blob: 11f0d0a5e0f853687744a0430ae6b8259daab87c [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/mempool.c
3 *
4 * memory buffer pool support. Such pools are mostly used
5 * for guaranteed, deadlock-free memory allocations during
6 * extreme VM load.
7 *
8 * started by Ingo Molnar, Copyright (C) 2001
9 */
10
11#include <linux/mm.h>
12#include <linux/slab.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040013#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/mempool.h>
15#include <linux/blkdev.h>
16#include <linux/writeback.h>
17
18static void add_element(mempool_t *pool, void *element)
19{
20 BUG_ON(pool->curr_nr >= pool->min_nr);
21 pool->elements[pool->curr_nr++] = element;
22}
23
24static void *remove_element(mempool_t *pool)
25{
26 BUG_ON(pool->curr_nr <= 0);
27 return pool->elements[--pool->curr_nr];
28}
29
30static void free_pool(mempool_t *pool)
31{
32 while (pool->curr_nr) {
33 void *element = remove_element(pool);
34 pool->free(element, pool->pool_data);
35 }
36 kfree(pool->elements);
37 kfree(pool);
38}
39
40/**
41 * mempool_create - create a memory pool
42 * @min_nr: the minimum number of elements guaranteed to be
43 * allocated for this pool.
44 * @alloc_fn: user-defined element-allocation function.
45 * @free_fn: user-defined element-freeing function.
46 * @pool_data: optional private data available to the user-defined functions.
47 *
48 * this function creates and allocates a guaranteed size, preallocated
Robert P. J. Day72fd4a32007-02-10 01:45:59 -080049 * memory pool. The pool can be used from the mempool_alloc() and mempool_free()
Linus Torvalds1da177e2005-04-16 15:20:36 -070050 * functions. This function might sleep. Both the alloc_fn() and the free_fn()
Robert P. J. Day72fd4a32007-02-10 01:45:59 -080051 * functions might sleep - as long as the mempool_alloc() function is not called
Linus Torvalds1da177e2005-04-16 15:20:36 -070052 * from IRQ contexts.
53 */
Christoph Lameter19460892005-06-23 00:08:19 -070054mempool_t *mempool_create(int min_nr, mempool_alloc_t *alloc_fn,
Linus Torvalds1da177e2005-04-16 15:20:36 -070055 mempool_free_t *free_fn, void *pool_data)
56{
Christoph Lameter19460892005-06-23 00:08:19 -070057 return mempool_create_node(min_nr,alloc_fn,free_fn, pool_data,-1);
58}
59EXPORT_SYMBOL(mempool_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
Christoph Lameter19460892005-06-23 00:08:19 -070061mempool_t *mempool_create_node(int min_nr, mempool_alloc_t *alloc_fn,
62 mempool_free_t *free_fn, void *pool_data, int node_id)
63{
64 mempool_t *pool;
Christoph Lameter94f60302007-07-17 04:03:29 -070065 pool = kmalloc_node(sizeof(*pool), GFP_KERNEL | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -070066 if (!pool)
67 return NULL;
Christoph Lameter19460892005-06-23 00:08:19 -070068 pool->elements = kmalloc_node(min_nr * sizeof(void *),
69 GFP_KERNEL, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -070070 if (!pool->elements) {
71 kfree(pool);
72 return NULL;
73 }
74 spin_lock_init(&pool->lock);
75 pool->min_nr = min_nr;
76 pool->pool_data = pool_data;
77 init_waitqueue_head(&pool->wait);
78 pool->alloc = alloc_fn;
79 pool->free = free_fn;
80
81 /*
82 * First pre-allocate the guaranteed number of buffers.
83 */
84 while (pool->curr_nr < pool->min_nr) {
85 void *element;
86
87 element = pool->alloc(GFP_KERNEL, pool->pool_data);
88 if (unlikely(!element)) {
89 free_pool(pool);
90 return NULL;
91 }
92 add_element(pool, element);
93 }
94 return pool;
95}
Christoph Lameter19460892005-06-23 00:08:19 -070096EXPORT_SYMBOL(mempool_create_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -070097
98/**
99 * mempool_resize - resize an existing memory pool
100 * @pool: pointer to the memory pool which was allocated via
101 * mempool_create().
