Hi YueHaibing,
On Fri, May 31, 2019 at 11:49 PM YueHaibing <yuehaibing@xxxxxxxxxx> wrote:
>
> mutex_lock(&pca->lock);
> - pwm = &pca->chip.pwms[offset];
> mutex_unlock(&pca->lock);
Thanks for noticing this issue. However it should be fixed differently.
This was introduced by Uwe's clean-up patch:
commit e926b12c611c2095c79 ("pwm: Clear chip_data in pwm_put()")
But Uwe did not realize that in this case, the pwm chip_data is used as a
synchronization mechanism between pwm and gpio. Moving the chip_data
clear out of the mutex breaks this mechanism.
I think the following would restore the mechanism:
> mutex_lock(&pca->lock);
> pwm = &pca->chip.pwms[offset];
> + pwm_set_chip_data(pwm, NULL);
> mutex_unlock(&pca->lock);
This would of course clear the pwm chip_data twice, once in the driver and
once in the core, but that's not a problem.
I'd like to hear Mika Westerberg's opinion, because he introduced this
synchronization mechanism back in 2016.
[Adding Mika]
Sven
