Skip to content

Unit Test Framework Package

About

This package provides unit test frameworks capable of building tests for multiple contexts including the UEFI shell environment and host-based environments. It allows for unit test development to focus on the tests and leave error logging, result formatting, context persistence, and test running to the framework. The unit test framework works well for low level unit tests as well as system level tests and fits easily in automation frameworks.

Framework

The first unit test framework is called Framework and is implemented as a set of EDK II libraries. The Framework supports both host-based unit tests and target-based unit tests that share the same source style, macros, and APIs. In some scenarios, the same unit test case sources can be built for both host-based unit test execution and target-based unit test execution. Host-based unit tests that require mocked interfaces can use the mocking infrastructure provided by cmocka that is included in the UnitTestFrameworkPkg as a submodule.

GoogleTest

The second unit test framework supported by the UnitTestFrameworkPkg is GoogleTest and can be used to implement host-based unit tests. GoogleTest on GitHub is included in the UnitTestFrameworkPkg as a submodule. Use of GoogleTest for target-based unit tests of EDK II components is not supported. Host-based unit tests that require mocked interfaces can use the mocking infrastructure included with GoogleTest called gMock. Note that the gMock framework does not directly support mocking of free (C style) functions, so the FunctionMockLib (containing a set of macros that wrap gMock's MOCK_METHOD macro) was created within the UnitTestFrameworkPkg to enable this support. The details and usage of these macros in the FunctionMockLib are described later.

GoogleTest requires less overhead to register test suites and test cases compared to the Framework. There are also a number of tools that layer on top of GoogleTest that improve developer productivity. One example is the VS Code extension C++ TestMate that may be used to implement, run, and debug unit tests implemented using GoogleTest.

If a component can be tested with host-based unit tests, then GoogleTest is recommended. The MdePkg contains a port of the BaseSafeIntLib unit tests in the GoogleTest style so the differences between GoogleTest and Framework unit tests can be reviewed. The paths to the BaseSafeIntLib unit tests are:

  • MdePkg/Test/UnitTest/Library/BaseSafeIntLib
  • MdePkg/Test/GoogleTest/Library/BaseSafeIntLib

Furthermore, the SecurityPkg contains unit tests for the SecureBootVariableLib using mocks in both the Framework/cmocka and GoogleTest/gMock style so the differences between cmocka and gMock can be reviewed. The paths to the SecureBootVariableLib unit tests are:

  • SecurityPkg/Library/SecureBootVariableLib/UnitTest
  • SecurityPkg/Library/SecureBootVariableLib/GoogleTest

Framework and GoogleTest Feature Comparison

Feature Framework GoogleTest
Host Based Unit Tests YES YES
Target Based Unit Tests YES NO
Unit Test Source Language C C++
Register Test Suite YES Auto
Register Test Case YES Auto
Death/Expected Assert Tests YES YES
Setup/Teardown Hooks YES YES
Value-Parameterized Tests NO YES
Typed Tests NO YES
Type-Parameterized Tests NO YES
Timeout Support NO YES
Mocking Support Cmocka gMock
JUNIT XML Reports YES YES
Execute subset of tests NO YES
VS Code Extensions NO YES

Framework Libraries

UnitTestLib

The main "framework" library. The core of the framework is the Framework object, which can have any number of test cases and test suites registered with it. The Framework object is also what drives test execution.

The Framework also provides helper macros and functions for checking test conditions and reporting errors. Status and error info will be logged into the test context. There are a number of Assert macros that make the unit test code friendly to view and easy to understand.

Finally, the Framework also supports logging strings during the test execution. This data is logged to the test context and will be available in the test reporting phase. This should be used for logging test details and helpful messages to resolve test failures.

UnitTestPersistenceLib

Persistence lib has the main job of saving and restoring test context to a storage medium so that for tests that require exiting the active process and then resuming state can be maintained. This is critical in supporting a system reboot in the middle of a test run.

UnitTestResultReportLib

Library provides function to run at the end of a framework test run and handles formatting the report. This is a common customization point and allows the unit test framework to fit its output reports into other test infrastructure. In this package simple library instances have been supplied to output test results to the console as plain text.

Framework Samples

There is a sample unit test provided as both an example of how to write a unit test and leverage many of the features of the framework. This sample can be found in the Test/UnitTest/Sample/SampleUnitTest directory.

The sample is provided in PEI, SMM, DXE, and UEFI App flavors. It also has a flavor for the HOST_APPLICATION build type, which can be run on a host system without needing a target.

Framework Usage

This section is built a lot like a "Getting Started". We'll go through some of the components that are needed when constructing a unit test and some of the decisions that are made by the test writer. We'll also describe how to check for expected conditions in test cases and a bit of the logging characteristics.

Most of these examples will refer to the SampleUnitTestUefiShell app found in this package.

Framework Requirements - INF

In our INF file, we'll need to bring in the UnitTestLib library. Conveniently, the interface header for the UnitTestLib is located in MdePkg, so you shouldn't need to depend on any other packages. As long as your DSC file knows where to find the lib implementation that you want to use, you should be good to go.

See this example in SampleUnitTestUefiShell.inf...

[Packages]
  MdePkg/MdePkg.dec

[LibraryClasses]
  UefiApplicationEntryPoint
  BaseLib
  DebugLib
  UnitTestLib
  PrintLib

Also, if you want your test to automatically be picked up by the Test Runner plugin, you will need to make sure that the module BASE_NAME contains the word Test...

