dmaengine: at_hdmac: drop useless LIST_HEAD
[linux-2.6-block.git] / mm / mempool.c
CommitLineData
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
LT
19#include <linux/mempool.h>
20#include <linux/blkdev.h>
21#include <linux/writeback.h>
e244c9e6 22#include "slab.h"
1da177e4 23
bdfedb76
DR
24#if defined(CONFIG_DEBUG_SLAB) || defined(CONFIG_SLUB_DEBUG_ON)
25static void poison_error(mempool_t *pool, void *element, size_t size,
26 size_t byte)
27{
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);
31 int i;
32
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 ? "..." : "");
39 dump_stack();
40}
41
42static void __check_element(mempool_t *pool, void *element, size_t size)
43{
44 u8 *obj = element;
45 size_t i;
46
47 for (i = 0; i < size; i++) {
48 u8 exp = (i < size - 1) ? POISON_FREE : POISON_END;
49
50 if (obj[i] != exp) {
51 poison_error(pool, element, size, i);
52 return;
53 }
54 }
55 memset(obj, POISON_INUSE, size);
56}
57
58static void check_element(mempool_t *pool, void *element)
59{
60 /* Mempools backed by slab allocator */
61 if (pool->free == mempool_free_slab || pool->free == mempool_kfree)
62 __check_element(pool, element, ksize(element));
63
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);
68
69 __check_element(pool, addr, 1UL << (PAGE_SHIFT + order));
70 kunmap_atomic(addr);
71 }
72}
73
74static void __poison_element(void *element, size_t size)
75{
76 u8 *obj = element;
77
78 memset(obj, POISON_FREE, size - 1);
79 obj[size - 1] = POISON_END;
80}
81
82static void poison_element(mempool_t *pool, void *element)
83{
84 /* Mempools backed by slab allocator */
85 if (pool->alloc == mempool_alloc_slab || pool->alloc == mempool_kmalloc)
86 __poison_element(element, ksize(element));
87
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);
92
93 __poison_element(addr, 1UL << (PAGE_SHIFT + order));
94 kunmap_atomic(addr);
95 }
96}
97#else /* CONFIG_DEBUG_SLAB || CONFIG_SLUB_DEBUG_ON */
98static inline void check_element(mempool_t *pool, void *element)
99{
100}
101static inline void poison_element(mempool_t *pool, void *element)
102{
103}
104#endif /* CONFIG_DEBUG_SLAB || CONFIG_SLUB_DEBUG_ON */
105
6860f634 106static __always_inline void kasan_poison_element(mempool_t *pool, void *element)
92393615 107{
9b75a867 108 if (pool->alloc == mempool_alloc_slab || pool->alloc == mempool_kmalloc)
6860f634 109 kasan_poison_kfree(element, _RET_IP_);
92393615
AR
110 if (pool->alloc == mempool_alloc_pages)
111 kasan_free_pages(element, (unsigned long)pool->pool_data);
112}
113
8cded866 114static void kasan_unpoison_element(mempool_t *pool, void *element)
92393615 115{
9b75a867
AR
116 if (pool->alloc == mempool_alloc_slab || pool->alloc == mempool_kmalloc)
117 kasan_unpoison_slab(element);
92393615
AR
118 if (pool->alloc == mempool_alloc_pages)
119 kasan_alloc_pages(element, (unsigned long)pool->pool_data);
120}
121
6860f634 122static __always_inline void add_element(mempool_t *pool, void *element)
1da177e4
LT
123{
124 BUG_ON(pool->curr_nr >= pool->min_nr);
bdfedb76 125 poison_element(pool, element);
92393615 126 kasan_poison_element(pool, element);
1da177e4
LT
127 pool->elements[pool->curr_nr++] = element;
128}
129
8cded866 130static void *remove_element(mempool_t *pool)
1da177e4 131{
bdfedb76
DR
132 void *element = pool->elements[--pool->curr_nr];
133
134 BUG_ON(pool->curr_nr < 0);
8cded866 135 kasan_unpoison_element(pool, element);
76401310 136 check_element(pool, element);
bdfedb76 137 return element;
1da177e4
LT
138}
139
c1a67fef
KO
140/**
141 * mempool_exit - exit a mempool initialized with mempool_init()
142 * @pool: pointer to the memory pool which was initialized with
143 * mempool_init().
