[linux-sunxi] Re: [PATCH v4 3/7] clk: sunxi: add generic multi-parent bus clock gates driver

[Chen-Yu Tsai]

On Tue, Aug 9, 2016 at 2:15 AM, Jean-Francois Moine <moinejf at free.fr> wrote:
> On Mon,  8 Aug 2016 18:21:45 +0100
> Andre Przywara <andre.przywara at arm.com> wrote:
>
>> The Allwinner H3 SoC introduced bus clock gates with potentially
>> different parents per clock gate register. The H3 driver chose to
>> hardcode the actual parent clock relation in the code.
>> Add a new driver (which has the potential to drive the H3 and also
>> the simple clock gates as well) which uses the power of DT to describe
>> this relationship in an elegant and flexible way.
>> Using one subnode for every parent clock we get away with a single
>> DT compatible match, which can be used as a fallback value in the
>> actual DTs without the need to add specific compatible strings to the
>> code.  This avoids adding a new driver or function for every new SoC.
>
> The 'parent's of the bus gates are of no interest.
> They are supposed to be (no clear documentation) apb1, apb2, ahb1 and
> ahb2, but, as you well noticed in the patch 5/7, these clocks are fixed
> and have no gate. Some of them are parents of real clocks, but they
> don't bring anything to the bus gates of the other clocks.

Yes they are. Some devices, such as UARTs and I2C controllers, need
to get the clock rate of the gate and calculate the proper internal
divider.

> As I wrote previously, the simplest is to ungate/gate the clocks in
> both the bus and clock registers on clk_prepare/unprepare.
> Then, your 'multi-bus-gates' would be simply a generic 'multi-gates'.

This is somewhat misleading. What "clock" registers are you referring
to? There are no "bus" registers. The reason we call them "bus clock
gates" is because they are mashed together, instead of having clearly
separated registers for each AHB/APB bus.

And if you want just a generic clock gates driver, we already have
the "simple-gates" driver.

Regards
ChenYu