[PATCH 3/8] vfio/mdev: simplify mdev_type handling

[Kirti Wankhede]

On 6/7/2022 11:20 AM, Christoph Hellwig wrote:
> On Tue, Jun 07, 2022 at 12:52:49AM +0530, Kirti Wankhede wrote:
>>>    	void (*remove)(struct mdev_device *dev);
>>> -	struct attribute_group **supported_type_groups;
>>>    	struct device_driver driver;
>>>    };
>>
>> mdev_type should be part of mdev_parent, separating it from mdev_parent
>> could result in more errors while using mdev framework.
> 
> Why?
> 
>> Similarly it should
>> be added as part of mdev_register_device. Below adding types is separated
>> from mdev_register_device which is more error prone.
> 
> How so?
> 

Jason has already pointed that about udev event.

>> What if driver
>> registering to mdev doesn't add mdev_types? - mdev framework is un-usable
>> in that case.
> 
> Yes, so it is if you don't add it to the supported_type_groups field
> in the current kernel.  Basic programmer error, and trivially caught.
> 

Current kernel version mandate supported_type_groups, otherwise mdev 
registration fails. But this behavior is being changed with this patch.


>> We had kept it together with mdev registration so that
>> mdev_types should be mandatory to be defined by driver during registration.
>> How would you mandate mdev_type by such separation?
> 
> I would not.  Registering a parent without types is perfectly valid from
> the code correctness perspective.  It just isn't very useful.  Just
> like say creating a kobject without attributes in the device model.

Creating kobject without kobj_type is not allowed in the kernel, 
similarly mdev registration should not be allowed without its type.

Instead of exporting mdev_type_add/mdev_type_remove, these functions 
might be called internally from registration function.

Thanks,
Kirti