Re: [PATCH 2/2] pwm: bcm-kona: apply pwm settings on enable

[Uwe Kleine-König]

Hello,

On Wed, Nov 07, 2018 at 10:36:13AM +0100, Clément Péron wrote:
> From: Suji Velupillai <suji.velupillai@xxxxxxxxxxxx>
> 
> When pwm_bl framework calls enable, a call to pwm_is_enabled(pwm) still
> return false, this prevents the backlight being turn on at boot time.
> 
> Signed-off-by: Suji Velupillai <suji.velupillai@xxxxxxxxxxxx>
> Signed-off-by: Clément Péron <peron.clem@xxxxxxxxx>
> ---
>  drivers/pwm/pwm-bcm-kona.c | 16 +++++++++++-----
>  1 file changed, 11 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/pwm/pwm-bcm-kona.c b/drivers/pwm/pwm-bcm-kona.c
> index 09a95aeb3a70..d991d53c4b38 100644
> --- a/drivers/pwm/pwm-bcm-kona.c
> +++ b/drivers/pwm/pwm-bcm-kona.c
> @@ -108,8 +108,8 @@ static void kona_pwmc_apply_settings(struct kona_pwmc *kp, unsigned int chan)
>  	ndelay(400);
>  }
>  
> -static int kona_pwmc_config(struct pwm_chip *chip, struct pwm_device *pwm,
> -			    int duty_ns, int period_ns)
> +static int __pwmc_config(struct pwm_chip *chip, struct pwm_device *pwm,
> +			 int duty_ns, int period_ns, bool pwmc_enabled)
>  {
>  	struct kona_pwmc *kp = to_kona_pwmc(chip);
>  	u64 val, div, rate;
> @@ -155,7 +155,7 @@ static int kona_pwmc_config(struct pwm_chip *chip, struct pwm_device *pwm,
>  	 * always calculated above to ensure the new values are
>  	 * validated immediately instead of on enable.
>  	 */
> -	if (pwm_is_enabled(pwm)) {
> +	if (pwm_is_enabled(pwm) || pwmc_enabled) {

Having pwm-API-calls in hw-drivers is ugly. Apart from not giving the
intended return code this function should IMHO be reserved to pwm
consumers. The underlaying problem is that pwm-bl does:

	pwm_config(pwm, duty_cycle, period);
	pwm_enable(pwm);

and expects that the duty_cycle and period is used then. Doesn't
everything works just fine if the if-block is always executed?

The better fix here would be to convert the driver to the atomic API
(i.e. implement .apply instead of .config, .set_polarity, .enable and
.disable).

Alternatively in .enable ensure that the hardware is programmed with the
parameters from pwm->state. (But converting to the atomic API is the
better approach.)

Best regards
Uwe

-- 
Pengutronix e.K.                           | Uwe Kleine-König            |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |