Re: [Question] sched: Should nr_uninterruptible be decremented in ttwu_do_activate()?

[Michael Wang <wangyun@linux.vnet.ibm.com>]

On 01/13/2012 01:08 AM, Rakib Mullick wrote:

> On Thu, Jan 12, 2012 at 1:25 PM, Peter Zijlstra <a.p.zijlstra@chello.nl> wrote:
>> On Thu, 2012-01-12 at 12:09 +0600, Rakib Mullick wrote:
>>> On Wed, Jan 11, 2012 at 11:29 PM, Peter Zijlstra <a.p.zijlstra@chello.nl> wrote:
>>>> On Wed, 2012-01-11 at 23:22 +0600, Rakib Mullick wrote:
>>>>> Hello all,
>>>>>
>>>>> In ttwu_do_activate(), we're decrementing nr_uninterruptible if
>>>>> p->sched_contributes_to_load (for SMP=y). But, we're also decrementing
>>>>> nr_uninterruptible from activate_task at the same path. Why we're
>>>>> doing it twice for a single task activation path?
>>>>
>>>> activate_task() does:
>>>>
>>>>  if (task_contributes_to_load(p))
>>>>   rq->nr_uninterruptible--;
>>>>
>>>> Now task_contributes_to_load() reads like:
>>>>
>>>> #define task_contributes_to_load(task)  \
>>>>                                ((task->state & TASK_UNINTERRUPTIBLE) != 0 && \
>>>>                                 (task->flags & PF_FREEZING) == 0)
>>>>
>>>> which will be false, since we've set TASK_WAKING.
>>>
>>> Enough confusing. TASK_WAKING will be set when called from
>>> try_to_wake_up(). ttwu_do_activate() gets called from other places:
>>> scheduler_ipi() and sched_ttwu_pending() (at the time of cpu goes
>>> down). TASK_WAKING will be not set at that time,
>>
>> Yes it will be, the only way to get on that list is throught
>> ttwu_queue_remote() at which point tasks are TASK_WAKING.
>>
>>>  moreover it is
>>> possible that, task has p->sched_contributes_to_load is set and latter
>>> on gets wake up by sched_ttwu_pending/scheduler_ipi() call.
>>
>> That's the entire point. But all ways to ttwu_queue_remote() explicitly
>> set ->sched_contributes_to_load.
> 
> That might be the case for scheduler_ipi(), but when
> sched_ttwu_pending() gets called when a cpu goes down, all tasks from
> wake_list of that cpu has been moved without TASK_WAKING is set. For a


I think the task in rq->wake_list should already have state:TASK_WAKING,
because it's a wake list.

> particular task it might be possible that when it ran previously it
> had p->sched_contributes_to_load is set. Latter, this task's cpu has
> been put down and calls sched_ttwu_pending(), then for that task
> p->sched_contributes_to_load is set and TASK_WAKING is not set.
> Couldn't be happen?
> 
> Thanks,
> Rakib
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/
>