Automated Analysis and Verification of UEFI BIOS Stress and Stability Testing

Abstract

Firmware can be defined as software embedded on hardware. Firmware can be used to initialize the hardware and give OS the ability to communicate to different devices through runtime services provided by the firmware. Firmware validation is a process of validating different features of the firmware and system. One such validation process is called UEFI/BIOS Stress and Stability Testing. Stress and Stability Testing includes the validation of the system functionalities and features by putting system in stringent test environment and observing the stability of the system during and after the validation process. This includes Power Management features' testing and much more. The testing process, executed by the internal tools, generates lot of logs and reports after completing the validation process. Reading and examining those files take lot of time and efforts which can be spent in other important and productive tasks. Therefore, to address this issue, an automation tool is developed in order to eliminate human intervention in this results analysis process. The tool is being developed in several sub-tools. Currently two different tools are developed in order to address this issue and those can be merged to serve the purpose of automation of the results analysis of UEFI/BIOS Stress and Stability Testing. Results and Advantages after using the developed tool are also described in this report. Also the execution ow of the tools are given so that the working of the tool and core concept can be understood well. As part of parallel research project, a project has been undertaken which is based on "Hyperspectral Image Classification using Semi-Supervised Learning with Label Propagation". The classification task for hyperspectral image is considered difficult due to less number of labeled samples available. Deep Learning algorithms require huge amount of labeled data which is not suitable for hyperspectral images as getting labeled data is costly. To mitigate this problem, we can employ semi-supervised learning techniques with label propagation that can address the issue of less labeled samples for training. By using this strategy, we can obtain comparative performance on hyperspectral data using very less number of labeled samples.