Re: Another dropped connection with iwl3945 and NM-0.6.5 on FC7

[Dan Williams <dcbw redhat com>]

On Thu, 2007-12-06 at 00:21 +0100, Christian Deckelmann wrote:
> On Wed, Dec 05, 2007 at 12:47:04PM -0500, Derek Atkins wrote:
> 
> > Dan Williams <dcbw redhat com> writes:
> > 
> > > Looks like the driver decided that the AP had gone away.  It probably
> > > missed too many beacons or probe responses due to the RF heavy
> > > environment or something?  Could need a driver tweak.
> > 
> > Maybe.  It's the iwl3945.  Maybe it was scanning? ;)
> > The IETF really is a great testbed.  Frustrating as a user,
> > but a great testbed.
> 
> That reminds me on a problem I have seen a few weeks ago.
> Something (maybe NM) triggered a scan every few (1-3) minutes.

This was likely NM, yes, doing its periodic scan.

> The iwl3945 driver had (or still has) a bug where no traffic has been 
> transfered while it was scanning.

This isn't actually a bug; drivers simply cannot transfer traffic while
scanning because they aren't on the same channel as the AP, and so any
traffic sent while the card was on a different channel during the scan
wouldn't be heard by the AP.

The solution is to:

1) Enter power-save poll mode so that the AP buffers incoming traffic
for you, do the scan, and return to the associated channel and exit
power-save poll mode

2) Buffer local outgoing traffic until the scan is done

> That caused a roaming event.
> IIRC NM even logged that in /var/log/NetworkManager.
> The scan also showed up in iwevent.
> 
> IIRC iwl has been fixed to not interrupt traffic while scanning.

Yeah, there have been a few fixes for scanning behavior lately, most
notable is the "don't scan while associating" patch which I can't
believe was overlooked in the first place...

Dan