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 throughztest_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 bytest_name
to a given suite bysuite_name
.ZTEST_P
(suite_name, test_name)
- Add a parameterized test to a given suite by specifying thesuite_name
andtest_name
. You can then access the passed parameter within the body of the test using thedata
pointer.ZTEST_USER
(suite_name, test_name)
- Which behaves the same asZTEST
, only that whenCONFIG_USERSPACE
is enabled, then the test will be run in a userspace thread.ZTEST_F
(suite_name, test_name)
- Which behaves the same asZTEST
, only that the test function will already include a variable namedfixture
with the type<suite_name>_fixture
.ZTEST_USER_F
(suite_name, test_name)
- Which combines the fixture feature ofZTEST_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:
CONFIG_ZTRESS_MAX_THREADS
- number of supported threads.
CONFIG_ZTRESS_STACK_SIZE
- Stack size of created threads.
CONFIG_ZTRESS_REPORT_PROGRESS_MS
- Test progress report interval.
API reference
Running tests
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
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
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")
:
Ztress
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::*"