Over the past few years, high resolution and stealthy attacks and their variants such as Flush+Reload, Flush+Flush, Prime+Probe, Spectre and Meltdown have completely exposed the vulnerabilities in modern computing architectures. Many effective mitigation techniques against such attacks are also being proposed that use system's behavioral parameters at run-time using Performance Counters (PCs) coupled with machine learning models. Although PCs, both in hardware and software, have shown promising results when used in the context of security, this paper provides experimental evaluation and analysis of the potential challenges, perils and pitfalls of using these counters in security.