Libmetal provides common user APIs to access devices, handle device interrupts and request memory across the following operating environments:
- Linux user space (based on UIO and VFIO support in the kernel)
- RTOS (with and without virtual memory)
- Bare-metal environments
For more details on the framework please refer to the OpenAMP Docs.
The configuration phase begins when the user invokes CMake. CMake begins by processing the CMakeLists.txt file and the cmake directory. Some cmake options are available to help user to customize the libmetal to their own project.
- WITH_DOC (default ON): Build with documentation. Add -DWITH_DOC=OFF in cmake command line to disable.
- WITH_EXAMPLES (default ON): Build with application examples. Add -DWITH_DOC=OFF in cmake command line to disable the option.
- WITH_TESTS (default ON): Build with application tests. Add -DWITH_DOC=OFF in cmake command line to disable the option.
- WITH_DEFAULT_LOGGER (default ON): Build with default trace logger. Add -DWITH_DEFAULT_LOGGER=OFF in cmake command line to disable the option.
- WITH_SHARED_LIB (default ON): Generate a shared library. Add -DWITH_SHARED_LIB=OFF in cmake command line to disable the option.
- WITH_STATIC_LIB (default ON): Generate a static library. Add -DWITH_STATIC_LIB=OFF in cmake command line to disable the option.
- WITH_ZEPHYR (default OFF): Build for Zephyr environment. Add -DWITH_ZEPHYR=ON in cmake command line to enable the the option.
$ git clone https://github.com/OpenAMP/libmetal.git
$ mkdir -p libmetal/<build directory>
$ cd libmetal/<build directory>
$ cmake ..
$ make VERBOSE=1 DESTDIR=<libmetal install location> install
Use meta-openamp to build libmetal library.
Use package libmetal in your Yocto config file.
To build on baremetal, you will need to provide a toolchain file. Here is an example toolchain file:
set (CMAKE_SYSTEM_PROCESSOR "arm" CACHE STRING "")
set (MACHINE "zynqmp_r5" CACHE STRING "")
set (CROSS_PREFIX "armr5-none-eabi-" CACHE STRING "")
set (CMAKE_C_FLAGS "-mfloat-abi=soft -mcpu=cortex-r5 -Wall -Werror -Wextra \
-flto -Os -I/ws/xsdk/r5_0_bsp/psu_cortexr5_0/include" CACHE STRING "")
SET(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -flto")
SET(CMAKE_AR "gcc-ar" CACHE STRING "")
SET(CMAKE_C_ARCHIVE_CREATE "<CMAKE_AR> qcs <TARGET> <LINK_FLAGS> <OBJECTS>")
SET(CMAKE_C_ARCHIVE_FINISH true)
include (cross-generic-gcc)
- Note: other toolchain files can be found in the
cmake/platforms/directory. - Compile with your toolchain file.
$ mkdir -p build-libmetal
$ cd build-libmetal
$ cmake <libmetal_source> -DCMAKE_TOOLCHAIN_FILE=<toolchain_file>
$ make VERBOSE=1 DESTDIR=<libmetal_install> install
- Note: When building baremetal for Xilinx 2018.3 or earlier environments, add -DXILINX_PRE_V2019 to your CMake invocation. This will include the xilmem and xilstandalone libraries in your build. These libraries were removed in 2019.1.
The zephyr-libmetal implements the libmetal for the Zephyr project. It is mainly a fork of this repository, with some add-ons for integration in the Zephyr project.
Following instruction is only to be able to run test application on a QEMU running a Zephyr environment.
As Zephyr uses CMake, we build libmetal library and test application as targets of Zephyr CMake project. Here is how to build libmetal for Zephyr:
$ export ZEPHYR_TOOLCHAIN_VARIANT=zephyr
$ export ZEPHYR_SDK_INSTALL_DIR=<where Zephyr SDK is installed>
$ source <git_clone_zephyr_project_source_root>/zephyr-env.sh
$ cmake <libmetal_source_root> -DWITH_ZEPHYR=on -DBOARD=qemu_cortex_m3 \
[-DWITH_TESTS=on]
$ make VERBOSE=1 all
# If we have turned on tests with "-DWITH_TESTS=on" when we run cmake,
# we launch libmetal test on Zephyr QEMU platform as follows:
$ make VERBOSE=1 run
The following subsections give an overview of interfaces provided by libmetal.
These interfaces do not need to be ported across to new operating systems.
The libmetal I/O region abstraction provides access to memory mapped I/O and shared memory regions. This includes:
- primitives to read and write memory with ordering constraints, and
- ability to translate between physical and virtual addressing on systems that support virtual memory.
The libmetal logging interface is used to plug log messages generated by libmetal into application specific logging mechanisms (e.g. syslog). This also provides basic message prioritization and filtering mechanisms.
This is a simple doubly linked list implementation used internally within libmetal, and also available for application use.
The following utilities are provided in lib/utilities.h:
- Min/max, round up/down, etc.
