[Ksummit-2012-discuss] [ATTEND] ARM virtualization

Marc Zyngier marc.zyngier at arm.com
Wed Jun 27 08:20:09 UTC 2012


On Tue, 26 Jun 2012 14:37:58 +0100, stefano.stabellini at eu.citrix.com
(Stefano Stabellini) wrote:
> Hi all,
> I would like to propose "ARM virtualization" as a topic for discussion
> at the next Kernel Summit:
> 
> 
> - Xen on ARM guests
> having the opportunity of starting from scratch, we would like to
> get rid of the distinction between PV guests and HVM guests that is
> present on Xen x86, keeping the best of both world.
> 
> - device tree support and a common boot sequence
> even though the "hardware" available in a Xen ARM guest is very
> different from a real platform, Xen is going to use a device tree to
> describe it, so that we can have the same boot sequence in a Xen ARM
> guest, in Dom0 and on native.
> Is the Linux device tree support flexible enough to allow it?
> 
> - calling convention
> I would like to discuss the introduction of a common hypercall calling
> convention across multiple hypervisors on ARM.
>
> - event channel notifications
> In the x86 world we use three different mechanisms: a direct upcall from
> the hypervisor in PV guests; a PCI interrupt for basic HVM guests; a
> direct vector injection in more advanced PV on HVM guests.  What is the
> most efficient and less disruptive way of introducing event
notifications
> on ARM? Do we need to allocate an SGI (or a PPI) for Xen usage?
> 
> - collaboration with KVM
> Aside from the hypercall calling convention, we might be able to share
> other pieces of the infrastructure. I would like to make sure that we
> think about them before writing any code.

I definitely support the above proposals, and would be very glad to
discuss all of this, as there is a number of points I'm working on for
KVM/ARM (boot sequence, DT, general kernel infrastructure).

Thanks,

        M.
-- 
Fast, cheap, reliable. Pick two.


More information about the Ksummit-2012-discuss mailing list