On Tue, Jun 24, 2003 at 04:54:34PM -0400, Rafal Boni wrote:

> 	* L2 cache on the R5k (we disable it, so R5kSC machines still work)

... but see below ...

> 	* on-board MACE NIC

... if anyone has any clue about the bloody mace MII errata, I would _love_
to hear from them ... currently, the only way to get the mec to talk to
the phy is to let ARCS initialize the mec via netbooting, which is _really_
suboptimal.  There used to be a SGI dev guy lurking on this list; if you're
still around, would you care to shine some light on this matter?

> 	* Most PCI NICs and probably some chunk of other PCI cards
> 	  for which there are MI drivers in NetBSD

I've personally tested:

	* fxp (Intel EtherExpress Pro).  No issues.
	* ex (3com 3c509b).  Somewhat frequent PCI overruns, but I think the
	  card may be bad.
	* rtk (no-name Realtek 8139 card).  Lousy performance, but that's
	  par for this particular chip.
	* url (USB next-generation Realtek NIC).  USB 1.1, but throughput
	  is pretty decent.

As Rafal said, USB is pretty stable on the O2 right now.  If we can't sort out
the audio/NIC/blah issues, the best thing to do is to drop an ehci board into
the machine and use USB stuff (it was pretty neat to play mp3s on my o2 via
USB speakers).

> The L2 cache stuff on R5k requires me to clean up some changes I have
> had sitting around for the MIPS pmap and check them in (conceptually,
> all the hard work is done, but I've had other stuff going on), but I
> have not had a lot of free time recently...

Rafal's cache hacks work very well for me (r5ksc).  In fact, they actually
address the cache deficiencies in the ip22 port as well -- with them, one no
longer needs to define L2_CACHE_ABSENT in the kernel configuration.  I'd very
much like to see them committed once Rafal is happy with them.

I can make available a diff against -current that incorporates the cache code
and the various ip32 fixes upon request.

-- Chris
	GPG key FEB9DE7F (91AF 4534 4529 4BCC 31A5  938E 023E EEFB FEB9 DE7F)