Re: [XEN PATCH] x86/hvm: make ACPI PM timer support optional

[Sergiy Kibrik <sergiy_kibrik@xxxxxxxx>]

18.09.24 12:41, Roger Pau Monné:
> On Wed, Sep 18, 2024 at 12:29:39PM +0300, Sergiy Kibrik wrote:
> > 16.09.24 22:57, Stefano Stabellini:
> > > On Mon, 16 Sep 2024, Roger Pau Monné wrote:
> > > > On Mon, Sep 16, 2024 at 09:37:57AM +0300, Sergiy Kibrik wrote:
> > > > > Introduce config option X86_PMTIMER so that pmtimer driver can be disabled on
> > > > > systems that don't need it.
> > > > Same comment as in the VGA patch, you need to handle the user passing
> > > > X86_EMU_PM.  It's not OK to just ignore the flag if the hypervisor is
> > > > built without ACPI PM timer support.
> > > I also think that the flag should not be ignored. I think that Xen
> > > should return error if a user is passing a domain feature not supported
> > > by a particular version of the Xen build. I don't think that libxl needs
> > > to be changed as part of this patch necessarily.
> > It looks like toolstack always leaves X86_EMU_PM bit enabled, so that part
> > may also require changes.
> I think you will be unable to create HVM guests, but that's kind of
> expected if ACPI PM emulation is removed from the hypervisor (it won't
> be an HVM guest anymore if it doesn't have ACPI PM).
>
> PVH guest don't set X86_EMU_PM so you should be able to create those
> fine.
would the check like this be enough?:

--- a/xen/arch/x86/domain.c
+++ b/xen/arch/x86/domain.c
@@ -758,6 +758,9 @@ static bool emulation_flags_ok(const struct domain 
*d, uint32_t emflags)
              (X86_EMU_ALL & ~(X86_EMU_VPCI | X86_EMU_USE_PIRQ)) &&
              emflags != X86_EMU_LAPIC )
             return false;
+        if ( !is_hardware_domain(d) &&
+             (emflags & X86_EMU_PM) && !IS_ENABLED(CONFIG_X86_PMTIMER))
+            return false;
     }
     else if ( emflags != 0 && emflags != X86_EMU_PIT )
     {


(probably with X86_PMTIMER option depending on PV)

  -Sergiy