Loading...
1/*
2 * Copyright (c) 2013 Red Hat, Inc. and Parallels Inc. All rights reserved.
3 * Authors: David Chinner and Glauber Costa
4 *
5 * Generic LRU infrastructure
6 */
7#include <linux/kernel.h>
8#include <linux/module.h>
9#include <linux/mm.h>
10#include <linux/list_lru.h>
11#include <linux/slab.h>
12#include <linux/mutex.h>
13#include <linux/memcontrol.h>
14
15#if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
16static LIST_HEAD(list_lrus);
17static DEFINE_MUTEX(list_lrus_mutex);
18
19static void list_lru_register(struct list_lru *lru)
20{
21 mutex_lock(&list_lrus_mutex);
22 list_add(&lru->list, &list_lrus);
23 mutex_unlock(&list_lrus_mutex);
24}
25
26static void list_lru_unregister(struct list_lru *lru)
27{
28 mutex_lock(&list_lrus_mutex);
29 list_del(&lru->list);
30 mutex_unlock(&list_lrus_mutex);
31}
32#else
33static void list_lru_register(struct list_lru *lru)
34{
35}
36
37static void list_lru_unregister(struct list_lru *lru)
38{
39}
40#endif /* CONFIG_MEMCG && !CONFIG_SLOB */
41
42#if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
43static inline bool list_lru_memcg_aware(struct list_lru *lru)
44{
45 /*
46 * This needs node 0 to be always present, even
47 * in the systems supporting sparse numa ids.
48 */
49 return !!lru->node[0].memcg_lrus;
50}
51
52static inline struct list_lru_one *
53list_lru_from_memcg_idx(struct list_lru_node *nlru, int idx)
54{
55 struct list_lru_memcg *memcg_lrus;
56 /*
57 * Either lock or RCU protects the array of per cgroup lists
58 * from relocation (see memcg_update_list_lru_node).
59 */
60 memcg_lrus = rcu_dereference_check(nlru->memcg_lrus,
61 lockdep_is_held(&nlru->lock));
62 if (memcg_lrus && idx >= 0)
63 return memcg_lrus->lru[idx];
64 return &nlru->lru;
65}
66
67static __always_inline struct mem_cgroup *mem_cgroup_from_kmem(void *ptr)
68{
69 struct page *page;
70
71 if (!memcg_kmem_enabled())
72 return NULL;
73 page = virt_to_head_page(ptr);
74 return page->mem_cgroup;
75}
76
77static inline struct list_lru_one *
78list_lru_from_kmem(struct list_lru_node *nlru, void *ptr)
79{
80 struct mem_cgroup *memcg;
81
82 if (!nlru->memcg_lrus)
83 return &nlru->lru;
84
85 memcg = mem_cgroup_from_kmem(ptr);
86 if (!memcg)
87 return &nlru->lru;
88
89 return list_lru_from_memcg_idx(nlru, memcg_cache_id(memcg));
90}
91#else
92static inline bool list_lru_memcg_aware(struct list_lru *lru)
93{
94 return false;
95}
96
97static inline struct list_lru_one *
98list_lru_from_memcg_idx(struct list_lru_node *nlru, int idx)
99{
100 return &nlru->lru;
101}
102
103static inline struct list_lru_one *
104list_lru_from_kmem(struct list_lru_node *nlru, void *ptr)
105{
106 return &nlru->lru;
107}
108#endif /* CONFIG_MEMCG && !CONFIG_SLOB */
109
110bool list_lru_add(struct list_lru *lru, struct list_head *item)
111{
112 int nid = page_to_nid(virt_to_page(item));
113 struct list_lru_node *nlru = &lru->node[nid];
114 struct list_lru_one *l;
115
116 spin_lock(&nlru->lock);
117 if (list_empty(item)) {
118 l = list_lru_from_kmem(nlru, item);
119 list_add_tail(item, &l->list);
120 l->nr_items++;
121 nlru->nr_items++;
122 spin_unlock(&nlru->lock);
123 return true;
124 }
125 spin_unlock(&nlru->lock);
126 return false;
127}
128EXPORT_SYMBOL_GPL(list_lru_add);
129
130bool list_lru_del(struct list_lru *lru, struct list_head *item)
131{
132 int nid = page_to_nid(virt_to_page(item));
133 struct list_lru_node *nlru = &lru->node[nid];
134 struct list_lru_one *l;
135
136 spin_lock(&nlru->lock);
137 if (!list_empty(item)) {
138 l = list_lru_from_kmem(nlru, item);
139 list_del_init(item);
140 l->nr_items--;
141 nlru->nr_items--;
142 spin_unlock(&nlru->lock);
143 return true;
144 }
145 spin_unlock(&nlru->lock);
146 return false;
147}
148EXPORT_SYMBOL_GPL(list_lru_del);
149
150void list_lru_isolate(struct list_lru_one *list, struct list_head *item)
151{
152 list_del_init(item);
153 list->nr_items--;
154}
155EXPORT_SYMBOL_GPL(list_lru_isolate);
156
157void list_lru_isolate_move(struct list_lru_one *list, struct list_head *item,
158 struct list_head *head)
159{
160 list_move(item, head);
161 list->nr_items--;
162}
163EXPORT_SYMBOL_GPL(list_lru_isolate_move);
164
165static unsigned long __list_lru_count_one(struct list_lru *lru,
166 int nid, int memcg_idx)
167{
168 struct list_lru_node *nlru = &lru->node[nid];
169 struct list_lru_one *l;
170 unsigned long count;
171
172 rcu_read_lock();
173 l = list_lru_from_memcg_idx(nlru, memcg_idx);
174 count = l->nr_items;
