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 <[email protected]> > > Cc: Nicholas Piggin <[email protected]> > > Acked-by: Benjamin Herrenschmidt <[email protected]> > > > Signed-off-by: Balbir Singh <[email protected]> > > --- > > �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; > > �
