Follow

Follow
Sharing Data Among Tasks in Rust Embassy: Synchronization Primitives

Sharing Data Among Tasks in Rust Embassy: Synchronization Primitives

Omar Hiari's photo
Omar Hiari
Β·Jan 9, 2023Β·

11 min read

Table of contents

Introduction

If you haven't experienced it yet, dealing with global variables in embedded Rust could be a rough experience. Though for good reason I would say. The challenge stems from making sure that variables are shared in a "safe" manner among threads to prevent cases like data races. However, using embassy, the experience is rectified through several synchronization primitives provided through the embassy-sync crate. Though the question is, why are there so many different primitives? and which should you use? The answer lies in how you plan to share the data. Meaning, do you want to only share it among tasks in a blocking manner? Do you want the primitive to support async? better yet, do you want the primitive to notify a task only when the data it holds changes?

Please note that the full code in this post is available on the apollolabsdev Nucleo-F401RE git repo. Additionally, the hardware used in this post is the Nucleo-F401RE board.

The List of Primitives

The embassy-sync crate provides the following primitives (along with the associated description from the documentation):

  • Channel - A Multiple Producer Multiple Consumer (MPMC) channel. Each message is only received by a single consumer.

  • PubSubChannel - A broadcast channel (publish-subscribe) channel. Each message is received by all consumers.

  • Signal - Signalling latest value to a single consumer.

  • Mutex - Mutex for synchronizing state between asynchronous tasks.

  • Pipe - Byte stream implementing embedded_io traits.

There are also Waker primitives which are listed below. I won't be covering these here, but probably in a later post. Though for now, it is sufficient to know that waker primitives are a way to signal the executor to poll a future.

  • WakerRegistration - Utility to register and wake a Waker.

  • AtomicWaker - A variant of WakerRegistration accessible using a non-mut API.

  • MultiWakerRegistration - Utility registering and waking multiple Waker’s.

So this all looks great, but how or when would I choose which to use?

The Use Cases

The use of the different primitives can be split into three cases:

  1. Reading/Writing from/to multiple tasks: This is a common case where all one would need is a simple read/write from/to a variable in multiple tasks.

  2. Reading/Writing across async tasks: These are cases where the value that is being held by a variable needs to be held across async tasks. Consequently, this means holding a lock while awaiting.

  3. Wait for Value Change: These are cases where more than read/write is needed. In particular, a receiving task waiting for a change in value.

In the following sections the constructs available under each category are covered.

Reading/Writing from Multiple Tasks

The AtomicU32 Type

You might have noticed that AtomicU32 is not in the list presented earlier. That is because AtomicU32 is not really an embassy-sync primitive but rather a core::sync primitive. I decided to add it to this post for the sake of completion as I found it really helpful myself. The AtomicU32 is relatively easy to use and is good enough if you just want to share a simple value among tasks. However, AtomicU32 works only for types that are u32 or less in size. If the size is larger, one needs to defer to using a global blocking Mutex which is presented after.

Below is an example of the usage of AtomicU32 . The new instance method is used to create and initialize a AtomicU32 type SHARED variable and initialize it to 0. Following that, store() and load() methods are used to store and load values to/from the global context.

In the code below, in the async_task task, the value in SHARED is incremented every 1 second. The load() method is used to retrieve the current value and the store() method is used to update the shared value in the global context.

