__scx_bpf_select_cpu_and(9)

📦 version 7.3.1, in nixpkgs unstable

💅 NAME 💅

__scx_bpf_select_cpu_and — arg-wrapped CPU selection, but make it cpumask 🎯 (and yes bestie, this is a section 9 entry from the Kernel API Manual, we're in kernel-land now)

✨ SYNOPSIS ✨

__bpf_kfunc s32 __scx_bpf_select_cpu_and (struct task_struct *p , const
struct cpumask *cpus_allowed , struct scx_bpf_select_cpu_and_args *args
, const struct bpf_prog_aux *aux );

okay so it's giving very long signature energy, but that's literally the whole point of this function, iykyk 🫶

🎀 ARGUMENTS 🎀

  • p — the task_struct we're picking a CPU for. main character of the operation, honestly 👑
  • cpus_allowed — a cpumask of the CPUs that are actually allowed. the guest list, if you will 🎟️
  • args — a struct holding the rest of the arguments, because we ran out of room (more on that below 😭):
    • args->prev_cpu — the CPU that p was on previously, i.e. where it's coming from
    • args->wake_flags — the SCX_WAKE_* flags
    • args->flags — the SCX_PICK_IDLE* flags
  • aux — the implicit BPF argument giving access to bpf_prog_aux, and it's hidden from BPF progs. you literally can't touch this one from your BPF program, it's for the kernel's eyes only 🙈

💖 DESCRIPTION 💖

So here's the tea, bestie 🍓: this is a wrapper kfunc that takes its arguments through a struct, and it does that for one extremely relatable reason — to work around BPF's 5 argument limit. like, five args, that's the whole budget, that's all BPF gives you 💀

BPF programs should not call this directly. instead use scx_bpf_select_cpu_and, which is provided as an inline wrapper in common.bpf.h. that's your entry point, periodt ✨

You can call it from ops.select_cpu, from ops.enqueue, or from an unlocked context such as a BPF test_run call — but only as long as built-in CPU selection is enabled, which means either:

  • ops.update_idle is missing, or
  • SCX_OPS_KEEP_BUILTIN_IDLE is set
no built-in idle selection? then this vibe check fails, fr fr. make sure one of those two conditions holds 🙏

Also super important matching situation: p, args->prev_cpu and args->wake_flags match ops.select_cpu. same crew, same semantics 🧚

🌸 What it gives back

It returns the selected idle CPU 💫 — and listen, that CPU gets automatically awakened upon returning from ops.select_cpu, and you can use it for direct dispatch. Or, if there's no idle CPU available, you get a negative value instead 😭 that's the "nobody's home" answer.

🌈 SEE ALSO 🌈

Kernel file ./kernel/sched/ext/idle.c, plus the whole friend group:

okay that's the whole page, slay 🎀 this one lives in ./kernel/sched/ext and it's dated October 2026 — so yeah, it's a little time-traveling queen from the future, we stan a forward-thinking manpage 👑💜

📦 where this page comes from