Loading...
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 * debugging by David Rientjes, Copyright (C) 2015
10 */
11
12#include <linux/mm.h>
13#include <linux/slab.h>
14#include <linux/highmem.h>
15#include <linux/kasan.h>
16#include <linux/kmemleak.h>
17#include <linux/export.h>
18#include <linux/mempool.h>
19#include <linux/blkdev.h>
20#include <linux/writeback.h>
21#include "slab.h"
22
23#if defined(CONFIG_DEBUG_SLAB) || defined(CONFIG_SLUB_DEBUG_ON)
24static 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
41static 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
57static void check_element(mempool_t *pool, void *element)
58{
59 /* Mempools backed by slab allocator */
60 if (pool->free == mempool_free_slab || pool->free == mempool_kfree)
61 __check_element(pool, element, ksize(element));
62
63 /* Mempools backed by page allocator */
64 if (pool->free == mempool_free_pages) {
65 int order = (int)(long)pool->pool_data;
66 void *addr = kmap_atomic((struct page *)element);
67
68 __check_element(pool, addr, 1UL << (PAGE_SHIFT + order));
69 kunmap_atomic(addr);
70 }
71}
72
73static void __poison_element(void *element, size_t size)
74{
75 u8 *obj = element;
76
77 memset(obj, POISON_FREE, size - 1);
78 obj[size - 1] = POISON_END;
79}
80
81static void poison_element(mempool_t *pool, void *element)
82{
83 /* Mempools backed by slab allocator */
84 if (pool->alloc == mempool_alloc_slab || pool->alloc == mempool_kmalloc)
85 __poison_element(element, ksize(element));
86
87 /* Mempools backed by page allocator */
88 if (pool->alloc == mempool_alloc_pages) {
89 int order = (int)(long)pool->pool_data;
90 void *addr = kmap_atomic((struct page *)element);
91
92 __poison_element(addr, 1UL << (PAGE_SHIFT + order));
93 kunmap_atomic(addr);
94 }
95}
96#else /* CONFIG_DEBUG_SLAB || CONFIG_SLUB_DEBUG_ON */
97static inline void check_element(mempool_t *pool, void *element)
98{
99}
100static inline void poison_element(mempool_t *pool, void *element)
101{
102}
103#endif /* CONFIG_DEBUG_SLAB || CONFIG_SLUB_DEBUG_ON */
104
105static void kasan_poison_element(mempool_t *pool, void *element)
106{
107 if (pool->alloc == mempool_alloc_slab)
108 kasan_slab_free(pool->pool_data, element);
109 if (pool->alloc == mempool_kmalloc)
110 kasan_kfree(element);
111 if (pool->alloc == mempool_alloc_pages)
112 kasan_free_pages(element, (unsigned long)pool->pool_data);
113}
114
115static void kasan_unpoison_element(mempool_t *pool, void *element, gfp_t flags)
116{
117 if (pool->alloc == mempool_alloc_slab)
118 kasan_slab_alloc(pool->pool_data, element, flags);
119 if (pool->alloc == mempool_kmalloc)
120 kasan_krealloc(element, (size_t)pool->pool_data, flags);
121 if (pool->alloc == mempool_alloc_pages)
122 kasan_alloc_pages(element, (unsigned long)pool->pool_data);
123}
124
125static void add_element(mempool_t *pool, void *element)
126{
127 BUG_ON(pool->curr_nr >= pool->min_nr);
128 poison_element(pool, element);
129 kasan_poison_element(pool, element);
130 pool->elements[pool->curr_nr++] = element;
131}
132
133static void *remove_element(mempool_t *pool, gfp_t flags)
134{
135 void *element = pool->elements[--pool->curr_nr];
136
137 BUG_ON(pool->curr_nr < 0);
138 kasan_unpoison_element(pool, element, flags);
139 check_element(pool, element);
140 return element;
141}
142
143/**
144 * mempool_destroy - deallocate a memory pool
145 * @pool: pointer to the memory pool which was allocated via
146 * mempool_create().
147 *
148 * Free all reserved elements in @pool and @pool itself. This function
149 * only sleeps if the free_fn() function sleeps.
