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

[Tom Lendacky]

On 6/3/21 1:15 PM, Borislav Petkov wrote:
> From f1e9f051c86b09fe660f49b0307bc7c6cec5e6f4 Mon Sep 17 00:00:00 2001
> From: Borislav Petkov <bp@xxxxxxx>
> Date: Thu, 3 Jun 2021 20:03:31 +0200
> Subject: Convert sme_active()
> 
>  	 */
> -	if (sme_active())
> +	if (protected_guest_has(VM_HOST_MEM_ENCRYPT))
>  		swiotlb = 1;

I still feel this is confusing. SME is a host/bare-metal technology, so
calling protected_guest_has() seems odd and using VM_HOST_MEM_ENCRYPT,
where I assume VM is short for virtual machine, also seems odd.

How about just protected_os_has()? Then you could have
- HOST_MEM_ENCRYPT  for host memory encryption
- GUEST_MEM_ENCRYPT for guest memory encryption
- MEM_ENCRYPT       for either host or guest memory encryption.

The first is analogous to sme_active(), the second to sev_active() and the
third to mem_encrypt_active(). Just my opinion, though...

>  
>  	return swiotlb;
> diff --git a/arch/x86/kernel/sev.c b/arch/x86/kernel/sev.c
> index 01a224fdb897..3aa2658ced52 100644
> --- a/arch/x86/kernel/sev.c
> +++ b/arch/x86/kernel/sev.c
> @@ -1409,6 +1409,11 @@ bool sev_protected_guest_has(unsigned long flag)
>  	case VM_MEM_ENCRYPT:
>  	case VM_MEM_ENCRYPT_ACTIVE:
>  		return true;
> +	case VM_HOST_MEM_ENCRYPT:
> +		return sme_me_mask && !sev_active();
> +	default:
> +		WARN_ON_ONCE(1);
> +		return false;

I don't think you want a WARN_ON_ONCE() here. The code will be written to
work with either SEV or TDX, so we shouldn't warn on a check for a TDX
supported feature when running on AMD (or vice-versa).

Thanks,
Tom