Re: [PATCH v4 1/4] dt-bindings: media: imx258: add bindings for IMX258 sensor

[Krzysztof Kozlowski <krzk@kernel.org>]

On Tue, Sep 29, 2020 at 02:02:55PM +0300, Sakari Ailus wrote:
> On Tue, Sep 29, 2020 at 11:46:36AM +0200, Krzysztof Kozlowski wrote:
> > On Tue, Sep 29, 2020 at 12:40:46PM +0300, Sakari Ailus wrote:
> > > On Tue, Sep 29, 2020 at 11:18:46AM +0200, Krzysztof Kozlowski wrote:
> > > > On Tue, 29 Sep 2020 at 11:15, Sakari Ailus <sakari.ailus@linux.intel.com> wrote:
> > > > >
> > > > > Hi Krzysztof,
> > > > >
> > > > > On Wed, Sep 23, 2020 at 05:21:26PM +0200, Krzysztof Kozlowski wrote:
> > > > > > Add bindings for the IMX258 camera sensor.  The bindings, just like the
> > > > > > driver, are quite limited, e.g. do not support regulator supplies.
> > > > > >
> > > > > > Signed-off-by: Krzysztof Kozlowski <krzk@kernel.org>
> > > > > >
> > > > > > ---
> > > > > >
> > > > > > Changes since v3:
> > > > > > 1. Document also two lane setup.
> > > > > >
> > > > > > Changes since v2:
> > > > > > 1. Remove clock-frequency, add reset GPIOs, add supplies.
> > > > >
> > > > > Oops. I missed this one.
> > > > >
> > > > > How does the driver know the appropriate clock frequency for the platform
> > > > > if it's not in DT? The sensor supports a range of frequencies, not a single
> > > > > frequency.
> > > > >
> > > > > Could you add clock-frequency back?
> > > > 
> > > > Not really, it was removed on Rob's request. The bindings do not
> > > > describe driver's behavior so how the driver gets frequency should not
> > > > be part of the bindings. Also it's not a real problem - the driver
> > > > just calls clk_get_rate().
> > > 
> > > How is the rate determined? I mean, many ISPs or CSI-2 receivers that
> > > provide the clock are also capable of using a variety of frequencies. But
> > > only one can be used on the platform in general.
> > 
> > Having "clock-frequency" property in DTS did not solve that. It has no
> > effect on actual frequency.
> 
> It's up to the driver to do what's needed, yes.
> 
> Please see examples in e.g. drivers/media/i2c/ov8856.c and
> Documentation/devicetree/bindings/media/i2c/ov8856.yaml .

It seems the ov8856 driver uses this property in different way than
imx258 driver. It is more appropriate and quite similar to clock
providers and buses - to set the desired frequency on input clock.

Therefore what is the point of using this DT property comparing to
assigned-clock-rates?

It's the same. So let's use generic (already documented)
assigned-clock-rates.  

For your question (not related to the bindings but to driver
implementation): "How is the rate determined?", simple: clk_get_rate.
The driver then uses it like this:
	if (clk_get_rate() != only_working_configuration_hz)
		return -EINVAL;

From the bindings point of view, the clock can be anything within a
range of sensor's accepted values. The clock frequency is a property of
a clock, not of a sensor. Therefore for HW description it should be
described in the clock bindings, not in the sensor bindings.

To summarize, the "clock-frequency" property of sensor node:
1. As a way to configure the clock should be replaced with
   assigned-clock properties,
2. As a way to describe the hardware is simply invalid. It is not a HW
   description, because HW requires just a clock of frequency within
   given range, not a fixed frequency clock.

Consider the example:
        camera@1a {
                compatible = "sony,imx258";
                reg = <0x1a>;

                clocks = <&imx258_clk>;
                clock-names = "clk";

                /* Oscillator on camera board */
                imx258_clk: clk {
                        compatible = "fixed-clock";
                        #clock-cells = <0>;
                        clock-frequency = <19200000>;
                };

                port {
			...
                };
        };

What is the point to add "clock-frequency" property to the camera
itself, since it is already clearly defined by the clock?

Or another example:

        camera@1a {
                compatible = "sony,imx258";
                reg = <0x1a>;

                clocks = <&iclk 0>;
                clock-names = "clk";
		assigned-clocks = <&clk 0>;
		assigned-clock-rates = <19200000>;

                port {
			...
                };
        };

Again, no reason for artificial clock-frequency property. It is not part
of HW description of the sensor.

Best regards,
Krzysztof