"After struggling with this issue for well over a year and really pushing hard to track it down in the last two months I was finally able to come up with a test case running 'cc1' from gcc-4.4 in a loop and get it to fail in less than 60 seconds. Prior to finding this case it would take anywhere up to 2 days with 48 cores fully loaded to reproduce the failure. AMD confirms. '...it isn't every day that a guy like me gets to find an honest-to-god hardware bug in a major cpu!'"
Post a Comment
Member since:
2005-09-27
Member since:2005-07-06
Member since:2005-09-03
Member since:2012-03-06
That depends, is it a silicon bug (actual hardware) or a microcode bug (base instruction execution code)?
I am no major-processor expert, but I am an EE and normally if a bug is present in hardware (which costs bazillions of dollars to fix) a patch is created in software (firmware, microcode or whatever) to work around it.
Member since:
2009-10-31
Member since:
2006-12-06
Member since:
2007-10-26
Member since:
2007-03-26
Member since:
2006-07-14
Its not a Joe Shmoe like Bill, Shooter of Bul that found this cpu bug, its Matthew, Father of DragonFly BSD and HammerFS, Dillon.
It would take someone like him, who's stubborn enough and secure enough in his own ability to fork off a major operating system, to find a CPU bug.
Me, I'm excited when I find a bug in my operating system ( if its not windows), database, or programming language and track it down in code. So I imagine he's feeling like that only x 1000
Member since:2006-03-03
Member since:2005-07-11
Member since:2011-01-28
phoenix,
"...so he knows/understands the inner workings of a CPU, assembler, etc. It's not like John Q. Public, some random OS developer, found the bug."
As a random OS developer myself, I feel totally dissed.
At what point did "OS developer" fall to the same level as a "script writer"? No disrespect 
. What's taken the top spot for prestigious technical occupation?
Member since:2010-03-08
Member since:2009-03-27
Member since:2011-08-13
Member since:
2006-02-12
https://en.wikipedia.org/wiki/Pentium_FDIV_bug
Hopefully AMD is looking close at what Intel did in this situation.
Member since:2011-01-28
akkad,
These are more common than you might realize. Having hundreds of "erratum" isn't unusual. For it's part, Amd already has experience with CPU faults, for example, AMD's Phenom processor line once exhibited caching errors and the solution for processors in the field was to disable certain caches.
http://techreport.com/articles.x/13741
Something I didn't know until searching today, is that Linux kernel patches were released to work around the processor bug...so there's another answer to the first poster's question "How do you patch a CPU?".