I had started the removal of semaphores in this driver without knowing
that Nicolas Saenz Julienne also worked on this. In case of the "remote
event" infrastructure, my solution seemed significantly better, so I'm
proposing this as a change on top.

The problem with using either semaphores or completions here is that
it's an overly complex way of waking up a thread, and it looks like the
'count' of the semaphore can easily get out of sync, even though I found
it hard to come up with a specific example.

Changing it to a 'wait_queue_head_t' instead of a completion simplifies
this by letting us wait directly on the 'event->fired' variable that is
set by the videocore.

Another simplification is passing the wait queue directly into the helper
functions instead of going through the fragile logic of recording the
offset inside of a structure as part of a shared memory variable. This
also avoids one uncached memory read and should be faster.

Note that I'm changing it back to 'killable' after the previous patch
changed 'killable' to 'interruptible', apparently based on a misunderstanding
of the subtle down_interruptible() macro override in vchiq_killable.h.

Fixes: f27e47bc ("staging: vchiq: use completions instead of semaphores")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>