Re: [PATCH] KVM: x86: Use fast path for Xen timer delivery

[Paul Durrant <xadimgnik@gmail.com>]

On 22/09/2023 10:29, David Woodhouse wrote:
> On Fri, 2023-09-22 at 10:22 +0100, Paul Durrant wrote:
>> On 22/09/2023 10:20, David Woodhouse wrote:
>>> From: David Woodhouse <dwmw@amazon.co.uk>
>>>
>>> Most of the time there's no need to kick the vCPU and deliver the timer
>>> event through kvm_xen_inject_timer_irqs(). Use kvm_xen_set_evtchn_fast()
>>> directly from the timer callback, and only fall back to the slow path
>>> when it's necessary to do so.
>>>
>>> This gives a significant improvement in timer latency testing (using
>>> nanosleep() for various periods and then measuring the actual time
>>> elapsed).
>>>
>>> Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
>>> ---
>>>    arch/x86/kvm/xen.c | 17 ++++++++++++++---
>>>    1 file changed, 14 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/arch/x86/kvm/xen.c b/arch/x86/kvm/xen.c
>>> index 40edf4d1974c..66c4cf93a55c 100644
>>> --- a/arch/x86/kvm/xen.c
>>> +++ b/arch/x86/kvm/xen.c
>>> @@ -134,12 +134,23 @@ static enum hrtimer_restart xen_timer_callback(struct hrtimer *timer)
>>>    {
>>>          struct kvm_vcpu *vcpu = container_of(timer, struct kvm_vcpu,
>>>                                               arch.xen.timer);
>>> +       struct kvm_xen_evtchn e;
>>> +       int rc;
>>> +
>>>          if (atomic_read(&vcpu->arch.xen.timer_pending))
>>>                  return HRTIMER_NORESTART;
>>>    
>>> -       atomic_inc(&vcpu->arch.xen.timer_pending);
>>> -       kvm_make_request(KVM_REQ_UNBLOCK, vcpu);
>>> -       kvm_vcpu_kick(vcpu);
>>> +       e.vcpu_id = vcpu->vcpu_id;
>>> +       e.vcpu_idx = vcpu->vcpu_idx;
>>> +       e.port = vcpu->arch.xen.timer_virq;
>>
>> Don't you need to check the port for validity? The VMM may not have set
>> it (and hence be delivering timer VIRQ events itself).
> 
> Nah, the kvm_xen_timer_enabled() check which gates all the kernel
> interception of timer hypercalls is already testing precisely that.
> 
> If the TIMER_VIRQ isn't set, none of this code runs. And when userspace
> tears down TIMER_VIRQ, the hrtimer is cancelled. So none of this code
> (then) runs.

Ok. Good :-)

Reviewed-by: Paul Durrant <paul@xen.org>