Remote graphical processing for dual display of RTOS and GPOS on an embedded hypervisor

Abstract

In this paper, we introduce a remote graphics library framework based on inter-virtual-machines communication in an embedded hypervisor. With this framework, there are no causality conflicts when multiple guest operating systems share one GPU. We adopted API remoting for GPU virtualization because it has relatively small overhead when connected with OpenGL ES standard library on embedded hypervisors. Interferences from the hypervisor during synchronization between front-end and back-end can be reduced by inter-VM commutation. To make improvement on size, weight and power for embedded systems, we opted for displaying both guest operating systems on a single display panel. The presented framework is applied to a real-world embedded hypervisor used for safety-critical systems. Our implementation runs an automotive digital instrument cluster on a real-time guest operating system and an in-vehicle infotainment application on a general purpose guest operating system within the hypervisor. We found it feasible for an embedded hypervisor to provide GPU service to heterogeneous industrial guest operating systems on a single hardware platform.