[Defines]
  BASE_NAME      = SampleUnitTestUefiShell

Framework Requirements - DSC

In our DSC file, we'll need to bring in the INF file that was just created into the [Components] section so that the unit tests will be built.

See this example in UnitTestFrameworkPkg.dsc...

[Components]
  UnitTestFrameworkPkg/Test/UnitTest/Sample/SampleUnitTest/SampleUnitTestUefiShell.inf

Also, based on the type of tests that are being created, the associated DSC include file from the UnitTestFrameworkPkg for Host or Target based tests should also be included at the top of the DSC file.

!include UnitTestFrameworkPkg/UnitTestFrameworkPkgTarget.dsc.inc

Lastly, in the case that the test build has specific dependent libraries associated with it, they should be added in the \<LibraryClasses> sub-section for the INF file in the [Components] section of the DSC file.

See this example in SecurityPkgHostTest.dsc...

[Components]
  SecurityPkg/Library/SecureBootVariableLib/UnitTest/SecureBootVariableLibUnitTest.inf {
    <LibraryClasses>
      SecureBootVariableLib|SecurityPkg/Library/SecureBootVariableLib/SecureBootVariableLib.inf
      UefiRuntimeServicesTableLib|SecurityPkg/Library/SecureBootVariableLib/UnitTest/MockUefiRuntimeServicesTableLib.inf
      PlatformPKProtectionLib|SecurityPkg/Library/SecureBootVariableLib/UnitTest/MockPlatformPKProtectionLib.inf
      UefiLib|SecurityPkg/Library/SecureBootVariableLib/UnitTest/MockUefiLib.inf
  }

Framework Requirements - Code

Not to state the obvious, but let's make sure we have the following include before getting too far along...

#include <Library/UnitTestLib.h>

Now that we've got that squared away, let's look at our 'Main()' routine (or DriverEntryPoint() or whatever).

Framework Configuration

Everything in the UnitTestFrameworkPkg framework is built around an object called -- conveniently -- the Framework. This Framework object will contain all the information about our test, the test suites and test cases associated with it, the current location within the test pass, and any results that have been recorded so far.

To get started with a test, we must first create a Framework instance. The function for this is InitUnitTestFramework. It takes in CHAR8 strings for the long name, short name, and test version. The long name and version strings are just for user presentation and relatively flexible. The short name will be used to name any cache files and/or test results, so should be a name that makes sense in that context. These strings are copied internally to the Framework, so using stack-allocated or literal strings is fine.

In the SampleUnitTestUefiShell app, the module name is used as the short name, so the initialization looks like this.

DEBUG(( DEBUG_INFO, "%a v%a\n", UNIT_TEST_APP_NAME, UNIT_TEST_APP_VERSION ));

//
// Start setting up the test framework for running the tests.
//
Status = InitUnitTestFramework( &Framework, UNIT_TEST_APP_NAME, gEfiCallerBaseName, UNIT_TEST_APP_VERSION );

The &Framework returned here is the handle to the Framework. If it's successfully returned, we can start adding test suites and test cases.

Test suites exist purely to help organize test cases and to differentiate the results in reports. If you're writing a small unit test, you can conceivably put all test cases into a single suite. However, if you end up with 20+ test cases, it may be beneficial to organize them according to purpose. You must have at least one test suite, even if it's just a catch-all. The function to create a test suite is CreateUnitTestSuite. It takes in a handle to the Framework object, a CHAR8 string for the suite title and package name, and optional function pointers for a setup function and a teardown function.

The suite title is for user presentation. The package name is for xUnit type reporting and uses a '.'-separated hierarchical format (see 'SampleUnitTestApp' for example). If provided, the setup and teardown functions will be called once at the start of the suite (before any tests have run) and once at the end of the suite (after all tests have run), respectively. If either or both of these are unneeded, pass NULL. The function prototypes are UNIT_TEST_SUITE_SETUP and UNIT_TEST_SUITE_TEARDOWN.

Looking at SampleUnitTestUefiShell app, you can see that the first test suite is created as below...

//
// Populate the SimpleMathTests Unit Test Suite.
//
Status = CreateUnitTestSuite( &SimpleMathTests, Fw, "Simple Math Tests", "Sample.Math", NULL, NULL );

This test suite has no setup or teardown functions. The &SimpleMathTests returned here is a handle to the suite and will be used when adding test cases.

Great! Now we've finished some of the cruft, red tape, and busy work. We're ready to add some tests. Adding a test to a test suite is accomplished with the -- you guessed it -- AddTestCase function. It takes in the suite handle; a CHAR8 string for the description and class name; a function pointer for the test case itself; additional, optional function pointers for prerequisite check and cleanup routines; and an optional pointer to a context structure.

Okay, that's a lot. Let's take it one piece at a time. The description and class name strings are very similar in usage to the suite title and package name strings in the test suites. The former is for user presentation and the latter is for xUnit parsing. The test case function pointer is what is executed as the "test" and the prototype should be UNIT_TEST_FUNCTION. The last three parameters require a little bit more explaining.

The prerequisite check function has a prototype of UNIT_TEST_PREREQUISITE and -- if provided -- will be called immediately before the test case. If this function returns any error, the test case will not be run and will be recorded as UNIT_TEST_ERROR_PREREQUISITE_NOT_MET. The cleanup function (prototype UNIT_TEST_CLEANUP) will be called immediately after the test case to provide an opportunity to reset any global state that may have been changed in the test case. In the event of a prerequisite failure, the cleanup function will also be skipped. If either of these functions is not needed, pass NULL.

