Hi, On 26/01/15 19:10, Andrew Lunn wrote: >> So... To me it's still slightly unclear when should one write a PWM >> driver and when a LED driver. But I would say that as the TLC591xx >> outputs a PWM signal, it should be a PWM driver. Then the different >> users of this PWM signal could be made on top of that (LED, backlight, GPO). >> >> What would be the technical drawbacks with having the TLC591xx driver as >> a PWM, instead of LED? > > Hi Tomi > > We have been through this once, but the big technical drawback is that > this hardware cannot do what the Linux Kernel defines as PWM. > > It cannot correctly implement the PMW call: > > int pwm_config(struct pwm_device *pwm, int duty_ns, int period_ns); > > This hardware has a fixed period, since it is clocked at 97-kHz. So > you cannot set the period. The duty is also somewhat restrictive, in > that it only allows 1/256 increments of the 97Khz.
Surely other PWM devices also have restrictions in the period or duty cycle?
> This hardware does however perfectly fit the LED API:
>
> enum led_brightness {
> LED_OFF = 0,
> LED_HALF = 127,
> LED_FULL = 255,
> };
>
> void (*brightness_set)(struct led_classdev *led_cdev,
> enum led_brightness brightness);
>
> So we can model it perfectly as an LED driver, or badly as a PWM
> driver.
Maybe so, but what does it mean in practice? If it's implemented as a
PWM driver, and the existing leds-pwm driver is used for the led
functionality, shall we miss some brightness values? Is the
configuration more difficult? Or what?
So my point here is that it outputs a PWM signal, so it makes sense to
have it as a PWM driver. It has restricted configurability compared to
more versatile PWM hardware, but I so far don't see why that would be an
issue.
If it is a PWM driver, we get backlight support for free via the
existing pwm_bl driver, and LED support via leds-pwm. And there has been
a clear acceptance on GPO functionality with PWM outputs (in the Peter's
mail thread), whereas I would bet that a LED based GPO functionality
would encounter resistance, because that doesn't quite make sense.
Tomi
signature.asc
Description: OpenPGP digital signature
