From owner-freebsd-smp  Sat Aug 30 13:05:36 1997
Return-Path: <owner-freebsd-smp>
Received: (from root@localhost)
          by hub.freebsd.org (8.8.7/8.8.7) id NAA00553
          for smp-outgoing; Sat, 30 Aug 1997 13:05:36 -0700 (PDT)
Received: from Ilsa.StevesCafe.com (Ilsa.StevesCafe.com [205.168.119.129])
          by hub.freebsd.org (8.8.7/8.8.7) with ESMTP id NAA00548
          for <smp@freebsd.org>; Sat, 30 Aug 1997 13:05:34 -0700 (PDT)
Received: from Ilsa.StevesCafe.com (localhost [127.0.0.1])
	by Ilsa.StevesCafe.com (8.8.7/8.8.5) with ESMTP id OAA16897;
	Sat, 30 Aug 1997 14:00:29 -0600 (MDT)
Message-Id: <199708302000.OAA16897@Ilsa.StevesCafe.com>
X-Mailer: exmh version 2.0gamma 1/27/96
From: Steve Passe <smp@csn.net>
To: Poul-Henning Kamp <phk@critter.freebsd.dk>
cc: smp@freebsd.org
Subject: Re: HEADS UP: another set of changes. 
In-reply-to: Your message of "Sat, 30 Aug 1997 21:44:49 +0200."
             <7453.872970289@critter.freebsd.dk> 
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Date: Sat, 30 Aug 1997 14:00:29 -0600
Sender: owner-freebsd-smp@freebsd.org
X-Loop: FreeBSD.org
Precedence: bulk

Hi,

> >could you elaborate?  Do you mean you expect my current disable_intr()
> >stuff to recursively lock?  I have run it for about 13 hours now, including
> >a complete buildworld without problem.
> 
> No, but I am pretty sure there are nested simple_lock() calls in the
> vnode code.

I am completely ignorant of the vnode code.  Is this belief based on
observed lockups?

I tried using your(?) recursive MPgetlock stuff in place of simple locks
for the disable_intr() code, but it locked b4 the first message came out
of the boot sequence.  I suspect its called well before cpuid is valid,
which is necessary for that flavor to work.

I'm pondering another variation of simplelock that looks something like:

top:
    attempt to get the binary lock
    if successful
        set holder_id and return
    if fail
        compare holder_id to cpuid
        if same
            panic, we shouldn't be holding and asking for a binary lock
	else if legal (0 < holder_id < mp_ncpus)
            goto top
        else
            if smp_active
                panic, this shouldn't be
            else
                ignore, we're just early

--
Steve Passe	| powered by
smp@csn.net	|            Symmetric MultiProcessor FreeBSD