Chuck, 

Removing the dependency on a fixed-sized compile-time-allocated table
altogether is the right fix, for sure.


But for 1.3.x and/or the Mach-derived VM, surely the Right Thing to do
here is to compute the default MAX_KMAPENT as, say,

	#ifndef MAX_KMAPENT
	#define MAX_KMAPENT (50 + 2 * MAXPROC)
	#endif

(where the 50 and x2 is based on Mark Allman's numbers)
rather than a hardwired constant?

We're already letting users specify MAXUSERS and NPROC ( at least in
theory) so surely we should compute a default MAX_KMAPENT that's
consistent with those config parameters.  If we'd done this, Mark
Allman's etup would have bumped MAX_KMAPENT when Mark bumped MAXUSERS,
and everything would Just Work.

Anyone object if we do this (and maybe even pull it up for 1.3.x)?