Message-ID: 
 <CAGwozwF_eWE+yQ9DReO58NruLAnaM_PDEdNYOzt4-UgOcdKWTg@mail.gmail.com>
Date: Fri, 17 Oct 2025 13:21:38 +0200
From: Antheas Kapenekakis <lkml@...heas.dev>
To: Denis Benato <benato.denis96@...il.com>
Cc: Ilpo Järvinen <ilpo.jarvinen@...ux.intel.com>,
	platform-driver-x86@...r.kernel.org, linux-input@...r.kernel.org,
	LKML <linux-kernel@...r.kernel.org>, Jiri Kosina <jikos@...nel.org>,
	Benjamin Tissoires <bentiss@...nel.org>,
 Corentin Chary <corentin.chary@...il.com>,
	"Luke D . Jones" <luke@...nes.dev>, Hans de Goede <hdegoede@...hat.com>
Subject: Re: [PATCH v6 0/7] HID: asus: Fix ASUS ROG Laptop's Keyboard
 backlight handling

On Fri, 17 Oct 2025 at 13:00, Denis Benato <benato.denis96@...il.com> wrote:
>
>
> On 10/17/25 09:54, Antheas Kapenekakis wrote:
> > On Thu, 16 Oct 2025 at 18:16, Antheas Kapenekakis <lkml@...heas.dev> wrote:
> >> On Thu, 16 Oct 2025 at 17:09, Ilpo Järvinen
> >> <ilpo.jarvinen@...ux.intel.com> wrote:
> >>> On Thu, 16 Oct 2025, Antheas Kapenekakis wrote:
> >>>> On Thu, 16 Oct 2025 at 13:57, Denis Benato <benato.denis96@...il.com> wrote:
> >>>>> On 10/13/25 22:15, Antheas Kapenekakis wrote:
> >>>>>> This is a two part series which does the following:
> >>>>>>   - Clean-up init sequence
> >>>>>>   - Unify backlight handling to happen under asus-wmi so that all Aura
> >>>>>>     devices have synced brightness controls and the backlight button works
> >>>>>>     properly when it is on a USB laptop keyboard instead of one w/ WMI.
> >>>>>>
> >>>>>> For more context, see cover letter of V1. Since V5, I removed some patches
> >>>>>> to make this easier to merge.
> >>>>>>
> >>>>>> All comments with these patches had been addressed since V4.
> >>>>> I have loaded this patchset for users of asus-linux project to try out.
> >>>>>
> >>>>> One of them opened a bug report about a kernel bug that happens
> >>>>> consistently when closing the lid of his laptop [1].
> >>>>>
> >>>>> He also sent another piece of kernel log, but didn't specify anything more
> >>>>> about this [2].
> >>>>>
> >>>>> [1] https://pastebin.com/akZx1w10
> >>>>> [2] https://pastebin.com/sKdczPgf
> >>>> Can you provide a link to the bug report? [2] seems unrelated.
> >>>>
> >>>> As for [1], it looks like a trace that stems from a sysfs write to
> >>>> brightness stemming from userspace that follows the same chain it
> >>>> would on a stock kernel and times out. Is it present on a stock
> >>>> kernel?
> >>>>
> >>>> Ilpo should know more about this, could the spinlock be interfering?
> >>> [1] certainly seems to do schedule() from do_kbd_led_set() so it's not
> >>> possible to use spinlock there.
> >>>
> >>> So we're back to what requires the spinlock? And what the spinlock
> >>> protects?
> >> For that invocation, since it is coming from the cdev device owned by
> >> asus_wmi, it protects asus_ref.listeners under do_kbd_led_set.
> >> asus_wmi is protected by the fact it is owned by that device. Spinlock
> >> is not required in this invocation due to not being an IRQ.
> >>
> >> Under asus_hid_event (second to last patch), which is called from an
> >> IRQ, a spinlock is required for protecting both listeners and the
> >> asus_ref.asus, and I suspect that scheduling from an IRQ is not
> >> allowed either. Is that correct?
> > So it is a bit tricky here. When the IRQ fires, it needs to know
> > whether asus-wmi will handle the keyboard brightness event so that it
> > falls back to emitting it.
> >
> > If we want it to know for sure, it needs to access asus_wmi, so it
> > needs a spinlock or an IRQ friendly lock. This way, currently,
> > asus_hid_event will return -EBUSY if there is no led device so the
> > event propagates through hid.
> >
> > If we say that it is good enough to know that it was compiled with
> > IS_REACHABLE(CONFIG_ASUS_WMI), ie the actual implementation of
> > asus_hid_event in asus-wmi will never return an error, then,
> > asus_hid_event can schedule a task to fire the event without a lock,
> > and that task can use a normal locking primitive.
> >
> > If the task needs to be assigned to a device or have a handle,
> > asus_hid_listener can be provided to asus_hid_event, so that it is
> > owned by the calling device.
> >
> > What would the appropriate locking primitive be in this case?
> The right place to look into appears to be Documentation/kernel-hacking/locking.rst
>
> I see mutex being used in various irq handlers, even bmi323-imu but that page has
> many alternatives for irq.
>
> There is rwlock_t but it appears to be using spinlock on certain configurations.
>
> Absolute worst-case scenario you might resort implementing an rwlock with mutexes only.
>
> I would avoid taking decisions based solely on the configuration because ASUS
> makes keyboards and reuses designs across various products, so it is very likely
> at least one model of those keyboards can be confused with a laptop one.
>
> Beside I am sure there must be at the very least one appropriate synchronization primitive,
> so I would discard the configure option.

