Re: [tg_shares_up rewrite v3 09/11] sched: demand based update_cfs_load()

From: Paul Turner
Date: Fri Nov 12 2010 - 21:48:24 EST


On Fri, Nov 12, 2010 at 5:01 PM, Paul Turner <pjt@xxxxxxxxxx> wrote:
> On Fri, Nov 12, 2010 at 2:53 AM, Peter Zijlstra <a.p.zijlstra@xxxxxxxxx> wrote:
>> On Thu, 2010-11-11 at 19:24 -0800, Paul Turner wrote:
>>> +#ifdef CONFIG_FAIR_GROUP_SCHED
>>> +       cfs_rq->load_unacc_exec_time += delta_exec;
>>> +       if (cfs_rq->load_unacc_exec_time > sysctl_sched_shares_window)
>>> {
>>> +               update_cfs_load(cfs_rq);
>>> +               update_cfs_shares(cfs_rq, 0);
>>
>> Why not:
>> +               cfs_rq->load_unacc_exec_time -= sysctl_sched_shares_window;
>>
>> (although you probably want to read the sysctl value into a local
>> variable using ACCESS_ONCE() and use that for both cases).
>>
>
> I think this is hard to do in a clean fashion (without a strange arg
> to update_cfs_load).
>
> I'm also not sure it's worth synchronizing on a shares_window change
> since, when we trigger an update from update_curr() is independent of
> folding that time into the load average anyway.
> (Note even: using sysctl_sched_shares_window is already a larger
> window than update_cfs_load will normally fold at since we fold at
> window/2 after the initial period, it just sets a grace period on
> computation without updates in the busy case.)
>

Let me know if you still think this warrants any changes and I'll (fix
accordingly) and repost with the "-p0" stripping fixed.

>>> +       }
>>> +#endif
>>>  }
>>>
>>>  static void update_curr(struct cfs_rq *cfs_rq)
>>> @@ -713,6 +724,7 @@ static void update_cfs_load(struct cfs_r
>>>         }
>>>
>>>         cfs_rq->load_stamp = now;
>>> +       cfs_rq->load_unacc_exec_time = 0;
>>
>> and drop this one?
>>
>
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/