Re: [PATCH] mm-vmscan-dont-mess-with-pgdat-flags-in-memcg-reclaim-v2-fix

[Andrey Ryabinin]

On 04/06/2018 05:37 PM, Shakeel Butt wrote:

>>
>> @@ -2482,7 +2494,7 @@ static inline bool should_continue_reclaim(struct pglist_data *pgdat,
>>  static bool pgdat_memcg_congested(pg_data_t *pgdat, struct mem_cgroup *memcg)
>>  {
>>         return test_bit(PGDAT_CONGESTED, &pgdat->flags) ||
>> -               (memcg && test_memcg_bit(PGDAT_CONGESTED, memcg));
>> +               (memcg && memcg_congested(pgdat, memcg));
> 
> I am wondering if we should check all ancestors for congestion as
> well. Maybe a parallel memcg reclaimer might have set some ancestor of
> this memcg to congested.
> 

Why? If ancestor is congested but its child (the one we currently reclaim) is not,
it could mean only 2 things:
 - Either child use mostly anon and inactive file lru is small (file_lru >> priority == 0)
   so it's not congested.
 - Or the child was congested recently (at the time when ancestor scanned this group),
   but not anymore. So the information from ancestor is simply outdated.