Le Wed, Mar 13, 2024 at 09:21:35PM +0100, Marco Felsch a écrit :
> Hi Kamel,
>
Hello Marco,
[...]
> > +static int axiom_i2c_probe(struct i2c_client *client)
> > +{
> > + struct device *dev = &client->dev;
> > + struct input_dev *input_dev;
> > + struct axiom_data *ts;
> > + u32 poll_interval;
> > + int target;
> > + int error;
> > +
> > + ts = devm_kzalloc(dev, sizeof(*ts), GFP_KERNEL);
> > + if (!ts)
> > + return -ENOMEM;
> > +
> > + i2c_set_clientdata(client, ts);
> > + ts->client = client;
> > + ts->dev = dev;
> > +
> > + ts->regmap = devm_regmap_init_i2c(client, &axiom_i2c_regmap_config);
> > + error = PTR_ERR_OR_ZERO(ts->regmap);
> > + if (error) {
> > + dev_err(dev, "Failed to initialize regmap: %d\n", error);
> > + return error;
> > + }
> > +
> > + ts->reset_gpio = devm_gpiod_get_optional(dev, "reset", GPIOD_OUT_HIGH);
> > + if (IS_ERR(ts->reset_gpio))
> > + return dev_err_probe(dev, PTR_ERR(ts->reset_gpio), "failed to get reset GPIO\n");
> > +
> > + if (ts->reset_gpio)
> > + axiom_reset(ts->reset_gpio);
>
> This seems useless, since you doing an reset without enabling the power
> supply (below). I know there are systems which do have the supply always
> connected or for ACPI the supply is managed via firmware, but the driver
> should implement the correct logic and for DT/OF case this is not
> correct.
>
> > +
> > + ts->vddi = devm_regulator_get_optional(dev, "vddi");
> > + if (!IS_ERR(ts->vddi)) {
> > + error = devm_regulator_get_enable(dev, "vddi");
>
> Regulators are ref counted and now you request the regulator twice. Also
> the regulator is not optional, it is required for the device to work.
> Same applies to the vdda below.
>
While it is true most of the time, it occurs that for x86 based boards,
adding a regulator entirely is not always possible.
In our particular case, the I2C controller for this touchscreen is
behind a CPLD (aka embedded controller) so I have no direct access to
the I2C controller and it isn't described in the ACPI table.
In a normal case, I would use ACPI override to pass regulator
properties, but here it's not possible.
Having a CPLD exposing this kind of controller is quite common on x86
based boards. So, we need a way to support the case when a regulator
can't be described. The optional regulator looked like a good option,
but if you have a better alternative, I am open to considering it.
--
Kamel Bouhara, Bootlin
Embedded Linux and kernel engineering
https://bootlin.com