The bug is not due to the spinlock. Its due to accessing WMI through
an IRQ. A work queue should be used. Using a spinlock is a result of
that omission.

asus-wmi already has workqueues for the other leds but not for the
keyboard one so we can add one.

If we want the IRQ that handles brightness events through
asus_hid_event to know whether a brightness handler is registered
before choosing whether to forward the event, then we will need to use
a spinlock. Since the spinlock protects objects that are setup during
asus-wmi init and removal, this spinlock needs to be in asus-wmi since
hid-asus has a child relationship.

Now, if we assume that hid-asus should be able to handle cases where
asus-wmi is not loaded with proper brightness control and in parallel
support cases where asus-wmi is loaded in a unified manner, then
that's a different class of problem because a device needs to own the
led handler and currently that would be the platform device.

> >> Antheas
> >>> Not related to this particular email in this thread, if the users are
> >>> testing something with different kernels, it's also important to make sure
> >>> that the lockdep configs are enabled in both. As it could be that in one
> >>> kernel lockdep is not enabled and thus it won't do the splat.
> >>>
> >>> --
> >>>  i.
> >>>
> >>>
> >>>> My testing on devices that have WMI led controls is a bit limited
> >>>> unfortunately. However, most of our asus users have been happy with
> >>>> this series for around half a year now.
> >>>>
> >>>>>> ---
> >>>>>> V5: https://lore.kernel.org/all/20250325184601.10990-1-lkml@antheas.dev/
> >>>>>> V4: https://lore.kernel.org/lkml/20250324210151.6042-1-lkml@antheas.dev/
> >>>>>> V3: https://lore.kernel.org/lkml/20250322102804.418000-1-lkml@antheas.dev/
> >>>>>> V2: https://lore.kernel.org/all/20250320220924.5023-1-lkml@antheas.dev/
> >>>>>> V1: https://lore.kernel.org/all/20250319191320.10092-1-lkml@antheas.dev/
> >>>>>>
> >>>>>> Changes since V5:
> >>>>>>   - It's been a long time
> >>>>>>   - Remove addition of RGB as that had some comments I need to work on
> >>>>>>   - Remove folio patch (already merged)
> >>>>>>   - Remove legacy fix patch 11 from V4. There is a small chance that
> >>>>>>     without this patch, some old NKEY keyboards might not respond to
> >>>>>>     RGB commands according to Luke, but the kernel driver does not do
> >>>>>>     RGB currently. The 0x5d init is done by Armoury crate software in
> >>>>>>     Windows. If an issue is found, we can re-add it or just remove patches
> >>>>>>     1/2 before merging. However, init could use the cleanup.
> >>>>>>
> >>>>>> Changes since V4:
> >>>>>>   - Fix KConfig (reported by kernel robot)
> >>>>>>   - Fix Ilpo's nits, if I missed anything lmk
> >>>>>>
> >>>>>> Changes since V3:
> >>>>>>   - Add initializer for 0x5d for old NKEY keyboards until it is verified
> >>>>>>     that it is not needed for their media keys to function.
> >>>>>>   - Cover init in asus-wmi with spinlock as per Hans
> >>>>>>   - If asus-wmi registers WMI handler with brightness, init the brightness
> >>>>>>     in USB Asus keyboards, per Hans.
> >>>>>>   - Change hid handler name to asus-UNIQ:rgb:peripheral to match led class
> >>>>>>   - Fix oops when unregistering asus-wmi by moving unregister outside of
> >>>>>>     the spin lock (but after the asus reference is set to null)
> >>>>>>