175 rcu_read_unlock();
176
177 return count;
178}
179
180unsigned long list_lru_count_one(struct list_lru *lru,
181 int nid, struct mem_cgroup *memcg)
182{
183 return __list_lru_count_one(lru, nid, memcg_cache_id(memcg));
184}
185EXPORT_SYMBOL_GPL(list_lru_count_one);
186
187unsigned long list_lru_count_node(struct list_lru *lru, int nid)
188{
189 struct list_lru_node *nlru;
190
191 nlru = &lru->node[nid];
192 return nlru->nr_items;
193}
194EXPORT_SYMBOL_GPL(list_lru_count_node);
195
196static unsigned long
197__list_lru_walk_one(struct list_lru *lru, int nid, int memcg_idx,
198 list_lru_walk_cb isolate, void *cb_arg,
199 unsigned long *nr_to_walk)
200{
201
202 struct list_lru_node *nlru = &lru->node[nid];
203 struct list_lru_one *l;
204 struct list_head *item, *n;
205 unsigned long isolated = 0;
206
207 spin_lock(&nlru->lock);
208 l = list_lru_from_memcg_idx(nlru, memcg_idx);
209restart:
210 list_for_each_safe(item, n, &l->list) {
211 enum lru_status ret;
212
213 /*
214 * decrement nr_to_walk first so that we don't livelock if we
215 * get stuck on large numbesr of LRU_RETRY items
216 */
217 if (!*nr_to_walk)
218 break;
219 --*nr_to_walk;
220
221 ret = isolate(item, l, &nlru->lock, cb_arg);
222 switch (ret) {
223 case LRU_REMOVED_RETRY:
224 assert_spin_locked(&nlru->lock);
225 /* fall through */
226 case LRU_REMOVED:
227 isolated++;
228 nlru->nr_items--;
229 /*
230 * If the lru lock has been dropped, our list
231 * traversal is now invalid and so we have to
232 * restart from scratch.
233 */
234 if (ret == LRU_REMOVED_RETRY)
235 goto restart;
236 break;
237 case LRU_ROTATE:
238 list_move_tail(item, &l->list);
239 break;
240 case LRU_SKIP:
241 break;
242 case LRU_RETRY:
243 /*
244 * The lru lock has been dropped, our list traversal is
245 * now invalid and so we have to restart from scratch.
246 */
247 assert_spin_locked(&nlru->lock);
248 goto restart;
249 default:
250 BUG();
251 }
252 }
253
254 spin_unlock(&nlru->lock);
255 return isolated;
256}
257
258unsigned long
259list_lru_walk_one(struct list_lru *lru, int nid, struct mem_cgroup *memcg,
260 list_lru_walk_cb isolate, void *cb_arg,
261 unsigned long *nr_to_walk)
262{
263 return __list_lru_walk_one(lru, nid, memcg_cache_id(memcg),
264 isolate, cb_arg, nr_to_walk);
265}
266EXPORT_SYMBOL_GPL(list_lru_walk_one);
267
268unsigned long list_lru_walk_node(struct list_lru *lru, int nid,
269 list_lru_walk_cb isolate, void *cb_arg,
270 unsigned long *nr_to_walk)
271{
272 long isolated = 0;
273 int memcg_idx;
274
275 isolated += __list_lru_walk_one(lru, nid, -1, isolate, cb_arg,
276 nr_to_walk);
277 if (*nr_to_walk > 0 && list_lru_memcg_aware(lru)) {
278 for_each_memcg_cache_index(memcg_idx) {
279 isolated += __list_lru_walk_one(lru, nid, memcg_idx,
280 isolate, cb_arg, nr_to_walk);
281 if (*nr_to_walk <= 0)
282 break;
283 }
284 }
285 return isolated;
286}
287EXPORT_SYMBOL_GPL(list_lru_walk_node);
288
289static void init_one_lru(struct list_lru_one *l)
290{
291 INIT_LIST_HEAD(&l->list);
292 l->nr_items = 0;
293}
294
295#if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
296static void __memcg_destroy_list_lru_node(struct list_lru_memcg *memcg_lrus,
297 int begin, int end)
298{
299 int i;
300
301 for (i = begin; i < end; i++)
302 kfree(memcg_lrus->lru[i]);
303}
304
305static int __memcg_init_list_lru_node(struct list_lru_memcg *memcg_lrus,
306 int begin, int end)
307{
308 int i;
309
310 for (i = begin; i < end; i++) {
311 struct list_lru_one *l;
312
313 l = kmalloc(sizeof(struct list_lru_one), GFP_KERNEL);
314 if (!l)
315 goto fail;
316
317 init_one_lru(l);
318 memcg_lrus->lru[i] = l;
319 }
320 return 0;
321fail:
322 __memcg_destroy_list_lru_node(memcg_lrus, begin, i - 1);
323 return -ENOMEM;
324}
325
326static int memcg_init_list_lru_node(struct list_lru_node *nlru)
327{
328 struct list_lru_memcg *memcg_lrus;
329 int size = memcg_nr_cache_ids;
330
331 memcg_lrus = kvmalloc(sizeof(*memcg_lrus) +
332 size * sizeof(void *), GFP_KERNEL);
333 if (!memcg_lrus)
334 return -ENOMEM;
335
336 if (__memcg_init_list_lru_node(memcg_lrus, 0, size)) {
337 kvfree(memcg_lrus);
338 return -ENOMEM;
339 }
340 RCU_INIT_POINTER(nlru->memcg_lrus, memcg_lrus);
341
342 return 0;
343}
344
345static void memcg_destroy_list_lru_node(struct list_lru_node *nlru)
346{
347 struct list_lru_memcg *memcg_lrus;
348 /*
349 * This is called when shrinker has already been unregistered,
350 * and nobody can use it. So, there is no need to use kvfree_rcu().
