On Mon, Dec 11, 2023 at 07:47:32AM +0100, Greg KH wrote:
> On Sun, Dec 10, 2023 at 09:39:30PM +0000, Léo Lam wrote:
> > Commit 4a7e92551618f3737b305f62451353ee05662f57 ("wifi: cfg80211: fix
> > CQM for non-range use" on 6.6.x) causes nl80211_set_cqm_rssi not to
> > release the wdev lock in some situations.
> >
> > Of course, the ensuing deadlock causes userland network managers to
> > break pretty badly, and on typical systems this also causes lockups on
> > on suspend, poweroff and reboot. See [1], [2], [3] for example reports.
> >
> > The upstream commit, 7e7efdda6adb385fbdfd6f819d76bc68c923c394
> > ("wifi: cfg80211: fix CQM for non-range use"), does not trigger this
> > issue because the wdev lock does not exist there.
> >
> > Fix the deadlock by releasing the lock before returning.
> >
> > [1] https://bugzilla.kernel.org/show_bug.cgi?id=218247
> > [2] https://bbs.archlinux.org/viewtopic.php?id=290976
> > [3] https://lore.kernel.org/all/87sf4belmm.fsf@xxxxxxxxxxxxx/
> >
> > Fixes: 4a7e92551618 ("wifi: cfg80211: fix CQM for non-range use")
> > Cc: stable@xxxxxxxxxxxxxxx
> > Signed-off-by: Léo Lam <leo@xxxxxxxxx>
> > ---
> > net/wireless/nl80211.c | 18 ++++++++++++------
> > 1 file changed, 12 insertions(+), 6 deletions(-)
>
> So this is only for the 6.6.y tree? If so, you should at least cc: the
> other wireless developers involved in the original fix, right?
>
> And what commit actually fixed this issue upstream, why not take that
> instead?
I've reverted the offending commit in the last 6.1.y and 6.6.y release,
so can you send this as a patch series, first one being the original
backport, and the second one this one, AFTER it has been tested?
thanks,
greg k-h
