Re: [PATCH] mm: split thp synchronously on MADV_DONTNEED

[Shakeel Butt <shakeelb@google.com>]

On Sun, Nov 21, 2021 at 4:50 PM Kirill A. Shutemov <kirill@shutemov.name> wrote:
>
[...]
>
> Have you considered impact on short-living tasks where paying splitting
> tax would hurt performace without any benefits? Maybe a sparete madvise
> opration needed? I donno.
>

Do you have a concrete example of such short-living applications doing
MADV_DONTNEED? I can try to get some numbers to measure the impact.

Regarding the new advice option, I did give some thought to it but
decided against it based on the reason that we should not be exposing
some low level kernel implementation detail to users through a stable
API.

[...]
> > diff --git a/include/linux/sched.h b/include/linux/sched.h
> > index 9d27fd0ce5df..a984bb6509d9 100644
> > --- a/include/linux/sched.h
> > +++ b/include/linux/sched.h
> > @@ -1412,6 +1412,10 @@ struct task_struct {
> >       struct mem_cgroup               *active_memcg;
> >  #endif
> >
> > +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
> > +     struct list_head                *deferred_split_list;
> > +#endif
> > +
> >  #ifdef CONFIG_BLK_CGROUP
> >       struct request_queue            *throttle_queue;
> >  #endif
>
> It looks dirty. We really don't have options to pass it down?
>
> Maybe passdown the list via zap_details and call a new rmap remove helper
> if the list is present?
>

We already have precedence on using this technique for other cases but
let me take a stab at passing the list through zap_details and see how
that looks.

> >
> > +void split_local_deferred_list(struct list_head *defer_list)
[...]
> Looks like a lot of copy-paste from deferred_split_scan(). Can we get them
> consolidated?

I will see what I can do.

Thanks for the review.
Shakeel