Am 18.09.2015 um 12:34 schrieb David Hildenbrand:
> We observed some performance degradation on s390x with dynamic
> halt polling. Until we can provide a proper fix, let's enable
> halt_poll_ns as default only for supported architectures.
>
> Architectures are now free to set their own halt_poll_ns
> default value.
>
> Signed-off-by: David Hildenbrand <dahi@xxxxxxxxxxxxxxxxxx>
> ---
[...]
> diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
> index eb4c9d2..a447c8c 100644
> --- a/virt/kvm/kvm_main.c
> +++ b/virt/kvm/kvm_main.c
> @@ -66,8 +66,8 @@
> MODULE_AUTHOR("Qumranet");
> MODULE_LICENSE("GPL");
>
> -/* halt polling only reduces halt latency by 5-7 us, 500us is enough */
> -static unsigned int halt_poll_ns = 500000;
> +/* Architectures should define their poll value according to the halt latency */
> +static unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
Yes, I prefer this over disabling it via Kconfig. There are benchmarks which
benefit from polling on s390. Furthermore it seems that the latency
strongly depends on timing of the architecture so making it per arch is
probably the right thing to do.
Acked-by: Christian Borntraeger <borntraeger@xxxxxxxxxx>
> module_param(halt_poll_ns, uint, S_IRUGO | S_IWUSR);
>
> /* Default doubles per-vcpu halt_poll_ns. */
>
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
