On Sun, 2016-07-17 at 09:11:10 -0700, Sören Brinkmann wrote:
> On Sun, 2016-07-17 at 00:15:23 +0900, Masahiro Yamada wrote:
> > The "clock-output-names" property is useful for generic clock
> > providers such as fixed-clock, fixed-factor-clock, etc.
> >
> > On the other hand, it should not be used for really SoC-specific
> > clock providers like this one. As you see in "enum zynq_clk" in
> > this driver, it is written as if it already knows all the clock
> > names. Besides, this is instantiated only once, so no clock name
> > conflict would happen even if the clock names are hard-coded in the
> > driver.
> >
> > The device tree (arch/arm/boot/dts/zynq-7000.dtsi) will be fixed
> > later.
> >
> > Signed-off-by: Masahiro Yamada <yamada.masah...@socionext.com>
>
> IIRC, this was the only way to allow circular clock routing. E.g. there
> are use-cases that route an FCLK back into the PS to use it as input for
> the GEM. With the introduction of deferred probing this might all be
> possible now but it would be good to verify that this kind of circular
> clock routing still works with this change.
Yep, this breaks this kind of routing. E.g. if you remove the
clk-output-names from the DT and add an EMIO clock input like this:
clocks = <&clkc 15>;
clock-names = "gem1_emio_clk";
the code will no longer be able to correctly obtain the clock name for that
input.
Sören
