Re: [PATCH] x86/coco, x86/sev: Use cpu_feature_enabled() to detect SEV guest flavor

[Jeremi Piotrowski <jpiotrowski@linux.microsoft.com>]

On 05/12/2023 17:00, Borislav Petkov wrote:
> On Tue, Dec 05, 2023 at 06:14:37PM +0300, Kirill A. Shutemov wrote:
>> My point is that if you need to check for SEV you need to check SEV, not
>> CC_ATTR. CC_ATTRs only make sense in generic code that deals with multiple
>> CoCo environments.
> 
> That makes more sense.

So given this series, what is the canonical way to expose sub-features of
TDX/SNP going forward? X86_FEATURE_xxx for every one that needs to be queried
in TDX/SNP specific code?

I see that in [1] X86_FEATURE_SNP_SECURE_TSC is being proposed. How about the
CC_ATTR_TDX_MODULE_CALLS (perhaps better called TDX_TDCALL or something) that
I'm proposing in the other thread [2]? VTOM? SVSM?

We can also export amd_cc_platform_has() and intel_cc_platform_has() for such cases.
But we really need is to agree which to use when (X86_FEATURE vs CC_ATTR).

[1]: https://lore.kernel.org/lkml/20231128125959.1810039-10-nikunj@amd.com/
[2]: https://lore.kernel.org/lkml/20231122170106.270266-2-jpiotrowski@linux.microsoft.com/

Jeremi

> 
> So that commit already says "If future support is added for other
> memory encryption technologies, the use of CC_ATTR_GUEST_MEM_ENCRYPT
> can be updated, as required."
> 
> And what this test needs to do is to check:
> 
> 	if (guest type >= SEV)
> 
> meaning SEV and -ES and -SNP.
> 
> I'm wondering if we should export amd_cc_platform_has() for such
> cases...
> 
> The logic being, we're calling a SEV-specific function so using
> cc_platform_has() in there is the wrong layer.
> 
> Tom?
>