On 10/11/2023 17:45, Borislav Petkov wrote:
> On Fri, Nov 10, 2023 at 04:51:43PM +0100, Jeremi Piotrowski wrote:
>> What's semi-correct about checking for CC_VENDOR_INTEL and then
>> printing Intel? I can post a v2 that checks CC_ATTR_GUEST_MEM_ENCRYPT
>> before printing "TDX".
>
> How is it that you're not seeing the conflict:
>
> Your TD partitioning guest *is* a TDX guest so X86_FEATURE_TDX_GUEST
> should be set there. But it isn't. Which means, that is already wrong.
> Or insufficient.
>
> if (cc_vendor == CC_VENDOR_INTEL)
>
> just *happens* to work for your case.
>
> What the detection code should do, rather, is:
>
> if (guest type == TD partioning)
> set bla;
> else if (TDX_CPUID_LEAF_ID)
> "normal" TDX guest;
>
> and those rules need to be spelled out so that everyone is on the same
> page as to how a TD partitioning guest is detected, how a normal TDX
> guest is detected, a SEV-ES, a SNP one, yadda yadda.
>
>> The paravisor *is* telling the guest it is running on one - using a CPUID leaf
>> (HYPERV_CPUID_ISOLATION_CONFIG). A paravisor is a hypervisor for a confidential
>> guest, that's why paravisor detection shares logic with hypervisor detection.
>>
>> tdx_early_init() runs extremely early, way before hypervisor(/paravisor) detection.
>
> What?
>
> Why can't tdx_early_init() run CPUID(HYPERV_CPUID_ISOLATION_CONFIG) if
> it can't find a valid TDX_CPUID_LEAF_ID and set X86_FEATURE_TDX_GUEST
> then?
I guess it can if no one has an issue with it.
Thank you for the review, I've posted a patchset that implements this idea here:
https://lore.kernel.org/lkml/[email protected]/T/#u
>
>> Additionally we'd need to sprinkle paravisor checks along with
>> existing X86_FEATURE_TDX_GUEST checks. And any time someone adds a new
>> feature that depends solely on X86_FEATURE_TDX_GUEST we'd run the
>> chance of it breaking things.
>
> Well, before anything, you'd need to define what exactly the guest kernel
> needs to do when running as a TD partitioning guest and how exactly that
> is going to be detected and checked using the current cc_* and
> cpufeatures infra. If it doesn't work with the current scheme, then the
> current scheme should be extended.
>
> Then, that should be properly written out:
>
> "if bit X is set, then that is a guest type Y"
> "if feature foo present, then so and so are given"
>
> If the current guest type detection is insufficient, then that should be
> extended/amended.
>
> That's the only viable way where the kernel would support properly and
> reliably a given guest type. There'll be no sprinkling of checks
> anywhere.
>
> Thx.
>
OK. In the new submission I've added CC_ATTR_TDX_MODULE_CALLS because that
is what we really need to guard against, and these guests can then have
X86_FEATURE_TDX_GUEST set normally.
Jeremi