Crate svd2rust

source ·
Expand description

Peripheral API generator from CMSIS-SVD files

A SVD file is an XML file that describes the hardware features of a microcontroller. In particular, it list all the peripherals available to the device, where the registers associated to each device are located in memory and what’s the function of each register.

svd2rust is a command line tool that transforms SVD files into crates that expose a type safe API to access the peripherals of the device.

Installation

$ cargo install svd2rust

Usage

svd2rust supports Cortex-M, MSP430 and RISCV microcontrollers. The generated crate can be tailored for either architecture using the --target flag. The flag accepts “cortex-m”, “msp430”, “riscv” and “none” as values. “none” can be used to generate a crate that’s architecture agnostic and that should work for architectures that svd2rust doesn’t currently know about like the Cortex-A architecture.

If the --target flag is omitted svd2rust assumes the target is the Cortex-M architecture.

target = cortex-m

When targeting the Cortex-M architecture svd2rust will generate three files in the current directory:

  • build.rs, build script that places device.x somewhere the linker can find.
  • device.x, linker script that weakly aliases all the interrupt handlers to the default exception handler (DefaultHandler).
  • lib.rs, the generated code.

All these files must be included in the same device crate. The lib.rs file contains several inlined modules and its not formatted. It’s recommend to split it out using the form tool and then format the output using rustfmt / cargo fmt:

$ svd2rust -i STM32F30x.svd

$ rm -rf src

$ form -i lib.rs -o src/ && rm lib.rs

$ cargo fmt

The resulting crate must provide an opt-in “rt” feature and depend on these crates: bare-metal v0.2.x, cortex-m v0.5.x, cortex-m-rt >=v0.6.5 and vcell v0.1.x. Furthermore the “device” feature of cortex-m-rt must be enabled when the “rt” feature is enabled. The Cargo.toml of the device crate will look like this:

[dependencies]
bare-metal = "0.2.0"
cortex-m = "0.5.8"
vcell = "0.1.0"

[dependencies.cortex-m-rt]
optional = true
version = "0.6.5"

[features]
rt = ["cortex-m-rt/device"]

target != cortex-m

When the target is msp430, riscv or none svd2rust will emit only the lib.rs file. Like in the cortex-m case we recommend you use form and rustfmt on the output.

The resulting crate must provide an opt-in “rt” feature and depend on these crates:

The *-rt dependencies must be optional only enabled when the “rt” feature is enabled. The Cargo.toml of the device crate will look like this for an msp430 target:

[dependencies]
bare-metal = "0.1.0"
msp430 = "0.1.0"
vcell = "0.1.0"

[dependencies.msp430-rt]
optional = true
version = "0.1.0"

[features]
rt = ["msp430-rt"]

Peripheral API

To use a peripheral first you must get an instance of the peripheral. All the device peripherals are modeled as singletons (there can only ever be, at most, one instance of any one of them) and the only way to get an instance of them is through the Peripherals::take method.

fn main() {
    let mut peripherals = stm32f30x::Peripherals::take().unwrap();
    peripherals.GPIOA.odr.write(|w| w.bits(1));
}

This method can only be successfully called once – that’s why the method returns an Option. Subsequent calls to the method will result in a None value being returned.

fn main() {
    let ok = stm32f30x::Peripherals::take().unwrap();
    let panics = stm32f30x::Peripherals::take().unwrap();
}

The singleton property can be unsafely bypassed using the ptr static method which is available on all the peripheral types. This method is a useful for implementing safe higher level abstractions.

struct PA0 { _0: () }
impl PA0 {
    fn is_high(&self) -> bool {
        // NOTE(unsafe) actually safe because this is an atomic read with no side effects
        unsafe { (*GPIOA::ptr()).idr.read().bits() & 1 != 0 }
    }

    fn is_low(&self) -> bool {
        !self.is_high()
    }
}
struct PA1 { _0: () }
// ..

fn configure(gpioa: GPIOA) -> (PA0, PA1, ..) {
    // configure all the PAx pins as inputs
    gpioa.moder.reset();
    // the GPIOA proxy is destroyed here now the GPIOA register block can't be modified
    // thus the configuration of the PAx pins is now frozen
    drop(gpioa);
    (PA0 { _0: () }, PA1 { _0: () }, ..)
}

Each peripheral proxy derefs to a RegisterBlock struct that represents a piece of device memory. Each field in this struct represents one register in the register block associated to the peripheral.