144 *
145 * Free all reserved elements in @pool and @pool itself. This function
146 * only sleeps if the free_fn() function sleeps.
147 *
148 * May be called on a zeroed but uninitialized mempool (i.e. allocated with
149 * kzalloc()).
150 */
151void mempool_exit(mempool_t *pool)
152{
153 while (pool->curr_nr) {
8cded866 154 void *element = remove_element(pool);
c1a67fef
KO
155 pool->free(element, pool->pool_data);
156 }
157 kfree(pool->elements);
158 pool->elements = NULL;
159}
160EXPORT_SYMBOL(mempool_exit);
161
0565d317
TH
162/**
163 * mempool_destroy - deallocate a memory pool
164 * @pool: pointer to the memory pool which was allocated via
165 * mempool_create().
166 *
167 * Free all reserved elements in @pool and @pool itself. This function
168 * only sleeps if the free_fn() function sleeps.
169 */
170void mempool_destroy(mempool_t *pool)
1da177e4 171{
4e3ca3e0
SS
172 if (unlikely(!pool))
173 return;
174
c1a67fef 175 mempool_exit(pool);
1da177e4
LT
176 kfree(pool);
177}
0565d317 178EXPORT_SYMBOL(mempool_destroy);
1da177e4 179
c1a67fef
KO
180int 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)
183{
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);
190
191 pool->elements = kmalloc_array_node(min_nr, sizeof(void *),
192 gfp_mask, node_id);
193 if (!pool->elements)
194 return -ENOMEM;
195
196 /*
197 * First pre-allocate the guaranteed number of buffers.
198 */
199 while (pool->curr_nr < pool->min_nr) {
200 void *element;
201
202 element = pool->alloc(gfp_mask, pool->pool_data);
203 if (unlikely(!element)) {
204 mempool_exit(pool);
205 return -ENOMEM;
206 }
207 add_element(pool, element);
208 }
209
210 return 0;
211}
212EXPORT_SYMBOL(mempool_init_node);
213
214/**
215 * mempool_init - initialize a memory pool
a3bf6ce3 216 * @pool: pointer to the memory pool that should be initialized
c1a67fef
KO
217 * @min_nr: the minimum number of elements guaranteed to be
218 * allocated for this pool.
219 * @alloc_fn: user-defined element-allocation function.
220 * @free_fn: user-defined element-freeing function.
221 * @pool_data: optional private data available to the user-defined functions.
222 *
223 * Like mempool_create(), but initializes the pool in (i.e. embedded in another
224 * structure).
225 */
226int mempool_init(mempool_t *pool, int min_nr, mempool_alloc_t *alloc_fn,
227 mempool_free_t *free_fn, void *pool_data)
228{
229 return mempool_init_node(pool, min_nr, alloc_fn, free_fn,
230 pool_data, GFP_KERNEL, NUMA_NO_NODE);
231
232}
233EXPORT_SYMBOL(mempool_init);
234
1da177e4
LT
235/**
236 * mempool_create - create a memory pool
237 * @min_nr: the minimum number of elements guaranteed to be
238 * allocated for this pool.
239 * @alloc_fn: user-defined element-allocation function.
240 * @free_fn: user-defined element-freeing function.
241 * @pool_data: optional private data available to the user-defined functions.
242 *
243 * this function creates and allocates a guaranteed size, preallocated
72fd4a35 244 * memory pool. The pool can be used from the mempool_alloc() and mempool_free()
1da177e4 245 * functions. This function might sleep. Both the alloc_fn() and the free_fn()
72fd4a35 246 * functions might sleep - as long as the mempool_alloc() function is not called
1da177e4
LT
247 * from IRQ contexts.
