xref: /openbmc/linux/mm/mempool.c (revision 85716a80)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  linux/mm/mempool.c
4  *
5  *  memory buffer pool support. Such pools are mostly used
6  *  for guaranteed, deadlock-free memory allocations during
7  *  extreme VM load.
8  *
9  *  started by Ingo Molnar, Copyright (C) 2001
10  *  debugging by David Rientjes, Copyright (C) 2015
11  */
12 
13 #include <linux/mm.h>
14 #include <linux/slab.h>
15 #include <linux/highmem.h>
16 #include <linux/kasan.h>
17 #include <linux/kmemleak.h>
18 #include <linux/export.h>
19 #include <linux/mempool.h>
20 #include <linux/writeback.h>
21 #include "slab.h"
22 
23 #if defined(CONFIG_DEBUG_SLAB) || defined(CONFIG_SLUB_DEBUG_ON)
24 static void poison_error(mempool_t *pool, void *element, size_t size,
25 			 size_t byte)
26 {
27 	const int nr = pool->curr_nr;
28 	const int start = max_t(int, byte - (BITS_PER_LONG / 8), 0);
29 	const int end = min_t(int, byte + (BITS_PER_LONG / 8), size);
30 	int i;
31 
32 	pr_err("BUG: mempool element poison mismatch\n");
33 	pr_err("Mempool %p size %zu\n", pool, size);
34 	pr_err(" nr=%d @ %p: %s0x", nr, element, start > 0 ? "... " : "");
35 	for (i = start; i < end; i++)
36 		pr_cont("%x ", *(u8 *)(element + i));
37 	pr_cont("%s\n", end < size ? "..." : "");
38 	dump_stack();
39 }
40 
41 static void __check_element(mempool_t *pool, void *element, size_t size)
42 {
43 	u8 *obj = element;
44 	size_t i;
45 
46 	for (i = 0; i < size; i++) {
47 		u8 exp = (i < size - 1) ? POISON_FREE : POISON_END;
48 
49 		if (obj[i] != exp) {
50 			poison_error(pool, element, size, i);
51 			return;
52 		}
53 	}
54 	memset(obj, POISON_INUSE, size);
55 }
56 
57 static void check_element(mempool_t *pool, void *element)
58 {
59 	/* Mempools backed by slab allocator */
60 	if (pool->free == mempool_kfree) {
61 		__check_element(pool, element, (size_t)pool->pool_data);
62 	} else if (pool->free == mempool_free_slab) {
63 		__check_element(pool, element, kmem_cache_size(pool->pool_data));
64 	} else if (pool->free == mempool_free_pages) {
65 		/* Mempools backed by page allocator */
66 		int order = (int)(long)pool->pool_data;
67 		void *addr = kmap_atomic((struct page *)element);
68 
69 		__check_element(pool, addr, 1UL << (PAGE_SHIFT + order));
70 		kunmap_atomic(addr);
71 	}
72 }
73 
74 static void __poison_element(void *element, size_t size)
75 {
76 	u8 *obj = element;
77 
78 	memset(obj, POISON_FREE, size - 1);
79 	obj[size - 1] = POISON_END;
80 }
81 
82 static void poison_element(mempool_t *pool, void *element)
83 {
84 	/* Mempools backed by slab allocator */
85 	if (pool->alloc == mempool_kmalloc) {
86 		__poison_element(element, (size_t)pool->pool_data);
87 	} else if (pool->alloc == mempool_alloc_slab) {
88 		__poison_element(element, kmem_cache_size(pool->pool_data));
89 	} else if (pool->alloc == mempool_alloc_pages) {
90 		/* Mempools backed by page allocator */
91 		int order = (int)(long)pool->pool_data;
92 		void *addr = kmap_atomic((struct page *)element);
93 
94 		__poison_element(addr, 1UL << (PAGE_SHIFT + order));
95 		kunmap_atomic(addr);
96 	}
97 }
98 #else /* CONFIG_DEBUG_SLAB || CONFIG_SLUB_DEBUG_ON */
99 static inline void check_element(mempool_t *pool, void *element)
100 {
101 }
102 static inline void poison_element(mempool_t *pool, void *element)
103 {
104 }
105 #endif /* CONFIG_DEBUG_SLAB || CONFIG_SLUB_DEBUG_ON */
106 
107 static __always_inline void kasan_poison_element(mempool_t *pool, void *element)
108 {
109 	if (pool->alloc == mempool_alloc_slab || pool->alloc == mempool_kmalloc)
110 		kasan_slab_free_mempool(element);
111 	else if (pool->alloc == mempool_alloc_pages)
112 		kasan_poison_pages(element, (unsigned long)pool->pool_data,
