Re: [PATCH] fix rcu vs hotplug race

[Paul E. McKenney]

On Fri, Jun 27, 2008 at 11:19:59AM +0530, Dhaval Giani wrote:
> On Fri, Jun 27, 2008 at 10:48:55AM +0530, Dipankar Sarma wrote:
> > On Fri, Jun 27, 2008 at 10:17:38AM +0530, Gautham R Shenoy wrote:
> > > IMHO the warning is a spurious one.
> > > Here's the timeline.
> > > CPU_A						 CPU_B
> > > --------------------------------------------------------------
> > > cpu_down():					.
> > > .					   	.
> > > .						.
> > > stop_machine(): /* disables preemption,		.
> > > 		 * and irqs */			.
> > > .						.
> > > .						.
> > > take_cpu_down();				.
> > > .						.
> > > .						.
> > > .						.
> > > cpu_disable(); /*this removes cpu 		.
> > > 		*from cpu_online_map 		.
> > > 		*/				.
> > > .						.
> > > .						.
> > > restart_machine(); /* enables irqs */		.
> > > ------WINDOW DURING WHICH rcp->cpumask is stale ---------------
> > > .						call_rcu();
> > > .						/* disables irqs here */
> > > .						.force_quiescent_state();
> > > .CPU_DEAD:					.for_each_cpu(rcp->cpumask)
> > > .						.   smp_send_reschedule();
> > > .						.
> > > .						.   WARN_ON() for offlined CPU!
> > > .
> > 
> > Exactly. The call_rcu()s are coming from a different subsystem
> > and can happen anytime during the CPU hotplug path. So, RCU subsystem
> > doesn't have anything to do to keep rcu->cpumask consistent.
> > It is *safe* even if we miss poking a cpu or two while
> > forcing quiescent state in all CPUs. The worst that can happen
> > is a delay in grace period. No correctness problem here.
> > 
> 
> One question. What is preventing a CPU from clearing its mask after we
> have checked whether it is online but before we have called into
> smp_send_reschedule?

This is my concern as well.  Gautham, at which point in the above
timeline is the offlining CPU marked DYING?  Before stop_machine(), right?

If so, can't we just disable irqs, check for DYING or DEAD, and invoke
smp_send_reschedule() only if not DYING or DEAD?

							Thanx, Paul
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/