Hi Neetigya,
Interrupt latencies introduced by EL2 are highly implementation-defined.
ARMv8 GIC3 provides a set of virtualization registers in hardware, but real efficiency varies among different hypervisor products due to different EL2 software designs. It's suggested to inquire hypervisor vendors about the committed performance.
Use cases are also an influential factor. For example, if only a few pending virtual interrupts holding in ICH_LR<n>_EL2 for a vPE, interrupt performance may be good. But when the total number of interrupts exceeds the maximum number of ICH_LR<n>_EL2, hypervisor software may save some interrupt entries to memory, which definitely increases interrupt firing time. Besides, multiple vPEs running on a single core also could cost the hypervisor more time to handle interrupts due to sharing the same hardware registers.
The hypervisor has many benefits but efficiency/performance is not one of them. If you have time-sensitive applications, you'd better measure critical metrics in a certain hypervisor software and real use cases.
It's up to users to choose multicore OS or hypervisor or both. Hypervisor provided similar functions as multicore OS, like VM scheduling. But virtualization is a key feature OS does not have, which brings benefits to isolation and cross-team cooperation. If there is a hypervisor presented, the need for OS could still be there, for example, AutoSAR standard compliance. So it really depends on the application and strategy.
