Re: Nohz_full on boot CPU is broken (was: Re: [PATCH v2 1/1] wq: Avoid using isolated cpus' timers on queue_delayed_work)

[Oleg Nesterov <oleg@redhat.com>]

Hi Nicholas,

On 04/10, Nicholas Piggin wrote:
>
> Thanks for this. Taking a while to page this back in, the intention is
> for housekeeping to be done by boot CPU until house keeper is awake, so
> returning smp_processor_id() seems like the right thing to do here for
> ephemeral jobs like timers and work, provided that CPU / mask is not
> stored somewhere long term by the caller.
>
> For things that set an affinity like kthread, sched, maybe managed
> irqs, and such.
>
> There are not many callers of housekeeping_any_cpu() so that's easy
> enough to verify. But similar like housekeeping_cpumask() and others
> could be an issue or at least a foot-gun, I'm not sure how well I
> convinced myself of those.
>
> Could you test like this?
>
>   WARN_ON_ONCE(system_state == SYSTEM_RUNNING ||
>                type != HK_TYPE_TIMER);
>
> With a comment to say other ephemeral mask types could be exempted if
> needed.

Sorry, I don't understand... Let me repeat, I know absolutely nothing
about nonhz/etc. I didn't even try to really fix the problem(s), I am
only trying to find a minimal/simple workaround to fix the problem we
hit in Red Hat.

This is what I was going to send:

	--- a/kernel/sched/isolation.c
	+++ b/kernel/sched/isolation.c
	@@ -46,7 +46,15 @@ int housekeeping_any_cpu(enum hk_type type)
				if (cpu < nr_cpu_ids)
					return cpu;
	 
	-			return cpumask_any_and(housekeeping.cpumasks[type], cpu_online_mask);
	+			cpu = cpumask_any_and(housekeeping.cpumasks[type], cpu_online_mask);
	+			if (likely(cpu < nr_cpu_ids))
	+				return cpu;
	+			/*
	+			 * Unless we have another problem this can only happen
	+			 * at boot time before start_secondary() brings the 1st
	+			 * housekeeping CPU up.
	+			 */
	+			WARN_ON_ONCE(system_state == SYSTEM_RUNNING);
			}
		}
		return smp_processor_id();

Yes, this fixes the symptom, not the problem. And yes, "another problem"
mentioned in the comment is very possible, say "maxcpus" kernel-parameter
can be less than the first housekeeping cpu. But in this case the user
should blame himself (and I am not sure the kernel will boot).

I don't understand why do you suggest to add "|| type != HK_TYPE_TIMER",
currently all the callers of housekeeping_any_cpu() use type == HK_TIMER.
But OK, I can add this check. I guess for the case it finds another user
with type != HK_TYPE_TIMER which can't use smp_processor_id() even at
boot time or stores the returned CPU for the long term.

Will you agree with the change above or what do you suggest instead as
a simple workaround?


> It would also be nice to warn for cases that would be bugs if the boot
> CPU was not in the HK mask. Could that be done by having a
> housekeepers_online() call after smp_init() (maybe at the start of
> sched_init_smp()) that could verify there is at least one online, and
> set a flag that could be used to create warnings.

Again, I am not sure I understand, but I too thought that something like

	housekeeping_check(void)
	{
		for_each_set_bit(type, &housekeeping.flags, HK_TYPE_MAX) {
			if (!cpumask_intersects(cpu_online, housekeeping.cpumasks[type]))
				panic();
		}

after bringup_nonboot_cpus(setup_max_cpus).

But I am not sure this is correct and this is another (although related) issue.

Oleg.