Hi Uwe,

    On 2021/7/16, 6:13 PM, "Uwe Kleine-K?nig" [off-list ref] wrote:

        On Fri, Jul 16, 2021 at 09:22:22AM +0000, Billy Tsai wrote:
        >> On 2021/7/16, 3:10 PM, "Uwe Kleine-K?nig" [off-list ref] wrote:
        >> 
        >>     On Fri, Jul 16, 2021 at 01:48:20AM +0000, Billy Tsai wrote:
        >>     >> On 2021/7/15, 11:06 PM, "Uwe Kleine-K?nig" [off-list ref]> wrote:
        >>     >>     > Another is: The PWM doesn't support duty_cycle 0, on such a request the
        >>     >>     > PWM is disabled which results in a constant inactive level.
        >>     >> 
        >>     >>     > (This is correct, is it? Or does it yield a constant 0 level?)
        >>     >> 
        >>     >> Our pwm can support duty_cycle 0 by unset CLK_ENABLE.
        >> 
        >>     > This has a slightly different semantic though. Some consumer might
        >>     > expect that the following sequence:
        >> 
        >>     >	pwm_apply(mypwm, { .period = 10000, .duty_cycle = 10000, .enabled = true })
        >>     >	pwm_apply(mypwm, { .period = 10000, .duty_cycle = 0, .enabled = true })
        >>     >	pwm_apply(mypwm, { .period = 10000, .duty_cycle = 10000, .enabled = true })
        >> 
        >>     > results in the output being low for an integer multiple of 10 ?s. This
        >>     > isn't given with setting CLK_ENABLE to zero, is it? (I didn't recheck,
        >>     > if the PWM doesn't complete periods on reconfiguration this doesn't
        >>     > matter much though.)
        >> Thanks for the explanation.
        >> Our hardware actually can only support duty from 1/256 to 256/256.
        >> For this situation I can do possible solution:
        >> We can though change polarity to meet this requirement. Inverse the pin and use
        >> duty_cycle 100. 
        >> But I think this is not a good solution for this problem right?

        > If this doesn't result in more glitches that would be fine for me.
        > (Assuming it is documented good enough in the code to be
        > understandable.)

    > The polarity of our pwm controller will affect the duty cycle range:
    > PWM_POLARITY_INVERSED : Support duty_cycle from 0% to 99%
    > PWM_POLARITY_NORMAL: Support duty_cycle from 1% to 100%
    > Dynamic change polarity will result in more glitches. Thus, this will become
    > a trade-off between 100% and 0% duty_cycle support for user to use our pwm device.
    > I will document it and send next patch.

For handling the situation that the user want to set the duty cycle to 0%, the driver can:
1. Just return the error.
2. Use the minimum duty cycle value.
I don't know which solution will be the better way or others.
I would be grateful if you can give me some suggestion about this problem.

Thanks

Best Regards,
Billy Tsai