Re: [PATCH v2 1/2] KVM: Refactor kvm_arch_vcpu_fault() to return a struct page pointer

From: Paolo Bonzini
Date: Thu Aug 12 2021 - 11:41:31 EST


On 12/08/21 11:04, David Hildenbrand wrote:

Reviewed-by: David Hildenbrand <david@xxxxxxxxxx>

But at the same time I wonder if we should just get rid of CONFIG_KVM_S390_UCONTROL and consequently kvm_arch_vcpu_fault().

In practice CONFIG_KVM_S390_UCONTROL, is never enabled in any reasonable kernel build and consequently it's never tested; further, exposing the sie_block to user space allows user space to generate random SIE validity intercepts.

CONFIG_KVM_S390_UCONTROL feels like something that should just be maintained out of tree by someone who really needs to hack deep into hw virtualization for testing purposes etc.

I have no preference either way. It should definitely have selftests, but in x86 land there are some features that are not covered by QEMU and were nevertheless accepted upstream with selftests.

Paolo