On 2/6/19 1:55 PM, Thierry Reding wrote:
> On Wed, Feb 06, 2019 at 09:54:05AM +0100, Uwe Kleine-König wrote:
>> On Wed, Feb 06, 2019 at 09:42:48AM +0100, Fabrice Gasnier wrote:
>>> If you agree with the current approach, I can send a V2 with Tomasz's
>>> suggestion to remove the ifdefs and use __maybe_unused instead.
>>
>> I think the suspend callback should have something like:
>>
>> if (is_still_enabled) {
>> /*
>> * The consumer didn't stop us, so refuse to suspend.
>> */
>> dev_err(dev, "The consumer didn't stop us, so refuse to suspend.\n");
>> return -EBUSY;
>> }
>>
>> This way there are no bad surprises if the pwm is suspended before its
>> consumer and it's obvious what is missing.
Thierry, Uwe,
When the pwm is suspended before its consumer, the bad surprise is the
suspend request will fail... I'm not sure a new attempt may be better.
So, it looks like the only way to have this clean is by implementing the
device link e.g. via pwm_get() ?
>
> Something that just occurred to me: perhaps as part of pwm_get() we
> should track where we were being requested from so that we could give
> hints in situations like this as to where the consumer is that forgot
> to suspend the PWM.
The current approach handles the situation where the consumer forgot to
suspend the PWM... I can add some warning about that in the suspend
routine, incl the label.
What do you think? What's the best approach ?
Please advise,
BR,
Fabrice
>
> I suppose we already have pwm_device.label to help with this, but
> perhaps we could improve things if we stored __builtin_return_address
> during pwm_get() to help users pinpoint where they need to look.
>
> Thierry
>
