]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
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> | |
b95f1b31 | 13 | #include <linux/export.h> |
1da177e4 LT |
14 | #include <linux/mempool.h> |
15 | #include <linux/blkdev.h> | |
16 | #include <linux/writeback.h> | |
17 | ||
18 | static 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 | ||
24 | static void *remove_element(mempool_t *pool) | |
25 | { | |
26 | BUG_ON(pool->curr_nr <= 0); | |
27 | return pool->elements[--pool->curr_nr]; | |
28 | } | |
29 | ||
0565d317 TH |
30 | /** |
31 | * mempool_destroy - deallocate a memory pool | |
32 | * @pool: pointer to the memory pool which was allocated via | |
33 | * mempool_create(). | |
34 | * | |
35 | * Free all reserved elements in @pool and @pool itself. This function | |
36 | * only sleeps if the free_fn() function sleeps. | |
37 | */ | |
38 | void mempool_destroy(mempool_t *pool) | |
1da177e4 LT |
39 | { |
40 | while (pool->curr_nr) { | |
41 | void *element = remove_element(pool); | |
42 | pool->free(element, pool->pool_data); | |
43 | } | |
44 | kfree(pool->elements); | |
45 | kfree(pool); | |
46 | } | |
0565d317 | 47 | EXPORT_SYMBOL(mempool_destroy); |
1da177e4 LT |
48 | |
49 | /** | |
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. | |
56 | * | |
57 | * this function creates and allocates a guaranteed size, preallocated | |
72fd4a35 | 58 | * memory pool. The pool can be used from the mempool_alloc() and mempool_free() |
1da177e4 | 59 | * functions. This function might sleep. Both the alloc_fn() and the free_fn() |
72fd4a35 | 60 | * functions might sleep - as long as the mempool_alloc() function is not called |
1da177e4 LT |
61 | * from IRQ contexts. |
62 | */ | |
1946089a | 63 | mempool_t *mempool_create(int min_nr, mempool_alloc_t *alloc_fn, |
1da177e4 LT |
64 | mempool_free_t *free_fn, void *pool_data) |
65 | { | |
1946089a CL |
66 | return mempool_create_node(min_nr,alloc_fn,free_fn, pool_data,-1); |
67 | } | |
68 | EXPORT_SYMBOL(mempool_create); | |
1da177e4 | 69 | |
1946089a CL |
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) | |
72 | { | |
73 | mempool_t *pool; | |
94f6030c | 74 | pool = kmalloc_node(sizeof(*pool), GFP_KERNEL | __GFP_ZERO, node_id); |
1da177e4 LT |
75 | if (!pool) |
76 | return NULL; | |
1946089a CL |
77 | pool->elements = kmalloc_node(min_nr * sizeof(void *), |
78 | GFP_KERNEL, node_id); | |
1da177e4 LT |
79 | if (!pool->elements) { |
80 | kfree(pool); | |
81 | return NULL; | |
82 | } | |
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; | |
88 | pool->free = free_fn; | |
89 | ||
90 | /* | |
91 | * First pre-allocate the guaranteed number of buffers. | |
92 | */ | |
93 | while (pool->curr_nr < pool->min_nr) { | |
94 | void *element; | |
95 | ||
96 | element = pool->alloc(GFP_KERNEL, pool->pool_data); | |
97 | if (unlikely(!element)) { | |
0565d317 | 98 | mempool_destroy(pool); |
1da177e4 LT |
99 | return NULL; |
100 | } | |
101 | add_element(pool, element); | |
102 | } | |
103 | return pool; | |
104 | } | |
1946089a | 105 | EXPORT_SYMBOL(mempool_create_node); |
1da177e4 LT |
106 | |
107 | /** | |
108 | * mempool_resize - resize an existing memory pool | |
109 | * @pool: pointer to the memory pool which was allocated via | |
110 | * mempool_create(). | |
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. | |
114 | * | |
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. | |
118 | * | |
