On Fri, Feb 5, 2021 at 5:42 AM Giuliano Procida <gprocida@xxxxxxxxxx> wrote:
>
> This is to avoid misaligned access to BTF type structs when
> memory-mapping ELF objects.
>
> Signed-off-by: Giuliano Procida <gprocida@xxxxxxxxxx>
> ---
I trust you did verify that it actually works in cases where
previously .BTF was mis-aligned?
Acked-by: Andrii Nakryiko <andrii@xxxxxxxxxx>
> libbtf.c | 9 +++++++++
> 1 file changed, 9 insertions(+)
>
> diff --git a/libbtf.c b/libbtf.c
> index 9f4abb3..6754a17 100644
> --- a/libbtf.c
> +++ b/libbtf.c
> @@ -744,6 +744,14 @@ static int btf_elf__write(const char *filename, struct btf *btf)
> goto out;
> }
>
> + /*
> + * We'll align .BTF to 8 bytes to cater for all architectures. It'd be
> + * nice if we could fetch this value from somewhere. The BTF
> + * specification does not discuss alignment and its trailing string
> + * table is not currently padded to any particular alignment.
> + */
> + const size_t btf_alignment = 8;
> +
> /*
> * First we check if there is already a .BTF section present.
> */
> @@ -821,6 +829,7 @@ static int btf_elf__write(const char *filename, struct btf *btf)
> elf_error("elf_getshdr(btf_scn) failed");
> goto out;
> }
> + btf_shdr.sh_addralign = btf_alignment;
> btf_shdr.sh_entsize = 0;
> btf_shdr.sh_flags = 0;
> if (dot_btf_offset)
> --
> 2.30.0.478.g8a0d178c01-goog
>
