xref: /openbmc/linux/arch/ia64/include/asm/bitops.h (revision 9344dade)
1 #ifndef _ASM_IA64_BITOPS_H
2 #define _ASM_IA64_BITOPS_H
3 
4 /*
5  * Copyright (C) 1998-2003 Hewlett-Packard Co
6  *	David Mosberger-Tang <davidm@hpl.hp.com>
7  *
8  * 02/06/02 find_next_bit() and find_first_bit() added from Erich Focht's ia64
9  * O(1) scheduler patch
10  */
11 
12 #ifndef _LINUX_BITOPS_H
13 #error only <linux/bitops.h> can be included directly
14 #endif
15 
16 #include <linux/compiler.h>
17 #include <linux/types.h>
18 #include <asm/intrinsics.h>
19 
20 /**
21  * set_bit - Atomically set a bit in memory
22  * @nr: the bit to set
23  * @addr: the address to start counting from
24  *
25  * This function is atomic and may not be reordered.  See __set_bit()
26  * if you do not require the atomic guarantees.
27  * Note that @nr may be almost arbitrarily large; this function is not
28  * restricted to acting on a single-word quantity.
29  *
30  * The address must be (at least) "long" aligned.
31  * Note that there are driver (e.g., eepro100) which use these operations to
32  * operate on hw-defined data-structures, so we can't easily change these
33  * operations to force a bigger alignment.
34  *
35  * bit 0 is the LSB of addr; bit 32 is the LSB of (addr+1).
36  */
37 static __inline__ void
38 set_bit (int nr, volatile void *addr)
39 {
40 	__u32 bit, old, new;
41 	volatile __u32 *m;
42 	CMPXCHG_BUGCHECK_DECL
43 
44 	m = (volatile __u32 *) addr + (nr >> 5);
45 	bit = 1 << (nr & 31);
46 	do {
47 		CMPXCHG_BUGCHECK(m);
48 		old = *m;
49 		new = old | bit;
50 	} while (cmpxchg_acq(m, old, new) != old);
51 }
52 
53 /**
54  * __set_bit - Set a bit in memory
55  * @nr: the bit to set
56  * @addr: the address to start counting from
57  *
58  * Unlike set_bit(), this function is non-atomic and may be reordered.
59  * If it's called on the same region of memory simultaneously, the effect
60  * may be that only one operation succeeds.
61  */
62 static __inline__ void
63 __set_bit (int nr, volatile void *addr)
64 {
65 	*((__u32 *) addr + (nr >> 5)) |= (1 << (nr & 31));
66 }
67 
68 /*
69  * clear_bit() has "acquire" semantics.
70  */
71 #define smp_mb__before_clear_bit()	smp_mb()
72 #define smp_mb__after_clear_bit()	do { /* skip */; } while (0)
73 
74 /**
75  * clear_bit - Clears a bit in memory
76  * @nr: Bit to clear
77  * @addr: Address to start counting from
78  *
79  * clear_bit() is atomic and may not be reordered.  However, it does
80  * not contain a memory barrier, so if it is used for locking purposes,
81  * you should call smp_mb__before_clear_bit() and/or smp_mb__after_clear_bit()
82  * in order to ensure changes are visible on other processors.
83  */
84 static __inline__ void
85 clear_bit (int nr, volatile void *addr)
86 {
87 	__u32 mask, old, new;
88 	volatile __u32 *m;
89 	CMPXCHG_BUGCHECK_DECL
90 
91 	m = (volatile __u32 *) addr + (nr >> 5);
92 	mask = ~(1 << (nr & 31));
93 	do {
94 		CMPXCHG_BUGCHECK(m);
95 		old = *m;
96 		new = old & mask;
97 	} while (cmpxchg_acq(m, old, new) != old);
98 }
99 
100 /**
101  * clear_bit_unlock - Clears a bit in memory with release
102  * @nr: Bit to clear
103  * @addr: Address to start counting from
104  *
105  * clear_bit_unlock() is atomic and may not be reordered.  It does
106  * contain a memory barrier suitable for unlock type operations.
