Re: [PATCH RFC 6/6] arm64: dts: qcom: sdm845-google: Add basic audio support

[David Heidelberg]

On 06/07/2026 14:00, Konrad Dybcio wrote:
> On 7/5/26 10:06 PM, David Heidelberg via B4 Relay wrote:
> > From: David Heidelberg <david@xxxxxxx>
> >
> > Introduce support for sound card and wire two CS35L36 audio codecs for
> > top and bottom speakers.
> >
> > Inspired by commit from Joel Selvaraj.
> >
> > Signed-off-by: David Heidelberg <david@xxxxxxx>
> > ---
>
> [...]
>
> > +	/* CS35L36, Bottom Speaker */
> > +	cs35l36_bottom: audio-codec@40 {
> > +		compatible = "cirrus,cs35l36";
> > +		reg = <0x40>;
> > +		reset-gpios = <&tlmm 112 GPIO_ACTIVE_HIGH>;
> > +		#sound-dai-cells = <0>;
> > +		sound-name-prefix = "Right"; /* Bottom */
>
> One comment's enough (either above the node or here)
>
> [...]
>
> > +	mm1-dai-link {
> > +		link-name = "MultiMedia1";
> > +
> > +		cpu {
> > +			sound-dai = <&q6asmdai MSM_FRONTEND_DAI_MULTIMEDIA1>;
> > +		};
> > +	};
> > +
> > +	cs35l36-dai-link {
> > +		link-name = "Speaker Playback";
> > +
> > +		cpu {
> > +			sound-dai = <&q6afedai QUATERNARY_TDM_RX_0>;
> > +		};
> > +
> > +		platform {
> > +			sound-dai = <&q6routing>;
> > +		};
> > +
> > +		codec {
> > +			sound-dai = <&cs35l36_bottom>, <&cs35l36_top>;
> > +		};
>
> nit: 'co'dec < 'cp'u < 'p'latform
>
> also 'c's35l36-dai-link < `m`m1-dai-link

... aaaand **sorting nodes** broke audio.

Moving `mm1-dai-link` behind `cs35l36-dai-link` broke the audio.

For now I'm keeping m1-dai-link at the begging and added the comment why is it 
there, but I guess would be nice to have this solved properly.

---
  What actually happens in the core. In soc_probe_component() the ASoC core 
creates a component's widgets, probes it, and immediately adds that component's own
  routes. Card-level routes, including the DT audio-routing ones, are only 
added after every component is probed. So a component whose routes point at another
  component's widgets, which is exactly what q6routing does with MM_DL* and 
MM_UL*, only works if the other component happens to be probed earlier. Probe 
order is
  the DT link order, so this becomes a hidden DT contract.

  Upstream does not actually sort these nodes. The binding matches .*-dai-link$ 
and the reference boards rely on the order: db845c lists mm1 to mm4 and then
  hdmi-dai-link, sm8250-mtp lists mm1 to mm3 and then wcd-..., wsa-..., 
va-dai-link. Neither is alphabetical. So "sorted by name" is not the convention 
being
  followed for these children, and cs35l36-dai-link sorting before mm1-dai-link 
is precisely why an alphabetical rework breaks.

  Where a fix would belong. Two options, in order of preference:

  - In the ASoC core: defer component routes until after 
soc_probe_link_components() finishes, next to where the card routes are added. 
That removes the dependency
    for every driver with cross-component routes. It is a small change, but it 
touches every card, so it needs a wide audience on alsa-devel.
  - In qcom_snd_parse_of(): it already knows which links are frontends (no 
codec and no platform child, marked dynamic) and which are backends (no_pcm). 
Two passes
    over the children, frontends first, make the order independent of the DT. 
That is local and would be acceptable to the Qualcomm audio maintainers, but it 
papers
    over a core problem.

> otherwise:
>
> Reviewed-by: Konrad Dybcio <konrad.dybcio@xxxxxxxxxxxxxxxx>
>   
> Konrad