xref: /openbmc/linux/arch/ia64/include/asm/bitops.h (revision 2a598d0b)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _ASM_IA64_BITOPS_H
3 #define _ASM_IA64_BITOPS_H
4 
5 /*
6  * Copyright (C) 1998-2003 Hewlett-Packard Co
7  *	David Mosberger-Tang <davidm@hpl.hp.com>
8  *
9  * 02/06/02 find_next_bit() and find_first_bit() added from Erich Focht's ia64
10  * O(1) scheduler patch
11  */
12 
13 #ifndef _LINUX_BITOPS_H
14 #error only <linux/bitops.h> can be included directly
15 #endif
16 
17 #include <linux/compiler.h>
18 #include <linux/types.h>
19 #include <asm/intrinsics.h>
20 #include <asm/barrier.h>
21 
22 /**
23  * set_bit - Atomically set a bit in memory
24  * @nr: the bit to set
25  * @addr: the address to start counting from
26  *
27  * This function is atomic and may not be reordered.  See __set_bit()
28  * if you do not require the atomic guarantees.
29  * Note that @nr may be almost arbitrarily large; this function is not
30  * restricted to acting on a single-word quantity.
31  *
32  * The address must be (at least) "long" aligned.
33  * Note that there are driver (e.g., eepro100) which use these operations to
34  * operate on hw-defined data-structures, so we can't easily change these
35  * operations to force a bigger alignment.
36  *
37  * bit 0 is the LSB of addr; bit 32 is the LSB of (addr+1).
38  */
39 static __inline__ void
40 set_bit (int nr, volatile void *addr)
41 {
42 	__u32 bit, old, new;
43 	volatile __u32 *m;
44 	CMPXCHG_BUGCHECK_DECL
45 
46 	m = (volatile __u32 *) addr + (nr >> 5);
47 	bit = 1 << (nr & 31);
48 	do {
49 		CMPXCHG_BUGCHECK(m);
50 		old = *m;
51 		new = old | bit;
52 	} while (cmpxchg_acq(m, old, new) != old);
53 }
54 
55 /**
56  * arch___set_bit - Set a bit in memory
57  * @nr: the bit to set
58  * @addr: the address to start counting from
59  *
60  * Unlike set_bit(), this function is non-atomic and may be reordered.
61  * If it's called on the same region of memory simultaneously, the effect
62  * may be that only one operation succeeds.
63  */
64 static __always_inline void
65 arch___set_bit(unsigned long nr, volatile unsigned long *addr)
66 {
67 	*((__u32 *) addr + (nr >> 5)) |= (1 << (nr & 31));
68 }
69 
70 /**
71  * clear_bit - Clears a bit in memory
72  * @nr: Bit to clear
73  * @addr: Address to start counting from
74  *
75  * clear_bit() is atomic and may not be reordered.  However, it does
76  * not contain a memory barrier, so if it is used for locking purposes,
77  * you should call smp_mb__before_atomic() and/or smp_mb__after_atomic()
78  * in order to ensure changes are visible on other processors.
79  */
80 static __inline__ void
81 clear_bit (int nr, volatile void *addr)
82 {
83 	__u32 mask, old, new;
84 	volatile __u32 *m;
85 	CMPXCHG_BUGCHECK_DECL
86 
87 	m = (volatile __u32 *) addr + (nr >> 5);
88 	mask = ~(1 << (nr & 31));
89 	do {
90 		CMPXCHG_BUGCHECK(m);
91 		old = *m;
92 		new = old & mask;
93 	} while (cmpxchg_acq(m, old, new) != old);
94 }
95 
96 /**
97  * clear_bit_unlock - Clears a bit in memory with release
98  * @nr: Bit to clear
99  * @addr: Address to start counting from
100  *
101  * clear_bit_unlock() is atomic and may not be reordered.  It does
102  * contain a memory barrier suitable for unlock type operations.
