[PATCH 1/2 v2] iio: magnetometer: ak8974: Break out measurement

[PATCH 1/2 v2] iio: magnetometer: ak8974: Break out measurement

[Linus Walleij]

This breaks out the measurement code to its own function
so we can handle this without swirling it up with the
bis switch() statement inside ak8974_read_raw().

Use an intermediary s16* variable since we read s16 but
the external API required an int* so this way we get
explicit casting.

Cc: Nick Reitemeyer <nick.reitemeyer@web.de>
Cc: Stephan Gerhold <stephan@gerhold.net>
Cc: Michał Mirosław <mirq-linux@rere.qmqm.pl>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
---
ChangeLog v1->v2:
- Break out as a separate patch.
---
 drivers/iio/magnetometer/ak8974.c | 51 +++++++++++++++++++------------
 1 file changed, 31 insertions(+), 20 deletions(-)

diff --git a/drivers/iio/magnetometer/ak8974.c b/drivers/iio/magnetometer/ak8974.c
index ade4ed8f67d2..5361647b9054 100644
--- a/drivers/iio/magnetometer/ak8974.c
+++ b/drivers/iio/magnetometer/ak8974.c
@@ -554,46 +554,57 @@ static int ak8974_detect(struct ak8974 *ak8974)
 	return 0;
 }
 
+static int ak8974_measure(struct ak8974 *ak8974, unsigned long address, s16 *val)
+{
+	__le16 hw_values[3];
+	int ret;
+
+	pm_runtime_get_sync(&ak8974->i2c->dev);
+	mutex_lock(&ak8974->lock);
+
+	ret = ak8974_trigmeas(ak8974);
+	if (ret)
+		goto out_unlock;
+	ret = ak8974_getresult(ak8974, hw_values);
+	if (ret)
+		goto out_unlock;
+	*val = (s16)le16_to_cpu(hw_values[address]);
+out_unlock:
+	mutex_unlock(&ak8974->lock);
+	pm_runtime_mark_last_busy(&ak8974->i2c->dev);
+	pm_runtime_put_autosuspend(&ak8974->i2c->dev);
+
+	return ret;
+}
+
 static int ak8974_read_raw(struct iio_dev *indio_dev,
 			   struct iio_chan_spec const *chan,
 			   int *val, int *val2,
 			   long mask)
 {
 	struct ak8974 *ak8974 = iio_priv(indio_dev);
-	__le16 hw_values[3];
 	int ret = -EINVAL;
-
-	pm_runtime_get_sync(&ak8974->i2c->dev);
-	mutex_lock(&ak8974->lock);
+	s16 outval;
 
 	switch (mask) {
 	case IIO_CHAN_INFO_RAW:
 		if (chan->address > 2) {
 			dev_err(&ak8974->i2c->dev, "faulty channel address\n");
 			ret = -EIO;
-			goto out_unlock;
+			goto out_err_read;
 		}
-		ret = ak8974_trigmeas(ak8974);
-		if (ret)
-			goto out_unlock;
-		ret = ak8974_getresult(ak8974, hw_values);
-		if (ret)
-			goto out_unlock;
-
 		/*
 		 * We read all axes and discard all but one, for optimized
 		 * reading, use the triggered buffer.
 		 */
-		*val = (s16)le16_to_cpu(hw_values[chan->address]);
-
+		ret = ak8974_measure(ak8974, chan->address, &outval);
+		if (ret)
+			goto out_err_read;
+		*val = outval;
 		ret = IIO_VAL_INT;
+		break;
 	}
-
- out_unlock:
-	mutex_unlock(&ak8974->lock);
-	pm_runtime_mark_last_busy(&ak8974->i2c->dev);
-	pm_runtime_put_autosuspend(&ak8974->i2c->dev);
-
+out_err_read:
 	return ret;
 }
 
-- 
2.21.1

[PATCH 2/2 v2] iio: magnetometer: ak8974: Provide scaling

[Linus Walleij]

The manual for the HSCDTD008A gives us a scaling for the
three axis as +/- 2.4mT per axis.

When I implement this the biggest axis indicates 0.59 Gauss
which is a reasonable measurement for the earths magnetic
which is in the range of 0.25 to 0.65 Gauss on the surface
according to Wikipedia.

