Re: linux-next: build failure after merge of the rcu tree

[Peter Zijlstra]

On Fri, Aug 11, 2017 at 11:14:34AM +0200, Peter Zijlstra wrote:
> On Thu, Aug 10, 2017 at 09:54:53PM -0700, Paul E. McKenney wrote:
> > On Fri, Aug 11, 2017 at 02:43:52PM +1000, Stephen Rothwell wrote:
> > 
> > Looks like I need to rebase my patch on top of a9668cd6ee28, and
> > than put an smp_mb__after_spinlock() between the lock and the unlock.
> > 
> > Peter, any objections to that approach?  Other suggestions?
> 
> Hurm.. I'll have to try and understand that comment there again it
> seems.

OK, so per commit b5740f4b2cb3 ("sched: Fix ancient race in do_exit()")
the race is with try_to_wake_up():

down_read()
        p->state = TASK_UNINTERRUPTIBLE;

                                                try_to_wake_up(p)
                                                        spin_lock(p->pi_lock);
                                                        /* sees 
TASK_UNINTERRUPTIBLE */
                                                        ttwu_remote()
        /* check stuff, no need to schedule() */
        p->state = TASK_RUNNING


p->state = TASK_DEAD

                                                                p->state = 
TASK_RUNNING /* whoops! */
                                                        spin_unlock(p->pi_lock);

__schedule(false);
BUG();




So given that, I think that:

  spin_lock(&current->pi_lock);
  spin_unlock(&current->pi_lock);

  current->state = TASK_DEAD;

is sufficient. I don't see a need for an additional smp_mb here.

Either the concurrent ttwu is finished and we must observe its RUNNING
store, or it will observe our RUNNING store.