On Wed, Mar 18, 2015 at 06:31:03PM +0100, Mattieu Baptiste wrote: > Anyone seeing this?
Yes, all the time. > I'll try to find this night the commit which is > causing this regression. If there was particular commit that made it much worse, that would be good to know. But I believe this problem has been present since the driver was added. I made sure to list frequent firmware timeouts in the man page as one of the outstanding bugs. Usually the firmware manages to recover after a few association attempts. The problem(s?) is/are in the transition from initial state to successful association. Once the driver manages to associate the connection is stable. I think this problem is caused by the net80211 state machine racing the firmware/hardware. I haven't yet found time to confirm this theory though. Testing the theory involves introducing synchronous register i/o operations in the driver. Currently they happen asynchronously to net80211 state changes (else the kernel panics very quickly). I don't know why the driver was written this way. Perhaps it's a remnant from the porting process from Linux. This driver has a peculiar history. OpenBSD adopted it but it wasn't originally written by us. In any case, please make sure you have 2 antennas connected to the device. The driver doesn't handle a missing antenna yet.