351 */
352 memcg_lrus = rcu_dereference_protected(nlru->memcg_lrus, true);
353 __memcg_destroy_list_lru_node(memcg_lrus, 0, memcg_nr_cache_ids);
354 kvfree(memcg_lrus);
355}
356
357static void kvfree_rcu(struct rcu_head *head)
358{
359 struct list_lru_memcg *mlru;
360
361 mlru = container_of(head, struct list_lru_memcg, rcu);
362 kvfree(mlru);
363}
364
365static int memcg_update_list_lru_node(struct list_lru_node *nlru,
366 int old_size, int new_size)
367{
368 struct list_lru_memcg *old, *new;
369
370 BUG_ON(old_size > new_size);
371
372 old = rcu_dereference_protected(nlru->memcg_lrus,
373 lockdep_is_held(&list_lrus_mutex));
374 new = kvmalloc(sizeof(*new) + new_size * sizeof(void *), GFP_KERNEL);
375 if (!new)
376 return -ENOMEM;
377
378 if (__memcg_init_list_lru_node(new, old_size, new_size)) {
379 kvfree(new);
380 return -ENOMEM;
381 }
382
383 memcpy(&new->lru, &old->lru, old_size * sizeof(void *));
384
385 /*
386 * The locking below allows readers that hold nlru->lock avoid taking
387 * rcu_read_lock (see list_lru_from_memcg_idx).
388 *
389 * Since list_lru_{add,del} may be called under an IRQ-safe lock,
390 * we have to use IRQ-safe primitives here to avoid deadlock.
391 */
392 spin_lock_irq(&nlru->lock);
393 rcu_assign_pointer(nlru->memcg_lrus, new);
394 spin_unlock_irq(&nlru->lock);
395
396 call_rcu(&old->rcu, kvfree_rcu);
397 return 0;
398}
399
400static void memcg_cancel_update_list_lru_node(struct list_lru_node *nlru,
401 int old_size, int new_size)
402{
403 struct list_lru_memcg *memcg_lrus;
404
405 memcg_lrus = rcu_dereference_protected(nlru->memcg_lrus,
406 lockdep_is_held(&list_lrus_mutex));
407 /* do not bother shrinking the array back to the old size, because we
408 * cannot handle allocation failures here */
409 __memcg_destroy_list_lru_node(memcg_lrus, old_size, new_size);
410}
411
412static int memcg_init_list_lru(struct list_lru *lru, bool memcg_aware)
413{
414 int i;
415
416 if (!memcg_aware)
417 return 0;
418
419 for_each_node(i) {
420 if (memcg_init_list_lru_node(&lru->node[i]))
421 goto fail;
422 }
423 return 0;
424fail:
425 for (i = i - 1; i >= 0; i--) {
426 if (!lru->node[i].memcg_lrus)
427 continue;
428 memcg_destroy_list_lru_node(&lru->node[i]);
429 }
430 return -ENOMEM;
431}
432
433static void memcg_destroy_list_lru(struct list_lru *lru)
434{
435 int i;
436
437 if (!list_lru_memcg_aware(lru))
438 return;
439
440 for_each_node(i)
441 memcg_destroy_list_lru_node(&lru->node[i]);
442}
443
444static int memcg_update_list_lru(struct list_lru *lru,
445 int old_size, int new_size)
446{
447 int i;
448
449 if (!list_lru_memcg_aware(lru))
450 return 0;
451
452 for_each_node(i) {
453 if (memcg_update_list_lru_node(&lru->node[i],
454 old_size, new_size))
455 goto fail;
456 }
457 return 0;
458fail:
459 for (i = i - 1; i >= 0; i--) {
460 if (!lru->node[i].memcg_lrus)
461 continue;
462
463 memcg_cancel_update_list_lru_node(&lru->node[i],
464 old_size, new_size);
465 }
466 return -ENOMEM;
467}
468
469static void memcg_cancel_update_list_lru(struct list_lru *lru,
470 int old_size, int new_size)
471{
472 int i;
473
474 if (!list_lru_memcg_aware(lru))
475 return;
476
477 for_each_node(i)
478 memcg_cancel_update_list_lru_node(&lru->node[i],
479 old_size, new_size);
480}
481
482int memcg_update_all_list_lrus(int new_size)
483{
484 int ret = 0;
485 struct list_lru *lru;
486 int old_size = memcg_nr_cache_ids;
487
488 mutex_lock(&list_lrus_mutex);
489 list_for_each_entry(lru, &list_lrus, list) {
490 ret = memcg_update_list_lru(lru, old_size, new_size);
491 if (ret)
492 goto fail;
493 }
494out:
495 mutex_unlock(&list_lrus_mutex);
496 return ret;
497fail:
498 list_for_each_entry_continue_reverse(lru, &list_lrus, list)
499 memcg_cancel_update_list_lru(lru, old_size, new_size);
500 goto out;
501}
502
503static void memcg_drain_list_lru_node(struct list_lru_node *nlru,
504 int src_idx, int dst_idx)
505{
506 struct list_lru_one *src, *dst;
507
508 /*
509 * Since list_lru_{add,del} may be called under an IRQ-safe lock,
510 * we have to use IRQ-safe primitives here to avoid deadlock.
