Some of the i82559 chips have a hardware bug. The symptom of the bug
is, approximatley, that when an fxp sends lots of small back-to-back
packets, the chip sometimes fails to correctly update the Tx buffer
descriptors.  (If there are several frames outstanding, the chip
sometimes updates a block of descriptors in the TxCB ring but
does not mark one descriptor in the middle of the block, even
though it has clearly sent that packet.)

I've probably garbled that; a full description is at:
     http://www.tux.org/hypermail/linux-eepro100/1999-Sep/0020.html

I'd really appreciate it if someone could confirm that either the fxp
driver is coded so as to never hit that problem; or if it is, that the
fxp reset routine recovers properly. (that  L-thing does not).


BTW: does our fxp driver handle dual-NIC fxp boards?