Messages in this thread | | | Date | Thu, 14 Jun 2012 17:07:30 +0200 (CEST) | From | Thomas Gleixner <> | Subject | Re: [RFC patch 2/5] smpboot: Provide infrastructure for percpu hotplug threads |
| |
On Thu, 14 Jun 2012, Peter Zijlstra wrote:
> On Thu, 2012-06-14 at 07:47 -0700, Paul E. McKenney wrote: > > RCU callback processing consumes the entire CPU in RCU_BOOST case where > > processing runs at real-time priority. This is analogous to RT throttling > > in the scheduler. > > But previously we can in non-preemptible softirq context, why would if > behave differently when done from a RT task?
softirqs are different. They loop ten times and then wake ksoftirqd which runs with sched_other.
Though that's wonky, because if an interrupt arrives before ksoftirqd can take over we loop another 10 times in irq_exit(). Rinse and repeat.....
And the main difference is that the scheduler does not yell on the softirq context, but it yells when an rt task monopolizes the cpu.
Thanks,
tglx
| |