On Tue, 24 Aug 2004 22:35:10 +0200 Bart De Schuymer <bdschuym@xxxxxxxxxx> wrote: > On Tuesday 24 August 2004 20:38, Stephen Hemminger wrote: > > How about this, it is basically Bart's patch with: > > * some more cleanups to handle_bridge > > * removing extra rcu_read_lock (already done in netif_receive_skb) > > * pass bridge port to hook since already dereferenced in handle_bridge > > Looks good. > I'm not sure if the rcu_dereference() is needed, however. Can the bridge port > be removed while a packet is being processed? I thought the rcu_read_lock() > protects against that (del_nbp() does call_rcu(&p->rcu, destroy_nbp_rcu)). > > cheers, > Bart > > rcu_read_lock is not really a lock, it just turns off preempt. So, rcu_dereference is needed on CPU with weak consitency like Alpha. It handles the case where a packet is processed between del_nbp() and destroy_nbp_rcu where the bridge port still exists, but is being deleted.
