Some CPUID features may be provided by KVM for some guests, independent of
processor support, for example TSC deadline or TSC adjust. They are not going
to be present in named models unless the vendor implements them in hardware,
but they will be present in "-cpu host".
If these bits are not supported by the confidential computing firmware,
however, the guest will fail to start, and indeed this is a problem when
you run SNP guests with "-cpu host". This series fixes the issue.
However, I am marking this as RFC because it's not future proof.
If in the future AMD processors do provide any of these bits, this is
going to break (tsc_deadline and tsc_adjust are the most likely one).
Including the bits if they are present in host CPUID is not super safe
either, since the firmware might not be updated to follow suit.
Michael, any ideas? Is there a way for the host to retrieve the supported
CPUID bits for SEV-SNP guests?
One possibility is to set up a fake guest---either in QEMU or when KVM
starts---to do a LAUNCH_UPDATE for the CPUID page, but even that is not
perfect. For example, I got
> function 0x7, index: 0x0 provided: edx: 0xbc000010, expected: edx: 0x00000000
even though the FSRM bit (0x10) is supported. That might be just a
firmware bug however.
Paolo
Based-on: <20240627140628.1025317-1-pbonz...@redhat.com>
Paolo Bonzini (4):
target/i386: add support for masking CPUID features in confidential
guests
target/i386/SEV: implement mask_cpuid_features
target/i386/confidential-guest.h | 24 ++++++++++++++++++++++++
target/i386/cpu.c | 9 +++++++++
target/i386/cpu.h | 4 ++++
target/i386/kvm/kvm.c | 5 +++++
target/i386/sev.c | 33 +++++++++++++++++++++++++++++++++
5 files changed, 75 insertions(+)
--
2.45.2
