Re: [PATCH v5 3/3] x86/tdx: Add Quote generation support

[Kirill A. Shutemov]

On Mon, May 09, 2022 at 07:14:20AM -0700, Dave Hansen wrote:
> On 5/9/22 05:09, Kirill A. Shutemov wrote:
> > Private->Shared conversion is destructive. You have to split SEPT, flush
> > TLB. Backward conversion even more costly.
> > 
> > Rule of thumb is avoid conversion where possible. DMA API is there for
> > you.
> 
> Kirill, I understand that the DMA API is a quick fix today.  But is it
> _really_ the right long-term interface?

Yes, I think so.

> There will surely come a time when TDX I/O devices won't be using fixed
> bounce buffers.  What will the quote generation code do then?  How will
> we know to come back around and fix this up?

VirtIO will not go away with TDX I/O in picture. TDX I/O will be addition
to existing stuff, not replacement.

And we have hooks in place to accommodate this: force_dma_unencrypted()
will return false for devices capable of TDX I/O. While the rest of
devices, including VirtIO and attestation, keep using existing paths with
swiotlb.

> Does SEV or the s390 ultravisor need anything like this?

At quick glance sev-guest.c uses set_memory_decrypted()/encrypted() for
allocation and freeing shared memory. I consider it inferior to using DMA
API.

-- 
 Kirill A. Shutemov