From owner-freebsd-stable  Sat Oct 28 11:41:41 1995
Return-Path: owner-stable
Received: (from root@localhost)
          by freefall.freebsd.org (8.6.12/8.6.6) id LAA07509
          for stable-outgoing; Sat, 28 Oct 1995 11:41:41 -0700
Received: from godzilla.zeta.org.au (godzilla.zeta.org.au [203.2.228.19])
          by freefall.freebsd.org (8.6.12/8.6.6) with ESMTP id LAA07488
          ; Sat, 28 Oct 1995 11:41:35 -0700
Received: (from bde@localhost) by godzilla.zeta.org.au (8.6.9/8.6.9) id FAA05152; Sun, 29 Oct 1995 05:39:01 +1100
Date: Sun, 29 Oct 1995 05:39:01 +1100
From: Bruce Evans <bde@zeta.org.au>
Message-Id: <199510281839.FAA05152@godzilla.zeta.org.au>
To: stable@freebsd.org, swallace@ece.uci.edu
Subject: Re: probs with latest stable snap
Cc: bugs@freebsd.org
Sender: owner-stable@freebsd.org
Precedence: bulk

>Ahh!  I have traced it to line 180 of microtime.s:
>	        divl    %ecx            # get value in usec

>Hmmm, it appears the code checks for pentium_mhz != 0, so it could not
>be div by 0.  However, I looked up div, and it says, "If the
>dividend is 0 or if the quotient is too large to fit in the result accumulator, a divide error fault (interrupt 0) occurs.

>So, if the pentium cycle count is > 2^59, then on my 90 MHz pentium
>(correctly reported), the fault will occur.  I do not know what originally
>caused the first panic, but it appears it put the pentium clock count
>to a large number.  Then it rebooted and gets into a panic div 0 loop.
>I noticed that after I hit reset, then it booted up just fine (resetting
>the counter?).

>I noticed in -current code it has a i586_ctr_bias long long which
>gets initialized and subtracts the cycle count before the division
>in microtime.s.  I think something like this should be put into -stable
>to prevent this panic.

Normally the counter gets set every clock tick.  This reduces the dividend
in much the same way as subtracting i586_ctr_bias, so there should be no
problem dividing.  However, microtime() is apparently sometimes called
before the counter has been reset after booting.  -current initializes
i586_ctr_bias early so there should be no problem.  The equivalent in
-stable would be resetting the counter early.

Bruce