]> err.no Git - linux-2.6/blobdiff - kernel/sched_rt.c
sched: clean up schedule_balance_rt()
[linux-2.6] / kernel / sched_rt.c
index ee9c8b6529e9bad0c0b09708713c861e13363675..4d0a60e47dfab5e06405cffbd803a1ededd4708e 100644 (file)
@@ -3,6 +3,52 @@
  * policies)
  */
 
+#ifdef CONFIG_SMP
+
+/*
+ * The "RT overload" flag: it gets set if a CPU has more than
+ * one runnable RT task.
+ */
+static cpumask_t rt_overload_mask;
+static atomic_t rto_count;
+
+static inline int rt_overloaded(void)
+{
+       return atomic_read(&rto_count);
+}
+
+static inline void rt_set_overload(struct rq *rq)
+{
+       rq->rt.overloaded = 1;
+       cpu_set(rq->cpu, rt_overload_mask);
+       /*
+        * Make sure the mask is visible before we set
+        * the overload count. That is checked to determine
+        * if we should look at the mask. It would be a shame
+        * if we looked at the mask, but the mask was not
+        * updated yet.
+        */
+       wmb();
+       atomic_inc(&rto_count);
+}
+
+static inline void rt_clear_overload(struct rq *rq)
+{
+       /* the order here really doesn't matter */
+       atomic_dec(&rto_count);
+       cpu_clear(rq->cpu, rt_overload_mask);
+       rq->rt.overloaded = 0;
+}
+
+static void update_rt_migration(struct rq *rq)
+{
+       if (rq->rt.rt_nr_migratory && (rq->rt.rt_nr_running > 1))
+               rt_set_overload(rq);
+       else
+               rt_clear_overload(rq);
+}
+#endif /* CONFIG_SMP */
+
 /*
  * Update the current task's runtime statistics. Skip current tasks that
  * are not in our scheduling class.
@@ -26,12 +72,54 @@ static void update_curr_rt(struct rq *rq)
        cpuacct_charge(curr, delta_exec);
 }
 
+static inline void inc_rt_tasks(struct task_struct *p, struct rq *rq)
+{
+       WARN_ON(!rt_task(p));
+       rq->rt.rt_nr_running++;
+#ifdef CONFIG_SMP
+       if (p->prio < rq->rt.highest_prio)
+               rq->rt.highest_prio = p->prio;
+       if (p->nr_cpus_allowed > 1)
+               rq->rt.rt_nr_migratory++;
+
+       update_rt_migration(rq);
+#endif /* CONFIG_SMP */
+}
+
+static inline void dec_rt_tasks(struct task_struct *p, struct rq *rq)
+{
+       WARN_ON(!rt_task(p));
+       WARN_ON(!rq->rt.rt_nr_running);
+       rq->rt.rt_nr_running--;
+#ifdef CONFIG_SMP
+       if (rq->rt.rt_nr_running) {
+               struct rt_prio_array *array;
+
+               WARN_ON(p->prio < rq->rt.highest_prio);
+               if (p->prio == rq->rt.highest_prio) {
+                       /* recalculate */
+                       array = &rq->rt.active;
+                       rq->rt.highest_prio =
+                               sched_find_first_bit(array->bitmap);
+               } /* otherwise leave rq->highest prio alone */
+       } else
+               rq->rt.highest_prio = MAX_RT_PRIO;
+       if (p->nr_cpus_allowed > 1)
+               rq->rt.rt_nr_migratory--;
+
+       update_rt_migration(rq);
+#endif /* CONFIG_SMP */
+}
+
 static void enqueue_task_rt(struct rq *rq, struct task_struct *p, int wakeup)
 {
        struct rt_prio_array *array = &rq->rt.active;
 
        list_add_tail(&p->run_list, array->queue + p->prio);
        __set_bit(p->prio, array->bitmap);
+       inc_cpu_load(rq, p->se.load.weight);
+
+       inc_rt_tasks(p, rq);
 }
 