102 * @new_min_nr: the new minimum number of elements guaranteed to be
103 * allocated for this pool.
104 * @gfp_mask: the usual allocation bitmask.
105 *
106 * This function shrinks/grows the pool. In the case of growing,
107 * it cannot be guaranteed that the pool will be grown to the new
108 * size immediately, but new mempool_free() calls will refill it.
109 *
110 * Note, the caller must guarantee that no mempool_destroy is called
111 * while this function is running. mempool_alloc() & mempool_free()
112 * might be called (eg. from IRQ contexts) while this function executes.
113 */
Al Virodd0fc662005-10-07 07:46:04 +0100114int mempool_resize(mempool_t *pool, int new_min_nr, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115{
116 void *element;
117 void **new_elements;
118 unsigned long flags;
119
120 BUG_ON(new_min_nr <= 0);
121
122 spin_lock_irqsave(&pool->lock, flags);
123 if (new_min_nr <= pool->min_nr) {
124 while (new_min_nr < pool->curr_nr) {
125 element = remove_element(pool);
126 spin_unlock_irqrestore(&pool->lock, flags);
127 pool->free(element, pool->pool_data);
128 spin_lock_irqsave(&pool->lock, flags);
129 }
130 pool->min_nr = new_min_nr;
131 goto out_unlock;
132 }
133 spin_unlock_irqrestore(&pool->lock, flags);
134
135 /* Grow the pool */
136 new_elements = kmalloc(new_min_nr * sizeof(*new_elements), gfp_mask);
137 if (!new_elements)
138 return -ENOMEM;
139
140 spin_lock_irqsave(&pool->lock, flags);
141 if (unlikely(new_min_nr <= pool->min_nr)) {
142 /* Raced, other resize will do our work */
143 spin_unlock_irqrestore(&pool->lock, flags);
144 kfree(new_elements);
145 goto out;
146 }
147 memcpy(new_elements, pool->elements,
148 pool->curr_nr * sizeof(*new_elements));
149 kfree(pool->elements);
150 pool->elements = new_elements;
151 pool->min_nr = new_min_nr;
152
153 while (pool->curr_nr < pool->min_nr) {
154 spin_unlock_irqrestore(&pool->lock, flags);
155 element = pool->alloc(gfp_mask, pool->pool_data);
156 if (!element)
157 goto out;
158 spin_lock_irqsave(&pool->lock, flags);
159 if (pool->curr_nr < pool->min_nr) {
160 add_element(pool, element);
161 } else {
162 spin_unlock_irqrestore(&pool->lock, flags);
163 pool->free(element, pool->pool_data); /* Raced */
164 goto out;
165 }
166 }
167out_unlock:
168 spin_unlock_irqrestore(&pool->lock, flags);
169out:
170 return 0;
171}
172EXPORT_SYMBOL(mempool_resize);
173
174/**
175 * mempool_destroy - deallocate a memory pool
176 * @pool: pointer to the memory pool which was allocated via
177 * mempool_create().
178 *
179 * this function only sleeps if the free_fn() function sleeps. The caller
180 * has to guarantee that all elements have been returned to the pool (ie:
181 * freed) prior to calling mempool_destroy().
182 */
183void mempool_destroy(mempool_t *pool)
184{
Eric Sesterhennf02e1fa2006-03-26 18:31:56 +0200185 /* Check for outstanding elements */
186 BUG_ON(pool->curr_nr != pool->min_nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187 free_pool(pool);
188}
189EXPORT_SYMBOL(mempool_destroy);
190
191/**
192 * mempool_alloc - allocate an element from a specific memory pool
193 * @pool: pointer to the memory pool which was allocated via
194 * mempool_create().
195 * @gfp_mask: the usual allocation bitmask.
