Subject: Re: [PATCH 000/002] Fix frequent reconnects caused by new
 conection monitor
From: Marcel Holtmann <marcel@holtmann.org>
To: Maxim Levitsky <maximlevitsky@gmail.com>
Cc: reinette chatre <reinette.chatre@intel.com>,
	linux-wireless <linux-wireless@vger.kernel.org>,
	linville <linville@tuxdriver.com>,
	Johannes Berg <johannes@sipsolutions.net>
In-Reply-To: <1249070739.25620.11.camel@maxim-laptop>
References: <1249056817.20593.1.camel@maxim-laptop>
	 <1249066338.30019.214.camel@rc-desk>
	 <1249067293.19089.10.camel@maxim-laptop>
	 <1249068449.23662.17.camel@localhost.localdomain>
	 <1249070739.25620.11.camel@maxim-laptop>
Content-Type: text/plain
Date: Sat, 01 Aug 2009 08:25:45 -0700
Message-Id: <1249140345.3491.5.camel@localhost.localdomain>
Mime-Version: 1.0
Sender: linux-wireless-owner@vger.kernel.org

Hi Maxim,

> > > > > Hi, here is the updated version of these two patches that fix the
> > > > > $SUBJECT issue.
> > > > > 
> > > > > I attach these (in case mailer mangles them), and reply with patches.
> > > > > 
> > > > > Tested both with low quality signal, and beacon loss.
> > > > > Lack of TX is found, every 30 seconds now, and quite reliable.
> > > > > Lack of beacons, triggers probe like it did every 2 seconds.
> > > > 
> > > > Thanks!
> > > > 
> > > > I've been running with this for two hours now with no disconnects. This
> > > > is where before the patches I would get disconnected after a few
> > > > minutes. I did get two "No probe response from AP xx:xx:xx:xx:xx:xx
> > > > after 500ms, try 1" messages in my log.
> > > This is normal, or at least can be normal, I patched the driver to
> > > display this message, when there is a probe timeout, but instead of
> > > disconnect, it retries, currently 5 times, but this can be even further
> > > increased is necessarily.
> > > (these messages are only in logs when verbose mac debugging is enabled)
> > > 
> > > I don't know exactly why probes aren't answered, but I strongly suspect
> > > that my AP sometimes 'goes out to lunch' and then answers, since
> > > typically after a failed probe it sends many replies.
> > > (Or it could be some buffering done by iwl3945 microcode). I currently
> > > can't monitor the connection from outside, but as soon as I can I see
> > > whether the above is true. Nevertheless if signal quality isn't great,
> > > there are valid reasons for probe loss, and it shouldn't cause all the
> > > fuzz (and since I use WPA2, every reconnection causes whole WPA
> > > handshake to be preformed, and this takes at least 2 seconds, and if a
> > > reconnection happens each 5 seconds, it gets very very annoying, and
> > > almost unusable.
> > 
> > I am testing your patches and so far so good. Seems to be working
> > perfectly fine. I see this in the logs:
> > 
> > [41027.333419] wlan0: detected beacon loss from AP - sending probe request
> > [41027.389260] wlan0: cancelling probereq poll due to a received beacon
> > [41027.793518] No probe response from AP 00:1c:f0:62:88:5b after 500ms, try 1
> > [41028.292731] No probe response from AP 00:1c:f0:62:88:5b after 500ms, try 2
> > 
> > Need to watch out if this pattern emerges and if the beacon loss trigger
> > might give us an indication. Maybe the ucode is just not ready then.
> Here (on my system) I see no beacon losses at all, like I said there
> could be many reasons behind packet losses, and best way to mitigate
> them is to retry.
> 
> Your logs indicate that beacons weren't recieved for 2 seconds, then
> mac80211 tried to send a probe, but a beacon is recieved before the
> probe answer, this probe is canceled (at least should be) then after a
> while, a probe request (same one?) is time outed, and retried twice,
> then finally answered.

it looked related, but it wasn't at all. I have this running for over 24
hours by now and the patches work perfectly fine. Today it saw for the
first time a try 4 message. Otherwise it only had to try up to three
times before it succeeded.

Tested-by: Marcel Holtmann <marcel@holtmann.org>

Regards

Marcel