511 */
512 spin_lock_irq(&nlru->lock);
513
514 src = list_lru_from_memcg_idx(nlru, src_idx);
515 dst = list_lru_from_memcg_idx(nlru, dst_idx);
516
517 list_splice_init(&src->list, &dst->list);
518 dst->nr_items += src->nr_items;
519 src->nr_items = 0;
520
521 spin_unlock_irq(&nlru->lock);
522}
523
524static void memcg_drain_list_lru(struct list_lru *lru,
525 int src_idx, int dst_idx)
526{
527 int i;
528
529 if (!list_lru_memcg_aware(lru))
530 return;
531
532 for_each_node(i)
533 memcg_drain_list_lru_node(&lru->node[i], src_idx, dst_idx);
534}
535
536void memcg_drain_all_list_lrus(int src_idx, int dst_idx)
537{
538 struct list_lru *lru;
539
540 mutex_lock(&list_lrus_mutex);
541 list_for_each_entry(lru, &list_lrus, list)
542 memcg_drain_list_lru(lru, src_idx, dst_idx);
543 mutex_unlock(&list_lrus_mutex);
544}
545#else
546static int memcg_init_list_lru(struct list_lru *lru, bool memcg_aware)
547{
548 return 0;
549}
550
551static void memcg_destroy_list_lru(struct list_lru *lru)
552{
553}
554#endif /* CONFIG_MEMCG && !CONFIG_SLOB */
555
556int __list_lru_init(struct list_lru *lru, bool memcg_aware,
557 struct lock_class_key *key)
558{
559 int i;
560 size_t size = sizeof(*lru->node) * nr_node_ids;
561 int err = -ENOMEM;
562
563 memcg_get_cache_ids();
564
565 lru->node = kzalloc(size, GFP_KERNEL);
566 if (!lru->node)
567 goto out;
568
569 for_each_node(i) {
570 spin_lock_init(&lru->node[i].lock);
571 if (key)
572 lockdep_set_class(&lru->node[i].lock, key);
573 init_one_lru(&lru->node[i].lru);
574 }
575
576 err = memcg_init_list_lru(lru, memcg_aware);
577 if (err) {
578 kfree(lru->node);
579 /* Do this so a list_lru_destroy() doesn't crash: */
580 lru->node = NULL;
581 goto out;
582 }
583
584 list_lru_register(lru);
585out:
586 memcg_put_cache_ids();
587 return err;
588}
589EXPORT_SYMBOL_GPL(__list_lru_init);
590
591void list_lru_destroy(struct list_lru *lru)
592{
593 /* Already destroyed or not yet initialized? */
594 if (!lru->node)
595 return;
596
597 memcg_get_cache_ids();
598
599 list_lru_unregister(lru);
600
601 memcg_destroy_list_lru(lru);
602 kfree(lru->node);
603 lru->node = NULL;
604
605 memcg_put_cache_ids();
606}
607EXPORT_SYMBOL_GPL(list_lru_destroy);
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (c) 2013 Red Hat, Inc. and Parallels Inc. All rights reserved.
