Linux kernel cgroup v2 CFS - CPU throttled_usec accounting?

[Alexandru Lazarev]

Hi Costa, Thx U very much for your response.

On Sat, Jan 11, 2025 at 7:03 PM Costa Shulyupin <costa.shul at redhat.com> wrote:
> <alexandru.lazarev at gmail.com> wrote:
> > In Linux kernel cgroup v2’s CFS scheduler, how is cpu.max throttled_usec accounted when a cgroup with multiple threads gets throttled during a single quota period? Specifically, is throttled_usec tracked as the total wall-clock time that the cgroup was throttled as a whole, or is it a sum of the throttled times of all individual threads?
> >
> > Kernel Version: "5.14.0-284.11.1.el9_2.x86_64 #1 SMP PREEMPT_DYNAMIC Tue May 9 11:41:53 PDT 2023 x86_64 x86_64 x86_64 GNU/Linux" Distro: Oracle Linux 9.x
>
> Hi
>
> When the cgroup is throttled, all tasks (threads) within the cgroup
> are simultaneously prevented from running. The `throttled_time`
> counter starts incrementing when the cgroup as a whole enters the
> throttled state and stops when the cgroup is no longer throttled.
>
> Quote:
>
> "throttled_time: The total time duration (in nanoseconds) for which
> entities of the group have been throttled."
>
>   https://docs.kernel.org/scheduler/sched-bwc.html#:~:text=throttled_time
>

Wanted experts' confirmations :), because IMHO I got confused by this
"The total time duration (in nanoseconds) for which entities".
Someone (those 'smart' AI chats) interpreted them as cumulative time
of all throttled threads/processes within a cgroup,
and also due to missing description of "throttled_usec" in cgroup v2
in https://docs.kernel.org/admin-guide/cgroup-v2.html#cgroup-v2-cpu:~:text=nr_throttled-,throttled_usec,-nr_bursts

> The source:
>
> void unthrottle_cfs_rq(struct cfs_rq *cfs_rq)
> {
>         ...
>         cfs_b->throttled_time += rq_clock(rq) - cfs_rq->throttled_clock;
>         ...
>
> static int cpu_extra_stat_show(struct seq_file *sf, struct
> cgroup_subsys_state *css)
> {
>         struct task_group *tg = css_tg(css);
>         struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
>         ...
>         throttled_usec = cfs_b->throttled_time;
>
> Costa
> Thanks
>

Regards,
AlexL