Re: RCU stall warnings...

From: Paul E. McKenney
Date: Mon Jul 24 2017 - 22:44:27 EST


On Mon, Jul 24, 2017 at 04:49:27PM -0700, Paul E. McKenney wrote:
> On Mon, Jul 24, 2017 at 04:34:58PM -0700, David Miller wrote:
> > From: "Paul E. McKenney" <paulmck@xxxxxxxxxxxxxxxxxx>
> > Date: Mon, 24 Jul 2017 16:20:33 -0700

[ . . . ]

> > That would take a while as it's hard to forcibly set this thing off.
>
> And my similar error can take awhile as well. But maybe I should try
> forcing nr_cpus=43 and maxcpus=8 on older versions to see what happens.
>
> A bisection would of course be quite helpful, depending of course on
> the value of "a while". ;-)

And if "a while" is too long, one alternative is to enable event tracing
for timers, as in why didn't they wake the grace-period kthread?
For example, it is possible that that kthread set its status (which
was printed out in the "rcu_sched kthread starved" message) but for
whatever reason didn't make it to the swait_event_idle_timeout()
immediately afterwards;

rsp->gp_state = RCU_GP_WAIT_FQS;
ret = swait_event_idle_timeout(rsp->gp_wq,
rcu_gp_fqs_check_wake(rsp, &gf), j);

Or that the timer-based wakeup didn't happen for whatever reason. Or...

Thanx, Paul

> > ====================
> > commit f92c734f02cbf10e40569facff82059ae9b61920
> > Author: Paul E. McKenney <paulmck@xxxxxxxxxxxxxxxxxx>
> > Date: Mon Apr 10 15:40:35 2017 -0700
> >
> > rcu: Prevent rcu_barrier() from starting needless grace periods
> >
> > Currently rcu_barrier() uses call_rcu() to enqueue new callbacks
> > on each CPU with a non-empty callback list. This works, but means
> > that rcu_barrier() forces grace periods that are not otherwise needed.
> > The key point is that rcu_barrier() never needs to wait for a grace
> > period, but instead only for all pre-existing callbacks to be invoked.
> > This means that rcu_barrier()'s new callbacks should be placed in
> > the callback-list segment containing the last pre-existing callback.
> >
> > This commit makes this change using the new rcu_segcblist_entrain()
> > function.
> >
> > Signed-off-by: Paul E. McKenney <paulmck@xxxxxxxxxxxxxxxxxx>
>
>