107  */
108 static __inline__ void
109 clear_bit_unlock (int nr, volatile void *addr)
110 {
111 	__u32 mask, old, new;
112 	volatile __u32 *m;
113 	CMPXCHG_BUGCHECK_DECL
114 
115 	m = (volatile __u32 *) addr + (nr >> 5);
116 	mask = ~(1 << (nr & 31));
117 	do {
118 		CMPXCHG_BUGCHECK(m);
119 		old = *m;
120 		new = old & mask;
121 	} while (cmpxchg_rel(m, old, new) != old);
122 }
123 
124 /**
125  * __clear_bit_unlock - Non-atomically clears a bit in memory with release
126  * @nr: Bit to clear
127  * @addr: Address to start counting from
128  *
129  * Similarly to clear_bit_unlock, the implementation uses a store
130  * with release semantics. See also arch_spin_unlock().
131  */
132 static __inline__ void
133 __clear_bit_unlock(int nr, void *addr)
134 {
135 	__u32 * const m = (__u32 *) addr + (nr >> 5);
136 	__u32 const new = *m & ~(1 << (nr & 31));
137 
138 	ia64_st4_rel_nta(m, new);
139 }
140 
141 /**
142  * __clear_bit - Clears a bit in memory (non-atomic version)
143  * @nr: the bit to clear
144  * @addr: the address to start counting from
145  *
146  * Unlike clear_bit(), this function is non-atomic and may be reordered.
147  * If it's called on the same region of memory simultaneously, the effect
148  * may be that only one operation succeeds.
149  */
150 static __inline__ void
151 __clear_bit (int nr, volatile void *addr)
152 {
153 	*((__u32 *) addr + (nr >> 5)) &= ~(1 << (nr & 31));
154 }
155 
156 /**
157  * change_bit - Toggle a bit in memory
158  * @nr: Bit to toggle
159  * @addr: Address to start counting from
160  *
161  * change_bit() is atomic and may not be reordered.
162  * Note that @nr may be almost arbitrarily large; this function is not
163  * restricted to acting on a single-word quantity.
164  */
165 static __inline__ void
166 change_bit (int nr, volatile void *addr)
167 {
168 	__u32 bit, old, new;
169 	volatile __u32 *m;
170 	CMPXCHG_BUGCHECK_DECL
171 
172 	m = (volatile __u32 *) addr + (nr >> 5);
173 	bit = (1 << (nr & 31));
174 	do {
175 		CMPXCHG_BUGCHECK(m);
176 		old = *m;
177 		new = old ^ bit;
178 	} while (cmpxchg_acq(m, old, new) != old);
179 }
180 
181 /**
182  * __change_bit - Toggle a bit in memory
183  * @nr: the bit to toggle
184  * @addr: the address to start counting from
185  *
186  * Unlike change_bit(), this function is non-atomic and may be reordered.
187  * If it's called on the same region of memory simultaneously, the effect
188  * may be that only one operation succeeds.
189  */
190 static __inline__ void
191 __change_bit (int nr, volatile void *addr)
192 {
193 	*((__u32 *) addr + (nr >> 5)) ^= (1 << (nr & 31));
194 }
195 
196 /**
197  * test_and_set_bit - Set a bit and return its old value
198  * @nr: Bit to set
199  * @addr: Address to count from
200  *
201  * This operation is atomic and cannot be reordered.
202  * It also implies the acquisition side of the memory barrier.
203  */
204 static __inline__ int
205 test_and_set_bit (int nr, volatile void *addr)
206 {
207 	__u32 bit, old, new;
208 	volatile __u32 *m;
209 	CMPXCHG_BUGCHECK_DECL
210 
211 	m = (volatile __u32 *) addr + (nr >> 5);
212 	bit = 1 << (nr & 31);
213 	do {
214 		CMPXCHG_BUGCHECK(m);
215 		old = *m;
216 		new = old | bit;
217 	} while (cmpxchg_acq(m, old, new) != old);
218 	return (old & bit) != 0;
219 }
220 
221 /**
222  * test_and_set_bit_lock - Set a bit and return its old value for lock
223  * @nr: Bit to set
224  * @addr: Address to count from
225  *
226  * This is the same as test_and_set_bit on ia64
227  */
228 #define test_and_set_bit_lock test_and_set_bit
229 
230 /**
231  * __test_and_set_bit - Set a bit and return its old value
232  * @nr: Bit to set
233  * @addr: Address to count from
234  *
235  * This operation is non-atomic and can be reordered.
