Crate embedded_hal

A Hardware Abstraction Layer (HAL) for embedded systems

NOTE This HAL is still is active development. Expect the traits presented here to be tweaked, split or be replaced wholesale before being stabilized, i.e. before hitting the 1.0.0 release. That being said there's a part of the HAL that's currently considered unproven and is hidden behind an "unproven" Cargo feature. This API is even more volatile and it's exempt from semver rules: it can change in a non-backward compatible fashion or even disappear in between patch releases.

Design goals

The HAL

• Must erase device specific details. Neither register, register blocks or magic values should appear in the API.

• Must be generic within a device and across devices. The API to use a serial interface must be the same regardless of whether the implementation uses the USART1 or UART4 peripheral of a device or the UART0 peripheral of another device.

• Where possible must not be tied to a specific asynchronous model. The API should be usable in blocking mode, with the futures model, with an async/await model or with a callback model. (cf. the nb crate)

• Must be minimal, and thus easy to implement and zero cost, yet highly composable. People that want higher level abstraction should prefer to use this HAL rather than re-implement register manipulation code.

• Serve as a foundation for building an ecosystem of platform agnostic drivers. Here driver means a library crate that lets a target platform interface an external device like a digital sensor or a wireless transceiver. The advantage of this system is that by writing the driver as a generic library on top of embedded-hal driver authors can support any number of target platforms (e.g. Cortex-M microcontrollers, AVR microcontrollers, embedded Linux, etc.). The advantage for application developers is that by adopting embedded-hal they can unlock all these drivers for their platform.

Out of scope

• Initialization and configuration stuff like "ensure this serial interface and that SPI interface are not using the same pins". The HAL will focus on doing I/O.

Reference implementation

The stm32f30x-hal crate contains a reference implementation of this HAL.

Platform agnostic drivers

You can find platform agnostic drivers built on top of embedded-hal on crates.io by searching for the embedded-hal keyword.

If you writing a platform agnostic driver yourself you are highly encouraged to add the embedded-hal keyword to your crate before publishing it!

Detailed design

Traits

The HAL is specified as traits to allow generic programming. These traits make use of the nb crate (please go read that crate documentation before continuing) to abstract over the asynchronous model and to also provide a blocking operation mode.

Here's how a HAL trait may look like:

extern crate nb;

/// A serial interface
pub trait Serial {
    /// Error type associated to this serial interface
    type Error;

    /// Reads a single byte
    fn read(&mut self) -> nb::Result<u8, Self::Error>;

    /// Writes a single byte
    fn write(&mut self, byte: u8) -> nb::Result<(), Self::Error>;
}

The nb::Result enum is used to add a WouldBlock variant to the errors of the serial interface. As explained in the documentation of the nb crate this single API, when paired with the macros in the nb crate, can operate in a blocking manner, or in a non-blocking manner compatible with futures and with the await! operator.

Some traits, like the one shown below, may expose possibly blocking APIs that can't fail. In those cases nb::Result<_, !> is used.

#![feature(never_type)]

extern crate nb;

/// A count down timer
pub trait CountDown {
    // ..

    /// "waits" until the count down is over
    fn wait(&mut self) -> nb::Result<(), !>;
}

Suggested implementation

The HAL traits should be implemented for device crates generated via svd2rust to maximize code reuse.

Shown below is an implementation of some of the HAL traits for the stm32f30x crate. This single implementation will work for any microcontroller in the STM32F30x family.

// crate: stm32f30x-hal
//! An implementation of the `embedded-hal` traits for STM32F30x microcontrollers

extern crate embedded_hal as hal;
extern crate nb;

// device crate
extern crate stm32f30x;

use stm32f30x::USART1;

/// A serial interface
// NOTE generic over the USART peripheral
pub struct Serial<USART> { usart: USART }

// convenience type alias
pub type Serial1 = Serial<USART1>;

/// Serial interface error
pub enum Error {
    /// Buffer overrun
    Overrun,
    // omitted: other error variants
}

impl hal::serial::Read<u8> for Serial<USART1> {
    type Error = Error;

    fn read(&mut self) -> nb::Result<u8, Error> {
        // read the status register
        let isr = self.usart.isr.read();

        if isr.ore().bit_is_set() {
            // Error: Buffer overrun
            Err(nb::Error::Other(Error::Overrun))
        }
        // omitted: checks for other errors
        else if isr.rxne().bit_is_set() {
            // Data available: read the data register
            Ok(self.usart.rdr.read().bits() as u8)
        } else {
            // No data available yet
            Err(nb::Error::WouldBlock)
        }
    }
}

impl hal::serial::Write<u8> for Serial<USART1> {
    type Error = Error;

    fn write(&mut self, byte: u8) -> nb::Result<(), Error> {
        // Similar to the `read` implementation
    }

    fn flush(&mut self) -> nb::Result<(), Error> {
        // Similar to the `read` implementation
    }
}

Intended usage

Thanks to the nb crate the HAL API can be used in a blocking manner, with futures or with the await operator using the block!, try_nb! and await! macros respectively.

