Comment # 14 on bug 986893 from Takashi Iwai

Thanks, the dmesg outputs are enough as a start.
The first one is at intel_idle driver while the second one looks different, at
first emitting ehci_suspend error (-EBUSY) then it got MCE.  Basically both
indicate something in a deeper level than the normal kernel crash.

One big question I forgot: is this a regression from earlier kernels?

In anyway, one easy thing to be tested is to disable the idle and cpu freq
drivers.  Try to pass the following boot options:
  intel_idle.max_cstate=1 intel_pstate=disable

These will disable the fine power tuning, so it would consume more power, but
it's interesting to see whether they influence on the behavior.