Device Trigger

[tyaglot]

When I install the driver I see the iio device under /sys/bus/iio/devices, but there isn't an associated trigger.

iio_device_is_trigger() returns false for the BMI260 device, and there are no other iio devices on my system.

Without the trigger I'm unable to create an iio buffer to read from the device. I am able to use the raw attribute on the accel and gyro x, y, and z channels to repeatedly poll data from the device, however this doesn't work for the timestamp channel as it has no attributes.

See https://libiio.github.io/libiio/master/libiio/index.html, under heading "Enabling channels and creating the Buffer object": "This call will fail if no channels have been enabled, or for triggered buffers, if the trigger has not been assigned."

Thank you for writing this driver, and for any direction you may be able to provide.

[justinweiss]

If IRQ triggers aren't supported with your device (most devices don't have them hooked up), you'll have to set up an hrtimer trigger yourself. You can follow along with how hhd sets it up here: https://github.com/hhd-dev/hhd/blob/master/src/hhd/controller/physical/imu.py#L473

You probably just need something like this:

mkdir /config/iio/triggers/hrtimer/instance1

Then you can use instance1 as the trigger (see also https://www.kernel.org/doc/Documentation/iio/iio_configfs.txt)

BTW, a version of this driver has been upstreamed, so you should not need this module on recent linux versions anymore.

[tyaglot]

Thank you for your prompt reply.
My kernel wasn't compiled with CONFIG_IIO_CONFIGFS, so I don't see an iio directory when mounting configfs.
Is there another interface to create a trigger without requiring recompilation of the kernel?

[justinweiss]

It wasn't built as a module or anything, right?

I'm not aware of another way to do that -- I haven't worked much with the hrtimer triggers because the main device I use has IRQ triggers, so I only know the basics there.

[tyaglot]

I built the driver as a DKMS module, but the CONFIG_IIO_CONFIGFS is a kernel parameter that is set before compiling the kernel.

Would it be possible to modify the timestamp channel to expose the value as a raw attribute? This wouldn't be appropriate for the driver at large, but it would work for my use case. Could you give me a pointer on how to do so?

[justinweiss]

Hm, changing the iio_chan_spec to use a different (unused) channel for the timestamp might work? I'm not sure.

Is there a particular reason you need the timestamp channel and can't simulate it? For example, inputplumber just polls the accel / gyro attributes and doesn't use a timestamp.

[tyaglot]

Currently I'm polling the attributes and using the system time as a timestamp, discarding any values that are the same as the last polled value. These timestamps are noisy and I wanted to see if the ones generated by the IMU were more consistent. I could just assume a constant rate of new data and generate synthetic timestamps at the operating frequency of the IMU like you're suggesting.
Procrastination has many forms...