Re: [PATCH v5 2/7] sched: remove SMP cover for runnable variables in cfs_rq

[Preeti U Murthy <preeti@linux.vnet.ibm.com>]

Hi Paul,

On 05/06/2013 02:38 PM, Paul Turner wrote:
> On Mon, May 6, 2013 at 1:57 AM, Alex Shi <alex.shi@intel.com> wrote:
>> On 05/06/2013 04:01 PM, Paul Turner wrote:
>>> On Sun, May 5, 2013 at 6:45 PM, Alex Shi <alex.shi@intel.com> wrote:
>>>> The following variables were covered under CONFIG_SMP in struct cfs_rq.
>>>> but similar runnable variables work for UP in struct rq and task_group.
>>>> like rq->avg, task_group->load_avg.
>>>> So move them out, they also can work with UP.
>>>
>>> Is there a proposed use-case for UP?  My apologies if I missed it in
>>> an alternate patch.
>>
>>> It would seem the only possibly useful thing there would the the
>>> per-rq average for p-state selection; but we can get that without the
>>> per-entity values already.
>>
>>
>> Do you mean to move the rq->avg and task_group->load_avg into CONFIG_SMP?
> 
> More generally:  Why do we need them in !CONFIG_SMP?
> 
> [ I was suggesting (potentially) using only rq->avg in the !CONFIG_SMP case. ]

If you were to have runnable_load_avg and blocked_load_avg  of cfs_rq
under CONFIG_SMP, how will tg->load_avg get updated? tg->load_avg is not
SMP dependent.

tg->load_avg in-turn is used to decide the CPU shares of the sched
entities on the processor right?

Thanks

Regards
Preeti U Murthy