Re: [PATCH v3 3/4] crypto: ccp - Add an API to return the supported SEV-SNP policy bits

[Tom Lendacky <thomas.lendacky@amd.com>]

On 10/23/25 11:37, Sean Christopherson wrote:
> On Wed, Oct 22, 2025, Tom Lendacky wrote:
>> Supported policy bits are dependent on the level of SEV firmware that is
>> currently running. Create an API to return the supported policy bits for
>> a given level of firmware. KVM will AND that value with the KVM supported
> 
> Given "KVM will AND" and the shortlog, I expected a _future_ patch to have the 
>            ^^^^
> KVM changes.
> 
> That's partly a PEBKAC on my end (I mean, it's literally the first diff), but I
> do think it highlights that we should probably separate the KVM change from the
> PSP support. 
> 
> Hmm, actually, the patch ordering is bad.  There shouldn't need to be a separate
> KVM change after this commit, because as things are ordered now, there will be an
> ABI change between patch 1 and this patch.
> 
> So I think what you want is:
> 
>   1. KVM: SEV: Consolidate the SEV policy bits in a single header file
>   2. crypto: ccp - Add an API to return the supported SEV-SNP policy bits
>   3. KVM: SEV: Publish supported SEV-SNP policy bits
>   4. KVM: SEV: Add known supported SEV-SNP policy bits
> 
> where #3 uses sev_get_snp_policy_bits() straightaway.

Ok, let me rearrange things, resend, and see how it looks.

> 
>> policy bits to generate the actual supported policy bits.
>>
>> Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
>> ---
>>  arch/x86/kvm/svm/sev.c       |  3 ++-
>>  drivers/crypto/ccp/sev-dev.c | 37 ++++++++++++++++++++++++++++++++++++
>>  include/linux/psp-sev.h      | 20 +++++++++++++++++++
>>  3 files changed, 59 insertions(+), 1 deletion(-)
>>
>> diff --git a/arch/x86/kvm/svm/sev.c b/arch/x86/kvm/svm/sev.c
>> index 45e87d756e15..24167178bf05 100644
>> --- a/arch/x86/kvm/svm/sev.c
>> +++ b/arch/x86/kvm/svm/sev.c
>> @@ -3099,7 +3099,8 @@ void __init sev_hardware_setup(void)
>>  			sev_snp_supported = is_sev_snp_initialized();
>>  
>>  		if (sev_snp_supported) {
>> -			snp_supported_policy_bits = KVM_SNP_POLICY_MASK_VALID;
>> +			snp_supported_policy_bits = sev_get_snp_policy_bits();
>> +			snp_supported_policy_bits &= KVM_SNP_POLICY_MASK_VALID;
> 
> I vote for:
> 
> 			snp_supported_policy_bits = sev_get_snp_policy_bits() &
> 						    KVM_SNP_POLICY_MASK_VALID;
> 
> which makes it visually easier to see the policy bits logic.

Will do.

Thanks,
Tom

> 	
>>  			nr_ciphertext_hiding_asids = init_args.max_snp_asid;
>>  		}