On Wed, 16 Sep 1998, Larry Kollar wrote:

> Scott Reynolds wrote: 
> 
> > I've just committed changes to the SCSI layer and machine-
> > independent 5380 controller driver that resolve the problem
> > with the sbc driver that crept in at the beginning of August.
> 
> Sounds good....  What was the nature of the problem?  And would
> those of us running 1.3 see any reliability improvements?

Probably not.  It seems unlikely that the problem could have caused any
significant data corruption, though I suppose it's remotely possible.

When the SCSI code was converted to use the pool allocator to allocate and
free certain structures, we uncovered a bug where you could reference a
structure that had already been marked as freed.  The pool allocator
modified the structure in such a way that the SCSI layer went to sleep
while waiting for a transfer to complete, when in fact the transfer had
_already_ completed.

If you'd like a more detailed description of the problem and its solution,
you'll want to check out the source-changes mailing list archive.

--scott