On Wed, 2007-09-05 at 10:23 -0400, Michael Wu wrote: > That's just a general rule of thumb. configure_filter holds the tx lock so > it's actually safe to do that there. Good point. > In other cases, it can be buggy (some tx > code can get stuck) and racy (frames can get enqueued after stopping the > queues). The first can be fixed in mac80211, but the second can't be. I think that we should be able to get all this resolved when we rip out the queue code and implement multiqueue devices. But if I want to do that I need to reverse engineer b43 QoS first to have multiple queues ;) > For > most cases, stopping the hardware queue(s) should be done instead and > possibly flushing the queue or waiting for it to empty, if you're trying to > adjust the radio. In this case it's about adjusting the MAC (filtering) and I'm not even sure the the USB device has a queue that can be stopped. johannes
Attachment:
signature.asc
Description: This is a digitally signed message part
