User a per cpu tasklet to schedule rebalancing

Turns out that tasklets have a flag that only allows one instance to run
on all processors. So we need a tasklet structure for each processor.

Signed-off-by: Christoph Lameter <clameter@sgi.com>

Index: linux-2.6.19-rc4-mm2/kernel/sched.c
===================================================================
--- linux-2.6.19-rc4-mm2.orig/kernel/sched.c	2006-11-06 13:58:38.000000000 -0600
+++ linux-2.6.19-rc4-mm2/kernel/sched.c	2006-11-07 13:05:56.236343144 -0600
@@ -2918,7 +2918,8 @@ static void rebalance_domains(unsigned l
 	this_rq->next_balance = next_balance;
 }
 
-DECLARE_TASKLET(rebalance, &rebalance_domains, 0L);
+static DEFINE_PER_CPU(struct tasklet_struct, rebalance) =
+		{ NULL, 0, ATOMIC_INIT(0), rebalance_domains, 0L };
 #else
 /*
  * on UP we do not need to balance between CPUs:
@@ -3171,7 +3172,7 @@ void scheduler_tick(void)
 #ifdef CONFIG_SMP
 	update_load(rq);
 	if (time_after_eq(jiffies, rq->next_balance))
-		tasklet_schedule(&rebalance);
+		tasklet_schedule(&__get_cpu_var(rebalance));
 #endif
 }