KVM: x86/xen: Use fast path for Xen timer delivery

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 if
delivering the timer would block, i.e. if kvm_xen_set_evtchn_fast()
returns -EWOULDBLOCK.  If delivery fails for any other reason, do nothing
and just let it fail silently, as that is what the slow path would end up
doing anyways.

This gives a significant improvement in timer latency testing (using
nanosleep() for various periods and then measuring the actual time
elapsed).

However, there was a reason[1] the fast path was dropped when this support
was first added. The current code holds vcpu->mutex for all operations on
the kvm->arch.timer_expires field, and the fast path introduces a
potential race condition. Avoid that race by ensuring the hrtimer is
(temporarily) cancelled before making changes in kvm_xen_start_timer(),
and also when reading the values out for KVM_XEN_VCPU_ATTR_TYPE_TIMER.

[1] https://lore.kernel.org/kvm/[email protected]

Signed-off-by: David Woodhouse <[email protected]>
Reviewed-by: Paul Durrant <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
[sean: massage changelog]
Signed-off-by: Sean Christopherson <[email protected]>

0 files changed, 0 insertions, 0 deletions