vmscan: per memory cgroup slab shrinkers
authorVladimir Davydov <vdavydov@parallels.com>
Thu, 12 Feb 2015 22:58:54 +0000 (14:58 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 13 Feb 2015 02:54:09 +0000 (18:54 -0800)
This patch adds SHRINKER_MEMCG_AWARE flag.  If a shrinker has this flag
set, it will be called per memory cgroup.  The memory cgroup to scan
objects from is passed in shrink_control->memcg.  If the memory cgroup
is NULL, a memcg aware shrinker is supposed to scan objects from the
global list.  Unaware shrinkers are only called on global pressure with
memcg=NULL.

Signed-off-by: Vladimir Davydov <vdavydov@parallels.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Greg Thelen <gthelen@google.com>
Cc: Glauber Costa <glommer@gmail.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Lameter <cl@linux.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Tejun Heo <tj@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/drop_caches.c
include/linux/memcontrol.h
include/linux/mm.h
include/linux/shrinker.h
mm/memcontrol.c
mm/memory-failure.c
mm/vmscan.c

index 2bc2c87f35e7a251628da1390f288b0cf8f2e721..5718cb9f7273e83a2fe6c7775dcf76c636dd908a 100644 (file)
@@ -37,20 +37,6 @@ static void drop_pagecache_sb(struct super_block *sb, void *unused)
        iput(toput_inode);
 }
 
-static void drop_slab(void)
-{
-       int nr_objects;
-
-       do {
-               int nid;
-
-               nr_objects = 0;
-               for_each_online_node(nid)
-                       nr_objects += shrink_node_slabs(GFP_KERNEL, nid,
-                                                       1000, 1000);
-       } while (nr_objects > 10);
-}
-
 int drop_caches_sysctl_handler(struct ctl_table *table, int write,
        void __user *buffer, size_t *length, loff_t *ppos)
 {
index 6cfd934c7c9b3ef766fdfc0fb9d75c40238e75d4..54992fe0959f541940c1f4c85fc847e3506cfab7 100644 (file)
@@ -413,6 +413,8 @@ static inline bool memcg_kmem_enabled(void)
        return static_key_false(&memcg_kmem_enabled_key);
 }
 
+bool memcg_kmem_is_active(struct mem_cgroup *memcg);
+
 /*
  * In general, we'll do everything in our power to not incur in any overhead
  * for non-memcg users for the kmem functions. Not even a function call, if we
@@ -542,6 +544,11 @@ static inline bool memcg_kmem_enabled(void)
        return false;
 }
 
+static inline bool memcg_kmem_is_active(struct mem_cgroup *memcg)
+{
+       return false;
+}
+
 static inline bool
 memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **memcg, int order)
 {
index a4d24f3c5430fefe78090cec27ff2bfe5992dcaa..af4ff88a11e0899e88778edf4b6bf142acb37d70 100644 (file)
@@ -2168,9 +2168,8 @@ int drop_caches_sysctl_handler(struct ctl_table *, int,
                                        void __user *, size_t *, loff_t *);
 #endif
 
-unsigned long shrink_node_slabs(gfp_t gfp_mask, int nid,
-                               unsigned long nr_scanned,
-                               unsigned long nr_eligible);
+void drop_slab(void);
+void drop_slab_node(int nid);
 
 #ifndef CONFIG_MMU
 #define randomize_va_space 0
index f4aee75f00b19449e63c2a48b864e36f11b10ce7..4fcacd915d45adaf914c770f7d04f8b12b2a504a 100644 (file)
@@ -20,6 +20,9 @@ struct shrink_control {
 
        /* current node being shrunk (for NUMA aware shrinkers) */
        int nid;
+
+       /* current memcg being shrunk (for memcg aware shrinkers) */
+       struct mem_cgroup *memcg;
 };
 
 #define SHRINK_STOP (~0UL)
@@ -61,7 +64,8 @@ struct shrinker {
 #define DEFAULT_SEEKS 2 /* A good number if you don't know better. */
 
