1 /* SPDX-License-Identifier: GPL-2.0-or-later */
2 /*
3  *	Definitions for the 'struct ptr_ring' datastructure.
4  *
5  *	Author:
6  *		Michael S. Tsirkin <mst@redhat.com>
7  *
8  *	Copyright (C) 2016 Red Hat, Inc.
9  *
10  *	This is a limited-size FIFO maintaining pointers in FIFO order, with
11  *	one CPU producing entries and another consuming entries from a FIFO.
12  *
13  *	This implementation tries to minimize cache-contention when there is a
14  *	single producer and a single consumer CPU.
15  */
16 
17 #ifndef _LINUX_PTR_RING_H
18 #define _LINUX_PTR_RING_H 1
19 
20 #ifdef __KERNEL__
21 #include <linux/spinlock.h>
22 #include <linux/cache.h>
23 #include <linux/types.h>
24 #include <linux/compiler.h>
25 #include <linux/slab.h>
26 #include <asm/errno.h>
27 #endif
28 
29 struct ptr_ring {
30 	int producer ____cacheline_aligned_in_smp;
31 	spinlock_t producer_lock;
32 	int consumer_head ____cacheline_aligned_in_smp; /* next valid entry */
33 	int consumer_tail; /* next entry to invalidate */
34 	spinlock_t consumer_lock;
35 	/* Shared consumer/producer data */
36 	/* Read-only by both the producer and the consumer */
37 	int size ____cacheline_aligned_in_smp; /* max entries in queue */
38 	int batch; /* number of entries to consume in a batch */
39 	void **queue;
40 };
41 
42 /* Note: callers invoking this in a loop must use a compiler barrier,
43  * for example cpu_relax().
44  *
45  * NB: this is unlike __ptr_ring_empty in that callers must hold producer_lock:
46  * see e.g. ptr_ring_full.
47  */
__ptr_ring_full(struct ptr_ring * r)48 static inline bool __ptr_ring_full(struct ptr_ring *r)
49 {
50 	return r->queue[r->producer];
51 }
52 
ptr_ring_full(struct ptr_ring * r)53 static inline bool ptr_ring_full(struct ptr_ring *r)
54 {
55 	bool ret;
56 
57 	spin_lock(&r->producer_lock);
58 	ret = __ptr_ring_full(r);
59 	spin_unlock(&r->producer_lock);
60 
61 	return ret;
62 }
63 
ptr_ring_full_irq(struct ptr_ring * r)64 static inline bool ptr_ring_full_irq(struct ptr_ring *r)
65 {
66 	bool ret;
67 
68 	spin_lock_irq(&r->producer_lock);
69 	ret = __ptr_ring_full(r);
70 	spin_unlock_irq(&r->producer_lock);
71 
72 	return ret;
73 }
74 
ptr_ring_full_any(struct ptr_ring * r)75 static inline bool ptr_ring_full_any(struct ptr_ring *r)
76 {
77 	unsigned long flags;
78 	bool ret;
79 
80 	spin_lock_irqsave(&r->producer_lock, flags);
81 	ret = __ptr_ring_full(r);
82 	spin_unlock_irqrestore(&r->producer_lock, flags);
83 
84 	return ret;
85 }
86 
ptr_ring_full_bh(struct ptr_ring * r)87 static inline bool ptr_ring_full_bh(struct ptr_ring *r)
88 {
89 	bool ret;
90 
91 	spin_lock_bh(&r->producer_lock);
92 	ret = __ptr_ring_full(r);
93 	spin_unlock_bh(&r->producer_lock);
94 
95 	return ret;
96 }
97 
98 /* Note: callers invoking this in a loop must use a compiler barrier,
99  * for example cpu_relax(). Callers must hold producer_lock.
100  * Callers are responsible for making sure pointer that is being queued
101  * points to a valid data.
102  */
__ptr_ring_produce(struct ptr_ring * r,void * ptr)103 static inline int __ptr_ring_produce(struct ptr_ring *r, void *ptr)
104 {
105 	if (unlikely(!r->size) || r->queue[r->producer])
106 		return -ENOSPC;
107 
108 	/* Make sure the pointer we are storing points to a valid data. */
109 	/* Pairs with smp_read_barrier_depends in __ptr_ring_consume. */
110 	smp_wmb();
111 
112 	WRITE_ONCE(r->queue[r->producer++], ptr);
113 	if (unlikely(r->producer >= r->size))
114 		r->producer = 0;
115 	return 0;
116 }
117 
118 /*
119  * Note: resize (below) nests producer lock within consumer lock, so if you
120  * consume in interrupt or BH context, you must disable interrupts/BH when
121  * calling this.
