xref: /linux/mm/list_lru.c (revision a1c3be890440a1769ed6f822376a3e3ab0d42994)
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
18 static LIST_HEAD(list_lrus);
19 static DEFINE_MUTEX(list_lrus_mutex);
20 
21 static 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 
28 static 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 
35 static int lru_shrinker_id(struct list_lru *lru)
36 {
37 	return lru->shrinker_id;
38 }
39 
40 static inline bool list_lru_memcg_aware(struct list_lru *lru)
41 {
42 	return lru->memcg_aware;
43 }
44 
45 static inline struct list_lru_one *
46 list_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 
60 static inline struct list_lru_one *
61 list_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));
75 out:
76 	if (memcg_ptr)
77 		*memcg_ptr = memcg;
78 	return l;
79 }
80 #else
81 static void list_lru_register(struct list_lru *lru)
82 {
83 }
84 
85 static void list_lru_unregister(struct list_lru *lru)
86 {
87 }
88 
89 static int lru_shrinker_id(struct list_lru *lru)
90 {
91 	return -1;
92 }
93 
94 static inline bool list_lru_memcg_aware(struct list_lru *lru)
95 {
96 	return false;
97 }
98 
99 static inline struct list_lru_one *
100 list_lru_from_memcg_idx(struct list_lru_node *nlru, int idx)
101 {
102 	return &nlru->lru;
103 }
104 
105 static inline struct list_lru_one *
106 list_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 
115 bool 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 }
137 EXPORT_SYMBOL_GPL(list_lru_add);
138 
139 bool 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 }
157 EXPORT_SYMBOL_GPL(list_lru_del);
158 
159 void list_lru_isolate(struct list_lru_one *list, struct list_head *item)
160 {
161 	list_del_init(item);
162 	list->nr_items--;
163 }
164 EXPORT_SYMBOL_GPL(list_lru_isolate);
165 
166 void 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 }
172 EXPORT_SYMBOL_GPL(list_lru_isolate_move);
173 
174 unsigned 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 }
188 EXPORT_SYMBOL_GPL(list_lru_count_one);
189 
190 unsigned 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 }
197 EXPORT_SYMBOL_GPL(list_lru_count_node);
198 
199 static 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);
210 restart:
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 
257 unsigned long
258 list_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 }
271 EXPORT_SYMBOL_GPL(list_lru_walk_one);
272 
273 unsigned long
274 list_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 
288 unsigned 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 }
313 EXPORT_SYMBOL_GPL(list_lru_walk_node);
314 
315 static 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
322 static 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 
331 static 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;
347 fail:
348 	__memcg_destroy_list_lru_node(memcg_lrus, begin, i);
349 	return -ENOMEM;
350 }
351 
352 static 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 
371 static 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().
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 
383 static int memcg_update_list_lru_node(struct list_lru_node *nlru,
384 				      int old_size, int new_size)
385 {
386 	struct list_lru_memcg *old, *new;
387 
388 	BUG_ON(old_size > new_size);
389 
390 	old = rcu_dereference_protected(nlru->memcg_lrus,
391 					lockdep_is_held(&list_lrus_mutex));
392 	new = kvmalloc(sizeof(*new) + new_size * sizeof(void *), GFP_KERNEL);
393 	if (!new)
394 		return -ENOMEM;
395 
396 	if (__memcg_init_list_lru_node(new, old_size, new_size)) {
397 		kvfree(new);
398 		return -ENOMEM;
399 	}
400 
401 	memcpy(&new->lru, &old->lru, old_size * sizeof(void *));
402 
403 	/*
404 	 * The locking below allows readers that hold nlru->lock avoid taking
405 	 * rcu_read_lock (see list_lru_from_memcg_idx).
406 	 *
407 	 * Since list_lru_{add,del} may be called under an IRQ-safe lock,
408 	 * we have to use IRQ-safe primitives here to avoid deadlock.
