On Mon, Mar 09, 2015 at 11:31:56AM +0100, Martin Pieuchot wrote: > On 07/03/15(Sat) 14:13, Remi Locherer wrote: > > On Fri, Mar 06, 2015 at 11:24:27PM +0100, Martin Pieuchot wrote: > > > [...] > > > Could you tell me if the diff below fixes your issue? > > > > I built a kernel with this diff and the diff from Stefan applied on a > > cvs checkout from yesterday. I could provoke an urtwn0 timeout with running > > speedtest-cli. > > > > * TX status and page fault trap: > > https://relo.ch/urtwn_freeze/IMG_20150307_134913.jpg > > * trace cpu 2,0,1,3: > > https://relo.ch/urtwn_freeze/IMG_20150307_135150.jpg > > > > I tested the same with urtwn plugged to the xhci port. The timeout also > > happens but I can unplug the device without the system dropping to ddb. > > Thanks Remi, the problem is that when a pipe is aborted the stack call > the corresponding *_abort() function even if the transfer has not yet > been transfered. In this case it is not in the list of pending transfer > and we should not try to remove it. > > Updated diff below to apply on top of a -current checkout.
This diff fixes the issue for me. I can now remove the urtwn device once the "urtwn0 timeout" happens and the system continuous to work. Remi