This reverts commit 928686b7.

The patch was an optimization of the old futex wake code where we woke
the waiter and then set q->lock_ptr to NULL. When the waiter preempted
the waker then we run into lock contention on q->lock_ptr
aka. hb->lock.

commit f1a11e (futex: remove the wait queue) changes the wakeup logic
by setting q->lock_ptr to NULL _before_ waking the task. It keeps a
reference on the task struct of the to be woken task to avoid an exit
race.

The combination of both patches resulted in different race on -RT:

    A is blocked on futex
    B calls futex_wake
    B sets q(A)->lock_ptr to NULL and puts A on the wake list
    B is preempted
    ...
    A wakes up (e.g. timer, signal)
    A detects q->lock_ptr = NULL and returns
    A waits on a different futex

    B is scheduled back in
    B wakes A
    A sees a spurious wake up
Reported-by: Blaise Gassend <blaise@willowgarage.com>
Debugged-by: Darren Hart <dvhltc@us.ibm.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

 enter the commit message for your changes. Lines starting