On Sat, May 01, 2021 at 06:09:43PM +0200, Uwe Kleine-König wrote:
> Without this change it can happen that if changing the polarity succeeded
> but changing duty_cycle and period failed pwm->state contains a mixture
> between the old and the requested state.
> 
> So remember the initial state before starting to modify the configuration
> and restore it when one of the required callback fails.
> 
> Compared to the previous implementation .disable() (if necessary) is called
> earlier to prevent a glitch.
> 
> Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
> ---
> Hello,
> 
> just a small optimisation: At the end of pwm_apply_legacy()
> state->enabled is known to be true, so simplify
> 
> 	if (state->enabled && !pwm->state.enabled) {
> 
> to
> 	if (!pwm->state.enabled) {
> 
> Best regards
> Uwe
> 
>  drivers/pwm/core.c | 139 +++++++++++++++++++++++++--------------------
>  1 file changed, 78 insertions(+), 61 deletions(-)

Applied, thanks. I made a few tiny changes such as dropping the apply
function pointer, which didn't seem very useful.

Thierry