Andi Kleen wrote:
> Ok thanks for the information. I will stare a bit at the patch.
> 
> It is very mysterious though. Even if the patch was somehow wrong
> the worst thing that could happen is that you end up with interrupts
> off when you shouldnt, and the NMI watchdog is very good 
> at catching that.

I found that bringing back the cli in retint_swapgs fixed the problem,
so I traced back paths that could get there with interrupts enabled
and found int_restore_rest -> int_with_check -> retint_swapgs.
Adding a cli to int_restore_rest fixes the problem for me. I hope this
helps.

Regards
Patrick