On 05/25/2020 02:25 AM, Johannes Berg wrote:
On Wed, 2020-05-13 at 06:56 -0700, Ben Greear wrote:
Hello,
I've been running some stress tests against OpenWrt APs. My AP partially died yesterday
during a test and did not recover. It looks like firmware became non responsive,
but then it could not recover due to what I think is bugs I fixed long ago in
the kernel that I normally use. The OpenWrt image is using my ath10k-ct driver and firmware,
but stock OpenWrt kernel/mac80211 as far as I know.
I think there are several issues:
1) The WARN-ON about check-sdata-in-driver should be a warn-on-once. Spamming serial
console logs like this is both very slow and also useless for debugging.
I posted a patch with this title some time back to mitigate this problem:
"mac80211: Don't spam kernel with sdata-in-driver failures."
2) I suspect that the ensuing lockup may be fixed by this patch I posted
back in 12/1/2016:
"mac80211: do not iterate active interfaces when in re-configure"
I have been running both of those patches since posting them to the list, so they have good
soak time in some strenuous wifi usage cases.
Huh, these are kinda old I guess.
Can you repost them?
For the warning log-spam, you could grab patch 3 and 4 from this series I posted late last year:
[PATCH 00/10] Ben's grab bag of mac80211 patches
I think patch 5 in there is also nice to have, we passed email about it, but not sure it ever
made it upstream.
I'll dig out the 'iterate active' patch and post it stand-alone.
Thanks,
Ben
johannes
--
Ben Greear <greearb@xxxxxxxxxxxxxxx>
Candela Technologies Inc http://www.candelatech.com
