Return-path: Received: from mail-iy0-f174.google.com ([209.85.210.174]:34832 "EHLO mail-iy0-f174.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1756383Ab0LPR6K convert rfc822-to-8bit (ORCPT ); Thu, 16 Dec 2010 12:58:10 -0500 Received: by iyi12 with SMTP id 12so1680211iyi.19 for ; Thu, 16 Dec 2010 09:58:09 -0800 (PST) MIME-Version: 1.0 In-Reply-To: <20101216174410.GB5677@tux> References: <1292482775.3542.8.camel@jlt3.sipsolutions.net> <1292499024.3612.3.camel@jlt3.sipsolutions.net> <1292501047.3612.9.camel@jlt3.sipsolutions.net> <1292502934.3612.15.camel@jlt3.sipsolutions.net> <1292504020.3612.17.camel@jlt3.sipsolutions.net> <20101216174410.GB5677@tux> From: "Luis R. Rodriguez" Date: Thu, 16 Dec 2010 09:49:17 -0800 Message-ID: Subject: Re: [PATCH] mac80211: Push idle state to driver before stop To: Paul Stewart Cc: Johannes Berg , "linux-wireless@vger.kernel.org" , Luis Rodriguez Content-Type: text/plain; charset=UTF-8 Sender: linux-wireless-owner@vger.kernel.org List-ID: On Thu, Dec 16, 2010 at 9:44 AM, Luis R. Rodriguez wrote: > On Thu, Dec 16, 2010 at 08:02:27AM -0800, Paul Stewart wrote: >> On Thu, Dec 16, 2010 at 4:53 AM, Johannes Berg >> wrote: >> > On Thu, 2010-12-16 at 04:46 -0800, Paul Stewart wrote: >> > Not really -- the driver may throw away all internal state, mac80211 >> > will (attempt to) restore it all through drv_config() with changed = ~0. >> > Evidently ath9k has some magic that makes this fail? >> >> The magic in question is as follows (ath9k_config()): >> >>     if (changed & IEEE80211_CONF_CHANGE_IDLE) { >>         bool enable_radio; >>         bool all_wiphys_idle; >>         bool idle = !!(conf->flags & IEEE80211_CONF_IDLE); >> >>         spin_lock_bh(&sc->wiphy_lock); >>         all_wiphys_idle =  ath9k_all_wiphys_idle(sc); >>        ath9k_set_wiphy_idle(aphy, idle); >> >>         enable_radio = (!idle && all_wiphys_idle); >> >>         /* >>          * After we unlock here its possible another wiphy >>          * can be re-renabled so to account for that we will >>          * only disable the radio toward the end of this routine >>          * if by then all wiphys are still idle. >>          */ >>         spin_unlock_bh(&sc->wiphy_lock); >> >>         if (enable_radio) { >>             sc->ps_idle = false; >>             ath_radio_enable(sc, hw); >>             ath_dbg(common, ATH_DBG_CONFIG, >>                     "not-idle: enabling radio\n"); >>         } >>     } >> >> So, ath9k enables and disables the radio based on ps_idle (the >> internal state), but that's not re-evaluated unless the wiphys >> actually change.  In this case they do not change because ath9k was >> never informed that the wiphy went idle (all_wiphys_idle evaluates to >> false).  One solution would be to do an ah9k_set_wiphy_idle(aphy, >> true) in ath9k_stop() instead of Luis' previous suggestion of setting >> sc->ps_idle directly.  Perhaps in January when I return from vacation >> I'll test this to make sure that performs as intended in the >> suspend-resume-while-associated case.  It's a little bit of a lie to >> do it in this case since in fact as far as mac80211 is concerned, the >> wiphy is not idle in this case. > > Try this: > > diff --git a/drivers/net/wireless/ath/ath9k/main.c b/drivers/net/wireless/ath/ath9k/main.c > index c68205d..3de3dbf 100644 > --- a/drivers/net/wireless/ath/ath9k/main.c > +++ b/drivers/net/wireless/ath/ath9k/main.c > @@ -1350,6 +1350,7 @@ static void ath9k_stop(struct ieee80211_hw *hw) >        ath9k_ps_restore(sc); > >        sc->ps_idle = true; > +       ath9k_set_wiphy_idle(aphy, true); >        ath_radio_disable(sc, hw); > >        sc->sc_flags |= SC_OP_INVALID; > diff --git a/drivers/net/wireless/ath/ath9k/pci.c b/drivers/net/wireless/ath/ath9k/pci.c > index 7ca8499..4538283 100644 > --- a/drivers/net/wireless/ath/ath9k/pci.c > +++ b/drivers/net/wireless/ath/ath9k/pci.c > @@ -310,6 +310,7 @@ static int ath_pci_resume(struct device *device) >        ath9k_hw_set_gpio(sc->sc_ah, sc->sc_ah->led_pin, 1); > >        sc->ps_idle = true; > +       ath9k_set_wiphy_idle(aphy, true); >        ath_radio_disable(sc, hw); > >        return 0; If this does not work then yes, I will argue that mac80211 is not turning us on when it needs to. I was concered about the resume case given that mac80211 only calls drv_start() when open_count is > 0. So we'll see. Luis