[PATCH] kernel/cpu.c: Move the CPU_DYING notifiers

[PATCH] kernel/cpu.c: Move the CPU_DYING notifiers

[Manfred Spraul]

When a cpu is taken offline, the CPU_DYING notifiers are called on the
dying cpu. According to <linux/notifiers.h>, the cpu should be "not
running any task, not handling interrupts, soon dead".

For the current implementation, this is not true:
- __cpu_disable can fail. If it fails, then the cpu will remain alive
  and happy.
- At least on x86, __cpu_disable() briefly enables the local interrupts
  to handle any outstanding interrupts.

What about moving CPU_DYING down a few lines, behind the __cpu_disable()
line?
There are only two CPU_DYING handlers in the kernel right now: one in
kvm, one in the scheduler. Both should work with the patch applied
[and: I'm not sure if either one handles a failing __cpu_disable()]

The patch survives simple offlining a cpu. kvm untested due to lack
of a test setup.

Signed-Off-By: Manfred Spraul <manfred@colorfullife.com>
---
 kernel/cpu.c |    5 +++--
 1 files changed, 3 insertions(+), 2 deletions(-)

diff --git a/kernel/cpu.c b/kernel/cpu.c
index e202a68..5b7c88f 100644
--- a/kernel/cpu.c
+++ b/kernel/cpu.c
@@ -199,13 +199,14 @@ static int __ref take_cpu_down(void *_param)
 	struct take_cpu_down_param *param = _param;
 	int err;
 
-	raw_notifier_call_chain(&cpu_chain, CPU_DYING | param->mod,
-				param->hcpu);
 	/* Ensure this CPU doesn't handle any more interrupts. */
 	err = __cpu_disable();
 	if (err < 0)
 		return err;
 
+	raw_notifier_call_chain(&cpu_chain, CPU_DYING | param->mod,
+				param->hcpu);
+
 	/* Force idle task to run as soon as we yield: it should
 	   immediately notice cpu is offline and die quickly. */
 	sched_idle_next();
-- 
1.5.5.1

Re: [PATCH] kernel/cpu.c: Move the CPU_DYING notifiers

[Paul E. McKenney]

On Sun, Aug 31, 2008 at 07:58:49PM +0200, Manfred Spraul wrote:
> When a cpu is taken offline, the CPU_DYING notifiers are called on the
> dying cpu. According to <linux/notifiers.h>, the cpu should be "not
> running any task, not handling interrupts, soon dead".
> 
> For the current implementation, this is not true:
> - __cpu_disable can fail. If it fails, then the cpu will remain alive
>   and happy.
> - At least on x86, __cpu_disable() briefly enables the local interrupts
>   to handle any outstanding interrupts.
> 
> What about moving CPU_DYING down a few lines, behind the __cpu_disable()
> line?
> There are only two CPU_DYING handlers in the kernel right now: one in
> kvm, one in the scheduler. Both should work with the patch applied
> [and: I'm not sure if either one handles a failing __cpu_disable()]
> 
> The patch survives simple offlining a cpu. kvm untested due to lack
> of a test setup.

Several architectures re-enable interrupts in __cpu_disable() or in
functions called from __cpu_disable(), which happens after CPU_DYING,
if I understand correctly.  :-(

							Thanx, Paul

> Signed-Off-By: Manfred Spraul <manfred@colorfullife.com>
> ---
>  kernel/cpu.c |    5 +++--
>  1 files changed, 3 insertions(+), 2 deletions(-)
> 
> diff --git a/kernel/cpu.c b/kernel/cpu.c
> index e202a68..5b7c88f 100644
> --- a/kernel/cpu.c
> +++ b/kernel/cpu.c
> @@ -199,13 +199,14 @@ static int __ref take_cpu_down(void *_param)
>  	struct take_cpu_down_param *param = _param;
>  	int err;
> 
> -	raw_notifier_call_chain(&cpu_chain, CPU_DYING | param->mod,
> -				param->hcpu);
>  	/* Ensure this CPU doesn't handle any more interrupts. */
>  	err = __cpu_disable();
>  	if (err < 0)
>  		return err;
> 
> +	raw_notifier_call_chain(&cpu_chain, CPU_DYING | param->mod,
> +				param->hcpu);
> +
>  	/* Force idle task to run as soon as we yield: it should
>  	   immediately notice cpu is offline and die quickly. */
>  	sched_idle_next();
> -- 
> 1.5.5.1
> 

Re: [PATCH] kernel/cpu.c: Move the CPU_DYING notifiers

[Paul E. McKenney]

On Sun, Aug 31, 2008 at 12:17:21PM -0700, Paul E. McKenney wrote:
> On Sun, Aug 31, 2008 at 07:58:49PM +0200, Manfred Spraul wrote:
> > When a cpu is taken offline, the CPU_DYING notifiers are called on the
> > dying cpu. According to <linux/notifiers.h>, the cpu should be "not
> > running any task, not handling interrupts, soon dead".
> > 
> > For the current implementation, this is not true:
> > - __cpu_disable can fail. If it fails, then the cpu will remain alive
> >   and happy.
> > - At least on x86, __cpu_disable() briefly enables the local interrupts
> >   to handle any outstanding interrupts.
> > 
> > What about moving CPU_DYING down a few lines, behind the __cpu_disable()
> > line?
> > There are only two CPU_DYING handlers in the kernel right now: one in
> > kvm, one in the scheduler. Both should work with the patch applied
> > [and: I'm not sure if either one handles a failing __cpu_disable()]
> > 
> > The patch survives simple offlining a cpu. kvm untested due to lack
> > of a test setup.
> 
> Several architectures re-enable interrupts in __cpu_disable() or in
> functions called from __cpu_disable(), which happens after CPU_DYING,
> if I understand correctly.  :-(

