Re: [PATCH v4 5/7] x86/sev: Register the guest with the SVSM APIC protocol

[Ard Biesheuvel]

On Thu, 15 Oct 2026, at 05:56, Melody Wang wrote:
> Hi Ard,
>
> On 10/14/26 2:43 AM, Ard Biesheuvel wrote:
>> I'm struggling to follow the logic here: doesn't this attempt to locate
>> the protocol on non-SEV/SNP systems as well?
>
> Alternate Injection is only working for SEV-SNP guest.
>
>> Wouldn't it be sufficient to do
>> 
>> if (!snp_vmpl ||
>>     !(sev_get_status() & MSR_AMD64_SNP_ALTERNATE_INJ))
>>         return 0;
>> 
>> and leave the early_is_sevsnp_guest() check alone?
>> 
>
> The thing is, snp_vmpl hasn't been initialized here yet. We read here a 
> 0. So I need early_is_sevsnp_guest() to initialize it to U8_MAX so that 
> I know that there's an SVSM present.
>
> Then I can enable AltInj.
>

OK.

But there are three conditions where there is no point in checking
whether the protocol exists:
- early_is_sevsnp_guest() -> false
- snp_vmpl == 0
- (sev_get_status() & MSR_AMD64_SNP_ALTERNATE_INJ) == 0

So this should be

>> if (!early_is_sevsnp_guest() || !snp_vmpl ||
>>     !(sev_get_status() & MSR_AMD64_SNP_ALTERNATE_INJ))
>>         return 0;

then?

Btw please don't use bare '1' and '2' as error codes, especially
since you don't actually check the value - just use a bool instead.