4 * memory buffer pool support. Such pools are mostly used
5 * for guaranteed, deadlock-free memory allocations during
8 * started by Ingo Molnar, Copyright (C) 2001
12 #include <linux/slab.h>
13 #include <linux/export.h>
14 #include <linux/mempool.h>
15 #include <linux/blkdev.h>
16 #include <linux/writeback.h>
18 static void add_element(mempool_t *pool, void *element)
20 BUG_ON(pool->curr_nr >= pool->min_nr);
21 pool->elements[pool->curr_nr++] = element;
24 static void *remove_element(mempool_t *pool)
26 BUG_ON(pool->curr_nr <= 0);
27 return pool->elements[--pool->curr_nr];
31 * mempool_destroy - deallocate a memory pool
32 * @pool: pointer to the memory pool which was allocated via
35 * Free all reserved elements in @pool and @pool itself. This function
36 * only sleeps if the free_fn() function sleeps.
38 void mempool_destroy(mempool_t *pool)
40 while (pool->curr_nr) {
41 void *element = remove_element(pool);
42 pool->free(element, pool->pool_data);
44 kfree(pool->elements);
47 EXPORT_SYMBOL(mempool_destroy);
50 * mempool_create - create a memory pool
51 * @min_nr: the minimum number of elements guaranteed to be
52 * allocated for this pool.
53 * @alloc_fn: user-defined element-allocation function.
54 * @free_fn: user-defined element-freeing function.
55 * @pool_data: optional private data available to the user-defined functions.
57 * this function creates and allocates a guaranteed size, preallocated
58 * memory pool. The pool can be used from the mempool_alloc() and mempool_free()
59 * functions. This function might sleep. Both the alloc_fn() and the free_fn()
60 * functions might sleep - as long as the mempool_alloc() function is not called
63 mempool_t *mempool_create(int min_nr, mempool_alloc_t *alloc_fn,
64 mempool_free_t *free_fn, void *pool_data)
66 return mempool_create_node(min_nr,alloc_fn,free_fn, pool_data,-1);
68 EXPORT_SYMBOL(mempool_create);
70 mempool_t *mempool_create_node(int min_nr, mempool_alloc_t *alloc_fn,
71 mempool_free_t *free_fn, void *pool_data, int node_id)
74 pool = kmalloc_node(sizeof(*pool), GFP_KERNEL | __GFP_ZERO, node_id);
77 pool->elements = kmalloc_node(min_nr * sizeof(void *),
79 if (!pool->elements) {
83 spin_lock_init(&pool->lock);
84 pool->min_nr = min_nr;
85 pool->pool_data = pool_data;
86 init_waitqueue_head(&pool->wait);
87 pool->alloc = alloc_fn;
91 * First pre-allocate the guaranteed number of buffers.
93 while (pool->curr_nr < pool->min_nr) {
96 element = pool->alloc(GFP_KERNEL, pool->pool_data);
97 if (unlikely(!element)) {
98 mempool_destroy(pool);
101 add_element(pool, element);
105 EXPORT_SYMBOL(mempool_create_node);
108 * mempool_resize - resize an existing memory pool
109 * @pool: pointer to the memory pool which was allocated via
111 * @new_min_nr: the new minimum number of elements guaranteed to be
112 * allocated for this pool.
113 * @gfp_mask: the usual allocation bitmask.
115 * This function shrinks/grows the pool. In the case of growing,
116 * it cannot be guaranteed that the pool will be grown to the new
117 * size immediately, but new mempool_free() calls will refill it.
119 * Note, the caller must guarantee that no mempool_destroy is called
120 * while this function is running. mempool_alloc() & mempool_free()
121 * might be called (eg. from IRQ contexts) while this function executes.
123 int mempool_resize(mempool_t *pool, int new_min_nr, gfp_t gfp_mask)
129 BUG_ON(new_min_nr <= 0);
131 spin_lock_irqsave(&pool->lock, flags);
132 if (new_min_nr <= pool->min_nr) {
133 while (new_min_nr < pool->curr_nr) {
134 element = remove_element(pool);
135 spin_unlock_irqrestore(&pool->lock, flags);
136 pool->free(element, pool->pool_data);
137 spin_lock_irqsave(&pool->lock, flags);
139 pool->min_nr = new_min_nr;
142 spin_unlock_irqrestore(&pool->lock, flags);
145 new_elements = kmalloc(new_min_nr * sizeof(*new_elements), gfp_mask);
149 spin_lock_irqsave(&pool->lock, flags);
150 if (unlikely(new_min_nr <= pool->min_nr)) {
151 /* Raced, other resize will do our work */
152 spin_unlock_irqrestore(&pool->lock, flags);
156 memcpy(new_elements, pool->elements,
157 pool->curr_nr * sizeof(*new_elements));
158 kfree(pool->elements);
159 pool->elements = new_elements;
160 pool->min_nr = new_min_nr;
162 while (pool->curr_nr < pool->min_nr) {
163 spin_unlock_irqrestore(&pool->lock, flags);
164 element = pool->alloc(gfp_mask, pool->pool_data);
167 spin_lock_irqsave(&pool->lock, flags);
168 if (pool->curr_nr < pool->min_nr) {
169 add_element(pool, element);
171 spin_unlock_irqrestore(&pool->lock, flags);
172 pool->free(element, pool->pool_data); /* Raced */
177 spin_unlock_irqrestore(&pool->lock, flags);
181 EXPORT_SYMBOL(mempool_resize);
184 * mempool_alloc - allocate an element from a specific memory pool
185 * @pool: pointer to the memory pool which was allocated via
187 * @gfp_mask: the usual allocation bitmask.
