Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 LT |
2 | /* |
3 | * linux/mm/mempool.c | |
4 | * | |
5 | * memory buffer pool support. Such pools are mostly used | |
6 | * for guaranteed, deadlock-free memory allocations during | |
7 | * extreme VM load. | |
8 | * | |
9 | * started by Ingo Molnar, Copyright (C) 2001 | |
bdfedb76 | 10 | * debugging by David Rientjes, Copyright (C) 2015 |
1da177e4 LT |
11 | */ |
12 | ||
13 | #include <linux/mm.h> | |
14 | #include <linux/slab.h> | |
bdfedb76 | 15 | #include <linux/highmem.h> |
92393615 | 16 | #include <linux/kasan.h> |
17411962 | 17 | #include <linux/kmemleak.h> |
b95f1b31 | 18 | #include <linux/export.h> |
1da177e4 | 19 | #include <linux/mempool.h> |
1da177e4 | 20 | #include <linux/writeback.h> |
e244c9e6 | 21 | #include "slab.h" |
1da177e4 | 22 | |
8c20b29d | 23 | #ifdef CONFIG_SLUB_DEBUG_ON |
bdfedb76 DR |
24 | static void poison_error(mempool_t *pool, void *element, size_t size, |
25 | size_t byte) | |
26 | { | |
27 | const int nr = pool->curr_nr; | |
28 | const int start = max_t(int, byte - (BITS_PER_LONG / 8), 0); | |
29 | const int end = min_t(int, byte + (BITS_PER_LONG / 8), size); | |
30 | int i; | |
31 | ||
32 | pr_err("BUG: mempool element poison mismatch\n"); | |
33 | pr_err("Mempool %p size %zu\n", pool, size); | |
34 | pr_err(" nr=%d @ %p: %s0x", nr, element, start > 0 ? "... " : ""); | |
35 | for (i = start; i < end; i++) | |
36 | pr_cont("%x ", *(u8 *)(element + i)); | |
37 | pr_cont("%s\n", end < size ? "..." : ""); | |
38 | dump_stack(); | |
39 | } | |
40 | ||
41 | static void __check_element(mempool_t *pool, void *element, size_t size) | |
42 | { | |
43 | u8 *obj = element; | |
44 | size_t i; | |
45 | ||
46 | for (i = 0; i < size; i++) { | |
47 | u8 exp = (i < size - 1) ? POISON_FREE : POISON_END; | |
48 | ||
49 | if (obj[i] != exp) { | |
50 | poison_error(pool, element, size, i); | |
51 | return; | |
52 | } | |
53 | } | |
54 | memset(obj, POISON_INUSE, size); | |
55 | } | |
56 | ||
57 | static void check_element(mempool_t *pool, void *element) | |
58 | { | |
7d4847de AK |
59 | /* Skip checking: KASAN might save its metadata in the element. */ |
60 | if (kasan_enabled()) | |
61 | return; | |
62 | ||
bdfedb76 | 63 | /* Mempools backed by slab allocator */ |
b2b23ba0 KC |
64 | if (pool->free == mempool_kfree) { |
65 | __check_element(pool, element, (size_t)pool->pool_data); | |
66 | } else if (pool->free == mempool_free_slab) { | |
67 | __check_element(pool, element, kmem_cache_size(pool->pool_data)); | |
544941d7 ML |
68 | } else if (pool->free == mempool_free_pages) { |
69 | /* Mempools backed by page allocator */ | |
bdfedb76 | 70 | int order = (int)(long)pool->pool_data; |
f2bcc99a | 71 | void *addr = kmap_local_page((struct page *)element); |
bdfedb76 DR |
72 | |
73 | __check_element(pool, addr, 1UL << (PAGE_SHIFT + order)); | |
f2bcc99a | 74 | kunmap_local(addr); |
bdfedb76 DR |
75 | } |
76 | } | |
77 | ||
78 | static void __poison_element(void *element, size_t size) | |
79 | { | |
80 | u8 *obj = element; | |
81 | ||
82 | memset(obj, POISON_FREE, size - 1); | |
83 | obj[size - 1] = POISON_END; | |
