Re: [RESEND][v2][PATCH] Fix a race between try_to_wake_up() and a woken up task

[Peter Zijlstra <peterz@infradead.org>]

On Mon, Sep 05, 2016 at 05:14:19PM +1000, Benjamin Herrenschmidt wrote:
> On Mon, 2016-09-05 at 13:16 +1000, Balbir Singh wrote:
> 
>  .../...
> > 
> > Cc: Peter Zijlstra <peterz@infradead.org>
> > Cc: Nicholas Piggin <npiggin@gmail.com>
> 
> Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
> 
> > Signed-off-by: Balbir Singh <bsingharora@gmail.com>
> > ---
> >  kernel/sched/core.c | 11 +++++++++++
> >  1 file changed, 11 insertions(+)
> > 
> > diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> > index 2a906f2..582c684 100644
> > --- a/kernel/sched/core.c
> > +++ b/kernel/sched/core.c
> > @@ -2016,6 +2016,17 @@ try_to_wake_up(struct task_struct *p, unsigned
> > int state, int wake_flags)
> >  	success = 1; /* we're going to change ->state */
> >  	cpu = task_cpu(p);
> >  
> > +	/*
> > +	 * Ensure we see on_rq and p_state consistently
> > +	 *
> > +	 * For example in __rwsem_down_write_failed(), we have
> > +	 *    [S] ->on_rq = 1				[L] ->state
> > +	 *    MB					 RMB
> > +	 *    [S] ->state = TASK_UNINTERRUPTIBLE	[L] ->on_rq
> > +	 * In the absence of the RMB p->on_rq can be observed to be 0
> > +	 * and we end up spinning indefinitely in while (p->on_cpu)
> > +	 */

So I did replace that comment with the one I proposed earlier. I checked
a fair number of architectures and many did not have an obvious barrier
in switch_to(). So that is not something we can rely on, nor do we need
to I think.

> > +	smp_rmb();
> >  	if (p->on_rq && ttwu_remote(p, wake_flags))
> >  		goto stat;
> >