Root/
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> |
13 | #include <linux/export.h> |
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 | |
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) |
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 | } |
47 | EXPORT_SYMBOL(mempool_destroy); |
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 |
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 |
61 | * from IRQ contexts. |
62 | */ |
63 | mempool_t *mempool_create(int min_nr, mempool_alloc_t *alloc_fn, |
64 | mempool_free_t *free_fn, void *pool_data) |
65 | { |
66 | return mempool_create_node(min_nr,alloc_fn,free_fn, pool_data, |
67 | GFP_KERNEL, NUMA_NO_NODE); |
68 | } |
69 | EXPORT_SYMBOL(mempool_create); |
70 | |
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) |
74 | { |
75 | mempool_t *pool; |
76 | pool = kzalloc_node(sizeof(*pool), gfp_mask, node_id); |
77 | if (!pool) |
78 | return NULL; |
79 | pool->elements = kmalloc_node(min_nr * sizeof(void *), |
80 | gfp_mask, node_id); |
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 | |
98 | element = pool->alloc(gfp_mask, pool->pool_data); |
99 | if (unlikely(!element)) { |
100 | mempool_destroy(pool); |
101 | return NULL; |
102 | } |
103 | add_element(pool, element); |
104 | } |
105 | return pool; |
106 | } |
107 | EXPORT_SYMBOL(mempool_create_node); |
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 | */ |
125 | int mempool_resize(mempool_t *pool, int new_min_nr, gfp_t gfp_mask) |
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 | |
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 | * |
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.) |
195 | */ |
196 | void * mempool_alloc(mempool_t *pool, gfp_t gfp_mask) |
197 | { |
198 | void *element; |
199 | unsigned long flags; |
200 | wait_queue_t wait; |
201 | gfp_t gfp_temp; |
202 | |
203 | might_sleep_if(gfp_mask & __GFP_WAIT); |
204 | |
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 */ |
208 | |
209 | gfp_temp = gfp_mask & ~(__GFP_WAIT|__GFP_IO); |
210 | |
211 | repeat_alloc: |
212 | |
213 | element = pool->alloc(gfp_temp, pool->pool_data); |
214 | if (likely(element != NULL)) |
215 | return element; |
216 | |
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 */ |
222 | smp_wmb(); |
223 | return element; |
224 | } |
225 | |
226 | /* |
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. |
229 | */ |
230 | if (gfp_temp != gfp_mask) { |
231 | spin_unlock_irqrestore(&pool->lock, flags); |
232 | gfp_temp = gfp_mask; |
233 | goto repeat_alloc; |
234 | } |
235 | |
236 | /* We must not sleep if !__GFP_WAIT */ |
237 | if (!(gfp_mask & __GFP_WAIT)) { |
238 | spin_unlock_irqrestore(&pool->lock, flags); |
239 | return NULL; |
240 | } |
241 | |
242 | /* Let's wait for someone else to return an element to @pool */ |
243 | init_wait(&wait); |
244 | prepare_to_wait(&pool->wait, &wait, TASK_UNINTERRUPTIBLE); |
245 | |
246 | spin_unlock_irqrestore(&pool->lock, flags); |
247 | |
248 | /* |
249 | * FIXME: this should be io_schedule(). The timeout is there as a |
250 | * workaround for some DM problems in 2.6.18. |
251 | */ |
252 | io_schedule_timeout(5*HZ); |
253 | |
254 | finish_wait(&pool->wait, &wait); |
255 | goto repeat_alloc; |
256 | } |
257 | EXPORT_SYMBOL(mempool_alloc); |
258 | |
259 | /** |
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 |
263 | * mempool_create(). |
264 | * |
265 | * this function only sleeps if the free_fn() function sleeps. |
266 | */ |
267 | void mempool_free(void *element, mempool_t *pool) |
268 | { |
269 | unsigned long flags; |
270 | |
271 | if (unlikely(element == NULL)) |
272 | return; |
273 | |
274 | /* |
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 |
280 | * barriers. |
281 | * |
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. |
287 | */ |
288 | smp_rmb(); |
289 | |
290 | /* |
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 |
294 | * is desirable. |
295 | * |
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. |
302 | * |
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. |
306 | */ |
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); |
313 | return; |
314 | } |
315 | spin_unlock_irqrestore(&pool->lock, flags); |
316 | } |
317 | pool->free(element, pool->pool_data); |
318 | } |
319 | EXPORT_SYMBOL(mempool_free); |
320 | |
321 | /* |
322 | * A commonly used alloc and free fn. |
323 | */ |
324 | void *mempool_alloc_slab(gfp_t gfp_mask, void *pool_data) |
325 | { |
326 | struct kmem_cache *mem = pool_data; |
327 | return kmem_cache_alloc(mem, gfp_mask); |
328 | } |
329 | EXPORT_SYMBOL(mempool_alloc_slab); |
330 | |
331 | void mempool_free_slab(void *element, void *pool_data) |
332 | { |
333 | struct kmem_cache *mem = pool_data; |
334 | kmem_cache_free(mem, element); |
335 | } |
336 | EXPORT_SYMBOL(mempool_free_slab); |
337 | |
338 | /* |
339 | * A commonly used alloc and free fn that kmalloc/kfrees the amount of memory |
340 | * specified by pool_data |
341 | */ |
342 | void *mempool_kmalloc(gfp_t gfp_mask, void *pool_data) |
343 | { |
344 | size_t size = (size_t)pool_data; |
345 | return kmalloc(size, gfp_mask); |
346 | } |
347 | EXPORT_SYMBOL(mempool_kmalloc); |
348 | |
349 | void mempool_kfree(void *element, void *pool_data) |
350 | { |
351 | kfree(element); |
352 | } |
353 | EXPORT_SYMBOL(mempool_kfree); |
354 | |
355 | /* |
356 | * A simple mempool-backed page allocator that allocates pages |
357 | * of the order specified by pool_data. |
358 | */ |
359 | void *mempool_alloc_pages(gfp_t gfp_mask, void *pool_data) |
360 | { |
361 | int order = (int)(long)pool_data; |
362 | return alloc_pages(gfp_mask, order); |
363 | } |
364 | EXPORT_SYMBOL(mempool_alloc_pages); |
365 | |
366 | void mempool_free_pages(void *element, void *pool_data) |
367 | { |
368 | int order = (int)(long)pool_data; |
369 | __free_pages(element, order); |
370 | } |
371 | EXPORT_SYMBOL(mempool_free_pages); |
372 |
Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master
Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9