Cc: Nick Reitemeyer <nick.reitemeyer@web.de>
Cc: Stephan Gerhold <stephan@gerhold.net>
Cc: Michał Mirosław <mirq-linux@rere.qmqm.pl>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
---
ChangeLog v1->v2:
- Split out the measurement refactoring.
---
 drivers/iio/magnetometer/ak8974.c | 15 ++++++++++++++-
 1 file changed, 14 insertions(+), 1 deletion(-)

diff --git a/drivers/iio/magnetometer/ak8974.c b/drivers/iio/magnetometer/ak8974.c
index 5361647b9054..effcdd93e650 100644
--- a/drivers/iio/magnetometer/ak8974.c
+++ b/drivers/iio/magnetometer/ak8974.c
@@ -603,6 +603,18 @@ static int ak8974_read_raw(struct iio_dev *indio_dev,
 		*val = outval;
 		ret = IIO_VAL_INT;
 		break;
+	case IIO_CHAN_INFO_SCALE:
+		/*
+		 * The datasheet for HSCDTF008A, page 3 specifies the
+		 * range of the sensor as +/- 2.4 mT per axis, which corresponds
+		 * to +/- 2400 uT = +/- 24 Gauss. So 0x7fff is 24 Gauss and
+		 * 0xffff is -24 Gauss. To account for the one missing value if
+		 * we multiply by 1/S16_MAX, instead multiply with 2/U16_MAX.
+		 */
+		*val = 24 * 2;
+		*val2 = U16_MAX;
+		ret = IIO_VAL_FRACTIONAL;
+		break;
 	}
 out_err_read:
 	return ret;
@@ -667,7 +679,8 @@ static const struct iio_chan_spec_ext_info ak8974_ext_info[] = {
 		.type = IIO_MAGN,					\
 		.modified = 1,						\
 		.channel2 = IIO_MOD_##axis,				\
-		.info_mask_separate = BIT(IIO_CHAN_INFO_RAW),		\
+		.info_mask_separate = BIT(IIO_CHAN_INFO_RAW) |		\
+			BIT(IIO_CHAN_INFO_SCALE),			\
 		.ext_info = ak8974_ext_info,				\
 		.address = index,					\
 		.scan_index = index,					\
-- 
2.21.1

Re: [PATCH 2/2 v2] iio: magnetometer: ak8974: Provide scaling

[Stephan Gerhold]

Hi Linus,

On Thu, Apr 16, 2020 at 04:09:17PM +0200, Linus Walleij wrote:
> The manual for the HSCDTD008A gives us a scaling for the
> three axis as +/- 2.4mT per axis.
> 

I wonder if we can really assume that this applies to
the other models (e.g. AK8974) as well?

> When I implement this the biggest axis indicates 0.59 Gauss
> which is a reasonable measurement for the earths magnetic
> which is in the range of 0.25 to 0.65 Gauss on the surface
> according to Wikipedia.
> 
> Cc: Nick Reitemeyer <nick.reitemeyer@web.de>
> Cc: Stephan Gerhold <stephan@gerhold.net>
> Cc: Michał Mirosław <mirq-linux@rere.qmqm.pl>
> Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
> ---
> ChangeLog v1->v2:
> - Split out the measurement refactoring.
> ---
>  drivers/iio/magnetometer/ak8974.c | 15 ++++++++++++++-
>  1 file changed, 14 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/iio/magnetometer/ak8974.c b/drivers/iio/magnetometer/ak8974.c
> index 5361647b9054..effcdd93e650 100644
> --- a/drivers/iio/magnetometer/ak8974.c
> +++ b/drivers/iio/magnetometer/ak8974.c
> @@ -603,6 +603,18 @@ static int ak8974_read_raw(struct iio_dev *indio_dev,
>  		*val = outval;
>  		ret = IIO_VAL_INT;
>  		break;
> +	case IIO_CHAN_INFO_SCALE:
> +		/*
> +		 * The datasheet for HSCDTF008A, page 3 specifies the
> +		 * range of the sensor as +/- 2.4 mT per axis, which corresponds
> +		 * to +/- 2400 uT = +/- 24 Gauss. So 0x7fff is 24 Gauss and
> +		 * 0xffff is -24 Gauss. To account for the one missing value if
> +		 * we multiply by 1/S16_MAX, instead multiply with 2/U16_MAX.
> +		 */

I just want to note that (according to the datasheet), HSCDTD008A
produces either 14-bit or 15-bit measurements (depending on
the HSCDTD008A_CTRL4_RANGE bit that we set by default).

