Re: [RFC] Refcounting of objects part of a lockfree collection

[Dipankar Sarma <dipankar@in.ibm.com>]

On Wed, Jul 14, 2004 at 10:03:37AM -0700, Greg KH wrote:
> On Wed, Jul 14, 2004 at 08:52:35PM +0530, Dipankar Sarma wrote:
> > And I suspect that Andrew thwak me for trying to increase dentry size :)
> > Anyway, the summary is this - Kiran is not trying to introduce
> > a new refcounting API.
> 
> Yes he is.  He's calling it refcount.h, and creating a typedef called
> refcount_t.  Sure looks like a new refcount API to me :)

OK, in terms of code, it is new API. Just the refcounting mechanism
is same as existing non-kref refcounts.

> > He is just adding lock-free support from an existing refcounting
> > mechanism that is used in VFS.
> 
> If this is true, then I strongly object to the naming of this file, and
> the name of the typedef (which shouldn't be a typedef at all) and this
> should be made a private data structure to the vfs so no one else tries
> to use it.  Otherwise it will be used.

I am reasonably sure that when that patch was done (months ago) kref wasn't
there. Now that kref.[ch] is around, everything can be put there.

Now, if struct kref is shrinked (want patch ? ;-), all this 
can possibly be nicely collapsed into one set of APIs for refcounting.
There aren't many users of kref yet, so this seems like a good
time to do it. Was there any objection to shrinking it ?


Thanks
Dipankar