On Sun, Jan 13, 2019 at 05:54:00AM +0000, Parav Pandit wrote: > In _ib_dev_change_net_ns() there are only two functions that can fail. > One of them rdmacg_register_device() always return 0. > I should supply a patch to simplify this path regardless. I'd be much happier if it can't fail.. That makes far more sense > > Well, no, the device_mutex is supposed to protect the device and other lists, > > it is not a lock to be used against the entire registration flow. If a lock for > > that is needed then we'll need a new lock > > Ok. let me rework on this part of the series with new lock that > protects these operations and reduce dependency on device mutex. Okay, let me send you some stuff.. Jason
