Re: [RFC PATCH v6 00/25] Hierarchical Constant Bandwidth Server

[luca abeni]

Hi,

On Mon, 15 Jun 2026 10:38:54 -1000
Tejun Heo <tj@xxxxxxxxxx> wrote:
[...]
> 2. root's cpu.rt.max: sched_rt_runtime_us already caps total DL/RT
>    bandwidth and rt-cgroups admit against the same pool, so what does
>    reserving the cgroup share separately at root add? It's also a
> writable control on root, which we otherwise keep off root.

Well, /proc/sys/kernel/sched_rt_{runtime,period}_us control the total
fraction of CPU time that can be reserved for dl and rt tasks; here, we
need a way to reserve a part of this fraction for rt cgropus only. We
need this because "regular" dl entities (the ones serving dl tasks) are
"global" (can migrate between different CPU cores), whereas dl entities
serving rt cgroups are pinned to their own CPU cores.

Combining the admission tests for these two different kinds of dl
entities is not simple, so we reserve a fixed fraction of CPU time for
rt cgroups, and the remaining part of
sched_rt_runtime_us/sched_rt_period_us is left to dl tasks.

If there is a better interface we can use to achieve this goal, we are
happy to switch to it.


			Thanks,
				Luca