On Sun, Aug 03, 2003 at 08:37:39PM +0900, Curt Sampson wrote:
> Well, it seems to be both disk and controller related. We're using
> at least fifteen disks of this series on machines other than the
> supermicro, and none of them have ever had any problems. Yet any
> Supermicro machine that we put these disks into will eventually have the
> problem if it does any reasonable amount of disk I/O. (I don't know if
> it happens under OSes other than NetBSD, however.)

This would be good to know.

> 
> We put a couple of Seagate Barracudas in a Supermicro box and beat on them
> real hard, and had no problems.
> 
> > If you can't get the disks remplaced (eventually with some from another
> > manufacturer), I'd try to force a lower UDMA mode with flags in the kernel
> > config file (see wd(4))
> 
> Yeah, we'll probably try that, and maybe see if another OS has the same
> problem. We're just about to free up a machine with a pair of these
> drives, so we can spend some time doing some testing. If you wanted
> access to it to try to help debug this, we could arrange that. (Or if
> anybody else wants to give it a go, too.) We don't have the expertise in
> house to try to fix the problem, unfortunately, but we're willing to try
> to help anybody who wants to give it a go.

We could look at a few things, but I fear it's really a hardware bug.
It may be dependant on the implementation of the motherboard, I've already
seen this (some drives with fails with the onboard promise controllers, but
works with an add-on promise - same promise chip, same revision), though 
in this case there was data transmission error on the bus, the drives
didn't report errors themselves. But you may have a drive firmware bug in
addition to other problems.

-- 
Manuel Bouyer <bouyer@antioche.eu.org>
     NetBSD: 24 ans d'experience feront toujours la difference
--