409 	 */
410 	spin_lock_irq(&nlru->lock);
411 	rcu_assign_pointer(nlru->memcg_lrus, new);
412 	spin_unlock_irq(&nlru->lock);
413 
414 	kvfree_rcu(old, rcu);
415 	return 0;
416 }
417 
418 static void memcg_cancel_update_list_lru_node(struct list_lru_node *nlru,
419 					      int old_size, int new_size)
420 {
421 	struct list_lru_memcg *memcg_lrus;
422 
423 	memcg_lrus = rcu_dereference_protected(nlru->memcg_lrus,
424 					       lockdep_is_held(&list_lrus_mutex));
425 	/* do not bother shrinking the array back to the old size, because we
426 	 * cannot handle allocation failures here */
427 	__memcg_destroy_list_lru_node(memcg_lrus, old_size, new_size);
428 }
429 
430 static int memcg_init_list_lru(struct list_lru *lru, bool memcg_aware)
431 {
432 	int i;
433 
434 	lru->memcg_aware = memcg_aware;
435 
436 	if (!memcg_aware)
437 		return 0;
438 
439 	for_each_node(i) {
440 		if (memcg_init_list_lru_node(&lru->node[i]))
441 			goto fail;
442 	}
443 	return 0;
444 fail:
445 	for (i = i - 1; i >= 0; i--) {
446 		if (!lru->node[i].memcg_lrus)
447 			continue;
448 		memcg_destroy_list_lru_node(&lru->node[i]);
449 	}
450 	return -ENOMEM;
451 }
452 
453 static void memcg_destroy_list_lru(struct list_lru *lru)
454 {
455 	int i;
456 
457 	if (!list_lru_memcg_aware(lru))
458 		return;
459 
460 	for_each_node(i)
461 		memcg_destroy_list_lru_node(&lru->node[i]);
462 }
463 
464 static int memcg_update_list_lru(struct list_lru *lru,
465 				 int old_size, int new_size)
466 {
467 	int i;
468 
469 	if (!list_lru_memcg_aware(lru))
470 		return 0;
471 
472 	for_each_node(i) {
473 		if (memcg_update_list_lru_node(&lru->node[i],
474 					       old_size, new_size))
475 			goto fail;
476 	}
477 	return 0;
478 fail:
479 	for (i = i - 1; i >= 0; i--) {
480 		if (!lru->node[i].memcg_lrus)
481 			continue;
482 
483 		memcg_cancel_update_list_lru_node(&lru->node[i],
484 						  old_size, new_size);
485 	}
486 	return -ENOMEM;
487 }
488 
489 static void memcg_cancel_update_list_lru(struct list_lru *lru,
490 					 int old_size, int new_size)
491 {
492 	int i;
493 
494 	if (!list_lru_memcg_aware(lru))
495 		return;
496 
497 	for_each_node(i)
498 		memcg_cancel_update_list_lru_node(&lru->node[i],
499 						  old_size, new_size);
500 }
501 
502 int memcg_update_all_list_lrus(int new_size)
503 {
504 	int ret = 0;
505 	struct list_lru *lru;
506 	int old_size = memcg_nr_cache_ids;
507 
508 	mutex_lock(&list_lrus_mutex);
509 	list_for_each_entry(lru, &list_lrus, list) {
510 		ret = memcg_update_list_lru(lru, old_size, new_size);
511 		if (ret)
512 			goto fail;
513 	}
514 out:
515 	mutex_unlock(&list_lrus_mutex);
516 	return ret;
517 fail:
518 	list_for_each_entry_continue_reverse(lru, &list_lrus, list)
519 		memcg_cancel_update_list_lru(lru, old_size, new_size);
520 	goto out;
521 }
522 
523 static void memcg_drain_list_lru_node(struct list_lru *lru, int nid,
524 				      int src_idx, struct mem_cgroup *dst_memcg)
525 {
526 	struct list_lru_node *nlru = &lru->node[nid];
527 	int dst_idx = dst_memcg->kmemcg_id;
528 	struct list_lru_one *src, *dst;
529 
530 	/*
531 	 * Since list_lru_{add,del} may be called under an IRQ-safe lock,
532 	 * we have to use IRQ-safe primitives here to avoid deadlock.
