You are on page 1of 6

Intel battles another CPU bug

Alexander Wolfe
Electronic Engineering Times.​ .953 (May 12, 1997): p1.
Copyright: COPYRIGHT 1997 United Business Media LLC - UBM. All rights reserved.
No part of this information may be reproduced, republished or redistributed without the
prior written consent of CMP Media, Inc.
http://ubmtechnology.com/

Resumo:

Intel is acknowledging a floating-point problem in its Pentium II and Pentium Pro


processors, promising a correction in the next versions of the chips. The company is
also developing software workarounds for the current chips. The bug was reported on
the Intel Secrets Web site. It has been named Dan-0411, after the person who reported
the bug. His actual identity is unknown, but he is believed to work in the semiconductor
industry. The bug was first described as occurring when overflow conditions occur in
floating-point addition, when an operand is stored as an integer. The processor should
indicate this as an invalid-operand error, but it does not. The chip actually sets a flag
that indicates an overflow has occurred. This is not the correct action. Thirteen software
vendors have made announcements that their applications are not affected by the flaws.

Texto completo:

Santa Clara, Calif. - Intel Corp. last week owned up to the floating-point bug involving its
Pentium II and Pentium Pro microprocessors, stating that it would fix the "flag erratum"
in future chip steppings. In the interim, software workarounds are being prepared.

In addition, 13 software vendors stated that their programs aren't affected by the flaw.

The news capped a weeklong cyberspace soap opera that took numerous twists and
that prompted intense scrutiny of the renegade "Intel Secrets" Web site, where first
reports of the flaw appeared last Monday.

The bug has been designated Dan-0411, after the person who uncovered the bug and
conveyed the information to Intel Secrets, according to Robert Collins, proprietor of the
site. Intense speculation has surrounded the question of Dan's identity; some sources
say he works in the semiconductor industry (see sidebar, page 4).

News of Collins's bug report spread rapidly after a story about it appeared on EE Times
Online (www.eet.com) last Monday.
At midweek, Collins abruptly revised his technical description of the bug, prompting
questions on some Internet discussion groups about the veracity of his initial claim. At
first, Collins said the bug could occur during overflow conditions involving floating-point
addition, when an operand is stored as an integer. "What the processor should do is
signal an invalid-operand error; it does not," he said at the time.

Instead, Collins said, the chip sets a flag indicating that an overflow has occurred. While
that might seem to be the correct action, Collins said it isn't.

Collins subsequently changed his report to say that the Pentium II and Pentium Pro
processors actually set the "precision-exception" (PE) bit-not the overflow flag-during
the affected operations.

"It's the PE bit that's getting set," said Collins. "I wrote a program that tests for the 'Dan'
bug. It loads 80-bit floating-point numbers and tries to store them as integers, and sees
if the error occurs." When that happens, the test program prints out the status word
containing the PE bit.

Throughout the week, many knowledgeable sources maintained that the purported bug
was little more than a temporary public-relations black eye. "I think it [the initial bug
report] is much less serious than the 1994 Pentium FDIV bug," said Richard Smith,
president of Phar Lap Software (Cambridge, Mass.), a vendor of X86 programming
tools.

The 1994 floating-point division bug, which was first reported by EE Times, was a
public-relations disaster for Intel that resulted in its first-ever chip recall and an eventual
charge against company earnings of $475 million.

"As I understand it, the [Dan-0411] bug only manifests when storing 80-bit values, which
almost nobody ever uses. We certainly don't," wrote John Carmack, a lead developer at
games vendor Id Software, in an e-mail posted to an Internet group discussing the
issue.

Bad timing

Still, talk of the bug couldn't have come at a worse time for Intel, which last week
formally unveiled its Pentium II CPU in separate events in New York, London and
Taipei. In England and Taiwan, talk of the bug nearly overshadowed the Pentium II
introduction. In New York, Intel stuck to its product-launch script and didn't field
questions from the floor.
But Intel officials told EE Times there that hundreds of engineers were at work
throughout the week investigating the bug report, in an effort to provide a quick and
quantitative response to the technical community.

"We've got test suites, we can write unique software tests and we can go in and do
specific hardware tests," said Mike Fister, the vice president and general manager of
Intel's microprocessor division (Hillsboro, Ore.).

"Since we got the report, we've been jumping all over it," added Richard Dracott, Intel's
marketing director for the Pentium II. "It normally takes weeks to characterize [a flaw].
We've greatly accelerated that process."

Dracott said he doesn't think news of the flaw will affect sales of the Pentium II, which
Intel is aiming at business-computing users.

Strike two

Intel officials last week also addressed questions about how lightning could apparently
strike the company twice-first with the infamous Pentium FDIV flaw of November 1994
and now the new Dan-0411 glitch.

