From owner-freebsd-arch  Fri Mar 16  7:52:40 2001
Delivered-To: freebsd-arch@freebsd.org
Received: from feral.com (feral.com [192.67.166.1])
	by hub.freebsd.org (Postfix) with ESMTP
	id 8FAC437B71C; Fri, 16 Mar 2001 07:52:36 -0800 (PST)
	(envelope-from mjacob@feral.com)
Received: from beppo (beppo [192.67.166.79])
	by feral.com (8.9.3/8.9.3) with ESMTP id HAA22797;
	Fri, 16 Mar 2001 07:52:38 -0800
Date: Fri, 16 Mar 2001 07:52:34 -0800 (PST)
From: Matthew Jacob <mjacob@feral.com>
Reply-To: mjacob@feral.com
To: Alfred Perlstein <bright@wintelcom.net>
Cc: Benno Rice <benno@FreeBSD.ORG>, Doug Rabson <dfr@nlsystems.com>,
	John Baldwin <jhb@FreeBSD.ORG>, arch@FreeBSD.ORG
Subject: page faults in AIX kernel 
In-Reply-To: <20010316073921.W29888@fw.wintelcom.net>
Message-ID: <Pine.BSF.4.21.0103160744260.53255-100000@beppo.feral.com>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Sender: owner-freebsd-arch@FreeBSD.ORG
Precedence: bulk
X-Loop: FreeBSD.ORG


[ different topic ]

On Fri, 16 Mar 2001, Alfred Perlstein wrote:

> * Matthew Jacob <mjacob@feral.com> [010316 07:28] wrote:
> > 
> > > > 
> > > > (in the PPC architecture I believe that page faults are interrupts)
> > > 
> > > Not quite.  Page faults and external interrupts are both classed as
> > > 'exceptions'.  External interrupts however can be turned on or off using the
> > > EE bit of the machine state register without disabling page fault (DSI or ISI)
> > > exceptions.
> > 
> > Ah! Thanks for the clarification.
> > 
> > What I had gotten my notion from was from AIX- if you have a spinlock
> > contested and take a page fault in the kernel, you lock up. 
> 
> I've heard that page faults in the AIX kernel are "ok" (obviously
> not when holding a mutex) because some of the kernel memory is
> actually pageable.
> 
> Any idea on what structures they keep in pageable memory?
> 
> (just wondering)

They have a concept of pinned and unpinned memory. IIRC (it's been a while
now- one of my former clients took back the 2 processor 300MHz PPC box about
a year ago and I hadn't done any work with it for 2-3 years) pinned memory
isn't unpageable while unpinned memory is.... Let's see if I can dig up some
examples... I can't remember exactly what other 'structures' they then
actually have as pageable... I guess pretty much anything that isn't pinned
(now isn't *that* a useful answer) :-)..

Okay- here's one:


        /*
         * If this is the first time through:
         *   Add entry to the device switch table to call this driver
         */
        if (nconfigured == 0) {
            result = devswadd(dev, &vtd_dsw);
            if (result != 0) {
                Trace1(0, "vtd_config: devswadd result: %d", result);
                break;
            }
            result = pincode((int (*) ()) vtd_timer);
            if (result) {
                Trace1(0, "vtd_config: pincode result: %d", result);
                devswdel(dev);
                break;
            }
        }
        nconfigured++;

Note that the timer funcion is pinned so that it can't be paged out when a
soft call happens (which would be embarrassing).



-matt



To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-arch" in the body of the message