[PATCH 2/6] pinctrl: Update clock handling for the pinctrl-nomadik GPIO driver

[Linus Walleij]

On Thu, Oct 25, 2012 at 2:41 PM, Linus Walleij <linus.walleij at linaro.org> wrote:
> On Wed, Oct 24, 2012 at 4:45 PM, Lee Jones <lee.jones at linaro.org> wrote:
>
>> The clock framework has changed somewhat and it's now better to
>> invoke clock_prepare_enable() and clk_disable_unprepare() rather
>> than the legacy clk_enable() and clk_disable() calls. This patch
>> converts the Nomadik Pin Control driver to the new framework.
>>
>> Signed-off-by: Lee Jones <lee.jones at linaro.org>
>
> I was convinced that this is a good change but no regression,
> so applied to the devel branch for 3.8.
>
> I also removed the initial clk_prepare() so the reference count
> may actually go down to 0 for the GPIO block and the peripheral
> cluster eventually gets relaxed.

Famous last words!

The good news is that this actually works, and the refcount
*does* go down to zero and gate off entire peripheral
clusters.

However that was not good because something vital in
some peripheral cluster died and killed the system :-D

Lee, could to to track down the reason and fix it so the patch
can be applied?

The only thing you need to do is to remove the superfluous
clk_prepare() right after the devm_clk_get() that hogs each
peripheral cluster.

Probably some driver is needing a clk_get() or a clk_get_sys() is
needs to be added somewhere to bring up some vital cluster,
or there may be some out-of-tree driver needed to bring up the
cluster properly I have no clue... Maybe some cluster just
cannot be declocked like that.

Yours,
Linus Walleij