4 * Kernel scheduler and related syscalls
6 * Copyright (C) 1991-2002 Linus Torvalds
8 * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and
9 * make semaphores SMP safe
10 * 1998-11-19 Implemented schedule_timeout() and related stuff
12 * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar:
13 * hybrid priority-list and round-robin design with
14 * an array-switch method of distributing timeslices
15 * and per-CPU runqueues. Cleanups and useful suggestions
16 * by Davide Libenzi, preemptible kernel bits by Robert Love.
17 * 2003-09-03 Interactivity tuning by Con Kolivas.
18 * 2004-04-02 Scheduler domains code by Nick Piggin
21 #include <linux/module.h>
22 #include <linux/nmi.h>
23 #include <linux/init.h>
24 #include <asm/uaccess.h>
25 #include <linux/highmem.h>
26 #include <linux/smp_lock.h>
27 #include <linux/pagemap.h>
28 #include <asm/mmu_context.h>
29 #include <linux/interrupt.h>
30 #include <linux/completion.h>
31 #include <linux/kernel_stat.h>
32 #include <linux/security.h>
33 #include <linux/notifier.h>
34 #include <linux/suspend.h>
35 #include <linux/blkdev.h>
36 #include <linux/delay.h>
37 #include <linux/smp.h>
38 #include <linux/timer.h>
39 #include <linux/rcupdate.h>
40 #include <linux/cpu.h>
41 #include <linux/percpu.h>
42 #include <linux/kthread.h>
44 #include <linux/vserver/sched.h>
45 #include <linux/vs_base.h>
47 #include <asm/unistd.h>
50 #define cpu_to_node_mask(cpu) node_to_cpumask(cpu_to_node(cpu))
52 #define cpu_to_node_mask(cpu) (cpu_online_map)
56 * Convert user-nice values [ -20 ... 0 ... 19 ]
57 * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ],
60 #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20)
61 #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20)
62 #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio)
65 * 'User priority' is the nice value converted to something we
66 * can work with better when scaling various scheduler parameters,
67 * it's a [ 0 ... 39 ] range.
69 #define USER_PRIO(p) ((p)-MAX_RT_PRIO)
70 #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio)
71 #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO))
72 #define AVG_TIMESLICE (MIN_TIMESLICE + ((MAX_TIMESLICE - MIN_TIMESLICE) *\
73 (MAX_PRIO-1-NICE_TO_PRIO(0))/(MAX_USER_PRIO - 1)))
76 * Some helpers for converting nanosecond timing to jiffy resolution
78 #define NS_TO_JIFFIES(TIME) ((TIME) / (1000000000 / HZ))
79 #define JIFFIES_TO_NS(TIME) ((TIME) * (1000000000 / HZ))
82 * These are the 'tuning knobs' of the scheduler:
84 * Minimum timeslice is 10 msecs, default timeslice is 100 msecs,
85 * maximum timeslice is 200 msecs. Timeslices get refilled after
88 #define MIN_TIMESLICE ( 10 * HZ / 1000)
89 #define MAX_TIMESLICE (200 * HZ / 1000)
90 #define ON_RUNQUEUE_WEIGHT 30
91 #define CHILD_PENALTY 95
92 #define PARENT_PENALTY 100
94 #define PRIO_BONUS_RATIO 25
95 #define MAX_BONUS (MAX_USER_PRIO * PRIO_BONUS_RATIO / 100)
96 #define INTERACTIVE_DELTA 2
97 #define MAX_SLEEP_AVG (AVG_TIMESLICE * MAX_BONUS)
98 #define STARVATION_LIMIT (MAX_SLEEP_AVG)
99 #define NS_MAX_SLEEP_AVG (JIFFIES_TO_NS(MAX_SLEEP_AVG))
100 #define CREDIT_LIMIT 100
103 * If a task is 'interactive' then we reinsert it in the active
104 * array after it has expired its current timeslice. (it will not
105 * continue to run immediately, it will still roundrobin with
106 * other interactive tasks.)
108 * This part scales the interactivity limit depending on niceness.
110 * We scale it linearly, offset by the INTERACTIVE_DELTA delta.
111 * Here are a few examples of different nice levels:
113 * TASK_INTERACTIVE(-20): [1,1,1,1,1,1,1,1,1,0,0]
114 * TASK_INTERACTIVE(-10): [1,1,1,1,1,1,1,0,0,0,0]
115 * TASK_INTERACTIVE( 0): [1,1,1,1,0,0,0,0,0,0,0]
116 * TASK_INTERACTIVE( 10): [1,1,0,0,0,0,0,0,0,0,0]
117 * TASK_INTERACTIVE( 19): [0,0,0,0,0,0,0,0,0,0,0]
119 * (the X axis represents the possible -5 ... 0 ... +5 dynamic
120 * priority range a task can explore, a value of '1' means the
121 * task is rated interactive.)
123 * Ie. nice +19 tasks can never get 'interactive' enough to be
124 * reinserted into the active array. And only heavily CPU-hog nice -20
125 * tasks will be expired. Default nice 0 tasks are somewhere between,
126 * it takes some effort for them to get interactive, but it's not
130 #define CURRENT_BONUS(p) \
131 (NS_TO_JIFFIES((p)->sleep_avg) * MAX_BONUS / \
135 #define TIMESLICE_GRANULARITY(p) (MIN_TIMESLICE * \
136 (1 << (((MAX_BONUS - CURRENT_BONUS(p)) ? : 1) - 1)) * \
139 #define TIMESLICE_GRANULARITY(p) (MIN_TIMESLICE * \
140 (1 << (((MAX_BONUS - CURRENT_BONUS(p)) ? : 1) - 1)))
143 #define SCALE(v1,v1_max,v2_max) \
144 (v1) * (v2_max) / (v1_max)
147 (SCALE(TASK_NICE(p), 40, MAX_BONUS) + INTERACTIVE_DELTA)
149 #define TASK_INTERACTIVE(p) \
150 ((p)->prio <= (p)->static_prio - DELTA(p))
152 #define INTERACTIVE_SLEEP(p) \
153 (JIFFIES_TO_NS(MAX_SLEEP_AVG * \
154 (MAX_BONUS / 2 + DELTA((p)) + 1) / MAX_BONUS - 1))
156 #define HIGH_CREDIT(p) \
157 ((p)->interactive_credit > CREDIT_LIMIT)
159 #define LOW_CREDIT(p) \
160 ((p)->interactive_credit < -CREDIT_LIMIT)
163 * BASE_TIMESLICE scales user-nice values [ -20 ... 19 ]
164 * to time slice values.
166 * The higher a thread's priority, the bigger timeslices
167 * it gets during one round of execution. But even the lowest
168 * priority thread gets MIN_TIMESLICE worth of execution time.
170 * task_timeslice() is the interface that is used by the scheduler.
173 #define BASE_TIMESLICE(p) (MIN_TIMESLICE + \
174 ((MAX_TIMESLICE - MIN_TIMESLICE) * \
175 (MAX_PRIO-1 - (p)->static_prio) / (MAX_USER_PRIO-1)))
177 static unsigned int task_timeslice(task_t *p)
179 return BASE_TIMESLICE(p);
182 #define task_hot(p, now, sd) ((now) - (p)->timestamp < (sd)->cache_hot_time)
185 * These are the runqueue data structures:
187 typedef struct runqueue runqueue_t;
189 #ifdef CONFIG_CKRM_CPU_SCHEDULE
190 #include <linux/ckrm_classqueue.h>
193 #ifdef CONFIG_CKRM_CPU_SCHEDULE
196 * if belong to different class, compare class priority
197 * otherwise compare task priority
199 #define TASK_PREEMPTS_CURR(p, rq) \
200 (((p)->cpu_class != (rq)->curr->cpu_class) && ((rq)->curr != (rq)->idle))? class_preempts_curr((p),(rq)->curr) : ((p)->prio < (rq)->curr->prio)
202 #define TASK_PREEMPTS_CURR(p, rq) \
203 ((p)->prio < (rq)->curr->prio)
207 * This is the main, per-CPU runqueue data structure.
209 * Locking rule: those places that want to lock multiple runqueues
210 * (such as the load balancing or the thread migration code), lock
211 * acquire operations must be ordered by ascending &runqueue.
217 * nr_running and cpu_load should be in the same cacheline because
218 * remote CPUs use both these fields when doing load calculation.
220 unsigned long nr_running;
221 #if defined(CONFIG_SMP)
222 unsigned long cpu_load;
224 unsigned long long nr_switches, nr_preempt;
225 unsigned long expired_timestamp, nr_uninterruptible;
226 unsigned long long timestamp_last_tick;
228 struct mm_struct *prev_mm;
229 #ifdef CONFIG_CKRM_CPU_SCHEDULE
230 unsigned long ckrm_cpu_load;
231 struct classqueue_struct classqueue;
233 prio_array_t *active, *expired, arrays[2];
235 int best_expired_prio;
239 struct sched_domain *sd;
241 /* For active balancing */
245 task_t *migration_thread;
246 struct list_head migration_queue;
248 struct list_head hold_queue;
252 static DEFINE_PER_CPU(struct runqueue, runqueues);
254 #define for_each_domain(cpu, domain) \
255 for (domain = cpu_rq(cpu)->sd; domain; domain = domain->parent)
257 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
258 #define this_rq() (&__get_cpu_var(runqueues))
259 #define task_rq(p) cpu_rq(task_cpu(p))
260 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
263 * Default context-switch locking:
265 #ifndef prepare_arch_switch
266 # define prepare_arch_switch(rq, next) do { } while (0)
267 # define finish_arch_switch(rq, next) spin_unlock_irq(&(rq)->lock)
268 # define task_running(rq, p) ((rq)->curr == (p))
271 #ifdef CONFIG_CKRM_CPU_SCHEDULE
272 #include <linux/ckrm_sched.h>
273 spinlock_t cvt_lock = SPIN_LOCK_UNLOCKED;
274 rwlock_t class_list_lock = RW_LOCK_UNLOCKED;
275 LIST_HEAD(active_cpu_classes); // list of active cpu classes; anchor
276 struct ckrm_cpu_class default_cpu_class_obj;
279 * the minimum CVT allowed is the base_cvt
280 * otherwise, it will starve others
282 CVT_t get_min_cvt(int cpu)
285 struct ckrm_local_runqueue * lrq;
288 node = classqueue_get_head(bpt_queue(cpu));
289 lrq = (node) ? class_list_entry(node) : NULL;
292 min_cvt = lrq->local_cvt;
300 * update the classueue base for all the runqueues
301 * TODO: we can only update half of the min_base to solve the movebackward issue
303 static inline void check_update_class_base(int this_cpu) {
304 unsigned long min_base = 0xFFFFFFFF;
308 if (! cpu_online(this_cpu)) return;
311 * find the min_base across all the processors
313 for_each_online_cpu(i) {
315 * I should change it to directly use bpt->base
317 node = classqueue_get_head(bpt_queue(i));
318 if (node && node->prio < min_base) {
319 min_base = node->prio;
322 if (min_base != 0xFFFFFFFF)
323 classqueue_update_base(bpt_queue(this_cpu),min_base);
326 static inline void ckrm_rebalance_tick(int j,int this_cpu)
328 #ifdef CONFIG_CKRM_CPU_SCHEDULE
329 read_lock(&class_list_lock);
330 if (!(j % CVT_UPDATE_TICK))
331 update_global_cvts(this_cpu);
333 #define CKRM_BASE_UPDATE_RATE 400
334 if (! (jiffies % CKRM_BASE_UPDATE_RATE))
335 check_update_class_base(this_cpu);
337 read_unlock(&class_list_lock);
341 static inline struct ckrm_local_runqueue *rq_get_next_class(struct runqueue *rq)
343 cq_node_t *node = classqueue_get_head(&rq->classqueue);
344 return ((node) ? class_list_entry(node) : NULL);
347 static inline struct task_struct * rq_get_next_task(struct runqueue* rq)
350 struct task_struct *next;
351 struct ckrm_local_runqueue *queue;
352 int cpu = smp_processor_id();
356 if ((queue = rq_get_next_class(rq))) {
357 array = queue->active;
358 //check switch active/expired queue
359 if (unlikely(!queue->active->nr_active)) {
360 queue->active = queue->expired;
361 queue->expired = array;
362 queue->expired_timestamp = 0;
364 if (queue->active->nr_active)
365 set_top_priority(queue,
366 find_first_bit(queue->active->bitmap, MAX_PRIO));
368 classqueue_dequeue(queue->classqueue,
369 &queue->classqueue_linkobj);
370 cpu_demand_event(get_rq_local_stat(queue,cpu),CPU_DEMAND_DEQUEUE,0);
373 goto retry_next_class;
375 BUG_ON(!queue->active->nr_active);
376 next = task_list_entry(array->queue[queue->top_priority].next);
381 static inline void rq_load_inc(runqueue_t *rq, struct task_struct *p) { rq->ckrm_cpu_load += cpu_class_weight(p->cpu_class); }
382 static inline void rq_load_dec(runqueue_t *rq, struct task_struct *p) { rq->ckrm_cpu_load -= cpu_class_weight(p->cpu_class); }
384 #else /*CONFIG_CKRM_CPU_SCHEDULE*/
386 static inline struct task_struct * rq_get_next_task(struct runqueue* rq)
389 struct list_head *queue;
393 if (unlikely(!array->nr_active)) {
395 * Switch the active and expired arrays.
397 rq->active = rq->expired;
400 rq->expired_timestamp = 0;
401 rq->best_expired_prio = MAX_PRIO;
404 idx = sched_find_first_bit(array->bitmap);
405 queue = array->queue + idx;
406 return list_entry(queue->next, task_t, run_list);
409 static inline void class_enqueue_task(struct task_struct* p, prio_array_t *array) { }
410 static inline void class_dequeue_task(struct task_struct* p, prio_array_t *array) { }
411 static inline void init_cpu_classes(void) { }
412 static inline void rq_load_inc(runqueue_t *rq, struct task_struct *p) { }
413 static inline void rq_load_dec(runqueue_t *rq, struct task_struct *p) { }
414 #endif /* CONFIG_CKRM_CPU_SCHEDULE */
418 * task_rq_lock - lock the runqueue a given task resides on and disable
419 * interrupts. Note the ordering: we can safely lookup the task_rq without
420 * explicitly disabling preemption.
422 runqueue_t *task_rq_lock(task_t *p, unsigned long *flags)
427 local_irq_save(*flags);
429 spin_lock(&rq->lock);
430 if (unlikely(rq != task_rq(p))) {
431 spin_unlock_irqrestore(&rq->lock, *flags);
432 goto repeat_lock_task;
437 void task_rq_unlock(runqueue_t *rq, unsigned long *flags)
439 spin_unlock_irqrestore(&rq->lock, *flags);
443 * rq_lock - lock a given runqueue and disable interrupts.
