[tulip] Netgear FA310TX rev D2 NIC

[Donald Becker]

On Fri, 11 Aug 2000, Eric Thelin wrote:
> On Tue, 8 Aug 2000, Donald Becker wrote:
> > On Tue, 8 Aug 2000, Bruce Bales wrote:
> > > I have a Netgear FA310TX rev D2 ethernet board with LC82C169C chip (not
> > > LC82C168c).  Netgear says
> > > board is based on DEC 21040 (tulip).
> > 
> > No, it's a LiteOn chip that has a data transfer engine with Tulip-format
> > descriptors, but the transceiver section isn't at all like the 21040.
> 
> I have this problem too where the card just seems to hang after heavy
...
> haven't tried the .92 version of the driver yet.  I was trying to just

You should try v0.92, although the usual difficult is with media selection,
not data transfer.  The v92 driver adds better PCI bus error reporting.

> I have also heard that the tulip drivers are no longer as well supported
> as others.

We support the tulip driver in Scyld driver set.  Those changes make it into
the kernel on an ad hoc basis.  I didn't want the situation to end up this
way, but that's the Way Things Are.

> Is there any truth to this?  Is it just these Netgear cards?

The claim is that the LiteOn is very sensitive to PCI bus timing that
doesn't meet the spec.  On some motherboard the Rx FIFO loses sync with the
data transfer engine, and packets are occasionally dropped.

I have not been able to reproduce this reported problem.  There is a very
ugly hack in the BSD driver that is supposed to work around this packet
corrupion, but I'm not going put in such obviously sleazy and risky code
without being certain that the error exists and knowing the work-around is
valid and does not result in corrupted data.


Donald Becker				becker@scyld.com
Scyld Computing Corporation		http://www.scyld.com
410 Severn Ave. Suite 210		Beowulf-II Cluster Distribution
Annapolis MD 21403