xref: /linux/mm/mempool.c (revision 918e25291ce95ef26c288234b088e9d433ecd94e)
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 <linux/static_key.h>
20 #include <linux/init.h>
21 #include "slab.h"
22 
23 static DECLARE_FAULT_ATTR(fail_mempool_alloc);
24 static DECLARE_FAULT_ATTR(fail_mempool_alloc_bulk);
25 
26 /*
27  * Debugging support for mempool using static key.
28  *
29  * This allows enabling mempool debug at boot time via:
30  *   mempool_debug
31  */
32 static DEFINE_STATIC_KEY_FALSE(mempool_debug_enabled);
33 
mempool_debug_setup(char * str)34 static int __init mempool_debug_setup(char *str)
35 {
36 	static_branch_enable(&mempool_debug_enabled);
37 	return 1;
38 }
39 __setup("mempool_debug", mempool_debug_setup);
40 
mempool_faul_inject_init(void)41 static int __init mempool_faul_inject_init(void)
42 {
43 	int error;
44 
45 	error = PTR_ERR_OR_ZERO(fault_create_debugfs_attr("fail_mempool_alloc",
46 			NULL, &fail_mempool_alloc));
47 	if (error)
48 		return error;
49 
50 	/* booting will fail on error return here, don't bother to cleanup */
51 	return PTR_ERR_OR_ZERO(
52 		fault_create_debugfs_attr("fail_mempool_alloc_bulk", NULL,
53 		&fail_mempool_alloc_bulk));
54 }
55 late_initcall(mempool_faul_inject_init);
56 
poison_error(struct mempool * pool,void * element,size_t size,size_t byte)57 static void poison_error(struct mempool *pool, void *element, size_t size,
58 			 size_t byte)
59 {
60 	const int nr = pool->curr_nr;
61 	const int start = max_t(int, byte - (BITS_PER_LONG / 8), 0);
62 	const int end = min_t(int, byte + (BITS_PER_LONG / 8), size);
63 	int i;
64 
65 	pr_err("BUG: mempool element poison mismatch\n");
66 	pr_err("Mempool %p size %zu\n", pool, size);
67 	pr_err(" nr=%d @ %p: %s0x", nr, element, start > 0 ? "... " : "");
68 	for (i = start; i < end; i++)
69 		pr_cont("%x ", *(u8 *)(element + i));
70 	pr_cont("%s\n", end < size ? "..." : "");
71 	dump_stack();
72 }
73 
__check_element(struct mempool * pool,void * element,size_t size)74 static void __check_element(struct mempool *pool, void *element, size_t size)
75 {
76 	u8 *obj = element;
77 	size_t i;
78 
79 	for (i = 0; i < size; i++) {
80 		u8 exp = (i < size - 1) ? POISON_FREE : POISON_END;
81 
82 		if (obj[i] != exp) {
83 			poison_error(pool, element, size, i);
84 			return;
85 		}
86 	}
87 	memset(obj, POISON_INUSE, size);
88 }
89 
check_element(struct mempool * pool,void * element)90 static void check_element(struct mempool *pool, void *element)
91 {
92 	/* Skip checking: KASAN might save its metadata in the element. */
93 	if (kasan_enabled())
94 		return;
95 
96 	/* Mempools backed by slab allocator */
97 	if (pool->free == mempool_kfree) {
98 		__check_element(pool, element, (size_t)pool->pool_data);
99 	} else if (pool->free == mempool_free_slab) {
100 		__check_element(pool, element, kmem_cache_size(pool->pool_data));
101 	} else if (pool->free == mempool_free_pages) {
102 		/* Mempools backed by page allocator */
103 		int order = (int)(long)pool->pool_data;
104 
105 #ifdef CONFIG_HIGHMEM
106 		for (int i = 0; i < (1 << order); i++) {
107 			struct page *page = (struct page *)element;
108 			void *addr = kmap_local_page(page + i);
109 
110 			__check_element(pool, addr, PAGE_SIZE);
111 			kunmap_local(addr);
112 		}
113 #else
114 		void *addr = page_address((struct page *)element);
115 
116 		__check_element(pool, addr, PAGE_SIZE << order);
117 #endif
