Re: Fix for kern/20700

[Manuel Bouyer <bouyer%antioche.eu.org@localhost>]

On Mon, Dec 05, 2011 at 06:08:00PM +0000, David Laight wrote:
> On Sun, Dec 04, 2011 at 10:32:02PM -0800, Brian Buhrow wrote:
> >     Hello.  I think a patch which restricts autonegotiation to 1000-base-T
> > when  1000-base-T is requested will cover all the cases I'm aware of:
> > 
> > 1.  10-base-T and 10-base-T full-duplex (works with media auto and
> > hard-coded)
> > 
> > 2.  100-base-T and 100-base-T full-duplex (works with media auto and
> > hard-coded)
> 
> For 10-base-T HDX and 100-base-T HDX a hub (or end system) might only
> support 'autodetect' instead of 'autonegotiate' - which might be why
> you recollect issues with some old cisco kit.
> 
> 'Autodetect' hubs will select between 10/100M HDX based on whether
> the link test pulses are singles, or in bursts.  Any autonegotiate
> attempts will be treated by such hubs as 100M.
> 
> This all gave some 'fun' when such hubs were common, because whether
> a link worked could depend on what is was plugged into before!
> 
> FWIW the most important think to avoid is HDX/FDX mismatch.

This is what I had with the cisco, more than 10/100 detection
(it was with earlier 2900 series, and some 10/100 modules in 6000 chassis,
it was definively a switch and not a hub).

-- 
Manuel Bouyer <bouyer%antioche.eu.org@localhost>
     NetBSD: 26 ans d'experience feront toujours la difference
--