Re: [PATCH] ASoC: stm32: sai: manage context in set_sysclk callback

[Olivier MOYSAN]

Hi Kuninori,

On 9/18/25 03:34, Kuninori Morimoto wrote:
> Hi Olivier
>
> Thank you for clarify the situation
>
> > Looking at the traces (see below), initially, we had one call with the
> > 'in' direction (kernel clock frequency) and one call with the 'out'
> > direction.
> (snip)
> > Now, both calls have the 'out' direction when the property
> > 'system-clock-direction-out' is set. This seems more consistent with the
> > changes from commit 5725bce709db1.
>
> Ah.. simple-card-utils calls snd_soc_dai_set_sysclk() from 3 functions
>
> 	- simple_init_dai()
> 	- simple_util_hw_params()
> 	- simple_util_shutdown()
>
> but its "direction" is...
>
> Step0:
> 	Before
> 	a728f56094e7cf60a1dc0642fe86901d1a4dfb4e
> 	 ("ASoC: make clock direction configurable in asoc-simple")
>
> 	- simple_init_dai()		fixed: 0
> 	- simple_util_hw_params()	fixed: SND_SOC_CLOCK_xx
> 	- simple_util_shutdown()	fixed: SND_SOC_CLOCK_xx
>
> Step1:
> 	After
> 	a728f56094e7cf60a1dc0642fe86901d1a4dfb4e
> 	 ("ASoC: make clock direction configurable in asoc-simple")
>
> 	and before
>
> 	5725bce709db1c001140d79398581e067e28c031
> 	("ASoC: simple-card-utils: Unify clock direction by clk_direction")
>
>
> 	- simple_init_dai()		dai->clk_direction
> 	- simple_util_hw_params()	fixed: SND_SOC_CLOCK_xx
> 	- simple_util_shutdown()	fixed: SND_SOC_CLOCK_xx
>
> Step2:
> 	After
> 	5725bce709db1c001140d79398581e067e28c031
> 	("ASoC: simple-card-utils: Unify clock direction by clk_direction")
>
> 	- simple_init_dai()		dai->clk_direction
> 	- simple_util_hw_params()	dai->clk_direction
> 	- simple_util_shutdown()	dai->clk_direction
>
> In Step0 and Step1, your dirver is called with both dir IN/OUT ?

Yes, based on the trace, this appears to be the case.

> > In my setup, the kernel clock is an input clock. Maybe it should be
> > tagged as an input clock. However, we have no
> > 'system-clock-direction-in' property yet.
> > Anyway, the patch in the STM32 SAI driver is still valid.
>
> OK
>
> Or maybe reset simple_dai->sysclk somehow (not sure...) ?
> (simple_init_dai() will call snd_soc_dai_set_sysclk() if it has
> simple_dai->sysclk)

Yes, resetting simple_dai->sysclk may also be an option, but I could not 
find a way to do it in practice. Requesting a reset of 
simple_dai->sysclk from the CPU DAI .probe callback might be the right 
time I think, but there is no service that allows this.
Another way would be to set "system-clock-frequency = <0>" in the CPU DT 
node. However, this is not possible, as it conflicts with the following 
check in the simple_set_clk_rate() function:
if (simple_dai->clk_fixed && rate != simple_dai->sysclk) {
   dev_err(...)
}

So, the current patch still seems like the better option to me.

Best regards
Olivier

> Thank you for your help !!
>
> Best regards
> ---
> Kuninori Morimoto