Re: [RFC PATCH try2] sysfs: add /sys/dev/{char,block} to lookup sysfs path by major:minor

Re: [RFC PATCH try2] sysfs: add /sys/dev/{char,block} to lookup sysfs path by major:minor

[Kay Sievers]

On Fri, 2008-04-18 at 12:33 -0700, Dan Williams wrote:
> --- a/Documentation/filesystems/sysfs.txt
> +++ b/Documentation/filesystems/sysfs.txt
> @@ -248,6 +248,7 @@ The top level sysfs directory looks like:
>  block/
>  bus/
>  class/
> +dev/
>  devices/

Lots of stuff in that file is pretty damn wrong, and at it should
probably just be deleted. Care to put this new documentation into:
  Documentation/ABI/

> --- a/drivers/base/class.c
> +++ b/drivers/base/class.c
> @@ -134,6 +134,18 @@ static void remove_class_attrs(struct class *cls)
>  	}
>  }
>  
> +static void class_set_dev_kobj(struct class *class)
> +{
> +#ifdef CONFIG_BLOCK
> +	if (class == &block_class) {
> +		class->dev_kobj = block_kobj;
> +		return;
> +	}
> +#endif
> +
> +	class->dev_kobj = char_kobj;
> +}
> +

If you do:
  struct class block_class = {
    .name = "block",
    .dev_kobj = block_kobj,
  }

in:
  block/genhd.c
and put:
  if (!class->dev_kobj)
          class->dev_kobj = char_kobj;
in:
  class_register()
we can get rid of all this logic and the BLOCK #ifdef, right?

Also:
  class_create(THIS_MODULE, "usb_device");
in:
  drivers/usb/core/devio.c
should probably clear the dev_kobj after registration, and we can get
rid of the "parent has the same dev_t" logic?

> --- a/include/linux/device.h
> +++ b/include/linux/device.h
> @@ -192,6 +192,7 @@ struct class {
>  	struct class_attribute		*class_attrs;
>  	struct class_device_attribute	*class_dev_attrs;
>  	struct device_attribute		*dev_attrs;
> +	struct kobject			*dev_kobj;
>  
>  	int (*uevent)(struct class_device *dev, struct kobj_uevent_env *env);
>  	int (*dev_uevent)(struct device *dev, struct kobj_uevent_env *env);
> @@ -204,6 +205,8 @@ struct class {
>  	int (*resume)(struct device *dev);
>  };
>  
> +extern struct kobject *block_kobj;
> +extern struct kobject *char_kobj;

These global symbols should get better names, to make it obvious that
it's "only" the /sys/dev/ stuff.


Thanks,
Kay

Re: [RFC PATCH try2] sysfs: add /sys/dev/{char,block} to lookupsysfs path by major:minor

[Kay Sievers]

On Fri, 2008-04-18 at 17:41 -0700, Dan Williams wrote:
> On Fri, 2008-04-18 at 13:53 -0700, Kay Sievers wrote:
> > On Fri, 2008-04-18 at 12:33 -0700, Dan Williams wrote:
> > Also:
> >   class_create(THIS_MODULE, "usb_device");
> > in:
> >   drivers/usb/core/devio.c
> > should probably clear the dev_kobj after registration, and we can get
> > rid of the "parent has the same dev_t" logic?
> 
> Greg had suspicions that there might be more cases like
> usb_classdev_class, so I added a dev_warn() to help debug those.

Yeah, it's Linux, you never know. :) But there are no currently known
other subsystem doing this. It is perfectly valid scenario though, to
have that, if someone comes up with a better idea as some current stuff
is doing it, and we need to deprecate something, and have duplicated
dev_t entries in sysfs for a while.

> I don't mind taking your word for it that there are no more such cases,
> but maybe the warning should stay in for a few -rc releases?

I think, we can remove that, as it will print a warning for the
"duplicate entry" anyway.

Looks all fine now from my point of view.

Thanks,
Kay