On 13.05.2018 19:47, Vladimir Davydov wrote:
> On Thu, May 10, 2018 at 12:52:36PM +0300, Kirill Tkhai wrote:
>> 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.
>>
>> Signed-off-by: Kirill Tkhai <ktkhai@xxxxxxxxxxxxx>
>> ---
>> include/linux/memcontrol.h | 21 ++++++++
>> mm/memcontrol.c | 116 ++++++++++++++++++++++++++++++++++++++++++++
>> mm/vmscan.c | 16 ++++++
>> 3 files changed, 152 insertions(+), 1 deletion(-)
>>
>> diff --git a/include/linux/memcontrol.h b/include/linux/memcontrol.h
>> index 6cbea2f25a87..e5e7e0fc7158 100644
>> --- a/include/linux/memcontrol.h
>> +++ b/include/linux/memcontrol.h
>> @@ -105,6 +105,17 @@ struct lruvec_stat {
>> long count[NR_VM_NODE_STAT_ITEMS];
>> };
>>
>> +#ifdef CONFIG_MEMCG_SHRINKER
>> +/*
>> + * 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];
>> +};
>> +#endif /* CONFIG_MEMCG_SHRINKER */
>> +
>
> AFAIR we don't normally ifdef structure definitions.
>
>> /*
>> * per-zone information in memory controller.
>> */
>> @@ -118,6 +129,9 @@ struct mem_cgroup_per_node {
>>
>> struct mem_cgroup_reclaim_iter iter[DEF_PRIORITY + 1];
>>
>> +#ifdef CONFIG_MEMCG_SHRINKER
>> + 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*/
>> @@ -1255,4 +1269,11 @@ static inline void memcg_put_cache_ids(void)
>>
>> #endif /* CONFIG_MEMCG && !CONFIG_SLOB */
>>
>> +#ifdef CONFIG_MEMCG_SHRINKER
>
>> +#define MEMCG_SHRINKER_MAP(memcg, nid) (memcg->nodeinfo[nid]->shrinker_map)
>
> I don't really like this helper macro. Accessing shrinker_map directly
> looks cleaner IMO.
>
>> +
>> +extern int memcg_shrinker_nr_max;
>
> As I've mentioned before, the capacity of shrinker map should be a
> private business of memcontrol.c IMHO. We shouldn't use it in vmscan.c
> as max shrinker id, instead we should introduce another variable for
> this, private to vmscan.c.
>
>> +extern int memcg_expand_shrinker_maps(int old_id, int id);
>
> ... Then this function would take just one argument, max id, and would
> update shrinker_map capacity if necessary in memcontrol.c under the
> corresponding mutex, which would look much more readable IMHO as all
> shrinker_map related manipulations would be isolated in memcontrol.c.
>
>> +#endif /* CONFIG_MEMCG_SHRINKER */
>> +
>> #endif /* _LINUX_MEMCONTROL_H */
>> diff --git a/mm/memcontrol.c b/mm/memcontrol.c
>> index 3df3efa7ff40..18e0fdf302a9 100644
>> --- a/mm/memcontrol.c
>> +++ b/mm/memcontrol.c
>> @@ -322,6 +322,116 @@ struct workqueue_struct *memcg_kmem_cache_wq;
>>
>> #endif /* !CONFIG_SLOB */
>>
>> +#ifdef CONFIG_MEMCG_SHRINKER
>> +int memcg_shrinker_nr_max;
>
> memcg_shrinker_map_capacity, may be?
>
>> +static DEFINE_MUTEX(shrinkers_nr_max_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)
>
> If you followed my advice and made the shrinker_map_capacity private to
> memcontrol.c, you wouldn't need to pass old_size here either, just max
> shrinker id.
>
>> +{
>> + struct memcg_shrinker_map *new, *old;
>> + int nid;
>> +
>> + lockdep_assert_held(&shrinkers_nr_max_mutex);
>> +
>> + for_each_node(nid) {
>> + old = rcu_dereference_protected(MEMCG_SHRINKER_MAP(memcg, nid), true);
>> + /* Not yet online memcg */
>> + if (old_size && !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);
>> + if (old)
>> + 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 (memcg == root_mem_cgroup)
>> + return;
>
> Nit: there's mem_cgroup_is_root() helper.
>
>> +
>> + mutex_lock(&shrinkers_nr_max_mutex);
>
> Why do you need to take the mutex here? You don't access shrinker map
> capacity here AFAICS.
Allocation of shrinkers map is in css_online() now, and this wants its pay.
memcg_expand_one_shrinker_map() must be able to differ mem cgroups with
allocated maps, mem cgroups with not allocated maps, and mem cgroups with
failed/failing css_online. So, the mutex is used for synchronization with
expanding. See "old_size && !old" check in memcg_expand_one_shrinker_map().
