2008-02-08 19:27:38

by Balaji Rao

Subject: [REGRESSION] kobject handling in cpufreq

Hi greg,

I encountered a regression in linux-2.6.git which prevented the system from shutting down. I found that it
was 'waiting_for_completion' on kobj_unregister.The same thing happened when I tried to take CPUs offline.

git-bisect pointed me to this commit :

commit c10997f6575f476ff38442fa18fd4a0d80345f9d
Author: Greg Kroah-Hartman <[email protected]>
Date: Thu Dec 20 08:13:05 2007 -0800

Kobject: convert drivers/* from kobject_unregister() to kobject_put()

In the file drivers/cpufreq/cpufreq.c, you have not replaced one occurrence of kobject_unregister with kobject_put.
Apparently its because of the kobject_put in the subsequent line.

@@ -1030,8 +1030,6 @@ static int __cpufreq_remove_dev (struct sys_device * sys_dev)

unlock_policy_rwsem_write(cpu);

- kobject_unregister(&data->kobj);
-
kobject_put(&data->kobj);

/* we need to make sure that the underlying kobj is actually

Now, this can be fixed by adding another kobject_put. But having two kobject_puts in consequent lines tells us that there
should be two references to the kobject at that point. So, I believe that it is safe to remove the kobject_get
corresponding to the existing kobject_put to fix this.

Signed-off-by: Balaji Rao <[email protected]>

diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c
index 64926aa..2158786 100644
--- a/drivers/cpufreq/cpufreq.c
+++ b/drivers/cpufreq/cpufreq.c
@@ -1004,17 +1004,6 @@ static int __cpufreq_remove_dev (struct sys_device * sys_dev)
unlock_policy_rwsem_write(cpu);
return 0;
}
-#endif
-
-
- if (!kobject_get(&data->kobj)) {
- spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
- cpufreq_debug_enable_ratelimit();
- unlock_policy_rwsem_write(cpu);
- return -EFAULT;
- }
-
-#ifdef CONFIG_SMP

#ifdef CONFIG_HOTPLUG_CPU
cpufreq_cpu_governor[cpu] = data->governor;
--
regards,
balaji rao

2008-02-08 19:41:47

by Greg KH

Subject: Re: [REGRESSION] kobject handling in cpufreq

On Sat, Feb 09, 2008 at 12:52:40AM +0530, Balaji Rao wrote:
> Hi greg,
>
> I encountered a regression in linux-2.6.git which prevented the system from shutting down. I found that it
> was 'waiting_for_completion' on kobj_unregister.The same thing happened when I tried to take CPUs offline.
>
> git-bisect pointed me to this commit :
>
> commit c10997f6575f476ff38442fa18fd4a0d80345f9d
> Author: Greg Kroah-Hartman <[email protected]>
> Date: Thu Dec 20 08:13:05 2007 -0800
>
> Kobject: convert drivers/* from kobject_unregister() to kobject_put()
>
> In the file drivers/cpufreq/cpufreq.c, you have not replaced one occurrence of kobject_unregister with kobject_put.
> Apparently its because of the kobject_put in the subsequent line.
>
> @@ -1030,8 +1030,6 @@ static int __cpufreq_remove_dev (struct sys_device * sys_dev)
>
> unlock_policy_rwsem_write(cpu);
>
> - kobject_unregister(&data->kobj);
> -
> kobject_put(&data->kobj);
>
> /* we need to make sure that the underlying kobj is actually
>
> Now, this can be fixed by adding another kobject_put. But having two kobject_puts in consequent lines tells us that there
> should be two references to the kobject at that point. So, I believe that it is safe to remove the kobject_get
> corresponding to the existing kobject_put to fix this.
>
> Signed-off-by: Balaji Rao <[email protected]>

This looks correct to me, nice catch.

Acked-by: Greg Kroah-Hartman <[email protected]>

thanks,

greg k-h