1 // SPDX-License-Identifier: GPL-2.0
5 * memory buffer pool support. Such pools are mostly used
6 * for guaranteed, deadlock-free memory allocations during
9 * started by Ingo Molnar, Copyright (C) 2001
10 * debugging by David Rientjes, Copyright (C) 2015
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/blkdev.h>
21 #include <linux/writeback.h>
24 #if defined(CONFIG_DEBUG_SLAB) || defined(CONFIG_SLUB_DEBUG_ON)
25 static void poison_error(mempool_t *pool, void *element, size_t size,
28 const int nr = pool->curr_nr;
29 const int start = max_t(int, byte - (BITS_PER_LONG / 8), 0);
30 const int end = min_t(int, byte + (BITS_PER_LONG / 8), size);
33 pr_err("BUG: mempool element poison mismatch\n");
34 pr_err("Mempool %p size %zu\n", pool, size);
35 pr_err(" nr=%d @ %p: %s0x", nr, element, start > 0 ? "... " : "");
36 for (i = start; i < end; i++)
37 pr_cont("%x ", *(u8 *)(element + i));
38 pr_cont("%s\n", end < size ? "..." : "");
42 static void __check_element(mempool_t *pool, void *element, size_t size)
47 for (i = 0; i < size; i++) {
48 u8 exp = (i < size - 1) ? POISON_FREE : POISON_END;
51 poison_error(pool, element, size, i);
55 memset(obj, POISON_INUSE, size);
58 static void check_element(mempool_t *pool, void *element)
60 /* Mempools backed by slab allocator */
61 if (pool->free == mempool_free_slab || pool->free == mempool_kfree)
62 __check_element(pool, element, ksize(element));
64 /* Mempools backed by page allocator */
65 if (pool->free == mempool_free_pages) {
66 int order = (int)(long)pool->pool_data;
67 void *addr = kmap_atomic((struct page *)element);
69 __check_element(pool, addr, 1UL << (PAGE_SHIFT + order));
74 static void __poison_element(void *element, size_t size)
78 memset(obj, POISON_FREE, size - 1);
79 obj[size - 1] = POISON_END;
82 static void poison_element(mempool_t *pool, void *element)
84 /* Mempools backed by slab allocator */
85 if (pool->alloc == mempool_alloc_slab || pool->alloc == mempool_kmalloc)
86 __poison_element(element, ksize(element));
88 /* Mempools backed by page allocator */
89 if (pool->alloc == mempool_alloc_pages) {
90 int order = (int)(long)pool->pool_data;
91 void *addr = kmap_atomic((struct page *)element);
93 __poison_element(addr, 1UL << (PAGE_SHIFT + order));
97 #else /* CONFIG_DEBUG_SLAB || CONFIG_SLUB_DEBUG_ON */
98 static inline void check_element(mempool_t *pool, void *element)
101 static inline void poison_element(mempool_t *pool, void *element)
104 #endif /* CONFIG_DEBUG_SLAB || CONFIG_SLUB_DEBUG_ON */
106 static __always_inline void kasan_poison_element(mempool_t *pool, void *element)
108 if (pool->alloc == mempool_alloc_slab || pool->alloc == mempool_kmalloc)
109 kasan_poison_kfree(element, _RET_IP_);
110 if (pool->alloc == mempool_alloc_pages)
111 kasan_free_pages(element, (unsigned long)pool->pool_data);
114 static void kasan_unpoison_element(mempool_t *pool, void *element)
116 if (pool->alloc == mempool_alloc_slab || pool->alloc == mempool_kmalloc)
117 kasan_unpoison_slab(element);
118 if (pool->alloc == mempool_alloc_pages)
119 kasan_alloc_pages(element, (unsigned long)pool->pool_data);
122 static __always_inline void add_element(mempool_t *pool, void *element)
124 BUG_ON(pool->curr_nr >= pool->min_nr);
125 poison_element(pool, element);
126 kasan_poison_element(pool, element);
127 pool->elements[pool->curr_nr++] = element;
130 static void *remove_element(mempool_t *pool)
132 void *element = pool->elements[--pool->curr_nr];
134 BUG_ON(pool->curr_nr < 0);
135 kasan_unpoison_element(pool, element);
136 check_element(pool, element);
141 * mempool_exit - exit a mempool initialized with mempool_init()
142 * @pool: pointer to the memory pool which was initialized with
145 * Free all reserved elements in @pool and @pool itself. This function
146 * only sleeps if the free_fn() function sleeps.
