[vortex] Strange errors from 3c509B driver

[Andrew Morton]

Bogdan Costescu wrote:
> 
> On Fri, 8 Dec 2000, Richard Gooch wrote:
> 
> > OK, why is this? I don't understand why if one end is fixed and the
> > other is auto, why the auto end can't sense the carrier and work it
> > out from there?
> 
> The carrier is sensed (look at the last line of mii-diag report) and
> gives the speed, but the duplex part is not, because this cannot be
> sensed. When autonegotiation is enabled, data about capabilities is
> exchanged in some special packets; but when one side doesn't want to
> exchange this packets (e.g. forced setting)...
> 
> More detailed info at http://www.scyld.com/expert/NWay.html
> 
> > Is there a way of doing this via ifconfig (my man page doesn't
> > document a 100 Mb FD option for media type)? I've got 3 NICs in this
> > box, and I don't want to end up turning off autonegotiation for the
> > other two, otherwise I'll just end up moving the problem down the
> > line.
> 
> Not that I know of. But http://www.scyld.com/network/vortex.html explains
> how to pass options for multiple cards. You should probably have:
> 
> alias eth1 3c59x
> options full_duplex=0,1,0
> 
> in your /etc/conf.modules, although if you use the driver in a recent
> 2.4-test, you might need to replace 0 with -1, as Andrew changed the logic
> to allow forcing HD mode (sorry, I know what the changes are, but not in
> which driver version they appear; try reading the beginning of the
> driver source, Andrew puts there some comments about changes).

 0 -> force half duplex
 1 -> force full duplex
-1 -> don't force duplex at all.

This was in the 2.2 driver.  I never submitted it because we were
at 2.2.18-pre16 and I thought that 2.2.18 was close.  That was
two months ago.  Sigh.

Similar story with the 2.4 driver.  I haven't even made that change
there.   Richard's using 2.4.  As you say, he should use

	full_duplex=0,1,0

to force eth1 to FD and leave the other NICs alone.