The context pointer is entirely case-specific. It will be passed to the test case upon execution. One of the purposes of the context pointer is to allow test case reuse with different input data. (Another use is for testing that wraps around a system reboot, but that's beyond the scope of this guide.) The test case must know how to interpret the context pointer, so it could be a simple value, or it could be a complex structure. If unneeded, pass NULL.

In SampleUnitTestUefiShell app, the first test case is added using the code below...

AddTestCase( SimpleMathTests, "Adding 1 to 1 should produce 2", "Addition", OnePlusOneShouldEqualTwo, NULL, NULL, NULL );

This test case calls the function OnePlusOneShouldEqualTwo and has no prerequisite, cleanup, or context.

Once all the suites and cases are added, it's time to run the Framework.

//
// Execute the tests.
//
Status = RunAllTestSuites( Framework );

Framework - A Simple Test Case

We'll take a look at the below test case from 'SampleUnitTestApp'...

UNIT_TEST_STATUS
EFIAPI
OnePlusOneShouldEqualTwo (
  IN UNIT_TEST_FRAMEWORK_HANDLE  Framework,
  IN UNIT_TEST_CONTEXT           Context
  )
{
  UINTN     A, B, C;

  A = 1;
  B = 1;
  C = A + B;

  UT_ASSERT_EQUAL(C, 2);
  return UNIT_TEST_PASSED;
} // OnePlusOneShouldEqualTwo()

The prototype for this function matches the UNIT_TEST_FUNCTION prototype. It takes in a handle to the Framework itself and the context pointer. The context pointer could be cast and interpreted as anything within this test case, which is why it's important to configure contexts carefully. The test case returns a value of UNIT_TEST_STATUS, which will be recorded in the Framework and reported at the end of all suites.

In this test case, the UT_ASSERT_EQUAL assertion is being used to establish that the business logic has functioned correctly. There are several assertion macros, and you are encouraged to use one that matches as closely to your intended test criterium as possible, because the logging is specific to the macro and more specific macros have more detailed logs. When in doubt, there are always UT_ASSERT_TRUE and UT_ASSERT_FALSE. Assertion macros that fail their test criterium will immediately return from the test case with UNIT_TEST_ERROR_TEST_FAILED and log an error string. Note that this early return can have implications for memory leakage.

At the end, if all test criteria pass, you should return UNIT_TEST_PASSED.

Framework - More Complex Cases

To write more advanced tests, first look at all the Assertion and Logging macros provided in the framework.

Beyond that, if you're writing host-based tests and want to take a dependency on the UnitTestFrameworkPkg, you can leverage the cmocka.h interface and write tests with all the features of the Cmocka framework.

Documentation for Cmocka can be found here.

GoogleTest Samples

There is a sample unit test provided as both an example of how to write a unit test and leverage many of the GoogleTest features. This sample can be found in the Test/GoogleTest/Sample/SampleGoogleTest directory.

The sample is provided for the HOST_APPLICATION build type, which can be run on a host system without needing a target.

There is also a sample unit test provided as both an example of how to write a unit test with mock functions and leverage some of the gMock features. This sample can be found in the SecurityPkg/Library/SecureBootVariableLib/GoogleTest directory.

It too is provided for the HOST_APPLICATION build type, which can be run on a host system without needing a target.

GoogleTest Usage

This section is built a lot like a "Getting Started". We'll go through some of the components that are needed when constructing a unit test and some of the decisions that are made by the test writer. We'll also describe how to check for expected conditions in test cases and a bit of the logging characteristics.

Most of these examples will refer to the SampleGoogleTestHost app found in this package, but the examples related to mock functions will refer to the SecureBootVariableLibGoogleTest app found in the SecurityPkg.

GoogleTest Requirements - INF

In our INF file, we'll need to bring in the GoogleTestLib library. Conveniently, the interface header for the GoogleTestLib is in UnitTestFrameworkPkg, so you shouldn't need to depend on any other packages. As long as your DSC file knows where to find the lib implementation that you want to use, you should be good to go.

See this example in SampleGoogleTestHost.inf...

[Packages]
  MdePkg/MdePkg.dec
  UnitTestFrameworkPkg/UnitTestFrameworkPkg.dec

[LibraryClasses]
  GoogleTestLib
  BaseLib
  DebugLib

Also, if you want your test to automatically be picked up by the Test Runner plugin, you will need to make sure that the module BASE_NAME contains the word Test...

[Defines]
  BASE_NAME      = SampleGoogleTestHost

GoogleTest Requirements - DSC

In our DSC file, we'll need to bring in the INF file that was just created into the [Components] section so that the unit tests will be built.

See this example in UnitTestFrameworkPkgHostTest.dsc...

#include <gtest/gtest.h>
extern "C" {
  #include <Uefi.h>
  #include <Library/BaseLib.h>
  #include <Library/DebugLib.h>
}

The first include brings in the GoogleTest definitions. Other EDK II related include files must be wrapped in extern "C" {} because they are C include files. Link failures will occur if this is not done.

Also, when using GoogleTest it is helpful to add a using declaration for its testing namespace. This using statement greatly reduces the amount of code you need to write in the tests when referencing the utilities within the testing namespace. For example, instead of writing ::testing::Return or ::testing::Test, you can just write Return or Test respectively, and these types of references occur numerous times within the tests.

Lastly, in the case that tests within a GoogleTest application require the usage of mock functions, it is also necessary to include the header files for those interfaces as well. As an example, the SecureBootVariableLibGoogleTest uses the mock versions of UefiLib and UefiRuntimeServicesTableLib. So its test file contains the below includes. Note that the using declaration mentioned above is also shown in the code below for completeness of the example.

