On Fri, 27 Feb 2009 15:20:56 -0800
David Brownell [off-list ref] wrote:

On Friday 27 February 2009, Andrew Morton wrote:

quoted

quoted

quoted

What has gone wrong here?

Not much I can see. __It's registering a platform_device,
but only if it could be used on this system.

Again, that all gets fixed if this is done the right way around. __Run
your probe function. __If the hardware is there, register with the core
and all the nodes appear. __If the hardware is not present: bale.

Right.  If there's no button hooked up to its "power button"
signal, don't register that "twl4030-pwrbutton" node.

quoted

The design of the whole subsystem appears to be upside down :(

You seem to be missing something.  Is it only the lack of that
tweak Felipe sent?

I generally just delete unchangelogged patches.

<goes back and finds it>


That doesn't address the problem at all.  A function called
"add_children" just shouldn't exist.  The general kernel design is for
client drivers to register themselves with the core, so the core does
not have any hard-wired knowledge of any client drivers.

IOW, twl4030_pwrbutton_probe() should call into twl4030-core,
registering the powerbutton driver.