7-STABLE Watchdog Timeout

[Pieter de Goeje]

On Tuesday 03 June 2008, Jeremy Karlson wrote:
> I'm still looking into my watchdog timeout with me re card.  I'm
> starting to wonder if my problem is in any way related to the
> discussion back in September 2006 starting with this post:
>
> "6.2 SHOWSTOPPER - em completely unusable on 6.2"
> http://lists.freebsd.org/pipermail/freebsd-stable/2006-September/028792.htm
>l
>
> It seems that under certain conditions and loads, a network interface
> with a shared interrupt would stop responding until the watchdog
> resets it.  This seems to be very similar to what I see.  At the time,
> they seemed mostly concerned about fixing the em driver; I'm using re.
>
> Unfortunately though, I can't seem to find what the resolution to this
> was, and if it could be related to the failure I'm seeing.  Does
> anyone know what happened with that problem?
>
> -- Jeremy

If I remember correctly, there were two problems: 1) there was a race in the 
interrupt handler (this affected all interrupt handlers), 2) em used the now 
obsolete if_timer to implement it's watchdog timeout. This timer is 
unreliable. Both items have been fixed, so this shouldn't be related to your 
problem. #2 wasn't fixed in all drivers, but if you are affected you should 
see a warning on startup stating the use of the obsolete timer. I believe 
if_re was converted some time ago.

Note that in theory watchdog timeouts could also mean broken hardware, bad 
connection to PCI bus or other intermittent hw failure. I would try reseating 
the card in another slot.

-- 
Pieter de Goeje