Make sure we disable preemption in try_to_wake_up and when changing the cpu
in set_cpus_allowed_ptr.

Rework the change in rr_interval with number of cpus to not go up quite so
quickly, scaling only by 50% every doubling of CPUs for better interactivity
on multicore machines. Throughput did not appear to decrease measurably with
this change.

-ck

---
 include/linux/sched.h |    2 +-
 kernel/sched_bfs.c    |   19 ++++++++++++++++---
 2 files changed, 17 insertions(+), 4 deletions(-)

Index: linux-2.6.33-bfs/include/linux/sched.h
===================================================================
--- linux-2.6.33-bfs.orig/include/linux/sched.h	2010-05-18 17:30:14.587299206 +1000
+++ linux-2.6.33-bfs/include/linux/sched.h	2010-05-18 17:30:42.824299835 +1000
@@ -1612,7 +1612,7 @@ static inline void tsk_cpus_current(stru
 
 static inline void print_scheduler_version(void)
 {
-	printk(KERN_INFO"BFS CPU scheduler v0.316 by Con Kolivas.\n");
+	printk(KERN_INFO"BFS CPU scheduler v0.318 by Con Kolivas.\n");
 }
 
 static inline int iso_task(struct task_struct *p)
Index: linux-2.6.33-bfs/kernel/sched_bfs.c
===================================================================
--- linux-2.6.33-bfs.orig/kernel/sched_bfs.c	2010-05-18 17:30:14.562298019 +1000
+++ linux-2.6.33-bfs/kernel/sched_bfs.c	2010-05-18 17:32:19.458298645 +1000
@@ -1260,6 +1260,8 @@ static int try_to_wake_up(struct task_st
 	unsigned long flags;
 	struct rq *rq;
 
+	get_cpu();
+
 	/* This barrier is undocumented, probably for p->state? くそ */
 	smp_wmb();
 
@@ -1294,6 +1296,8 @@ out_running:
 	p->state = TASK_RUNNING;
 out_unlock:
 	task_grq_unlock(&flags);
+	put_cpu();
+
 	return success;
 }
 
@@ -4331,8 +4335,11 @@ int set_cpus_allowed_ptr(struct task_str
 		/* Task is running on the wrong cpu now, reschedule it. */
 		set_tsk_need_resched(p);
 		running_wrong = 1;
-	} else
+	} else {
+		get_cpu();
 		set_task_cpu(p, cpumask_any_and(cpu_active_mask, new_mask));
+		put_cpu();
+	}
 
 out:
 	if (queued)
@@ -6270,7 +6277,7 @@ static int cache_cpu_idle(unsigned long 
 void __init sched_init_smp(void)
 {
 	struct sched_domain *sd;
-	int cpu;
+	int cpu, i, cpu_scale;
 
 	cpumask_var_t non_isolated_cpus;
 
@@ -6309,7 +6316,13 @@ void __init sched_init_smp(void)
 	 * allowing us to increase the base rr_interval, but in a non linear
 	 * fashion.
 	 */
-	rr_interval *= 1 + ilog2(num_online_cpus());
+	cpu_scale = ilog2(num_online_cpus());
+	rr_interval *= 100;
+	for (i = 0; i < cpu_scale; i++) {
+		rr_interval *= 3;
+		rr_interval /= 2;
+	}
+	rr_interval /= 100;
 
 	grq_lock_irq();
 	/*