Re: [RFC PATCH] arm64: defconfig: Disable fine-grained task level IRQ time accounting

From: Vladimir Oltean
Date: Mon Aug 03 2020 - 07:45:44 EST


Hi Thomas,

On Mon, Aug 03, 2020 at 12:49:36PM +0200, Thomas Gleixner wrote:
> Kurt,
>
> Kurt Kanzenbach <kurt.kanzenbach@xxxxxxxxxxxxx> writes:
> > On Thu Jul 30 2020, Vladimir Oltean wrote:
> > OK. I've reproduced it on a Marvell Armada SoC with v5.6 mainline. See
> > splats below. Running with irq time accounting enabled, kills the
> > machine immediately. However, I'm not getting the possible deadlock
> > warnings in 8250 as you did. So that might be unrelated.
> >
> > Unfortunately I have no idea what to debug here.
>
> lets look at the facts we have so far.
>
> 1) When irq accounting is disabled, RT throttling kicks in as
> expected.
>
> 2) With irq accounting the RT throttler does not kick in and the RCU
> stall/lockups happen.
>
> Not much, but there is clearly interaction between irq time accounting
> and scheduler accounting.
>
> Can you please reduce /proc/sys/kernel/sched_rt_runtime_us in steps of
> 50000? At least in theory, reduction should bring the throttling back.
>
> Thanks,
>
> tglx
>

We already know from my original report that reducing
sched_rt_runtime_us to 80% (800000) gives us a stable system.
https://lkml.org/lkml/2020/6/4/1062

What is this telling us?

Thanks,
-Vladimir