xref: /linux/mm/mempool.c (revision 69050f8d6d075dc01af7a5f2f550a8067510366f)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  memory buffer pool support. Such pools are mostly used
4  *  for guaranteed, deadlock-free memory allocations during
5  *  extreme VM load.
6  *
7  *  started by Ingo Molnar, Copyright (C) 2001
8  *  debugging by David Rientjes, Copyright (C) 2015
9  */
10 #include <linux/fault-inject.h>
11 #include <linux/mm.h>
12 #include <linux/slab.h>
13 #include <linux/highmem.h>
14 #include <linux/kasan.h>
15 #include <linux/kmemleak.h>
16 #include <linux/export.h>
17 #include <linux/mempool.h>
18 #include <linux/writeback.h>
19 #include "slab.h"
20 
21 static DECLARE_FAULT_ATTR(fail_mempool_alloc);
22 static DECLARE_FAULT_ATTR(fail_mempool_alloc_bulk);
23 
24 static int __init mempool_faul_inject_init(void)
25 {
26 	int error;
27 
28 	error = PTR_ERR_OR_ZERO(fault_create_debugfs_attr("fail_mempool_alloc",
29 			NULL, &fail_mempool_alloc));
30 	if (error)
31 		return error;
32 
33 	/* booting will fail on error return here, don't bother to cleanup */
34 	return PTR_ERR_OR_ZERO(
35 		fault_create_debugfs_attr("fail_mempool_alloc_bulk", NULL,
36 		&fail_mempool_alloc_bulk));
37 }
38 late_initcall(mempool_faul_inject_init);
39 
40 #ifdef CONFIG_SLUB_DEBUG_ON
41 static void poison_error(struct mempool *pool, void *element, size_t size,
42 			 size_t byte)
43 {
44 	const int nr = pool->curr_nr;
45 	const int start = max_t(int, byte - (BITS_PER_LONG / 8), 0);
46 	const int end = min_t(int, byte + (BITS_PER_LONG / 8), size);
47 	int i;
48 
49 	pr_err("BUG: mempool element poison mismatch\n");
50 	pr_err("Mempool %p size %zu\n", pool, size);
51 	pr_err(" nr=%d @ %p: %s0x", nr, element, start > 0 ? "... " : "");
52 	for (i = start; i < end; i++)
53 		pr_cont("%x ", *(u8 *)(element + i));
54 	pr_cont("%s\n", end < size ? "..." : "");
55 	dump_stack();
56 }
57 
58 static void __check_element(struct mempool *pool, void *element, size_t size)
59 {
60 	u8 *obj = element;
61 	size_t i;
62 
63 	for (i = 0; i < size; i++) {
64 		u8 exp = (i < size - 1) ? POISON_FREE : POISON_END;
65 
66 		if (obj[i] != exp) {
67 			poison_error(pool, element, size, i);
68 			return;
69 		}
70 	}
71 	memset(obj, POISON_INUSE, size);
72 }
73 
74 static void check_element(struct mempool *pool, void *element)
75 {
76 	/* Skip checking: KASAN might save its metadata in the element. */
77 	if (kasan_enabled())
78 		return;
79 
80 	/* Mempools backed by slab allocator */
81 	if (pool->free == mempool_kfree) {
82 		__check_element(pool, element, (size_t)pool->pool_data);
83 	} else if (pool->free == mempool_free_slab) {
84 		__check_element(pool, element, kmem_cache_size(pool->pool_data));
85 	} else if (pool->free == mempool_free_pages) {
86 		/* Mempools backed by page allocator */
87 		int order = (int)(long)pool->pool_data;
88 
89 #ifdef CONFIG_HIGHMEM
90 		for (int i = 0; i < (1 << order); i++) {
91 			struct page *page = (struct page *)element;
92 			void *addr = kmap_local_page(page + i);
93 
94 			__check_element(pool, addr, PAGE_SIZE);
95 			kunmap_local(addr);
96 		}
97 #else
98 		void *addr = page_address((struct page *)element);
99 
100 		__check_element(pool, addr, PAGE_SIZE << order);
101 #endif
102 	}
103 }
104 
105 static void __poison_element(void *element, size_t size)
106 {
107 	u8 *obj = element;
108 
109 	memset(obj, POISON_FREE, size - 1);
110 	obj[size - 1] = POISON_END;
111 }
112 
113 static void poison_element(struct mempool *pool, void *element)
114 {
115 	/* Skip poisoning: KASAN might save its metadata in the element. */