- Bitmap operations
- Helper to compute container structure pointers
- ... and more ...
The libmetal version interface allows user to get the version of the library. The version increment follows the set of rule proposed in Semantic Versioning specification.
The users will need to call two top level interfaces to use libmetal APIs:
- metal_init - initialize the libmetal resource
- metal_finish - release libmetal resource
Each system needs to have their own implementation inside libmetal for these two APIs to call:
- metal_sys_init
- metal_sys_finish
For the current release, libmetal provides Linux userspace and bare-metal implementation for metal_sys_init and metal_sys_finish.
For Linux userspace, metal_sys_init sets up a table for available shared pages, checks whether UIO/VFIO drivers are avail, and starts interrupt handling thread.
Please note that on Linux, to access device's memory that is not page aligned,
an offset has to be added to the pointer returned by mmap(). This offset,
although it can be read from the device tree property exposed by the uio driver,
is not handled yet by the library.
For bare-metal, metal_sys_init and metal_sys_finish just returns.
The libmetal atomic operations API is consistent with the C11/C++11 stdatomics interface. The stdatomics interface is commonly provided by recent toolchains including GCC and LLVM/Clang. When porting to a different toolchain, it may be necessary to provide an stdatomic compatible implementation if the toolchain does not already provide one.
libmetal provides memory allocation and release APIs.
libmetal provides the following locking APIs.
libmetal has a generic mutex implementation which is a busy wait. It is recommended to have OS specific implementation for mutex.
The Linux userspace mutex implementation uses futex to wait for the lock and wakeup a waiter.
libmetal condition variable APIs provide "wait" for user applications to wait on some condition to be met, and "signal" to indicate a particular even occurs.
libmetal spinlock APIs provides busy waiting mechanism to acquire a lock.
libmetal has a generic static shared memory implementation. If your OS has a global shared memory allocation, you will need to port it for the OS.
The Linux userspace shmem implementation uses libhugetlbfs to support huge page sizes.
libmetal has a static generic implementation. If your OS has a driver model implementation, you will need to port it for the OS.
The Linux userspace abstraction binds the devices to UIO or VFIO driver. The user applications specify which device to use, e.g. bus "platform" bus, device "f8000000.slcr", and then the abstraction will check if platform UIO driver or platform VFIO driver is there. If platform VFIO driver exists, it will bind the device to the platform VFIO driver, otherwise, if UIO driver exists, it will bind the device to the platform UIO driver.
The VFIO support is not yet implemented.
libmetal provides APIs to register an interrupt, disable interrupts and restore interrupts.
The Linux userspace implementation will use a thread to call select() function to listen to the file descriptors of the devices to see if there is an interrupt triggered. If there is an interrupt triggered, it will call the interrupt handler registered by the user application.
libmetal provides APIs to flush and invalidate caches.
The cache APIs for Linux userspace are empty functions for now as cache operations system calls are not available for all architectures.
libmetal DMA APIs provide DMA map and unmap implementation.
After calling DMA map, the DMA device will own the memory. After calling DMA unmap, the cpu will own the memory.
For Linux userspace, it only supports to use UIO device memory as DMA memory for this release.
libmetal time APIs provide getting timestamp implementation.
libmetal sleep APIs provide getting delay execution implementation.
This API is for compiler dependent functions. For this release, there is only a GCC implementation, and compiler specific code is limited to atomic operations.
- AMD-Xilinx Zynq7000. Last known stable release - v2023.04
As an open-source project, we welcome and encourage the community to submit patches directly to the project. As a contributor you should be familiar with common developer tools such as Git and CMake, and platforms such as GitHub.
Then following points should be rescpected to facilitate the review process.
Code is contributed to OpenAMP under a number of licenses, but all code must be compatible with version the BSD License, which is the license covering the OpenAMP distribution as a whole. In practice, use the following tag instead of the full license text in the individual files:
```
SPDX-License-Identifier: BSD-3-Clause
```
Commit messages must contain Signed-off-by: line and your email must match the change authorship information. Make sure your .gitconfig is set up correctly:
```
git config --global user.name "first-name Last-Namer"
git config --global user.email "[email protected]"
```
Before you submit a pull request to the project, verify your commit messages meet the requirements. The check can be performed locally using the the gitlint command.
Run gitlint locally in your tree and branch where your patches have been committed:
```gitlint```
Note, gitlint only checks HEAD (the most recent commit), so you should run it after each commit, or use the --commits option to specify a commit range covering all the development patches to be submitted.
In general, follow the Linux kernel coding style, with the following exceptions:
- Use /** */ for doxygen comments that need to appear in the documentation.
The Linux kernel GPL-licensed tool checkpatch is used to check coding style conformity. Checkpatch is available in the scripts directory.
To check your <n> commits in your git branch:
./scripts/checkpatch.pl --strict -g HEAD-<n>
We use standard GitHub mechanism for pull request. Please refer to GitHub documentation for help.
Subscribe to the OpenAMP mailing list ([email protected]).