84 | } | |
85 | ||
86 | static void poison_element(mempool_t *pool, void *element) | |
87 | { | |
7d4847de AK |
88 | /* Skip poisoning: KASAN might save its metadata in the element. */ |
89 | if (kasan_enabled()) | |
90 | return; | |
91 | ||
bdfedb76 | 92 | /* Mempools backed by slab allocator */ |
b2b23ba0 KC |
93 | if (pool->alloc == mempool_kmalloc) { |
94 | __poison_element(element, (size_t)pool->pool_data); | |
95 | } else if (pool->alloc == mempool_alloc_slab) { | |
96 | __poison_element(element, kmem_cache_size(pool->pool_data)); | |
544941d7 ML |
97 | } else if (pool->alloc == mempool_alloc_pages) { |
98 | /* Mempools backed by page allocator */ | |
bdfedb76 | 99 | int order = (int)(long)pool->pool_data; |
f2bcc99a | 100 | void *addr = kmap_local_page((struct page *)element); |
bdfedb76 DR |
101 | |
102 | __poison_element(addr, 1UL << (PAGE_SHIFT + order)); | |
f2bcc99a | 103 | kunmap_local(addr); |
bdfedb76 DR |
104 | } |
105 | } | |
8c20b29d | 106 | #else /* CONFIG_SLUB_DEBUG_ON */ |
bdfedb76 DR |
107 | static inline void check_element(mempool_t *pool, void *element) |
108 | { | |
109 | } | |
110 | static inline void poison_element(mempool_t *pool, void *element) | |
111 | { | |
112 | } | |
8c20b29d | 113 | #endif /* CONFIG_SLUB_DEBUG_ON */ |
bdfedb76 | 114 | |
413643f3 | 115 | static __always_inline bool kasan_poison_element(mempool_t *pool, void *element) |
92393615 | 116 | { |
9b75a867 | 117 | if (pool->alloc == mempool_alloc_slab || pool->alloc == mempool_kmalloc) |
413643f3 | 118 | return kasan_mempool_poison_object(element); |
544941d7 | 119 | else if (pool->alloc == mempool_alloc_pages) |
413643f3 AK |
120 | return kasan_mempool_poison_pages(element, |
121 | (unsigned long)pool->pool_data); | |
122 | return true; | |
92393615 AR |
123 | } |
124 | ||
8cded866 | 125 | static void kasan_unpoison_element(mempool_t *pool, void *element) |
92393615 | 126 | { |
b2b23ba0 | 127 | if (pool->alloc == mempool_kmalloc) |
413643f3 | 128 | kasan_mempool_unpoison_object(element, (size_t)pool->pool_data); |
b2b23ba0 | 129 | else if (pool->alloc == mempool_alloc_slab) |
413643f3 AK |
130 | kasan_mempool_unpoison_object(element, |
131 | kmem_cache_size(pool->pool_data)); | |
544941d7 | 132 | else if (pool->alloc == mempool_alloc_pages) |
413643f3 AK |
133 | kasan_mempool_unpoison_pages(element, |
134 | (unsigned long)pool->pool_data); | |
92393615 AR |
135 | } |
136 | ||
6860f634 | 137 | static __always_inline void add_element(mempool_t *pool, void *element) |
1da177e4 LT |
138 | { |
139 | BUG_ON(pool->curr_nr >= pool->min_nr); | |
bdfedb76 | 140 | poison_element(pool, element); |
413643f3 AK |
141 | if (kasan_poison_element(pool, element)) |
142 | pool->elements[pool->curr_nr++] = element; | |
1da177e4 LT |
143 | } |
144 | ||
8cded866 | 145 | static void *remove_element(mempool_t *pool) |
1da177e4 | 146 | { |
bdfedb76 DR |
147 | void *element = pool->elements[--pool->curr_nr]; |
148 | ||
149 | BUG_ON(pool->curr_nr < 0); | |
8cded866 | 150 | kasan_unpoison_element(pool, element); |
76401310 | 151 | check_element(pool, element); |
bdfedb76 | 152 | return element; |
1da177e4 LT |
