One more data point then: I moved the DE500-BA (21143-PC) and the 
QLC1020 cards around a few times to different PCI slots (32-bit, 
64-bit, etc) to see if that made any difference - no soap.

I've also tried using the older de* driver, and it gives the same 
results: "master abort". (hmm, smells like hardware).

Then I changed out the DE500-BA for a no-name 4-port 10/100 card with 
an Intel (DEC) 21152 PCI bridge to four Intel (DEC) 21143 ethernet 
chips, with PHYs that we don't recognize explicitly (ukphy* matches), 
and Lo! Everything works.

This new card is not recognized by SRM v5.5 as an ew?* so I can't 
netboot off of it (interestingly, the DE500-BA is only 
half-recognized; if you type "show device" it shows an ewa0 with the 
ethernet address, but if you type "show config" it doesn't report the 
card as being there!).

While I've got your attention, there is one other thing that it seems 
both the tlp* and de* drivers don't quite get right with the DE500: 
Nway negotiation with a 10/100 FDX switch (in this case a NetGear 
FS524). SRM also produces some pretty funky results.

	Erik <fair@clock.org>