Hello,

On Fri, Jun 16, 2023 at 05:29:08PM +0200, Artur Weber wrote:

On 14/06/2023 10:39, Uwe Kleine-König wrote:

quoted

On Sat, Apr 29, 2023 at 12:45:32PM +0200, Artur Weber wrote:

quoted

Also deprecate the pwm-period DT property, as it is now redundant
(pwms property already contains period value).

Signed-off-by: Artur Weber <aweber.kernel@gmail.com>
---
 drivers/video/backlight/lp855x_bl.c | 48 ++++++++++++++++-------------
 1 file changed, 26 insertions(+), 22 deletions(-)

diff --git a/drivers/video/backlight/lp855x_bl.c b/drivers/video/backlight/lp855x_bl.c
index 81012bf29baf..21eb4943ed56 100644
--- a/drivers/video/backlight/lp855x_bl.c
+++ b/drivers/video/backlight/lp855x_bl.c

...

@@ -472,11 +456,31 @@ static int lp855x_probe(struct i2c_client *cl)
 	lp->enable = devm_regulator_get_optional(dev, "enable");
 	if (IS_ERR(lp->enable)) {
 		ret = PTR_ERR(lp->enable);
-		if (ret == -ENODEV) {
+		if (ret == -ENODEV)
 			lp->enable = NULL;
-		} else {
+		else
 			return dev_err_probe(dev, ret, "getting enable regulator\n");
-		}
+	}
+
+	lp->pwm = devm_pwm_get(lp->dev, lp->chipname);
+	if (IS_ERR(lp->pwm)) {
+		ret = PTR_ERR(lp->pwm);
+		if (ret == -ENODEV || ret == -EINVAL)

Why would you ignore EINVAL?

EINVAL is returned when the pwms property is not found in the DT node
for the backlight. Not sure if there's a better way of separately
detecting whether it's present (especially when taking into
consideration non-DT platforms that might use the driver). Would be nice
to have something like devm_regulator_get_optional but for PWMs...

Ah, that is because of_pwm_get() calls of_property_match_string(np,
"pwm-names", con_id) which returns -EINVAL if there is no pwm-names
property. This is different for clocks. I wonder if pwm should adapt
accordingly? Thierry?

Still, someone who's setting up the driver could check the debug
messages to see if the backlight was set up in PWM mode or register mode.

quoted

...

quoted

+		pwm_init_state(lp->pwm, &pwmstate);
+		/* Legacy platform data compatibility */
+		if (lp->pdata->period_ns > 0)
+			pwmstate.period = lp->pdata->period_ns;
+		pwm_apply_state(lp->pwm, &pwmstate);

This is a change in behaviour. Before lp855x_probe() didn't modify the
state the bootloader left the backlight in. Now you're disabling it (I
think). Is this intended?

I didn't really consider the implication of this in this way, as on the
device I was testing this on (Exynos4212-based tablet) the PWM state
would get reset during PWM chip initialization in the kernel anyways,

Which chip driver is in use here? That's a patch opportunity.

meaning that the state from the bootloader would be lost regardless of
this change. Either way, there's no guarantee that this would be the
same on other devices, though I'd assume that in most cases it's not
noticeable anyways as brightness is usually set somewhere in userspace
(or even earlier, in the driver, if the init-brt property is set).
Nonetheless, that's an oversight on my part.

As for the reasoning for this change in behavior - the previous behavior
was to silently fail if, while setting the brightness, the PWM could not

This sounds wrong.

be set up. This seemed rather confusing to me (I encountered this while
I was initially working on the tablet, I added a "pwm" property instead
of "pwms" and was wondering why the backlight didn't work...)

Of course, that could be fixed by adding error detection in the
brightness set function, but since I was already working on it, it made
more sense to me for the PWM to be set up during the probing process,
given that this way we could 1. warn about errors early, and 2. catch
deferred probes and defer the backlight's probe if we're still waiting
for the PWM. That's why it's done the way it is in this patch.

If this is undesired behavior, let me know and I'll submit another patch
to revert it.

Best regards
Uwe

-- 
Pengutronix e.K.                           | Uwe Kleine-König            |
Industrial Linux Solutions                 | https://www.pengutronix.de/ |