diff --git a/tf_adapter/kernels/geop_npu.cc b/tf_adapter/kernels/geop_npu.cc
index 3f086db4f39a96acc39064fa7f12f26fabcf35f4..1f637e69db60ce9db467eb63f7dd4a0658f2a227 100644
--- a/tf_adapter/kernels/geop_npu.cc
+++ b/tf_adapter/kernels/geop_npu.cc
@@ -1050,6 +1050,7 @@ Status GeOp::DoGraphParser(ge::ComputeGraphPtr &compute_graph, FunctionLibraryDe
 }
 
 void GeOp::ComputeAsync(OpKernelContext *ctx, DoneCallback done) {
+  run_mtx_.lock();
   // ctx is not nullptr
   OP_REQUIRES_ASYNC(ctx, init_flag_, errors::InvalidArgument("GeOp not Initialize success."), done);
   if (!sess_init_flag_) {
@@ -1318,9 +1319,10 @@ void GeOp::ComputeAsync(OpKernelContext *ctx, DoneCallback done) {
   }
 
   int64 run_start_time = InferShapeUtil::GetCurrentTimestap();
-  auto callback = [done, ctx, run_start_time](ge::Status ge_status, std::vector<ge::Tensor> &outputs) {
+  auto callback = [done, ctx, run_start_time, this](ge::Status ge_status, std::vector<ge::Tensor> &outputs) {
     if (ge_status == ge::SUCCESS) {
       if (BuildOutputTensorInfo(ctx, outputs) != Status::OK()) {
+        run_mtx_.unlock();
         ADP_LOG(FATAL) << ctx->op_kernel().name() << " GEOP::DoRunAsync get output failed.";
         std::string error_message = ge::GEGetErrorMsg();
         std::stringstream ss;
@@ -1335,6 +1337,7 @@ void GeOp::ComputeAsync(OpKernelContext *ctx, DoneCallback done) {
       ADP_LOG(WARNING) << "[GEOP] Out of range: End of sequence.";
       LOG(WARNING) << "[GEOP] Out of range: End of sequence.";
     } else if (ge_status != ge::SUCCESS) {
+      run_mtx_.unlock();
       std::this_thread::sleep_for(std::chrono::milliseconds(kFatalSleepTime));
       ADP_LOG(FATAL) << ctx->op_kernel().name() << "GEOP::::DoRunAsync Failed";
       std::string error_message = ge::GEGetErrorMsg();
@@ -1348,6 +1351,7 @@ void GeOp::ComputeAsync(OpKernelContext *ctx, DoneCallback done) {
     ADP_LOG(INFO) << "[GEOP] RunGraphAsync callback, status:" << ge_status
                    << ", kernel_name:" << ctx->op_kernel().name() << "[ " << (run_end_time - run_start_time) << "us]";
     done();
+    run_mtx_.unlock();
   };
 
   // call ge session runGraphAsync api
diff --git a/tf_adapter/kernels/geop_npu.h b/tf_adapter/kernels/geop_npu.h
index bf102b2ff9990790a9f848ffedcf54a7d5ae368e..01ca513c062946336881511b4c4ab3d57113f42d 100644
--- a/tf_adapter/kernels/geop_npu.h
+++ b/tf_adapter/kernels/geop_npu.h
@@ -247,6 +247,7 @@ public:
   AoeSetTuningGraphInputFunc aoe_set_tuning_graph_input_;
   // accelerate train
   AccelerateInfo accelerate_info_;
+  std::mutex run_mtx_;
 };
 }  // namespace tensorflow
 #endif  // TENSORFLOW_KERNELS_GEOP_NPU_H_