Blocking mode

An example of sending a string over the serial interface in a blocking fashion:

extern crate embedded_hal;
#[macro_use(block)]
extern crate nb;

use stm32f30x_hal::Serial1;
use embedded_hal::serial::Write;

let mut serial: Serial1 = {
    // ..
};

for byte in b"Hello, world!" {
    // NOTE `block!` blocks until `serial.write()` completes and returns
    // `Result<(), Error>`
    block!(serial.write(*byte)).unwrap();
}

futures

An example of running two tasks concurrently. First task: blink an LED every second. Second task: loop back data over the serial interface.

#![feature(conservative_impl_trait)]
#![feature(never_type)]

extern crate embedded_hal as hal;
extern crate futures;

#[macro_use(try_nb)]
extern crate nb;

use hal::prelude::*;
use futures::{
    future,
    Async,
    Future,
};
use futures::future::Loop;
use stm32f30x_hal::{Led, Serial1, Timer6};

/// `futures` version of `CountDown.wait`
///
/// This returns a future that must be polled to completion
fn wait<T>(mut timer: T) -> impl Future<Item = T, Error = !>
where
    T: hal::timer::CountDown,
{
    let mut timer = Some(timer);
    future::poll_fn(move || {
        try_nb!(timer.as_mut().unwrap().wait());

        Ok(Async::Ready(timer.take().unwrap()))
    })
}

/// `futures` version of `Serial.read`
///
/// This returns a future that must be polled to completion
fn read<S>(mut serial: S) -> impl Future<Item = (S, u8), Error = S::Error>
where
    S: hal::serial::Read<u8>,
{
    let mut serial = Some(serial);
    future::poll_fn(move || {
        let byte = try_nb!(serial.as_mut().unwrap().read());

        Ok(Async::Ready((serial.take().unwrap(), byte)))
    })
}

/// `futures` version of `Serial.write`
///
/// This returns a future that must be polled to completion
fn write<S>(mut serial: S, byte: u8) -> impl Future<Item = S, Error = S::Error>
where
    S: hal::serial::Write<u8>,
{
    let mut serial = Some(serial);
    future::poll_fn(move || {
        try_nb!(serial.as_mut().unwrap().write(byte));

        Ok(Async::Ready(serial.take().unwrap()))
    })
}

fn main() {
    // HAL implementers
    let timer: Timer6 = {
        // ..
    };
    let serial: Serial1 = {
        // ..
    };
    let led: Led = {
        // ..
    };

    // Tasks
    let mut blinky = future::loop_fn::<_, (), _, _>(
        (led, timer, true),
        |(mut led, mut timer, state)| {
            wait(timer).map(move |timer| {
                if state {
                    led.on();
                } else {
                    led.off();
                }

                Loop::Continue((led, timer, !state))
            })
        });

    let mut loopback = future::loop_fn::<_, (), _, _>(serial, |mut serial| {
        read(serial).and_then(|(serial, byte)| {
            write(serial, byte)
        }).map(|serial| {
            Loop::Continue(serial)
        })
    });

    // Event loop
    loop {
        blinky.poll().unwrap(); // NOTE(unwrap) E = !
        loopback.poll().unwrap();
    }
}

await

Same example as above but using await! instead of futures.

#![feature(generator_trait)]
#![feature(generators)]

extern crate embedded_hal as hal;

#[macro_use(await)]
extern crate nb;

use std::ops::Generator;

use hal::prelude::*;
use stm32f30x_hal::{Led, Serial1, Timer6};

fn main() {
    // HAL implementers
    let mut timer: Timer6 = {
        // ..
    };
    let mut serial: Serial1 = {
        // ..
    };
    let mut led: Led = {
        // ..
    };

    // Tasks
    let mut blinky = (move || {
        let mut state = false;
        loop {
            // `await!` means "suspend / yield here" instead of "block until
            // completion"
            await!(timer.wait()).unwrap(); // NOTE(unwrap) E = !

            state = !state;

            if state {
                led.on();
            } else {
                led.off();
            }
        }
    });

    let mut loopback = (move || {
        loop {
            let byte = await!(serial.read()).unwrap();
            await!(serial.write(byte)).unwrap();
        }
    });

    // Event loop
    loop {
        blinky.resume();
        loopback.resume();
    }
}

Generic programming and higher level abstractions

The core of the HAL has been kept minimal on purpose to encourage building generic higher level abstractions on top of it. Some higher level abstractions that pick an asynchronous model or that have blocking behavior and that are deemed useful to build other abstractions can be found in the blocking module and, in the future, in the futures and async modules.

