Re: [PATCH v6 03/12] dt-bindings: arm: mediatek: sgmiisys: Convert to DT schema

[Rob Herring <robh@kernel.org>]

On Wed, Feb 15, 2023 at 3:16 PM Russell King (Oracle)
<linux@armlinux.org.uk> wrote:
>
> On Wed, Feb 15, 2023 at 02:43:18PM -0600, Rob Herring wrote:
> > On Mon, Feb 13, 2023 at 09:34:43PM +0000, Daniel Golle wrote:
> > > Convert mediatek,sgmiiisys bindings to DT schema format.
> > > Add maintainer Matthias Brugger, no maintainers were listed in the
> > > original documentation.
> > > As this node is also referenced by the Ethernet controller and used
> > > as SGMII PCS add this fact to the description.
> > >
> > > Signed-off-by: Daniel Golle <daniel@makrotopia.org>
> > > ---
> > >  .../arm/mediatek/mediatek,sgmiisys.txt        | 27 ----------
> > >  .../arm/mediatek/mediatek,sgmiisys.yaml       | 49 +++++++++++++++++++
> > >  2 files changed, 49 insertions(+), 27 deletions(-)
> > >  delete mode 100644 Documentation/devicetree/bindings/arm/mediatek/mediatek,sgmiisys.txt
> > >  create mode 100644 Documentation/devicetree/bindings/arm/mediatek/mediatek,sgmiisys.yaml
> >
> > If you respin or as a follow-up, can you move this to bindings/clock/?
>
> I'm not sure that's appropriate. Let's take the MT7622 as an example,
> here is the extract from the device tree for this:
>
>         sgmiisys: sgmiisys@1b128000 {
>                 compatible = "mediatek,mt7622-sgmiisys",
>                              "syscon";
>                 reg = <0 0x1b128000 0 0x3000>;
>                 #clock-cells = <1>;
>         };
>
> This makes it look primarily like a clock controller, but when I look
> at the MT7622 documentation, this region is described as the
> "Serial Gigabit Media Independent Interface".
>
> If we delve a little deeper and look at the code we have in the kernel,
> yes, there is a clock driver, but there is also the SGMII code which is
> wrapped up into the mtk_eth_soc driver - and the only user of the
> clocks provided by the sgmiisys is the ethernet driver.
>
> To me, this looks very much like a case of "lets use the clock API
> because it says we have clocks inside this module" followed by "now
> how can we make it work with DT with a separate clock driver".
>
> In other words, I believe that describing this hardware as something
> that is primarily to do with clocks is wrong. It looks to me more
> like the hardware is primarily a PCS that happens to provide some
> clocks to the ethernet subsystem that is attached to it.
>
> Why do I say this? There are 23 documented PCS registers in the
> 0x1b128000 block, and there is one single register which has a bunch
> of bits that enable the various clocks that is used by its clock
> driver.
>
> Hence, I put forward that:
>
> "The MediaTek SGMIISYS controller provides various clocks to the system."
>
> is quite misleading, and it should be described as:

Indeed I was...

>
> "The MediaTek SGMIISYS controller provides a SGMII PCS and some clocks
> to the ethernet subsystem to which it is attached."

+1

> and a PCS providing clocks to the ethernet subsystem is nothing
> really new - we just don't use the clk API to describe them, and
> thus don't normally need to throw a syscon thing in there to share
> the register space between two drivers.

Humm, yes. Just like phys that provide clocks.

If PCS is the main function, then it should go in the PCS directory:
bindings/net/pcs/

> So, in summary, I don't think moving this to "bindings/clock/" makes
> any sense what so ever, and that is probably being based on a
> misleading description of what this hardware is and the code structure
> adopted in the kernel.
>
> Yes, DT describes the hardware. That's exactly the point I'm making.
> It seems that the decision here to classify it has a clock driver is
> being made based off the kernel implementation, not what the hardware
> actually is.

Right. I'm just trying to get misc blocks out of bindings/arm/.

Rob