On Tue, Dec 7, 2010 at 7:23 AM, Johannes Berg <[email protected]> wrote: > On Tue, 2010-12-07 at 07:20 -0800, Paul Stewart wrote: >> On Tue, Dec 7, 2010 at 1:38 AM, Johannes Berg <[email protected]> >> wrote: >> > On Mon, 2010-12-06 at 18:48 -0800, Luis R. Rodriguez wrote: >> >> --- a/net/mac80211/pm.c >> >> +++ b/net/mac80211/pm.c >> >> @@ -301,7 +301,7 @@ static void __ieee80211_scan_completed_finish(struct >> >> ieee80211_hw *hw, >> >> } >> >> >> >> mutex_lock(&local->mtx); >> >> - ieee80211_recalc_idle(local); >> >> + ieee80211_recalc_idle_force(local); >> > >> > Does this really occur afterwards closing interfaces? >> >> It appears possible using wpa_supplicant with nl80211, at least using >> the new DBus API. > > Actually I believe it happens _during_ ifdown, but when the running bit > is already cleared, so I guess this is about right -- although it > shouldn't matter since the recalc_idle in do_stop should catch it later.
So the scenario is as follows. If you down an interface ieee80211_do_stop() first decrements local->open_count, then later calls __ieee80211_recalc_idle, et al. This means that later in this call graph if ieee80211_hw_config() finally gets called, open_count is already 0 and therefore drv_config is never called. Luis' reference to scan_completed() above is probable in reference to older versions of compat where ieee80211_stop() called ieee80211_scan_cancel() directly. My admittedly casual look at wireless-testing seems to imply that path may not exist in current compat, although my tests seem to indicate the same class of problem still exists. -- Paul _______________________________________________ stable mailing list [email protected] http://linux.kernel.org/mailman/listinfo/stable
