RE: [alsa-devel] [PATCH] ASoC: AMD: Configure master codec on all playback/capture cases

From: Adam Thomson
Date: Wed Feb 20 2019 - 04:30:09 EST


On 8 February 2019 18:18, Mark Brown wrote:

> On Fri, Feb 15, 2019 at 07:06:15AM +0000, Agrawal, Akshu wrote:
>
> > +static static int cz_da7219_hw_params(struct snd_pcm_substream
> *substream,
> > + struct snd_pcm_hw_params *params) {
> > + /* da7219 Codec is clock master so setup as per the needs */
> > + if (codec_dai->driver->ops->hw_params)
> > + return codec_dai->driver->ops->hw_params(substream, params,
> > + codec_dai);
> > + return 0;
> > +}
>
> No, this is a bad idea - you shouldn't be callings ops in drivers outside of the
> framework, something is going to break on you there.
> Either the drivers or the frameworks are going to get surprised by things getting
> called out of sequence, for example I'd not rely on the fact that the CODEC would
> be outputing a clock without active audio.
> This probably means that you're going to have to extend the framework in some
> way, or refactor things so that the BCLK is exported as a clock by the CODEC and
> reflect the relationships via the clock API. I need to think a bit more but I think
> the clock API is going to be the the best solution.
>
> I also suspect that some work is needed to propagate constraints between the
> various DAIs in the system.

So this was a suggestion from my side to avoid the scenario where the system
starts, the DA7219 codec hasn't yet been used (hw_params() not been called), and
one of the other audio devices is needed for playback/capture. The codec will be
at a default DAI configuration rather than the configuration that's required. If
the DA7219 codec had been used first then of course there's no problem and the
system runs as expected from then on.

This AMD platform limits all DAIs to 48KHz, 2 channel so the configuration is
fixed, as I understand it. DA7219 will output DAI clocks without data so there's
no problem there. I initially suggested using 'soc_dai_hw_params()' but the
function isn't a global symbol, probably intentionally, so for this more unusual
config I felt the above approach might be an option. Absolutely understand your
concerns though.

I considered the CCF approach but I'm not 100% confident that the clocking
mechanism of the codec will fit ideally into that framework. There's already
a clock gate for the DAI clocks, in the driver, but nothing more. The problem
here is that the BCLK rate is intrinsically linked to the WCLK rate, as well as
the word length, so changing the desired BCLK rate would affect the WCLK rate,
and vice versa. I will take a look again, but if you have any suggestions then
they'd be greatly received.