Re: [PATCH RFC] KVM: x86: emulate wait-for-SIPI and SIPI-VMExit

[Sean Christopherson <sean.j.christopherson@intel.com>]

On Thu, Sep 17, 2020 at 10:56:18AM +0200, Paolo Bonzini wrote:
> On 17/09/20 04:25, yadong.qi@intel.com wrote:
> > From: Yadong Qi <yadong.qi@intel.com>
> > 
> > Background: We have a lightweight HV, it needs INIT-VMExit and
> > SIPI-VMExit to wake-up APs for guests since it do not monitoring
> > the Local APIC. But currently virtual wait-for-SIPI(WFS) state
> > is not supported in KVM, so when running on top of KVM, the L1
> > HV cannot receive the INIT-VMExit and SIPI-VMExit which cause
> > the L2 guest cannot wake up the APs.
> > 
> > This patch is incomplete, it emulated wait-for-SIPI state by halt
> > the vCPU and emulated SIPI-VMExit to L1 when trapped SIPI signal
> > from L2. I am posting it RFC to gauge whether or not upstream
> > KVM is interested in emulating wait-for-SIPI state before
> > investing the time to finish the full support.
> 
> Yes, the patch makes sense and is a good addition.  What exactly is
> missing?  (Apart from test cases in kvm-unit-tests!)

nested_vmx_run() puts the vCPU into KVM_MP_STATE_HALTED instead of properly
transitioning to INIT_RECEIVED, e.g. events that arrive while the vCPU is
supposed to be in WFS will be incorrectly recognized.  I suspect there are
other gotchas lurking, but that's the big one.