Skip to contentSkip to conversation

issue with (parent)class for exported channels in sysfs

PATCH Bart De Vos 2 participants 4 messages Mar 20, 2019 PWM

lore
Share & tools
lore
#0 lore raw

Hi,

The commit below causes the following side-effect:

symlinks are created for each individual pwm, as follows (conceptually):

/sys/class/pwm/pwm0 -> /sys/class/pwm/pwmchip0/pwm0

This fails for exporting the next pwmchip, since pwmchips enumerate their channels from 0 ... <npwm>.

I can work around it by putting the assignment of the class around a CONFIG_PWM_PARENT_CLASS, since I don't use uevents on my embedded system for pwms.

Yet I believe both use-cases should be possible, i.e. set the right properties to support proper uevents _AND_
export multiple channels from multiple chips over sysfs.

What are your thoughts on a good way forward?

Thanks in advance for your feedback.

Kr,
Bart.

commit 7e5d1fd75c3dde9fc10c4472b9368089d1b81d00

Author: Gottfried Haider <[email protected]>
Date:   Tue Sep 26 11:59:51 2017 +0000

pwm: Set class for exported channels in sysfs

Notifications for devices without bus or class set get dropped by dev_uevent_filter(). Adding the class to the exported child matches what the GPIO subsystem is doing.

With this change exporting a channel triggers a udev event, which gives userspace a chance to fixup permissions and makes it possible for non-root users to make use of the PWM subsystem.

drivers/pwm/sysfs.c +1 -0
index a813239300c3..83f2b0b15712 100644
@@ -263,6 +263,7 @@ static int pwm_export_child(struct device *parent, struct pwm_device *pwm)
        export->pwm = pwm;
        mutex_init(&export->lock);
 
+       export->child.class = parent->class;
        export->child.release = pwm_export_release;
        export->child.parent = parent;
        export->child.devt = MKDEV(0, 0);
#1 lore raw
Quoted 12 lines · Bart De Vos

Hi,

The commit below causes the following side-effect:

symlinks are created for each individual pwm, as follows (conceptually):

/sys/class/pwm/pwm0 -> /sys/class/pwm/pwmchip0/pwm0

This fails for exporting the next pwmchip, since pwmchips enumerate their channels from 0 ... <npwm>.

I can work around it by putting the assignment of the class around a CONFIG_PWM_PARENT_CLASS, since I don't use uevents on my embedded system for pwms.

Quoted 8 lines

Yet I believe both use-cases should be possible, i.e. set the right properties to support proper uevents _AND_
export multiple channels from multiple chips over sysfs.

What are your thoughts on a good way forward?

Thanks in advance for your feedback.

Hello Bart,

What kernel version you use? It seems it must be something older than v4.19 as this issue has already been fixed and the commit you reffer to has been reverted in v4.19-rc1. See:

See commit 552c02e3e7cf ("pwm: Send a uevent on the pwmchip device upon channel sysfs (un)export") and commit c289d6625237 ("Revert "pwm: Set class for exported channels in sysfs"").

So a good way forward is to use more recent kernel :)

All the best,
Michal

Quoted 35 lines

commit 7e5d1fd75c3dde9fc10c4472b9368089d1b81d00

Author: Gottfried Haider <[email protected]>
Date:   Tue Sep 26 11:59:51 2017 +0000

pwm: Set class for exported channels in sysfs

Notifications for devices without bus or class set get dropped by dev_uevent_filter(). Adding the class to the exported child matches what the GPIO subsystem is doing.

With this change exporting a channel triggers a udev event, which gives userspace a chance to fixup permissions and makes it possible for non-root users to make use of the PWM subsystem.

drivers/pwm/sysfs.c +1 -0
index a813239300c3..83f2b0b15712 100644
@@ -263,6 +263,7 @@ static int pwm_export_child(struct device *parent, struct pwm_device *pwm)
         export->pwm = pwm;
         mutex_init(&export->lock);
 
+       export->child.class = parent->class;
         export->child.release = pwm_export_release;
         export->child.parent = parent;
         export->child.devt = MKDEV(0, 0);
#2 lore raw

Okay,

I saw in the meantime it has been solved in mainstream, and could easily backport that part.

but ...

Your mentioning on versions confuses me: I'm running 4.19.28, I would expect that commit 552c02e3e7cf is part of that.

Anyhow, issue solved, thanks for your quick reply.

Kr,
Bart.

Quoted 74 lines · Michal Vokáč
Quoted 12 lines · Bart De Vos

Hi,

The commit below causes the following side-effect:

symlinks are created for each individual pwm, as follows (conceptually):

/sys/class/pwm/pwm0 -> /sys/class/pwm/pwmchip0/pwm0

This fails for exporting the next pwmchip, since pwmchips enumerate their channels from 0 ... <npwm>.

I can work around it by putting the assignment of the class around a CONFIG_PWM_PARENT_CLASS, since I don't use uevents on my embedded system for pwms.

Quoted 8 lines

Yet I believe both use-cases should be possible, i.e. set the right properties to support proper uevents _AND_
export multiple channels from multiple chips over sysfs.

What are your thoughts on a good way forward?

Thanks in advance for your feedback.

Hello Bart,

What kernel version you use? It seems it must be something older than v4.19 as this issue has already been fixed and the commit you reffer to has been reverted in v4.19-rc1. See:

See commit 552c02e3e7cf ("pwm: Send a uevent on the pwmchip device upon channel sysfs (un)export") and commit c289d6625237 ("Revert "pwm: Set class for exported channels in sysfs"").

So a good way forward is to use more recent kernel :)

All the best,
Michal

Quoted 36 lines

commit 7e5d1fd75c3dde9fc10c4472b9368089d1b81d00
Author: Gottfried Haider <[email protected]> Date:   Tue Sep 26 11:59:51 2017 +0000

     pwm: Set class for exported channels in sysfs

     Notifications for devices without bus or class set get dropped by dev_uevent_filter(). Adding the class to the exported child matches what the GPIO subsystem is doing.

     With this change exporting a channel triggers a udev event, which gives userspace a chance to fixup permissions and makes it possible for non-root users to make use of the PWM subsystem.

drivers/pwm/sysfs.c +1 -0
index a813239300c3..83f2b0b15712 100644
@@ -263,6 +263,7 @@ static int pwm_export_child(struct device *parent, 
 struct pwm_device *pwm)
          export->pwm = pwm;
          mutex_init(&export->lock);
 
+       export->child.class = parent->class;
          export->child.release = pwm_export_release;
          export->child.parent = parent;

         export->child.devt = MKDEV(0, 0);

#3 lore raw
Quoted 7 lines · Bart De Vos

Okay,

I saw in the meantime it has been solved in mainstream, and could easily backport that part.

but ...

Your mentioning on versions confuses me: I'm running 4.19.28, I would expect that commit 552c02e3e7cf is part of that.

My bad, git describe 552c02e3e7cf returned: v4.19-rc1-15-g552c02e3e7cf.

That does not mean it is in v4.19. It is actually fixed in v4.20. Sorry for the confusion.

Quoted 1 line

Anyhow, issue solved, thanks for your quick reply.

You're welcome.
Michal