Marcus Haebler wrote:
> thanks for the patches! I am on an Intel P965/ICH8R.

I see.  That can happen too.  There was a race window where in-flight
r/w command which left SCSI midlayer but pending on libata gets executed
in the wrong mode.  If possible, please verify that it doesn't happen
with the patches applied.  I'm attaching combined patch against v2.6.20.

Thanks.

-- 
tejun