From owner-freebsd-current  Sun Sep 20 11:33:54 1998
Return-Path: <owner-freebsd-current@FreeBSD.ORG>
Received: (from majordom@localhost)
          by hub.freebsd.org (8.8.8/8.8.8) id LAA11126
          for freebsd-current-outgoing; Sun, 20 Sep 1998 11:33:54 -0700 (PDT)
          (envelope-from owner-freebsd-current@FreeBSD.ORG)
Received: from godzilla.zeta.org.au (godzilla.zeta.org.au [203.15.68.22])
          by hub.freebsd.org (8.8.8/8.8.8) with ESMTP id LAA11078
          for <current@FreeBSD.ORG>; Sun, 20 Sep 1998 11:33:36 -0700 (PDT)
          (envelope-from bde@godzilla.zeta.org.au)
Received: (from bde@localhost)
	by godzilla.zeta.org.au (8.8.7/8.8.7) id EAA31365;
	Mon, 21 Sep 1998 04:33:06 +1000
Date: Mon, 21 Sep 1998 04:33:06 +1000
From: Bruce Evans <bde@zeta.org.au>
Message-Id: <199809201833.EAA31365@godzilla.zeta.org.au>
To: jhay@mikom.csir.co.za, jkh@time.cdrom.com
Subject: Re: [CAM] Broken tagged queuing drive?
Cc: current@FreeBSD.ORG, gibbs@plutotech.com, kato@ganko.eps.nagoya-u.ac.jp,
        mantar@netcom.com
Sender: owner-freebsd-current@FreeBSD.ORG
Precedence: bulk
X-Loop: FreeBSD.ORG

>I also see it here on my dual P5 machine. It seems to be caused by the
>latest version of isa/clock.c. If I go to the previous version (while
>keeping the rest of the source the same), the panic disappear. It only
>happens on the dual P5 machine, the 486 runs happily with the latest
>version of isa/clock.c.

Apparently it wasn't safe to use disable_intr()/enable_intr() under
SMP, or SMP locking doesn't nest properly :(.  The interrupt nesting is:

clock interrupt -> normal interrupt masking, whatever that is for SMP
                   ...
                   disable cpu interrupts (known to be enabled to begin with)
                   lock clock
                   sometimes:
                     i8254_get_timecount()  -> save cpu interrupt mask
                                               disable cpu interrupts (nop)
                                               lock clock (oops?)
                                               ...
                                               unlock clock
                                               restore cpu interrupt mask
                   unlock clock             <- return
                   enable cpu interrupts

i8254_get_timecount(), getit() and set_timer_freq() are supposed to be
callable with interrupts disabled.  Apparently this never worked.

It's probably unnecessary to lock the clock in clkintr() - the giant
lock suffices.  I should never have suggested turning disable_intr()
into a macro to hide the locking.

Bruce

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