Re: Severe performance regression w/ 4.4+ on Android due to cgroup locking changes

[Paul E. McKenney]

On Wed, Jul 13, 2016 at 11:13:26AM -0700, Dmitry Shmidt wrote:
> On Wed, Jul 13, 2016 at 7:42 AM, Paul E. McKenney
> <paulmck@xxxxxxxxxxxxxxxxxx> wrote:
> > On Wed, Jul 13, 2016 at 10:21:12AM +0200, Peter Zijlstra wrote:
> >> On Tue, Jul 12, 2016 at 05:00:04PM -0700, John Stultz wrote:
> >> > Hey Tejun,
> >> >
> >> >   So Dmitry Shmidt recently noticed that with 4.4 based systems we're
> >> > seeing quite a bit of performance overhead from
> >> > __cgroup_procs_write().
> >> >
> >> > With 4.4 tree as it stands, we're seeing __cgroup_procs_write() quite
> >> > often take 10s of miliseconds to execute (with max times up in the
> >> > 80ms range).
> >> >
> >> > While with 4.1 it was quite often in the single usec range, and max
> >> > time values still in in sub-milisecond range.
> >> >
> >> > The majority of these performance regressions seem to come from the
> >> > locking changes in:
> >> >
> >> > 3014dde762f6 ("cgroup: simplify threadgroup locking")
> >> > and
> >> > 1ed1328792ff  ("sched, cgroup: replace signal_struct->group_rwsem with
> >> > a global percpu_rwsem")
> >> >
> >> > Dmitry has found that by reverting these two changes (which don't
> >> > revert easiliy), we can get back down to tens 10-100 usec range for
> >> > most calls, with max values occasionally spiking to ~18ms.
> >> >
> >> > Those two commits do talk about performance regressions, that were
> >> > supposedly alleviated by percpu_rwsem changes, but I'm not sure we are
> >> > seeing this.
> >>
> >> Do you have 'funny' RCU options that quickly force a grace period when
> >> you go idle or something?
> >>
> >> But yes, it does not surprise me to find this commit is causing
> >> problems.
> >
> > Hmmm...  Looks like RCU is present both before and after.  But please
> > do send along your .config.
> 
> Attached

No funny RCU Kconfig options set -- vanilla preemptible RCU.

> > Speaking of .config, is CONFIG_PREEMPT=y?  If so, does the workload
> > feature preemption and migration?  If that is the case, you might be
> > seeing contention on the per-CPU cgroup_threadgroup_rwsem, given that
> > the second patch seems to be adding acquisitions.
> 
> CONFIG_PREEMPT=y is set.
> We see this issue during the boot, so it supposes to be enough CPU load to
> cause preemption and migration.

How early during boot?  Presumably after the scheduler has started...

							Thanx, Paul