On Mon, May 30, 2022 at 01:07:25PM +0300, Kalle Valo wrote: > Sascha Hauer <s.hauer@xxxxxxxxxxxxxx> writes: > > > On Mon, May 30, 2022 at 12:25:13PM +0300, Kalle Valo wrote: > >> Sascha Hauer <s.hauer@xxxxxxxxxxxxxx> writes: > >> > >> > Another problem to address is that the driver uses > >> > ieee80211_iterate_stations_atomic() and > >> > ieee80211_iterate_active_interfaces_atomic() and does register accesses > >> > in the iterator. This doesn't work with USB, so iteration is done in two > >> > steps now: The ieee80211_iterate_*_atomic() functions are only used to > >> > collect the stations/interfaces on a list which is then iterated over > >> > non-atomically in the second step. The implementation for this is > >> > basically the one suggested by Ping-Ke here: > >> > > >> > https://lore.kernel.org/lkml/423f474e15c948eda4db5bc9a50fd391@xxxxxxxxxxx/ > >> > >> Isn't this racy? What guarantees that vifs are not deleted after > >> ieee80211_iterate_active_interfaces_atomic() call? > > > > The driver mutex &rtwdev->mutex is acquired during the whole > > collection/iteration process. For deleting an interface > > ieee80211_ops::remove_interface would have to be called, right? > > That would acquire &rtwdev->mutex as well, so I think this should be > > safe. > > Can you add a comment to the code explaining this? Sure. > And > lockdep_assert_held() is a good way to guarantee that the mutex is > really held. Yes, Ping-Ke already pointed that out. Will add in the next round. Sascha -- Pengutronix e.K. | | Steuerwalder Str. 21 | http://www.pengutronix.de/ | 31137 Hildesheim, Germany | Phone: +49-5121-206917-0 | Amtsgericht Hildesheim, HRA 2686 | Fax: +49-5121-206917-5555 |