116 	if (kasan_enabled())
117 		return;
118 
119 	/* Mempools backed by slab allocator */
120 	if (pool->alloc == mempool_kmalloc) {
121 		__poison_element(element, (size_t)pool->pool_data);
122 	} else if (pool->alloc == mempool_alloc_slab) {
123 		__poison_element(element, kmem_cache_size(pool->pool_data));
124 	} else if (pool->alloc == mempool_alloc_pages) {
125 		/* Mempools backed by page allocator */
126 		int order = (int)(long)pool->pool_data;
127 
128 #ifdef CONFIG_HIGHMEM
129 		for (int i = 0; i < (1 << order); i++) {
130 			struct page *page = (struct page *)element;
131 			void *addr = kmap_local_page(page + i);
132 
133 			__poison_element(addr, PAGE_SIZE);
134 			kunmap_local(addr);
135 		}
136 #else
137 		void *addr = page_address((struct page *)element);
138 
139 		__poison_element(addr, PAGE_SIZE << order);
140 #endif
141 	}
142 }
143 #else /* CONFIG_SLUB_DEBUG_ON */
144 static inline void check_element(struct mempool *pool, void *element)
145 {
146 }
147 static inline void poison_element(struct mempool *pool, void *element)
148 {
149 }
150 #endif /* CONFIG_SLUB_DEBUG_ON */
151 
152 static __always_inline bool kasan_poison_element(struct mempool *pool,
153 		void *element)
154 {
155 	if (pool->alloc == mempool_alloc_slab || pool->alloc == mempool_kmalloc)
156 		return kasan_mempool_poison_object(element);
157 	else if (pool->alloc == mempool_alloc_pages)
158 		return kasan_mempool_poison_pages(element,
159 						(unsigned long)pool->pool_data);
160 	return true;
161 }
162 
163 static void kasan_unpoison_element(struct mempool *pool, void *element)
164 {
165 	if (pool->alloc == mempool_kmalloc)
166 		kasan_mempool_unpoison_object(element, (size_t)pool->pool_data);
167 	else if (pool->alloc == mempool_alloc_slab)
168 		kasan_mempool_unpoison_object(element,
169 					      kmem_cache_size(pool->pool_data));
170 	else if (pool->alloc == mempool_alloc_pages)
171 		kasan_mempool_unpoison_pages(element,
172 					     (unsigned long)pool->pool_data);
173 }
174 
175 static __always_inline void add_element(struct mempool *pool, void *element)
176 {
177 	BUG_ON(pool->min_nr != 0 && pool->curr_nr >= pool->min_nr);
178 	poison_element(pool, element);
179 	if (kasan_poison_element(pool, element))
180 		pool->elements[pool->curr_nr++] = element;
181 }
182 
183 static void *remove_element(struct mempool *pool)
184 {
185 	void *element = pool->elements[--pool->curr_nr];
186 
187 	BUG_ON(pool->curr_nr < 0);
188 	kasan_unpoison_element(pool, element);
189 	check_element(pool, element);
190 	return element;
191 }
192 
193 /**
194  * mempool_exit - exit a mempool initialized with mempool_init()
195  * @pool:      pointer to the memory pool which was initialized with
196  *             mempool_init().
197  *
198  * Free all reserved elements in @pool and @pool itself.  This function
199  * only sleeps if the free_fn() function sleeps.
200  *
201  * May be called on a zeroed but uninitialized mempool (i.e. allocated with
202  * kzalloc()).
203  */
204 void mempool_exit(struct mempool *pool)
205 {
206 	while (pool->curr_nr) {
207 		void *element = remove_element(pool);
208 		pool->free(element, pool->pool_data);
209 	}
210 	kfree(pool->elements);
211 	pool->elements = NULL;
212 }
213 EXPORT_SYMBOL(mempool_exit);
214 
215 /**
216  * mempool_destroy - deallocate a memory pool
217  * @pool:      pointer to the memory pool which was allocated via
218  *             mempool_create().
219  *
220  * Free all reserved elements in @pool and @pool itself.  This function
221  * only sleeps if the free_fn() function sleeps.
