On Mon, May 6, 2013 at 2:21 AM, Alex Shi <alex....@intel.com> wrote:
> On 05/06/2013 04:19 PM, Paul Turner wrote:
>> On Sun, May 5, 2013 at 6:45 PM, Alex Shi <alex....@intel.com> wrote:
>>> We need initialize the se.avg.{decay_count, load_avg_contrib} for a
>>> new forked task.
>>> Otherwise random values of above variables cause mess when do new task
>>> enqueue:
>>> enqueue_task_fair
>>> enqueue_entity
>>> enqueue_entity_load_avg
>>>
>>> and make forking balancing imbalance since incorrect load_avg_contrib.
>>>
>>> set avg.decay_count = 0, and avg.load_avg_contrib = se->load.weight to
>>> resolve such issues.
>>>
>>> Signed-off-by: Alex Shi <alex....@intel.com>
>>> ---
>>> kernel/sched/core.c | 6 ++++++
>>> kernel/sched/fair.c | 4 ++++
>>> 2 files changed, 10 insertions(+)
>>>
>>> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
>>> index c8db984..ecec7f1 100644
>>> --- a/kernel/sched/core.c
>>> +++ b/kernel/sched/core.c
>>> @@ -1566,6 +1566,7 @@ static void __sched_fork(struct task_struct *p)
>>> #ifdef CONFIG_SMP
>>> p->se.avg.runnable_avg_period = 0;
>>> p->se.avg.runnable_avg_sum = 0;
>>> + p->se.avg.decay_count = 0;
>>> #endif
>>> #ifdef CONFIG_SCHEDSTATS
>>> memset(&p->se.statistics, 0, sizeof(p->se.statistics));
>>> @@ -1653,6 +1654,11 @@ void sched_fork(struct task_struct *p)
>>> p->sched_reset_on_fork = 0;
>>> }
>>>
>>> + /* New forked task assumed with full utilization */
>>> +#if defined(CONFIG_SMP)
>>> + p->se.avg.load_avg_contrib = p->se.load.weight;
>>
>> This is missing a scale_load() right? Further: Why not put this in
>> __sched_fork?
>
> scale_load is not working now. Anyway I can add this.
I believe someone tracked down a plausible cause for this:
A governor was examining the values and making a mess with the scaled
ones. I'm sorry, I don't have the post off hand.
You actually likely ideally want this _on_ for these patches; the
available resolution with SCHED_LOAD_SHIFT=10 disappears really
quickly and scaling by runnable_avg only further accelerates that.
We should try to get this generally turned on by default again.
>
>>
>> We should also charge a minimum period to make the numbers a little
>> more kosher, e.g.:
>> + p->se.avg.runnable_avg_period = 1024;
>> + p->se.avg.runnable_avg_sum = 1024;
>>
>> Rather than exposing the representation of load_avg_contrib to
>> __sched_fork it might also be better to call:
>> __update_task_entity_contrib(&p->se)
>> After the initialization above; this would also avoid potential bugs
>> like the missing scale_load() above.
>
> Above simple change can not work.
> We had talked this solution months ago. And get agreement on this patch.
> https://lkml.org/lkml/2013/2/20/48 :)
>
> --
> Thanks
> Alex
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