 /*
@@ -46,6 +134,9 @@ static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int sleep)
        list_del(&p->run_list);
        if (list_empty(array->queue + p->prio))
                __clear_bit(p->prio, array->bitmap);
+       dec_cpu_load(rq, p->se.load.weight);
+
+       dec_rt_tasks(p, rq);
 }
 
 /*
@@ -65,6 +156,45 @@ yield_task_rt(struct rq *rq)
        requeue_task_rt(rq, rq->curr);
 }
 
+#ifdef CONFIG_SMP
+static int find_lowest_rq(struct task_struct *task);
+
+static int select_task_rq_rt(struct task_struct *p, int sync)
+{
+       struct rq *rq = task_rq(p);
+
+       /*
+        * If the current task is an RT task, then
+        * try to see if we can wake this RT task up on another
+        * runqueue. Otherwise simply start this RT task
+        * on its current runqueue.
+        *
+        * We want to avoid overloading runqueues. Even if
+        * the RT task is of higher priority than the current RT task.
+        * RT tasks behave differently than other tasks. If
+        * one gets preempted, we try to push it off to another queue.
+        * So trying to keep a preempting RT task on the same
+        * cache hot CPU will force the running RT task to
+        * a cold CPU. So we waste all the cache for the lower
+        * RT task in hopes of saving some of a RT task
+        * that is just being woken and probably will have
+        * cold cache anyway.
+        */
+       if (unlikely(rt_task(rq->curr)) &&
+           (p->nr_cpus_allowed > 1)) {
+               int cpu = find_lowest_rq(p);
+
+               return (cpu == -1) ? task_cpu(p) : cpu;
+       }
+
+       /*
+        * Otherwise, just let it ride on the affined RQ and the
+        * post-schedule router will push the preempted task away
+        */
+       return task_cpu(p);
+}
+#endif /* CONFIG_SMP */
+
 /*
  * Preempt the current task with a newly woken task if needed:
  */
@@ -100,76 +230,508 @@ static void put_prev_task_rt(struct rq *rq, struct task_struct *p)
 }
 
 #ifdef CONFIG_SMP
-/*
- * Load-balancing iterator. Note: while the runqueue stays locked
- * during the whole iteration, the current task might be
- * dequeued so the iterator has to be dequeue-safe. Here we
- * achieve that by always pre-iterating before returning
- * the current task:
- */
-static struct task_struct *load_balance_start_rt(void *arg)
+/* Only try algorithms three times */
+#define RT_MAX_TRIES 3
+
+static int double_lock_balance(struct rq *this_rq, struct rq *busiest);
+static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep);
+
+static int pick_rt_task(struct rq *rq, struct task_struct *p, int cpu)
+{
+       if (!task_running(rq, p) &&
+           (cpu < 0 || cpu_isset(cpu, p->cpus_allowed)) &&
+           (p->nr_cpus_allowed > 1))
+               return 1;
+       return 0;
+}
+
+/* Return the second highest RT task, NULL otherwise */
+static struct task_struct *pick_next_highest_task_rt(struct rq *rq, int cpu)
 {
-       struct rq *rq = arg;
        struct rt_prio_array *array = &rq->rt.active;
-       struct list_head *head, *curr;
-       struct task_struct *p;
+       struct task_struct *next;
+       struct list_head *queue;
        int idx;
 
+       if (likely(rq->rt.rt_nr_running < 2))
+               return NULL;
+
        idx = sched_find_first_bit(array->bitmap);
-       if (idx >= MAX_RT_PRIO)
+       if (unlikely(idx >= MAX_RT_PRIO)) {
+               WARN_ON(1); /* rt_nr_running is bad */
                return NULL;
+       }
 
-       head = array->queue + idx;
-       curr = head->prev;
+       queue = array->queue + idx;
+       BUG_ON(list_empty(queue));
 
