GPU security is a vital area of mobile security highlighted both by public security research as well as by in-the-wild attacks. Due to the high complexity of the GPU software/firmware along with a widely available attack surface, issues in GPU provide strong exploitation primitives for local privilege escalation attacks by the code running in unprivileged context. In this talk, we will focus our research on the Qualcomm Adreno GPU, which is a very popular GPU implementation in mobile devices. We will do a deep dive into Adreno GPU kernel module implementation focusing on the most recent GPU versions, reveal its complex and new attack surfaces, and discuss vulnerabilities we discovered in this component.