Re: [PATCH V3 04/10] sched/core: Prevent race condition between cpuset and __sched_setscheduler()

From: Juri Lelli
Date: Wed Feb 14 2018 - 11:31:55 EST


On 14/02/18 08:33, Mathieu Poirier wrote:
> On 14 February 2018 at 04:27, Juri Lelli <juri.lelli@xxxxxxxxxx> wrote:
> > On 14/02/18 11:49, Juri Lelli wrote:
> >> On 14/02/18 11:36, Juri Lelli wrote:
> >> > Hi Mathieu,
> >> >
> >> > On 13/02/18 13:32, Mathieu Poirier wrote:
> >> > > No synchronisation mechanism exist between the cpuset subsystem and calls
> >> > > to function __sched_setscheduler(). As such it is possible that new root
> >> > > domains are created on the cpuset side while a deadline acceptance test
> >> > > is carried out in __sched_setscheduler(), leading to a potential oversell
> >> > > of CPU bandwidth.
> >> > >
> >> > > By making available the cpuset_mutex to the core scheduler it is possible
> >> > > to prevent situations such as the one described above from happening.
> >> > >
> >> > > Signed-off-by: Mathieu Poirier <mathieu.poirier@xxxxxxxxxx>
> >> > > ---
> >> >
> >> > [...]
> >> >
> >> > > diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> >> > > index f727c3d0064c..0d8badcf1f0f 100644
> >> > > --- a/kernel/sched/core.c
> >> > > +++ b/kernel/sched/core.c
> >> > > @@ -4176,6 +4176,13 @@ static int __sched_setscheduler(struct task_struct *p,
> >> > > }
> >> > >
> >> > > /*
> >> > > + * Make sure we don't race with the cpuset subsystem where root
> >> > > + * domains can be rebuilt or modified while operations like DL
> >> > > + * admission checks are carried out.
> >> > > + */
> >> > > + cpuset_lock();
> >> > > +
> >> > > + /*
> >> >
> >> > Mmm, I'm afraid we can't do this. __sched_setscheduler might be called
> >> > from interrupt contex by normalize_rt_tasks().
> >>
> >> Maybe conditionally grabbing it if pi is true could do? I guess we don't
> >> care much about domains when sysrq.
> >
> > Ops.. just got this. :/
>
>
> Arrghhh... Back to the drawing board.
>

Eh.. even though the warning simply happens because unlocking of
cpuset lock is missing

--->8---
@@ -4312,6 +4312,7 @@ static int __sched_setscheduler(struct task_struct *p,
/* Avoid rq from going away on us: */
preempt_disable();
task_rq_unlock(rq, p, &rf);
+ cpuset_unlock();

if (pi)
rt_mutex_adjust_pi(p);
--->8---

Still grabbing it is a no-go, as do_sched_setscheduler calls
sched_setscheduler from inside an RCU read-side critical section.

So, back to the drawing board indeed. :/

Thanks,

- Juri