153 | } |
154 | ||
c1a67fef KO |
155 | /** |
156 | * mempool_exit - exit a mempool initialized with mempool_init() | |
157 | * @pool: pointer to the memory pool which was initialized with | |
158 | * mempool_init(). | |
159 | * | |
160 | * Free all reserved elements in @pool and @pool itself. This function | |
161 | * only sleeps if the free_fn() function sleeps. | |
162 | * | |
163 | * May be called on a zeroed but uninitialized mempool (i.e. allocated with | |
164 | * kzalloc()). | |
165 | */ | |
166 | void mempool_exit(mempool_t *pool) | |
167 | { | |
168 | while (pool->curr_nr) { | |
8cded866 | 169 | void *element = remove_element(pool); |
c1a67fef KO |
170 | pool->free(element, pool->pool_data); |
171 | } | |
172 | kfree(pool->elements); | |
173 | pool->elements = NULL; | |
174 | } | |
175 | EXPORT_SYMBOL(mempool_exit); | |
176 | ||
0565d317 TH |
177 | /** |
178 | * mempool_destroy - deallocate a memory pool | |
179 | * @pool: pointer to the memory pool which was allocated via | |
180 | * mempool_create(). | |
181 | * | |
182 | * Free all reserved elements in @pool and @pool itself. This function | |
183 | * only sleeps if the free_fn() function sleeps. | |
184 | */ | |
185 | void mempool_destroy(mempool_t *pool) | |
1da177e4 | 186 | { |
4e3ca3e0 SS |
187 | if (unlikely(!pool)) |
188 | return; | |
189 | ||
c1a67fef | 190 | mempool_exit(pool); |
1da177e4 LT |
191 | kfree(pool); |
192 | } | |
0565d317 | 193 | EXPORT_SYMBOL(mempool_destroy); |
1da177e4 | 194 | |
c1a67fef KO |
195 | int mempool_init_node(mempool_t *pool, int min_nr, mempool_alloc_t *alloc_fn, |
196 | mempool_free_t *free_fn, void *pool_data, | |
197 | gfp_t gfp_mask, int node_id) | |
198 | { | |
199 | spin_lock_init(&pool->lock); | |
200 | pool->min_nr = min_nr; | |
201 | pool->pool_data = pool_data; | |
202 | pool->alloc = alloc_fn; | |
203 | pool->free = free_fn; | |
204 | init_waitqueue_head(&pool->wait); | |
205 | ||
206 | pool->elements = kmalloc_array_node(min_nr, sizeof(void *), | |
207 | gfp_mask, node_id); | |
208 | if (!pool->elements) | |
209 | return -ENOMEM; | |
210 | ||
211 | /* | |
212 | * First pre-allocate the guaranteed number of buffers. | |
213 | */ | |
214 | while (pool->curr_nr < pool->min_nr) { | |
215 | void *element; | |
216 | ||
217 | element = pool->alloc(gfp_mask, pool->pool_data); | |
218 | if (unlikely(!element)) { | |
219 | mempool_exit(pool); | |
220 | return -ENOMEM; | |
221 | } | |
222 | add_element(pool, element); | |
223 | } | |
224 | ||
225 | return 0; | |
226 | } | |
227 | EXPORT_SYMBOL(mempool_init_node); | |
228 | ||
229 | /** | |
230 | * mempool_init - initialize a memory pool | |
a3bf6ce3 | 231 | * @pool: pointer to the memory pool that should be initialized |
c1a67fef KO |
232 | * @min_nr: the minimum number of elements guaranteed to be |
233 | * allocated for this pool. | |
234 | * @alloc_fn: user-defined element-allocation function. | |
235 | * @free_fn: user-defined element-freeing function. | |
236 | * @pool_data: optional private data available to the user-defined functions. | |
237 | * | |
238 | * Like mempool_create(), but initializes the pool in (i.e. embedded in another | |
