Subject: Re: i386 booting, take 2
To: Phil Nelson <email@example.com>
From: Charles M. Hannum <firstname.lastname@example.org>
Date: 09/21/1998 02:24:42
> >* Incorporate the method I suggested of matching BIOS geometries to
> > disks, so that the defaults we choose are right for almost everyone.
> No, I don't agree with this method for the following reasons:
> a) It may force 4k frags
Quite frankly, that's a *PATHETIC* reason to make the user deal with
multiple geometries. Fix sysinst/newfs/whatever; don't make life
arbitrarily difficult for people!
> b) I have seen many times that the BIOS geometries do not
> address the real capacity of the disks. Currently,
> sysinst allows the user to choose a fake geometry that
> addresses more than the reported SCSI geometries address.
> If it is appropriate to do something similar for lba
> IDE drives then we should do it.
That's a complete non-issue. The geometry is only used for booting;
after that the `capacity' in the disklabel is used to bound transfers.
> c) The only reason to need the BIOS geometry is to make sure
> the MBR has the proper values in it so the boot process
> can get direct access to NetBSD boot program.
...and, currently, to read the kernel out of the file system. (I
covered this exhaustively.)
Actually, reading the above, I'm convinced that the entire disk setup
in sysinst as it currently exists needs to be scrapped. It makes the
user do far too much work to get a working system.
As I said, the changes I propose (and perhaps a fix for newfs(8))
allow turnkey installation for most people -- in fact, on *all* `off
the shelf' machines. What you suggest does not even come close.
Personally, I thought the goal was to make it easy.