Re: [PATCH v3 1/2] KVM: VMX: FIXED+PHYSICAL mode single target IPI fastpath

[Wanpeng Li <kernellwp@gmail.com>]

On Thu, 21 Nov 2019 at 07:55, Liran Alon <liran.alon@oracle.com> wrote:
>
>
>
> > On 21 Nov 2019, at 1:36, Liran Alon <liran.alon@oracle.com> wrote:
> >
> >
> >
> >> On 20 Nov 2019, at 5:42, Wanpeng Li <kernellwp@gmail.com> wrote:
> >>
> >> From: Wanpeng Li <wanpengli@tencent.com>
> >>
> >> ICR and TSCDEADLINE MSRs write cause the main MSRs write vmexits in
> >> our product observation, multicast IPIs are not as common as unicast
> >> IPI like RESCHEDULE_VECTOR and CALL_FUNCTION_SINGLE_VECTOR etc.
> >
> > Have you also had the chance to measure non-Linux workloads. Such as Windows?
> >
> >>
> >> This patch tries to optimize x2apic physical destination mode, fixed
> >> delivery mode single target IPI. The fast path is invoked at
> >> ->handle_exit_irqoff() to emulate only the effect of the ICR write
> >> itself, i.e. the sending of IPIs.  Sending IPIs early in the VM-Exit
> >> flow reduces the latency of virtual IPIs by avoiding the expensive bits
> >> of transitioning from guest to host, e.g. reacquiring KVM's SRCU lock.
> >> Especially when running guest w/ KVM_CAP_X86_DISABLE_EXITS capability
> >> enabled or guest can keep running, IPI can be injected to target vCPU
> >> by posted-interrupt immediately.
> >
> > May I suggest an alternative phrasing? Something such as:
> >
> > “””
> > This patch introduce a mechanism to handle certain performance-critical WRMSRs
> > in a very early stage of KVM VMExit handler.
> >
> > This mechanism is specifically used for accelerating writes to x2APIC ICR that
> > attempt to send a virtual IPI with physical destination-mode, fixed delivery-mode
> > and single target. Which was found as one of the main causes of VMExits for
> > Linux workloads.
> >
> > The reason this mechanism significantly reduce the latency of such virtual IPIs
> > is by sending the physical IPI to the target vCPU in a very early stage of KVM
> > VMExit handler, before host interrupts are enabled and before expensive
> > operations such as reacquiring KVM’s SRCU lock.
> > Latency is reduced even more when KVM is able to use APICv posted-interrupt
> > mechanism (which allows to deliver the virtual IPI directly to target vCPU without
> > the need to kick it to host).
> > “””
> >
> >>
> >> Testing on Xeon Skylake server:
> >>
> >> The virtual IPI latency from sender send to receiver receive reduces
> >> more than 200+ cpu cycles.
> >>
> >> Cc: Paolo Bonzini <pbonzini@redhat.com>
> >> Cc: Radim Krčmář <rkrcmar@redhat.com>
> >> Cc: Sean Christopherson <sean.j.christopherson@intel.com>
> >> Cc: Vitaly Kuznetsov <vkuznets@redhat.com>
> >> Cc: Liran Alon <liran.alon@oracle.com>
> >> Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
> >
> > I see you used the code I provided my reply to v2. :)
> > I had only some very minor comments inline below. Therefore:
> > Reviewed-by: Liran Alon <liran.alon@oracle.com>
>
> Oh there is a small thing pointed by Sean I agree with (and missed in my review) before this Reviewed-by can be given.
> You should avoid performing accelerated WRMSR handling in case vCPU is in guest-mode.
> (From obvious reasons as L1 for example could intercept this WRMSR…)

Yes, I add !is_guest_mode(vcpu) checking in v1, but careless lose it
in further versions.

    Wanpeng