118 	}
119 }
120 
__poison_element(void * element,size_t size)121 static void __poison_element(void *element, size_t size)
122 {
123 	u8 *obj = element;
124 
125 	memset(obj, POISON_FREE, size - 1);
126 	obj[size - 1] = POISON_END;
127 }
128 
poison_element(struct mempool * pool,void * element)129 static void poison_element(struct mempool *pool, void *element)
130 {
131 	/* Skip poisoning: KASAN might save its metadata in the element. */
132 	if (kasan_enabled())
133 		return;
134 
135 	/* Mempools backed by slab allocator */
136 	if (pool->alloc == mempool_kmalloc) {
137 		__poison_element(element, (size_t)pool->pool_data);
138 	} else if (pool->alloc == mempool_alloc_slab) {
139 		__poison_element(element, kmem_cache_size(pool->pool_data));
140 	} else if (pool->alloc == mempool_alloc_pages) {
141 		/* Mempools backed by page allocator */
142 		int order = (int)(long)pool->pool_data;
143 
144 #ifdef CONFIG_HIGHMEM
145 		for (int i = 0; i < (1 << order); i++) {
146 			struct page *page = (struct page *)element;
147 			void *addr = kmap_local_page(page + i);
148 
149 			__poison_element(addr, PAGE_SIZE);
150 			kunmap_local(addr);
151 		}
152 #else
153 		void *addr = page_address((struct page *)element);
154 
155 		__poison_element(addr, PAGE_SIZE << order);
156 #endif
157 	}
158 }
159 
kasan_poison_element(struct mempool * pool,void * element)160 static __always_inline bool kasan_poison_element(struct mempool *pool,
161 		void *element)
162 {
163 	if (pool->alloc == mempool_alloc_slab || pool->alloc == mempool_kmalloc)
164 		return kasan_mempool_poison_object(element);
165 	else if (pool->alloc == mempool_alloc_pages)
166 		return kasan_mempool_poison_pages(element,
167 						(unsigned long)pool->pool_data);
168 	return true;
169 }
170 
kasan_unpoison_element(struct mempool * pool,void * element)171 static void kasan_unpoison_element(struct mempool *pool, void *element)
172 {
173 	if (pool->alloc == mempool_kmalloc)
174 		kasan_mempool_unpoison_object(element, (size_t)pool->pool_data);
175 	else if (pool->alloc == mempool_alloc_slab)
176 		kasan_mempool_unpoison_object(element,
177 					      kmem_cache_size(pool->pool_data));
178 	else if (pool->alloc == mempool_alloc_pages)
179 		kasan_mempool_unpoison_pages(element,
180 					     (unsigned long)pool->pool_data);
181 }
182 
add_element(struct mempool * pool,void * element)183 static __always_inline void add_element(struct mempool *pool, void *element)
184 {
185 	BUG_ON(pool->min_nr != 0 && pool->curr_nr >= pool->min_nr);
186 
187 	if (static_branch_unlikely(&mempool_debug_enabled))
188 		poison_element(pool, element);
189 
190 	if (kasan_poison_element(pool, element))
191 		pool->elements[pool->curr_nr++] = element;
192 }
193 
remove_element(struct mempool * pool)194 static void *remove_element(struct mempool *pool)
195 {
196 	void *element = pool->elements[--pool->curr_nr];
197 
198 	BUG_ON(pool->curr_nr < 0);
199 	kasan_unpoison_element(pool, element);
200 
201 	if (static_branch_unlikely(&mempool_debug_enabled))
202 		check_element(pool, element);
203 	return element;
204 }
205 
206 /**
207  * mempool_exit - exit a mempool initialized with mempool_init()
208  * @pool:      pointer to the memory pool which was initialized with
209  *             mempool_init().
210  *
211  * Free all reserved elements in @pool and @pool itself.  This function
212  * only sleeps if the free_fn() function sleeps.
