Re: [PATCH v5 04/12] x86/tdx: Add protected guest support for TDX guest

[Sean Christopherson <seanjc@google.com>]

On Wed, Aug 04, 2021, Kuppuswamy, Sathyanarayanan wrote:
> 
> On 8/4/21 3:03 PM, Dave Hansen wrote:
> > > > +#include <asm/processor.h>
> > > > +#include <asm/tdx.h>
> > > > +
> > > >   #ifndef __ASSEMBLY__
> > > >   static inline bool prot_guest_has(unsigned int attr)
> > > >   {
> > > >   	if (sme_me_mask)
> > > >   		return amd_prot_guest_has(attr);
> > > > +	else if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL)
> > > Why not "boot_cpu_has(X86_FEATURE_TDX_GUEST)"?
> > Even better: cpu_feature_enabled(X86_FEATURE_TDX_GUEST).  That gets you
> > both static patching*and*  compile-time optimization if you hook

Ah, I keep forgetting it can be compiled out.

> > X86_FEATURE_TDX_GUEST into disabled-features.h.
> 
> This is how Borislav preferred it. tdx_prot_guest_has() internally uses
> cpu_feature_enabled(X86_FEATURE_TDX_GUEST) to return the status.
> 
> I think the intention is to keep the first call generic (non TDX
> specific). So that it can be extended for other use cases.

What other possible use case is there for invoking tdx_prot_guest_has() beyond
running as a TDX guest?  If it were e.g. intel_prot_guest_has() then I would at
least understand the code, if not agree with the sub-optimal approach, but as is
it makes no sense.

Given amd_prot_guest_has(), my guess is Boris intended intel_prot_guest_has()...