How to generate `bpftool`? about libbpf-bootstrap HOT 3 CLOSED

No, not really. You generally need "recent enough" bpftool, but it will handle older kernels just fine. I suspect in this case bpftool built from 5.8 kernel sources just doesn't handle BTF_KIND_FLOAT type that was introduced after 5.8. If you build bpftool from 5.15 and run it against 5.8 kernel it should work fine.

But generally, bpftool version is independent. We now also have Github mirror for bpftool, so you can build it from there: https://github.com/libbpf/bpftool/

I want to stop including prebuilt version of bpftool (to allow non-x86-64 builds of libbpf-bootstrap) and instead build it from sources (using the above Github repo). Just like we are going to do it for libbpf-tools. See iovisor/bcc@03163a6 for reference. If you'd like to contribute, wiring up bpftool from Github repo would be a great help!

As for vmlinux.h, you generally don't have to re-generate it very often or from very recent kernel. It just depends on types you are expecting to program against. If some latest kernel has some field that you'd like to reference, it would be more convenient to generate vmlinux.h from most recent kernel. But often old kernel's vmlinux.h is totally fine even for much more recent kernels. Please refer to https://nakryiko.com/posts/bpf-core-reference-guide/#defining-own-co-re-relocatable-type-definitions (and one section after the above one) for some more understanding of the role of vmlinux.h and how you can "complement" it with your own type definitions.

from libbpf-bootstrap.

I am very much open to contribute. From what I understand by making https://github.com/libbpf/bpftool/ a submodule, when bpftool is being built it will use libraries that are specific to an architecture so you wouldn't have to include multiple binaries in the version control? So, the steps are to exactly replicate how iovisor/bcc@03163a6 is being built.
Thanks for response!

from libbpf-bootstrap.

yes, exactly

from libbpf-bootstrap.