Re: [PATCH 4/8][KVM nVMX]: Check "load IA32_PERF_GLOBAL_CTRL" VM-exit control on vmentry of nested guests

[Jim Mattson <jmattson@google.com>]

On Fri, May 17, 2019 at 1:34 PM Sean Christopherson
<sean.j.christopherson@intel.com> wrote:
>
> On Thu, May 16, 2019 at 03:07:48PM -0700, Krish Sadhukhan wrote:
> >
> >
> > On 05/13/2019 12:00 PM, Sean Christopherson wrote:
> > >On Wed, Apr 24, 2019 at 07:17:20PM -0400, Krish Sadhukhan wrote:
> > >>According to section "Checks on Host Control Registers and MSRs" in Intel
> > >>SDM vol 3C, the following check is performed on vmentry of nested guests:
> > >>
> > >>     "If the "load IA32_PERF_GLOBAL_CTRL" VM-exit control is 1, bits reserved
> > >>     in the IA32_PERF_GLOBAL_CTRL MSR must be 0 in the field for that
> > >>     register."
> > >>
> > >>Signed-off-by: Krish Sadhukhan <krish.sadhukhan@oracle.com>
> > >>Reviewed-by: Karl Heubaum <karl.heubaum@oracle.com>
> > >>---
> > >>  arch/x86/kvm/vmx/nested.c | 5 +++++
> > >>  1 file changed, 5 insertions(+)
> > >>
> > >>diff --git a/arch/x86/kvm/vmx/nested.c b/arch/x86/kvm/vmx/nested.c
> > >>index 83cd887638cb..d2067370e288 100644
> > >>--- a/arch/x86/kvm/vmx/nested.c
> > >>+++ b/arch/x86/kvm/vmx/nested.c
> > >>@@ -2595,6 +2595,11 @@ static int nested_check_host_control_regs(struct kvm_vcpu *vcpu,
> > >>        !nested_host_cr4_valid(vcpu, vmcs12->host_cr4) ||
> > >>        !nested_cr3_valid(vcpu, vmcs12->host_cr3))
> > >>            return -EINVAL;
> > >>+
> > >>+   if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL &&
> > >>+      !kvm_valid_perf_global_ctrl(vmcs12->host_ia32_perf_global_ctrl))
> > >If vmcs12->host_ia32_perf_global_ctrl were ever actually consumed, this
> > >needs to ensure L1 isn't able to take control of counters that are owned
> > >by the host.
> >
> > Sorry, I didn't understand your concern. Could you please explain how L1 can
> > control L0's counters ?
>
> MSR_CORE_PERF_GLOBAL_CTRL isn't virtualized in the sense that there is
> only one MSR in hardware (per logical CPU).  Loading an arbitrary value
> into hardware on a nested VM-Exit via vmcs12->host_ia32_perf_global_ctrl
> means L1 could toggle counters on/off without L0's knowledge.  See
> intel_guest_get_msrs() and intel_pmu_set_msr().
>
> Except that your patches don't actually do anything functional with
> vmcs12->host_ia32_perf_global_ctrl, hence my confusion over what you're
> trying to accomplish.  If KVM advertises VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL
> to L1, then L1 will think it can use the VMCS field to handle
> MSR_CORE_PERF_GLOBAL_CTRL when running L2, e.g. instead of adding
> MSR_CORE_PERF_GLOBAL_CTRL to the MSR load lists, which will break L1
> (assuming nested PMU works today).

I think nested PMU should work today if the vmcs12 VM-entry and
VM-exit MSR-load lists are used to swap IA32_PERF_GLOBAL_CTRL values
between L1 and L2. Note that atomic_switch_perf_msrs() is called on
every vmx_vcpu_run(), for vmcs02 as well as vmcs01.

This change set should aim to provide equivalent functionality with
the new(er) VM-entry and VM-exit controls for "load
IA32_PERF_GLOBAL_CTRL," so that L1 doesn't have to resort to the
MSR-load lists.