Re: [PATCH] [XEND] alignment of vtpm support in xenapi, documentation and libxen

[Ewan Mellor <ewan@xensource.com>]

On Tue, Jan 30, 2007 at 10:53:24AM -0500, Stefan Berger wrote:

> Ewan Mellor <ewan@xensource.com> wrote on 01/30/2007 10:32:40 AM:
> 
> > On Tue, Jan 30, 2007 at 10:23:13AM -0500, Stefan Berger wrote:
> >
> > > Ewan Mellor <ewan@xensource.com> wrote on 01/30/2007 10:12:10 AM:
> > >
> > > > On Tue, Jan 30, 2007 at 09:29:47AM -0500, Stefan Berger wrote:
> > > >
> > > > > This patch aligns vTPM support in the Xen-API, documentation and
> > > lib-xen
> > > > > (after the recent changes).
> > > > >
> > > > > Signed-off-by: Stefan Berger <stefanb@us.ibm.com>
> > > >
> > > > What's the intention here?  You've removed the get_instance calls,
> but
> > > not the
> > > > instance field from the documentation.  Assuming that that's just a
> > > mistake,
> > > > and you meant to remove the instance field, we're left with a VTPM
> class
> > > that
> > > > has nothing other than a reference to a VM and a reference to a
> backend
> > > > domain.  What are the semantics of that object now?
> > >
> > > The instance will remain to be assigned by the hotplug scripts. In the
> > > old-style of VM configuration file one could still pass it as
> parameter,
> > > but its ignored. I rather not have it passed in as a parameter by the
> > > Xen-API, either. From what I can see a getter for it is not useful,
> > > either, since I want the instance number to be hidden from management
> > > software.
> >
> > What we're left with seems like a pretty expensive way of saying
> "VTPM_backend
> > = N".  Is there really nothing else that's configurable?  We could just
> put
> > this into VM.other_config if that's the only thing that you need, which
> would
> > make configuring a VTPM a lot easier.
> 
> I would like to treat the vTPM as a device like VIF and VBDs with create
> and destroy methods exported to management software so that a TPM device
> can be added to a VM similar to other devices and possibly removed when
> the VM is not running.

Well you'll certainly be able to remove it, whichever way it's modelled.  I'm
not sure that treating the VTPM as a device is worth the cost, but if you
prefer it that way, that's fine by me.

I'll just remove that instance field from the docs, and leave it at that.

Ewan.