Hi,
Greg KH <greg@xxxxxxxxx> writes:
>> Here's what we have for most UDCs (net2280.c included):
>>
>> struct my_udc {
>> struct gadget gadget;
>> [...]
>> };
>>
>> probe()
>> {
>> struct my_udc *u;
>>
>> u = kzalloc(sizeof(*u), GFP_KERNEL);
>> [...]
>> return 0;
>> }
>>
>> Now, if this kzalloc() would be replaced with devm_kzalloc() wouldn't
>> this result on a functionally equivalent execution to the patch I
>> proposed above?
>>
>> Iff we change struct gadget to contain a struct device *dev instead of a
>> struct device dev, then sure, we will need to cope with proper
>> ->release() implementations.
>>
>> As it is, it brings nothing to the table, IMO.
>
> You always have to have a release function for a kobject, no matter
> where it is, as it is being reference counted. To not do so, is a huge
> indication of a problem in the design.
okay, this goes all the way back to when Dave B wrote the API, it has
always had empty ->release() functions (not all UDCs, though). I'll make
sure to review all of this for v4.13.
--
balbi
Attachment:
signature.asc
Description: PGP signature
