Re: [RFC] Make need_resched() return true when rcu_urgent_qs requested

["Paul E. McKenney" <paulmck@linux.vnet.ibm.com>]

On Mon, Jul 09, 2018 at 09:45:45PM +0100, David Woodhouse wrote:
> On Mon, 2018-07-09 at 13:42 -0700, Paul E. McKenney wrote:
> > On Mon, Jul 09, 2018 at 09:35:38PM +0100, David Woodhouse wrote:
> > > 
> > > 
> > > On Mon, 2018-07-09 at 13:34 -0700, Paul E. McKenney wrote:
> > > > 
> > > > So here are the possible code paths when .rcu_urgent_qs is set to true:
> > > > 
> > > > 1.      A context switch will record the quiescent state and clear
> > > >         .rcu_urgent_qs.  (The failure to do the clearing in current -rcu
> > > >         for PREEMPT builds is a performance bug that I need to fix.)
> > > 
> > > What if there's nothing else runnable and there is no actual context
> > > switch?
> > 
> > The scheduler invokes rcu_note_context_switch() before looking to see
> > if there really will or won't be a context switch.
> > 
> > I am sure that Peter will correct me if I am confused on this point.  ;-)
> 
> Ah, OK. Yes, that looks correct. Thanks.

Here is hoping!

> I'll give your patch a spin tomorrow, unless Marius beats me to it.

Please see below for the version that I eventually queued.  Should Marius
have a Reported-by?  If so, please tell me his full name so I can add that.

							Thanx, Paul

------------------------------------------------------------------------

commit e1e91fd0796dc544a77ddfaa5afbfc1f5fb42ecd
Author: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Date:   Mon Jul 9 13:47:30 2018 -0700

    rcu: Make need_resched() respond to urgent RCU-QS needs
    
    The per-CPU rcu_dynticks.rcu_urgent_qs variable communicates an urgent
    need for an RCU quiescent state from the force-quiescent-state processing
    within the grace-period kthread to context switches and to cond_resched().
    Unfortunately, such urgent needs are not communicated to need_resched(),
    which is sometimes used to decide when to invoke cond_resched(), for
    but one example, within the KVM vcpu_run() function.  As of v4.15, this
    can result in synchronize_sched() being delayed by up to ten seconds,
    which can be problematic, to say nothing of annoying.
    
    This commit therefore checks rcu_dynticks.rcu_urgent_qs from within
    rcu_check_callbacks(), which is invoked from the scheduling-clock
    interrupt handler.  If the current task is not an idle task and is
    not executing in usermode, a context switch is forced, and either way,
    the rcu_dynticks.rcu_urgent_qs variable is set to false.  If the current
    task is an idle task, then RCU's dyntick-idle code will detect the
    quiescent state, so no further action is required.  Similarly, if the
    task is executing in usermode, other code in rcu_check_callbacks() and
    its called functions will report the corresponding quiescent state.
    
    Reported-by: David Woodhouse <dwmw2@infradead.org>
    Suggested-by: Peter Zijlstra <peterz@infradead.org>
    Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>

diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
index 51919985f6cf..ea756bb64eb3 100644
--- a/kernel/rcu/tree.c
+++ b/kernel/rcu/tree.c
@@ -2496,6 +2496,15 @@ void rcu_check_callbacks(int user)
 {
 	trace_rcu_utilization(TPS("Start scheduler-tick"));
 	raw_cpu_inc(rcu_data.ticks_this_gp);
+	/* The load-acquire pairs with the store-release setting to true. */
+	if (smp_load_acquire(this_cpu_ptr(&rcu_dynticks.rcu_urgent_qs))) {
+		/* Idle and userspace execution already are quiescent states. */
+		if (!is_idle_task(current) && !user) {
+			set_tsk_need_resched(current);
+			set_preempt_need_resched();
+		}
+		__this_cpu_write(rcu_dynticks.rcu_urgent_qs, false);
+	}
 	rcu_flavor_check_callbacks(user);
 	if (rcu_pending())
 		invoke_rcu_core();