diff --git a/arch/x86/kvm/mmu/tdp_mmu.c b/arch/x86/kvm/mmu/tdp_mmu.c
index af60922906ef0cfd1b052134bc0ac5c9d1c6e54e..7f63e1a704e3b27cd156b97c06ccbce051e1244e 100644
--- a/arch/x86/kvm/mmu/tdp_mmu.c
+++ b/arch/x86/kvm/mmu/tdp_mmu.c
@@ -941,13 +941,17 @@ static bool tdp_mmu_zap_leafs(struct kvm *kvm, struct kvm_mmu_page *root,
 		flush = true;
 	}
 
-	rcu_read_unlock();
-
 	/*
-	 * Because this flow zaps _only_ leaf SPTEs, the caller doesn't need
-	 * to provide RCU protection as no 'struct kvm_mmu_page' will be freed.
+	 * Need to flush before releasing RCU.  TODO: do it only if intermediate
+	 * page tables were zapped; there is no need to flush under RCU protection
+	 * if no 'struct kvm_mmu_page' is freed.
 	 */
-	return flush;
+	if (flush)
+		kvm_flush_remote_tlbs_with_address(kvm, start, end - start);
+
+	rcu_read_unlock();
+
+	return false;
 }
 
 /*