Re: [Qemu-arm] [PATCH RFC 5/7] ARM64: KVM: Support cross type vCPU

[Christoffer Dall]

On Sat, Jan 28, 2017 at 03:47:54PM +0100, Andrew Jones wrote:
> On Mon, Jan 16, 2017 at 05:33:32PM +0800, Shannon Zhao wrote:
> > From: Shannon Zhao <address@hidden>
> > 
> > Add a capability to tell userspace that KVM supports cross type vCPU.
> > Add a cpu feature for userspace to set when it doesn't use host type
> > vCPU and kvm_vcpu_preferred_target return the host MIDR register value
> > so that userspace can check whether its requested vCPU type macthes the
> > one of physical CPU and if so, KVM will not trap ID registers even
> > though userspace doesn't specify -cpu host.
> > Guest accesses MIDR through VPIDR_EL2 so we save/restore it no matter
> > it's a cross type vCPU.
> > 
> > Signed-off-by: Shannon Zhao <address@hidden>
> > ---
> >  arch/arm/kvm/arm.c                   | 10 ++++++++--
> >  arch/arm64/include/asm/kvm_emulate.h |  3 +++
> >  arch/arm64/include/asm/kvm_host.h    |  3 ++-
> >  arch/arm64/include/uapi/asm/kvm.h    |  1 +
> >  arch/arm64/kvm/guest.c               | 17 ++++++++++++++++-
> >  arch/arm64/kvm/hyp/sysreg-sr.c       |  2 ++
> >  include/uapi/linux/kvm.h             |  1 +
> >  7 files changed, 33 insertions(+), 4 deletions(-)
> > 
> > diff --git a/arch/arm/kvm/arm.c b/arch/arm/kvm/arm.c
> > index 1167678..bdceb19 100644
> > --- a/arch/arm/kvm/arm.c
> > +++ b/arch/arm/kvm/arm.c
> > @@ -207,6 +207,7 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long 
> > ext)
> >     case KVM_CAP_ARM_PSCI_0_2:
> >     case KVM_CAP_READONLY_MEM:
> >     case KVM_CAP_MP_STATE:
> > +   case KVM_CAP_ARM_CROSS_VCPU:
> >             r = 1;
> >             break;
> >     case KVM_CAP_COALESCED_MMIO:
> > @@ -809,8 +810,9 @@ static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
> >  {
> >     unsigned int i;
> >     int phys_target = kvm_target_cpu();
> > +   bool cross_vcpu = kvm_vcpu_has_feature_cross_cpu(init);
> >  
> > -   if (init->target != phys_target)
> > +   if (!cross_vcpu && init->target != phys_target)
> >             return -EINVAL;
> 
> I'm not sure we need the vcpu feature bit. I think qemu should be
> allowed to try any target (if using -cpu host it will try the
> kvm preferred target). kvm should check that the input target is
> a known target and that it is compatible with the phys_target,
> otherwise -EINVAL.
> 

I agree. I think we just need to advertise the capability to user space
instead.

Thanks,
-Christoffer