103  */
104 static __inline__ void
105 clear_bit_unlock (int nr, volatile void *addr)
106 {
107 	__u32 mask, old, new;
108 	volatile __u32 *m;
109 	CMPXCHG_BUGCHECK_DECL
110 
111 	m = (volatile __u32 *) addr + (nr >> 5);
112 	mask = ~(1 << (nr & 31));
113 	do {
114 		CMPXCHG_BUGCHECK(m);
115 		old = *m;
116 		new = old & mask;
117 	} while (cmpxchg_rel(m, old, new) != old);
118 }
119 
120 /**
121  * __clear_bit_unlock - Non-atomically clears a bit in memory with release
122  * @nr: Bit to clear
123  * @addr: Address to start counting from
124  *
125  * Similarly to clear_bit_unlock, the implementation uses a store
126  * with release semantics. See also arch_spin_unlock().
127  */
128 static __inline__ void
129 __clear_bit_unlock(int nr, void *addr)
130 {
131 	__u32 * const m = (__u32 *) addr + (nr >> 5);
132 	__u32 const new = *m & ~(1 << (nr & 31));
133 
134 	ia64_st4_rel_nta(m, new);
135 }
136 
137 /**
138  * arch___clear_bit - Clears a bit in memory (non-atomic version)
139  * @nr: the bit to clear
140  * @addr: the address to start counting from
141  *
142  * Unlike clear_bit(), this function is non-atomic and may be reordered.
143  * If it's called on the same region of memory simultaneously, the effect
144  * may be that only one operation succeeds.
145  */
146 static __always_inline void
147 arch___clear_bit(unsigned long nr, volatile unsigned long *addr)
148 {
149 	*((__u32 *) addr + (nr >> 5)) &= ~(1 << (nr & 31));
150 }
151 
152 /**
153  * change_bit - Toggle a bit in memory
154  * @nr: Bit to toggle
155  * @addr: Address to start counting from
156  *
157  * change_bit() is atomic and may not be reordered.
158  * Note that @nr may be almost arbitrarily large; this function is not
159  * restricted to acting on a single-word quantity.
160  */
161 static __inline__ void
162 change_bit (int nr, volatile void *addr)
163 {
164 	__u32 bit, old, new;
165 	volatile __u32 *m;
166 	CMPXCHG_BUGCHECK_DECL
167 
168 	m = (volatile __u32 *) addr + (nr >> 5);
169 	bit = (1 << (nr & 31));
170 	do {
171 		CMPXCHG_BUGCHECK(m);
172 		old = *m;
173 		new = old ^ bit;
174 	} while (cmpxchg_acq(m, old, new) != old);
175 }
176 
177 /**
178  * arch___change_bit - Toggle a bit in memory
179  * @nr: the bit to toggle
180  * @addr: the address to start counting from
181  *
182  * Unlike change_bit(), this function is non-atomic and may be reordered.
183  * If it's called on the same region of memory simultaneously, the effect
184  * may be that only one operation succeeds.
185  */
186 static __always_inline void
187 arch___change_bit(unsigned long nr, volatile unsigned long *addr)
188 {
189 	*((__u32 *) addr + (nr >> 5)) ^= (1 << (nr & 31));
190 }
191 
192 /**
193  * test_and_set_bit - Set a bit and return its old value
194  * @nr: Bit to set
195  * @addr: Address to count from
196  *
197  * This operation is atomic and cannot be reordered.
198  * It also implies the acquisition side of the memory barrier.
199  */
200 static __inline__ int
201 test_and_set_bit (int nr, volatile void *addr)
202 {
203 	__u32 bit, old, new;
204 	volatile __u32 *m;
205 	CMPXCHG_BUGCHECK_DECL
206 
207 	m = (volatile __u32 *) addr + (nr >> 5);
208 	bit = 1 << (nr & 31);
209 	do {
210 		CMPXCHG_BUGCHECK(m);
211 		old = *m;
212 		new = old | bit;
213 	} while (cmpxchg_acq(m, old, new) != old);
214 	return (old & bit) != 0;
215 }
216 
217 /**
218  * test_and_set_bit_lock - Set a bit and return its old value for lock
219  * @nr: Bit to set
220  * @addr: Address to count from
221  *
222  * This is the same as test_and_set_bit on ia64
223  */
224 #define test_and_set_bit_lock test_and_set_bit
225 
226 /**
227  * arch___test_and_set_bit - Set a bit and return its old value
228  * @nr: Bit to set
229  * @addr: Address to count from
230  *
231  * This operation is non-atomic and can be reordered.