445 static runqueue_t *this_rq_lock(void)
451 spin_lock(&rq->lock);
456 static inline void rq_unlock(runqueue_t *rq)
458 spin_unlock_irq(&rq->lock);
462 * Adding/removing a task to/from a priority array:
464 void dequeue_task(struct task_struct *p, prio_array_t *array)
468 list_del(&p->run_list);
469 if (list_empty(array->queue + p->prio))
470 __clear_bit(p->prio, array->bitmap);
471 class_dequeue_task(p,array);
474 void enqueue_task(struct task_struct *p, prio_array_t *array)
476 list_add_tail(&p->run_list, array->queue + p->prio);
477 __set_bit(p->prio, array->bitmap);
480 class_enqueue_task(p,array);
484 * Used by the migration code - we pull tasks from the head of the
485 * remote queue so we want these tasks to show up at the head of the
488 static inline void enqueue_task_head(struct task_struct *p, prio_array_t *array)
490 list_add(&p->run_list, array->queue + p->prio);
491 __set_bit(p->prio, array->bitmap);
494 class_enqueue_task(p,array);
498 * effective_prio - return the priority that is based on the static
499 * priority but is modified by bonuses/penalties.
501 * We scale the actual sleep average [0 .... MAX_SLEEP_AVG]
502 * into the -5 ... 0 ... +5 bonus/penalty range.
504 * We use 25% of the full 0...39 priority range so that:
506 * 1) nice +19 interactive tasks do not preempt nice 0 CPU hogs.
507 * 2) nice -20 CPU hogs do not get preempted by nice 0 tasks.
509 * Both properties are important to certain workloads.
511 static int effective_prio(task_t *p)
518 bonus = CURRENT_BONUS(p) - MAX_BONUS / 2;
520 prio = p->static_prio - bonus;
521 if (__vx_task_flags(p, VXF_SCHED_PRIO, 0))
522 prio += effective_vavavoom(p, MAX_USER_PRIO);
524 if (prio < MAX_RT_PRIO)
526 if (prio > MAX_PRIO-1)
532 * __activate_task - move a task to the runqueue.
534 static inline void __activate_task(task_t *p, runqueue_t *rq)
536 enqueue_task(p, rq_active(p,rq));
542 * __activate_idle_task - move idle task to the _front_ of runqueue.
544 static inline void __activate_idle_task(task_t *p, runqueue_t *rq)
546 enqueue_task_head(p, rq_active(p,rq));
551 static void recalc_task_prio(task_t *p, unsigned long long now)
553 unsigned long long __sleep_time = now - p->timestamp;
554 unsigned long sleep_time;
556 if (__sleep_time > NS_MAX_SLEEP_AVG)
557 sleep_time = NS_MAX_SLEEP_AVG;
559 sleep_time = (unsigned long)__sleep_time;
561 if (likely(sleep_time > 0)) {
563 * User tasks that sleep a long time are categorised as
564 * idle and will get just interactive status to stay active &
565 * prevent them suddenly becoming cpu hogs and starving
568 if (p->mm && p->activated != -1 &&
569 sleep_time > INTERACTIVE_SLEEP(p)) {
570 p->sleep_avg = JIFFIES_TO_NS(MAX_SLEEP_AVG -
573 p->interactive_credit++;
576 * The lower the sleep avg a task has the more
577 * rapidly it will rise with sleep time.
579 sleep_time *= (MAX_BONUS - CURRENT_BONUS(p)) ? : 1;
582 * Tasks with low interactive_credit are limited to
583 * one timeslice worth of sleep avg bonus.
586 sleep_time > JIFFIES_TO_NS(task_timeslice(p)))
587 sleep_time = JIFFIES_TO_NS(task_timeslice(p));
590 * Non high_credit tasks waking from uninterruptible
591 * sleep are limited in their sleep_avg rise as they
592 * are likely to be cpu hogs waiting on I/O
594 if (p->activated == -1 && !HIGH_CREDIT(p) && p->mm) {
595 if (p->sleep_avg >= INTERACTIVE_SLEEP(p))
597 else if (p->sleep_avg + sleep_time >=
598 INTERACTIVE_SLEEP(p)) {
599 p->sleep_avg = INTERACTIVE_SLEEP(p);
605 * This code gives a bonus to interactive tasks.
607 * The boost works by updating the 'average sleep time'
608 * value here, based on ->timestamp. The more time a
609 * task spends sleeping, the higher the average gets -
610 * and the higher the priority boost gets as well.
612 p->sleep_avg += sleep_time;
614 if (p->sleep_avg > NS_MAX_SLEEP_AVG) {
615 p->sleep_avg = NS_MAX_SLEEP_AVG;
617 p->interactive_credit++;
622 p->prio = effective_prio(p);
626 * activate_task - move a task to the runqueue and do priority recalculation
628 * Update all the scheduling statistics stuff. (sleep average
629 * calculation, priority modifiers, etc.)
631 static void activate_task(task_t *p, runqueue_t *rq, int local)
633 unsigned long long now;
638 /* Compensate for drifting sched_clock */
639 runqueue_t *this_rq = this_rq();
640 now = (now - this_rq->timestamp_last_tick)
641 + rq->timestamp_last_tick;
645 recalc_task_prio(p, now);
648 * This checks to make sure it's not an uninterruptible task
649 * that is now waking up.
653 * Tasks which were woken up by interrupts (ie. hw events)
654 * are most likely of interactive nature. So we give them
655 * the credit of extending their sleep time to the period
656 * of time they spend on the runqueue, waiting for execution
657 * on a CPU, first time around:
663 * Normal first-time wakeups get a credit too for
664 * on-runqueue time, but it will be weighted down:
671 __activate_task(p, rq);
675 * deactivate_task - remove a task from the runqueue.
677 static void deactivate_task(struct task_struct *p, runqueue_t *rq)
681 if (p->state == TASK_UNINTERRUPTIBLE)
682 rq->nr_uninterruptible++;
683 dequeue_task(p, p->array);
688 * resched_task - mark a task 'to be rescheduled now'.
690 * On UP this means the setting of the need_resched flag, on SMP it
691 * might also involve a cross-CPU call to trigger the scheduler on
695 static void resched_task(task_t *p)
697 int need_resched, nrpolling;
700 /* minimise the chance of sending an interrupt to poll_idle() */
701 nrpolling = test_tsk_thread_flag(p,TIF_POLLING_NRFLAG);
702 need_resched = test_and_set_tsk_thread_flag(p,TIF_NEED_RESCHED);
703 nrpolling |= test_tsk_thread_flag(p,TIF_POLLING_NRFLAG);
705 if (!need_resched && !nrpolling && (task_cpu(p) != smp_processor_id()))
706 smp_send_reschedule(task_cpu(p));
710 static inline void resched_task(task_t *p)
712 set_tsk_need_resched(p);
717 * task_curr - is this task currently executing on a CPU?
718 * @p: the task in question.
720 inline int task_curr(const task_t *p)
722 return cpu_curr(task_cpu(p)) == p;
732 struct list_head list;
733 enum request_type type;
735 /* For REQ_MOVE_TASK */
739 /* For REQ_SET_DOMAIN */
740 struct sched_domain *sd;
742 struct completion done;
746 * The task's runqueue lock must be held.
747 * Returns true if you have to wait for migration thread.
749 static int migrate_task(task_t *p, int dest_cpu, migration_req_t *req)
751 runqueue_t *rq = task_rq(p);
754 * If the task is not on a runqueue (and not running), then
755 * it is sufficient to simply update the task's cpu field.
757 if (!p->array && !task_running(rq, p)) {
758 set_task_cpu(p, dest_cpu);
762 init_completion(&req->done);
763 req->type = REQ_MOVE_TASK;
765 req->dest_cpu = dest_cpu;
766 list_add(&req->list, &rq->migration_queue);
771 * wait_task_inactive - wait for a thread to unschedule.
773 * The caller must ensure that the task *will* unschedule sometime soon,
774 * else this function might spin for a *long* time. This function can't
775 * be called with interrupts off, or it may introduce deadlock with
776 * smp_call_function() if an IPI is sent by the same process we are
777 * waiting to become inactive.
779 void wait_task_inactive(task_t * p)
786 rq = task_rq_lock(p, &flags);
787 /* Must be off runqueue entirely, not preempted. */
788 if (unlikely(p->array)) {
789 /* If it's preempted, we yield. It could be a while. */
790 preempted = !task_running(rq, p);
791 task_rq_unlock(rq, &flags);
797 task_rq_unlock(rq, &flags);
801 * kick_process - kick a running thread to enter/exit the kernel
802 * @p: the to-be-kicked thread
804 * Cause a process which is running on another CPU to enter
805 * kernel-mode, without any delay. (to get signals handled.)
807 void kick_process(task_t *p)
813 if ((cpu != smp_processor_id()) && task_curr(p))
814 smp_send_reschedule(cpu);
818 EXPORT_SYMBOL_GPL(kick_process);
821 * Return a low guess at the load of a migration-source cpu.
823 * We want to under-estimate the load of migration sources, to
824 * balance conservatively.
826 static inline unsigned long source_load(int cpu)
828 runqueue_t *rq = cpu_rq(cpu);
829 unsigned long load_now = rq->nr_running * SCHED_LOAD_SCALE;
831 return min(rq->cpu_load, load_now);
835 * Return a high guess at the load of a migration-target cpu
837 static inline unsigned long target_load(int cpu)
839 runqueue_t *rq = cpu_rq(cpu);
840 unsigned long load_now = rq->nr_running * SCHED_LOAD_SCALE;
842 return max(rq->cpu_load, load_now);
848 * wake_idle() is useful especially on SMT architectures to wake a
849 * task onto an idle sibling if we would otherwise wake it onto a
852 * Returns the CPU we should wake onto.
854 #if defined(ARCH_HAS_SCHED_WAKE_IDLE)
855 static int wake_idle(int cpu, task_t *p)
858 runqueue_t *rq = cpu_rq(cpu);
859 struct sched_domain *sd;
866 if (!(sd->flags & SD_WAKE_IDLE))
869 cpus_and(tmp, sd->span, cpu_online_map);
870 cpus_and(tmp, tmp, p->cpus_allowed);
872 for_each_cpu_mask(i, tmp) {
880 static inline int wake_idle(int cpu, task_t *p)
887 * try_to_wake_up - wake up a thread
888 * @p: the to-be-woken-up thread
889 * @state: the mask of task states that can be woken
890 * @sync: do a synchronous wakeup?
892 * Put it on the run-queue if it's not already there. The "current"
893 * thread is always on the run-queue (except when the actual
894 * re-schedule is in progress), and as such you're allowed to do
895 * the simpler "current->state = TASK_RUNNING" to mark yourself
896 * runnable without the overhead of this.
898 * returns failure only if the task is already active.
900 static int try_to_wake_up(task_t * p, unsigned int state, int sync)
902 int cpu, this_cpu, success = 0;
907 unsigned long load, this_load;
908 struct sched_domain *sd;
912 rq = task_rq_lock(p, &flags);
913 old_state = p->state;
914 if (!(old_state & state))
921 this_cpu = smp_processor_id();
924 if (unlikely(task_running(rq, p)))
929 if (cpu == this_cpu || unlikely(!cpu_isset(this_cpu, p->cpus_allowed)))
932 load = source_load(cpu);
933 this_load = target_load(this_cpu);
936 * If sync wakeup then subtract the (maximum possible) effect of
937 * the currently running task from the load of the current CPU:
940 this_load -= SCHED_LOAD_SCALE;
942 /* Don't pull the task off an idle CPU to a busy one */
943 if (load < SCHED_LOAD_SCALE/2 && this_load > SCHED_LOAD_SCALE/2)
946 new_cpu = this_cpu; /* Wake to this CPU if we can */
949 * Scan domains for affine wakeup and passive balancing
952 for_each_domain(this_cpu, sd) {
953 unsigned int imbalance;
955 * Start passive balancing when half the imbalance_pct
958 imbalance = sd->imbalance_pct + (sd->imbalance_pct - 100) / 2;
960 if ( ((sd->flags & SD_WAKE_AFFINE) &&
961 !task_hot(p, rq->timestamp_last_tick, sd))
962 || ((sd->flags & SD_WAKE_BALANCE) &&
963 imbalance*this_load <= 100*load) ) {
965 * Now sd has SD_WAKE_AFFINE and p is cache cold in sd
966 * or sd has SD_WAKE_BALANCE and there is an imbalance
968 if (cpu_isset(cpu, sd->span))
973 new_cpu = cpu; /* Could not wake to this_cpu. Wake to cpu instead */
975 new_cpu = wake_idle(new_cpu, p);
976 if (new_cpu != cpu && cpu_isset(new_cpu, p->cpus_allowed)) {
977 set_task_cpu(p, new_cpu);
978 task_rq_unlock(rq, &flags);
979 /* might preempt at this point */
980 rq = task_rq_lock(p, &flags);
981 old_state = p->state;
982 if (!(old_state & state))
987 this_cpu = smp_processor_id();
992 #endif /* CONFIG_SMP */
993 if (old_state == TASK_UNINTERRUPTIBLE) {
994 rq->nr_uninterruptible--;
996 * Tasks on involuntary sleep don't earn
997 * sleep_avg beyond just interactive state.
1003 * Sync wakeups (i.e. those types of wakeups where the waker
1004 * has indicated that it will leave the CPU in short order)
1005 * don't trigger a preemption, if the woken up task will run on
1006 * this cpu. (in this case the 'I will reschedule' promise of
1007 * the waker guarantees that the freshly woken up task is going
1008 * to be considered on this CPU.)
1010 activate_task(p, rq, cpu == this_cpu);
1011 if (!sync || cpu != this_cpu) {
1012 if (TASK_PREEMPTS_CURR(p, rq))
1013 resched_task(rq->curr);
1018 p->state = TASK_RUNNING;
1020 task_rq_unlock(rq, &flags);
1025 int fastcall wake_up_process(task_t * p)
1027 return try_to_wake_up(p, TASK_STOPPED |
1028 TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE, 0);
1031 EXPORT_SYMBOL(wake_up_process);
1033 int fastcall wake_up_state(task_t *p, unsigned int state)
1035 return try_to_wake_up(p, state, 0);
1039 * Perform scheduler related setup for a newly forked process p.
1040 * p is forked by current.
1042 void fastcall sched_fork(task_t *p)
1045 * We mark the process as running here, but have not actually
1046 * inserted it onto the runqueue yet. This guarantees that
1047 * nobody will actually run it, and a signal or other external
1048 * event cannot wake it up and insert it on the runqueue either.
