> From: "A.Arnold" <arnold@zam275.zam.kfa-juelich.de>
> Date: Fri, 10 May 1996 16:22:53 +0200

> Hello altogether,
> 
> I am quite new to NetBSD and this list, so hopefully I do not bother anyone
> with questions that have been asked a thousand times before...
> 
> A week or so ago, I successfully installes NetBSD 1.1 on my Sun3.  It is a

Congratulations!

> Sun 3/60 with 8MBytes RAM, 340MBytes HD (about 40M swap) and an old QIC-24
> Wangtek streamer recycled from a broken 3/280.  The systems seems to work
> fine basically, but a few questions are open:
> 
> 1. Why do the cursor keys not generate any characters when the system is in
>    console mode (makes an editor a bit hard to use ;-) ?  They work in an xterm
>    when X11R6 is running, and I also have access to a 3/160 with SunOS 4.0 .
>    Under SunOS, the cursor keys work also in console mode.

That version of the driver did not support the keypad.
The "current" version does arrow keys, capslock, etc.

> 2. The 3/60 has an option to plug in a 68030 instead of the original 68020
>    CPU.  I soldered in the required socket and did the replacement.  The system
>    still works fine, but without any speed improvement.  However, there should
>    be one due to the 68030's additional data cache.  Could it be that NetBSD
>    doesn't enable the cache?

Sorry, can't help.  I've never done that (but it sounds like fun 8^)

> 3. In 2 out of 3 cases, X11R6 will immediately die after startup with xinit,
>    xterm and xclock generating core dumps.  Once this has happened, further
>    tries will result in an immediate coredump for xinit.  Even doing an 'ldd'
>    for any X-based application will result in a signal 10 or 11 afterwards.
>    I have to reboot the machine to get it back to its state before the first
>    crash happened :-(  I do not think that my X binaries/libraries are cor-
>    rupted, as I reinstalled them two or three times...

That is a kernel bug which I've been hunting for quite a while.
It could be the scsi driver, or the pmap code... I'm not sure.
On my machine, it almost never happens if I run with si_options=1
(instead of 3) so DMA transfer completion is polled.

Strange... my sun3/260 works fine with si_options=3, but the sun3/60
almost always get the "wrong page in shared library" bug if I run
with si_options=3 (using si_options=1 works better).  I've been
adding some instrumentation to the scsi driver to do full logging
of all transfers (and the XOR sums) to help me compare the activity
between the failing and working cases.  This is a really hard bug,
which I strongly suspect is some weird hardware mis-behavior...

Gordon