150 */
151void mempool_destroy(mempool_t *pool)
152{
153 if (unlikely(!pool))
154 return;
155
156 while (pool->curr_nr) {
157 void *element = remove_element(pool, GFP_KERNEL);
158 pool->free(element, pool->pool_data);
159 }
160 kfree(pool->elements);
161 kfree(pool);
162}
163EXPORT_SYMBOL(mempool_destroy);
164
165/**
166 * mempool_create - create a memory pool
167 * @min_nr: the minimum number of elements guaranteed to be
168 * allocated for this pool.
169 * @alloc_fn: user-defined element-allocation function.
170 * @free_fn: user-defined element-freeing function.
171 * @pool_data: optional private data available to the user-defined functions.
172 *
173 * this function creates and allocates a guaranteed size, preallocated
174 * memory pool. The pool can be used from the mempool_alloc() and mempool_free()
175 * functions. This function might sleep. Both the alloc_fn() and the free_fn()
176 * functions might sleep - as long as the mempool_alloc() function is not called
177 * from IRQ contexts.
178 */
179mempool_t *mempool_create(int min_nr, mempool_alloc_t *alloc_fn,
180 mempool_free_t *free_fn, void *pool_data)
181{
182 return mempool_create_node(min_nr,alloc_fn,free_fn, pool_data,
183 GFP_KERNEL, NUMA_NO_NODE);
184}
185EXPORT_SYMBOL(mempool_create);
186
187mempool_t *mempool_create_node(int min_nr, mempool_alloc_t *alloc_fn,
188 mempool_free_t *free_fn, void *pool_data,
189 gfp_t gfp_mask, int node_id)
190{
191 mempool_t *pool;
192 pool = kzalloc_node(sizeof(*pool), gfp_mask, node_id);
193 if (!pool)
194 return NULL;
195 pool->elements = kmalloc_node(min_nr * sizeof(void *),
196 gfp_mask, node_id);
197 if (!pool->elements) {
198 kfree(pool);
199 return NULL;
200 }
201 spin_lock_init(&pool->lock);
202 pool->min_nr = min_nr;
203 pool->pool_data = pool_data;
204 init_waitqueue_head(&pool->wait);
205 pool->alloc = alloc_fn;
206 pool->free = free_fn;
207
208 /*
209 * First pre-allocate the guaranteed number of buffers.
210 */
211 while (pool->curr_nr < pool->min_nr) {
212 void *element;
213
214 element = pool->alloc(gfp_mask, pool->pool_data);
215 if (unlikely(!element)) {
216 mempool_destroy(pool);
217 return NULL;
218 }
219 add_element(pool, element);
220 }
221 return pool;
222}
223EXPORT_SYMBOL(mempool_create_node);
224
225/**
226 * mempool_resize - resize an existing memory pool
227 * @pool: pointer to the memory pool which was allocated via
228 * mempool_create().
229 * @new_min_nr: the new minimum number of elements guaranteed to be
230 * allocated for this pool.
231 *
232 * This function shrinks/grows the pool. In the case of growing,
233 * it cannot be guaranteed that the pool will be grown to the new
234 * size immediately, but new mempool_free() calls will refill it.
235 * This function may sleep.
236 *
237 * Note, the caller must guarantee that no mempool_destroy is called
238 * while this function is running. mempool_alloc() & mempool_free()
239 * might be called (eg. from IRQ contexts) while this function executes.