1050 p->state = TASK_RUNNING;
1051 INIT_LIST_HEAD(&p->run_list);
1053 spin_lock_init(&p->switch_lock);
1054 #ifdef CONFIG_PREEMPT
1056 * During context-switch we hold precisely one spinlock, which
1057 * schedule_tail drops. (in the common case it's this_rq()->lock,
1058 * but it also can be p->switch_lock.) So we compensate with a count
1059 * of 1. Also, we want to start with kernel preemption disabled.
1061 p->thread_info->preempt_count = 1;
1064 * Share the timeslice between parent and child, thus the
1065 * total amount of pending timeslices in the system doesn't change,
1066 * resulting in more scheduling fairness.
1068 local_irq_disable();
1069 p->time_slice = (current->time_slice + 1) >> 1;
1071 * The remainder of the first timeslice might be recovered by
1072 * the parent if the child exits early enough.
1074 p->first_time_slice = 1;
1075 current->time_slice >>= 1;
1076 p->timestamp = sched_clock();
1077 if (!current->time_slice) {
1079 * This case is rare, it happens when the parent has only
1080 * a single jiffy left from its timeslice. Taking the
1081 * runqueue lock is not a problem.
1083 current->time_slice = 1;
1085 scheduler_tick(0, 0);
1093 * wake_up_forked_process - wake up a freshly forked process.
1095 * This function will do some initial scheduler statistics housekeeping
1096 * that must be done for every newly created process.
1098 void fastcall wake_up_forked_process(task_t * p)
1100 unsigned long flags;
1101 runqueue_t *rq = task_rq_lock(current, &flags);
1103 BUG_ON(p->state != TASK_RUNNING);
1106 * We decrease the sleep average of forking parents
1107 * and children as well, to keep max-interactive tasks
1108 * from forking tasks that are max-interactive.
1110 current->sleep_avg = JIFFIES_TO_NS(CURRENT_BONUS(current) *
1111 PARENT_PENALTY / 100 * MAX_SLEEP_AVG / MAX_BONUS);
1113 p->sleep_avg = JIFFIES_TO_NS(CURRENT_BONUS(p) *
1114 CHILD_PENALTY / 100 * MAX_SLEEP_AVG / MAX_BONUS);
1116 p->interactive_credit = 0;
1118 p->prio = effective_prio(p);
1119 set_task_cpu(p, smp_processor_id());
1121 if (unlikely(!current->array))
1122 __activate_task(p, rq);
1124 p->prio = current->prio;
1125 list_add_tail(&p->run_list, ¤t->run_list);
1126 p->array = current->array;
1127 p->array->nr_active++;
1131 task_rq_unlock(rq, &flags);
1135 * Potentially available exiting-child timeslices are
1136 * retrieved here - this way the parent does not get
1137 * penalized for creating too many threads.
1139 * (this cannot be used to 'generate' timeslices
1140 * artificially, because any timeslice recovered here
1141 * was given away by the parent in the first place.)
1143 void fastcall sched_exit(task_t * p)
1145 unsigned long flags;
1148 local_irq_save(flags);
1149 if (p->first_time_slice) {
1150 p->parent->time_slice += p->time_slice;
1151 if (unlikely(p->parent->time_slice > MAX_TIMESLICE))
1152 p->parent->time_slice = MAX_TIMESLICE;
1154 local_irq_restore(flags);
1156 * If the child was a (relative-) CPU hog then decrease
1157 * the sleep_avg of the parent as well.
1159 rq = task_rq_lock(p->parent, &flags);
1160 if (p->sleep_avg < p->parent->sleep_avg)
1161 p->parent->sleep_avg = p->parent->sleep_avg /
1162 (EXIT_WEIGHT + 1) * EXIT_WEIGHT + p->sleep_avg /
1164 task_rq_unlock(rq, &flags);
1168 * finish_task_switch - clean up after a task-switch
1169 * @prev: the thread we just switched away from.
1171 * We enter this with the runqueue still locked, and finish_arch_switch()
1172 * will unlock it along with doing any other architecture-specific cleanup
1175 * Note that we may have delayed dropping an mm in context_switch(). If
1176 * so, we finish that here outside of the runqueue lock. (Doing it
1177 * with the lock held can cause deadlocks; see schedule() for
1180 static void finish_task_switch(task_t *prev)
1182 runqueue_t *rq = this_rq();
1183 struct mm_struct *mm = rq->prev_mm;
1184 unsigned long prev_task_flags;
1189 * A task struct has one reference for the use as "current".
1190 * If a task dies, then it sets TASK_ZOMBIE in tsk->state and calls
1191 * schedule one last time. The schedule call will never return,
1192 * and the scheduled task must drop that reference.
1193 * The test for TASK_ZOMBIE must occur while the runqueue locks are
1194 * still held, otherwise prev could be scheduled on another cpu, die
1195 * there before we look at prev->state, and then the reference would
1197 * Manfred Spraul <manfred@colorfullife.com>
1199 prev_task_flags = prev->flags;
1200 finish_arch_switch(rq, prev);
1203 if (unlikely(prev_task_flags & PF_DEAD))
1204 put_task_struct(prev);
1208 * schedule_tail - first thing a freshly forked thread must call.
1209 * @prev: the thread we just switched away from.
1211 asmlinkage void schedule_tail(task_t *prev)
1213 finish_task_switch(prev);
1215 if (current->set_child_tid)
1216 put_user(current->pid, current->set_child_tid);
1220 * context_switch - switch to the new MM and the new
1221 * thread's register state.
1224 task_t * context_switch(runqueue_t *rq, task_t *prev, task_t *next)
1226 struct mm_struct *mm = next->mm;
1227 struct mm_struct *oldmm = prev->active_mm;
1229 if (unlikely(!mm)) {
1230 next->active_mm = oldmm;
1231 atomic_inc(&oldmm->mm_count);
1232 enter_lazy_tlb(oldmm, next);
1234 switch_mm(oldmm, mm, next);
1236 if (unlikely(!prev->mm)) {
1237 prev->active_mm = NULL;
1238 WARN_ON(rq->prev_mm);
1239 rq->prev_mm = oldmm;
1242 /* Here we just switch the register state and the stack. */
1243 switch_to(prev, next, prev);
1249 * nr_running, nr_uninterruptible and nr_context_switches:
1251 * externally visible scheduler statistics: current number of runnable
1252 * threads, current number of uninterruptible-sleeping threads, total
1253 * number of context switches performed since bootup.
1255 unsigned long nr_running(void)
1257 unsigned long i, sum = 0;
1260 sum += cpu_rq(i)->nr_running;
1265 unsigned long nr_uninterruptible(void)
1267 unsigned long i, sum = 0;
1269 for_each_online_cpu(i)
1270 sum += cpu_rq(i)->nr_uninterruptible;
1275 unsigned long long nr_context_switches(void)
1277 unsigned long long i, sum = 0;
1279 for_each_online_cpu(i)
1280 sum += cpu_rq(i)->nr_switches;
1285 unsigned long nr_iowait(void)
1287 unsigned long i, sum = 0;
1289 for_each_online_cpu(i)
1290 sum += atomic_read(&cpu_rq(i)->nr_iowait);
1296 * double_rq_lock - safely lock two runqueues
1298 * Note this does not disable interrupts like task_rq_lock,
1299 * you need to do so manually before calling.
1301 static void double_rq_lock(runqueue_t *rq1, runqueue_t *rq2)
1304 spin_lock(&rq1->lock);
1307 spin_lock(&rq1->lock);
1308 spin_lock(&rq2->lock);
1310 spin_lock(&rq2->lock);
1311 spin_lock(&rq1->lock);
1317 * double_rq_unlock - safely unlock two runqueues
1319 * Note this does not restore interrupts like task_rq_unlock,
1320 * you need to do so manually after calling.
1322 static void double_rq_unlock(runqueue_t *rq1, runqueue_t *rq2)
1324 spin_unlock(&rq1->lock);
1326 spin_unlock(&rq2->lock);
1329 unsigned long long nr_preempt(void)
1331 unsigned long long i, sum = 0;
1333 for_each_online_cpu(i)
1334 sum += cpu_rq(i)->nr_preempt;
1349 * find_idlest_cpu - find the least busy runqueue.
1351 static int find_idlest_cpu(struct task_struct *p, int this_cpu,
1352 struct sched_domain *sd)
1354 unsigned long load, min_load, this_load;
1359 min_load = ULONG_MAX;
1361 cpus_and(mask, sd->span, cpu_online_map);
1362 cpus_and(mask, mask, p->cpus_allowed);
1364 for_each_cpu_mask(i, mask) {
1365 load = target_load(i);
1367 if (load < min_load) {
1371 /* break out early on an idle CPU: */
1377 /* add +1 to account for the new task */
1378 this_load = source_load(this_cpu) + SCHED_LOAD_SCALE;
1381 * Would with the addition of the new task to the
1382 * current CPU there be an imbalance between this
1383 * CPU and the idlest CPU?
1385 * Use half of the balancing threshold - new-context is
1386 * a good opportunity to balance.
1388 if (min_load*(100 + (sd->imbalance_pct-100)/2) < this_load*100)
1395 * wake_up_forked_thread - wake up a freshly forked thread.
1397 * This function will do some initial scheduler statistics housekeeping
1398 * that must be done for every newly created context, and it also does
1399 * runqueue balancing.
1401 void fastcall wake_up_forked_thread(task_t * p)
1403 unsigned long flags;
1404 int this_cpu = get_cpu(), cpu;
1405 struct sched_domain *tmp, *sd = NULL;
1406 runqueue_t *this_rq = cpu_rq(this_cpu), *rq;
1409 * Find the largest domain that this CPU is part of that
1410 * is willing to balance on clone:
1412 for_each_domain(this_cpu, tmp)
1413 if (tmp->flags & SD_BALANCE_CLONE)
1416 cpu = find_idlest_cpu(p, this_cpu, sd);
1420 local_irq_save(flags);
1423 double_rq_lock(this_rq, rq);
1425 BUG_ON(p->state != TASK_RUNNING);
1428 * We did find_idlest_cpu() unlocked, so in theory
1429 * the mask could have changed - just dont migrate
1432 if (unlikely(!cpu_isset(cpu, p->cpus_allowed))) {
1434 double_rq_unlock(this_rq, rq);
1438 * We decrease the sleep average of forking parents
1439 * and children as well, to keep max-interactive tasks
1440 * from forking tasks that are max-interactive.
1442 current->sleep_avg = JIFFIES_TO_NS(CURRENT_BONUS(current) *
1443 PARENT_PENALTY / 100 * MAX_SLEEP_AVG / MAX_BONUS);
1445 p->sleep_avg = JIFFIES_TO_NS(CURRENT_BONUS(p) *
1446 CHILD_PENALTY / 100 * MAX_SLEEP_AVG / MAX_BONUS);
1448 p->interactive_credit = 0;
1450 p->prio = effective_prio(p);
1451 set_task_cpu(p, cpu);
1453 if (cpu == this_cpu) {
1454 if (unlikely(!current->array))
1455 __activate_task(p, rq);
1457 p->prio = current->prio;
1458 list_add_tail(&p->run_list, ¤t->run_list);
1459 p->array = current->array;
1460 p->array->nr_active++;
1465 /* Not the local CPU - must adjust timestamp */
1466 p->timestamp = (p->timestamp - this_rq->timestamp_last_tick)
1467 + rq->timestamp_last_tick;
1468 __activate_task(p, rq);
1469 if (TASK_PREEMPTS_CURR(p, rq))
1470 resched_task(rq->curr);
1473 double_rq_unlock(this_rq, rq);
1474 local_irq_restore(flags);
1479 * If dest_cpu is allowed for this process, migrate the task to it.
1480 * This is accomplished by forcing the cpu_allowed mask to only
1481 * allow dest_cpu, which will force the cpu onto dest_cpu. Then
1482 * the cpu_allowed mask is restored.
1484 static void sched_migrate_task(task_t *p, int dest_cpu)
1486 migration_req_t req;
1488 unsigned long flags;
1490 rq = task_rq_lock(p, &flags);
1491 if (!cpu_isset(dest_cpu, p->cpus_allowed)
1492 || unlikely(cpu_is_offline(dest_cpu)))
1495 /* force the process onto the specified CPU */
1496 if (migrate_task(p, dest_cpu, &req)) {
1497 /* Need to wait for migration thread (might exit: take ref). */
1498 struct task_struct *mt = rq->migration_thread;
1499 get_task_struct(mt);
1500 task_rq_unlock(rq, &flags);
1501 wake_up_process(mt);
1502 put_task_struct(mt);
1503 wait_for_completion(&req.done);
1507 task_rq_unlock(rq, &flags);
1511 * sched_balance_exec(): find the highest-level, exec-balance-capable
1512 * domain and try to migrate the task to the least loaded CPU.
1514 * execve() is a valuable balancing opportunity, because at this point
1515 * the task has the smallest effective memory and cache footprint.
1517 void sched_balance_exec(void)
1519 struct sched_domain *tmp, *sd = NULL;
1520 int new_cpu, this_cpu = get_cpu();
1522 /* Prefer the current CPU if there's only this task running */
1523 if (this_rq()->nr_running <= 1)
1526 for_each_domain(this_cpu, tmp)
1527 if (tmp->flags & SD_BALANCE_EXEC)
1531 new_cpu = find_idlest_cpu(current, this_cpu, sd);
1532 if (new_cpu != this_cpu) {
1534 sched_migrate_task(current, new_cpu);
1543 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1545 static void double_lock_balance(runqueue_t *this_rq, runqueue_t *busiest)
1547 if (unlikely(!spin_trylock(&busiest->lock))) {
1548 if (busiest < this_rq) {
1549 spin_unlock(&this_rq->lock);
1550 spin_lock(&busiest->lock);
1551 spin_lock(&this_rq->lock);
1553 spin_lock(&busiest->lock);
1558 * pull_task - move a task from a remote runqueue to the local runqueue.
1559 * Both runqueues must be locked.
1562 void pull_task(runqueue_t *src_rq, prio_array_t *src_array, task_t *p,
1563 runqueue_t *this_rq, prio_array_t *this_array, int this_cpu)
1565 dequeue_task(p, src_array);
1566 src_rq->nr_running--;
1567 rq_load_dec(src_rq,p);
1569 set_task_cpu(p, this_cpu);
1570 this_rq->nr_running++;
1571 rq_load_inc(this_rq,p);
1572 enqueue_task(p, this_array);
1574 p->timestamp = (p->timestamp - src_rq->timestamp_last_tick)
1575 + this_rq->timestamp_last_tick;
1577 * Note that idle threads have a prio of MAX_PRIO, for this test
1578 * to be always true for them.
1580 if (TASK_PREEMPTS_CURR(p, this_rq))
1581 resched_task(this_rq->curr);
1585 * can_migrate_task - may task p from runqueue rq be migrated to this_cpu?
