On 2/28/25 4:18 PM, Uwe Kleine-König wrote:
Hey David,
On Fri, Feb 28, 2025 at 11:09:31AM +0100, David Jander wrote:
On Fri, 28 Feb 2025 10:37:48 +0100
Krzysztof Kozlowski <krzk@xxxxxxxxxx> wrote:
On 28/02/2025 10:22, David Jander wrote:
+
+ motion,pwm-inverted:
+ $ref: /schemas/types.yaml#/definitions/flag
And PWM flag does not work?
I have seen PWM controllers that don't seem to support the
PWM_POLARITY_INVERTED flag and those where it just doesn't work. Should all
Shouldn't the controllers be fixed? Or let's rephrase the question: why
only this PWM consumer needs this property and none of others need it?
CCing Uwe Kleine-Koenig and linux-pwm mailing list.
I know that at least in kernel 6.11 the pwm-stm32.c PWM driver doesn't
properly invert the PWM signal when specifying PWM_POLARITY_INVERTED. I agree
this is a probably bug that needs fixing if still present in 6.14-rc. Besides
that, if linux-pwm agrees that every single PWM driver _must_ properly support
this flag, I will drop this consumer flag an start fixing broken PWM drivers
that I encounter. I agree that it makes more sense this way, but I wanted to
be sure.
Some hardwares cannot support PWM_POLARITY_INVERTED. Affected drivers
include:
pwm-adp5585
pwm-ntxec
pwm-raspberrypi-poe
pwm-rz-mtu3 (software limitation only)
pwm-sunplus
pwm-twl-led (not completely sure, that one is strange)
. ISTR that there is a driver that does only support inverted polarity,
but I don't find it. For an overview I recommend reading through the
output of:
The only one that I know of is the opencores pwm driver that the
starfive jh71xx uses, I remember talking with you there. That one does
still need a proper review I believe:
https://lore.kernel.org/all/20250106103540.10079-1-william.qiu@xxxxxxxxxxxxxxxx/
It is kind of in a limbo right now
for f in drivers/pwm/pwm-*; do
echo $f;
sed -rn '/Limitations:/,/\*\/?$/p' $f;
echo;
done | less
. (Note not all drivers have commentary in the right format to unveil
their limitations.)
For most use-cases you can just do
.duty_cycle = .period - .duty_cycle
instead of inverting polarity, but there is no abstraction in the PWM
bindings for that and also no helpers in the PWM framework. The problem
is more or less ignored, so if you have a device with
pwms = <&pwm0 0 PWM_POLARITY_INVERTED>;
and the PWM chip in question doesn't support that, the pwm API functions
will fail. So the system designer better makes sure that the PWM
hardware can cope with the needed polarity.
Best regards
Uwe
