Date: Mon, 13 Dec 2010 12:19:00 +0100
From: Markus Trippelsdorf <markus@trippelsdorf.de>
To: Simon Kirby <sim@hostway.ca>
Cc: linux-kernel@vger.kernel.org, john stultz <johnstul@us.ibm.com>,
        Thomas Gleixner <tglx@linutronix.de>, Borislav Petkov <bp@alien8.de>,
        Linus Torvalds <torvalds@linux-foundation.org>
Subject: Re: [2.6.37-rc5] Timer or ICE1724 issues, HZ=250, NO_HZ=y
Message-ID: <20101213111900.GA2310@gentoo.trippels.de>
References: <20101213105440.GA12768@hostway.ca>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20101213105440.GA12768@hostway.ca>
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1737
Lines: 40

On 2010.12.13 at 02:54 -0800, Simon Kirby wrote:
> With Linus git HEAD just compiled earlier today, I noticed a weird issue
> just now while in the middle of some coding.  I think this may have
> happened before (maybe in 2.6.36), but not as obvious as this time.
> 
> The music that was playing stopped, and the key I had just pressed did
> not show up in rxvt.  I figured the machine had hung, but I hit another
> key and the music played again for another second and my _previous_
> keystroke output appeared.
> 
> Looking at /proc/interrupts, "timer" had completely stopped.  "perf top"
> showed "snd_pcm_hwsync" was at the top of the profile with about 80%
> usage.  Spinning in "snd_pcm_stream_lock_irq" from other CPUs, probably..
> 
> dmesg showed:
> 
> [13995.498393] Clocksource tsc unstable (delta = -8589909196 ns)
> [13995.511452] Switching to clocksource hpet
> [14039.525074] NOHZ: local_softirq_pending 100
> [14039.882183] hrtimer: interrupt took 30800 ns
> [14134.540724] NOHZ: local_softirq_pending 100

Welcome to the club. This is a known problem on newer AMD systems see:
http://thread.gmane.org/gmane.linux.kernel/1054283
and the following patch:
http://thread.gmane.org/gmane.linux.kernel/1069458

But the real problem is that no patch found its way to Linus yet and
2.6.37 will be released in a few days. 

I think Thomas' solution (with a hardcoded hpet_min_tick):
     return res < 128 ? -ETIME : 0;
is the only sane option left this late in the series.
-- 
Markus
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/