From owner-freebsd-current  Thu May 30 11:17:14 1996
Return-Path: owner-current
Received: (from root@localhost)
          by freefall.freebsd.org (8.7.5/8.7.3) id LAA05894
          for current-outgoing; Thu, 30 May 1996 11:17:14 -0700 (PDT)
Received: from dyson.iquest.net (dyson.iquest.net [198.70.144.127])
          by freefall.freebsd.org (8.7.5/8.7.3) with ESMTP id LAA05888;
          Thu, 30 May 1996 11:17:10 -0700 (PDT)
Received: (from root@localhost) by dyson.iquest.net (8.7.5/8.6.9) id NAA24087; Thu, 30 May 1996 13:16:37 -0500 (EST)
From: John Dyson <dyson@dyson.iquest.net>
Message-Id: <199605301816.NAA24087@dyson.iquest.net>
Subject: Re: VM problems 05/29/96
To: fcurrent@jraynard.demon.co.uk (James Raynard)
Date: Thu, 30 May 1996 13:16:37 -0500 (EST)
Cc: freebsd-current@freebsd.org, dyson@freebsd.org
In-Reply-To: <199605300109.BAA10515@jraynard.demon.co.uk> from "James Raynard" at May 30, 96 01:09:48 am
X-Mailer: ELM [version 2.4 PL24 ME8]
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Sender: owner-current@freebsd.org
X-Loop: FreeBSD.org
Precedence: bulk

> 
> 
> > 	Sorry I was not able to send you the output over the weekend.
> > But, I just got done testing your VM changes you submitted on 5/29 and
> > things are getting better.
> 
> I'm still having the same problems with Emacs 8-(
> 
Here is a fairly detailed status report of the solution to the
problems that you and some others have been having:

There has been several bugs that both DG and I have found.

1)	DG has found that there is a case that can block that can allow
	the page queue can change in the deactivate-->free loop in
	the pageout daemon (just fixed.)

The following were found last night by me, but not fixed in the tree yet:
2)	The inactive, free, active AND cache queues can be modified
	by vfs_bio at splbio interrupt time :-(.  (Have a fix on
	my machine.)

3)	The active queue can get modified by in the active->deactive/cache
	loop in the pageout daemon by vm_page_protect(m, VM_PROT_NONE).
	(Have a fix on my machine.)

Aall of the above has caused queue corruption in the pageout daemon,
and since DG and I both have at least 32MByte of ram, the problem is
not manifest often for us to have trouble.  Each fix has made my system
running in 4-8MBytes "better", and have been I have often been "tricked"
by the problem appearing to go away with each succesive fix.

There is still a problem when using 4MByte memory of my system locking up under
X windows, but the queue corruption is gone now.  (This (1,2) above could be
the instability that some people have seen running their system has an MMAPped
news server.)  (3) above only manifests itself when the page that is unmapped
from a process is just prior to it's pagetable page in the active queue
(believe it or not, it happens often...)

DG and I have found *numerous* bugs in the pageout daemon recently.

I plan to commit the above fixes on the night of 30May.

John