> > <-- snip --> > > Are you sure thats safe? The lock is a mutex, and it appears that we get > called > from a few contexts which may not be safe to grab a mutex in (from > cnic_netdev_event for instance: > cnic_netdev_event > cnic_ulp_exit > cnic_exit > bnx2fc_ulp_exit > bnx2fc_if_destroy > fc_exch_mgr_free > > I agree that using the lport lock seems like the easy solution and one > that > would be free of the race vasu described, but I'd hate to trade in what > we have > now for lockdep warnings and potential deadlock. > Neil The calling path you have here is already grabbing some mutex, e.g., bnx2fc_dev_lock in bnx2fc_ulp_exit, similarly in fcoe_if_destroy that calls fc_exch_mgr_free(), we grab rtnl_lcok, therefore, I would think we should be fine, it will be nested...I do agree with you we want to be sure about this. I will dig more.
However, the issue is more on what Vasu is asking, is it probably somehow the corresponding exch/seq is gone in fc_fcp_complete_locked()? yi _______________________________________________ devel mailing list devel@open-fcoe.org https://lists.open-fcoe.org/mailman/listinfo/devel