-       p = list_entry(curr, struct task_struct, run_list);
+       next = list_entry(queue->next, struct task_struct, run_list);
+       if (unlikely(pick_rt_task(rq, next, cpu)))
+               goto out;
+
+       if (queue->next->next != queue) {
+               /* same prio task */
+               next = list_entry(queue->next->next, struct task_struct,
+                                 run_list);
+               if (pick_rt_task(rq, next, cpu))
+                       goto out;
+       }
 
-       curr = curr->prev;
+ retry:
+       /* slower, but more flexible */
+       idx = find_next_bit(array->bitmap, MAX_RT_PRIO, idx+1);
+       if (unlikely(idx >= MAX_RT_PRIO))
+               return NULL;
 
-       rq->rt.rt_load_balance_idx = idx;
-       rq->rt.rt_load_balance_head = head;
-       rq->rt.rt_load_balance_curr = curr;
+       queue = array->queue + idx;
+       BUG_ON(list_empty(queue));
+
+       list_for_each_entry(next, queue, run_list) {
+               if (pick_rt_task(rq, next, cpu))
+                       goto out;
+       }
 
-       return p;
+       goto retry;
+
+ out:
+       return next;
 }
 
-static struct task_struct *load_balance_next_rt(void *arg)
+static DEFINE_PER_CPU(cpumask_t, local_cpu_mask);
+
+static int find_lowest_cpus(struct task_struct *task, cpumask_t *lowest_mask)
 {
-       struct rq *rq = arg;
-       struct rt_prio_array *array = &rq->rt.active;
-       struct list_head *head, *curr;
-       struct task_struct *p;
-       int idx;
+       int       lowest_prio = -1;
+       int       lowest_cpu  = -1;
+       int       count       = 0;
+       int       cpu;
+
+       cpus_and(*lowest_mask, cpu_online_map, task->cpus_allowed);
+
+       /*
+        * Scan each rq for the lowest prio.
+        */
+       for_each_cpu_mask(cpu, *lowest_mask) {
+               struct rq *rq = cpu_rq(cpu);
+
+               /* We look for lowest RT prio or non-rt CPU */
+               if (rq->rt.highest_prio >= MAX_RT_PRIO) {
+                       /*
+                        * if we already found a low RT queue
+                        * and now we found this non-rt queue
+                        * clear the mask and set our bit.
+                        * Otherwise just return the queue as is
+                        * and the count==1 will cause the algorithm
+                        * to use the first bit found.
+                        */
+                       if (lowest_cpu != -1) {
+                               cpus_clear(*lowest_mask);
+                               cpu_set(rq->cpu, *lowest_mask);
+                       }
+                       return 1;
+               }
+
+               /* no locking for now */
+               if ((rq->rt.highest_prio > task->prio)
+                   && (rq->rt.highest_prio >= lowest_prio)) {
+                       if (rq->rt.highest_prio > lowest_prio) {
+                               /* new low - clear old data */
+                               lowest_prio = rq->rt.highest_prio;
+                               lowest_cpu = cpu;
+                               count = 0;
+                       }
+                       count++;
+               } else
+                       cpu_clear(cpu, *lowest_mask);
+       }
+
+       /*
+        * Clear out all the set bits that represent
+        * runqueues that were of higher prio than
+        * the lowest_prio.
+        */
+       if (lowest_cpu > 0) {
+               /*
+                * Perhaps we could add another cpumask op to
+                * zero out bits. Like cpu_zero_bits(cpumask, nrbits);
+                * Then that could be optimized to use memset and such.
+                */
+               for_each_cpu_mask(cpu, *lowest_mask) {
+                       if (cpu >= lowest_cpu)
+                               break;
+                       cpu_clear(cpu, *lowest_mask);
+               }
+       }
+
+       return count;
+}
 
