xref: /linux/mm/list_lru.c (revision b24413180f5600bcb3bb70fbed5cf186b60864bd)
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 	/*
56 	 * The lock protects the array of per cgroup lists from relocation
57 	 * (see memcg_update_list_lru_node).
58 	 */
59 	lockdep_assert_held(&nlru->lock);
60 	if (nlru->memcg_lrus && idx >= 0)
61 		return nlru->memcg_lrus->lru[idx];
62 
63 	return &nlru->lru;
64 }
65 
66 static __always_inline struct mem_cgroup *mem_cgroup_from_kmem(void *ptr)
67 {
68 	struct page *page;
69 
70 	if (!memcg_kmem_enabled())
71 		return NULL;
72 	page = virt_to_head_page(ptr);
73 	return page->mem_cgroup;
74 }
75 
76 static inline struct list_lru_one *
77 list_lru_from_kmem(struct list_lru_node *nlru, void *ptr)
78 {
79 	struct mem_cgroup *memcg;
80 
81 	if (!nlru->memcg_lrus)
82 		return &nlru->lru;
83 
84 	memcg = mem_cgroup_from_kmem(ptr);
85 	if (!memcg)
86 		return &nlru->lru;
87 
88 	return list_lru_from_memcg_idx(nlru, memcg_cache_id(memcg));
89 }
90 #else
91 static inline bool list_lru_memcg_aware(struct list_lru *lru)
92 {
93 	return false;
94 }
95 
96 static inline struct list_lru_one *
97 list_lru_from_memcg_idx(struct list_lru_node *nlru, int idx)
98 {
99 	return &nlru->lru;
100 }
101 
102 static inline struct list_lru_one *
103 list_lru_from_kmem(struct list_lru_node *nlru, void *ptr)
104 {
105 	return &nlru->lru;
106 }
107 #endif /* CONFIG_MEMCG && !CONFIG_SLOB */
108 
109 bool list_lru_add(struct list_lru *lru, struct list_head *item)
110 {
111 	int nid = page_to_nid(virt_to_page(item));
112 	struct list_lru_node *nlru = &lru->node[nid];
113 	struct list_lru_one *l;
114 
115 	spin_lock(&nlru->lock);
116 	if (list_empty(item)) {
117 		l = list_lru_from_kmem(nlru, item);
118 		list_add_tail(item, &l->list);
119 		l->nr_items++;
120 		nlru->nr_items++;
121 		spin_unlock(&nlru->lock);
122 		return true;
123 	}
124 	spin_unlock(&nlru->lock);
125 	return false;
126 }
127 EXPORT_SYMBOL_GPL(list_lru_add);
128 
129 bool list_lru_del(struct list_lru *lru, struct list_head *item)
130 {
131 	int nid = page_to_nid(virt_to_page(item));
132 	struct list_lru_node *nlru = &lru->node[nid];
133 	struct list_lru_one *l;
134 
135 	spin_lock(&nlru->lock);
136 	if (!list_empty(item)) {
137 		l = list_lru_from_kmem(nlru, item);
138 		list_del_init(item);
139 		l->nr_items--;
140 		nlru->nr_items--;
141 		spin_unlock(&nlru->lock);
142 		return true;
143 	}
144 	spin_unlock(&nlru->lock);
145 	return false;
146 }
147 EXPORT_SYMBOL_GPL(list_lru_del);
148 
149 void list_lru_isolate(struct list_lru_one *list, struct list_head *item)
150 {
151 	list_del_init(item);
152 	list->nr_items--;
153 }
154 EXPORT_SYMBOL_GPL(list_lru_isolate);
155 
156 void list_lru_isolate_move(struct list_lru_one *list, struct list_head *item,
157 			   struct list_head *head)
158 {
159 	list_move(item, head);
160 	list->nr_items--;
161 }
162 EXPORT_SYMBOL_GPL(list_lru_isolate_move);
163 
164 static unsigned long __list_lru_count_one(struct list_lru *lru,
165 					  int nid, int memcg_idx)
166 {
167 	struct list_lru_node *nlru = &lru->node[nid];
168 	struct list_lru_one *l;
169 	unsigned long count;
170 
171 	spin_lock(&nlru->lock);
172 	l = list_lru_from_memcg_idx(nlru, memcg_idx);
173 	count = l->nr_items;
174 	spin_unlock(&nlru->lock);
175 
176 	return count;
177 }
178 
179 unsigned long list_lru_count_one(struct list_lru *lru,
180 				 int nid, struct mem_cgroup *memcg)
181 {
