On Wed, 16 Sept 2026 at 10:12, Piotr Kubaj <[email protected]> wrote:
>
> From: Piotr Kubaj <[email protected]>
>
> rkcommon_verify_header() and rkcommon_print_header() compare the first
> four bytes of the image against RK_MAGIC_V2 using a raw host-endian read:
>
> if ((*(uint32_t *)buf) == RK_MAGIC_V2)
>
> The v2 header is written little-endian (rkcommon_set_header0_v2() uses
> cpu_to_le32(RK_MAGIC_V2)) and rkcommon_parse_header() already reads it
> back with le32_to_cpu(). On a big-endian build host the two raw reads
> above never match, so a valid v2 image (e.g. for RK3568) is not
> recognised as v2: verification falls through to the v1 path and fails
> with "Error: image verification failed".
>
> Read the magic with le32_to_cpu() at both sites so mkimage works when
> built on a big-endian host. This is a no-op on little-endian hosts.
>
> Signed-off-by: Piotr Kubaj <[email protected]>
> ---
> tools/rkcommon.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/tools/rkcommon.c b/tools/rkcommon.c
> index b39777f..0e456ae 100644
> --- a/tools/rkcommon.c
> +++ b/tools/rkcommon.c
> @@ -473,7 +473,7 @@ int rkcommon_verify_header(unsigned char *buf, int size,
> int ret;
>
> /* spl_hdr is abandon on header_v2 */
> - if ((*(uint32_t *)buf) == RK_MAGIC_V2)
> + if (le32_to_cpu(*(uint32_t *)buf) == RK_MAGIC_V2)
If you would instead cpu_to_le32() the RK_MAGIC_V2 constant, then
you'd give the compiler a fighting chance to do the work at compile
time.
--Philipp.
> return 0;
>
> ret = rkcommon_parse_header(buf, &header0, &img_spl_info);
> @@ -508,7 +508,7 @@ void rkcommon_print_header(const void *buf, struct
> image_tool_params *params)
> uint8_t image_type;
> int ret, boot_size, init_size;
>
> - if ((*(uint32_t *)buf) == RK_MAGIC_V2) {
> + if (le32_to_cpu(*(uint32_t *)buf) == RK_MAGIC_V2) {
> ret = rkcommon_parse_header_v2(buf, &header0_v2);
>
> if (ret < 0) {
> --
> 2.52.0
>
