On Mon, Feb 8, 2021 at 4:22 PM Matteo Croce [off-list ref] wrote:

On Mon, Feb 8, 2021 at 9:13 PM Bruce Ashfield [off-list ref] wrote:

quoted

On Mon, Feb 8, 2021 at 1:18 PM Matteo Croce [off-list ref] wrote:

quoted

From: Matteo Croce <redacted>

Add a recipe to build libbpf from https://github.com/libbpf/libbpf
The only patch fixes a build issue, and it's already merged upstream.

Thanks for the submission! I have a few comments / questions.

To get this into oe-core, we should be commenting / documenting why it
should be in core, versus another layer. The standard criteria is that
there are enough varied users and that the functionality is common
enough, that it belongs in core.

For sure bcc and bpftrace can use it, and maybe also perf.
In future even iproute2 will use it, as it has been ported to libbpf recently.
Feel free to propose another layer, in case you know a better one.

That's not for me to propose . that's for you to sort out.

A quick check of the layer index would show that other bpf tools are
in meta-oe.

quoted

There should also be some sort of oe-selftest for the functionality,
otherwise, it is hard to detect breakages. Some sort of application
that uses the library and that can be executed in qemu would be
enough.

That's doable.

quoted

What are the kernel requirements ? CONFIG_BPF is selected by other
kernel configs (it has no menu entry, so it must be), is it that, or
something else that is the requirement (classic BFP?). If that option
is now always on, is that true for the reference kernel versions in
master (5.4 and 5.10).

I'd say BPF_SYSCALL, which is the single entry point for al the eBPF routines.

Yes, that's the core support, and a selftest would ensure that the
reference kernels can support the package (they can, but we still need
the test) and implicitly document that requirement.

quoted

Finally, does this work across all the supported architectures ? if
not, we'll need compatibility settings.

I tested it only x86 and aarch64, but it should be arch independent.

Then it should be limited to where it has been tested, otherwise, the
burden falls to the oe-core maintainer, which we don't want.

Bruce

Thanks!
--
per aspera ad upstream

-- 
- Thou shalt not follow the NULL pointer, for chaos and madness await
thee at its end
- "Use the force Harry" - Gandalf, Star Trek II