RE: [Xen-devel] tsc_mode == 1 and hvm guests

[Stefano Stabellini <stefano.stabellini@xxxxxxxxxxxxx>]

On Mon, 10 May 2010, Dan Magenheimer wrote:
> > On Mon, 10 May 2010, Dan Magenheimer wrote:
> > > Is it an objective for hybrid VMs to work on 4.0.1?
> > > I think the fix only is required for hybrid VM's
> > > and the rest is cleanup.  Is that correct, Stefano?
> > 
> > it is required to use VIRQ_TIMER on HVM, an hybrid VM would work anyway
> > falling back to the emulated timer.
> 
> OK, so not even required for hybrid.
> 
> > Also I believe that migrating to a guest that has a different tsc rate
> > was broken (on HVM) before my patch.
> 
> I'm sure I tested this last December and it worked properly, though
> I haven't tested it recently.  Could you confirm that it was
> broken in 4.0.0 (without your patch)?
> 
> Keir, if you are putting all of the hybrid VM patches
> into 4.0.1, then this should go in as well.  If not,
> please hold off until we confirm there was a fix (and
> I don't have multiple machines free to test it right now,
> so if Stefano confirms it was broken in 4.0.0, I'm
> OK with the fix).
> 

the problem is that if the guest is not using the VIRQ_TIMER it
wouldn't hang even with a different tsc rate and I cannot use the
VIRQ_TIMER at all on HVM without the patch.
Therefore I don't have a simple way to test if the scaling is working or
not.

However I can tell you that without the patch when vtsc == 1
hvm_gtsc_need_scale returns 0 therefore scaling is never done.
For this reason I cannot see how the tsc frequency would remain the
same when vtsc == 1 and the underling hardware has a different tsc rate.
The offset would be OK though.

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