μtest

Unit #[test]ing for microcontrollers and other no_std systems

WARNING This crate relies on #[test] / rustc implementation details and could break at any time.

Table of Contents

• Features

• Limitations

• Testing on an emulated Cortex-M processor

• Testing on a real Cortex-M microcontroller

  • Requirements
  • Steps

• Building a custom test runner

  • Hooks
    • __test_start
    • __test_ignored
    • __test_before_run
    • __test_failed
    • __test_success
    • __test_summary
    • __test_panic_fmt

• How does this work without unwinding?

• License

  • Contribution

Features

• Doesn't depend on std.

• Fully configurable, through hooks.

Limitations

• Tests are executed sequentially. This is required to support bare metal systems where threads may not be implemented.

• All tests will print to stdout / stderr as they progress.

• panic!s outside the crate under test will NOT mark the unit test as failed; those panics will likely abort the test runner but this is implementation defined. (more about this later)

• #[bench] is not supported.

• No colorized output

Testing on an emulated Cortex-M processor

Using the utest-cortex-m-qemu test runner.

This uses QEMU user emulation to emulate a Cortex-M processor that has access to the host Linux kernel thus you can do stuff like using the WRITE system call to print to the host console.

The downside of this approach is that the QEMU user emulation doesn't emulate the peripherals of a Cortex-M microcontroller so this is mainly useful to test pure functions / functions that don't do embedded I/O (by embedded I/O, I mean I2C, Serial, PWM, etc.).

0. Start with a no_std library crate.

$ cargo new --lib foo && cd $_

$ edit src/lib.rs && cat $_

#![no_std]

#[test]
fn assert() {
    assert!(true);
}

#[test]
fn assert_failed() {
    assert!(false, "oh noes");
}

#[test]
fn assert_eq() {
    assert_eq!(1 + 1, 2);
}

#[test]
fn assert_eq_failed() {
    let answer = 24;
    assert_eq!(answer, 42, "The answer was 42!");
}

#[test]
fn it_works() {}

#[test]
#[should_panic]
fn should_panic() {
    panic!("Let's panic!")
}

1. Append this to your crate's Cargo.toml

[target.thumbv7m-linux-eabi.dev-dependencies.utest-macros]
git = "https://github.com/japaric/utest"

[target.thumbv7m-linux-eabi.dev-dependencies.test]
git = "https://github.com/japaric/utest"

[target.thumbv7m-linux-eabi.dev-dependencies.utest-cortex-m-qemu]
git = "https://github.com/japaric/utest"

NOTE Change thumbv7m-linux-eabi as necessary. The other options are thumbv6m-linux-eabi, thumbv7em-linux-eabi and thumbv7em-linux-eabihf. (Yes, linux not none)

2. Add this to your src/lib.rs

// test runner
#[cfg(all(target_arch = "arm",
          not(any(target_env = "gnu", target_env = "musl")),
          target_os = "linux",
          test))]
extern crate utest_cortex_m_qemu;

// overrides `panic!`
#[cfg(all(target_arch = "arm",
          not(any(target_env = "gnu", target_env = "musl")),
          target_os = "linux",
          test))]
#[macro_use]
extern crate utest_macros;

#[cfg(all(target_arch = "arm",
          not(any(target_env = "gnu", target_env = "musl")),
          target_os = "linux",
          test))]
macro_rules! panic {
    ($($tt:tt)*) => {
        upanic!($($tt)*);
    };
}

3. Create the target specification file.

Start with this target specification

$ rustc \
  -Z unstable-options \
  --print target-spec-json \
  --target thumbv7m-none-eabi \
  | tee thumbv7m-linux-eabi.json

{
  "abi-blacklist": [
    "stdcall",
    "fastcall",
    "vectorcall",
    "win64",
    "sysv64"
  ],
  "arch": "arm",
  "data-layout": "e-m:e-p:32:32-i64:64-v128:64:128-a:0:32-n32-S64",
  "env": "",
  "executables": true,
  "is-builtin": true,
  "linker": "arm-none-eabi-gcc",
  "llvm-target": "thumbv7m-none-eabi",
  "max-atomic-width": 32,
  "os": "none",
  "panic-strategy": "abort",
  "relocation-model": "static",
  "target-endian": "little",
  "target-pointer-width": "32",
  "vendor": ""
}

