Scheduling in OS kernel is an important role in improving application performance. The kernel performs scheduling based on low-level context such as the system calls and the process state. However, it is difficult to infer high-level context such as GPU utilization from the system calls. Thus, current kernels are not aware of such high-level process state. Therefore, this paper proposes a method to acquire process state by monitoring library access so that the scheduler can use it. As an example, we show a method for acquisition of GPU access information by hooking into CUDA. We also show that our method can detect GPU access in a more timely manner.