1a38e4082SDave Chinner /* 2a38e4082SDave Chinner * Copyright (c) 2013 Red Hat, Inc. and Parallels Inc. All rights reserved. 3a38e4082SDave Chinner * Authors: David Chinner and Glauber Costa 4a38e4082SDave Chinner * 5a38e4082SDave Chinner * Generic LRU infrastructure 6a38e4082SDave Chinner */ 7a38e4082SDave Chinner #ifndef _LRU_LIST_H 8a38e4082SDave Chinner #define _LRU_LIST_H 9a38e4082SDave Chinner 10a38e4082SDave Chinner #include <linux/list.h> 113b1d58a4SDave Chinner #include <linux/nodemask.h> 12503c358cSVladimir Davydov #include <linux/shrinker.h> 13a38e4082SDave Chinner 1460d3fd32SVladimir Davydov struct mem_cgroup; 1560d3fd32SVladimir Davydov 16a38e4082SDave Chinner /* list_lru_walk_cb has to always return one of those */ 17a38e4082SDave Chinner enum lru_status { 18a38e4082SDave Chinner LRU_REMOVED, /* item removed from list */ 19449dd698SJohannes Weiner LRU_REMOVED_RETRY, /* item removed, but lock has been 20449dd698SJohannes Weiner dropped and reacquired */ 21a38e4082SDave Chinner LRU_ROTATE, /* item referenced, give another pass */ 22a38e4082SDave Chinner LRU_SKIP, /* item cannot be locked, skip */ 23a38e4082SDave Chinner LRU_RETRY, /* item not freeable. May drop the lock 24a38e4082SDave Chinner internally, but has to return locked. */ 25a38e4082SDave Chinner }; 26a38e4082SDave Chinner 2760d3fd32SVladimir Davydov struct list_lru_one { 28a38e4082SDave Chinner struct list_head list; 29a38e4082SDave Chinner /* kept as signed so we can catch imbalance bugs */ 30a38e4082SDave Chinner long nr_items; 3160d3fd32SVladimir Davydov }; 3260d3fd32SVladimir Davydov 3360d3fd32SVladimir Davydov struct list_lru_memcg { 3460d3fd32SVladimir Davydov /* array of per cgroup lists, indexed by memcg_cache_id */ 3560d3fd32SVladimir Davydov struct list_lru_one *lru[0]; 3660d3fd32SVladimir Davydov }; 3760d3fd32SVladimir Davydov 3860d3fd32SVladimir Davydov struct list_lru_node { 3960d3fd32SVladimir Davydov /* protects all lists on the node, including per cgroup */ 4060d3fd32SVladimir Davydov spinlock_t lock; 4160d3fd32SVladimir Davydov /* global list, used for the root cgroup in cgroup aware lrus */ 4260d3fd32SVladimir Davydov struct list_lru_one lru; 4360d3fd32SVladimir Davydov #ifdef CONFIG_MEMCG_KMEM 4460d3fd32SVladimir Davydov /* for cgroup aware lrus points to per cgroup lists, otherwise NULL */ 4560d3fd32SVladimir Davydov struct list_lru_memcg *memcg_lrus; 4660d3fd32SVladimir Davydov #endif 473b1d58a4SDave Chinner } ____cacheline_aligned_in_smp; 483b1d58a4SDave Chinner 493b1d58a4SDave Chinner struct list_lru { 505ca302c8SGlauber Costa struct list_lru_node *node; 51c0a5b560SVladimir Davydov #ifdef CONFIG_MEMCG_KMEM 52c0a5b560SVladimir Davydov struct list_head list; 53c0a5b560SVladimir Davydov #endif 54a38e4082SDave Chinner }; 55a38e4082SDave Chinner 565ca302c8SGlauber Costa void list_lru_destroy(struct list_lru *lru); 5760d3fd32SVladimir Davydov int __list_lru_init(struct list_lru *lru, bool memcg_aware, 5860d3fd32SVladimir Davydov struct lock_class_key *key); 5960d3fd32SVladimir Davydov 6060d3fd32SVladimir Davydov #define list_lru_init(lru) __list_lru_init((lru), false, NULL) 6160d3fd32SVladimir Davydov #define list_lru_init_key(lru, key) __list_lru_init((lru), false, (key)) 6260d3fd32SVladimir Davydov #define list_lru_init_memcg(lru) __list_lru_init((lru), true, NULL) 6360d3fd32SVladimir Davydov 6460d3fd32SVladimir Davydov int memcg_update_all_list_lrus(int num_memcgs); 65a38e4082SDave Chinner 66a38e4082SDave Chinner /** 67a38e4082SDave Chinner * list_lru_add: add an element to the lru list's tail 68a38e4082SDave Chinner * @list_lru: the lru pointer 69a38e4082SDave Chinner * @item: the item to be added. 