
22
23#if __GNUC__ < 4 || (__GNUC__ == 4 && __GNUC_MINOR__ < 1)
24/* Technically wrong, but this avoids compilation errors on some gcc
25 versions. */
1a750e0c 26#define BITOP_ADDR(x) "=m" (*(volatile long *) (x))
1c54d770 27#else
1a750e0c 28#define BITOP_ADDR(x) "+m" (*(volatile long *) (x))
1c54d770
JF
29#endif
30
7dbceaf9 31#define ADDR BITOP_ADDR(addr)
1a750e0c
LT
32
33/*
34 * We do the locked ops that don't return the old value as
35 * a mask operation on a byte.
36 */
7dbceaf9
IM
37#define IS_IMMEDIATE(nr) (__builtin_constant_p(nr))
38#define CONST_MASK_ADDR(nr, addr) BITOP_ADDR((void *)(addr) + ((nr)>>3))
39#define CONST_MASK(nr) (1 << ((nr) & 7))
1a750e0c 40
1c54d770
JF
41/**
42 * set_bit - Atomically set a bit in memory
43 * @nr: the bit to set
44 * @addr: the address to start counting from
45 *
46 * This function is atomic and may not be reordered. See __set_bit()
47 * if you do not require the atomic guarantees.
48 *
49 * Note: there are no guarantees that this function will not be reordered
50 * on non x86 architectures, so if you are writing portable code,
51 * make sure not to rely on its reordering guarantees.
52 *
53 * Note that @nr may be almost arbitrarily large; this function is not
54 * restricted to acting on a single-word quantity.
55 */
1a750e0c 56static inline void set_bit(unsigned int nr, volatile unsigned long *addr)
1c54d770 57{
7dbceaf9
IM
58 if (IS_IMMEDIATE(nr)) {
59 asm volatile(LOCK_PREFIX "orb %1,%0"
60 : CONST_MASK_ADDR(nr, addr)
61 : "i" (CONST_MASK(nr))
62 : "memory");
63 } else {
64 asm volatile(LOCK_PREFIX "bts %1,%0"
65 : BITOP_ADDR(addr) : "Ir" (nr) : "memory");
66 }
1c54d770
JF
67}
68
69/**
70 * __set_bit - Set a bit in memory
71 * @nr: the bit to set
72 * @addr: the address to start counting from
73 *
74 * Unlike set_bit(), this function is non-atomic and may be reordered.
75 * If it's called on the same region of memory simultaneously, the effect
76 * may be that only one operation succeeds.
77 */
5136dea5 78static inline void __set_bit(int nr, volatile unsigned long *addr)
1c54d770 79{
f19dcf4a 80 asm volatile("bts %1,%0" : ADDR : "Ir" (nr) : "memory");
1c54d770
JF
81}
82
1c54d770
JF
83/**
84 * clear_bit - Clears a bit in memory
85 * @nr: Bit to clear
86 * @addr: Address to start counting from
87 *
88 * clear_bit() is atomic and may not be reordered. However, it does
89 * not contain a memory barrier, so if it is used for locking purposes,
90 * you should call smp_mb__before_clear_bit() and/or smp_mb__after_clear_bit()
91 * in order to ensure changes are visible on other processors.
92 */
5136dea5 93static inline void clear_bit(int nr, volatile unsigned long *addr)
1c54d770 94{
7dbceaf9
IM
95 if (IS_IMMEDIATE(nr)) {
96 asm volatile(LOCK_PREFIX "andb %1,%0"
97 : CONST_MASK_ADDR(nr, addr)
98 : "i" (~CONST_MASK(nr)));
99 } else {
100 asm volatile(LOCK_PREFIX "btr %1,%0"
101 : BITOP_ADDR(addr)
102 : "Ir" (nr));
103 }
1c54d770
JF
104}
105
106/*
107 * clear_bit_unlock - Clears a bit in memory
108 * @nr: Bit to clear
109 * @addr: Address to start counting from
110 *
111 * clear_bit() is atomic and implies release semantics before the memory
112 * operation. It can be used for an unlock.
