iio/imu/adis16XXX sysfs scale precision lost

[tisyang]

This is not adis16XXX driver bug, actualy is IIO subsystem bug.

I use adis16505-2 on armv7l soc with spi, kernel version 6.1. In iio sysfs, anglvel and accel scale is:

# anglvel scale
cat /sys/bus/iio/devices/iio\:device1/in_anglvel_scale 
0.000000006
# accel scale
cat /sys/bus/iio/devices/iio\:device1/in_accel_scale 
0.000000037

But ref the datasheet and driver code , anglvel(gyro) scale and accel scale should be

gyro:  1/IIO_RAD_TO_DEGREE(40 << 16) => 6.657902725198096e-09
accel(driver): 78.0 /(32000 << 16) => 3.719329833984375e-08
accel(datasheet): 2.45 * 1E-3 / (2**16) => 3.7384033203125005e-08

Compare to these, the sysfs scale lost much precision, especially gyro lost 10% precision.

The sysfs output ref function __iio_format_value of type IIO_VAL_FRACTIONAL :

linux/drivers/iio/industrialio-core.c

Lines 671 to 678 in 10d2a5d

	case IIO_VAL_FRACTIONAL:
	tmp2 = div_s64((s64)vals[0] * 1000000000LL, vals[1]);
	tmp0 = (int)div_s64_rem(tmp2, 1000000000, &tmp1);
	if ((tmp2 < 0) && (tmp0 == 0))
	return sysfs_emit_at(buf, offset, "-0.%09u", abs(tmp1));
	else
	return sysfs_emit_at(buf, offset, "%d.%09u", tmp0,
	abs(tmp1));

Only preserve 9 number after zero point, not enough for adis16505-2 scale.

I don't know how to fix it properly， kernel has not float point, I suppose using scientific notation for fractional values will has no probem.

I know it is not analog driver bug, but it affect usability of adis16xxx imu with iio drivers.

[nunojsa]

Only preserve 9 number after zero point, not enough for adis16505-2 scale.

I guess the most straightforward thing would be to introduce (propose) a new type like IIO_VAL_FRACTIONAL_PICO (or with a better name) where we would add 12 decimals (or even more). The tricky part is that this would be an ABI breakage so it might not be so easy to propose a fix.

[tisyang]

I guess the most straightforward thing would be to introduce (propose) a new type like IIO_VAL_FRACTIONAL_PICO (or with a better name) where we would add 12 decimals (or even more). The tricky part is that this would be an ABI breakage so it might not be so easy to propose a fix.

Maybe better introducing a new type to pass both the numerator and the denominator of the scale directly via sysfs, allowing user space to perform the precise calculation.