248 */
1946089a 249mempool_t *mempool_create(int min_nr, mempool_alloc_t *alloc_fn,
1da177e4
LT
250 mempool_free_t *free_fn, void *pool_data)
251{
a91a5ac6
TH
252 return mempool_create_node(min_nr,alloc_fn,free_fn, pool_data,
253 GFP_KERNEL, NUMA_NO_NODE);
1946089a
CL
254}
255EXPORT_SYMBOL(mempool_create);
1da177e4 256
1946089a 257mempool_t *mempool_create_node(int min_nr, mempool_alloc_t *alloc_fn,
a91a5ac6
TH
258 mempool_free_t *free_fn, void *pool_data,
259 gfp_t gfp_mask, int node_id)
1946089a
CL
260{
261 mempool_t *pool;
c1a67fef 262
7b5219db 263 pool = kzalloc_node(sizeof(*pool), gfp_mask, node_id);
1da177e4
LT
264 if (!pool)
265 return NULL;
c1a67fef
KO
266
267 if (mempool_init_node(pool, min_nr, alloc_fn, free_fn, pool_data,
268 gfp_mask, node_id)) {
1da177e4
LT
269 kfree(pool);
270 return NULL;
271 }
1da177e4 272
1da177e4
LT
273 return pool;
274}
1946089a 275EXPORT_SYMBOL(mempool_create_node);
1da177e4
LT
276
277/**
278 * mempool_resize - resize an existing memory pool
279 * @pool: pointer to the memory pool which was allocated via
280 * mempool_create().
281 * @new_min_nr: the new minimum number of elements guaranteed to be
282 * allocated for this pool.
1da177e4
LT
283 *
284 * This function shrinks/grows the pool. In the case of growing,
285 * it cannot be guaranteed that the pool will be grown to the new
286 * size immediately, but new mempool_free() calls will refill it.
11d83360 287 * This function may sleep.
1da177e4
LT
288 *
289 * Note, the caller must guarantee that no mempool_destroy is called
290 * while this function is running. mempool_alloc() & mempool_free()
291 * might be called (eg. from IRQ contexts) while this function executes.
292 */
11d83360 293int mempool_resize(mempool_t *pool, int new_min_nr)
1da177e4
LT
294{
295 void *element;
296 void **new_elements;
297 unsigned long flags;
298
299 BUG_ON(new_min_nr <= 0);
11d83360 300 might_sleep();
1da177e4
LT
301
302 spin_lock_irqsave(&pool->lock, flags);
303 if (new_min_nr <= pool->min_nr) {
304 while (new_min_nr < pool->curr_nr) {
8cded866 305 element = remove_element(pool);
1da177e4
LT
306 spin_unlock_irqrestore(&pool->lock, flags);
307 pool->free(element, pool->pool_data);
308 spin_lock_irqsave(&pool->lock, flags);
309 }
310 pool->min_nr = new_min_nr;
311 goto out_unlock;
312 }
313 spin_unlock_irqrestore(&pool->lock, flags);
314
315 /* Grow the pool */
11d83360
DR
316 new_elements = kmalloc_array(new_min_nr, sizeof(*new_elements),
317 GFP_KERNEL);
1da177e4
LT
318 if (!new_elements)
319 return -ENOMEM;
320
321 spin_lock_irqsave(&pool->lock, flags);
322 if (unlikely(new_min_nr <= pool->min_nr)) {
323 /* Raced, other resize will do our work */
324 spin_unlock_irqrestore(&pool->lock, flags);
325 kfree(new_elements);
326 goto out;
327 }
328 memcpy(new_elements, pool->elements,
329 pool->curr_nr * sizeof(*new_elements));
330 kfree(pool->elements);
331 pool->elements = new_elements;
332 pool->min_nr = new_min_nr;
333
334 while (pool->curr_nr < pool->min_nr) {
335 spin_unlock_irqrestore(&pool->lock, flags);
11d83360 336 element = pool->alloc(GFP_KERNEL, pool->pool_data);
1da177e4
LT
337 if (!element)
338 goto out;
339 spin_lock_irqsave(&pool->lock, flags);
340 if (pool->curr_nr < pool->min_nr) {
341 add_element(pool, element);
342 } else {
343 spin_unlock_irqrestore(&pool->lock, flags);
344 pool->free(element, pool->pool_data); /* Raced */
345 goto out;
346 }
347 }
348out_unlock:
349 spin_unlock_irqrestore(&pool->lock, flags);
350out:
351 return 0;
352}
353EXPORT_SYMBOL(mempool_resize);
354
1da177e4
LT
355/**
356 * mempool_alloc - allocate an element from a specific memory pool
357 * @pool: pointer to the memory pool which was allocated via
358 * mempool_create().
