On 07-06-18, 16:11, Daniel Lezcano wrote: > >> I'm wondering if we can have a CPU hotplugged right after the > >> 'put_online_cpus', resulting in the 'should park' flag set and then the > >> thread goes in the kthread_parkme instead of jumping back the idle > >> injection function and decrease the count, leading up to the timer not > >> being set again. > > > > True. That looks like a valid problem to me as well. > > > > What about starting the hrtimer right from this routine itself, after taking > > into account running-time, idle-time, delay, etc ? That would get rid of the > > count stuff, this get_online_cpus(), etc. > > > > Even if we restart the next play-idle cycle a bit early or with some delay, > > sky > > wouldn't fall :) > > We won't be able to call completion() in this case.
I was just having a look at smpboot.h and saw the park/unpark callbacks. I think you can very much use them to align things with CPU hotplug. -- viresh