/// Inter-integrated circuit
pub mod i2c1 {
    /// Register block
    pub struct RegisterBlock {
        /// 0x00 - Control register 1
        pub cr1: CR1,
        /// 0x04 - Control register 2
        pub cr2: CR2,
        /// 0x08 - Own address register 1
        pub oar1: OAR1,
        /// 0x0c - Own address register 2
        pub oar2: OAR2,
        /// 0x10 - Timing register
        pub timingr: TIMINGR,
        /// Status register 1
        pub timeoutr: TIMEOUTR,
        /// Interrupt and Status register
        pub isr: ISR,
        /// 0x1c - Interrupt clear register
        pub icr: ICR,
        /// 0x20 - PEC register
        pub pecr: PECR,
        /// 0x24 - Receive data register
        pub rxdr: RXDR,
        /// 0x28 - Transmit data register
        pub txdr: TXDR,
    }
}

read / modify / write API

Each register in the register block, e.g. the cr1 field in the I2C struct, exposes a combination of the read, modify and write methods. Which methods exposes each register depends on whether the register is read-only, read-write or write-only:

  • read-only registers only expose the read method.
  • write-only registers only expose the write method.
  • read-write registers expose all the methods: read, modify and write.

This is signature of each of these methods:

(using I2C’s CR2 register as an example)

impl CR2 {
    /// Modifies the contents of the register
    pub fn modify<F>(&self, f: F)
    where
        for<'w> F: FnOnce(&R, &'w mut W) -> &'w mut W
    {
        ..
    }

    /// Reads the contents of the register
    pub fn read(&self) -> R { .. }

    /// Writes to the register
    pub fn write<F>(&mut self, f: F)
    where
        F: FnOnce(&mut W) -> &mut W,
    {
        ..
    }
}

The read method “reads” the register using a single, volatile LDR instruction and returns a proxy R struct that allows access to only the readable bits (i.e. not to the reserved or write-only bits) of the CR2 register:

/// Value read from the register
impl R {
    /// Bit 0 - Slave address bit 0 (master mode)
    pub fn sadd0(&self) -> SADD0R { .. }

    /// Bits 1:7 - Slave address bit 7:1 (master mode)
    pub fn sadd1(&self) -> SADD1R { .. }

    (..)
}

Usage looks like this:

// is the SADD0 bit of the CR2 register set?
if i2c1.c2r.read().sadd0().bit() {
    // yes
} else {
    // no
}

On the other hand, the write method writes some value to the register using a single, volatile STR instruction. This method involves a W struct that only allows constructing valid states of the CR2 register.

The only constructor that W provides is reset_value which returns the value of the CR2 register after a reset. The rest of W methods are “builder-like” and can be used to modify the writable bitfields of the CR2 register.

impl CR2W {
    /// Reset value
    pub fn reset_value() -> Self {
        CR2W { bits: 0 }
    }

    /// Bits 1:7 - Slave address bit 7:1 (master mode)
    pub fn sadd1(&mut self) -> _SADD1W { .. }

    /// Bit 0 - Slave address bit 0 (master mode)
    pub fn sadd0(&mut self) -> SADD0R { .. }
}

The write method takes a closure with signature (&mut W) -> &mut W. If the “identity closure”, |w| w, is passed then the write method will set the CR2 register to its reset value. Otherwise, the closure specifies how the reset value will be modified before it’s written to CR2.

