On Tue, 2024-11-26 at 13:25 -0300, Arnaldo Carvalho de Melo wrote:
[...]
Hi Arnaldo,
> I think I saw instructions in one of the messages in this thread to get
> hold of a vmlinux for s390 and test it. Right?
Yes, in cover letter. Full vmlinux is not needed, a vmlinux binary for
s390 would be sufficient for testing. Repeating the recipe for convenience:
To reproduce the bug:
- follow the instructions in [0] to build an s390 vmlinux;
- generate BTF requesting declaration tags for kfuncs:
$ pahole --btf_features_strict=decl_tag_kfuncs,decl_tag \
--btf_encode_detached=test.btf vmlinux
- observe that no kfuncs are generated:
$ bpftool btf dump file test.btf format c | grep __ksym
[0] https://docs.kernel.org/bpf/s390.html
> One extra question: this solves the BTF encoder case, the loader already
> supported loading BTF from a different endianness, right? Lemme
> check.
>
> cus__load_btf()
> cu->little_endian = btf__endianness(btf) == BTF_LITTLE_ENDIAN;
>
> enum btf_endianness btf__endianness(const struct btf *btf)
> {
> if (is_host_big_endian())
> return btf->swapped_endian ? BTF_LITTLE_ENDIAN : BTF_BIG_ENDIAN;
> else
> return btf->swapped_endian ? BTF_BIG_ENDIAN : BTF_LITTLE_ENDIAN;
> }
I can switch to is_host_big_endian() instead of `BYTE_ORDER` macro
if you think that's better.
> So we have parts of BTF byte swapping happening in libbpf and with this
> patch, parts of it done in pahole, have you tought about doing this in
> libbpf instead?
BTF id lists handling is currently not a part of libbpf.
Thanks,
Eduard