70a38e4082SDave Chinner * 71a38e4082SDave Chinner * If the element is already part of a list, this function returns doing 72a38e4082SDave Chinner * nothing. Therefore the caller does not need to keep state about whether or 73a38e4082SDave Chinner * not the element already belongs in the list and is allowed to lazy update 74a38e4082SDave Chinner * it. Note however that this is valid for *a* list, not *this* list. If 75a38e4082SDave Chinner * the caller organize itself in a way that elements can be in more than 76a38e4082SDave Chinner * one type of list, it is up to the caller to fully remove the item from 77a38e4082SDave Chinner * the previous list (with list_lru_del() for instance) before moving it 78a38e4082SDave Chinner * to @list_lru 79a38e4082SDave Chinner * 80a38e4082SDave Chinner * Return value: true if the list was updated, false otherwise 81a38e4082SDave Chinner */ 82a38e4082SDave Chinner bool list_lru_add(struct list_lru *lru, struct list_head *item); 83a38e4082SDave Chinner 84a38e4082SDave Chinner /** 85a38e4082SDave Chinner * list_lru_del: delete an element to the lru list 86a38e4082SDave Chinner * @list_lru: the lru pointer 87a38e4082SDave Chinner * @item: the item to be deleted. 88a38e4082SDave Chinner * 89a38e4082SDave Chinner * This function works analogously as list_lru_add in terms of list 90a38e4082SDave Chinner * manipulation. The comments about an element already pertaining to 91a38e4082SDave Chinner * a list are also valid for list_lru_del. 92a38e4082SDave Chinner * 93a38e4082SDave Chinner * Return value: true if the list was updated, false otherwise 94a38e4082SDave Chinner */ 95a38e4082SDave Chinner bool list_lru_del(struct list_lru *lru, struct list_head *item); 96a38e4082SDave Chinner 97a38e4082SDave Chinner /** 9860d3fd32SVladimir Davydov * list_lru_count_one: return the number of objects currently held by @lru 99a38e4082SDave Chinner * @lru: the lru pointer. 1006a4f496fSGlauber Costa * @nid: the node id to count from. 10160d3fd32SVladimir Davydov * @memcg: the cgroup to count from. 102a38e4082SDave Chinner * 103a38e4082SDave Chinner * Always return a non-negative number, 0 for empty lists. There is no 104a38e4082SDave Chinner * guarantee that the list is not updated while the count is being computed. 105a38e4082SDave Chinner * Callers that want such a guarantee need to provide an outer lock. 106a38e4082SDave Chinner */ 10760d3fd32SVladimir Davydov unsigned long list_lru_count_one(struct list_lru *lru, 10860d3fd32SVladimir Davydov int nid, struct mem_cgroup *memcg); 1096a4f496fSGlauber Costa unsigned long list_lru_count_node(struct list_lru *lru, int nid); 110503c358cSVladimir Davydov 111503c358cSVladimir Davydov static inline unsigned long list_lru_shrink_count(struct list_lru *lru, 112503c358cSVladimir Davydov struct shrink_control *sc) 113503c358cSVladimir Davydov { 11460d3fd32SVladimir Davydov return list_lru_count_one(lru, sc->nid, sc->memcg); 115503c358cSVladimir Davydov } 116503c358cSVladimir Davydov 1176a4f496fSGlauber Costa static inline unsigned long list_lru_count(struct list_lru *lru) 1186a4f496fSGlauber Costa { 1196a4f496fSGlauber Costa long count = 0; 1206a4f496fSGlauber Costa int nid; 1216a4f496fSGlauber Costa 122ff0b67efSVladimir Davydov for_each_node_state(nid, N_NORMAL_MEMORY) 1236a4f496fSGlauber Costa count += list_lru_count_node(lru, nid); 1246a4f496fSGlauber Costa 1256a4f496fSGlauber Costa return count; 1266a4f496fSGlauber Costa } 127a38e4082SDave Chinner 128*3f97b163SVladimir Davydov void list_lru_isolate(struct list_lru_one *list, struct list_head *item); 129*3f97b163SVladimir Davydov void list_lru_isolate_move(struct list_lru_one *list, struct list_head *item, 130*3f97b163SVladimir Davydov struct list_head *head); 131*3f97b163SVladimir Davydov 132*3f97b163SVladimir Davydov typedef enum lru_status (*list_lru_walk_cb)(struct list_head *item, 133*3f97b163SVladimir Davydov struct list_lru_one *list, spinlock_t *lock, void *cb_arg); 134*3f97b163SVladimir Davydov 135a38e4082SDave Chinner /** 13660d3fd32SVladimir Davydov * list_lru_walk_one: walk a list_lru, isolating and disposing freeable items. 