Re: [RFC PATCH 1/6] kvm: add device control API

[Marc Zyngier <maz@misterjones.org>]

On Tue, 19 Feb 2013 18:16:53 -0800, Christoffer Dall
<cdall@cs.columbia.edu> wrote:
> On Tue, Feb 19, 2013 at 12:16 PM, Scott Wood <scottwood@freescale.com>
> wrote:


>> We at least need the numberspace to not be architecture-specific if we
>> want
>> to retain the possibility of changing later -- not to mention what
>> happens
>> if architectures merge.  I see that "arm" and "arm64" are separate,
>> despite
>> the fact that other architectures that used to be split this way have
>> since
>> merged.  Maybe "arm64" is too different from "arm" for that to happen,
>> but
>> who knows...
>>
> 
> Fair point, nobody knows.

This is unlikely to happen soon.

>> ...and if they don't merge, wouldn't that be a likely case for devices
>> shared across architectures?  Does arm64 use gic/vgic?  This post
>> suggests
>> that there is at least something in common (the bit about "once the GIC
>> code
>> is shared between
>> arm and arm64"):
>>
http://lists.infradead.org/pipermail/linux-arm-kernel/2012-December/135836.html
>>
> 
> I'm not sure how much of that is public at this point, or even
> determined. But KVM already shares code between arm64 and arm, so I
> guess I thought of this as a single architecture from the point of
> view of virt/kvm/kvm_main.c, but that may be incorrect actually.

As I am the sad bastard who wrote most of the VGIC stuff, and also the one
who maintains KVM on arm64, I feel a need to chime in:

arm64 very much uses GIC/VGIC. Actually, all the "in-kernel device" code
is shared between arm and arm64. That means VGIC, timers and PSCI. Which
probably means that code should be moved out of arch/arm/kvm, but that's a
separate story.

It is my intention to keep the interface as close as possible to arm, as
long as that doesn't cause any major issue (arm64 has some slightly
different requirements).

Cheers,

        M.
-- 
Who you jivin' with that Cosmik Debris?