On 01/25/21 09:26, Vincent Guittot wrote: > On Sat, 23 Jan 2021 at 19:31, Qais Yousef <qais.yousef@xxxxxxx> wrote: > > > > On 01/20/21 15:46, Vincent Guittot wrote: > > > > +.TP > > > > +For example: > > > > +.B uclampset \-s > > > > +.I -m\ 512 > > > > +.I -M\ 700 > > > > +.PP > > > > +will prevent any task from being boosted higher than 512. And all tasks in the > > > > > > will prevent any task from being boosted higher than 700 > > > > 700 or 512 you mean here? > > > > > or > > > will ensure any task to be boosted higher than 512 > > > > It is certainly not that. That's the confusion I was trying to clear up in this > > note. The system values are _constraints_. It sets an upper bound on the > > maximum uclamp_min/max value the system will apply. > > > > For example if a task::uclamp_min=1024 but the system::uclamp_min=512 512, then > > effectively the task will only be boosted to 512. Once this restriction is left > > by setting system::uclamp_min=1024 again, then the task will get the boost > > value it requested. > > ok, I got confused because uclamp -p PID -min 512 - M 700 and uclamp > -s -min 512 - M 700 look quite similar but their behaviors are quite > different. > > I mean > > uclamp -p PID -min 512 - M 700 will ensure that the task's utilization > will be always reported in the range [512:700] > The behavior is the same for the cgroup with cpu.uclamp.min and cpu.uclamp.max The cgroup behavior actually works like system. It is a constraint of your max allowed util_min/max value. There's an exception to that where cgroup will behave like '-p'... I avoided commenting on cgroup behavior here. I plan to send a patch to document util clamp in the kernel. For this tool since we don't interact with cgroup I omitted talking about it. I intend to send a patch documenting util clamp in the kernel. I think It's better to discuss this details there? > > whereas > > uclamp -s -min 512 - M 700 will do the opposite; i.e. the min of > task's utilization range will never be above 512 > > TBH I don't know how to make it clear that the behavior of -min is the > opposite between -p and -s Hopefully it'll become common knowledge at some point.. Thanks for having a look. Cheers -- Qais Yousef
