The objective of using Kconfig in RIOT is to configure software modules at compile-time. This means having a standard way of:
Modules in RIOT expose their configurable parameters via Kconfig files (for more information on Kconfig syntax check the specification). In these files documentation, restrictions, default values and dependencies can be expressed.
Kconfig files are structured through the file system mirroring the current module distribution. In time, all modules will have Kconfig files to make themselves configurable through this system.
The user can assign values to the exposed parameters, either by manually writing '.config' files or using an interface such as Menuconfig. Parameters with no assigned values will take the default ones. For a detailed distinction between Kconfig and '.config' files see Appendix B.
Using '.config' and Kconfig files the build system takes care of doing the necessary checks on the values according to the parameter definition. After that, the
autoconf.h header file is generated, it contains all the configurations in the form of (
CONFIG_ prefixed) macros.
In order to use the graphical interface menuconfig to configure the application, run
make menuconfig in the application's folder. All available configurations (based on the used modules) for the particular platform will be presented. By default, the configuration of a module via Kconfig is not enabled. In order to activate the configuration via Kconfig the corresponding option should be selected. That will enable the configuration of all inner options, if available.
Once the desired configuration is achieved save the configuration to the default proposed path and exit. The saved configuration will be applied when the code is compiled (
If the current configuration will be used in the future it can be saved in the application's folder as
user.config, using the 'Save' option in menuconfig. This way it will be persistent after cleaning the application directory (
The second way to configure the application is by directly writing '.config' files. Two files will be sources of configuration during the generation of the final header file:
user.config, which should be placed inside the application's folder.
app.config sets default configuration values for the particular application, the user can override them by setting them in
user.config. Additionally, further
.config files can be added to the variable
KCONFIG_ADD_CONFIG, which will be applied after default CPU and board configurations,
user.config. This means that they will have priority.
Let's say that the
SOCK_UTIL_SCHEME_MAXLEN symbol in
sock_util module needs to be configured. The
user.config file could look like:
In this case, there is no need for using menuconfig. It's enough just to call
make all in the application folder, as this configuration will be read and applied. Note that if any dependency issue occurs, warnings will be generated (e.g. not enabling the configuration of a module via Kconfig).
To expose application-specific configuration options a
Kconfig file can be placed in the application's folder. For an example of this you can check the tests/kconfig application.
For easy debugging of configuration or testing new modules by compiling them into existing applications, one can also use environment variables prefixed by
RIOT_CONFIG_. To achieve the same configuration exemplified in Configure using '.config' files, e.g., you could also use
All the checks that apply for
.config files also are done with this approach.
Mind that this is only meant to be used during development. In production, please set the configuration via
When a certain module is being configured via Kconfig the configuration macro will no longer be overridable by means of CFLAGS (e.g. set on the compilation command or on a Makefile). Consider this if you are getting a 'redefined warning'.
The integration of Kconfig into the build system is mainly done in
Currently, the resolution of module dependencies is performed by the build system where all the used modules and packages end up listed in the
USEMODULE make variables. In the next phases of integration we plan to resolve dependencies using Kconfig.
The list of modules needed for the particular build is dumped into the
$ (GENERATED_DIR)/Kconfig.dep file, where each module is translated into a Kconfig symbol as documented in Appendix A.
In this step configuration values are taken from multiple sources and merged into a single
out.config configuration file. This file is temporary and is removed on clean. If the user needs to save a particular configuration set, a backup has to be saved (this can be done using the menuconfig interface) so it can be loaded later in this step.
To accomplish merging of multiple input files, the
genconfig script is used. Note that the order matters: existing configuration values are merged in the order expressed in the input section, where the last value assigned to a parameter has the highest priority. If no configuration files are available all default values will be applied.
out.config is the only configuration input for the
autoconf.h in the generation step.
Additionally this step generates a file
out.config.d which holds the information of all the used Kconfig files in Makefile format. This file is included by the build system and allows to re-trigger the generation of
out.conf whenever a Kconfig file is modified.
$ (APPDIR)/app.config: Application specific default configurations.
$ (APPDIR)/user.config: Configurations saved by user.
Menuconfig is a graphical interface for software configuration. It is used for the configuration of the Linux kernel. This section explains the process that occurs when RIOT is being configured using the menuconfig interface.
Kconfig file is used in this step to show the configurable parameters of the system. Kconfig will filter inapplicable parameters (i.e. parameters exposed by modules that are not being used) based on the file
$ (GENERATED_DIR)/Kconfig.dep generated in step 1.
During the transition phase, the user needs to enable Kconfig explicitly per module, by setting the corresponding option. If using
menuconfig a checkbox with a submenu has to be selected, if using
.config files a
CONFIG_KCONFIG_MODULE_ prefixed option has to be set to
y. For more information see Making configuration via Kconfig optional.