122  */
ptr_ring_produce(struct ptr_ring * r,void * ptr)123 static inline int ptr_ring_produce(struct ptr_ring *r, void *ptr)
124 {
125 	int ret;
126 
127 	spin_lock(&r->producer_lock);
128 	ret = __ptr_ring_produce(r, ptr);
129 	spin_unlock(&r->producer_lock);
130 
131 	return ret;
132 }
133 
ptr_ring_produce_irq(struct ptr_ring * r,void * ptr)134 static inline int ptr_ring_produce_irq(struct ptr_ring *r, void *ptr)
135 {
136 	int ret;
137 
138 	spin_lock_irq(&r->producer_lock);
139 	ret = __ptr_ring_produce(r, ptr);
140 	spin_unlock_irq(&r->producer_lock);
141 
142 	return ret;
143 }
144 
ptr_ring_produce_any(struct ptr_ring * r,void * ptr)145 static inline int ptr_ring_produce_any(struct ptr_ring *r, void *ptr)
146 {
147 	unsigned long flags;
148 	int ret;
149 
150 	spin_lock_irqsave(&r->producer_lock, flags);
151 	ret = __ptr_ring_produce(r, ptr);
152 	spin_unlock_irqrestore(&r->producer_lock, flags);
153 
154 	return ret;
155 }
156 
ptr_ring_produce_bh(struct ptr_ring * r,void * ptr)157 static inline int ptr_ring_produce_bh(struct ptr_ring *r, void *ptr)
158 {
159 	int ret;
160 
161 	spin_lock_bh(&r->producer_lock);
162 	ret = __ptr_ring_produce(r, ptr);
163 	spin_unlock_bh(&r->producer_lock);
164 
165 	return ret;
166 }
167 
__ptr_ring_peek(struct ptr_ring * r)168 static inline void *__ptr_ring_peek(struct ptr_ring *r)
169 {
170 	if (likely(r->size))
171 		return READ_ONCE(r->queue[r->consumer_head]);
172 	return NULL;
173 }
174 
175 /*
176  * Test ring empty status without taking any locks.
177  *
178  * NB: This is only safe to call if ring is never resized.
179  *
180  * However, if some other CPU consumes ring entries at the same time, the value
181  * returned is not guaranteed to be correct.
182  *
183  * In this case - to avoid incorrectly detecting the ring
184  * as empty - the CPU consuming the ring entries is responsible
185  * for either consuming all ring entries until the ring is empty,
186  * or synchronizing with some other CPU and causing it to
187  * re-test __ptr_ring_empty and/or consume the ring enteries
188  * after the synchronization point.
189  *
190  * Note: callers invoking this in a loop must use a compiler barrier,
191  * for example cpu_relax().
192  */
__ptr_ring_empty(struct ptr_ring * r)193 static inline bool __ptr_ring_empty(struct ptr_ring *r)
194 {
195 	if (likely(r->size))
196 		return !r->queue[READ_ONCE(r->consumer_head)];
197 	return true;
198 }
199 
ptr_ring_empty(struct ptr_ring * r)200 static inline bool ptr_ring_empty(struct ptr_ring *r)
201 {
202 	bool ret;
203 
204 	spin_lock(&r->consumer_lock);
205 	ret = __ptr_ring_empty(r);
206 	spin_unlock(&r->consumer_lock);
207 
208 	return ret;
209 }
210 
ptr_ring_empty_irq(struct ptr_ring * r)211 static inline bool ptr_ring_empty_irq(struct ptr_ring *r)
212 {
213 	bool ret;
214 
215 	spin_lock_irq(&r->consumer_lock);
216 	ret = __ptr_ring_empty(r);
217 	spin_unlock_irq(&r->consumer_lock);
218 
219 	return ret;
220 }
221 
ptr_ring_empty_any(struct ptr_ring * r)222 static inline bool ptr_ring_empty_any(struct ptr_ring *r)
223 {
224 	unsigned long flags;
225 	bool ret;
226 
227 	spin_lock_irqsave(&r->consumer_lock, flags);
228 	ret = __ptr_ring_empty(r);
229 	spin_unlock_irqrestore(&r->consumer_lock, flags);
230 
231 	return ret;
232 }
233 
ptr_ring_empty_bh(struct ptr_ring * r)234 static inline bool ptr_ring_empty_bh(struct ptr_ring *r)
235 {
236 	bool ret;
237 
238 	spin_lock_bh(&r->consumer_lock);
239 	ret = __ptr_ring_empty(r);
240 	spin_unlock_bh(&r->consumer_lock);
241 
242 	return ret;
243 }
244 
245 /* Must only be called after __ptr_ring_peek returned !NULL */
__ptr_ring_discard_one(struct ptr_ring * r)246 static inline void __ptr_ring_discard_one(struct ptr_ring *r)
247 {
248 	/* Fundamentally, what we want to do is update consumer
249 	 * index and zero out the entry so producer can reuse it.
