Abstract

Most modern smart phones are equipped with a Graphics Processing Unit (GPU), which takes care of rendering graphics, taking the workload off the CPU . For efficiency reasons, GPUs get Direct Memory Access (DMA), so that they can access any part of both kernel and user space memory. DMA attacks via FireWire and other PCI devices are well known and publicised. In this talk, I present a DMA attack using the GPU . The theoretical model holds for any device with GPU . Errors in the driver implementation of the Mali GPU on Samsung Galaxy SII phones allowed me to create a proof-of-concept exploit, which I will describe. I start off by describing the GPU programming model and processing pipeline, followed by an analysis of Mali Memory Management, and the flaws that enabled me to perform the DMA attack on Samsung Galaxy SII phones. I will explain different approaches for creating an exploit.