Oh, I'm sorry, I should have responded... I found the problem...
the problem is really that st_interpret_sense is getting into the
game for the check condition on the initial test unit ready...

If there happens to be an EOM detection or some other error as well,
st_intepret_sense will incorrectly mark an error rather than letting
the generic handler code return the ERESTART that neeeds to happen..

I was trying to test this, but I've spent most of the day trying to
figure how to get around periodic alignment panics in the nfs code 
plust other stuff....

I'll be checking in a fix tomorrow...


Michael- that second check condition on a write- that's really something
busted in the drive f/w. That particular drive is a IBM branded Archive
DAT drive, but that's a Vendor Unique ASC/ASCQ (or garbage- the distinction
is sometimes small) so it's hard to know what to do when this happens.

(got a manual for this puppy?)