Re: Fwd: sysfs: cannot create duplicate filename .../system76_acpi::kbd_backlight/color

[Takashi Iwai <tiwai@suse.de>]

On Tue, 21 Nov 2023 10:19:03 +0100,
Thorsten Leemhuis wrote:
> 
> Takashi, Jean-Jacques Hiblot, Lee,
> 
> On 20.11.23 14:53, Takashi Iwai wrote:
> > On Mon, 06 Nov 2023 14:19:08 +0100,
> > Bagas Sanjaya wrote:
> >> On Sat, Nov 04, 2023 at 01:01:56PM +0100, Jean-Jacques Hiblot wrote:
> >>> On 29/10/2023 02:48, Bagas Sanjaya wrote:
> >>>> On Thu, Oct 26, 2023 at 02:55:06PM +0700, Bagas Sanjaya wrote:
> >>>>> The culprit seems to be commit c7d80059b086c4986cd994a1973ec7a5d75f8eea, which introduces a new 'color' attribute for led sysfs class devices. The problem is that the system76-acpi platform driver tries to create the exact same sysfs attribute itself for the system76_acpi::kbd_backlight device, leading to the conflict. For testing purposes, I've just rebuilt the kernel with the system76-apci color attribute renamed to kb_color, and that fixes the issue.
> >>>>
> >>>> Jean-Jacques Hiblot, would you like to take a look on this regression,
> >>>> since you authored the culprit?
> >
> >>> The offending commit stores the color in struct led_classdev and exposes it
> >>> via sysfs. It was part of a series that create a RGB leds from multiple
> >>> single-color LEDs. for this series, we need the color information but we
> >>> don't really need to expose it it via sysfs. In order to fix the issue, we
> >>> can remove the 'color' attribute from the sysfs.
> >>
> >> OK, see you in the patch!
> > 
> > Is there a patch available?
> 
> Not that I know of. Could not find anything on lore either.
> 
> > This bug hits for a few Logitech keyboard models, too, and it makes
> > 6.6 kernel unsable for them, as hid-lg-g15 driver probe fails due to
> > this bug:
> >   https://bugzilla.kernel.org/show_bug.cgi?id=218155
> > 
> > We need a quick fix for 6.6.x.
> 
> Given that Jean-Jacques Hiblot (the author of the culprit) and Lee (who
> committed it and sent it to Linus) know about this for a while already
> without doing anything about it, I wonder if someone should just send a
> revert to Linus (unless of course that is likely to introduce a
> regression on its own).
> 
> Takashi, could you maybe do this, unless a fix shows up real soon?

I can, but we need to decide which way to go.
There are several options:

1. Revert the commit c7d80059b086;
   this drops led class color sysfs entries.  Also the store of
   led_cdev->color from fwnode is dropped, too.

2. Drop only led class color sysfs entries;
   a partial revert of c7d80059b086 above

3. Rename conflicting sysfs entries in drivers;
   e.g. color -> kb_color for hid-lg-g15 and system76_acpi

In either way, we'd break user-space (sysfs).

IMO, 2 would be the least harm, as the class sysfs entry was
introduced since 6.6.  I guess this is what Jean-Jacques suggested.
But I'm not sure how important this new class sysfs entry is; it has
to be clarified from leds people who actually use / wanted the
feature.

3 was already confirmed to work on both bug reports.  OTOH, it's not
clear whether we can neglect (potentially) already existing users.


Takashi

> 
> Ciao, Thorsten (wearing his 'the Linux kernel's regression tracker' hat)
> 
> P.S.: /me should have followed up on this earlier... :-/
> --
> Everything you wanna know about Linux kernel regression tracking:
> https://linux-regtracking.leemhuis.info/about/#tldr
> If I did something stupid, please tell me, as explained on that page.
>