Re: [PATCH v2 03/10] cgroup: net_cls: Protect access to task_cls_classid() when built as module

On Sat, Aug 25, 2012 at 06:56:29PM +0200, Daniel Wagner wrote:

On 25.08.2012 01:26, Tejun Heo wrote:

quoted

On Fri, Aug 24, 2012 at 04:01:37PM +0200, Daniel Wagner wrote:

quoted

@@ -306,6 +312,11 @@ static void __exit exit_cgroup_cls(void)
 	synchronize_rcu();
 #endif

+#if IS_MODULE(CONFIG_NET_CLS_CGROUP)
+	static_key_slow_dec(&cgroup_cls_enabled);
+	rcu_barrier();

Why is this rcu_barrier() necessary?

I have read the rcubarrier.txt document and I got from that that an
rcu_barrier() is needed when unloading a module. But maybe I got it
wrong.

So the idea after disabling the jump lables all pending readers in
task_cls_classid() have left. THe same thing is done in the old code
with the dynamic id part. With the difference that synchronize_rcu()
is used.

FWIW, the rcu_barrier() is needed only if the module uses call_rcu().
In that case it is required to ensure that all the resulting callbacks
execute before the module's .text is freed up.

							Thanx, Paul

quoted

In general, please explain what
synchronization is going on when using sync constructs which aren't
obvious - e.g. memory barriers, rcu barriers.

Sure, I will keep this in mind.

--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html