> > There's an alternate approach which is probably less drastic, which is
> > for the boot program to pass the pathname of the kernel to the kernel
> > when it starts; the kernel can pass this to init, and init can pass it
> > to /etc/rc, which can use it for the kvm_mkdb stuff...
> This has some problems.
> Consider the case where you boot a kernel from a device that isn't the
> root device (e.g. a partition that's only optionally mounted).
> You could also boot a kernel from a device that isn't even supported by
> this kernel.

Or heck, imagine booting from a kernel on a floppy that you're removed and put
back in its little sleeve (anybody do that? ;-).

Actually, Marc Wandschneider commented in email that "Adam Glass has been trying 
to get [him] to write the sysctl() system call," so Marc, consider yourself
encouraged :-).  (The CMU implementation of table() is certainly grotesque, at
least the copy I've seen as filtered through OSF is, so I'm not attached to
table() as such...)

------------------------------------------------------------------------------