On Fri, May 27, 2016 at 06:52:11PM +0200, Sebastian Andrzej Siewior wrote:
> On 2016-05-19 14:21:48 [+0200], Peter Zijlstra wrote:
> > > +static void futex_populate_hash(unsigned int hash_bits)
> > > +{
> …
> > > + raw_spin_lock(&mm->futex_hash.lock);
> > > + /* We might have raced with another task allocating the hash. */
> > > + if (!mm->futex_hash.hash) {
> > > + mm->futex_hash.hash_bits = hash_bits;
> > > + /*
> > > + * Ensure that the above is visible before we store
> > > + * the pointer.
> > > + */
> > > + smp_wmb(); /* (A0) Pairs with (B) */
> > > + mm->futex_hash.hash = hb;
> >
> > smp_store_release(&mm->futex_hash.hash, hb); ?
>
> just to be clear: You suggest to use "smp_store_release()" instead
> smp_wmb() followed by the assignment?
Yes, smp_store_release() is the most natural way to publish things like
this. Note that rcu_assign_pointer() also switched to using that. See
commit: 88c1863066cc ("rcu: Define rcu_assign_pointer() in terms of
smp_store_release()") for detail on the difference.
