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 | { | |
a91a5ac6 TH |
66 | return mempool_create_node(min_nr,alloc_fn,free_fn, pool_data, |
67 | GFP_KERNEL, NUMA_NO_NODE); | |
1946089a CL |
68 | } |
69 | EXPORT_SYMBOL(mempool_create); | |
1da177e4 | 70 | |
1946089a | 71 | mempool_t *mempool_create_node(int min_nr, mempool_alloc_t *alloc_fn, |
a91a5ac6 TH |
72 | mempool_free_t *free_fn, void *pool_data, |
73 | gfp_t gfp_mask, int node_id) | |
1946089a CL |
74 | { |
75 | mempool_t *pool; | |
7b5219db | 76 | pool = kzalloc_node(sizeof(*pool), gfp_mask, node_id); |
1da177e4 LT |
77 | if (!pool) |
78 | return NULL; | |
1946089a | 79 | pool->elements = kmalloc_node(min_nr * sizeof(void *), |
a91a5ac6 | 80 | gfp_mask, node_id); |
1da177e4 LT |
81 | if (!pool->elements) { |
82 | kfree(pool); | |
83 | return NULL; | |
84 | } | |
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; | |
90 | pool->free = free_fn; | |
91 | ||
92 | /* | |
93 | * First pre-allocate the guaranteed number of buffers. | |
94 | */ | |
95 | while (pool->curr_nr < pool->min_nr) { | |
96 | void *element; | |
97 | ||
a91a5ac6 | 98 | element = pool->alloc(gfp_mask, pool->pool_data); |
1da177e4 | 99 | if (unlikely(!element)) { |
0565d317 | 100 | mempool_destroy(pool); |
1da177e4 LT |
101 | return NULL; |
102 | } | |
103 | add_element(pool, element); | |
104 | } | |
105 | return pool; | |
106 | } | |
1946089a | 107 | EXPORT_SYMBOL(mempool_create_node); |
1da177e4 LT |
108 | |
109 | /** | |
110 | * mempool_resize - resize an existing memory pool | |
111 | * @pool: pointer to the memory pool which was allocated via | |
112 | * mempool_create(). | |
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. | |
116 | * | |
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. | |
120 | * | |
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. | |
124 | */ | |
dd0fc66f | 125 | int mempool_resize(mempool_t *pool, int new_min_nr, gfp_t gfp_mask) |
1da177e4 LT |
126 | { |
127 | void *element; | |
128 | void **new_elements; | |
129 | unsigned long flags; | |
130 | ||
131 | BUG_ON(new_min_nr <= 0); | |
132 | ||
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); | |
140 | } | |
141 | pool->min_nr = new_min_nr; | |
142 | goto out_unlock; | |
143 | } | |
144 | spin_unlock_irqrestore(&pool->lock, flags); | |
145 | ||
146 | /* Grow the pool */ | |
147 | new_elements = kmalloc(new_min_nr * sizeof(*new_elements), gfp_mask); | |
148 | if (!new_elements) | |
149 | return -ENOMEM; | |
150 | ||
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); | |
155 | kfree(new_elements); | |
156 | goto out; | |
157 | } | |
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; | |
163 | ||
164 | while (pool->curr_nr < pool->min_nr) { | |
165 | spin_unlock_irqrestore(&pool->lock, flags); | |
166 | element = pool->alloc(gfp_mask, pool->pool_data); | |
167 | if (!element) | |
168 | goto out; | |
169 | spin_lock_irqsave(&pool->lock, flags); | |
170 | if (pool->curr_nr < pool->min_nr) { | |
171 | add_element(pool, element); | |
172 | } else { | |
173 | spin_unlock_irqrestore(&pool->lock, flags); | |
174 | pool->free(element, pool->pool_data); /* Raced */ | |
175 | goto out; | |
176 | } | |
177 | } | |
178 | out_unlock: | |
179 | spin_unlock_irqrestore(&pool->lock, flags); | |
180 | out: | |
181 | return 0; | |
182 | } | |
183 | EXPORT_SYMBOL(mempool_resize); | |
184 | ||
1da177e4 LT |
185 | /** |
186 | * mempool_alloc - allocate an element from a specific memory pool | |
187 | * @pool: pointer to the memory pool which was allocated via | |
188 | * mempool_create(). | |
189 | * @gfp_mask: the usual allocation bitmask. | |
190 | * | |
