On Fri 26-02-21 16:56:28, Tim Chen wrote:

On 2/26/21 12:52 AM, Michal Hocko wrote:

quoted

quoted

Michal,

Let's take an extreme case where memcg 1 always generate the
first event and memcg 2 generates the rest of 128*8-1 events
and the pattern repeat.

I do not follow. Events are per-memcg, aren't they?
	__this_cpu_read(memcg->vmstats_percpu->targets[target]);
	[...]
	__this_cpu_write(memcg->vmstats_percpu->targets[target], next);

You are right. My previous reasoning is incorrect as the sampling is done per memcg.
I'll do some additional debugging on why memcg is not on the tree.

OK, thanks for the confirmation. I think we want to do 2 things. Remove
the soft limit specific threshold and stay with a single one and
recognize THPs.
-- 
Michal Hocko
SUSE Labs