213  *
214  * May be called on a zeroed but uninitialized mempool (i.e. allocated with
215  * kzalloc()).
216  */
mempool_exit(struct mempool * pool)217 void mempool_exit(struct mempool *pool)
218 {
219 	while (pool->curr_nr) {
220 		void *element = remove_element(pool);
221 		pool->free(element, pool->pool_data);
222 	}
223 	kfree(pool->elements);
224 	pool->elements = NULL;
225 }
226 EXPORT_SYMBOL(mempool_exit);
227 
228 /**
229  * mempool_destroy - deallocate a memory pool
230  * @pool:      pointer to the memory pool which was allocated via
231  *             mempool_create().
232  *
233  * Free all reserved elements in @pool and @pool itself.  This function
234  * only sleeps if the free_fn() function sleeps.
235  */
mempool_destroy(struct mempool * pool)236 void mempool_destroy(struct mempool *pool)
237 {
238 	if (unlikely(!pool))
239 		return;
240 
241 	mempool_exit(pool);
242 	kfree(pool);
243 }
244 EXPORT_SYMBOL(mempool_destroy);
245 
mempool_init_node(struct mempool * pool,int min_nr,mempool_alloc_t * alloc_fn,mempool_free_t * free_fn,void * pool_data,gfp_t gfp_mask,int node_id)246 int mempool_init_node(struct mempool *pool, int min_nr,
247 		mempool_alloc_t *alloc_fn, mempool_free_t *free_fn,
248 		void *pool_data, gfp_t gfp_mask, int node_id)
249 {
250 	spin_lock_init(&pool->lock);
251 	pool->min_nr	= min_nr;
252 	pool->pool_data = pool_data;
253 	pool->alloc	= alloc_fn;
254 	pool->free	= free_fn;
255 	init_waitqueue_head(&pool->wait);
256 	/*
257 	 * max() used here to ensure storage for at least 1 element to support
258 	 * zero minimum pool
259 	 */
260 	pool->elements = kmalloc_array_node(max(1, min_nr), sizeof(void *),
261 					    gfp_mask, node_id);
262 	if (!pool->elements)
263 		return -ENOMEM;
264 
265 	/*
266 	 * First pre-allocate the guaranteed number of buffers,
267 	 * also pre-allocate 1 element for zero minimum pool.
268 	 */
269 	while (pool->curr_nr < max(1, pool->min_nr)) {
270 		void *element;
271 
272 		element = pool->alloc(gfp_mask, pool->pool_data);
273 		if (unlikely(!element)) {
274 			mempool_exit(pool);
275 			return -ENOMEM;
276 		}
277 		add_element(pool, element);
278 	}
279 
280 	return 0;
281 }
282 EXPORT_SYMBOL(mempool_init_node);
283 
284 /**
285  * mempool_init - initialize a memory pool
286  * @pool:      pointer to the memory pool that should be initialized
287  * @min_nr:    the minimum number of elements guaranteed to be
288  *             allocated for this pool.
289  * @alloc_fn:  user-defined element-allocation function.
290  * @free_fn:   user-defined element-freeing function.
291  * @pool_data: optional private data available to the user-defined functions.
292  *
293  * Like mempool_create(), but initializes the pool in (i.e. embedded in another
294  * structure).
295  *
296  * Return: %0 on success, negative error code otherwise.
297  */
mempool_init_noprof(struct mempool * pool,int min_nr,mempool_alloc_t * alloc_fn,mempool_free_t * free_fn,void * pool_data)298 int mempool_init_noprof(struct mempool *pool, int min_nr,
299 		mempool_alloc_t *alloc_fn, mempool_free_t *free_fn,
300 		void *pool_data)
301 {
302 	return mempool_init_node(pool, min_nr, alloc_fn, free_fn,
303 				 pool_data, GFP_KERNEL, NUMA_NO_NODE);
304 
305 }
306 EXPORT_SYMBOL(mempool_init_noprof);
307 
308 /**
309  * mempool_create_node - create a memory pool
310  * @min_nr:    the minimum number of elements guaranteed to be
311  *             allocated for this pool.
