Subject: Re: Possible (but Evil) work-around for Pentium bug?
To: Peter Seebach <>
From: Eric Haszlakiewicz <haszlaki@UAccess.NET>
List: current-users
Date: 11/12/1997 17:23:36
> >upgradable microcode.  This would explain the SCO fix - it fixes the
> >floating point problem.  Does anyone know which problem BSDI has fixed?
> It's the F0 bug.
> >I'm not sure if the F00FC7C8 bug can properly be fixed in software
> >(without diabling the L1 cache) - but it appears the floating point bug is
> >being fixed by a microcode update.
> I have no clue how we do it, and, as they say, we are not at liberty to
> discuss it.  You could ask Intel - we have announced that we have a fix,
> and yes, it apparently works.
> I personally disbelieve it - it's patently, and obviously, physically
> impossible.  But it's apparently working.

	Although I don't know offhand, and I don't have any books on how
the pentium handles things like this, I have heard that there is a workaround.
This could be completely bogus, but here goes:
	CPL = 3 for user process.
	if IOPL < CPL then the the LOCK instruction (F0) will cause an
	exception/trap.  If this happens then you do the obvious.  (check
	for the bad bytes)

Does this sound right?  Anyone have some protected mode books handy?