359 * @gfp_mask: the usual allocation bitmask.
360 *
72fd4a35 361 * this function only sleeps if the alloc_fn() function sleeps or
1da177e4
LT
362 * returns NULL. Note that due to preallocation, this function
363 * *never* fails when called from process contexts. (it might
364 * fail if called from an IRQ context.)
4e390b2b 365 * Note: using __GFP_ZERO is not supported.
1da177e4 366 */
f9054c70 367void *mempool_alloc(mempool_t *pool, gfp_t gfp_mask)
1da177e4
LT
368{
369 void *element;
370 unsigned long flags;
ac6424b9 371 wait_queue_entry_t wait;
6daa0e28 372 gfp_t gfp_temp;
20a77776 373
8bf8fcb0 374 VM_WARN_ON_ONCE(gfp_mask & __GFP_ZERO);
d0164adc 375 might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM);
b84a35be 376
4e390b2b 377 gfp_mask |= __GFP_NOMEMALLOC; /* don't allocate emergency reserves */
b84a35be
NP
378 gfp_mask |= __GFP_NORETRY; /* don't loop in __alloc_pages */
379 gfp_mask |= __GFP_NOWARN; /* failures are OK */
1da177e4 380
d0164adc 381 gfp_temp = gfp_mask & ~(__GFP_DIRECT_RECLAIM|__GFP_IO);
20a77776 382
1da177e4 383repeat_alloc:
20a77776
NP
384
385 element = pool->alloc(gfp_temp, pool->pool_data);
1da177e4
LT
386 if (likely(element != NULL))
387 return element;
388
1da177e4
LT
389 spin_lock_irqsave(&pool->lock, flags);
390 if (likely(pool->curr_nr)) {
8cded866 391 element = remove_element(pool);
1da177e4 392 spin_unlock_irqrestore(&pool->lock, flags);
5b990546
TH
393 /* paired with rmb in mempool_free(), read comment there */
394 smp_wmb();
17411962
CM
395 /*
396 * Update the allocation stack trace as this is more useful
397 * for debugging.
398 */
399 kmemleak_update_trace(element);
1da177e4
LT
400 return element;
401 }
1da177e4 402
1ebb7044 403 /*
d0164adc 404 * We use gfp mask w/o direct reclaim or IO for the first round. If
1ebb7044
TH
405 * alloc failed with that and @pool was empty, retry immediately.
406 */
4e390b2b 407 if (gfp_temp != gfp_mask) {
1ebb7044
TH
408 spin_unlock_irqrestore(&pool->lock, flags);
409 gfp_temp = gfp_mask;
410 goto repeat_alloc;
411 }
412
d0164adc
MG
413 /* We must not sleep if !__GFP_DIRECT_RECLAIM */
414 if (!(gfp_mask & __GFP_DIRECT_RECLAIM)) {
5b990546 415 spin_unlock_irqrestore(&pool->lock, flags);
1da177e4 416 return NULL;
5b990546 417 }
1da177e4 418
5b990546 419 /* Let's wait for someone else to return an element to @pool */
01890a4c 420 init_wait(&wait);
1da177e4 421 prepare_to_wait(&pool->wait, &wait, TASK_UNINTERRUPTIBLE);
1da177e4 422
5b990546
TH
423 spin_unlock_irqrestore(&pool->lock, flags);
424
425 /*
426 * FIXME: this should be io_schedule(). The timeout is there as a
427 * workaround for some DM problems in 2.6.18.
428 */
429 io_schedule_timeout(5*HZ);
430
431 finish_wait(&pool->wait, &wait);
1da177e4
LT
432 goto repeat_alloc;
433}
434EXPORT_SYMBOL(mempool_alloc);
435
436/**
437 * mempool_free - return an element to the pool.
438 * @element: pool element pointer.
439 * @pool: pointer to the memory pool which was allocated via
440 * mempool_create().
441 *
442 * this function only sleeps if the free_fn() function sleeps.
