On 2019-02-04 22:57, Stijn Segers wrote: > > > Op zondag 3 februari 2019 om 14:54 schreef Felix Fietkau <n...@nbd.name>: >> On 2019-02-02 19:08, Stijn Segers wrote: >>> Hi Felix, >>> >>> I built 18.06 HEAD with your 2019-01-30 bump, and with the >>> mt76-disable-edcca.patch applied. That seemed very stable (up since >>> 30/01 evening, very high throughput - never seen anything close to >>> 50 >>> MBps on my 2x2 client before; I barely got close to 25 MBps >>> before). No >>> laggy feeling, so didn't check ping. Wireless wasn't used in the >>> morning or the day, but I have tried to stress it in the evening >>> (31/01) -transferred about 65 GiB of data over the WLAN (scp). No >>> hiccups whatsoever. >>> >>> I then reverted to vanilla 18.06 HEAD 31/01, so none of the patches >>> applied, and the connection kept working but page loading in the >>> browser e.g. would occasionally be slow (nothing else chewing on my >>> network connection). I checked ping, and it was all over the place >>> again (from sub-milliseconds to 10 ms, occasionally 100ms to other >>> devices on the LAN). I suppose there's no real value in those >>> numbers, >>> but it does quantify the feeling of the wireless 'choking' >>> momentarily, >>> then picking up again. After like half an hour, wireless would hang >>> on >>> my laptop: I was unable to ping anything, web pages stopped loading >>> altogether. Same thing on my smartphone. However, when I disabled >>> wireless on both clients, then turned it on again, the connection >>> started working once more; that didn't happen on the previous mt76 >>> bumps. There the radio would just seem to die like explained in the >>> bug >>> report. >> Thanks, that information was very helpful. Did you run those tests on >> 2.4 GHz or 5 GHz? > > Hi Felix, > > 5 GHz only (I don't use the 2,4 GHz radio). > > >> >>> Since that hang already happened within the hour, I ran a new build >>> with the most recent mt76 bump (2019-01-31) together with the >>> 'disable >>> edcca' patch. That has been working fine so far (although I'm not >>> hitting 50 MBps anymore :-P ). It's been up since Friday afternoon, >>> so >>> almost 30 hours now. The radio gets shut down at night, but I've >>> been >>> using it for hours and no issues as far as I can tell. >>> >>> Do you want me to try the second patch, or is this OK >> No need. I've pushed a whole bunch of ED/CCA related patches to >> mt76.git >> (haven't bumped OpenWrt git yet). >> Could you please test that latest version (commit >> e301f2349a191c845b6bd2d487e64591ac2ed64d)? > > I just rebuilt your 2019-02-03 bump (copied the Makefile from master to > the 18.06 package feed), installed the kmods on top of my existing > 18.06 build and rebooted. > Initially it seemed fine, although the connection seemed a bit laggy at > first but that stabilised. My smartphone then just disconnected after > an hour (router uptime 1:15' +-) and the AP was not visible anymore > despite 'wifi status radio0' saying it was still up. Dmesg was > completely silent (nothing related to mt76 or the AP dying). > > 'Wifi reload' brought it back temporarily - it then only took a few > minutes to go down again though, for the AP to disappear again and this > second time the MCU timeout messages reappeared like in the bug report > on Flyspray. > > I didn't have either of your previous patches applied (so 'vanilla' > mt76 2019-02-03). I'm rolling back to vanilla 18.06 + your ED/CCA patch > for now. If you need any other info/need me to test/check other things, > let me know. (I hope this is software and not dying hardware, I > wouldn't be able to tell the difference.) > > Thanks for your help already so far. Thanks for testing! Does this issue happen for you only on 2.4 GHz, or also on 5 GHz? My patches seem to have resolved the hang for at least one other user. I suspect that this is way more likely to happen on 2.4 GHz because of all the extra noise, so I'm considering disabling ED/CCA for 2.4 GHz.
- Felix _______________________________________________ openwrt-devel mailing list openwrt-devel@lists.openwrt.org https://lists.openwrt.org/mailman/listinfo/openwrt-devel
