Johannes Berg <johannes@xxxxxxxxxxxxxxxx> writes: > On Wed, 2024-04-24 at 12:50 +0300, Kalle Valo wrote: >> Johannes Berg <johannes@xxxxxxxxxxxxxxxx> writes: >> >> > On Wed, 2024-04-24 at 12:39 +0300, Kalle Valo wrote: >> > > >> > > Exactly. Swithing to use wiphy_lock() definitely one of my longterm >> > > goals in ath12k. I already started working on switching ath12k to use >> > > wiphy_lock() but IIRC I got blocked by not being able to call >> > > wiphy_delayed_work_cancel() without taking wiphy_lock. >> > >> > That's because I didn't want to have an async version, but theoretically >> > we could have a version of it that just cancels the timer. If you don't >> > hold the wiphy mutex already you could even wait for it to finish. It >> > just adds more complexity there, and I was trying to make it all a lot >> > more obvious :) >> >> Yeah, understandable. I think changing ath12k WMI event handling to use >> wiphy_work makes sense, running them in tasklet context feels overkill. > > Maybe. Note that the wiphy lock _can_ create delays etc. here, if you're > doing other configuration work at the same time, or similar. Though I > guess eventually you need locking there anyway, to access the HW. So it > can be a bit of a trade-off. > > I expect this will evolve over time even in mac80211, though so far we > haven't seen any bad effect from it. Good point. And now that I think of this more I don't think we can use wiphy_work with WMI events as we are waiting some events while holding wiphy_lock, that would end up as deadlock. So most likely we need to use a normal workqueue with WMI events and take wiphy_lock manually in certain cases. But I'll need to investigate more and do some experiments. -- https://patchwork.kernel.org/project/linux-wireless/list/ https://wireless.wiki.kernel.org/en/developers/documentation/submittingpatches
