From: Wanpeng Li <[email protected]>
The downside of guest side polling is that polling is performed even
with other runnable tasks in the host. However, even if poll in kvm
can aware whether or not other runnable tasks in the same pCPU, it
can still incur extra overhead in over-subscribe scenario. Now we can
just enable guest polling when dedicated pCPUs are available.
Acked-by: Paolo Bonzini <[email protected]>
Cc: Rafael J. Wysocki <[email protected]>
Cc: Paolo Bonzini <[email protected]>
Cc: Radim Krčmář <[email protected]>
Signed-off-by: Wanpeng Li <[email protected]>
--
v1 -> v2:
* export kvm_arch_para_hints to fix haltpoll driver build as module error
* just disable haltpoll driver instead of both driver and governor
since KVM_HINTS_REALTIME is not defined in other arches, and governor
doesn't depend on x86, to fix the warning on powerpc
arch/x86/kernel/kvm.c | 1 +
drivers/cpuidle/cpuidle-haltpoll.c | 3 ++-
2 files changed, 3 insertions(+), 1 deletion(-)
diff --git a/arch/x86/kernel/kvm.c b/arch/x86/kernel/kvm.c
index f48401b..68463c1 100644
--- a/arch/x86/kernel/kvm.c
+++ b/arch/x86/kernel/kvm.c
@@ -711,6 +711,7 @@ unsigned int kvm_arch_para_hints(void)
{
return cpuid_edx(kvm_cpuid_base() | KVM_CPUID_FEATURES);
}
+EXPORT_SYMBOL_GPL(kvm_arch_para_hints);
static uint32_t __init kvm_detect(void)
{
diff --git a/drivers/cpuidle/cpuidle-haltpoll.c b/drivers/cpuidle/cpuidle-haltpoll.c
index 9ac093d..7aee38a 100644
--- a/drivers/cpuidle/cpuidle-haltpoll.c
+++ b/drivers/cpuidle/cpuidle-haltpoll.c
@@ -53,7 +53,8 @@ static int __init haltpoll_init(void)
cpuidle_poll_state_init(drv);
- if (!kvm_para_available())
+ if (!kvm_para_available() ||
+ !kvm_para_has_hint(KVM_HINTS_REALTIME))
return 0;
ret = cpuidle_register(&haltpoll_driver, NULL);
--
2.7.4
On 8/29/2019 10:49 AM, Wanpeng Li wrote:
> From: Wanpeng Li <[email protected]>
>
> The downside of guest side polling is that polling is performed even
> with other runnable tasks in the host. However, even if poll in kvm
> can aware whether or not other runnable tasks in the same pCPU, it
> can still incur extra overhead in over-subscribe scenario. Now we can
> just enable guest polling when dedicated pCPUs are available.
>
> Acked-by: Paolo Bonzini <[email protected]>
> Cc: Rafael J. Wysocki <[email protected]>
> Cc: Paolo Bonzini <[email protected]>
> Cc: Radim Krčmář <[email protected]>
> Signed-off-by: Wanpeng Li <[email protected]>
As stated before, I'm going to queue up this change for 5.4, with the
Paolo's ACK.
BTW, in the future please CC power management changes to
[email protected] for easier handling.
> --
> v1 -> v2:
> * export kvm_arch_para_hints to fix haltpoll driver build as module error
> * just disable haltpoll driver instead of both driver and governor
> since KVM_HINTS_REALTIME is not defined in other arches, and governor
> doesn't depend on x86, to fix the warning on powerpc
>
> arch/x86/kernel/kvm.c | 1 +
> drivers/cpuidle/cpuidle-haltpoll.c | 3 ++-
> 2 files changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/arch/x86/kernel/kvm.c b/arch/x86/kernel/kvm.c
> index f48401b..68463c1 100644
> --- a/arch/x86/kernel/kvm.c
> +++ b/arch/x86/kernel/kvm.c
> @@ -711,6 +711,7 @@ unsigned int kvm_arch_para_hints(void)
> {
> return cpuid_edx(kvm_cpuid_base() | KVM_CPUID_FEATURES);
> }
> +EXPORT_SYMBOL_GPL(kvm_arch_para_hints);
>
> static uint32_t __init kvm_detect(void)
> {
> diff --git a/drivers/cpuidle/cpuidle-haltpoll.c b/drivers/cpuidle/cpuidle-haltpoll.c
> index 9ac093d..7aee38a 100644
> --- a/drivers/cpuidle/cpuidle-haltpoll.c
> +++ b/drivers/cpuidle/cpuidle-haltpoll.c
> @@ -53,7 +53,8 @@ static int __init haltpoll_init(void)
>
> cpuidle_poll_state_init(drv);
>
> - if (!kvm_para_available())
> + if (!kvm_para_available() ||
> + !kvm_para_has_hint(KVM_HINTS_REALTIME))
> return 0;
>
> ret = cpuidle_register(&haltpoll_driver, NULL);
On Wed, 4 Sep 2019 at 17:48, Rafael J. Wysocki
<[email protected]> wrote:
>
> On 8/29/2019 10:49 AM, Wanpeng Li wrote:
> > From: Wanpeng Li <[email protected]>
> >
> > The downside of guest side polling is that polling is performed even
> > with other runnable tasks in the host. However, even if poll in kvm
> > can aware whether or not other runnable tasks in the same pCPU, it
> > can still incur extra overhead in over-subscribe scenario. Now we can
> > just enable guest polling when dedicated pCPUs are available.
