Path: utzoo!utgpu!water!watmath!clyde!att!ihlpa!ihnp4!cbmvax!jesup
From: jesup@cbmvax.UUCP (Randell Jesup)
Newsgroups: comp.sys.amiga
Subject: Re: 2090 dies with FFS
Message-ID: <4113@cbmvax.UUCP>
Date: 26 Jun 88 21:42:09 GMT
References: <7528@watdragon.waterloo.edu>
Reply-To: jesup@cbmvax.UUCP (Randell Jesup)
Organization: Commodore Technology, West Chester, PA
Lines: 47

In article <7528@watdragon.waterloo.edu> mwjones@lion.waterloo.edu (Morgan Jones) writes:
>The hardware is a B2000 with 2090 and 28ms SCSI hard drive.  Naturally, the
>system starts generating errors when the screen is in overscan mode. But,
>under the FFS, the problem doesn't stop here - the drive also generates 
>errors in hi-res or interlaced with more than 4 colors.

	This sounds like the problem that was corrected very recently
in the driver.  The November version of the driver fixed this for st506,
the latest version (available on bix and I think other networks as well)
fixes it for SCSI as well.  The new version is in an ARC file on bix named
I believe HD34.4.  It may not totally solve the problem, but should reduce
it considerably.

	The real cause for this is the FIFO overflowing on the controller.
When the system is in man-color hi-res modes, the processor may not get around
to granting the DMA request until the end of the scan line, or even frame
in severe overscan.  If there's too big a delay, the FIFO overflows (I think
it's 64 bytes on the 2090.)  Any controller that uses DMA might have this
problem.  The real trick is to identify that it happened, and restart the
transfer without putting up a read/write error.  FFS may add to the likely-
hood of this due to the fact that it tends to start multi-block reads instead
of reading 1 block at a time.

	This may be compounded on some of the old dealer-demo rev 3.9
motherboards, though I'm not sure (I'm a software person).  I believe the
Buster chip (main difference between 3.9 and 4.x) is responsible for DMA
control.

	As a temporary bandage (that may or may not work): try adding a
"	Maxtransfer=512;" to the mountlist for the FFS partitions.  This
hopefully will stop it from using multi-block reads, and reduce the likely-
hood of a r/w error until you get a copy of the new driver.  It will cost
you in speed, however.

>The dealer called several controller manufacturers to see if they knew of
>this problem.  CA said that they'd never heard of the problem but would
>look into it.  SCA and ASDG both (after much digging) admitted that they
>knew of the problem (I got the impression that they admitted that their
>controllers suffered the same problem).  The dealer is now ordering a

	SCA???? As in viruses???  Also, ASDG dropped their HD interface
design some time back, it was never finished.

DISCLAIMER: I'm not a hardware guy, and I didn't write/work on any of the
software involved in this problem, and this is not official by any means.
-- 
Randell Jesup, Commodore Engineering {uunet|rutgers|allegra}!cbmvax!jesup