Re: [RFC PATCH 1/2] soundwire: add support for static port mapping

[Srinivas Kandagatla]

Thanks Pierre for your inputs,

On 20/01/2021 22:15, Pierre-Louis Bossart wrote:
> On 1/20/21 12:01 PM, Srinivas Kandagatla wrote:
> > Some of the soundwire controllers can have static functions assigned
> > to each port, like some ports can only do PCM or PDM. This is the 
> > situation
> > with some of the Qualcomm Controllers.
> >
> > In such cases its not correct to assign/map any free port on master
> > during streaming.
> >
> > So, this patch provides a way to pass mapped port number along
> > with the port config, so that master can assign correct ports based
> > on the provided static mapping.
>
> I am not sure I understand the problem or what's different between Intel 
> and Qualcomm.
>
> On the Intel side we also have fixed-function ports, some for PDM and 
> some for PCM. They are not interchangeable, and they are also dedicated 
> for each link.
That is good to know!

> We don't dynamically allocate ports on the master side, the mapping is 
> defined by the dai->id and is static. There is a 1:1 relationship 
> between dai->id and port_number. See intel_register_dai() and 
> intel_hw_params() in drivers/soundwire/intel.c
>
> In the machine driver we make use of specific master DAIs in the dailink 
> definitions, just like regular ASoC solutions, so which DAIs you use in 
> the machine driver defines what ports end-up being used. There is 
> nothing fancy or dynamic here, the dai/port allocation is defined by the 
> dailinks. This is a static/worst-case allocation, we don't reassign 
> ports depending on use-cases, etc.
>
> The only thing that is dynamic is that the programming of each port is 
> handled based on the bandwidth needs of that port, i.e if you play 16 or 
> 24 bits you'd get fewer or more bitSlots allocated to that dai/port, and 
> the DPn registers are updated if you have concurrent streaming on other 
> ports. If you only have a fixed set of payloads, as in the existing 
> amplifier cases, you can hard-code this allocation as well.
Yes, it will work for the existing WSA881x amplifier case.

Am preparing patches for a new QCOM codec driver WCD938x (TX and RX) 
connected via Soundwire,

Port allocations are something like this:

RX: (Simple)
Port 1 -> HPH L/R
Port 2 -> CLASS H Amp
Port 3 -> COMP
Port 4 -> DSD.

TX: (This get bit more complicated)
Port 1: PCM
Port 2: ADC 1 & 2
Port 3: ADC 3 & 4
Port 4: DMIC-0, DMIC-1, DIMC-2 , DMIC-3 and MBHC
Port 5: DMIC-4, DMIC-5, DMIC-6 and DMIC-7

We handle the port allocation dynamically based on mixer and dapm 
widgets in my code! Also channel allocations are different for each 
function!

> Does this help and can you align on what Intel started with?

Firstly, This is where the issue comes, if we go with the 
suggested(dai->id) solution, we would end up with a long list of 
dai-links with different combinations of both inputs/output connections 
and usecases. Again we have to deal with limited DSP resources too!

Secondly, The check [1] in stream.c will not allow more than one master 
port config to be added to master runtime. Ex: RX Port 1, 2, 3 is used 
for Headset Playback.

But if we have a static mapping table of the ports then this will 
provide more flexibility to codec driver! And we can dynamically select 
ports based on the usecase or active dapm path!

--srini

[1]: 
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/soundwire/stream.c?h=v5.11-rc4#n1294
>