137a38e4082SDave Chinner * @lru: the lru pointer. 1386a4f496fSGlauber Costa * @nid: the node id to scan from. 13960d3fd32SVladimir Davydov * @memcg: the cgroup to scan from. 140a38e4082SDave Chinner * @isolate: callback function that is resposible for deciding what to do with 141a38e4082SDave Chinner * the item currently being scanned 142a38e4082SDave Chinner * @cb_arg: opaque type that will be passed to @isolate 143a38e4082SDave Chinner * @nr_to_walk: how many items to scan. 144a38e4082SDave Chinner * 145a38e4082SDave Chinner * This function will scan all elements in a particular list_lru, calling the 146a38e4082SDave Chinner * @isolate callback for each of those items, along with the current list 147a38e4082SDave Chinner * spinlock and a caller-provided opaque. The @isolate callback can choose to 148a38e4082SDave Chinner * drop the lock internally, but *must* return with the lock held. The callback 149a38e4082SDave Chinner * will return an enum lru_status telling the list_lru infrastructure what to 150a38e4082SDave Chinner * do with the object being scanned. 151a38e4082SDave Chinner * 152a38e4082SDave Chinner * Please note that nr_to_walk does not mean how many objects will be freed, 153a38e4082SDave Chinner * just how many objects will be scanned. 154a38e4082SDave Chinner * 155a38e4082SDave Chinner * Return value: the number of objects effectively removed from the LRU. 156a38e4082SDave Chinner */ 15760d3fd32SVladimir Davydov unsigned long list_lru_walk_one(struct list_lru *lru, 15860d3fd32SVladimir Davydov int nid, struct mem_cgroup *memcg, 15960d3fd32SVladimir Davydov list_lru_walk_cb isolate, void *cb_arg, 16060d3fd32SVladimir Davydov unsigned long *nr_to_walk); 1616a4f496fSGlauber Costa unsigned long list_lru_walk_node(struct list_lru *lru, int nid, 1626a4f496fSGlauber Costa list_lru_walk_cb isolate, void *cb_arg, 1636a4f496fSGlauber Costa unsigned long *nr_to_walk); 1646a4f496fSGlauber Costa 1656a4f496fSGlauber Costa static inline unsigned long 166503c358cSVladimir Davydov list_lru_shrink_walk(struct list_lru *lru, struct shrink_control *sc, 167503c358cSVladimir Davydov list_lru_walk_cb isolate, void *cb_arg) 168503c358cSVladimir Davydov { 16960d3fd32SVladimir Davydov return list_lru_walk_one(lru, sc->nid, sc->memcg, isolate, cb_arg, 170503c358cSVladimir Davydov &sc->nr_to_scan); 171503c358cSVladimir Davydov } 172503c358cSVladimir Davydov 173503c358cSVladimir Davydov static inline unsigned long 1746a4f496fSGlauber Costa list_lru_walk(struct list_lru *lru, list_lru_walk_cb isolate, 1756a4f496fSGlauber Costa void *cb_arg, unsigned long nr_to_walk) 1766a4f496fSGlauber Costa { 1776a4f496fSGlauber Costa long isolated = 0; 1786a4f496fSGlauber Costa int nid; 1796a4f496fSGlauber Costa 180ff0b67efSVladimir Davydov for_each_node_state(nid, N_NORMAL_MEMORY) { 1816a4f496fSGlauber Costa isolated += list_lru_walk_node(lru, nid, isolate, 1826a4f496fSGlauber Costa cb_arg, &nr_to_walk); 1836a4f496fSGlauber Costa if (nr_to_walk <= 0) 1846a4f496fSGlauber Costa break; 1856a4f496fSGlauber Costa } 1866a4f496fSGlauber Costa return isolated; 1876a4f496fSGlauber Costa } 188a38e4082SDave Chinner #endif /* _LRU_LIST_H */ 189