A few people offered to test the change to the ncr_si driver.
The change, by the way, is to hit the DMA engine with "extra"
resets at a few key places (clues taken from the Sprite code).
The driver does DMA on my 3/50 just fine, but that machine has
never shown me the corruption problem, so I don't know if this
change really fixes the problem.  Anyway, get the kernel here:

	ftp.netbsd.org
	cd /pub/NetBSD/arch/sun3/install-11-p2
	binary
	get netbsd-gen.gz

Note that this might not be the "final" patch-2 binary, because
there were some other changes for patch-2 that I still need to
pull into my final build area.  This kernel is nearly identical
to the generic kernel in the 1.1 release, except for the "extra
reset" stuff in the SCSI driver for the OBIO "si" DMA hardware.

In case anyone missed it, here is a good way to test the driver
on a single disk machine: (without risking disk corruption)

> Date: Mon, 12 Feb 1996 10:13:31 +1030
> From: Ian.Dall@dsto.defence.gov.au

> [...]  I have a methodology which seems to work without corrupting
> the disk. I have a disk with a small partition (I split off some swap
> space) which is "disposable". I boot single user, mount everything
> read only. Then I can test the IO to the small scratch partition.

Note, if you boot single-user, you can generally get away with
tromping around in the swap area, so you don't need a separate
"spare" disk partition.  Just do things like:
	b sd() -s
	dd if=/dev/rsd0b of=/dev/null
	dd of=/dev/rsd0b if=/dev/zero
and experiment with block sizes, etc.

Thanks, everyone for helping save those poor old 3/50 machines!
Gordon