1588 int can_migrate_task(task_t *p, runqueue_t *rq, int this_cpu,
1589 struct sched_domain *sd, enum idle_type idle)
1592 * We do not migrate tasks that are:
1593 * 1) running (obviously), or
1594 * 2) cannot be migrated to this CPU due to cpus_allowed, or
1595 * 3) are cache-hot on their current CPU.
1597 if (task_running(rq, p))
1599 if (!cpu_isset(this_cpu, p->cpus_allowed))
1602 /* Aggressive migration if we've failed balancing */
1603 if (idle == NEWLY_IDLE ||
1604 sd->nr_balance_failed < sd->cache_nice_tries) {
1605 if (task_hot(p, rq->timestamp_last_tick, sd))
1612 #ifdef CONFIG_CKRM_CPU_SCHEDULE
1614 struct ckrm_cpu_class *find_unbalanced_class(int busiest_cpu, int this_cpu, unsigned long *cls_imbalance)
1616 struct ckrm_cpu_class *most_unbalanced_class = NULL;
1617 struct ckrm_cpu_class *clsptr;
1618 int max_unbalance = 0;
1620 list_for_each_entry(clsptr,&active_cpu_classes,links) {
1621 struct ckrm_local_runqueue *this_lrq = get_ckrm_local_runqueue(clsptr,this_cpu);
1622 struct ckrm_local_runqueue *busiest_lrq = get_ckrm_local_runqueue(clsptr,busiest_cpu);
1623 int unbalance_degree;
1625 unbalance_degree = (local_queue_nr_running(busiest_lrq) - local_queue_nr_running(this_lrq)) * cpu_class_weight(clsptr);
1626 if (unbalance_degree >= *cls_imbalance)
1627 continue; // already looked at this class
1629 if (unbalance_degree > max_unbalance) {
1630 max_unbalance = unbalance_degree;
1631 most_unbalanced_class = clsptr;
1634 *cls_imbalance = max_unbalance;
1635 return most_unbalanced_class;
1640 * find_busiest_queue - find the busiest runqueue among the cpus in cpumask.
1642 static int find_busiest_cpu(runqueue_t *this_rq, int this_cpu, int idle,
1645 int cpu_load, load, max_load, i, busiest_cpu;
1646 runqueue_t *busiest, *rq_src;
1649 /*Hubertus ... the concept of nr_running is replace with cpu_load */
1650 cpu_load = this_rq->ckrm_cpu_load;
1656 for_each_online_cpu(i) {
1658 load = rq_src->ckrm_cpu_load;
1660 if ((load > max_load) && (rq_src != this_rq)) {
1667 if (likely(!busiest))
1670 *imbalance = max_load - cpu_load;
1672 /* It needs an at least ~25% imbalance to trigger balancing. */
1673 if (!idle && ((*imbalance)*4 < max_load)) {
1678 double_lock_balance(this_rq, busiest);
1680 * Make sure nothing changed since we checked the
1683 if (busiest->ckrm_cpu_load <= cpu_load) {
1684 spin_unlock(&busiest->lock);
1688 return (busiest ? busiest_cpu : -1);
1691 static int load_balance(int this_cpu, runqueue_t *this_rq,
1692 struct sched_domain *sd, enum idle_type idle)
1696 runqueue_t *busiest;
1697 prio_array_t *array;
1698 struct list_head *head, *curr;
1700 struct ckrm_local_runqueue * busiest_local_queue;
1701 struct ckrm_cpu_class *clsptr;
1703 unsigned long cls_imbalance; // so we can retry other classes
1705 // need to update global CVT based on local accumulated CVTs
1706 read_lock(&class_list_lock);
1707 busiest_cpu = find_busiest_cpu(this_rq, this_cpu, idle, &imbalance);
1708 if (busiest_cpu == -1)
1711 busiest = cpu_rq(busiest_cpu);
1714 * We only want to steal a number of tasks equal to 1/2 the imbalance,
1715 * otherwise we'll just shift the imbalance to the new queue:
1719 /* now find class on that runqueue with largest inbalance */
1720 cls_imbalance = 0xFFFFFFFF;
1723 clsptr = find_unbalanced_class(busiest_cpu, this_cpu, &cls_imbalance);
1727 busiest_local_queue = get_ckrm_local_runqueue(clsptr,busiest_cpu);
1728 weight = cpu_class_weight(clsptr);
1731 * We first consider expired tasks. Those will likely not be
1732 * executed in the near future, and they are most likely to
1733 * be cache-cold, thus switching CPUs has the least effect
1736 if (busiest_local_queue->expired->nr_active)
1737 array = busiest_local_queue->expired;
1739 array = busiest_local_queue->active;
1742 /* Start searching at priority 0: */
1746 idx = sched_find_first_bit(array->bitmap);
1748 idx = find_next_bit(array->bitmap, MAX_PRIO, idx);
1749 if (idx >= MAX_PRIO) {
1750 if (array == busiest_local_queue->expired && busiest_local_queue->active->nr_active) {
1751 array = busiest_local_queue->active;
1754 goto retry_other_class;
1757 head = array->queue + idx;
1760 tmp = list_entry(curr, task_t, run_list);
1764 if (!can_migrate_task(tmp, busiest, this_cpu, sd,idle)) {
1770 pull_task(busiest, array, tmp, this_rq, rq_active(tmp,this_rq),this_cpu);
1772 * tmp BUG FIX: hzheng
1773 * load balancing can make the busiest local queue empty
1774 * thus it should be removed from bpt
1776 if (! local_queue_nr_running(busiest_local_queue)) {
1777 classqueue_dequeue(busiest_local_queue->classqueue,&busiest_local_queue->classqueue_linkobj);
1778 cpu_demand_event(get_rq_local_stat(busiest_local_queue,busiest_cpu),CPU_DEMAND_DEQUEUE,0);
1781 imbalance -= weight;
1782 if (!idle && (imbalance>0)) {
1789 spin_unlock(&busiest->lock);
1791 read_unlock(&class_list_lock);
1796 static inline void idle_balance(int this_cpu, runqueue_t *this_rq)
1799 #else /* CONFIG_CKRM_CPU_SCHEDULE */
1801 * move_tasks tries to move up to max_nr_move tasks from busiest to this_rq,
1802 * as part of a balancing operation within "domain". Returns the number of
1805 * Called with both runqueues locked.
1807 static int move_tasks(runqueue_t *this_rq, int this_cpu, runqueue_t *busiest,
1808 unsigned long max_nr_move, struct sched_domain *sd,
1809 enum idle_type idle)
1811 prio_array_t *array, *dst_array;
1812 struct list_head *head, *curr;
1813 int idx, pulled = 0;
1816 if (max_nr_move <= 0 || busiest->nr_running <= 1)
1820 * We first consider expired tasks. Those will likely not be
1821 * executed in the near future, and they are most likely to
1822 * be cache-cold, thus switching CPUs has the least effect
1825 if (busiest->expired->nr_active) {
1826 array = busiest->expired;
1827 dst_array = this_rq->expired;
1829 array = busiest->active;
1830 dst_array = this_rq->active;
1834 /* Start searching at priority 0: */
1838 idx = sched_find_first_bit(array->bitmap);
1840 idx = find_next_bit(array->bitmap, MAX_PRIO, idx);
1841 if (idx >= MAX_PRIO) {
1842 if (array == busiest->expired && busiest->active->nr_active) {
1843 array = busiest->active;
1844 dst_array = this_rq->active;
1850 head = array->queue + idx;
1853 tmp = list_entry(curr, task_t, run_list);
1857 if (!can_migrate_task(tmp, busiest, this_cpu, sd, idle)) {
1863 pull_task(busiest, array, tmp, this_rq, dst_array, this_cpu);
1866 /* We only want to steal up to the prescribed number of tasks. */
1867 if (pulled < max_nr_move) {
1878 * find_busiest_group finds and returns the busiest CPU group within the
1879 * domain. It calculates and returns the number of tasks which should be
1880 * moved to restore balance via the imbalance parameter.
1882 static struct sched_group *
1883 find_busiest_group(struct sched_domain *sd, int this_cpu,
1884 unsigned long *imbalance, enum idle_type idle)
1886 struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups;
1887 unsigned long max_load, avg_load, total_load, this_load, total_pwr;
1889 max_load = this_load = total_load = total_pwr = 0;
1897 local_group = cpu_isset(this_cpu, group->cpumask);
1899 /* Tally up the load of all CPUs in the group */
1901 cpus_and(tmp, group->cpumask, cpu_online_map);
1902 if (unlikely(cpus_empty(tmp)))
1905 for_each_cpu_mask(i, tmp) {
1906 /* Bias balancing toward cpus of our domain */
1908 load = target_load(i);
1910 load = source_load(i);
1919 total_load += avg_load;
1920 total_pwr += group->cpu_power;
1922 /* Adjust by relative CPU power of the group */
1923 avg_load = (avg_load * SCHED_LOAD_SCALE) / group->cpu_power;
1926 this_load = avg_load;
1929 } else if (avg_load > max_load) {
1930 max_load = avg_load;
1934 group = group->next;
1935 } while (group != sd->groups);
1937 if (!busiest || this_load >= max_load)
1940 avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr;
1942 if (this_load >= avg_load ||
1943 100*max_load <= sd->imbalance_pct*this_load)
1947 * We're trying to get all the cpus to the average_load, so we don't
1948 * want to push ourselves above the average load, nor do we wish to
1949 * reduce the max loaded cpu below the average load, as either of these
1950 * actions would just result in more rebalancing later, and ping-pong
1951 * tasks around. Thus we look for the minimum possible imbalance.
1952 * Negative imbalances (*we* are more loaded than anyone else) will
1953 * be counted as no imbalance for these purposes -- we can't fix that
1954 * by pulling tasks to us. Be careful of negative numbers as they'll
1955 * appear as very large values with unsigned longs.
1957 *imbalance = min(max_load - avg_load, avg_load - this_load);
1959 /* How much load to actually move to equalise the imbalance */
1960 *imbalance = (*imbalance * min(busiest->cpu_power, this->cpu_power))
1963 if (*imbalance < SCHED_LOAD_SCALE - 1) {
1964 unsigned long pwr_now = 0, pwr_move = 0;
1967 if (max_load - this_load >= SCHED_LOAD_SCALE*2) {
1973 * OK, we don't have enough imbalance to justify moving tasks,
1974 * however we may be able to increase total CPU power used by
1978 pwr_now += busiest->cpu_power*min(SCHED_LOAD_SCALE, max_load);
1979 pwr_now += this->cpu_power*min(SCHED_LOAD_SCALE, this_load);
1980 pwr_now /= SCHED_LOAD_SCALE;
1982 /* Amount of load we'd subtract */
1983 tmp = SCHED_LOAD_SCALE*SCHED_LOAD_SCALE/busiest->cpu_power;
1985 pwr_move += busiest->cpu_power*min(SCHED_LOAD_SCALE,
1988 /* Amount of load we'd add */
1989 tmp = SCHED_LOAD_SCALE*SCHED_LOAD_SCALE/this->cpu_power;
1992 pwr_move += this->cpu_power*min(SCHED_LOAD_SCALE, this_load + tmp);
1993 pwr_move /= SCHED_LOAD_SCALE;
1995 /* Move if we gain another 8th of a CPU worth of throughput */
1996 if (pwr_move < pwr_now + SCHED_LOAD_SCALE / 8)
2003 /* Get rid of the scaling factor, rounding down as we divide */
2004 *imbalance = (*imbalance + 1) / SCHED_LOAD_SCALE;
2009 if (busiest && (idle == NEWLY_IDLE ||
2010 (idle == IDLE && max_load > SCHED_LOAD_SCALE)) ) {
2020 * find_busiest_queue - find the busiest runqueue among the cpus in group.
2022 static runqueue_t *find_busiest_queue(struct sched_group *group)
2025 unsigned long load, max_load = 0;
2026 runqueue_t *busiest = NULL;
2029 cpus_and(tmp, group->cpumask, cpu_online_map);
2030 for_each_cpu_mask(i, tmp) {
2031 load = source_load(i);
2033 if (load > max_load) {
2035 busiest = cpu_rq(i);
2043 * Check this_cpu to ensure it is balanced within domain. Attempt to move
2044 * tasks if there is an imbalance.
2046 * Called with this_rq unlocked.
2048 static int load_balance(int this_cpu, runqueue_t *this_rq,
2049 struct sched_domain *sd, enum idle_type idle)
2051 struct sched_group *group;
2052 runqueue_t *busiest;
2053 unsigned long imbalance;
2056 spin_lock(&this_rq->lock);
2058 group = find_busiest_group(sd, this_cpu, &imbalance, idle);
2062 busiest = find_busiest_queue(group);
2066 * This should be "impossible", but since load
2067 * balancing is inherently racy and statistical,
2068 * it could happen in theory.
2070 if (unlikely(busiest == this_rq)) {
2076 if (busiest->nr_running > 1) {
2078 * Attempt to move tasks. If find_busiest_group has found
2079 * an imbalance but busiest->nr_running <= 1, the group is
2080 * still unbalanced. nr_moved simply stays zero, so it is
2081 * correctly treated as an imbalance.
2083 double_lock_balance(this_rq, busiest);
2084 nr_moved = move_tasks(this_rq, this_cpu, busiest,
2085 imbalance, sd, idle);
2086 spin_unlock(&busiest->lock);
2088 spin_unlock(&this_rq->lock);
2091 sd->nr_balance_failed++;
2093 if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) {
2096 spin_lock(&busiest->lock);
2097 if (!busiest->active_balance) {
2098 busiest->active_balance = 1;
2099 busiest->push_cpu = this_cpu;
2102 spin_unlock(&busiest->lock);
2104 wake_up_process(busiest->migration_thread);
2107 * We've kicked active balancing, reset the failure
2110 sd->nr_balance_failed = sd->cache_nice_tries;
2113 sd->nr_balance_failed = 0;
2115 /* We were unbalanced, so reset the balancing interval */
2116 sd->balance_interval = sd->min_interval;
2121 spin_unlock(&this_rq->lock);
2123 /* tune up the balancing interval */
2124 if (sd->balance_interval < sd->max_interval)
2125 sd->balance_interval *= 2;
2131 * Check this_cpu to ensure it is balanced within domain. Attempt to move
2132 * tasks if there is an imbalance.
2134 * Called from schedule when this_rq is about to become idle (NEWLY_IDLE).
2135 * this_rq is locked.
2137 static int load_balance_newidle(int this_cpu, runqueue_t *this_rq,
2138 struct sched_domain *sd)
2140 struct sched_group *group;
2141 runqueue_t *busiest = NULL;
2142 unsigned long imbalance;
2145 group = find_busiest_group(sd, this_cpu, &imbalance, NEWLY_IDLE);
2149 busiest = find_busiest_queue(group);
2150 if (!busiest || busiest == this_rq)
2153 /* Attempt to move tasks */
2154 double_lock_balance(this_rq, busiest);
2156 nr_moved = move_tasks(this_rq, this_cpu, busiest,
2157 imbalance, sd, NEWLY_IDLE);
2159 spin_unlock(&busiest->lock);
2166 * idle_balance is called by schedule() if this_cpu is about to become
2167 * idle. Attempts to pull tasks from other CPUs.
