Patchwork powerpc/sysfs: disable hotplug for the boot cpu

login
register
mail settings
Submitter chenhui zhao
Date May 28, 2013, 7:59 a.m.
Message ID <1369727984-21505-2-git-send-email-chenhui.zhao@freescale.com>
Download mbox | patch
Permalink /patch/246723/
State Changes Requested
Headers show

Comments

chenhui zhao - May 28, 2013, 7:59 a.m.
Some features depend on the boot cpu, for instance, hibernate/suspend.
So disable hotplug for the boot cpu.

Signed-off-by: Zhao Chenhui <chenhui.zhao@freescale.com>
---
 arch/powerpc/kernel/sysfs.c |    4 +++-
 1 files changed, 3 insertions(+), 1 deletions(-)
Benjamin Herrenschmidt - May 31, 2013, 9:49 p.m.
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.

> 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);
chenhui zhao - June 3, 2013, 10:43 a.m.
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);
> 
> 
>
Benjamin Herrenschmidt - June 12, 2013, 3:25 a.m.
On Mon, 2013-06-03 at 18:43 +0800, Zhao Chenhui wrote:
> 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;

But the code in pseries/hotplug-cpu.c doesn't, we just "move" the boot
CPU around when that happens. Any reason we can't do that generically ?

Cheers,
Ben.

> 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);
> > 
> > 
> >
chenhui zhao - June 13, 2013, 11:25 a.m.
On Wed, Jun 12, 2013 at 01:25:22PM +1000, Benjamin Herrenschmidt wrote:
> On Mon, 2013-06-03 at 18:43 +0800, Zhao Chenhui wrote:
> > 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;
> 
> But the code in pseries/hotplug-cpu.c doesn't, we just "move" the boot
> CPU around when that happens. Any reason we can't do that generically ?
> 
> Cheers,
> Ben.
> 

Some multicore SoCs firstly boot up the cpu0 after warm reset.
In some suspend/resume cases, SoC will do a warm reset when resuming.
In order to ensure that the suspending and resuming is running
on a same cpu, cpu0 should be the last cpu to suspend. Here, cpu0 is
the boot_cpuid.

-Chenhui

> > 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
> >
Benjamin Herrenschmidt - June 14, 2013, 10:53 p.m.
On Thu, 2013-06-13 at 19:25 +0800, Zhao Chenhui wrote:
> Some multicore SoCs firstly boot up the cpu0 after warm reset.
> In some suspend/resume cases, SoC will do a warm reset when resuming.
> In order to ensure that the suspending and resuming is running
> on a same cpu, cpu0 should be the last cpu to suspend. Here, cpu0 is
> the boot_cpuid.

Well, so:

  - In any case, your patch will break pseries, so it's not acceptable.

  - Why does it have to absolutely resume from the same CPU it suspended
from ? Can't you have a little bit of code on the resuming CPU that
checks if it's not online, poke an online one and goes back to sleep ?

Cheers,
Ben.

Patch

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);