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. | |
122 | */ | |
dd0fc66f | 123 | int mempool_resize(mempool_t *pool, int new_min_nr, gfp_t gfp_mask) |
1da177e4 LT |
124 | { |
125 | void *element; | |
126 | void **new_elements; | |
127 | unsigned long flags; | |
128 | ||
129 | BUG_ON(new_min_nr <= 0); | |
130 | ||
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); | |
138 | } | |
139 | pool->min_nr = new_min_nr; | |
140 | goto out_unlock; | |
141 | } | |
142 | spin_unlock_irqrestore(&pool->lock, flags); | |
143 | ||
144 | /* Grow the pool */ | |
145 | new_elements = kmalloc(new_min_nr * sizeof(*new_elements), gfp_mask); | |
146 | if (!new_elements) | |
147 | return -ENOMEM; | |
148 | ||
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); | |
153 | kfree(new_elements); | |
154 | goto out; | |
155 | } | |
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; | |
161 | ||
162 | while (pool->curr_nr < pool->min_nr) { | |
163 | spin_unlock_irqrestore(&pool->lock, flags); | |
164 | element = pool->alloc(gfp_mask, pool->pool_data); | |
165 | if (!element) | |
166 | goto out; | |
167 | spin_lock_irqsave(&pool->lock, flags); | |
168 | if (pool->curr_nr < pool->min_nr) { | |
169 | add_element(pool, element); | |
170 | } else { | |
171 | spin_unlock_irqrestore(&pool->lock, flags); | |
172 | pool->free(element, pool->pool_data); /* Raced */ | |
173 | goto out; | |
174 | } | |
175 | } | |
176 | out_unlock: | |
177 | spin_unlock_irqrestore(&pool->lock, flags); | |
178 | out: | |
179 | return 0; | |
180 | } | |
181 | EXPORT_SYMBOL(mempool_resize); | |
182 | ||
1da177e4 LT |
183 | /** |
184 | * mempool_alloc - allocate an element from a specific memory pool | |
185 | * @pool: pointer to the memory pool which was allocated via | |
186 | * mempool_create(). | |
187 | * @gfp_mask: the usual allocation bitmask. | |
188 | * | |
72fd4a35 | 189 | * this function only sleeps if the alloc_fn() function sleeps or |
1da177e4 LT |
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.) | |
193 | */ | |
dd0fc66f | 194 | void * mempool_alloc(mempool_t *pool, gfp_t gfp_mask) |
1da177e4 LT |
195 | { |
196 | void *element; | |
197 | unsigned long flags; | |
01890a4c | 198 | wait_queue_t wait; |
6daa0e28 | 199 | gfp_t gfp_temp; |
20a77776 NP |
200 | |
201 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
b84a35be NP |
202 | |
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 */ | |
1da177e4 | 206 | |
20a77776 NP |
207 | gfp_temp = gfp_mask & ~(__GFP_WAIT|__GFP_IO); |
208 | ||
1da177e4 | 209 | repeat_alloc: |
20a77776 NP |
210 | |
211 | element = pool->alloc(gfp_temp, pool->pool_data); | |
1da177e4 LT |
212 | if (likely(element != NULL)) |
213 | return element; | |
214 | ||
1da177e4 LT |
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); | |
5b990546 TH |
219 | /* paired with rmb in mempool_free(), read comment there */ |
220 | smp_wmb(); | |
1da177e4 LT |
221 | return element; |
222 | } | |
1da177e4 LT |
223 | |
224 | /* We must not sleep in the GFP_ATOMIC case */ | |
5b990546 TH |
225 | if (!(gfp_mask & __GFP_WAIT)) { |
226 | spin_unlock_irqrestore(&pool->lock, flags); | |
1da177e4 | 227 | return NULL; |
5b990546 | 228 | } |
1da177e4 | 229 | |
5b990546 | 230 | /* Let's wait for someone else to return an element to @pool */ |
20a77776 | 231 | gfp_temp = gfp_mask; |
01890a4c | 232 | init_wait(&wait); |
1da177e4 | 233 | prepare_to_wait(&pool->wait, &wait, TASK_UNINTERRUPTIBLE); |
