Skip to contentSkip to conversation

platform/x86: int3472: support the POWER1 GPIO type

PATCH Sergey Lebedev 3 participants 5 messages Aug 30 – Aug 31

Subsystems (2)
lore
Share & tools
lore

Sergey Lebedev

author Tested-by
#0 lore raw

Hans pointed me at this from a report I sent this morning about the same GPIO type on a Surface Pro 11 - thank you, and sorry for the duplicate question. I have now tested this patch on that machine, which is a third model and, more usefully, a different sensor. Result below, with the part that is still missing for this sensor family.

What the patch fixes here
-------------------------

Built on 7.0.0-30 (Ubuntu 26.04). The warning is gone and the rail is mapped:

before: int3472-discrete INT3472:00: GPIO type 0x08 unknown;
the sensor may not work
after : no int3472 messages at all

/sys/class/regulator:

    regulator.1  INT3472:00-avdd
    regulator.2  INT3472:00-dvdd     <- new, from this patch
    regulator.3  INT3472:01-avdd
    regulator.4  INT3472:01-dovdd
    regulator.5  INT3472:02-avdd

Nothing else regressed: audio, Secure Boot and module signing unaffected, no failed units.

What it does not fix, and why that is not this patch's fault
------------------------------------------------------------

The camera is exactly as dead as before:

ov13858 i2c-OVTID858:00: failed to find sensor: -5 every regulator: num_users=0, state=disabled /dev/media0: 0 entities

The rear sensor here is an OV13858, and the in-tree ov13858 driver requests no regulators and touches no GPIOs at all - zero regulator and zero gpiod references in drivers/media/i2c/ov13858.c. So INT3472:00-dvdd is registered and then never claimed by anyone, and the sensor is still held in reset because nothing releases it.

That is exactly the difference between your machine and this one. ov8865 asks for "dvdd", "dovdd" and "avdd" by name, so mapping POWER1 to "dvdd" completes the picture for the Surface Pro 7+. ov13858 asks for nothing.

The same conclusion was reached independently on the Surface Pro 10, which carries the same OV13858:

https://github.com/linux-surface/linux-surface/issues/2153

There they had to add reset-GPIO handling to ov13858_probe() and force the regulators on, and describe the latter as too broad for upstream.

So: this patch is correct and necessary, and for the OV13858 machines it is not sufficient. The remaining work is in the sensor driver rather than in int3472, which seems worth stating explicitly so nobody expects the Pro 10 or Pro 11 rear camera to start working when this lands.

If it would help, I am happy to test a patch teaching ov13858 to request its supplies and release reset - it is the same shape as what ov8865 already does. The machine is here and I can build and boot kernels on it.

One note for anyone reproducing this out-of-tree: the module build uses /usr/src/linux-headers-<ver>/include/, not the patched source tree, so patching only the tree gives 'INT3472_GPIO_TYPE_POWER1' undeclared. The installed header has to be patched too.

Thanks,
Sergey

#1 lore raw

Hi Sergey, thank you for testing it on your SP11

Does any of the maintainers want me to send out a v4 of this patch where I include a commit message with this limitation related to ov13858?

Best,
Jakob

Quoted 72 lines · Sergey Lebedev

Hans pointed me at this from a report I sent this morning about the same GPIO type on a Surface Pro 11 - thank you, and sorry for the duplicate question. I have now tested this patch on that machine, which is a third model and, more usefully, a different sensor. Result below, with the part that is still missing for this sensor family.

What the patch fixes here
-------------------------

Built on 7.0.0-30 (Ubuntu 26.04). The warning is gone and the rail is mapped:

before: int3472-discrete INT3472:00: GPIO type 0x08 unknown;
the sensor may not work
after : no int3472 messages at all

/sys/class/regulator:

    regulator.1  INT3472:00-avdd
    regulator.2  INT3472:00-dvdd     <- new, from this patch
    regulator.3  INT3472:01-avdd
    regulator.4  INT3472:01-dovdd
    regulator.5  INT3472:02-avdd

Nothing else regressed: audio, Secure Boot and module signing unaffected, no failed units.

What it does not fix, and why that is not this patch's fault
------------------------------------------------------------

The camera is exactly as dead as before:

ov13858 i2c-OVTID858:00: failed to find sensor: -5 every regulator: num_users=0, state=disabled /dev/media0: 0 entities

The rear sensor here is an OV13858, and the in-tree ov13858 driver requests no regulators and touches no GPIOs at all - zero regulator and zero gpiod references in drivers/media/i2c/ov13858.c. So INT3472:00-dvdd is registered and then never claimed by anyone, and the sensor is still held in reset because nothing releases it.

That is exactly the difference between your machine and this one. ov8865 asks for "dvdd", "dovdd" and "avdd" by name, so mapping POWER1 to "dvdd" completes the picture for the Surface Pro 7+. ov13858 asks for nothing.