#include <Library/GoogleTestLib.h>
#include <GoogleTest/Library/MockUefiLib.h>
#include <GoogleTest/Library/MockUefiRuntimeServicesTableLib.h>

extern "C" {
  #include <Uefi.h>
  ...
}

using namespace testing;

Now that we've got that squared away, let's look at our 'Main()' routine (or DriverEntryPoint() or whatever).

GoogleTest Configuration

Unlike the Framework, GoogleTest does not require test suites or test cases to be registered. Instead, the test cases declare the test suite name and test case name as part of their implementation. The only requirement for GoogleTest is to have a main() function that initializes the GoogleTest infrastructure and calls the service RUN_ALL_TESTS() to run all the unit tests.

int main(int argc, char* argv[]) {
  testing::InitGoogleTest(&argc, argv);
  return RUN_ALL_TESTS();
}

However, while GoogleTest does not require test suites or test cases to be registered, there is still one rule within EDK II that currently needs to be followed. This rule is that all tests for a given GoogleTest application must be contained within the same source file that contains the main() function shown above. These tests can be written directly in the file or a #include can be used to add them into the file indirectly.

The reason for this is due to EDK II taking the host application INF file and first compiling all of its source files into a static library. This static library is then linked into the final host application. The problem with this method is that only the tests in the object file containing the main() function are linked into the final host application. This is because the other tests are contained in their own object files within the static library and they have no symbols in them that the final host application depends on, so those object files are not linked into the final host application.

GoogleTest - A Simple Test Case

Below is a sample test case from SampleGoogleTestHost.

TEST(SimpleMathTests, OnePlusOneShouldEqualTwo) {
  UINTN  A;
  UINTN  B;
  UINTN  C;

  A = 1;
  B = 1;
  C = A + B;

  ASSERT_EQ (C, 2);
}

This uses the simplest form of a GoogleTest unit test using TEST() that declares the test suite name and the unit test name within that test suite. The unit test performs actions and typically makes calls to the code under test and contains test assertions to verify that the code under test behaves as expected for the given inputs.

In this test case, the ASSERT_EQ assertion is being used to establish that the business logic has functioned correctly. There are several assertion macros, and you are encouraged to use one that matches as closely to your intended test criterium as possible, because the logging is specific to the macro and more specific macros have more detailed logs. When in doubt, there are always ASSERT_TRUE and ASSERT_FALSE. Assertion macros that fail their test criterium will immediately return from the test case with a failed status and log an error string. Note that this early return can have implications for memory leakage.

For most ASSERT macros in GoogleTest there is also an equivalent EXPECT macro. Both macro versions will ultimately cause the TEST to fail if the check fails. However, the difference between the two macro versions is that when the check fails, the ASSERT version immediately returns from the TEST while the EXPECT version continues running the TEST.

There is no return status from a GooglTest unit test. If no assertions (or expectations) are triggered then the unit test has a passing status.

GoogleTest - A gMock Test Case

Below is a sample test case from SecureBootVariableLibGoogleTest. Although actually, the test case is not written exactly like this in the test file, but more on that in a bit.

TEST(SetSecureBootModeTest, SetVarError) {
  MockUefiRuntimeServicesTableLib RtServicesMock;
  UINT8                           SecureBootMode;
  EFI_STATUS                      Status;

  // Any random magic number can be used for these tests
  SecureBootMode = 0xAB;

  EXPECT_CALL(RtServicesMock, gRT_SetVariable)
    .WillOnce(Return(EFI_INVALID_PARAMETER));

  Status = SetSecureBootMode(SecureBootMode);
  EXPECT_EQ(Status, EFI_INVALID_PARAMETER);
}

Keep in mind that this test is written to verify that SetSecureBootMode() will return EFI_INVALID_PARAMETER when the call to gRT->SetVariable() within the implementation of SetSecureBootMode() returns EFI_INVALID_PARAMETER. With that in mind, let's discuss how a mock function is used to accomplish this in the test.

In this test case, the MockUefiRuntimeServicesTableLib interface is instantiated as RtServicesMock which enables its associated mock functions. These interface instantiations that contain the mock functions are very important for mock function based unit tests because without these instantiations, the mock functions within that interface will not exist and can not be used.

The next line of interest is the EXPECT_CALL, which is a standard part of the gMock framework. This macro is telling the test that a call is expected to occur to a specific function on a specific interface. The first argument is the name of the interface object that was instantiated in this test, and the second argument is the name of the mock function within that interface that is expected to be called. The WillOnce(Return(EFI_INVALID_PARAMETER)) associated with this EXPECT_CALL states that the gRT_SetVariable() function (remember from earlier in this documentation that this refers to the gRT->SetVariable() function) will be called once during this test, and when it does get called, we want it to return EFI_INVALID_PARAMETER.

Once this EXPECT_CALL has been setup, the call to SetSecureBootMode() occurs in the test, and its return value is saved in Status so that it can be tested. Based on the EXPECT_CALL that was setup earlier, when SetSecureBootMode() internally calls gRT->SetVariable(), it returns EFI_INVALID_PARAMETER. This value should then be returned by SetSecureBootMode(), and the EXPECT_EQ(Status, EFI_INVALID_PARAMETER) verifies this is the case.

There is much more that can be done with EXPECT_CALL and mock functions, but we will leave those details to be explained in the gMock documentation.

Now it was mentioned earlier that this test case is not written exactly like this in the test file, and the next section describes how this test is slightly refactored to reduce the total amount of code in the entire test suite.