1da177e4 | 234 | |
5b990546 TH |
235 | spin_unlock_irqrestore(&pool->lock, flags); |
236 | ||
237 | /* | |
238 | * FIXME: this should be io_schedule(). The timeout is there as a | |
239 | * workaround for some DM problems in 2.6.18. | |
240 | */ | |
241 | io_schedule_timeout(5*HZ); | |
242 | ||
243 | finish_wait(&pool->wait, &wait); | |
1da177e4 LT |
244 | goto repeat_alloc; |
245 | } | |
246 | EXPORT_SYMBOL(mempool_alloc); | |
247 | ||
248 | /** | |
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 | |
252 | * mempool_create(). | |
253 | * | |
254 | * this function only sleeps if the free_fn() function sleeps. | |
255 | */ | |
256 | void mempool_free(void *element, mempool_t *pool) | |
257 | { | |
258 | unsigned long flags; | |
259 | ||
c80e7a82 RR |
260 | if (unlikely(element == NULL)) |
261 | return; | |
262 | ||
5b990546 TH |
263 | /* |
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 | |
269 | * barriers. | |
270 | * | |
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. | |
276 | */ | |
277 | smp_rmb(); | |
278 | ||
279 | /* | |
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 | |
283 | * is desirable. | |
284 | * | |
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. | |
291 | * | |
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. | |
295 | */ | |
1da177e4 LT |
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); | |
302 | return; | |
303 | } | |
304 | spin_unlock_irqrestore(&pool->lock, flags); | |
305 | } | |
306 | pool->free(element, pool->pool_data); | |
307 | } | |
308 | EXPORT_SYMBOL(mempool_free); | |
309 | ||
310 | /* | |
311 | * A commonly used alloc and free fn. | |
312 | */ | |
dd0fc66f | 313 | void *mempool_alloc_slab(gfp_t gfp_mask, void *pool_data) |
1da177e4 | 314 | { |
fcc234f8 | 315 | struct kmem_cache *mem = pool_data; |
1da177e4 LT |
316 | return kmem_cache_alloc(mem, gfp_mask); |
317 | } | |
318 | EXPORT_SYMBOL(mempool_alloc_slab); | |
319 | ||
320 | void mempool_free_slab(void *element, void *pool_data) | |
321 | { | |
fcc234f8 | 322 | struct kmem_cache *mem = pool_data; |
1da177e4 LT |
323 | kmem_cache_free(mem, element); |
324 | } | |
325 | EXPORT_SYMBOL(mempool_free_slab); | |
6e0678f3 | 326 | |
53184082 MD |
327 | /* |
328 | * A commonly used alloc and free fn that kmalloc/kfrees the amount of memory | |
183ff22b | 329 | * specified by pool_data |
53184082 MD |
330 | */ |
331 | void *mempool_kmalloc(gfp_t gfp_mask, void *pool_data) | |
332 | { | |
5e2f89b5 | 333 | size_t size = (size_t)pool_data; |
53184082 MD |
334 | return kmalloc(size, gfp_mask); |
335 | } | |
336 | EXPORT_SYMBOL(mempool_kmalloc); | |
337 | ||
338 | void mempool_kfree(void *element, void *pool_data) | |
339 | { | |
340 | kfree(element); | |
341 | } | |
342 | EXPORT_SYMBOL(mempool_kfree); | |
343 | ||
6e0678f3 MD |
344 | /* |
345 | * A simple mempool-backed page allocator that allocates pages | |
346 | * of the order specified by pool_data. | |
347 | */ | |
348 | void *mempool_alloc_pages(gfp_t gfp_mask, void *pool_data) | |
349 | { | |
350 | int order = (int)(long)pool_data; | |
351 | return alloc_pages(gfp_mask, order); | |
352 | } | |
353 | EXPORT_SYMBOL(mempool_alloc_pages); | |
354 | ||
355 | void mempool_free_pages(void *element, void *pool_data) | |
356 | { | |
357 | int order = (int)(long)pool_data; | |
358 | __free_pages(element, order); | |
359 | } | |
360 | EXPORT_SYMBOL(mempool_free_pages); |