Subject: Re: comments on i386 -1.2BETA snapshot
To: Jukka Marin <jmarin@pyy.jmp.fi>
From: Justin T. Gibbs <gibbs@freefall.freebsd.org>
List: port-i386
Date: 08/27/1996 08:31:40
>> The two busses are multiplexed by a single controller.  I wouldn't expect
>> much of a difference in splitting drives across the busses.  You should
>> also be aware that the controller only allows 4 concurrent transactions
>> and that you will most certainly run into the "can't sleep waiting for
>> openings" problem in NetBSD's SCSI system if you play with CCD because
>> of this small number of controller resources.
>
>Is there any hope of improvement in the future?  This is a real problem
>for us.  CCD is just great, especially on news servers, but the SCSI
>driver problems limit the number of disks to two. :-(
>
>  -jm

As I told Luke, FreeBSD-current has a "workaround" to the problem that
entails tagging all operations that happen from an interrupt context so
that the SCSI system can sleep when it has a process context.  However, it
is still possible to get into trouble with this approach.  I've also
started my revamp of the MI SCSI layer and have completely solved this
problem with a queue based approach.  As soon as Stefan completes the
necessary changes to the NCR driver, these changes will go into
FreeBSD-current.  The current patch set can be found as:

ftp://Freefall.FreeBSD.org/incoming/scsi-diffs.96-815.gz

I have a lot of other changes in the pipe and will probably release another
snapshot today or tomorrow.  The plan is to port the entire MI layer to
NetBSD once I've incorperated all of NetBSD's improvements and the
interfaces have settled down a bit, but if someone wants to step in and track
these changes as they happen, be my guest.

>-- 
>
>                     ---> http://www.jmp.fi/~jmarin/ <---

--
Justin T. Gibbs
===========================================
  FreeBSD: Turning PCs into workstations
===========================================