148 * May be called on a zeroed but uninitialized mempool (i.e. allocated with
151 void mempool_exit(mempool_t *pool)
153 while (pool->curr_nr) {
154 void *element = remove_element(pool);
155 pool->free(element, pool->pool_data);
157 kfree(pool->elements);
158 pool->elements = NULL;
160 EXPORT_SYMBOL(mempool_exit);
163 * mempool_destroy - deallocate a memory pool
164 * @pool: pointer to the memory pool which was allocated via
167 * Free all reserved elements in @pool and @pool itself. This function
168 * only sleeps if the free_fn() function sleeps.
170 void mempool_destroy(mempool_t *pool)
178 EXPORT_SYMBOL(mempool_destroy);
180 int mempool_init_node(mempool_t *pool, int min_nr, mempool_alloc_t *alloc_fn,
181 mempool_free_t *free_fn, void *pool_data,
182 gfp_t gfp_mask, int node_id)
184 spin_lock_init(&pool->lock);
185 pool->min_nr = min_nr;
186 pool->pool_data = pool_data;
187 pool->alloc = alloc_fn;
188 pool->free = free_fn;
189 init_waitqueue_head(&pool->wait);
191 pool->elements = kmalloc_array_node(min_nr, sizeof(void *),
197 * First pre-allocate the guaranteed number of buffers.
199 while (pool->curr_nr < pool->min_nr) {
202 element = pool->alloc(gfp_mask, pool->pool_data);
203 if (unlikely(!element)) {
207 add_element(pool, element);
212 EXPORT_SYMBOL(mempool_init_node);
215 * mempool_init - initialize a memory pool
216 * @min_nr: the minimum number of elements guaranteed to be
217 * allocated for this pool.
218 * @alloc_fn: user-defined element-allocation function.
219 * @free_fn: user-defined element-freeing function.
220 * @pool_data: optional private data available to the user-defined functions.
222 * Like mempool_create(), but initializes the pool in (i.e. embedded in another
225 int mempool_init(mempool_t *pool, int min_nr, mempool_alloc_t *alloc_fn,
226 mempool_free_t *free_fn, void *pool_data)
228 return mempool_init_node(pool, min_nr, alloc_fn, free_fn,
229 pool_data, GFP_KERNEL, NUMA_NO_NODE);
232 EXPORT_SYMBOL(mempool_init);
235 * mempool_create - create a memory pool
236 * @min_nr: the minimum number of elements guaranteed to be
237 * allocated for this pool.
238 * @alloc_fn: user-defined element-allocation function.
239 * @free_fn: user-defined element-freeing function.
240 * @pool_data: optional private data available to the user-defined functions.
242 * this function creates and allocates a guaranteed size, preallocated
243 * memory pool. The pool can be used from the mempool_alloc() and mempool_free()
244 * functions. This function might sleep. Both the alloc_fn() and the free_fn()
245 * functions might sleep - as long as the mempool_alloc() function is not called
248 mempool_t *mempool_create(int min_nr, mempool_alloc_t *alloc_fn,
249 mempool_free_t *free_fn, void *pool_data)
251 return mempool_create_node(min_nr,alloc_fn,free_fn, pool_data,
252 GFP_KERNEL, NUMA_NO_NODE);
254 EXPORT_SYMBOL(mempool_create);
256 mempool_t *mempool_create_node(int min_nr, mempool_alloc_t *alloc_fn,
257 mempool_free_t *free_fn, void *pool_data,
258 gfp_t gfp_mask, int node_id)
262 pool = kzalloc_node(sizeof(*pool), gfp_mask, node_id);
266 if (mempool_init_node(pool, min_nr, alloc_fn, free_fn, pool_data,
267 gfp_mask, node_id)) {
274 EXPORT_SYMBOL(mempool_create_node);
277 * mempool_resize - resize an existing memory pool
278 * @pool: pointer to the memory pool which was allocated via
280 * @new_min_nr: the new minimum number of elements guaranteed to be
281 * allocated for this pool.
