Test Framework

The Zephyr Test Framework (Ztest) provides a simple testing framework intended to be used during development. It provides basic assertion macros and a generic test structure.

The framework can be used in two ways, either as a generic framework for integration testing, or for unit testing specific modules.

Creating a test suite

Using Ztest to create a test suite is as easy as calling the ZTEST_SUITE. The macro accepts the following arguments:

  • suite_name - The name of the suite. This name must be unique within a single binary.

  • ztest_suite_predicate_t - An optional predicate function to allow choosing when the test will run. The predicate will get a pointer to the global state passed in through ztest_run_all() and should return a boolean to decide if the suite should run.

  • ztest_suite_setup_t - An optional setup function which returns a test fixture. This will be called and run once per test suite run.

  • ztest_suite_before_t - An optional before function which will run before every single test in this suite.

  • ztest_suite_after_t - An optional after function which will run after every single test in this suite.

  • ztest_suite_teardown_t - An optional teardown function which will run at the end of all the tests in the suite.

Below is an example of a test suite using a predicate:

#include <zephyr/ztest.h>
#include "test_state.h"

static bool predicate(const void *global_state)
{
     return ((const struct test_state*)global_state)->x == 5;
}

ZTEST_SUITE(alternating_suite, predicate, NULL, NULL, NULL, NULL);

Adding tests to a suite

There are 5 macros used to add a test to a suite, they are:

  • ZTEST (suite_name, test_name) - Which can be used to add a test by test_name to a given suite by suite_name.

  • ZTEST_P (suite_name, test_name) - Add a parameterized test to a given suite by specifying the suite_name and test_name. You can then access the passed parameter within the body of the test using the data pointer.

  • ZTEST_USER (suite_name, test_name) - Which behaves the same as ZTEST, only that when CONFIG_USERSPACE is enabled, then the test will be run in a userspace thread.

  • ZTEST_F (suite_name, test_name) - Which behaves the same as ZTEST, only that the test function will already include a variable named fixture with the type <suite_name>_fixture.

  • ZTEST_USER_F (suite_name, test_name) - Which combines the fixture feature of ZTEST_F with the userspace threading for the test.

Test fixtures

Test fixtures can be used to help simplify repeated test setup operations. In many cases, tests in the same suite will require some initial setup followed by some form of reset between each test. This is achieved via fixtures in the following way:

#include <zephyr/ztest.h>

struct my_suite_fixture {
     size_t max_size;
     size_t size;
     uint8_t buff[1];
};

static void *my_suite_setup(void)
{
     /* Allocate the fixture with 256 byte buffer */
   struct my_suite_fixture *fixture = malloc(sizeof(struct my_suite_fixture) + 255);

     zassume_not_null(fixture, NULL);
     fixture->max_size = 256;

     return fixture;
}

static void my_suite_before(void *f)
{
     struct my_suite_fixture *fixture = (struct my_suite_fixture *)f;
     memset(fixture->buff, 0, fixture->max_size);
     fixture->size = 0;
}

static void my_suite_teardown(void *f)
{
   free(f);
}

ZTEST_SUITE(my_suite, NULL, my_suite_setup, my_suite_before, NULL, my_suite_teardown);

ZTEST_F(my_suite, test_feature_x)
{
     zassert_equal(0, fixture->size);
     zassert_equal(256, fixture->max_size);
}

Using memory allocated by a test fixture in a userspace thread, such as during execution of ZTEST_USER or ZTEST_USER_F, requires that memory to be declared userspace accessible. This is because the fixture memory is owned and initialized by kernel space. The Ztest framework provides the ZTEST_DMEM and ZTEST_BMEM macros for use of such user/kernel space shared memory.

Advanced features

Test result expectations

Some tests were made to be broken. In cases where the test is expected to fail or skip due to the nature of the code, it’s possible to annotate the test as such. For example:

#include <zephyr/ztest.h>

ZTEST_SUITE(my_suite, NULL, NULL, NULL, NULL, NULL);

ZTEST_EXPECT_FAIL(my_suite, test_fail);
ZTEST(my_suite, test_fail)
{
  /** This will fail the test */
  zassert_true(false, NULL);
}

ZTEST_EXPECT_SKIP(my_suite, test_skip);
ZTEST(my_suite, test_skip)
{
  /** This will skip the test */
  zassume_true(false, NULL);
}

In this example, the above tests should be marked as failed and skipped respectively. Instead, Ztest will mark both as passed due to the expectation.

