]>
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> | |
18 | ||
19 | static void add_element(mempool_t *pool, void *element) | |
20 | { | |
21 | BUG_ON(pool->curr_nr >= pool->min_nr); | |
22 | pool->elements[pool->curr_nr++] = element; | |
23 | } | |
24 | ||
25 | static void *remove_element(mempool_t *pool) | |
26 | { | |
27 | BUG_ON(pool->curr_nr <= 0); | |
28 | return pool->elements[--pool->curr_nr]; | |
29 | } | |
30 | ||
0565d317 TH |
31 | /** |
32 | * mempool_destroy - deallocate a memory pool | |
33 | * @pool: pointer to the memory pool which was allocated via | |
34 | * mempool_create(). | |
35 | * | |
36 | * Free all reserved elements in @pool and @pool itself. This function | |
37 | * only sleeps if the free_fn() function sleeps. | |
38 | */ | |
39 | void mempool_destroy(mempool_t *pool) | |
1da177e4 LT |
40 | { |
41 | while (pool->curr_nr) { | |
42 | void *element = remove_element(pool); | |
43 | pool->free(element, pool->pool_data); | |
44 | } | |
45 | kfree(pool->elements); | |
46 | kfree(pool); | |
47 | } | |
0565d317 | 48 | EXPORT_SYMBOL(mempool_destroy); |
1da177e4 LT |
49 | |
50 | /** | |
51 | * mempool_create - create a memory pool | |
52 | * @min_nr: the minimum number of elements guaranteed to be | |
53 | * allocated for this pool. | |
54 | * @alloc_fn: user-defined element-allocation function. | |
55 | * @free_fn: user-defined element-freeing function. | |
56 | * @pool_data: optional private data available to the user-defined functions. | |
57 | * | |
58 | * this function creates and allocates a guaranteed size, preallocated | |
72fd4a35 | 59 | * memory pool. The pool can be used from the mempool_alloc() and mempool_free() |
1da177e4 | 60 | * functions. This function might sleep. Both the alloc_fn() and the free_fn() |
72fd4a35 | 61 | * functions might sleep - as long as the mempool_alloc() function is not called |
1da177e4 LT |
62 | * from IRQ contexts. |
63 | */ | |
1946089a | 64 | mempool_t *mempool_create(int min_nr, mempool_alloc_t *alloc_fn, |
1da177e4 LT |
65 | mempool_free_t *free_fn, void *pool_data) |
66 | { | |
a91a5ac6 TH |
67 | return mempool_create_node(min_nr,alloc_fn,free_fn, pool_data, |
68 | GFP_KERNEL, NUMA_NO_NODE); | |
1946089a CL |
69 | } |
70 | EXPORT_SYMBOL(mempool_create); | |
1da177e4 | 71 | |
1946089a | 72 | mempool_t *mempool_create_node(int min_nr, mempool_alloc_t *alloc_fn, |
a91a5ac6 TH |
73 | mempool_free_t *free_fn, void *pool_data, |
74 | gfp_t gfp_mask, int node_id) | |
1946089a CL |
75 | { |
76 | mempool_t *pool; | |
7b5219db | 77 | pool = kzalloc_node(sizeof(*pool), gfp_mask, node_id); |
1da177e4 LT |
78 | if (!pool) |
79 | return NULL; | |
1946089a | 80 | pool->elements = kmalloc_node(min_nr * sizeof(void *), |
a91a5ac6 | 81 | gfp_mask, node_id); |
1da177e4 LT |
82 | if (!pool->elements) { |
83 | kfree(pool); | |
84 | return NULL; | |
85 | } | |
86 | spin_lock_init(&pool->lock); | |
87 | pool->min_nr = min_nr; | |
88 | pool->pool_data = pool_data; | |
89 | init_waitqueue_head(&pool->wait); | |
90 | pool->alloc = alloc_fn; | |
91 | pool->free = free_fn; | |
92 | ||
93 | /* | |
94 | * First pre-allocate the guaranteed number of buffers. | |
95 | */ | |
96 | while (pool->curr_nr < pool->min_nr) { | |
97 | void *element; | |
98 | ||
a91a5ac6 | 99 | element = pool->alloc(gfp_mask, pool->pool_data); |
1da177e4 | 100 | if (unlikely(!element)) { |
0565d317 | 101 | mempool_destroy(pool); |
1da177e4 LT |
102 | return NULL; |
103 | } | |
104 | add_element(pool, element); | |
105 | } | |
106 | return pool; | |
107 | } | |
1946089a | 108 | EXPORT_SYMBOL(mempool_create_node); |