182 	return __list_lru_count_one(lru, nid, memcg_cache_id(memcg));
183 }
184 EXPORT_SYMBOL_GPL(list_lru_count_one);
185 
186 unsigned long list_lru_count_node(struct list_lru *lru, int nid)
187 {
188 	struct list_lru_node *nlru;
189 
190 	nlru = &lru->node[nid];
191 	return nlru->nr_items;
192 }
193 EXPORT_SYMBOL_GPL(list_lru_count_node);
194 
195 static unsigned long
196 __list_lru_walk_one(struct list_lru *lru, int nid, int memcg_idx,
197 		    list_lru_walk_cb isolate, void *cb_arg,
198 		    unsigned long *nr_to_walk)
199 {
200 
201 	struct list_lru_node *nlru = &lru->node[nid];
202 	struct list_lru_one *l;
203 	struct list_head *item, *n;
204 	unsigned long isolated = 0;
205 
206 	spin_lock(&nlru->lock);
207 	l = list_lru_from_memcg_idx(nlru, memcg_idx);
208 restart:
209 	list_for_each_safe(item, n, &l->list) {
210 		enum lru_status ret;
211 
212 		/*
213 		 * decrement nr_to_walk first so that we don't livelock if we
214 		 * get stuck on large numbesr of LRU_RETRY items
215 		 */
216 		if (!*nr_to_walk)
217 			break;
218 		--*nr_to_walk;
219 
220 		ret = isolate(item, l, &nlru->lock, cb_arg);
221 		switch (ret) {
222 		case LRU_REMOVED_RETRY:
223 			assert_spin_locked(&nlru->lock);
224 		case LRU_REMOVED:
225 			isolated++;
226 			nlru->nr_items--;
227 			/*
228 			 * If the lru lock has been dropped, our list
229 			 * traversal is now invalid and so we have to
230 			 * restart from scratch.
231 			 */
232 			if (ret == LRU_REMOVED_RETRY)
233 				goto restart;
234 			break;
235 		case LRU_ROTATE:
236 			list_move_tail(item, &l->list);
237 			break;
238 		case LRU_SKIP:
239 			break;
240 		case LRU_RETRY:
241 			/*
242 			 * The lru lock has been dropped, our list traversal is
243 			 * now invalid and so we have to restart from scratch.
244 			 */
245 			assert_spin_locked(&nlru->lock);
246 			goto restart;
247 		default:
248 			BUG();
249 		}
250 	}
251 
252 	spin_unlock(&nlru->lock);
253 	return isolated;
254 }
255 
256 unsigned long
257 list_lru_walk_one(struct list_lru *lru, int nid, struct mem_cgroup *memcg,
258 		  list_lru_walk_cb isolate, void *cb_arg,
259 		  unsigned long *nr_to_walk)
260 {
261 	return __list_lru_walk_one(lru, nid, memcg_cache_id(memcg),
262 				   isolate, cb_arg, nr_to_walk);
263 }
264 EXPORT_SYMBOL_GPL(list_lru_walk_one);
265 
266 unsigned long list_lru_walk_node(struct list_lru *lru, int nid,
267 				 list_lru_walk_cb isolate, void *cb_arg,
268 				 unsigned long *nr_to_walk)
269 {
270 	long isolated = 0;
271 	int memcg_idx;
272 
273 	isolated += __list_lru_walk_one(lru, nid, -1, isolate, cb_arg,
274 					nr_to_walk);
275 	if (*nr_to_walk > 0 && list_lru_memcg_aware(lru)) {
276 		for_each_memcg_cache_index(memcg_idx) {
277 			isolated += __list_lru_walk_one(lru, nid, memcg_idx,
278 						isolate, cb_arg, nr_to_walk);
279 			if (*nr_to_walk <= 0)
280 				break;
281 		}
282 	}
283 	return isolated;
284 }
285 EXPORT_SYMBOL_GPL(list_lru_walk_node);
286 
287 static void init_one_lru(struct list_lru_one *l)
288 {
289 	INIT_LIST_HEAD(&l->list);
290 	l->nr_items = 0;
291 }
292 
293 #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
294 static void __memcg_destroy_list_lru_node(struct list_lru_memcg *memcg_lrus,
295 					  int begin, int end)
296 {
297 	int i;
298 
299 	for (i = begin; i < end; i++)
300 		kfree(memcg_lrus->lru[i]);
301 }
302 
303 static int __memcg_init_list_lru_node(struct list_lru_memcg *memcg_lrus,
304 				      int begin, int end)
305 {
306 	int i;
307 