GoogleTest - A gMock Test Case (refactored)

The sample test case from SecureBootVariableLibGoogleTest in the prior section is actually written as shown below.

class SetSecureBootModeTest : public Test {
  protected:
    MockUefiRuntimeServicesTableLib RtServicesMock;
    UINT8       SecureBootMode;
    EFI_STATUS  Status;

    void SetUp() override {
      // Any random magic number can be used for these tests
      SecureBootMode = 0xAB;
    }
};

TEST_F(SetSecureBootModeTest, SetVarError) {
  EXPECT_CALL(RtServicesMock, gRT_SetVariable)
    .WillOnce(Return(EFI_INVALID_PARAMETER));

  Status = SetSecureBootMode(SecureBootMode);
  EXPECT_EQ(Status, EFI_INVALID_PARAMETER);
}

This code may at first seem more complicated, but you will notice that the code with in it is still the same. There is still a MockUefiRuntimeServicesTableLib instantiation, there is still a SecureBootMode and Status variable defined, there is still an EXPECT_CALL, and etc. However, the benefit of constructing the test this way is that the new TEST_F() requires less code than the prior TEST().

This is made possible by the usage of what GoogleTest calls a test fixture. This concept of a test fixture allows multiple tests to use (or more specifically inherit from a base class) a common set of variables and initial conditions. Notice that using TEST_F() requires the first argument to be a name that aligns with a test fixture (in this case SetSecureBootModeTest), and the second argument is the name of the test (just like in the TEST() macro).

All TEST_F() tests that use a specific test fixture can be thought of as having all of that test fixture's variables automatically defined in the test as well as having that text fixture's SetUp() function called before entering the test.

This means that another TEST_F() can be written without needing to worry about defining a bunch of variables or instantiating a bunch of interfaces for mock functions. For example, the below test (also in SecureBootVariableLibGoogleTest) uses the same test fixture and makes use of its RtServicesMock, Status, and SecureBootMode variables.

TEST_F(SetSecureBootModeTest, PropogateModeToSetVar) {
  EXPECT_CALL(RtServicesMock,
    gRT_SetVariable(
      Char16StrEq(EFI_CUSTOM_MODE_NAME),
      BufferEq(&gEfiCustomModeEnableGuid, sizeof(EFI_GUID)),
      EFI_VARIABLE_NON_VOLATILE | EFI_VARIABLE_BOOTSERVICE_ACCESS,
      sizeof(SecureBootMode),
      BufferEq(&SecureBootMode, sizeof(SecureBootMode))))
    .WillOnce(Return(EFI_SUCCESS));

  Status = SetSecureBootMode(SecureBootMode);
  EXPECT_EQ(Status, EFI_SUCCESS);
}

The biggest benefit is that the TEST_F() code can now focus on what is being tested and not worry about any repetitive setup. There is more that can be done with test fixtures, but we will leave those details to be explained in the gMock documentation.

Now, as for what is in the above test, it is slightly more complicated than the first test. So let's explain this added complexity and what it is actually testing. In this test, there is still an EXPECT_CALL for the gRT_SetVariable() function. However, in this test we are stating that we expect the input arguments passed to gRT_SetVariable() be specific values. The order they are provided in the EXPECT_CALL align with the order of the arguments in the gRT_SetVariable() function. In this case the order of the gRT_SetVariable() arguments is as shown below.

IN  CHAR16                       *VariableName,
IN  EFI_GUID                     *VendorGuid,
IN  UINT32                       Attributes,
IN  UINTN                        DataSize,
IN  VOID                         *Data

So in the EXPECT_CALL we are stating that the call to gRT_SetVariable() will be made with the below input argument values.

  1. VariableName is equal to the EFI_CUSTOM_MODE_NAME string
  2. VendorGuid is equal to the gEfiCustomModeEnableGuid GUID (which has a byte length of sizeof(EFI_GUID))
  3. Attributes is equal to EFI_VARIABLE_NON_VOLATILE | EFI_VARIABLE_BOOTSERVICE_ACCESS
  4. DataSize is equal to sizeof(SecureBootMode)
  5. Data is equal to SecureBootMode (which has a byte length of sizeof(SecureBootMode))

If any one of these input arguments does not match in the actual call to gRT_SetVariable() in the design, then the test will fail. There is much more that can be done with EXPECT_CALL and mock functions, but again we will leave those details to be explained in the gMock documentation.

GoogleTest - More Complex Cases

To write more advanced tests, take a look at the GoogleTest User's Guide.

Development

Iterating on a Single Test

When using the EDK2 Pytools for CI testing, the host-based unit tests will be built and run on any build that includes the NOOPT build target.

If you are trying to iterate on a single test, a convenient pattern is to build only that test module. For example, the following command will build only the SafeIntLib host-based test from the MdePkg...

stuart_ci_build -c .pytool/CISettings.py TOOL_CHAIN_TAG=VS2017 -p MdePkg -t NOOPT BUILDMODULE=MdePkg/Test/UnitTest/Library/BaseSafeIntLib/TestBaseSafeIntLib.inf

Hooking BaseLib

Most unit test mocking can be performed by the functions provided in the UnitTestFrameworkPkg libraries, but since BaseLib is consumed by the Framework itself, it requires different techniques to substitute parts of the functionality.

To solve some of this, the UnitTestFrameworkPkg consumes a special implementation of BaseLib for host-based tests. This implementation contains a hook table that can be used to substitute test functionality for any of the BaseLib functions. By default, this implementation will use the underlying BaseLib implementation, so the unit test writer only has to supply minimal code to test a particular case.