113 				   false);
114 }
115 
116 static void kasan_unpoison_element(mempool_t *pool, void *element)
117 {
118 	if (pool->alloc == mempool_kmalloc)
119 		kasan_unpoison_range(element, (size_t)pool->pool_data);
120 	else if (pool->alloc == mempool_alloc_slab)
121 		kasan_unpoison_range(element, kmem_cache_size(pool->pool_data));
122 	else if (pool->alloc == mempool_alloc_pages)
123 		kasan_unpoison_pages(element, (unsigned long)pool->pool_data,
124 				     false);
125 }
126 
127 static __always_inline void add_element(mempool_t *pool, void *element)
128 {
129 	BUG_ON(pool->curr_nr >= pool->min_nr);
130 	poison_element(pool, element);
131 	kasan_poison_element(pool, element);
132 	pool->elements[pool->curr_nr++] = element;
133 }
134 
135 static void *remove_element(mempool_t *pool)
136 {
137 	void *element = pool->elements[--pool->curr_nr];
138 
139 	BUG_ON(pool->curr_nr < 0);
140 	kasan_unpoison_element(pool, element);
141 	check_element(pool, element);
142 	return element;
143 }
144 
145 /**
146  * mempool_exit - exit a mempool initialized with mempool_init()
147  * @pool:      pointer to the memory pool which was initialized with
148  *             mempool_init().
149  *
150  * Free all reserved elements in @pool and @pool itself.  This function
151  * only sleeps if the free_fn() function sleeps.
152  *
153  * May be called on a zeroed but uninitialized mempool (i.e. allocated with
154  * kzalloc()).
155  */
156 void mempool_exit(mempool_t *pool)
157 {
158 	while (pool->curr_nr) {
159 		void *element = remove_element(pool);
160 		pool->free(element, pool->pool_data);
161 	}
162 	kfree(pool->elements);
163 	pool->elements = NULL;
164 }
165 EXPORT_SYMBOL(mempool_exit);
166 
167 /**
168  * mempool_destroy - deallocate a memory pool
169  * @pool:      pointer to the memory pool which was allocated via
170  *             mempool_create().
171  *
172  * Free all reserved elements in @pool and @pool itself.  This function
173  * only sleeps if the free_fn() function sleeps.
174  */
175 void mempool_destroy(mempool_t *pool)
176 {
177 	if (unlikely(!pool))
178 		return;
179 
180 	mempool_exit(pool);
181 	kfree(pool);
182 }
183 EXPORT_SYMBOL(mempool_destroy);
184 
185 int mempool_init_node(mempool_t *pool, int min_nr, mempool_alloc_t *alloc_fn,
186 		      mempool_free_t *free_fn, void *pool_data,
187 		      gfp_t gfp_mask, int node_id)
188 {
189 	spin_lock_init(&pool->lock);
190 	pool->min_nr	= min_nr;
191 	pool->pool_data = pool_data;
192 	pool->alloc	= alloc_fn;
193 	pool->free	= free_fn;
194 	init_waitqueue_head(&pool->wait);
195 
196 	pool->elements = kmalloc_array_node(min_nr, sizeof(void *),
197 					    gfp_mask, node_id);
198 	if (!pool->elements)
199 		return -ENOMEM;
200 
201 	/*
202 	 * First pre-allocate the guaranteed number of buffers.
203 	 */
204 	while (pool->curr_nr < pool->min_nr) {
205 		void *element;
206 
207 		element = pool->alloc(gfp_mask, pool->pool_data);
208 		if (unlikely(!element)) {
209 			mempool_exit(pool);
210 			return -ENOMEM;
211 		}
212 		add_element(pool, element);
213 	}
214 
215 	return 0;
216 }
217 EXPORT_SYMBOL(mempool_init_node);
218 
219 /**
220  * mempool_init - initialize a memory pool
221  * @pool:      pointer to the memory pool that should be initialized
222  * @min_nr:    the minimum number of elements guaranteed to be
223  *             allocated for this pool.
224  * @alloc_fn:  user-defined element-allocation function.
225  * @free_fn:   user-defined element-freeing function.
226  * @pool_data: optional private data available to the user-defined functions.
227  *
228  * Like mempool_create(), but initializes the pool in (i.e. embedded in another
229  * structure).
