On 03/13/2017 07:00 AM, Luca Coelho wrote: > On Fri, 2017-03-10 at 08:41 -0700, Jens Axboe wrote: >> On 03/10/2017 08:36 AM, Luca Coelho wrote: >>> On Fri, 2017-03-10 at 08:23 -0700, Jens Axboe wrote: >>>> On 03/10/2017 05:01 AM, Luca Coelho wrote: >>>>> Hi Jens, >>>>> >>>>> On Thu, 2017-03-09 at 21:41 -0700, Jens Axboe wrote: >>>>>> On 03/01/2017 09:10 PM, Jens Axboe wrote: >>>>>>> On 03/01/2017 08:33 PM, Luca Coelho wrote: >>>>>>>> Hi Jens, >>>>>>>> >>>>>>>> On Mar 1, 2017 20:25, Jens Axboe <axboe@xxxxxxxxx> wrote: >>>>>>>> >>>>>>>> Not that folks have been jumping all over this, but in case someone is >>>>>>>> curious, it triggered twice here today. For those two times, the value >>>>>>>> of mvm->pending_frames[sta_id] was 80 and 39, respectively. >>>>>>>> >>>>>>>> Sorry for the delay, I'm on vacation now with limited internet access. >>>>>>>> But we'll take a look into this early next week at the latest. >>>>>>>> >>>>>>>> Thanks a lot for the detailed report! >>>>>>> >>>>>>> No worries, thanks for responding. I just wanted to ensure this wasn't >>>>>>> dropped on the floor. >>>>>>> >>>>>>> BTW, a few more values of ->pending_frames[sta_id]: >>>>>>> >>>>>>> $ dmesg | grep "ret=" >>>>>>> [ 2334.308254] ret=39 >>>>>>> [ 7915.311828] ret=80 >>>>>>> [31602.317204] ret=41 >>>>>>> [32139.510993] ret=54 >>>>>>> [33292.917759] ret=96 >>>>>>> >>>>>>> it seems to often happen around resume. >>>>>> >>>>>> This is still happening all the time in current -git. >>>>> >>>>> Could you collect traces with trace-cmd, as explained in our wiki[1]? >>>>> This will probably help point out the problem. I know it's a bit >>>>> difficult because it appears to happen randomly for you, but it's worth >>>>> trying. >>>> >>>> Sure I can, but honestly I'm a little puzzled that nobody else can >>>> reproduce this, it happens every time I resume of switch access points. >>>> Is anyone trying to reproduce this? >>>> >>>> I'll have to recompile with iwlwifi tracing enabled, then I'll send a trace >>>> when it happens. >>> >>> Are you using 4.11-rc1? Or linus' master? Or...? >> >> The trace I just sent is tip of Linus' tree. It's happened continually >> since the commit I mentioned in my initial report was merged: >> >> commit 94c3e614df2117626fccfac8f821c66e30556384 >> Author: Sara Sharon <sara.sharon@xxxxxxxxx> >> Date: Wed Dec 7 15:04:37 2016 +0200 >> >> iwlwifi: mvm: fix pending frame counter calculation > > I found the patch that fixes this issue in our internal tree. I'll send > it out for you to try now. > > The reason is that in DQA (Dynamic Queue Allocation) mode that we > introduced recently, we should not be counting the frames in the same > way as before. The warning was introduced exactly to catch this kind of > problems. > > Please let me know if it works for you! Seems to work for me, thanks! You can add my Tested-by: Jens Axboe <axboe@xxxxxx> to the patch. -- Jens Axboe