And perform these modifications

     "data-layout": "e-m:e-p:32:32-i64:64-v128:64:128-a:0:32-n32-S64",
     "env": "",
     "executables": true,
-    "is-builtin": true,
     "linker": "arm-none-eabi-gcc",
     "llvm-target": "thumbv7m-none-eabi",
     "max-atomic-width": 32,
-    "os": "none",
+    "os": "linux",
     "panic-strategy": "abort",
+    "pre-link-args": ["-nostartfiles"],
     "relocation-model": "static",
     "target-endian": "little",
     "target-pointer-width": "32",

4. Built the test runner

$ export RUST_TARGET_PATH=$(pwd)

$ xargo test --target thumbv7m-linux-eabi --no-run

5. Execute the test runner using QEMU

$ qemu-arm target/thumbv7m-linux-eabi/debug/deps/foo-aacd724200d968b7
running 6 tests
test assert ... OK
test assert_failed ...
panicked at 'oh noes', src/lib.rs:23
FAILED
test assert_eq ... OK
test assert_eq_failed ...
panicked at 'assertion failed: `(left == right)` (left: `24`, right: `42`): The answer was 42!', src/lib.rs:34
FAILED
test it_works ... OK
test should_panic ...
panicked at 'Let's panic!', src/lib.rs:43
OK

Testing on a real Cortex-M microcontroller

Using the utest-cortex-m-semihosting test runner.

Requirements

• Your target crate must support vanilla fn main(). This means that the start lang item must be defined somewhere in your crate dependency graph.

• The panic_fmt lang item must be defined in your crate dependency graph. Hitting panic_fmt while running the test suite is considered a fatal error so it doesn't matter how you have implemented it.

These two requirements can be fulfilled if your crate is based on the cortex-m-template.

• You should be able to use GDB to run / debug a Rust program. GDB is required for semihosting.

Steps

0. Start with a crate that meets the requirements and some unit tests.

$ cargo new vl --template https://github.com/japaric/cortex-m-template

$ cd vl

$ edit memory.x && head $_
MEMORY
{
  FLASH : ORIGIN = 0x08000000, LENGTH = 128K
  RAM : ORIGIN = 0x20000000, LENGTH = 8K
}

$ edit tests/foo.rs && cat $_

#![no_std]

extern crate vl;

use core::ptr;

#[test]
fn assert() {
    assert!(true);
}

#[test]
fn assert_failed() {
    assert!(false, "oh noes");
}

#[test]
fn assert_eq() {
    assert_eq!(1 + 1, 2);
}

#[test]
fn assert_eq_failed() {
    let answer = 24;
    assert_eq!(answer, 42, "The answer was 42!");
}

// STM32F103xx = medium density device -> DEVICE_ID = 0x410
// See section 31.6.1 of the reference manual
// (http://www.st.com/resource/en/reference_manual/cd00171190.pdf)
#[test]
fn device_id() {
    assert_eq!(unsafe { ptr::read_volatile(0xe004_2000 as *const u32) } &
               ((1 << 12) - 1),
               0x410);
}

#[ignore]
#[test]
fn ignored() {}

#[test]
#[should_panic]
fn should_panic() {
    panic!("Let's panic!")
}

1. Append this to your Cargo.toml

[target.thumbv7m-none-eabi.dev-dependencies.test]
git = "https://github.com/japaric/utest"

[target.thumbv7m-none-eabi.dev-dependencies.utest-macros]
git = "https://github.com/japaric/utest"

[target.thumbv7m-none-eabi.dev-dependencies.utest-cortex-m-semihosting]
git = "https://github.com/japaric/utest"

2. Add this to you your integration test file (tests/foo.rs as per our example)

#[cfg(all(target_arch = "arm",
          not(any(target_env = "gnu", target_env = "musl"))))]
#[macro_use]
extern crate utest_macros;

#[cfg(all(target_arch = "arm",
          not(any(target_env = "gnu", target_env = "musl"))))]
extern crate utest_cortex_m_semihosting;

#[cfg(all(target_arch = "arm",
          not(any(target_env = "gnu", target_env = "musl")),
          target_os = "linux",
          test))]
macro_rules! panic {
    ($($tt:tt)*) => {
        upanic!($($tt)*);
    };
}

3. If required (this is required for cortex-m-template based crates), define how exceptions and interrupts are handled. In our example, add this to tests/foo.rs.

use vl::exceptions::{self, Exceptions};
use vl::interrupts::{self, Interrupts};

#[no_mangle]
pub static _EXCEPTIONS: Exceptions =
    Exceptions { ..exceptions::DEFAULT_HANDLERS };

#[no_mangle]
pub static _INTERRUPTS: Interrupts =
    Interrupts { ..interrupts::DEFAULT_HANDLERS };

4. Build the test runner

$ xargo test --target thumbv7m-none-eabi --test foo --no-run

5. Flash the test runner and execute the program using GDB.

NOTE These steps assume OpenOCD support.

