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#include "internal.h"
17
18#ifdef CONFIG_MEMCG
19static LIST_HEAD(memcg_list_lrus);
20static DEFINE_MUTEX(list_lrus_mutex);
21
22static inline bool list_lru_memcg_aware(struct list_lru *lru)
23{
24 return lru->memcg_aware;
25}
26
27static void list_lru_register(struct list_lru *lru)
28{
29 if (!list_lru_memcg_aware(lru))
30 return;
31
32 mutex_lock(&list_lrus_mutex);
33 list_add(&lru->list, &memcg_list_lrus);
34 mutex_unlock(&list_lrus_mutex);
35}
36
37static void list_lru_unregister(struct list_lru *lru)
38{
39 if (!list_lru_memcg_aware(lru))
40 return;
41
42 mutex_lock(&list_lrus_mutex);
43 list_del(&lru->list);
44 mutex_unlock(&list_lrus_mutex);
45}
46
47static int lru_shrinker_id(struct list_lru *lru)
48{
49 return lru->shrinker_id;
50}
51
52static inline struct list_lru_one *
53list_lru_from_memcg_idx(struct list_lru *lru, int nid, int idx)
54{
55 if (list_lru_memcg_aware(lru) && idx >= 0) {
56 struct list_lru_memcg *mlru = xa_load(&lru->xa, idx);
57
58 return mlru ? &mlru->node[nid] : NULL;
59 }
60 return &lru->node[nid].lru;
61}
62
63static inline struct list_lru_one *
64lock_list_lru_of_memcg(struct list_lru *lru, int nid, struct mem_cgroup *memcg,
65 bool irq, bool skip_empty)
66{
67 struct list_lru_one *l;
68 long nr_items;
69
70 rcu_read_lock();
71again:
72 l = list_lru_from_memcg_idx(lru, nid, memcg_kmem_id(memcg));
73 if (likely(l)) {
74 if (irq)
75 spin_lock_irq(&l->lock);
76 else
77 spin_lock(&l->lock);
78 nr_items = READ_ONCE(l->nr_items);
79 if (likely(nr_items != LONG_MIN)) {
80 rcu_read_unlock();
81 return l;
82 }
83 if (irq)
84 spin_unlock_irq(&l->lock);
85 else
86 spin_unlock(&l->lock);
87 }
88 /*
89 * Caller may simply bail out if raced with reparenting or
90 * may iterate through the list_lru and expect empty slots.
91 */
92 if (skip_empty) {
93 rcu_read_unlock();
94 return NULL;
95 }
96 VM_WARN_ON(!css_is_dying(&memcg->css));
97 memcg = parent_mem_cgroup(memcg);
98 goto again;
99}
100
101static inline void unlock_list_lru(struct list_lru_one *l, bool irq_off)
102{
103 if (irq_off)
104 spin_unlock_irq(&l->lock);
105 else
106 spin_unlock(&l->lock);
107}
108#else
109static void list_lru_register(struct list_lru *lru)
110{
111}
112
113static void list_lru_unregister(struct list_lru *lru)
114{
115}
116
117static int lru_shrinker_id(struct list_lru *lru)
118{
119 return -1;
120}
121
122static inline bool list_lru_memcg_aware(struct list_lru *lru)
123{
124 return false;
125}
126
127static inline struct list_lru_one *
128list_lru_from_memcg_idx(struct list_lru *lru, int nid, int idx)
129{
130 return &lru->node[nid].lru;
131}
132
133static inline struct list_lru_one *
134lock_list_lru_of_memcg(struct list_lru *lru, int nid, struct mem_cgroup *memcg,
135 bool irq, bool skip_empty)
136{
137 struct list_lru_one *l = &lru->node[nid].lru;
138
139 if (irq)
140 spin_lock_irq(&l->lock);
141 else
142 spin_lock(&l->lock);
143
144 return l;