240 */
241int mempool_resize(mempool_t *pool, int new_min_nr)
242{
243 void *element;
244 void **new_elements;
245 unsigned long flags;
246
247 BUG_ON(new_min_nr <= 0);
248 might_sleep();
249
250 spin_lock_irqsave(&pool->lock, flags);
251 if (new_min_nr <= pool->min_nr) {
252 while (new_min_nr < pool->curr_nr) {
253 element = remove_element(pool, GFP_KERNEL);
254 spin_unlock_irqrestore(&pool->lock, flags);
255 pool->free(element, pool->pool_data);
256 spin_lock_irqsave(&pool->lock, flags);
257 }
258 pool->min_nr = new_min_nr;
259 goto out_unlock;
260 }
261 spin_unlock_irqrestore(&pool->lock, flags);
262
263 /* Grow the pool */
264 new_elements = kmalloc_array(new_min_nr, sizeof(*new_elements),
265 GFP_KERNEL);
266 if (!new_elements)
267 return -ENOMEM;
268
269 spin_lock_irqsave(&pool->lock, flags);
270 if (unlikely(new_min_nr <= pool->min_nr)) {
271 /* Raced, other resize will do our work */
272 spin_unlock_irqrestore(&pool->lock, flags);
273 kfree(new_elements);
274 goto out;
275 }
276 memcpy(new_elements, pool->elements,
277 pool->curr_nr * sizeof(*new_elements));
278 kfree(pool->elements);
279 pool->elements = new_elements;
280 pool->min_nr = new_min_nr;
281
282 while (pool->curr_nr < pool->min_nr) {
283 spin_unlock_irqrestore(&pool->lock, flags);
284 element = pool->alloc(GFP_KERNEL, pool->pool_data);
285 if (!element)
286 goto out;
287 spin_lock_irqsave(&pool->lock, flags);
288 if (pool->curr_nr < pool->min_nr) {
289 add_element(pool, element);
290 } else {
291 spin_unlock_irqrestore(&pool->lock, flags);
292 pool->free(element, pool->pool_data); /* Raced */
293 goto out;
294 }
295 }
296out_unlock:
297 spin_unlock_irqrestore(&pool->lock, flags);
298out:
299 return 0;
300}
301EXPORT_SYMBOL(mempool_resize);
302
303/**
304 * mempool_alloc - allocate an element from a specific memory pool
305 * @pool: pointer to the memory pool which was allocated via
306 * mempool_create().
307 * @gfp_mask: the usual allocation bitmask.
308 *
309 * this function only sleeps if the alloc_fn() function sleeps or
310 * returns NULL. Note that due to preallocation, this function
311 * *never* fails when called from process contexts. (it might
312 * fail if called from an IRQ context.)
313 * Note: neither __GFP_NOMEMALLOC nor __GFP_ZERO are supported.
314 */
315void *mempool_alloc(mempool_t *pool, gfp_t gfp_mask)
316{
317 void *element;
318 unsigned long flags;
319 wait_queue_t wait;
320 gfp_t gfp_temp;
321
322 /* If oom killed, memory reserves are essential to prevent livelock */
323 VM_WARN_ON_ONCE(gfp_mask & __GFP_NOMEMALLOC);
324 /* No element size to zero on allocation */
325 VM_WARN_ON_ONCE(gfp_mask & __GFP_ZERO);
326
327 might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM);
328
329 gfp_mask |= __GFP_NORETRY; /* don't loop in __alloc_pages */
330 gfp_mask |= __GFP_NOWARN; /* failures are OK */
331
332 gfp_temp = gfp_mask & ~(__GFP_DIRECT_RECLAIM|__GFP_IO);
333
334repeat_alloc:
335 if (likely(pool->curr_nr)) {
336 /*
337 * Don't allocate from emergency reserves if there are
338 * elements available. This check is racy, but it will
339 * be rechecked each loop.
340 */
341 gfp_temp |= __GFP_NOMEMALLOC;
342 }
343
344 element = pool->alloc(gfp_temp, pool->pool_data);
345 if (likely(element != NULL))
346 return element;
347
348 spin_lock_irqsave(&pool->lock, flags);
349 if (likely(pool->curr_nr)) {
350 element = remove_element(pool, gfp_temp);
351 spin_unlock_irqrestore(&pool->lock, flags);
352 /* paired with rmb in mempool_free(), read comment there */
353 smp_wmb();
354 /*
355 * Update the allocation stack trace as this is more useful
356 * for debugging.
357 */
358 kmemleak_update_trace(element);
359 return element;
360 }
361
362 /*
363 * We use gfp mask w/o direct reclaim or IO for the first round. If
364 * alloc failed with that and @pool was empty, retry immediately.