>> + for_each_node(nid) {
>> + pn = mem_cgroup_nodeinfo(memcg, nid);
>> + map = rcu_dereference_protected(pn->shrinker_map, true);
>> + if (map)
>> + call_rcu(&map->rcu, memcg_free_shrinker_map_rcu);
>> + rcu_assign_pointer(pn->shrinker_map, NULL);
>> + }
>> + mutex_unlock(&shrinkers_nr_max_mutex);
>> +}
>> +
>> +static int memcg_alloc_shrinker_maps(struct mem_cgroup *memcg)
>> +{
>> + int ret, size = memcg_shrinker_nr_max/BITS_PER_BYTE;
>> +
>> + if (memcg == root_mem_cgroup)
>> + return 0;
>
> Nit: mem_cgroup_is_root().
>
>> +
>> + mutex_lock(&shrinkers_nr_max_mutex);
>
>> + ret = memcg_expand_one_shrinker_map(memcg, size, 0);
>
> I don't think it's worth reusing the function designed for reallocating
> shrinker maps for initial allocation. Please just fold the code here -
> it will make both 'alloc' and 'expand' easier to follow IMHO.
These function will have 80% code the same. What are the reasons to duplicate
the same functionality? Two functions are more difficult for support, and
everywhere in kernel we try to avoid this IMHO.
>> + mutex_unlock(&shrinkers_nr_max_mutex);
>> +
>> + if (ret)
>> + memcg_free_shrinker_maps(memcg);
>> +
>> + return ret;
>> +}
>> +
>
>> +static struct idr mem_cgroup_idr;
>
> Stray change.
>
>> +
>> +int memcg_expand_shrinker_maps(int old_nr, int nr)
>> +{
>> + int size, old_size, ret = 0;
>> + struct mem_cgroup *memcg;
>> +
>> + old_size = old_nr / BITS_PER_BYTE;
>> + size = nr / BITS_PER_BYTE;
>> +
>> + mutex_lock(&shrinkers_nr_max_mutex);
>> +
>
>> + if (!root_mem_cgroup)
>> + goto unlock;
>
> This wants a comment.
Which comment does this want? "root_mem_cgroup is not initialized, so it does not have child mem cgroups"?
>> +
>> + for_each_mem_cgroup(memcg) {
>> + if (memcg == root_mem_cgroup)
>
> Nit: mem_cgroup_is_root().
>
>> + continue;
>> + ret = memcg_expand_one_shrinker_map(memcg, size, old_size);
>> + if (ret)
>> + goto unlock;
>> + }
>> +unlock:
>> + mutex_unlock(&shrinkers_nr_max_mutex);
>> + return ret;
>> +}
>> +#else /* CONFIG_MEMCG_SHRINKER */
>> +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_SHRINKER */
>> +
>> /**
>> * mem_cgroup_css_from_page - css of the memcg associated with a page
>> * @page: page of interest
>> @@ -4471,6 +4581,11 @@ static int mem_cgroup_css_online(struct cgroup_subsys_state *css)
>> {
>> struct mem_cgroup *memcg = mem_cgroup_from_css(css);
>>
>> + 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);
>> @@ -4522,6 +4637,7 @@ 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);
>> }
>> diff --git a/mm/vmscan.c b/mm/vmscan.c
>> index d691beac1048..d8a2870710e0 100644
>> --- a/mm/vmscan.c
>> +++ b/mm/vmscan.c
>> @@ -174,12 +174,26 @@ static DEFINE_IDR(shrinker_idr);
>>
>> static int prealloc_memcg_shrinker(struct shrinker *shrinker)
>> {
>> - int id, ret;
>> + int id, nr, ret;
>>
>> down_write(&shrinker_rwsem);
>> ret = id = idr_alloc(&shrinker_idr, shrinker, 0, 0, GFP_KERNEL);
>> if (ret < 0)
>> goto unlock;
>> +
>> + if (id >= memcg_shrinker_nr_max) {
>> + nr = memcg_shrinker_nr_max * 2;
>> + if (nr == 0)
>> + nr = BITS_PER_BYTE;
>> + BUG_ON(id >= nr);
>
> The logic defining shrinker map capacity growth should be private to
> memcontrol.c IMHO.
>
>> +
>> + if (memcg_expand_shrinker_maps(memcg_shrinker_nr_max, nr)) {
>> + idr_remove(&shrinker_idr, id);
>> + goto unlock;
>> + }
>> + memcg_shrinker_nr_max = nr;
>> + }
>> +
>> shrinker->id = id;
>> ret = 0;
>> unlock:
>>
