Re: [PATCHSET] Announce: High-res timers, tickless/dyntick and dynamic HZ -V4

From: Con Kolivas <kernel@kolivas.org>
To: linux-kernel@vger.kernel.org
Cc: Robert Hancock <hancockr@shaw.ca>,
	tglx@timesys.com, Ingo Molnar <mingo@elte.hu>,
	Andrew Morton <akpm@osdl.org>,
	Michal Piotrowski <michal.k.k.piotrowski@gmail.com>
Subject: Re: [PATCHSET] Announce: High-res timers, tickless/dyntick and	dynamic HZ -V4
Date: Sat, 24 Jun 2006 11:09:55 +1000	[thread overview]
Message-ID: <200606241109.56414.kernel@kolivas.org> (raw)
In-Reply-To: <449C8C7E.1040500@shaw.ca>

On Saturday 24 June 2006 10:51, Robert Hancock wrote:
> Thomas Gleixner wrote:
> > On Thu, 2006-06-22 at 21:31 -0600, Robert Hancock wrote:
> >> Thomas Gleixner wrote:
> >>> An updated patchset is available from:
> >>>
> >>> http://www.tglx.de/projects/hrtimers/2.6.17/patch-2.6.17-hrt-dyntick4.p
> >>>atch
> >>
> >> On my Compaq Presario X1050 laptop running Fedora Core 5 I get:
> >>
> >> Disabling NO_HZ and high resolution timers due to timer broadcasting
> >>
> >> Not sure exactly what this is indicating or what's triggered this, but
> >> I'm assuming the patch isn't doing much on this machine?
> >
> > The system is configured for SMP, but this is an UP machine and the APIC
> > is disabled in the BIOS. Linux uses then the PIT and an IPI mechanism to
> > broadcast timer events. We need to do the event reprogramming per CPU,
> > so we switch off in that situation.
> >
> > Solution: Either use an UP kernel, or enable Local APIC in the BIOS (is
> > not possible in most BIOSes), or add "lapic" to the kernel command line.
> >
> > Also for an UP kernel adding "lapic" to the commandline is good, as the
> > APIC is faster accessible than the PIT.
>
> Tried that, still no dice:
>
> Kernel command line: ro root=/dev/VolGroup00/LogVol00 lapic
> Local APIC disabled by BIOS -- reenabling.
> Found and enabled local APIC!
>
> ...
>
> Disabling NO_HZ and high resolution timers due to timer broadcasting
>
> This isn't a viable solution for all machines anyway - some laptops
> disable the local APIC and the BIOS expects it to remain that way, and
> blows up if it gets turned on.

One thing I did discover with my dynticks was that APIC on UP i386 was nigh on 
broken for the most part. Virtually all BIOSs disable it and even if you 
force enable it you may have to play with enable_timer_pin to get it to work 
(sometimes 1, sometimes 0) and there was no pattern to when it would be 
required since the BIOS manufacturer cared not. This is why for UP I disabled 
APIC when dynticks was enabled, and would test for one online cpu on SMP and 
do the same. This obviously isn't a solution when cpu hotplug is enabled and 
only one cpu is online.

-- 
-ck