As shown as well, the store() and load() methods require an Ordering enum argument to be specified. The Ordering enum refers to the way atomic operations synchronize memory which I chose Relaxed. For more detail, one can refer to the full list of options here. There is even a more detailed explanation of ordering here.

static SHARED: AtomicU32 = AtomicU32::new(0);

#[embassy_executor::task]
async fn async_task() {
    loop {
        // Load value from global context, modify and store
        let shared_var = SHARED.load(Ordering::Relaxed);
        SHARED.store(shared_var.wrapping_add(1), Ordering::Relaxed);
        Timer::after(Duration::from_millis(1000)).await;
    }
}

#[embassy_executor::main]
async fn main(spawner: Spawner) {
    // Initialize and create handle for devicer peripherals
    let p = embassy_stm32::init(Default::default());
    //Configure UART
    let mut usart = UartTx::new(p.USART2, p.PA2, NoDma, Config::default());
    // Create empty String for message
    let mut msg: String<8> = String::new();
    // Spawn async blinking task
    spawner.spawn(async_task()).unwrap();

    loop {
        // Load value from global context
        let shared = SHARED.load(Ordering::Relaxed);
        // Wait 1 second
        Timer::after(Duration::from_millis(1000)).await;
        // Format value for printing
        core::writeln!(&mut msg, "{:02}", shared).unwrap();
        // Transmit Message
        usart.blocking_write(msg.as_bytes()).unwrap();
        msg.clear();
    }
}

The Blocking Mutex Type

The use case of blocking Mutex is similar to AtomicU32. The difference is that the blocking Mutex supports values larger than u32. Keep in mind that the blocking Mutex will block threads waiting for the lock to become available. Additionally, the blocking Mutex does not hold the lock between await points.

The code below repeats the earlier example of AtomicU32 only using the blocking Mutex instead. In the below code note the following differences:

  • When instantiating the shared global variable, a RawMutex type needs to be defined. Here I chose ThreadModeMutex. What to chose depends on the context in which you’re using the mutex. In our context which we choose does not matter much. More detail can be found here.

  • In the global variable installation, the u32 type inside the Mutex is wrapped with a RefCell. This is to allow interior mutability of the value that is wrapped in the Mutex.

  • To access SHARED the lock method is used to obtain a lock to the global shared value. lock provides access to the locked variable in a closure where the value is mutated.

use core::cell::RefCell;
use embassy_sync::blocking_mutex::Mutex;

static SHARED: Mutex<ThreadModeRawMutex, RefCell<u32>> = Mutex::new(RefCell::new(0));

#[embassy_executor::task]
async fn async_task() {
    loop {
        // Load value from global context, modify and store
        SHARED.lock(|f| {
            let val = f.borrow_mut().wrapping_add(1);
            f.replace(val);
        });
        Timer::after(Duration::from_millis(1000)).await;
    }
}

#[embassy_executor::main]
async fn main(spawner: Spawner) {
    // Initialize and create handle for device peripherals
    let p = embassy_stm32::init(Default::default());
    //Configure UART
    let mut usart = UartTx::new(p.USART2, p.PA2, NoDma, Config::default());
    // Create empty String for message
    let mut msg: String<8> = String::new();
    // Spawn async blinking task
    spawner.spawn(async_task()).unwrap();

    loop {
        // Wait 1 second
        Timer::after(Duration::from_millis(1000)).await;
        // Obtain updated value from global context
        let shared = SHARED.lock(|f| f.clone().into_inner());
        core::writeln!(&mut msg, "{:02}", shared).unwrap();
        // Transmit Message
        usart.blocking_write(msg.as_bytes()).unwrap();
        msg.clear();
    }
}

Reading/Writing Across async Tasks

The async Mutex Type

So how does a blocking Mutex differ from an async Mutex? A blocking Mutex lock is not held across await points/async tasks. Alternatively, with an async Mutex, one can await while holding the lock and other tasks will wait accordingly.