113 */
5136dea5 114static inline void clear_bit_unlock(unsigned nr, volatile unsigned long *addr)
1c54d770
JF
115{
116 barrier();
117 clear_bit(nr, addr);
118}
119
5136dea5 120static inline void __clear_bit(int nr, volatile unsigned long *addr)
1c54d770 121{
eb2b4e68 122 asm volatile("btr %1,%0" : ADDR : "Ir" (nr));
1c54d770
JF
123}
124
125/*
126 * __clear_bit_unlock - Clears a bit in memory
127 * @nr: Bit to clear
128 * @addr: Address to start counting from
129 *
130 * __clear_bit() is non-atomic and implies release semantics before the memory
131 * operation. It can be used for an unlock if no other CPUs can concurrently
132 * modify other bits in the word.
133 *
134 * No memory barrier is required here, because x86 cannot reorder stores past
135 * older loads. Same principle as spin_unlock.
136 */
5136dea5 137static inline void __clear_bit_unlock(unsigned nr, volatile unsigned long *addr)
1c54d770
JF
138{
139 barrier();
140 __clear_bit(nr, addr);
141}
142
143#define smp_mb__before_clear_bit() barrier()
144#define smp_mb__after_clear_bit() barrier()
145
146/**
147 * __change_bit - Toggle a bit in memory
148 * @nr: the bit to change
149 * @addr: the address to start counting from
150 *
151 * Unlike change_bit(), this function is non-atomic and may be reordered.
152 * If it's called on the same region of memory simultaneously, the effect
153 * may be that only one operation succeeds.
154 */
5136dea5 155static inline void __change_bit(int nr, volatile unsigned long *addr)
1c54d770 156{
eb2b4e68 157 asm volatile("btc %1,%0" : ADDR : "Ir" (nr));
1c54d770
JF
158}
159
160/**
161 * change_bit - Toggle a bit in memory
162 * @nr: Bit to change
163 * @addr: Address to start counting from
164 *
165 * change_bit() is atomic and may not be reordered.
166 * Note that @nr may be almost arbitrarily large; this function is not
167 * restricted to acting on a single-word quantity.
168 */
5136dea5 169static inline void change_bit(int nr, volatile unsigned long *addr)
1c54d770 170{
eb2b4e68 171 asm volatile(LOCK_PREFIX "btc %1,%0" : ADDR : "Ir" (nr));
1c54d770
JF
172}
173
174/**
175 * test_and_set_bit - Set a bit and return its old value
176 * @nr: Bit to set
177 * @addr: Address to count from
178 *
179 * This operation is atomic and cannot be reordered.
180 * It also implies a memory barrier.
181 */
5136dea5 182static inline int test_and_set_bit(int nr, volatile unsigned long *addr)
1c54d770
JF
183{
184 int oldbit;
185
186 asm volatile(LOCK_PREFIX "bts %2,%1\n\t"
286275c9 187 "sbb %0,%0" : "=r" (oldbit), ADDR : "Ir" (nr) : "memory");
1c54d770
JF
188
189 return oldbit;
190}
191
192/**
193 * test_and_set_bit_lock - Set a bit and return its old value for lock
194 * @nr: Bit to set
195 * @addr: Address to count from
196 *
197 * This is the same as test_and_set_bit on x86.
198 */
5136dea5 199static inline int test_and_set_bit_lock(int nr, volatile unsigned long *addr)
1c54d770
JF
200{
201 return test_and_set_bit(nr, addr);
202}
203
204/**
205 * __test_and_set_bit - Set a bit and return its old value
206 * @nr: Bit to set
207 * @addr: Address to count from
208 *
209 * This operation is non-atomic and can be reordered.
210 * If two examples of this operation race, one can appear to succeed
211 * but actually fail. You must protect multiple accesses with a lock.
212 */
5136dea5 213static inline int __test_and_set_bit(int nr, volatile unsigned long *addr)
1c54d770
JF
214{
215 int oldbit;
216
eb2b4e68
SHT
217 asm("bts %2,%1\n\t"
218 "sbb %0,%0"
219 : "=r" (oldbit), ADDR
220 : "Ir" (nr));
1c54d770
JF
221 return oldbit;
222}
223
224/**
225 * test_and_clear_bit - Clear a bit and return its old value
226 * @nr: Bit to clear
227 * @addr: Address to count from
228 *
229 * This operation is atomic and cannot be reordered.
