Skip to contentSkip to conversation

pwm: pca9685: Fix regression for GPIO use

PATCH In review Andy Shevchenko 3 participants 3 messages Aug 30, 2019 – Sep 14, 2019 PWM

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 pwm: pca9685: Fix regression for GPIO use

The commit e926b12c611c ("pwm: Clear chip_data in pwm_put()") breaks GPIO usage by actually removing a crucial call to pwm_set_chip_data(), which is being used as a flag to understand the mode (GPIO vs. PWM) in which pin is currently working. Now, if once I requested GPIO, I may not be able to free and request it again.

Fixes e926b12c611c ("pwm: Clear chip_data in pwm_put()")
Cc Mika Westerberg <[email protected]>
Cc Uwe Kleine-König <[email protected]>

---
v2: drop leftovers in Fixes tag

 drivers/pwm/pwm-pca9685.c | 1 +
 1 file changed, 1 insertion(+)
drivers/pwm/pwm-pca9685.c +1 -0
index 168684b02ebc..7cdec56ab3ec 100644
@@ -165,6 +165,7 @@ static void pca9685_pwm_gpio_free(struct gpio_chip *gpio, unsigned int offset)
 	pm_runtime_put(pca->chip.dev);
 	mutex_lock(&pca->lock);
 	pwm = &pca->chip.pwms[offset];
+	pwm_set_chip_data(pwm, NULL);
 	mutex_unlock(&pca->lock);
 }
 
Signature · 1 line
2.23.0.rc1
#1 lore raw
Quoted 5 lines · Andy Shevchenko

The commit e926b12c611c ("pwm: Clear chip_data in pwm_put()") breaks GPIO usage by actually removing a crucial call to pwm_set_chip_data(), which is being used as a flag to understand the mode (GPIO vs. PWM) in which pin is currently working. Now, if once I requested GPIO, I may not be able to free and request it again.

There was another patch from Sven (cc'd) fixing the same issue here:

https://lore.kernel.org/patchwork/patch/1084046/

But I don't think it was ever merged anywhere. Maybe you can try it out and give your tested-by?

Quoted 23 lines
Fixes e926b12c611c ("pwm: Clear chip_data in pwm_put()")
Cc Mika Westerberg <[email protected]>
Cc Uwe Kleine-König <[email protected]>

---
v2: drop leftovers in Fixes tag

 drivers/pwm/pwm-pca9685.c | 1 +
 1 file changed, 1 insertion(+)
drivers/pwm/pwm-pca9685.c +1 -0
index 168684b02ebc..7cdec56ab3ec 100644
@@ -165,6 +165,7 @@ static void pca9685_pwm_gpio_free(struct gpio_chip *gpio, unsigned int offset)
 	pm_runtime_put(pca->chip.dev);
 	mutex_lock(&pca->lock);
 	pwm = &pca->chip.pwms[offset];
+	pwm_set_chip_data(pwm, NULL);
 	mutex_unlock(&pca->lock);
 }
 
Signature · 1 line
2.23.0.rc1
#2 lore raw

Hi Andy,

Quoted 6 lines · Mika Westerberg

There was another patch from Sven (cc'd) fixing the same issue here:
https://lore.kernel.org/patchwork/patch/1084046/

But I don't think it was ever merged anywhere. Maybe you can try it out and give your tested-by?

When the issue came up, Mika and I considered restoring the old synchronization. But we suspect that the old synchronization is broken, or has become broken over time.
See: https://lkml.org/lkml/2019/6/2/73

The patch above was reviewed by Mika, but could not be merged - there was no-one available to test it out on actual h/w.

I did however successfully test the logic. I disconnected the driver from its h/w by
giving it a dummy regmap.
See: https://lkml.org/lkml/2019/6/6/666

I am of course open to alternative solutions.

Sven