drivers: can: implement native Linux SocketCAN driver

[martinjaeger]

Background Information

This is a new attempt to add a CAN driver for the native_posix board to interface with a Linux host system using the SocketCAN API. The driver itself is called native_linux, as it supports only Linux hosts and no other POSIX-based systems.

A similar driver was introduced in #37124. However, the CAN implementation in Zephyr changed significantly since that PR, so the driver was pretty much rewritten from scratch. Comments by @gmarull and @alexanderwachter in previous PR are addressed.

Details about the Linux SocketCAN implementation can be found in the Linux kernel CAN documentation and API header file.

Testing

In order to test the driver, a virtual CAN interface has to be set up on the Linux host:

sudo ip link add dev vcan0 type vcan
sudo ip link set up vcan0
sudo ip link property add dev vcan0 altname zcan0

In addition to that, the following overlay (or update of native_posix.dts) is required:

/ {
	chosen {
		zephyr,canbus = &can0;
	};
};

&can0 {
	status = "okay";
};

Afterwards, the driver passes all CAN driver tests:

scripts/twister -p native_posix -T tests/drivers/can

The driver does currently not pass the ISO-TP tests in tests/subsys/canbus/isotp. The ISO-TP tests are known to be brittle (e.g. they also fail for STM32G4, see #47014) and there is most probably a timing issue in the ISO-TP implementation itself. Fixing that is not within the scope of this PR.

[henrikbrixandersen]

Please see my comment about reworking the binding to the host CAN interface name. You could submit the two first commits in separate PRs and get them approved quickly in order to cut down on the changes here.

[martinjaeger]

The first commit will be removed once #49907 is merged.

[jukkar]

Nice work! Couple of years ago I had some plans to implement this but got side tracked and plans changed.
Anyway, there is a helper script at zephyrproject-rtos/net-tools#26 that I created at that point. It tries to setup canbus things in linux side, similar ways as the net-setup.sh script is doing in networking. Not sure if that script contains anything useful but feel free to re-use it if needed.

[henrikbrixandersen]

Looks good. One minor change proposal as discussed on Discord:

[henrikbrixandersen]

Thanks!

[str4t0m]

Thanks for this addition! I am not very familiar with the socket can api on Linux, but still have a few comments.

[martinjaeger]

Thanks @str4t0m for your very thorough review. I think all your points are valid and I'll come up with a solution latest tomorrow.

[martinjaeger]

The remarks by @str4t0m are now considered. For easier review you can take a look at the changes which were squashed into the commit in this PR here: eef547f

I have also reduced the sleep time in the RX thread from 10ms to 1ms so that we can receive frames with less latency.

[str4t0m]

two more comments.