[tulip-bug] tulip not correctly identifying LNE100TX

Dave Platt dplatt@radagast.org
Tue Jul 9 15:35:01 2002


> The ADMtek Centaur is an all-in-one chip.  There is no way for a board
> designer to screw it up, so there no reason to identify the specific
> board.

#laughing out loud#

I recall one of the very first boards I bought which had an AMD PCNet-PCI
chip on it.  The manufacturer (Boca) had ignored Appendix B in the
data sheet, and had installed less than half of the ceramic-chip and
tantalum bypass caps which AMD said were essential for reliable
operation.

The board worked fine, up until the moment when a particular data pattern
caused the chip's analog inputs to suffer from enough ground-bounce and
Vcc ringing to drop bits.  The chip would detect an FCS mismatch and drop
the packet, the sending system would time out and resend it, the
re-sent packet would hit the same error, etc.

It was quite maddening... FTP connections would freeze up, while a TELNET
session to the same host worked fine.

Granted, this isn't a screwup which would require (or be helped by) any sort
of software configuration fix, but it did teach me a valuable lesson:
there's no design so foolproof that an idiot board designer can't make a mess
of it.  [Naturally, an analogous situation exists in software design!]