72fd4a35 | 191 | * this function only sleeps if the alloc_fn() function sleeps or |
1da177e4 LT |
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.) | |
8bf8fcb0 | 195 | * Note: using __GFP_ZERO is not supported. |
1da177e4 | 196 | */ |
dd0fc66f | 197 | void * mempool_alloc(mempool_t *pool, gfp_t gfp_mask) |
1da177e4 LT |
198 | { |
199 | void *element; | |
200 | unsigned long flags; | |
01890a4c | 201 | wait_queue_t wait; |
6daa0e28 | 202 | gfp_t gfp_temp; |
20a77776 | 203 | |
8bf8fcb0 | 204 | VM_WARN_ON_ONCE(gfp_mask & __GFP_ZERO); |
20a77776 | 205 | might_sleep_if(gfp_mask & __GFP_WAIT); |
b84a35be NP |
206 | |
207 | gfp_mask |= __GFP_NOMEMALLOC; /* don't allocate emergency reserves */ | |
208 | gfp_mask |= __GFP_NORETRY; /* don't loop in __alloc_pages */ | |
209 | gfp_mask |= __GFP_NOWARN; /* failures are OK */ | |
1da177e4 | 210 | |
20a77776 NP |
211 | gfp_temp = gfp_mask & ~(__GFP_WAIT|__GFP_IO); |
212 | ||
1da177e4 | 213 | repeat_alloc: |
20a77776 NP |
214 | |
215 | element = pool->alloc(gfp_temp, pool->pool_data); | |
1da177e4 LT |
216 | if (likely(element != NULL)) |
217 | return element; | |
218 | ||
1da177e4 LT |
219 | spin_lock_irqsave(&pool->lock, flags); |
220 | if (likely(pool->curr_nr)) { | |
221 | element = remove_element(pool); | |
222 | spin_unlock_irqrestore(&pool->lock, flags); | |
5b990546 TH |
223 | /* paired with rmb in mempool_free(), read comment there */ |
224 | smp_wmb(); | |
1da177e4 LT |
225 | return element; |
226 | } | |
1da177e4 | 227 | |
1ebb7044 TH |
228 | /* |
229 | * We use gfp mask w/o __GFP_WAIT or IO for the first round. If | |
230 | * alloc failed with that and @pool was empty, retry immediately. | |
231 | */ | |
232 | if (gfp_temp != gfp_mask) { | |
233 | spin_unlock_irqrestore(&pool->lock, flags); | |
234 | gfp_temp = gfp_mask; | |
235 | goto repeat_alloc; | |
236 | } | |
237 | ||
238 | /* We must not sleep if !__GFP_WAIT */ | |
5b990546 TH |
239 | if (!(gfp_mask & __GFP_WAIT)) { |
240 | spin_unlock_irqrestore(&pool->lock, flags); | |
1da177e4 | 241 | return NULL; |
5b990546 | 242 | } |
1da177e4 | 243 | |
5b990546 | 244 | /* Let's wait for someone else to return an element to @pool */ |
01890a4c | 245 | init_wait(&wait); |
1da177e4 | 246 | prepare_to_wait(&pool->wait, &wait, TASK_UNINTERRUPTIBLE); |
1da177e4 | 247 | |
5b990546 TH |
248 | spin_unlock_irqrestore(&pool->lock, flags); |
249 | ||
250 | /* | |
251 | * FIXME: this should be io_schedule(). The timeout is there as a | |
252 | * workaround for some DM problems in 2.6.18. | |
253 | */ | |
254 | io_schedule_timeout(5*HZ); | |
255 | ||
256 | finish_wait(&pool->wait, &wait); | |
1da177e4 LT |
257 | goto repeat_alloc; |
258 | } | |
259 | EXPORT_SYMBOL(mempool_alloc); | |
260 | ||
261 | /** | |
262 | * mempool_free - return an element to the pool. | |
263 | * @element: pool element pointer. | |
264 | * @pool: pointer to the memory pool which was allocated via | |
265 | * mempool_create(). | |
266 | * | |
267 | * this function only sleeps if the free_fn() function sleeps. | |
268 | */ | |
269 | void mempool_free(void *element, mempool_t *pool) | |
270 | { | |
271 | unsigned long flags; | |
272 | ||
c80e7a82 RR |
273 | if (unlikely(element == NULL)) |
274 | return; | |
275 | ||
5b990546 TH |
276 | /* |
277 | * Paired with the wmb in mempool_alloc(). The preceding read is | |
278 | * for @element and the following @pool->curr_nr. This ensures | |
279 | * that the visible value of @pool->curr_nr is from after the | |
280 | * allocation of @element. This is necessary for fringe cases | |
281 | * where @element was passed to this task without going through | |
282 | * barriers. | |
283 | * | |