I think this isn't exposed correctly in the AK8974_AXIS_CHANNEL() macro
(realbits is 16 instead of 15), so this might need special casing for
hscdt008a?

The reason I mention this is because I think it would also affect the
scaling that you implement here. With 15-bit output it produces values
from +16383 (0x3fff) (= 2.4 mT?) to -16384 (0xc000) (= -2.4 mT?).

So it would never reach the 0x7fff and 0xffff you mention
in your comment.

> +		*val = 24 * 2;
> +		*val2 = U16_MAX;
> +		ret = IIO_VAL_FRACTIONAL;
> +		break;
>  	}
>  out_err_read:
>  	return ret;
> @@ -667,7 +679,8 @@ static const struct iio_chan_spec_ext_info ak8974_ext_info[] = {
>  		.type = IIO_MAGN,					\
>  		.modified = 1,						\
>  		.channel2 = IIO_MOD_##axis,				\
> -		.info_mask_separate = BIT(IIO_CHAN_INFO_RAW),		\
> +		.info_mask_separate = BIT(IIO_CHAN_INFO_RAW) |		\
> +			BIT(IIO_CHAN_INFO_SCALE),			\
>  		.ext_info = ak8974_ext_info,				\
>  		.address = index,					\
>  		.scan_index = index,					\
> -- 
> 2.21.1
> 

Re: [PATCH 2/2 v2] iio: magnetometer: ak8974: Provide scaling

[Linus Walleij]

On Thu, Apr 16, 2020 at 5:32 PM Stephan Gerhold <stephan@gerhold.net> wrote:
> On Thu, Apr 16, 2020 at 04:09:17PM +0200, Linus Walleij wrote:

> > The manual for the HSCDTD008A gives us a scaling for the
> > three axis as +/- 2.4mT per axis.
>
> I wonder if we can really assume that this applies to
> the other models (e.g. AK8974) as well?

Patches are for testing :D

I have a Ux500 reference board with the AK8974 vanilla
variant mounted so I will check with that one.

> > +     case IIO_CHAN_INFO_SCALE:
> > +             /*
> > +              * The datasheet for HSCDTF008A, page 3 specifies the
> > +              * range of the sensor as +/- 2.4 mT per axis, which corresponds
> > +              * to +/- 2400 uT = +/- 24 Gauss. So 0x7fff is 24 Gauss and
> > +              * 0xffff is -24 Gauss. To account for the one missing value if
> > +              * we multiply by 1/S16_MAX, instead multiply with 2/U16_MAX.
> > +              */
>
> I just want to note that (according to the datasheet), HSCDTD008A
> produces either 14-bit or 15-bit measurements (depending on
> the HSCDTD008A_CTRL4_RANGE bit that we set by default).

Argh OK I will fix this. I try to get an AMI datasheet as well.

> I think this isn't exposed correctly in the AK8974_AXIS_CHANNEL() macro
> (realbits is 16 instead of 15), so this might need special casing for
> hscdt008a?

Yes definately. It's a bug. I'll make a separate patch for this.

> The reason I mention this is because I think it would also affect the
> scaling that you implement here. With 15-bit output it produces values
> from +16383 (0x3fff) (= 2.4 mT?) to -16384 (0xc000) (= -2.4 mT?).
>
> So it would never reach the 0x7fff and 0xffff you mention
> in your comment.

You're right. What I need to do is put the HSCDTD008A and
AK8974 side by side and compare.

Yours,
Linus Walleij

Re: [PATCH 1/2 v2] iio: magnetometer: ak8974: Break out measurement

[Michał Mirosław]

