From: Jason Thorpe <thorpej@nas.nasa.gov>
   Date: Sat, 03 Feb 1996 23:58:38 -0800

   ...except which "default" mapping do you want to propose be fixed into 
   the kernel?  A fixed mapping for the sun4c is totally wrong for the sun4...

Well, that's why there should be (and is) a separate sun4c kernel.

   sun4:
   	sd0	id0 lun0
   	sd1	id0 lun1
   	sd2	id1 lun0
   	sd3	id1 lun1

Say what?  I used to have a 4/110, and it didn't do anything like this...

   It gets even weirder ... some hp400 BOOTROMs expect:

Well, then there should be HP kernels distributed with that mapping too.

It's not a whole lot of work to produce these kernels, is it?  And they don't
take up a lot of space on the FTP server (or they wouldn't if compressed).  So
what's the problem?

   There's another sticky problem with wiring down the scsi mappings...you 
   have to know about all scsi busses, etc.  I think it's neat that NetBSD 
   boots on a random machine and finds all of its drives, no matter where 
   they're located.

Well, to each their own.  Personally, my feeling is that I'm not interested in
using the dynamic mapping feature anyway, so that if I had a machine with
multiple controllers (which I don't, and quite possibly never will) I would
have to config a kernel for that machine anyway.

   So, I don't really see the benefit in solving one or two specific cases 
   with wired down mappings for SCSI when the general case code works and 
   its behavior is documented...

The benefit, as I said, is that it's less hassle for people installing their
first NetBSD.

It's always better to remove a source of confusion, when possible, than to
document it.  People are used to thinking of this drive as sd0, this other
drive as sd1, etc. etc., so why force them to change?  There are plenty of
other, less avoidable hassles to deal with when switching over.

Okay, how about this.  You're already distributing one fully dynamic kernel
and one that swaps 3 and 0 but leaves 1 and 2 dynamic.  How about just
changing the latter to map 1 and 2 statically, and adding one that does a
straight-through static mapping?

   From: "Stefan Monnier" <stefan.monnier@lia.di.epfl.ch>
   Date: Sun, 04 Feb 1996 19:54:51 +0100

   >  > Of course, dropping that hardcoded root drive (and swap drvies) would be good
   >  > too.
   > Huh?  I'm not quite sure what you're saying here...

   I'm talking about having to reompile the kernel when you want to add a second
   swap drive, or the (supposed) inability of the normal kernel to mount anything
   else than /dev/sd0a on /. The root drive should be the same as the boot drive
   rather than being hardcoded in the kernel.

Truly!  This would also be a big help to first-time installers (not only is
it the way SunOS works, it's just plain more convenient to be able to install
the distributed kernel on any drive).  [This hasn't been done yet?  I haven't
gotten around to installing 1.1 -- indeed I haven't used NetBSD since about
last May, when I concluded that 1.0 wasn't quite soup yet.]

-- Scott