Why not SCSI on Unix PC?

[Michael Thompson]

Greetings fellow Unix PC'ers,

I have recently been flipping through some back issues of BYTE magazine
and I have come across two articles from Ciarcia's Circuit Cellar
(May & June of 1986) which describe how to build a SCSI interface to
his SB180 computer.  What intrigued me is that he uses the NCR 5380 which
is a single chip SCSI controller.  It seems to me that this chip would
be very easy to connect to the Unix PCs bus with minimal support chips.
I believe that only some address decoding and interrupt interface chips
would be needed.  I admit that I have not looked at the schematics for
the Unix PC, but I can't imagen AT&T (through Convergent) have done 
anything too bizzare to the bus to prevent hooking up the NCR 5380 in
a relatively straight forward manner.  

What could be the greater unknown is the coding of a loadable device 
driver to control the SCSI controller chip.  I have written a character
based loadable driver for a Sun 386i before, but it operated on a relatively
simple level and did not implement a protocal as complicated as the
SCSI protocal.  I also understand that block oriented device drivers
are more complicated to write too.  It also may be desireable to make
a more complicated type interface circuit where buffer memory is used
and the Unix PC be able to perform DMA transfers into that buffer
memory and the NCR 5830 transfer to and from that memory.  This will 
keep the SCSI interface chip from constantly yanking on the Unix PC 
bus interrupt to request more data.

I would encourage anyone interested in perhaps joining forces to build
as SCSI interface, read the Circuit Cellar articles from the May
and June issue of BYTE magazine and see if they also think that it
would be possible to adapt the SCSI controller for use on the Unix PC.
I would also appreciate it if anyone could point me to a good reference
source on the SCSI protocal since these articles were a little skimpy
in these areas.  These articles are also available in Volume VI of
Ciarcia's Circuit Cellar book. 

I think that building a SCSI board based on this chip would be a 4 on
a scale of 1 to 10 in difficulty and writing a SCSI block device 
driver would be a 8 or 9 on a scale of 1 to 10.  What do you think?  
I'm I crazy and dreaming of pie in the sky or is there a possibility 
that I (or we) could pull it off.  I am pretty handy at building prototype 
digital hardware and even better at coding, I just don't have access to all 
the information to really tell if a SCSI adapter board would be beyond my
modest abilities.

I can envision a few of us hardware types communicating together to come
up with a SCSI interface card which could be prototyped and then a
few of us software types communicating and coming up with a workable device
driver.  Once we got a working prototype (hopefully more than one) we 
could release all the information to the Unix PC community so anyone would
be able to build such a device.  Am I really thinking crazy here or not. 

The ultimate goal would be to have a SCSI board which could be attached to
commonly available SCSI hard drives and tape drives. 

I welcome any comments.  Just as long as they don't recommend that I be
hauled away in a straight jacket.

Mike Thompson
=========================================================================
Michael P. Thompson                                Bell Northern Research 
Member Scientific Staff                          685A E. Middlefield Road
(415) 940-2575                                          Mountain View, CA
amdahl!bnrmtv!miket                                            94039-7277