On Wed, Feb 10, 2021 at 11:39 PM Mark Brown [off-list ref] wrote:

On Wed, Feb 10, 2021 at 02:35:29PM +0800, Shengjiu Wang wrote:

quoted

On Wed, Feb 10, 2021 at 6:30 AM Mark Brown [off-list ref] wrote:

quoted

quoted

Like I say I'd actually recommend moving this control to DAPM.

quoted

I may understand your point, you suggest to use the .set_bias_level
interface. But in my case I need to enable the clock in earlier stage
and keep the clock on when system go to suspend.

The device can be kept alive over system suspend if that's needed, or
possibly it sounds like runtime PM is a better fit?  There's callbacks
in the core to keep the device runtime PM enabled while it's open which
is probably about the time range you're looking for.

Before enabling the clock, I need to reparent the clock according to
the sample rate,  Maybe the hw_params is the right place to do
these things.

Can I add a flag:
"rpmsg->mclk_streams & BIT(substream->stream)"
for avoiding multiple calls of hw_params function before enabling
clock?