"I think people need to come to terms with the fact that all microprocessors have errata.
As soon as we understand them, we take them public," said Intel's Dracott.

"Microprocessors are a difficult technology. I don't think any one is without anomalies,"
said Jeff Edson, vice president of Intergraph Computer Systems (Huntsville, Ala.). "We
expect Intel to respond quickly."

"People need to understand that all microprocessors have bugs," said Linley Gwennap,
editor-in-chief of the Microprocessor Report newsletter, echoing Dracott's comment.

Indeed, at midweek Intel was proactive in releasing the first bug list
(developer.intel.com/design/pentiumii/update/ppiisu.htm) for the Pentium II. That
document confirmed that the purported Dan-0411 flaw wasn't one that was already
familiar to Intel.

However, of the 18 Pentium II errata for which immediate fixes aren't planned, two
involve floating-point calculations. Those two glitches also apply to the standard
Pentium Pro CPU. The standard part is a cousin to the Pentium II but lacks MMX
multimedia extensions and uses old-style Socket 7 packaging.

Specifically, the first flaw is that a "floating-point operand pointer may be incorrectly
calculated after floating-point access which wraps the 64-kbyte boundary in 16-bit
code."

[Intel's detailed description noted that the FP operand pointer is the effective address of
the operand associated with the last non-control FP instruction executed by the
machine.]

[If an 80-bit FP access occurs in a 16-bit mode (other than protected mode), if the
memory access wraps a 64-kbyte boundary and the FP environment is subsequently
saved in 32-bit mode, then the subtraction routine used to calculate the FP operand
pointer will assume the FP access was in 32-bit mode. It will also incorrectly fill out the
high word of the address with 1's.]

The second flaw is that "floating-point exception flag may not be set" in certain obscure
scenarios.

Pentium II details

As for the technology of the new chip itself, the Pentium II combines the upgraded
architecture of the Pentium Pro with the extra multimedia instructions of the Pentium
MMX. Intel said it has already shipped 100,000 CPUs to its OEMs.

Packaging the chip with a larger second-level cache in a 5- x 2-inch single-edge


cartridge allows 233- to 300-MHz clock speeds. The 233- and 266-MHz versions are
$636 and $775, respectively, in 1,000-unit quantities. A 300-MHz version is due in the
third quarter at a unit price of $1,981 in similar quantities.

Other details at the launch of the Pentium II revealed it to be more of a physical


packaging exercise than a vehicle for architectural change. The packaging allows the
chip to be clocked at 300 MHz and beyond, which makes it the fastest Intel processor
yet. Intel produced several benchmark results to back that up.

Copyright 1997 CMP Media Inc.

Related Article Santa Clara, Calif. - Industry speculation surrounding the Intel
floating-point flaw centered on the identity of "Dan," the mysterious source who
reportedly found the bug affecting Intel's Pentium II and Pentium Pro processors and
conveyed his information to Robert Collins of the "Intel Secrets" Web site.

Dan's motives were also questioned, via postings in cyberspace discussion groups
opining that he works within the semiconductor industry. If that's true, Dan could have
an interest in raining on Intel's parade by releasing a bug report just weeks before the
launch of the Pentium II.

According to knowledgeable sources, Dan is employed at a company making a Pentium


Pro-class clone, though it's not Advanced Micro Devices or Cyrix. Moreover, Dan's first
name is really Dan.

"Dan's motives aren't bad," said Cleve Moler, co-founder of Mathworks Inc. "He's trying
to carefully model the behavior of the processor."

"The only people who would run across this are people who write test suites for a
living," said Richard Smith, president of programming-tools vendor Phar Lap Software
(Cambridge, Mass.). "My guess would be somebody at a chip or a compiler
company-somebody doing low-level coding. I think it'd be worth investigating who this
'Dan' guy is."

Don't know him

Both Moler and Smith stressed that they do not know Dan.

For its part, Intel said it doesn't believe a competitor is behind the bug report. And
Advanced Micro Devices categorically denied any link to Dan. "Absolutely not," said a
company spokesman. "We have nothing whatsoever to do with Dan. We learned about
this [bug] when everybody else did."

"Intel Secrets" proprietor Collins declined to identify "Dan."

Copyright 1997 CMP Media Inc.

Citação da fonte​ (MLA 8​th​ Edition)

Wolfe, Alexander. "Intel battles another CPU bug." ​Electronic Engineering Times,​
12 May 1997, p. 1+. ​Academic OneFile,​
go.galegroup.com/ps/i.do?p=AONE&sw=w&u=capes&v=2.1&id=GALE%7CA19433
514&it=r&asid=7b68d0ec6a435fbe539b792d26eec1e9. Accessed 11 Sept. 2017.
Número do documento Gale: ​ GALE|A19433514

You might also like