Re: [PATCH 6/6] KVM: nVMX: support RDRAND and RDSEED exiting

From: Paolo Bonzini
Date: Tue Mar 14 2017 - 05:56:05 EST

On 13/03/2017 19:27, Jim Mattson wrote:
> I would expect that any reasonable CPU will support "RDRAND exiting"
> iff it supports RDRAND (i.e. CPUID.01H:ECX.RDRAND[bit 30]).
> Similarly, any reasonable CPU will support "RDSEED exiting" iff it
> supports RDSEED (i.e. CPUID.(EAX=07H, ECX=0H):EBX.RDSEED[bit 18]).
> Shouldn't there be some code in vmx_cpuid_update to adjust the
> vmx->nested.nested_vmx_secondary_ctls_high bits for "RDRAND exiting"
> and "RDSEED exiting," as there already is for "enable RDTSCP"?

Enable RDTSCP is a bit different in that the instruction itself is
disabled rather than the VMEXIT, still it would probably be a good idea.

One thing to be careful about is any conflicts with David Matlack's work
to customize the MSRs for processor models, since there's no prescribed
ordering of KVM_SET_CPUID2 and KVM_SET_MSR. Maybe
nested_vmx_setup_ctls_msrs can save the "default" nested_*_ctls_*; then
vmx_restore_control_msr can refer to those in order to check the written
value against the supported bits.