Skip to contentSkip to conversation

"leds: gpio: Use generic support for composing LED names" breaks gpio debugfs

DISCUSSION Russell King - ARM Linux admin 3 participants 4 messages Dec 4, 2019 – Dec 5, 2019 LED

lore
Share & tools
lore
#0 lore raw

Hi,

Patch d7235f5feaa0 ("leds: gpio: Use generic support for composing LED names") changed the way labels are used.

devm_fwnode_get_gpiod_from_child() is passed led.name, which used to be set to the label or node name, resulting in /sys/kernel/debug/gpio containing useful information. For example:

 gpio-487 (                    |udpu:green:power    ) out lo
 gpio-488 (                    |udpu:red:power      ) out hi
 gpio-489 (                    |udpu:green:network  ) out hi
 gpio-490 (                    |udpu:red:network    ) out hi
 gpio-491 (                    |udpu:green:alarm    ) out hi
 gpio-492 (                    |udpu:red:alarm      ) out hi

After this commit, it now contains a rather less useful:

 gpio-487 (                    |?                   ) out hi ACTIVE LOW
 gpio-488 (                    |?                   ) out hi ACTIVE LOW
 gpio-489 (                    |?                   ) out hi ACTIVE LOW
 gpio-490 (                    |?                   ) out hi ACTIVE LOW
 gpio-491 (                    |?                   ) out hi ACTIVE LOW
 gpio-492 (                    |?                   ) out hi ACTIVE LOW

This is because led.name is now NULL prior to the call to devm_fwnode_get_gpiod_from_child().

While this is not an API, it is useful for debugging, and given the dynamic allocation of GPIOs to physical GPIOs, having the labels present is an advantage.

Can this be fixed / old behaviour restored?

Thanks.

Signature · 3 lines
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line in suburbia: sync at 12.1Mbps down 622kbps up
According to speedtest.net: 11.9Mbps down 500kbps up
#1 lore raw

Hi Russel,

Quoted 25 lines · Russell King - ARM Linux admin

Hi,

Patch d7235f5feaa0 ("leds: gpio: Use generic support for composing LED names") changed the way labels are used.

devm_fwnode_get_gpiod_from_child() is passed led.name, which used to be set to the label or node name, resulting in /sys/kernel/debug/gpio containing useful information. For example:

 gpio-487 (                    |udpu:green:power    ) out lo
 gpio-488 (                    |udpu:red:power      ) out hi
 gpio-489 (                    |udpu:green:network  ) out hi
 gpio-490 (                    |udpu:red:network    ) out hi
 gpio-491 (                    |udpu:green:alarm    ) out hi
 gpio-492 (                    |udpu:red:alarm      ) out hi

After this commit, it now contains a rather less useful:

 gpio-487 (                    |?                   ) out hi ACTIVE LOW
 gpio-488 (                    |?                   ) out hi ACTIVE LOW
 gpio-489 (                    |?                   ) out hi ACTIVE LOW
 gpio-490 (                    |?                   ) out hi ACTIVE LOW
 gpio-491 (                    |?                   ) out hi ACTIVE LOW
 gpio-492 (                    |?                   ) out hi ACTIVE LOW
Quoted 2 lines

This is because led.name is now NULL prior to the call to devm_fwnode_get_gpiod_from_child().

Ugh, my bad. But it actually has been wrong since the patch a96aa64cb5723 ("leds/led-class: Handle LEDs with the same name"), which enabled LED core to amend the LED name just beforehand the registration of a LED class device.

It seem that the only proper solution would be introducing a new pre_register_cdev(const char *name) op to the LED core, that would allow drivers to come up with their implementation thereof. In this particular case leds-gpio driver would need to put there gpiod initialization.

The pre_register_device op would need to be called (when available) from the led_classdev_register_ext() after calling led_classdev_next_name() and before device_create_with_groups(). The op would have to be passed the final_name variable then.

Quoted 5 lines

While this is not an API, it is useful for debugging, and given the dynamic allocation of GPIOs to physical GPIOs, having the labels present is an advantage.

Can this be fixed / old behaviour restored?

If there are no objections I can come up with a fix, but no sooner than at the weekend.

Signature · 2 lines
Best regards,
Jacek Anaszewski
#2 lore raw
Quoted 4 lines · Jacek Anaszewski

It seem that the only proper solution would be introducing a new pre_register_cdev(const char *name) op to the LED core, that would allow drivers to come up with their implementation thereof. In this particular case leds-gpio driver would need to put there gpiod initialization.

I don't know the particulars but if it helps you can at any point call:
int gpiod_set_consumer_name(struct gpio_desc *desc, const char *name);

For any successfully obtained descriptor, to set the consumer name, i.e. the text appearing in debugfs and the GPIO character device.

Yours,
Linus Walleij

#3 lore raw
Quoted 8 lines · Linus Walleij
Quoted 4 lines · Jacek Anaszewski

It seem that the only proper solution would be introducing a new pre_register_cdev(const char *name) op to the LED core, that would allow drivers to come up with their implementation thereof. In this particular case leds-gpio driver would need to put there gpiod initialization.

Quoted 6 lines

I don't know the particulars but if it helps you can at any point call:
int gpiod_set_consumer_name(struct gpio_desc *desc, const char *name);

For any successfully obtained descriptor, to set the consumer name, i.e. the text appearing in debugfs and the GPIO character device.

Thanks for the pointer. I've just used that to fix the issue.

Signature · 2 lines
Best regards,
Jacek Anaszewski