2169 static inline void idle_balance(int this_cpu, runqueue_t *this_rq)
2171 struct sched_domain *sd;
2173 for_each_domain(this_cpu, sd) {
2174 if (sd->flags & SD_BALANCE_NEWIDLE) {
2175 if (load_balance_newidle(this_cpu, this_rq, sd)) {
2176 /* We've pulled tasks over so stop searching */
2184 * active_load_balance is run by migration threads. It pushes a running
2185 * task off the cpu. It can be required to correctly have at least 1 task
2186 * running on each physical CPU where possible, and not have a physical /
2187 * logical imbalance.
2189 * Called with busiest locked.
2191 static void active_load_balance(runqueue_t *busiest, int busiest_cpu)
2193 struct sched_domain *sd;
2194 struct sched_group *group, *busy_group;
2197 if (busiest->nr_running <= 1)
2200 for_each_domain(busiest_cpu, sd)
2201 if (cpu_isset(busiest->push_cpu, sd->span))
2209 while (!cpu_isset(busiest_cpu, group->cpumask))
2210 group = group->next;
2219 if (group == busy_group)
2222 cpus_and(tmp, group->cpumask, cpu_online_map);
2223 if (!cpus_weight(tmp))
2226 for_each_cpu_mask(i, tmp) {
2232 rq = cpu_rq(push_cpu);
2235 * This condition is "impossible", but since load
2236 * balancing is inherently a bit racy and statistical,
2237 * it can trigger.. Reported by Bjorn Helgaas on a
2240 if (unlikely(busiest == rq))
2242 double_lock_balance(busiest, rq);
2243 move_tasks(rq, push_cpu, busiest, 1, sd, IDLE);
2244 spin_unlock(&rq->lock);
2246 group = group->next;
2247 } while (group != sd->groups);
2250 #endif /* CONFIG_CKRM_CPU_SCHEDULE*/
2253 * rebalance_tick will get called every timer tick, on every CPU.
2255 * It checks each scheduling domain to see if it is due to be balanced,
2256 * and initiates a balancing operation if so.
2258 * Balancing parameters are set up in arch_init_sched_domains.
2261 /* Don't have all balancing operations going off at once */
2262 #define CPU_OFFSET(cpu) (HZ * cpu / NR_CPUS)
2264 static void rebalance_tick(int this_cpu, runqueue_t *this_rq,
2265 enum idle_type idle)
2267 unsigned long old_load, this_load;
2268 unsigned long j = jiffies + CPU_OFFSET(this_cpu);
2269 struct sched_domain *sd;
2271 ckrm_rebalance_tick(j,this_cpu);
2273 /* Update our load */
2274 old_load = this_rq->cpu_load;
2275 this_load = this_rq->nr_running * SCHED_LOAD_SCALE;
2277 * Round up the averaging division if load is increasing. This
2278 * prevents us from getting stuck on 9 if the load is 10, for
2281 if (this_load > old_load)
2283 this_rq->cpu_load = (old_load + this_load) / 2;
2285 for_each_domain(this_cpu, sd) {
2286 unsigned long interval = sd->balance_interval;
2289 interval *= sd->busy_factor;
2291 /* scale ms to jiffies */
2292 interval = msecs_to_jiffies(interval);
2293 if (unlikely(!interval))
2296 if (j - sd->last_balance >= interval) {
2297 if (load_balance(this_cpu, this_rq, sd, idle)) {
2298 /* We've pulled tasks over so no longer idle */
2301 sd->last_balance += interval;
2307 * on UP we do not need to balance between CPUs:
2309 static inline void rebalance_tick(int cpu, runqueue_t *rq, enum idle_type idle)
2311 ckrm_rebalance_tick(jiffies,cpu);
2314 static inline void idle_balance(int cpu, runqueue_t *rq)
2319 static inline int wake_priority_sleeper(runqueue_t *rq)
2321 #ifdef CONFIG_SCHED_SMT
2323 * If an SMT sibling task has been put to sleep for priority
2324 * reasons reschedule the idle task to see if it can now run.
2326 if (rq->nr_running) {
2327 resched_task(rq->idle);
2334 DEFINE_PER_CPU(struct kernel_stat, kstat) = { { 0 } };
2336 EXPORT_PER_CPU_SYMBOL(kstat);
2339 * We place interactive tasks back into the active array, if possible.
2341 * To guarantee that this does not starve expired tasks we ignore the
2342 * interactivity of a task if the first expired task had to wait more
2343 * than a 'reasonable' amount of time. This deadline timeout is
2344 * load-dependent, as the frequency of array switched decreases with
2345 * increasing number of running tasks. We also ignore the interactivity
2346 * if a better static_prio task has expired:
2349 #ifndef CONFIG_CKRM_CPU_SCHEDULE
2350 #define EXPIRED_STARVING(rq) \
2351 ((STARVATION_LIMIT && ((rq)->expired_timestamp && \
2352 (jiffies - (rq)->expired_timestamp >= \
2353 STARVATION_LIMIT * ((rq)->nr_running) + 1))) || \
2354 ((rq)->curr->static_prio > (rq)->best_expired_prio))
2356 #define EXPIRED_STARVING(rq) \
2357 (STARVATION_LIMIT && ((rq)->expired_timestamp && \
2358 (jiffies - (rq)->expired_timestamp >= \
2359 STARVATION_LIMIT * (local_queue_nr_running(rq)) + 1)))
2363 * This function gets called by the timer code, with HZ frequency.
2364 * We call it with interrupts disabled.
2366 * It also gets called by the fork code, when changing the parent's
2369 void scheduler_tick(int user_ticks, int sys_ticks)
2371 int cpu = smp_processor_id();
2372 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
2373 runqueue_t *rq = this_rq();
2374 task_t *p = current;
2376 rq->timestamp_last_tick = sched_clock();
2378 if (rcu_pending(cpu))
2379 rcu_check_callbacks(cpu, user_ticks);
2381 /* note: this timer irq context must be accounted for as well */
2382 if (hardirq_count() - HARDIRQ_OFFSET) {
2383 cpustat->irq += sys_ticks;
2385 } else if (softirq_count()) {
2386 cpustat->softirq += sys_ticks;
2390 if (p == rq->idle) {
2391 if (!--rq->idle_tokens && !list_empty(&rq->hold_queue))
2394 if (atomic_read(&rq->nr_iowait) > 0)
2395 cpustat->iowait += sys_ticks;
2397 cpustat->idle += sys_ticks;
2398 if (wake_priority_sleeper(rq))
2400 rebalance_tick(cpu, rq, IDLE);
2403 if (TASK_NICE(p) > 0)
2404 cpustat->nice += user_ticks;
2406 cpustat->user += user_ticks;
2407 cpustat->system += sys_ticks;
2409 /* Task might have expired already, but not scheduled off yet */
2410 if (p->array != rq_active(p,rq)) {
2411 set_tsk_need_resched(p);
2414 spin_lock(&rq->lock);
2416 * The task was running during this tick - update the
2417 * time slice counter. Note: we do not update a thread's
2418 * priority until it either goes to sleep or uses up its
2419 * timeslice. This makes it possible for interactive tasks
2420 * to use up their timeslices at their highest priority levels.
2422 if (unlikely(rt_task(p))) {
2424 * RR tasks need a special form of timeslice management.
2425 * FIFO tasks have no timeslices.
2427 if ((p->policy == SCHED_RR) && !--p->time_slice) {
2428 p->time_slice = task_timeslice(p);
2429 p->first_time_slice = 0;
2430 set_tsk_need_resched(p);
2432 /* put it at the end of the queue: */
2433 dequeue_task(p, rq_active(p,rq));
2434 enqueue_task(p, rq_active(p,rq));
2438 #warning MEF PLANETLAB: "if (vx_need_resched(p)) was if (!--p->time_slice) */"
2439 if (vx_need_resched(p)) {
2440 #ifdef CONFIG_CKRM_CPU_SCHEDULE
2441 /* Hubertus ... we can abstract this out */
2442 struct ckrm_local_runqueue* rq = get_task_class_queue(p);
2444 dequeue_task(p, rq->active);
2445 set_tsk_need_resched(p);
2446 p->prio = effective_prio(p);
2447 p->time_slice = task_timeslice(p);
2448 p->first_time_slice = 0;
2450 if (!rq->expired_timestamp)
2451 rq->expired_timestamp = jiffies;
2452 if (!TASK_INTERACTIVE(p) || EXPIRED_STARVING(rq)) {
2453 enqueue_task(p, rq->expired);
2454 if (p->static_prio < this_rq()->best_expired_prio)
2455 this_rq()->best_expired_prio = p->static_prio;
2457 enqueue_task(p, rq->active);
2460 * Prevent a too long timeslice allowing a task to monopolize
2461 * the CPU. We do this by splitting up the timeslice into
2464 * Note: this does not mean the task's timeslices expire or
2465 * get lost in any way, they just might be preempted by
2466 * another task of equal priority. (one with higher
2467 * priority would have preempted this task already.) We
2468 * requeue this task to the end of the list on this priority
2469 * level, which is in essence a round-robin of tasks with
2472 * This only applies to tasks in the interactive
2473 * delta range with at least TIMESLICE_GRANULARITY to requeue.
2475 if (TASK_INTERACTIVE(p) && !((task_timeslice(p) -
2476 p->time_slice) % TIMESLICE_GRANULARITY(p)) &&
2477 (p->time_slice >= TIMESLICE_GRANULARITY(p)) &&
2478 (p->array == rq_active(p,rq))) {
2480 dequeue_task(p, rq_active(p,rq));
2481 set_tsk_need_resched(p);
2482 p->prio = effective_prio(p);
2483 enqueue_task(p, rq_active(p,rq));
2487 spin_unlock(&rq->lock);
2489 rebalance_tick(cpu, rq, NOT_IDLE);
2492 #ifdef CONFIG_SCHED_SMT
2493 static inline void wake_sleeping_dependent(int cpu, runqueue_t *rq)
2496 struct sched_domain *sd = rq->sd;
2497 cpumask_t sibling_map;
2499 if (!(sd->flags & SD_SHARE_CPUPOWER))
2502 cpus_and(sibling_map, sd->span, cpu_online_map);
2503 for_each_cpu_mask(i, sibling_map) {
2512 * If an SMT sibling task is sleeping due to priority
2513 * reasons wake it up now.
2515 if (smt_rq->curr == smt_rq->idle && smt_rq->nr_running)
2516 resched_task(smt_rq->idle);
2520 static inline int dependent_sleeper(int cpu, runqueue_t *rq, task_t *p)
2522 struct sched_domain *sd = rq->sd;
2523 cpumask_t sibling_map;
2526 if (!(sd->flags & SD_SHARE_CPUPOWER))
2529 cpus_and(sibling_map, sd->span, cpu_online_map);
2530 for_each_cpu_mask(i, sibling_map) {
2538 smt_curr = smt_rq->curr;
2541 * If a user task with lower static priority than the
2542 * running task on the SMT sibling is trying to schedule,
2543 * delay it till there is proportionately less timeslice
2544 * left of the sibling task to prevent a lower priority
2545 * task from using an unfair proportion of the
2546 * physical cpu's resources. -ck
2548 if (((smt_curr->time_slice * (100 - sd->per_cpu_gain) / 100) >
2549 task_timeslice(p) || rt_task(smt_curr)) &&
2550 p->mm && smt_curr->mm && !rt_task(p))
2554 * Reschedule a lower priority task on the SMT sibling,
2555 * or wake it up if it has been put to sleep for priority
2558 if ((((p->time_slice * (100 - sd->per_cpu_gain) / 100) >
2559 task_timeslice(smt_curr) || rt_task(p)) &&
2560 smt_curr->mm && p->mm && !rt_task(smt_curr)) ||
2561 (smt_curr == smt_rq->idle && smt_rq->nr_running))
2562 resched_task(smt_curr);
2567 static inline void wake_sleeping_dependent(int cpu, runqueue_t *rq)
2571 static inline int dependent_sleeper(int cpu, runqueue_t *rq, task_t *p)
2578 * schedule() is the main scheduler function.
2580 asmlinkage void __sched schedule(void)
2583 task_t *prev, *next;
2585 prio_array_t *array;
2586 unsigned long long now;
2587 unsigned long run_time;
2589 #ifdef CONFIG_VSERVER_HARDCPU
2590 struct vx_info *vxi;
2594 //WARN_ON(system_state == SYSTEM_BOOTING);
2596 * Test if we are atomic. Since do_exit() needs to call into
2597 * schedule() atomically, we ignore that path for now.
2598 * Otherwise, whine if we are scheduling when we should not be.
2600 if (likely(!(current->state & (TASK_DEAD | TASK_ZOMBIE)))) {
2601 if (unlikely(in_atomic())) {
2602 printk(KERN_ERR "bad: scheduling while atomic!\n");
2612 release_kernel_lock(prev);
2613 now = sched_clock();
2614 if (likely(now - prev->timestamp < NS_MAX_SLEEP_AVG))
2615 run_time = now - prev->timestamp;
2617 run_time = NS_MAX_SLEEP_AVG;
2620 * Tasks with interactive credits get charged less run_time
2621 * at high sleep_avg to delay them losing their interactive
2624 if (HIGH_CREDIT(prev))
2625 run_time /= (CURRENT_BONUS(prev) ? : 1);
2627 spin_lock_irq(&rq->lock);
2630 * if entering off of a kernel preemption go straight
2631 * to picking the next task.