365 */
366 if ((gfp_temp & ~__GFP_NOMEMALLOC) != gfp_mask) {
367 spin_unlock_irqrestore(&pool->lock, flags);
368 gfp_temp = gfp_mask;
369 goto repeat_alloc;
370 }
371 gfp_temp = gfp_mask;
372
373 /* We must not sleep if !__GFP_DIRECT_RECLAIM */
374 if (!(gfp_mask & __GFP_DIRECT_RECLAIM)) {
375 spin_unlock_irqrestore(&pool->lock, flags);
376 return NULL;
377 }
378
379 /* Let's wait for someone else to return an element to @pool */
380 init_wait(&wait);
381 prepare_to_wait(&pool->wait, &wait, TASK_UNINTERRUPTIBLE);
382
383 spin_unlock_irqrestore(&pool->lock, flags);
384
385 /*
386 * FIXME: this should be io_schedule(). The timeout is there as a
387 * workaround for some DM problems in 2.6.18.
388 */
389 io_schedule_timeout(5*HZ);
390
391 finish_wait(&pool->wait, &wait);
392 goto repeat_alloc;
393}
394EXPORT_SYMBOL(mempool_alloc);
395
396/**
397 * mempool_free - return an element to the pool.
398 * @element: pool element pointer.
399 * @pool: pointer to the memory pool which was allocated via
400 * mempool_create().
401 *
402 * this function only sleeps if the free_fn() function sleeps.
403 */
404void mempool_free(void *element, mempool_t *pool)
405{
406 unsigned long flags;
407
408 if (unlikely(element == NULL))
409 return;
410
411 /*
412 * Paired with the wmb in mempool_alloc(). The preceding read is
413 * for @element and the following @pool->curr_nr. This ensures
414 * that the visible value of @pool->curr_nr is from after the
415 * allocation of @element. This is necessary for fringe cases
416 * where @element was passed to this task without going through
417 * barriers.
418 *
419 * For example, assume @p is %NULL at the beginning and one task
420 * performs "p = mempool_alloc(...);" while another task is doing
421 * "while (!p) cpu_relax(); mempool_free(p, ...);". This function
422 * may end up using curr_nr value which is from before allocation
423 * of @p without the following rmb.
424 */
425 smp_rmb();
426
427 /*
428 * For correctness, we need a test which is guaranteed to trigger
429 * if curr_nr + #allocated == min_nr. Testing curr_nr < min_nr
430 * without locking achieves that and refilling as soon as possible
431 * is desirable.
432 *
433 * Because curr_nr visible here is always a value after the
434 * allocation of @element, any task which decremented curr_nr below
435 * min_nr is guaranteed to see curr_nr < min_nr unless curr_nr gets
436 * incremented to min_nr afterwards. If curr_nr gets incremented
437 * to min_nr after the allocation of @element, the elements
438 * allocated after that are subject to the same guarantee.
439 *
440 * Waiters happen iff curr_nr is 0 and the above guarantee also
441 * ensures that there will be frees which return elements to the
442 * pool waking up the waiters.
443 */
444 if (unlikely(pool->curr_nr < pool->min_nr)) {
445 spin_lock_irqsave(&pool->lock, flags);
446 if (likely(pool->curr_nr < pool->min_nr)) {
447 add_element(pool, element);
448 spin_unlock_irqrestore(&pool->lock, flags);
449 wake_up(&pool->wait);
450 return;
451 }
452 spin_unlock_irqrestore(&pool->lock, flags);
453 }
454 pool->free(element, pool->pool_data);
455}
456EXPORT_SYMBOL(mempool_free);
457
458/*
459 * A commonly used alloc and free fn.
460 */
461void *mempool_alloc_slab(gfp_t gfp_mask, void *pool_data)
462{
463 struct kmem_cache *mem = pool_data;
464 VM_BUG_ON(mem->ctor);
465 return kmem_cache_alloc(mem, gfp_mask);
466}
467EXPORT_SYMBOL(mempool_alloc_slab);
468
469void mempool_free_slab(void *element, void *pool_data)
470{
471 struct kmem_cache *mem = pool_data;
472 kmem_cache_free(mem, element);
473}
474EXPORT_SYMBOL(mempool_free_slab);
475
476/*
477 * A commonly used alloc and free fn that kmalloc/kfrees the amount of memory
478 * specified by pool_data
479 */
480void *mempool_kmalloc(gfp_t gfp_mask, void *pool_data)
481{
482 size_t size = (size_t)pool_data;
483 return kmalloc(size, gfp_mask);
484}
485EXPORT_SYMBOL(mempool_kmalloc);
486
487void mempool_kfree(void *element, void *pool_data)
488{
489 kfree(element);
490}
491EXPORT_SYMBOL(mempool_kfree);
492
493/*
494 * A simple mempool-backed page allocator that allocates pages
495 * of the order specified by pool_data.
