]>
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> | |
17411962 | 13 | #include <linux/kmemleak.h> |
b95f1b31 | 14 | #include <linux/export.h> |
1da177e4 LT |
15 | #include <linux/mempool.h> |
16 | #include <linux/blkdev.h> | |
17 | #include <linux/writeback.h> | |
e244c9e6 | 18 | #include "slab.h" |
1da177e4 LT |
19 | |
20 | static void add_element(mempool_t *pool, void *element) | |
21 | { | |
22 | BUG_ON(pool->curr_nr >= pool->min_nr); | |
23 | pool->elements[pool->curr_nr++] = element; | |
24 | } | |
25 | ||
26 | static void *remove_element(mempool_t *pool) | |
27 | { | |
28 | BUG_ON(pool->curr_nr <= 0); | |
29 | return pool->elements[--pool->curr_nr]; | |
30 | } | |
31 | ||
0565d317 TH |
32 | /** |
33 | * mempool_destroy - deallocate a memory pool | |
34 | * @pool: pointer to the memory pool which was allocated via | |
35 | * mempool_create(). | |
36 | * | |
37 | * Free all reserved elements in @pool and @pool itself. This function | |
38 | * only sleeps if the free_fn() function sleeps. | |
39 | */ | |
40 | void mempool_destroy(mempool_t *pool) | |
1da177e4 LT |
41 | { |
42 | while (pool->curr_nr) { | |
43 | void *element = remove_element(pool); | |
44 | pool->free(element, pool->pool_data); | |
45 | } | |
46 | kfree(pool->elements); | |
47 | kfree(pool); | |
48 | } | |
0565d317 | 49 | EXPORT_SYMBOL(mempool_destroy); |
1da177e4 LT |
50 | |
51 | /** | |
52 | * mempool_create - create a memory pool | |
53 | * @min_nr: the minimum number of elements guaranteed to be | |
54 | * allocated for this pool. | |
55 | * @alloc_fn: user-defined element-allocation function. | |
56 | * @free_fn: user-defined element-freeing function. | |
57 | * @pool_data: optional private data available to the user-defined functions. | |
58 | * | |
59 | * this function creates and allocates a guaranteed size, preallocated | |
72fd4a35 | 60 | * memory pool. The pool can be used from the mempool_alloc() and mempool_free() |
1da177e4 | 61 | * functions. This function might sleep. Both the alloc_fn() and the free_fn() |
72fd4a35 | 62 | * functions might sleep - as long as the mempool_alloc() function is not called |
1da177e4 LT |
63 | * from IRQ contexts. |
64 | */ | |
1946089a | 65 | mempool_t *mempool_create(int min_nr, mempool_alloc_t *alloc_fn, |
1da177e4 LT |
66 | mempool_free_t *free_fn, void *pool_data) |
67 | { | |
a91a5ac6 TH |
68 | return mempool_create_node(min_nr,alloc_fn,free_fn, pool_data, |
69 | GFP_KERNEL, NUMA_NO_NODE); | |
1946089a CL |
70 | } |
71 | EXPORT_SYMBOL(mempool_create); | |
1da177e4 | 72 | |
1946089a | 73 | mempool_t *mempool_create_node(int min_nr, mempool_alloc_t *alloc_fn, |
a91a5ac6 TH |
74 | mempool_free_t *free_fn, void *pool_data, |
75 | gfp_t gfp_mask, int node_id) | |
1946089a CL |
76 | { |
77 | mempool_t *pool; | |
7b5219db | 78 | pool = kzalloc_node(sizeof(*pool), gfp_mask, node_id); |
1da177e4 LT |
79 | if (!pool) |
80 | return NULL; | |
1946089a | 81 | pool->elements = kmalloc_node(min_nr * sizeof(void *), |
a91a5ac6 | 82 | gfp_mask, node_id); |
1da177e4 LT |
83 | if (!pool->elements) { |
84 | kfree(pool); | |
85 | return NULL; | |
86 | } | |
87 | spin_lock_init(&pool->lock); | |
88 | pool->min_nr = min_nr; | |
89 | pool->pool_data = pool_data; | |
90 | init_waitqueue_head(&pool->wait); | |
91 | pool->alloc = alloc_fn; | |
92 | pool->free = free_fn; | |
93 | ||
94 | /* | |
95 | * First pre-allocate the guaranteed number of buffers. | |
96 | */ | |
97 | while (pool->curr_nr < pool->min_nr) { | |
98 | void *element; | |
99 | ||
a91a5ac6 | 100 | element = pool->alloc(gfp_mask, pool->pool_data); |
1da177e4 | 101 | if (unlikely(!element)) { |
0565d317 | 102 | mempool_destroy(pool); |
