[PATCH 0 of 7] [RFC] kvm ppc refactoring

[PATCH 0 of 7] [RFC] kvm ppc refactoring

[Hollis Blanchard]

These patches refactor the KVM PPC code to support non-Book E processors. The
Book E code might need a little more work to support e500, but probably not
much. Beyond this, we still need to split struct kvm_vcpu_arch using
container_of() macros.

There's a lot of code movement, so it might be easier to review by examining
the final result. Basically we have powerpc.c and emulate.c calling
core-specific functions, some of which are provided by booke.c and some by
44x*.c.

Comments?

-Hollis

Re: [PATCH 0 of 7] [RFC] kvm ppc refactoring

[Alexander Graf]

On 29.09.2008, at 20:36, Hollis Blanchard wrote:

> These patches refactor the KVM PPC code to support non-Book E  
> processors. The
> Book E code might need a little more work to support e500, but  
> probably not
> much. Beyond this, we still need to split struct kvm_vcpu_arch using
> container_of() macros.
>
> There's a lot of code movement, so it might be easier to review by  
> examining
> the final result. Basically we have powerpc.c and emulate.c calling
> core-specific functions, some of which are provided by booke.c and  
> some by
> 44x*.c.
>
> Comments?

Does non-Book E processors mean 97x series :-)?

Alex

Re: [PATCH 0 of 7] [RFC] kvm ppc refactoring

[Hollis Blanchard]

On Mon, 2008-09-29 at 20:40 +0200, Alexander Graf wrote:
> On 29.09.2008, at 20:36, Hollis Blanchard wrote:
> 
> > These patches refactor the KVM PPC code to support non-Book E  
> > processors. The
> > Book E code might need a little more work to support e500, but  
> > probably not
> > much. Beyond this, we still need to split struct kvm_vcpu_arch using
> > container_of() macros.
> >
> > There's a lot of code movement, so it might be easier to review by  
> > examining
> > the final result. Basically we have powerpc.c and emulate.c calling
> > core-specific functions, some of which are provided by booke.c and  
> > some by
> > 44x*.c.
> >
> > Comments?
> 
> Does non-Book E processors mean 97x series :-)?

That's not something I'm personally looking at right now, but this patch
series would get you closer. :)

Probably the most feasible would be to ignore 970's hypervisor mode
(since you can't use it on any generally available platforms anyways),
and implement 440-style software-only virtualization instead. I've heard
people dreaming about doing exactly that for Cell (same problem).

If you want to chat about how exactly this could be implemented, I'd be
happy to brainstorm with you... :)

-- 
Hollis Blanchard
IBM Linux Technology Center