Chris Leech wrote: > On Wed, Jul 08, 2009 at 02:13:29PM -0700, Love, Robert W wrote: >> On Wed, 2009-07-08 at 10:40 -0700, Chris Leech wrote: >>> Rather than rely on the hostlist_lock to be held while creating exchange >>> managers, serialize fcoe instance creation and destruction with a mutex. >>> This will allow the hostlist addition to be moved out of fcoe_if_create(), >>> which will simplify NPIV support. >>> >> Is there a reason that we don't use the rtnl_lock for protection? It's >> already held when we get the NETDEV_UNREGISTER event (added in patch >> 13/13). I believe that doing so would remove the need to defer the >> fcoe_destroy() call since the deferral is only needed so that we don't >> grab the rtnl_lock twice. > > Mostly I'm just not a fan of putting other people locks/mutexs to our own > use. I think it confuses the issue of what is the mutex protecting, but > I don't see any technical barriers to doing it that way. > > I thought there'd be more changes required; but I thought > dev_get_by_name() would be taking the rtnl mutex, it's not there's the > dev_base_lock protecting the list. > > unregister_netdevice_notifier() takes rtnl, but I don't see an issue > with re-taking the mutex after calling that in fcoe_exit().
Let's not drop and re-take it. Instead do the unregister before ever taking rtnl_lock. > We take rtnl around a few netdev related calls, but those code paths are > either pretty clearly in the create/destroy path or not. > _______________________________________________ > devel mailing list > [email protected] > http://www.open-fcoe.org/mailman/listinfo/devel _______________________________________________ devel mailing list [email protected] http://www.open-fcoe.org/mailman/listinfo/devel
