Re: [Patch -v4 1/4] Migrate shutdown/reboot to boot cpu.

[Robin Holt]

On Wed, Apr 17, 2013 at 09:48:35AM +0200, Ingo Molnar wrote:
> 
> * Robin Holt <holt@xxxxxxx> wrote:
> 
> > On Tue, Apr 16, 2013 at 09:18:07PM +0530, Srivatsa S. Bhat wrote:
> > > On 04/16/2013 05:36 PM, Robin Holt wrote:
> > > > On Tue, Apr 16, 2013 at 01:32:56PM +0200, Ingo Molnar wrote:
> > > >>
> > > >> * Robin Holt <holt@xxxxxxx> wrote:
> > > >>
> > > >>> We recently noticed that reboot of a 1024 cpu machine takes approx 16
> > > >>> minutes of just stopping the cpus.  The slowdown was tracked to commit
> > > >>> f96972f.
> > > >>>
> > > >>> The current implementation does all the work of hot removing the cpus
> > > >>> before halting the system.  We are switching to just migrating to the
> > > >>> boot cpu and then continuing with shutdown/reboot.
> > > >>>
> > > >>> This also has the effect of not breaking x86's command line parameter for
> > > >>> specifying the reboot cpu.  Note, this code was shamelessly copied from
> > > >>> arch/x86/kernel/reboot.c with bits removed pertaining to the reboot_cpu
> > > >>> command line parameter.
> > > >>>
> > > >>> Signed-off-by: Robin Holt <holt@xxxxxxx>
> > > >>> Tested-by: Shawn Guo <shawn.guo@xxxxxxxxxx>
> > > >>> To: Ingo Molnar <mingo@xxxxxxxxxx>
> > > >>> To: Russ Anderson <rja@xxxxxxx>
> > > >>> To: Oleg Nesterov <oleg@xxxxxxxxxx>
> > > >>> Cc: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
> > > >>> Cc: "H. Peter Anvin" <hpa@xxxxxxxxx>
> > > >>> Cc: Lai Jiangshan <laijs@xxxxxxxxxxxxxx>
> > > >>> Cc: Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx>
> > > >>> Cc: Linux Kernel Mailing List <linux-kernel@xxxxxxxxxxxxxxx>
> > > >>> Cc: Michel Lespinasse <walken@xxxxxxxxxx>
> > > >>> Cc: Oleg Nesterov <oleg@xxxxxxxxxx>
> > > >>> Cc: "Paul E. McKenney" <paulmck@xxxxxxxxxxxxxxxxxx>
> > > >>> Cc: Paul Mackerras <paulus@xxxxxxxxx>
> > > >>> Cc: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
> > > >>> Cc: Robin Holt <holt@xxxxxxx>
> > > >>> Cc: "rusty@xxxxxxxxxxxxxxx" <rusty@xxxxxxxxxxxxxxx>
> > > >>> Cc: Tejun Heo <tj@xxxxxxxxxx>
> > > >>> Cc: the arch/x86 maintainers <x86@xxxxxxxxxx>
> > > >>> Cc: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
> > > >>> Cc: <stable@xxxxxxxxxxxxxxx>
> > > >>>
> > > >>> ---
> > > >>>
> > > >>> Changes since -v1.
> > > >>> - Set PF_THREAD_BOUND before migrating to eliminate potential race.
> > > >>> - Modified kernel_power_off to also migrate instead of using
> > > >>>   disable_nonboot_cpus().
> > > >>> ---
> > > >>>  kernel/sys.c | 22 +++++++++++++++++++---
> > > >>>  1 file changed, 19 insertions(+), 3 deletions(-)
> > > >>>
> > > >>> diff --git a/kernel/sys.c b/kernel/sys.c
> > > >>> index 0da73cf..5ef7aa2 100644
> > > >>> --- a/kernel/sys.c
> > > >>> +++ b/kernel/sys.c
> > > >>> @@ -357,6 +357,22 @@ int unregister_reboot_notifier(struct notifier_block *nb)
> > > >>>  }
> > > >>>  EXPORT_SYMBOL(unregister_reboot_notifier);
> > > >>>  
> > > >>> +void migrate_to_reboot_cpu(void)
> > > >>
> > > >> It appears to be file-scope, so should be static I guess?
> > > > 
> > > > Done.
> > > > 
> > > >>> +{
> > > >>> +	/* The boot cpu is always logical cpu 0 */
> > > >>> +	int reboot_cpu_id = 0;
> > > >>> +
> > > >>> +	/* Make certain the cpu I'm about to reboot on is online */
> > > >>> +	if (!cpu_online(reboot_cpu_id))
> > > >>> +		reboot_cpu_id = smp_processor_id();
> > > >>
> > > >> Shouldn't we pick the first online CPU instead, to make it deterministic?
> > > > 
> > > > Done.
> > > > 
> > > > 		reboot_cpu_id = cpumask_first(cpu_online_mask);
> > > > 
> > > 
> > > Let me ask again: if CPU 0 (or whatever the preferred reboot cpu is)
> > > is offline, then why should we even bother pinning the task to (another)
> > > CPU? Why not just proceed with the reboot?
> > 
> > No idea.  I copied it from the arch/x86 code.  I can not defend it.
> 
> I'd say it's a quality of implementation improvement if the choice of the CPU is 
> deterministic, as long as the current configuration of CPUs is deterministic.
> 
> I.e. instead of 'reboot on the first CPU, or a random CPU', make the rule 'reboot 
> on the first online CPU'. That's a simple rule to think about.
> 
> ( On most architectures CPU#0 cannot be unplugged, so the rule will effectively be 
>   'reboot on CPU#0'. Like the current upstream behavior. )

Would you be more comfortable with:

static void migrate_to_reboot_cpu(void)
{
        /* The boot cpu is always logical cpu 0 */
        int reboot_cpu_id = reboot_cpuid;

        get_online_cpus();

        /* Make certain the cpu I'm about to reboot on is online */
        if (!cpu_online(reboot_cpu_id))
                reboot_cpu_id = 0;
        if (!cpu_online(reboot_cpu_id))
                reboot_cpu_id = cpumask_first(cpu_online_mask);

        /* Prevent races with other tasks migrating this task */
        current->flags |= PF_THREAD_BOUND;

        /* Make certain I only run on the appropriate processor */
        set_cpus_allowed_ptr(current, cpumask_of(reboot_cpu_id));

        put_online_cpus();
}

Robin
--
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/