Xen is a powerful virtualization solution which enables multiple virtual machines to run in a single physical machine isolatedly. Since the hardware resources of a physical machine (e.g., the CPU) are limited and they have to be shared by multiple virtual machines, the performance of virtual machines might be degraded. How to allocate the underlying hardware resources to virtual machines appropriately has become an active research topic. In this paper, we are interested in CPU scheduling algorithm for a Xen virtualized system. The default CPU scheduling algorithm of Xen, called Credit scheduler, is a fairness strategy which provides proportional fair shared CPU time to each virtual machine based on its predefined weight value. However, the Credit scheduler might deteriorate the performance of a virtual machine when its predefined weight value cannot meet its current requirement of the physical CPU. It is because the dynamic workload of a virtual machine is unpredictable. In this paper, a credit-based CPU scheduling algorithm, called load-awareness credit (LA-Credit), is proposed to adjust the weight values of virtual machines dynamically such that the overall system performance could be improved. In particular, the LA-Credit algorithm monitors the current workload of each virtual machine and then it calculates and adjusts an appropriate weight for each virtual machine so that a virtual machine with heavy workload can obtain more CPU share. We have implemented the LA-Credit algorithm in Xen and evaluated the performance, for which some encouraging results were obtained.