4 * Authors: David Chinner and Glauber Costa
5 *
6 * Generic LRU infrastructure
7 */
8#include <linux/kernel.h>
9#include <linux/module.h>
10#include <linux/mm.h>
11#include <linux/list_lru.h>
12#include <linux/slab.h>
13#include <linux/mutex.h>
14#include <linux/memcontrol.h>
15#include "slab.h"
16
17#ifdef CONFIG_MEMCG_KMEM
18static LIST_HEAD(list_lrus);
19static DEFINE_MUTEX(list_lrus_mutex);
20
21static void list_lru_register(struct list_lru *lru)
22{
23 mutex_lock(&list_lrus_mutex);
24 list_add(&lru->list, &list_lrus);
25 mutex_unlock(&list_lrus_mutex);
26}
27
28static void list_lru_unregister(struct list_lru *lru)
29{
30 mutex_lock(&list_lrus_mutex);
31 list_del(&lru->list);
32 mutex_unlock(&list_lrus_mutex);
33}
34
35static int lru_shrinker_id(struct list_lru *lru)
36{
37 return lru->shrinker_id;
38}
39
40static inline bool list_lru_memcg_aware(struct list_lru *lru)
41{
42 return lru->memcg_aware;
43}
44
45static inline struct list_lru_one *
46list_lru_from_memcg_idx(struct list_lru_node *nlru, int idx)
47{
48 struct list_lru_memcg *memcg_lrus;
49 /*
50 * Either lock or RCU protects the array of per cgroup lists
51 * from relocation (see memcg_update_list_lru_node).
52 */
53 memcg_lrus = rcu_dereference_check(nlru->memcg_lrus,
54 lockdep_is_held(&nlru->lock));
55 if (memcg_lrus && idx >= 0)
56 return memcg_lrus->lru[idx];
57 return &nlru->lru;
58}
59
60static inline struct list_lru_one *
61list_lru_from_kmem(struct list_lru_node *nlru, void *ptr,
62 struct mem_cgroup **memcg_ptr)
63{
64 struct list_lru_one *l = &nlru->lru;
65 struct mem_cgroup *memcg = NULL;
66
67 if (!nlru->memcg_lrus)
68 goto out;
69
70 memcg = mem_cgroup_from_obj(ptr);
71 if (!memcg)
72 goto out;
73
74 l = list_lru_from_memcg_idx(nlru, memcg_cache_id(memcg));
75out:
76 if (memcg_ptr)
77 *memcg_ptr = memcg;
78 return l;
79}
80#else
81static void list_lru_register(struct list_lru *lru)
82{
83}
84
85static void list_lru_unregister(struct list_lru *lru)
86{
87}
88
89static int lru_shrinker_id(struct list_lru *lru)
90{
91 return -1;
92}
93
94static inline bool list_lru_memcg_aware(struct list_lru *lru)
95{
96 return false;
97}
98
99static inline struct list_lru_one *
100list_lru_from_memcg_idx(struct list_lru_node *nlru, int idx)
101{
102 return &nlru->lru;
103}
104
105static inline struct list_lru_one *
106list_lru_from_kmem(struct list_lru_node *nlru, void *ptr,
107 struct mem_cgroup **memcg_ptr)
108{
109 if (memcg_ptr)
110 *memcg_ptr = NULL;
111 return &nlru->lru;
112}
113#endif /* CONFIG_MEMCG_KMEM */
114
115bool list_lru_add(struct list_lru *lru, struct list_head *item)
116{
117 int nid = page_to_nid(virt_to_page(item));
118 struct list_lru_node *nlru = &lru->node[nid];
119 struct mem_cgroup *memcg;
120 struct list_lru_one *l;
121
122 spin_lock(&nlru->lock);
123 if (list_empty(item)) {
124 l = list_lru_from_kmem(nlru, item, &memcg);
125 list_add_tail(item, &l->list);
126 /* Set shrinker bit if the first element was added */
127 if (!l->nr_items++)
128 memcg_set_shrinker_bit(memcg, nid,
129 lru_shrinker_id(lru));
130 nlru->nr_items++;
131 spin_unlock(&nlru->lock);
132 return true;
133 }
134 spin_unlock(&nlru->lock);
135 return false;
136}
137EXPORT_SYMBOL_GPL(list_lru_add);
138
139bool list_lru_del(struct list_lru *lru, struct list_head *item)
140{
141 int nid = page_to_nid(virt_to_page(item));
142 struct list_lru_node *nlru = &lru->node[nid];
143 struct list_lru_one *l;
144
145 spin_lock(&nlru->lock);
146 if (!list_empty(item)) {
147 l = list_lru_from_kmem(nlru, item, NULL);
148 list_del_init(item);
149 l->nr_items--;
150 nlru->nr_items--;
151 spin_unlock(&nlru->lock);
152 return true;
153 }
154 spin_unlock(&nlru->lock);
155 return false;
156}
157EXPORT_SYMBOL_GPL(list_lru_del);
158
159void list_lru_isolate(struct list_lru_one *list, struct list_head *item)
160{
161 list_del_init(item);
162 list->nr_items--;
163}
164EXPORT_SYMBOL_GPL(list_lru_isolate);
165
166void list_lru_isolate_move(struct list_lru_one *list, struct list_head *item,
167 struct list_head *head)
168{
169 list_move(item, head);
170 list->nr_items--;
171}
172EXPORT_SYMBOL_GPL(list_lru_isolate_move);
173
174unsigned long list_lru_count_one(struct list_lru *lru,
175 int nid, struct mem_cgroup *memcg)
176{
177 struct list_lru_node *nlru = &lru->node[nid];
178 struct list_lru_one *l;
179 unsigned long count;
180
181 rcu_read_lock();
182 l = list_lru_from_memcg_idx(nlru, memcg_cache_id(memcg));
183 count = READ_ONCE(l->nr_items);
184 rcu_read_unlock();
185
186 return count;
187}
188EXPORT_SYMBOL_GPL(list_lru_count_one);
189
190unsigned long list_lru_count_node(struct list_lru *lru, int nid)
191{
192 struct list_lru_node *nlru;