On Thu, Apr 16, 2020 at 04:09:16PM +0200, Linus Walleij wrote:
> This breaks out the measurement code to its own function
> so we can handle this without swirling it up with the
> bis switch() statement inside ak8974_read_raw().
> 
> Use an intermediary s16* variable since we read s16 but
> the external API required an int* so this way we get
> explicit casting.
> 
> Cc: Nick Reitemeyer <nick.reitemeyer@web.de>
> Cc: Stephan Gerhold <stephan@gerhold.net>
> Cc: Michał Mirosław <mirq-linux@rere.qmqm.pl>
> Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
> ---
> ChangeLog v1->v2:
> - Break out as a separate patch.
> ---
>  drivers/iio/magnetometer/ak8974.c | 51 +++++++++++++++++++------------
>  1 file changed, 31 insertions(+), 20 deletions(-)
> 
> diff --git a/drivers/iio/magnetometer/ak8974.c b/drivers/iio/magnetometer/ak8974.c
> index ade4ed8f67d2..5361647b9054 100644
> --- a/drivers/iio/magnetometer/ak8974.c
> +++ b/drivers/iio/magnetometer/ak8974.c
> @@ -554,46 +554,57 @@ static int ak8974_detect(struct ak8974 *ak8974)
>  	return 0;
>  }
>  
> +static int ak8974_measure(struct ak8974 *ak8974, unsigned long address, s16 *val)
> +{
> +	__le16 hw_values[3];
> +	int ret;
> +
> +	pm_runtime_get_sync(&ak8974->i2c->dev);
> +	mutex_lock(&ak8974->lock);
> +
> +	ret = ak8974_trigmeas(ak8974);
> +	if (ret)
> +		goto out_unlock;
> +	ret = ak8974_getresult(ak8974, hw_values);
> +	if (ret)
> +		goto out_unlock;
> +	*val = (s16)le16_to_cpu(hw_values[address]);

This will still work if the 'val' argument was 'int *'. And it would make
'outval' in the caller redundant.

> +out_unlock:
> +	mutex_unlock(&ak8974->lock);
> +	pm_runtime_mark_last_busy(&ak8974->i2c->dev);
> +	pm_runtime_put_autosuspend(&ak8974->i2c->dev);
> +
> +	return ret;
> +}
> +
>  static int ak8974_read_raw(struct iio_dev *indio_dev,
>  			   struct iio_chan_spec const *chan,
>  			   int *val, int *val2,
>  			   long mask)
>  {
>  	struct ak8974 *ak8974 = iio_priv(indio_dev);
> -	__le16 hw_values[3];
>  	int ret = -EINVAL;
> -
> -	pm_runtime_get_sync(&ak8974->i2c->dev);
> -	mutex_lock(&ak8974->lock);
> +	s16 outval;
>  
>  	switch (mask) {
>  	case IIO_CHAN_INFO_RAW:
>  		if (chan->address > 2) {
>  			dev_err(&ak8974->i2c->dev, "faulty channel address\n");
>  			ret = -EIO;
> -			goto out_unlock;
> +			goto out_err_read;

You could get rid of the gotos to shorten and clear up the code.

>  		}
> -		ret = ak8974_trigmeas(ak8974);
> -		if (ret)
> -			goto out_unlock;
> -		ret = ak8974_getresult(ak8974, hw_values);
> -		if (ret)
> -			goto out_unlock;
> -
>  		/*
>  		 * We read all axes and discard all but one, for optimized
>  		 * reading, use the triggered buffer.
>  		 */
> -		*val = (s16)le16_to_cpu(hw_values[chan->address]);
> -
> +		ret = ak8974_measure(ak8974, chan->address, &outval);
> +		if (ret)
> +			goto out_err_read;
> +		*val = outval;
>  		ret = IIO_VAL_INT;
> +		break;
>  	}
> -
> - out_unlock:
> -	mutex_unlock(&ak8974->lock);
> -	pm_runtime_mark_last_busy(&ak8974->i2c->dev);
> -	pm_runtime_put_autosuspend(&ak8974->i2c->dev);
> -
> +out_err_read:
>  	return ret;
>  }
[...]

Best Regards,
Michał Mirosław

Re: [PATCH 2/2 v2] iio: magnetometer: ak8974: Provide scaling

[Michał Mirosław]

