As originator downgraded question goes to Alexey. Can you reproduce problem at will? If so, can you collect 'info reg' at crash point? In the context it's more valuable than so detailed backtraces... And what's your /proc/cpuinfo?