Note that if Kconfig is not used to configure a module, the corresponding header files default values will be used.
out.config is one of the inputs for menuconfig. This means that any configuration that the application defines in the
app.config or a backup configuration from the user in
user.config are taken into account on the first run (see Appendix C).
In this step the user chooses configuration values (or selects the minimal configuration) and saves it to the
out.config file. Here the user can choose to save a backup configuration file for later at a different location (e.g. a
user.config file in the application folder).
If any changes occur to
out.config, the generation of autoconf.h is executed automatically.
$ (GENERATED_DIR)/out.config.oldbackup file.
With the addition of Kconfig a dependency has been added to the build process: the
$ (GENERATED_DIR)/autoconf.h header file. This header file is the main output from the Kconfig configuration system. It holds all the macros that should be used to configure modules in RIOT:
In order to generate the
autoconf.h file the
genconfig script is used. Inputs for this script are the main
Kconfig file and
out.config configuration file, which holds the selected values for the exposed parameters.
Kconfigfile exposing configuration of modules.
$ (GENERATED_DIR)/autoconf.hconfiguration header file.
$ (GENERATED_DIR)/deps/*/*.hheader files that allow incremental builds
These files are defined in
|Defines configuration options of modules.|
|Holds a list of the modules that are being compiled.|
|Holds default application configuration values.|
|Holds configuration values applied by the user.|
|Configuration file containing all the symbols defined in |
|Dependency file of |
|Header file containing the macros that applied the selected configuration.|
As '.config' files have Makefile syntax they can be included when building, which allows to access the applied configuration from the build system and, in the future, to check for enabled modules.
During migration this is also useful, as it gives the ability to check if a parameter is being configured via Kconfig or a default value via
CFLAGS could be injected. For example:
Symbols will have the same name as the configuration macros (thus will always have the
CONFIG_ prefix). As the configuration file is loaded in
Makefile.include care should be taken when performing checks in the application's Makefile. The symbols will not be defined until after including
During transition to the usage of Kconfig as the main configuration tool for RIOT, the default behavior will be the traditional one: expose configuration options in header files and use CFLAGS as inputs. To allow optional configuration via Kconfig, a convention will be used when writing Kconfig files.
Modules should be contained in their own
menuconfig entries, this way the user can choose to enable the configuration via Kconfig for an specific module. These entries should define a dependency on the module they configure (see Appendix A to see how to check if a module is being used).
The module configuration then can be enabled either via the menuconfig interface:
or by means of a '.config' file:
During the current migration phase architectures, CPUs, boards and provided features are being modelled in Kconfig. The following is a guide on how to organize and name the symbols.
Features must be modelled as hidden boolean symbols with the prefix
HAS_. They must contain a
help attribute clearly specifying what providing that feature means. The location of the symbol declaration depends on the type of feature. Features that are not platform-specific (e.g.
cpp) must be placed in
/kconfigs/Kconfig.features. If a feature is specific to a certain CPU family or vendor, it should be placed in the correspondent Kconfig file (e.g.
esp_wifi_enterprise). Features related to modules should be placed in the Kconfig file of that module.
arduino is placed in
/kconfigs/Kconfig.features and modelled like:
The proposed hierarchy for the classification of CPUs is as follows:
Where each hierarchy is defined as:
CPU_MODEL: The specific identifier of the used CPU, used for some CPU implementations to differentiate between different memory layouts.
CPU_FAM: An intermediate identifier between CPU and CPU_MODEL that represents a sub-group of a Manufacturers CPU's.
CPU_CORE: The specific identifier of the core present in the CPU.
CPU_ARCH: The specific identifier of the architecture of the core defined in
In order to model the hierarchies, a hidden boolean symbol must be declared for each. The name of the symbol must begin with the correspondent prefix and must be followed by the specific value. For instance, the 'samd21' family symbol is named
In addition, a default value to the correspondent common symbol must be defined. The default value must be guarded by the boolean symbol correspondent to the hierarchy.
Features may be provided by any hierarchy symbol. Usually symbols are selected from more specific to less specific. This means that a
CPU_MODEL_<model> symbol usually would select the correspondent
CPU_FAM_<family> symbol, which would in turn select the
CPU_CORE_<core>. This may change in some cases where
CPU_COMMON_ symbols are defined to avoid repetition. For convenience and if it makes sense within a CPU vendor family, it's also allowed to use intermediate grouping levels, like
CPU_LINE_<xxx> used for STM32.
In addition to the symbols of the hierarchy described above, a default value to the
CPU symbol should be assigned, which will match the value of the
CPU Makefile variable in the build system.
The declaration of the symbols should be placed in a
Kconfig file in the folder that corresponds to the hierarchy. When the symbols are scattered into multiple files, it is responsibility of file containing the most specific symbols to
source the less specific. Keep in mind that only the file located in
/cpu/<CPU>/Kconfig will be included by the root
Boards must be modelled as hidden boolean symbols with the prefix
BOARD_ which default to
y and are placed in
/boards/<BOARD>/Kconfig. This file will be
sourced from the main
/Kconfig file. The board symbol must select the
CPU_MODEL_<model> symbol that corresponds to the CPU model present on the board. The board symbol must also select the symbols that correspond to the features it provides.
