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

[Dipankar Sarma]

On Thu, Jul 15, 2004 at 11:51:04AM +0530, Ravikiran G Thirumalai wrote:
> Now it would have been nice if we did not have to use cmpxchg 
> for refcount_get_rcu, or atleast if all arches implemented cmpxchg in
> hardware.  Since neither is true, for arches with no cmpxchg, we use
> the hashed spinlock method.  When we use hashed spinlock for 
> refcount_get_rcu method, we need to use the same spinlock to protect the
> refcount for the usual refcount_gets too.  So no atomic_inc for
> refcount_gets on arches with no cmpxchg.  
> Now why refcount_get when you are using refcount_get_rcu for a refcount one 
> might ask. With the fd patch, there are places where the refcount_get 
> happens with traditional serialisation, and places where refcount_get 
> happens lockfree.  So it makes sense for performance sake to have a 
> simple atomic_inc (refcount_get) instead of and the whole cmpxchg shebang
> for the same refcounter when traditional locking is held.  Hence the
> refcount_get infrastructure.  The whole refcount infrastructure 
> provided is _not_ meant for traditinal refcounting ...which kref 
> accomplishes.  If it is used for traditional refcounting (i.e refcount_get
> on a refcounter is used without any refcount_get_rcus for the same
> counter), arches with nocmpxchg will end up using hashed spinlocks for
> simple refcount_gets :).  A Note was included in the header file as
> a warning:

Would this work -

kref_get - just atomic inc
kref_put - just atomic dec
kref_get_rcu - cmpxchg if arch has or hashed spinlock
kref_put_rcu - dec if has cmpxchg else hashed spinlock

If the object has lock-free look-up, it must use kref_get_rcu.
You can use kref_get on such an object if the look-up is being
done with lock. Is there a situation that is not covered by this ?

Thanks
Dipankar
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/