496 */
497void *mempool_alloc_pages(gfp_t gfp_mask, void *pool_data)
498{
499 int order = (int)(long)pool_data;
500 return alloc_pages(gfp_mask, order);
501}
502EXPORT_SYMBOL(mempool_alloc_pages);
503
504void mempool_free_pages(void *element, void *pool_data)
505{
506 int order = (int)(long)pool_data;
507 __free_pages(element, order);
508}
509EXPORT_SYMBOL(mempool_free_pages);
1// SPDX-License-Identifier: GPL-2.0
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
10 * debugging by David Rientjes, Copyright (C) 2015
11 */
12
13#include <linux/mm.h>
14#include <linux/slab.h>
15#include <linux/highmem.h>
16#include <linux/kasan.h>
17#include <linux/kmemleak.h>
18#include <linux/export.h>
19#include <linux/mempool.h>
20#include <linux/writeback.h>
21#include "slab.h"
22
23#ifdef CONFIG_SLUB_DEBUG_ON
24static 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
41static 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
57static void check_element(mempool_t *pool, void *element)
58{
59 /* Skip checking: KASAN might save its metadata in the element. */
60 if (kasan_enabled())
61 return;
62
63 /* Mempools backed by slab allocator */
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));
68 } else if (pool->free == mempool_free_pages) {
69 /* Mempools backed by page allocator */
70 int order = (int)(long)pool->pool_data;
71 void *addr = kmap_local_page((struct page *)element);
72
73 __check_element(pool, addr, 1UL << (PAGE_SHIFT + order));
74 kunmap_local(addr);
75 }
76}
77
78static 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
86static void poison_element(mempool_t *pool, void *element)
87{
88 /* Skip poisoning: KASAN might save its metadata in the element. */
89 if (kasan_enabled())
90 return;
91
92 /* Mempools backed by slab allocator */
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));
97 } else if (pool->alloc == mempool_alloc_pages) {
98 /* Mempools backed by page allocator */
99 int order = (int)(long)pool->pool_data;
100 void *addr = kmap_local_page((struct page *)element);
101
102 __poison_element(addr, 1UL << (PAGE_SHIFT + order));
103 kunmap_local(addr);
104 }
105}
106#else /* CONFIG_SLUB_DEBUG_ON */
107static inline void check_element(mempool_t *pool, void *element)
108{
109}
110static inline void poison_element(mempool_t *pool, void *element)
111{
112}
113#endif /* CONFIG_SLUB_DEBUG_ON */
114
115static __always_inline bool kasan_poison_element(mempool_t *pool, void *element)
116{
117 if (pool->alloc == mempool_alloc_slab || pool->alloc == mempool_kmalloc)
118 return kasan_mempool_poison_object(element);
119 else if (pool->alloc == mempool_alloc_pages)
120 return kasan_mempool_poison_pages(element,
121 (unsigned long)pool->pool_data);
122 return true;
123}
124
125static void kasan_unpoison_element(mempool_t *pool, void *element)
126{
127 if (pool->alloc == mempool_kmalloc)
128 kasan_mempool_unpoison_object(element, (size_t)pool->pool_data);
129 else if (pool->alloc == mempool_alloc_slab)
130 kasan_mempool_unpoison_object(element,
131 kmem_cache_size(pool->pool_data));
132 else if (pool->alloc == mempool_alloc_pages)
133 kasan_mempool_unpoison_pages(element,
134 (unsigned long)pool->pool_data);
135}
136
137static __always_inline void add_element(mempool_t *pool, void *element)
138{
139 BUG_ON(pool->curr_nr >= pool->min_nr);
140 poison_element(pool, element);
141 if (kasan_poison_element(pool, element))
142 pool->elements[pool->curr_nr++] = element;
143}
144
145static void *remove_element(mempool_t *pool)
146{
147 void *element = pool->elements[--pool->curr_nr];
148
149 BUG_ON(pool->curr_nr < 0);
150 kasan_unpoison_element(pool, element);
151 check_element(pool, element);
152 return element;
153}
154
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 */
166void mempool_exit(mempool_t *pool)
167{
168 while (pool->curr_nr) {
169 void *element = remove_element(pool);
170 pool->free(element, pool->pool_data);
171 }
172 kfree(pool->elements);
173 pool->elements = NULL;
174}
175EXPORT_SYMBOL(mempool_exit);
176
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 */
185void mempool_destroy(mempool_t *pool)
186{
187 if (unlikely(!pool))
188 return;
189
190 mempool_exit(pool);
191 kfree(pool);
192}
193EXPORT_SYMBOL(mempool_destroy);
194
195int 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}
227EXPORT_SYMBOL(mempool_init_node);
228
229/**
230 * mempool_init - initialize a memory pool
231 * @pool: pointer to the memory pool that should be initialized
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).
