On Thu, Nov 06, 2014 at 11:57:51AM +0100, Christoffer Dall wrote:

On Fri, Oct 31, 2014 at 05:26:35PM +0000, Andre Przywara wrote:

[...]

quoted

Please review and test.
I would be grateful for people to test for GICv2 regressions also
(so on a GICv2 host with current kvmtool/qemu), as there is quite
some refactoring on that front.

So looking at the final result, we have a very strange flow with the
vgic_create() and kvm_vgic_create() functions.  I lost track in all the
rewrite patches how this happened exactly, but what I think you want to
end up with is:

one exported function:
int kvm_vgic_create(struct kvm_device *dev, u32 type);

which calls a static function:
static int vgic_create(struct kvm *kvm, u32 type);

or simply inline the static one in the exported one, I can't seem to
find other callers.

Strike that, my cscope setup was messed up.

What I think you want is a static vgic_v3_create in vgic-v3-emul.c that
digs out the struct kvm pointer from the struct kvm_device and calls
kvm_vgic_create() and also just copy that single kfree(dev) line into
vgic_v3_destroy in vgic-v3-emul.c and the same for v2.

Thanks,
-Christoffer