284 | * For example, assume @p is %NULL at the beginning and one task | |
285 | * performs "p = mempool_alloc(...);" while another task is doing | |
286 | * "while (!p) cpu_relax(); mempool_free(p, ...);". This function | |
287 | * may end up using curr_nr value which is from before allocation | |
288 | * of @p without the following rmb. | |
289 | */ | |
290 | smp_rmb(); | |
291 | ||
292 | /* | |
293 | * For correctness, we need a test which is guaranteed to trigger | |
294 | * if curr_nr + #allocated == min_nr. Testing curr_nr < min_nr | |
295 | * without locking achieves that and refilling as soon as possible | |
296 | * is desirable. | |
297 | * | |
298 | * Because curr_nr visible here is always a value after the | |
299 | * allocation of @element, any task which decremented curr_nr below | |
300 | * min_nr is guaranteed to see curr_nr < min_nr unless curr_nr gets | |
301 | * incremented to min_nr afterwards. If curr_nr gets incremented | |
302 | * to min_nr after the allocation of @element, the elements | |
303 | * allocated after that are subject to the same guarantee. | |
304 | * | |
305 | * Waiters happen iff curr_nr is 0 and the above guarantee also | |
306 | * ensures that there will be frees which return elements to the | |
307 | * pool waking up the waiters. | |
308 | */ | |
eb9a3c62 | 309 | if (unlikely(pool->curr_nr < pool->min_nr)) { |
1da177e4 | 310 | spin_lock_irqsave(&pool->lock, flags); |
eb9a3c62 | 311 | if (likely(pool->curr_nr < pool->min_nr)) { |
1da177e4 LT |
312 | add_element(pool, element); |
313 | spin_unlock_irqrestore(&pool->lock, flags); | |
314 | wake_up(&pool->wait); | |
315 | return; | |
316 | } | |
317 | spin_unlock_irqrestore(&pool->lock, flags); | |
318 | } | |
319 | pool->free(element, pool->pool_data); | |
320 | } | |
321 | EXPORT_SYMBOL(mempool_free); | |
322 | ||
323 | /* | |
324 | * A commonly used alloc and free fn. | |
325 | */ | |
dd0fc66f | 326 | void *mempool_alloc_slab(gfp_t gfp_mask, void *pool_data) |
1da177e4 | 327 | { |
fcc234f8 | 328 | struct kmem_cache *mem = pool_data; |
1da177e4 LT |
329 | return kmem_cache_alloc(mem, gfp_mask); |
330 | } | |
331 | EXPORT_SYMBOL(mempool_alloc_slab); | |
332 | ||
333 | void mempool_free_slab(void *element, void *pool_data) | |
334 | { | |
fcc234f8 | 335 | struct kmem_cache *mem = pool_data; |
1da177e4 LT |
336 | kmem_cache_free(mem, element); |
337 | } | |
338 | EXPORT_SYMBOL(mempool_free_slab); | |
6e0678f3 | 339 | |
53184082 MD |
340 | /* |
341 | * A commonly used alloc and free fn that kmalloc/kfrees the amount of memory | |
183ff22b | 342 | * specified by pool_data |
53184082 MD |
343 | */ |
344 | void *mempool_kmalloc(gfp_t gfp_mask, void *pool_data) | |
345 | { | |
5e2f89b5 | 346 | size_t size = (size_t)pool_data; |
53184082 MD |
347 | return kmalloc(size, gfp_mask); |
348 | } | |
349 | EXPORT_SYMBOL(mempool_kmalloc); | |
350 | ||
351 | void mempool_kfree(void *element, void *pool_data) | |
352 | { | |
353 | kfree(element); | |
354 | } | |
355 | EXPORT_SYMBOL(mempool_kfree); | |
356 | ||
6e0678f3 MD |
357 | /* |
358 | * A simple mempool-backed page allocator that allocates pages | |
359 | * of the order specified by pool_data. | |
360 | */ | |
361 | void *mempool_alloc_pages(gfp_t gfp_mask, void *pool_data) | |
362 | { | |
363 | int order = (int)(long)pool_data; | |
364 | return alloc_pages(gfp_mask, order); | |
365 | } | |
366 | EXPORT_SYMBOL(mempool_alloc_pages); | |
367 | ||
368 | void mempool_free_pages(void *element, void *pool_data) | |
369 | { | |
370 | int order = (int)(long)pool_data; | |
371 | __free_pages(element, order); | |
372 | } | |
373 | EXPORT_SYMBOL(mempool_free_pages); |