Re: [BUG] iwl4965: huge latency and errors when unblocking the hardware rfkill

[Dan Williams <dcbw@redhat.com>]

On Mon, 2008-07-07 at 16:01 -0400, Andrew Lutomirski wrote:
> On Mon, Jul 7, 2008 at 3:45 PM, Dan Williams <dcbw@redhat.com> wrote:
> > On Mon, 2008-07-07 at 14:56 -0400, Andrew Lutomirski wrote:
> >> This is wireless-testing c80200cd38c265da90f0d9d031ace84aa56b0453, pulled today.
> >>
> >> I have a Lenovo X61s with a physical rfkill switch (the kind that
> >> slides between blocked and unblocked).
> >>
> >> If I turn off the hardware rfkill switch (set to block) while
> >> associated, I lose the connection (obviously).  If I turn it back on
> >> again (set to unblock), I get a lot of latency (my mouse stops moving
> >> for a second or two) and I sometimes get errors in the syslog like
> >> this:
> >>
> >> iwl4965: Error sending REPLY_CT_KILL_CONFIG_CMD: time out after 500ms.
> >>
> >> The card fails to associate afterwards.  If I tell network-manager to
> >> turn off wireless, then block rfkill, then wait a few seconds, then
> >> unblock it, then turn nm back on, everything works again.
> >>
> >> This seems like at least two different bugs:
> >>
> >>  - iwl4965 causes latency.  This latency issue has been reproducible
> >> every time for me on 2.6.24-ubuntu_something, 2.5.25, wireless-compat
> >> (recent), and current wireless-testing.
> >>  - iwl4965 doesn't work right after hard unblocking rfkill.
> >>
> >> I'm happy to do further troubleshooting and/or test patches.
> >
> > It depends on what HAL reports for the killswitch's status.  NM asks HAL
> > what the radio's rfkill state is every 6 seconds (since we can't get
> > event notifications when the state changes until 2.6.27).  If HAL
> > reports the radio is killed, NM won't touch the device.  It sounds like
> > something is trying to poke the device before it's ready, which may mean
> > that either NM isn't aware you have a killswitch, or HAL isn't correctly
> > talking to the iwlwifi killswitch bits (either old-style or new-style
> > with Henrique's patches).
> 
> OK.  But IMO iwl4965 should still avoid freaking out when this happens.
> 
> >
> > Can you post some logs from /var/log/daemon.log around when flipping the
> > killswitch to BLOCKED, waiting about 20 seconds, then flipping it back
> > to UNBLOCKED and waiting 10 seconds?  That will tell whether NM is
> > correctly talking to HAL or not.
> 
> Attached.  While the connection was failing to happen after
> re-enabling the radio, dmesg showed a lot of crap like:

Ok, so with the patches HAL no longer knows about the killswitch and NM
is unable to poll the switch for the radio state.  So NM doesn't stop
trying to poke the radio and connect to stuff because it doesn't know
the radio is off.

Yes, iwlwifi shouldn't freak out.  We also need to get HAL fixed up to
recognize the new rfkill stuff as a switch.

Dan