Re: [PATCH v3 0/5] Add SW_TABLET_MODE support for AMD SFH convertibles

On Mon, 3 Aug 2026, Basavaraj Natikar wrote:

> Add SW_TABLET_MODE support for AMD convertibles whose operating mode
> (laptop or tablet) is reported by the Sensor Fusion Hub.
> 
> Patches 1 and 2 are preparation: the first tracks the MP2 version
> explicitly and the second serializes access to the shared emp2 pointer.
> The third adds a way to read the operating mode. The fourth registers an
> auxiliary device when the sensor is present, and the fifth is a small
> input driver that binds to it and reports the posture to userspace.
> Keeping the input handling in its own driver avoids pulling input/evdev
> into the sensor transport driver.
> 
> Changes in v3:
> - Destroy the tablet-mode auxiliary device before the hardware teardown
>   in .remove(), instead of from a devres action that runs afterwards,
>   closing a use-after-teardown window (Sashiko review).
> - Move the Makefile entry to its correct alphabetical position
>   (Dmitry Torokhov).
> 
> Changes in v2 (address the Sashiko review of v1):
> - Read the operating-mode register only on confirmed MP2 v2, by tracking
>   the MP2 version explicitly. On v1.0 that register overlaps the 64-bit
>   DMA address programmed through C2P, so the earlier unconditional read
>   could return DMA bits instead of the operating mode.
> - Serialize the shared emp2 pointer and all exported readers under a
>   mutex, so a concurrent unbind can no longer free the device while a
>   reader is still dereferencing it.
> - Publish emp2 only after initialization has populated the client data,
>   and clear it early on the MP2 v2 remove path instead of from a
>   prematurely registered devres action; this removes both the
>   publish-before-init data race and the unconditional global clear on a
>   probe failure.
> - Destroy the tablet-mode auxiliary device and release its id on removal,
>   and allocate the id dynamically so multiple instances cannot collide.
> - Report the initial tablet-mode state before registering the input
>   device, so userspace never observes a stale default posture.
> - Rename the error variable from 'rc' to 'error' to follow the
>   input-subsystem convention.
> 
> v2: https://lore.kernel.org/all/[email protected]/
> v1: https://lore.kernel.org/all/[email protected]/

Now queued in hid.git#for-7.3/amd-sfh, sorry for the delay. Myself and 
Benjamin have been on a few weeks on leave in parallel, and then didn't 
really help with the throughput ... still recovering :)

-- 
Jiri Kosina
SUSE Labs