Test rules

Test rules are a way to run the same logic for every test and every suite. There are a lot of cases where you might want to reset some state for every test in the binary (regardless of which suite is currently running). As an example, this could be to reset mocks, reset emulators, flush the UART, etc.:

#include <zephyr/fff.h>
#include <zephyr/ztest.h>

#include "test_mocks.h"

DEFINE_FFF_GLOBALS;

DEFINE_FAKE_VOID_FUN(my_weak_func);

static void fff_reset_rule_before(const struct ztest_unit_test *test, void *fixture)
{
     ARG_UNUSED(test);
     ARG_UNUSED(fixture);

     RESET_FAKE(my_weak_func);
}

ZTEST_RULE(fff_reset_rule, fff_reset_rule_before, NULL);

A custom test_main

While the Ztest framework provides a default test_main() function, it’s possible that some applications will want to provide custom behavior. This is particularly true if there’s some global state that the tests depend on and that state either cannot be replicated or is difficult to replicate without starting the process over. For example, one such state could be a power sequence. Assuming there’s a board with several steps in the power-on sequence a test suite can be written using the predicate to control when it would run. In that case, the test_main() function can be written as follows:

#include <zephyr/ztest.h>

#include "my_test.h"

void test_main(void)
{
     struct power_sequence_state state;

     /* Only suites that use a predicate checking for phase == PWR_PHASE_0 will run. */
     state.phase = PWR_PHASE_0;
     ztest_run_all(&state, false, 1, 1);

     /* Only suites that use a predicate checking for phase == PWR_PHASE_1 will run. */
     state.phase = PWR_PHASE_1;
     ztest_run_all(&state, false, 1, 1);

     /* Only suites that use a predicate checking for phase == PWR_PHASE_2 will run. */
     state.phase = PWR_PHASE_2;
     ztest_run_all(&state, false, 1, 1);

     /* Check that all the suites in this binary ran at least once. */
     ztest_verify_all_test_suites_ran();
}

Quick start - Integration testing

A simple working base is located at samples/subsys/testsuite/integration. To make a test application for the bar component of foo, you should copy the sample folder to tests/foo/bar and edit files there adjusting for your test application’s purposes.

To build and execute all applicable test scenarios defined in your test application use the Twister tool, for example:

./scripts/twister -T tests/foo/bar/

To select just one of the test scenarios, run Twister with --scenario command:

./scripts/twister --scenario tests/foo/bar/your.test.scenario.name

In the command line above tests/foo/bar is the path to your test application and your.test.scenario.name references a test scenario defined in testcase.yaml file, which is like sample.testing.ztest in the boilerplate test suite sample.

See Twister test project diagram for more details on how Twister deals with Ztest application.

The sample contains the following files:

CMakeLists.txt

1# SPDX-License-Identifier: Apache-2.0
2
3cmake_minimum_required(VERSION 3.20.0)
4find_package(Zephyr REQUIRED HINTS $ENV{ZEPHYR_BASE})
5project(integration)
6
7FILE(GLOB app_sources src/*.c)
8target_sources(app PRIVATE ${app_sources})

testcase.yaml

 1tests:
 2  # section.subsection
 3  sample.testing.ztest:
 4    build_only: true
 5    platform_allow:
 6      - native_posix
 7      - native_sim
 8    integration_platforms:
 9      - native_sim
10    tags: test_framework

prj.conf

1CONFIG_ZTEST=y

src/main.c (see best practices)

 1/*
 2 * Copyright (c) 2016 Intel Corporation
 3 *
 4 * SPDX-License-Identifier: Apache-2.0
 5 */
 6
 7#include <zephyr/ztest.h>
 8
 9
10ZTEST_SUITE(framework_tests, NULL, NULL, NULL, NULL, NULL);
11
12/**
13 * @brief Test Asserts
14 *
15 * This test verifies various assert macros provided by ztest.
16 *
17 */
18ZTEST(framework_tests, test_assert)
19{
20	zassert_true(1, "1 was false");
21	zassert_false(0, "0 was true");
22	zassert_is_null(NULL, "NULL was not NULL");
23	zassert_not_null("foo", "\"foo\" was NULL");
24	zassert_equal(1, 1, "1 was not equal to 1");
25	zassert_equal_ptr(NULL, NULL, "NULL was not equal to NULL");
26}

A test application may consist of multiple test suites that either can be testing functionality or APIs. Functions implementing a test case should follow the guidelines below:

  • Test cases function names should be prefixed with test_

  • Test cases should be documented using doxygen

  • Test case function names should be unique within the section or component being tested

For example:

/**
 * @brief Test Asserts
 *
 * This test case verifies the zassert_true macro.
 */
ZTEST(my_suite, test_assert)
{
        zassert_true(1, "1 was false");
}

Listing Tests

Tests (test applications) in the Zephyr tree consist of many test scenarios that run as part of a project and test similar functionality, for example an API or a feature. The twister script can parse the test scenarios, suites and cases in all test applications or a subset of them, and can generate reports on a granular level, i.e. if test cases have passed or failed or if they were blocked or skipped.

