Re: [PATCH 3/6] sched/deadline: Allow DL tasks on empty (cgroup v2) cpusets

[Juri Lelli <juri.lelli@redhat.com>]

Hi,

On 12/01/21 16:53, Daniel Bristot de Oliveira wrote:
> cgroups v2 allows the cpuset controller to be enabled/disabled on
> demand. On Fedora 32, cpuset is disabled by default. To enable it,
> a user needs to:
> 
>   # cd /sys/fs/cgroup/
>   # echo +cpuset > cgroup.subtree_control
> 
> Existing cgroups will expose the cpuset interface (e.g., cpuset.cpus
> file). By default, cpuset.cpus has no CPU assigned, which means that
> existing tasks will move to a cpuset without cpus.

This is kind of confusing, though. Isn't it?

> Initially, I thought about returning an error and blocking the
> operation. However, that is indeed not needed. The cpuset without
> CPUs assigned will be a non-root cpuset, hence its cpu mask will
> be the same as the root one. So, the bandwidth was already accounted,
> and the task can proceed.
> 
> Signed-off-by: Daniel Bristot de Oliveira <bristot@redhat.com>
> Cc: Ingo Molnar <mingo@redhat.com>
> Cc: Peter Zijlstra <peterz@infradead.org>
> Cc: Juri Lelli <juri.lelli@redhat.com>
> Cc: Vincent Guittot <vincent.guittot@linaro.org>
> Cc: Dietmar Eggemann <dietmar.eggemann@arm.com>
> Cc: Steven Rostedt <rostedt@goodmis.org>
> Cc: Ben Segall <bsegall@google.com>
> Cc: Mel Gorman <mgorman@suse.de>
> Cc: Daniel Bristot de Oliveira <bristot@redhat.com>
> Cc: Li Zefan <lizefan@huawei.com>
> Cc: Tejun Heo <tj@kernel.org>
> Cc: Johannes Weiner <hannes@cmpxchg.org>
> Cc: Valentin Schneider <valentin.schneider@arm.com>
> Cc: linux-kernel@vger.kernel.org
> Cc: cgroups@vger.kernel.org
> ---
>  kernel/sched/deadline.c | 7 +++++++
>  1 file changed, 7 insertions(+)
> 
> diff --git a/kernel/sched/deadline.c b/kernel/sched/deadline.c
> index 943aa32cc1bc..788a391657a5 100644
> --- a/kernel/sched/deadline.c
> +++ b/kernel/sched/deadline.c
> @@ -2871,6 +2871,13 @@ int dl_task_can_attach(struct task_struct *p,
>  	bool overflow;
>  	int ret;
>  
> +	/*
> +	 * The cpuset has no cpus assigned, so the thread will not
> +	 * change its affinity.

Is this always the case also in the presence of deeper hierarchies?

Thanks,
Juri