2633 switch_count = &prev->nivcsw;
2634 if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) {
2635 switch_count = &prev->nvcsw;
2636 if (unlikely((prev->state & TASK_INTERRUPTIBLE) &&
2637 unlikely(signal_pending(prev))))
2638 prev->state = TASK_RUNNING;
2640 deactivate_task(prev, rq);
2643 cpu = smp_processor_id();
2644 #ifdef CONFIG_VSERVER_HARDCPU
2645 if (!list_empty(&rq->hold_queue)) {
2646 struct list_head *l, *n;
2650 list_for_each_safe(l, n, &rq->hold_queue) {
2651 next = list_entry(l, task_t, run_list);
2652 if (vxi == next->vx_info)
2655 vxi = next->vx_info;
2656 ret = vx_tokens_recalc(vxi);
2657 // tokens = vx_tokens_avail(next);
2660 list_del(&next->run_list);
2661 next->state &= ~TASK_ONHOLD;
2662 recalc_task_prio(next, now);
2663 __activate_task(next, rq);
2664 // printk("··· unhold %p\n", next);
2667 if ((ret < 0) && (maxidle < ret))
2671 rq->idle_tokens = -maxidle;
2675 if (unlikely(!rq->nr_running)) {
2676 idle_balance(cpu, rq);
2677 if (!rq->nr_running) {
2679 rq->expired_timestamp = 0;
2680 wake_sleeping_dependent(cpu, rq);
2685 next = rq_get_next_task(rq);
2686 if (next == rq->idle)
2689 if (dependent_sleeper(cpu, rq, next)) {
2694 #ifdef CONFIG_VSERVER_HARDCPU
2695 vxi = next->vx_info;
2696 if (vxi && __vx_flags(vxi->vx_flags,
2697 VXF_SCHED_PAUSE|VXF_SCHED_HARD, 0)) {
2698 int ret = vx_tokens_recalc(vxi);
2700 if (unlikely(ret <= 0)) {
2701 if (ret && (rq->idle_tokens > -ret))
2702 rq->idle_tokens = -ret;
2703 deactivate_task(next, rq);
2704 list_add_tail(&next->run_list, &rq->hold_queue);
2705 next->state |= TASK_ONHOLD;
2711 if (!rt_task(next) && next->activated > 0) {
2712 unsigned long long delta = now - next->timestamp;
2714 if (next->activated == 1)
2715 delta = delta * (ON_RUNQUEUE_WEIGHT * 128 / 100) / 128;
2717 array = next->array;
2718 dequeue_task(next, array);
2719 recalc_task_prio(next, next->timestamp + delta);
2720 enqueue_task(next, array);
2722 next->activated = 0;
2725 if (test_and_clear_tsk_thread_flag(prev,TIF_NEED_RESCHED))
2727 RCU_qsctr(task_cpu(prev))++;
2729 #ifdef CONFIG_CKRM_CPU_SCHEDULE
2730 if (prev != rq->idle) {
2731 unsigned long long run = now - prev->timestamp;
2732 cpu_demand_event(get_task_local_stat(prev),CPU_DEMAND_DESCHEDULE,run);
2733 update_local_cvt(prev, run);
2737 prev->sleep_avg -= run_time;
2738 if ((long)prev->sleep_avg <= 0) {
2739 prev->sleep_avg = 0;
2740 if (!(HIGH_CREDIT(prev) || LOW_CREDIT(prev)))
2741 prev->interactive_credit--;
2743 add_delay_ts(prev,runcpu_total,prev->timestamp,now);
2744 prev->timestamp = now;
2746 if (likely(prev != next)) {
2747 add_delay_ts(next,waitcpu_total,next->timestamp,now);
2748 inc_delay(next,runs);
2749 next->timestamp = now;
2754 prepare_arch_switch(rq, next);
2755 prev = context_switch(rq, prev, next);
2758 finish_task_switch(prev);
2760 spin_unlock_irq(&rq->lock);
2762 reacquire_kernel_lock(current);
2763 preempt_enable_no_resched();
2764 if (test_thread_flag(TIF_NEED_RESCHED))
2768 EXPORT_SYMBOL(schedule);
2770 #ifdef CONFIG_PREEMPT
2772 * this is is the entry point to schedule() from in-kernel preemption
2773 * off of preempt_enable. Kernel preemptions off return from interrupt
2774 * occur there and call schedule directly.
2776 asmlinkage void __sched preempt_schedule(void)
2778 struct thread_info *ti = current_thread_info();
2781 * If there is a non-zero preempt_count or interrupts are disabled,
2782 * we do not want to preempt the current task. Just return..
2784 if (unlikely(ti->preempt_count || irqs_disabled()))
2788 ti->preempt_count = PREEMPT_ACTIVE;
2790 ti->preempt_count = 0;
2792 /* we could miss a preemption opportunity between schedule and now */
2794 if (unlikely(test_thread_flag(TIF_NEED_RESCHED)))
2798 EXPORT_SYMBOL(preempt_schedule);
2799 #endif /* CONFIG_PREEMPT */
2801 int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, void *key)
2803 task_t *p = curr->task;
2804 return try_to_wake_up(p, mode, sync);
2807 EXPORT_SYMBOL(default_wake_function);
2810 * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just
2811 * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve
2812 * number) then we wake all the non-exclusive tasks and one exclusive task.
2814 * There are circumstances in which we can try to wake a task which has already
2815 * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns
2816 * zero in this (rare) case, and we handle it by continuing to scan the queue.
2818 static void __wake_up_common(wait_queue_head_t *q, unsigned int mode,
2819 int nr_exclusive, int sync, void *key)
2821 struct list_head *tmp, *next;
2823 list_for_each_safe(tmp, next, &q->task_list) {
2826 curr = list_entry(tmp, wait_queue_t, task_list);
2827 flags = curr->flags;
2828 if (curr->func(curr, mode, sync, key) &&
2829 (flags & WQ_FLAG_EXCLUSIVE) &&
2836 * __wake_up - wake up threads blocked on a waitqueue.
2838 * @mode: which threads
2839 * @nr_exclusive: how many wake-one or wake-many threads to wake up
2841 void fastcall __wake_up(wait_queue_head_t *q, unsigned int mode,
2842 int nr_exclusive, void *key)
2844 unsigned long flags;
2846 spin_lock_irqsave(&q->lock, flags);
2847 __wake_up_common(q, mode, nr_exclusive, 0, key);
2848 spin_unlock_irqrestore(&q->lock, flags);
2851 EXPORT_SYMBOL(__wake_up);
2854 * Same as __wake_up but called with the spinlock in wait_queue_head_t held.
2856 void fastcall __wake_up_locked(wait_queue_head_t *q, unsigned int mode)
2858 __wake_up_common(q, mode, 1, 0, NULL);
2862 * __wake_up - sync- wake up threads blocked on a waitqueue.
2864 * @mode: which threads
2865 * @nr_exclusive: how many wake-one or wake-many threads to wake up
2867 * The sync wakeup differs that the waker knows that it will schedule
2868 * away soon, so while the target thread will be woken up, it will not
2869 * be migrated to another CPU - ie. the two threads are 'synchronized'
2870 * with each other. This can prevent needless bouncing between CPUs.
2872 * On UP it can prevent extra preemption.
2874 void fastcall __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive)
2876 unsigned long flags;
2882 if (unlikely(!nr_exclusive))
2885 spin_lock_irqsave(&q->lock, flags);
2886 __wake_up_common(q, mode, nr_exclusive, sync, NULL);
2887 spin_unlock_irqrestore(&q->lock, flags);
2889 EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */
2891 void fastcall complete(struct completion *x)
2893 unsigned long flags;
2895 spin_lock_irqsave(&x->wait.lock, flags);
2897 __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2899 spin_unlock_irqrestore(&x->wait.lock, flags);
2901 EXPORT_SYMBOL(complete);
2903 void fastcall complete_all(struct completion *x)
2905 unsigned long flags;
2907 spin_lock_irqsave(&x->wait.lock, flags);
2908 x->done += UINT_MAX/2;
2909 __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2911 spin_unlock_irqrestore(&x->wait.lock, flags);
2913 EXPORT_SYMBOL(complete_all);
2915 void fastcall __sched wait_for_completion(struct completion *x)
2918 spin_lock_irq(&x->wait.lock);
2920 DECLARE_WAITQUEUE(wait, current);
2922 wait.flags |= WQ_FLAG_EXCLUSIVE;
2923 __add_wait_queue_tail(&x->wait, &wait);
2925 __set_current_state(TASK_UNINTERRUPTIBLE);
2926 spin_unlock_irq(&x->wait.lock);
2928 spin_lock_irq(&x->wait.lock);
2930 __remove_wait_queue(&x->wait, &wait);
2933 spin_unlock_irq(&x->wait.lock);
2935 EXPORT_SYMBOL(wait_for_completion);
2937 #define SLEEP_ON_VAR \
2938 unsigned long flags; \
2939 wait_queue_t wait; \
2940 init_waitqueue_entry(&wait, current);
2942 #define SLEEP_ON_HEAD \
2943 spin_lock_irqsave(&q->lock,flags); \
2944 __add_wait_queue(q, &wait); \
2945 spin_unlock(&q->lock);
2947 #define SLEEP_ON_TAIL \
2948 spin_lock_irq(&q->lock); \
2949 __remove_wait_queue(q, &wait); \
2950 spin_unlock_irqrestore(&q->lock, flags);
2952 #define SLEEP_ON_BKLCHECK \
2953 if (unlikely(!kernel_locked()) && \
2954 sleep_on_bkl_warnings < 10) { \
2955 sleep_on_bkl_warnings++; \
2959 static int sleep_on_bkl_warnings;
2961 void fastcall __sched interruptible_sleep_on(wait_queue_head_t *q)
2967 current->state = TASK_INTERRUPTIBLE;
2974 EXPORT_SYMBOL(interruptible_sleep_on);
2976 long fastcall __sched interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout)
2982 current->state = TASK_INTERRUPTIBLE;
2985 timeout = schedule_timeout(timeout);
2991 EXPORT_SYMBOL(interruptible_sleep_on_timeout);
2993 long fastcall __sched sleep_on_timeout(wait_queue_head_t *q, long timeout)
2999 current->state = TASK_UNINTERRUPTIBLE;
3002 timeout = schedule_timeout(timeout);
3008 EXPORT_SYMBOL(sleep_on_timeout);
3010 void set_user_nice(task_t *p, long nice)
3012 unsigned long flags;
3013 prio_array_t *array;
3015 int old_prio, new_prio, delta;
3017 if (TASK_NICE(p) == nice || nice < -20 || nice > 19)
3020 * We have to be careful, if called from sys_setpriority(),
3021 * the task might be in the middle of scheduling on another CPU.
3023 rq = task_rq_lock(p, &flags);
3025 * The RT priorities are set via setscheduler(), but we still
3026 * allow the 'normal' nice value to be set - but as expected
3027 * it wont have any effect on scheduling until the task is
3031 p->static_prio = NICE_TO_PRIO(nice);
3036 dequeue_task(p, array);
3039 new_prio = NICE_TO_PRIO(nice);
3040 delta = new_prio - old_prio;
3041 p->static_prio = NICE_TO_PRIO(nice);
3045 enqueue_task(p, array);
3047 * If the task increased its priority or is running and
3048 * lowered its priority, then reschedule its CPU:
3050 if (delta < 0 || (delta > 0 && task_running(rq, p)))
3051 resched_task(rq->curr);
3054 task_rq_unlock(rq, &flags);
3057 EXPORT_SYMBOL(set_user_nice);
3059 #ifdef __ARCH_WANT_SYS_NICE
3062 * sys_nice - change the priority of the current process.
3063 * @increment: priority increment
3065 * sys_setpriority is a more generic, but much slower function that
3066 * does similar things.
3068 asmlinkage long sys_nice(int increment)
3074 * Setpriority might change our priority at the same moment.
3075 * We don't have to worry. Conceptually one call occurs first
3076 * and we have a single winner.
3078 if (increment < 0) {
3079 if (!capable(CAP_SYS_NICE))
3081 if (increment < -40)
3087 nice = PRIO_TO_NICE(current->static_prio) + increment;
3093 retval = security_task_setnice(current, nice);
3097 set_user_nice(current, nice);
3104 * task_prio - return the priority value of a given task.
3105 * @p: the task in question.
3107 * This is the priority value as seen by users in /proc.
3108 * RT tasks are offset by -200. Normal tasks are centered
3109 * around 0, value goes from -16 to +15.
3111 int task_prio(const task_t *p)
3113 return p->prio - MAX_RT_PRIO;
3117 * task_nice - return the nice value of a given task.
3118 * @p: the task in question.
3120 int task_nice(const task_t *p)
3122 return TASK_NICE(p);
3125 EXPORT_SYMBOL(task_nice);
3128 * idle_cpu - is a given cpu idle currently?
3129 * @cpu: the processor in question.
3131 int idle_cpu(int cpu)
3133 return cpu_curr(cpu) == cpu_rq(cpu)->idle;
3136 EXPORT_SYMBOL_GPL(idle_cpu);
3139 * find_process_by_pid - find a process with a matching PID value.
3140 * @pid: the pid in question.
3142 static inline task_t *find_process_by_pid(pid_t pid)
3144 return pid ? find_task_by_pid(pid) : current;
3147 /* Actually do priority change: must hold rq lock. */
3148 static void __setscheduler(struct task_struct *p, int policy, int prio)
3152 p->rt_priority = prio;
3153 if (policy != SCHED_NORMAL)
3154 p->prio = MAX_USER_RT_PRIO-1 - p->rt_priority;
3156 p->prio = p->static_prio;
3160 * setscheduler - change the scheduling policy and/or RT priority of a thread.
3162 static int setscheduler(pid_t pid, int policy, struct sched_param __user *param)
3164 struct sched_param lp;
3165 int retval = -EINVAL;
3167 prio_array_t *array;
3168 unsigned long flags;
3172 if (!param || pid < 0)
3176 if (copy_from_user(&lp, param, sizeof(struct sched_param)))
3180 * We play safe to avoid deadlocks.
3182 read_lock_irq(&tasklist_lock);
3184 p = find_process_by_pid(pid);
3188 goto out_unlock_tasklist;
3191 * To be able to change p->policy safely, the apropriate
3192 * runqueue lock must be held.
3194 rq = task_rq_lock(p, &flags);
3200 if (policy != SCHED_FIFO && policy != SCHED_RR &&
3201 policy != SCHED_NORMAL)
3206 * Valid priorities for SCHED_FIFO and SCHED_RR are
3207 * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL is 0.
3210 if (lp.sched_priority < 0 || lp.sched_priority > MAX_USER_RT_PRIO-1)
3212 if ((policy == SCHED_NORMAL) != (lp.sched_priority == 0))
3216 if ((policy == SCHED_FIFO || policy == SCHED_RR) &&
3217 !capable(CAP_SYS_NICE))
3219 if ((current->euid != p->euid) && (current->euid != p->uid) &&
3220 !capable(CAP_SYS_NICE))
3223 retval = security_task_setscheduler(p, policy, &lp);
3229 deactivate_task(p, task_rq(p));
3232 __setscheduler(p, policy, lp.sched_priority);
3234 __activate_task(p, task_rq(p));
3236 * Reschedule if we are currently running on this runqueue and
3237 * our priority decreased, or if we are not currently running on
3238 * this runqueue and our priority is higher than the current's
3240 if (task_running(rq, p)) {
3241 if (p->prio > oldprio)
3242 resched_task(rq->curr);
3243 } else if (TASK_PREEMPTS_CURR(p, rq))
3244 resched_task(rq->curr);
3248 task_rq_unlock(rq, &flags);
3249 out_unlock_tasklist:
3250 read_unlock_irq(&tasklist_lock);
3257 * sys_sched_setscheduler - set/change the scheduler policy and RT priority
3258 * @pid: the pid in question.