> >
> > Acked-by: Paolo Bonzini <[email protected]>
> > Cc: Rafael J. Wysocki <[email protected]>
> > Cc: Paolo Bonzini <[email protected]>
> > Cc: Radim Krčmář <[email protected]>
> > Signed-off-by: Wanpeng Li <[email protected]>
>
> As stated before, I'm going to queue up this change for 5.4, with the
> Paolo's ACK.
>
> BTW, in the future please CC power management changes to
> [email protected] for easier handling.
Ok, thanks.
Wanpeng
On 04/09/19 11:48, Rafael J. Wysocki wrote:
> On 8/29/2019 10:49 AM, Wanpeng Li wrote:
>> From: Wanpeng Li <[email protected]>
>>
>> The downside of guest side polling is that polling is performed even
>> with other runnable tasks in the host. However, even if poll in kvm
>> can aware whether or not other runnable tasks in the same pCPU, it
>> can still incur extra overhead in over-subscribe scenario. Now we can
>> just enable guest polling when dedicated pCPUs are available.
>>
>> Acked-by: Paolo Bonzini <[email protected]>
>> Cc: Rafael J. Wysocki <[email protected]>
>> Cc: Paolo Bonzini <[email protected]>
>> Cc: Radim Krčmář <[email protected]>
>> Signed-off-by: Wanpeng Li <[email protected]>
>
> As stated before, I'm going to queue up this change for 5.4, with the
> Paolo's ACK.
>
> BTW, in the future please CC power management changes to
> [email protected] for easier handling.
Thanks. This patch makes sense to me and I don't know what
"limitations" are there in KVM_HINTS_REALTIME that Marcelo mentioned.
Any improvements that Marcelo can discuss can be made on top of this
during the 5.4 merge window, via the KVM tree.
Thanks Rafael for handling the reviewing and merging of this series.
Paolo
>
>> --
>> v1 -> v2:
>> * export kvm_arch_para_hints to fix haltpoll driver build as module
>> error
>> * just disable haltpoll driver instead of both driver and governor
>> since KVM_HINTS_REALTIME is not defined in other arches, and governor
>> doesn't depend on x86, to fix the warning on powerpc
>>
>> arch/x86/kernel/kvm.c | 1 +
>> drivers/cpuidle/cpuidle-haltpoll.c | 3 ++-
>> 2 files changed, 3 insertions(+), 1 deletion(-)
>>
>> diff --git a/arch/x86/kernel/kvm.c b/arch/x86/kernel/kvm.c
>> index f48401b..68463c1 100644
>> --- a/arch/x86/kernel/kvm.c
>> +++ b/arch/x86/kernel/kvm.c
>> @@ -711,6 +711,7 @@ unsigned int kvm_arch_para_hints(void)
>> {
>> return cpuid_edx(kvm_cpuid_base() | KVM_CPUID_FEATURES);
>> }
>> +EXPORT_SYMBOL_GPL(kvm_arch_para_hints);
>> static uint32_t __init kvm_detect(void)
>> {
>> diff --git a/drivers/cpuidle/cpuidle-haltpoll.c
>> b/drivers/cpuidle/cpuidle-haltpoll.c
>> index 9ac093d..7aee38a 100644
>> --- a/drivers/cpuidle/cpuidle-haltpoll.c
>> +++ b/drivers/cpuidle/cpuidle-haltpoll.c
>> @@ -53,7 +53,8 @@ static int __init haltpoll_init(void)
>> cpuidle_poll_state_init(drv);
>> - if (!kvm_para_available())
>> + if (!kvm_para_available() ||
>> + !kvm_para_has_hint(KVM_HINTS_REALTIME))
>> return 0;
>> ret = cpuidle_register(&haltpoll_driver, NULL);
>
>