250 	 * Doing it naively at each consume would be as simple as:
251 	 *       consumer = r->consumer;
252 	 *       r->queue[consumer++] = NULL;
253 	 *       if (unlikely(consumer >= r->size))
254 	 *               consumer = 0;
255 	 *       r->consumer = consumer;
256 	 * but that is suboptimal when the ring is full as producer is writing
257 	 * out new entries in the same cache line.  Defer these updates until a
258 	 * batch of entries has been consumed.
259 	 */
260 	/* Note: we must keep consumer_head valid at all times for __ptr_ring_empty
261 	 * to work correctly.
262 	 */
263 	int consumer_head = r->consumer_head;
264 	int head = consumer_head++;
265 
266 	/* Once we have processed enough entries invalidate them in
267 	 * the ring all at once so producer can reuse their space in the ring.
268 	 * We also do this when we reach end of the ring - not mandatory
269 	 * but helps keep the implementation simple.
270 	 */
271 	if (unlikely(consumer_head - r->consumer_tail >= r->batch ||
272 		     consumer_head >= r->size)) {
273 		/* Zero out entries in the reverse order: this way we touch the
274 		 * cache line that producer might currently be reading the last;
275 		 * producer won't make progress and touch other cache lines
276 		 * besides the first one until we write out all entries.
277 		 */
278 		while (likely(head >= r->consumer_tail))
279 			r->queue[head--] = NULL;
280 		r->consumer_tail = consumer_head;
281 	}
282 	if (unlikely(consumer_head >= r->size)) {
283 		consumer_head = 0;
284 		r->consumer_tail = 0;
285 	}
286 	/* matching READ_ONCE in __ptr_ring_empty for lockless tests */
287 	WRITE_ONCE(r->consumer_head, consumer_head);
288 }
289 
__ptr_ring_consume(struct ptr_ring * r)290 static inline void *__ptr_ring_consume(struct ptr_ring *r)
291 {
292 	void *ptr;
293 
294 	/* The READ_ONCE in __ptr_ring_peek guarantees that anyone
295 	 * accessing data through the pointer is up to date. Pairs
296 	 * with smp_wmb in __ptr_ring_produce.
297 	 */
298 	ptr = __ptr_ring_peek(r);
299 	if (ptr)
300 		__ptr_ring_discard_one(r);
301 
302 	return ptr;
303 }
304 
__ptr_ring_consume_batched(struct ptr_ring * r,void ** array,int n)305 static inline int __ptr_ring_consume_batched(struct ptr_ring *r,
306 					     void **array, int n)
307 {
308 	void *ptr;
309 	int i;
310 
311 	for (i = 0; i < n; i++) {
312 		ptr = __ptr_ring_consume(r);
313 		if (!ptr)
314 			break;
315 		array[i] = ptr;
316 	}
317 
318 	return i;
319 }
320 
321 /*
322  * Note: resize (below) nests producer lock within consumer lock, so if you
323  * call this in interrupt or BH context, you must disable interrupts/BH when
324  * producing.
