Hi,

at the moment I'm trying to get Leo Weppelman's new ncr driver working.
Actually the driver works pretty well and is *much* faster then the
old one. Esp. it uses much less CPU time.
One problem still remains. From time to time my machine starts dumping core
on random programs (or e.g. on all shared linked programs). This happens
*only* with the new driver. At first I suspected my pdma code, but I
never had filesystem problems and the bug almost exclusivly shows up
during heavy paging. So I did a little experiment. I wrote a small program
that mmaps a file and then writes the mapped space to another file.
Now I waited for the bug to show up. Next time when all my programs linked
shared dumped core I did the following:

cp /usr/lib/libc.so.12.3 /var/tmp
cp /usr/libexec/ld.so /var/tmp
cmp /usr/lib/libc.so.12.3 /var/tmp/libc.so.12.3 # File did not differ
cmp /usr/libexec/ld.so /var/tmp/ld.so		# File did not differ
msave </usr/lib/libc.so.12.3 >/var/tmp/libc.so.12.3
msave </usr/libexec/ld.so >/var/tmp/ld.so
cmp /usr/lib/libc.so.12.3 /var/tmp/libc.so.12.3	# File did not differ
cmp /usr/libexec/ld.so /var/tmp/ld.so		# *GOT IT*

The ld.so copied with msave did differ from the original ld.so within
one page. From looking at the damaged page it looks like the page was
a gcc data page. Usually I can fix the machine by doing a
"dd if=/dev/zero of=/dev/null bs=4m count=1"

Now I suspect that the problem is not in the ncr driver but somewhere
in the pc532's vm code. Note that this is the first time that we are
running with an interrupt driven disk driver. Any hint where to start
searching for the problem?

	Thank's - Matthias