The same conclusion was reached independently on the Surface Pro 10, which carries the same OV13858:

https://github.com/linux-surface/linux-surface/issues/2153

There they had to add reset-GPIO handling to ov13858_probe() and force the regulators on, and describe the latter as too broad for upstream.

So: this patch is correct and necessary, and for the OV13858 machines it is not sufficient. The remaining work is in the sensor driver rather than in int3472, which seems worth stating explicitly so nobody expects the Pro 10 or Pro 11 rear camera to start working when this lands.

If it would help, I am happy to test a patch teaching ov13858 to request its supplies and release reset - it is the same shape as what ov8865 already does. The machine is here and I can build and boot kernels on it.

One note for anyone reproducing this out-of-tree: the module build uses /usr/src/linux-headers-<ver>/include/, not the patched source tree, so patching only the tree gives 'INT3472_GPIO_TYPE_POWER1' undeclared. The installed header has to be patched too.

Thanks,
Sergey

maintainer
#2 lore raw

Hi,

Quoted 3 lines · Jakob Berg Jespersen

Hi Sergey, thank you for testing it on your SP11

Does any of the maintainers want me to send out a v4 of this patch where I include a commit message with this limitation related to ov13858?

No there is no need for a v4, the ov13858 problem is not a problem with the INT3472 driver, but rather with the ov13858 driver.

Regards,

Hans

#3 lore raw

Thanks for clarifying Hans,
I will leave v3 as it is.

Jakob

Quoted 16 lines · Hans de Goede

Hi,

Quoted 3 lines · Jakob Berg Jespersen

Hi Sergey, thank you for testing it on your SP11

Does any of the maintainers want me to send out a v4 of this patch where I include a commit message with this limitation related to ov13858?

No there is no need for a v4, the ov13858 problem is not a problem with the INT3472 driver, but rather with the ov13858 driver.

Regards,

Hans

maintainer
#4 lore raw
In reply to Sergey Lebedev

Hi,

Quoted 55 lines · Sergey Lebedev

Hans pointed me at this from a report I sent this morning about the same GPIO type on a Surface Pro 11 - thank you, and sorry for the duplicate question. I have now tested this patch on that machine, which is a third model and, more usefully, a different sensor. Result below, with the part that is still missing for this sensor family.

What the patch fixes here
-------------------------

Built on 7.0.0-30 (Ubuntu 26.04). The warning is gone and the rail is mapped:

before: int3472-discrete INT3472:00: GPIO type 0x08 unknown;
the sensor may not work
after : no int3472 messages at all

/sys/class/regulator:

    regulator.1  INT3472:00-avdd
    regulator.2  INT3472:00-dvdd     <- new, from this patch
    regulator.3  INT3472:01-avdd
    regulator.4  INT3472:01-dovdd
    regulator.5  INT3472:02-avdd

Nothing else regressed: audio, Secure Boot and module signing unaffected, no failed units.

What it does not fix, and why that is not this patch's fault
------------------------------------------------------------

The camera is exactly as dead as before:

ov13858 i2c-OVTID858:00: failed to find sensor: -5 every regulator: num_users=0, state=disabled /dev/media0: 0 entities

The rear sensor here is an OV13858, and the in-tree ov13858 driver requests no regulators and touches no GPIOs at all - zero regulator and zero gpiod references in drivers/media/i2c/ov13858.c. So INT3472:00-dvdd is registered and then never claimed by anyone, and the sensor is still held in reset because nothing releases it.

That is exactly the difference between your machine and this one. ov8865 asks for "dvdd", "dovdd" and "avdd" by name, so mapping POWER1 to "dvdd" completes the picture for the Surface Pro 7+. ov13858 asks for nothing.

The same conclusion was reached independently on the Surface Pro 10, which carries the same OV13858:

https://github.com/linux-surface/linux-surface/issues/2153

There they had to add reset-GPIO handling to ov13858_probe() and force the regulators on, and describe the latter as too broad for upstream.

Quoted 8 lines

So: this patch is correct and necessary, and for the OV13858 machines it is not sufficient. The remaining work is in the sensor driver rather than in int3472, which seems worth stating explicitly so nobody expects the Pro 10 or Pro 11 rear camera to start working when this lands.

If it would help, I am happy to test a patch teaching ov13858 to request its supplies and release reset - it is the same shape as what ov8865 already does. The machine is here and I can build and boot kernels on it.

Right. Someone needs to write 2 patches for the ov13858 driver to add:

1. Regulator support, this should list the 3 standard:

static const char * const ov02c10_supply_names[] = {

        "dovdd",        /* Digital I/O power */
        "avdd",         /* Analog power */
        "dvdd",         /* Digital core power */

};

OV sensor supply names. This should use the bulk regulator API, request these at probe and turn them on / off at stream start / stop time.

The ov2c10 driver in drivers/media/i2c/ov2c10 with its ov02c10_get_pm_resources() ov02c10_power_off() and ov02c10_power_on() functions is a good example of how to handle this.

2. Add reset GPIO support, again the ov02c10 driver and the 3 mentioned helper functions there are a good example to copy for this.

Regards,

Hans