Re: [PATCH 3/4] rhashtable: use bit_spin_locks to protect hash bucket.

[Paul E. McKenney]

On Tue, Mar 26, 2019 at 01:03:20PM +0800, Herbert Xu wrote:
> On Mon, Mar 25, 2019 at 04:05:39PM +1100, NeilBrown wrote:
> >
> > + * Sometimes we unlock a bucket by writing a new pointer there.  In that
> > + * case we don't need to unlock, but we do need to reset state such as
> > + * local_bh. For that we have rht_unlocked().  This doesn't include
> > + * the memory barrier that bit_spin_unlock() provides, but rcu_assign_pointer()
> > + * will have provided that.
> 
> Hmm, are you sure that's enough? IIRC rcu_assign_pointer only
> provides a write barrier compared to the more complete (but one-way)
> barrier that a spin-lock provides.

Not seeing the code, I have no opinion on the safety in this case,
but I did want to mention that rcu_assign_pointer() has been upgraded
to a release store, so that it orders all prior accesses from the
viewpoint of some other thread that just picked up the stored pointer
via rcu_dereference().

But you are quite right, rcu_assign_pointer() used to just do an
smp_wmb().  It is now new and improved!  ;-)

							Thanx, Paul