The push-to-talk demo used Zephyr’s microphone driver as it is. The board’s IMU went the other way.
The XIAO nRF54L15 Sense has an LSM6DS3TR-C accelerometer and gyro. The chip has a tap engine: set it up, and it
pulses a pin when it feels a tap, with the CPU asleep. Zephyr has a driver for the chip family (st,lsm6dsl), but
its only trigger is data-ready. It has no tap detection. That left three choices:
- add tap support to Zephyr’s C driver;
- read samples at 416 Hz and detect taps in software, which keeps the CPU awake;
- turn Zephyr’s driver off and let Toit own the chip.
ztoit took the third. All it needed from C was one generic I2C primitive (35 lines): write, read, or write then read,
on any Zephyr I2C bus. On top of that, zephyr.I2c (57 lines) implements Toit’s serial.Device, so register access
is Toit’s own serial.Registers. The tap driver is 42 lines of Toit; six register writes set up the tap engine.
The interrupt needed no new C at all. The IMU’s INT1 pin is declared in the devicetree overlay as a gpio-keys
button, so Toit waits on it the same way as on the board’s real button: zephyr.Key "imu_int1" and wait-for-press.
The task blocks, Zephyr’s idle thread runs, and the pin’s interrupt wakes the CPU.
What broke
- No tap reached Toit. The chip saw every tap, but its pulse lasts a few milliseconds, and
gpio-keyswaits out a 30 ms debounce before reading the pin. A 1 ms debounce on that one node fixed it. - The board rings. At a threshold low enough to catch every tap, one sharp tap gave several events over about a second. One rule in Toit: anything within 1000 ms of a tap is the same tap.
- Starting up reads as a tap (0 to 1 g), so the driver waits 50 ms before routing taps to the pin.
- The IMU stays powered across a reflash, so the driver starts with a software reset.
Zephyr’s driver against Toit’s
| Zephyr C sample + driver | Toit demo + driver | |
|---|---|---|
| Source | 133 code lines (the sample) | 61 code lines (demo and driver) |
| Taps | not supported | 6 register writes |
| The interrupt | a trigger callback in C | wait-for-press on a key |
| A wrong address | an error code to check | an exception |
| Flash | — | +13 KB over the bare VM |
The C that went in is generic: an I2C primitive and a pin. What is specific to this chip, the register settings, the timings and the fixes for ringing and start-up, is all in Toit.
When to write the driver in Toit
Use Zephyr’s driver when it does what you need: the microphone’s dmic driver did, and Toit wraps it in three
primitives. Write the driver in Toit when Zephyr’s driver lacks the feature, as here. Then C only has to provide the
bus, and the chip-specific logic stays in a language where an error is an exception.
The driver (app/lsm6ds3.toit) and the journal entry are in ktau; the I2C primitive is in ztoit.