145}
146
147static inline void unlock_list_lru(struct list_lru_one *l, bool irq_off)
148{
149 if (irq_off)
150 spin_unlock_irq(&l->lock);
151 else
152 spin_unlock(&l->lock);
153}
154#endif /* CONFIG_MEMCG */
155
156/* The caller must ensure the memcg lifetime. */
157bool list_lru_add(struct list_lru *lru, struct list_head *item, int nid,
158 struct mem_cgroup *memcg)
159{
160 struct list_lru_node *nlru = &lru->node[nid];
161 struct list_lru_one *l;
162
163 l = lock_list_lru_of_memcg(lru, nid, memcg, false, false);
164 if (!l)
165 return false;
166 if (list_empty(item)) {
167 list_add_tail(item, &l->list);
168 /* Set shrinker bit if the first element was added */
169 if (!l->nr_items++)
170 set_shrinker_bit(memcg, nid, lru_shrinker_id(lru));
171 unlock_list_lru(l, false);
172 atomic_long_inc(&nlru->nr_items);
173 return true;
174 }
175 unlock_list_lru(l, false);
176 return false;
177}
178
179bool list_lru_add_obj(struct list_lru *lru, struct list_head *item)
180{
181 bool ret;
182 int nid = page_to_nid(virt_to_page(item));
183
184 if (list_lru_memcg_aware(lru)) {
185 rcu_read_lock();
186 ret = list_lru_add(lru, item, nid, mem_cgroup_from_slab_obj(item));
187 rcu_read_unlock();
188 } else {
189 ret = list_lru_add(lru, item, nid, NULL);
190 }
191
192 return ret;
193}
194EXPORT_SYMBOL_GPL(list_lru_add_obj);
195
196/* The caller must ensure the memcg lifetime. */
197bool list_lru_del(struct list_lru *lru, struct list_head *item, int nid,
198 struct mem_cgroup *memcg)
199{
200 struct list_lru_node *nlru = &lru->node[nid];
201 struct list_lru_one *l;
202 l = lock_list_lru_of_memcg(lru, nid, memcg, false, false);
203 if (!l)
204 return false;
205 if (!list_empty(item)) {
206 list_del_init(item);
207 l->nr_items--;
208 unlock_list_lru(l, false);
209 atomic_long_dec(&nlru->nr_items);
210 return true;
211 }
212 unlock_list_lru(l, false);
213 return false;
214}
215
216bool list_lru_del_obj(struct list_lru *lru, struct list_head *item)
217{
218 bool ret;
219 int nid = page_to_nid(virt_to_page(item));
220
221 if (list_lru_memcg_aware(lru)) {
222 rcu_read_lock();
223 ret = list_lru_del(lru, item, nid, mem_cgroup_from_slab_obj(item));
224 rcu_read_unlock();
225 } else {
226 ret = list_lru_del(lru, item, nid, NULL);
227 }
228
229 return ret;
230}
231EXPORT_SYMBOL_GPL(list_lru_del_obj);
232
233void list_lru_isolate(struct list_lru_one *list, struct list_head *item)
234{
235 list_del_init(item);
236 list->nr_items--;
237}
238EXPORT_SYMBOL_GPL(list_lru_isolate);
239
240void list_lru_isolate_move(struct list_lru_one *list, struct list_head *item,
241 struct list_head *head)
242{
243 list_move(item, head);
244 list->nr_items--;
245}
246EXPORT_SYMBOL_GPL(list_lru_isolate_move);
247
248unsigned long list_lru_count_one(struct list_lru *lru,
249 int nid, struct mem_cgroup *memcg)
250{
251 struct list_lru_one *l;
252 long count;
253
254 rcu_read_lock();
255 l = list_lru_from_memcg_idx(lru, nid, memcg_kmem_id(memcg));
