]> git.proxmox.com Git - mirror_ubuntu-kernels.git/commitdiff
mm, memcg: assign memcg-aware shrinkers bitmap to memcg
authorKirill Tkhai <ktkhai@virtuozzo.com>
Fri, 17 Aug 2018 22:47:37 +0000 (15:47 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 17 Aug 2018 23:20:30 +0000 (16:20 -0700)
Imagine a big node with many cpus, memory cgroups and containers.  Let
we have 200 containers, every container has 10 mounts, and 10 cgroups.
All container tasks don't touch foreign containers mounts.  If there is
intensive pages write, and global reclaim happens, a writing task has to
iterate over all memcgs to shrink slab, before it's able to go to
shrink_page_list().

Iteration over all the memcg slabs is very expensive: the task has to
visit 200 * 10 = 2000 shrinkers for every memcg, and since there are
2000 memcgs, the total calls are 2000 * 2000 = 4000000.

So, the shrinker makes 4 million do_shrink_slab() calls just to try to
isolate SWAP_CLUSTER_MAX pages in one of the actively writing memcg via
shrink_page_list().  I've observed a node spending almost 100% in
kernel, making useless iteration over already shrinked slab.

This patch adds bitmap of memcg-aware shrinkers to memcg.  The size of
the bitmap depends on bitmap_nr_ids, and during memcg life it's
maintained to be enough to fit bitmap_nr_ids shrinkers.  Every bit in
the map is related to corresponding shrinker id.

Next patches will maintain set bit only for really charged memcg.  This
will allow shrink_slab() to increase its performance in significant way.
See the last patch for the numbers.

[ktkhai@virtuozzo.com: v9]
Link: http://lkml.kernel.org/r/153112549031.4097.3576147070498769979.stgit@localhost.localdomain
[ktkhai@virtuozzo.com: add comment to mem_cgroup_css_online()]
Link: http://lkml.kernel.org/r/521f9e5f-c436-b388-fe83-4dc870bfb489@virtuozzo.com
Link: http://lkml.kernel.org/r/153063056619.1818.12550500883688681076.stgit@localhost.localdomain
Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Vladimir Davydov <vdavydov.dev@gmail.com>
Tested-by: Shakeel Butt <shakeelb@google.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Guenter Roeck <linux@roeck-us.net>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Josef Bacik <jbacik@fb.com>
Cc: Li RongQing <lirongqing@baidu.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Matthias Kaehlcke <mka@chromium.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Philippe Ombredanne <pombredanne@nexb.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Sahitya Tummala <stummala@codeaurora.org>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Waiman Long <longman@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
include/linux/memcontrol.h
mm/memcontrol.c
mm/vmscan.c

index f3c026df7443de7ffb707a7a74c47d365c3d89eb..2cccbb9e1b3ec9790cbaed3f9c82de7f7f2bd0a4 100644 (file)
@@ -111,6 +111,15 @@ struct lruvec_stat {
        long count[NR_VM_NODE_STAT_ITEMS];
 };
 
+/*
+ * Bitmap of shrinker::id corresponding to memcg-aware shrinkers,
+ * which have elements charged to this memcg.
+ */
+struct memcg_shrinker_map {
+       struct rcu_head rcu;
+       unsigned long map[0];
+};
+
 /*
  * per-zone information in memory controller.
  */
@@ -124,6 +133,9 @@ struct mem_cgroup_per_node {
 
        struct mem_cgroup_reclaim_iter  iter[DEF_PRIORITY + 1];
 
+#ifdef CONFIG_MEMCG_KMEM
+       struct memcg_shrinker_map __rcu *shrinker_map;
+#endif
        struct rb_node          tree_node;      /* RB tree node */
        unsigned long           usage_in_excess;/* Set to the value by which */
                                                /* the soft limit is exceeded*/
@@ -1262,6 +1274,8 @@ static inline int memcg_cache_id(struct mem_cgroup *memcg)
        return memcg ? memcg->kmemcg_id : -1;
 }
 
+extern int memcg_expand_shrinker_maps(int new_id);
+
 #else
 #define for_each_memcg_cache_index(_idx)       \
        for (; NULL; )
index 313355dddf660ccc3a09928bcc1acdbcb032f2ce..827c9e87ca081a4ebfcb6b1f466fb8cb8d52f1a4 100644 (file)
@@ -320,6 +320,119 @@ EXPORT_SYMBOL(memcg_kmem_enabled_key);
 
 struct workqueue_struct *memcg_kmem_cache_wq;
 