222  */
223 void mempool_destroy(struct mempool *pool)
224 {
225 	if (unlikely(!pool))
226 		return;
227 
228 	mempool_exit(pool);
229 	kfree(pool);
230 }
231 EXPORT_SYMBOL(mempool_destroy);
232 
233 int mempool_init_node(struct mempool *pool, int min_nr,
234 		mempool_alloc_t *alloc_fn, mempool_free_t *free_fn,
235 		void *pool_data, gfp_t gfp_mask, int node_id)
236 {
237 	spin_lock_init(&pool->lock);
238 	pool->min_nr	= min_nr;
239 	pool->pool_data = pool_data;
240 	pool->alloc	= alloc_fn;
241 	pool->free	= free_fn;
242 	init_waitqueue_head(&pool->wait);
243 	/*
244 	 * max() used here to ensure storage for at least 1 element to support
245 	 * zero minimum pool
246 	 */
247 	pool->elements = kmalloc_array_node(max(1, min_nr), sizeof(void *),
248 					    gfp_mask, node_id);
249 	if (!pool->elements)
250 		return -ENOMEM;
251 
252 	/*
253 	 * First pre-allocate the guaranteed number of buffers,
254 	 * also pre-allocate 1 element for zero minimum pool.
255 	 */
256 	while (pool->curr_nr < max(1, pool->min_nr)) {
257 		void *element;
258 
259 		element = pool->alloc(gfp_mask, pool->pool_data);
260 		if (unlikely(!element)) {
261 			mempool_exit(pool);
262 			return -ENOMEM;
263 		}
264 		add_element(pool, element);
265 	}
266 
267 	return 0;
268 }
269 EXPORT_SYMBOL(mempool_init_node);
270 
271 /**
272  * mempool_init - initialize a memory pool
273  * @pool:      pointer to the memory pool that should be initialized
274  * @min_nr:    the minimum number of elements guaranteed to be
275  *             allocated for this pool.
276  * @alloc_fn:  user-defined element-allocation function.
277  * @free_fn:   user-defined element-freeing function.
278  * @pool_data: optional private data available to the user-defined functions.
279  *
280  * Like mempool_create(), but initializes the pool in (i.e. embedded in another
281  * structure).
282  *
283  * Return: %0 on success, negative error code otherwise.
284  */
285 int mempool_init_noprof(struct mempool *pool, int min_nr,
286 		mempool_alloc_t *alloc_fn, mempool_free_t *free_fn,
287 		void *pool_data)
288 {
289 	return mempool_init_node(pool, min_nr, alloc_fn, free_fn,
290 				 pool_data, GFP_KERNEL, NUMA_NO_NODE);
291 
292 }
293 EXPORT_SYMBOL(mempool_init_noprof);
294 
295 /**
296  * mempool_create_node - create a memory pool
297  * @min_nr:    the minimum number of elements guaranteed to be
298  *             allocated for this pool.
299  * @alloc_fn:  user-defined element-allocation function.
300  * @free_fn:   user-defined element-freeing function.
301  * @pool_data: optional private data available to the user-defined functions.
302  * @gfp_mask:  memory allocation flags
303  * @node_id:   numa node to allocate on
304  *
305  * this function creates and allocates a guaranteed size, preallocated
306  * memory pool. The pool can be used from the mempool_alloc() and mempool_free()
307  * functions. This function might sleep. Both the alloc_fn() and the free_fn()
308  * functions might sleep - as long as the mempool_alloc() function is not called
309  * from IRQ contexts.
310  *
311  * Return: pointer to the created memory pool object or %NULL on error.
312  */
313 struct mempool *mempool_create_node_noprof(int min_nr,
314 		mempool_alloc_t *alloc_fn, mempool_free_t *free_fn,
315 		void *pool_data, gfp_t gfp_mask, int node_id)
316 {
317 	struct mempool *pool;
318 
319 	pool = kmalloc_node_noprof(sizeof(*pool), gfp_mask | __GFP_ZERO, node_id);
320 	if (!pool)
321 		return NULL;
322 
323 	if (mempool_init_node(pool, min_nr, alloc_fn, free_fn, pool_data,
324 			      gfp_mask, node_id)) {
325 		kfree(pool);
326 		return NULL;
327 	}
328 
329 	return pool;
330 }
331 EXPORT_SYMBOL(mempool_create_node_noprof);
332 
333 /**
334  * mempool_resize - resize an existing memory pool
335  * @pool:       pointer to the memory pool which was allocated via
336  *              mempool_create().
337  * @new_min_nr: the new minimum number of elements guaranteed to be
338  *              allocated for this pool.
339  *
340  * This function shrinks/grows the pool. In the case of growing,
341  * it cannot be guaranteed that the pool will be grown to the new
342  * size immediately, but new mempool_free() calls will refill it.
343  * This function may sleep.
344  *
345  * Note, the caller must guarantee that no mempool_destroy is called
346  * while this function is running. mempool_alloc() & mempool_free()
347  * might be called (eg. from IRQ contexts) while this function executes.
348  *
349  * Return: %0 on success, negative error code otherwise.
350  */
351 int mempool_resize(struct mempool *pool, int new_min_nr)
352 {
353 	void *element;
354 	void **new_elements;
355 	unsigned long flags;
356 
357 	BUG_ON(new_min_nr <= 0);
358 	might_sleep();
359 
360 	spin_lock_irqsave(&pool->lock, flags);
361 	if (new_min_nr <= pool->min_nr) {
362 		while (new_min_nr < pool->curr_nr) {
363 			element = remove_element(pool);
364 			spin_unlock_irqrestore(&pool->lock, flags);
365 			pool->free(element, pool->pool_data);
366 			spin_lock_irqsave(&pool->lock, flags);
367 		}
368 		pool->min_nr = new_min_nr;
369 		goto out_unlock;
370 	}
371 	spin_unlock_irqrestore(&pool->lock, flags);
372 
373 	/* Grow the pool */
374 	new_elements = kmalloc_objs(*new_elements, new_min_nr, GFP_KERNEL);
375 	if (!new_elements)
376 		return -ENOMEM;
377 
378 	spin_lock_irqsave(&pool->lock, flags);
379 	if (unlikely(new_min_nr <= pool->min_nr)) {
380 		/* Raced, other resize will do our work */
381 		spin_unlock_irqrestore(&pool->lock, flags);
382 		kfree(new_elements);
383 		goto out;
384 	}
385 	memcpy(new_elements, pool->elements,
386 			pool->curr_nr * sizeof(*new_elements));
387 	kfree(pool->elements);
388 	pool->elements = new_elements;
389 	pool->min_nr = new_min_nr;
390 
391 	while (pool->curr_nr < pool->min_nr) {
392 		spin_unlock_irqrestore(&pool->lock, flags);
393 		element = pool->alloc(GFP_KERNEL, pool->pool_data);
394 		if (!element)
395 			goto out;
396 		spin_lock_irqsave(&pool->lock, flags);
397 		if (pool->curr_nr < pool->min_nr) {
398 			add_element(pool, element);
399 		} else {
400 			spin_unlock_irqrestore(&pool->lock, flags);
401 			pool->free(element, pool->pool_data);	/* Raced */
402 			goto out;
403 		}
404 	}
405 out_unlock:
406 	spin_unlock_irqrestore(&pool->lock, flags);
407 out:
408 	return 0;
409 }
410 EXPORT_SYMBOL(mempool_resize);
411 
412 static unsigned int mempool_alloc_from_pool(struct mempool *pool, void **elems,
413 		unsigned int count, unsigned int allocated,
414 		gfp_t gfp_mask)
415 {
416 	unsigned long flags;
417 	unsigned int i;
418 
419 	spin_lock_irqsave(&pool->lock, flags);
420 	if (unlikely(pool->curr_nr < count - allocated))
421 		goto fail;
422 	for (i = 0; i < count; i++) {
423 		if (!elems[i]) {
424 			elems[i] = remove_element(pool);
425 			allocated++;
426 		}
427 	}
428 	spin_unlock_irqrestore(&pool->lock, flags);
429 
430 	/* Paired with rmb in mempool_free(), read comment there. */
431 	smp_wmb();
432 
433 	/*
434 	 * Update the allocation stack trace as this is more useful for
435 	 * debugging.
436 	 */
437 	for (i = 0; i < count; i++)
438 		kmemleak_update_trace(elems[i]);
439 	return allocated;
440 
441 fail:
442 	if (gfp_mask & __GFP_DIRECT_RECLAIM) {
443 		DEFINE_WAIT(wait);
444 
445 		prepare_to_wait(&pool->wait, &wait, TASK_UNINTERRUPTIBLE);
446 		spin_unlock_irqrestore(&pool->lock, flags);
447 
448 		/*
449 		 * Wait for someone else to return an element to @pool, but wake
450 		 * up occasionally as memory pressure might have reduced even
451 		 * and the normal allocation in alloc_fn could succeed even if
452 		 * no element was returned.
