Re: [PATCH 1/1] psi: do not require setsched permission from the trigger creator

[Suren Baghdasaryan <surenb@google.com>]

Hi Peter,
Thanks for sharing your thoughts. I understand your point and I tend
to agree with it. I originally designed this using watchdog as the
example of a critical system health signal and in the context of
mobile device memory pressure is critical but I agree that there are
more important things in life. I checked and your proposal to change
it to FIFO-1 should still work for our purposes. Will test to make
sure and reply to your patch. Couple clarifications in-line.

On Thu, Aug 1, 2019 at 2:51 AM Peter Zijlstra <peterz@infradead.org> wrote:
>
> On Tue, Jul 30, 2019 at 10:44:51AM -0700, Suren Baghdasaryan wrote:
> > On Tue, Jul 30, 2019 at 1:11 AM Peter Zijlstra <peterz@infradead.org> wrote:
> > >
> > > On Mon, Jul 29, 2019 at 06:33:10PM -0700, Suren Baghdasaryan wrote:
> > > > When a process creates a new trigger by writing into /proc/pressure/*
> > > > files, permissions to write such a file should be used to determine whether
> > > > the process is allowed to do so or not. Current implementation would also
> > > > require such a process to have setsched capability. Setting of psi trigger
> > > > thread's scheduling policy is an implementation detail and should not be
> > > > exposed to the user level. Remove the permission check by using _nocheck
> > > > version of the function.
> > > >
> > > > Suggested-by: Nick Kralevich <nnk@google.com>
> > > > Signed-off-by: Suren Baghdasaryan <surenb@google.com>
> > > > ---
> > > >  kernel/sched/psi.c | 2 +-
> > > >  1 file changed, 1 insertion(+), 1 deletion(-)
> > > >
> > > > diff --git a/kernel/sched/psi.c b/kernel/sched/psi.c
> > > > index 7acc632c3b82..ed9a1d573cb1 100644
> > > > --- a/kernel/sched/psi.c
> > > > +++ b/kernel/sched/psi.c
> > > > @@ -1061,7 +1061,7 @@ struct psi_trigger *psi_trigger_create(struct psi_group *group,
> > > >                       mutex_unlock(&group->trigger_lock);
> > > >                       return ERR_CAST(kworker);
> > > >               }
> > > > -             sched_setscheduler(kworker->task, SCHED_FIFO, &param);
> > > > +             sched_setscheduler_nocheck(kworker->task, SCHED_FIFO, &param);
> > >
> > > ARGGH, wtf is there a FIFO-99!! thread here at all !?
> >
> > We need psi poll_kworker to be an rt-priority thread so that psi
>
> There is a giant difference between 'needs to be higher than OTHER' and
> FIFO-99.
>
> > notifications are delivered to the userspace without delay even when
> > the CPUs are very congested. Otherwise it's easy to delay psi
> > notifications by running a simple CPU hogger executing "chrt -f 50 dd
> > if=/dev/zero of=/dev/null". Because these notifications are
>
> So what; at that point that's exactly what you're asking for. Using RT
> is for those who know what they're doing.
>
> > time-critical for reacting to memory shortages we can't allow for such
> > delays.
>
> Furthermore, actual RT programs will have pre-allocated and locked any
> memory they rely on. They don't give a crap about your pressure
> nonsense.
>

This signal is used not to protect other RT tasks but to monitor
overall system memory health for the sake of system responsiveness.

> > Notice that this kworker is created only if userspace creates a psi
> > trigger. So unless you are using psi triggers you will never see this
> > kthread created.
>
> By marking it FIFO-99 you're in effect saying that your stupid
> statistics gathering is more important than your life. It will preempt
> the task that's in control of the band-saw emergency break, it will
> preempt the task that's adjusting the electromagnetic field containing
> this plasma flow.
>
> That's insane.

IMHO an opt-in feature stops being "stupid" as soon as the user opted
in to use it, therefore explicitly indicating interest in it. However
I assume you are using "stupid" here to indicate that it's "less
important" rather than it's "useless".

> I'm going to queue a patch to reduce this to FIFO-1, that will preempt
> regular OTHER tasks but will not perturb (much) actual RT bits.
>

Thanks for posting the fix.

> --
> To unsubscribe from this group and stop receiving emails from it, send an email to kernel-team+unsubscribe@android.com.
>