Re: [Xen-ia64-devel] SMP-guest status

[Tristan Gingold <Tristan.Gingold@xxxxxxxx>]

Le Jeudi 20 Avril 2006 05:47, Alex Williamson a Ãcrit :
> On Wed, 2006-04-19 at 11:38 +0200, Tristan Gingold wrote:
> > +       /* Fill a platform feature.  ITC are (almost!) synchronized. .
> > */
> > +       sal_feature->type = SAL_DESC_PLATFORM_FEATURE;
> > +       sal_feature->feature_mask =
> > IA64_SAL_PLATFORM_FEATURE_ITC_DRIFT;

>    Is the jitter protection in the time interpolator sufficient for
> ignoring this?  Drift is really meant to indicate the ITCs are driven
> from different time sources so may run at slightly different clock
> frequencies.  Seems we should only need to provide that flag to the
> guest if the platform firmware set it.  As long as the ITCs are nearly
> synchronized, the jitter protection in the ITC interpolator will prevent
> time from going backwards.  This would then get rid of the change in
> time.c.  Thanks,
The change in time.c is just to work around a kernel bug.  Linux kernel 
requires at least an interpolator. [Hence I think there is no platform 
without ITC drift].

I agree this area is more or less kludgy.

The problems are:
- if Xen doesn't set ITC_DRIFT, kernels tries to sync itc.  It doesn't really 
work *yet* as itc cannot be set by domains.  I will test this again.

- If Xen sets ITC_DRIFT, it lies (a little or a lot).

Tristan.



_______________________________________________
Xen-ia64-devel mailing list
Xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-ia64-devel