1da177e4 LT |
109 | |
110 | /** | |
111 | * mempool_resize - resize an existing memory pool | |
112 | * @pool: pointer to the memory pool which was allocated via | |
113 | * mempool_create(). | |
114 | * @new_min_nr: the new minimum number of elements guaranteed to be | |
115 | * allocated for this pool. | |
1da177e4 LT |
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. | |
11d83360 | 120 | * This function may sleep. |
1da177e4 LT |
121 | * |
122 | * Note, the caller must guarantee that no mempool_destroy is called | |
123 | * while this function is running. mempool_alloc() & mempool_free() | |
124 | * might be called (eg. from IRQ contexts) while this function executes. | |
125 | */ | |
11d83360 | 126 | int mempool_resize(mempool_t *pool, int new_min_nr) |
1da177e4 LT |
127 | { |
128 | void *element; | |
129 | void **new_elements; | |
130 | unsigned long flags; | |
131 | ||
132 | BUG_ON(new_min_nr <= 0); | |
11d83360 | 133 | might_sleep(); |
1da177e4 LT |
134 | |
135 | spin_lock_irqsave(&pool->lock, flags); | |
136 | if (new_min_nr <= pool->min_nr) { | |
137 | while (new_min_nr < pool->curr_nr) { | |
138 | element = remove_element(pool); | |
139 | spin_unlock_irqrestore(&pool->lock, flags); | |
140 | pool->free(element, pool->pool_data); | |
141 | spin_lock_irqsave(&pool->lock, flags); | |
142 | } | |
143 | pool->min_nr = new_min_nr; | |
144 | goto out_unlock; | |
145 | } | |
146 | spin_unlock_irqrestore(&pool->lock, flags); | |
147 | ||
148 | /* Grow the pool */ | |
11d83360 DR |
149 | new_elements = kmalloc_array(new_min_nr, sizeof(*new_elements), |
150 | GFP_KERNEL); | |
1da177e4 LT |
151 | if (!new_elements) |
152 | return -ENOMEM; | |
153 | ||
154 | spin_lock_irqsave(&pool->lock, flags); | |
155 | if (unlikely(new_min_nr <= pool->min_nr)) { | |
156 | /* Raced, other resize will do our work */ | |
157 | spin_unlock_irqrestore(&pool->lock, flags); | |
158 | kfree(new_elements); | |
159 | goto out; | |
160 | } | |
161 | memcpy(new_elements, pool->elements, | |
162 | pool->curr_nr * sizeof(*new_elements)); | |
163 | kfree(pool->elements); | |
164 | pool->elements = new_elements; | |
165 | pool->min_nr = new_min_nr; | |
166 | ||
167 | while (pool->curr_nr < pool->min_nr) { | |
168 | spin_unlock_irqrestore(&pool->lock, flags); | |
11d83360 | 169 | element = pool->alloc(GFP_KERNEL, pool->pool_data); |
1da177e4 LT |
170 | if (!element) |
171 | goto out; | |
172 | spin_lock_irqsave(&pool->lock, flags); | |
173 | if (pool->curr_nr < pool->min_nr) { | |
174 | add_element(pool, element); | |
175 | } else { | |
176 | spin_unlock_irqrestore(&pool->lock, flags); | |
177 | pool->free(element, pool->pool_data); /* Raced */ | |
178 | goto out; | |
179 | } | |
180 | } | |
181 | out_unlock: | |
182 | spin_unlock_irqrestore(&pool->lock, flags); | |
183 | out: | |
184 | return 0; | |
185 | } | |
186 | EXPORT_SYMBOL(mempool_resize); | |
187 | ||
1da177e4 LT |
188 | /** |
189 | * mempool_alloc - allocate an element from a specific memory pool | |
190 | * @pool: pointer to the memory pool which was allocated via | |
191 | * mempool_create(). | |
192 | * @gfp_mask: the usual allocation bitmask. | |
193 | * | |
72fd4a35 | 194 | * this function only sleeps if the alloc_fn() function sleeps or |
1da177e4 LT |
195 | * returns NULL. Note that due to preallocation, this function |
196 | * *never* fails when called from process contexts. (it might | |
197 | * fail if called from an IRQ context.) | |
8bf8fcb0 | 198 | * Note: using __GFP_ZERO is not supported. |
1da177e4 | 199 | */ |
dd0fc66f | 200 | void * mempool_alloc(mempool_t *pool, gfp_t gfp_mask) |