236  * If two examples of this operation race, one can appear to succeed
237  * but actually fail.  You must protect multiple accesses with a lock.
238  */
239 static __inline__ int
240 __test_and_set_bit (int nr, volatile void *addr)
241 {
242 	__u32 *p = (__u32 *) addr + (nr >> 5);
243 	__u32 m = 1 << (nr & 31);
244 	int oldbitset = (*p & m) != 0;
245 
246 	*p |= m;
247 	return oldbitset;
248 }
249 
250 /**
251  * test_and_clear_bit - Clear a bit and return its old value
252  * @nr: Bit to clear
253  * @addr: Address to count from
254  *
255  * This operation is atomic and cannot be reordered.
256  * It also implies the acquisition side of the memory barrier.
257  */
258 static __inline__ int
259 test_and_clear_bit (int nr, volatile void *addr)
260 {
261 	__u32 mask, old, new;
262 	volatile __u32 *m;
263 	CMPXCHG_BUGCHECK_DECL
264 
265 	m = (volatile __u32 *) addr + (nr >> 5);
266 	mask = ~(1 << (nr & 31));
267 	do {
268 		CMPXCHG_BUGCHECK(m);
269 		old = *m;
270 		new = old & mask;
271 	} while (cmpxchg_acq(m, old, new) != old);
272 	return (old & ~mask) != 0;
273 }
274 
275 /**
276  * __test_and_clear_bit - Clear a bit and return its old value
277  * @nr: Bit to clear
278  * @addr: Address to count from
279  *
280  * This operation is non-atomic and can be reordered.
281  * If two examples of this operation race, one can appear to succeed
282  * but actually fail.  You must protect multiple accesses with a lock.
283  */
284 static __inline__ int
285 __test_and_clear_bit(int nr, volatile void * addr)
286 {
287 	__u32 *p = (__u32 *) addr + (nr >> 5);
288 	__u32 m = 1 << (nr & 31);
289 	int oldbitset = (*p & m) != 0;
290 
291 	*p &= ~m;
292 	return oldbitset;
293 }
294 
295 /**
296  * test_and_change_bit - Change a bit and return its old value
297  * @nr: Bit to change
298  * @addr: Address to count from
299  *
300  * This operation is atomic and cannot be reordered.
301  * It also implies the acquisition side of the memory barrier.
302  */
303 static __inline__ int
304 test_and_change_bit (int nr, volatile void *addr)
305 {
306 	__u32 bit, old, new;
307 	volatile __u32 *m;
308 	CMPXCHG_BUGCHECK_DECL
309 
310 	m = (volatile __u32 *) addr + (nr >> 5);
311 	bit = (1 << (nr & 31));
312 	do {
313 		CMPXCHG_BUGCHECK(m);
314 		old = *m;
315 		new = old ^ bit;
316 	} while (cmpxchg_acq(m, old, new) != old);
317 	return (old & bit) != 0;
318 }
319 
320 /**
321  * __test_and_change_bit - Change a bit and return its old value
322  * @nr: Bit to change
323  * @addr: Address to count from
324  *
325  * This operation is non-atomic and can be reordered.
326  */
327 static __inline__ int
328 __test_and_change_bit (int nr, void *addr)
329 {
330 	__u32 old, bit = (1 << (nr & 31));
331 	__u32 *m = (__u32 *) addr + (nr >> 5);
332 
333 	old = *m;
334 	*m = old ^ bit;
335 	return (old & bit) != 0;
336 }
337 
338 static __inline__ int
339 test_bit (int nr, const volatile void *addr)
340 {
341 	return 1 & (((const volatile __u32 *) addr)[nr >> 5] >> (nr & 31));
342 }
343 
344 /**
345  * ffz - find the first zero bit in a long word
346  * @x: The long word to find the bit in
347  *
348  * Returns the bit-number (0..63) of the first (least significant) zero bit.
