Re: Any insight on this ubuntu bug?

[Dan Williams <dcbw redhat com>]

On Tue, 2010-01-26 at 16:31 -0500, Tony Espy wrote:
> Dan Williams wrote:
> > On Tue, 2010-01-26 at 15:44 -0500, Tom Sutherland wrote:
> >> https://bugs.launchpad.net/ubuntu/+source/linux/+bug/352228
> >>
> >> Wireless gets disconnected every so often and it appears the only
> >> workaround is to do a "sudo restart network-manager"
> >>
> >> It makes sense that this is a crappy firmware/driver issues with
> >> specific Intel cards, but is it possible network-manager isn't coping
> >> well?
> > 
> > I seem to recall a kernel patch for this bug being submitted in the past
> > week or two.
> > 
> > Basically, it's a driver bug.  I'd be very skeptical of a simple NM
> > restart fixing this as that doesn't do anything reinitialize or fix the
> > driver.  An rmmod/modprobe is necessary in situations like that to
> > re-load the driver and get it talking to the card again.
> 
> I just added a comment to the bug pretty much echoing Dan.
> 
> That said, I also think there's a possibility that the system calls made 
> by NM and wpa_supplicant during a restart might be sufficient to jog the 
> driver into working again.  Just a theory...

Yeah, the only interesting thing there would be IFF_UP/IFF_DOWN.  And
that may actually be enough, since I think on IFF_DOWN the iwl driver
will unload firmware?  I forget what it does.

A thought:

Down the device in NMDeviceWifi's activation_failure_handler() and
schedule an idle handler for 250ms from now to up the device.  Clear
that idle handler in stage1_device_prepare, and then bring the device up
there too.  Otherwise let the idle handler fire and re-up the device.
Also add a check to scanning_allowed() so that if the re-up idle handler
is non-zero, we don't allow scans (most devices can't scan when they are
down).

That might be a workaround.  Of coruse before adding something like that
we should probably see if it makes any difference.

Dan