283 * This function shrinks/grows the pool. In the case of growing,
284 * it cannot be guaranteed that the pool will be grown to the new
285 * size immediately, but new mempool_free() calls will refill it.
286 * This function may sleep.
288 * Note, the caller must guarantee that no mempool_destroy is called
289 * while this function is running. mempool_alloc() & mempool_free()
290 * might be called (eg. from IRQ contexts) while this function executes.
292 int mempool_resize(mempool_t *pool, int new_min_nr)
298 BUG_ON(new_min_nr <= 0);
301 spin_lock_irqsave(&pool->lock, flags);
302 if (new_min_nr <= pool->min_nr) {
303 while (new_min_nr < pool->curr_nr) {
304 element = remove_element(pool);
305 spin_unlock_irqrestore(&pool->lock, flags);
306 pool->free(element, pool->pool_data);
307 spin_lock_irqsave(&pool->lock, flags);
309 pool->min_nr = new_min_nr;
312 spin_unlock_irqrestore(&pool->lock, flags);
315 new_elements = kmalloc_array(new_min_nr, sizeof(*new_elements),
320 spin_lock_irqsave(&pool->lock, flags);
321 if (unlikely(new_min_nr <= pool->min_nr)) {
322 /* Raced, other resize will do our work */
323 spin_unlock_irqrestore(&pool->lock, flags);
327 memcpy(new_elements, pool->elements,
328 pool->curr_nr * sizeof(*new_elements));
329 kfree(pool->elements);
330 pool->elements = new_elements;
331 pool->min_nr = new_min_nr;
333 while (pool->curr_nr < pool->min_nr) {
334 spin_unlock_irqrestore(&pool->lock, flags);
335 element = pool->alloc(GFP_KERNEL, pool->pool_data);
338 spin_lock_irqsave(&pool->lock, flags);
339 if (pool->curr_nr < pool->min_nr) {
340 add_element(pool, element);
342 spin_unlock_irqrestore(&pool->lock, flags);
343 pool->free(element, pool->pool_data); /* Raced */
348 spin_unlock_irqrestore(&pool->lock, flags);
352 EXPORT_SYMBOL(mempool_resize);
355 * mempool_alloc - allocate an element from a specific memory pool
356 * @pool: pointer to the memory pool which was allocated via
358 * @gfp_mask: the usual allocation bitmask.
360 * this function only sleeps if the alloc_fn() function sleeps or
361 * returns NULL. Note that due to preallocation, this function
362 * *never* fails when called from process contexts. (it might
363 * fail if called from an IRQ context.)
364 * Note: using __GFP_ZERO is not supported.
366 void *mempool_alloc(mempool_t *pool, gfp_t gfp_mask)
370 wait_queue_entry_t wait;
373 VM_WARN_ON_ONCE(gfp_mask & __GFP_ZERO);
374 might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM);
376 gfp_mask |= __GFP_NOMEMALLOC; /* don't allocate emergency reserves */
377 gfp_mask |= __GFP_NORETRY; /* don't loop in __alloc_pages */
378 gfp_mask |= __GFP_NOWARN; /* failures are OK */
380 gfp_temp = gfp_mask & ~(__GFP_DIRECT_RECLAIM|__GFP_IO);
384 element = pool->alloc(gfp_temp, pool->pool_data);
385 if (likely(element != NULL))
388 spin_lock_irqsave(&pool->lock, flags);
389 if (likely(pool->curr_nr)) {
390 element = remove_element(pool);
391 spin_unlock_irqrestore(&pool->lock, flags);
392 /* paired with rmb in mempool_free(), read comment there */
395 * Update the allocation stack trace as this is more useful
398 kmemleak_update_trace(element);
403 * We use gfp mask w/o direct reclaim or IO for the first round. If
404 * alloc failed with that and @pool was empty, retry immediately.
