Re: [PATCH v3 2/4] soc: qcom: rpmhpd: Do proper power off when state synced

[Bjorn Andersson]

On Mon, Mar 27, 2023 at 10:38:27PM +0300, Abel Vesa wrote:
> Instead of aggregating different corner values on sync state callback,
> call the genpd API for queuing up the power off. This will also mark the
> domain as powered off in the debugfs genpd summary. Also, until sync
> state has been reached, return busy on power off request, in order to
> allow genpd core to know that the actual domain hasn't been powered of
> from the "disable unused" late initcall.
> 
> Signed-off-by: Abel Vesa <abel.vesa@xxxxxxxxxx>
> Reviewed-by: Ulf Hansson <ulf.hansson@xxxxxxxxxx>
> ---
>  drivers/soc/qcom/rpmhpd.c | 19 +++++++------------
>  1 file changed, 7 insertions(+), 12 deletions(-)
> 
> diff --git a/drivers/soc/qcom/rpmhpd.c b/drivers/soc/qcom/rpmhpd.c
> index f20e2a49a669..ec7926820772 100644
> --- a/drivers/soc/qcom/rpmhpd.c
> +++ b/drivers/soc/qcom/rpmhpd.c
> @@ -649,8 +649,12 @@ static int rpmhpd_power_off(struct generic_pm_domain *domain)
>  	mutex_lock(&rpmhpd_lock);
>  
>  	ret = rpmhpd_aggregate_corner(pd, 0);
> -	if (!ret)
> -		pd->enabled = false;
> +	if (!ret) {
> +		if (!pd->state_synced)
> +			ret = -EBUSY;
> +		else
> +			pd->enabled = false;
> +	}
>  
>  	mutex_unlock(&rpmhpd_lock);
>  
> @@ -810,10 +814,8 @@ static void rpmhpd_sync_state(struct device *dev)
>  {
>  	const struct rpmhpd_desc *desc = of_device_get_match_data(dev);
>  	struct rpmhpd **rpmhpds = desc->rpmhpds;
> -	unsigned int corner;
>  	struct rpmhpd *pd;
>  	unsigned int i;
> -	int ret;
>  
>  	mutex_lock(&rpmhpd_lock);
>  	for (i = 0; i < desc->num_pds; i++) {
> @@ -822,14 +824,7 @@ static void rpmhpd_sync_state(struct device *dev)
>  			continue;
>  
>  		pd->state_synced = true;
> -		if (pd->enabled)
> -			corner = max(pd->corner, pd->enable_corner);

Note that the intent of this line is to lower the corner from max to
either a requested performance_state or the lowest non-disabled corner.
I don't think your solution maintains this behavior?

> -		else
> -			corner = 0;
> -
> -		ret = rpmhpd_aggregate_corner(pd, corner);
> -		if (ret)
> -			dev_err(dev, "failed to sync %s\n", pd->res_name);
> +		pm_genpd_queue_power_off(&pd->pd);

In the event that the power-domain has a single device attached, and no
subdomains, wouldn't pm_genpd_queue_power_off() pass straight through
all checks and turn off the power domain? Perhaps I'm just missing
something?

Regards,
Bjorn

>  	}
>  	mutex_unlock(&rpmhpd_lock);
>  }
> -- 
> 2.34.1
>