Re: [PATCH v2 2/3] Input: aw86938 - add driver for Awinic AW86938

[Dmitry Torokhov <dmitry.torokhov@gmail.com>]

On Mon, Feb 02, 2026 at 04:11:51PM +0100, Konrad Dybcio wrote:
> On 2/2/26 12:04 PM, Dmitry Torokhov wrote:
> > On Mon, Feb 02, 2026 at 11:19:36AM +0100, Konrad Dybcio wrote:
> >> On 2/2/26 11:14 AM, Luca Weiss wrote:
> >>> Hi Konrad,
> >>>
> >>> On Mon Feb 2, 2026 at 11:12 AM CET, Konrad Dybcio wrote:
> >>>> On 2/1/26 2:49 AM, Dmitry Torokhov wrote:
> >>>>> Hi Griffin,
> >>>>>
> >>>>> On Wed, Jan 28, 2026 at 04:51:14PM +0100, Griffin Kroah-Hartman wrote:
> >>>>>> @@ -717,9 +746,19 @@ static int aw86927_detect(struct aw86927_data *haptics)
> >>>>>>  
> >>>>>>  	chip_id = be16_to_cpu(read_buf);
> >>>>>>  
> >>>>>> -	if (chip_id != AW86927_CHIPID) {
> >>>>>> -		dev_err(haptics->dev, "Unexpected CHIPID value 0x%x\n", chip_id);
> >>>>>> -		return -ENODEV;
> >>>>>> +	switch (haptics->model) {
> >>>>>> +	case AW86927:
> >>>>>> +		if (chip_id != AW86927_CHIPID) {
> >>>>>> +			dev_err(haptics->dev, "Unexpected CHIPID value 0x%x\n", chip_id);
> >>>>>> +			return -ENODEV;
> >>>>>> +		}
> >>>>>
> >>>>> If we are able to query chip ID why do we need to have separate
> >>>>> compatibles? I would define chip data structure with differences between
> >>>>> variants and assign and use it instead of having separate compatible.
> >>>>
> >>>> dt-bindings guidelines explicitly call for this, a chipid comparison
> >>>> then works as a safety net
> >>>
> >>> Are you saying, that
> >>>
> >>> 1. we should enforce dt-bindings == CHIP_ID (what's currently done)
> >>
> >> This
> > 
> > No. If there is a compatible chip with different ID (for whatever reason
> > - maybe there is additional functionality that either board does not
> > need or the driver does not implement) we absolutely should not refuse
> > to bind the driver.
> > 
> > Hint: this thing is called _compatible_ for a reason.
> 
> Right, the reason you have in mind is fulfilled by fallback compatibles
> 
> (i.e. "vendor,actualchipname", "vendor,similarchipname" where the driver
> only considers the latter becuase the software interface hasn't changed)

And having chip_id checks will break this...

> 
> > 
> >>
> >>>
> >>> or
> >>>
> >>> 2. we should have both compatibles with no handling based on compatible,
> >>>    but only use CHIP_ID at runtime to change behavior
> >>
> >> This is spaghetti
> > 
> > I really do not understand the aversion of DT maintainers to generic
> > compatibles. We see this in I2C HID where we keep adding compatibles
> > for what could be described via device properties.
> 
> This is because it's the only way to allow for retroactive changes that
> do not require changing firmware. That's why ACPI carries new identifiers
> for even very slightly different devices too. Once the firmware containing
> (ACPI tables / DTB) is put on a production device, it is generally not
> going to ever change.

They are actually solving slightly different problem. In ACPI world they
allocate a new ID to represent a peripheral in a given design, down to
it's firmware behavior. It encodes much more than chip ID that DT
maintainers want to key off of.

> 
> CHIP_ID registers are a good tool to validate that the author of the
> firmware table is doing the right thing, but solely relying on them
> encourages creating a "vendor,haptic" compatible, which I'm sure you'll
> agree is totally meaningless.

Is it? If a piece of hardware speaks i2c-hid protocol why do I need to
know the exact chip that is being used? Depending on the chassis and the
size of the sensing element and the version of the firmware that is
loaded into it the behavior and timings of the same chip may be very
different.

> 
> That's especially if the naming scheme makes no sense and you can't
> even factor out a common wildcard-name (which also happens to be the case
> quite often)
> 
> Plus a compatible is used to restrict/modify the set of allowed/required
> properties, so having an "actual" compatible is required for schema
> validation to work

Yes, in cases where there is not a common set of properties having
different compatibles makes sense. But in cases when the device is
supposed to have vendor-agnostic behavior insisting on myriad
compatibles makes little sense.

Thanks.

-- 
Dmitry