312  * @alloc_fn:  user-defined element-allocation function.
313  * @free_fn:   user-defined element-freeing function.
314  * @pool_data: optional private data available to the user-defined functions.
315  * @gfp_mask:  memory allocation flags
316  * @node_id:   numa node to allocate on
317  *
318  * this function creates and allocates a guaranteed size, preallocated
319  * memory pool. The pool can be used from the mempool_alloc() and mempool_free()
320  * functions. This function might sleep. Both the alloc_fn() and the free_fn()
321  * functions might sleep - as long as the mempool_alloc() function is not called
322  * from IRQ contexts.
323  *
324  * Return: pointer to the created memory pool object or %NULL on error.
325  */
mempool_create_node_noprof(int min_nr,mempool_alloc_t * alloc_fn,mempool_free_t * free_fn,void * pool_data,gfp_t gfp_mask,int node_id)326 struct mempool *mempool_create_node_noprof(int min_nr,
327 		mempool_alloc_t *alloc_fn, mempool_free_t *free_fn,
328 		void *pool_data, gfp_t gfp_mask, int node_id)
329 {
330 	struct mempool *pool;
331 
332 	pool = kmalloc_node_noprof(sizeof(*pool), gfp_mask | __GFP_ZERO, node_id);
333 	if (!pool)
334 		return NULL;
335 
336 	if (mempool_init_node(pool, min_nr, alloc_fn, free_fn, pool_data,
337 			      gfp_mask, node_id)) {
338 		kfree(pool);
339 		return NULL;
340 	}
341 
342 	return pool;
343 }
344 EXPORT_SYMBOL(mempool_create_node_noprof);
345 
346 /**
347  * mempool_resize - resize an existing memory pool
348  * @pool:       pointer to the memory pool which was allocated via
349  *              mempool_create().
350  * @new_min_nr: the new minimum number of elements guaranteed to be
351  *              allocated for this pool.
352  *
353  * This function shrinks/grows the pool. In the case of growing,
354  * it cannot be guaranteed that the pool will be grown to the new
355  * size immediately, but new mempool_free() calls will refill it.
356  * This function may sleep.
357  *
358  * Note, the caller must guarantee that no mempool_destroy is called
359  * while this function is running. mempool_alloc() & mempool_free()
360  * might be called (eg. from IRQ contexts) while this function executes.
361  *
362  * Return: %0 on success, negative error code otherwise.
363  */
mempool_resize(struct mempool * pool,int new_min_nr)364 int mempool_resize(struct mempool *pool, int new_min_nr)
365 {
366 	void *element;
367 	void **new_elements;
368 	unsigned long flags;
369 
370 	BUG_ON(new_min_nr <= 0);
371 	might_sleep();
372 
373 	spin_lock_irqsave(&pool->lock, flags);
374 	if (new_min_nr <= pool->min_nr) {
375 		while (new_min_nr < pool->curr_nr) {
376 			element = remove_element(pool);
377 			spin_unlock_irqrestore(&pool->lock, flags);
378 			pool->free(element, pool->pool_data);
379 			spin_lock_irqsave(&pool->lock, flags);
380 		}
381 		pool->min_nr = new_min_nr;
382 		goto out_unlock;
383 	}
384 	spin_unlock_irqrestore(&pool->lock, flags);
385 
386 	/* Grow the pool */
387 	new_elements = kmalloc_objs(*new_elements, new_min_nr);
388 	if (!new_elements)
389 		return -ENOMEM;
390 
391 	spin_lock_irqsave(&pool->lock, flags);
392 	if (unlikely(new_min_nr <= pool->min_nr)) {
393 		/* Raced, other resize will do our work */
394 		spin_unlock_irqrestore(&pool->lock, flags);
395 		kfree(new_elements);
396 		goto out;
397 	}
398 	memcpy(new_elements, pool->elements,
399 			pool->curr_nr * sizeof(*new_elements));
