Re: New, fundamental Intel Pentium flaw?

Robert S. Thau (rst@ai.mit.edu)
Mon, 10 Nov 1997 16:36:54 -0500 (EST)


This is bad C code, but that's NO excuse for it to crash out a chip.
This *can't* have been in there all along, could it? Shit!

Yes it can. See discussion on the linux-kernel mailing list and numerous
other places. This apparently shuts the chip down cold, literally: the
internal logic locks up and stops generating heat, reducing the chip to
room temperature. There is (according to letters on linux-kernel) a similar
bug in at least some Cyrix CPUs, which causes the chips to seize up, but
keeps them burning power; this has been summarized as "Pentium dies; Cyrix
goes into a coma".

Of course, all this does is denial of service; a bug of this ilk which
put the chip into supervisor code would be much, much more dangerous, and
might still be floating out there.

What amazes me is that it took so long to find the bug, with all those
linux folks supposedly running crashme (which generates random code and
branches to it; a procedure which can and does cause a lot of commercial
unices to panic) as part of their stress tests.

rst