Re: Regression: SYNA3602 I2C touchpad broken in Linux 6.17.7 (works in 6.17.6 and previous versions)

[Benjamin Tissoires <benjamin.tissoires@redhat.com>]

Hi,

On Fri, Nov 28, 2025 at 7:40 AM Vijay <vijayg0127@gmail.com> wrote:
>
> Hello,
>
> I would like to report a regression in the Linux kernel affecting I2C-HID
> touchpads that run through the Intel ISH + DesignWare I2C controller.
>
> Hardware:
> - Laptop: Infinix Y4 Max
> - CPU: Intel (13th gen core i5)
> - Touchpad: SYNA3602:00 093A:35ED (I2C HID)
> - Bus path: SYNA3602 → i2c_designware → Intel ISH → HID
> - OS: Linux (Arch/CachyOS)
> - Kernel config: Default distro config
>
> Regression summary:
> - Touchpad works perfectly in Linux 6.17.6 and below versions
> - Touchpad stops working in Linux 6.17.7 and all newer versions (6.17.8, 6.17.9, etc.)
> - Desktop environment does not matter (Hyprland/GNOME both fail)
> - The failure happens before userspace loads
> - Touchpad also works fine in Linux 6.12 LTS
>
> This is a kernel-level regression introduced between:
>     Good: Linux 6.17.6
>     Bad:  Linux 6.17.7
>
> **Dmesg logs from broken kernel (6.17.7 and newer):**
>
>     i2c-SYNA3602:00: can't add hid device: -110
>     hid_sensor_hub: reading report descriptor failed
>     intel-hid INTC1078:00: failed to enable HID power button

Looks like i2c-hid can't even communicate with any I2C device, so this
is slightly worrying.

>
> And the DesignWare I2C controller logs around the failure:
>     i2c_designware 0000:00:15.0: controller timed out
>     i2c_designware 0000:00:15.0: lost arbitration
>     i2c_designware 0000:00:15.0: transfer aborted (status = -110)
>
> These errors appear only on 6.17.7+ and not on 6.17.6.
>
> On working versions (6.17.6 and 6.12 LTS), the touchpad initializes normally:
>
>     input: SYNA3602:00 093A:35ED Touchpad as /devices/.../input/inputX
>     hid-multitouch: I2C HID v1.00 device initialized
>     i2c_designware 0000:00:15.0: controller operating normally
>
> This narrow regression window should make it possible to identify the offending
> change in either:
> - HID core
> - I2C-HID
> - Intel ISH HID
> - DesignWare I2C controller
> - ACPI timing changes
>
> I can provide:
> - Full dmesg (working and broken)
> - acpidump

Are you running on a full vanilla kernel?

The changelog between 6.17.6 and 6.17.7 is rather small, so it should
be easy enough to bisect and get the offending commit.

I have my suspicions on:
f1971d5ba2ef ("genirq/manage: Add buslock back in to enable_irq()")
b990b4c6ea6b ("genirq/manage: Add buslock back in to __disable_irq_nosync()")
3c97437239df ("genirq/chip: Add buslock back in to irq_set_handler()")

Because anything else is unrelated to any component involved in i2c-hid.
(But that's also assuming you are running vanilla kernels without any
extra patches.)

OTOH, I've booted a 6.17.8 and 6.17.7 shipped by Fedora and I don't
see any issues related to i2c-hid, so those 3 commits might not be the
culprits.


>
> Please let me know what additional data is needed.

Can you do a bisect between v6.17.7 and v6.17.6?

Cheers,
Benjamin

>
> Thank you,
> Vijay.