On 1 July 2012 14:37, Paolo Bonzini <pbonz...@redhat.com> wrote: > Il 23/06/2012 12:30, Peter Maydell ha scritto: >>> > Can't it go in hw/arm/kvm (mimicking the final desired place, which >>> > will be target-arm/hw/kvm)? And hw/kvm can be moved to hw/i386/kvm, or >>> > we can leave it there for now until we're ready to move it to >>> > target-i386/hw/kvm. >> Why's the final desired place target-arm/hw/kvm ? That doesn't >> make much sense to me... > > Doesn't it have some dependency on target-arm/kvm.c?
Well, it does at the moment, but I'm not entirely sure it should (this is on my list of things to look at). I would expect that "insert interrupt into the KVM kernel irqchip" should be a generic interface the same way that KVM hooks into cpu_interrupt(), only it doesn't seem to be handled that way for eg PPC. Having device models in hw/ make direct calls to per-target KVM functions in target-*/kvm.c seems like a bit of a layering violation to me. -- PMM