Re: Xen SMP DomU status (from cherry-xenmp)

[Manuel Bouyer <bouyer%antioche.eu.org@localhost>]

On Wed, Oct 26, 2011 at 08:43:02PM -0400, David Howland wrote:
> On 10/26/2011 5:56 AM, Manuel Bouyer wrote:
> >Hello,
> >I've done some tests with the code from the cherry-xenmp branch.
> >After fixing some interrupt issues I could reenable using HYPERVISOR_block()
> >in the idle loop, so that domUs are not spinning when idle any more.
> >
> >I've tested a i386 XEN3PAE_DOMU and amd64 XEN3_DOMU kernel.
> >Both could complete a build.sh -j12 release on a 8-CPU system, with
> >xen 4.1.1 and a NetBSD 5.1_STABLE dom0.
> >The i386 XEN3PAE_DOMU performances are good. It will eventually trigger
> >a bug in the hypervisor, I need to see if xen 4.1.2 is better in this
> >area.
> >The amd64 performances are not so good. Lots of time is spent in the
> >kernel and there's a hunge number of IPIs. I suspect this is related
> >to the APDP use in amd64, it may be possible to improve this.
> >
> 
> By the way, I get this message non-stop in my console with -current in domU.
> 
> pmap_kenter_pa: mapping already present
> pmap_kenter_pa: mapping already present
> pmap_kenter_pa: mapping already present

I'm seeing this too. pmap_kenter_pa() is called to change
flags on a mapping. Someone added this warning, I don't know why
he think this is a problem.

-- 
Manuel Bouyer <bouyer%antioche.eu.org@localhost>
     NetBSD: 26 ans d'experience feront toujours la difference
--