I noted the same problem and have been working on pinning down the exact
date of when things broke.
I have been using cvs to obtain sources on various dates.
Last night I built a kernel with 6/25 which did not have this problem.
I thought I had built a kernel with 6/26 which did have problems.
I'm going to try to verify this today.
It does take a while to build a kernel on these machines when running
things via nfs.
My impression is that the problem might be related to reading/writing to
kernel space rather than user space.  This probably could be
verified/disproved by writing a userland program to read/write/verify
various things to the disk with varying buffer sizes - might be an issue
of size of dma buffer.

I'll post when I find out something more useful.
johnh...
On Thu, 28 Sep 2000, Andrew Gillham wrote:

> Date: Thu, 28 Sep 2000 00:00:47 -0400 (EDT)
> From: Andrew Gillham <gillham@vaultron.com>
> To: Matthias Buelow <token@cip.informatik.uni-wuerzburg.de>
> Cc: port-vax@netbsd.org
> Subject: Re: 1.5_ALPHA2 on VS3100/m30 and external SCSI
> 
> Matthias Buelow writes:
> > Hi,
> > 
> > does anybody know of problems using an external SCSI disk with a
> > VAXstation 3100/m30 with 1.5_ALPHA2?
> > I have attached a 4GB disk on the external port and the kernel
> > recognizes it as sd1:
> > 
> > sd1 at scsibus1 target 1 lun 0: <QUANTUM, XP34301, 1071> SCSI2 0/direct fixed
> > sd1: 4106 MB, 4076 cyl, 20 head, 103 sec, 512 bytes/sect x 8410200 sectors
> > 
> > However, the disk doesn't seem to be able to hold a disklabel, once
> > written to it it always seems to "disappear" after a few moments.
> 
> This is the exact problem I have.  I have been whining about it for several
> weeks, but I haven't collected enough information to file a PR on it.
> :-)
> 
> I have a kernel from June 29 that does not have the problem.  Any kernel
> after that (actually after the vsbus.c rototillage) is broken.
> 
> Unfortunately the build was broken from about June 30 to July 30 so I am
> having trouble isolating the changes via CVS binary search. :-)
> 
> Anyway, I am very glad to hear someone else has this problem.  It shows
> I'm not going crazy. (at least on this issue)
> 
> I guess a PR should be filed anyway, but I was trying to isolate the dates
> involved better.
> 
> -Andrew
>