Debugging the Framework Itself

While most of the tests that are produced by the UnitTestFrameworkPkg are easy to step through in a debugger, the Framework itself consumes code (mostly Cmocka) that sets its own build flags. These flags cause parts of the Framework to not export symbols and captures exceptions, and as such are harder to debug. We have provided a Stuart parameter to force symbolic debugging to be enabled.

You can run a build by adding the BLD_*_UNIT_TESTING_DEBUG=TRUE parameter to enable this build option.

stuart_ci_build -c .pytool/CISettings.py TOOL_CHAIN_TAG=VS2019 -p MdePkg -t NOOPT BLD_*_UNIT_TESTING_DEBUG=TRUE

Building and Running Host-Based Tests

The EDK2 CI infrastructure provides a convenient way to run all host-based tests -- in the the entire tree or just selected packages -- and aggregate all the reports, including highlighting any failures. This functionality is provided through the Stuart build system (published by EDK2-PyTools) and the NOOPT build target. The sections that follow use Framework examples. Unit tests based on GoogleTest are built and run the same way. The text output and JUNIT XML output format have small differences.

Building Locally

First, to make sure you're working with the latest PyTools, run the following command:

# Would recommend running this in a Python venv, but that's out of scope for this doc.
python -m pip install --upgrade -r ./pip-requirements.txt

After that, the following commands will set up the build and run the host-based tests.

# Setup repo for building
# stuart_setup -c ./.pytool/CISettings.py TOOL_CHAIN_TAG=<GCC5, VS2019, etc.>
stuart_setup -c ./.pytool/CISettings.py TOOL_CHAIN_TAG=VS2019

# Mu specific step to clone mu repos required for ci check
# stuart_ci_setup -c ./.pytool/CISettings.py TOOL_CHAIN_TAG=<GCC5, VS2019, etc.>
stuart_ci_setup -c ./.pytool/CISettings.py TOOL_CHAIN_TAG=VS2019

# Update all binary dependencies
# stuart_update -c ./.pytool/CISettings.py TOOL_CHAIN_TAG=<GCC5, VS2019, etc.>
stuart_update -c ./.pytool/CISettings.py TOOL_CHAIN_TAG=VS2019

# Build and run the tests
# stuart_ci_build -c ./.pytool/CISettings.py TOOL_CHAIN_TAG=<GCC5, VS2019, etc.> -t NOOPT [-p <Package Name>]
stuart_ci_build -c ./.pytool/CISettings.py TOOL_CHAIN_TAG=VS2019 -t NOOPT -p MdePkg

Evaluating the Results

In your immediate output, any build failures will be highlighted. You can see these below as "WARNING" and "ERROR" messages.

(edk_env) PS C:\_uefi\edk2> stuart_ci_build -c .\.pytool\CISettings.py TOOL_CHAIN_TAG=VS2019 -t NOOPT -p MdePkg

SECTION - Init SDE
SECTION - Loading Plugins
SECTION - Start Invocable Tool
SECTION - Getting Environment
SECTION - Loading plugins
SECTION - Building MdePkg Package
PROGRESS - --Running MdePkg: Host Unit Test Compiler Plugin NOOPT --
WARNING - Allowing Override for key TARGET_ARCH
PROGRESS - Start time: 2020-07-27 17:18:08.521672
PROGRESS - Setting up the Environment
PROGRESS - Running Pre Build
PROGRESS - Running Build NOOPT
PROGRESS - Running Post Build
SECTION - Run Host based Unit Tests
SUBSECTION - Testing for architecture: X64
WARNING - TestBaseSafeIntLibHost.exe Test Failed
WARNING -   Test SafeInt8ToUint8 - UT_ASSERT_EQUAL(0x5b:5b, Result:5c)
c:\_uefi\edk2\MdePkg\Test\UnitTest\Library\BaseSafeIntLib\TestBaseSafeIntLib.c:35: error: Failure!
ERROR - Plugin Failed: Host-Based Unit Test Runner returned 1
CRITICAL - Post Build failed
PROGRESS - End time: 2020-07-27 17:18:19.792313  Total time Elapsed: 0:00:11
ERROR - --->Test Failed: Host Unit Test Compiler Plugin NOOPT returned 1
ERROR - Overall Build Status: Error
PROGRESS - There were 1 failures out of 1 attempts
SECTION - Summary
ERROR - Error

(edk_env) PS C:\_uefi\edk2>

If a test fails, you can run it manually to get more details...

(edk_env) PS C:\_uefi\edk2> .\Build\MdePkg\HostTest\NOOPT_VS2019\X64\TestBaseSafeIntLibHost.exe

Int Safe Lib Unit Test Application v0.1
---------------------------------------------------------
------------     RUNNING ALL TEST SUITES   --------------
---------------------------------------------------------
---------------------------------------------------------
RUNNING TEST SUITE: Int Safe Conversions Test Suite
---------------------------------------------------------
[==========] Running 71 test(s).
[ RUN      ] Test SafeInt8ToUint8
[  ERROR   ] --- UT_ASSERT_EQUAL(0x5b:5b, Result:5c)
[   LINE   ] --- c:\_uefi\edk2\MdePkg\Test\UnitTest\Library\BaseSafeIntLib\TestBaseSafeIntLib.c:35: error: Failure!
[  FAILED  ] Test SafeInt8ToUint8
[ RUN      ] Test SafeInt8ToUint16
[       OK ] Test SafeInt8ToUint16
[ RUN      ] Test SafeInt8ToUint32
[       OK ] Test SafeInt8ToUint32
[ RUN      ] Test SafeInt8ToUintn
[       OK ] Test SafeInt8ToUintn
...

