Hi Alex,
On Thu, Mar 05, 2020 at 03:36:28PM +0100, Alex Riesen wrote:
> Geert Uytterhoeven, Mon, Mar 02, 2020 17:13:30 +0100:
> > On Mon, Mar 2, 2020 at 5:09 PM Alex Riesen <alexander.riesen@xxxxxxxxxxx> wrote:
> > > Geert Uytterhoeven, Mon, Mar 02, 2020 16:32:32 +0100:
> > > >
> > > > The #clock-cells should be in the main video-receiver node.
> > > > Probably there is more than one clock output, so #clock-cells may be 1?
> > >
> > > AFAICS, the device can provide only this one clock line (audio master clock
> > > for I2S output)... I shall re-check, just in case.
>
> And you're right, of course: the audio output formatting module of the ADV748x
> devices provides a set of clock lines related to the I2S pins: the already
> discussed master clock, left-right channel clock and the serial clock (bit
> clock?).
I don't think we need to model the last two clocks through CCF though,
they're part of the I2S protocol, not clock sources that need to be
explicitly controlled (or queried).
> > > > There is no need for a fixed-clock compatible, nor for clock-frequency
> > > > and clock-output-names.
> > > >
> > > > But most important: this should be documented in the adv748x DT bindings,
> > > > and implemented in the adv748x driver.
> > >
> > > So if the driver is to export that clock for the kernel (like in this case),
> > > it must implement its support?
> >
> > Exactly. Unless that pin is hardcoded to output a fixed clock, in which case
> > you can just override the existing audio_clk_c rate.
>
> Just to try it out (I'll set #clock-cells to 1), I registered a fixed rate
> clock in the driver, added a clock provider:
>
> adv748x_probe:
>
> clk = clk_register_fixed_rate(state->dev,
> "clk-hdmi-i2s-mclk",
> NULL /* parent_name */,
> 0 /* flags */,
> 12288000 /* rate */);
> of_clk_add_provider(state->dev->of_node, of_clk_src_simple_get, clk);
>
> And removed the audio_clk_c frequency setting. I also replaced the audio_clk_c
> in the list of input clocks of the R-Car-side sound card with the phandle of
> the adv7482 main node:
>
> salvator-common.dtsi:
>
> &i2c4 {
> status = "okay";
>
> adv7482_hdmi_decoder: video-receiver@70 {
> #clock-cells = <0>; // to be replaced with <1>
> };
> };
>
> &rcar_sound {
> clocks = ..., <&adv7482_hdmi_decoder>, ...;
> };
>
> As everything continues to work as before, I assume that at least the clock
> dependencies were resolved.
This looks good to me.
> Is there a way to verify that the added input clock is actually used?
> IOW, if its frequency is actually has been programmed into the ssi4 (R-Car
> receiving hardware) registers, and not just a left-over from previuos attempts
> or plain default setting?
>
> As the ADV748x devices seem to provide also the clocks for video outputs, will
> it make any sense to place the clock definition into the port node?
> Or should all provided clocks be indexed in the main device node?
Those clocks are part of the CSI-2 protocol and also don't need to be
explicitly controlled. As far as I can tell from a quick check of the
ADV7482 documentation, only the I2S MCLK is a general-purpose clock that
needs to be exposed.
--
Regards,
Laurent Pinchart
