> the memory barrier is really necessary (as pointed out earlier in the list
> by Nishanth Aravamudan)?  if not, __set_current_state does the job.
> 
> I'd appreciate some notes/thoughts/tips/whatever on which conditions a
> memory barrier is really needed, or, more clearly, when the absence of a
> memory barrier breaks stuff.
> 
> TIA :)

As said in the TODO:

"...
Don't forget to set_current_state(TASK_{,UN}INTERRUPTIBLE); before
schedule*().
..."

And in this case, it is just before schedule, so I considered to change it.
Maybe the use of __set_current is for places where there is not schedule?
(Just a try :p)