You can also, if you are so inclined, read the output from the exact instance of the test that was run during stuart_ci_build. The output file can be found on a path that looks like:

Build/<Package>/HostTest/<Arch>/<TestName>.<TestSuiteName>.<Arch>.result.xml

A sample of this output looks like:

<!--
  Excerpt taken from:
  Build\MdePkg\HostTest\NOOPT_VS2019\X64\TestBaseSafeIntLibHost.exe.Int Safe Conversions Test Suite.X64.result.xml
  -->
<?xml version="1.0" encoding="UTF-8" ?>
<testsuites>
  <testsuite name="Int Safe Conversions Test Suite" time="0.000" tests="71" failures="1" errors="0" skipped="0" >
    <testcase name="Test SafeInt8ToUint8" time="0.000" >
      <failure><![CDATA[UT_ASSERT_EQUAL(0x5c:5c, Result:5b)
c:\_uefi\MdePkg\Test\UnitTest\Library\BaseSafeIntLib\TestBaseSafeIntLib.c:35: error: Failure!]]></failure>
    </testcase>
    <testcase name="Test SafeInt8ToUint16" time="0.000" >
    </testcase>
    <testcase name="Test SafeInt8ToUint32" time="0.000" >
    </testcase>
    <testcase name="Test SafeInt8ToUintn" time="0.000" >
    </testcase>

XML Reporting Mode

Unit test applications using Framework are built using Cmocka that requires the following environment variables to be set to generate structured XML output rather than text:

CMOCKA_MESSAGE_OUTPUT=xml
CMOCKA_XML_FILE=<absolute or relative path to output file>

Unit test applications using GoogleTest require the following environment variable to be set to generate structured XML output rather than text:

GTEST_OUTPUT=xml:<absolute or relative path to output file>

This mode is used by the test running plugin to aggregate the results for CI test status reporting in the web view.

Code Coverage

Code coverage can be enabled for Host based Unit Tests with CODE_COVERAGE=TRUE, which generates a cobertura report per package tested, and combined cobertura report for all packages tested. The per-package cobertura report will be present at Build/<Pkg>/HostTest/<Target_Toolchain>/<Pkg>_coverage.xml. The overall cobertura report will be present at Build/coverage.xml

Code coverage generation has two config knobs:

  1. CC_FULL: If set to TRUE, will generate zero'd out coverage data for untested source files in the package.
  2. CC_FLATTEN: If Set to TRUE, will group all source files together, rather than by INF.

**TIP: CC_FLATTEN=TRUE/FALSE will produce different coverage percentage results as TRUE de-duplicates source files that are consumed by multiple INFs.

For Windows, this is primarily leverage for pipeline builds, but this can be leveraged locally using the OpenCppCoverage windows tool to parse coverage data to cobertura xml format.

Prerequisites

In addition to required prerequisites to build and test, there are additional requirements for calculating code coverage files as noted below.

Examples

stuart_ci_build -c .pytool/CISettings.py -t NOOPT TOOL_CHAIN_TAG=VS2019 -p MdeModulePkg CODE_COVERAGE=TRUE
stuart_ci_build -c .pytool/CISettings.py -t NOOPT TOOL_CHAIN_TAG=VS2019 CODE_COVERAGE=TRUE CC_FLATTEN=TRUE CC_FULL=FALSE

How to see code coverage data on IDE Visual Studio

Open Visual Studio VS2019 or above version
Click "Tools" -> "OpenCppCoverage Settings"
Fill your execute file into "Program to run:"
Click "Tools" -> "Run OpenCppCoverage"

Additional Tools

There are a plethora of open source tools for generating reports from a Cobertura file, which is why it was selected as the output file format. Tools such as pycobertura (pip install pycobertura) and reportgenerator can be utilized to generate different report types, such as local html reports. VSCode Extensions such as Coverage Gutters can highlight coverage results directly in the file, and cloud tools such as CodeCov can consume cobertura files to provide PR checks and general code coverage statistics for the repository.

*** REMINDER: During CI builds, use the `CODE_COVERAGE=TRUE flag to generate the code coverage XML files, and additionally use the CC_FLATTEN=TRUE or CC_FULL=TRUE flags to customize coverage results.

Important Note

This works on both Windows and Linux but is currently limited to x64 architectures. Working on getting others, but we also welcome contributions.

Framework Known Limitations

PEI, DXE, SMM

While sample tests have been provided for these execution environments, only cursory build validation has been performed. Care has been taken while designing the frameworks to allow for execution during boot phases, but only UEFI Shell and host-based tests have been thoroughly evaluated. Full support for PEI, DXE, and SMM is forthcoming, but should be considered beta/staging for now.

Host-Based Support vs Other Tests

The host-based test framework is powered internally by the Cmocka framework. As such, it has abilities that the target-based tests don't (yet). It would be awesome if this meant that it was a super set of the target-based tests, and it worked just like the target-based tests but with more features. Unfortunately, this is not the case. While care has been taken to keep them as close as possible, there are a few known inconsistencies that we're still ironing out. For example, the logging messages in the target-based tests are cached internally and associated with the running test case. They can be saved later as part of the reporting lib. This isn't currently possible with host-based. Only the assertion failures are logged.

We will continue trying to make these as similar as possible.

