docs/sched_ext: document that cgroup CPU knobs are scheduler-dependent
Subsystems (3)
PATCH v2 1/1 docs/sched_ext: document that cgroup CPU knobs are scheduler-dependent
From: Tao Cui <[email protected]>
The fair class enforces cpu controller knobs such as cpu.max, cpu.weight and cpu.idle in the kernel. sched_ext only passes them to the BPF scheduler through the ops.cgroup_set_*() callbacks. Whether and how a knob takes effect is up to the loaded scheduler: if it doesn't implement the corresponding callback, the knob is ignored. For example, none of scx_simple, scx_flatcg and scx_central implements ops.cgroup_set_bandwidth(), so with them cpu.max has no effect -- the cgroup runs at unlimited CPU with nr_throttled staying at 0.
The same applies to other knobs like nice levels.
Document this in the basics section so users and container orchestrators know what to expect from a BPF scheduler.
---
RFC -> v2: v1 added a pr_warn_once() for the missing ops.cgroup_set_bandwidth() implementation. Drop the warning and document the scheduler-dependent behavior in sched-ext.rst instead, as suggested in review.
v1: https://lore.kernel.org/r/[email protected]
Documentation/scheduler/sched-ext.rst | 18 ++++++++++++++++++ 1 file changed, 18 insertions(+)
Documentation/scheduler/sched-ext.rst +18 -0
@@ -242,6 +242,24 @@ optional. The following modified excerpt is from .name = "simple", }; +Scheduler-Dependent Knobs +------------------------- + +The fair class enforces cpu controller knobs such as ``cpu.max``, +``cpu.weight`` and ``cpu.idle`` in the kernel. sched_ext only passes +them to the BPF scheduler through ``ops.cgroup_set_weight()``, +``ops.cgroup_set_idle()``, ``ops.cgroup_set_bandwidth()`` and friends. +Whether and how a knob takes effect is up to the loaded scheduler: if +it doesn't implement the corresponding callback, the knob is ignored. +For example, none of scx_simple, scx_flatcg and scx_central implements +``ops.cgroup_set_bandwidth()``, so with them ``cpu.max`` has no effect +-- the cgroup runs at unlimited CPU with ``nr_throttled`` staying +at 0. + +The same applies to other knobs like nice levels: the scheduler may +honor them partially or not at all. When relying on these knobs, check +the documentation or source of the loaded scheduler. + Dispatch Queues ---------------
Signature · 1 line
2.43.0
Tejun Heo
reviewerHello,
On Wed, Aug 19, 2026 at 09:21:57AM +0800, Tao Cui wrote:
...
Documentation/scheduler/sched-ext.rst optional. The following modified excerpt is from · line 257
Quoted 10 lines
Documentation/scheduler/sched-ext.rst +10
-0+The fair class enforces cpu controller knobs such as ``cpu.max``, +``cpu.weight`` and ``cpu.idle`` in the kernel. sched_ext only passes +them to the BPF scheduler through ``ops.cgroup_set_weight()``, +``ops.cgroup_set_idle()``, ``ops.cgroup_set_bandwidth()`` and friends. +Whether and how a knob takes effect is up to the loaded scheduler: if +it doesn't implement the corresponding callback, the knob is ignored. +For example, none of scx_simple, scx_flatcg and scx_central implements +``ops.cgroup_set_bandwidth()``, so with them ``cpu.max`` has no effect +-- the cgroup runs at unlimited CPU with ``nr_throttled`` staying +at 0.
SCX doesn't have nr_throttled routing right now even for scheds that implement cpu.max (scx_lavd is the only one right now AFAIK). I'd keep the documentation more concise and generic. I don't think listing schedulers that don't implement the feature or specifically pointing to nr_throttled is all that useful given that most schedulers don't implement it right now.
Thanks.
Signature · 1 line
tejun