 /* Flags */
-#define SHRINKER_NUMA_AWARE (1 << 0)
+#define SHRINKER_NUMA_AWARE    (1 << 0)
+#define SHRINKER_MEMCG_AWARE   (1 << 1)
 
 extern int register_shrinker(struct shrinker *);
 extern void unregister_shrinker(struct shrinker *);
index 095c1f96fbecf34c13ea9762140337fdbf316421..3c2a1a8286aca3e89a81e31d60fc1cd57152f94a 100644 (file)
@@ -352,7 +352,7 @@ struct mem_cgroup {
 };
 
 #ifdef CONFIG_MEMCG_KMEM
-static bool memcg_kmem_is_active(struct mem_cgroup *memcg)
+bool memcg_kmem_is_active(struct mem_cgroup *memcg)
 {
        return memcg->kmemcg_id >= 0;
 }
index feb803bf344364132e92bc10d575ba26bdd0b45d..1a735fad2a13120cd9543b926140af6724858e42 100644 (file)
@@ -242,15 +242,8 @@ void shake_page(struct page *p, int access)
         * Only call shrink_node_slabs here (which would also shrink
         * other caches) if access is not potentially fatal.
         */
-       if (access) {
-               int nr;
-               int nid = page_to_nid(p);
-               do {
-                       nr = shrink_node_slabs(GFP_KERNEL, nid, 1000, 1000);
-                       if (page_count(p) == 1)
-                               break;
-               } while (nr > 10);
-       }
+       if (access)
+               drop_slab_node(page_to_nid(p));
 }
 EXPORT_SYMBOL_GPL(shake_page);
 
index 8e645ee520455432e583deb2614199d84e422066..803886b8e35368d9e03adf02e90262f94518515c 100644 (file)
@@ -232,10 +232,10 @@ EXPORT_SYMBOL(unregister_shrinker);
 
 #define SHRINK_BATCH 128
 
-static unsigned long shrink_slabs(struct shrink_control *shrinkctl,
-                                 struct shrinker *shrinker,
-                                 unsigned long nr_scanned,
-                                 unsigned long nr_eligible)
+static unsigned long do_shrink_slab(struct shrink_control *shrinkctl,
+                                   struct shrinker *shrinker,
+                                   unsigned long nr_scanned,
+                                   unsigned long nr_eligible)
 {
        unsigned long freed = 0;
        unsigned long long delta;
@@ -344,9 +344,10 @@ static unsigned long shrink_slabs(struct shrink_control *shrinkctl,
 }
 
 /**
- * shrink_node_slabs - shrink slab caches of a given node
+ * shrink_slab - shrink slab caches
  * @gfp_mask: allocation context
  * @nid: node whose slab caches to target
+ * @memcg: memory cgroup whose slab caches to target
  * @nr_scanned: pressure numerator
  * @nr_eligible: pressure denominator
  *
@@ -355,6 +356,12 @@ static unsigned long shrink_slabs(struct shrink_control *shrinkctl,
  * @nid is passed along to shrinkers with SHRINKER_NUMA_AWARE set,
  * unaware shrinkers will receive a node id of 0 instead.
  *
+ * @memcg specifies the memory cgroup to target. If it is not NULL,
+ * only shrinkers with SHRINKER_MEMCG_AWARE set will be called to scan
+ * objects from the memory cgroup specified. Otherwise all shrinkers
+ * are called, and memcg aware shrinkers are supposed to scan the
+ * global list then.
+ *
  * @nr_scanned and @nr_eligible form a ratio that indicate how much of
  * the available objects should be scanned.  Page reclaim for example
  * passes the number of pages scanned and the number of pages on the
@@ -365,13 +372,17 @@ static unsigned long shrink_slabs(struct shrink_control *shrinkctl,
  *
  * Returns the number of reclaimed slab objects.
  */
-unsigned long shrink_node_slabs(gfp_t gfp_mask, int nid,
-                               unsigned long nr_scanned,
-                               unsigned long nr_eligible)
+static unsigned long shrink_slab(gfp_t gfp_mask, int nid,
+                                struct mem_cgroup *memcg,
+                                unsigned long nr_scanned,
+                                unsigned long nr_eligible)
 {
        struct shrinker *shrinker;
        unsigned long freed = 0;
 
+       if (memcg && !memcg_kmem_is_active(memcg))
+               return 0;
+
        if (nr_scanned == 0)
                nr_scanned = SWAP_CLUSTER_MAX;
 
