linux/Documentation/scheduler/sched-design-CFS.rst

Imported from _research/manual-study-linux/file-notes/linux__Documentation__scheduler__sched-design-CFS.rst.md.

File Notes: Documentation/scheduler/sched-design-CFS.rst

Status: reviewed.

Purpose

Documents the CFS model, its virtual-runtime accounting, RB-tree runqueue, and the scheduler-class abstraction that lets the core delegate policy.

Key Points

CFS models an ideal multitasking CPU by tracking each task’s p->se.vruntime. On real hardware only one task runs at a time, so virtual runtime records how much CPU service a task has received relative to its fair share. The old CFS pick rule was the leftmost entity in a time-ordered RB tree: the task with the smallest virtual runtime.

The document also explains scheduling classes. sched/fair.c implements CFS and EEVDF-era fair scheduling, sched/rt.c implements FIFO/RR, and the core calls class hooks for enqueue, dequeue, wakeup preemption, picking, setting the next task, and ticks.

Rust Translation Guidance

Treat virtual time as an explicit accounting domain. A Rust scheduler should not hide fairness accounting inside timers; it should have scheduling entities with service accounting, queue membership, and policy hooks.

AI-Native Systems Guidance

Agent scheduling should expose a similar class model. Normal jobs can be fair jobs, batch jobs can use longer slices, and latency-sensitive jobs can request more responsive behavior through policy rather than unbounded priority.

Evidence

• CFS models an ideal multitasking CPU and introduces virtual runtime at Documentation/scheduler/sched-design-CFS.rst:18-30.
• p->se.vruntime is the tracked value, and old CFS picked the smallest virtual runtime at Documentation/scheduler/sched-design-CFS.rst:37-49.
• CFS uses an RB tree and tracks min_vruntime and load at Documentation/scheduler/sched-design-CFS.rst:57-80.
• The scheduler-class abstraction and hook list are documented at Documentation/scheduler/sched-design-CFS.rst:151-207.