In some designs, multiple operating systems are required to run on the APU MPCore. Running multiple guest operating systems on a CPU cluster requires hardware virtualization support to virtualize the processor system into multiple virtual machines (VM) to allow each guest operating system to run on its VM.
Operating systems are generally designed to run on native hardware. The system expects to be executing in the most privileged mode and assumes total control over the whole system. In a virtualized environment, it is the VM that runs in privileged mode, while the operating system is executing at a lower privilege level.
When booting, a typical operating system configures the processor, memories, I/O devices, and peripherals. When executing, it expects exclusive access to such devices, including changing peripherals' configuration dynamically, directly managing the interrupt controller, replacing MMU page table entries (PTE), and initiating DMA transfers.
When running de-privileged inside a virtual machine, the guest operating system is not able to execute the privileged instructions necessary to configure and drive the hardware directly.
The VM must manage these functions. In addition, the VM could be hosting multiple guest operating systems. Therefore, direct modification of shared devices and memory requires cautious arbitration schemes.
The level of abstraction required to address this, and the inherent software complexity and performance overhead, are specific to the characteristics of the architecture, the hardware, and the guest operating systems. The main approaches can be broadly categorized in two groups.
In full virtualization, the guest operating system is not aware that it is virtualized, and it does not require any modification. The VM traps and handles all privileged and sensitive instruction sequences, while user-level instructions run unmodified at native speed.
In paravirtualization the guest operating system is modified to have direct access to the VM through hyper-calls or hypervisor calls. A special API is exposed by the VM to allow guest operating systems to execute privileged and sensitive instruction sequences.
The Arm Cortex-A53 exception level-2 (EL2) provides processor virtualization. The Arm v8 supports virtualization extension to achieve full virtualization with near native guest operating system performance.
There are four key hardware components for virtualization.
•System Memory Virtualization Using SMMU Address Translation