-       idx = rq->rt.rt_load_balance_idx;
-       head = rq->rt.rt_load_balance_head;
-       curr = rq->rt.rt_load_balance_curr;
+static inline int pick_optimal_cpu(int this_cpu, cpumask_t *mask)
+{
+       int first;
+
+       /* "this_cpu" is cheaper to preempt than a remote processor */
+       if ((this_cpu != -1) && cpu_isset(this_cpu, *mask))
+               return this_cpu;
+
+       first = first_cpu(*mask);
+       if (first != NR_CPUS)
+               return first;
+
+       return -1;
+}
+
+static int find_lowest_rq(struct task_struct *task)
+{
+       struct sched_domain *sd;
+       cpumask_t *lowest_mask = &__get_cpu_var(local_cpu_mask);
+       int this_cpu = smp_processor_id();
+       int cpu      = task_cpu(task);
+       int count    = find_lowest_cpus(task, lowest_mask);
+
+       if (!count)
+               return -1; /* No targets found */
+
+       /*
+        * There is no sense in performing an optimal search if only one
+        * target is found.
+        */
+       if (count == 1)
+               return first_cpu(*lowest_mask);
+
+       /*
+        * At this point we have built a mask of cpus representing the
+        * lowest priority tasks in the system.  Now we want to elect
+        * the best one based on our affinity and topology.
+        *
+        * We prioritize the last cpu that the task executed on since
+        * it is most likely cache-hot in that location.
+        */
+       if (cpu_isset(cpu, *lowest_mask))
+               return cpu;
+
+       /*
+        * Otherwise, we consult the sched_domains span maps to figure
+        * out which cpu is logically closest to our hot cache data.
+        */
+       if (this_cpu == cpu)
+               this_cpu = -1; /* Skip this_cpu opt if the same */
+
+       for_each_domain(cpu, sd) {
+               if (sd->flags & SD_WAKE_AFFINE) {
+                       cpumask_t domain_mask;
+                       int       best_cpu;
+
+                       cpus_and(domain_mask, sd->span, *lowest_mask);
+
+                       best_cpu = pick_optimal_cpu(this_cpu,
+                                                   &domain_mask);
+                       if (best_cpu != -1)
+                               return best_cpu;
+               }
+       }
 
        /*
-        * If we arrived back to the head again then
-        * iterate to the next queue (if any):
+        * And finally, if there were no matches within the domains
+        * just give the caller *something* to work with from the compatible
+        * locations.
         */
-       if (unlikely(head == curr)) {
-               int next_idx = find_next_bit(array->bitmap, MAX_RT_PRIO, idx+1);
+       return pick_optimal_cpu(this_cpu, lowest_mask);
+}
+
+/* Will lock the rq it finds */
+static struct rq *find_lock_lowest_rq(struct task_struct *task, struct rq *rq)
+{
+       struct rq *lowest_rq = NULL;
+       int tries;
+       int cpu;
+
+       for (tries = 0; tries < RT_MAX_TRIES; tries++) {
+               cpu = find_lowest_rq(task);
+
+               if ((cpu == -1) || (cpu == rq->cpu))
+                       break;
+
+               lowest_rq = cpu_rq(cpu);
+
+               /* if the prio of this runqueue changed, try again */
+               if (double_lock_balance(rq, lowest_rq)) {
+                       /*
+                        * We had to unlock the run queue. In
+                        * the mean time, task could have
+                        * migrated already or had its affinity changed.
+                        * Also make sure that it wasn't scheduled on its rq.
+                        */
+                       if (unlikely(task_rq(task) != rq ||
+                                    !cpu_isset(lowest_rq->cpu,
+                                               task->cpus_allowed) ||
+                                    task_running(rq, task) ||
+                                    !task->se.on_rq)) {
+
+                               spin_unlock(&lowest_rq->lock);
+                               lowest_rq = NULL;
+                               break;
+                       }
+               }
+
+               /* If this rq is still suitable use it. */
+               if (lowest_rq->rt.highest_prio > task->prio)
+                       break;
+
+               /* try again */
+               spin_unlock(&lowest_rq->lock);
+               lowest_rq = NULL;
+       }
 
