Eric Saxe wrote:
> Bill Holler wrote:
>   
>> Madhavan Venkataraman wrote:
>>   
>>     
>>> Bill Holler wrote:
>>>     
>>>       
>>>> How often will callouts cause a cross-call to another cpu?
>>>> Will this be eliminated with the new per-cpu callout infrastructure?
>>>>
>>>>   
>>>>       
>>>>         
>>> Callouts, by themselves, don't use X-calls. However, normal callouts are
>>> handled via taskq threads. So, waking up a taskq thread will involve
>>> the customary preemption actions which will involve X-calls.
>>> Also, realtime callouts are typically used to wake up threads.
>>> This wake up can involve the same kind of X-calls.
>>> There isn't a whole lot we can do about it.
>>>     
>>>       
>> Thinking out loud: Callouts will often cause threads to be
>> dispatched on cpus other then the one the callout/cyclic
>> occurred on.  We do not currently know how to account
>> for this in idle cpu sleep time prediction....
>> Sounds like a job for the Power Aware Dispatcher.  :-b
>>   
>>     
> Yes, the dispatcher can potentially help here....for example, perhaps 
> some FIFO behavior where the most recently busy CPUs tend to be used 
> sooner...in which case those CPUs could know that they are in the 
> "bullpen" or not.
>   

Would this information be encapsulated in the power processor
groups?  The idle cpu will check these when it predicts which
c-state to attempt to enter.

Bill

> -Eric
> _______________________________________________
> tesla-dev mailing list
> tesla-dev at opensolaris.org
> http://mail.opensolaris.org/mailman/listinfo/tesla-dev
>   


Reply via email to