193
194 nlru = &lru->node[nid];
195 return nlru->nr_items;
196}
197EXPORT_SYMBOL_GPL(list_lru_count_node);
198
199static unsigned long
200__list_lru_walk_one(struct list_lru_node *nlru, int memcg_idx,
201 list_lru_walk_cb isolate, void *cb_arg,
202 unsigned long *nr_to_walk)
203{
204
205 struct list_lru_one *l;
206 struct list_head *item, *n;
207 unsigned long isolated = 0;
208
209 l = list_lru_from_memcg_idx(nlru, memcg_idx);
210restart:
211 list_for_each_safe(item, n, &l->list) {
212 enum lru_status ret;
213
214 /*
215 * decrement nr_to_walk first so that we don't livelock if we
216 * get stuck on large numbers of LRU_RETRY items
217 */
218 if (!*nr_to_walk)
219 break;
220 --*nr_to_walk;
221
222 ret = isolate(item, l, &nlru->lock, cb_arg);
223 switch (ret) {
224 case LRU_REMOVED_RETRY:
225 assert_spin_locked(&nlru->lock);
226 fallthrough;
227 case LRU_REMOVED:
228 isolated++;
229 nlru->nr_items--;
230 /*
231 * If the lru lock has been dropped, our list
232 * traversal is now invalid and so we have to
233 * restart from scratch.
234 */
235 if (ret == LRU_REMOVED_RETRY)
236 goto restart;
237 break;
238 case LRU_ROTATE:
239 list_move_tail(item, &l->list);
240 break;
241 case LRU_SKIP:
242 break;
243 case LRU_RETRY:
244 /*
245 * The lru lock has been dropped, our list traversal is
246 * now invalid and so we have to restart from scratch.
247 */
248 assert_spin_locked(&nlru->lock);
249 goto restart;
250 default:
251 BUG();
252 }
253 }
254 return isolated;
255}
256
257unsigned long
258list_lru_walk_one(struct list_lru *lru, int nid, struct mem_cgroup *memcg,
259 list_lru_walk_cb isolate, void *cb_arg,
260 unsigned long *nr_to_walk)
261{
262 struct list_lru_node *nlru = &lru->node[nid];
263 unsigned long ret;
264
265 spin_lock(&nlru->lock);
266 ret = __list_lru_walk_one(nlru, memcg_cache_id(memcg), isolate, cb_arg,
267 nr_to_walk);
268 spin_unlock(&nlru->lock);
269 return ret;
270}
271EXPORT_SYMBOL_GPL(list_lru_walk_one);
272
273unsigned long
274list_lru_walk_one_irq(struct list_lru *lru, int nid, struct mem_cgroup *memcg,
275 list_lru_walk_cb isolate, void *cb_arg,
276 unsigned long *nr_to_walk)
277{
278 struct list_lru_node *nlru = &lru->node[nid];
279 unsigned long ret;
280
281 spin_lock_irq(&nlru->lock);
282 ret = __list_lru_walk_one(nlru, memcg_cache_id(memcg), isolate, cb_arg,
283 nr_to_walk);
284 spin_unlock_irq(&nlru->lock);
285 return ret;
286}
287
288unsigned long list_lru_walk_node(struct list_lru *lru, int nid,
289 list_lru_walk_cb isolate, void *cb_arg,
290 unsigned long *nr_to_walk)
291{
292 long isolated = 0;
293 int memcg_idx;
294
295 isolated += list_lru_walk_one(lru, nid, NULL, isolate, cb_arg,
296 nr_to_walk);
297 if (*nr_to_walk > 0 && list_lru_memcg_aware(lru)) {
298 for_each_memcg_cache_index(memcg_idx) {
299 struct list_lru_node *nlru = &lru->node[nid];
300
301 spin_lock(&nlru->lock);
302 isolated += __list_lru_walk_one(nlru, memcg_idx,
303 isolate, cb_arg,
304 nr_to_walk);
305 spin_unlock(&nlru->lock);
306
307 if (*nr_to_walk <= 0)
308 break;
309 }
310 }
311 return isolated;
312}
313EXPORT_SYMBOL_GPL(list_lru_walk_node);
314
315static void init_one_lru(struct list_lru_one *l)
316{
317 INIT_LIST_HEAD(&l->list);
318 l->nr_items = 0;
319}
320
321#ifdef CONFIG_MEMCG_KMEM
322static void __memcg_destroy_list_lru_node(struct list_lru_memcg *memcg_lrus,
323 int begin, int end)
324{
325 int i;
326
327 for (i = begin; i < end; i++)
328 kfree(memcg_lrus->lru[i]);
329}
330
331static int __memcg_init_list_lru_node(struct list_lru_memcg *memcg_lrus,
332 int begin, int end)
333{
334 int i;
335
336 for (i = begin; i < end; i++) {
337 struct list_lru_one *l;
338
339 l = kmalloc(sizeof(struct list_lru_one), GFP_KERNEL);
340 if (!l)
341 goto fail;
342
343 init_one_lru(l);
344 memcg_lrus->lru[i] = l;
345 }
346 return 0;
347fail:
348 __memcg_destroy_list_lru_node(memcg_lrus, begin, i);
349 return -ENOMEM;
350}
351
352static int memcg_init_list_lru_node(struct list_lru_node *nlru)
353{
354 struct list_lru_memcg *memcg_lrus;
355 int size = memcg_nr_cache_ids;
356
357 memcg_lrus = kvmalloc(sizeof(*memcg_lrus) +
358 size * sizeof(void *), GFP_KERNEL);
359 if (!memcg_lrus)
360 return -ENOMEM;
361
362 if (__memcg_init_list_lru_node(memcg_lrus, 0, size)) {
363 kvfree(memcg_lrus);
364 return -ENOMEM;
365 }
366 RCU_INIT_POINTER(nlru->memcg_lrus, memcg_lrus);
367
368 return 0;
369}
370
371static void memcg_destroy_list_lru_node(struct list_lru_node *nlru)
372{
373 struct list_lru_memcg *memcg_lrus;
374 /*
375 * This is called when shrinker has already been unregistered,
376 * and nobody can use it. So, there is no need to use kvfree_rcu_local().
