[ale] Intel confirms bug (fwd)

[Mike Kachline]

	If you own a P5, you want to read this. If you own a P5 with 
hostile users, you *definately* want to read this.
							- Mike
============================================================================
Michael Kachline - CS, Georgia Tech
kachline at cc.gatech.edu
http://brightstar.gt.ed.net/kachline/
============================================================================

                Intel (INTC) today confirmed that a bug can
                crash its Pentium processors, and the chip
                giant was rushing to come up with a fix. 

                Intel acknowledged the "F0" bug this
                afternoon, saying that it is now looking into a
                work-around. A spokesperson said the
                company would know more about the
                work-around "within a week." 

                Though completely different in nature, the F0
                bug evokes memories of a calculation glitch
                in the Pentium found by a mathematics
                professor back in 1994, which cost Intel close
                to $500 million dollars. 

                "Yes, we are confirming this is an errata. This
                was not on our published list of [processor
                bugs]. We didn't know it existed until Friday,"
                the spokesperson said today. 

                The bug was first reported by CNET's
                NEWS.COM Friday afternoon. Intel is an
                investor in CNET: The Computer Network. 

                The Pentium F0 bug can freeze up Pentium
                MMX and "classic" Pentium (non-MMX)
                computers, machines that number in the
                hundreds of millions worldwide. 

                A computer affected by the bug freezes
                completely and must be restarted. A
                "ctrl-alt-delete" keyboard operation, which
                can generally be used for "rebooting" a
                computer, will not work. The computer must
                be restarted. 

                Intel publishes a list of Pentium, Pentium Pro,
                and Pentium II bugs. These problems,
                referred to as "errata," are supposed to
                cover all known bugs. 

                Intel confirmed that the bug occurs only when
                the processor receives an illegal, one-line
                instruction. "The result is the system could
                freeze," Intel said. 

                "This bug will not be found in commercial
                software," Intel added. The bug does not
                affect Pentium Pro or Pentium II processors. 

                Nathan Brookwood, an analyst at marketing
                research firm Dataquest, said the bug is
                similar to glitch found in the IBM 790
                computer back in the early 1960s. In that
                case, a single errant instruction was able to
                bring down an IBM computer. 

                "This is dangerous if you're in a multiuser
                environment. If somebody does it to himself,
                fine. I don't care. But if somebody takes a
                bomb onto a plane I'm flying on, then I care,"
                he said. 

                One interesting aspect of the bug is that, in a
                crash, because the processor is actually still
                running, Intel may be able to "discover the
                state" of the processor and thereby come up
                with a software fix, Brookwood said. 

                "They may be able to set up an error-handling
                condition...That's what they are working on
                now within Intel and with key operating
                system suppliers," he said. 

                Brookwood said the strategy will probably be
                to go to operating system vendors and say,
                "Here's the problem and here's the
                work-around, and go figure out how to
                implement it." 


Darrin's Vegas News and Links
http://www.angelfire.com/nj/kitaro
 
 

---------- Forwarded message ----------
Date: Sun, 9 Nov 1997 22:31:19 -0600
 From: Mark Lin <mlin at ECE.MsState.Edu>
To: js1 at microwave.ph.msstate.edu, pde at nwu.edu
Subject: Fwd: Bug in the Pentium

Subject: Bug in the Pentium
 From: Tim Showalter <tjs+bp at andrew.cmu.edu>
X-Newsreader: WWW (128.2.232.95)
Message-ID: <rkP$mOA78GA.181 at newsgroups.intel.com>
Newsgroups: intel.microprocessors.pentium
Date: Fri, 07 Nov 1997 21:03:29 -0800
Lines: 31
Path: newsgroups.intel.com

Since I haven't seen any messages on your newsgroups, I'll ask.  The bugtraq
mailing list and others report that there is a bug tickled in the Intel
Pentium(R) processor by the following C program.  (It's not my code, I didn't
find it, I'm just worried about it.)

unsigned char x [5] = { 0xf0, 0x0f, 0xc7, 0xc8 };
main ()
{
       void (*f)() = (void (*)()) x;
}

This program should cause the OS to complain about an invalid instruction.
But it doesn't.  Aparently what it does is gives the CPU a very special
invalid instruction, causing the CPU to come to a screeching halt.  I'm
told that it works on any Pentium running any OS.  It does not need the
CPU to be in privileged mode; it can be run in user space on operating
systems that know the difference.

I can confirm it halts two Pentium systems under Linux.  I've heard it, uh,
runs under NT and 95.

This is not a serious problem for single user machines where crashing is only
punishing one user, but could be nasty on multi-user machines.

I'd appreciate a response from the Intel folks.  How will you be addressing
this problem?  (Please respond via personal mail as I don't read this
newsgroup.)

Tim Showalter