Re: [PATCH 09/10] mm: Iterate only over charged shrinkers during memcg shrink_slab()

[Kirill Tkhai]

On 24.03.2018 23:11, Vladimir Davydov wrote:
> On Wed, Mar 21, 2018 at 04:22:51PM +0300, Kirill Tkhai wrote:
>> Using the preparations made in previous patches, in case of memcg
>> shrink, we may avoid shrinkers, which are not set in memcg's shrinkers
>> bitmap. To do that, we separate iterations over memcg-aware and
>> !memcg-aware shrinkers, and memcg-aware shrinkers are chosen
>> via for_each_set_bit() from the bitmap. In case of big nodes,
>> having many isolated environments, this gives significant
>> performance growth. See next patch for the details.
>>
>> Note, that the patch does not respect to empty memcg shrinkers,
>> since we never clear the bitmap bits after we set it once.
>> Their shrinkers will be called again, with no shrinked objects
>> as result. This functionality is provided by next patch.
>>
>> Signed-off-by: Kirill Tkhai <ktkhai@xxxxxxxxxxxxx>
>> ---
>>  mm/vmscan.c |   54 +++++++++++++++++++++++++++++++++++++++++-------------
>>  1 file changed, 41 insertions(+), 13 deletions(-)
>>
>> diff --git a/mm/vmscan.c b/mm/vmscan.c
>> index 265cf069b470..e1fd16bc7a9b 100644
>> --- a/mm/vmscan.c
>> +++ b/mm/vmscan.c
>> @@ -327,6 +327,8 @@ static int alloc_shrinker_id(struct shrinker *shrinker)
>>  
>>  	if (!(shrinker->flags & SHRINKER_MEMCG_AWARE))
>>  		return 0;
>> +	BUG_ON(!(shrinker->flags & SHRINKER_NUMA_AWARE));
>> +
>>  retry:
>>  	ida_pre_get(&bitmap_id_ida, GFP_KERNEL);
>>  	down_write(&bitmap_rwsem);
>> @@ -366,7 +368,8 @@ static void add_shrinker(struct shrinker *shrinker)
>>  	down_write(&shrinker_rwsem);
>>  	if (shrinker->flags & SHRINKER_MEMCG_AWARE)
>>  		mcg_shrinkers[shrinker->id] = shrinker;
>> -	list_add_tail(&shrinker->list, &shrinker_list);
>> +	else
>> +		list_add_tail(&shrinker->list, &shrinker_list);
> 
> I don't think we should remove per-memcg shrinkers from the global
> shrinker list - this is confusing. It won't be critical if we iterate
> over all shrinkers on global reclaim, will it?

It depends on number of all shrinkers. And this is excess actions, we have to do.
Accessing their memory we flush cpu caches in some way. So, if there is no a reason
we really need them, I'd removed them from the list.

>>  	up_write(&shrinker_rwsem);
>>  }
>>  
>> @@ -701,6 +705,39 @@ static unsigned long shrink_slab(gfp_t gfp_mask, int nid,
>>  	if (!down_read_trylock(&shrinker_rwsem))
>>  		goto out;
>>  
>> +#if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
>> +	if (!memcg_kmem_enabled() || memcg) {
>> +		struct shrinkers_map *map;
>> +		int i;
>> +
>> +		map = rcu_dereference_protected(SHRINKERS_MAP(memcg), true);
>> +		if (map) {
>> +			for_each_set_bit(i, map->map[nid], bitmap_nr_ids) {
>> +				struct shrink_control sc = {
>> +					.gfp_mask = gfp_mask,
>> +					.nid = nid,
>> +					.memcg = memcg,
>> +				};
>> +
>> +				shrinker = mcg_shrinkers[i];
>> +				if (!shrinker) {
>> +					clear_bit(i, map->map[nid]);
>> +					continue;
>> +				}
>> +				freed += do_shrink_slab(&sc, shrinker, priority);
>> +
>> +				if (rwsem_is_contended(&shrinker_rwsem)) {
>> +					freed = freed ? : 1;
>> +					goto unlock;
>> +				}
>> +			}
>> +		}
>> +
>> +		if (memcg_kmem_enabled() && memcg)
>> +			goto unlock;
> 
> May be, factor this out to a separate function, say shrink_slab_memcg?
> Just for the sake of code legibility.

Good idea, thanks.

Kirill