Pavel Machek [off-list ref] writes:

quoted

For mobile devices it is not acceptable to filter events away at some
upper SW layer depending on the system state. The HW which generates
those events may not generate events at all to allow longer CPU sleep
periods.

Ok.

quoted

In ideal world it would be nice to control device states based on for
example user count. However, there are several listeners for input
devices and it is hard or impossible to have them all to follow overall
state transition (screen blanked etc.). Instead, there is some
system

So you have mobile device; why is it impossible to just close the
device when you do not want the events? I guess it is hard for generic
distros, but on your phone, you should be able to modify Xserver to
close touchscreen/keypad device when it is not needed... right?

We'd had this discussion before... cf. eg.
http://thread.gmane.org/gmane.linux.kernel.input/9266/focus=9767

The problem was that several processes may have a device open, while
another process should be able to control the state of the device.
Maybe this could be solved by making the controlling process a proxy,
and having all "user" processes going though it.  Then the controlling
(proxy) process could open/close the device as it wants, letting the
runtime PM do its job.  But this would mean duplicating some kernel
functionality (at least multiplexing) in user space.
-- 
Regards,
Feri.