On Mon, May 27, 2002 at 01:58:11PM -0700, Wolfgang Rupprecht wrote:
> 
> I'm seeing a handful of new disk errors under -current/i386.  At first
> I thought that the problem was that my 2-year old IBM disk was
> failing, so I bought two brand new Seagates Barracuda IV's.  After
> only three days I'm seeing errors again.  My guess would normally be
> bad wdc hardware or bad cables.  The weird part is that the disks were
> written a few days ago using a "tar|tar" to clone the old disks.  At
> that time a tar back off of the newly written disk worked just fine
> without errors, so the data was written ok.  Now a few days later I
> get 3 uncorrectable errors and 2 correctable ones from files that I
> know verified ok.
> 
> Anyone know what gives?  The disks are cool to the touch so it isn't a
> heat problem.  The only groping-for-straws guess I have is a problem
> with powering NetBSD-current down, eg something is scribbling on the
> disk as the power supply is turned off.  That doesn't seem likely
> since I have a two second power-off-delay compiled into the kernel.

Maybe an issue with your power supply ? I've seen failing disks ressurect
with a new power supply.

> 
> Does NetBSD "clean" soft errors by writing back the corrected data for
> the badly read block?  Would this be a good idea?  So far I've been
> trying to find which file the bad data is in and copy back a good copy
> of the file from archives just so that I know the disk data is ok.  It
> would also be nice to know if the disk block is ok too, or If I just
> put a defective block back on the freelist.

From the IDE driver "soft" errors are transfer errors on the wire, not errors
from the disk itself. Such transfers are retryed.

-- 
Manuel Bouyer <bouyer@antioche.eu.org>
--