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@xxxxxxxxxxxxx> 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@xxxxxxxxxx> > > > Signed-off-by: Suren Baghdasaryan <surenb@xxxxxxxxxx> > > > --- > > > 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, ¶m); > > > + sched_setscheduler_nocheck(kworker->task, SCHED_FIFO, ¶m); > > > > 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. > 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. 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.
