On Sun, 2006-04-09 at 15:17 +1000, Herbert Xu wrote: > Right. You can't stop the queue for a period longer than the watchdog > timeout (default is 5s) on the device. In fact there is a bug here in > that neither bcm43xx nor the softmac layer is setting trans_start.
Hah. Alright, I'll look into that. I don't quite understand all the intricacies of that netdev stuff yet, so I'll probably have to pick your brain a couple of times more. > Without that the message will trigger any time when you stop the queue > and the watchdog timer fires while it's stopped. Yah that makes sense. Thing is that bcm43xx then resets the device which gets rid of everything you've done before. > So I suggest that you first of all make sure that trans_start is set. > Then you need to ask yourself is this going to take a longer time than > a reasonable TX timeout. If the answer is no then you're fine. > Otherwise you need to do something more drastic to stop the traffic. Well we reduced the dwell time to what, 20ms? So scanning shouldn't take longer than something like 250ms. Or 1.5 seconds if we go with a dwell time of 100ms. So that should be fine. I'll try to look into it, if anyone wants to beat me to it feel free to ;) Thanks, johannes
signature.asc
Description: This is a digitally signed message part