400 	kfree(pool->elements);
401 	pool->elements = new_elements;
402 	pool->min_nr = new_min_nr;
403 
404 	while (pool->curr_nr < pool->min_nr) {
405 		spin_unlock_irqrestore(&pool->lock, flags);
406 		element = pool->alloc(GFP_KERNEL, pool->pool_data);
407 		if (!element)
408 			goto out;
409 		spin_lock_irqsave(&pool->lock, flags);
410 		if (pool->curr_nr < pool->min_nr) {
411 			add_element(pool, element);
412 		} else {
413 			spin_unlock_irqrestore(&pool->lock, flags);
414 			pool->free(element, pool->pool_data);	/* Raced */
415 			goto out;
416 		}
417 	}
418 out_unlock:
419 	spin_unlock_irqrestore(&pool->lock, flags);
420 out:
421 	return 0;
422 }
423 EXPORT_SYMBOL(mempool_resize);
424 
mempool_alloc_from_pool(struct mempool * pool,void ** elems,unsigned int count,unsigned int allocated,gfp_t gfp_mask)425 static unsigned int mempool_alloc_from_pool(struct mempool *pool, void **elems,
426 		unsigned int count, unsigned int allocated,
427 		gfp_t gfp_mask)
428 {
429 	unsigned long flags;
430 	unsigned int i;
431 
432 	spin_lock_irqsave(&pool->lock, flags);
433 	if (unlikely(pool->curr_nr < count - allocated))
434 		goto fail;
435 	while (allocated < count)
436 		elems[allocated++] = remove_element(pool);
437 	spin_unlock_irqrestore(&pool->lock, flags);
438 
439 	/* Paired with rmb in mempool_free(), read comment there. */
440 	smp_wmb();
441 
442 	/*
443 	 * Update the allocation stack trace as this is more useful for
444 	 * debugging.
445 	 */
446 	for (i = 0; i < count; i++)
447 		kmemleak_update_trace(elems[i]);
448 	return allocated;
449 
450 fail:
451 	if (gfp_mask & __GFP_DIRECT_RECLAIM) {
452 		DEFINE_WAIT(wait);
453 
454 		prepare_to_wait(&pool->wait, &wait, TASK_UNINTERRUPTIBLE);
455 		spin_unlock_irqrestore(&pool->lock, flags);
456 
457 		/*
458 		 * Wait for someone else to return an element to @pool, but wake
459 		 * up occasionally as memory pressure might have reduced even
460 		 * and the normal allocation in alloc_fn could succeed even if
461 		 * no element was returned.
462 		 */
463 		io_schedule_timeout(5 * HZ);
464 		finish_wait(&pool->wait, &wait);
465 	} else {
466 		/* We must not sleep if __GFP_DIRECT_RECLAIM is not set. */
467 		spin_unlock_irqrestore(&pool->lock, flags);
468 	}
469 
470 	return allocated;
471 }
472 
473 /*
474  * Adjust the gfp flags for mempool allocations, as we never want to dip into
475  * the global emergency reserves or retry in the page allocator.
476  *
477  * The first pass also doesn't want to go reclaim, but the next passes do, so
478  * return a separate subset for that first iteration.
479  */
mempool_adjust_gfp(gfp_t * gfp_mask)480 static inline gfp_t mempool_adjust_gfp(gfp_t *gfp_mask)
481 {
482 	*gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
483 	return *gfp_mask & ~(__GFP_DIRECT_RECLAIM | __GFP_IO);
484 }
485 
486 /**
487  * mempool_alloc_bulk - allocate multiple elements from a memory pool
488  * @pool:	pointer to the memory pool
489  * @elems:	partially or fully populated elements array
490  * @count:	number of entries in @elem that need to be allocated
491  *
492  * Allocate @count elements into @elems.  This is done by first calling into the
493  * alloc_fn supplied at pool initialization time, and dipping into the reserved
494  * pool when alloc_fn fails to allocate an element.
495  *
496  * On return all @count elements in @elems will be populated.
