On 11/22/2012 06:52 AM, Tomi Valkeinen wrote:

On 2012-11-22 02:20, Ricardo Neri wrote:

...

Hi Tomi, Mark,

On 11/19/2012 07:15 PM, Mark Brown wrote:

...

On Mon, Nov 19, 2012 at 02:58:41PM +0200, Tomi Valkeinen wrote:

...

I still don't understand why the codec and machine drivers need to be created in the board file. That just forces us to replicate the same code for all OMAP boards that have OMAP HDMI output. Why not create the devices in some common code, for example arch/arm/mach-omap2/display.c?

Yes, this would be more sensible if there's no board specifics involved.

I think they are truly board-specific. For instance, there could be some

I don't =).

...

board that does not have the OMAP HDMI IP wired to an external connector. We don't want the drivers to be probed in that case. If they are in common code, the devices will be created even if a board does not have HDMI output.

The HDMI devices are still there in the HW even if we don't have a HDMI connector. I don't see any problem with probing the HDMI audio driver even in that case.

But of course the user space shouldn't see a usable HDMI display/audio if there's no connector. For display side this is managed so that the HDMI IP driver is always loaded, but a HDMI panel driver is only there if the board file tells that we have a connector.

I guess this could be optimized by having a "disabled" flag for HDMI IP driver, so that it wouldn't even need to allocate any private data structures of such. But the savings would be quite minimal.

Maybe, just create the devices for ASoC only if it sees a HDMI dss device in the omapdss_board_info?

...

...

...

With DT this should be similar: OMAP's hdmi devices should be presented in the omap4.dtsi file, not in each individual board dts. Although the DT data should represent the hardware, and if the code and machine devices are not really there in the HW, then... I don't know =).

Well, in a case like this where the sound card is essentially autoprobed based on the detection of the hardware at runtime the sound card probably shouldn't appear in the device tree at all - you'll probably want something to say there's a physical HDMI port it's worth looking at there but everything else should be figured out at runtime.

Yes, I was planning to rely on the DSS DT entries in the omap4.dtsi file. However, no HDMI audio support should be probed if the board does not have an HDMI connector. Also, the TPD chip should appear on the Pandaboard/SDP4430 dts files. Only if both conditions are met, probe the HDMI audio drivers, this conditions will be checked at run time by the ASoC HDMI machine driver.

Something like this:

 sound_hdmi {
     compatible = "ti,omap-hdmi-card-audio";
     ti,model = "OMAP4HDMI";

     ti,hdmi_audio = <&hdmi>;
     ti,level_shifter = <&tpd12s015>;
 };

The ASoC machine driver would create the platform device for the HDMI codec if the DT has the required nodes.

The TPD chip really shouldn't be here in. It's an external component, not related to audio in any way. I think the HDMI audio driver should be only concerned about the HDMI IP.

OK. So display code will handle all the DT details for the audio drivers to use.

The HDMI IP video driver shouldn't care about TPD chip either, but for now we need to manage it somewhere, and that's the easiest place for it.

BTW, I have a draft for this, but more urgent things have been consuming my bandwidth. :/

BR,

Ricardo

So... I'm not sure how this should be managed, but I am 99% sure that there's nothing board specific with HDMI audio, and thus it should be managed in common .c files in arch code or dss code, or .dts files. If we add the hdmi display in the .dts files, I think the audio should just work.

Or is there something in ASoC that forces us to represent it in the .dts? I don't think there's really anything related to HW to describe there related to HDMI audio. If we have HDMI video output, we also have the audio, as simple as that.

Tomi