Re: [RFC 1/1] mm: page_alloc: replace mm_percpu_wq with kthreads in drain_all_pages

[Petr Mladek <pmladek@suse.com>]

On Tue 2022-03-01 13:12:19, Suren Baghdasaryan wrote:
> On Tue, Mar 1, 2022 at 4:25 AM Petr Mladek <pmladek@suse.com> wrote:
> >
> > On Thu 2022-02-24 17:28:19, Suren Baghdasaryan wrote:
> > > Sending as an RFC to confirm if this is the right direction and to
> > > clarify if other tasks currently executed on mm_percpu_wq should be
> > > also moved to kthreads. The patch seems stable in testing but I want
> > > to collect more performance data before submitting a non-RFC version.
> > >
> > >
> > > Currently drain_all_pages uses mm_percpu_wq to drain pages from pcp
> > > list during direct reclaim. The tasks on a workqueue can be delayed
> > > by other tasks in the workqueues using the same per-cpu worker pool.
> > > This results in sizable delays in drain_all_pages when cpus are highly
> > > contended.
> > > Memory management operations designed to relieve memory pressure should
> > > not be allowed to block by other tasks, especially if the task in direct
> > > reclaim has higher priority than the blocking tasks.
> > > Replace the usage of mm_percpu_wq with per-cpu low priority FIFO
> > > kthreads to execute draining tasks.
> > >
> > > Suggested-by: Petr Mladek <pmladek@suse.com>
> > > Signed-off-by: Suren Baghdasaryan <surenb@google.com>
> >
> > The patch looks good to me. See few comments below about things
> > where I was in doubts. But I do not see any real problem with
> > this approach.
> 
> Thanks for the review, Petr. One question inline.

Answering just this question.

> > > diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> > > index 3589febc6d31..c9ab2cf4b05b 100644
> > > --- a/mm/page_alloc.c
> > > +++ b/mm/page_alloc.c
> > > +static void __init init_drain_workers(void)
> > > +{
> > > +     unsigned int cpu;
> > > +
> > > +     for_each_online_cpu(cpu)
> > > +             alloc_drain_worker(cpu);
> >
> > I though whether this need to be called under cpus_read_lock();
> > And I think that the code should be safe as it is. There
> > is this call chain:
> >
> >   + kernel_init_freeable()
> >     + page_alloc_init_late()
> >       + init_drain_workers()
> >
> > It is called after smp_init() but before the init process
> > is executed. I guess that nobody could trigger CPU hotplug
> > at this state. So there there is no need to synchronize
> > against it.
> 
> Should I add a comment here to describe why we don't need
> cpus_read_lock here (due to init process not being active at this
> time)?

I would add the comment. That said, I hope that I am right and
lock is not really needed ;-)

Best Regards,
Petr