Using eBPF per-CPU Cgroup local storage for low overhead accounting

Linux eBPF ecosystem is growing fast with new features coming with every
release cycle. The spirit remains the same. At the core, eBPF is like
a safe way to script the Linux kernel by attaching to various pre-defines hooks
and exported functions. For kernel / user spaces communication, the eBPF comes
with various “map” types.

Every few kernel versions, new “map” types are introduced or significantly
improved and 3 years ago BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE was
introduced and is now available in recent distribution kernels.