Build error with BMI270 on Thingy:53 — ZERO_OR_COMPILE_ERROR in DEVICE_DT_GET_ONE

[Renan Vinicius]

Hi, I’m trying to build the github.com/.../neuton-nordic-thingy53-ble-remotecontrol project using nRF Connect SDK v2.6.2.
I added the BMI270 sensor on SPI3 in my DeviceTree overlay:

&spi3 { bmi270: spi-dev-bmi270@1 { status = "okay"; }; };

In my code (bsp_imu.c), I try to initialize the device like this:

imu_ctx_.dev = DEVICE_DT_GET_ONE(bosch_bmi270);

But during build I get the following error:

error: size of unnamed array is negative zephyr/include/zephyr/sys/util.h:94:55: note: in expansion of macro 'ZERO_OR_COMPILE_ERROR' zephyr/include/zephyr/device.h:286:22: note: in expansion of macro 'DEVICE_DT_GET_ONE' bsp_imu.c:38:22: note: in expansion of macro 'DEVICE_DT_GET_ONE'

It looks like DEVICE_DT_GET_ONE(bosch_bmi270) doesn’t find any compatible in the DeviceTree, which causes the compile-time error.

[Amanda Hsieh]

Hi, 

Do you use the toolchain v3.0.0. with SDK v2.6.2 as the Setup firmware project section?

That project already includes support for BMI 270, so you don't have to modify anything.  

https://github.com/Neuton-tinyML/neuton-nordic-thingy53-ble-remotecontrol/blob/bbeca10014ac76825fc44739cca75f683c95548f/thingy53_nrf5340_cpuapp.overlay#L14

https://github.com/Neuton-tinyML/neuton-nordic-thingy53-ble-remotecontrol/blob/bbeca10014ac76825fc44739cca75f683c95548f/src/bsp/sensor/imu/bsp_imu.c#L38

https://github.com/Neuton-tinyML/neuton-nordic-thingy53-ble-remotecontrol/blob/bbeca10014ac76825fc44739cca75f683c95548f/prj.conf#L21 

Please refer to the instructions in the Setup firmware project section to build the project. 

Regards,
Amanda H. 

[SergeiA]

Hi,

This answer doesn't really help. In my application I am getting the same problem after I did the same changes as in remotecontrol project. I am using toolchain 3.1.1 and SDK 3.1.1, but result with 3.0.2 is similar.

My overaly file is the same:

&spi3 {
	bmi270: spi-dev-bmi270@1 {
		status = "okay";
	};
};

prj.config has the same settings, related to bmi270:

# SPI configuration
CONFIG_SPI=y

# Sensor configuration
CONFIG_SENSOR=y

# Shell support
CONFIG_SHELL=y
CONFIG_SHELL_BACKEND_SERIAL=y

# Enable sensor driver
CONFIG_ADXL362=n
CONFIG_BMI270=y

In sensor creation code I am checking, if sensor is available in devicetree

    const struct device *dev = NULL;
#if defined(CONFIG_ADXL362)
    dev = DEVICE_DT_GET_ONE(adi_adxl362);
#endif
#if defined(CONFIG_BMI270)
#if !DT_NODE_EXISTS(bosh_bmi270)
#error("BMI270 is not enabled in devicetree") // << I am getting this error 
#endif
    dev = DEVICE_DT_GET_ONE(bosh_bmi270);
#endif

When CONFIG_ADXL362 is defined, everything compiles, but when CONFIG_BMI270 is enabled, I am getting build error "BMI270 is not enabled in devtree" and later DEVICE_DT_GET_ONE() fails with usual cryptic "error: size of unnamed array is negative", which is expected after first one. So something else is missing, could you help with this?

Thank you,

Sergei

[SergeiA]

Ok, I moved forward a bit with this issue. When I replaced macro DEVICE_DT_GET_ONE() in the code above by direct API call

    const char * sensor_name = "spi-dev-bmi270@1";
    dev = device_get_binding(sensor_name);

I was able to debug this (why do we still using opaque macros anyway? Ah, that's because Linux people refused to code in C++...). Initial attempt with name "bosh_bmi270" failed because in device tree this sensor known under name "spi-dev-bmi270@1" -- debugger helped to find it out. During enumerating devices it returns pointer to devices, where it possible to check name and status. In my case status was not ready, which means that driver wasn't able to initialize sensor, and this is different problem, and it happens way before I am trying to instantiate this device.

It is still unclear why DEVICE_DT_GET_ONE() keeps failing the build no matter what name I am using, but at least workaround is available.

--Sergei