Porting px4 to the evl4 kernel of aarch64 architecture

[yexuanyang]

Hello everyone! I am trying to port px4 to evl4 kernel. Now it can run with PX4_SIM_MODEL=shell environment variable set, I will add some benchmarks to measure the jitter in Tasks and WorkItems in the future.

The following are the operations that can be run on my machine:

• Create a cmake build directory:

  /usr/bin/cmake -DCMAKE_BUILD_TYPE:STRING=RelWithDebInfo -DCONFIG:STRING=px4_sitl_evl4 -DCMAKE_EXPORT_COMPILE_COMMANDS:BOOL=TRUE -DCMAKE_C_COMPILER:FILEPATH=/usr/bin/aarch64-linux-gnu-gcc -DCMAKE_CXX_COMPILER:FILEPATH=/usr/bin/aarch64-linux-gnu-g++ --no-warn-unused-cli -S/root/PX4-Autopilot -B/root/PX4-Autopilot/build/px4_sitl_evl4 -G Ninja

• cmake starts building:

  /usr/bin/cmake --build /root/PX4-Autopilot/build/px4_sitl_evl4 --config RelWithDebInfo --target all --

• Run px4:

  cd /root/PX4-Autopilot/build/px4_sitl_evl4/rootfs
  export PX4_SIM_MODEL=shell
  ../bin/px4

[ruiqurm]

Awesome job on this PR! I’ve left a few comments with some questions.
One thing I noticed is about naming — currently, nullptr is passed to all named elements like mutex, thread, etc. Would it be possible to assign meaningful names to these elements? That would make debugging easier.
No rush though — this can definitely be addressed in a future update.

[ruiqurm]

Note that evl_usleep cannot sleep for durations longer than one second. You may want to try evl_sleep_until instead.(I am not sure the evl_sleep_until if the API can be used to sleep for more than one second either)

[yexuanyang]

OK, I will change it to a loop using system_usleep.

[ruiqurm]

Could you please check the purpose of this macro? Should we use a real-time sleep function here, or is it acceptable to use a standard sleep instead?

[yexuanyang]

px4_sleep use the system_sleep when lockstepscheduler doesn't set absolute time, it is the only place calling system_sleep, but px4_sleep is commonly used. I think it is only use when lockstepscheduler is not start. But it will use in drivers while loop, like gps driver. GPS driver use px4_sleep in GPS::run().
I use a real-time sleep here, just in case.

[ruiqurm]

So the most use case is in debuging? I think just use usleep is ok.

[yexuanyang]

Awesome job on this PR! I’ve left a few comments with some questions. One thing I noticed is about naming — currently, nullptr is passed to all named elements like mutex, thread, etc. Would it be possible to assign meaningful names to these elements? That would make debugging easier. No rush though — this can definitely be addressed in a future update.

Well, when the mutex has a name, it must be an unique name. Obviously, it is easy to think about using getpid() to get so called unique name, but, some threads need to create more than one mutex or event, so the name may not be unique.
I just pick nullptr to avoid the problem, if in some scenes need to debug with these name, I will try to find a solution to give them an suffix UUID with meaningful prefix string.