Re: [PATCH v4 2/2] media: i2c: Add GC08A3 image sensor driver

[Laurent Pinchart]

Hi Zhi,

On Tue, Feb 20, 2024 at 02:12:26AM +0000, Zhi Mao (毛智) wrote:
> On Tue, 2024-02-06 at 20:45 +0200, Laurent Pinchart wrote:
> > On Sun, Feb 04, 2024 at 02:15:38PM +0800, Zhi Mao wrote:
> > > Add a V4L2 sub-device driver for Galaxycore GC08A3 image sensor.
> > >
> > > Signed-off-by: Zhi Mao <zhi.mao@xxxxxxxxxxxx>
> > > ---
> > >  drivers/media/i2c/Kconfig  |   10 +
> > >  drivers/media/i2c/Makefile |    1 +
> > >  drivers/media/i2c/gc08a3.c | 1448 ++++++++++++++++++++++++++++++++++++
> > >  3 files changed, 1459 insertions(+)
> > >  create mode 100644 drivers/media/i2c/gc08a3.c

[snip]

> > > diff --git a/drivers/media/i2c/gc08a3.c b/drivers/media/i2c/gc08a3.c
> > > new file mode 100644
> > > index 000000000000..3fc7fffb815c
> > > --- /dev/null
> > > +++ b/drivers/media/i2c/gc08a3.c
> > > @@ -0,0 +1,1448 @@

[snip]

> > > +static int gc08a3_power_on(struct device *dev)
> > > +{
> > > +	struct i2c_client *client = to_i2c_client(dev);
> > > +	struct v4l2_subdev *sd = i2c_get_clientdata(client);
> > > +	struct gc08a3 *gc08a3 = to_gc08a3(sd);
> > > +	int ret;
> > > +
> > > +	ret = regulator_bulk_enable(ARRAY_SIZE(gc08a3_supply_name),
> > > +				    gc08a3->supplies);
> > > +	if (ret < 0) {
> > > +		dev_err(gc08a3->dev, "failed to enable regulators: %d\n", ret);
> > > +		return ret;
> > > +	}
> > > +
> > > +	ret = clk_prepare_enable(gc08a3->xclk);
> > > +	if (ret < 0) {
> > > +		regulator_bulk_disable(ARRAY_SIZE(gc08a3_supply_name),
> > > +				       gc08a3->supplies);
> > > +		dev_err(gc08a3->dev, "clk prepare enable failed\n");
> > > +		return ret;
> > > +	}
> > > +
> > > +	usleep_range(GC08A3_MIN_SLEEP_US, GC08A3_MAX_SLEEP_US);
> > > +
> > > +	gpiod_set_value_cansleep(gc08a3->reset_gpio, 1);
> >
> > Are you asserting reset when powering on ? That sounds wrong, you should
> > de-assert reset here (and acquire the reset gpio in probe() with
> > GPIOD_OUT_HIGH). Drivers should use logical levels for GPIOs, setting a
> > GPIO named "reset" to 1 should assert the reset signal, even if the
> > physical signal is active low. You may have the wrong polarity in the
> > device tree.
>
> According to the sensor power sequence sepc, "reset" pin should be pull
> from low to high after "dovdd/dvdd/avdd" power on, so I follow this
> power sequece to pull "reset" pin high in software flow.

>From a hardware point of view that's right, but the Linux kernel handles
logical level of GPIOs. If a GPIO is named "reset", it is expected that
calling

	gpiod_set_value_cansleep(gc08a3->reset_gpio, 1);

will "assert" the reset signal, setting it to a logical "reset = true"
level. This maps to the hardware 0V output level, as the signal is
active-low. To achieve this, define the reset GPIO as active low in DT,
and the GPIO framework will invert the signal for you. You should then
call

	gpiod_set_value_cansleep(gc08a3->reset_gpio, 1);

in the driver when you want to assert reset (set it to 0V), and

	gpiod_set_value_cansleep(gc08a3->reset_gpio, 0);

when you want to deassert it (set it to 3.3V, or whatever the I/O
voltage for the signal is).

This way all driver use logical states, and the inversion is handled in
DT.

> > > +	usleep_range(GC08A3_MIN_SLEEP_US, GC08A3_MAX_SLEEP_US);
> > > +
> > > +	return 0;
> > > +}

[snip]

-- 
Regards,

Laurent Pinchart