1da177e4 LT |
201 | { |
202 | void *element; | |
203 | unsigned long flags; | |
01890a4c | 204 | wait_queue_t wait; |
6daa0e28 | 205 | gfp_t gfp_temp; |
20a77776 | 206 | |
8bf8fcb0 | 207 | VM_WARN_ON_ONCE(gfp_mask & __GFP_ZERO); |
20a77776 | 208 | might_sleep_if(gfp_mask & __GFP_WAIT); |
b84a35be NP |
209 | |
210 | gfp_mask |= __GFP_NOMEMALLOC; /* don't allocate emergency reserves */ | |
211 | gfp_mask |= __GFP_NORETRY; /* don't loop in __alloc_pages */ | |
212 | gfp_mask |= __GFP_NOWARN; /* failures are OK */ | |
1da177e4 | 213 | |
20a77776 NP |
214 | gfp_temp = gfp_mask & ~(__GFP_WAIT|__GFP_IO); |
215 | ||
1da177e4 | 216 | repeat_alloc: |
20a77776 NP |
217 | |
218 | element = pool->alloc(gfp_temp, pool->pool_data); | |
1da177e4 LT |
219 | if (likely(element != NULL)) |
220 | return element; | |
221 | ||
1da177e4 LT |
222 | spin_lock_irqsave(&pool->lock, flags); |
223 | if (likely(pool->curr_nr)) { | |
224 | element = remove_element(pool); | |
225 | spin_unlock_irqrestore(&pool->lock, flags); | |
5b990546 TH |
226 | /* paired with rmb in mempool_free(), read comment there */ |
227 | smp_wmb(); | |
17411962 CM |
228 | /* |
229 | * Update the allocation stack trace as this is more useful | |
230 | * for debugging. | |
231 | */ | |
232 | kmemleak_update_trace(element); | |
1da177e4 LT |
233 | return element; |
234 | } | |
1da177e4 | 235 | |
1ebb7044 TH |
236 | /* |
237 | * We use gfp mask w/o __GFP_WAIT or IO for the first round. If | |
238 | * alloc failed with that and @pool was empty, retry immediately. | |
239 | */ | |
240 | if (gfp_temp != gfp_mask) { | |
241 | spin_unlock_irqrestore(&pool->lock, flags); | |
242 | gfp_temp = gfp_mask; | |
243 | goto repeat_alloc; | |
244 | } | |
245 | ||
246 | /* We must not sleep if !__GFP_WAIT */ | |
5b990546 TH |
247 | if (!(gfp_mask & __GFP_WAIT)) { |
248 | spin_unlock_irqrestore(&pool->lock, flags); | |
1da177e4 | 249 | return NULL; |
5b990546 | 250 | } |
1da177e4 | 251 | |
5b990546 | 252 | /* Let's wait for someone else to return an element to @pool */ |
01890a4c | 253 | init_wait(&wait); |
1da177e4 | 254 | prepare_to_wait(&pool->wait, &wait, TASK_UNINTERRUPTIBLE); |
1da177e4 | 255 | |
5b990546 TH |
256 | spin_unlock_irqrestore(&pool->lock, flags); |
257 | ||
258 | /* | |
259 | * FIXME: this should be io_schedule(). The timeout is there as a | |
260 | * workaround for some DM problems in 2.6.18. | |
261 | */ | |
262 | io_schedule_timeout(5*HZ); | |
263 | ||
264 | finish_wait(&pool->wait, &wait); | |
1da177e4 LT |
265 | goto repeat_alloc; |
266 | } | |
267 | EXPORT_SYMBOL(mempool_alloc); | |
268 | ||
269 | /** | |
270 | * mempool_free - return an element to the pool. | |
271 | * @element: pool element pointer. | |
272 | * @pool: pointer to the memory pool which was allocated via | |
273 | * mempool_create(). | |
274 | * | |
275 | * this function only sleeps if the free_fn() function sleeps. | |
276 | */ | |
277 | void mempool_free(void *element, mempool_t *pool) | |
278 | { | |
279 | unsigned long flags; | |
280 | ||
c80e7a82 RR |
281 | if (unlikely(element == NULL)) |
282 | return; | |
283 | ||
5b990546 TH |
284 | /* |
285 | * Paired with the wmb in mempool_alloc(). The preceding read is | |
286 | * for @element and the following @pool->curr_nr. This ensures | |
287 | * that the visible value of @pool->curr_nr is from after the | |
288 | * allocation of @element. This is necessary for fringe cases | |
289 | * where @element was passed to this task without going through | |
290 | * barriers. | |
291 | * | |
292 | * For example, assume @p is %NULL at the beginning and one task | |