Twister parses the source files looking for test case names, so you can list all kernel test cases, for example, by running:

./scripts/twister --list-tests -T tests/kernel

Skipping Tests

Special- or architecture-specific tests cannot run on all platforms and architectures, however we still want to count those and report them as being skipped. Because the test inventory and the list of tests is extracted from the code, adding conditionals inside the test suite is sub-optimal. Tests that need to be skipped for a certain platform or feature need to explicitly report a skip using ztest_test_skip() or Z_TEST_SKIP_IFDEF. If the test runs, it needs to report either a pass or fail. For example:

#ifdef CONFIG_TEST1
ZTEST(common, test_test1)
{
     zassert_true(1, "true");
}
#else
ZTEST(common, test_test1)
{
     ztest_test_skip();
}
#endif

ZTEST(common, test_test2)
{
     Z_TEST_SKIP_IFDEF(CONFIG_BUGxxxxx);
     zassert_equal(1, 0, NULL);
}

ZTEST_SUITE(common, NULL, NULL, NULL, NULL, NULL);

Quick start - Unit testing

Ztest can be used for unit testing. This means that rather than including the entire Zephyr OS for testing a single function, you can focus the testing efforts into the specific module in question. This will speed up testing since only the module will have to be compiled in, and the tested functions will be called directly.

Examples of unit tests can be found in the tests/unit/ folder. In order to declare the unit tests present in a source folder, you need to add the relevant source files to the testbinary target from the CMake unittest component. See a minimal example below:

cmake_minimum_required(VERSION 3.20.0)

project(app)
find_package(Zephyr COMPONENTS unittest REQUIRED HINTS $ENV{ZEPHYR_BASE})
target_sources(testbinary PRIVATE main.c)

Since you won’t be including basic kernel data structures that most code depends on, you have to provide function stubs in the test. Ztest provides some helpers for mocking functions, as demonstrated below.

In a unit test, mock objects can simulate the behavior of complex real objects and are used to decide whether a test failed or passed by verifying whether an interaction with an object occurred, and if required, to assert the order of that interaction.

Best practices for declaring the test suite

twister and other validation tools need to obtain the list of test cases that a Zephyr ztest test image will expose.

Rationale

This all is for the purpose of traceability. It’s not enough to have only a semaphore test application. We also need to show that we have testpoints for all APIs and functionality, and we trace back to documentation of the API, and functional requirements.

The idea is that test reports show results for every test case as passed, failed, blocked, or skipped. Reporting on only the high-level test application, particularly when tests do too many things, is too vague.

Other questions:

  • Why not pre-scan with CPP and then parse? or post scan the ELF file?

    If C pre-processing or building fails because of any issue, then we won’t be able to tell the subcases.

  • Why not declare them in the YAML test configuration?

    A separate test case description file would be harder to maintain than just keeping the information in the test source files themselves – only one file to update when changes are made eliminates duplication.

Stress test framework

Zephyr stress test framework (Ztress) provides an environment for executing user functions in multiple priority contexts. It can be used to validate that code is resilient to preemptions. The framework tracks the number of executions and preemptions for each context. Execution can have various completion conditions like timeout, number of executions or number of preemptions.

The framework is setting up the environment by creating the requested number of threads (each on different priority), optionally starting a timer. For each context, a user function (different for each context) is called and then the context sleeps for a randomized amount of system ticks. The framework is tracking CPU load and adjusts sleeping periods to achieve higher CPU load. In order to increase the probability of preemptions, the system clock frequency should be relatively high. The default 100 Hz on QEMU x86 is much too low and it is recommended to increase it to 100 kHz.

The stress test environment is setup and executed using ZTRESS_EXECUTE which accepts a variable number of arguments. Each argument is a context that is specified by ZTRESS_TIMER or ZTRESS_THREAD macros. Contexts are specified in priority descending order. Each context specifies completion conditions by providing the minimum number of executions and preemptions. When all conditions are met and the execution has completed, an execution report is printed and the macro returns. Note that while the test is executing, a progress report is periodically printed.

Execution can be prematurely completed by specifying a test timeout (ztress_set_timeout()) or an explicit abort (ztress_abort()).

User function parameters contains an execution counter and a flag indicating if it is the last execution.

