I've copied a first cut of an Intel Pro/1000 Gigabit driver for freebsd to:
http://www.freebsd.org/~mjacob/FreeBSD_Intel_Gige.patch.gz
This is a patch against -current's /usr/src/sys as of today. I have run
it on a 2xPPro Intel 440FX and a Alpha PC164 (432Mhz).
This is a very first cut, but seems fairly clean to me. It lacks many many
things, but certainly is functional. I'd like to get some feedback about
this soon before I drop it into the tree.
Known Problems:
Link management is 'primitive'. I don't have a switch, so I've
only tested full duplex point2point. Right now, up'ing the
interface will fail if link isn't seen.
Known MetaProblems:
Performance is abysmal (~130Mbit). That's next to figure out
what's up and what I've done stupidly. Feel free to comment..
Some (most) of the stuff is based upon the released Linux driver
because this driver was done w/o recourse to officially released
Intel chip specs. In particular, accomodating early revision
boards (the only two I have) is based on information out of the
Linux driver which has no explicable collateral (so should be
looked at more carefully). Also there appears to be some issue
with some PCI chipsets that means that using the feature of
reporting a packet sent when it gets into the chip's xmit fifo
can't be used (have to wait for it to get onto the wire).
Futures:
Support for JumboGrams (requires thwacking the driver to handle
more than one receive descriptor per packet).
Statistics gathering.
Vlan tagging (if I can get good documentation on this)
Better multicast support (I'm only using the extra 15 addresses in
the address match array before going to Multicast Promiscuous, and
not using the MTA hash array).
All feedback gratefully solicited- I'm not, primarily, a NetWork driver
writer, so any pointers/comments would be appreciated.
-matt
To Unsubscribe: send mail to [EMAIL PROTECTED]
with "unsubscribe freebsd-hackers" in the body of the message
