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

From: Herbert Xu <herbert@gondor.apana.org.au>
To: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Cc: "David S. Miller" <davem@davemloft.net>,
	netdev@vger.kernel.org, Stephen Hemminger <shemminger@vyatta.com>
Subject: Re: [PATCH 6/13] bridge: Add core IGMP snooping support
Date: Sun, 7 Mar 2010 10:45:00 +0800	[thread overview]
Message-ID: <20100307024500.GA20126@gondor.apana.org.au> (raw)
In-Reply-To: <20100306190000.GA24445@linux.vnet.ibm.com>

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.

Cheers,
-- 
Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt