Re: [RFC][PATCH -mm 1/7] PM: Remove pm_parent from struct dev_pm_info

On Mon, 11 Jun 2007, Rafael J. Wysocki wrote:

> From: Rafael J. Wysocki <[email protected]>
> 
> The pm_parent member of struct dev_pm_info (defined in include/linux/pm.h) is
> only used to check if the device's parent is in the right state while the
> device is being suspended or resumed.  However, this can be done just as well
> with the help of the parent pointer in struct device, so pm_parent can be
> removed along with some code that handles it.

> @@ -61,21 +40,26 @@ int device_pm_add(struct device * dev)
>  		 kobject_name(&dev->kobj));
>  	mutex_lock(&dpm_list_mtx);
>  	list_add_tail(&dev->power.entry, &dpm_active);
> -	device_pm_set_parent(dev, dev->parent);
> -	if ((error = dpm_sysfs_add(dev)))
> +	/*
> +	 * The device's parent must not be released until the device itself is
> +	 * removed from the dpm_active list.
> +	 */
> +	get_device(dev->parent);
> +	error = dpm_sysfs_add(dev);
> +	if (error)
>  		list_del(&dev->power.entry);
>  	mutex_unlock(&dpm_list_mtx);
>  	return error;
>  }

The error pathway here does an unbalanced get_device on dev->parent.

Anyway, I don't think you need to do this get_device at all (or the
coresponding put in device_pm_remove).  As long as a device is
registered it retains a reference to its parent, and unregistration
always calls device_pm_remove.  The reason it was there in the first 
place was because people recognized that dev->power.pm_parent wouldn't 
be one of dev's ancestors in the device hierarchy.

Alan Stern