1da177e4 LT |
103 | return NULL; |
104 | } | |
105 | add_element(pool, element); | |
106 | } | |
107 | return pool; | |
108 | } | |
1946089a | 109 | EXPORT_SYMBOL(mempool_create_node); |
1da177e4 LT |
110 | |
111 | /** | |
112 | * mempool_resize - resize an existing memory pool | |
113 | * @pool: pointer to the memory pool which was allocated via | |
114 | * mempool_create(). | |
115 | * @new_min_nr: the new minimum number of elements guaranteed to be | |
116 | * allocated for this pool. | |
1da177e4 LT |
117 | * |
118 | * This function shrinks/grows the pool. In the case of growing, | |
119 | * it cannot be guaranteed that the pool will be grown to the new | |
120 | * size immediately, but new mempool_free() calls will refill it. | |
11d83360 | 121 | * This function may sleep. |
1da177e4 LT |
122 | * |
123 | * Note, the caller must guarantee that no mempool_destroy is called | |
124 | * while this function is running. mempool_alloc() & mempool_free() | |
125 | * might be called (eg. from IRQ contexts) while this function executes. | |
126 | */ | |
11d83360 | 127 | int mempool_resize(mempool_t *pool, int new_min_nr) |
1da177e4 LT |
128 | { |
129 | void *element; | |
130 | void **new_elements; | |
131 | unsigned long flags; | |
132 | ||
133 | BUG_ON(new_min_nr <= 0); | |
11d83360 | 134 | might_sleep(); |
1da177e4 LT |
135 | |
136 | spin_lock_irqsave(&pool->lock, flags); | |
137 | if (new_min_nr <= pool->min_nr) { | |
138 | while (new_min_nr < pool->curr_nr) { | |
139 | element = remove_element(pool); | |
140 | spin_unlock_irqrestore(&pool->lock, flags); | |
141 | pool->free(element, pool->pool_data); | |
142 | spin_lock_irqsave(&pool->lock, flags); | |
143 | } | |
144 | pool->min_nr = new_min_nr; | |
145 | goto out_unlock; | |
146 | } | |
147 | spin_unlock_irqrestore(&pool->lock, flags); | |
148 | ||
149 | /* Grow the pool */ | |
11d83360 DR |
150 | new_elements = kmalloc_array(new_min_nr, sizeof(*new_elements), |
151 | GFP_KERNEL); | |
1da177e4 LT |
152 | if (!new_elements) |
153 | return -ENOMEM; | |
154 | ||
155 | spin_lock_irqsave(&pool->lock, flags); | |
156 | if (unlikely(new_min_nr <= pool->min_nr)) { | |
157 | /* Raced, other resize will do our work */ | |
158 | spin_unlock_irqrestore(&pool->lock, flags); | |
159 | kfree(new_elements); | |
160 | goto out; | |
161 | } | |
162 | memcpy(new_elements, pool->elements, | |
163 | pool->curr_nr * sizeof(*new_elements)); | |
164 | kfree(pool->elements); | |
165 | pool->elements = new_elements; | |
166 | pool->min_nr = new_min_nr; | |
167 | ||
168 | while (pool->curr_nr < pool->min_nr) { | |
169 | spin_unlock_irqrestore(&pool->lock, flags); | |
11d83360 | 170 | element = pool->alloc(GFP_KERNEL, pool->pool_data); |
1da177e4 LT |
171 | if (!element) |
172 | goto out; | |
173 | spin_lock_irqsave(&pool->lock, flags); | |
174 | if (pool->curr_nr < pool->min_nr) { | |
175 | add_element(pool, element); | |
176 | } else { | |
177 | spin_unlock_irqrestore(&pool->lock, flags); | |
178 | pool->free(element, pool->pool_data); /* Raced */ | |
179 | goto out; | |
180 | } | |
181 | } | |
182 | out_unlock: | |
183 | spin_unlock_irqrestore(&pool->lock, flags); | |
184 | out: | |
185 | return 0; | |
186 | } | |
187 | EXPORT_SYMBOL(mempool_resize); | |
188 | ||
1da177e4 LT |
189 | /** |
190 | * mempool_alloc - allocate an element from a specific memory pool | |
191 | * @pool: pointer to the memory pool which was allocated via | |
192 | * mempool_create(). | |
193 | * @gfp_mask: the usual allocation bitmask. | |
194 | * | |
72fd4a35 | 195 | * this function only sleeps if the alloc_fn() function sleeps or |
1da177e4 LT |
196 | * returns NULL. Note that due to preallocation, this function |