325  */
ptr_ring_consume(struct ptr_ring * r)326 static inline void *ptr_ring_consume(struct ptr_ring *r)
327 {
328 	void *ptr;
329 
330 	spin_lock(&r->consumer_lock);
331 	ptr = __ptr_ring_consume(r);
332 	spin_unlock(&r->consumer_lock);
333 
334 	return ptr;
335 }
336 
ptr_ring_consume_irq(struct ptr_ring * r)337 static inline void *ptr_ring_consume_irq(struct ptr_ring *r)
338 {
339 	void *ptr;
340 
341 	spin_lock_irq(&r->consumer_lock);
342 	ptr = __ptr_ring_consume(r);
343 	spin_unlock_irq(&r->consumer_lock);
344 
345 	return ptr;
346 }
347 
ptr_ring_consume_any(struct ptr_ring * r)348 static inline void *ptr_ring_consume_any(struct ptr_ring *r)
349 {
350 	unsigned long flags;
351 	void *ptr;
352 
353 	spin_lock_irqsave(&r->consumer_lock, flags);
354 	ptr = __ptr_ring_consume(r);
355 	spin_unlock_irqrestore(&r->consumer_lock, flags);
356 
357 	return ptr;
358 }
359 
ptr_ring_consume_bh(struct ptr_ring * r)360 static inline void *ptr_ring_consume_bh(struct ptr_ring *r)
361 {
362 	void *ptr;
363 
364 	spin_lock_bh(&r->consumer_lock);
365 	ptr = __ptr_ring_consume(r);
366 	spin_unlock_bh(&r->consumer_lock);
367 
368 	return ptr;
369 }
370 
ptr_ring_consume_batched(struct ptr_ring * r,void ** array,int n)371 static inline int ptr_ring_consume_batched(struct ptr_ring *r,
372 					   void **array, int n)
373 {
374 	int ret;
375 
376 	spin_lock(&r->consumer_lock);
377 	ret = __ptr_ring_consume_batched(r, array, n);
378 	spin_unlock(&r->consumer_lock);
379 
380 	return ret;
381 }
382 
ptr_ring_consume_batched_irq(struct ptr_ring * r,void ** array,int n)383 static inline int ptr_ring_consume_batched_irq(struct ptr_ring *r,
384 					       void **array, int n)
385 {
386 	int ret;
387 
388 	spin_lock_irq(&r->consumer_lock);
389 	ret = __ptr_ring_consume_batched(r, array, n);
390 	spin_unlock_irq(&r->consumer_lock);
391 
392 	return ret;
393 }
394 
ptr_ring_consume_batched_any(struct ptr_ring * r,void ** array,int n)395 static inline int ptr_ring_consume_batched_any(struct ptr_ring *r,
396 					       void **array, int n)
397 {
398 	unsigned long flags;
399 	int ret;
400 
401 	spin_lock_irqsave(&r->consumer_lock, flags);
402 	ret = __ptr_ring_consume_batched(r, array, n);
403 	spin_unlock_irqrestore(&r->consumer_lock, flags);
404 
405 	return ret;
406 }
407 
ptr_ring_consume_batched_bh(struct ptr_ring * r,void ** array,int n)408 static inline int ptr_ring_consume_batched_bh(struct ptr_ring *r,
409 					      void **array, int n)
410 {
411 	int ret;
412 
413 	spin_lock_bh(&r->consumer_lock);
414 	ret = __ptr_ring_consume_batched(r, array, n);
415 	spin_unlock_bh(&r->consumer_lock);
416 
417 	return ret;
418 }
419 
420 /* Cast to structure type and call a function without discarding from FIFO.
421  * Function must return a value.
422  * Callers must take consumer_lock.
423  */
424 #define __PTR_RING_PEEK_CALL(r, f) ((f)(__ptr_ring_peek(r)))
425 
426 #define PTR_RING_PEEK_CALL(r, f) ({ \
427 	typeof((f)(NULL)) __PTR_RING_PEEK_CALL_v; \
428 	\
429 	spin_lock(&(r)->consumer_lock); \
430 	__PTR_RING_PEEK_CALL_v = __PTR_RING_PEEK_CALL(r, f); \
431 	spin_unlock(&(r)->consumer_lock); \
432 	__PTR_RING_PEEK_CALL_v; \
433 })
434 
435 #define PTR_RING_PEEK_CALL_IRQ(r, f) ({ \
436 	typeof((f)(NULL)) __PTR_RING_PEEK_CALL_v; \
437 	\
438 	spin_lock_irq(&(r)->consumer_lock); \
439 	__PTR_RING_PEEK_CALL_v = __PTR_RING_PEEK_CALL(r, f); \
440 	spin_unlock_irq(&(r)->consumer_lock); \
441 	__PTR_RING_PEEK_CALL_v; \
442 })
443 
444 #define PTR_RING_PEEK_CALL_BH(r, f) ({ \
445 	typeof((f)(NULL)) __PTR_RING_PEEK_CALL_v; \
446 	\
447 	spin_lock_bh(&(r)->consumer_lock); \
448 	__PTR_RING_PEEK_CALL_v = __PTR_RING_PEEK_CALL(r, f); \
449 	spin_unlock_bh(&(r)->consumer_lock); \
450 	__PTR_RING_PEEK_CALL_v; \
451 })
452 
453 #define PTR_RING_PEEK_CALL_ANY(r, f) ({ \
454 	typeof((f)(NULL)) __PTR_RING_PEEK_CALL_v; \
455 	unsigned long __PTR_RING_PEEK_CALL_f;\
456 	\
457 	spin_lock_irqsave(&(r)->consumer_lock, __PTR_RING_PEEK_CALL_f); \
458 	__PTR_RING_PEEK_CALL_v = __PTR_RING_PEEK_CALL(r, f); \
459 	spin_unlock_irqrestore(&(r)->consumer_lock, __PTR_RING_PEEK_CALL_f); \
460 	__PTR_RING_PEEK_CALL_v; \
461 })
462 
463 /* Not all gfp_t flags (besides GFP_KERNEL) are allowed. See
464  * documentation for vmalloc for which of them are legal.
