Subject: UVM and overcommitting...
To: None <current-users@netbsd.org>
From: Peter Seebach <seebs@plethora.net>
List: current-users
Date: 06/06/2000 10:22:28
So, I am wondering how hard it would be for UVM to get a "really guarantee
this memory" flag.  Overcommit-and-kill-if-you-run-out works fine for most
software, but there are occasional rare programs where:
	1.  The program may deal gracefully with failure.  I know, it's
	rare, but it happens...
	2.  The program may be much better off being able to save its work
	when it runs out of memory, rather than being killed abruptly.

So, how hard is it to get a flag saying "do not overcommit ever", or at the
very least "always really-reserve space for this process"?

This is arguably a conformance issue; there's nothing in the C spec saying
that malloc is allowed to return a non-null pointer to memory which will cause
the program to die if you access it.

-s