On Thu, Sep 01, 2022 at 12:45:49PM +0100, Jonathan Cameron wrote:
> On Thu, 1 Sep 2022 06:11:45 +0200
> Oleksij Rempel <o.rempel@xxxxxxxxxxxxxx> wrote:
>
> > If VREF pin is attached, we should use external VREF source instead of
> > the internal. Otherwise we will get wrong measurements on some of channel
> > types.
> >
> > Signed-off-by: Oleksij Rempel <o.rempel@xxxxxxxxxxxxxx>
> Hi Oleksij,
>
> I'm not sure why I didn't review this patch in v1...
>
> Anyhow, some comments below.
>
> Jonathan
>
> > ---
> > drivers/iio/adc/ti-tsc2046.c | 64 +++++++++++++++++++++++++++++++-----
> > 1 file changed, 55 insertions(+), 9 deletions(-)
> >
> > diff --git a/drivers/iio/adc/ti-tsc2046.c b/drivers/iio/adc/ti-tsc2046.c
> > index 0d9436a69cbfb..bbc8b4137b0b1 100644
> > --- a/drivers/iio/adc/ti-tsc2046.c
> > +++ b/drivers/iio/adc/ti-tsc2046.c
> > @@ -8,6 +8,7 @@
> > #include <linux/bitfield.h>
> > #include <linux/delay.h>
> > #include <linux/module.h>
> > +#include <linux/regulator/consumer.h>
> > #include <linux/spi/spi.h>
> >
> > #include <asm/unaligned.h>
> > @@ -175,6 +176,9 @@ struct tsc2046_adc_priv {
> > u32 time_per_bit_ns;
> >
> > struct tsc2046_adc_ch_cfg ch_cfg[TI_TSC2046_MAX_CHAN];
> > + bool use_internal_vref;
> > + unsigned int vref_mv;
> > + struct regulator *vref_reg;
> > };
> >
> > #define TI_TSC2046_V_CHAN(index, bits, name) \
> > @@ -252,7 +256,9 @@ static u8 tsc2046_adc_get_cmd(struct tsc2046_adc_priv *priv, int ch_idx,
> > case TI_TSC2046_ADDR_AUX:
> > case TI_TSC2046_ADDR_VBAT:
> > case TI_TSC2046_ADDR_TEMP0:
> > - pd |= TI_TSC2046_SER | TI_TSC2046_PD1_VREF_ON;
> > + pd |= TI_TSC2046_SER;
> > + if (priv->use_internal_vref)
> > + pd |= TI_TSC2046_PD1_VREF_ON;
> > }
> >
> > return TI_TSC2046_START | FIELD_PREP(TI_TSC2046_ADDR, ch_idx) | pd;
> > @@ -468,7 +474,7 @@ static int tsc2046_adc_read_raw(struct iio_dev *indio_dev,
> > * So, it is better to use external voltage-divider driver
> > * instead, which is calculating complete chain.
> > */
> > - *val = TI_TSC2046_INT_VREF;
> > + *val = priv->vref_mv;
> > *val2 = chan->scan_type.realbits;
> > return IIO_VAL_FRACTIONAL_LOG2;
> > }
> > @@ -781,22 +787,42 @@ static int tsc2046_adc_probe(struct spi_device *spi)
> > indio_dev->num_channels = dcfg->num_channels;
> > indio_dev->info = &tsc2046_adc_info;
> >
> > + priv->vref_reg = devm_regulator_get_optional(&spi->dev, "vref");
> > + if (!IS_ERR(priv->vref_reg)) {
> > + ret = regulator_enable(priv->vref_reg);
> > + if (ret)
> > + return ret;
> > +
> > + ret = regulator_get_voltage(priv->vref_reg);
> > + if (ret < 0)
> > + goto err_regulator_disable;
>
> Whilst regulators voltages of references rarely change at runtime
> they are allowed to, so it is logically better to query the
> voltage at the point of use. Requests for scale should be
> rare (unless there is a consumer that keeps querying this?)
> so the slightly overhead there
> shouldn't matter.
The voltage scale is related to actual adc measurements. If it variates,
we will have different scale of challenges.
> > +
> > + priv->vref_mv = ret / 1000;
> > + priv->use_internal_vref = false;
> > + } else {
> > + /* Use internal reference */
> > + priv->vref_mv = TI_TSC2046_INT_VREF;
> > + priv->use_internal_vref = true;
> > + }
> > +
> > tsc2046_adc_parse_fwnode(priv);
> >
> > ret = tsc2046_adc_setup_spi_msg(priv);
> > if (ret)
> > - return ret;
> > + goto err_regulator_disable;
> >
> > mutex_init(&priv->slock);
> >
> > ret = devm_request_irq(dev, spi->irq, &tsc2046_adc_irq,
> > IRQF_NO_AUTOEN, indio_dev->name, indio_dev);
> > if (ret)
> > - return ret;
> > + goto err_regulator_disable;
> >
> > trig = devm_iio_trigger_alloc(dev, "touchscreen-%s", indio_dev->name);
> > - if (!trig)
> > - return -ENOMEM;
> > + if (!trig) {
> > + ret = -ENOMEM;
> > + goto err_regulator_disable;
> > + }
> >
> > priv->trig = trig;
> > iio_trigger_set_drvdata(trig, indio_dev);
> > @@ -811,20 +837,39 @@ static int tsc2046_adc_probe(struct spi_device *spi)
> > ret = devm_iio_trigger_register(dev, trig);
> > if (ret) {
> > dev_err(dev, "failed to register trigger\n");
> > - return ret;
> > + goto err_regulator_disable;
>
> Please don't mix devm and non devm calls. It makes it much harder to reason about
> the correctness of ordering.
> Use devm_add_action_or_reset() to register a callback to disable the
> vref regulator.
>
> Alternative is back out the devm_ based registration of everything after
> the regulator enable.
ok.
Regards,
Oleksij
--
Pengutronix e.K. | |
Steuerwalder Str. 21 | http://www.pengutronix.de/ |
31137 Hildesheim, Germany | Phone: +49-5121-206917-0 |
Amtsgericht Hildesheim, HRA 2686 | Fax: +49-5121-206917-5555 |