@@ -390,12 +401,16 @@ unsigned long shrink_node_slabs(gfp_t gfp_mask, int nid,
                struct shrink_control sc = {
                        .gfp_mask = gfp_mask,
                        .nid = nid,
+                       .memcg = memcg,
                };
 
+               if (memcg && !(shrinker->flags & SHRINKER_MEMCG_AWARE))
+                       continue;
+
                if (!(shrinker->flags & SHRINKER_NUMA_AWARE))
                        sc.nid = 0;
 
-               freed += shrink_slabs(&sc, shrinker, nr_scanned, nr_eligible);
+               freed += do_shrink_slab(&sc, shrinker, nr_scanned, nr_eligible);
        }
 
        up_read(&shrinker_rwsem);
@@ -404,6 +419,29 @@ out:
        return freed;
 }
 
+void drop_slab_node(int nid)
+{
+       unsigned long freed;
+
+       do {
+               struct mem_cgroup *memcg = NULL;
+
+               freed = 0;
+               do {
+                       freed += shrink_slab(GFP_KERNEL, nid, memcg,
+                                            1000, 1000);
+               } while ((memcg = mem_cgroup_iter(NULL, memcg, NULL)) != NULL);
+       } while (freed > 10);
+}
+
+void drop_slab(void)
+{
+       int nid;
+
+       for_each_online_node(nid)
+               drop_slab_node(nid);
+}
+
 static inline int is_page_cache_freeable(struct page *page)
 {
        /*
@@ -2276,6 +2314,7 @@ static inline bool should_continue_reclaim(struct zone *zone,
 static bool shrink_zone(struct zone *zone, struct scan_control *sc,
                        bool is_classzone)
 {
+       struct reclaim_state *reclaim_state = current->reclaim_state;
        unsigned long nr_reclaimed, nr_scanned;
        bool reclaimable = false;
 
@@ -2294,6 +2333,7 @@ static bool shrink_zone(struct zone *zone, struct scan_control *sc,
                memcg = mem_cgroup_iter(root, NULL, &reclaim);
                do {
                        unsigned long lru_pages;
+                       unsigned long scanned;
                        struct lruvec *lruvec;
                        int swappiness;
 
@@ -2305,10 +2345,16 @@ static bool shrink_zone(struct zone *zone, struct scan_control *sc,
 
                        lruvec = mem_cgroup_zone_lruvec(zone, memcg);
                        swappiness = mem_cgroup_swappiness(memcg);
+                       scanned = sc->nr_scanned;
 
                        shrink_lruvec(lruvec, swappiness, sc, &lru_pages);
                        zone_lru_pages += lru_pages;
 
+                       if (memcg && is_classzone)
+                               shrink_slab(sc->gfp_mask, zone_to_nid(zone),
+                                           memcg, sc->nr_scanned - scanned,
+                                           lru_pages);
+
                        /*
                         * Direct reclaim and kswapd have to scan all memory
                         * cgroups to fulfill the overall scan target for the
@@ -2330,19 +2376,14 @@ static bool shrink_zone(struct zone *zone, struct scan_control *sc,
                 * Shrink the slab caches in the same proportion that
                 * the eligible LRU pages were scanned.
                 */
-               if (global_reclaim(sc) && is_classzone) {
-                       struct reclaim_state *reclaim_state;
-
-                       shrink_node_slabs(sc->gfp_mask, zone_to_nid(zone),
-                                         sc->nr_scanned - nr_scanned,
-                                         zone_lru_pages);
-
-                       reclaim_state = current->reclaim_state;
-                       if (reclaim_state) {
-                               sc->nr_reclaimed +=
-                                       reclaim_state->reclaimed_slab;
-                               reclaim_state->reclaimed_slab = 0;
-                       }
+               if (global_reclaim(sc) && is_classzone)
+                       shrink_slab(sc->gfp_mask, zone_to_nid(zone), NULL,
+                                   sc->nr_scanned - nr_scanned,
+                                   zone_lru_pages);
+
+               if (reclaim_state) {
+                       sc->nr_reclaimed += reclaim_state->reclaimed_slab;
+                       reclaim_state->reclaimed_slab = 0;
                }
 
                vmpressure(sc->gfp_mask, sc->target_mem_cgroup,