Re: [PATCH v2 1/3] wifi: ath12k: fixes to improve MLO station stability

[Jose Ignacio Tornos Martinez <jtornosm@redhat.com>]

Hi Baochen,

I agree that per-callback guards don't scale well long-term. Johannes
can comment better because both alternatives need mac80211 changes,
but let me share my thoughts.

Option 1 (recovery inside .start()) holds rtnl for ~1.3s, that seems
too long to hold.

Option 2 (split ieee80211_restart_hw() into early barrier + deferred
restart) looks like the right direction. It would benefit any driver
with async recovery, not just ath12k. But it is a mac80211 core change
and a bigger effort, and it still needs in_reconfig checks added to
callbacks that don't have them today (scan, for one), so the guarding
moves to mac80211 rather than disappearing.

In the meantime, the crashes this series fixes are real: without it,
an MLO firmware crash gives NULL derefs and 20+ second recovery. The
guards here cover the callbacks that actually run during that window,
not a speculative list.

I am happy to work on the mac80211 side (option 2), but would it make
sense for you to land this series as an interim fix while that is
worked out as a follow-up?

Thanks

Best regards
José Ignacio