The code below achieves the same as the earlier code with a slight modification. First, note that the RefCell is no longer required and that the shared variable is not accessed through a closure. Instead, a lock methods awaits until the lock is acquired and doesn't let it go until the scope ends. If you notice the scope of the lock in the async_task there is a Timer::after(Duration::from_millis(1000)).await; line that is inserted. It would be interesting to play around with the delay value in this line and see how the code behaves. The point is to demonstrate that as you increase the value, the lock will remain held by async_task preventing SHARED in the main task from being accessed.

use embassy_sync::blocking_mutex::raw::ThreadModeRawMutex;
use embassy_sync::mutex::Mutex;

static SHARED: Mutex<ThreadModeRawMutex, u32> = Mutex::new(0);

#[embassy_executor::task]
async fn async_task() {
    loop {
        {
            let mut shared = SHARED.lock().await;
            *shared = shared.wrapping_add(1);
            Timer::after(Duration::from_millis(1000)).await;
        }
        Timer::after(Duration::from_millis(1000)).await;
    }
}

#[embassy_executor::main]
async fn main(spawner: Spawner) {
    // Initialize and create handle for device peripherals
    let p = embassy_stm32::init(Default::default());
    //Configure UART
    let mut usart = UartTx::new(p.USART2, p.PA2, NoDma, Config::default());
    // Create empty String for message
    let mut msg: String<8> = String::new();
    // Spawn async blinking task
    spawner.spawn(async_task()).unwrap();

    loop {
        // Wait 1 second
        Timer::after(Duration::from_millis(1000)).await;
        // Obtain updated value from global context
        let shared = SHARED.lock().await;
        core::writeln!(&mut msg, "{:02}", *shared).unwrap();
        // Transmit Message
        usart.blocking_write(msg.as_bytes()).unwrap();
        msg.clear();
    }
}

Wait for Value Change

The Signal Type

Signal provides for a simple case where one value needs to be buffered/sent to another task. This is done by sending a "Signal" that a new value is available.

The code below shows a usage example where a value of 5 is sent every second from the async_task to the main task. The signal method updates the value and sets up a signal. The wait method in the main task awaits until a signal is received and updates val and prints it to the console. In this case, the new value signaled is always the same which is 5. Note here a difference with the AtomicU32 is that before I didn't check or wait for the value to change/update. I only grabbed whatever was stored in the shared global variable.

use embassy_sync::signal::Signal;
static SHARED: Signal<ThreadModeRawMutex, u32> = Signal::new();

#[embassy_executor::task]
async fn async_task() {
    loop {
        SHARED.signal(5);
        Timer::after(Duration::from_millis(1000)).await;
    }
}

#[embassy_executor::main]
async fn main(spawner: Spawner) {
    // Initialize and create handle for device peripherals
    let p = embassy_stm32::init(Default::default());
    //Configure UART
    let mut usart = UartTx::new(p.USART2, p.PA2, NoDma, Config::default());
    // Create empty String for message
    let mut msg: String<16> = String::new();
    // Spawn async blinking task
    spawner.spawn(async_task()).unwrap();

    loop {
        let val = SHARED.wait().await;
        core::writeln!(&mut msg, "Signal Detected {:02}", val).unwrap();
        //Transmit Message
        usart.blocking_write(msg.as_bytes()).unwrap();
        msg.clear();
    }
}

The Channel Type

Channel expands on Signal in which it allows for multiple values to be buffered in a queue. As such, when instantiating a Channelthe queue size needs to be defined. A Channel allows for multiple producers to write to a queue and multiple consumers to read. However, there can be only one reader for a value. This means that it's a first come first serve type of construct that once the first consumer reads a value, it's not available for other consumers anymore.

In the example below, a value of 2 is chosen for the Channel size. This is because I am going to allow two tasks to buffer/publish values to the Channel and one task to retrieve values. In the code example below note how the tasks async_task_one and async_task_two use the send method to send the values of 1 and 2 to the SHARED Channel. The main task, on the other, hand uses the recv method to retrieve the values buffered in SHARED and prints them to the serial monitor. If you run this code, you'll notice that the value of 1 will be printed twice to the serial monitor followed by a single instance of 2. This is because the timer delay in async_task_one is half a second and the delay in async_task_two is a second.

use embassy_sync::channel::Channel;

//Declare a channel of 2 u32s
static SHARED: Channel<ThreadModeRawMutex, u32, 2> = Channel::new();

