Re: [PATCH] sched: Add logic to handle parallel try_to_wake_up() of the same task

[Kirill Tkhai <tkhai@yandex.ru>]

16.07.2013, 00:19, "Peter Zijlstra" <peterz@infradead.org>:
> On Mon, Jul 15, 2013 at 06:14:34PM +0400, Kirill Tkhai wrote:
>
>>>>    #ifdef CONFIG_SMP
>>>>   + p->state = TASK_WAKING;
>>>>   + smp_wmb();
>>>>   +
>>>  This too is broken; the loop below needs to be completed first,
>>>  otherwise we change p->state while the task is still on the CPU and it
>>>  might read the wrong p->state.
>>  This place is below (on_rq && ttwu_remote) check, so the task
>>  either 'dequeued and on_cpu == 0'
>>  or it's in the middle of schedule() on arch, which wants unlocked
>>  context switch.
>>
>>  Nobody scheduler's probes p->state between prepare_lock_switch() and
>>  finish_lock_switch(). Archs with unlocked ctx switch (mips and ia64)
>>  don't change or probe state of previous process during context_switch.
>
> It means its after deactivate_task(), but before context_switch(). It so
> happens that
> context_switch()->prepare_task_switch()->trace_sched_switch() inspects
> p->state.
>
> Even if this was not the case, touching a task that is 'life' on another
> CPU is very _very_ bad practise.

Thanks for the explanation.

Kirill