> >>>>>> Changes since V2:
> >>>>>>   - Check lazy init succeds in asus-wmi before setting register variable
> >>>>>>   - make explicit check in asus_hid_register_listener for listener existing
> >>>>>>     to avoid re-init
> >>>>>>   - rename asus_brt to asus_hid in most places and harmonize everything
> >>>>>>   - switch to a spinlock instead of a mutex to avoid kernel ooops
> >>>>>>   - fixup hid device quirks to avoid multiple RGB devices while still exposing
> >>>>>>     all input vendor devices. This includes moving rgb init to probe
> >>>>>>     instead of the input_configured callbacks.
> >>>>>>   - Remove fan key (during retest it appears to be 0xae that is already
> >>>>>>     supported by hid-asus)
> >>>>>>   - Never unregister asus::kbd_backlight while asus-wmi is active, as that
> >>>>>>   - removes fds from userspace and breaks backlight functionality. All
> >>>>>>   - current mainline drivers do not support backlight hotplugging, so most
> >>>>>>     userspace software (e.g., KDE, UPower) is built with that assumption.
> >>>>>>     For the Ally, since it disconnects its controller during sleep, this
> >>>>>>     caused the backlight slider to not work in KDE.
> >>>>>>
> >>>>>> Changes since V1:
> >>>>>>   - Add basic RGB support on hid-asus, (Z13/Ally) tested in KDE/Z13
> >>>>>>   - Fix ifdef else having an invalid signature (reported by kernel robot)
> >>>>>>   - Restore input arguments to init and keyboard function so they can
> >>>>>>     be re-used for RGB controls.
> >>>>>>   - Remove Z13 delay (it did not work to fix the touchpad) and replace it
> >>>>>>     with a HID_GROUP_GENERIC quirk to allow hid-multitouch to load. Squash
> >>>>>>     keyboard rename into it.
> >>>>>>   - Unregister brightness listener before removing work queue to avoid
> >>>>>>     a race condition causing corruption
> >>>>>>   - Remove spurious mutex unlock in asus_brt_event
> >>>>>>   - Place mutex lock in kbd_led_set after LED_UNREGISTERING check to avoid
> >>>>>>     relocking the mutex and causing a deadlock when unregistering leds
> >>>>>>   - Add extra check during unregistering to avoid calling unregister when
> >>>>>>     no led device is registered.
> >>>>>>   - Temporarily HID_QUIRK_INPUT_PER_APP from the ROG endpoint as it causes
> >>>>>>     the driver to create 4 RGB handlers per device. I also suspect some
> >>>>>>     extra events sneak through (KDE had the @@@@@@).
> >>>>>>
> >>>>>> Antheas Kapenekakis (7):
> >>>>>>   HID: asus: refactor init sequence per spec
> >>>>>>   HID: asus: prevent binding to all HID devices on ROG
> >>>>>>   platform/x86: asus-wmi: Add support for multiple kbd RGB handlers
> >>>>>>   HID: asus: listen to the asus-wmi brightness device instead of
> >>>>>>     creating one
> >>>>>>   platform/x86: asus-wmi: remove unused keyboard backlight quirk
> >>>>>>   platform/x86: asus-wmi: add keyboard brightness event handler
> >>>>>>   HID: asus: add support for the asus-wmi brightness handler
> >>>>>>
> >>>>>>  drivers/hid/hid-asus.c                     | 235 +++++++++++----------
> >>>>>>  drivers/platform/x86/asus-wmi.c            | 157 ++++++++++++--
> >>>>>>  include/linux/platform_data/x86/asus-wmi.h |  69 +++---
> >>>>>>  3 files changed, 291 insertions(+), 170 deletions(-)
> >>>>>>
> >>>>>>
> >>>>>> base-commit: 3a8660878839faadb4f1a6dd72c3179c1df56787
> >>>
>

Powered by blists - more mailing lists