Dynamic Memory Access Monitoring Based on VMM

Memory access monitoring is an important fine-grained monitoring method, which monitors system status at single instruction execution and data access level. However, this kind of fine-grained monitoring method based on VMM (Virtual Machine Monitor) causes significant performance overhead, especially when monitoring dynamic data. This is because safety critical data and non-critical data is located mixed in memory, while the hardware only supports monitoring central memory regions. In this paper, we will propose a dynamic and flexible memory access monitoring method, named DynMon, which can adjust the timing to monitor and select the critical data to be monitored automatically. DynMon monitors safety-critical data and significantly reduces performance overhead by triggering monitoring only when it is highly likely that these data will be accessed. Besides, DynMon can be easily deployed on systems with support of hardware-assisted virtualization, while it does not need any modification of the monitored system. In the evaluation, the results show that DynMon can reduce 22.23% performance overhead. Besides, the performance cost can be tolerated, when monitoring the safety-critical data set in out prototype.