Re: [Xen-devel] [PATCH 3/4] x86/traps: undo the PVHv1 bodge in NMI delivery path

[Roger Pau Monné <roger.pau@xxxxxxxxxx>]

On Fri, Mar 17, 2017 at 11:29:38AM +0000, Wei Liu wrote:
> On Fri, Mar 17, 2017 at 05:26:07AM -0600, Jan Beulich wrote:
> > >>> On 17.03.17 at 12:14, <wei.liu2@xxxxxxxxxx> wrote:
> > > --- a/xen/arch/x86/traps.c
> > > +++ b/xen/arch/x86/traps.c
> > > @@ -3537,7 +3537,7 @@ static void nmi_hwdom_report(unsigned int 
> > > reason_idx)
> > >  {
> > >      struct domain *d = hardware_domain;
> > >  
> > > -    if ( !d || !d->vcpu || !d->vcpu[0] || !is_pv_domain(d) /* PVH fixme 
> > > */ )
> > > +    if ( !d || !d->vcpu || !d->vcpu[0] || !is_pv_domain(d) )
> > >          return;
> > 
> > But why would you remove the comment then? This needs fixing,
> > after all.
> > 
> 
> Isn't PVHv2 going to use the native path? Now that there is no PVHv1,
> what is there to fix?

The code in nmi_hwdom_report assumes that Dom0 is a PV guest, and injects a PV
NMI (or whatever it's called) by setting the reason in the shared info struct.
On PVHv2 we sould use the native (HVM) way of injecting NMIs IMHO.

Roger.

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel