xref: /openbmc/linux/mm/list_lru.c (revision 2359ccdd)
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)
16 static LIST_HEAD(list_lrus);
17 static DEFINE_MUTEX(list_lrus_mutex);
18 
19 static 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 
26 static 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
33 static void list_lru_register(struct list_lru *lru)
34 {
35 }
36 
37 static 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)
43 static 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 
52 static inline struct list_lru_one *
53 list_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 
67 static __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 
77 static inline struct list_lru_one *
78 list_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
92 static inline bool list_lru_memcg_aware(struct list_lru *lru)
93 {
94 	return false;
95 }
96 
97 static inline struct list_lru_one *
98 list_lru_from_memcg_idx(struct list_lru_node *nlru, int idx)
99 {
100 	return &nlru->lru;
101 }
102 
103 static inline struct list_lru_one *
104 list_lru_from_kmem(struct list_lru_node *nlru, void *ptr)
105 {
106 	return &nlru->lru;
107 }
108 #endif /* CONFIG_MEMCG && !CONFIG_SLOB */
109 
110 bool 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 }
128 EXPORT_SYMBOL_GPL(list_lru_add);
129 
130 bool 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 }
148 EXPORT_SYMBOL_GPL(list_lru_del);
149 
150 void list_lru_isolate(struct list_lru_one *list, struct list_head *item)
151 {
152 	list_del_init(item);
153 	list->nr_items--;
154 }
155 EXPORT_SYMBOL_GPL(list_lru_isolate);
156 
157 void 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 }
163 EXPORT_SYMBOL_GPL(list_lru_isolate_move);
164 
165 static 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 
180 unsigned 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 }
185 EXPORT_SYMBOL_GPL(list_lru_count_one);
186 
187 unsigned 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 }
194 EXPORT_SYMBOL_GPL(list_lru_count_node);
195 
196 static 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);
209 restart:
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 
258 unsigned long
259 list_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 }
266 EXPORT_SYMBOL_GPL(list_lru_walk_one);
267 
268 unsigned 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 }
287 EXPORT_SYMBOL_GPL(list_lru_walk_node);
288 
289 static 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)
296 static 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 
305 static 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;
321 fail:
322 	__memcg_destroy_list_lru_node(memcg_lrus, begin, i - 1);
323 	return -ENOMEM;
324 }
325 
326 static 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 
345 static 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 
357 static 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 
365 static 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 
400 static 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 
412 static 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;
424 fail:
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 
433 static 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 
444 static 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;
458 fail:
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 
469 static 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 
482 int 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 	}
494 out:
495 	mutex_unlock(&list_lrus_mutex);
496 	return ret;
497 fail:
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 
503 static 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 
524 static 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 
536 void 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
546 static int memcg_init_list_lru(struct list_lru *lru, bool memcg_aware)
547 {
548 	return 0;
549 }
550 
551 static void memcg_destroy_list_lru(struct list_lru *lru)
552 {
553 }
554 #endif /* CONFIG_MEMCG && !CONFIG_SLOB */
555 
556 int __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);
585 out:
586 	memcg_put_cache_ids();
587 	return err;
588 }
589 EXPORT_SYMBOL_GPL(__list_lru_init);
590 
591 void 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 }
607 EXPORT_SYMBOL_GPL(list_lru_destroy);
608