Re: [PATCH v4 0/3] KVM: Introduce dedicated vCPUs hint KVM_HINTS_DEDICATED

[Wanpeng Li]

2018-03-06 22:39 GMT+08:00 Eduardo Valentin :
> Folks,
>
> On Tue, Feb 13, 2018 at 09:05:39AM +0800, Wanpeng Li wrote:
>> This patchset introduces dedicated vCPUs(vCPU pinning, and there is no
>> vCPU over-commitment) hint KVM_HINTS_DEDICATED, it has two users now:
>>
>> 1) Waiman Long mentioned that:
>>
>> Generally speaking, unfair lock performs well for VMs with a small
>> number of vCPUs. Native qspinlock may perform better than pvqspinlock
>> if there is vCPU pinning and there is no vCPU over-commitment.
>>
>> 2) vCPUs are very unlikely to get preempted when they are the only task
>>running on a CPU. PV TLB flush is slower that the native flush in that
>>case.
>>
>> v3 -> v4:
>>  * update feature bit document
>> v2 -> v3:
>>  * a separate table for CPUID[0x4001].EDX bits
>>  * a new kvm_hint_has_feature macro
>> v1 -> v2:
>>  * update to KVM_HINTS_DEDICATED
>
> Great to see this has finally moved forward! Thanks for picking it up Wanpeng.
>
> I would just request to next time at least reference the original thread and
> copy the original authors of the patch/idea.

Thanks for the original idea, Eduardo!

Regards,
Wanpeng Li

Re: [PATCH v4 0/3] KVM: Introduce dedicated vCPUs hint KVM_HINTS_DEDICATED

[Wanpeng Li]

2018-03-06 22:39 GMT+08:00 Eduardo Valentin :
> Folks,
>
> On Tue, Feb 13, 2018 at 09:05:39AM +0800, Wanpeng Li wrote:
>> This patchset introduces dedicated vCPUs(vCPU pinning, and there is no
>> vCPU over-commitment) hint KVM_HINTS_DEDICATED, it has two users now:
>>
>> 1) Waiman Long mentioned that:
>>
>> Generally speaking, unfair lock performs well for VMs with a small
>> number of vCPUs. Native qspinlock may perform better than pvqspinlock
>> if there is vCPU pinning and there is no vCPU over-commitment.
>>
>> 2) vCPUs are very unlikely to get preempted when they are the only task
>>running on a CPU. PV TLB flush is slower that the native flush in that
>>case.
>>
>> v3 -> v4:
>>  * update feature bit document
>> v2 -> v3:
>>  * a separate table for CPUID[0x4001].EDX bits
>>  * a new kvm_hint_has_feature macro
>> v1 -> v2:
>>  * update to KVM_HINTS_DEDICATED
>
> Great to see this has finally moved forward! Thanks for picking it up Wanpeng.
>
> I would just request to next time at least reference the original thread and
> copy the original authors of the patch/idea.

Thanks for the original idea, Eduardo!

Regards,
Wanpeng Li

Re: [PATCH v4 0/3] KVM: Introduce dedicated vCPUs hint KVM_HINTS_DEDICATED

[Eduardo Valentin]

Folks,

On Tue, Feb 13, 2018 at 09:05:39AM +0800, Wanpeng Li wrote:
> This patchset introduces dedicated vCPUs(vCPU pinning, and there is no 
> vCPU over-commitment) hint KVM_HINTS_DEDICATED, it has two users now:
> 
> 1) Waiman Long mentioned that:
> 
> Generally speaking, unfair lock performs well for VMs with a small
> number of vCPUs. Native qspinlock may perform better than pvqspinlock
> if there is vCPU pinning and there is no vCPU over-commitment.
> 
> 2) vCPUs are very unlikely to get preempted when they are the only task 
>running on a CPU. PV TLB flush is slower that the native flush in that 
>case.
> 
> v3 -> v4:
>  * update feature bit document
> v2 -> v3:
>  * a separate table for CPUID[0x4001].EDX bits
>  * a new kvm_hint_has_feature macro
> v1 -> v2:
>  * update to KVM_HINTS_DEDICATED

Great to see this has finally moved forward! Thanks for picking it up Wanpeng.

I would just request to next time at least reference the original thread and
copy the original authors of the patch/idea.

Thanks a lot.

-- 
All the best,
Eduardo Valentin

Re: [PATCH v4 0/3] KVM: Introduce dedicated vCPUs hint KVM_HINTS_DEDICATED

[Eduardo Valentin]

Folks,

On Tue, Feb 13, 2018 at 09:05:39AM +0800, Wanpeng Li wrote:
> This patchset introduces dedicated vCPUs(vCPU pinning, and there is no 
> vCPU over-commitment) hint KVM_HINTS_DEDICATED, it has two users now:
> 
> 1) Waiman Long mentioned that:
> 
> Generally speaking, unfair lock performs well for VMs with a small
> number of vCPUs. Native qspinlock may perform better than pvqspinlock
> if there is vCPU pinning and there is no vCPU over-commitment.
> 
> 2) vCPUs are very unlikely to get preempted when they are the only task 
>running on a CPU. PV TLB flush is slower that the native flush in that 
>case.
> 
> v3 -> v4:
>  * update feature bit document
> v2 -> v3:
>  * a separate table for CPUID[0x4001].EDX bits
>  * a new kvm_hint_has_feature macro
> v1 -> v2:
>  * update to KVM_HINTS_DEDICATED

Great to see this has finally moved forward! Thanks for picking it up Wanpeng.

I would just request to next time at least reference the original thread and
copy the original authors of the patch/idea.

Thanks a lot.

-- 
All the best,
Eduardo Valentin