KVM: VMX: Mark VMXArea with revision_id of physical CPU even when eVMCS enabled
authorLiran Alon <liran.alon@oracle.com>
Fri, 29 Jun 2018 19:59:04 +0000 (22:59 +0300)
committerPaolo Bonzini <pbonzini@redhat.com>
Wed, 18 Jul 2018 09:31:28 +0000 (11:31 +0200)
When eVMCS is enabled, all VMCS allocated to be used by KVM are marked
with revision_id of KVM_EVMCS_VERSION instead of revision_id reported
by MSR_IA32_VMX_BASIC.

However, even though not explictly documented by TLFS, VMXArea passed
as VMXON argument should still be marked with revision_id reported by
physical CPU.

This issue was found by the following setup:
* L0 = KVM which expose eVMCS to it's L1 guest.
* L1 = KVM which consume eVMCS reported by L0.
This setup caused the following to occur:
1) L1 execute hardware_enable().
2) hardware_enable() calls kvm_cpu_vmxon() to execute VMXON.
3) L0 intercept L1 VMXON and execute handle_vmon() which notes
vmxarea->revision_id != VMCS12_REVISION and therefore fails with
nested_vmx_failInvalid() which sets RFLAGS.CF.
4) L1 kvm_cpu_vmxon() don't check RFLAGS.CF for failure and therefore
hardware_enable() continues as usual.
5) L1 hardware_enable() then calls ept_sync_global() which executes
INVEPT.
6) L0 intercept INVEPT and execute handle_invept() which notes
!vmx->nested.vmxon and thus raise a #UD to L1.
7) Raised #UD caused L1 to panic.

Reviewed-by: Krish Sadhukhan <krish.sadhukhan@oracle.com>
Cc: stable@vger.kernel.org
Fixes: 773e8a0425c923bc02668a2d6534a5ef5a43cc69
Signed-off-by: Liran Alon <liran.alon@oracle.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
arch/x86/kvm/vmx.c

index ba98145..c3c8590 100644 (file)
@@ -4108,11 +4108,7 @@ static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
        vmcs_conf->order = get_order(vmcs_conf->size);
        vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
 
-       /* KVM supports Enlightened VMCS v1 only */
-       if (static_branch_unlikely(&enable_evmcs))
-               vmcs_conf->revision_id = KVM_EVMCS_VERSION;
-       else
-               vmcs_conf->revision_id = vmx_msr_low;
+       vmcs_conf->revision_id = vmx_msr_low;
 
        vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
        vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
@@ -4182,7 +4178,13 @@ static struct vmcs *alloc_vmcs_cpu(int cpu)
                return NULL;
        vmcs = page_address(pages);
        memset(vmcs, 0, vmcs_config.size);
-       vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
+
+       /* KVM supports Enlightened VMCS v1 only */
+       if (static_branch_unlikely(&enable_evmcs))
+               vmcs->revision_id = KVM_EVMCS_VERSION;
+       else
+               vmcs->revision_id = vmcs_config.revision_id;
+
        return vmcs;
 }
 
@@ -4341,6 +4343,19 @@ static __init int alloc_kvm_area(void)
                        return -ENOMEM;
                }
 
+               /*
+                * When eVMCS is enabled, alloc_vmcs_cpu() sets
+                * vmcs->revision_id to KVM_EVMCS_VERSION instead of
+                * revision_id reported by MSR_IA32_VMX_BASIC.
+                *
+                * However, even though not explictly documented by
+                * TLFS, VMXArea passed as VMXON argument should
+                * still be marked with revision_id reported by
+                * physical CPU.
+                */
+               if (static_branch_unlikely(&enable_evmcs))
+                       vmcs->revision_id = vmcs_config.revision_id;
+
                per_cpu(vmxarea, cpu) = vmcs;
        }
        return 0;