On Thu, Jun 02, 2016 at 11:11:07PM +0800, Boqun Feng wrote:
> On Thu, Jun 02, 2016 at 04:44:24PM +0200, Peter Zijlstra wrote:
> > Let me go ponder that some :/
> >
>
> An intial thought of the fix is making queued_spin_unlock_wait() an
> atomic-nop too:
>
> static inline void queued_spin_unlock_wait(struct qspinlock *lock)
> {
> struct __qspinlock *l = (struct __qspinlock *)lock;
>
> while (!cmpxchg(&l->locked, 0, 0))
> cpu_relax();
> }
>
> This could make queued_spin_unlock_wait() a WRITE, with a smp_mb()
> preceding it, it would act like a RELEASE, which can be paired with
> spin_lock().
>
> Just food for thought. ;-)
Not sure that'll actually work. The qspinlock store is completely
unordered and not part of a ll/sc or anything like that.
Doing competing stores might even result in loosing it entirely.
But I think I got something.. Lemme go test it :-)
--
To unsubscribe from this list: send the line "unsubscribe netfilter-devel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