-               if (next_idx >= MAX_RT_PRIO)
-                       return NULL;
+       return lowest_rq;
+}
+
+/*
+ * If the current CPU has more than one RT task, see if the non
+ * running task can migrate over to a CPU that is running a task
+ * of lesser priority.
+ */
+static int push_rt_task(struct rq *rq)
+{
+       struct task_struct *next_task;
+       struct rq *lowest_rq;
+       int ret = 0;
+       int paranoid = RT_MAX_TRIES;
+
+       if (!rq->rt.overloaded)
+               return 0;
+
+       next_task = pick_next_highest_task_rt(rq, -1);
+       if (!next_task)
+               return 0;
+
+ retry:
+       if (unlikely(next_task == rq->curr)) {
+               WARN_ON(1);
+               return 0;
+       }
 
-               idx = next_idx;
-               head = array->queue + idx;
-               curr = head->prev;
+       /*
+        * It's possible that the next_task slipped in of
+        * higher priority than current. If that's the case
+        * just reschedule current.
+        */
+       if (unlikely(next_task->prio < rq->curr->prio)) {
+               resched_task(rq->curr);
+               return 0;
+       }
 
-               rq->rt.rt_load_balance_idx = idx;
-               rq->rt.rt_load_balance_head = head;
+       /* We might release rq lock */
+       get_task_struct(next_task);
+
+       /* find_lock_lowest_rq locks the rq if found */
+       lowest_rq = find_lock_lowest_rq(next_task, rq);
+       if (!lowest_rq) {
+               struct task_struct *task;
+               /*
+                * find lock_lowest_rq releases rq->lock
+                * so it is possible that next_task has changed.
+                * If it has, then try again.
+                */
+               task = pick_next_highest_task_rt(rq, -1);
+               if (unlikely(task != next_task) && task && paranoid--) {
+                       put_task_struct(next_task);
+                       next_task = task;
+                       goto retry;
+               }
+               goto out;
        }
 
-       p = list_entry(curr, struct task_struct, run_list);
+       deactivate_task(rq, next_task, 0);
+       set_task_cpu(next_task, lowest_rq->cpu);
+       activate_task(lowest_rq, next_task, 0);
 
-       curr = curr->prev;
+       resched_task(lowest_rq->curr);
 
-       rq->rt.rt_load_balance_curr = curr;
+       spin_unlock(&lowest_rq->lock);
 
