Re: [PATCH] perf: Fix race between close() and fork()

[Mark Rutland]

On Fri, Jun 28, 2019 at 10:46:08PM +0200, Peter Zijlstra wrote:
> On Fri, Jun 28, 2019 at 05:50:03PM +0100, Mark Rutland wrote:
> > > +		/*
> > > +		 * Wake any perf_event_free_task() waiting for this event to be
> > > +		 * freed.
> > > +		 */
> > > +		smp_mb(); /* pairs with wait_var_event() */
> > > +		wake_up_var(var);
> > 
> > Huh, so wake_up_var() doesn't imply a RELEASE?
> > 
> > As an aside, doesn't that mean all callers of wake_up_var() have to do
> > likewise to ensure it isn't re-ordered with whatever prior stuff they're
> > trying to notify waiters about? Several do an smp_store_release() then a
> > wake_up_var(), but IIUC the wake_up_var() could get pulled before that
> > release...
> 
> Yah,...
> 
>   https://lkml.kernel.org/r/20190624165012.GH3436@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

That gets me a 404, so I'll assume that's a speculative store. ;)

> I needs to get back to that.

Ouch; sorry for reminding you of this mess!

Thanks,
Mark.