256 count = l ? READ_ONCE(l->nr_items) : 0;
257 rcu_read_unlock();
258
259 if (unlikely(count < 0))
260 count = 0;
261
262 return count;
263}
264EXPORT_SYMBOL_GPL(list_lru_count_one);
265
266unsigned long list_lru_count_node(struct list_lru *lru, int nid)
267{
268 struct list_lru_node *nlru;
269
270 nlru = &lru->node[nid];
271 return atomic_long_read(&nlru->nr_items);
272}
273EXPORT_SYMBOL_GPL(list_lru_count_node);
274
275static unsigned long
276__list_lru_walk_one(struct list_lru *lru, int nid, struct mem_cgroup *memcg,
277 list_lru_walk_cb isolate, void *cb_arg,
278 unsigned long *nr_to_walk, bool irq_off)
279{
280 struct list_lru_node *nlru = &lru->node[nid];
281 struct list_lru_one *l = NULL;
282 struct list_head *item, *n;
283 unsigned long isolated = 0;
284
285restart:
286 l = lock_list_lru_of_memcg(lru, nid, memcg, irq_off, true);
287 if (!l)
288 return isolated;
289 list_for_each_safe(item, n, &l->list) {
290 enum lru_status ret;
291
292 /*
293 * decrement nr_to_walk first so that we don't livelock if we
294 * get stuck on large numbers of LRU_RETRY items
295 */
296 if (!*nr_to_walk)
297 break;
298 --*nr_to_walk;
299
300 ret = isolate(item, l, cb_arg);
301 switch (ret) {
302 /*
303 * LRU_RETRY, LRU_REMOVED_RETRY and LRU_STOP will drop the lru
304 * lock. List traversal will have to restart from scratch.
305 */
306 case LRU_RETRY:
307 goto restart;
308 case LRU_REMOVED_RETRY:
309 fallthrough;
310 case LRU_REMOVED:
311 isolated++;
312 atomic_long_dec(&nlru->nr_items);
313 if (ret == LRU_REMOVED_RETRY)
314 goto restart;
315 break;
316 case LRU_ROTATE:
317 list_move_tail(item, &l->list);
318 break;
319 case LRU_SKIP:
320 break;
321 case LRU_STOP:
322 goto out;
323 default:
324 BUG();
325 }
326 }
327 unlock_list_lru(l, irq_off);
328out:
329 return isolated;
330}
331
332unsigned long
333list_lru_walk_one(struct list_lru *lru, int nid, struct mem_cgroup *memcg,
334 list_lru_walk_cb isolate, void *cb_arg,
335 unsigned long *nr_to_walk)
336{
337 return __list_lru_walk_one(lru, nid, memcg, isolate,
338 cb_arg, nr_to_walk, false);
339}
340EXPORT_SYMBOL_GPL(list_lru_walk_one);
341
342unsigned long
343list_lru_walk_one_irq(struct list_lru *lru, int nid, struct mem_cgroup *memcg,
344 list_lru_walk_cb isolate, void *cb_arg,
345 unsigned long *nr_to_walk)
346{
347 return __list_lru_walk_one(lru, nid, memcg, isolate,
348 cb_arg, nr_to_walk, true);
349}
350
351unsigned long list_lru_walk_node(struct list_lru *lru, int nid,
352 list_lru_walk_cb isolate, void *cb_arg,
353 unsigned long *nr_to_walk)
354{
355 long isolated = 0;
356
357 isolated += list_lru_walk_one(lru, nid, NULL, isolate, cb_arg,
358 nr_to_walk);
359
360#ifdef CONFIG_MEMCG
361 if (*nr_to_walk > 0 && list_lru_memcg_aware(lru)) {
362 struct list_lru_memcg *mlru;
363 struct mem_cgroup *memcg;
364 unsigned long index;
365
366 xa_for_each(&lru->xa, index, mlru) {
367 rcu_read_lock();
368 memcg = mem_cgroup_from_id(index);
