Re: [rfc patch v2] rt,nohz_full: fix nohz_full for PREEMPT_RT_FULL

[Mike Galbraith <umgwanakikbuti@gmail.com>]

On Mon, 2015-03-16 at 21:24 +0100, Sebastian Andrzej Siewior wrote:
> * Mike Galbraith | 2015-03-13 05:53:25 [+0100]:
> 
> >First of all, a task being ticked and trying to shut the tick down will
> >fail to do so due to having just awakened ksoftirqd, so let ksoftirqd
> >try to do that after SOFTIRQ_TIMER processing.  Secondly, should the
> >tick be shut down, we may livelock in hrtimer-cancel() because in -rt
> >a callback may be running.  Break the loop, and let tick_nohz_restart()
> >know that the timer is busy so it can bail.
> 
> I am a bit undecided on that one. I included it in the series but did
> not enable it yet.
> Just so we are on the same page here: you boot your machine with
> something like
>     "isolcpus=1-31 rcu_nocbs=1-31 nohz_full=1-31"
> and pin all kernel threads to CPU0, right?

No, I only declare the nohz_full set, do the isolation via cpusets.

> What you do is that you accept the fact that the timer-softirq is
> scheduled for no reason and then you try to disable the tick from within
> the timer-softirq. I assumed that it would work get the "expired timer"
> somehow.

Yup, it works around that otherwise crippling wakeup.  If I re-apply..

     timers-do-not-raise-softirq-unconditionally.patch

..the workaround is not needed of course, but the livelock fix still is.
I haven't yet tested that in 3.18-rt though, only 4.0-rt, but I presume
it'll be the same deal there when I do.

	-Mike