Comment # 23 on bug 1184924 from Ivan Ivanov

(In reply to Torsten Duwe from comment #22)
> (In reply to Ivan Ivanov from comment #20)
> > To be on the safe side :-). 
> > 
> > Torsten, add_hwgenerator_randomness() is not supposed to be 
> > called early during kernel init, right? 
> 
> add_hwgenerator_randomness() should be callable any time, from any context.
> Of course it would be a bad idea to execute an expensive hash function in an
> interrupt context, but otherwise I see no restrictions.
> 

Yep, sure.

> > Please, see backtrace in comment #14.

But what about early kernel init. I see NULL pointer deference to 
current->set_child_tid in kthread_should_stop().