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,
67 GFP_KERNEL, NUMA_NO_NODE);
69 EXPORT_SYMBOL(mempool_create);
71 mempool_t *mempool_create_node(int min_nr, mempool_alloc_t *alloc_fn,
72 mempool_free_t *free_fn, void *pool_data,
73 gfp_t gfp_mask, int node_id)
76 pool = kmalloc_node(sizeof(*pool), gfp_mask | __GFP_ZERO, node_id);
79 pool->elements = kmalloc_node(min_nr * sizeof(void *),
81 if (!pool->elements) {
85 spin_lock_init(&pool->lock);
86 pool->min_nr = min_nr;
87 pool->pool_data = pool_data;
88 init_waitqueue_head(&pool->wait);
89 pool->alloc = alloc_fn;
93 * First pre-allocate the guaranteed number of buffers.
95 while (pool->curr_nr < pool->min_nr) {
98 element = pool->alloc(gfp_mask, pool->pool_data);
99 if (unlikely(!element)) {
100 mempool_destroy(pool);
103 add_element(pool, element);
107 EXPORT_SYMBOL(mempool_create_node);
110 * mempool_resize - resize an existing memory pool
111 * @pool: pointer to the memory pool which was allocated via
113 * @new_min_nr: the new minimum number of elements guaranteed to be
114 * allocated for this pool.
115 * @gfp_mask: the usual allocation bitmask.
117 * This function shrinks/grows the pool. In the case of growing,
118 * it cannot be guaranteed that the pool will be grown to the new
119 * size immediately, but new mempool_free() calls will refill it.
121 * Note, the caller must guarantee that no mempool_destroy is called
122 * while this function is running. mempool_alloc() & mempool_free()
123 * might be called (eg. from IRQ contexts) while this function executes.
125 int mempool_resize(mempool_t *pool, int new_min_nr, gfp_t gfp_mask)
131 BUG_ON(new_min_nr <= 0);
133 spin_lock_irqsave(&pool->lock, flags);
134 if (new_min_nr <= pool->min_nr) {
135 while (new_min_nr < pool->curr_nr) {
136 element = remove_element(pool);
137 spin_unlock_irqrestore(&pool->lock, flags);
138 pool->free(element, pool->pool_data);
139 spin_lock_irqsave(&pool->lock, flags);
141 pool->min_nr = new_min_nr;
144 spin_unlock_irqrestore(&pool->lock, flags);
147 new_elements = kmalloc(new_min_nr * sizeof(*new_elements), gfp_mask);
151 spin_lock_irqsave(&pool->lock, flags);
152 if (unlikely(new_min_nr <= pool->min_nr)) {
153 /* Raced, other resize will do our work */
154 spin_unlock_irqrestore(&pool->lock, flags);
158 memcpy(new_elements, pool->elements,
159 pool->curr_nr * sizeof(*new_elements));
160 kfree(pool->elements);
161 pool->elements = new_elements;
162 pool->min_nr = new_min_nr;
164 while (pool->curr_nr < pool->min_nr) {
165 spin_unlock_irqrestore(&pool->lock, flags);
166 element = pool->alloc(gfp_mask, pool->pool_data);
169 spin_lock_irqsave(&pool->lock, flags);
170 if (pool->curr_nr < pool->min_nr) {
171 add_element(pool, element);
173 spin_unlock_irqrestore(&pool->lock, flags);
174 pool->free(element, pool->pool_data); /* Raced */
179 spin_unlock_irqrestore(&pool->lock, flags);
183 EXPORT_SYMBOL(mempool_resize);
186 * mempool_alloc - allocate an element from a specific memory pool
187 * @pool: pointer to the memory pool which was allocated via
189 * @gfp_mask: the usual allocation bitmask.
191 * this function only sleeps if the alloc_fn() function sleeps or
192 * returns NULL. Note that due to preallocation, this function
193 * *never* fails when called from process contexts. (it might
194 * fail if called from an IRQ context.)
196 void * mempool_alloc(mempool_t *pool, gfp_t gfp_mask)
203 might_sleep_if(gfp_mask & __GFP_WAIT);
205 gfp_mask |= __GFP_NOMEMALLOC; /* don't allocate emergency reserves */
206 gfp_mask |= __GFP_NORETRY; /* don't loop in __alloc_pages */
207 gfp_mask |= __GFP_NOWARN; /* failures are OK */
209 gfp_temp = gfp_mask & ~(__GFP_WAIT|__GFP_IO);
213 element = pool->alloc(gfp_temp, pool->pool_data);
214 if (likely(element != NULL))
217 spin_lock_irqsave(&pool->lock, flags);
218 if (likely(pool->curr_nr)) {
219 element = remove_element(pool);
220 spin_unlock_irqrestore(&pool->lock, flags);
221 /* paired with rmb in mempool_free(), read comment there */
227 * We use gfp mask w/o __GFP_WAIT or IO for the first round. If
228 * alloc failed with that and @pool was empty, retry immediately.
