Re: [patch RFC 24/38] x86/xen: Consolidate XEN-MSI init

[Jürgen Groß <jgross@xxxxxxxx>]

On 24.08.20 23:21, Thomas Gleixner wrote:
> On Mon, Aug 24 2020 at 06:59, Jürgen Groß wrote:
> > On 21.08.20 02:24, Thomas Gleixner wrote:
> > > +static __init void xen_setup_pci_msi(void)
> > > +{
> > > +       if (xen_initial_domain()) {
> > > +               x86_msi.setup_msi_irqs = xen_initdom_setup_msi_irqs;
> > > +               x86_msi.teardown_msi_irqs = xen_teardown_msi_irqs;
> > > +               x86_msi.restore_msi_irqs = xen_initdom_restore_msi_irqs;
> > > +               pci_msi_ignore_mask = 1;
> > This is wrong, as a PVH initial domain shouldn't do the pv settings.
> >
> > The "if (xen_initial_domain())" should be inside the pv case, like:
> >
> > if (xen_pv_domain()) {
> >         if (xen_initial_domain()) {
> >                 ...
> >         } else {
> >                 ...
> >         }
> > } else if (xen_hvm_domain()) {
> >         ...
> I still think it does the right thing depending on the place it is
> called from, but even if so, it's completely unreadable gunk. I'll fix
> that proper.
The main issue is that xen_initial_domain() and xen_pv_domain() are
orthogonal to each other. So xen_initial_domain() can either be true
for xen_pv_domain() (the "classic" pv dom0) or for xen_hvm_domain()
(the new PVH dom0).


Juergen