Issues using simple-audio-card driver with Xilinx Audio Formatter

[Robert Hancock]

On Fri, 2021-07-09 at 13:38 +0100, Mark Brown wrote:
> On Fri, Jul 09, 2021 at 10:16:45AM +0900, Kuninori Morimoto wrote:
> 
> > > So the next issue I'm now facing is that the MCLK to SCLK divider is not
> > > being
> > > set properly in either the Audio Formatter (MM2S Fs Multiplier register)
> > > or in
> > > the I2S Transmitter (I2S Timing Control register). The xlnx_i2s driver
> > > has a
> > > set_clkdiv function defined in its snd_soc_dai_ops structure, however
> > > that
> > > doesn't appear to be getting called. And the xlnx_formatter_pcm driver
> > > doesn't
> > > seem to have any code to set XLNX_AUD_FS_MULTIPLIER at all.
> > > In this case I have a sample rate to MCLK divider of 256, so it looks
> > > like I
> > > should add mclk-fs = <256> into the dai-link nodes in the device tree,
> > > but
> > > there will need to be some code added to the xlnx_formatter_pcm to do
> > > something
> > > with that information? And then should that driver have code to trigger
> > > the
> > > call to set_clkdiv on the CPU DAI as well?
> > Hmm... clock is one of difficult point to be generic, I guess.
> > audio-graph / audio-graph2 has customize feature in such case,
> > but simple-card doesn't.
> > 	- create generic clock handling way on simple-card ?
> > 	- add customize feature to simple-card ?
> > 	- switch to audio-graph / audio-graph2, and use customize feature ?
> > Thank you for your help !!
> 
> For something like this I think the driver should be able to figure out
> the ratio based on the configured MCLK and sample rate.  For the most
> part set_clkdiv() should be a legacy thing, it's very manual and hard to
> see why a system would do something different to the obvious ratio
> usually.

Possibly the I2S transmitter should be implementing set_sysclk rather than
set_clkdiv then? simple-audio-card seems like it would already propagate that
through into the CPU DAI in asoc_simple_hw_params and then it could figure out
the right divider value to use.

The tricky part is that the Audio Formatter (used as the "plat" component here)
also needs to know what the mclk-fs value is. (I really don't know why it
cares, I would think it would just push data to the output stream as fast as it
was accepted, but indeed it does have a register to set the sample rate to MCLK
divider, and if it's not set properly the I2S transmitter downstream seems to
constantly get underruns.) I'm not sure if there's any mechanism for it to
determine the value right now, or if something new would need to be added?

Our actual product using this Xilinx logic is only using capture functionality
- the playback side (which is where these issues come in) is just being used as
a loopback to the capture side for testing purposes - but I can poke at this
further if there's some spare time..

-- 
Robert Hancock
Senior Hardware Designer, Calian Advanced Technologies
www.calian.com