From mboxrd@z Thu Jan  1 00:00:00 1970
From: tglx@linutronix.de (Thomas Gleixner)
Date: Fri, 3 Jul 2015 16:17:13 +0200 (CEST)
Subject: Possible regression due to "tick: broadcast: Prevent livelock
 from event handler"
In-Reply-To: <20150703133245.GA19284@katana>
References: <20150703024044.GB24695@verge.net.au>
 <20150703105441.GF1521@katana> <alpine.DEB.2.11.1507031511180.3916@nanos>
 <20150703133245.GA19284@katana>
Message-ID: <alpine.DEB.2.11.1507031612410.3916@nanos>
To: linux-arm-kernel@lists.infradead.org
List-Id: linux-arm-kernel.lists.infradead.org

On Fri, 3 Jul 2015, Wolfram Sang wrote:

> 
> > So with high res timers it boots. Can you please provide the output of
> > /proc/timer_list for that case?

> Tick Device: mode:     1
> Per CPU device: 0
> Clock Event Device: e0180000.timer
>  max_delta_ns:   131071523464982
>  min_delta_ns:   61035
>  mult:           70369
>  shift:          31
>  mode:           3
>  next_event:     40760000000 nsecs
>  set_next_event: em_sti_clock_event_next
>  set_mode:       em_sti_clock_event_mode
>  event_handler:  hrtimer_interrupt
>  retries:        0

So this is a single core machine and uses the em_sti timer w/o the
broadcast nonsense. In Simons case it looks like em_sti is used as
broadcast device.

Though the issues you see in the highres=n case might be the same as
the ones Simon is observing.

So in that nohz=y highres=n case, does adding idle=poll on the command
line fix the issue?

Thanks,

	tglx