On Tue, Dec 17, 2019 at 07:56:28AM -0600, Pierre-Louis Bossart wrote:
> On 12/17/19 6:13 AM, Mark Brown wrote:
> > On Tue, Dec 17, 2019 at 01:41:09PM +0900, Kuninori Morimoto wrote:

> > > Can we use snd_soc_dapm_stream_stop() for these, too ?

> > That does seem like an oversight...

> What would the rationale for deferring a FE shutdown though?

> IIRC in the Intel machine drivers we only use .pm_ignore_downtime to
> backends to keep some clocks on, and even that is questionable (the clock
> dependencies should be properly modeled instead of hidden with an obscure
> dailink setting). When I asked why exactly this flag was set in most cases
> the reason why 'because others did so in the past'...

It's there as a hack to keep clocks live or to avoid pops and clicks as
the datastream goes to zero during shutdown if the CODEC doesn't support
mute or otherwise handle the input going away cleanly (which is just a
more general case of keeping the clocks live usually).  You'd do it for
front ends as well to make sure things are consistent.