Re: [External] Re: [PATCH v4 4/6] mm: hugetlb: retry dissolve page when hitting race

[Michal Hocko]

On Wed 13-01-21 19:11:06, Muchun Song wrote:
> On Wed, Jan 13, 2021 at 6:38 PM Michal Hocko <mhocko@xxxxxxxx> wrote:
[...]
> > > I just want the fix patch to be small enough.
> > > So I do the retry in this patch. If you do not agree with this. I
> > > will fold this into the previous patch.
> > >
> > > Do you mean this?
> > >
> > > cpu_relax();
> > > cond_resched();
> > > cpu_relax();
> >
> > No, I am questiong the use of cpu_relax. What is the point?
> 
> If there is no task to be scheduled. Here is just a while loop.
> The cpu_relax is a good thing to insert into busy-wait loops,
> right?

Well in an ideal world we would simply have a way to block and wait for
the particular page. This is probably an overkill for a rare event like
this. And while you are right that theoretically there might be nobody
else to run but I find it rather unlikely considering that this path is
racing with somebody. Sure there is even less likely possibility that
the race is actually waiting for worker context but really I would just
make it simple retry loop. If we ever hit a real busy loop then this
would be pretty straightforward to spot and fix up.

It's not like I am against the patch with cpu_relax but I find it
excessive for this purpose. If you feel strongly then just keep it.

Once the two patches are squashed I will ack it.
-- 
Michal Hocko
SUSE Labs