On Mon, 5 Feb 1996 kpneal@eos.ncsu.edu wrote:

> >    Date: Thu, 1 Feb 1996 23:21:11 -0800
> >    From: monroe@teleport.com (Monroe Williams)
> > 
> >    BTW, at least on the mac68k port (which, I believe, doesn't yet run on 
> >    any machines with multiple scsi busses), would it make more sense to 
> >    configure the distributed kernel this way?  Most people seem to know what
> >    scsi IDs their devices have, but I've seen a lot of confusion on the list
> >    about the scsi id -> device node mapping system.
> > 
> > I second this.  If you folks want NetBSD to be used as widely as possible, it
> > will help if you make converting to it as easy as possible.  If users, once
> > they get up and running, want to turn on the dynamic assignments, they can do
> > that easily enough; there's just no reason to add this to the hurdles they
> > have to leap to get started.
> > 
> > Personally, I don't think it's much of a feature to have to edit /etc/fstab
> > every time I add or remove a disk that doesn't have the highest SCSI ID, but
> > de gustibus non disputandum and all that.  Those who want it can enable it;
> > those who don't shouldn't have to deal with it.
> > 
> > I would add that there should be a distributed SPARC kernel that mimics
> > the sun4c's 3 <-> 0 swap (bizarre though it is).
> 
> Sigh, not this argument again.
> 
> I think the argument for config sd* on * goes like this:
> You can nail down the first 7, but if the user has more than
> one SCSI card, or whatever, or more than 1 drive per lun like
> the ESDI-SCSI cards found on some (mostly) Suns, then you might
> nail down the drives that might not be there, and miss a few
> that are there. 
> 
> So you config sd* on * and catch all of the drives in the
> GENERIC case. 
>
> The user is encouraged to compile their own kernel once they
> get their machine running, and put drives where they want them.

???

This is the first I've heard of this suggestion. I agree it's good for 
experienced users, but I wonder about it for the rank novices. On the 
mac68k list, we get a lot of people who are really new to admining a UN*X 
box. Often times, they have little disk space. Having them compile a 
kernel right after they set up the machine sounds like we'd be asking for 
LOTS of "Help me"'s. :-(

But I see the flip-side of the argument, especially the possibility of 
missing an existing drive.

> Perhaps putting the "nailed down" lines in the GENERIC kernel
> config file, commented out? So the new user with a running
> NetBSD machine just has to comment out 1 line and uncomment
> 7 or whatever?

Sounds like a good first step. Though couldn't we have the hard-coded 
(commented out or not to taste) before the sd* line? That way if you do 
have an extra scsi bus, or a dual-lun drive, it does get configured 
(though possibly as sd8). ?

Also, since we have a different GENERIC for each port, and we try to 
respect the hints of the native OS, could we just have some ports 
by default hard-code sd0-sd6 while others not? I suspect most mac users 
would expect the SCSI ID of the drive to be in the sd's name. Sounds like 
the Sun default would be (should be) to number drives as they are found.

Or maybe make two "GENERIC" kernels, one w/ & one w/o hard coding? They'd 
need different names, but I can't think of another good one.

Take care,

Bill