Re: [PATCH v2 4/5] iio:magnetometer:ak8975: mounting matrix support

Re: [PATCH v2 4/5] iio:magnetometer:ak8975: mounting matrix support

[Jonathan Cameron]

On 09/03/16 10:56, Gregor Boirie wrote:
> 
> On 03/05/2016 04:36 PM, Jonathan Cameron wrote:
>> On 03/03/16 10:44, Gregor Boirie wrote:
>>> Expose a rotation matrix to indicate userspace the chip placement with
>>> respect to the overall hardware system. This is needed to adjust
>>> coordinates sampled from a magnetometer chip when its position deviates
>>> from the main hardware system.
>>>
>>> Final coordinates computation is delegated to userspace since:
>>> * computation may involve floating point arithmetics ;
>>> * it allows an application to combine adjustments with arbitrary
>>>    transformations.
>>>
>>> This 3 dimentional space rotation matrix is expressed as 3x3 array of
>>> strings to support floating point numbers. It may be retrieved from a
>>> "in_magn_matrix" sysfs attribute file. It is declared into ak8975 DTS
>>> entry as a "matrix" property.
>>>
>>> Signed-off-by: Gregor Boirie <gregor.boirie-ITF29qwbsa/QT0dZR+AlfA@public.gmane.org>
>> This one definitely wants to go to the device tree list. I'd also like a few
>> IIO related comments as well. I was sure we'd been through this before and
>> ended up with some sort of solution, but now I can't find it... Oh well.
> I took inv_mpu_core as an example. However matrices are exposed as integers
> and we really need floating point support.
> A search on the device-tree mailing list archive does not show up with a
> solution to encode floating point numbers into DTS either.
Certainly non obvious how to do this.

For anyone device tree related who picks this up (I've added a CC)
we are basically talking about passing a rotation matrix from the device tree
straight through to userspace to describe the relative orientation of
some sensor.  An array of strings clearly works, but is this the best
way to do it?

(btw as this touches device tree bindings the list and maintainers
should have been cc'd from the start - another one of those kernel
conventions that are non obvious to people who haven't hit it before.
Doing this was a big news item a few years back when a policy to keep
bindings reviewed in a manageable way was decided, but if you weren't
active at the time then you wouldn't necessarily know).
> 
>>
>> Definitely needs ABI docs in Documentation/ABI/testing/sysfs-bus-iio
>> as well. We need the reference directions to be well specified as
>> well - how do we define the main hardware orientation?  on yours presumably
>> the usual approach of using the screen is irrelevant so that definition
>> could be a little 'interesting' to write down?
> The standard orientation for us is defined with respect to the drone's head
> and the propellers plane, which does not suit the nominal use cases indeed...
> Although it really deserves some more detailed documentation, I wonder if the
> question is relevant.
> Given that the DTS is board specific (correct me if I'm wrong), why not
> consider the reference orientation as board specific as well (and make the
> whole thing implicit) ?
I think that is fine for your sort of usecase - though might be nice to keep
a bit of documentation somewhere (perhaps as comments in the DTS), or perhaps
more centrally - would be nice for userspace libraries if at least all aircraft
had a similar 'default orientation'.

For the more 'standard' device with a screen I think we would want to standardise
if at all possible.  Most hand held devices have a 'natural' orientation 
(front facing camera at the top for example).  In those cases we are basically looking
at whether the sensor is mounted on the back or the front of the main circuit board and
then of course it can be at any rotation in plane but typically is at 90 degrees
(hence the inv mpu limitations)

I'm guessing there are existing userspace mappings for this.  Perhaps in the Gnome
sensors stuff?