#[embassy_executor::task]
async fn async_task_one() {
    loop {
        SHARED.send(1).await;
        Timer::after(Duration::from_millis(500)).await;
    }
}

#[embassy_executor::task]
async fn async_task_two() {
    loop {
        SHARED.send(2).await;
        Timer::after(Duration::from_millis(1000)).await;
    }
}

#[embassy_executor::main]
async fn main(spawner: Spawner) {
    // Initialize and create handle for device peripherals
    let p = embassy_stm32::init(Default::default());
    //Configure UART
    let mut usart = UartTx::new(p.USART2, p.PA2, NoDma, Config::default());
    // Create empty String for message
    let mut msg: String<16> = String::new();
    // Spawn async blinking task
    spawner.spawn(async_task_one()).unwrap();
    spawner.spawn(async_task_two()).unwrap();

    loop {
        let val = SHARED.recv().await;
        core::writeln!(&mut msg, "{:02}", val).unwrap();
        //Transmit Message
        usart.blocking_write(msg.as_bytes()).unwrap();
        msg.clear();
    }
}

The PubSubChannel Type

The PubSubChannel is an expansion of the Channel type in which multiple consumers can now access the value. In the Channel type, if one consumer read a value, it's no longer available for others. With the PubSubChannel this is no longer an issue. However, what actually could happen in a PubSubChannel is that a certain consumer/subscriber can miss out on a value in the queue because a new value was pushed in. If that happens, the PubSubChannel provides an error signaling that occurrence.

The example below repeats the Channel example above, though using a PubSubChannel instead. Note how pub1 and pub2 are declared as publishers to SHARED and use the publish_immediate method to send/publish values to the global queue. Correspondingly, sub is declared as a subscriber to SHARED and uses the next_message_pure method to access the values in the global queue.


use embassy_sync::pubsub::PubSubChannel;

//Declare a pubsub channel with a capcity of 2 and 1 subscriber and 2 publishers
static SHARED: PubSubChannel<ThreadModeRawMutex, u32, 2, 2, 2> = PubSubChannel::new();

#[embassy_executor::task]
async fn async_task_one() {
    let pub1 = SHARED.publisher().unwrap();
    loop {
        pub1.publish_immediate(1);
        Timer::after(Duration::from_millis(500)).await;
    }
}

#[embassy_executor::task]
async fn async_task_two() {
    let pub2 = SHARED.publisher().unwrap();
    loop {
        pub2.publish_immediate(2);
        Timer::after(Duration::from_millis(1000)).await;
    }
}

#[embassy_executor::main]
async fn main(spawner: Spawner) {
    // Initialize and create handle for device peripherals
    let p = embassy_stm32::init(Default::default());
    //Configure UART
    let mut usart = UartTx::new(p.USART2, p.PA2, NoDma, Config::default());
    // Create empty String for message
    let mut msg: String<16> = String::new();
    // Spawn async blinking task
    spawner.spawn(async_task_one()).unwrap();
    spawner.spawn(async_task_two()).unwrap();

    let mut sub = SHARED.subscriber().unwrap();

    loop {
        let val = sub.next_message_pure().await;
        core::writeln!(&mut msg, "{:02}", val).unwrap();
        //Transmit Message
        usart.blocking_write(msg.as_bytes()).unwrap();
        msg.clear();
    }
}

The Pipe Type

A Pipe is more or less just like a Channel though restricted to buffering of u8 types.

Conclusion

Dealing with global variables shared among threads in embedded Rust (or Rust in general) can be a hassle normally. In embassy, however different types are provided through the embassy-sync crate to facilitate value sharing among threads. This post furnishes the different types available through embassy-sync and how they can be used. Have any questions/comments? Share your thoughts in the comments below πŸ‘‡.

Did you find this article valuable?

Support Omar Hiari by becoming a sponsor. Any amount is appreciated!

Learn more about Hashnode Sponsors
Β 
Share this