3259 * @policy: new policy
3260 * @param: structure containing the new RT priority.
3262 asmlinkage long sys_sched_setscheduler(pid_t pid, int policy,
3263 struct sched_param __user *param)
3265 return setscheduler(pid, policy, param);
3269 * sys_sched_setparam - set/change the RT priority of a thread
3270 * @pid: the pid in question.
3271 * @param: structure containing the new RT priority.
3273 asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param)
3275 return setscheduler(pid, -1, param);
3279 * sys_sched_getscheduler - get the policy (scheduling class) of a thread
3280 * @pid: the pid in question.
3282 asmlinkage long sys_sched_getscheduler(pid_t pid)
3284 int retval = -EINVAL;
3291 read_lock(&tasklist_lock);
3292 p = find_process_by_pid(pid);
3294 retval = security_task_getscheduler(p);
3298 read_unlock(&tasklist_lock);
3305 * sys_sched_getscheduler - get the RT priority of a thread
3306 * @pid: the pid in question.
3307 * @param: structure containing the RT priority.
3309 asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param)
3311 struct sched_param lp;
3312 int retval = -EINVAL;
3315 if (!param || pid < 0)
3318 read_lock(&tasklist_lock);
3319 p = find_process_by_pid(pid);
3324 retval = security_task_getscheduler(p);
3328 lp.sched_priority = p->rt_priority;
3329 read_unlock(&tasklist_lock);
3332 * This one might sleep, we cannot do it with a spinlock held ...
3334 retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0;
3340 read_unlock(&tasklist_lock);
3345 * sys_sched_setaffinity - set the cpu affinity of a process
3346 * @pid: pid of the process
3347 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
3348 * @user_mask_ptr: user-space pointer to the new cpu mask
3350 asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len,
3351 unsigned long __user *user_mask_ptr)
3357 if (len < sizeof(new_mask))
3360 if (copy_from_user(&new_mask, user_mask_ptr, sizeof(new_mask)))
3364 read_lock(&tasklist_lock);
3366 p = find_process_by_pid(pid);
3368 read_unlock(&tasklist_lock);
3369 unlock_cpu_hotplug();
3374 * It is not safe to call set_cpus_allowed with the
3375 * tasklist_lock held. We will bump the task_struct's
3376 * usage count and then drop tasklist_lock.
3379 read_unlock(&tasklist_lock);
3382 if ((current->euid != p->euid) && (current->euid != p->uid) &&
3383 !capable(CAP_SYS_NICE))
3386 retval = set_cpus_allowed(p, new_mask);
3390 unlock_cpu_hotplug();
3395 * Represents all cpu's present in the system
3396 * In systems capable of hotplug, this map could dynamically grow
3397 * as new cpu's are detected in the system via any platform specific
3398 * method, such as ACPI for e.g.
3401 cpumask_t cpu_present_map;
3402 EXPORT_SYMBOL(cpu_present_map);
3405 cpumask_t cpu_online_map = CPU_MASK_ALL;
3406 cpumask_t cpu_possible_map = CPU_MASK_ALL;
3410 * sys_sched_getaffinity - get the cpu affinity of a process
3411 * @pid: pid of the process
3412 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
3413 * @user_mask_ptr: user-space pointer to hold the current cpu mask
3415 asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len,
3416 unsigned long __user *user_mask_ptr)
3418 unsigned int real_len;
3423 real_len = sizeof(mask);
3428 read_lock(&tasklist_lock);
3431 p = find_process_by_pid(pid);
3436 cpus_and(mask, p->cpus_allowed, cpu_possible_map);
3439 read_unlock(&tasklist_lock);
3440 unlock_cpu_hotplug();
3443 if (copy_to_user(user_mask_ptr, &mask, real_len))
3449 * sys_sched_yield - yield the current processor to other threads.
3451 * this function yields the current CPU by moving the calling thread
3452 * to the expired array. If there are no other threads running on this
3453 * CPU then this function will return.
3455 asmlinkage long sys_sched_yield(void)
3457 runqueue_t *rq = this_rq_lock();
3458 prio_array_t *array = current->array;
3459 prio_array_t *target = rq_expired(current,rq);
3462 * We implement yielding by moving the task into the expired
3465 * (special rule: RT tasks will just roundrobin in the active
3468 if (unlikely(rt_task(current)))
3469 target = rq_active(current,rq);
3471 dequeue_task(current, array);
3472 enqueue_task(current, target);
3475 * Since we are going to call schedule() anyway, there's
3476 * no need to preempt or enable interrupts:
3478 _raw_spin_unlock(&rq->lock);
3479 preempt_enable_no_resched();
3486 void __sched __cond_resched(void)
3488 #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP
3489 __might_sleep(__FILE__, __LINE__, 0);
3492 * The system_state check is somewhat ugly but we might be
3493 * called during early boot when we are not yet ready to reschedule.
3495 if (need_resched() && system_state >= SYSTEM_BOOTING_SCHEDULER_OK) {
3496 set_current_state(TASK_RUNNING);
3501 EXPORT_SYMBOL(__cond_resched);
3503 void __sched __cond_resched_lock(spinlock_t * lock)
3505 if (need_resched()) {
3506 _raw_spin_unlock(lock);
3507 preempt_enable_no_resched();
3508 set_current_state(TASK_RUNNING);
3514 EXPORT_SYMBOL(__cond_resched_lock);
3517 * yield - yield the current processor to other threads.
3519 * this is a shortcut for kernel-space yielding - it marks the
3520 * thread runnable and calls sys_sched_yield().
3522 void __sched yield(void)
3524 set_current_state(TASK_RUNNING);
3528 EXPORT_SYMBOL(yield);
3531 * This task is about to go to sleep on IO. Increment rq->nr_iowait so
3532 * that process accounting knows that this is a task in IO wait state.
3534 * But don't do that if it is a deliberate, throttling IO wait (this task
3535 * has set its backing_dev_info: the queue against which it should throttle)
3537 void __sched io_schedule(void)
3539 struct runqueue *rq = this_rq();
3540 def_delay_var(dstart);
3542 start_delay_set(dstart,PF_IOWAIT);
3543 atomic_inc(&rq->nr_iowait);
3545 atomic_dec(&rq->nr_iowait);
3546 add_io_delay(dstart);
3549 EXPORT_SYMBOL(io_schedule);
3551 long __sched io_schedule_timeout(long timeout)
3553 struct runqueue *rq = this_rq();
3555 def_delay_var(dstart);
3557 start_delay_set(dstart,PF_IOWAIT);
3558 atomic_inc(&rq->nr_iowait);
3559 ret = schedule_timeout(timeout);
3560 atomic_dec(&rq->nr_iowait);
3561 add_io_delay(dstart);
3566 * sys_sched_get_priority_max - return maximum RT priority.
3567 * @policy: scheduling class.
3569 * this syscall returns the maximum rt_priority that can be used
3570 * by a given scheduling class.
3572 asmlinkage long sys_sched_get_priority_max(int policy)
3579 ret = MAX_USER_RT_PRIO-1;
3589 * sys_sched_get_priority_min - return minimum RT priority.
3590 * @policy: scheduling class.
3592 * this syscall returns the minimum rt_priority that can be used
3593 * by a given scheduling class.
3595 asmlinkage long sys_sched_get_priority_min(int policy)
3611 * sys_sched_rr_get_interval - return the default timeslice of a process.
3612 * @pid: pid of the process.
3613 * @interval: userspace pointer to the timeslice value.
3615 * this syscall writes the default timeslice value of a given process
3616 * into the user-space timespec buffer. A value of '0' means infinity.
3619 long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval)
3621 int retval = -EINVAL;
3629 read_lock(&tasklist_lock);
3630 p = find_process_by_pid(pid);
3634 retval = security_task_getscheduler(p);
3638 jiffies_to_timespec(p->policy & SCHED_FIFO ?
3639 0 : task_timeslice(p), &t);
3640 read_unlock(&tasklist_lock);
3641 retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0;
3645 read_unlock(&tasklist_lock);
3649 static inline struct task_struct *eldest_child(struct task_struct *p)
3651 if (list_empty(&p->children)) return NULL;
3652 return list_entry(p->children.next,struct task_struct,sibling);
3655 static inline struct task_struct *older_sibling(struct task_struct *p)
3657 if (p->sibling.prev==&p->parent->children) return NULL;
3658 return list_entry(p->sibling.prev,struct task_struct,sibling);
3661 static inline struct task_struct *younger_sibling(struct task_struct *p)
3663 if (p->sibling.next==&p->parent->children) return NULL;
3664 return list_entry(p->sibling.next,struct task_struct,sibling);
3667 static void show_task(task_t * p)
3671 unsigned long free = 0;
3672 static const char *stat_nam[] = { "R", "S", "D", "T", "Z", "W" };
3674 printk("%-13.13s ", p->comm);
3675 state = p->state ? __ffs(p->state) + 1 : 0;
3676 if (state < ARRAY_SIZE(stat_nam))
3677 printk(stat_nam[state]);
3680 #if (BITS_PER_LONG == 32)
3681 if (state == TASK_RUNNING)
3682 printk(" running ");
3684 printk(" %08lX ", thread_saved_pc(p));
3686 if (state == TASK_RUNNING)
3687 printk(" running task ");
3689 printk(" %016lx ", thread_saved_pc(p));
3691 #ifdef CONFIG_DEBUG_STACK_USAGE
3693 unsigned long * n = (unsigned long *) (p->thread_info+1);
3696 free = (unsigned long) n - (unsigned long)(p->thread_info+1);
3699 printk("%5lu %5d %6d ", free, p->pid, p->parent->pid);
3700 if ((relative = eldest_child(p)))
3701 printk("%5d ", relative->pid);
3704 if ((relative = younger_sibling(p)))
3705 printk("%7d", relative->pid);
3708 if ((relative = older_sibling(p)))
3709 printk(" %5d", relative->pid);
3713 printk(" (L-TLB)\n");
3715 printk(" (NOTLB)\n");
3717 if (state != TASK_RUNNING)
3718 show_stack(p, NULL);
3721 void show_state(void)
3725 #if (BITS_PER_LONG == 32)
3728 printk(" task PC pid father child younger older\n");
3732 printk(" task PC pid father child younger older\n");
3734 read_lock(&tasklist_lock);
3735 do_each_thread(g, p) {
3737 * reset the NMI-timeout, listing all files on a slow
3738 * console might take alot of time:
3740 touch_nmi_watchdog();
3742 } while_each_thread(g, p);
3744 read_unlock(&tasklist_lock);
3747 EXPORT_SYMBOL_GPL(show_state);
3749 void __devinit init_idle(task_t *idle, int cpu)
3751 runqueue_t *idle_rq = cpu_rq(cpu), *rq = cpu_rq(task_cpu(idle));
3752 unsigned long flags;
3754 local_irq_save(flags);
3755 double_rq_lock(idle_rq, rq);
3757 idle_rq->curr = idle_rq->idle = idle;
3758 deactivate_task(idle, rq);
3760 idle->prio = MAX_PRIO;
3761 idle->state = TASK_RUNNING;
3762 set_task_cpu(idle, cpu);
3763 double_rq_unlock(idle_rq, rq);
3764 set_tsk_need_resched(idle);
3765 local_irq_restore(flags);
3767 /* Set the preempt count _outside_ the spinlocks! */
3768 #ifdef CONFIG_PREEMPT
3769 idle->thread_info->preempt_count = (idle->lock_depth >= 0);
3771 idle->thread_info->preempt_count = 0;
3776 * In a system that switches off the HZ timer nohz_cpu_mask
3777 * indicates which cpus entered this state. This is used
3778 * in the rcu update to wait only for active cpus. For system
3779 * which do not switch off the HZ timer nohz_cpu_mask should
3780 * always be CPU_MASK_NONE.
3782 cpumask_t nohz_cpu_mask = CPU_MASK_NONE;
3786 * This is how migration works:
3788 * 1) we queue a migration_req_t structure in the source CPU's
3789 * runqueue and wake up that CPU's migration thread.
3790 * 2) we down() the locked semaphore => thread blocks.
3791 * 3) migration thread wakes up (implicitly it forces the migrated
3792 * thread off the CPU)
3793 * 4) it gets the migration request and checks whether the migrated
3794 * task is still in the wrong runqueue.
3795 * 5) if it's in the wrong runqueue then the migration thread removes
3796 * it and puts it into the right queue.
3797 * 6) migration thread up()s the semaphore.
3798 * 7) we wake up and the migration is done.
3802 * Change a given task's CPU affinity. Migrate the thread to a
3803 * proper CPU and schedule it away if the CPU it's executing on
3804 * is removed from the allowed bitmask.
3806 * NOTE: the caller must have a valid reference to the task, the
3807 * task must not exit() & deallocate itself prematurely. The
3808 * call is not atomic; no spinlocks may be held.
3810 int set_cpus_allowed(task_t *p, cpumask_t new_mask)
3812 unsigned long flags;
3814 migration_req_t req;
3817 rq = task_rq_lock(p, &flags);
3818 if (!cpus_intersects(new_mask, cpu_online_map)) {
3823 p->cpus_allowed = new_mask;
3824 /* Can the task run on the task's current CPU? If so, we're done */
3825 if (cpu_isset(task_cpu(p), new_mask))
3828 if (migrate_task(p, any_online_cpu(new_mask), &req)) {
3829 /* Need help from migration thread: drop lock and wait. */
3830 task_rq_unlock(rq, &flags);
3831 wake_up_process(rq->migration_thread);
3832 wait_for_completion(&req.done);
3833 tlb_migrate_finish(p->mm);
3837 task_rq_unlock(rq, &flags);
3841 EXPORT_SYMBOL_GPL(set_cpus_allowed);
3844 * Move (not current) task off this cpu, onto dest cpu. We're doing
3845 * this because either it can't run here any more (set_cpus_allowed()
3846 * away from this CPU, or CPU going down), or because we're
3847 * attempting to rebalance this task on exec (sched_balance_exec).
3849 * So we race with normal scheduler movements, but that's OK, as long
3850 * as the task is no longer on this CPU.
3852 static void __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu)
3854 runqueue_t *rq_dest, *rq_src;
3856 if (unlikely(cpu_is_offline(dest_cpu)))
3859 rq_src = cpu_rq(src_cpu);
3860 rq_dest = cpu_rq(dest_cpu);
3862 double_rq_lock(rq_src, rq_dest);
3863 /* Already moved. */
3864 if (task_cpu(p) != src_cpu)
3866 /* Affinity changed (again). */
3867 if (!cpu_isset(dest_cpu, p->cpus_allowed))
3870 set_task_cpu(p, dest_cpu);
3873 * Sync timestamp with rq_dest's before activating.
3874 * The same thing could be achieved by doing this step
3875 * afterwards, and pretending it was a local activate.