-       return p;
+       ret = 1;
+out:
+       put_task_struct(next_task);
+
+       return ret;
+}
+
+/*
+ * TODO: Currently we just use the second highest prio task on
+ *       the queue, and stop when it can't migrate (or there's
+ *       no more RT tasks).  There may be a case where a lower
+ *       priority RT task has a different affinity than the
+ *       higher RT task. In this case the lower RT task could
+ *       possibly be able to migrate where as the higher priority
+ *       RT task could not.  We currently ignore this issue.
+ *       Enhancements are welcome!
+ */
+static void push_rt_tasks(struct rq *rq)
+{
+       /* push_rt_task will return true if it moved an RT */
+       while (push_rt_task(rq))
+               ;
+}
+
+static int pull_rt_task(struct rq *this_rq)
+{
+       int this_cpu = this_rq->cpu, ret = 0, cpu;
+       struct task_struct *p, *next;
+       struct rq *src_rq;
+
+       /*
+        * If cpusets are used, and we have overlapping
+        * run queue cpusets, then this algorithm may not catch all.
+        * This is just the price you pay on trying to keep
+        * dirtying caches down on large SMP machines.
+        */
+       if (likely(!rt_overloaded()))
+               return 0;
+
+       next = pick_next_task_rt(this_rq);
+
+       for_each_cpu_mask(cpu, rt_overload_mask) {
+               if (this_cpu == cpu)
+                       continue;
+
+               src_rq = cpu_rq(cpu);
+               if (unlikely(src_rq->rt.rt_nr_running <= 1)) {
+                       /*
+                        * It is possible that overlapping cpusets
+                        * will miss clearing a non overloaded runqueue.
+                        * Clear it now.
+                        */
+                       if (double_lock_balance(this_rq, src_rq)) {
+                               /* unlocked our runqueue lock */
+                               struct task_struct *old_next = next;
+
+                               next = pick_next_task_rt(this_rq);
+                               if (next != old_next)
+                                       ret = 1;
+                       }
+                       if (likely(src_rq->rt.rt_nr_running <= 1)) {
+                               /*
+                                * Small chance that this_rq->curr changed
+                                * but it's really harmless here.
+                                */
+                               rt_clear_overload(this_rq);
+                       } else {
+                               /*
+                                * Heh, the src_rq is now overloaded, since
+                                * we already have the src_rq lock, go straight
+                                * to pulling tasks from it.
+                                */
+                               goto try_pulling;
+                       }
+                       spin_unlock(&src_rq->lock);
+                       continue;
+               }
+
+               /*
+                * We can potentially drop this_rq's lock in
+                * double_lock_balance, and another CPU could
+                * steal our next task - hence we must cause
+                * the caller to recalculate the next task
+                * in that case:
+                */
+               if (double_lock_balance(this_rq, src_rq)) {
+                       struct task_struct *old_next = next;
+
+                       next = pick_next_task_rt(this_rq);
+                       if (next != old_next)
+                               ret = 1;
+               }
+
+               /*
+                * Are there still pullable RT tasks?
+                */
+               if (src_rq->rt.rt_nr_running <= 1) {
+                       spin_unlock(&src_rq->lock);
+                       continue;
+               }
+
+ try_pulling:
+               p = pick_next_highest_task_rt(src_rq, this_cpu);
+
+               /*
+                * Do we have an RT task that preempts
+                * the to-be-scheduled task?
+                */
+               if (p && (!next || (p->prio < next->prio))) {
+                       WARN_ON(p == src_rq->curr);
+                       WARN_ON(!p->se.on_rq);
+
+                       /*
+                        * There's a chance that p is higher in priority
+                        * than what's currently running on its cpu.
+                        * This is just that p is wakeing up and hasn't
+                        * had a chance to schedule. We only pull
+                        * p if it is lower in priority than the
+                        * current task on the run queue or
+                        * this_rq next task is lower in prio than
+                        * the current task on that rq.
+                        */
+                       if (p->prio < src_rq->curr->prio ||
+                           (next && next->prio < src_rq->curr->prio))
+                               goto out;
+
+                       ret = 1;
+
+                       deactivate_task(src_rq, p, 0);
+                       set_task_cpu(p, this_cpu);
+                       activate_task(this_rq, p, 0);
+                       /*
+                        * We continue with the search, just in
+                        * case there's an even higher prio task
+                        * in another runqueue. (low likelyhood
+                        * but possible)
+                        *
+                        * Update next so that we won't pick a task
+                        * on another cpu with a priority lower (or equal)
+                        * than the one we just picked.
+                        */
+                       next = p;
+
+               }
+ out:
+               spin_unlock(&src_rq->lock);
+       }
+
+       return ret;
+}
+
+static void schedule_balance_rt(struct rq *rq, struct task_struct *prev)
+{
+       /* Try to pull RT tasks here if we lower this rq's prio */
+       if (unlikely(rt_task(prev)) && rq->rt.highest_prio > prev->prio)
+               pull_rt_task(rq);
+}
+
+static void schedule_tail_balance_rt(struct rq *rq)
+{
+       /*
+        * If we have more than one rt_task queued, then
+        * see if we can push the other rt_tasks off to other CPUS.
+        * Note we may release the rq lock, and since
+        * the lock was owned by prev, we need to release it
+        * first via finish_lock_switch and then reaquire it here.
+        */
+       if (unlikely(rq->rt.overloaded)) {
+               spin_lock_irq(&rq->lock);
+               push_rt_tasks(rq);
+               spin_unlock_irq(&rq->lock);
+       }
+}
+
+
+static void wakeup_balance_rt(struct rq *rq, struct task_struct *p)
+{
+       if (unlikely(rt_task(p)) &&
+           !task_running(rq, p) &&
+           (p->prio >= rq->rt.highest_prio) &&
+           rq->rt.overloaded)
+               push_rt_tasks(rq);
 }
 
