Date: Wed, 5 Oct 2011 13:15:08 +0200
From: Stanislaw Gruszka <sgruszka@redhat.com>
To: Johannes Berg <johannes@sipsolutions.net>
Cc: linux-wireless@vger.kernel.org,
	Ben Greear <greearb@candelatech.com>
Subject: Re: [RFC] mac80211: remove per band sta supported rates
Message-ID: <20111005111507.GA2184@redhat.com> (sfid-20111005_131732_330818_E6F18326)
References: <1317121970-3638-1-git-send-email-sgruszka@redhat.com>
 <1317123289.4082.12.camel@jlt3.sipsolutions.net>
 <20110929150018.GA4554@redhat.com>
 <1317740824.6741.22.camel@jlt3.sipsolutions.net>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
In-Reply-To: <1317740824.6741.22.camel@jlt3.sipsolutions.net>
Sender: linux-wireless-owner@vger.kernel.org

On Tue, Oct 04, 2011 at 05:07:04PM +0200, Johannes Berg wrote:
> On Thu, 2011-09-29 at 17:00 +0200, Stanislaw Gruszka wrote:
> > On Tue, Sep 27, 2011 at 01:34:49PM +0200, Johannes Berg wrote:
> > > On Tue, 2011-09-27 at 13:12 +0200, Stanislaw Gruszka wrote:
> > > > It is not possible to connect to remote station on two bands
> > > > at once, or I'm wrong?
> > > 
> > > I don't think it is, but there could be channel switches maybe?
> > 
> > If AP would like to change to different band, it will need provide new
> > Supported Rates element, or operate on older one.
> > 
> > So this could be simplified, but from other hand this helps to catch
> > bugs, so maybe would be better to keep it. As long some other warning
> > would be added to check that we send on proper channel.
> 
> Yeah I was going to say ... the main purpose of this these days seems to
> be catching this tx-on-wrong-band (which should be channel) ...

There is a old patch from Luis, which add channel to sta_info,
http://www.spinics.net/lists/linux-wireless/msg56399.html
I could incorporate it into my sta->supp_rates[BAND] removal patch,
so we will not loose debuggability, but event extend it.
 
> > > > Which may happen when we are just after disassociation and changed
> > > > channel/band, but still want send some frames (namely delBA) to old
> > > > sta. Right fix should prevent to change channel before we fully
> > > > dissassocate, or prevent to send frames after connection is lost,
> > > > or both, but I don't know how to correctly do this so far.
> > > 
> > > Well, either we should simply not send the frame, or send it before
> > > disassoc, no?
> > 
> > I think I can fix this that way, but I'm not sure if that would be right
> > fix either. We have few instances of warning:
> 
> I think that would be the right fix. In some cases, we may want to
> reorder things instead of just not sending frames.
> 
> > 1) started at ieee80211_sta_connection_lost()
> > https://bugzilla.redhat.com/show_bug.cgi?id=731365#c0
> > could be fixed by:
> > ieee80211_set_disassoc(..., false);
> > ieee80211_send_deauth_disassoc(, false);
> 
> Why would this trigger the problem? Can we somehow lose connection while
> already trying to connect to a new AP?

Not sure if I understand. I think that change should not influence
new AP connection. Also if we really loose connection to old AP, not
sending deauth/disassoc frames does not matter. It can have matter if
for some reason old AP start to become available again, so we properly
disassociate. But for such corner case we should rather not care.

> > 2) started at ieee80211_tx_status()
> > https://bugzilla.redhat.com/show_bug.cgi?id=731365#c11
> > could be fixed by adding association check before 
> > ieee80211_send_bar()
> 
> That seems reasonable. No use in trying to tear down BA sessions when
> we're already disconnected.
> 
> > 3) started at ieee80211_offchannel_return()
> > https://bugzilla.redhat.com/show_bug.cgi?id=737993#c0
> > no idea how to fix
> 
> Hm, seems like maybe offchannel return is telling the AP that we woke
> up, but maybe the offchannel work disconnected or something?
> 
> > All these problems looks like channel switching issue -
> > - we changed channel, whereas we should still operate on old one.
> > Fedora 15 users start to report that WARNING after update to 3.0
> > from 2.6.38, so this could be related to Ben offchannel work.
> 
> That's very well possible.
> 
> Thanks for looking into all this!

Unfortunately problem is not reproducible for me, so I can not debug.
I'm currently review channel switching on mac80211, to try to understand
how things work. Hopefully I'll be able to send reasonable patches
soon.

Stanislaw