230 if (gfp_temp != gfp_mask) {
231 spin_unlock_irqrestore(&pool->lock, flags);
236 /* We must not sleep if !__GFP_WAIT */
237 if (!(gfp_mask & __GFP_WAIT)) {
238 spin_unlock_irqrestore(&pool->lock, flags);
242 /* Let's wait for someone else to return an element to @pool */
244 prepare_to_wait(&pool->wait, &wait, TASK_UNINTERRUPTIBLE);
246 spin_unlock_irqrestore(&pool->lock, flags);
249 * FIXME: this should be io_schedule(). The timeout is there as a
250 * workaround for some DM problems in 2.6.18.
252 io_schedule_timeout(5*HZ);
254 finish_wait(&pool->wait, &wait);
257 EXPORT_SYMBOL(mempool_alloc);
260 * mempool_free - return an element to the pool.
261 * @element: pool element pointer.
262 * @pool: pointer to the memory pool which was allocated via
265 * this function only sleeps if the free_fn() function sleeps.
267 void mempool_free(void *element, mempool_t *pool)
271 if (unlikely(element == NULL))
275 * Paired with the wmb in mempool_alloc(). The preceding read is
276 * for @element and the following @pool->curr_nr. This ensures
277 * that the visible value of @pool->curr_nr is from after the
278 * allocation of @element. This is necessary for fringe cases
279 * where @element was passed to this task without going through
282 * For example, assume @p is %NULL at the beginning and one task
283 * performs "p = mempool_alloc(...);" while another task is doing
284 * "while (!p) cpu_relax(); mempool_free(p, ...);". This function
285 * may end up using curr_nr value which is from before allocation
286 * of @p without the following rmb.
291 * For correctness, we need a test which is guaranteed to trigger
292 * if curr_nr + #allocated == min_nr. Testing curr_nr < min_nr
293 * without locking achieves that and refilling as soon as possible
296 * Because curr_nr visible here is always a value after the
297 * allocation of @element, any task which decremented curr_nr below
298 * min_nr is guaranteed to see curr_nr < min_nr unless curr_nr gets
299 * incremented to min_nr afterwards. If curr_nr gets incremented
300 * to min_nr after the allocation of @element, the elements
301 * allocated after that are subject to the same guarantee.
303 * Waiters happen iff curr_nr is 0 and the above guarantee also
304 * ensures that there will be frees which return elements to the
305 * pool waking up the waiters.
307 if (pool->curr_nr < pool->min_nr) {
308 spin_lock_irqsave(&pool->lock, flags);
309 if (pool->curr_nr < pool->min_nr) {
310 add_element(pool, element);
311 spin_unlock_irqrestore(&pool->lock, flags);
312 wake_up(&pool->wait);
315 spin_unlock_irqrestore(&pool->lock, flags);
317 pool->free(element, pool->pool_data);
319 EXPORT_SYMBOL(mempool_free);
322 * A commonly used alloc and free fn.
324 void *mempool_alloc_slab(gfp_t gfp_mask, void *pool_data)
326 struct kmem_cache *mem = pool_data;
327 return kmem_cache_alloc(mem, gfp_mask);
329 EXPORT_SYMBOL(mempool_alloc_slab);
331 void mempool_free_slab(void *element, void *pool_data)
333 struct kmem_cache *mem = pool_data;
334 kmem_cache_free(mem, element);
336 EXPORT_SYMBOL(mempool_free_slab);
339 * A commonly used alloc and free fn that kmalloc/kfrees the amount of memory
340 * specified by pool_data
342 void *mempool_kmalloc(gfp_t gfp_mask, void *pool_data)
344 size_t size = (size_t)pool_data;
345 return kmalloc(size, gfp_mask);
347 EXPORT_SYMBOL(mempool_kmalloc);
349 void mempool_kfree(void *element, void *pool_data)
353 EXPORT_SYMBOL(mempool_kfree);
356 * A simple mempool-backed page allocator that allocates pages
357 * of the order specified by pool_data.
359 void *mempool_alloc_pages(gfp_t gfp_mask, void *pool_data)
361 int order = (int)(long)pool_data;
362 return alloc_pages(gfp_mask, order);
364 EXPORT_SYMBOL(mempool_alloc_pages);
366 void mempool_free_pages(void *element, void *pool_data)
368 int order = (int)(long)pool_data;
369 __free_pages(element, order);
371 EXPORT_SYMBOL(mempool_free_pages);