Re: [PATCH v3 4/4] dt-bindings: audio-graph-port: add ch-maps property

[Conor Dooley <conor@kernel.org>]

[-- Attachment #1: Type: text/plain, Size: 1529 bytes --]

On Mon, Oct 16, 2023 at 12:46:29AM +0000, Kuninori Morimoto wrote:
> 
> Hi Conor
> 
> > > Some of this explanation needs to go into the binding - someone reading
> > > the binding should really be able to figure out what numbers to put in
> > > there without looking at the code.
> > 
> > Absolutely :)
> 
> Indeed :) will do in v4
> 
> > > > ch-maps = <0 0 1> means, 
> > > > 	cpu0 <-> codec0
> > > > 	cpu1 <-> codec0
> > > > 	cpu2 <-> codec1
> > 
> > What happens when you want to convey that codec0 & codec1 are both
> > connected to cpu0 & codec2 is connected to cpu1?
> > How would that be described in a DT?
> > Or is that not something anyone would even want to do?
> 
> In such case, ch-maps is from codec. it will be like below.
> It is judged by number of cpu vs codec. [PATCH 3/4] has both case sample.
> 
> 	cpu >= codec : CPU   base
> 	cpu <  codec : Codec base
> 
> 	ch-maps = <0 0 1>
> 	codec0 <-> cpu0
> 	codec1 <-> cpu0
> 	codec2 <-> cpu1

That seems like a very unintuitive interface. If there were 32 CPUs and
30 codecs, then it'd be very inconvenient for a human reader to grok the
configuration. CPUs were to be disabled in the DT, could the meaning of
the property invert?

I am not really the best when it comes to audio (or media) bindings, but
I am wondering if a phandle based approach would be better, where the
codecs have phandles for their associated CPUs. Maybe Mark, Rob etc could
comment if doing that sort of thing is not feasible.

Cheers,
Conor.


[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 228 bytes --]