]> git.karo-electronics.de Git - linux-beck.git/commit
rcu: Define rcu_assign_pointer() in terms of smp_store_release()
authorPaul E. McKenney <paulmck@linux.vnet.ibm.com>
Mon, 16 Dec 2013 21:24:32 +0000 (13:24 -0800)
committerPaul E. McKenney <paulmck@linux.vnet.ibm.com>
Mon, 17 Feb 2014 23:02:00 +0000 (15:02 -0800)
commit88c1863066ccfa456797e12c5d8b4631aa1ad0d0
tree4f456425e3af718957113bd9c4b019b10c496061
parent0adab9b9aa18d7e90337d43567f1eec3d5401b81
rcu: Define rcu_assign_pointer() in terms of smp_store_release()

The new smp_store_release() function provides better guarantees than did
rcu_assign_pointer(), and potentially less overhead on some architectures.
The guarantee that smp_store_release() provides that rcu_assign_pointer()
does that is obscure, but its lack could cause considerable confusion.
This guarantee is illustrated by the following code fragment:

struct foo {
int a;
int b;
int c;
struct foo *next;
};
struct foo foo1;
struct foo foo2;
struct foo __rcu *foop;

...

foo2.a = 1;
foo2.b = 2;
BUG_ON(foo2.c);
rcu_assign_pointer(foop, &foo);

...

fp = rcu_dereference(foop);
fp.c = 3;

The current rcu_assign_pointer() semantics permit the BUG_ON() to
trigger because rcu_assign_pointer()'s smp_wmb() is not guaranteed to
order prior reads against later writes.  This commit therefore upgrades
rcu_assign_pointer() from smp_wmb() to smp_store_release() to avoid this
counter-intuitive outcome.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
include/linux/rcupdate.h