+static int memcg_shrinker_map_size;
+static DEFINE_MUTEX(memcg_shrinker_map_mutex);
+
+static void memcg_free_shrinker_map_rcu(struct rcu_head *head)
+{
+       kvfree(container_of(head, struct memcg_shrinker_map, rcu));
+}
+
+static int memcg_expand_one_shrinker_map(struct mem_cgroup *memcg,
+                                        int size, int old_size)
+{
+       struct memcg_shrinker_map *new, *old;
+       int nid;
+
+       lockdep_assert_held(&memcg_shrinker_map_mutex);
+
+       for_each_node(nid) {
+               old = rcu_dereference_protected(
+                       mem_cgroup_nodeinfo(memcg, nid)->shrinker_map, true);
+               /* Not yet online memcg */
+               if (!old)
+                       return 0;
+
+               new = kvmalloc(sizeof(*new) + size, GFP_KERNEL);
+               if (!new)
+                       return -ENOMEM;
+
+               /* Set all old bits, clear all new bits */
+               memset(new->map, (int)0xff, old_size);
+               memset((void *)new->map + old_size, 0, size - old_size);
+
+               rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, new);
+               call_rcu(&old->rcu, memcg_free_shrinker_map_rcu);
+       }
+
+       return 0;
+}
+
+static void memcg_free_shrinker_maps(struct mem_cgroup *memcg)
+{
+       struct mem_cgroup_per_node *pn;
+       struct memcg_shrinker_map *map;
+       int nid;
+
+       if (mem_cgroup_is_root(memcg))
+               return;
+
+       for_each_node(nid) {
+               pn = mem_cgroup_nodeinfo(memcg, nid);
+               map = rcu_dereference_protected(pn->shrinker_map, true);
+               if (map)
+                       kvfree(map);
+               rcu_assign_pointer(pn->shrinker_map, NULL);
+       }
+}
+
+static int memcg_alloc_shrinker_maps(struct mem_cgroup *memcg)
+{
+       struct memcg_shrinker_map *map;
+       int nid, size, ret = 0;
+
+       if (mem_cgroup_is_root(memcg))
+               return 0;
+
+       mutex_lock(&memcg_shrinker_map_mutex);
+       size = memcg_shrinker_map_size;
+       for_each_node(nid) {
+               map = kvzalloc(sizeof(*map) + size, GFP_KERNEL);
+               if (!map) {
+                       memcg_free_shrinker_maps(memcg);
+                       ret = -ENOMEM;
+                       break;
+               }
+               rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, map);
+       }
+       mutex_unlock(&memcg_shrinker_map_mutex);
+
+       return ret;
+}
+
+int memcg_expand_shrinker_maps(int new_id)
+{
+       int size, old_size, ret = 0;
+       struct mem_cgroup *memcg;
+
+       size = DIV_ROUND_UP(new_id + 1, BITS_PER_LONG) * sizeof(unsigned long);
+       old_size = memcg_shrinker_map_size;
+       if (size <= old_size)
+               return 0;
+
+       mutex_lock(&memcg_shrinker_map_mutex);
+       if (!root_mem_cgroup)
+               goto unlock;
+
+       for_each_mem_cgroup(memcg) {
+               if (mem_cgroup_is_root(memcg))
+                       continue;
+               ret = memcg_expand_one_shrinker_map(memcg, size, old_size);
+               if (ret)
+                       goto unlock;
+       }
+unlock:
+       if (!ret)
+               memcg_shrinker_map_size = size;
+       mutex_unlock(&memcg_shrinker_map_mutex);
+       return ret;
+}
+#else /* CONFIG_MEMCG_KMEM */
+static int memcg_alloc_shrinker_maps(struct mem_cgroup *memcg)
+{
+       return 0;
+}
+static void memcg_free_shrinker_maps(struct mem_cgroup *memcg) { }
 #endif /* CONFIG_MEMCG_KMEM */
 
 /**
@@ -4356,6 +4469,16 @@ static int mem_cgroup_css_online(struct cgroup_subsys_state *css)
 {
        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
 
+       /*
+        * A memcg must be visible for memcg_expand_shrinker_maps()
+        * by the time the maps are allocated. So, we allocate maps
+        * here, when for_each_mem_cgroup() can't skip it.
+        */
+       if (memcg_alloc_shrinker_maps(memcg)) {
+               mem_cgroup_id_remove(memcg);
+               return -ENOMEM;
+       }
+
        /* Online state pins memcg ID, memcg ID pins CSS */
        atomic_set(&memcg->id.ref, 1);
        css_get(css);
@@ -4408,6 +4531,7 @@ static void mem_cgroup_css_free(struct cgroup_subsys_state *css)
        vmpressure_cleanup(&memcg->vmpressure);
        cancel_work_sync(&memcg->high_work);
        mem_cgroup_remove_from_trees(memcg);
+       memcg_free_shrinker_maps(memcg);
        memcg_free_kmem(memcg);
        mem_cgroup_free(memcg);
 }
index 5cb4f779ea4a95169fbdf8c510de5848da4aa1ef..db0970ba340dedd3fafbfc3a04b96357b12a7398 100644 (file)
@@ -183,8 +183,14 @@ static int prealloc_memcg_shrinker(struct shrinker *shrinker)
        if (id < 0)
                goto unlock;
 
-       if (id >= shrinker_nr_max)
+       if (id >= shrinker_nr_max) {
+               if (memcg_expand_shrinker_maps(id)) {
+                       idr_remove(&shrinker_idr, id);
+                       goto unlock;
+               }
+
                shrinker_nr_max = id + 1;
+       }
        shrinker->id = id;
        ret = 0;
 unlock: