Re: [PATCH] platform/x86: int3472: handle GPIO type 0x08 as a power regulator

[Hans de Goede]

Hi,

On 20-Jul-26 00:00, Sakari Ailus wrote:
> Hi Jakob,
> 
> Thanks for the patch.
> 
> On Sun, Jul 19, 2026 at 08:15:12PM +0200, Jakob Berg Jespersen wrote:
>> On the Microsoft Surface Pro 7+, the rear camera's INT3472 device
>> (INT3472:01, sensor INT347A / ov8865) enumerates a GPIO of type 0x08
>> in its _DSM which the driver does not recognise, so it is left
>> unmapped:
>>
>>   int3472-discrete INT3472:01: GPIO type 0x08 unknown; the sensor may
>>   not work
>>
>> The DSDT shows this pin (0x0100AF08, pin 175, active-high) sitting
>> alongside a normal type-0x0b power-enable GPIO: it gates a second
>> sensor power rail. Because the type is unknown the rail is never
>> enabled, the ov8865's "dvdd" supply resolves to a dummy regulator, and
>> the sensor never probes.
>>
>> Map type 0x08 for INT347A to a POWER_ENABLE regulator with con_id
>> "dvdd", the supply the in-tree ov8865 driver already requests, so no
>> sensor-driver change is needed. ("dvdd" rather than "avdd" because the
>> type-0x0b power-enable pin on this INT3472 is already registered as
>> "avdd".)
>>
>> Type 0x08 is not in the driver's documented function list and is not
>> otherwise handled by mainline; based on the sensor's supply set and the
>> con_id that brings it up, it appears to be the DVDD (digital core
>> voltage) rail, the counterpart to the existing DOVDD (0x10) type.
>>
>> Related out-of-tree work approached the same rail by adding a new "pwr1"
>> supply instead of the sensor's existing "dvdd" (linux-surface PR #1867
>> for the Surface Pro 9, PR #2201 for the Pro 7+ ov8865); neither was sent
>> upstream. Mapping to "dvdd" keeps the change contained to int3472.
>>
>> With this change the SP7+ rear camera probes and streams reliably,
>> validated by over a month of use.
>>
>> Link: https://github.com/linux-surface/linux-surface/pull/1867
>> Link: https://github.com/linux-surface/linux-surface/pull/2201
>> Cc: Tooraj Taraz <tooraj.taraz@xxxxxxxxx>
>> Cc: Joseph V. Lavigne <jlavig88@xxxxxxxxx>
>> Signed-off-by: Jakob Berg Jespersen <dev@xxxxxxx>
>> ---
>> 0x08 is left as a bare literal rather than a named constant here; if
>> its canonical meaning is known I'm happy to add e.g.
>> INT3472_GPIO_TYPE_DVDD and/or handle it generically like DOVDD in v2.
>>
>> This is my first kernel contribution, so please bear with me on any
>> process mistakes.
>> ---
>>  drivers/platform/x86/intel/int3472/discrete.c | 13 +++++++++++++
>>  1 file changed, 13 insertions(+)
>>
>> diff --git a/drivers/platform/x86/intel/int3472/discrete.c b/drivers/platform/x86/intel/int3472/discrete.c
>> index 115bb37577a1..98c7b9bc393f 100644
>> --- a/drivers/platform/x86/intel/int3472/discrete.c
>> +++ b/drivers/platform/x86/intel/int3472/discrete.c
>> @@ -164,6 +164,19 @@ static const struct int3472_gpio_map int3472_gpio_map[] = {
>>  		.con_id = "dvdd",
>>  		.enable_time_us = 45 * USEC_PER_MSEC,
>>  	},
>> +	{	/*
>> +		 * Surface Pro 7+ ov8865 (rear camera): type 0x08 is an
>> +		 * undocumented sensor power rail; left undriven the sensor
>> +		 * stays in a failed power state. "dvdd" rather than "avdd"
>> +		 * because this INT3472 also has a type 0x0b (power enable)
>> +		 * pin already registered as "avdd".
>> +		 */
> 
> GPIO type 8 is in fact "POWER1" (type 7 being "POWER0"). This hasn't been
> seen anywhere yet so support for it hasn't been added either.
> 
> I understand the POWER_ENABLE GPIO is for executing the entire power
> sequence and isn't expected to be found with the other regulators, hence an
> apparent conflict here wouldn't be an issue.
> 
> I think for now we could map type POWER1 (8) to "dvdd", however not using a
> quirk but doing it for all devices, like mapping POWER_ENABLE to "avdd".
> That'd leave mapping POWER0 to "avdd" for later. The GPIO definition for
> POWER0 could be added in any case.

I was wondering what the internal Intel headers said for type 8, so thank
you for clarifying that that is called POWER1. Adding a generic mapping
of POWER1 to "dvdd" instead of using a quirk sounds good to me.

Regards,

Hans