Linux Ethernet Bridging - Nov 2007 - [Bridge] [PATCH] Remove rcu_assign_pointer() penalty for NULL pointers

Hello!

The rcu_assign_pointer() primitive currently unconditionally executes
a memory barrier, even when a NULL pointer is being assigned.  This
has lead some to avoid using rcu_assign_pointer() for NULL pointers,
which loses the self-documenting advantages of rcu_assign_pointer()
This patch uses __builtin_const_p() to omit needless memory barriers
for NULL-pointer assignments at compile time with no runtime penalty,
as discussed in the following thread:

	http://www.mail-archive.com/netdev@vger.kernel.org/msg54852.html

Tested on x86_64 and ppc64, also compiled the four cases (NULL/non-NULL
and const/non-const) with gcc version 4.1.2, and hand-checked the
assembly output.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
---

 rcupdate.h |   11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)

diff -urpNa -X dontdiff linux-2.6.24-rc1-ego/include/linux/rcupdate.h
linux-2.6.24-rc1-egoxu/include/linux/rcupdate.h
--- linux-2.6.24-rc1-ego/include/linux/rcupdate.h	2007-11-06 15:30:02.000000000
-0800
+++ linux-2.6.24-rc1-egoxu/include/linux/rcupdate.h	2007-11-30
09:06:11.000000000 -0800
@@ -191,10 +191,13 @@ static inline void rcu_preempt_boost(voi
  * code.
  */
 
-#define rcu_assign_pointer(p, v)	({ \
-						smp_wmb(); \
-						(p) = (v); \
-					})
+#define rcu_assign_pointer(p, v) \
+	({ \
+		if (!__builtin_constant_p(v) || \
+		    ((v) != NULL)) \
+			smp_wmb(); \
+		(p) = (v); \
+	})
 
 /**
  * synchronize_sched - block until all CPUs have exited any non-preemptive

On Fri, Nov 30, 2007 at 04:37:21PM -0800, Paul E. McKenney
wrote:
> 
> The rcu_assign_pointer() primitive currently unconditionally executes
> a memory barrier, even when a NULL pointer is being assigned.  This
> has lead some to avoid using rcu_assign_pointer() for NULL pointers,
> which loses the self-documenting advantages of rcu_assign_pointer()
> This patch uses __builtin_const_p() to omit needless memory barriers
> for NULL-pointer assignments at compile time with no runtime penalty,
> as discussed in the following thread:
> 
> 	http://www.mail-archive.com/netdev@vger.kernel.org/msg54852.html
> 
> Tested on x86_64 and ppc64, also compiled the four cases (NULL/non-NULL
> and const/non-const) with gcc version 4.1.2, and hand-checked the
> assembly output.
> 
> Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Acked-by: Herbert Xu <herbert@gondor.apana.org.au>

Thanks a lot for following through with this Paul!
-- 
Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt

On Sat, Dec 01, 2007 at 12:07:52PM +1100, Herbert Xu
wrote:
> On Fri, Nov 30, 2007 at 04:37:21PM -0800, Paul E. McKenney wrote:
> > 
> > The rcu_assign_pointer() primitive currently unconditionally executes
> > a memory barrier, even when a NULL pointer is being assigned.  This
> > has lead some to avoid using rcu_assign_pointer() for NULL pointers,
> > which loses the self-documenting advantages of rcu_assign_pointer()
> > This patch uses __builtin_const_p() to omit needless memory barriers
> > for NULL-pointer assignments at compile time with no runtime penalty,
> > as discussed in the following thread:
> > 
> > 	http://www.mail-archive.com/netdev@vger.kernel.org/msg54852.html
> > 
> > Tested on x86_64 and ppc64, also compiled the four cases
(NULL/non-NULL
> > and const/non-const) with gcc version 4.1.2, and hand-checked the
> > assembly output.
> > 
> > Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
> 
> Acked-by: Herbert Xu <herbert@gondor.apana.org.au>
> 
> Thanks a lot for following through with this Paul!
No problem -- after all, it is not every day that one gets the opportunity
to make a simple change that speeds things up and makes kernel hackers
lives a bit simpler.  ;-)

						Thanx, Paul