230 * It also implies a memory barrier.
231 */
5136dea5 232static inline int test_and_clear_bit(int nr, volatile unsigned long *addr)
1c54d770
JF
233{
234 int oldbit;
235
236 asm volatile(LOCK_PREFIX "btr %2,%1\n\t"
237 "sbb %0,%0"
286275c9 238 : "=r" (oldbit), ADDR : "Ir" (nr) : "memory");
1c54d770
JF
239
240 return oldbit;
241}
242
243/**
244 * __test_and_clear_bit - Clear a bit and return its old value
245 * @nr: Bit to clear
246 * @addr: Address to count from
247 *
248 * This operation is non-atomic and can be reordered.
249 * If two examples of this operation race, one can appear to succeed
250 * but actually fail. You must protect multiple accesses with a lock.
251 */
5136dea5 252static inline int __test_and_clear_bit(int nr, volatile unsigned long *addr)
1c54d770
JF
253{
254 int oldbit;
255
eb2b4e68 256 asm volatile("btr %2,%1\n\t"
1c54d770 257 "sbb %0,%0"
eb2b4e68
SHT
258 : "=r" (oldbit), ADDR
259 : "Ir" (nr));
1c54d770
JF
260 return oldbit;
261}
262
263/* WARNING: non atomic and it can be reordered! */
5136dea5 264static inline int __test_and_change_bit(int nr, volatile unsigned long *addr)
1c54d770
JF
265{
266 int oldbit;
267
eb2b4e68 268 asm volatile("btc %2,%1\n\t"
1c54d770 269 "sbb %0,%0"
eb2b4e68
SHT
270 : "=r" (oldbit), ADDR
271 : "Ir" (nr) : "memory");
1c54d770
JF
272
273 return oldbit;
274}
275
276/**
277 * test_and_change_bit - Change a bit and return its old value
278 * @nr: Bit to change
279 * @addr: Address to count from
280 *
281 * This operation is atomic and cannot be reordered.
282 * It also implies a memory barrier.
283 */
5136dea5 284static inline int test_and_change_bit(int nr, volatile unsigned long *addr)
1c54d770
JF
285{
286 int oldbit;
287
288 asm volatile(LOCK_PREFIX "btc %2,%1\n\t"
289 "sbb %0,%0"
286275c9 290 : "=r" (oldbit), ADDR : "Ir" (nr) : "memory");
1c54d770
JF
291
292 return oldbit;
293}
294
5136dea5 295static inline int constant_test_bit(int nr, const volatile unsigned long *addr)
1c54d770 296{
26996dd2
GOC
297 return ((1UL << (nr % BITS_PER_LONG)) &
298 (((unsigned long *)addr)[nr / BITS_PER_LONG])) != 0;
1c54d770
JF
299}
300
5136dea5 301static inline int variable_test_bit(int nr, volatile const unsigned long *addr)
1c54d770
JF
302{
303 int oldbit;
304
eb2b4e68 305 asm volatile("bt %2,%1\n\t"
1c54d770
JF
306 "sbb %0,%0"
307 : "=r" (oldbit)
eb2b4e68 308 : "m" (*(unsigned long *)addr), "Ir" (nr));
1c54d770
JF
309
310 return oldbit;
311}
312
313#if 0 /* Fool kernel-doc since it doesn't do macros yet */
314/**
315 * test_bit - Determine whether a bit is set
316 * @nr: bit number to test
317 * @addr: Address to start counting from
318 */
319static int test_bit(int nr, const volatile unsigned long *addr);
320#endif
321
f19dcf4a
JP
322#define test_bit(nr, addr) \
323 (__builtin_constant_p((nr)) \
324 ? constant_test_bit((nr), (addr)) \
325 : variable_test_bit((nr), (addr)))
1c54d770 326
12d9c842
AH
327/**
328 * __ffs - find first set bit in word
329 * @word: The word to search
330 *
331 * Undefined if no bit exists, so code should check against 0 first.