230  *
231  * Return: %0 on success, negative error code otherwise.
232  */
233 int mempool_init(mempool_t *pool, int min_nr, mempool_alloc_t *alloc_fn,
234 		 mempool_free_t *free_fn, void *pool_data)
235 {
236 	return mempool_init_node(pool, min_nr, alloc_fn, free_fn,
237 				 pool_data, GFP_KERNEL, NUMA_NO_NODE);
238 
239 }
240 EXPORT_SYMBOL(mempool_init);
241 
242 /**
243  * mempool_create - create a memory pool
244  * @min_nr:    the minimum number of elements guaranteed to be
245  *             allocated for this pool.
246  * @alloc_fn:  user-defined element-allocation function.
247  * @free_fn:   user-defined element-freeing function.
248  * @pool_data: optional private data available to the user-defined functions.
249  *
250  * this function creates and allocates a guaranteed size, preallocated
251  * memory pool. The pool can be used from the mempool_alloc() and mempool_free()
252  * functions. This function might sleep. Both the alloc_fn() and the free_fn()
253  * functions might sleep - as long as the mempool_alloc() function is not called
254  * from IRQ contexts.
255  *
256  * Return: pointer to the created memory pool object or %NULL on error.
257  */
258 mempool_t *mempool_create(int min_nr, mempool_alloc_t *alloc_fn,
259 				mempool_free_t *free_fn, void *pool_data)
260 {
261 	return mempool_create_node(min_nr, alloc_fn, free_fn, pool_data,
262 				   GFP_KERNEL, NUMA_NO_NODE);
263 }
264 EXPORT_SYMBOL(mempool_create);
265 
266 mempool_t *mempool_create_node(int min_nr, mempool_alloc_t *alloc_fn,
267 			       mempool_free_t *free_fn, void *pool_data,
268 			       gfp_t gfp_mask, int node_id)
269 {
270 	mempool_t *pool;
271 
272 	pool = kzalloc_node(sizeof(*pool), gfp_mask, node_id);
273 	if (!pool)
274 		return NULL;
275 
276 	if (mempool_init_node(pool, min_nr, alloc_fn, free_fn, pool_data,
277 			      gfp_mask, node_id)) {
278 		kfree(pool);
279 		return NULL;
280 	}
281 
282 	return pool;
283 }
284 EXPORT_SYMBOL(mempool_create_node);
285 
286 /**
287  * mempool_resize - resize an existing memory pool
288  * @pool:       pointer to the memory pool which was allocated via
289  *              mempool_create().
290  * @new_min_nr: the new minimum number of elements guaranteed to be
291  *              allocated for this pool.
292  *
293  * This function shrinks/grows the pool. In the case of growing,
294  * it cannot be guaranteed that the pool will be grown to the new
295  * size immediately, but new mempool_free() calls will refill it.
296  * This function may sleep.
297  *
298  * Note, the caller must guarantee that no mempool_destroy is called
299  * while this function is running. mempool_alloc() & mempool_free()
300  * might be called (eg. from IRQ contexts) while this function executes.
301  *
302  * Return: %0 on success, negative error code otherwise.
303  */
304 int mempool_resize(mempool_t *pool, int new_min_nr)
305 {
306 	void *element;
307 	void **new_elements;
308 	unsigned long flags;
309 
310 	BUG_ON(new_min_nr <= 0);
311 	might_sleep();
312 
313 	spin_lock_irqsave(&pool->lock, flags);
314 	if (new_min_nr <= pool->min_nr) {
315 		while (new_min_nr < pool->curr_nr) {
316 			element = remove_element(pool);
317 			spin_unlock_irqrestore(&pool->lock, flags);
318 			pool->free(element, pool->pool_data);
319 			spin_lock_irqsave(&pool->lock, flags);
320 		}
321 		pool->min_nr = new_min_nr;
322 		goto out_unlock;
323 	}
324 	spin_unlock_irqrestore(&pool->lock, flags);
325 
326 	/* Grow the pool */
327 	new_elements = kmalloc_array(new_min_nr, sizeof(*new_elements),
328 				     GFP_KERNEL);
329 	if (!new_elements)
330 		return -ENOMEM;
331 
332 	spin_lock_irqsave(&pool->lock, flags);
333 	if (unlikely(new_min_nr <= pool->min_nr)) {
334 		/* Raced, other resize will do our work */
335 		spin_unlock_irqrestore(&pool->lock, flags);
336 		kfree(new_elements);
337 		goto out;
338 	}
339 	memcpy(new_elements, pool->elements,
340 			pool->curr_nr * sizeof(*new_elements));
341 	kfree(pool->elements);
342 	pool->elements = new_elements;
343 	pool->min_nr = new_min_nr;
344 
345 	while (pool->curr_nr < pool->min_nr) {
346 		spin_unlock_irqrestore(&pool->lock, flags);
347 		element = pool->alloc(GFP_KERNEL, pool->pool_data);
348 		if (!element)
349 			goto out;
350 		spin_lock_irqsave(&pool->lock, flags);
351 		if (pool->curr_nr < pool->min_nr) {
352 			add_element(pool, element);
353 		} else {
354 			spin_unlock_irqrestore(&pool->lock, flags);
355 			pool->free(element, pool->pool_data);	/* Raced */
356 			goto out;
357 		}
358 	}
359 out_unlock:
360 	spin_unlock_irqrestore(&pool->lock, flags);
361 out:
362 	return 0;
363 }
364 EXPORT_SYMBOL(mempool_resize);
365 
366 /**
367  * mempool_alloc - allocate an element from a specific memory pool
368  * @pool:      pointer to the memory pool which was allocated via
369  *             mempool_create().
