Ath11k: memory leak seemingly fixed on latest Snapshot build

Follow-up to this: OpenWrt 24.10.0-rc7 - Seventh Release Candidate - #56 by dspalu32

Presumably related to the change to 6.12 kernel 2 days ago, an ongoing qualcommax/ath11k memory leak has suddenly disappeared. It was still present on snapshot 27/04/2025, but is now good on a build from 20/05/2025

I used to have a 3-day reboot on cron to avoid OOM panics, now no longer necessary and with no other config change:

Every snapshot and release since late-2022 has suffered similar memory leaks on this unit. I imagine it's something specific about my multi-AP+WDS ath11k config. My other two AX3600s didn't have any such issues on the same builds for the past years, but with more basic ath11k use.

Thanks @robimarko!

1 Like

most likely, I'm still using a Feb 2023 snapshot on two WRX36 AP's, uptime is currently 101 days.
last reboot was caused by a power outage, not by an issue with the units.

Feedback after 1 week: the memory leak is still absent. No upward trend whatsoever. All good.

2 Likes

Which commit do you use?
I’m running bf917ea792d890f7ea431bd3dffb763c231857e0. I’m seeing the ‘same’ memory leak on my GL.Inet-AXT1800 (qualcommax/ath11k).

I’ve built and used snapshots on 20/05, 03/06, 26/06, 06/07 and 30/07. The memory leak is absent on each - I’ve been actively checking and logging RRD stats. The reported version of 30/07 is r30625-d802e6310a

Do you mind sharing your OpenWRT config as well as you WiFi config with me?

You’re likely starting a wild goose chase, but I’ll dig it out when next home. The received wisdom is that this is all in interplay between the ath11k driver and opaque firmware blob - and I believe that too.

I have three AX3600s: one as WAN router+AP, two as ‘dumb’ WDS-AP repeaters. All run the same firmware builds - generally snapshots - and have near-identical WiFi configs with the exception of WDS master vs. slave.

In the early days one of the dumb APs with lower WiFi traffic had a memory leak. A workaround was to generate artificial traffic with an iperf cron job. This was fixed in properly through ath11k/kernel at some point in 2022.

In 2023 the busy WAN AP developed the memory leak I link to above. I tried configuration changes, ath11k module/firmware reloads, periodic traffic, etc. all to no avail. I made zero progress tracing why, even having the benefit of my side-by-side A vs B router setup. Maybe you’ll have more luck.

Not to sound negative: aside from these past inconveniences, ath11k has been rock solid and performant for me. Sustained ~50% total RAM usage on these 512Mb AX3600s.