On Fri, 2009-05-15 at 10:22 +0200, Johannes Berg wrote: > > +++ b/net/mac80211/pm.c > > @@ -18,6 +18,10 @@ int __ieee80211_suspend(struct ieee80211_hw *hw) > > > > flush_workqueue(local->hw.workqueue); > > > > + /* Don't stuff our workqueue during suspend->resume cycle */ > > + del_timer_sync(&local->dynamic_ps_timer); > > + cancel_work_sync(&local->dynamic_ps_enable_work); > > + > > ack, but you forgot to add a comment why dynamic_ps_disable_work doesn't > need to be handled (actually it does need to handled with the current > code but shouldn't be necessary... up for a challenge? -- why? and how > to fix that better than handling it here?) > > If you ever suspend in AP mode there's also an issue with each station's > cleanup timer, and for mesh each station's plink timer. Hmm, actually, this isn't safe. We might still be receiving frames at this point, which would start the timer again. > I don't think that helps, like I said the timers are still set up, you > need to actually queue the work now. However that kinda sucks for small > suspend times. Not really sure what to do. Instead of doing the setup_timer dance you were doing, we should simply do add_timer /iff/ del_timer_sync() returned 1, because that means we actually cancelled the timer. johannes
Attachment:
signature.asc
Description: This is a digitally signed message part