293 | * performs "p = mempool_alloc(...);" while another task is doing | |
294 | * "while (!p) cpu_relax(); mempool_free(p, ...);". This function | |
295 | * may end up using curr_nr value which is from before allocation | |
296 | * of @p without the following rmb. | |
297 | */ | |
298 | smp_rmb(); | |
299 | ||
300 | /* | |
301 | * For correctness, we need a test which is guaranteed to trigger | |
302 | * if curr_nr + #allocated == min_nr. Testing curr_nr < min_nr | |
303 | * without locking achieves that and refilling as soon as possible | |
304 | * is desirable. | |
305 | * | |
306 | * Because curr_nr visible here is always a value after the | |
307 | * allocation of @element, any task which decremented curr_nr below | |
308 | * min_nr is guaranteed to see curr_nr < min_nr unless curr_nr gets | |
309 | * incremented to min_nr afterwards. If curr_nr gets incremented | |
310 | * to min_nr after the allocation of @element, the elements | |
311 | * allocated after that are subject to the same guarantee. | |
312 | * | |
313 | * Waiters happen iff curr_nr is 0 and the above guarantee also | |
314 | * ensures that there will be frees which return elements to the | |
315 | * pool waking up the waiters. | |
316 | */ | |
eb9a3c62 | 317 | if (unlikely(pool->curr_nr < pool->min_nr)) { |
1da177e4 | 318 | spin_lock_irqsave(&pool->lock, flags); |
eb9a3c62 | 319 | if (likely(pool->curr_nr < pool->min_nr)) { |
1da177e4 LT |
320 | add_element(pool, element); |
321 | spin_unlock_irqrestore(&pool->lock, flags); | |
322 | wake_up(&pool->wait); | |
323 | return; | |
324 | } | |
325 | spin_unlock_irqrestore(&pool->lock, flags); | |
326 | } | |
327 | pool->free(element, pool->pool_data); | |
328 | } | |
329 | EXPORT_SYMBOL(mempool_free); | |
330 | ||
331 | /* | |
332 | * A commonly used alloc and free fn. | |
333 | */ | |
dd0fc66f | 334 | void *mempool_alloc_slab(gfp_t gfp_mask, void *pool_data) |
1da177e4 | 335 | { |
fcc234f8 | 336 | struct kmem_cache *mem = pool_data; |
1da177e4 LT |
337 | return kmem_cache_alloc(mem, gfp_mask); |
338 | } | |
339 | EXPORT_SYMBOL(mempool_alloc_slab); | |
340 | ||
341 | void mempool_free_slab(void *element, void *pool_data) | |
342 | { | |
fcc234f8 | 343 | struct kmem_cache *mem = pool_data; |
1da177e4 LT |
344 | kmem_cache_free(mem, element); |
345 | } | |
346 | EXPORT_SYMBOL(mempool_free_slab); | |
6e0678f3 | 347 | |
53184082 MD |
348 | /* |
349 | * A commonly used alloc and free fn that kmalloc/kfrees the amount of memory | |
183ff22b | 350 | * specified by pool_data |
53184082 MD |
351 | */ |
352 | void *mempool_kmalloc(gfp_t gfp_mask, void *pool_data) | |
353 | { | |
5e2f89b5 | 354 | size_t size = (size_t)pool_data; |
53184082 MD |
355 | return kmalloc(size, gfp_mask); |
356 | } | |
357 | EXPORT_SYMBOL(mempool_kmalloc); | |
358 | ||
359 | void mempool_kfree(void *element, void *pool_data) | |
360 | { | |
361 | kfree(element); | |
362 | } | |
363 | EXPORT_SYMBOL(mempool_kfree); | |
364 | ||
6e0678f3 MD |
365 | /* |
366 | * A simple mempool-backed page allocator that allocates pages | |
367 | * of the order specified by pool_data. | |
368 | */ | |
369 | void *mempool_alloc_pages(gfp_t gfp_mask, void *pool_data) | |
370 | { | |
371 | int order = (int)(long)pool_data; | |
372 | return alloc_pages(gfp_mask, order); | |
373 | } | |
374 | EXPORT_SYMBOL(mempool_alloc_pages); | |
375 | ||
376 | void mempool_free_pages(void *element, void *pool_data) | |
377 | { | |
378 | int order = (int)(long)pool_data; | |
379 | __free_pages(element, order); | |
380 | } | |
381 | EXPORT_SYMBOL(mempool_free_pages); |