Titolo della tesi: Methodologies and techniques for on-line exploitation of Performance Monitor Units in modern computing systems
We are living in a hot period in which the intense competition among chipmakers is turning into an explosion of new hardware-level features.
Among the innovations, modern processors embed a hardware support able to identify the footprint on the hardware caused by the execution of software.
In this thesis, we propose novel techniques to leverage the Performance Monitor Units (PMUs) hardware facilities offered by modern off-the-shelf CPUs to gather hardware-level footprint data at runtime, and exploit this information on-the fly.
Our solutions aim to collect hardware-level information while minimizing the perturbation experienced by the software applications and foster its online analysis and exploitation.
Throughout this thesis, we show how our methodology can be employed to address different problems in different domains. In particular, we provide the operating system with a module to master this integrated hardware support in order to perform dynamic corrective operations in malware detection, CPU-scheduling consolidation and functional hardware-accelerated tasks.
Moreover, working at the lowest software layer has the extra benefit of enforcing the transparency property of our approach, strengthening both the adaptability and the dependability of the whole solution.
Beyond demonstrating how PMUs can be exploited for security, performance and functional activities, in this thesis we, additionally present our profiling infrastructure, which implements versatile mechanisms for the exploitation of PMUs in modern contexts.