Not quite running in userspace. To the best of my own understanding:
The new kernel feature is to allow writing schedulers in eBPF, a “language” the kernel runs in kernelspace that is heavily restricted.
For example, all eBPF programs must complete in bounded time, and the kernel’s static checker must be able to verify that before the program can even begin executing. eBPF is a rare language that is not touring complete.
“For scx_simple, suspending the scheduler process doesn’t affect scheduling behavior because all that the userspace component does is print statistics. This doesn’t hold for all schedulers.”
So, it may be that eBPF also makes it easier to write a truly userspace scheduler, but that’s not the primary purpose, and it’s not what is being done with scx_simple.
Not quite running in userspace. To the best of my own understanding:
The new kernel feature is to allow writing schedulers in eBPF, a “language” the kernel runs in kernelspace that is heavily restricted.
For example, all eBPF programs must complete in bounded time, and the kernel’s static checker must be able to verify that before the program can even begin executing. eBPF is a rare language that is not touring complete.
“For scx_simple, suspending the scheduler process doesn’t affect scheduling behavior because all that the userspace component does is print statistics. This doesn’t hold for all schedulers.”
So, it may be that eBPF also makes it easier to write a truly userspace scheduler, but that’s not the primary purpose, and it’s not what is being done with scx_simple.
https://lwn.net/Articles/909095/ for more about (e)BPF.
Thank you for the correction! Reading up on eBPF is fascinating.
Additional resource that adds to your secondary point that this is more than just allowing schedulers to be run in userland: https://github.com/sched-ext/scx/blob/main/scheds/c/README.md
Is eBPF dependently typed?