On Sep 26,  2:32pm, Martin Husemann wrote:
} On Thu, May 06, 2004 at 10:27:10AM -0700, Dave wrote:
} > Since both cards on the network were the same make and model and both had
} > identical /etc/ifconfig.tlp0 contents, that can't be right.  Using
} > "auto" in place of "100baseTX" cleared up all the problems.
} 
} AFAICT NetBSD only does proper NWAY negotiation if media is set to auto.
} So the switch needs to be hard-wired (if it can even do that) if you
} hardwire the mode on the NetBSD side.
} 
} This is a bug, and it should be fixed.

     This is debatable.  The standards only define "auto" and don't say
what happens when you don't use it, i.e. using something other then
"auto" results in undefined behaviour.  Some devices will still do
NWAY, but will only present the configured options, and some don't do
NWAY at all.  On the flip side, some devices will only do full duplex
if they receive an NWAY signal saying that full duplex can be done
regardless of their configuration; whereas, others will do whatever
they are configured to do regardless of what NWAY says.  Behaviours can
differ between different models of devices from the same manufacturer.
With so many different variations in behaviour the only safe thing to
do with modern networking equipment is to set everything to "auto" and
let the hardware work it out.  If you want to hard code things then it
is only safe to specify half duplex unless you know the exact behaviour
of all devices involved, and finding out how any given switch will
behave will be very difficult.

     It has been a very long time since I've written a NIC driver.  How
much of NWAY is under driver control?  Does this vary with different
PHYs?

}-- End of excerpt from Martin Husemann