369 if (!mem_cgroup_tryget(memcg)) {
370 rcu_read_unlock();
371 continue;
372 }
373 rcu_read_unlock();
374 isolated += __list_lru_walk_one(lru, nid, memcg,
375 isolate, cb_arg,
376 nr_to_walk, false);
377 mem_cgroup_put(memcg);
378
379 if (*nr_to_walk <= 0)
380 break;
381 }
382 }
383#endif
384
385 return isolated;
386}
387EXPORT_SYMBOL_GPL(list_lru_walk_node);
388
389static void init_one_lru(struct list_lru *lru, struct list_lru_one *l)
390{
391 INIT_LIST_HEAD(&l->list);
392 spin_lock_init(&l->lock);
393 l->nr_items = 0;
394#ifdef CONFIG_LOCKDEP
395 if (lru->key)
396 lockdep_set_class(&l->lock, lru->key);
397#endif
398}
399
400#ifdef CONFIG_MEMCG
401static struct list_lru_memcg *memcg_init_list_lru_one(struct list_lru *lru, gfp_t gfp)
402{
403 int nid;
404 struct list_lru_memcg *mlru;
405
406 mlru = kmalloc(struct_size(mlru, node, nr_node_ids), gfp);
407 if (!mlru)
408 return NULL;
409
410 for_each_node(nid)
411 init_one_lru(lru, &mlru->node[nid]);
412
413 return mlru;
414}
415
416static inline void memcg_init_list_lru(struct list_lru *lru, bool memcg_aware)
417{
418 if (memcg_aware)
419 xa_init_flags(&lru->xa, XA_FLAGS_LOCK_IRQ);
420 lru->memcg_aware = memcg_aware;
421}
422
423static void memcg_destroy_list_lru(struct list_lru *lru)
424{
425 XA_STATE(xas, &lru->xa, 0);
426 struct list_lru_memcg *mlru;
427
428 if (!list_lru_memcg_aware(lru))
429 return;
430
431 xas_lock_irq(&xas);
432 xas_for_each(&xas, mlru, ULONG_MAX) {
433 kfree(mlru);
434 xas_store(&xas, NULL);
435 }
436 xas_unlock_irq(&xas);
437}
438
439static void memcg_reparent_list_lru_one(struct list_lru *lru, int nid,
440 struct list_lru_one *src,
441 struct mem_cgroup *dst_memcg)
442{
443 int dst_idx = dst_memcg->kmemcg_id;
444 struct list_lru_one *dst;
445
446 spin_lock_irq(&src->lock);
447 dst = list_lru_from_memcg_idx(lru, nid, dst_idx);
448 spin_lock_nested(&dst->lock, SINGLE_DEPTH_NESTING);
449
450 list_splice_init(&src->list, &dst->list);
451 if (src->nr_items) {
452 WARN_ON(src->nr_items < 0);
453 dst->nr_items += src->nr_items;
454 set_shrinker_bit(dst_memcg, nid, lru_shrinker_id(lru));
455 }
456 /* Mark the list_lru_one dead */
457 src->nr_items = LONG_MIN;
458
459 spin_unlock(&dst->lock);
460 spin_unlock_irq(&src->lock);
461}
462
463void memcg_reparent_list_lrus(struct mem_cgroup *memcg, struct mem_cgroup *parent)
464{
465 struct list_lru *lru;
466 int i;
467
468 mutex_lock(&list_lrus_mutex);
469 list_for_each_entry(lru, &memcg_list_lrus, list) {
470 struct list_lru_memcg *mlru;
471 XA_STATE(xas, &lru->xa, memcg->kmemcg_id);
472
473 /*
474 * Lock the Xarray to ensure no on going list_lru_memcg
475 * allocation and further allocation will see css_is_dying().
476 */
477 xas_lock_irq(&xas);
478 mlru = xas_store(&xas, NULL);
479 xas_unlock_irq(&xas);
480 if (!mlru)
481 continue;
482
483 /*
484 * With Xarray value set to NULL, holding the lru lock below
485 * prevents list_lru_{add,del,isolate} from touching the lru,
486 * safe to reparent.