232  * If two examples of this operation race, one can appear to succeed
233  * but actually fail.  You must protect multiple accesses with a lock.
234  */
235 static __always_inline bool
236 arch___test_and_set_bit(unsigned long nr, volatile unsigned long *addr)
237 {
238 	__u32 *p = (__u32 *) addr + (nr >> 5);
239 	__u32 m = 1 << (nr & 31);
240 	int oldbitset = (*p & m) != 0;
241 
242 	*p |= m;
243 	return oldbitset;
244 }
245 
246 /**
247  * test_and_clear_bit - Clear a bit and return its old value
248  * @nr: Bit to clear
249  * @addr: Address to count from
250  *
251  * This operation is atomic and cannot be reordered.
252  * It also implies the acquisition side of the memory barrier.
253  */
254 static __inline__ int
255 test_and_clear_bit (int nr, volatile void *addr)
256 {
257 	__u32 mask, old, new;
258 	volatile __u32 *m;
259 	CMPXCHG_BUGCHECK_DECL
260 
261 	m = (volatile __u32 *) addr + (nr >> 5);
262 	mask = ~(1 << (nr & 31));
263 	do {
264 		CMPXCHG_BUGCHECK(m);
265 		old = *m;
266 		new = old & mask;
267 	} while (cmpxchg_acq(m, old, new) != old);
268 	return (old & ~mask) != 0;
269 }
270 
271 /**
272  * arch___test_and_clear_bit - Clear a bit and return its old value
273  * @nr: Bit to clear
274  * @addr: Address to count from
275  *
276  * This operation is non-atomic and can be reordered.
277  * If two examples of this operation race, one can appear to succeed
278  * but actually fail.  You must protect multiple accesses with a lock.
279  */
280 static __always_inline bool
281 arch___test_and_clear_bit(unsigned long nr, volatile unsigned long *addr)
282 {
283 	__u32 *p = (__u32 *) addr + (nr >> 5);
284 	__u32 m = 1 << (nr & 31);
285 	int oldbitset = (*p & m) != 0;
286 
287 	*p &= ~m;
288 	return oldbitset;
289 }
290 
291 /**
292  * test_and_change_bit - Change a bit and return its old value
293  * @nr: Bit to change
294  * @addr: Address to count from
295  *
296  * This operation is atomic and cannot be reordered.
297  * It also implies the acquisition side of the memory barrier.
298  */
299 static __inline__ int
300 test_and_change_bit (int nr, volatile void *addr)
301 {
302 	__u32 bit, old, new;
303 	volatile __u32 *m;
304 	CMPXCHG_BUGCHECK_DECL
305 
306 	m = (volatile __u32 *) addr + (nr >> 5);
307 	bit = (1 << (nr & 31));
308 	do {
309 		CMPXCHG_BUGCHECK(m);
310 		old = *m;
311 		new = old ^ bit;
312 	} while (cmpxchg_acq(m, old, new) != old);
313 	return (old & bit) != 0;
314 }
315 
316 /**
317  * arch___test_and_change_bit - Change a bit and return its old value
318  * @nr: Bit to change
319  * @addr: Address to count from
320  *
321  * This operation is non-atomic and can be reordered.