465  */
__ptr_ring_init_queue_alloc(unsigned int size,gfp_t gfp)466 static inline void **__ptr_ring_init_queue_alloc(unsigned int size, gfp_t gfp)
467 {
468 	if (size > KMALLOC_MAX_SIZE / sizeof(void *))
469 		return NULL;
470 	return kvmalloc_array(size, sizeof(void *), gfp | __GFP_ZERO);
471 }
472 
__ptr_ring_set_size(struct ptr_ring * r,int size)473 static inline void __ptr_ring_set_size(struct ptr_ring *r, int size)
474 {
475 	r->size = size;
476 	r->batch = SMP_CACHE_BYTES * 2 / sizeof(*(r->queue));
477 	/* We need to set batch at least to 1 to make logic
478 	 * in __ptr_ring_discard_one work correctly.
479 	 * Batching too much (because ring is small) would cause a lot of
480 	 * burstiness. Needs tuning, for now disable batching.
481 	 */
482 	if (r->batch > r->size / 2 || !r->batch)
483 		r->batch = 1;
484 }
485 
ptr_ring_init(struct ptr_ring * r,int size,gfp_t gfp)486 static inline int ptr_ring_init(struct ptr_ring *r, int size, gfp_t gfp)
487 {
488 	r->queue = __ptr_ring_init_queue_alloc(size, gfp);
489 	if (!r->queue)
490 		return -ENOMEM;
491 
492 	__ptr_ring_set_size(r, size);
493 	r->producer = r->consumer_head = r->consumer_tail = 0;
494 	spin_lock_init(&r->producer_lock);
495 	spin_lock_init(&r->consumer_lock);
496 
497 	return 0;
498 }
499 
500 /*
501  * Return entries into ring. Destroy entries that don't fit.
502  *
503  * Note: this is expected to be a rare slow path operation.
504  *
505  * Note: producer lock is nested within consumer lock, so if you
506  * resize you must make sure all uses nest correctly.
507  * In particular if you consume ring in interrupt or BH context, you must
508  * disable interrupts/BH when doing so.
509  */
ptr_ring_unconsume(struct ptr_ring * r,void ** batch,int n,void (* destroy)(void *))510 static inline void ptr_ring_unconsume(struct ptr_ring *r, void **batch, int n,
511 				      void (*destroy)(void *))
512 {
513 	unsigned long flags;
514 	int head;
515 
516 	spin_lock_irqsave(&r->consumer_lock, flags);
517 	spin_lock(&r->producer_lock);
518 
519 	if (!r->size)
520 		goto done;
521 
522 	/*
523 	 * Clean out buffered entries (for simplicity). This way following code
524 	 * can test entries for NULL and if not assume they are valid.
525 	 */
526 	head = r->consumer_head - 1;
527 	while (likely(head >= r->consumer_tail))
528 		r->queue[head--] = NULL;
529 	r->consumer_tail = r->consumer_head;
530 
531 	/*
532 	 * Go over entries in batch, start moving head back and copy entries.
533 	 * Stop when we run into previously unconsumed entries.
