Re: 5.2-RC oerrs and collisions on dc0

On Tue, 30 Dec 2003, Dejan Lesjak wrote:

> I'm glad to hear that reporting of errors might be false. So I setup some
> monitoring of ports on both ends of cable, put apache on server with dc card
> and made a cron job on another machine (on same switch) to fetch a file of
> around 100MB every 5 minutes from server.
> So things go thusly - on server side there is constant flow of output errors
> and collisions with a peak at reboot. Collisions amount about 10% of output
> errors, which about 0.1% of output traffic. On switch side there is only a
> peak of input errors at server reboot, then there are no more errors.
> I'm not entirely sure if I can make conclusions based on this, but this can
> probably be false alarm then. Tell me if there is something more that I can
> do to confirm that these are in fact false collisions and errors. Also if
> seing graphs would help, they are on http://niobe.ijs.si/ewok/
>
> Thanks everybody for help,
> 	Dejan

Hm, you may have a real problem.  I see 0 output errors here, although my
collision count is constantly increasing.

If you'd like to "fix" the false collisions, find the following line in
if_dc.c, and remove it:

                ifp->if_collisions += (txstat & DC_TXSTAT_COLLCNT) >> 3;

I say "fix" because that count used to be accurate, we apparently broke
something somewhere else in the driver which is causing the false
collision reports.  Of course, whatever is causing the false collisions
may be causing false errors on your card, so it's hard to tell. :)

Good luck,

Mike "Silby" Silbersack