On Mon, Apr 07, 2014 at 12:35:47PM +0100, Russell King - ARM Linux wrote: > On Mon, Apr 07, 2014 at 01:10:05PM +0200, Thierry Reding wrote: > > On Mon, Apr 07, 2014 at 10:46:51AM +0200, Alexandre Belloni wrote: > > > (adding Thierry Reding) > > > > > > Hi, > > > > > > On 06/04/2014 at 23:20:18 +0100, Russell King wrote : > > > > Some PWM outputs are wired such that the LED they're controlling is > > > > connected to supply rather than ground. These PWMs may not support > > > > output inversion, or when they do, disabling the PWM may set the > > > > PWM output low, causing a "brightness" value of zero to turn the LED > > > > fully on. > > > > > > > > The platform data for this driver already indicates that this was > > > > thought about, and we have the "active_low" property there already. > > > > However, the implementation for this is missing. > > > > > > > > Add the trivial implementation for this feature. > > > > > > > > Signed-off-by: Russell King <rmk+kernel@xxxxxxxxxxxxxxxx> > > > > --- > > > > drivers/leds/leds-pwm.c | 4 ++++ > > > > 1 file changed, 4 insertions(+) > > > > > > > > diff --git a/drivers/leds/leds-pwm.c b/drivers/leds/leds-pwm.c > > > > index e1b4c23a409a..1d47742c551f 100644 > > > > --- a/drivers/leds/leds-pwm.c > > > > +++ b/drivers/leds/leds-pwm.c > > > > @@ -70,6 +70,10 @@ static void led_pwm_set(struct led_classdev *led_cdev, > > > > > > > > duty *= brightness; > > > > do_div(duty, max); > > > > + > > > > + if (led_dat->active_low) > > > > + duty = led_dat->period - duty; > > > > + > > > > led_dat->duty = duty; > > > > > > > > > > This will conflict with my patch (which is still lacking proper review) > > > there: > > > http://thread.gmane.org/gmane.linux.leds/482 > > > > > > I would say that it is better to hide the polarity inversion in the PWM > > > driver for your specific PWM. Else we will end up with all the drivers > > > using PWMs trying to detect whether the PWM supports inversion and if it > > > is not the case, calculating the inverted duty cycle. > > > > If the PWM hardware really does support inversion of the polarity, then > > by all means that's what you should be using. However, and I've said > > this a few times already, polarity inversion is not the same as > > reversing the duty-cycle. The *effect* will be the same for LEDs and > > backlights, but the signal is not in fact inverted. > > Sorry, for the general case, you're talking rubbish there. The PWM layer > does not define the starting point in the PWM cycle, so there's no precise > definition of the exact waveform. Therefore: > > Here's a 10% duty cycle: ~_________~_________~_________ > Here's another 10% duty cycle: _________~_________~_________~ > which are both equally valid implementations of a 10% PWM. > > Here's a 90% duty cycle: ~~~~~~~~~_~~~~~~~~~_~~~~~~~~~_ > Here's another 90% duty cycle: _~~~~~~~~~_~~~~~~~~~_~~~~~~~~~ > which are again both equally valid implementations of a 90% PWM. > > Here's the first 10% duty cycle, > inverted: _~~~~~~~~~_~~~~~~~~~_~~~~~~~~~ > which is the same as the second example of a 90% PWM signal. > > Here's the second 10% duty cycle, > inverted: ~~~~~~~~~_~~~~~~~~~_~~~~~~~~~_ > which is the same as the first example of a 90% PWM signal. Good, I'm glad we agree on the basics here. > Now, if PWM did define the starting point of the wave (which it doesn't, > or if it does, it's not documented which means we probably have loads of > buggy drivers) then yes, there would be some grounds to object. PWM does in fact define the starting point. And it's even documented (see enum pwm_polarity in include/linux/pwm.h). The fact that you thought it wasn't documented probably indicates that it's the wrong place, so perhaps you can suggest a better location. One where you would've looked. The convention defined there does match the example signals that you gave above. > In any case, we probably already have drivers which implement both > variants of the 10% signal, so really there's no grounds to say "a 90% > duty cycle is not the same as an inverted 10% duty cycle signal". Like you said, since it's documented any of the drivers that implement it wrongly should then be considered buggy. After all the documentation was there from the beginning (commit 0aa0869c3c9b "pwm: Add support for configuring the PWM polarity"). One of the reasons that I insisted on having this defined rigorously is that all the way back when I took on maintainership there was a fair bit of discussion and somebody (unfortunately I no longer recall where or who) did mentioned that PWMs are used to trigger processes (I vaguely remember synchronization of multiple processes being an issue as well). One of the points made during the discussion was that polarity was indeed important for certain use-cases and hence the rigorous definition of enum pwm_polarity. There's of course the issue that there could be hardware that doesn't allow changing the polarity but doesn't produce the signal as expected by the "normal" polarity. The only way I can think of handling that situation would be to allow drivers to provide a .get_polarity() and have client drivers chicken out if they can't cope with it. Nobody's ever had any incentive to be that rigorous, probably because all drivers really care about is the power of the signal. > The electronic engineer in me says you're talking rubbish too, from the > point of view of designing stuff to produce a PWM signal. Can you please elaborate. I don't understand what you're saying. Thierry
Attachment:
pgpPlKUh_iBp6.pgp
Description: PGP signature