> 
>>
>> I'm also unsure of whether the right thing to do in the 'not provided'
>> cases is to not export the setting, or to assume that the chip is at some
>> 'standard' orientation and provide that?
> As a self-explanatory interface, I suppose sysfs convention would expect to expose
> this even in the case of an identity matrix.
> However, userspace would need to either:
> * systematically apply rotation matrix ;
> * or detect identity matrix and skip rotation which may require an a-priori
>   knowledge of reference orientation (breaking the statement mentioned above).
> 
> Personally, I'd prefer not exporting the property at all when aligned with
> the main hardware orientation : this makes application simpler and allows for
> better performances.
Make sense.  Much as we don't bother exporting scales of 1 or offsets of 0.
> 
> What do you think ?
> Grégor.
> 
>>
>> Jonathan
>>
>>> ---
>>>   .../bindings/iio/magnetometer/ak8975.txt           | 10 +++++
>>>   drivers/iio/magnetometer/ak8975.c                  | 46 +++++++++++++++++++++-
>>>   2 files changed, 55 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/Documentation/devicetree/bindings/iio/magnetometer/ak8975.txt b/Documentation/devicetree/bindings/iio/magnetometer/ak8975.txt
>>> index 34a3206..f936f86 100644
>>> --- a/Documentation/devicetree/bindings/iio/magnetometer/ak8975.txt
>>> +++ b/Documentation/devicetree/bindings/iio/magnetometer/ak8975.txt
>>> @@ -9,6 +9,7 @@ Optional properties:
>>>       - gpios : should be device tree identifier of the magnetometer DRDY pin
>>>     - vdd-supply: an optional regulator that needs to be on to provide VDD
>>> +  - matrix: an optional 3x3 mounting rotation matrix
>>>     Example:
>>>   @@ -17,4 +18,13 @@ ak8975@0c {
>>>           reg = <0x0c>;
>>>           gpios = <&gpj0 7 0>;
>>>           vdd-supply = <&ldo_3v3_gnss>;
>>> +        matrix = "-0.984807753012208",  /* x0 */
>>> +                 "0",                   /* y0 */
>>> +                 "-0.173648177666930",  /* z0 */
>>> +                 "0",                   /* x1 */
>>> +                 "-1",                  /* y1 */
>>> +                 "0",                   /* z1 */
>>> +                 "-0.173648177666930",  /* x2 */
>>> +                 "0",                   /* y2 */
>>> +                 "0.984807753012208";   /* z2 */
>>>   };
>>> diff --git a/drivers/iio/magnetometer/ak8975.c b/drivers/iio/magnetometer/ak8975.c
>>> index 72c03d9..95c68952 100644
>>> --- a/drivers/iio/magnetometer/ak8975.c
>>> +++ b/drivers/iio/magnetometer/ak8975.c
>>> @@ -370,6 +370,7 @@ struct ak8975_data {
>>>       wait_queue_head_t    data_ready_queue;
>>>       unsigned long        flags;
>>>       u8            cntl_cache;
>>> +    const char        *matrix[9];
>>>   };
>>>     /*
>>> @@ -697,6 +698,28 @@ static int ak8975_read_raw(struct iio_dev *indio_dev,
>>>       return -EINVAL;
>>>   }
>>>   +static ssize_t ak8975_show_matrix(struct device *dev,
>>> +                  struct device_attribute *attr,
>>> +                  char *buf)
>>> +{
>>> +    const struct ak8975_data *data = iio_priv(dev_to_iio_dev(dev));
>>> +    const char * const *m = data->matrix;
>>> +
>>> +    return sprintf(buf, "%s, %s, %s; %s, %s, %s; %s, %s, %s\n",
>>> +               m[0], m[1], m[2], m[3], m[4], m[5], m[6], m[7], m[8]);
>>> +}
>>> +
>>> +static IIO_DEVICE_ATTR(in_magn_matrix, S_IRUGO, ak8975_show_matrix, NULL, -1);
>>> +
>>> +static struct attribute *ak8975_attrs[] = {
>>> +    &iio_dev_attr_in_magn_matrix.dev_attr.attr,
>>> +    NULL
>>> +};
>>> +
>>> +static const struct attribute_group ak8975_attrs_group = {
>>> +    .attrs = ak8975_attrs
>>> +};
>>> +
>>>   #define AK8975_CHANNEL(axis, index)                    \
>>>       {                                \
>>>           .type = IIO_MAGN,                    \
>>> @@ -716,6 +739,12 @@ static const struct iio_info ak8975_info = {
>>>       .driver_module = THIS_MODULE,
>>>   };
>>>   +static const struct iio_info ak8975_matrix_info = {
>>> +    .read_raw = &ak8975_read_raw,
>>> +    .attrs = &ak8975_attrs_group,
>>> +    .driver_module = THIS_MODULE,
>>> +};
>>> +
>>>   static const struct acpi_device_id ak_acpi_match[] = {
>>>       {"AK8975", AK8975},
>>>       {"AK8963", AK8963},
>>> @@ -785,6 +814,22 @@ static int ak8975_probe(struct i2c_client *client,
>>>       data->eoc_gpio = eoc_gpio;
>>>       data->eoc_irq = 0;
>>>   +    /*
>>> +     * Rotation matrix is expressed as an array of 3x3 strings to be able
>>> +     * to represent floating point numbers.
>>> +     */
>>> +    err = of_property_read_string_array(client->dev.of_node, "matrix",
>>> +                        data->matrix,
>>> +                        ARRAY_SIZE(data->matrix));
>>> +    if (err == ARRAY_SIZE(data->matrix))
>>> +        indio_dev->info = &ak8975_matrix_info;
>>> +    else if (err == -EINVAL)
>>> +        indio_dev->info = &ak8975_info;
>>> +    else if (err >= 0)
>>> +        return -EINVAL;
>>> +    else
>>> +        return err;
>>> +
>>>       /* id will be NULL when enumerated via ACPI */
>>>       if (id) {
>>>           chipset = (enum asahi_compass_chipset)(id->driver_data);
>>> @@ -822,7 +867,6 @@ static int ak8975_probe(struct i2c_client *client,
>>>       indio_dev->dev.parent = &client->dev;
>>>       indio_dev->channels = ak8975_channels;
>>>       indio_dev->num_channels = ARRAY_SIZE(ak8975_channels);
>>> -    indio_dev->info = &ak8975_info;
>>>       indio_dev->modes = INDIO_DIRECT_MODE;
>>>       indio_dev->name = name;
>>>       return devm_iio_device_register(&client->dev, indio_dev);
>>>
> 
> -- 
> To unsubscribe from this list: send the line "unsubscribe linux-iio" in
> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2 4/5] iio:magnetometer:ak8975: mounting matrix support

