Re: [PATCH v3 6/8] x86/sev: Register the guest with the SVSM APIC protocol

[Melody Wang <huibo.wang@amd.com>]

Hi all,

On 9/20/26 11:16 AM, Melody Wang wrote:
> diff --git a/drivers/firmware/efi/libstub/x86-stub.c b/drivers/firmware/efi/libstub/x86-stub.c
> index 49388524a61f..35b8641119b9 100644
> --- a/drivers/firmware/efi/libstub/x86-stub.c
> +++ b/drivers/firmware/efi/libstub/x86-stub.c
> @@ -793,6 +793,9 @@ static bool sev_prepare(void)
>   			unsupported);
>   		return true;
>   	}
> +
> +	svsm_register_alt_inj();
> +

Sashiko said this:

"Does this code handle non-EFI direct kernel boots correctly?
When an SEV-SNP guest is booted directly and bypasses the EFI stub (such as
when using QEMU's -kernel with a setup_data CC blob), sev_prepare() is
skipped entirely.
This completely bypasses the svsm_register_alt_inj() call.
The main kernel checks for Alternate Injection support via
cc_platform_has(CC_ATTR_SNP_ALTERNATE_INJECTION). This check relies
strictly on the sev_status MSR hardware bit set by the SVSM and ignores
whether the guest successfully registered the APIC emulation.
Could this cause the APIC probe to issue SVSM APIC calls via
__svsm_apic_msr_rw() that are rejected by the SVSM, leading to an
unconditional crash via sev_es_terminate() early in the boot sequence?"

Alternate Injection does not work without OVMF support so no need to 
handle the non-EFI direct kernel boots.

Thanks,
Melody
>   	return false;
>   }
>   

-- 
Thanks,
Melody