Re: [PATCH v2 6/8] x86/sev: Register the guest with the SVSM APIC protocol

[Ard Biesheuvel]

On Wed, 16 Sep 2026, at 20:45, Melody Wang wrote:
> Hi Ard,
>
> On 9/15/26 12:44 AM, Ard Biesheuvel wrote:
>> So please add it to the EFI stub, and code it against the boot services
>> (efi_err() for error messages, efi_exit() when aborting and returning to
>> the firmware, etc). Implementing this in a hybrid manner seems entirely
>> unnecessary.
>
> I moved sev_prepare() into the EFI stub, and I wrote a wrapper for 
> registering Alternate Injection in the EFI stub but kept the code in 
> compressed, otherwise I have to export early_is_sevsnp_guest(), 
> boot_svsm_caa_pa, and svsm_call_msr_protocol() so that it builds.
>

early_is_sevsnp_guest() needs to be exported anyway, as I am doing in [0]

> How about this below:
>
> This is in the stub:
>
> static bool sev_prepare(void)
> {
>          u64 unsupported;
>
>          unsupported = snp_get_unsupported_features(sev_get_status());
>          if (unsupported) {
>                  efi_err("Unsupported SEV-SNP features detected: 0x%llx\n",
>                          unsupported);
>                  return true;
>          }
>
>          register_alternate_injection();
>
>          return false;
> }
>
> and the stub calls this function:
>
> +void register_alternate_injection(void)
> +{
> +       if (early_is_sevsnp_guest() && snp_vmpl) {
> +               struct svsm_call call = {};
> +               int ret;
> +
> +               if (!(sev_get_status() & MSR_AMD64_SNP_ALTERNATE_INJ))
> +                       return;
> +
> +               call.caa = (struct svsm_ca *)boot_svsm_caa_pa;
> +               call.rax = SVSM_APIC_CALL(SVSM_APIC_CONFIG_EMULATION);
> +               call.rcx = SVSM_AI_REGISTER;
> +
> +               ret = svsm_call_msr_protocol(&call);
> +               if (ret)
> +                       sev_es_terminate(SEV_TERM_SET_LINUX, 
> GHCB_TERM_ALT_INJ_FAIL);
> +       }
> +}
>
> which is kept in sev.c to avoid cross-dependencies.
>

Yes, that seems fine. The important thing is that the SEV support
code that is shared between the EFI stub, the legacy decompressor
and the kernel proper has no dependencies on the underlying
execution context, e.g., EFI boot services or decompressor APIs.

> I can send it as patches too if it is easier to review this way.
>

Naturally.

[0] https://lore.kernel.org/all/20260914183745.37538-5-ardb@xxxxxxxxxx/