Am 08.03.2017 um 10:13 schrieb Rafał Miłecki:
> Hi,
>
> I backported commit f363a870eaef ("leds: core: use deferred probing if
> default trigger isn't available yet") to my standard kernel and
> noticed some problems when using it.
>
> First of all my LEDs stopped working despite having all triggers
> enabled. It has appeared to be caused by my DTS files using invalid
> "default-off" trigger. I already sent patches fixing this:
> [1/2] ARM: dts: BCM53573: Don't use nonexistent "default-off" LED trigger
> [2/2] ARM: dts: BCM5301X: Don't use nonexistent "default-off" LED trigger
> but maybe there are more devices regressed by this change? Does it
> count as breaking DT files compatibility?
>
> Related problem is that with this change user won't get any LED
> working even if there is only a single trigger missing. People may
> decide not to use some triggers for various reasons. Let's say I don't
> care about USB support and I disable whole USB subsystem including
> "usbport" trigger. If my DTS file includes "userport" default trigger
> for any LED, I won't get any LED registered at all. I'd definitely
> want to have other LEDs working.
>
> Finally this commit introduces delay when e.g. mixing built-in with
> modules. Thanks to DTS files a lot of devices can share a single
> kernel. I don't want to bloat it (the kernel) so I keep subsystem
> modules & extra triggers as modules. I built-in leds-gpio and
> ledtrig-default-on as they are used by 99% of devices and are needed
> early to indicate device/kernel is alive. With the commit in question
> LEDs don't get registered/enabled until are triggers are ready and
> some triggers need to be loaded as modules from rootfs. This delays
> settings LEDs to some initial state indicating e.g. device is powered
> on.
>
> Because of these problems I'd like to suggest reverting that commit
> and look for a different solution.
>
> Unfortunately first of all I need to ask for a better explanation of
> the problem. I read commit description but this load order doesn't
> make sense to me. If you take a look at led_trigger_register, you'll
> see it iterates over registered LEDs and enables just-loaded trigger
> for these using it as default one. I also just tested this feature and
> it works as expected for me. I loaded ledtrig-timer.ko and LED with
> linux,default-trigger = "timer" started blinking nicely!
>
Uuhh, you're absolutely right, my fault.
Based on the code in led_classdev_register I assumed that there must
be an issue with triggers exported by drivers loaded later.
And I only tested that there was no such issue after applying the patch.
But I didn't test that the issue actually existed before.
So yes, the patch can be reverted.
But there's at least something good with it: It helped to clean up
some DT's ..
Sorry again and thanks for the analysis, Heiner
> Heiner: can you describe what exactly was a problem for you? Why you
> needed deffer solution and why you decided to keep setting trigger in
> led_trigger_register function? They seems like 2 competing solutions
> of the same problem.
>
