Re: [PATCH 6/13] bridge: Add core IGMP snooping support

["Paul E. McKenney" <paulmck@linux.vnet.ibm.com>]

On Sun, Mar 07, 2010 at 10:45:00AM +0800, Herbert Xu wrote:
> On Sat, Mar 06, 2010 at 11:00:00AM -0800, Paul E. McKenney wrote:
> >
> > > Hmmm...  rcu_barrier() definitely does -not- imply rcu_barrier_bh(),
> > > because there are separate sets of callbacks whose execution can
> > > be throttled separately.  So, while you would expect RCU-bh grace
> > > periods to complete more quickly, if there was a large number of
> > > RCU-bh callbacks on a given CPU but very few RCU callbacks, it might
> > > well take longer for the RCU-bh callbacks to be invoked.
> > > 
> > > With TREE_PREEMPT_RCU, if there were no RCU readers but one long-running
> > > RCU-bh reader, then synchronize_rcu_bh() could return before
> > > synchronize_rcu() does.
> 
> OK, then we definitely do have some issues under net/ with respect
> to the two types of RCU usage.  As you can see, we use the RCU-BH
> variant on the read-side in various places, and call_rcu_bh on the
> write-side too, but we only ever use the non-BH version of the
> functions rcu_barrier and synchronize_rcu.
> 
> Now there is a possibility that the places where we use synchronize
> and rcu_barrier don't really care about the BH variant, but an
> audit wouldn't hurt.
> 
> > You really are talking about code like the following, correct?
> > 
> > 	rcu_read_lock();
> > 	p = rcu_dereference(global_p);
> > 	do_something_with(p);
> > 	rcu_read_unlock();
> > 
> > 	. . .
> > 
> > 	rcu_read_lock_bh();
> > 	p = rcu_dereference(global_p);
> > 	do_something_else_with(p);
> > 	rcu_read_unlock_bh();
> > 
> > 	. . . 
> > 
> > 	spin_lock(&my_lock);
> > 	p = global_p;
> > 	rcu_assign_pointer(global_p, NULL);
> > 	synchronize_rcu();  /* BUG -- also need synchronize_rcu_bh(). */
> > 	kfree(p);
> > 	spin_unlock(&my_lock);
> > 
> > In other words, different readers traversing the same data structure
> > under different flavors of RCU protection, but then using only one
> > flavor of RCU grace period during the update?
> 
> We usually don't use synchronize_rcu/rcu_barrier on the update side,
> but rather they are used in the tear-down process.
> 
> But otherwise yes this is exactly my concern.
> 
> Note that we may have a problem on the update side too if we used
> the wrong call_rcu variant, but it would require a thorough audit
> to reveal those.

OK, just re-checked your patch, and it looks OK.

Also adding Arnd to CC.

Arnd, would it be reasonable to extend your RCU-sparse changes to have
four different pointer namespaces, one for each flavor of RCU?  (RCU,
RCU-bh, RCU-sched, and SRCU)?  Always a fan of making the computer do
the auditing where reasonable.  ;-)

This could potentially catch the mismatched call_rcu()s, at least if the
rcu_head could be labeled.

Other thoughts?

							Thanx, Paul