The example below presents how to setup and run 3 contexts (one of which is k_timer interrupt handler context). Completion criteria is set to at least 10000 executions of each context and 1000 preemptions of the lowest priority context. Additionally, the timeout is configured to complete after 10 seconds if those conditions are not met. The last argument of each context is the initial sleep time which will be adjusted throughout the test to achieve the highest CPU load.

ztress_set_timeout(K_MSEC(10000));
ZTRESS_EXECUTE(ZTRESS_TIMER(foo_0, user_data_0, 10000, Z_TIMEOUT_TICKS(20)),
               ZTRESS_THREAD(foo_1, user_data_1, 10000, 0, Z_TIMEOUT_TICKS(20)),
               ZTRESS_THREAD(foo_2, user_data_2, 10000, 1000, Z_TIMEOUT_TICKS(20)));

Configuration

Static configuration of Ztress contains:

API reference

Running tests

Ztest testing macros

Assertions

These macros will instantly fail the test if the related assertion fails. When an assertion fails, it will print the current file, line and function, alongside a reason for the failure and an optional message. If the config CONFIG_ZTEST_ASSERT_VERBOSE is 0, the assertions will only print the file and line numbers, reducing the binary size of the test.

Example output for a failed macro from zassert_equal(buf->ref, 2, "Invalid refcount"):

Assertion failed at main.c:62: test_get_single_buffer: Invalid refcount (buf->ref not equal to 2)
Aborted at unit test function
Ztest assertion macros

Expectations

These macros will continue test execution if the related expectation fails and subsequently fail the test at the end of its execution. When an expectation fails, it will print the current file, line, and function, alongside a reason for the failure and an optional message but continue executing the test. If the config CONFIG_ZTEST_ASSERT_VERBOSE is 0, the expectations will only print the file and line numbers, reducing the binary size of the test.

For example, if the following expectations fail:

zexpect_equal(buf->ref, 2, "Invalid refcount");
zexpect_equal(buf->ref, 1337, "Invalid refcount");

The output will look something like:

START - test_get_single_buffer
    Expectation failed at main.c:62: test_get_single_buffer: Invalid refcount (buf->ref not equal to 2)
    Expectation failed at main.c:63: test_get_single_buffer: Invalid refcount (buf->ref not equal to 1337)
 FAIL - test_get_single_buffer in 0.0 seconds
Ztest expectation macros

Assumptions

These macros will instantly skip the test or suite if the related assumption fails. When an assumption fails, it will print the current file, line, and function, alongside a reason for the failure and an optional message. If the config CONFIG_ZTEST_ASSERT_VERBOSE is 0, the assumptions will only print the file and line numbers, reducing the binary size of the test.

Example output for a failed macro from zassume_equal(buf->ref, 2, "Invalid refcount"):

Ztest assumption macros

Ztress

Ztest ztress macros

Mocking via FFF

Zephyr has integrated with FFF for mocking. See FFF for documentation. To use it, include the relevant header:

#include <zephyr/fff.h>

Zephyr provides several FFF-based fake drivers which can be used as either stubs or mocks. Fake driver instances are configured via Devicetree and Configuration System (Kconfig). See the following devicetree bindings for more information:

Zephyr also has defined extensions to FFF for simplified declarations of fake functions. See FFF Extensions.

Customizing Test Output

Customization is enabled by setting CONFIG_ZTEST_TC_UTIL_USER_OVERRIDE to “y” and adding a file tc_util_user_override.h with your overrides.

Add the line zephyr_include_directories(my_folder) to your project’s CMakeLists.txt to let Zephyr find your header file during builds.

See the file subsys/testsuite/include/zephyr/tc_util.h to see which macros and/or defines can be overridden. These will be surrounded by blocks such as:

#ifndef SOMETHING
#define SOMETHING <default implementation>
#endif /* SOMETHING */

Shuffling Test Sequence

By default the tests are sorted and ran in alphanumerical order. Test cases may be dependent on this sequence. Enable CONFIG_ZTEST_SHUFFLE to randomize the order. The output from the test will display the seed for failed tests. For native simulator builds you can provide the seed as an argument to twister with --seed.

Repeating Tests

By default the tests are executed once. The test cases and test suites may be executed multiple times. Enable CONFIG_ZTEST_REPEAT to execute the tests multiple times. By default the multiplication factors are 3, which means every test suite is executed 3 times and every test case is executed 3 times. This can be changed by the CONFIG_ZTEST_SUITE_REPEAT_COUNT and CONFIG_ZTEST_TEST_REPEAT_COUNT Kconfig options.

Test Selection

For tests built for native simulator, use command line arguments to list or select tests to run. The test argument expects a comma separated list of suite::test . You can substitute the test name with an * to run all tests within a suite.

For example

$ zephyr.exe -list
$ zephyr.exe -test="fixture_tests::test_fixture_pointer,framework_tests::test_assert_mem_equal"
$ zephyr.exe -test="framework_tests::*"

FFF Extensions

FFF extensions