189 * this function only sleeps if the alloc_fn() function sleeps or
190 * returns NULL. Note that due to preallocation, this function
191 * *never* fails when called from process contexts. (it might
192 * fail if called from an IRQ context.)
194 void * mempool_alloc(mempool_t *pool, gfp_t gfp_mask)
201 might_sleep_if(gfp_mask & __GFP_WAIT);
203 gfp_mask |= __GFP_NOMEMALLOC; /* don't allocate emergency reserves */
204 gfp_mask |= __GFP_NORETRY; /* don't loop in __alloc_pages */
205 gfp_mask |= __GFP_NOWARN; /* failures are OK */
207 gfp_temp = gfp_mask & ~(__GFP_WAIT|__GFP_IO);
211 element = pool->alloc(gfp_temp, pool->pool_data);
212 if (likely(element != NULL))
215 spin_lock_irqsave(&pool->lock, flags);
216 if (likely(pool->curr_nr)) {
217 element = remove_element(pool);
218 spin_unlock_irqrestore(&pool->lock, flags);
219 /* paired with rmb in mempool_free(), read comment there */
224 /* We must not sleep in the GFP_ATOMIC case */
225 if (!(gfp_mask & __GFP_WAIT)) {
226 spin_unlock_irqrestore(&pool->lock, flags);
230 /* Let's wait for someone else to return an element to @pool */
233 prepare_to_wait(&pool->wait, &wait, TASK_UNINTERRUPTIBLE);
235 spin_unlock_irqrestore(&pool->lock, flags);
238 * FIXME: this should be io_schedule(). The timeout is there as a
239 * workaround for some DM problems in 2.6.18.
241 io_schedule_timeout(5*HZ);
243 finish_wait(&pool->wait, &wait);
246 EXPORT_SYMBOL(mempool_alloc);
249 * mempool_free - return an element to the pool.
250 * @element: pool element pointer.
251 * @pool: pointer to the memory pool which was allocated via
254 * this function only sleeps if the free_fn() function sleeps.
256 void mempool_free(void *element, mempool_t *pool)
260 if (unlikely(element == NULL))
264 * Paired with the wmb in mempool_alloc(). The preceding read is
265 * for @element and the following @pool->curr_nr. This ensures
266 * that the visible value of @pool->curr_nr is from after the
267 * allocation of @element. This is necessary for fringe cases
268 * where @element was passed to this task without going through
271 * For example, assume @p is %NULL at the beginning and one task
272 * performs "p = mempool_alloc(...);" while another task is doing
273 * "while (!p) cpu_relax(); mempool_free(p, ...);". This function
274 * may end up using curr_nr value which is from before allocation
275 * of @p without the following rmb.
280 * For correctness, we need a test which is guaranteed to trigger
281 * if curr_nr + #allocated == min_nr. Testing curr_nr < min_nr
282 * without locking achieves that and refilling as soon as possible
285 * Because curr_nr visible here is always a value after the
286 * allocation of @element, any task which decremented curr_nr below
287 * min_nr is guaranteed to see curr_nr < min_nr unless curr_nr gets
288 * incremented to min_nr afterwards. If curr_nr gets incremented
289 * to min_nr after the allocation of @element, the elements
290 * allocated after that are subject to the same guarantee.
292 * Waiters happen iff curr_nr is 0 and the above guarantee also
293 * ensures that there will be frees which return elements to the
294 * pool waking up the waiters.
296 if (pool->curr_nr < pool->min_nr) {
297 spin_lock_irqsave(&pool->lock, flags);
298 if (pool->curr_nr < pool->min_nr) {
299 add_element(pool, element);
300 spin_unlock_irqrestore(&pool->lock, flags);
301 wake_up(&pool->wait);
304 spin_unlock_irqrestore(&pool->lock, flags);
306 pool->free(element, pool->pool_data);
308 EXPORT_SYMBOL(mempool_free);
311 * A commonly used alloc and free fn.
313 void *mempool_alloc_slab(gfp_t gfp_mask, void *pool_data)
315 struct kmem_cache *mem = pool_data;
316 return kmem_cache_alloc(mem, gfp_mask);
318 EXPORT_SYMBOL(mempool_alloc_slab);
320 void mempool_free_slab(void *element, void *pool_data)
322 struct kmem_cache *mem = pool_data;
323 kmem_cache_free(mem, element);
325 EXPORT_SYMBOL(mempool_free_slab);
328 * A commonly used alloc and free fn that kmalloc/kfrees the amount of memory
329 * specified by pool_data
331 void *mempool_kmalloc(gfp_t gfp_mask, void *pool_data)
333 size_t size = (size_t)pool_data;
334 return kmalloc(size, gfp_mask);
336 EXPORT_SYMBOL(mempool_kmalloc);
338 void mempool_kfree(void *element, void *pool_data)
342 EXPORT_SYMBOL(mempool_kfree);
345 * A simple mempool-backed page allocator that allocates pages
346 * of the order specified by pool_data.
348 void *mempool_alloc_pages(gfp_t gfp_mask, void *pool_data)
350 int order = (int)(long)pool_data;
351 return alloc_pages(gfp_mask, order);
353 EXPORT_SYMBOL(mempool_alloc_pages);
355 void mempool_free_pages(void *element, void *pool_data)
357 int order = (int)(long)pool_data;
358 __free_pages(element, order);
360 EXPORT_SYMBOL(mempool_free_pages);