The key reason for using an RTOS is determinism, which typically means precise management or time and space resources. Unfortunately, many problems require more complex solutions, including virtual memory and virtual-machine support.
Virtual-memory systems drive RTOS programmers nuts, because a single page fault can mess up timing. In most instances, the trick is to lock down applications that require real-time support. This sometimes means not using the caches associated with a virtualmemory system.
More often, though, the chips can lock down portions of the cache for use by the kernel or device drivers. As a result, other portions of the application can take advantage of virtual-memory support. These applications could still be real time, but their cycle times must be longer to account for the possibility of page faults.
Virtual memory provides memory-access protection in addition to virtualization support. Often, just the protection support is required. In this case, platforms that only provide memory-access protection can be used. The RTOS operation in this case isn’t much different than a virtual-memory operating system.
Virtual-machine support is more sophisticated than virtual-memory support. At this point, all virtual-machine platforms implement virtual memory. This can be a challenge for programmers needing determinism, but RTOS support is possible.
A small hypervisor sits between the real hardware and the rest of the system (see the figure). The hypervisor normally runs virtual machines that are isolated. The virtual machines can run a single application or another operating system. Sometimes a host operating system will run directly atop the hypervisor.
Running an RTOS on the hypervisor is one way to provide real-time support. A more challenging approach is to run an RTOS inside a virtual machine. This is possible, and several vendors deliver this type of virtual-machine support. The hard part is determining the timing associated with the RTOS and its applications. Hardware support can minimize the overhead associated with virtual machine support. Locking down cache contents is an option, too, allowing virtualized applications to perform close to what would occur with bare, non-virtualized hardware
