http://news.yahoo.com/s/pcworld/20080714/tc_pcworld/148353

The article makes it sound like it's specific to Intel chips (AMD chips may not have this flaw)

=====================================

Researcher to Demonstrate Attack Code for Intel Chips Sumner Lemon, IDG News Service
Mon Jul 14, 5:30 AM ET

Security researcher and author Kris Kaspersky plans to demonstrate how an attacker can target flaws in Intel's microprocessors to remotely attack a computer using JavaScript or TCP/IP packets, regardless of what operating system the computer is running.
Kaspersky will demonstrate how such an attack can be made in a presentation at the upcoming Hack In The Box (HITB) Security Conference in Kuala Lumpur, Malaysia, during October. The proof-of-concept attacks will show how processor bugs, called errata, can be exploited using certain instruction sequences and a knowledge of how Java compilers work, allowing an attacker to take control of the compiler.
"I'm going to show real working code...and make it publicly available," Kaspersky said, adding that CPU bugs are a growing threat and malware is being written that targets these vulnerabilities.
Different bugs will allow hackers to do different things on the attacked computers. "Some bugs just crash the system, some allow a hacker to gain full control on the kernel level. Some just help to attack Vista, disabling security protections," he said.
The demonstrated attack will be made against fully patched computers running a range of operating systems, including Windows XP, Vista, Windows Server 2003, Windows Server 2008, Linux and BSD, Kaspersky said, adding that the demonstration of an attack against a Mac is also a possibility.
Processors contain hundreds of millions of transistors and errata in these chips are relatively common. While some errata can affect a chip's ability to function properly-- such as the errata that last year forced Advanced Micro Devices to push back volume shipments of its quad-core Opteron processors-- many others exist unnoticed by users.
For example, the Silverthorne version of Intel's Atom processor, which lies at the heart of the Centrino Atom chip platform, contains 35 errata, according to a June specification update released by Intel.
"It's possible to fix most of the bugs, and Intel provides workarounds to the major BIOS vendors," Kaspersky said, referring to the code that controls the most basic functions of a PC. "However, not every vendor uses it and some bugs have no workarounds."

Microcode patches fix they, no?

Microcode patches can't fix all errata.

As far as specific to Intel chips, well, yeah. They exploit errata and the odds of the same errata affecting processors with different underlying architectures in the same way is about as close to zero as you can get.

I suppose it could happen. But, I suppose I could also win the lottery.

Which is not to say that AMD chips don't have similar exploits. They might not, but they could. There is no way to tell based on this. The fact there is more than one exploit for the Intel chips raises the odds that AMD chips have similar things lurking around.

Way back in the 1970s I remember reading a Byte magazine article on undocumented 6502 opcodes. Of the 256 possible values, a lot were unallocated, and some of them actually did useful, interesting things. I don't remember what they were now, but I remember actually using a few in some of my assembly code. They were great for copy protection schemes, because disassemblers didn't recognize them, and simulators ignored them.

In the 1980s it was synthetic programming on HP calculators, first the HP41CV and then HP15C.

Even with no errors in the implementation of the documented instructions, you could still generate byte sequences that had bizarre, interesting effects. Today on the x86 there are literally billions of byte sequences with unknown effects, just because of the crappy layout of the instruction encoding. The fact that someone found a security exploit on any x86 chip shouldn't surprise anyone. There are far more undocumented instruction sequences than there are documented; what should be surprising is that more people aren't running into these things every day. As long as the x86 IS encoding continues in its current form, with extensions being added thru prefix bytes and such, the problem will never go away.

The M68K wasn't perfectly orthogonal, but it was pretty clean, and most undefined byte codes would generate an Illegal Instruction trap.

As long as x86 is the dominant ISA in the world, mainstream computers will always be insecure.