kernel: em2: watchdog timeout -- resetting

[Barney Cordoba]

--- On Tue, 8/25/09, Jack Vogel <jfvogel at gmail.com> wrote:

> From: Jack Vogel <jfvogel at gmail.com>
> Subject: Re: kernel: em2: watchdog timeout -- resetting
> To: "Scott Ullrich" <sullrich at gmail.com>
> Cc: "FreeBSD Current" <freebsd-current at freebsd.org>
> Date: Tuesday, August 25, 2009, 5:17 PM
> Hmm, I do not know what this hardware
> looks like, but its a wired interface,
> not wireless, so if it has link (and I'm guessing it does)
> it will have
> informed
> the stack its running, and thus will get traffic.
> 
> Sounds like this is something the Wimax box should handle,
> at least that's
> my first take on the matter. This is the first I've
> encountered a setup like
> this however.
> 
> Jack
> 
> 
> On Tue, Aug 25, 2009 at 2:12 PM, Scott Ullrich <sullrich at gmail.com>
> wrote:
> 
> > On Tue, Aug 25, 2009 at 5:08 PM, Jack Vogel<jfvogel at gmail.com>
> wrote:
> > > Well, there's no magic switch, you could change
> the driver code to remove
> > > it,
> > > however its really indicative of some kind of
> problem, it shouldn't
> > happen,
> > > so
> > > perhaps you want to pursue that. Does it only
> happen on em2, what is
> > special
> > > about that path, etc etc..
> > >
> > > You can look at the debug/stats data of the
> interface for clues, maybe
> > its a
> > > mbuf exhaustion, perhaps descriptor.
> >
> > After looking at the interface closer it appears the
> device it is
> > plugged into does not have power.  In this case
> it is a Wimax wireless
> > radio.
> >
> > Should a device with NOCARRIER cause this error
> non-stop in the logs?
> >  Can I get you more information that might help
> understand why the
> > watchdog is being tripped?
> >
> > Thanks for your help!
> >
> > Scott
> >
Its amazing that it doesn't affect traffic, since the controller is
getting reset each time.

Althought the watchdog isn't coded very cleanly in the driver, the only
time you should get such behavior would be if interrupts weren't working
correctly. Its a pretty unusual part; a special low power device that
is probably rarely used, so its not impossible that there's some required
code missing. I wonder how much testing Jack has done with an 82541GI?

Barney