On Thu, Apr 16, 2020 at 04:09:17PM +0200, Linus Walleij wrote:
> The manual for the HSCDTD008A gives us a scaling for the
> three axis as +/- 2.4mT per axis.
> 
> When I implement this the biggest axis indicates 0.59 Gauss
> which is a reasonable measurement for the earths magnetic
> which is in the range of 0.25 to 0.65 Gauss on the surface
> according to Wikipedia.
> 
> Cc: Nick Reitemeyer <nick.reitemeyer@web.de>
> Cc: Stephan Gerhold <stephan@gerhold.net>
> Cc: Michał Mirosław <mirq-linux@rere.qmqm.pl>
> Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
> ---
> ChangeLog v1->v2:
> - Split out the measurement refactoring.
> ---
>  drivers/iio/magnetometer/ak8974.c | 15 ++++++++++++++-
>  1 file changed, 14 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/iio/magnetometer/ak8974.c b/drivers/iio/magnetometer/ak8974.c
> index 5361647b9054..effcdd93e650 100644
> --- a/drivers/iio/magnetometer/ak8974.c
> +++ b/drivers/iio/magnetometer/ak8974.c
> @@ -603,6 +603,18 @@ static int ak8974_read_raw(struct iio_dev *indio_dev,
>  		*val = outval;
>  		ret = IIO_VAL_INT;
>  		break;
> +	case IIO_CHAN_INFO_SCALE:
> +		/*
> +		 * The datasheet for HSCDTF008A, page 3 specifies the
> +		 * range of the sensor as +/- 2.4 mT per axis, which corresponds
> +		 * to +/- 2400 uT = +/- 24 Gauss. So 0x7fff is 24 Gauss and
> +		 * 0xffff is -24 Gauss. To account for the one missing value if
> +		 * we multiply by 1/S16_MAX, instead multiply with 2/U16_MAX.
> +		 */
> +		*val = 24 * 2;
> +		*val2 = U16_MAX;
> +		ret = IIO_VAL_FRACTIONAL;
> +		break;

Hi,

The comment seems wrong. DS mentions that the measurement values are
S16, but limited in range (-k to +k, with k = 2^13 or 2^14). So the
denominator should be 2^13 or 2^14, and numerator 2.4mT.

Best Regards,
Michał Mirosław

Re: [PATCH 2/2 v2] iio: magnetometer: ak8974: Provide scaling

[Michał Mirosław]

On Thu, Apr 16, 2020 at 08:17:46PM +0200, Michał Mirosław wrote:
> On Thu, Apr 16, 2020 at 04:09:17PM +0200, Linus Walleij wrote:
> > The manual for the HSCDTD008A gives us a scaling for the
> > three axis as +/- 2.4mT per axis.
> > 
> > When I implement this the biggest axis indicates 0.59 Gauss
> > which is a reasonable measurement for the earths magnetic
> > which is in the range of 0.25 to 0.65 Gauss on the surface
> > according to Wikipedia.
> > 
> > Cc: Nick Reitemeyer <nick.reitemeyer@web.de>
> > Cc: Stephan Gerhold <stephan@gerhold.net>
> > Cc: Michał Mirosław <mirq-linux@rere.qmqm.pl>
> > Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
> > ---
> > ChangeLog v1->v2:
> > - Split out the measurement refactoring.
> > ---
> >  drivers/iio/magnetometer/ak8974.c | 15 ++++++++++++++-
> >  1 file changed, 14 insertions(+), 1 deletion(-)
> > 
> > diff --git a/drivers/iio/magnetometer/ak8974.c b/drivers/iio/magnetometer/ak8974.c
> > index 5361647b9054..effcdd93e650 100644
> > --- a/drivers/iio/magnetometer/ak8974.c
> > +++ b/drivers/iio/magnetometer/ak8974.c
> > @@ -603,6 +603,18 @@ static int ak8974_read_raw(struct iio_dev *indio_dev,
> >  		*val = outval;
> >  		ret = IIO_VAL_INT;
> >  		break;
> > +	case IIO_CHAN_INFO_SCALE:
> > +		/*
> > +		 * The datasheet for HSCDTF008A, page 3 specifies the
> > +		 * range of the sensor as +/- 2.4 mT per axis, which corresponds
> > +		 * to +/- 2400 uT = +/- 24 Gauss. So 0x7fff is 24 Gauss and
> > +		 * 0xffff is -24 Gauss. To account for the one missing value if
> > +		 * we multiply by 1/S16_MAX, instead multiply with 2/U16_MAX.
> > +		 */
> > +		*val = 24 * 2;
> > +		*val2 = U16_MAX;
> > +		ret = IIO_VAL_FRACTIONAL;
> > +		break;
> 
> Hi,
> 
> The comment seems wrong. DS mentions that the measurement values are
> S16, but limited in range (-k to +k, with k = 2^13 or 2^14). So the
> denominator should be 2^13 or 2^14, and numerator 2.4mT.

Ok, this is what you actually did. :-) The error is in 0xffff in the
comment: it should be 0x8000 (== -0x8000). After applying the limit
in range it would be 0x3fff to -0x4000 (== 0xC000).

I would actually use 2^14 as denominator because that's a "round" number,
the difference is insignificant, and we don't really know which one is
correct. (BTW, the DS mentions 0.15uT/LSB resolution, so it would not be
exact 2.4mT for the max value.)

Best Regards,
Michał Mirosław