OSDN Git Service

KVM: nVMX: Flush TLB entries tagged by dest EPTP on L1<->L2 transitions
authorLiran Alon <liran.alon@oracle.com>
Mon, 8 Oct 2018 20:42:17 +0000 (23:42 +0300)
committerPaolo Bonzini <pbonzini@redhat.com>
Sat, 13 Oct 2018 10:00:53 +0000 (12:00 +0200)
If L1 and L2 share VPID (because L1 don't use VPID or we haven't allocated
a vpid02), we need to flush TLB on L1<->L2 transitions.

Before this patch, this TLB flushing was done by vmx_flush_tlb().
If L0 use EPT, this will translate into INVEPT(active_eptp);
However, if L1 use EPT, in L1->L2 VMEntry, active EPTP is EPTP01 but
TLB entries populated by L2 are tagged with EPTP02.
Therefore we should delay vmx_flush_tlb() until active_eptp is EPTP02.

To achieve this, instead of directly calling vmx_flush_tlb() we request
it to be called by KVM_REQ_TLB_FLUSH which is evaluated after
KVM_REQ_LOAD_CR3 which sets the active_eptp to EPTP02 as required.

Similarly, on L2->L1 VMExit, active EPTP is EPTP02 but TLB entries
populated by L1 are tagged with EPTP01 and therefore we should delay
vmx_flush_tlb() until active_eptp is EPTP01.

Reviewed-by: Mihai Carabas <mihai.carabas@oracle.com>
Reviewed-by: Darren Kenny <darren.kenny@oracle.com>
Reviewed-by: Nikita Leshenko <nikita.leshchenko@oracle.com>
Signed-off-by: Liran Alon <liran.alon@oracle.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
arch/x86/kvm/vmx.c

index 022642d..a4fa15b 100644 (file)
@@ -12290,7 +12290,15 @@ static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
                                __vmx_flush_tlb(vcpu, vmx->nested.vpid02, true);
                        }
                } else {
-                       vmx_flush_tlb(vcpu, true);
+                       /*
+                        * If L1 use EPT, then L0 needs to execute INVEPT on
+                        * EPTP02 instead of EPTP01. Therefore, delay TLB
+                        * flush until vmcs02->eptp is fully updated by
+                        * KVM_REQ_LOAD_CR3. Note that this assumes
+                        * KVM_REQ_TLB_FLUSH is evaluated after
+                        * KVM_REQ_LOAD_CR3 in vcpu_enter_guest().
+                        */
+                       kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
                }
        }
 
@@ -13188,10 +13196,14 @@ static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
         * Therefore, flush TLB only in case vmcs01 uses VPID and
         * vmcs12 don't use VPID as in this case L1 & L2 TLB entries
         * are both tagged with vmx->vpid.
+        *
+        * If vmcs12 uses EPT, we need to execute this flush on EPTP01
+        * and therefore we request the TLB flush to happen only after VMCS EPTP
+        * has been set by KVM_REQ_LOAD_CR3.
         */
        if (enable_vpid &&
            !(nested_cpu_has_vpid(vmcs12) && to_vmx(vcpu)->nested.vpid02)) {
-               vmx_flush_tlb(vcpu, true);
+               kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
        }
 
        vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);