497  *
498  * Return: Always 0.  If it wasn't for %$#^$ alloc tags, it would return void.
499  */
mempool_alloc_bulk_noprof(struct mempool * pool,void ** elems,unsigned int count)500 int mempool_alloc_bulk_noprof(struct mempool *pool, void **elems,
501 		unsigned int count)
502 {
503 	gfp_t gfp_mask = GFP_KERNEL;
504 	gfp_t gfp_temp = mempool_adjust_gfp(&gfp_mask);
505 	unsigned int allocated = 0;
506 
507 	VM_WARN_ON_ONCE(count > pool->min_nr);
508 	might_alloc(gfp_mask);
509 
510 	/*
511 	 * If an error is injected, fail all elements in a bulk allocation so
512 	 * that we stress the multiple elements missing path.
513 	 */
514 	if (should_fail_ex(&fail_mempool_alloc_bulk, 1, FAULT_NOWARN)) {
515 		pr_info("forcing mempool usage for %pS\n",
516 				(void *)_RET_IP_);
517 		goto use_pool;
518 	}
519 
520 repeat_alloc:
521 	/*
522 	 * Try to allocate the elements using the allocation callback first as
523 	 * that might succeed even when the caller's bulk allocation did not.
524 	 */
525 	while (allocated < count) {
526 		elems[allocated] = pool->alloc(gfp_temp, pool->pool_data);
527 		if (unlikely(!elems[allocated]))
528 			goto use_pool;
529 		allocated++;
530 	}
531 
532 	return 0;
533 
534 use_pool:
535 	allocated = mempool_alloc_from_pool(pool, elems, count, allocated,
536 			gfp_temp);
537 	gfp_temp = gfp_mask;
538 	goto repeat_alloc;
539 }
540 EXPORT_SYMBOL_GPL(mempool_alloc_bulk_noprof);
541 
542 /**
543  * mempool_alloc - allocate an element from a memory pool
544  * @pool:	pointer to the memory pool
545  * @gfp_mask:	GFP_* flags.  %__GFP_ZERO is not supported.
546  *
547  * Allocate an element from @pool.  This is done by first calling into the
548  * alloc_fn supplied at pool initialization time, and dipping into the reserved
549  * pool when alloc_fn fails to allocate an element.
550  *
551  * This function only sleeps if the alloc_fn callback sleeps, or when waiting
552  * for elements to become available in the pool.
553  *
554  * Return: pointer to the allocated element or %NULL when failing to allocate
555  * an element.  Allocation failure can only happen when @gfp_mask does not
556  * include %__GFP_DIRECT_RECLAIM.
557  */
mempool_alloc_noprof(struct mempool * pool,gfp_t gfp_mask)558 void *mempool_alloc_noprof(struct mempool *pool, gfp_t gfp_mask)
559 {
560 	gfp_t gfp_temp = mempool_adjust_gfp(&gfp_mask);
561 	void *element;
562 
563 	VM_WARN_ON_ONCE(gfp_mask & __GFP_ZERO);
564 	might_alloc(gfp_mask);
565 
566 repeat_alloc:
567 	if (should_fail_ex(&fail_mempool_alloc, 1, FAULT_NOWARN)) {
568 		pr_info("forcing mempool usage for %pS\n",
569 				(void *)_RET_IP_);
570 		element = NULL;
571 	} else {
572 		element = pool->alloc(gfp_temp, pool->pool_data);
573 	}
574 
575 	if (unlikely(!element)) {
576 		/*
577 		 * Try to allocate an element from the pool.
578 		 *
579 		 * The first pass won't have __GFP_DIRECT_RECLAIM and won't
580 		 * sleep in mempool_alloc_from_pool.  Retry the allocation
581 		 * with all flags set in that case.
