OPNsense Forum

English Forums => 26.7, 26.10 Series => Topic started by: cercle on November 21, 2026, 05:30:51 PM

Title: Kernel panic ath0 / WLE200NX (ieee80211_beacon_construct) still present in 26.7.
Post by: cercle on November 21, 2026, 05:30:51 PM
Hi all,

Follow-up of https://forum.opnsense.org/index.php?topic=41816.0 (topic locked).

OPNsense 26.7.9_3 (FreeBSD 15.1-RELEASE-p6, stable/26.7), same hardware as
in that topic: PC Engines APU6 + Compex WLE200NX (AR9280), two hostap VAPs on
ath0 (two SSIDs, one bridged into LAN, one routed on its own subnet).

Note: channel is already set to Auto (the workaround found in that topic).
It prevents the panic at interface configuration, but not when a VAP changes
state under client load.

What happens

Same panic and same backtrace as in the old topic, in the "ath0 net80211
taskq" thread, always with fault virtual address 0xffff:

Fatal trap 12: page fault while in kernel mode
fault virtual address = 0xffff
fault code            = supervisor read data, page not present
current process       = 0 (ath0 net80211 taskq)

ieee80211_beacon_construct() at ieee80211_beacon_construct+0x7d
ieee80211_beacon_alloc() at ieee80211_beacon_alloc+0xb3
ath_beacon_alloc() at ath_beacon_alloc+0x84
ath_newstate() at ath_newstate+0x3f0
ieee80211_newstate_cb() at ieee80211_newstate_cb+0x1f8
taskqueue_run_locked() at taskqueue_run_locked+0x182
taskqueue_thread_loop() at taskqueue_thread_loop+0xc2

It is now reproducible on demand: the panic triggers when a VAP changes state
while clients are active. Today I got 2 panics out of 2 attempts with client
traffic (a Mac pinging the firewall over wifi + phones idle in power-save):

- during a normal "shutdown -r now", right after the 'config' stop script
- with a custom early stop script doing "pkill hostapd" + "ifconfig
  ath0_wlanX down": the panic simply moved into that step

I also had 2 runtime panics with the same backtrace (2026-09-09 and
2026-10-14), after repeated "stuck beacon" resets. hw.ath.bstuck=8 reduced
the reset frequency but did not prevent the panic.

New detail supporting Franco's locking theory

In the old topic, Franco suggested kern.smp.disabled=1 as a diagnostic and
mentioned a possible locking issue. In one textdump, another CPU was
concurrently in the RX / power-save path of the same driver at the time of
the panic:

Tracing command kernel pid 0 tid 100065 (ath0 taskq) (CPU 0)
lock_delay() at lock_delay+0x12
ieee80211_set_tim() at ieee80211_set_tim+0xfa
ieee80211_node_pwrsave() at ieee80211_node_pwrsave+0xd2
hostap_input() at hostap_input+0x916
ieee80211_input_mimo() at ieee80211_input_mimo+0x1f0
ath_rx_pkt() at ath_rx_pkt+0x656
ath_rx_proc() at ath_rx_proc+0x26d
taskqueue_run_locked() at taskqueue_run_locked+0x182

So it really looks like a race between the beacon (re)allocation on VAP
state change (CPU 3) and RX power-save / TIM handling (CPU 0), probably made
worse by having two VAPs sharing the beacon slots of one radio.

Side note on firmware

With the stock PC Engines coreboot (v4.19.0.1) the panic is written to swap
and the box reboots by itself. With a self-built Dasharo SeaBIOS
(v26.09.00.01) the same panic left the box frozen, without a dump, until a
power cycle. I went back to the PC Engines ROM and also enabled the hardware
watchdog (amdsbwd + watchdogd -t 128 -x 300) as a safety net.

Attachments

- textdump-0-redacted.tar: panic during "shutdown -r now" (after the
  'config' stop script)
- textdump-1-redacted.tar: panic during "pkill hostapd" + "ifconfig
  ath0_wlanX down" (early stop script)

WAN IP redacted in both. I'm happy to test a patch or a debug kernel on
this box.

Thanks in advance for any help or pointers!

cercle
Title: Re: Kernel panic ath0 / WLE200NX (ieee80211_beacon_construct) still present in 26.7.
Post by: nero355 on November 21, 2026, 07:53:18 PM
I think this still stands : https://forum.opnsense.org/index.php?msg=208662

Recently I have discovered that the by TP-Link EOL-ed EAP615-Wall v1.0 has a nice OpenWRT port : https://openwrt.org/toh/tp-link/eap615-wall
But for stock firmware has a nice webGUI too with almost everything you might need :)

I just wish the price was a lot lower than it currently is, but after 2 weeks of e-mailing with TP-Link I think that's not going to happen...