239 | * structure). | |
a862f68a MR |
240 | * |
241 | * Return: %0 on success, negative error code otherwise. | |
c1a67fef KO |
242 | */ |
243 | int mempool_init(mempool_t *pool, int min_nr, mempool_alloc_t *alloc_fn, | |
244 | mempool_free_t *free_fn, void *pool_data) | |
245 | { | |
246 | return mempool_init_node(pool, min_nr, alloc_fn, free_fn, | |
247 | pool_data, GFP_KERNEL, NUMA_NO_NODE); | |
248 | ||
249 | } | |
250 | EXPORT_SYMBOL(mempool_init); | |
251 | ||
1da177e4 LT |
252 | /** |
253 | * mempool_create - create a memory pool | |
254 | * @min_nr: the minimum number of elements guaranteed to be | |
255 | * allocated for this pool. | |
256 | * @alloc_fn: user-defined element-allocation function. | |
257 | * @free_fn: user-defined element-freeing function. | |
258 | * @pool_data: optional private data available to the user-defined functions. | |
259 | * | |
260 | * this function creates and allocates a guaranteed size, preallocated | |
72fd4a35 | 261 | * memory pool. The pool can be used from the mempool_alloc() and mempool_free() |
1da177e4 | 262 | * functions. This function might sleep. Both the alloc_fn() and the free_fn() |
72fd4a35 | 263 | * functions might sleep - as long as the mempool_alloc() function is not called |
1da177e4 | 264 | * from IRQ contexts. |
a862f68a MR |
265 | * |
266 | * Return: pointer to the created memory pool object or %NULL on error. | |
1da177e4 | 267 | */ |
1946089a | 268 | mempool_t *mempool_create(int min_nr, mempool_alloc_t *alloc_fn, |
1da177e4 LT |
269 | mempool_free_t *free_fn, void *pool_data) |
270 | { | |
68d68ff6 | 271 | return mempool_create_node(min_nr, alloc_fn, free_fn, pool_data, |
a91a5ac6 | 272 | GFP_KERNEL, NUMA_NO_NODE); |
1946089a CL |
273 | } |
274 | EXPORT_SYMBOL(mempool_create); | |
1da177e4 | 275 | |
1946089a | 276 | mempool_t *mempool_create_node(int min_nr, mempool_alloc_t *alloc_fn, |
a91a5ac6 TH |
277 | mempool_free_t *free_fn, void *pool_data, |
278 | gfp_t gfp_mask, int node_id) | |
1946089a CL |
279 | { |
280 | mempool_t *pool; | |
c1a67fef | 281 | |
7b5219db | 282 | pool = kzalloc_node(sizeof(*pool), gfp_mask, node_id); |
1da177e4 LT |
283 | if (!pool) |
284 | return NULL; | |
c1a67fef KO |
285 | |
286 | if (mempool_init_node(pool, min_nr, alloc_fn, free_fn, pool_data, | |
287 | gfp_mask, node_id)) { | |
1da177e4 LT |
288 | kfree(pool); |
289 | return NULL; | |
290 | } | |
1da177e4 | 291 | |
1da177e4 LT |
292 | return pool; |
293 | } | |
1946089a | 294 | EXPORT_SYMBOL(mempool_create_node); |
1da177e4 LT |
295 | |
296 | /** | |
297 | * mempool_resize - resize an existing memory pool | |
298 | * @pool: pointer to the memory pool which was allocated via | |
299 | * mempool_create(). | |
300 | * @new_min_nr: the new minimum number of elements guaranteed to be | |
301 | * allocated for this pool. | |
1da177e4 LT |
302 | * |
303 | * This function shrinks/grows the pool. In the case of growing, | |
304 | * it cannot be guaranteed that the pool will be grown to the new | |
305 | * size immediately, but new mempool_free() calls will refill it. | |
11d83360 | 306 | * This function may sleep. |
1da177e4 LT |
307 | * |
308 | * Note, the caller must guarantee that no mempool_destroy is called | |
309 | * while this function is running. mempool_alloc() & mempool_free() | |
310 | * might be called (eg. from IRQ contexts) while this function executes. | |
a862f68a MR |
311 | * |
312 | * Return: %0 on success, negative error code otherwise. | |
1da177e4 | 313 | */ |
11d83360 | 314 | int mempool_resize(mempool_t *pool, int new_min_nr) |
1da177e4 LT |
315 | { |
316 | void *element; | |
317 | void **new_elements; | |
318 | unsigned long flags; | |
319 | ||
320 | BUG_ON(new_min_nr <= 0); | |
11d83360 | 321 | might_sleep(); |
1da177e4 LT |
322 | |
323 | spin_lock_irqsave(&pool->lock, flags); | |
324 | if (new_min_nr <= pool->min_nr) { | |
325 | while (new_min_nr < pool->curr_nr) { | |
8cded866 | 326 | element = remove_element(pool); |
1da177e4 LT |
327 | spin_unlock_irqrestore(&pool->lock, flags); |
328 | pool->free(element, pool->pool_data); | |
329 | spin_lock_irqsave(&pool->lock, flags); | |
330 | } | |
331 | pool->min_nr = new_min_nr; | |
332 | goto out_unlock; | |
333 | } | |
334 | spin_unlock_irqrestore(&pool->lock, flags); | |
335 | ||
336 | /* Grow the pool */ | |
11d83360 DR |
337 | new_elements = kmalloc_array(new_min_nr, sizeof(*new_elements), |
338 | GFP_KERNEL); | |
1da177e4 LT |
339 | if (!new_elements) |
340 | return -ENOMEM; | |
341 | ||
342 | spin_lock_irqsave(&pool->lock, flags); | |
343 | if (unlikely(new_min_nr <= pool->min_nr)) { | |
344 | /* Raced, other resize will do our work */ | |
345 | spin_unlock_irqrestore(&pool->lock, flags); | |
346 | kfree(new_elements); | |
347 | goto out; | |
348 | } | |
349 | memcpy(new_elements, pool->elements, | |
350 | pool->curr_nr * sizeof(*new_elements)); | |
351 | kfree(pool->elements); | |
352 | pool->elements = new_elements; | |
353 | pool->min_nr = new_min_nr; | |
354 | ||
355 | while (pool->curr_nr < pool->min_nr) { | |
356 | spin_unlock_irqrestore(&pool->lock, flags); | |
11d83360 | 357 | element = pool->alloc(GFP_KERNEL, pool->pool_data); |
1da177e4 LT |
358 | if (!element) |
359 | goto out; | |
360 | spin_lock_irqsave(&pool->lock, flags); | |
361 | if (pool->curr_nr < pool->min_nr) { | |
362 | add_element(pool, element); | |
363 | } else { | |
364 | spin_unlock_irqrestore(&pool->lock, flags); | |
365 | pool->free(element, pool->pool_data); /* Raced */ | |
366 | goto out; | |
367 | } | |
368 | } | |
369 | out_unlock: | |
370 | spin_unlock_irqrestore(&pool->lock, flags); | |
371 | out: | |
372 | return 0; | |
373 | } | |
374 | EXPORT_SYMBOL(mempool_resize); | |
375 | ||
1da177e4 LT |
376 | /** |
377 | * mempool_alloc - allocate an element from a specific memory pool | |
378 | * @pool: pointer to the memory pool which was allocated via | |
379 | * mempool_create(). | |
380 | * @gfp_mask: the usual allocation bitmask. | |
381 | * | |
72fd4a35 | 382 | * this function only sleeps if the alloc_fn() function sleeps or |
1da177e4 LT |
383 | * returns NULL. Note that due to preallocation, this function |
384 | * *never* fails when called from process contexts. (it might | |
385 | * fail if called from an IRQ context.) | |
4e390b2b | 386 | * Note: using __GFP_ZERO is not supported. |
a862f68a MR |
387 | * |
388 | * Return: pointer to the allocated element or %NULL on error. | |
1da177e4 | 389 | */ |
f9054c70 | 390 | void *mempool_alloc(mempool_t *pool, gfp_t gfp_mask) |
1da177e4 LT |
391 | { |
392 | void *element; | |
393 | unsigned long flags; | |
ac6424b9 | 394 | wait_queue_entry_t wait; |
6daa0e28 | 395 | gfp_t gfp_temp; |
20a77776 | 396 | |
8bf8fcb0 | 397 | VM_WARN_ON_ONCE(gfp_mask & __GFP_ZERO); |
21bfe8db | 398 | might_alloc(gfp_mask); |
b84a35be | 399 | |
4e390b2b | 400 | gfp_mask |= __GFP_NOMEMALLOC; /* don't allocate emergency reserves */ |
b84a35be NP |
401 | gfp_mask |= __GFP_NORETRY; /* don't loop in __alloc_pages */ |
402 | gfp_mask |= __GFP_NOWARN; /* failures are OK */ | |
1da177e4 | 403 | |
d0164adc | 404 | gfp_temp = gfp_mask & ~(__GFP_DIRECT_RECLAIM|__GFP_IO); |
20a77776 | 405 | |
1da177e4 | 406 | repeat_alloc: |
20a77776 NP |
407 | |
408 | element = pool->alloc(gfp_temp, pool->pool_data); | |
1da177e4 LT |
409 | if (likely(element != NULL)) |
410 | return element; | |
411 | ||
1da177e4 LT |
412 | spin_lock_irqsave(&pool->lock, flags); |
413 | if (likely(pool->curr_nr)) { | |
8cded866 | 414 | element = remove_element(pool); |
1da177e4 | 415 | spin_unlock_irqrestore(&pool->lock, flags); |
5b990546 TH |
416 | /* paired with rmb in mempool_free(), read comment there */ |
417 | smp_wmb(); | |
17411962 CM |
418 | /* |
419 | * Update the allocation stack trace as this is more useful | |
420 | * for debugging. | |
421 | */ | |
422 | kmemleak_update_trace(element); | |
1da177e4 LT |
423 | return element; |
424 | } | |
1da177e4 | 425 | |
1ebb7044 | 426 | /* |
d0164adc | 427 | * We use gfp mask w/o direct reclaim or IO for the first round. If |
1ebb7044 TH |
428 | * alloc failed with that and @pool was empty, retry immediately. |
429 | */ | |
4e390b2b | 430 | if (gfp_temp != gfp_mask) { |
1ebb7044 TH |
431 | spin_unlock_irqrestore(&pool->lock, flags); |
432 | gfp_temp = gfp_mask; | |
433 | goto repeat_alloc; | |
434 | } | |
435 | ||
d0164adc MG |
436 | /* We must not sleep if !__GFP_DIRECT_RECLAIM */ |
437 | if (!(gfp_mask & __GFP_DIRECT_RECLAIM)) { | |
5b990546 | 438 | spin_unlock_irqrestore(&pool->lock, flags); |
1da177e4 | 439 | return NULL; |
5b990546 | 440 | } |
1da177e4 | 441 | |
5b990546 | 442 | /* Let's wait for someone else to return an element to @pool */ |
01890a4c | 443 | init_wait(&wait); |
1da177e4 | 444 | prepare_to_wait(&pool->wait, &wait, TASK_UNINTERRUPTIBLE); |
1da177e4 | 445 | |
5b990546 TH |
446 | spin_unlock_irqrestore(&pool->lock, flags); |
447 | ||
448 | /* | |
449 | * FIXME: this should be io_schedule(). The timeout is there as a | |
450 | * workaround for some DM problems in 2.6.18. | |
451 | */ | |
452 | io_schedule_timeout(5*HZ); | |
453 | ||
454 | finish_wait(&pool->wait, &wait); | |
1da177e4 LT |
455 | goto repeat_alloc; |
456 | } | |
457 | EXPORT_SYMBOL(mempool_alloc); | |
458 | ||
37dcc69a AK |
459 | /** |
460 | * mempool_alloc_preallocated - allocate an element from preallocated elements | |
461 | * belonging to a specific memory pool | |
