Re: [PATCH v2 1/5] kho: Do not translate the handover FDT address in is_kho_boot()

[Mike Rapoport <rppt@kernel.org>]

Hi Chen,

On Tue, Nov 03, 2026 at 04:09:22PM +0800, Chen Pei wrote:
> is_kho_boot() only has to say whether a handover FDT was recorded, but
> it asks kho_get_fdt() for one and so converts kho_in.fdt_phys with
> phys_to_virt() just to throw the pointer away. Both callers use it as a
> boolean and never dereference the result.
> 
> On riscv64 that conversion is not valid yet when efi_init() reaches it:
> setup_vm() leaves kernel_map.va_pa_offset at 0 and setup_bootmem()
> assigns it later, so a handover boot that came through UEFI panics in
> linear_mapping_pa_to_va(). Only debug builds fail -- without
> CONFIG_DEBUG_VIRTUAL __va() degenerates to "phys + 0" and the boolean
> stays correct -- but the panic also goes unnoticed without an earlycon.
> 
> Check the physical address instead. Nothing needs the translation, and
> architectures stay free to bring their linear mapping up whenever they
> like.
> 
> Fixes: d6d511639185 ("kexec: introduce is_kho_boot()")
> Signed-off-by: Chen Pei <cp0613@linux.alibaba.com>
> ---

There was very similar patch: 

https://lore.kernel.org/all/20261023054518.3258350-1-gthelen@google.com

and while both Greg's and your patch are correct they both have that wall
of text with too many details commit messages that I don't like.
 
> Tested on qemu-system-riscv64 (virt, -cpu max, TCG): edk2
> RISCV_VIRT_CODE.fd/VARS.fd on pflash with OpenSBI fw_dynamic.bin below
> it, and a GPT disk holding an ESP with grub plus the kernel and an ext4
> partition as rootfs.  Kernel config is riscv defconfig plus
> DEBUG_KERNEL, DEBUG_VIRTUAL, KEXEC_FILE, KEXEC_HANDOVER and the kho
> selftest fragment; the outgoing kernel gets "kho=on" and the kho
> selftest init does kexec_file_load() followed by reboot(RB_KEXEC).
> 
> Before, the first handover panics.  earlycon= is needed to see it: the
> BUG precedes console registration, so without an earlycon the machine
> goes silent right after kexec.
> 
>   KHO: found kexec handover data.
>   efi: EFI v2.7 by EDK II
>   kernel BUG at arch/riscv/mm/physaddr.c:47!
>   epc : linear_mapping_pa_to_va+0x1c/0x1e
>    ra : is_kho_boot+0x18/0x28
> 
> After, the same setup keeps handing over until the window closes, 29
> consecutive boots and no BUG.
> 
>   KHO: found kexec handover data.
>   KHO: exec from: 7.4.0-rc1-next-20261102-00005-gfca49e548231 (count 29)
> 
>  kernel/liveupdate/kexec_handover.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/kernel/liveupdate/kexec_handover.c b/kernel/liveupdate/kexec_handover.c
> index 6f1f55fe3181..6ab181de345c 100644
> --- a/kernel/liveupdate/kexec_handover.c
> +++ b/kernel/liveupdate/kexec_handover.c
> @@ -1643,7 +1643,7 @@ EXPORT_SYMBOL_GPL(kho_restore_free);
>   */
>  bool is_kho_boot(void)
>  {
> -	return !!kho_get_fdt();
> +	return kho_in.fdt_phys != 0;
>  }
>  EXPORT_SYMBOL_GPL(is_kho_boot);
>  
> -- 
> 2.50.1
> 

-- 
Sincerely yours,
Mike.