308 	for (i = begin; i < end; i++) {
309 		struct list_lru_one *l;
310 
311 		l = kmalloc(sizeof(struct list_lru_one), GFP_KERNEL);
312 		if (!l)
313 			goto fail;
314 
315 		init_one_lru(l);
316 		memcg_lrus->lru[i] = l;
317 	}
318 	return 0;
319 fail:
320 	__memcg_destroy_list_lru_node(memcg_lrus, begin, i - 1);
321 	return -ENOMEM;
322 }
323 
324 static int memcg_init_list_lru_node(struct list_lru_node *nlru)
325 {
326 	int size = memcg_nr_cache_ids;
327 
328 	nlru->memcg_lrus = kvmalloc(size * sizeof(void *), GFP_KERNEL);
329 	if (!nlru->memcg_lrus)
330 		return -ENOMEM;
331 
332 	if (__memcg_init_list_lru_node(nlru->memcg_lrus, 0, size)) {
333 		kvfree(nlru->memcg_lrus);
334 		return -ENOMEM;
335 	}
336 
337 	return 0;
338 }
339 
340 static void memcg_destroy_list_lru_node(struct list_lru_node *nlru)
341 {
342 	__memcg_destroy_list_lru_node(nlru->memcg_lrus, 0, memcg_nr_cache_ids);
343 	kvfree(nlru->memcg_lrus);
344 }
345 
346 static int memcg_update_list_lru_node(struct list_lru_node *nlru,
347 				      int old_size, int new_size)
348 {
349 	struct list_lru_memcg *old, *new;
350 
351 	BUG_ON(old_size > new_size);
352 
353 	old = nlru->memcg_lrus;
354 	new = kvmalloc(new_size * sizeof(void *), GFP_KERNEL);
355 	if (!new)
356 		return -ENOMEM;
357 
358 	if (__memcg_init_list_lru_node(new, old_size, new_size)) {
359 		kvfree(new);
360 		return -ENOMEM;
361 	}
362 
363 	memcpy(new, old, old_size * sizeof(void *));
364 
365 	/*
366 	 * The lock guarantees that we won't race with a reader
367 	 * (see list_lru_from_memcg_idx).
368 	 *
369 	 * Since list_lru_{add,del} may be called under an IRQ-safe lock,
370 	 * we have to use IRQ-safe primitives here to avoid deadlock.
371 	 */
372 	spin_lock_irq(&nlru->lock);
373 	nlru->memcg_lrus = new;
374 	spin_unlock_irq(&nlru->lock);
375 
376 	kvfree(old);
377 	return 0;
378 }
379 
380 static void memcg_cancel_update_list_lru_node(struct list_lru_node *nlru,
381 					      int old_size, int new_size)
382 {
383 	/* do not bother shrinking the array back to the old size, because we
384 	 * cannot handle allocation failures here */
385 	__memcg_destroy_list_lru_node(nlru->memcg_lrus, old_size, new_size);
386 }
387 
388 static int memcg_init_list_lru(struct list_lru *lru, bool memcg_aware)
389 {
390 	int i;
391 
392 	if (!memcg_aware)
393 		return 0;
394 
395 	for_each_node(i) {
396 		if (memcg_init_list_lru_node(&lru->node[i]))
397 			goto fail;
398 	}
399 	return 0;
400 fail:
401 	for (i = i - 1; i >= 0; i--) {
402 		if (!lru->node[i].memcg_lrus)
403 			continue;
404 		memcg_destroy_list_lru_node(&lru->node[i]);
405 	}
406 	return -ENOMEM;
407 }
408 
409 static void memcg_destroy_list_lru(struct list_lru *lru)
410 {
411 	int i;
412 
413 	if (!list_lru_memcg_aware(lru))
414 		return;
415 
416 	for_each_node(i)
417 		memcg_destroy_list_lru_node(&lru->node[i]);
418 }
419 
420 static int memcg_update_list_lru(struct list_lru *lru,
421 				 int old_size, int new_size)
422 {
423 	int i;
424 
425 	if (!list_lru_memcg_aware(lru))
426 		return 0;
427 
428 	for_each_node(i) {
429 		if (memcg_update_list_lru_node(&lru->node[i],
430 					       old_size, new_size))
431 			goto fail;
432 	}
433 	return 0;
434 fail:
435 	for (i = i - 1; i >= 0; i--) {
436 		if (!lru->node[i].memcg_lrus)
437 			continue;
438 
439 		memcg_cancel_update_list_lru_node(&lru->node[i],
440 						  old_size, new_size);
441 	}
442 	return -ENOMEM;
443 }
444 
445 static void memcg_cancel_update_list_lru(struct list_lru *lru,
