On Thu, Sep 09, 2010 at 05:14:43PM +0200, Arnd Bergmann wrote:

On Tuesday 07 September 2010, Paul E. McKenney wrote:

quoted

On Tue, Sep 07, 2010 at 10:00:58PM +0200, Arnd Bergmann wrote:

quoted

On Tuesday 07 September 2010 21:21:55 Kulikov Vasiliy wrote:

quoted

#define __rcu_dereference_check(p, c, space) \
        ({ \
                typeof(*p) *_________p1 = (typeof(*p)*__force )ACCESS_ONCE(p); \
                                                                   ^
                rcu_lockdep_assert(c); \
                (void) (((typeof (*p) space *)p) == p); \
                                      ^     ^
                smp_read_barrier_depends(); \
                ((typeof(*p) __force __kernel *)(_________p1)); \
        })

If I understand this, it is evaluated three times, right?

Yes, that looks like my own fault, I added that :(

This patch seems to fix it, but I need to think about it some more
to make sure it still does everything we need.

Let me know when you are satisfied with it, and then I will pick it up.

I guess it would be good to put it in now. I haven't had the time
to try out all cases, but the current code in -next is definitely
broken, so please put the fix in now.

Hmmm...  One approach would be have a secondary macro that was:

	#define __rcu_dereference_check_sparse(p, space) \
		(void) (((typeof (*p) space *)p) == p);

when running sparse and:

	#define __rcu_dereference_check_sparse(p, space)

otherwise.

Would that do the trick?

							Thanx, Paul