Towards Fully Exploiting LLM Internal States to Enhance Knowledge Boundary Perception