Re: [patch] scheduler fixes, 2.5.32-BK

[Ingo Molnar]

we used to have the global semaphore_lock - which, if used separately from
the waitqueue lock, indeed can cause the unuse of the semaphore structure
before the spin_unlock in wakeup() completes.

but since 2.5.25 or so we use the semaphore waitqueue's spinlock for
semaphore locking - this also neatly solves the semaphore-unuse problem.  
Four architectures, sparc, ia64, arm and x86-64 still use the global
semaphore_lock, but the other 13 architectures use the waitqueue spinlock
already.

(unless there's something else i missed.)

	Ingo