Messages in this thread | | | Date | Mon, 8 Jun 2009 09:49:34 +0530 | From | Gautham R Shenoy <> | Subject | Re: [PATCH] cpuhotplug: introduce try_get_online_cpus() take 2 |
| |
Hi Paul,
On Fri, Jun 05, 2009 at 08:37:14AM -0700, Paul E. McKenney wrote: > > One question for Gautham Shenoy -- are non-atomic CPU-hotplug notifiers > always invoked from the task that did the cpu_hotplug_begin()?
Except for the notifiers handling two events, rest of the notifiers are always invoked from the task that did the cpu_hotplug_begin().
The two events are CPU_DYING which is called from the context of the stop_machine_thread and CPU_STARTING which is called from the context of the idle thread on the CPU that has just come up. The notifiers handling these two events are expected to be atomic.
> If so, well and good. If not, then it would not be possible to > expedite RCU grace periods from within CPU-hotplug notifiers.
I hope this would be good enough :-) > > Thanx, Paul >
-- Thanks and Regards gautham
| |