On Thu, Jul 22, 2021 at 10:31:27AM -0700, Marc Orr wrote:

IMHO, we need to be completely certain that guest data cannot be
compromised if we're going to remove the requirement that guest memory
only be validated once in a certain state (e.g., from within a crash
kernel). Perhaps it is the case that we're certain that guest data
cannot be compromised from within a crash kernel -- but it's not what
I read in the email exchange.

Right, at least SNP has a strict requirement that no memory could be
validated or invalidated twice without giving up security guarantees for
that memory.

Regards,

	Jörg