Re: [PATCH v3 05/10] pwm: add PWM mode to pwm_config()

[Claudiu Beznea]

On 27.02.2018 17:38, Daniel Thompson wrote:
> On Tue, Feb 27, 2018 at 01:40:58PM +0200, Claudiu Beznea wrote:
>> On 27.02.2018 12:54, Daniel Thompson wrote:
>>> On Mon, Feb 26, 2018 at 04:24:15PM +0200, Claudiu Beznea wrote:
>>>> On 26.02.2018 11:57, Jani Nikula wrote:
>>>>> On Thu, 22 Feb 2018, Daniel Thompson <daniel.thompson@xxxxxxxxxx> wrote:
>>>>>> On Thu, Feb 22, 2018 at 02:01:16PM +0200, Claudiu Beznea wrote:
>>>>>>> Add PWM mode to pwm_config() function. The drivers which uses pwm_config()
>>>>>>> were adapted to this change.
>>>>>>>
>>>>>>> Signed-off-by: Claudiu Beznea <claudiu.beznea@xxxxxxxxxxxxx>
>>>>>>> ---
>>>>>>>  arch/arm/mach-s3c24xx/mach-rx1950.c  | 11 +++++++++--
>>>>>>>  drivers/bus/ts-nbus.c                |  2 +-
>>>>>>>  drivers/clk/clk-pwm.c                |  3 ++-
>>>>>>>  drivers/gpu/drm/i915/intel_panel.c   | 17 ++++++++++++++---
>>>>>>>  drivers/hwmon/pwm-fan.c              |  2 +-
>>>>>>>  drivers/input/misc/max77693-haptic.c |  2 +-
>>>>>>>  drivers/input/misc/max8997_haptic.c  |  6 +++++-
>>>>>>>  drivers/leds/leds-pwm.c              |  5 ++++-
>>>>>>>  drivers/media/rc/ir-rx51.c           |  5 ++++-
>>>>>>>  drivers/media/rc/pwm-ir-tx.c         |  5 ++++-
>>>>>>>  drivers/video/backlight/lm3630a_bl.c |  4 +++-
>>>>>>>  drivers/video/backlight/lp855x_bl.c  |  4 +++-
>>>>>>>  drivers/video/backlight/lp8788_bl.c  |  5 ++++-
>>>>>>>  drivers/video/backlight/pwm_bl.c     | 11 +++++++++--
>>>>>>>  drivers/video/fbdev/ssd1307fb.c      |  3 ++-
>>>>>>>  include/linux/pwm.h                  |  6 ++++--
>>>>>>>  16 files changed, 70 insertions(+), 21 deletions(-)
>>>>>>>
>>>>>>> diff --git a/drivers/video/backlight/lm3630a_bl.c b/drivers/video/backlight/lm3630a_bl.c
>>>>>>> index 2030a6b77a09..696fa25dafd2 100644
>>>>>>> --- a/drivers/video/backlight/lm3630a_bl.c
>>>>>>> +++ b/drivers/video/backlight/lm3630a_bl.c
>>>>>>> @@ -165,8 +165,10 @@ static void lm3630a_pwm_ctrl(struct lm3630a_chip *pchip, int br, int br_max)
>>>>>>>  {
>>>>>>>  	unsigned int period = pchip->pdata->pwm_period;
>>>>>>>  	unsigned int duty = br * period / br_max;
>>>>>>> +	struct pwm_caps caps = { };
>>>>>>>  
>>>>>>> -	pwm_config(pchip->pwmd, duty, period);
>>>>>>> +	pwm_get_caps(pchip->pwmd->chip, pchip->pwmd, &caps);
>>>>>>> +	pwm_config(pchip->pwmd, duty, period, BIT(ffs(caps.modes) - 1));
>>>>>>
>>>>>> Well... I admit I've only really looked at the patches that impact 
>>>>>> backlight but dispersing this really odd looking bit twiddling 
>>>>>> throughout the kernel doesn't strike me a great API design.
>>>>>>
>>>>>> IMHO callers should not be required to find the first set bit in
>>>>>> some specially crafted set of capability bits simply to get sane 
>>>>>> default behaviour.
>>>>>
>>>>> Agreed. IMHO the regular use case becomes rather tedious, ugly, and
>>>>> error prone.
>>>>
>>>> Using simply PWM_MODE(NORMAL) instead of BIT(ffs(caps.modes) - 1) would be OK
>>>> from your side?
>>>>
>>>> Or, what about using a function like pwm_mode_first() to get the first supported
>>>> mode by PWM channel?
>>>>
>>>> Or, would you prefer to solve this inside pwm_config() function, let's say, in
>>>> case an invalid mode is passed as argument, to let pwm_config() to choose the
>>>> first available PWM mode for PWM channel passed as argument?
>>>
>>> What is it that actually needs solving?
>>>
>>> If a driver requests normal mode and the PWM driver cannot support it
>>> why not just return an error an move on.
>> Because, simply, I wasn't aware of what these PWM client drivers needs for.
> 
> I'm afraid you have confused me here.
> 
> Didn't you just *add* the whole concept of PWM caps with your patches?
> How could any existing call site expect anything except normal mode.
> Until now there has been no possiblity to request anything else.
Agree. And agree I was confusing in previous email, sorry about that. And
agree that there was nothing before and everything should work with PWM
normal mode.

When I choose to have BIT(ffs(caps.modes)) instead of PWM_MODE(NORMAL) I
was thinking at having these pwm_config() calls working all the time having
in mind that in future the PWM controllers that these drivers use, might
change in terms of PWM supported modes.

Thank you,
Claudiu Beznea

> 
> 
>>> Put another way, what is the use case for secretly adopting a mode the
>>> caller didn't want? Under what circumstances is this a good thing?
>> No one... But I wasn't aware of what the PWM clients needs for from their PWM
>> controllers. At this moment having BIT(ffs(caps.modes)) instead of
>> PWM_MODE(NORMAL) is mostly the same since all the driver that has not explicitly
>> registered PWM caps will use PWM normal mode.
>>
>> I will use PWM_MODE(NORMAL) instead of this in all the cases if this is OK from
>> your side.
>>
>> Thank you,
>> Claudiu Beznea
>>>
>>>
>>> Daniel.
>>>
>