406 if (gfp_temp != gfp_mask) {
407 spin_unlock_irqrestore(&pool->lock, flags);
412 /* We must not sleep if !__GFP_DIRECT_RECLAIM */
413 if (!(gfp_mask & __GFP_DIRECT_RECLAIM)) {
414 spin_unlock_irqrestore(&pool->lock, flags);
418 /* Let's wait for someone else to return an element to @pool */
420 prepare_to_wait(&pool->wait, &wait, TASK_UNINTERRUPTIBLE);
422 spin_unlock_irqrestore(&pool->lock, flags);
425 * FIXME: this should be io_schedule(). The timeout is there as a
426 * workaround for some DM problems in 2.6.18.
428 io_schedule_timeout(5*HZ);
430 finish_wait(&pool->wait, &wait);
433 EXPORT_SYMBOL(mempool_alloc);
436 * mempool_free - return an element to the pool.
437 * @element: pool element pointer.
438 * @pool: pointer to the memory pool which was allocated via
441 * this function only sleeps if the free_fn() function sleeps.
443 void mempool_free(void *element, mempool_t *pool)
447 if (unlikely(element == NULL))
451 * Paired with the wmb in mempool_alloc(). The preceding read is
452 * for @element and the following @pool->curr_nr. This ensures
453 * that the visible value of @pool->curr_nr is from after the
454 * allocation of @element. This is necessary for fringe cases
455 * where @element was passed to this task without going through
458 * For example, assume @p is %NULL at the beginning and one task
459 * performs "p = mempool_alloc(...);" while another task is doing
460 * "while (!p) cpu_relax(); mempool_free(p, ...);". This function
461 * may end up using curr_nr value which is from before allocation
462 * of @p without the following rmb.
467 * For correctness, we need a test which is guaranteed to trigger
468 * if curr_nr + #allocated == min_nr. Testing curr_nr < min_nr
469 * without locking achieves that and refilling as soon as possible
472 * Because curr_nr visible here is always a value after the
473 * allocation of @element, any task which decremented curr_nr below
474 * min_nr is guaranteed to see curr_nr < min_nr unless curr_nr gets
475 * incremented to min_nr afterwards. If curr_nr gets incremented
476 * to min_nr after the allocation of @element, the elements
477 * allocated after that are subject to the same guarantee.
479 * Waiters happen iff curr_nr is 0 and the above guarantee also
480 * ensures that there will be frees which return elements to the
481 * pool waking up the waiters.
483 if (unlikely(pool->curr_nr < pool->min_nr)) {
484 spin_lock_irqsave(&pool->lock, flags);
485 if (likely(pool->curr_nr < pool->min_nr)) {
486 add_element(pool, element);
487 spin_unlock_irqrestore(&pool->lock, flags);
488 wake_up(&pool->wait);
491 spin_unlock_irqrestore(&pool->lock, flags);
493 pool->free(element, pool->pool_data);
495 EXPORT_SYMBOL(mempool_free);
498 * A commonly used alloc and free fn.
500 void *mempool_alloc_slab(gfp_t gfp_mask, void *pool_data)
502 struct kmem_cache *mem = pool_data;
503 VM_BUG_ON(mem->ctor);
504 return kmem_cache_alloc(mem, gfp_mask);
506 EXPORT_SYMBOL(mempool_alloc_slab);
508 void mempool_free_slab(void *element, void *pool_data)
510 struct kmem_cache *mem = pool_data;
511 kmem_cache_free(mem, element);
513 EXPORT_SYMBOL(mempool_free_slab);
516 * A commonly used alloc and free fn that kmalloc/kfrees the amount of memory
517 * specified by pool_data
519 void *mempool_kmalloc(gfp_t gfp_mask, void *pool_data)
521 size_t size = (size_t)pool_data;
522 return kmalloc(size, gfp_mask);
524 EXPORT_SYMBOL(mempool_kmalloc);
526 void mempool_kfree(void *element, void *pool_data)
530 EXPORT_SYMBOL(mempool_kfree);
533 * A simple mempool-backed page allocator that allocates pages
534 * of the order specified by pool_data.
536 void *mempool_alloc_pages(gfp_t gfp_mask, void *pool_data)
538 int order = (int)(long)pool_data;
539 return alloc_pages(gfp_mask, order);
541 EXPORT_SYMBOL(mempool_alloc_pages);
543 void mempool_free_pages(void *element, void *pool_data)
545 int order = (int)(long)pool_data;
546 __free_pages(element, order);
548 EXPORT_SYMBOL(mempool_free_pages);