Re: Doesn't always work after resume

[Dan Williams <dcbw redhat com>]

On Mon, 2006-08-14 at 15:17 -0400, Robert G. Brown wrote:
> On Sun, 13 Aug 2006, Pat Suwalski wrote:
> 
> > Stian Jordet wrote:
> >> I have a problem that NetworkManager often (but not always) don't find
> >> my wireless network after a resume. It always works after a fresh boot,
> >> it sometimes work after resume. It always finds my neighbours
> >> unencrypted network everytime, but I don't know if my WEP-encryption
> >> really has anything to do about it.
> >
> > I have a similar issue with madwifi. To work around it, I right-click on
> > the applet and uncheck "Wireless" then recheck it. It seems to work
> > afterwards.
> >
> > I wonder if that is the case on ipw as well?
> 
> Absolutely.  In fact, I've encountered a variety of "interesting" hangs
> and errors when I suspend/resume a lot, including a message that I've
> "exceeded the allowed number" of some resource and refusing to work at
> all.  Other times it finds the network but ignores the key in the
> keyring manager and prompts me for the key all over again.  When this
> happens maybe half the time it ends up looping on this and never
> getting a connection, sometimes it hooks right up after I enter a
> password I shouldn't have had to enter.

This is likely due to artificial D-Bus limits on the number of allowed
pending calls.  If you ever have more than 30-some wireless networks in
the scan list, you'll run into this.  There's not a good solution at the
moment other than (a) increasing the dbus limit or (b) rate-limiting
pending calls in NM or (c) using a private dbus connection for the
applet.  The last two will be a bunch of code, the first is easy but we
just run into the problem again when you have, say, 64 networks in the
scan list.  This will have to eventually get fixed in some way of
course.

> I have a variety of ad hoc workarounds for the stuff I encounter along
> these lines.  Disabling/reenabling wireless sometimes works.  Restarting
> NetworkManager sometimes works.  Removing the ipw2200 kernel module and
> re-inserting it (sandwiched by NetworkManager stop/start) sometimes
> works.  Sometimes nothing works and I reboot (logging out and logging in
> again usually don't seem to work).

Sometimes the ipw2200 kernel thread starts pegging the CPU for me, but
that's not necessarily due to suspend/resume, just random driver issues.
An rmmod/modprobe usually works here, but if the card becomes
unresponsive after a while and CPU utilization goes to 100%, that might
be the problem.

> I think that what I'm seeing is a variety of bugs at different levels --
> in the ipw2200 driver (as sometimes re-initializing it altogether is
> essential to success or there are messages in the log that lead me to
> believe that it is having problems), in NM's interface, in the gnome
> keyring (where you have to sometimes delete altogether a key that worked
> fine for a week, re-insert it, and go into the manager interface and
> re-enable it to work without a password).  Maybe in wpa_supplicant and
> NM's back end stuff.  This is all pretty complex stuff.

Quite right, and bugs in some pieces sometimes magnify the bugs in other
places.  The only way to work it out is to gradually fix the bugs in
various places.

Dan

> It does seem to gradually be improving though.  I have periods of weeks
> in a row where it just works, exactly the way it is supposed to, across
> suspend/resume and all else.  I just dislike the non-deterministic/state
> dependent nature of the crashes/failures, as they are really hard to
> debug, and causes appear to be spread out over several layers of
> responsibility at the software layer.
> 
>    rgb
> 
> >
> > --Pat
> > _______________________________________________
> > NetworkManager-list mailing list
> > NetworkManager-list gnome org
> > http://mail.gnome.org/mailman/listinfo/networkmanager-list
> >
>