453 		 */
454 		io_schedule_timeout(5 * HZ);
455 		finish_wait(&pool->wait, &wait);
456 	} else {
457 		/* We must not sleep if __GFP_DIRECT_RECLAIM is not set. */
458 		spin_unlock_irqrestore(&pool->lock, flags);
459 	}
460 
461 	return allocated;
462 }
463 
464 /*
465  * Adjust the gfp flags for mempool allocations, as we never want to dip into
466  * the global emergency reserves or retry in the page allocator.
467  *
468  * The first pass also doesn't want to go reclaim, but the next passes do, so
469  * return a separate subset for that first iteration.
470  */
471 static inline gfp_t mempool_adjust_gfp(gfp_t *gfp_mask)
472 {
473 	*gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
474 	return *gfp_mask & ~(__GFP_DIRECT_RECLAIM | __GFP_IO);
475 }
476 
477 /**
478  * mempool_alloc_bulk - allocate multiple elements from a memory pool
479  * @pool:	pointer to the memory pool
480  * @elems:	partially or fully populated elements array
481  * @count:	number of entries in @elem that need to be allocated
482  * @allocated:	number of entries in @elem already allocated
483  *
484  * Allocate elements for each slot in @elem that is non-%NULL. This is done by
485  * first calling into the alloc_fn supplied at pool initialization time, and
486  * dipping into the reserved pool when alloc_fn fails to allocate an element.
487  *
488  * On return all @count elements in @elems will be populated.
489  *
490  * Return: Always 0.  If it wasn't for %$#^$ alloc tags, it would return void.
491  */
492 int mempool_alloc_bulk_noprof(struct mempool *pool, void **elems,
493 		unsigned int count, unsigned int allocated)
494 {
495 	gfp_t gfp_mask = GFP_KERNEL;
496 	gfp_t gfp_temp = mempool_adjust_gfp(&gfp_mask);
497 	unsigned int i = 0;
498 
499 	VM_WARN_ON_ONCE(count > pool->min_nr);
500 	might_alloc(gfp_mask);
501 
502 	/*
503 	 * If an error is injected, fail all elements in a bulk allocation so
504 	 * that we stress the multiple elements missing path.
505 	 */
506 	if (should_fail_ex(&fail_mempool_alloc_bulk, 1, FAULT_NOWARN)) {
507 		pr_info("forcing mempool usage for %pS\n",
508 				(void *)_RET_IP_);
509 		goto use_pool;
510 	}
511 
512 repeat_alloc:
513 	/*
514 	 * Try to allocate the elements using the allocation callback first as
515 	 * that might succeed even when the caller's bulk allocation did not.
516 	 */
517 	for (i = 0; i < count; i++) {
518 		if (elems[i])
519 			continue;
520 		elems[i] = pool->alloc(gfp_temp, pool->pool_data);
521 		if (unlikely(!elems[i]))
522 			goto use_pool;
523 		allocated++;
524 	}
525 
526 	return 0;
527 
528 use_pool:
529 	allocated = mempool_alloc_from_pool(pool, elems, count, allocated,
530 			gfp_temp);
531 	gfp_temp = gfp_mask;
532 	goto repeat_alloc;
533 }
534 EXPORT_SYMBOL_GPL(mempool_alloc_bulk_noprof);
535 
536 /**
537  * mempool_alloc - allocate an element from a memory pool
538  * @pool:	pointer to the memory pool
539  * @gfp_mask:	GFP_* flags.  %__GFP_ZERO is not supported.
540  *
541  * Allocate an element from @pool.  This is done by first calling into the
542  * alloc_fn supplied at pool initialization time, and dipping into the reserved
543  * pool when alloc_fn fails to allocate an element.
544  *
545  * This function only sleeps if the alloc_fn callback sleeps, or when waiting
546  * for elements to become available in the pool.
547  *
548  * Return: pointer to the allocated element or %NULL when failing to allocate
549  * an element.  Allocation failure can only happen when @gfp_mask does not
550  * include %__GFP_DIRECT_RECLAIM.
