Re: [RFC v2-fix-v2 1/1] x86: Introduce generic protected guest abstractionn

[Kuppuswamy, Sathyanarayanan]

On 6/7/21 1:14 PM, Borislav Petkov wrote:
> On Mon, Jun 07, 2021 at 10:55:44PM +0300, Kirill A. Shutemov wrote:
> > I think conversions like this are wrong: relocate_kernel(), which got
> > called here, only knows how to deal with SME, not how to handle some
> > generic case.
>
> What do you mean wrong? Wrong for TDX?
>
> If so, then that can be
>
> protected_guest_has(SME)
>
> or so, which would be false on Intel.

I agree. Since most of the code changed in this patch is
not applicable to TDX, it might need product specific or
new function specific flags.

> And this patch was only a mechanical conversion to see how it would look
> like.
>
> > If code is written to handle a specific technology we need to stick
> > with a check that makes it clear. Trying to make sound generic only
> > leads to confusion.
>
> Sure, fine by me.
>
> And I don't want a zoo of gazillion small checking functions per
> technology. sev_<something>, tdx_<something>, yadda yadda.
>
> So stuff better be unified. Even if you'd have vendor-specific defines
> you hand into that function - and you will have such - it is still much
> saner than what it turns into with the AMD side of things.

Agree. Currently we share code with AMD SEV in memory encryption support and
string I/O handling code. So defining common flag for such code is
useful.

>

--
Sathyanarayanan Kuppuswamy
Linux Kernel Developer