462 | * @pool: pointer to the memory pool which was allocated via | |
463 | * mempool_create(). | |
464 | * | |
465 | * This function is similar to mempool_alloc, but it only attempts allocating | |
466 | * an element from the preallocated elements. It does not sleep and immediately | |
467 | * returns if no preallocated elements are available. | |
468 | * | |
469 | * Return: pointer to the allocated element or %NULL if no elements are | |
470 | * available. | |
471 | */ | |
472 | void *mempool_alloc_preallocated(mempool_t *pool) | |
473 | { | |
474 | void *element; | |
475 | unsigned long flags; | |
476 | ||
477 | spin_lock_irqsave(&pool->lock, flags); | |
478 | if (likely(pool->curr_nr)) { | |
479 | element = remove_element(pool); | |
480 | spin_unlock_irqrestore(&pool->lock, flags); | |
481 | /* paired with rmb in mempool_free(), read comment there */ | |
482 | smp_wmb(); | |
483 | /* | |
484 | * Update the allocation stack trace as this is more useful | |
485 | * for debugging. | |
486 | */ | |
487 | kmemleak_update_trace(element); | |
488 | return element; | |
489 | } | |
490 | spin_unlock_irqrestore(&pool->lock, flags); | |
491 | ||
492 | return NULL; | |
493 | } | |
494 | EXPORT_SYMBOL(mempool_alloc_preallocated); | |
495 | ||
1da177e4 LT |
496 | /** |
497 | * mempool_free - return an element to the pool. | |
498 | * @element: pool element pointer. | |
499 | * @pool: pointer to the memory pool which was allocated via | |
500 | * mempool_create(). | |
501 | * | |
502 | * this function only sleeps if the free_fn() function sleeps. | |
503 | */ | |
504 | void mempool_free(void *element, mempool_t *pool) | |
505 | { | |
506 | unsigned long flags; | |
507 | ||
c80e7a82 RR |
508 | if (unlikely(element == NULL)) |
509 | return; | |
510 | ||
5b990546 TH |
511 | /* |
512 | * Paired with the wmb in mempool_alloc(). The preceding read is | |
513 | * for @element and the following @pool->curr_nr. This ensures | |
514 | * that the visible value of @pool->curr_nr is from after the | |
515 | * allocation of @element. This is necessary for fringe cases | |
516 | * where @element was passed to this task without going through | |
517 | * barriers. | |
518 | * | |
519 | * For example, assume @p is %NULL at the beginning and one task | |
520 | * performs "p = mempool_alloc(...);" while another task is doing | |
521 | * "while (!p) cpu_relax(); mempool_free(p, ...);". This function | |
522 | * may end up using curr_nr value which is from before allocation | |
523 | * of @p without the following rmb. | |
524 | */ | |
525 | smp_rmb(); | |
526 | ||
527 | /* | |
528 | * For correctness, we need a test which is guaranteed to trigger | |
529 | * if curr_nr + #allocated == min_nr. Testing curr_nr < min_nr | |
530 | * without locking achieves that and refilling as soon as possible | |
531 | * is desirable. | |
532 | * | |
533 | * Because curr_nr visible here is always a value after the | |
534 | * allocation of @element, any task which decremented curr_nr below | |
535 | * min_nr is guaranteed to see curr_nr < min_nr unless curr_nr gets | |
536 | * incremented to min_nr afterwards. If curr_nr gets incremented | |