443 */
444void mempool_free(void *element, mempool_t *pool)
445{
446 unsigned long flags;
447
c80e7a82
RR
448 if (unlikely(element == NULL))
449 return;
450
5b990546
TH
451 /*
452 * Paired with the wmb in mempool_alloc(). The preceding read is
453 * for @element and the following @pool->curr_nr. This ensures
454 * that the visible value of @pool->curr_nr is from after the
455 * allocation of @element. This is necessary for fringe cases
456 * where @element was passed to this task without going through
457 * barriers.
458 *
459 * For example, assume @p is %NULL at the beginning and one task
460 * performs "p = mempool_alloc(...);" while another task is doing
461 * "while (!p) cpu_relax(); mempool_free(p, ...);". This function
462 * may end up using curr_nr value which is from before allocation
463 * of @p without the following rmb.
464 */
465 smp_rmb();
466
467 /*
468 * For correctness, we need a test which is guaranteed to trigger
469 * if curr_nr + #allocated == min_nr. Testing curr_nr < min_nr
470 * without locking achieves that and refilling as soon as possible
471 * is desirable.
472 *
473 * Because curr_nr visible here is always a value after the
474 * allocation of @element, any task which decremented curr_nr below
475 * min_nr is guaranteed to see curr_nr < min_nr unless curr_nr gets
476 * incremented to min_nr afterwards. If curr_nr gets incremented
477 * to min_nr after the allocation of @element, the elements
478 * allocated after that are subject to the same guarantee.
479 *
480 * Waiters happen iff curr_nr is 0 and the above guarantee also
481 * ensures that there will be frees which return elements to the
482 * pool waking up the waiters.
483 */
eb9a3c62 484 if (unlikely(pool->curr_nr < pool->min_nr)) {
1da177e4 485 spin_lock_irqsave(&pool->lock, flags);
eb9a3c62 486 if (likely(pool->curr_nr < pool->min_nr)) {
1da177e4
LT
487 add_element(pool, element);
488 spin_unlock_irqrestore(&pool->lock, flags);
489 wake_up(&pool->wait);
490 return;
491 }
492 spin_unlock_irqrestore(&pool->lock, flags);
493 }
494 pool->free(element, pool->pool_data);
495}
496EXPORT_SYMBOL(mempool_free);
497
498/*
499 * A commonly used alloc and free fn.
500 */
dd0fc66f 501void *mempool_alloc_slab(gfp_t gfp_mask, void *pool_data)
1da177e4 502{
fcc234f8 503 struct kmem_cache *mem = pool_data;
e244c9e6 504 VM_BUG_ON(mem->ctor);
1da177e4
LT
505 return kmem_cache_alloc(mem, gfp_mask);
506}
507EXPORT_SYMBOL(mempool_alloc_slab);
508
509void mempool_free_slab(void *element, void *pool_data)
510{
fcc234f8 511 struct kmem_cache *mem = pool_data;
1da177e4
LT
512 kmem_cache_free(mem, element);
513}
514EXPORT_SYMBOL(mempool_free_slab);
6e0678f3 515
53184082
MD
516/*
517 * A commonly used alloc and free fn that kmalloc/kfrees the amount of memory
183ff22b 518 * specified by pool_data
53184082
MD
519 */
520void *mempool_kmalloc(gfp_t gfp_mask, void *pool_data)
521{
5e2f89b5 522 size_t size = (size_t)pool_data;
53184082
MD
523 return kmalloc(size, gfp_mask);
524}
525EXPORT_SYMBOL(mempool_kmalloc);
526
527void mempool_kfree(void *element, void *pool_data)
528{
529 kfree(element);
530}
531EXPORT_SYMBOL(mempool_kfree);
532
6e0678f3
MD
533/*
534 * A simple mempool-backed page allocator that allocates pages
535 * of the order specified by pool_data.
536 */
537void *mempool_alloc_pages(gfp_t gfp_mask, void *pool_data)
538{
539 int order = (int)(long)pool_data;
540 return alloc_pages(gfp_mask, order);
541}
542EXPORT_SYMBOL(mempool_alloc_pages);
543
544void mempool_free_pages(void *element, void *pool_data)
545{
546 int order = (int)(long)pool_data;
547 __free_pages(element, order);
548}
549EXPORT_SYMBOL(mempool_free_pages);