534 	 */
535 	while (n) {
536 		head = r->consumer_head - 1;
537 		if (head < 0)
538 			head = r->size - 1;
539 		if (r->queue[head]) {
540 			/* This batch entry will have to be destroyed. */
541 			goto done;
542 		}
543 		r->queue[head] = batch[--n];
544 		r->consumer_tail = head;
545 		/* matching READ_ONCE in __ptr_ring_empty for lockless tests */
546 		WRITE_ONCE(r->consumer_head, head);
547 	}
548 
549 done:
550 	/* Destroy all entries left in the batch. */
551 	while (n)
552 		destroy(batch[--n]);
553 	spin_unlock(&r->producer_lock);
554 	spin_unlock_irqrestore(&r->consumer_lock, flags);
555 }
556 
__ptr_ring_swap_queue(struct ptr_ring * r,void ** queue,int size,gfp_t gfp,void (* destroy)(void *))557 static inline void **__ptr_ring_swap_queue(struct ptr_ring *r, void **queue,
558 					   int size, gfp_t gfp,
559 					   void (*destroy)(void *))
560 {
561 	int producer = 0;
562 	void **old;
563 	void *ptr;
564 
565 	while ((ptr = __ptr_ring_consume(r)))
566 		if (producer < size)
567 			queue[producer++] = ptr;
568 		else if (destroy)
569 			destroy(ptr);
570 
571 	if (producer >= size)
572 		producer = 0;
573 	__ptr_ring_set_size(r, size);
574 	r->producer = producer;
575 	r->consumer_head = 0;
576 	r->consumer_tail = 0;
577 	old = r->queue;
578 	r->queue = queue;
579 
580 	return old;
581 }
582 
583 /*
584  * Note: producer lock is nested within consumer lock, so if you
585  * resize you must make sure all uses nest correctly.
586  * In particular if you consume ring in interrupt or BH context, you must
587  * disable interrupts/BH when doing so.
588  */
ptr_ring_resize(struct ptr_ring * r,int size,gfp_t gfp,void (* destroy)(void *))589 static inline int ptr_ring_resize(struct ptr_ring *r, int size, gfp_t gfp,
590 				  void (*destroy)(void *))
591 {
592 	unsigned long flags;
593 	void **queue = __ptr_ring_init_queue_alloc(size, gfp);
594 	void **old;
595 
596 	if (!queue)
597 		return -ENOMEM;
598 
599 	spin_lock_irqsave(&(r)->consumer_lock, flags);
600 	spin_lock(&(r)->producer_lock);
601 
602 	old = __ptr_ring_swap_queue(r, queue, size, gfp, destroy);
603 
604 	spin_unlock(&(r)->producer_lock);
605 	spin_unlock_irqrestore(&(r)->consumer_lock, flags);
606 
607 	kvfree(old);
608 
609 	return 0;
610 }
611 
612 /*
613  * Note: producer lock is nested within consumer lock, so if you
614  * resize you must make sure all uses nest correctly.
615  * In particular if you consume ring in interrupt or BH context, you must
616  * disable interrupts/BH when doing so.
617  */
ptr_ring_resize_multiple(struct ptr_ring ** rings,unsigned int nrings,int size,gfp_t gfp,void (* destroy)(void *))618 static inline int ptr_ring_resize_multiple(struct ptr_ring **rings,
619 					   unsigned int nrings,
620 					   int size,
621 					   gfp_t gfp, void (*destroy)(void *))
622 {
623 	unsigned long flags;
624 	void ***queues;
625 	int i;
626 
627 	queues = kmalloc_array(nrings, sizeof(*queues), gfp);
628 	if (!queues)
629 		goto noqueues;
630 
631 	for (i = 0; i < nrings; ++i) {
632 		queues[i] = __ptr_ring_init_queue_alloc(size, gfp);
633 		if (!queues[i])
634 			goto nomem;
635 	}
636 
637 	for (i = 0; i < nrings; ++i) {
638 		spin_lock_irqsave(&(rings[i])->consumer_lock, flags);
639 		spin_lock(&(rings[i])->producer_lock);
640 		queues[i] = __ptr_ring_swap_queue(rings[i], queues[i],
641 						  size, gfp, destroy);
642 		spin_unlock(&(rings[i])->producer_lock);
643 		spin_unlock_irqrestore(&(rings[i])->consumer_lock, flags);
644 	}
645 
646 	for (i = 0; i < nrings; ++i)
647 		kvfree(queues[i]);
648 
649 	kfree(queues);
650 
651 	return 0;
652 
653 nomem:
654 	while (--i >= 0)
655 		kvfree(queues[i]);
656 
657 	kfree(queues);
658 
659 noqueues:
660 	return -ENOMEM;
661 }
662 
ptr_ring_cleanup(struct ptr_ring * r,void (* destroy)(void *))663 static inline void ptr_ring_cleanup(struct ptr_ring *r, void (*destroy)(void *))
664 {
665 	void *ptr;
666 
667 	if (destroy)
668 		while ((ptr = ptr_ring_consume(r)))
669 			destroy(ptr);
670 	kvfree(r->queue);
671 }
672 
673 #endif /* _LINUX_PTR_RING_H  */
674