377 */
378 memcg_lrus = rcu_dereference_protected(nlru->memcg_lrus, true);
379 __memcg_destroy_list_lru_node(memcg_lrus, 0, memcg_nr_cache_ids);
380 kvfree(memcg_lrus);
381}
382
383static void kvfree_rcu_local(struct rcu_head *head)
384{
385 struct list_lru_memcg *mlru;
386
387 mlru = container_of(head, struct list_lru_memcg, rcu);
388 kvfree(mlru);
389}
390
391static int memcg_update_list_lru_node(struct list_lru_node *nlru,
392 int old_size, int new_size)
393{
394 struct list_lru_memcg *old, *new;
395
396 BUG_ON(old_size > new_size);
397
398 old = rcu_dereference_protected(nlru->memcg_lrus,
399 lockdep_is_held(&list_lrus_mutex));
400 new = kvmalloc(sizeof(*new) + new_size * sizeof(void *), GFP_KERNEL);
401 if (!new)
402 return -ENOMEM;
403
404 if (__memcg_init_list_lru_node(new, old_size, new_size)) {
405 kvfree(new);
406 return -ENOMEM;
407 }
408
409 memcpy(&new->lru, &old->lru, old_size * sizeof(void *));
410
411 /*
412 * The locking below allows readers that hold nlru->lock avoid taking
413 * rcu_read_lock (see list_lru_from_memcg_idx).
414 *
415 * Since list_lru_{add,del} may be called under an IRQ-safe lock,
416 * we have to use IRQ-safe primitives here to avoid deadlock.
417 */
418 spin_lock_irq(&nlru->lock);
419 rcu_assign_pointer(nlru->memcg_lrus, new);
420 spin_unlock_irq(&nlru->lock);
421
422 call_rcu(&old->rcu, kvfree_rcu_local);
423 return 0;
424}
425
426static void memcg_cancel_update_list_lru_node(struct list_lru_node *nlru,
427 int old_size, int new_size)
428{
429 struct list_lru_memcg *memcg_lrus;
430
431 memcg_lrus = rcu_dereference_protected(nlru->memcg_lrus,
432 lockdep_is_held(&list_lrus_mutex));
433 /* do not bother shrinking the array back to the old size, because we
434 * cannot handle allocation failures here */
435 __memcg_destroy_list_lru_node(memcg_lrus, old_size, new_size);
436}
437
438static int memcg_init_list_lru(struct list_lru *lru, bool memcg_aware)
439{
440 int i;
441
442 lru->memcg_aware = memcg_aware;
443
444 if (!memcg_aware)
445 return 0;
446
447 for_each_node(i) {
448 if (memcg_init_list_lru_node(&lru->node[i]))
449 goto fail;
450 }
451 return 0;
452fail:
453 for (i = i - 1; i >= 0; i--) {
454 if (!lru->node[i].memcg_lrus)
455 continue;
456 memcg_destroy_list_lru_node(&lru->node[i]);
457 }
458 return -ENOMEM;
459}
460
461static void memcg_destroy_list_lru(struct list_lru *lru)
462{
463 int i;
464
465 if (!list_lru_memcg_aware(lru))
466 return;
467
468 for_each_node(i)
469 memcg_destroy_list_lru_node(&lru->node[i]);
470}
471
472static int memcg_update_list_lru(struct list_lru *lru,
473 int old_size, int new_size)
474{
475 int i;
476
477 if (!list_lru_memcg_aware(lru))
478 return 0;
479
480 for_each_node(i) {
481 if (memcg_update_list_lru_node(&lru->node[i],
482 old_size, new_size))
483 goto fail;
484 }
485 return 0;
486fail:
487 for (i = i - 1; i >= 0; i--) {
488 if (!lru->node[i].memcg_lrus)
489 continue;
490
491 memcg_cancel_update_list_lru_node(&lru->node[i],
492 old_size, new_size);
493 }
494 return -ENOMEM;
495}
496
497static void memcg_cancel_update_list_lru(struct list_lru *lru,
498 int old_size, int new_size)
499{
500 int i;
501
502 if (!list_lru_memcg_aware(lru))
503 return;
504
505 for_each_node(i)
506 memcg_cancel_update_list_lru_node(&lru->node[i],
507 old_size, new_size);
508}
509
510int memcg_update_all_list_lrus(int new_size)
511{
512 int ret = 0;
513 struct list_lru *lru;