197 | * *never* fails when called from process contexts. (it might | |
198 | * fail if called from an IRQ context.) | |
8bf8fcb0 | 199 | * Note: using __GFP_ZERO is not supported. |
1da177e4 | 200 | */ |
dd0fc66f | 201 | void * mempool_alloc(mempool_t *pool, gfp_t gfp_mask) |
1da177e4 LT |
202 | { |
203 | void *element; | |
204 | unsigned long flags; | |
01890a4c | 205 | wait_queue_t wait; |
6daa0e28 | 206 | gfp_t gfp_temp; |
20a77776 | 207 | |
8bf8fcb0 | 208 | VM_WARN_ON_ONCE(gfp_mask & __GFP_ZERO); |
20a77776 | 209 | might_sleep_if(gfp_mask & __GFP_WAIT); |
b84a35be NP |
210 | |
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 */ | |
1da177e4 | 214 | |
20a77776 NP |
215 | gfp_temp = gfp_mask & ~(__GFP_WAIT|__GFP_IO); |
216 | ||
1da177e4 | 217 | repeat_alloc: |
20a77776 NP |
218 | |
219 | element = pool->alloc(gfp_temp, pool->pool_data); | |
1da177e4 LT |
220 | if (likely(element != NULL)) |
221 | return element; | |
222 | ||
1da177e4 LT |
223 | spin_lock_irqsave(&pool->lock, flags); |
224 | if (likely(pool->curr_nr)) { | |
225 | element = remove_element(pool); | |
226 | spin_unlock_irqrestore(&pool->lock, flags); | |
5b990546 TH |
227 | /* paired with rmb in mempool_free(), read comment there */ |
228 | smp_wmb(); | |
17411962 CM |
229 | /* |
230 | * Update the allocation stack trace as this is more useful | |
231 | * for debugging. | |
232 | */ | |
233 | kmemleak_update_trace(element); | |
1da177e4 LT |
234 | return element; |
235 | } | |
1da177e4 | 236 | |
1ebb7044 TH |
237 | /* |
238 | * We use gfp mask w/o __GFP_WAIT or IO for the first round. If | |
239 | * alloc failed with that and @pool was empty, retry immediately. | |
240 | */ | |
241 | if (gfp_temp != gfp_mask) { | |
242 | spin_unlock_irqrestore(&pool->lock, flags); | |
243 | gfp_temp = gfp_mask; | |
244 | goto repeat_alloc; | |
245 | } | |
246 | ||
247 | /* We must not sleep if !__GFP_WAIT */ | |
5b990546 TH |
248 | if (!(gfp_mask & __GFP_WAIT)) { |
249 | spin_unlock_irqrestore(&pool->lock, flags); | |
1da177e4 | 250 | return NULL; |
5b990546 | 251 | } |
1da177e4 | 252 | |
5b990546 | 253 | /* Let's wait for someone else to return an element to @pool */ |
01890a4c | 254 | init_wait(&wait); |
1da177e4 | 255 | prepare_to_wait(&pool->wait, &wait, TASK_UNINTERRUPTIBLE); |
1da177e4 | 256 | |
5b990546 TH |
257 | spin_unlock_irqrestore(&pool->lock, flags); |
258 | ||
259 | /* | |
260 | * FIXME: this should be io_schedule(). The timeout is there as a | |
261 | * workaround for some DM problems in 2.6.18. | |
262 | */ | |
263 | io_schedule_timeout(5*HZ); | |
264 | ||
265 | finish_wait(&pool->wait, &wait); | |
1da177e4 LT |
266 | goto repeat_alloc; |
267 | } | |
268 | EXPORT_SYMBOL(mempool_alloc); | |
269 | ||
270 | /** | |
271 | * mempool_free - return an element to the pool. | |
272 | * @element: pool element pointer. | |
273 | * @pool: pointer to the memory pool which was allocated via | |
274 | * mempool_create(). | |
275 | * | |
276 | * this function only sleeps if the free_fn() function sleeps. | |
277 | */ | |
278 | void mempool_free(void *element, mempool_t *pool) | |
279 | { | |
280 | unsigned long flags; | |
281 | ||
c80e7a82 RR |
282 | if (unlikely(element == NULL)) |
283 | return; | |
284 | ||
5b990546 TH |
285 | /* |
286 | * Paired with the wmb in mempool_alloc(). The preceding read is | |
287 | * for @element and the following @pool->curr_nr. This ensures | |
288 | * that the visible value of @pool->curr_nr is from after the | |
289 | * allocation of @element. This is necessary for fringe cases | |
290 | * where @element was passed to this task without going through | |
291 | * barriers. | |
292 | * | |
