On Fri, 2024-09-20 at 00:49 -0700, Tony Ambardar wrote:
> Initially, the .BTF_ids section was created zero-filled and then patched
> with BTF IDs by resolve_btfids on the build host. Patching was done in
> native endianness and thus failed to work for cross-endian compile targets.
> This was fixed in [1] by using libelf-based translation to output patched
> data in target byte order.
>
> The addition of 8-byte BTF sets in [2] lead to .BTF_ids creation with both
> target-endian values and zero-filled data to be later patched. This again
> broke cross-endian compilation as the already-correct target-endian values
> were translated on output by libelf [1]. The problem was worked around [3]
> by manually converting BTF SET8 values to native endianness, so that final
> libelf output translation yields data in target byte order.
>
> Simplify and make the code more robust against future changes like [2] by
> employing libelf-based endian translation on both input and output, which
> is typical of libelf usage.
>
> [1]: 61e8aeda9398 ("bpf: Fix libelf endian handling in resolv_btfids")
> [2]: ef2c6f370a63 ("tools/resolve_btfids: Add support for 8-byte BTF sets")
> [3]: 903fad439466 ("tools/resolve_btfids: Fix cross-compilation to non-host endianness")
>
> CC: Viktor Malik <vmalik@xxxxxxxxxx>
> Signed-off-by: Tony Ambardar <tony.ambardar@xxxxxxxxx>
> ---
Acked-by: Eduard Zingerman <eddyz87@xxxxxxxxx>
[...]
