<   You wrote:
<> And in fact the VS2K manual does make it clear that there are
<> interactions between the NCR and the disk controller (e.g. they share
<> buffers).

This is never been adaquately explained.  I believe it to be a 
misstatement or contextual error with regard to the VS2K manual.
The 9224 UDC has it's own internally dedicated DMA logic that cannot 
be used for any other device (LANCE is similar). That is indeed 
curious as the 5380 does not have DMA (internally) at all so that it's 
dependent on either PIO or some external DMA controller as it is not 
possible to share a dma channel of either the LANCE or the UDC(9224). 
If that is true how then would a memory buffer conflicting between the 
disks and the SCSI device??  This is not a ka4xx thing as the raw chips 
aka NCR5380, LANCE and UDC(9224) impose this structure on the system.

Now if you mean that this "buffer" is a pool of memory that is not tied 
to hardware as the DMA range of the LANCE and UDC is 16mb(24bits) and the 
poor SCSI chips is DMAless so the PIO driver has whatever access it may 
need. No conflict there unless drivers used the same pool by error or
need.

The 3100 series members are distinct enough from each other that each 
must be looked at for commonality and also the specific differeces must 
be identified.  This should bring some reality to SCSI/FDC/MFM drivers for 
the small box vaxen.

The above features should make a SCSI driver for the vs2k much simpler.

The rest of the noise is pure speculation.

Allison