370  * @gfp_mask:  the usual allocation bitmask.
371  *
372  * this function only sleeps if the alloc_fn() function sleeps or
373  * returns NULL. Note that due to preallocation, this function
374  * *never* fails when called from process contexts. (it might
375  * fail if called from an IRQ context.)
376  * Note: using __GFP_ZERO is not supported.
377  *
378  * Return: pointer to the allocated element or %NULL on error.
379  */
380 void *mempool_alloc(mempool_t *pool, gfp_t gfp_mask)
381 {
382 	void *element;
383 	unsigned long flags;
384 	wait_queue_entry_t wait;
385 	gfp_t gfp_temp;
386 
387 	VM_WARN_ON_ONCE(gfp_mask & __GFP_ZERO);
388 	might_alloc(gfp_mask);
389 
390 	gfp_mask |= __GFP_NOMEMALLOC;	/* don't allocate emergency reserves */
391 	gfp_mask |= __GFP_NORETRY;	/* don't loop in __alloc_pages */
392 	gfp_mask |= __GFP_NOWARN;	/* failures are OK */
393 
394 	gfp_temp = gfp_mask & ~(__GFP_DIRECT_RECLAIM|__GFP_IO);
395 
396 repeat_alloc:
397 
398 	element = pool->alloc(gfp_temp, pool->pool_data);
399 	if (likely(element != NULL))
400 		return element;
401 
402 	spin_lock_irqsave(&pool->lock, flags);
403 	if (likely(pool->curr_nr)) {
404 		element = remove_element(pool);
405 		spin_unlock_irqrestore(&pool->lock, flags);
406 		/* paired with rmb in mempool_free(), read comment there */
407 		smp_wmb();
408 		/*
409 		 * Update the allocation stack trace as this is more useful
410 		 * for debugging.
411 		 */
412 		kmemleak_update_trace(element);
413 		return element;
414 	}
415 
416 	/*
417 	 * We use gfp mask w/o direct reclaim or IO for the first round.  If
418 	 * alloc failed with that and @pool was empty, retry immediately.
419 	 */
420 	if (gfp_temp != gfp_mask) {
421 		spin_unlock_irqrestore(&pool->lock, flags);
422 		gfp_temp = gfp_mask;
423 		goto repeat_alloc;
424 	}
425 
426 	/* We must not sleep if !__GFP_DIRECT_RECLAIM */
427 	if (!(gfp_mask & __GFP_DIRECT_RECLAIM)) {
428 		spin_unlock_irqrestore(&pool->lock, flags);
429 		return NULL;
430 	}
431 
432 	/* Let's wait for someone else to return an element to @pool */
433 	init_wait(&wait);
434 	prepare_to_wait(&pool->wait, &wait, TASK_UNINTERRUPTIBLE);
435 
436 	spin_unlock_irqrestore(&pool->lock, flags);
437 
438 	/*
439 	 * FIXME: this should be io_schedule().  The timeout is there as a
440 	 * workaround for some DM problems in 2.6.18.
441 	 */
442 	io_schedule_timeout(5*HZ);
443 
444 	finish_wait(&pool->wait, &wait);
445 	goto repeat_alloc;
446 }
447 EXPORT_SYMBOL(mempool_alloc);
448 
449 /**
450  * mempool_free - return an element to the pool.