322  */
323 static __always_inline bool
324 arch___test_and_change_bit(unsigned long nr, volatile unsigned long *addr)
325 {
326 	__u32 old, bit = (1 << (nr & 31));
327 	__u32 *m = (__u32 *) addr + (nr >> 5);
328 
329 	old = *m;
330 	*m = old ^ bit;
331 	return (old & bit) != 0;
332 }
333 
334 #define arch_test_bit generic_test_bit
335 #define arch_test_bit_acquire generic_test_bit_acquire
336 
337 /**
338  * ffz - find the first zero bit in a long word
339  * @x: The long word to find the bit in
340  *
341  * Returns the bit-number (0..63) of the first (least significant) zero bit.
342  * Undefined if no zero exists, so code should check against ~0UL first...
343  */
344 static inline unsigned long
345 ffz (unsigned long x)
346 {
347 	unsigned long result;
348 
349 	result = ia64_popcnt(x & (~x - 1));
350 	return result;
351 }
352 
353 /**
354  * __ffs - find first bit in word.
355  * @x: The word to search
356  *
357  * Undefined if no bit exists, so code should check against 0 first.
358  */
359 static __inline__ unsigned long
360 __ffs (unsigned long x)
361 {
362 	unsigned long result;
363 
364 	result = ia64_popcnt((x-1) & ~x);
365 	return result;
366 }
367 
368 #ifdef __KERNEL__
369 
370 /*
371  * Return bit number of last (most-significant) bit set.  Undefined
372  * for x==0.  Bits are numbered from 0..63 (e.g., ia64_fls(9) == 3).
373  */
374 static inline unsigned long
375 ia64_fls (unsigned long x)
376 {
377 	long double d = x;
378 	long exp;
379 
380 	exp = ia64_getf_exp(d);
381 	return exp - 0xffff;
382 }
383 
384 /*
385  * Find the last (most significant) bit set.  Returns 0 for x==0 and
386  * bits are numbered from 1..32 (e.g., fls(9) == 4).
387  */
388 static inline int fls(unsigned int t)
389 {
390 	unsigned long x = t & 0xffffffffu;
391 
392 	if (!x)
393 		return 0;
394 	x |= x >> 1;
395 	x |= x >> 2;
396 	x |= x >> 4;
397 	x |= x >> 8;
398 	x |= x >> 16;
399 	return ia64_popcnt(x);
400 }
401 
402 /*
403  * Find the last (most significant) bit set.  Undefined for x==0.
404  * Bits are numbered from 0..63 (e.g., __fls(9) == 3).
405  */
406 static inline unsigned long
407 __fls (unsigned long x)
408 {
409 	x |= x >> 1;
410 	x |= x >> 2;
411 	x |= x >> 4;
412 	x |= x >> 8;
413 	x |= x >> 16;
414 	x |= x >> 32;
415 	return ia64_popcnt(x) - 1;
416 }
417 
418 #include <asm-generic/bitops/fls64.h>
419 
420 #include <asm-generic/bitops/builtin-ffs.h>
421 
422 /*
423  * hweightN: returns the hamming weight (i.e. the number
424  * of bits set) of a N-bit word
425  */
426 static __inline__ unsigned long __arch_hweight64(unsigned long x)
427 {
428 	unsigned long result;
429 	result = ia64_popcnt(x);
430 	return result;
431 }
432 
433 #define __arch_hweight32(x) ((unsigned int) __arch_hweight64((x) & 0xfffffffful))
434 #define __arch_hweight16(x) ((unsigned int) __arch_hweight64((x) & 0xfffful))
435 #define __arch_hweight8(x)  ((unsigned int) __arch_hweight64((x) & 0xfful))
436 
437 #include <asm-generic/bitops/const_hweight.h>
438 
439 #endif /* __KERNEL__ */
440 
441 #ifdef __KERNEL__
442 
443 #include <asm-generic/bitops/non-instrumented-non-atomic.h>
444 
445 #include <asm-generic/bitops/le.h>
446 
447 #include <asm-generic/bitops/ext2-atomic-setbit.h>
448 
449 #include <asm-generic/bitops/sched.h>
450 
451 #endif /* __KERNEL__ */
452 
453 #endif /* _ASM_IA64_BITOPS_H */
454