514 int old_size = memcg_nr_cache_ids;
515
516 mutex_lock(&list_lrus_mutex);
517 list_for_each_entry(lru, &list_lrus, list) {
518 ret = memcg_update_list_lru(lru, old_size, new_size);
519 if (ret)
520 goto fail;
521 }
522out:
523 mutex_unlock(&list_lrus_mutex);
524 return ret;
525fail:
526 list_for_each_entry_continue_reverse(lru, &list_lrus, list)
527 memcg_cancel_update_list_lru(lru, old_size, new_size);
528 goto out;
529}
530
531static void memcg_drain_list_lru_node(struct list_lru *lru, int nid,
532 int src_idx, struct mem_cgroup *dst_memcg)
533{
534 struct list_lru_node *nlru = &lru->node[nid];
535 int dst_idx = dst_memcg->kmemcg_id;
536 struct list_lru_one *src, *dst;
537 bool set;
538
539 /*
540 * Since list_lru_{add,del} may be called under an IRQ-safe lock,
541 * we have to use IRQ-safe primitives here to avoid deadlock.
542 */
543 spin_lock_irq(&nlru->lock);
544
545 src = list_lru_from_memcg_idx(nlru, src_idx);
546 dst = list_lru_from_memcg_idx(nlru, dst_idx);
547
548 list_splice_init(&src->list, &dst->list);
549 set = (!dst->nr_items && src->nr_items);
550 dst->nr_items += src->nr_items;
551 if (set)
552 memcg_set_shrinker_bit(dst_memcg, nid, lru_shrinker_id(lru));
553 src->nr_items = 0;
554
555 spin_unlock_irq(&nlru->lock);
556}
557
558static void memcg_drain_list_lru(struct list_lru *lru,
559 int src_idx, struct mem_cgroup *dst_memcg)
560{
561 int i;
562
563 if (!list_lru_memcg_aware(lru))
564 return;
565
566 for_each_node(i)
567 memcg_drain_list_lru_node(lru, i, src_idx, dst_memcg);
568}
569
570void memcg_drain_all_list_lrus(int src_idx, struct mem_cgroup *dst_memcg)
571{
572 struct list_lru *lru;
573
574 mutex_lock(&list_lrus_mutex);
575 list_for_each_entry(lru, &list_lrus, list)
576 memcg_drain_list_lru(lru, src_idx, dst_memcg);
577 mutex_unlock(&list_lrus_mutex);
578}
579#else
580static int memcg_init_list_lru(struct list_lru *lru, bool memcg_aware)
581{
582 return 0;
583}
584
585static void memcg_destroy_list_lru(struct list_lru *lru)
586{
587}
588#endif /* CONFIG_MEMCG_KMEM */
589
590int __list_lru_init(struct list_lru *lru, bool memcg_aware,
591 struct lock_class_key *key, struct shrinker *shrinker)
592{
593 int i;
594 int err = -ENOMEM;
595
596#ifdef CONFIG_MEMCG_KMEM
597 if (shrinker)
598 lru->shrinker_id = shrinker->id;
599 else
600 lru->shrinker_id = -1;
601#endif
602 memcg_get_cache_ids();
603
604 lru->node = kcalloc(nr_node_ids, sizeof(*lru->node), GFP_KERNEL);
605 if (!lru->node)
606 goto out;
607
608 for_each_node(i) {
609 spin_lock_init(&lru->node[i].lock);
610 if (key)
611 lockdep_set_class(&lru->node[i].lock, key);
612 init_one_lru(&lru->node[i].lru);
613 }
614
615 err = memcg_init_list_lru(lru, memcg_aware);
616 if (err) {
617 kfree(lru->node);
618 /* Do this so a list_lru_destroy() doesn't crash: */
619 lru->node = NULL;
620 goto out;
621 }
622
623 list_lru_register(lru);
624out:
625 memcg_put_cache_ids();
626 return err;
627}
628EXPORT_SYMBOL_GPL(__list_lru_init);
629
630void list_lru_destroy(struct list_lru *lru)
631{
632 /* Already destroyed or not yet initialized? */
633 if (!lru->node)
634 return;
635
636 memcg_get_cache_ids();
637
638 list_lru_unregister(lru);
639
640 memcg_destroy_list_lru(lru);
641 kfree(lru->node);
642 lru->node = NULL;
643
644#ifdef CONFIG_MEMCG_KMEM
645 lru->shrinker_id = -1;
646#endif
647 memcg_put_cache_ids();
648}
649EXPORT_SYMBOL_GPL(list_lru_destroy);