Skip to contentSkip to conversation

leds: fix regression in usbport led trigger

PATCH Superseded Greg Kroah-Hartman 1 participant 1 message Jan 18, 2019 USB

lore
Share & tools
lore

Greg Kroah-Hartman

author maintainer
#0 lore raw

PATCH v2 1/1 leds: fix regression in usbport led trigger

Quoted 17 lines · Uwe Kleine-König
Quoted 16 lines · Greg Kroah-Hartman
Quoted 14 lines · Christian Lamparter

The patch "usb: simplify usbport trigger" together with "leds: triggers:
add device attribute support" caused an regression for the usbport trigger. it will no longer enumerate any active usb hub ports under the "ports" directory in the sysfs class directory, if the usb host drivers are fully initialized before the usbport trigger was loaded.

The reason is that the usbport driver tries to register the sysfs entries during the activate() callback. And this will fail with -2 / ENOENT because the patch "leds: triggers: add device attribute support" made it so that the sysfs "ports" group was only being added after the activate() callback succeeded.

This version of the patch reverts parts of the "usb: simplify usbport trigger" patch and restores usbport trigger's functionality.

Quoted 6 lines
Quoted 3 lines

This feels like going backwards, as a driver should not be adding and removing sysfs groups, because you race with userspace. Userspace has no idea that the new sysfs files were added or removed, right?

I think this is not an issue as the led trigger core code calls kobject_uevent_env() when the trigger is set.

Ok, that helps a lot then.

Quoted 8 lines

But I still agree that there seems to be something "unusual" about the usb led trigger ...

Instead of providing a sysfs-file per port to make said port trigger the led, I'd prefer a single file that you can use to add and remove ports from the trigger. With this change the driver can properly benefit from the attribute handling of the led-trigger core and is more in line with the other triggers.

But how do you know how many ports are present? And this feels like it ends up being a "custom api" for each different type of led that is present in the system. Or is that already the case?

thanks,

greg k-h