551  */
552 void *mempool_alloc_noprof(struct mempool *pool, gfp_t gfp_mask)
553 {
554 	gfp_t gfp_temp = mempool_adjust_gfp(&gfp_mask);
555 	void *element;
556 
557 	VM_WARN_ON_ONCE(gfp_mask & __GFP_ZERO);
558 	might_alloc(gfp_mask);
559 
560 repeat_alloc:
561 	if (should_fail_ex(&fail_mempool_alloc, 1, FAULT_NOWARN)) {
562 		pr_info("forcing mempool usage for %pS\n",
563 				(void *)_RET_IP_);
564 		element = NULL;
565 	} else {
566 		element = pool->alloc(gfp_temp, pool->pool_data);
567 	}
568 
569 	if (unlikely(!element)) {
570 		/*
571 		 * Try to allocate an element from the pool.
572 		 *
573 		 * The first pass won't have __GFP_DIRECT_RECLAIM and won't
574 		 * sleep in mempool_alloc_from_pool.  Retry the allocation
575 		 * with all flags set in that case.
576 		 */
577 		if (!mempool_alloc_from_pool(pool, &element, 1, 0, gfp_temp)) {
578 			if (gfp_temp != gfp_mask) {
579 				gfp_temp = gfp_mask;
580 				goto repeat_alloc;
581 			}
582 			if (gfp_mask & __GFP_DIRECT_RECLAIM) {
583 				goto repeat_alloc;
584 			}
585 		}
586 	}
587 
588 	return element;
589 }
590 EXPORT_SYMBOL(mempool_alloc_noprof);
591 
592 /**
593  * mempool_alloc_preallocated - allocate an element from preallocated elements
594  *                              belonging to a memory pool
595  * @pool:	pointer to the memory pool
596  *
597  * This function is similar to mempool_alloc(), but it only attempts allocating
598  * an element from the preallocated elements. It only takes a single spinlock_t
599  * and immediately returns if no preallocated elements are available.
600  *
601  * Return: pointer to the allocated element or %NULL if no elements are
602  * available.
603  */
604 void *mempool_alloc_preallocated(struct mempool *pool)
605 {
606 	void *element = NULL;
607 
608 	mempool_alloc_from_pool(pool, &element, 1, 0, GFP_NOWAIT);
609 	return element;
610 }
611 EXPORT_SYMBOL(mempool_alloc_preallocated);
612 
613 /**
614  * mempool_free_bulk - return elements to a mempool
615  * @pool:	pointer to the memory pool
616  * @elems:	elements to return
617  * @count:	number of elements to return
618  *
619  * Returns a number of elements from the start of @elem to @pool if @pool needs
620  * replenishing and sets their slots in @elem to NULL.  Other elements are left
621  * in @elem.
622  *
623  * Return: number of elements transferred to @pool.  Elements are always
624  * transferred from the beginning of @elem, so the return value can be used as
625  * an offset into @elem for the freeing the remaining elements in the caller.
626  */
627 unsigned int mempool_free_bulk(struct mempool *pool, void **elems,
628 		unsigned int count)
629 {
630 	unsigned long flags;
631 	unsigned int freed = 0;
632 	bool added = false;
633 
634 	/*
635 	 * Paired with the wmb in mempool_alloc().  The preceding read is
636 	 * for @element and the following @pool->curr_nr.  This ensures
637 	 * that the visible value of @pool->curr_nr is from after the
638 	 * allocation of @element.  This is necessary for fringe cases
639 	 * where @element was passed to this task without going through
640 	 * barriers.
641 	 *
642 	 * For example, assume @p is %NULL at the beginning and one task
643 	 * performs "p = mempool_alloc(...);" while another task is doing
644 	 * "while (!p) cpu_relax(); mempool_free(p, ...);".  This function
645 	 * may end up using curr_nr value which is from before allocation
646 	 * of @p without the following rmb.
647 	 */
648 	smp_rmb();
649 
650 	/*
651 	 * For correctness, we need a test which is guaranteed to trigger
652 	 * if curr_nr + #allocated == min_nr.  Testing curr_nr < min_nr
653 	 * without locking achieves that and refilling as soon as possible
654 	 * is desirable.
655 	 *
656 	 * Because curr_nr visible here is always a value after the
657 	 * allocation of @element, any task which decremented curr_nr below
658 	 * min_nr is guaranteed to see curr_nr < min_nr unless curr_nr gets
659 	 * incremented to min_nr afterwards.  If curr_nr gets incremented
660 	 * to min_nr after the allocation of @element, the elements
661 	 * allocated after that are subject to the same guarantee.
