Eric Saxe <> wrote:
> Hi Aubrey,
>
> Li, Aubrey wrote:
>> The policy seems not to be good.
>> It prevents the core from going into idle. When the new kthread is
>> idle thread, the policy will change the power state to low power
>> state, this calls speedstep power interface on IA platform, which is
>> implemented by xcall. the core may spend some time to grab highest
>> mutex here.
>>
> I'd like to see this done in a way that doesn't require the xcall. Is
> it possible to do the P-State transition such that one CPU (the
> current
> CPU), affects the whole P-state domain, eliminating the need for the
> xcall?
>
>> P-state policy implemented in cmt thread switch here seems not to be
>> a good idea.
> Are you saying this because a xcall would be needed?
It's dependent on the domain coordination method, currently there are three 
kinds of domain coordination:
1) Software all mode: to transit a domain into a specific P-state, software 
needs to write to P-state control register on every CPU in the domain.
2) Software any mode:  to transit a domain into a specific P-state, software 
only needs to write to P-state control registre on one of the CPUs in the 
domain.
3) Hardware coordination mode: each CPU in the domain writes what it wants to 
P-state control register and hardware/BIOS will choose the correct P-state for 
the domain.
so the answer to your question is it depends and a domain controller is needed 
to support different coordination methods.

>
>> We can't upgrade or downgrade the speed level just by the old
>> thread or
>> the new thread is idle thread.
> That's the implementation of the current (rather naive) policy. It's
> really about utilization of the P-state domain, and currently that
> utilization is tracked by looking at transitions between idle and
> non-idle threads.
>
>> We probably still need change it
>> according to the
>> system workload.
>>
> Why not on a (P-state) domain by domain basis? We're trying to made
> some headway towards optimizing for larger systems that are only
> partially utilized....which is difficult to do considering the system
> wide workload...
>
> Thanks,
> -Eric
>
>> Thanks,
>> -Aubrey
>>
>>
>
> _______________________________________________
> tesla-dev mailing list
> tesla-dev at opensolaris.org
> http://mail.opensolaris.org/mailman/listinfo/tesla-dev

Reply via email to