On Wed, May 21, 2025 at 11:21 PM Tony Ambardar <tony.ambardar@xxxxxxxxx> wrote:
>
> Update btf_new_empty() to copy the pointer size from a provided base BTF.
> This ensures split BTF works properly and fixes test failures seen on
> 32-bit targets:
>
> root@qemu-armhf:/usr/libexec/kselftests-bpf# ./test_progs -a btf_split
> __test_btf_split:PASS:empty_main_btf 0 nsec
> __test_btf_split:PASS:main_ptr_sz 0 nsec
> __test_btf_split:PASS:empty_split_btf 0 nsec
> __test_btf_split:FAIL:inherit_ptr_sz unexpected inherit_ptr_sz: actual 4 != expected 8
> [...]
> #41/1 btf_split/single_split:FAIL
>
Hm... can you debug it a little bit, please? I see that
btf__pointer_size() on split BTF will do determine_ptr_size() call,
which will do
if (btf->base_btf && btf->base_btf->ptr_sz > 0)
return btf->base_btf->ptr_sz;
So it looks intentional (though I can't claim I remember much of the
details by now) that we don't proactively cache btf->ptr_sz when
creating a new split BTF, but it should have resolved into base's
pointer size. And if it doesn't, let's try to understand why?
> Fixes: ba451366bf44 ("libbpf: Implement basic split BTF support")
> Signed-off-by: Tony Ambardar <tony.ambardar@xxxxxxxxx>
> ---
> tools/lib/bpf/btf.c | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/tools/lib/bpf/btf.c b/tools/lib/bpf/btf.c
> index 8d0d0b645a75..b1977888b35e 100644
> --- a/tools/lib/bpf/btf.c
> +++ b/tools/lib/bpf/btf.c
> @@ -995,6 +995,7 @@ static struct btf *btf_new_empty(struct btf *base_btf)
>
> if (base_btf) {
> btf->base_btf = base_btf;
> + btf->ptr_sz = base_btf->ptr_sz;
> btf->start_id = btf__type_cnt(base_btf);
> btf->start_str_off = base_btf->hdr->str_len + base_btf->start_str_off;
> btf->swapped_endian = base_btf->swapped_endian;
> --
> 2.34.1
>
