Hard
Syscall transition cost under KPTI
Both io_uring and the vDSO exist partly to avoid the same underlying cost: a plain syscall's user-to-kernel mode transition. On a modern x86-64 Linux system with Meltdown mitigations (KPTI, kernel page-table isolation) enabled, why is that transition frequently significantly MORE expensive than 'just' the CPU switching privilege rings, and how do io_uring and the vDSO each sidestep that specific extra cost in a different way?
41 of 53 in Operating Systems