The Sun3 port has a problem in the le(4) driver for the original 
Am7990 LANCE; revision C of that chip has a serious bug. We've got a 
work-around (options LANCE_REVC_BUG) but the work-around disables 
multicast as a consequence (it compares the destination address in 
the packet against the kernel-understood MAC address to detect 
mangled packets). We can:

1. add multicast support to the work-around (more instructions per 
packet on a slow platform; ouch). If we do this, we should instrument 
the work-around to see how often it is invoked - it may be possible 
to adaptively turn it off if measurement indicates a high enough 
probability that you don't have the bug.

2. see if we can get AMD to cough up the information that would allow 
a user to visually identify the Revision C chips by their fabrication 
date code. If they don't have the bad chip, the work-around option 
(default on for the sun3 GENERIC) can be turned off, with the happy 
result of better network performance.

I tried to find the information required by #2 to add to the le(4) 
man page a year ago, but didn't get very far. On the other hand, I 
didn't exhaustively pursue it, and a more persistent effort may 
succeed where I failed.

	Erik <fair@clock.org>