446 					 int old_size, int new_size)
447 {
448 	int i;
449 
450 	if (!list_lru_memcg_aware(lru))
451 		return;
452 
453 	for_each_node(i)
454 		memcg_cancel_update_list_lru_node(&lru->node[i],
455 						  old_size, new_size);
456 }
457 
458 int memcg_update_all_list_lrus(int new_size)
459 {
460 	int ret = 0;
461 	struct list_lru *lru;
462 	int old_size = memcg_nr_cache_ids;
463 
464 	mutex_lock(&list_lrus_mutex);
465 	list_for_each_entry(lru, &list_lrus, list) {
466 		ret = memcg_update_list_lru(lru, old_size, new_size);
467 		if (ret)
468 			goto fail;
469 	}
470 out:
471 	mutex_unlock(&list_lrus_mutex);
472 	return ret;
473 fail:
474 	list_for_each_entry_continue_reverse(lru, &list_lrus, list)
475 		memcg_cancel_update_list_lru(lru, old_size, new_size);
476 	goto out;
477 }
478 
479 static void memcg_drain_list_lru_node(struct list_lru_node *nlru,
480 				      int src_idx, int dst_idx)
481 {
482 	struct list_lru_one *src, *dst;
483 
484 	/*
485 	 * Since list_lru_{add,del} may be called under an IRQ-safe lock,
486 	 * we have to use IRQ-safe primitives here to avoid deadlock.
487 	 */
488 	spin_lock_irq(&nlru->lock);
489 
490 	src = list_lru_from_memcg_idx(nlru, src_idx);
491 	dst = list_lru_from_memcg_idx(nlru, dst_idx);
492 
493 	list_splice_init(&src->list, &dst->list);
494 	dst->nr_items += src->nr_items;
495 	src->nr_items = 0;
496 
497 	spin_unlock_irq(&nlru->lock);
498 }
499 
500 static void memcg_drain_list_lru(struct list_lru *lru,
501 				 int src_idx, int dst_idx)
502 {
503 	int i;
504 
505 	if (!list_lru_memcg_aware(lru))
506 		return;
507 
508 	for_each_node(i)
509 		memcg_drain_list_lru_node(&lru->node[i], src_idx, dst_idx);
510 }
511 
512 void memcg_drain_all_list_lrus(int src_idx, int dst_idx)
513 {
514 	struct list_lru *lru;
515 
516 	mutex_lock(&list_lrus_mutex);
517 	list_for_each_entry(lru, &list_lrus, list)
518 		memcg_drain_list_lru(lru, src_idx, dst_idx);
519 	mutex_unlock(&list_lrus_mutex);
520 }
521 #else
522 static int memcg_init_list_lru(struct list_lru *lru, bool memcg_aware)
523 {
524 	return 0;
525 }
526 
527 static void memcg_destroy_list_lru(struct list_lru *lru)
528 {
529 }
530 #endif /* CONFIG_MEMCG && !CONFIG_SLOB */
531 
532 int __list_lru_init(struct list_lru *lru, bool memcg_aware,
533 		    struct lock_class_key *key)
534 {
535 	int i;
536 	size_t size = sizeof(*lru->node) * nr_node_ids;
537 	int err = -ENOMEM;
538 
539 	memcg_get_cache_ids();
540 
541 	lru->node = kzalloc(size, GFP_KERNEL);
542 	if (!lru->node)
543 		goto out;
544 
545 	for_each_node(i) {
546 		spin_lock_init(&lru->node[i].lock);
547 		if (key)
548 			lockdep_set_class(&lru->node[i].lock, key);
549 		init_one_lru(&lru->node[i].lru);
550 	}
551 
552 	err = memcg_init_list_lru(lru, memcg_aware);
553 	if (err) {
554 		kfree(lru->node);
555 		/* Do this so a list_lru_destroy() doesn't crash: */
556 		lru->node = NULL;
557 		goto out;
558 	}
559 
560 	list_lru_register(lru);
561 out:
562 	memcg_put_cache_ids();
563 	return err;
564 }
565 EXPORT_SYMBOL_GPL(__list_lru_init);
566 
567 void list_lru_destroy(struct list_lru *lru)
568 {
569 	/* Already destroyed or not yet initialized? */
570 	if (!lru->node)
571 		return;
572 
573 	memcg_get_cache_ids();
574 
575 	list_lru_unregister(lru);
576 
577 	memcg_destroy_list_lru(lru);
578 	kfree(lru->node);
579 	lru->node = NULL;
580 
581 	memcg_put_cache_ids();
582 }
583 EXPORT_SYMBOL_GPL(list_lru_destroy);
584