349  * Undefined if no zero exists, so code should check against ~0UL first...
350  */
351 static inline unsigned long
352 ffz (unsigned long x)
353 {
354 	unsigned long result;
355 
356 	result = ia64_popcnt(x & (~x - 1));
357 	return result;
358 }
359 
360 /**
361  * __ffs - find first bit in word.
362  * @x: The word to search
363  *
364  * Undefined if no bit exists, so code should check against 0 first.
365  */
366 static __inline__ unsigned long
367 __ffs (unsigned long x)
368 {
369 	unsigned long result;
370 
371 	result = ia64_popcnt((x-1) & ~x);
372 	return result;
373 }
374 
375 #ifdef __KERNEL__
376 
377 /*
378  * Return bit number of last (most-significant) bit set.  Undefined
379  * for x==0.  Bits are numbered from 0..63 (e.g., ia64_fls(9) == 3).
380  */
381 static inline unsigned long
382 ia64_fls (unsigned long x)
383 {
384 	long double d = x;
385 	long exp;
386 
387 	exp = ia64_getf_exp(d);
388 	return exp - 0xffff;
389 }
390 
391 /*
392  * Find the last (most significant) bit set.  Returns 0 for x==0 and
393  * bits are numbered from 1..32 (e.g., fls(9) == 4).
394  */
395 static inline int
396 fls (int t)
397 {
398 	unsigned long x = t & 0xffffffffu;
399 
400 	if (!x)
401 		return 0;
402 	x |= x >> 1;
403 	x |= x >> 2;
404 	x |= x >> 4;
405 	x |= x >> 8;
406 	x |= x >> 16;
407 	return ia64_popcnt(x);
408 }
409 
410 /*
411  * Find the last (most significant) bit set.  Undefined for x==0.
412  * Bits are numbered from 0..63 (e.g., __fls(9) == 3).
413  */
414 static inline unsigned long
415 __fls (unsigned long x)
416 {
417 	x |= x >> 1;
418 	x |= x >> 2;
419 	x |= x >> 4;
420 	x |= x >> 8;
421 	x |= x >> 16;
422 	x |= x >> 32;
423 	return ia64_popcnt(x) - 1;
424 }
425 
426 #include <asm-generic/bitops/fls64.h>
427 
428 /*
429  * ffs: find first bit set. This is defined the same way as the libc and
430  * compiler builtin ffs routines, therefore differs in spirit from the above
431  * ffz (man ffs): it operates on "int" values only and the result value is the
432  * bit number + 1.  ffs(0) is defined to return zero.
433  */
434 #define ffs(x)	__builtin_ffs(x)
435 
436 /*
437  * hweightN: returns the hamming weight (i.e. the number
438  * of bits set) of a N-bit word
439  */
440 static __inline__ unsigned long __arch_hweight64(unsigned long x)
441 {
442 	unsigned long result;
443 	result = ia64_popcnt(x);
444 	return result;
445 }
446 
447 #define __arch_hweight32(x) ((unsigned int) __arch_hweight64((x) & 0xfffffffful))
448 #define __arch_hweight16(x) ((unsigned int) __arch_hweight64((x) & 0xfffful))
449 #define __arch_hweight8(x)  ((unsigned int) __arch_hweight64((x) & 0xfful))
450 
451 #include <asm-generic/bitops/const_hweight.h>
452 
453 #endif /* __KERNEL__ */
454 
455 #include <asm-generic/bitops/find.h>
456 
457 #ifdef __KERNEL__
458 
459 #include <asm-generic/bitops/le.h>
460 
461 #include <asm-generic/bitops/ext2-atomic-setbit.h>
462 
463 #include <asm-generic/bitops/sched.h>
464 
465 #endif /* __KERNEL__ */
466 
467 #endif /* _ASM_IA64_BITOPS_H */
468