[Commotion-dev] Alternate mac80211 / hostapd from nbd for ath9k driver freeze issue
Ben West
ben at gowasabi.net
Tue Jan 14 21:05:58 UTC 2014
Howdy,
Giving you a heads up about a problem I've encountered on UBNT Nanostation
Loco M2, Nanostation M5, and TP-Link TL-MR3020. WLAN Slovenija also saw it
on other TP-Link products. Good chance Commotion-OpenWRT might be seeing
this driver freeze, too.
Under heavy load conditions, the ath9k driver bundled with recent versions
of AA may freeze, and all broadcast SSIDs disappear. For me, this was also
causing additional weird side-effects like hostapd/wpa_supplicant crashing,
and olsrd complaining " OLSR: sendto IPv4 Resource temporarily
unavailable," depending on which vif happened to lock up.
The lockup does appear to correlate well with heavy traffic load, i.e.
never see this problem on nodes with very light traffic. Also, it's very
intermittent. Nodes with a weekly reboot schedule might see this lockup a
couple times a month or so, in my experience.
This is the ticket filed in OpenWRT issue queue:
https://dev.openwrt.org/ticket/13681
This is the updated version of mac80211 and hostapd which nbd has offered
to address the driver lockup issue under AA. I've reflashed my affected
problem nodes over the past couple weeks (AA r39154), and so far so good.
I presume these versions will eventually be backported into AA.
http://nbd.name/gitweb.cgi?p=aa-mac80211.git;a=summary
Finally, this is the script from WLAN Slovenija to detect the freeze and
trigger a reboot. I've incorporated this script as-is in WasabiNet
firmware for Nanostation Loco M2s and the TL-MR3020s.
https://github.com/wlanslovenija/firmware-packages-opkg/blob/master/util/nodewatcher-watchdog/files/check_ath9k_freeze
For Nanostation M5s, the string the check for is "qdepth: 0 ampdu-depth: 0
pending: 124 stopped: 1".
--
Ben West
http://gowasabi.net
ben at gowasabi.net
314-246-9434
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.chambana.net/pipermail/commotion-dev/attachments/20140114/dd84cd8d/attachment.html>
More information about the Commotion-dev
mailing list