537 | * to min_nr after the allocation of @element, the elements | |
538 | * allocated after that are subject to the same guarantee. | |
539 | * | |
540 | * Waiters happen iff curr_nr is 0 and the above guarantee also | |
541 | * ensures that there will be frees which return elements to the | |
542 | * pool waking up the waiters. | |
543 | */ | |
abe1de42 | 544 | if (unlikely(READ_ONCE(pool->curr_nr) < pool->min_nr)) { |
1da177e4 | 545 | spin_lock_irqsave(&pool->lock, flags); |
eb9a3c62 | 546 | if (likely(pool->curr_nr < pool->min_nr)) { |
1da177e4 LT |
547 | add_element(pool, element); |
548 | spin_unlock_irqrestore(&pool->lock, flags); | |
549 | wake_up(&pool->wait); | |
550 | return; | |
551 | } | |
552 | spin_unlock_irqrestore(&pool->lock, flags); | |
553 | } | |
554 | pool->free(element, pool->pool_data); | |
555 | } | |
556 | EXPORT_SYMBOL(mempool_free); | |
557 | ||
558 | /* | |
559 | * A commonly used alloc and free fn. | |
560 | */ | |
dd0fc66f | 561 | void *mempool_alloc_slab(gfp_t gfp_mask, void *pool_data) |
1da177e4 | 562 | { |
fcc234f8 | 563 | struct kmem_cache *mem = pool_data; |
e244c9e6 | 564 | VM_BUG_ON(mem->ctor); |
1da177e4 LT |
565 | return kmem_cache_alloc(mem, gfp_mask); |
566 | } | |
567 | EXPORT_SYMBOL(mempool_alloc_slab); | |
568 | ||
569 | void mempool_free_slab(void *element, void *pool_data) | |
570 | { | |
fcc234f8 | 571 | struct kmem_cache *mem = pool_data; |
1da177e4 LT |
572 | kmem_cache_free(mem, element); |
573 | } | |
574 | EXPORT_SYMBOL(mempool_free_slab); | |
6e0678f3 | 575 | |
53184082 MD |
576 | /* |
577 | * A commonly used alloc and free fn that kmalloc/kfrees the amount of memory | |
183ff22b | 578 | * specified by pool_data |
53184082 MD |
579 | */ |
580 | void *mempool_kmalloc(gfp_t gfp_mask, void *pool_data) | |
581 | { | |
5e2f89b5 | 582 | size_t size = (size_t)pool_data; |
53184082 MD |
583 | return kmalloc(size, gfp_mask); |
584 | } | |
585 | EXPORT_SYMBOL(mempool_kmalloc); | |
586 | ||
587 | void mempool_kfree(void *element, void *pool_data) | |
588 | { | |
589 | kfree(element); | |
590 | } | |
591 | EXPORT_SYMBOL(mempool_kfree); | |
592 | ||
0225bdfa KO |
593 | void *mempool_kvmalloc(gfp_t gfp_mask, void *pool_data) |
594 | { | |
595 | size_t size = (size_t)pool_data; | |
596 | return kvmalloc(size, gfp_mask); | |
597 | } | |
598 | EXPORT_SYMBOL(mempool_kvmalloc); | |
599 | ||
600 | void mempool_kvfree(void *element, void *pool_data) | |
601 | { | |
602 | kvfree(element); | |
603 | } | |
604 | EXPORT_SYMBOL(mempool_kvfree); | |
605 | ||
6e0678f3 MD |
606 | /* |
607 | * A simple mempool-backed page allocator that allocates pages | |
608 | * of the order specified by pool_data. | |
609 | */ | |
610 | void *mempool_alloc_pages(gfp_t gfp_mask, void *pool_data) | |
611 | { | |
612 | int order = (int)(long)pool_data; | |
613 | return alloc_pages(gfp_mask, order); | |
614 | } | |
615 | EXPORT_SYMBOL(mempool_alloc_pages); | |
616 | ||
617 | void mempool_free_pages(void *element, void *pool_data) | |
618 | { | |
619 | int order = (int)(long)pool_data; | |
620 | __free_pages(element, order); | |
621 | } | |
622 | EXPORT_SYMBOL(mempool_free_pages); |