Unit Test Location/Layout Rules

Code/Test Location
Host-Based Unit Tests for a Library/Protocol/PPI/GUID Interface If what's being tested is an interface (e.g. a library with a public header file, like DebugLib) and the test is agnostic to a specific implementation, then the test should be scoped to the parent package.
Example: MdePkg/Test/UnitTest/[Library/Protocol/Ppi/Guid]/

A real-world example of this is the BaseSafeIntLib test in MdePkg.
MdePkg/Test/UnitTest/Library/BaseSafeIntLib/TestBaseSafeIntLibHost.inf
Host-Based Unit Tests for a Library/Driver (PEI/DXE/SMM) implementation If what's being tested is a specific implementation (e.g. BaseDebugLibSerialPort for DebugLib), then the test should be scoped to the implementation directory itself, in a UnitTest (or GoogleTest) subdirectory.

Module Example: MdeModulePkg/Universal/EsrtFmpDxe/UnitTest/
Library Example: MdePkg/Library/BaseMemoryLib/UnitTest/
Library Example (GoogleTest): SecurityPkg/Library/SecureBootVariableLib/GoogleTest/
Host-Based Tests for a Functionality or Feature If you're writing a functional test that operates at the module level (i.e. if it's more than a single file or library), the test should be located in the package-level Tests directory under the HostFuncTest subdirectory.
For example, if you were writing a test for the entire FMP Device Framework, you might put your test in:
FmpDevicePkg/Test/HostFuncTest/FmpDeviceFramework

If the feature spans multiple packages, it's location should be determined by the package owners related to the feature.
Non-Host-Based (PEI/DXE/SMM/UefiShell) Tests for a Functionality or Feature Similar to Host-Based, if the feature is in one package, should be located in the *Pkg/Test/[UefiShell/Dxe/Smm/Pei]Test directory.

If the feature spans multiple packages, it's location should be determined by the package owners related to the feature.

USAGE EXAMPLES
PEI Example: MP_SERVICE_PPI. Or check MTRR configuration in a notification function.
SMM Example: a test in a protocol callback function. (It is different with the solution that SmmAgent+ShellApp)
DXE Example: a test in a UEFI event call back to check SPI/SMRAM status.
Shell Example: the SMM handler audit test has a shell-based app that interacts with an SMM handler to get information. The SMM paging audit test gathers information about both DXE and SMM. And the SMM paging functional test actually forces errors into SMM via a DXE driver.

Example Directory Tree

<PackageName>Pkg/
  ComponentY/
    ComponentY.inf
    ComponentY.c
    GoogleTest/
      ComponentYHostGoogleTest.inf    # Host-Based Test for Driver Module
      ComponentYGoogleTest.cpp
    UnitTest/
      ComponentYUnitTestHost.inf      # Host-Based Test for Driver Module
      ComponentYUnitTest.c

  Library/
    GeneralPurposeLibBase/
      ...

    GeneralPurposeLibSerial/
      ...

    SpecificLibDxe/
      SpecificLibDxe.c
      SpecificLibDxe.inf
      GoogleTest/                    # Host-Based Test for Specific Library Implementation
        SpecificLibDxeHostGoogleTest.cpp
        SpecificLibDxeHostGoogleTest.inf
      UnitTest/                      # Host-Based Test for Specific Library Implementation
        SpecificLibDxeUnitTest.c
        SpecificLibDxeUnitTestHost.inf
  Test/
    <Package>HostTest.dsc             # Host-Based Test Apps
    GoogleTest/
      InterfaceX
        InterfaceXHostGoogleTest.inf  # Host-Based App (should be in Test/<Package>HostTest.dsc)
        InterfaceXUnitTest.cpp        # Test Logic

      GeneralPurposeLib/              # Host-Based Test for any implementation of GeneralPurposeLib
        GeneralPurposeLibTest.cpp
        GeneralPurposeLibHostUnitTest.inf

    UnitTest/
      InterfaceX
        InterfaceXUnitTestHost.inf    # Host-Based App (should be in Test/<Package>HostTest.dsc)
        InterfaceXUnitTestPei.inf     # PEIM Target-Based Test (if applicable)
        InterfaceXUnitTestDxe.inf     # DXE Target-Based Test (if applicable)
        InterfaceXUnitTestSmm.inf     # SMM Target-Based Test (if applicable)
        InterfaceXUnitTestUefiShell.inf # Shell App Target-Based Test (if applicable)
        InterfaceXUnitTest.c          # Test Logic

      GeneralPurposeLib/              # Host-Based Test for any implementation of GeneralPurposeLib
        GeneralPurposeLibTest.c
        GeneralPurposeLibUnitTestHost.inf
    Mock/
      Include/
        GoogleTest/
          Library/
            MockGeneralPurposeLib.h

      Library/
        GoogleTest/
          MockGeneralPurposeLib/
            MockGeneralPurposeLib.cpp
            MockGeneralPurposeLib.inf

  <Package>Pkg.dsc          # Standard Modules and any Target-Based Test Apps (including in Test/)

Future Locations in Consideration

We don't know if these types will exist or be applicable yet, but if you write a support library or module that matches the following, please make sure they live in the correct place.

Code/Test Location
Host-Based Library Implementations Host-Based Implementations of common libraries (eg. MemoryAllocationLibHost) should live in the same package that declares the library interface in its .DEC file in the *Pkg/Test/Library directory. Should have 'Host' in the name.
Host-Based Mocks and Stubs Mock and Stub libraries that require test infrastructure should live in the UefiTestFrameworkPkg/Library with either 'Mock' or 'Stub' in the library name.

If still in doubt

Hop on GitHub and ask @corthon, @mdkinney, or @spbrogan. ;)

Copyright © Microsoft Corporation. SPDX-License-Identifier: BSD-2-Clause-Patent