451  * @element:   pool element pointer.
452  * @pool:      pointer to the memory pool which was allocated via
453  *             mempool_create().
454  *
455  * this function only sleeps if the free_fn() function sleeps.
456  */
457 void mempool_free(void *element, mempool_t *pool)
458 {
459 	unsigned long flags;
460 
461 	if (unlikely(element == NULL))
462 		return;
463 
464 	/*
465 	 * Paired with the wmb in mempool_alloc().  The preceding read is
466 	 * for @element and the following @pool->curr_nr.  This ensures
467 	 * that the visible value of @pool->curr_nr is from after the
468 	 * allocation of @element.  This is necessary for fringe cases
469 	 * where @element was passed to this task without going through
470 	 * barriers.
471 	 *
472 	 * For example, assume @p is %NULL at the beginning and one task
473 	 * performs "p = mempool_alloc(...);" while another task is doing
474 	 * "while (!p) cpu_relax(); mempool_free(p, ...);".  This function
475 	 * may end up using curr_nr value which is from before allocation
476 	 * of @p without the following rmb.
477 	 */
478 	smp_rmb();
479 
480 	/*
481 	 * For correctness, we need a test which is guaranteed to trigger
482 	 * if curr_nr + #allocated == min_nr.  Testing curr_nr < min_nr
483 	 * without locking achieves that and refilling as soon as possible
484 	 * is desirable.
485 	 *
486 	 * Because curr_nr visible here is always a value after the
487 	 * allocation of @element, any task which decremented curr_nr below
488 	 * min_nr is guaranteed to see curr_nr < min_nr unless curr_nr gets
489 	 * incremented to min_nr afterwards.  If curr_nr gets incremented
490 	 * to min_nr after the allocation of @element, the elements
491 	 * allocated after that are subject to the same guarantee.
492 	 *
493 	 * Waiters happen iff curr_nr is 0 and the above guarantee also
494 	 * ensures that there will be frees which return elements to the
495 	 * pool waking up the waiters.
496 	 */
497 	if (unlikely(READ_ONCE(pool->curr_nr) < pool->min_nr)) {
498 		spin_lock_irqsave(&pool->lock, flags);
499 		if (likely(pool->curr_nr < pool->min_nr)) {
500 			add_element(pool, element);
501 			spin_unlock_irqrestore(&pool->lock, flags);
502 			wake_up(&pool->wait);
503 			return;
504 		}
505 		spin_unlock_irqrestore(&pool->lock, flags);
506 	}
507 	pool->free(element, pool->pool_data);
508 }
509 EXPORT_SYMBOL(mempool_free);
510 
511 /*
512  * A commonly used alloc and free fn.
513  */
514 void *mempool_alloc_slab(gfp_t gfp_mask, void *pool_data)
515 {
516 	struct kmem_cache *mem = pool_data;
517 	VM_BUG_ON(mem->ctor);
518 	return kmem_cache_alloc(mem, gfp_mask);
519 }
520 EXPORT_SYMBOL(mempool_alloc_slab);
521 
522 void mempool_free_slab(void *element, void *pool_data)
523 {
524 	struct kmem_cache *mem = pool_data;
525 	kmem_cache_free(mem, element);
526 }
527 EXPORT_SYMBOL(mempool_free_slab);
528 
529 /*
530  * A commonly used alloc and free fn that kmalloc/kfrees the amount of memory
531  * specified by pool_data
532  */
533 void *mempool_kmalloc(gfp_t gfp_mask, void *pool_data)
534 {
535 	size_t size = (size_t)pool_data;
536 	return kmalloc(size, gfp_mask);
537 }
538 EXPORT_SYMBOL(mempool_kmalloc);
539 
540 void mempool_kfree(void *element, void *pool_data)
541 {
542 	kfree(element);
543 }
544 EXPORT_SYMBOL(mempool_kfree);
545 
546 /*
547  * A simple mempool-backed page allocator that allocates pages
548  * of the order specified by pool_data.
549  */
550 void *mempool_alloc_pages(gfp_t gfp_mask, void *pool_data)
551 {
552 	int order = (int)(long)pool_data;
553 	return alloc_pages(gfp_mask, order);
554 }
555 EXPORT_SYMBOL(mempool_alloc_pages);
556 
557 void mempool_free_pages(void *element, void *pool_data)
558 {
559 	int order = (int)(long)pool_data;
560 	__free_pages(element, order);
561 }
562 EXPORT_SYMBOL(mempool_free_pages);
563