[OpenWrt-Devel] ath10k QCA9880 firmware TX hang

[David Bauer]

Hello,

while preparing the switch to 19.07 for the mesh-framework Gluon, we found an issue
with ath10k, which results in an overflowing TX queue in the kernel. The QCA9880 does
not send TX_DONE or TX_FAILED messages back to the driver, which results in the SKBs never
being freed. Both HTT_COOKIES counters are also frozen when this happens.

This happens when using 10.2.4-1.0-00045 (current in linux-firmware) and
10.2.4-1.0-00037 firmware (shipped with OpenWrt 18.06). However, it seems to be fixed in
10.2.4-1.0-00047.

We have some statistical data about this issue here. [1]
As you can see, the issue started with the switch to OpenWrt on 26. September and
was resolved on 5th October (we did some testing afterwards, so we didn't reach a continuous
uptime)

We've also tried ath10k-ct firmware (which does not seem to show the issue), however this is
an option for us, as we need 802.11s meshing on our devices (the described issue however does also
happen when using a single AP VIF).

As an interim fix, i would propose to switch to the 10.2.4-1.0-00047 firmware for the QCA9880/QCA9887.
This would include switching to ath10k-firmware repository from kvalo for both firmware files.

Any objections on this?

[1] https://stats.darmstadt.freifunk.net/d/000000021/router-meshviewer-export?orgId=1&var-node=f09fc2dec4c5&from=1569466967823&to=1570394540260

Best wishes
David



_______________________________________________
openwrt-devel mailing list
openwrt-devel at lists.openwrt.org
https://lists.openwrt.org/mailman/listinfo/openwrt-devel