332 */
333static inline unsigned long __ffs(unsigned long word)
334{
f19dcf4a
JP
335 asm("bsf %1,%0"
336 : "=r" (word)
337 : "rm" (word));
12d9c842
AH
338 return word;
339}
340
341/**
342 * ffz - find first zero bit in word
343 * @word: The word to search
344 *
345 * Undefined if no zero exists, so code should check against ~0UL first.
346 */
347static inline unsigned long ffz(unsigned long word)
348{
f19dcf4a
JP
349 asm("bsf %1,%0"
350 : "=r" (word)
351 : "r" (~word));
12d9c842
AH
352 return word;
353}
354
355/*
356 * __fls: find last set bit in word
357 * @word: The word to search
358 *
359 * Undefined if no zero exists, so code should check against ~0UL first.
360 */
361static inline unsigned long __fls(unsigned long word)
362{
f19dcf4a
JP
363 asm("bsr %1,%0"
364 : "=r" (word)
365 : "rm" (word));
12d9c842
AH
366 return word;
367}
368
369#ifdef __KERNEL__
370/**
371 * ffs - find first set bit in word
372 * @x: the word to search
373 *
374 * This is defined the same way as the libc and compiler builtin ffs
375 * routines, therefore differs in spirit from the other bitops.
376 *
377 * ffs(value) returns 0 if value is 0 or the position of the first
378 * set bit if value is nonzero. The first (least significant) bit
379 * is at position 1.
380 */
381static inline int ffs(int x)
382{
383 int r;
384#ifdef CONFIG_X86_CMOV
f19dcf4a
JP
385 asm("bsfl %1,%0\n\t"
386 "cmovzl %2,%0"
387 : "=r" (r) : "rm" (x), "r" (-1));
12d9c842 388#else
f19dcf4a
JP
389 asm("bsfl %1,%0\n\t"
390 "jnz 1f\n\t"
391 "movl $-1,%0\n"
392 "1:" : "=r" (r) : "rm" (x));
12d9c842
AH
393#endif
394 return r + 1;
395}
396
397/**
398 * fls - find last set bit in word
399 * @x: the word to search
400 *
401 * This is defined in a similar way as the libc and compiler builtin
402 * ffs, but returns the position of the most significant set bit.
403 *
404 * fls(value) returns 0 if value is 0 or the position of the last
405 * set bit if value is nonzero. The last (most significant) bit is
406 * at position 32.
407 */
408static inline int fls(int x)
409{
410 int r;
411#ifdef CONFIG_X86_CMOV
f19dcf4a
JP
412 asm("bsrl %1,%0\n\t"
413 "cmovzl %2,%0"
414 : "=&r" (r) : "rm" (x), "rm" (-1));
12d9c842 415#else
f19dcf4a
JP
416 asm("bsrl %1,%0\n\t"
417 "jnz 1f\n\t"
418 "movl $-1,%0\n"
419 "1:" : "=r" (r) : "rm" (x));
12d9c842
AH
420#endif
421 return r + 1;
422}
423#endif /* __KERNEL__ */
424
1c54d770
JF
425#undef ADDR
426
d66462f5
AH
427static inline void set_bit_string(unsigned long *bitmap,
428 unsigned long i, int len)
429{
430 unsigned long end = i + len;
431 while (i < end) {
432 __set_bit(i, bitmap);
433 i++;
434 }
435}
436
437#ifdef __KERNEL__
438
439#include <asm-generic/bitops/sched.h>
440
441#define ARCH_HAS_FAST_MULTIPLIER 1
442
443#include <asm-generic/bitops/hweight.h>
1c54d770 444
d66462f5
AH
445#endif /* __KERNEL__ */
446
447#include <asm-generic/bitops/fls64.h>
448
449#ifdef __KERNEL__
450
451#include <asm-generic/bitops/ext2-non-atomic.h>
452
453#define ext2_set_bit_atomic(lock, nr, addr) \
454 test_and_set_bit((nr), (unsigned long *)(addr))
455#define ext2_clear_bit_atomic(lock, nr, addr) \
456 test_and_clear_bit((nr), (unsigned long *)(addr))
457
458#include <asm-generic/bitops/minix.h>
459
460#endif /* __KERNEL__ */
1c54d770 461#endif /* _ASM_X86_BITOPS_H */