On 04/30/09 21:23, Garrett D'Amore wrote:
Earlier this week I got yet another request for qfe support for x86
systems. I get these requests fairly periodically. I guess there are a
lot of 4-port qfe cards still out there.
This last one finally gave me the impetus to do the work to do the
port. What I want to do is poll the community on this, because I'm
going to need help:
1) testing -- my last SPARC system with onboard hme died a while back
I still have some. Plus an assortment of various qfe and hme PCI cards
(Freshchoice and Freshchoice Lite). No SBus systems or cards though.
2) testing -- I have no sbus hme ports -- a Sun Ultra 2 or some
other system with sbus qfe or sunswift would be helpful here.
3) review -- as part of this effort I've removed all the legacy dvma
and on-demand DMA resource allocation, and replaced it with a very
simple "preallocated" DMA region/bcopy -- this is typical for most
simple 1G and slower ethernet NICs today. (This eliminated about 1000
lines from the driver.)
4) perf. testing -- the above changes "potentially" have performance
impact. I think on any supported CPU that these cards run on, the
simplifications will greatly outweigh the savings formerly afforded by
the "private" dvma interface this driver used to use
Testing on an x86 port of the DLPI qfe driver some years ago showed that
a driver that used preallocated DMA/bcopy was as fast as the classic
complicated driver even on slow systems, so I'm all for the simplification.
Menno
--
Menno Lageman - Sun Microsystems - http://blogs.sun.com/menno
_______________________________________________
driver-discuss mailing list
driver-discuss@opensolaris.org
http://mail.opensolaris.org/mailman/listinfo/driver-discuss