 static unsigned long
@@ -178,36 +740,55 @@ load_balance_rt(struct rq *this_rq, int this_cpu, struct rq *busiest,
                struct sched_domain *sd, enum cpu_idle_type idle,
                int *all_pinned, int *this_best_prio)
 {
-       struct rq_iterator rt_rq_iterator;
-
-       rt_rq_iterator.start = load_balance_start_rt;
-       rt_rq_iterator.next = load_balance_next_rt;
-       /* pass 'busiest' rq argument into
-        * load_balance_[start|next]_rt iterators
-        */
-       rt_rq_iterator.arg = busiest;
-
-       return balance_tasks(this_rq, this_cpu, busiest, max_load_move, sd,
-                            idle, all_pinned, this_best_prio, &rt_rq_iterator);
+       /* don't touch RT tasks */
+       return 0;
 }
 
 static int
 move_one_task_rt(struct rq *this_rq, int this_cpu, struct rq *busiest,
                 struct sched_domain *sd, enum cpu_idle_type idle)
 {
-       struct rq_iterator rt_rq_iterator;
+       /* don't touch RT tasks */
+       return 0;
+}
+
+static void set_cpus_allowed_rt(struct task_struct *p, cpumask_t *new_mask)
+{
+       int weight = cpus_weight(*new_mask);
 
-       rt_rq_iterator.start = load_balance_start_rt;
-       rt_rq_iterator.next = load_balance_next_rt;
-       rt_rq_iterator.arg = busiest;
+       BUG_ON(!rt_task(p));
 
-       return iter_move_one_task(this_rq, this_cpu, busiest, sd, idle,
-                                 &rt_rq_iterator);
+       /*
+        * Update the migration status of the RQ if we have an RT task
+        * which is running AND changing its weight value.
+        */
+       if (p->se.on_rq && (weight != p->nr_cpus_allowed)) {
+               struct rq *rq = task_rq(p);
+
+               if ((p->nr_cpus_allowed <= 1) && (weight > 1)) {
+                       rq->rt.rt_nr_migratory++;
+               } else if ((p->nr_cpus_allowed > 1) && (weight <= 1)) {
+                       BUG_ON(!rq->rt.rt_nr_migratory);
+                       rq->rt.rt_nr_migratory--;
+               }
+
+               update_rt_migration(rq);
+       }
+
+       p->cpus_allowed    = *new_mask;
+       p->nr_cpus_allowed = weight;
 }
-#endif
+
+#else /* CONFIG_SMP */
+# define schedule_tail_balance_rt(rq)  do { } while (0)
+# define schedule_balance_rt(rq, prev) do { } while (0)
+# define wakeup_balance_rt(rq, p)      do { } while (0)
+#endif /* CONFIG_SMP */
 
 static void task_tick_rt(struct rq *rq, struct task_struct *p)
 {
+       update_curr_rt(rq);
+
        /*
         * RR tasks need a special form of timeslice management.
         * FIFO tasks have no timeslices.
@@ -242,6 +823,9 @@ const struct sched_class rt_sched_class = {
        .enqueue_task           = enqueue_task_rt,
        .dequeue_task           = dequeue_task_rt,
        .yield_task             = yield_task_rt,
+#ifdef CONFIG_SMP
+       .select_task_rq         = select_task_rq_rt,
+#endif /* CONFIG_SMP */
 
        .check_preempt_curr     = check_preempt_curr_rt,
 
@@ -251,6 +835,7 @@ const struct sched_class rt_sched_class = {
 #ifdef CONFIG_SMP
        .load_balance           = load_balance_rt,
        .move_one_task          = move_one_task_rt,
+       .set_cpus_allowed       = set_cpus_allowed_rt,
 #endif
 
        .set_curr_task          = set_curr_task_rt,