Re: meson-aiu: HDMI codec .prepare() callback not called

[Jerome Brunet <jbrunet@baylibre.com>]

On Tue 31 Dec 2024 at 19:44, Martin Blumenstingl <martin.blumenstingl@googlemail.com> wrote:

> Hi Jerome,
>
> I am still working on a HDMI controller driver for Meson8/8b/8m2.
> HDMI output would be incomplete without audio.
> There's a great series from Dmitry [0] which simplifies the HDMI
> controller driver implementation by moving all the hdmi-codec related
> management to a generic framework.
>
> I tried integrating Dmitry's work [0] into my HDMI controller driver.
> While testing I found that hdmi-codec's .prepare() callback is not
> called at all and asked Dmitry for help [1]. All other callbacks
> (.hw_params, .startup, .shutdown, ...) however are called.
> On his board (Qualcomm SDM845, sdm845-db845c.dts) hdmi_codec_prepare()
> is called by snd_soc_pcm_dai_prepare() (core sound framework, not
> platform specific).
> However, on my Odroid-C1 this is not happening.
>
> Looking further snd_soc_pcm_dai_prepare() I can see that
> for_each_rtd_dais() has three entries:
>   snd_soc_pcm_dai_prepare(), dai name=I2S Encoder, id=2
>   snd_soc_pcm_dai_prepare(), dai name=CODEC CTRL HDMI I2S IN, id=0
>   snd_soc_pcm_dai_prepare(), dai name=I2S FIFO, id=0
> What I'm missing here is a dai name "i2s-hifi" (from
> sound/soc/codecs/hdmi-codec.c, see hdmi_i2s_dai)
>
> My hdmi_tx node looks like this (abbreviated):
>   hdmi_tx: hdmi-tx@42000 {
>       compatible = "amlogic,meson8b-hdmi-tx";
>       reg = <0x42000 0xc>;
>       #sound-dai-cells = <1>;
>        ...
>   };
>
> Then I have a "amlogic,gx-sound-card" instance with the following dai-links:
>   dai-link-0 {
>       sound-dai = <&aiu AIU_CPU CPU_I2S_FIFO>;
>   };
>   dai-link-1 {
>       sound-dai = <&aiu AIU_CPU CPU_I2S_ENCODER>;
>       dai-format = "i2s";
>       mclk-fs = <256>;
>
>       codec-0 {
>           sound-dai = <&aiu AIU_HDMI CTRL_I2S>;
>     };
>   };
>   dai-link-2 {
>       sound-dai = <&aiu AIU_HDMI CTRL_OUT>;
>
>       codec-0 {
>           sound-dai = <&hdmi_tx 0>;
>       };
>   };
> So apart from the additional cell in the sound-dai towards hdmi_tx
> (Meson8/8b/8m2's HDMI controller has two inputs: I2C and SPDIF. I2C is
> the first one, hence &hdmi_tx 0) this is identical to
> arch/arm64/boot/dts/amlogic/meson-gxbb-odroidc2.dts.
>
> I have further verified that the gx-card parsing does find the HDMi
> controller and links it correctly.
> To me it's odd that only the .prepare() callback is not called, all
> others (as mentioned above: .hw_params, .startup, ...) are working
> fine.

I think the problem you are seeing comes from the quirk of
codec-to-codec links. The hdmi codec link is such a link on Amlogic
because further digital routing is required after the backend.

Those type of links are not used much beside some
CPU offloading on Samsung and Amlogic, as far as I know.
It is possible, even likely, that things are still missing there.

So those C2C links are operated by the DAPM events, not the regualar
ASoC code. You can start here:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/sound/soc/soc-dapm.c#n3995

You'll see that .prepare() is not called, same as .trigger()
That should propably be fixed :/

Side Note:
While this type of link is not used much, I think that - fully
implemented - it could hold the key to proper DAI chaining ... possibly
allowing to get rid of DPCM.

>
> I'm not familiar with the sound stack and I'm hoping that you have
> some ideas on how to either debug or fix it.
>
>
> Thanks in advance and best regards,
> Martin
>
>
> [0] https://lore.kernel.org/dri-devel/20241224-drm-bridge-hdmi-connector-v10-0-dc89577cd438@linaro.org/
> [1] https://lore.kernel.org/dri-devel/20241231004311.2574720-1-martin.blumenstingl@googlemail.com/
> [2] https://lore.kernel.org/dri-devel/l3u3wtnxyhrwjynevkwfjwarisc4yt4xy2rbzf5kb7k5l5dw3n@lxqtimymyjg6/
>
> [2. text/x-patch; sound-soc-soc-dai-snd_soc_pcm_dai_prepare-info.diff]...

-- 
Jerome