Re: [PATCH 0/7] convert semaphore to mutex in struct class

["Dave Young" <hidave.darkstar@gmail.com>]

On Jan 10, 2008 9:23 PM, Cornelia Huck <cornelia.huck@de.ibm.com> wrote:
> On Thu, 10 Jan 2008 17:48:43 +0800,
> Dave Young <hidave.darkstar@gmail.com> wrote:
>
> Please add a kerneldoc comment for each of the new interfaces.
Will do.
>
> > +int class_for_each_device(struct class *class, void *data,
> > +                        int (*fn)(struct device *, void *))
> > +{
> > +     struct device *dev;
> > +     int error = 0;
> > +
> > +     if (!class)
> > +             return -EINVAL;
> > +     mutex_lock(&class->mutex);
> > +     list_for_each_entry(dev, &class->devices, node) {
> > +             error = fn(dev, data);
>
> Hm, the equivalent _for_each_device() functions all elevate the
> device's refcount while calling fn(). I wonder whether we want this
> here as well?

Thanks for comment.
Hm, I'm not sure about this. Greg, what's your opinion?

>
> > +             if (error)
> > +                     break;
> > +     }
> > +     mutex_unlock(&class->mutex);
> > +
> > +     return error;
> > +}
> > +EXPORT_SYMBOL_GPL(class_for_each_device);
> > +
> > +struct device *class_find_device(struct class *class, void *data,
> > +                                int (*match)(struct device *, void *))
> > +{
> > +     struct device *dev;
> > +
> > +     if (!class)
> > +             return NULL;
> > +
> > +     mutex_lock(&class->mutex);
> > +     list_for_each_entry(dev, &class->devices, node)
> > +             if (match(dev, data) && get_device(dev))
>
> First get the reference and then drop it if the match function returns
> 0 to make this analogous to the other _find_device() functions?

It's just like other _find_device() functions. Are these more get/put
really needed?

>
> > +                     break;
> > +     mutex_unlock(&class->mutex);
> > +
> > +     return dev;
> > +}
> > +EXPORT_SYMBOL_GPL(class_find_device);
> > +
> > +int class_for_each_child(struct class *class, void *data,
> > +                        int (*fn)(struct class_device *, void *))
>
> Haven't looked at the callers, but isn't it better to convert them to
> use struct device instead so we don't need to introduce new
> class_device api?

The drivers/scsi/hosts.c need it.

>
> > +{
> > +     struct class_device *dev;
> > +     int error = 0;
> > +
> > +     if (!class)
> > +             return -EINVAL;
> > +     mutex_lock(&class->mutex);
> > +     list_for_each_entry(dev, &class->children, node) {
> > +             error = fn(dev, data);
>
> Same comment as above concerning reference counts.
>
> > +             if (error)
> > +                     break;
> > +     }
> > +     mutex_unlock(&class->mutex);
> > +
> > +     return error;
> > +}
> > +EXPORT_SYMBOL_GPL(class_for_each_child);
> > +
> > +struct class_device *class_find_child(struct class *class, void *data,
> > +                                int (*match)(struct class_device *, void *))
> > +{
> > +     struct class_device *dev;
> > +
> > +     if (!class)
> > +             return NULL;
> > +
> > +     mutex_lock(&class->mutex);
> > +     list_for_each_entry(dev, &class->children, node)
> > +             if (match(dev, data) && class_device_get(dev))
>
> And here.
>
>
> > +                     break;
> > +     mutex_unlock(&class->mutex);
> > +
> > +     return dev;
> > +}
> > +EXPORT_SYMBOL_GPL(class_find_child);
> > +
>