Avian’s Blog

A couple of days ago I got the following error in the logs of my home server:

Dec 13 18:53:41 chandra kernel: MCE: The hardware reports a non fatal, correctable incident occurred on CPU 0.
Dec 13 18:53:41 chandra kernel: Bank 2: 940040000000017a

MCE is short for Machine Check Exception - it's a way for the CPU to tell the software that internal error checking has detected a bit flip in one of the CPU's registers or execution units. Since the error was correctable in this case, nothing happened as far as software was concerned, except for a log message being generated.

One common case when this might happen is if an energetic particle hits the circuit. It creates an ionization path in the semiconductor, which causes a temporary short-circuit and a flip-flop storing one bit of information flips from one state into the other.

A lot of such particles come in form of cosmic rays from space, where they are among other things created by supernovae.

On the other hand, the thing that caused my CPU to hick up could also come from a much closer source, like a wholly unfashionable radioactive decay of an atom in the CPU's package.