Re: [PATCH] powerpc/sysfs: disable hotplug for the boot cpu

[Zhao Chenhui <chenhui.zhao@freescale.com>]

On Sat, Jun 01, 2013 at 07:49:44AM +1000, Benjamin Herrenschmidt wrote:
> On Tue, 2013-05-28 at 15:59 +0800, Zhao Chenhui wrote:
> > Some features depend on the boot cpu, for instance, hibernate/suspend.
> > So disable hotplug for the boot cpu.
> 
> Don't we have code to "move" the boot CPU around when that happens ?
> 
> Ben.
> 

Currently, the code in generic_cpu_disable() likes this:

        if (cpu == boot_cpuid)                                                 
	                return -EBUSY;

If the dying cpu is the boot cpu, it will return -EBUSY. In the subsequent error handling,
cpu_notify_nofail(CPU_DOWN_FAILED) in _cpu_down() will be called. Unfortunately, some
cpu notifier callbacks handled CPU_DOWN_PREPARE, but not CPU_DOWN_FAILED, such as sched_cpu_inactive().
So it will cause issues.

If we set the hotpluggable for the boot cpu, we can prevent user applications from disabling the boot cpu.

-Chenhui

> > Signed-off-by: Zhao Chenhui <chenhui.zhao@freescale.com>
> > ---
> >  arch/powerpc/kernel/sysfs.c |    4 +++-
> >  1 files changed, 3 insertions(+), 1 deletions(-)
> > 
> > diff --git a/arch/powerpc/kernel/sysfs.c b/arch/powerpc/kernel/sysfs.c
> > index e68a845..294b1c4e 100644
> > --- a/arch/powerpc/kernel/sysfs.c
> > +++ b/arch/powerpc/kernel/sysfs.c
> > @@ -655,8 +655,10 @@ static int __init topology_init(void)
> >  		 * CPU.  For instance, the boot cpu might never be valid
> >  		 * for hotplugging.
> >  		 */
> > -		if (ppc_md.cpu_die)
> > +		if (ppc_md.cpu_die && cpu != boot_cpuid)
> >  			c->hotpluggable = 1;
> > +		else
> > +			c->hotpluggable = 0;
> >  
> >  		if (cpu_online(cpu) || c->hotpluggable) {
> >  			register_cpu(c, cpu);
> 
> 
>