An intriguing problem

[utzoo!duke!cincy!chris]

I have been tinkering with some drivers on our (V6) system
and have unearthed an interesting phenomenon. Perhaps one
of you can explain it to me.
We have a Ramtek RM9200 Display Processor hooked up to our
UNIBUS through an interface that looks almost like a DR11B.
Until recently, the driver stole a system buffer and copied
the user's data into it, and then did DMA from the system buffer.
It worked fine -- except that it seemed slow. I thought it
would be better to use physio to do DMA direct from the user's
buffer. So I rewrote the driver and installed it. It seemed
to work fine. Then the system got busy, and I started
getting data late errors from our RK drive. We swap there,
and the system has since crashed numerous times from
I/O error in swap. I don't understand why this has cropped
up all of a sudden. The UNIBUS is laid out as follows:

	11/60
	KL-11W (console)
	RK11	spl5
	RP11	spl5
	Ramtek	spl5
	DZ	spl4
	DZ	spl4
	DR11K	spl4
	TM11	spl5

I get no complaints from the RP11 (it's really a Plessey controller
--by the way, don't ever buy one-- hooked to a 300 MB CDC
drive that emulates 6 rp03s), but the RK goes nuts.
Any suggestions? I would like to leave the Ramtek at spl5,
because we do highly interactive work with it, and the
general feeling is that it is as important if not
more important than the disks.