Usage looks like this:

// Starting from the reset value, `0x0000_0000`, change the bitfields SADD0
// and SADD1 to `1` and `0b0011110` respectively and write that to the
// register CR2.
i2c1.cr2.write(|w| unsafe { w.sadd0().bit(true).sadd1().bits(0b0011110) });
// NOTE ^ unsafe because you could be writing a reserved bit pattern into
// the register. In this case, the SVD doesn't provide enough information to
// check whether that's the case.

// NOTE The argument to `bits` will be *masked* before writing it to the
// bitfield. This makes it impossible to write, for example, `6` to a 2-bit
// field; instead, `6 & 3` (i.e. `2`) will be written to the bitfield.

Finally, the modify method performs a single read-modify-write operation that involves one read (LDR) to the register, modifying the value and then a single write (STR) of the modified value to the register. This method accepts a closure that specifies how the CR2 register will be modified (the w argument) and also provides access to the state of the register before it’s modified (the r argument).

Usage looks like this:

// Set the START bit to 1 while KEEPING the state of the other bits intact
i2c1.cr2.modify(|_, w| unsafe { w.start().bit(true) });

// TOGGLE the STOP bit, all the other bits will remain untouched
i2c1.cr2.modify(|r, w| w.stop().bit(!r.stop().bit()));

enumeratedValues

If your SVD uses the <enumeratedValues> feature, then the API will be extended to provide even more type safety. This extension is backward compatible with the original version so you could “upgrade” your SVD by adding, yourself, <enumeratedValues> to it and then use svd2rust to re-generate a better API that doesn’t break the existing code that uses that API.

The new read API returns an enum that you can match:

match gpioa.dir.read().pin0() {
    gpioa::dir::PIN0R::Input => { .. },
    gpioa::dir::PIN0R::Output => { .. },
}

or test for equality

if gpioa.dir.read().pin0() == gpio::dir::PIN0R::Input {
    ..
}

It also provides convenience methods to check for a specific variant without having to import the enum:

if gpioa.dir.read().pin0().is_input() {
    ..
}

if gpioa.dir.read().pin0().is_output() {
    ..
}

The original bits method is available as well:

if gpioa.dir.read().pin0().bits() == 0 {
    ..
}

And the new write API provides similar additions as well: variant lets you pick the value to write from an enumeration of the possible ones:

// enum DIRW { Input, Output }
gpioa.dir.write(|w| w.pin0().variant(gpio::dir::PIN0W::Output));

There are convenience methods to pick one of the variants without having to import the enum:

gpioa.dir.write(|w| w.pin0().output());

The bits (or bit) method is still available but will become safe if it’s impossible to write a reserved bit pattern into the register:

// safe because there are only two options: `0` or `1`
gpioa.dir.write(|w| w.pin0().bit(true));

Interrupt API

SVD files also describe the device interrupts. svd2rust generated crates expose an enumeration of the device interrupts as an Interrupt enum in the root of the crate. This enum can be used with the cortex-m crate NVIC API.

extern crate cortex_m;
extern crate stm32f30x;

use cortex_m::interrupt;
use cortex_m::peripheral::Peripherals;
use stm32f30x::Interrupt;

let p = Peripherals::take().unwrap();
let mut nvic = p.NVIC;

nvic.enable(Interrupt::TIM2);
nvic.enable(Interrupt::TIM3);

the “rt” feature

If the “rt” Cargo feature of the svd2rust generated crate is enabled the crate will populate the part of the vector table that contains the interrupt vectors and provide an interrupt! macro (non Cortex-M targets) or interrupt attribute (Cortex-M) that can be used to register interrupt handlers.

the --nightly flag

The --nightly flag can be passed to svd2rust to enable features in the generated api that are only available to a nightly compiler. These features are

  • #[feature(untagged_unions)] for overlapping/overloaded registers

Macros

Assigns a handler to an interrupt