rcu: Throttle NOCB kthread grace-period rate

rcu: Throttle NOCB kthread grace-period rate

On a system with high CPU utilization and high context-switch rates,
grace periods tend to complete quite quickly.  If all CPUs are no-CBs
CPUs, and all are producing at least one callback per grace period, all
the rcuo kthreads will need to be awakened on every grace period, which
on large systems can be an excessive number of wakeups.  This commit
therefore throttles the rcuo kthreads to prevent them from needing
a grace-period wakeup more often than the interval between a pair of
force-quiescent-state scans.

Reported-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>