[Rob Herring]

On Wed, Mar 9, 2016 at 2:46 PM, Jonathan Cameron <jic23-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org> wrote:
> On 09/03/16 10:56, Gregor Boirie wrote:
>>
>> On 03/05/2016 04:36 PM, Jonathan Cameron wrote:
>>> On 03/03/16 10:44, Gregor Boirie wrote:
>>>> Expose a rotation matrix to indicate userspace the chip placement with
>>>> respect to the overall hardware system. This is needed to adjust
>>>> coordinates sampled from a magnetometer chip when its position deviates
>>>> from the main hardware system.
>>>>
>>>> Final coordinates computation is delegated to userspace since:
>>>> * computation may involve floating point arithmetics ;
>>>> * it allows an application to combine adjustments with arbitrary
>>>>    transformations.
>>>>
>>>> This 3 dimentional space rotation matrix is expressed as 3x3 array of
>>>> strings to support floating point numbers. It may be retrieved from a
>>>> "in_magn_matrix" sysfs attribute file. It is declared into ak8975 DTS
>>>> entry as a "matrix" property.
>>>>
>>>> Signed-off-by: Gregor Boirie <gregor.boirie-ITF29qwbsa/QT0dZR+AlfA@public.gmane.org>
>>> This one definitely wants to go to the device tree list. I'd also like a few
>>> IIO related comments as well. I was sure we'd been through this before and
>>> ended up with some sort of solution, but now I can't find it... Oh well.
>> I took inv_mpu_core as an example. However matrices are exposed as integers
>> and we really need floating point support.
>> A search on the device-tree mailing list archive does not show up with a
>> solution to encode floating point numbers into DTS either.
> Certainly non obvious how to do this.
>
> For anyone device tree related who picks this up (I've added a CC)
> we are basically talking about passing a rotation matrix from the device tree
> straight through to userspace to describe the relative orientation of
> some sensor.  An array of strings clearly works, but is this the best
> way to do it?

That's probably okay. I'd format the dts and sysfs to be arranged in a
3x3 matrix rather than a linear list though, but that is just style.

However, there's nothing really preventing us from having floats in
DT. Data is not typed, so it would be a matter of adding the dtc
parsing support. The other aspect would be the limited use of float
types in the kernel (though that should not influence binding design).
Are there cases of the kernel ABI using floats?

