Re: [PATCH] memory ordering in __kfifo primitives

["Paul E. McKenney" <paulmck@us.ibm.com>]

On Thu, Aug 10, 2006 at 04:26:53PM +0200, Stelian Pop wrote:
> Le jeudi 10 août 2006 à 06:41 -0700, Paul E. McKenney a écrit :
> 
> > I am happy to go either way -- the patch with the memory barriers
> > (which does have the side-effect of slowing down kfifo_get() and
> > kfifo_put(), by the way), or a patch removing the comments saying
> > that it is OK to invoke __kfifo_get() and __kfifo_put() without
> > locking.
> > 
> > Any other thoughts on which is better?  (1) the memory barriers or
> > (2) requiring the caller hold appropriate locks across calls to
> > __kfifo_get() and __kfifo_put()?
> 
> If someone wants to use explicit locking, he/she can go with kfifo_get()
> instead of the __ version.

However, the kfifo_get()/kfifo_put() interfaces use the internal lock,
which cannot be used by the caller to protect other code surrounding
the call to kfifo_get()/kfifo_put().  See for example the ISCSI use,
where they have a session->lock that, among other things, protects their
__kfifo_get()/__kfifo_put() calls.

> I'd rather keep the __kfifo_get() and __kfifo_put() functions lockless,
> so I say go for (1) even if there is a tiny price to pay for corectness.

If we require the caller to supply the locks for __kfifo_get() and
__kfifo_put(), then we have -both- correctness -and- better performance.
And the only current user of __kfifo_get()/__kfifo_put() stated that
they could easily expand their session->lock to cover all such calls,
and that doing so would not hurt their performance.

So, are you sure?  And if so, why?

						Thanx, Paul