If testing a crate based on the cortex-m-template, you'll only have to launch OpenOCD.

# Terminal 1
$ openocd -f interface/stlink-v1.cfg -f target/stm32f1x.cfg

and then launch GDB.

# Terminal 2
$ arm-none-eabi-gdb ./target/thumbv7m-none-eabi/debug/foo-87b629153685d76f

(gdb) continue

You should see this in the OpenOCD output

# Terminal 1
running 7 tests
test assert ... OK
test assert_failed ...
panicked at 'oh noes', tests/foo.rs:26
FAILED
test assert_eq ... OK
test assert_eq_failed ...
panicked at 'assertion failed: `(left == right)` (left: `24`, right: `42`): The answer was 42!', tests/foo.rs:37
FAILED
test device_id ... OK
test ignored ... ignored
test should_panic ...
panicked at 'Let's panic!', tests/foo.rs:57
OK

test result: FAILED. 4 passed; 2 failed; 1 ignored

If you are not using a cortex-m-template based crate, then make sure you enable semihosting from the GDB command line.

(gdb) monitor arm semihosting enable

Building a custom test runner

You can create a custom test runner that, for example, doesn't require executing the test runner under GDB and that instead reports the tests results via Serial port or ITM.

The best way to implement a custom test runner is to base your implementation on the implementation of the two tests runners shown above.

• utest-cortex-m-qemu

• utest-cortex-m-semihosting

But in a nutshell you'll have to define all these "hook" functions:

Hooks

Hooks are just vanilla functions with predefined symbol names that configure the behavior of the test runner.

__test_start

Runs before the unit tests are executed.

Signature:

/// `ntests`, number of unit tests (functions marked with `#[test]`)
#[no_mangle]
pub fn __test_start(ntests: usize) {
    ..
}

__test_ignored

Runs when a test if marked as #[ignore]d.

Signature:

/// `name`, name of the ignored test
#[no_mangle]
pub fn __test_ignored(name: &'static str) {
    ..
}

__test_before_run

Runs right before an unit test gets executed.

Signature:

/// `name`, name of the test that's about to be executed
#[no_mangle]
pub fn __test_before_run(name: &'static str) {
    ..
}

__test_failed

Runs if the unit test failed

Signature:

/// `name`, name of the test that failed
#[no_mangle]
pub fn __test_failed(name: &'static str) {
    ..
}

__test_success

Runs if the unit test succeeded

Signature:

/// `name`, name of the test that "passed"
#[no_mangle]
pub fn __test_success(name: &'static str) {
    ..
}

__test_summary

Runs after all the unit tests have been executed.

Signature:

/// `passed`, number of unit tests that passed
/// `failed`, number of unit tests that failed
/// `ignored`, number of unit tests that were ignored
#[no_mangle]
pub fn __test_summary(passed: usize, failed: usize, ignored: usize) {
    ..
}

__test_panic_fmt

Runs when utest-macros's panic! macro is called.

Signature:

/// Signature matches the signature of the `panic_fmt` lang item
#[no_mangle]
pub fn __test_panic_fmt(args: ::core::fmt::Arguments,
                        file: &'static str,
                        line: u32) {
    ..
}

How does this work without unwinding?

std unit tests rely heavily on unwinding. Each unit test is run inside a catch_unwind block and if the unit test panics then the panic is caught and the test is marked as failed (or as passed if the unit test was marked with #[should_panic]).

We attempt to emulate this behavior by overriding the panic! macro to mark the test failed and then early return instead of unwind. Of course, this emulation breaks down if the panic! originates from outside the crate under test, because panic is not overridden in that scope. So this setup is certainly not perfect.

License

Licensed under either of

• Apache License, Version 2.0 (LICENSE-APACHE or http://www.apache.org/licenses/LICENSE-2.0)

• MIT license (LICENSE-MIT or http://opensource.org/licenses/MIT)

at your option.

Contribution

Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.