Re: [RFC PATCH 06/15] virt: tdx-guest: Support TDI bind and unbind operations
From: Alexey Kardashevskiy
Date: Wed Nov 18 2026 - 06:12:31 EST
On 18/11/26 01:24, Jason Gunthorpe wrote:
On Tue, Nov 17, 2026 at 10:16:58PM +1100, Alexey Kardashevskiy wrote:
You should still clear the sDTE and that should be a non-failable
operation.
In AMD GHCB, the host forwards guest requests to the PSP, and the
host can screw this up. The guest still gets to know if the host
tried to meddle. I do not expect it to fail normally but I saw some
of this while debugging. Thanks,
Like I said, it is completely bad you don't have a fail safe way to
block DMA again.
Who does? I mean, 100% guaranteed to not have software bugs?
If the guest tries to do this operation and it fails
then it must panic.
Alright.
I still don't know how your "host controlls dma enable" works with anypossible security. Is crypto involved somehow?
The host forwards encrypted messages between the VM and the PSP, one of the messages is SDTE_WRITE (which is enable/disable *trusted* DMA, pretty much). The "trusted dma enable" bit lives in SDTE which is in the encrypted memory accessible by the PSP.
Jason
--
Alexey