240 *
241 * Return: %0 on success, negative error code otherwise.
242 */
243int 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}
250EXPORT_SYMBOL(mempool_init);
251
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
261 * memory pool. The pool can be used from the mempool_alloc() and mempool_free()
262 * functions. This function might sleep. Both the alloc_fn() and the free_fn()
263 * functions might sleep - as long as the mempool_alloc() function is not called
264 * from IRQ contexts.
265 *
266 * Return: pointer to the created memory pool object or %NULL on error.
267 */
268mempool_t *mempool_create(int min_nr, mempool_alloc_t *alloc_fn,
269 mempool_free_t *free_fn, void *pool_data)
270{
271 return mempool_create_node(min_nr, alloc_fn, free_fn, pool_data,
272 GFP_KERNEL, NUMA_NO_NODE);
273}
274EXPORT_SYMBOL(mempool_create);
275
276mempool_t *mempool_create_node(int min_nr, mempool_alloc_t *alloc_fn,
277 mempool_free_t *free_fn, void *pool_data,
278 gfp_t gfp_mask, int node_id)
279{
280 mempool_t *pool;
281
282 pool = kzalloc_node(sizeof(*pool), gfp_mask, node_id);
283 if (!pool)
284 return NULL;
285
286 if (mempool_init_node(pool, min_nr, alloc_fn, free_fn, pool_data,
287 gfp_mask, node_id)) {
288 kfree(pool);
289 return NULL;
290 }
291
292 return pool;
293}
294EXPORT_SYMBOL(mempool_create_node);
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.
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.
306 * This function may sleep.
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.
311 *
312 * Return: %0 on success, negative error code otherwise.
313 */
314int mempool_resize(mempool_t *pool, int new_min_nr)
315{
316 void *element;
317 void **new_elements;
318 unsigned long flags;
319
320 BUG_ON(new_min_nr <= 0);
321 might_sleep();
322
323 spin_lock_irqsave(&pool->lock, flags);
324 if (new_min_nr <= pool->min_nr) {
325 while (new_min_nr < pool->curr_nr) {
326 element = remove_element(pool);
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 */
337 new_elements = kmalloc_array(new_min_nr, sizeof(*new_elements),
338 GFP_KERNEL);
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);
357 element = pool->alloc(GFP_KERNEL, pool->pool_data);
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 }
369out_unlock:
370 spin_unlock_irqrestore(&pool->lock, flags);
371out:
372 return 0;
373}
374EXPORT_SYMBOL(mempool_resize);
375
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 *
382 * this function only sleeps if the alloc_fn() function sleeps or
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.)
386 * Note: using __GFP_ZERO is not supported.
387 *
388 * Return: pointer to the allocated element or %NULL on error.
