Re: [PATCH 0/13] Time: Generic Timeofday Subsystem (v B10)

[john stultz <johnstul@us.ibm.com>]

On Tue, 2005-11-15 at 21:53 +0100, Ingo Molnar wrote:
> * john stultz <johnstul@us.ibm.com> wrote:
> 
> > > idle=poll does seem to fix the major clock drift problem.  There may
> > > still be an issue, but it's much smaller:
> > > 
> > > 2.6.14-mm2-todb10:
> > > 14 Nov 21:50:57      offset: -0.025373       drift: -22404.0 ppm
> > > 14 Nov 21:51:59      offset: -1.577053       drift: -24985.4603175 ppm
> > > 14 Nov 21:53:00      offset: -3.104569       drift: -25012.9032258 ppm
> > > 
> > > 2.6.14-mm2-todb10 with idle=poll:
> > > 14 Nov 21:37:59      offset: 5.9e-05         drift: 63.0 ppm
> > > 14 Nov 21:39:00      offset: 0.003207        drift: 51.7903225806 ppm
> > 
> > Hmm. It seems the c3 compensation is triggering when it shouldn't, or 
> > maybe its over compensating.
> > 
> > I can't reproduce it on my laptop. Do you recall if in previous tests 
> > you saw anything like this? I'm trying to narrow down if its just a 
> > difference in hardware or if something in the c3 idle code changed.
> 
> it's with an earlier queue of yours, but maybe it's related: i have a 
> report that HPET causes HRT inaccuracies (e.g. sleeps for 20 msecs last 
> 21 msecs). If all HPET options are turned off in the .config then 
> everything is fine and accurate.

Hmm. HPET would be separate from this, I believe. I suspect that could
be related to the HPET legacy replacement functionality, where the HPET
is triggering the timer interrupts. Not something I changed, but
probably should be looked into.

thanks
-john