196 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800197 * this function only sleeps if the alloc_fn() function sleeps or
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 * returns NULL. Note that due to preallocation, this function
199 * *never* fails when called from process contexts. (it might
200 * fail if called from an IRQ context.)
201 */
Al Virodd0fc662005-10-07 07:46:04 +0100202void * mempool_alloc(mempool_t *pool, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203{
204 void *element;
205 unsigned long flags;
Benjamin LaHaise01890a42005-06-23 00:10:01 -0700206 wait_queue_t wait;
Al Viro6daa0e22005-10-21 03:18:50 -0400207 gfp_t gfp_temp;
Nick Piggin20a77772005-05-01 08:58:37 -0700208
209 might_sleep_if(gfp_mask & __GFP_WAIT);
Nick Pigginb84a35b2005-05-01 08:58:36 -0700210
211 gfp_mask |= __GFP_NOMEMALLOC; /* don't allocate emergency reserves */
212 gfp_mask |= __GFP_NORETRY; /* don't loop in __alloc_pages */
213 gfp_mask |= __GFP_NOWARN; /* failures are OK */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214
Nick Piggin20a77772005-05-01 08:58:37 -0700215 gfp_temp = gfp_mask & ~(__GFP_WAIT|__GFP_IO);
216
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217repeat_alloc:
Nick Piggin20a77772005-05-01 08:58:37 -0700218
219 element = pool->alloc(gfp_temp, pool->pool_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 if (likely(element != NULL))
221 return element;
222
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 spin_lock_irqsave(&pool->lock, flags);
224 if (likely(pool->curr_nr)) {
225 element = remove_element(pool);
226 spin_unlock_irqrestore(&pool->lock, flags);
Tejun Heo5b990542012-01-10 15:08:23 -0800227 /* paired with rmb in mempool_free(), read comment there */
228 smp_wmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 return element;
230 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231
232 /* We must not sleep in the GFP_ATOMIC case */
Tejun Heo5b990542012-01-10 15:08:23 -0800233 if (!(gfp_mask & __GFP_WAIT)) {
234 spin_unlock_irqrestore(&pool->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 return NULL;
Tejun Heo5b990542012-01-10 15:08:23 -0800236 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237
Tejun Heo5b990542012-01-10 15:08:23 -0800238 /* Let's wait for someone else to return an element to @pool */
Nick Piggin20a77772005-05-01 08:58:37 -0700239 gfp_temp = gfp_mask;
Benjamin LaHaise01890a42005-06-23 00:10:01 -0700240 init_wait(&wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 prepare_to_wait(&pool->wait, &wait, TASK_UNINTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242
Tejun Heo5b990542012-01-10 15:08:23 -0800243 spin_unlock_irqrestore(&pool->lock, flags);
244
245 /*
246 * FIXME: this should be io_schedule(). The timeout is there as a
247 * workaround for some DM problems in 2.6.18.
248 */
249 io_schedule_timeout(5*HZ);
250
251 finish_wait(&pool->wait, &wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 goto repeat_alloc;
253}
254EXPORT_SYMBOL(mempool_alloc);
255
256/**
257 * mempool_free - return an element to the pool.
258 * @element: pool element pointer.
259 * @pool: pointer to the memory pool which was allocated via
260 * mempool_create().
261 *
262 * this function only sleeps if the free_fn() function sleeps.
263 */
264void mempool_free(void *element, mempool_t *pool)
265{
266 unsigned long flags;
267
Rusty Russellc80e7a82007-07-15 23:42:00 -0700268 if (unlikely(element == NULL))
269 return;
270
Tejun Heo5b990542012-01-10 15:08:23 -0800271 /*
272 * Paired with the wmb in mempool_alloc(). The preceding read is
273 * for @element and the following @pool->curr_nr. This ensures
274 * that the visible value of @pool->curr_nr is from after the
275 * allocation of @element. This is necessary for fringe cases
276 * where @element was passed to this task without going through
277 * barriers.
