On Wed, Sep 13, 2017 at 10:02:20AM +0200, Arnaud Mouiche wrote:

quoted

Could you please give me a few set of examples of how you set
set_sysclk(), set_tdm_slot() with the current driver? The idea
here is to figure out a way to calculate the bclk in hw_params
without getting set_sysclk() involved any more.

Here is one, where a bclk = 4*16*fs is expected

In another setup, there are 8 x 16 bits slots, whatever the number
of active channels is.
In this case bclk = 128 * fs
The number of slots is completely arbitrary. Some slots can even be
reserved for communication between codecs that don't communicate
with linux.

In summary, bclk = sample rate * slots * slot_width;

I will update my patch soon.

quoted

Unfortunately, it looks like a work around to me. I understand
the idea of leaving set_sysclk() out there to override the bit
clock is convenient, but it is not a standard ALSA design and
may eventually introduce new problems like today.

I agree. I'm not conservative at all concerning this question.
I don't see a way to remove set_sysclk without breaking current TDM
users anyway, at least for those who don't have their code
upstreamed.

Which TDM case would be broken by this removal? The only impact
that I can see is that the ASoC core returns an ENOTSUPP for a
set_sysclk() call now, which is something that a dai-link driver
should have taken care of anyway.

All information provided through snd_soc_dai_set_tdm_slot( cpu_dai,
mask, mask, slots, width ) should be enough
In this case, for TDM users

   bclk = slots * width * fs   (where slots is != channels)

will manage 99 % of the cases.
And the remaining 1% will concern people who need to hack the kernel
so widely they don't care about the set_sysclk removal.

A patch from those people will be always welcome.

- fsl-asoc-card.c : *something will break since
snd_soc_dai_set_sysclk returned code is checked*

I've already submitted a patch to ignore all ENOTSUPP.