Re: nfe0 watchdog timeout

On Tue, May 13, 2008 at 07:38:09AM +0200, Pascal Hofstee wrote:
> On Tue, 13 May 2008 11:09:27 +0900
> Pyun YongHyeon <pyunyh_at_gmail.com> wrote:
> 
> > Can you see link state change reports of nfe(4) in your dmesg file?
> > If you find any entries related with nfe(4) please show them to me.
> > It may have a couple of link UP/DOWN messages and watchdog timeouts.
...
> > When you know nfe(4) is not working anymore, would you check Rx MAC
> > is still able to see incoming packets by invoking tcpdump on
> > nfe(4) interface?
> 
> Once it happens again i'll make sure to check on it.
>  
> > It would be even better I can see verbosed boot messages related
> > with nfe(4).
> 
> Will provide once i can reboot the system again (it's currently in the
> middle of a somewhat lengthy build process).

related to nfe(4) i recently noticed on one of my systems a tendency
of the receiver to stall under high link and CPU load.
A patch to address the problem is at
http://info.iet.unipi.it/~luigi/FreeBSD/#nfe
though not ideal (because it calls the nfe_init routine on a
suspect stall) it does the job here, so maybe you can try it
on your system too.

At first i too saw the problem on an AMD64 system, but it turns
out that i386 uses the exact same code and the problem was
repeatable on i386 platforms too.

A way to trigger the stall, in my case, was to start an scp of
a long file from a local machine (100Mbit in my case) to the nfe-equipped
system. At the same time, start playing with X11, run a compile or do
some other CPU intensive work, which should cause CPU shortage for
the receive task to a sufficient level for the stall to occur

	cheers
	luigi