One of my servers threw another MCE this morning (Machine Check Error), and I guess it is being caused by ECC memory. I was able to capture the error codes:
CPU 0: Machine Check Exception: 0000000000000004 Bank 4: b63e200200080813 at 00000000048d7bc0 Kernel panic - not syncing: CPU context corrupt
And using mceparse, this is the explanation:
CPU 0 Status: (4) Machine Check in progress. Restart IP invalid. parsebank(4): b63e200200080813 @ 48d7bc0 External tag parity error Uncorrectable ECC error CPU state corrupt. Restart not possible Address in addr register valid Error enabled in control register Error not corrected. Bus and interconnect error Participation: Local processor originated request Timeout: Request did not timeout Request: Generic error Transaction type : Instruction Memory/IO : Other
Probably not a good idea, but I've added nomce to my boot options.