Never mind -- you are moving CPU_DYING after __cpu_disable().  :-/

							Thanx, Paul

> > Signed-Off-By: Manfred Spraul <manfred@colorfullife.com>
> > ---
> >  kernel/cpu.c |    5 +++--
> >  1 files changed, 3 insertions(+), 2 deletions(-)
> > 
> > diff --git a/kernel/cpu.c b/kernel/cpu.c
> > index e202a68..5b7c88f 100644
> > --- a/kernel/cpu.c
> > +++ b/kernel/cpu.c
> > @@ -199,13 +199,14 @@ static int __ref take_cpu_down(void *_param)
> >  	struct take_cpu_down_param *param = _param;
> >  	int err;
> > 
> > -	raw_notifier_call_chain(&cpu_chain, CPU_DYING | param->mod,
> > -				param->hcpu);
> >  	/* Ensure this CPU doesn't handle any more interrupts. */
> >  	err = __cpu_disable();
> >  	if (err < 0)
> >  		return err;
> > 
> > +	raw_notifier_call_chain(&cpu_chain, CPU_DYING | param->mod,
> > +				param->hcpu);
> > +
> >  	/* Force idle task to run as soon as we yield: it should
> >  	   immediately notice cpu is offline and die quickly. */
> >  	sched_idle_next();
> > -- 
> > 1.5.5.1
> > 

Re: [PATCH] kernel/cpu.c: Move the CPU_DYING notifiers

[Ingo Molnar]

* Manfred Spraul <manfred@colorfullife.com> wrote:

> -	raw_notifier_call_chain(&cpu_chain, CPU_DYING | param->mod,
> -				param->hcpu);
>  	/* Ensure this CPU doesn't handle any more interrupts. */
>  	err = __cpu_disable();
>  	if (err < 0)
>  		return err;
>  
> +	raw_notifier_call_chain(&cpu_chain, CPU_DYING | param->mod,
> +				param->hcpu);

hm, doesnt this break things like CPU cross-calls done in CPU_DYING 
callbacks?

	Ingo

Re: [PATCH] kernel/cpu.c: Move the CPU_DYING notifiers

[Manfred Spraul]

Ingo Molnar wrote:
> * Manfred Spraul <manfred@colorfullife.com> wrote:
>   
>> -	raw_notifier_call_chain(&cpu_chain, CPU_DYING | param->mod,
>> -				param->hcpu);
>>  	/* Ensure this CPU doesn't handle any more interrupts. */
>>  	err = __cpu_disable();
>>  	if (err < 0)
>>  		return err;
>>  
>> +	raw_notifier_call_chain(&cpu_chain, CPU_DYING | param->mod,
>> +				param->hcpu);
>>     
>
> hm, doesnt this break things like CPU cross-calls done in CPU_DYING 
> callbacks?
>   
We are within stop_machine(). No other cpu is running. As fas as I can 
see no cross-calls are possible.

Which scenario do you think about?

--
    Manfred

Re: [PATCH] kernel/cpu.c: Move the CPU_DYING notifiers

[Ingo Molnar]

* Manfred Spraul <manfred@colorfullife.com> wrote:

> Ingo Molnar wrote:
>> * Manfred Spraul <manfred@colorfullife.com> wrote:
>>   
>>> -	raw_notifier_call_chain(&cpu_chain, CPU_DYING | param->mod,
>>> -				param->hcpu);
>>>  	/* Ensure this CPU doesn't handle any more interrupts. */
>>>  	err = __cpu_disable();
>>>  	if (err < 0)
>>>  		return err;
>>>  +	raw_notifier_call_chain(&cpu_chain, CPU_DYING | param->mod,
>>> +				param->hcpu);
>>>     
>>
>> hm, doesnt this break things like CPU cross-calls done in CPU_DYING  
>> callbacks?
>>   
>
> We are within stop_machine(). No other cpu is running. As fas as I can 
> see no cross-calls are possible.

ah, ok - my bad. I was confusing it with the much more common 
CPU_DOWN_PREPARE type of callbacks which do use various cross-CPU APIs.

applied to tip/sched/devel, thanks Manfred!

	Ingo

Re: [PATCH] kernel/cpu.c: Move the CPU_DYING notifiers

[Avi Kivity]

Manfred Spraul wrote:
> When a cpu is taken offline, the CPU_DYING notifiers are called on the
> dying cpu. According to <linux/notifiers.h>, the cpu should be "not
> running any task, not handling interrupts, soon dead".
>
> For the current implementation, this is not true:
> - __cpu_disable can fail. If it fails, then the cpu will remain alive
>   and happy.
> - At least on x86, __cpu_disable() briefly enables the local interrupts
>   to handle any outstanding interrupts.
>
> What about moving CPU_DYING down a few lines, behind the __cpu_disable()
> line?
> There are only two CPU_DYING handlers in the kernel right now: one in
> kvm, one in the scheduler. Both should work with the patch applied
> [and: I'm not sure if either one handles a failing __cpu_disable()]
>
> The patch survives simple offlining a cpu. kvm untested due to lack
> of a test setup.
>
>   

kvm should work with this patch.

-- 
I have a truly marvellous patch that fixes the bug which this
signature is too narrow to contain.