293 | * For example, assume @p is %NULL at the beginning and one task | |
294 | * performs "p = mempool_alloc(...);" while another task is doing | |
295 | * "while (!p) cpu_relax(); mempool_free(p, ...);". This function | |
296 | * may end up using curr_nr value which is from before allocation | |
297 | * of @p without the following rmb. | |
298 | */ | |
299 | smp_rmb(); | |
300 | ||
301 | /* | |
302 | * For correctness, we need a test which is guaranteed to trigger | |
303 | * if curr_nr + #allocated == min_nr. Testing curr_nr < min_nr | |
304 | * without locking achieves that and refilling as soon as possible | |
305 | * is desirable. | |
306 | * | |
307 | * Because curr_nr visible here is always a value after the | |
308 | * allocation of @element, any task which decremented curr_nr below | |
309 | * min_nr is guaranteed to see curr_nr < min_nr unless curr_nr gets | |
310 | * incremented to min_nr afterwards. If curr_nr gets incremented | |
311 | * to min_nr after the allocation of @element, the elements | |
312 | * allocated after that are subject to the same guarantee. | |
313 | * | |
314 | * Waiters happen iff curr_nr is 0 and the above guarantee also | |
315 | * ensures that there will be frees which return elements to the | |
316 | * pool waking up the waiters. | |
317 | */ | |
eb9a3c62 | 318 | if (unlikely(pool->curr_nr < pool->min_nr)) { |
1da177e4 | 319 | spin_lock_irqsave(&pool->lock, flags); |
eb9a3c62 | 320 | if (likely(pool->curr_nr < pool->min_nr)) { |
1da177e4 LT |
321 | add_element(pool, element); |
322 | spin_unlock_irqrestore(&pool->lock, flags); | |
323 | wake_up(&pool->wait); | |
324 | return; | |
325 | } | |
326 | spin_unlock_irqrestore(&pool->lock, flags); | |
327 | } | |
328 | pool->free(element, pool->pool_data); | |
329 | } | |
330 | EXPORT_SYMBOL(mempool_free); | |
331 | ||
332 | /* | |
333 | * A commonly used alloc and free fn. | |
334 | */ | |
dd0fc66f | 335 | void *mempool_alloc_slab(gfp_t gfp_mask, void *pool_data) |
1da177e4 | 336 | { |
fcc234f8 | 337 | struct kmem_cache *mem = pool_data; |
e244c9e6 | 338 | VM_BUG_ON(mem->ctor); |
1da177e4 LT |
339 | return kmem_cache_alloc(mem, gfp_mask); |
340 | } | |
341 | EXPORT_SYMBOL(mempool_alloc_slab); | |
342 | ||
343 | void mempool_free_slab(void *element, void *pool_data) | |
344 | { | |
fcc234f8 | 345 | struct kmem_cache *mem = pool_data; |
1da177e4 LT |
346 | kmem_cache_free(mem, element); |
347 | } | |
348 | EXPORT_SYMBOL(mempool_free_slab); | |
6e0678f3 | 349 | |
53184082 MD |
350 | /* |
351 | * A commonly used alloc and free fn that kmalloc/kfrees the amount of memory | |
183ff22b | 352 | * specified by pool_data |
53184082 MD |
353 | */ |
354 | void *mempool_kmalloc(gfp_t gfp_mask, void *pool_data) | |
355 | { | |
5e2f89b5 | 356 | size_t size = (size_t)pool_data; |
53184082 MD |
357 | return kmalloc(size, gfp_mask); |
358 | } | |
359 | EXPORT_SYMBOL(mempool_kmalloc); | |
360 | ||
361 | void mempool_kfree(void *element, void *pool_data) | |
362 | { | |
363 | kfree(element); | |
364 | } | |
365 | EXPORT_SYMBOL(mempool_kfree); | |
366 | ||
6e0678f3 MD |
367 | /* |
368 | * A simple mempool-backed page allocator that allocates pages | |
369 | * of the order specified by pool_data. | |
370 | */ | |
371 | void *mempool_alloc_pages(gfp_t gfp_mask, void *pool_data) | |
372 | { | |
373 | int order = (int)(long)pool_data; | |
374 | return alloc_pages(gfp_mask, order); | |
375 | } | |
376 | EXPORT_SYMBOL(mempool_alloc_pages); | |
377 | ||
378 | void mempool_free_pages(void *element, void *pool_data) | |
379 | { | |
380 | int order = (int)(long)pool_data; | |
381 | __free_pages(element, order); | |
382 | } | |
383 | EXPORT_SYMBOL(mempool_free_pages); |