662 	 *
663 	 * Waiters happen iff curr_nr is 0 and the above guarantee also
664 	 * ensures that there will be frees which return elements to the
665 	 * pool waking up the waiters.
666 	 *
667 	 * For zero-minimum pools, curr_nr < min_nr (0 < 0) never succeeds,
668 	 * so waiters sleeping on pool->wait would never be woken by the
669 	 * wake-up path of previous test. This explicit check ensures the
670 	 * allocation of element when both min_nr and curr_nr are 0, and
671 	 * any active waiters are properly awakened.
672 	 */
673 	if (unlikely(READ_ONCE(pool->curr_nr) < pool->min_nr)) {
674 		spin_lock_irqsave(&pool->lock, flags);
675 		while (pool->curr_nr < pool->min_nr && freed < count) {
676 			add_element(pool, elems[freed++]);
677 			added = true;
678 		}
679 		spin_unlock_irqrestore(&pool->lock, flags);
680 	} else if (unlikely(pool->min_nr == 0 &&
681 		     READ_ONCE(pool->curr_nr) == 0)) {
682 		/* Handle the min_nr = 0 edge case: */
683 		spin_lock_irqsave(&pool->lock, flags);
684 		if (likely(pool->curr_nr == 0)) {
685 			add_element(pool, elems[freed++]);
686 			added = true;
687 		}
688 		spin_unlock_irqrestore(&pool->lock, flags);
689 	}
690 
691 	if (unlikely(added) && wq_has_sleeper(&pool->wait))
692 		wake_up(&pool->wait);
693 
694 	return freed;
695 }
696 EXPORT_SYMBOL_GPL(mempool_free_bulk);
697 
698 /**
699  * mempool_free - return an element to the pool.
700  * @element:	element to return
701  * @pool:	pointer to the memory pool
702  *
703  * Returns @element to @pool if it needs replenishing, else frees it using
704  * the free_fn callback in @pool.
705  *
706  * This function only sleeps if the free_fn callback sleeps.
707  */
708 void mempool_free(void *element, struct mempool *pool)
709 {
710 	if (likely(element) && !mempool_free_bulk(pool, &element, 1))
711 		pool->free(element, pool->pool_data);
712 }
713 EXPORT_SYMBOL(mempool_free);
714 
715 /*
716  * A commonly used alloc and free fn.
717  */
718 void *mempool_alloc_slab(gfp_t gfp_mask, void *pool_data)
719 {
720 	struct kmem_cache *mem = pool_data;
721 	VM_BUG_ON(mem->ctor);
722 	return kmem_cache_alloc_noprof(mem, gfp_mask);
723 }
724 EXPORT_SYMBOL(mempool_alloc_slab);
725 
726 void mempool_free_slab(void *element, void *pool_data)
727 {
728 	struct kmem_cache *mem = pool_data;
729 	kmem_cache_free(mem, element);
730 }
731 EXPORT_SYMBOL(mempool_free_slab);
732 
733 /*
734  * A commonly used alloc and free fn that kmalloc/kfrees the amount of memory
735  * specified by pool_data
736  */
737 void *mempool_kmalloc(gfp_t gfp_mask, void *pool_data)
738 {
739 	size_t size = (size_t)pool_data;
740 	return kmalloc_noprof(size, gfp_mask);
741 }
742 EXPORT_SYMBOL(mempool_kmalloc);
743 
744 void mempool_kfree(void *element, void *pool_data)
745 {
746 	kfree(element);
747 }
748 EXPORT_SYMBOL(mempool_kfree);
749 
750 /*
751  * A simple mempool-backed page allocator that allocates pages
752  * of the order specified by pool_data.
753  */
754 void *mempool_alloc_pages(gfp_t gfp_mask, void *pool_data)
755 {
756 	int order = (int)(long)pool_data;
757 	return alloc_pages_noprof(gfp_mask, order);
758 }
759 EXPORT_SYMBOL(mempool_alloc_pages);
760 
761 void mempool_free_pages(void *element, void *pool_data)
762 {
763 	int order = (int)(long)pool_data;
764 	__free_pages(element, order);
765 }
766 EXPORT_SYMBOL(mempool_free_pages);
767