Re: [Resend PATCH 1/6] mm/memcg: warning on !memcg after readahead page charged

[Michal Hocko]

On Fri 21-08-20 15:48:44, Michal Hocko wrote:
> On Fri 21-08-20 08:39:37, Qian Cai wrote:
> > On Fri, Aug 21, 2020 at 10:01:27AM +0200, Michal Hocko wrote:
> > > On Thu 20-08-20 10:58:51, Qian Cai wrote:
> > > > On Tue, Aug 11, 2020 at 07:10:27PM +0800, Alex Shi wrote:
> > > > > Since readahead page is charged on memcg too, in theory we don't have to
> > > > > check this exception now. Before safely remove them all, add a warning
> > > > > for the unexpected !memcg.
> > > > > 
> > > > > Signed-off-by: Alex Shi <alex.shi@xxxxxxxxxxxxxxxxx>
> > > > > Acked-by: Michal Hocko <mhocko@xxxxxxxx>
> > > > 
> > > > This will trigger,
> > > 
> > > Thanks for the report!
> > >  
> > > > [ 1863.916499] LTP: starting move_pages12
> > > > [ 1863.946520] page:000000008ccc1062 refcount:1 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x1fd3c0
> > > > [ 1863.946553] head:000000008ccc1062 order:5 compound_mapcount:0 compound_pincount:0
> > > 
> > > Hmm, this is really unexpected. How did we get order-5 page here? Is
> > > this some special mappaing that sys_move_pages should just ignore?
> > 
> > Well, I thought everybody should be able to figure out where to find the LTP
> > tests source code at this stage to see what it does. Anyway, the test simply
> > migrate hugepages while soft offlining, so order 5 is expected as that is 2M
> > hugepage on powerpc (also reproduced on x86 below). It might be easier to
> > reproduce using our linux-mm random bug collection on NUMA systems.
> 
> OK, I must have missed that this was on ppc. The order makes more sense
> now. I will have a look at this next week.

OK, so I've had a look and I know what's going on there. The
move_pages12 is migrating hugetlb pages. Those are not charged to any
memcg. We have completely missed this case. There are two ways going
around that. Drop the warning and update the comment so that we do not
forget about that or special case hugetlb pages.

I think the first option is better.
-- 
Michal Hocko
SUSE Labs