278 *
279 * For example, assume @p is %NULL at the beginning and one task
280 * performs "p = mempool_alloc(...);" while another task is doing
281 * "while (!p) cpu_relax(); mempool_free(p, ...);". This function
282 * may end up using curr_nr value which is from before allocation
283 * of @p without the following rmb.
284 */
285 smp_rmb();
286
287 /*
288 * For correctness, we need a test which is guaranteed to trigger
289 * if curr_nr + #allocated == min_nr. Testing curr_nr < min_nr
290 * without locking achieves that and refilling as soon as possible
291 * is desirable.
292 *
293 * Because curr_nr visible here is always a value after the
294 * allocation of @element, any task which decremented curr_nr below
295 * min_nr is guaranteed to see curr_nr < min_nr unless curr_nr gets
296 * incremented to min_nr afterwards. If curr_nr gets incremented
297 * to min_nr after the allocation of @element, the elements
298 * allocated after that are subject to the same guarantee.
299 *
300 * Waiters happen iff curr_nr is 0 and the above guarantee also
301 * ensures that there will be frees which return elements to the
302 * pool waking up the waiters.
303 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304 if (pool->curr_nr < pool->min_nr) {
305 spin_lock_irqsave(&pool->lock, flags);
306 if (pool->curr_nr < pool->min_nr) {
307 add_element(pool, element);
308 spin_unlock_irqrestore(&pool->lock, flags);
309 wake_up(&pool->wait);
310 return;
311 }
312 spin_unlock_irqrestore(&pool->lock, flags);
313 }
314 pool->free(element, pool->pool_data);
315}
316EXPORT_SYMBOL(mempool_free);
317
318/*
319 * A commonly used alloc and free fn.
320 */
Al Virodd0fc662005-10-07 07:46:04 +0100321void *mempool_alloc_slab(gfp_t gfp_mask, void *pool_data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322{
Pekka Enbergfcc234f2006-03-22 00:08:13 -0800323 struct kmem_cache *mem = pool_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 return kmem_cache_alloc(mem, gfp_mask);
325}
326EXPORT_SYMBOL(mempool_alloc_slab);
327
328void mempool_free_slab(void *element, void *pool_data)
329{
Pekka Enbergfcc234f2006-03-22 00:08:13 -0800330 struct kmem_cache *mem = pool_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 kmem_cache_free(mem, element);
332}
333EXPORT_SYMBOL(mempool_free_slab);
Matthew Dobson6e0678f2006-03-26 01:37:44 -0800334
335/*
Matthew Dobson53184082006-03-26 01:37:46 -0800336 * A commonly used alloc and free fn that kmalloc/kfrees the amount of memory
Simon Arlott183ff222007-10-20 01:27:18 +0200337 * specified by pool_data
Matthew Dobson53184082006-03-26 01:37:46 -0800338 */
339void *mempool_kmalloc(gfp_t gfp_mask, void *pool_data)
340{
Figo.zhang5e2f89b2009-08-08 21:01:22 +0800341 size_t size = (size_t)pool_data;
Matthew Dobson53184082006-03-26 01:37:46 -0800342 return kmalloc(size, gfp_mask);
343}
344EXPORT_SYMBOL(mempool_kmalloc);
345
346void mempool_kfree(void *element, void *pool_data)
347{
348 kfree(element);
349}
350EXPORT_SYMBOL(mempool_kfree);
351
352/*
Matthew Dobson6e0678f2006-03-26 01:37:44 -0800353 * A simple mempool-backed page allocator that allocates pages
354 * of the order specified by pool_data.
355 */
356void *mempool_alloc_pages(gfp_t gfp_mask, void *pool_data)
357{
358 int order = (int)(long)pool_data;
359 return alloc_pages(gfp_mask, order);
360}
361EXPORT_SYMBOL(mempool_alloc_pages);
362
363void mempool_free_pages(void *element, void *pool_data)
364{
365 int order = (int)(long)pool_data;
366 __free_pages(element, order);
367}
368EXPORT_SYMBOL(mempool_free_pages);