Re: [qemu-s390x] [PATCH v8 3/6] s390x/kvm: enable/disable AP instruction interpretation for guest

[Halil Pasic]

On 09/18/2018 06:59 PM, Tony Krowiak wrote:
> I've discussed this with Halil -- Pierre is out until next week. We
> are in agreement that while these changes are viable, they result in
> a slightly more complicated implementation compared to previous versions (e.g.
> kernel v9 QEMU v7), and locks us into a model that may limit design choices
> down the road if/when virtual and/or emulated AP devices are implemented.

Just want to confirm, that I did slightly change my mind compared to some weeks
ago regarding the introducing the dynamic toggle in this stage. Back then I was
like it's good, because it makes the interface complete. After thinking some
more, it appears to me alone it does not buy us (or userspace) a thing. If
userspace wants to emulate the ap instructions in the first place it can just
clear away the KVM_S390_VM_CPU_FEAT_AP bit (if offered) before applying the cpu
model. And I can't think of an situation where dynamically switching one way or
the other would make sense without additional kernel interfaces.

But in general my position stands. Any of the proposed solutions will work, and
I'm fine with any of them getting merged. That said, let me state my
preferences.  The most preferred would be a bulk operation based solution like
in v9, for the reasons pointed out by Tony. Next by preference is getting rid
of KVM_S390_VM_CPU_FEAT_AP as you proposed, for the reasons already stated
here. The least preferable would be the v10 solution, but it's also a working
solution, and that's what I care the most about at the moment. And about
vfio-ap getting merged ASAP.

Regards,
Halil