Re: [PATCH v5 0/3] Amlogic 32-bit Meson SoC SDHC MMC controller driver

[Ulf Hansson]

Jerome, Martin,

On Mon, 27 Apr 2020 at 18:46, Jerome Brunet <jbrunet@xxxxxxxxxxxx> wrote:
>
>
> On Mon 27 Apr 2020 at 18:23, Martin Blumenstingl <martin.blumenstingl@xxxxxxxxxxxxxx> wrote:
>
> > Hi Jerome,
> >
> > On Mon, Apr 27, 2020 at 10:56 AM Jerome Brunet <jbrunet@xxxxxxxxxxxx> wrote:
> > [...]
> >> > Changes since v3 at [3]:
> >> > - split the clock bits into a separate clock controller driver because
> >> >   of two reasons: 1) it keeps the MMC controller driver mostly clean of
> >> >   the clock bits
> >>
> >> If the register is in the MMC controller register space and the MMC
> >> driver is the driver using these clocks, it is where the clocks belong.
> >> I don't get why it could be an issue ?
> >>
> >> Is the clock block is shared with another device, like on the Gx family ?
> > no, it is not shared with another device (to my knowledge).
> >
> >> > 2) the pure clock controller can use
> >> >   devm_clk_hw_register() (instead of devm_clk_register(), which is
> >> >   deprecated) and the MMC controller can act as a pure clock consumer.
> >>
> >> Why can't you use devm_clk_hw_register in an MMC driver ?
> >> Unless I missed something, it is provided by clk-provider.h, which can be
> >> included by any driver.
> > indeed, I could use devm_clk_hw_register in the MMC driver.
> > Ulfs concern was that a lot of code was needed for managing the clocks
> > and I agree with him. so this is my way of keeping those details away
> > from the MMC driver and have two separate drivers which are better to
> > understand overall.
>
> Martin, Ulf,
>
> I understand that CCF code might seems verbose and I'm happy to help
> review it if necessary but I don't think every driver out there should
> register some kind of fake clock controller driver everytime they wish
> to use CCF API.
>
> Yes the it might make the driver code cleaner but the overall
> architecture is harder to follow.
>
> CCF was made so driver from any subsystem *may* use it. Creating a
> controller for a single register is overkill. The HW architecture of
> this particular device does not justify it.

I fully understand your point and I agree with it.

If I recall correctly, my point in the earlier review phase was that I
wanted the driver to be nicely split into a clock provider part and
into a mmc host driver part. I also raised the point of using
devm_clk_hw_register() rather than the deprecated devm_clk_register().
I still think this makes sense.

That said, perhaps a reasonable split could be to have two separate
c-files (one for clock provider and one for mmc host), but both in the
mmc subsystem.

Kind regards
Uffe