*BSD News Article 40609


Return to BSD News archive

Path: sserve!newshost.anu.edu.au!munnari.oz.au!goanna.cs.rmit.edu.au!aggedor.rmit.EDU.AU!harbinger.cc.monash.edu.au!msunews!uwm.edu!math.ohio-state.edu!jussieu.fr!univ-lyon1.fr!swidir.switch.ch!scsing.switch.ch!news.dfn.de!rrz.uni-koeln.de!RRZ.Uni-Koeln.DE!RRZ.Uni-Koeln.DE!news
From: se@fileserv1.MI.Uni-Koeln.DE (Stefan Esser)
Newsgroups: comp.os.386bsd.questions
Subject: Re: NCR 53C810 SCSI problem....
Date: 6 Jan 1995 22:41:23 GMT
Organization: Institute for Mathematics, University of Cologne, Germany
Lines: 80
Distribution: world
Message-ID: <3ekgujINN23nn@rs1.rrz.Uni-Koeln.DE>
References: <3edfbd$d25@ibridge.iohk.com>
NNTP-Posting-Host: fileserv1.mi.uni-koeln.de

In article <3edfbd$d25@ibridge.iohk.com>, jvong@igate.iohk.com (Jeffrey Vong) writes:
|> 
|> I got a ASUS PVI-486AP with 486DX2-66 and 16M ram.
|> 
|> When I boot up the machine with the boot disk, I got a message like that
|> 
|> CACHE test : failed ...... host write -1
|> CACHE test : failed .....  ncr write -1
|> .... etc

Yes. True. The ASUS PVI-486AP is the only PCI 
board we know about, that definitely doesn't 
like the NCR 53c810 driver (or is it the other
way around ?).

Is it possible to send me the full message ?
(I.e. all values written and read back, and who 
wrote and who read them. E.g. "CPU 1 NCR 0, ...")


|> The the system can't detact any scsi drive nor any partations whatsoever.

No, after failing the cache test, the system 
doesn't initialize the NCR chip at all.


|> By the way the SCSI card and all the SCSI devices are test okay with DOS 
|> and as a NOVELL server. The jummper setting and the terminators on the
|> cables are tested okay as well.

Yes. The DOS and NOVELL drivers don't try to
make best use of the NCR. The NCR is in fact
a small CPU with SCSI support features, but
it can be used in a dumb mode where only one
command is executed by  that CPU at a time, 
and then the status read back by the 486.

The BSD driver has the NCR work as an independent
CPU, that checks a shared memory region for new
commands and writes back the status after the 
command has completed.

Now there is a problem, if the shared memory 
isn't seen coherently by the 486 and the NCR.
This is usually the result of a buggy cache
implementation, which is not uncommon on PC 
motherboards that try to reduce cost at any 
price ...

The cache test consists of memory writes and
reads by the 486 and NCR. If either one doesn't 
immediately see the values written by the other
one, or if the cache writes back outdated values
(because it lacks a dirty tag RAM), this will
show up as test failures.

The ASUS PVI uses the Aries chip set, and that
chip set is capable of driving a write back second 
level cache.

If you can set the cache to write through, I'd
like to know the difference. (The CACHE test 
messages, and the behaviour of the system with
the cache test disabled, if neccessary).

We can try to debug this, but it will require
kernel compiles, and it would be easier if you
had a system to build test kernels on.

(We don't have any ASUS PVI system here, and I
don't want to buy one for obvious reasons.)


Regards, STefan
-- 
 Stefan Esser				Internet:	<se@ZPR.Uni-Koeln.DE>
 Zentrum fuer Paralleles Rechnen	Tel:		+49 221 4706010
 Universitaet zu Koeln			FAX:		+49 221 4705160
 Weyertal 80
 50931 Koeln