Re: [PATCH v2] locking/rwsem: add acquire barrier to read_slowpath exit when queue is empty

[Peter Zijlstra]

On Thu, Jul 18, 2019 at 06:50:52AM -0400, Jan Stancek wrote:
> > In writing this, I also noticed that we don't have any explicit ordering
> > at the end of the reader slowpath when we wait on the queue but get woken
> > immediately:
> > 
> >     if (!waiter.task)
> >             break;
> > 
> > Am I missing something?
> 
> I'm assuming this isn't problem, because set_current_state() on line above
> is using smp_store_mb().


X = 0;

                                                                X = 1;
        rwsem_down_read()                                       
rwsem_up_write();

          for (;;) {
            set_current_state(TASK_UNINTERRUPTIBLE);

                                                                  
rwsem_mark_wake()
                                                                    
atomic_long_add(adjustment, &sem->count);
                                                                    
smp_store_release(&waiter->task, NULL);

            if (!waiter.task)
              break;

            ...
          }


        r = X;


can I think result in r==0 just fine, because there's nothing ordering
the load of waiter.task with the store of X.

It is exceedingly unlikely, but not impossible afaict.