Rob

Re: [PATCH v2 4/5] iio:magnetometer:ak8975: mounting matrix support

[Jonathan Cameron]

On 10/03/16 02:15, Rob Herring wrote:
> On Wed, Mar 9, 2016 at 2:46 PM, Jonathan Cameron <jic23-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org> wrote:
>> On 09/03/16 10:56, Gregor Boirie wrote:
>>>
>>> On 03/05/2016 04:36 PM, Jonathan Cameron wrote:
>>>> On 03/03/16 10:44, Gregor Boirie wrote:
>>>>> Expose a rotation matrix to indicate userspace the chip placement with
>>>>> respect to the overall hardware system. This is needed to adjust
>>>>> coordinates sampled from a magnetometer chip when its position deviates
>>>>> from the main hardware system.
>>>>>
>>>>> Final coordinates computation is delegated to userspace since:
>>>>> * computation may involve floating point arithmetics ;
>>>>> * it allows an application to combine adjustments with arbitrary
>>>>>    transformations.
>>>>>
>>>>> This 3 dimentional space rotation matrix is expressed as 3x3 array of
>>>>> strings to support floating point numbers. It may be retrieved from a
>>>>> "in_magn_matrix" sysfs attribute file. It is declared into ak8975 DTS
>>>>> entry as a "matrix" property.
>>>>>
>>>>> Signed-off-by: Gregor Boirie <gregor.boirie-ITF29qwbsa/QT0dZR+AlfA@public.gmane.org>
>>>> This one definitely wants to go to the device tree list. I'd also like a few
>>>> IIO related comments as well. I was sure we'd been through this before and
>>>> ended up with some sort of solution, but now I can't find it... Oh well.
>>> I took inv_mpu_core as an example. However matrices are exposed as integers
>>> and we really need floating point support.
>>> A search on the device-tree mailing list archive does not show up with a
>>> solution to encode floating point numbers into DTS either.
>> Certainly non obvious how to do this.
>>
>> For anyone device tree related who picks this up (I've added a CC)
>> we are basically talking about passing a rotation matrix from the device tree
>> straight through to userspace to describe the relative orientation of
>> some sensor.  An array of strings clearly works, but is this the best
>> way to do it?
> 
> That's probably okay. I'd format the dts and sysfs to be arranged in a
> 3x3 matrix rather than a linear list though, but that is just style.
> 
> However, there's nothing really preventing us from having floats in
> DT. Data is not typed, so it would be a matter of adding the dtc
> parsing support. The other aspect would be the limited use of float
> types in the kernel (though that should not influence binding design).
> Are there cases of the kernel ABI using floats?
Not many cases of true floats, but certainly in IIO we use a lot of
pseudo floats in which an enum is combined with two integers.
Enum effectively specifies the 'fixed point' location and the
pair of integers typically the integer part and the decimal part
(as we often end up with for example scales that are close to 1)
It's a bit hideous in some ways, but seeing as we are doing fixed
point maths with it in the kernel it actually works out reasonably
cleanly internally.

Right now I don't think any of those are passed in from device tree
but it's certainly possible in the future.

Jonathan
> 
> Rob
> --
> To unsubscribe from this list: send the line "unsubscribe linux-iio" in
> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> 

Re: [PATCH v2 4/5] iio:magnetometer:ak8975: mounting matrix support

[Gregor Boirie]

On 03/09/2016 09:46 PM, Jonathan Cameron wrote:

[snip...]
> Definitely needs ABI docs in Documentation/ABI/testing/sysfs-bus-iio
> as well. We need the reference directions to be well specified as
> well - how do we define the main hardware orientation?  on yours presumably
> the usual approach of using the screen is irrelevant so that definition
> could be a little 'interesting' to write down?
>> The standard orientation for us is defined with respect to the drone's head
>> and the propellers plane, which does not suit the nominal use cases indeed...
>> Although it really deserves some more detailed documentation, I wonder if the
>> question is relevant.
>> Given that the DTS is board specific (correct me if I'm wrong), why not
>> consider the reference orientation as board specific as well (and make the
>> whole thing implicit) ?
> I think that is fine for your sort of usecase - though might be nice to keep
> a bit of documentation somewhere (perhaps as comments in the DTS), or perhaps
> more centrally - would be nice for userspace libraries if at least all aircraft
> had a similar 'default orientation'.
As you said earlier, things are never that simple. The standard orientation
may vary with types of UAVs (quad-rotor, fixed wings, bizarre home made 
UFOs...)
> For the more 'standard' device with a screen I think we would want to standardise
> if at all possible.  Most hand held devices have a 'natural' orientation
> (front facing camera at the top for example).  In those cases we are basically looking
> at whether the sensor is mounted on the back or the front of the main circuit board and
> then of course it can be at any rotation in plane but typically is at 90 degrees
> (hence the inv mpu limitations)
>
> I'm guessing there are existing userspace mappings for this.  Perhaps in the Gnome
> sensors stuff?
[snip...]
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html