On Thu, Feb 17, 2022 at 11:04:59AM +0300, Dan Carpenter wrote:

Salah sent a bunch of these.  The reasoning was explained in this email.

https://www.spinics.net/lists/kernel/msg4026672.html

When he resent the patch, Greg said that taking the reference wasn't
needed so the patch wasn't applied.  (Also it had the same reference
leak so that's a second reason it wasn't applied).

Indeed, the kerneldoc for usb_get_intf() does say that each reference 
held by a driver must be refcounted.  And there's nothing wrong with 
doing that, _provided_ you do it correctly.

But if you know the extra refcount will never be needed (because the 
reference will be dropped before the usb_interface in question is 
removed), fiddling with the reference count is unnecessary.  I guess 
whether or not to do it could be considered a matter of taste.

On the other hand, it wouldn't hurt to update the kerneldoc for 
usb_get_intf() (and usb_get_dev() also).  We could point out that if a 
driver does not access the usb_interface structure after its disconnect 
routine returns, incrementing the refcount isn't mandatory.

Greg, any opinion on this?

Alan Stern