On Monday 03 December 2012 21:09:06 Antonio Quartulli wrote:
But still we have the problem in batman-adv (as Sven pointed out in a previous email) that tries to unregister a device in that "critical window".
Exporting __rtnl_unlock() would solve the issue in this case.
If you think the bridge code should not end up in such situation, what if Simon resends the patch with only the __rtnl_unlock() exportation and the change in batman-adv?
I personally have big doubts about the removal of the second half of the unregister. Doesn't sound like the best idea. This would result in side effects... one of them for example would be the possible deadlock scenario moved to the other users of rtnl_trylock which don't unregister a device inside their critical section.... so either you do it always this way when using rtnl_trylock or it will break. I don't want to imagine other problems caused by this change.
Kind regards, Sven