389 */
390void *mempool_alloc(mempool_t *pool, gfp_t gfp_mask)
391{
392 void *element;
393 unsigned long flags;
394 wait_queue_entry_t wait;
395 gfp_t gfp_temp;
396
397 VM_WARN_ON_ONCE(gfp_mask & __GFP_ZERO);
398 might_alloc(gfp_mask);
399
400 gfp_mask |= __GFP_NOMEMALLOC; /* don't allocate emergency reserves */
401 gfp_mask |= __GFP_NORETRY; /* don't loop in __alloc_pages */
402 gfp_mask |= __GFP_NOWARN; /* failures are OK */
403
404 gfp_temp = gfp_mask & ~(__GFP_DIRECT_RECLAIM|__GFP_IO);
405
406repeat_alloc:
407
408 element = pool->alloc(gfp_temp, pool->pool_data);
409 if (likely(element != NULL))
410 return element;
411
412 spin_lock_irqsave(&pool->lock, flags);
413 if (likely(pool->curr_nr)) {
414 element = remove_element(pool);
415 spin_unlock_irqrestore(&pool->lock, flags);
416 /* paired with rmb in mempool_free(), read comment there */
417 smp_wmb();
418 /*
419 * Update the allocation stack trace as this is more useful
420 * for debugging.
421 */
422 kmemleak_update_trace(element);
423 return element;
424 }
425
426 /*
427 * We use gfp mask w/o direct reclaim or IO for the first round. If
428 * alloc failed with that and @pool was empty, retry immediately.
429 */
430 if (gfp_temp != gfp_mask) {
431 spin_unlock_irqrestore(&pool->lock, flags);
432 gfp_temp = gfp_mask;
433 goto repeat_alloc;
434 }
435
436 /* We must not sleep if !__GFP_DIRECT_RECLAIM */
437 if (!(gfp_mask & __GFP_DIRECT_RECLAIM)) {
438 spin_unlock_irqrestore(&pool->lock, flags);
439 return NULL;
440 }
441
442 /* Let's wait for someone else to return an element to @pool */
443 init_wait(&wait);
444 prepare_to_wait(&pool->wait, &wait, TASK_UNINTERRUPTIBLE);
445
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);
455 goto repeat_alloc;
456}
457EXPORT_SYMBOL(mempool_alloc);
458
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 */
472void *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}
494EXPORT_SYMBOL(mempool_alloc_preallocated);
495
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 */
504void mempool_free(void *element, mempool_t *pool)
505{
506 unsigned long flags;
507
508 if (unlikely(element == NULL))
509 return;
510
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 */
544 if (unlikely(READ_ONCE(pool->curr_nr) < pool->min_nr)) {
545 spin_lock_irqsave(&pool->lock, flags);
546 if (likely(pool->curr_nr < pool->min_nr)) {
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}
556EXPORT_SYMBOL(mempool_free);
557
558/*
559 * A commonly used alloc and free fn.
560 */
561void *mempool_alloc_slab(gfp_t gfp_mask, void *pool_data)
562{
563 struct kmem_cache *mem = pool_data;
564 VM_BUG_ON(mem->ctor);
565 return kmem_cache_alloc(mem, gfp_mask);
566}
567EXPORT_SYMBOL(mempool_alloc_slab);
568
569void mempool_free_slab(void *element, void *pool_data)
570{
571 struct kmem_cache *mem = pool_data;
572 kmem_cache_free(mem, element);
573}
574EXPORT_SYMBOL(mempool_free_slab);
575
576/*
577 * A commonly used alloc and free fn that kmalloc/kfrees the amount of memory
578 * specified by pool_data
579 */
580void *mempool_kmalloc(gfp_t gfp_mask, void *pool_data)
581{
582 size_t size = (size_t)pool_data;
583 return kmalloc(size, gfp_mask);
584}
585EXPORT_SYMBOL(mempool_kmalloc);
586
587void mempool_kfree(void *element, void *pool_data)
588{
589 kfree(element);
590}
591EXPORT_SYMBOL(mempool_kfree);
592
593void *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}
598EXPORT_SYMBOL(mempool_kvmalloc);
599
600void mempool_kvfree(void *element, void *pool_data)
601{
602 kvfree(element);
603}
604EXPORT_SYMBOL(mempool_kvfree);
605
606/*
607 * A simple mempool-backed page allocator that allocates pages
608 * of the order specified by pool_data.
609 */
610void *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}
615EXPORT_SYMBOL(mempool_alloc_pages);
616
617void mempool_free_pages(void *element, void *pool_data)
618{
619 int order = (int)(long)pool_data;
620 __free_pages(element, order);
621}
622EXPORT_SYMBOL(mempool_free_pages);