Re: [PREEMPT-RT] [PATCH] s390/cpum_sf: Remove superfluous SMP function call

[Heiko Carstens <heiko.carstens@de.ibm.com>]

On Tue, Apr 05, 2016 at 01:23:36PM +0200, Heiko Carstens wrote:
> On Tue, Apr 05, 2016 at 01:13:06PM +0200, Sebastian Andrzej Siewior wrote:
> > On 04/05/2016 12:49 PM, Heiko Carstens wrote:
> > >> --- a/arch/s390/kernel/perf_cpum_sf.c
> > >> +++ b/arch/s390/kernel/perf_cpum_sf.c
> > >> @@ -1510,7 +1510,6 @@ static void cpumf_measurement_alert(stru
> > >>  static int cpumf_pmu_notifier(struct notifier_block *self,
> > >>  			      unsigned long action, void *hcpu)
> > >>  {
> > >> -	unsigned int cpu = (long) hcpu;
> > >>  	int flags;
> > >>
> > >>  	/* Ignore the notification if no events are scheduled on the PMU.
> > >> @@ -1523,11 +1522,15 @@ static int cpumf_pmu_notifier(struct not
> > >>  	case CPU_ONLINE:
> > >>  	case CPU_DOWN_FAILED:
> > >>  		flags = PMC_INIT;
> > >> -		smp_call_function_single(cpu, setup_pmc_cpu, &flags, 1);
> > >> +		local_irq_disable();
> > >> +		setup_pmc_cpu(&flags);
> > >> +		local_irq_enable();
> > >>  		break;
> > > 
> > > ...but at least the CPU_DOWN_FAILED callback will not necessarily be called
> > > on the cpu that couldn't be brought offline.
> > 
> > I don't follow.
> 
> I was trying to say that if bringing a cpu down fails, then the cpu hotplug
> notifier with CPU_DOWN_FAILED might be called on a cpu that is _not_ the
> same cpu that was supposed to be brought offline.
> 
> Subsequently, in this case, the setup_pmc_cpu() call will be executed on
> the wrong cpu.

.. or to illustrate this behaviour: the following patch (white space
damaged due to copy-paste) results in the following:

# chcpu -d 2
[console] failed cpu: 2 - this cpu: 1

diff --git a/arch/s390/kernel/smp.c b/arch/s390/kernel/smp.c
index 40a6b4f9c36c..48d417abfdae 100644
--- a/arch/s390/kernel/smp.c
+++ b/arch/s390/kernel/smp.c
@@ -852,6 +852,7 @@ int __cpu_disable(void)
 {
	unsigned long cregs[16];
 
+	return -EBUSY;
	/* Handle possible pending IPIs */
	smp_handle_ext_call();
	set_cpu_online(smp_processor_id(), false);
@@ -1065,6 +1066,10 @@ static int smp_cpu_notify(struct notifier_block *self, unsigned long action,
	int err = 0;
 
	switch (action & ~CPU_TASKS_FROZEN) {
+	case CPU_DOWN_FAILED:
+		printk("failed cpu: %d - this cpu: %d\n", cpu, get_cpu());
+		put_cpu();
+		break;
	case CPU_ONLINE:
		err = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
		break;