In the same
Kconfig file a default value must be assigned to the common
BOARD symbol. It must be guarded by the board's symbol, so it only applies in that case.
There are cases when grouping common code for multiple boards helps to avoid unnecessary repetition. In the case features are provided in a common board folder (e.g.
/boards/common/arduino-atmega) a symbol should be declared to model this in Kconfig. Symbols for common boards must have the
BOARD_COMMON_ prefix, and must select the common provided features.
The samr21-xpro has a
samr21g18a CPU and provides multiple features. Its symbol is modelled as following:
Boards, common board directories, CPUs and common CPU directories may need to override default configuration values. Visible configuration symbols are configurable by the user and show on the menuconfig interface.
.config files are used to set their values. To allow multiple sources of
.config files, there are two Makefile variables developers should use:
KCONFIG_CPU_CONFIG for sources added by the CPU or common CPU directories, and
KCONFIG_BOARD_CONFIG for sources added by the board or common board directories. This ensures the correct priority of the configurations.
Makefile.features infrastructure is used to populate the configuration sources. As the order in which
.config files are merged matters, configuration sources should be ordered from more generic to more specific. Because board's
Makefile.features is included before CPU's
Makefile.features it is important to utilize two different lists of configuration sources. For instance, if
cpu/cortexm_common adds its configuration,
cpu/stm32 should add its configuration after it, and
boards/stm32f769i-disco after it.
The following symbol prefixes have been assigned particular semantics and are reserved for the cases described below:
|Models a board|
|Used for common symbols used by multiple boards|
|Models a CPU architecture|
|Used for common symbols used by multiple CPUs|
|Models a CPU core|
|Models a family of CPUs|
|Models a particular model of CPU|
|Models a feature|
|Used during transition to enable configuration of a module via Kconfig|
|Used during transition to enable configuration of a package via Kconfig|
|Models a feature not present in makefiles, will be unified in the future|
|Models a RIOT module. Generated from |
|Models an external package. Generated from |
In order to show only the relevant configuration parameters to the user with respect to a given application and board selection, Kconfig needs knowledge about all modules and packages to be used for a compilation. Currently dependency handling among modules is performed by the build system (via
Makefile.dep files). The interface defined to declared the used modules and packages is the
$ (GENERATED_DIR)/Kconfig.dep file.
Kconfig.dep is a Kconfig file that will define symbols of the form:
There will be a symbol for every used module (i.e. every module in
USEMODULE make variable) and package. The names in the symbols will be uppercase and separated by
_. Based on these symbols configurability is decided. Modules and packages symbols will have
PKG_ prefixes respectively.
The following is an example of how to use these symbols in Kconfig files to enable/disable a configuration menu:
Then, every configuration option for the previous module would be modeled like:
Kconfig files describe a configuration database, which is a collection of configuration options organized in a tree structure. Configuration options may have dependencies (among other attributes), which are used to determine their visibility.
Kconfig files are written in Kconfig language defined in the Linux kernel. Configuration options have attributes such as types, prompts and default values.
On the other hand configuration files contain assignment of values to configuration options and use Makefile syntax. They can also be used to save a set of configuration values as backup.
In other words: Kconfig files describe configuration options and '.config' files assign their values.
In the current configuration flow the user can choose to configure RIOT using the menuconfig graphical interface or writing '.config' files by hand.
As explained in the 'Configuration sources merging step' of the configuration process, configuration from multiple sources are loaded to create a single
out.config file, and the order of merging matters: last file has priority.
While editing values directly via '.config' files
out.config will be re-built. The user can also use menuconfig interface to modify the configuration file (this is the recommended way, as it gives access to much more information regarding dependencies and default values of the symbols). Menuconfig will change
out.config directly (a backup file
out.config.old will be kept).
It is recommended to save backups of the configurations, as any change on the configuration sources would re-trigger the merging process and overwrite
A macro that holds a 0 or 1 is modelled in Kconfig as a
bool symbol. References to this macro can then make use of IS_ACTIVE macro from kernel_defines.h with C conditionals for conditional compilation. FXOS8700 driver exposure to Kconfig can be considered as an example. If the macro is defined as
TRUE by default, a new symbol gets introduced to invert the semantics. The recommended practice is to add a new symbol and expose it to Kconfig while the old one is tagged to be deprecated. The process is documented in this commit
There may be cases where a macro is expected to hold only specific values, e.g. 'GNRC_IPV6_MSG_QUEUE_SIZE' expressed as the power of two. These may be modelled in such a way that a new macro is introduced to hold the restricted figures while operators are added to arrive at the desired value. The process is documented in this pull request.