Thanks, Sebastian, for your reply. Yes, it is about RT_GROUP_SCHED (and related parameters such as cpu.rt_runtime_us and cpu.rt_period_us). The documentation (such as https://www.kernel.org/doc/Documentation/scheduler/sched-rt-group.txt ;) is fuzzy on details and is not dated, so it was not clear if we were missing anything. Anyhow, thanks for the confirmation. I think we will go back to the original plan and avoid using cgroups for managing CPU time for RT tasks. Perhaps with time and experience on the implementation, we may have a better view on how to properly enable this. Regards Elana From: Sebastian Andrzej Siewior <bigeasy@xxxxxxxxxxxxx> Sent: Tuesday, May 26, 2020 5:33 PM To: Elana Copperman <Elana.Copperman@xxxxxxxxxxxx> Cc: Steven Rostedt <rostedt@xxxxxxxxxxx>; linux-rt-users <linux-rt-users@xxxxxxxxxxxxxxx>; Thomas Gleixner <tglx@xxxxxxxxxxxxx>; Vladimir Kondratiev (Mobileye) <63dc4b5d-c467-4bfb-bce5-74485907c850@xxxxxxxxxxxxxxxxxxxxxxxxxx> Subject: Re: cgroups CPU time management + PREEMPT_RT On 2020-05-12 13:41:03 [+0000], Elana Copperman wrote: > > I think you missed to write a reply or I didn't find it. So if this is about RT_GROUP_SCHED then it is disabled in RT because it was once broken. Since then people asked about it and I told them to please test it and report back what the results are but I never got a reply back. Sebastian
