References: <199403032004.UAA02603@wonderland.org> <199403040046.QAA19428@erewhon.CS.Berkeley.EDU>

cgd@postgres.berkeley.edu (Chris G. Demetriou) writes:

>> If there a swap leak somewhere:

>then you certainly couldn't tell from what you sent...  8-)

Which is not to say that there isn't a rather unpleasant swap leak present
in NetBSD.  It's not for nothing that my machine has this in its root
crontab:

	# restart VM-problematic daemons
	57 9 * * *      exec /root/bin/sendmail.restart
	57 10 * * *     exec /usr/libexec/news/ctlinnd xexec ''

It's a pity I have not had time to finish the fix I started working on over
a year ago... :-/  If anyone would like to see my (currently broken)
vm_object_collapse()-only fix to the problem, I'd be happy to dig it up and
discuss how we might get it working.

FreeBSD now includes a reasonable looking fix in the form of a
swap_pager_copy() and associated mods to vm_object_collapse().  The former
would need to rmap-ified before it could be added to NetBSD's swap_pager.c.
Any swap-pager gurus with some spare time? :)

Andrew

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