Some examples:

NOTE All the functions shown below could have been written as trait methods with default implementation to allow specialization, but they have been written as functions to keep things simple.

• Write a whole buffer to a serial device in blocking a fashion.

extern crate embedded_hal as hal;
#[macro_use(block)]
extern crate nb;

use hal::prelude::*;

fn write_all<S>(serial: &mut S, buffer: &[u8]) -> Result<(), S::Error>
where
    S: hal::serial::Write<u8>
{
    for &byte in buffer {
        block!(serial.write(byte))?;
    }

    Ok(())
}

• Blocking serial read with timeout

extern crate embedded_hal as hal;
extern crate nb;

use hal::prelude::*;

enum Error<E> {
    /// Serial interface error
    Serial(E),
    TimedOut,
}

fn read_with_timeout<S, T>(
    serial: &mut S,
    timer: &mut T,
    timeout: T::Time,
) -> Result<u8, Error<S::Error>>
where
    T: hal::timer::CountDown,
    S: hal::serial::Read<u8>,
{
    timer.start(timeout);

    loop {
        match serial.read() {
            // raise error
            Err(nb::Error::Other(e)) => return Err(Error::Serial(e)),
            Err(nb::Error::WouldBlock) => {
                // no data available yet, check the timer below
            },
            Ok(byte) => return Ok(byte),
        }

        match timer.wait() {
            Err(nb::Error::Other(e)) => {
                // The error type specified by `timer.wait()` is `!`, which
                // means no error can actually occur. The Rust compiler
                // still forces us to provide this match arm, though.
                unreachable!()
            },
            // no timeout yet, try again
            Err(nb::Error::WouldBlock) => continue,
            Ok(()) => return Err(Error::TimedOut),
        }
    }
}

• Asynchronous SPI transfer

#![feature(conservative_impl_trait)]
#![feature(generators)]
#![feature(generator_trait)]

extern crate embedded_hal as hal;
#[macro_use(await)]
extern crate nb;

use std::ops::Generator;

/// Transfers a byte buffer of size N
///
/// Returns the same byte buffer but filled with the data received from the
/// slave device
fn transfer<S, B>(
    mut spi: S,
    mut buffer: [u8; 16], // NOTE this should be generic over the size of the array
) -> impl Generator<Return = Result<(S, [u8; 16]), S::Error>, Yield = ()>
where
    S: hal::spi::FullDuplex<u8>,
{
    move || {
        let n = buffer.len();
        for i in 0..n {
            await!(spi.send(buffer[i]))?;
            buffer[i] = await!(spi.read())?;
        }

        Ok((spi, buffer))
    }
}

• Buffered serial interface with periodic flushing in interrupt handler

extern crate embedded_hal as hal;
extern crate nb;

use hal::prelude::*;

fn flush<S>(serial: &mut S, cb: &mut CircularBuffer)
where
    S: hal::serial::Write<u8, Error = !>,
{
    loop {
        if let Some(byte) = cb.peek() {
            match serial.write(*byte) {
                Err(nb::Error::Other(_)) => unreachable!(),
                Err(nb::Error::WouldBlock) => return,
                Ok(()) => {}, // keep flushing data
            }
        }

        cb.pop();
    }
}

// The stuff below could be in some other crate

/// Global singleton
pub struct BufferedSerial1;

// NOTE private
static BUFFER1: Mutex<CircularBuffer> = {
    // ..
};
static SERIAL1: Mutex<Serial1> = {
    // ..
};

impl BufferedSerial1 {
    pub fn write(&self, byte: u8) {
        self.write_all(&[byte])
    }

    pub fn write_all(&self, bytes: &[u8]) {
        let mut buffer = BUFFER1.lock();
        for byte in bytes {
            buffer.push(*byte).expect("buffer overrun");
        }
        // omitted: pend / enable interrupt_handler
    }
}

fn interrupt_handler() {
    let mut serial = SERIAL1.lock();
    let mut buffer = BUFFER1.lock();

    flush(&mut *serial, &mut buffer);
}

Modules

blocking	Blocking API
digital	Digital I/O
prelude	The prelude is a collection of all the traits in this crate
serial	Serial interface
spi	Serial Peripheral Interface
timer	Timers

Traits

PwmPin

A single PWM channel / pin