Re: Q: force_quiescent_state && cpu_online_map

[Paul E. McKenney]

On Tue, Nov 11, 2008 at 06:03:27PM +0100, Oleg Nesterov wrote:
> I don't think this matters, but still...
> 
> 	force_quiescent_state:
> 
> 			 * cpu_online_map is updated by the _cpu_down()
> 			 * using __stop_machine(). Since we're in irqs disabled
> 			 * section, __stop_machine() is not exectuting, hence
> 			 * the cpu_online_map is stable.
> 			 *
> 			 * However,  a cpu might have been offlined _just_ before
> 			 * we disabled irqs while entering here.
> 			 * And rcu subsystem might not yet have handled the CPU_DEAD
> 			 * notification, leading to the offlined cpu's bit
> 			 * being set in the rcp->cpumask.
> 			 *
> 			 * Hence cpumask = (rcp->cpumask & cpu_online_map) to prevent
> 			 * sending smp_reschedule() to an offlined CPU.
> 			 */
> 			cpus_and(cpumask, rcp->cpumask, cpu_online_map);
> 			cpu_clear(rdp->cpu, cpumask);
> 			for_each_cpu_mask_nr(cpu, cpumask)
> 				smp_send_reschedule(cpu);
> 
> However,
> 
> 	// called by __stop_machine take_cpu_down()
> 	arch/x86/kernel/smpboot.c:cpu_disable_common()
> 
> 		/*
> 		 * HACK:
> 		 * Allow any queued timer interrupts to get serviced
> 		 * This is only a temporary solution until we cleanup
> 		 * fixup_irqs as we do for IA64.
> 		 */
> 		local_irq_enable();
> 		mdelay(1);
> 		local_irq_disable();
> 		...
> 		remove_cpu_from_maps(cpu);
> 
> So it is possible to send the ipi to the dying CPU. I know nothing
> about this low-level irq code, most probably this is harmless. We
> already did clear_local_APIC(), but I don't understand what it does.

Indeed, some of the things I am doing as part of the hierarchical RCU
implementation need to be applied to preemptable RCU.  :-/

							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/