582 		 */
583 		if (!mempool_alloc_from_pool(pool, &element, 1, 0, gfp_temp)) {
584 			if (gfp_temp != gfp_mask) {
585 				gfp_temp = gfp_mask;
586 				goto repeat_alloc;
587 			}
588 			if (gfp_mask & __GFP_DIRECT_RECLAIM) {
589 				goto repeat_alloc;
590 			}
591 		}
592 	}
593 
594 	return element;
595 }
596 EXPORT_SYMBOL(mempool_alloc_noprof);
597 
598 /**
599  * mempool_alloc_preallocated - allocate an element from preallocated elements
600  *                              belonging to a memory pool
601  * @pool:	pointer to the memory pool
602  *
603  * This function is similar to mempool_alloc(), but it only attempts allocating
604  * an element from the preallocated elements. It only takes a single spinlock_t
605  * and immediately returns if no preallocated elements are available.
606  *
607  * Return: pointer to the allocated element or %NULL if no elements are
608  * available.
609  */
mempool_alloc_preallocated(struct mempool * pool)610 void *mempool_alloc_preallocated(struct mempool *pool)
611 {
612 	void *element = NULL;
613 
614 	mempool_alloc_from_pool(pool, &element, 1, 0, GFP_NOWAIT);
615 	return element;
616 }
617 EXPORT_SYMBOL(mempool_alloc_preallocated);
618 
619 /**
620  * mempool_free_bulk - return elements to a mempool
621  * @pool:	pointer to the memory pool
622  * @elems:	elements to return
623  * @count:	number of elements to return
624  *
625  * Returns a number of elements from the start of @elem to @pool if @pool needs
626  * replenishing and sets their slots in @elem to NULL.  Other elements are left
627  * in @elem.
628  *
629  * Return: number of elements transferred to @pool.  Elements are always
630  * transferred from the beginning of @elem, so the return value can be used as
631  * an offset into @elem for the freeing the remaining elements in the caller.
632  */
mempool_free_bulk(struct mempool * pool,void ** elems,unsigned int count)633 unsigned int mempool_free_bulk(struct mempool *pool, void **elems,
634 		unsigned int count)
635 {
636 	unsigned long flags;
637 	unsigned int freed = 0;
638 	bool added = false;
639 
640 	/*
641 	 * Paired with the wmb in mempool_alloc().  The preceding read is
642 	 * for @element and the following @pool->curr_nr.  This ensures
643 	 * that the visible value of @pool->curr_nr is from after the
644 	 * allocation of @element.  This is necessary for fringe cases
645 	 * where @element was passed to this task without going through
646 	 * barriers.
647 	 *
648 	 * For example, assume @p is %NULL at the beginning and one task
649 	 * performs "p = mempool_alloc(...);" while another task is doing
650 	 * "while (!p) cpu_relax(); mempool_free(p, ...);".  This function
651 	 * may end up using curr_nr value which is from before allocation
652 	 * of @p without the following rmb.
653 	 */
654 	smp_rmb();
655 
656 	/*
657 	 * For correctness, we need a test which is guaranteed to trigger
658 	 * if curr_nr + #allocated == min_nr.  Testing curr_nr < min_nr
659 	 * without locking achieves that and refilling as soon as possible
660 	 * is desirable.
661 	 *
662 	 * Because curr_nr visible here is always a value after the
663 	 * allocation of @element, any task which decremented curr_nr below
664 	 * min_nr is guaranteed to see curr_nr < min_nr unless curr_nr gets
665 	 * incremented to min_nr afterwards.  If curr_nr gets incremented
666 	 * to min_nr after the allocation of @element, the elements
667 	 * allocated after that are subject to the same guarantee.
668 	 *
669 	 * Waiters happen iff curr_nr is 0 and the above guarantee also
670 	 * ensures that there will be frees which return elements to the
671 	 * pool waking up the waiters.
672 	 *
673 	 * For zero-minimum pools, curr_nr < min_nr (0 < 0) never succeeds,
674 	 * so waiters sleeping on pool->wait would never be woken by the
675 	 * wake-up path of previous test. This explicit check ensures the
676 	 * allocation of element when both min_nr and curr_nr are 0, and
677 	 * any active waiters are properly awakened.
