Re: [PATCH] platform/x86: int3472: handle GPIO type 0x08 as a power regulator
From: Dan Scally
Date: Mon Jul 20 2026 - 17:22:25 EST
Hi all
On 19/07/2026 23:00, 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 wonder what Hans or Dan think.
Yep, sounds like a good idea to me.
Thanks
Dan
+ .hid = "INT347A",
+ .type_from = 0x08,
+ .type_to = INT3472_GPIO_TYPE_POWER_ENABLE,
+ .con_id = "dvdd",
+ .enable_time_us = GPIO_REGULATOR_ENABLE_TIME,
+ },
};
static void int3472_get_con_id_and_polarity(struct int3472_discrete_device *int3472, u8 *type,