487 */
488 for_each_node(i)
489 memcg_reparent_list_lru_one(lru, i, &mlru->node[i], parent);
490
491 /*
492 * Here all list_lrus corresponding to the cgroup are guaranteed
493 * to remain empty, we can safely free this lru, any further
494 * memcg_list_lru_alloc() call will simply bail out.
495 */
496 kvfree_rcu(mlru, rcu);
497 }
498 mutex_unlock(&list_lrus_mutex);
499}
500
501static inline bool memcg_list_lru_allocated(struct mem_cgroup *memcg,
502 struct list_lru *lru)
503{
504 int idx = memcg->kmemcg_id;
505
506 return idx < 0 || xa_load(&lru->xa, idx);
507}
508
509int memcg_list_lru_alloc(struct mem_cgroup *memcg, struct list_lru *lru,
510 gfp_t gfp)
511{
512 unsigned long flags;
513 struct list_lru_memcg *mlru;
514 struct mem_cgroup *pos, *parent;
515 XA_STATE(xas, &lru->xa, 0);
516
517 if (!list_lru_memcg_aware(lru) || memcg_list_lru_allocated(memcg, lru))
518 return 0;
519
520 gfp &= GFP_RECLAIM_MASK;
521 /*
522 * Because the list_lru can be reparented to the parent cgroup's
523 * list_lru, we should make sure that this cgroup and all its
524 * ancestors have allocated list_lru_memcg.
525 */
526 do {
527 /*
528 * Keep finding the farest parent that wasn't populated
529 * until found memcg itself.
530 */
531 pos = memcg;
532 parent = parent_mem_cgroup(pos);
533 while (!memcg_list_lru_allocated(parent, lru)) {
534 pos = parent;
535 parent = parent_mem_cgroup(pos);
536 }
537
538 mlru = memcg_init_list_lru_one(lru, gfp);
539 if (!mlru)
540 return -ENOMEM;
541 xas_set(&xas, pos->kmemcg_id);
542 do {
543 xas_lock_irqsave(&xas, flags);
544 if (!xas_load(&xas) && !css_is_dying(&pos->css)) {
545 xas_store(&xas, mlru);
546 if (!xas_error(&xas))
547 mlru = NULL;
548 }
549 xas_unlock_irqrestore(&xas, flags);
550 } while (xas_nomem(&xas, gfp));
551 if (mlru)
552 kfree(mlru);
553 } while (pos != memcg && !css_is_dying(&pos->css));
554
555 return xas_error(&xas);
556}
557#else
558static inline void memcg_init_list_lru(struct list_lru *lru, bool memcg_aware)
559{
560}
561
562static void memcg_destroy_list_lru(struct list_lru *lru)
563{
564}
565#endif /* CONFIG_MEMCG */
566
567int __list_lru_init(struct list_lru *lru, bool memcg_aware, struct shrinker *shrinker)
568{
569 int i;
570
571#ifdef CONFIG_MEMCG
572 if (shrinker)
573 lru->shrinker_id = shrinker->id;
574 else
575 lru->shrinker_id = -1;
576
577 if (mem_cgroup_kmem_disabled())
578 memcg_aware = false;
579#endif
580
581 lru->node = kcalloc(nr_node_ids, sizeof(*lru->node), GFP_KERNEL);
582 if (!lru->node)
583 return -ENOMEM;
584
585 for_each_node(i)
586 init_one_lru(lru, &lru->node[i].lru);
587
588 memcg_init_list_lru(lru, memcg_aware);
589 list_lru_register(lru);
590
591 return 0;
592}
593EXPORT_SYMBOL_GPL(__list_lru_init);
594
595void list_lru_destroy(struct list_lru *lru)
596{
597 /* Already destroyed or not yet initialized? */
598 if (!lru->node)
599 return;
600
601 list_lru_unregister(lru);
602
603 memcg_destroy_list_lru(lru);
604 kfree(lru->node);
605 lru->node = NULL;
606
607#ifdef CONFIG_MEMCG
608 lru->shrinker_id = -1;
609#endif
610}
611EXPORT_SYMBOL_GPL(list_lru_destroy);