On Fri, Sep 12, 2025 at 02:29:16PM +0100, Daniel P. Berrangé wrote:
> On Fri, Sep 12, 2025 at 03:25:08PM +0200, Gerd Hoffmann wrote:
> > On Wed, Jul 09, 2025 at 02:30:07PM +0200, Gerd Hoffmann wrote:
> > > Implement a ConfidentialGuestSupportClass for non-confidential VMs.
> > > This allows the igvm support code work without sev/tdx.
> > > 
> > > RfC: Not fully sure this is the best way to implement this.
> > > Alternatively we could add this directly into the igvm backend and run
> > > it in case no confidential guest support object is present.
> > 
> > Started to look at this again.  Noticed that at least simple native
> > igvm files (with memory regions only) boot just fine without an
> > ConfidentialGuestSupportClass.  Which kind-of underlines that going for
> > a pseudo-ConfidentialGuestSupportClass for native mode is maybe not the
> > best design idea ...
> > 
> > Nevertheless we'll go need target-specific code for some IGVM features.
> > That is obviously the case for loading some custom initial CPU state.
> > But also things like the memory map are not the same across targets,
> > even though some targets might share an implementation (e820 on x86,
> > fdt elsewhere).
> > 
> > Suggestions how to wire that up best?
> 
> I'm not familiar with the details of IGVM wrt varying technologies.
> Consider that x86 can boot a guest non-confidential, SEV-SNP or TDX.
> Do we have a single IGVM file that provides enough data for all
> three scenarios, or does each deployment technology need a separate
> IGVM file to be provided ?

You can have a single igvm file supporting all platforms.

take care,
  Gerd


Reply via email to