Allen Briggs writes:
>> Yep. There's a bug somewhere in the scsi driver which goes away when
>> splimp() gets set tp spl4(), not spl2().
>
>Actually somewhere in the sbc scsi driver, apparently not in the ncrscsi
>driver.  It's going back to spl2 in the release branch, too.

Okay, I'm confused now.  :/

Until very recently, I was running a -current kernel a couple of months
old with splimp() == spl2() and the ncrscsi driver.  This kernel would
randomly hang somewhere in scsi interrupt handling code.  Last week I
supped and recompiled a kernel with the sbc driver and splimp() ==
spl4().  The new kernel doesn't seem to have the scsi hang, but it has
pretty bad serial overflow problems.  (Transfer rates are maybe 1/2
what I got with the last kernel).

At this point, do I have to live with one or the other of these
problems, or has someone figured out how to fix both of them at the
same time?

-- monroe
------------------------------------------------------------------------
Monroe Williams                                      monroe@teleport.com