On Wed, Jul 12, 2023 at 09:45:08AM +0800, Guiting Shen wrote:
> On Wed, Jul 12, 2023 at 04:30:17AM GMT+8, Uwe Kleine-König wrote:
> > Hello,
> > 
> > On Wed, Jul 12, 2023 at 04:09:05AM +0800, Guiting Shen wrote:
> >> +static int atmel_pwm_enable_clk_if_on(struct atmel_pwm_chip *atmel_pwm)
> >> +{
> >> +	unsigned int i;
> >> +	int err;
> >> +	u32 sr;
> >> +
> >> +	sr = atmel_pwm_readl(atmel_pwm, PWM_SR);
> >> +	if (!sr)
> >> +		return 0;
> >> +
> >> +	for (i = 0; i < atmel_pwm->chip.npwm; i++) {
> >> +		if (!(sr & (1 << i)))
> >> +			continue;
> >> +
> >> +		err = clk_enable(atmel_pwm->clk);
> >> +		if (err) {
> >> +			dev_err(atmel_pwm->chip.dev,
> >> +				"failed to enable clock: %pe\n", ERR_PTR(err));
> > 
> > Here you leak possibly a few enables. While it's not likely that the
> > (say) third enable goes wrong, it's also not that hard to handle?!
> 
> The driver used the enable_count member of struct clk_core to count the
> PWM channels(4 channels). It will enable hardware clock only when one of
> the PWM channels becomed on from all PWM channels off which maybe return
> error. And in second/third/fourth times to clk_enable(), it just
> increased the enable_count of struct clk_core which would never return
> error.
> 
> It maybe confused at first time to view the code.
> Do it need to add something like that: ?
> 
> for (i = 0; i < atmel_pwm->chip.npwm; i++) {
> 	if (!(sr & (1 << i)))
> 		continue;
> 
> 	err = clk_enable(atmel_pwm->clk);
> 	if (err) {
> 		dev_err(atmel_pwm->chip.dev,
> 			"failed to enable clock: %pe\n", ERR_PTR(err));
> 
> 		for (i = 0; i < cnt; i++)
> 			clk_disable(atmel_pwm->clk);
> 		return err;
> 	}
> 	cnt++;

You can also achieve this by decrementing i back to zero, that way you
avoid the additional variable and you get a more natural unwinding of
what you did before.

So something like:

	while (i--)
		clk_disable(atmel_pwm->clk);

should do the same thing.

Thierry