On Thu, Sep 18, 2025 at 07:06:15PM +0200, Pratyush Yadav wrote:
> kho_fill_kimage() only checks for KHO being enabled before filling in
> the FDT to the image. KHO being enabled does not mean that the kernel
> has data to hand over. That happens when KHO is finalized.
>
> When a kexec is done with KHO enabled but not finalized, the FDT page is
> allocated but not initialized. FDT initialization happens after
> finalize. This means the KHO segment is filled in but the FDT contains
> garbage data.
>
> This leads to the below error messages in the next kernel:
>
> [ 0.000000] KHO: setup: handover FDT (0x10116b000) is invalid: -9
> [ 0.000000] KHO: disabling KHO revival: -22
>
> There is no problem in practice, and the next kernel boots and works
> fine. But this still leads to misleading error messages and garbage
> being handed over.
>
> Only fill in KHO segment when KHO is finalized. When KHO is not enabled,
> the debugfs interface is not created and there is no way to finalize it
> anyway. So the check for kho_enable is not needed, and kho_out.finalize
> alone is enough.
>
> Fixes: 3bdecc3c93f9f ("kexec: add KHO support to kexec file loads")
> Signed-off-by: Pratyush Yadav <[email protected]>
Nice catch!
Reviewed-by: Mike Rapoport (Microsoft) <[email protected]>
> ---
> kernel/kexec_handover.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/kernel/kexec_handover.c b/kernel/kexec_handover.c
> index 8079fc4b9189..5ff959a90165 100644
> --- a/kernel/kexec_handover.c
> +++ b/kernel/kexec_handover.c
> @@ -1242,7 +1242,7 @@ int kho_fill_kimage(struct kimage *image)
> int err = 0;
> struct kexec_buf scratch;
>
> - if (!kho_enable)
> + if (!kho_out.finalized)
> return 0;
>
> image->kho.fdt = page_to_phys(kho_out.ser.fdt);
> --
> 2.47.3
>
--
Sincerely yours,
Mike.
