Re: LIS331DLH accelerometer driver, IIO or not?

[Darren Hart <dvhart@linux.intel.com>]

On 05/27/2012 02:32 AM, Jonathan Cameron wrote:
> On 05/27/2012 04:14 AM, Darren Hart wrote:
>>
>>
>> On 05/26/2012 10:40 AM, Jonathan Cameron wrote:
>>> On 05/26/2012 12:53 PM, Éric Piel wrote:
>>>> On 25-05-12 07:10, Greg Kroah-Hartman wrote:
>>>>> On Thu, May 24, 2012 at 09:29:53PM -0700, Darren Hart wrote:
>>>>>> I'm working to enable the LIS331DLH accelerometer on the Fish River
>>>>>> Island II embedded atom development kit.
>>>>>>
>>>>>> I am more interested in enabling people to do bizarre and interesting
>>>>>> things with the device, so I'm leaning toward continuing with my IIO
>>>>>> implementation.
>>>>>
>>>>> Make it an IIO driver and then we can delete the misc driver, which
>>>>> shouldn't have snuck in there in the first place :)
>>>>>
>>>>
>>>> To be more fair to the misc driver, I wouldn't say it snucked in there,
>>>> but more "it ended up there as the least worse place" ;-) Actually, the
>>>> main problem is that there seemed to be no maintainer interested in
>>>> taking care of accelerometer devices. Now that the IIO subsystem is out
>>>> of staging, it might be a right place. That said, I don't know much
>>>> about the user interface to IIO. I know that I liked the idea of having
>>>> an joystick device created for an accelerometer because that allows to
>>>> get many programs to access the device almost without any modifications.
>>> I agree that this sort of device should have an input interface. Not
>>> sure if a joystick is the right option, but that's more one for Dmitry
>>> to comment on.
>>>>
>>>> I'd happy to help merge the lis3lv02d driver into IIO. IMHO, the main
>>>> steps are:
>>>>  * make sure all the various buses are supported (e.g., I²C, SPI, and
>>>> also "ACPI-HP")
>>>>  * ensure the various versions of the accelerometer are supported (there
>>>> are 3 supported currently)
>>>>  * check that the driver is automatically loaded on HP laptops (via ACPI
>>>> entry)
>>>>  * for each of the current interfaces decide if they should be ported or
>>>> dropped (/dev/js*, /dev/freefall, sysfs...)
>>>>
>>>> What do you think Jonathan?
>>> You've laid it out extremely clearly.  Thanks, I agree with these steps,
>>> though they may occur from a slightly different angle given Darren is
>>> interested in a part that is not (I believe) currently supported by
>>> your existing driver.   Hence he may initially want to do a separate
>>> driver with that (keeping in mind the aim of mergining in the existing
>>> driver).  I have an ancient driver for the lis3l02dq alone (in
>>> iio from the start) that will also get eaten up by Daren's new driver
>>> (and the ability to test on that part on spi).
>>> staging/iio/accel/lis3l02dq*.c
>>
>> I think I should start with getting the lis331dlh support completed, if
>> for no other reason than to keep the scope manageable as I write my
>> first real driver. From that I would like to merge in Jonathan's IIO
>> lis3l02dq driver to get the multi-chip support part right. Then we
>> should look at expanding the scope of the interface and finally merging
>> with misc/lis3102dq. I believe that should meet with everyone's suggestions.
>>
>> One thing I would like to understand better is what sort of interface
>> does userspace current expect. Phone Gap, for example, provides a very
>> high level interface to applications in m/s^2 for each axis. Is there
>> some interface we should ensure all accelerometer driver's implement?
>>
>> I suspect a /dev/accel interface that reads out xyz values in ms/s^2
>> would make sense. We would need to ensure that allows for polled as well
>> as event driven. Thoughts?
> 
> Immediate comment is don't call it /dev/accel.  That's just taken out
> using it for all the combined gyro/ accel units out there.

Sure, that was just off the cuff.

> Also doing scaling in kernel is expensive, fiddly and inefficient
> (no floating point - and lots of users don't actually care about
> absolute scale).

I was thinking about that yesterday as well. I suppose all that is
needed is to expose the range and the scale (e.g. max=32M, scale=8G)

> Then we get the question of what this gains us over
> existing options (input or /dev/iio/iio:device0) beyond a memorable
> location?  Maybe we leave this question for now...

My only complaint with the iio:device0 naming is the ":" which has to be
escaped and is very irritating to use from the shell!

-- 
Darren Hart
Intel Open Source Technology Center
Yocto Project - Linux Kernel