533 	 */
534 	spin_lock_irq(&nlru->lock);
535 
536 	src = list_lru_from_memcg_idx(nlru, src_idx);
537 	dst = list_lru_from_memcg_idx(nlru, dst_idx);
538 
539 	list_splice_init(&src->list, &dst->list);
540 
541 	if (src->nr_items) {
542 		dst->nr_items += src->nr_items;
543 		memcg_set_shrinker_bit(dst_memcg, nid, lru_shrinker_id(lru));
544 		src->nr_items = 0;
545 	}
546 
547 	spin_unlock_irq(&nlru->lock);
548 }
549 
550 static void memcg_drain_list_lru(struct list_lru *lru,
551 				 int src_idx, struct mem_cgroup *dst_memcg)
552 {
553 	int i;
554 
555 	if (!list_lru_memcg_aware(lru))
556 		return;
557 
558 	for_each_node(i)
559 		memcg_drain_list_lru_node(lru, i, src_idx, dst_memcg);
560 }
561 
562 void memcg_drain_all_list_lrus(int src_idx, struct mem_cgroup *dst_memcg)
563 {
564 	struct list_lru *lru;
565 
566 	mutex_lock(&list_lrus_mutex);
567 	list_for_each_entry(lru, &list_lrus, list)
568 		memcg_drain_list_lru(lru, src_idx, dst_memcg);
569 	mutex_unlock(&list_lrus_mutex);
570 }
571 #else
572 static int memcg_init_list_lru(struct list_lru *lru, bool memcg_aware)
573 {
574 	return 0;
575 }
576 
577 static void memcg_destroy_list_lru(struct list_lru *lru)
578 {
579 }
580 #endif /* CONFIG_MEMCG_KMEM */
581 
582 int __list_lru_init(struct list_lru *lru, bool memcg_aware,
583 		    struct lock_class_key *key, struct shrinker *shrinker)
584 {
585 	int i;
586 	int err = -ENOMEM;
587 
588 #ifdef CONFIG_MEMCG_KMEM
589 	if (shrinker)
590 		lru->shrinker_id = shrinker->id;
591 	else
592 		lru->shrinker_id = -1;
593 #endif
594 	memcg_get_cache_ids();
595 
596 	lru->node = kcalloc(nr_node_ids, sizeof(*lru->node), GFP_KERNEL);
597 	if (!lru->node)
598 		goto out;
599 
600 	for_each_node(i) {
601 		spin_lock_init(&lru->node[i].lock);
602 		if (key)
603 			lockdep_set_class(&lru->node[i].lock, key);
604 		init_one_lru(&lru->node[i].lru);
605 	}
606 
607 	err = memcg_init_list_lru(lru, memcg_aware);
608 	if (err) {
609 		kfree(lru->node);
610 		/* Do this so a list_lru_destroy() doesn't crash: */
611 		lru->node = NULL;
612 		goto out;
613 	}
614 
615 	list_lru_register(lru);
616 out:
617 	memcg_put_cache_ids();
618 	return err;
619 }
620 EXPORT_SYMBOL_GPL(__list_lru_init);
621 
622 void list_lru_destroy(struct list_lru *lru)
623 {
624 	/* Already destroyed or not yet initialized? */
625 	if (!lru->node)
626 		return;
627 
628 	memcg_get_cache_ids();
629 
630 	list_lru_unregister(lru);
631 
632 	memcg_destroy_list_lru(lru);
633 	kfree(lru->node);
634 	lru->node = NULL;
635 
636 #ifdef CONFIG_MEMCG_KMEM
637 	lru->shrinker_id = -1;
638 #endif
639 	memcg_put_cache_ids();
640 }
641 EXPORT_SYMBOL_GPL(list_lru_destroy);
642