678 	 */
679 	if (unlikely(READ_ONCE(pool->curr_nr) < pool->min_nr)) {
680 		spin_lock_irqsave(&pool->lock, flags);
681 		while (pool->curr_nr < pool->min_nr && freed < count) {
682 			add_element(pool, elems[freed++]);
683 			added = true;
684 		}
685 		spin_unlock_irqrestore(&pool->lock, flags);
686 	} else if (unlikely(pool->min_nr == 0 &&
687 		     READ_ONCE(pool->curr_nr) == 0)) {
688 		/* Handle the min_nr = 0 edge case: */
689 		spin_lock_irqsave(&pool->lock, flags);
690 		if (likely(pool->curr_nr == 0)) {
691 			add_element(pool, elems[freed++]);
692 			added = true;
693 		}
694 		spin_unlock_irqrestore(&pool->lock, flags);
695 	}
696 
697 	if (unlikely(added) && wq_has_sleeper(&pool->wait))
698 		wake_up(&pool->wait);
699 
700 	return freed;
701 }
702 EXPORT_SYMBOL_GPL(mempool_free_bulk);
703 
704 /**
705  * mempool_free - return an element to the pool.
706  * @element:	element to return
707  * @pool:	pointer to the memory pool
708  *
709  * Returns @element to @pool if it needs replenishing, else frees it using
710  * the free_fn callback in @pool.
711  *
712  * This function only sleeps if the free_fn callback sleeps.
713  */
mempool_free(void * element,struct mempool * pool)714 void mempool_free(void *element, struct mempool *pool)
715 {
716 	if (likely(element) && !mempool_free_bulk(pool, &element, 1))
717 		pool->free(element, pool->pool_data);
718 }
719 EXPORT_SYMBOL(mempool_free);
720 
721 /*
722  * A commonly used alloc and free fn.
723  */
mempool_alloc_slab(gfp_t gfp_mask,void * pool_data)724 void *mempool_alloc_slab(gfp_t gfp_mask, void *pool_data)
725 {
726 	struct kmem_cache *mem = pool_data;
727 	VM_BUG_ON(mem->ctor);
728 	return kmem_cache_alloc_noprof(mem, gfp_mask);
729 }
730 EXPORT_SYMBOL(mempool_alloc_slab);
731 
mempool_free_slab(void * element,void * pool_data)732 void mempool_free_slab(void *element, void *pool_data)
733 {
734 	struct kmem_cache *mem = pool_data;
735 	kmem_cache_free(mem, element);
736 }
737 EXPORT_SYMBOL(mempool_free_slab);
738 
739 /*
740  * A commonly used alloc and free fn that kmalloc/kfrees the amount of memory
741  * specified by pool_data
742  */
mempool_kmalloc(gfp_t gfp_mask,void * pool_data)743 void *mempool_kmalloc(gfp_t gfp_mask, void *pool_data)
744 {
745 	size_t size = (size_t)pool_data;
746 	return kmalloc_noprof(size, gfp_mask);
747 }
748 EXPORT_SYMBOL(mempool_kmalloc);
749 
mempool_kfree(void * element,void * pool_data)750 void mempool_kfree(void *element, void *pool_data)
751 {
752 	kfree(element);
753 }
754 EXPORT_SYMBOL(mempool_kfree);
755 
756 /*
757  * A simple mempool-backed page allocator that allocates pages
758  * of the order specified by pool_data.
759  */
mempool_alloc_pages(gfp_t gfp_mask,void * pool_data)760 void *mempool_alloc_pages(gfp_t gfp_mask, void *pool_data)
761 {
762 	int order = (int)(long)pool_data;
763 	return alloc_pages_noprof(gfp_mask, order);
764 }
765 EXPORT_SYMBOL(mempool_alloc_pages);
766 
mempool_free_pages(void * element,void * pool_data)767 void mempool_free_pages(void *element, void *pool_data)
768 {
769 	int order = (int)(long)pool_data;
770 	__free_pages(element, order);
771 }
772 EXPORT_SYMBOL(mempool_free_pages);
773