3876 * This way is cleaner and logically correct.
3878 p->timestamp = p->timestamp - rq_src->timestamp_last_tick
3879 + rq_dest->timestamp_last_tick;
3880 deactivate_task(p, rq_src);
3881 activate_task(p, rq_dest, 0);
3882 if (TASK_PREEMPTS_CURR(p, rq_dest))
3883 resched_task(rq_dest->curr);
3887 double_rq_unlock(rq_src, rq_dest);
3891 * migration_thread - this is a highprio system thread that performs
3892 * thread migration by bumping thread off CPU then 'pushing' onto
3895 static int migration_thread(void * data)
3898 int cpu = (long)data;
3901 BUG_ON(rq->migration_thread != current);
3903 set_current_state(TASK_INTERRUPTIBLE);
3904 while (!kthread_should_stop()) {
3905 struct list_head *head;
3906 migration_req_t *req;
3908 if (current->flags & PF_FREEZE)
3909 refrigerator(PF_FREEZE);
3911 spin_lock_irq(&rq->lock);
3913 if (cpu_is_offline(cpu)) {
3914 spin_unlock_irq(&rq->lock);
3918 if (rq->active_balance) {
3919 #ifndef CONFIG_CKRM_CPU_SCHEDULE
3920 active_load_balance(rq, cpu);
3922 rq->active_balance = 0;
3925 head = &rq->migration_queue;
3927 if (list_empty(head)) {
3928 spin_unlock_irq(&rq->lock);
3930 set_current_state(TASK_INTERRUPTIBLE);
3933 req = list_entry(head->next, migration_req_t, list);
3934 list_del_init(head->next);
3936 if (req->type == REQ_MOVE_TASK) {
3937 spin_unlock(&rq->lock);
3938 __migrate_task(req->task, smp_processor_id(),
3941 } else if (req->type == REQ_SET_DOMAIN) {
3943 spin_unlock_irq(&rq->lock);
3945 spin_unlock_irq(&rq->lock);
3949 complete(&req->done);
3951 __set_current_state(TASK_RUNNING);
3955 /* Wait for kthread_stop */
3956 set_current_state(TASK_INTERRUPTIBLE);
3957 while (!kthread_should_stop()) {
3959 set_current_state(TASK_INTERRUPTIBLE);
3961 __set_current_state(TASK_RUNNING);
3965 #ifdef CONFIG_HOTPLUG_CPU
3966 /* migrate_all_tasks - function to migrate all tasks from the dead cpu. */
3967 static void migrate_all_tasks(int src_cpu)
3969 struct task_struct *tsk, *t;
3973 write_lock_irq(&tasklist_lock);
3975 /* watch out for per node tasks, let's stay on this node */
3976 node = cpu_to_node(src_cpu);
3978 do_each_thread(t, tsk) {
3983 if (task_cpu(tsk) != src_cpu)
3986 /* Figure out where this task should go (attempting to
3987 * keep it on-node), and check if it can be migrated
3988 * as-is. NOTE that kernel threads bound to more than
3989 * one online cpu will be migrated. */
3990 mask = node_to_cpumask(node);
3991 cpus_and(mask, mask, tsk->cpus_allowed);
3992 dest_cpu = any_online_cpu(mask);
3993 if (dest_cpu == NR_CPUS)
3994 dest_cpu = any_online_cpu(tsk->cpus_allowed);
3995 if (dest_cpu == NR_CPUS) {
3996 cpus_setall(tsk->cpus_allowed);
3997 dest_cpu = any_online_cpu(tsk->cpus_allowed);
3999 /* Don't tell them about moving exiting tasks
4000 or kernel threads (both mm NULL), since
4001 they never leave kernel. */
4002 if (tsk->mm && printk_ratelimit())
4003 printk(KERN_INFO "process %d (%s) no "
4004 "longer affine to cpu%d\n",
4005 tsk->pid, tsk->comm, src_cpu);
4008 __migrate_task(tsk, src_cpu, dest_cpu);
4009 } while_each_thread(t, tsk);
4011 write_unlock_irq(&tasklist_lock);
4014 /* Schedules idle task to be the next runnable task on current CPU.
4015 * It does so by boosting its priority to highest possible and adding it to
4016 * the _front_ of runqueue. Used by CPU offline code.
4018 void sched_idle_next(void)
4020 int cpu = smp_processor_id();
4021 runqueue_t *rq = this_rq();
4022 struct task_struct *p = rq->idle;
4023 unsigned long flags;
4025 /* cpu has to be offline */
4026 BUG_ON(cpu_online(cpu));
4028 /* Strictly not necessary since rest of the CPUs are stopped by now
4029 * and interrupts disabled on current cpu.
4031 spin_lock_irqsave(&rq->lock, flags);
4033 __setscheduler(p, SCHED_FIFO, MAX_RT_PRIO-1);
4034 /* Add idle task to _front_ of it's priority queue */
4035 __activate_idle_task(p, rq);
4037 spin_unlock_irqrestore(&rq->lock, flags);
4039 #endif /* CONFIG_HOTPLUG_CPU */
4042 * migration_call - callback that gets triggered when a CPU is added.
4043 * Here we can start up the necessary migration thread for the new CPU.
4045 static int migration_call(struct notifier_block *nfb, unsigned long action,
4048 int cpu = (long)hcpu;
4049 struct task_struct *p;
4050 struct runqueue *rq;
4051 unsigned long flags;
4054 case CPU_UP_PREPARE:
4055 p = kthread_create(migration_thread, hcpu, "migration/%d",cpu);
4058 p->flags |= PF_NOFREEZE;
4059 kthread_bind(p, cpu);
4060 /* Must be high prio: stop_machine expects to yield to it. */
4061 rq = task_rq_lock(p, &flags);
4062 __setscheduler(p, SCHED_FIFO, MAX_RT_PRIO-1);
4063 task_rq_unlock(rq, &flags);
4064 cpu_rq(cpu)->migration_thread = p;
4067 /* Strictly unneccessary, as first user will wake it. */
4068 wake_up_process(cpu_rq(cpu)->migration_thread);
4070 #ifdef CONFIG_HOTPLUG_CPU
4071 case CPU_UP_CANCELED:
4072 /* Unbind it from offline cpu so it can run. Fall thru. */
4073 kthread_bind(cpu_rq(cpu)->migration_thread,smp_processor_id());
4074 kthread_stop(cpu_rq(cpu)->migration_thread);
4075 cpu_rq(cpu)->migration_thread = NULL;
4078 migrate_all_tasks(cpu);
4080 kthread_stop(rq->migration_thread);
4081 rq->migration_thread = NULL;
4082 /* Idle task back to normal (off runqueue, low prio) */
4083 rq = task_rq_lock(rq->idle, &flags);
4084 deactivate_task(rq->idle, rq);
4085 rq->idle->static_prio = MAX_PRIO;
4086 __setscheduler(rq->idle, SCHED_NORMAL, 0);
4087 task_rq_unlock(rq, &flags);
4088 BUG_ON(rq->nr_running != 0);
4090 /* No need to migrate the tasks: it was best-effort if
4091 * they didn't do lock_cpu_hotplug(). Just wake up
4092 * the requestors. */
4093 spin_lock_irq(&rq->lock);
4094 while (!list_empty(&rq->migration_queue)) {
4095 migration_req_t *req;
4096 req = list_entry(rq->migration_queue.next,
4097 migration_req_t, list);
4098 BUG_ON(req->type != REQ_MOVE_TASK);
4099 list_del_init(&req->list);
4100 complete(&req->done);
4102 spin_unlock_irq(&rq->lock);
4109 /* Register at highest priority so that task migration (migrate_all_tasks)
4110 * happens before everything else.
4112 static struct notifier_block __devinitdata migration_notifier = {
4113 .notifier_call = migration_call,
4117 int __init migration_init(void)
4119 void *cpu = (void *)(long)smp_processor_id();
4120 /* Start one for boot CPU. */
4121 migration_call(&migration_notifier, CPU_UP_PREPARE, cpu);
4122 migration_call(&migration_notifier, CPU_ONLINE, cpu);
4123 register_cpu_notifier(&migration_notifier);
4129 * The 'big kernel lock'
4131 * This spinlock is taken and released recursively by lock_kernel()
4132 * and unlock_kernel(). It is transparently dropped and reaquired
4133 * over schedule(). It is used to protect legacy code that hasn't
4134 * been migrated to a proper locking design yet.
4136 * Don't use in new code.
4138 * Note: spinlock debugging needs this even on !CONFIG_SMP.
4140 spinlock_t kernel_flag __cacheline_aligned_in_smp = SPIN_LOCK_UNLOCKED;
4141 EXPORT_SYMBOL(kernel_flag);
4144 /* Attach the domain 'sd' to 'cpu' as its base domain */
4145 void cpu_attach_domain(struct sched_domain *sd, int cpu)
4147 migration_req_t req;
4148 unsigned long flags;
4149 runqueue_t *rq = cpu_rq(cpu);
4154 spin_lock_irqsave(&rq->lock, flags);
4156 if (cpu == smp_processor_id() || !cpu_online(cpu)) {
4159 init_completion(&req.done);
4160 req.type = REQ_SET_DOMAIN;
4162 list_add(&req.list, &rq->migration_queue);
4166 spin_unlock_irqrestore(&rq->lock, flags);
4169 wake_up_process(rq->migration_thread);
4170 wait_for_completion(&req.done);
4173 unlock_cpu_hotplug();
4176 #ifdef ARCH_HAS_SCHED_DOMAIN
4177 extern void __init arch_init_sched_domains(void);
4179 static struct sched_group sched_group_cpus[NR_CPUS];
4180 static DEFINE_PER_CPU(struct sched_domain, cpu_domains);
4182 static struct sched_group sched_group_nodes[MAX_NUMNODES];
4183 static DEFINE_PER_CPU(struct sched_domain, node_domains);
4184 static void __init arch_init_sched_domains(void)
4187 struct sched_group *first_node = NULL, *last_node = NULL;
4189 /* Set up domains */
4191 int node = cpu_to_node(i);
4192 cpumask_t nodemask = node_to_cpumask(node);
4193 struct sched_domain *node_sd = &per_cpu(node_domains, i);
4194 struct sched_domain *cpu_sd = &per_cpu(cpu_domains, i);
4196 *node_sd = SD_NODE_INIT;
4197 node_sd->span = cpu_possible_map;
4198 node_sd->groups = &sched_group_nodes[cpu_to_node(i)];
4200 *cpu_sd = SD_CPU_INIT;
4201 cpus_and(cpu_sd->span, nodemask, cpu_possible_map);
4202 cpu_sd->groups = &sched_group_cpus[i];
4203 cpu_sd->parent = node_sd;
4207 for (i = 0; i < MAX_NUMNODES; i++) {
4208 cpumask_t tmp = node_to_cpumask(i);
4210 struct sched_group *first_cpu = NULL, *last_cpu = NULL;
4211 struct sched_group *node = &sched_group_nodes[i];
4214 cpus_and(nodemask, tmp, cpu_possible_map);
4216 if (cpus_empty(nodemask))
4219 node->cpumask = nodemask;
4220 node->cpu_power = SCHED_LOAD_SCALE * cpus_weight(node->cpumask);
4222 for_each_cpu_mask(j, node->cpumask) {
4223 struct sched_group *cpu = &sched_group_cpus[j];
4225 cpus_clear(cpu->cpumask);
4226 cpu_set(j, cpu->cpumask);
4227 cpu->cpu_power = SCHED_LOAD_SCALE;
4232 last_cpu->next = cpu;
4235 last_cpu->next = first_cpu;
4240 last_node->next = node;
4243 last_node->next = first_node;
4247 struct sched_domain *cpu_sd = &per_cpu(cpu_domains, i);
4248 cpu_attach_domain(cpu_sd, i);
4252 #else /* !CONFIG_NUMA */
4253 static void __init arch_init_sched_domains(void)
4256 struct sched_group *first_cpu = NULL, *last_cpu = NULL;
4258 /* Set up domains */
4260 struct sched_domain *cpu_sd = &per_cpu(cpu_domains, i);
4262 *cpu_sd = SD_CPU_INIT;
4263 cpu_sd->span = cpu_possible_map;
4264 cpu_sd->groups = &sched_group_cpus[i];
4267 /* Set up CPU groups */
4268 for_each_cpu_mask(i, cpu_possible_map) {
4269 struct sched_group *cpu = &sched_group_cpus[i];
4271 cpus_clear(cpu->cpumask);
4272 cpu_set(i, cpu->cpumask);
4273 cpu->cpu_power = SCHED_LOAD_SCALE;
4278 last_cpu->next = cpu;
4281 last_cpu->next = first_cpu;
4283 mb(); /* domains were modified outside the lock */
4285 struct sched_domain *cpu_sd = &per_cpu(cpu_domains, i);
4286 cpu_attach_domain(cpu_sd, i);
4290 #endif /* CONFIG_NUMA */
4291 #endif /* ARCH_HAS_SCHED_DOMAIN */
4293 #define SCHED_DOMAIN_DEBUG
4294 #ifdef SCHED_DOMAIN_DEBUG
4295 void sched_domain_debug(void)
4300 runqueue_t *rq = cpu_rq(i);
4301 struct sched_domain *sd;
4306 printk(KERN_DEBUG "CPU%d: %s\n",
4307 i, (cpu_online(i) ? " online" : "offline"));
4312 struct sched_group *group = sd->groups;
4313 cpumask_t groupmask;
4315 cpumask_scnprintf(str, NR_CPUS, sd->span);
4316 cpus_clear(groupmask);
4319 for (j = 0; j < level + 1; j++)
4321 printk("domain %d: span %s\n", level, str);
4323 if (!cpu_isset(i, sd->span))
4324 printk(KERN_DEBUG "ERROR domain->span does not contain CPU%d\n", i);
4325 if (!cpu_isset(i, group->cpumask))
4326 printk(KERN_DEBUG "ERROR domain->groups does not contain CPU%d\n", i);
4327 if (!group->cpu_power)
4328 printk(KERN_DEBUG "ERROR domain->cpu_power not set\n");
4331 for (j = 0; j < level + 2; j++)
4336 printk(" ERROR: NULL");
4340 if (!cpus_weight(group->cpumask))
4341 printk(" ERROR empty group:");
4343 if (cpus_intersects(groupmask, group->cpumask))
4344 printk(" ERROR repeated CPUs:");
4346 cpus_or(groupmask, groupmask, group->cpumask);
4348 cpumask_scnprintf(str, NR_CPUS, group->cpumask);
4351 group = group->next;
4352 } while (group != sd->groups);
4355 if (!cpus_equal(sd->span, groupmask))
4356 printk(KERN_DEBUG "ERROR groups don't span domain->span\n");
4362 if (!cpus_subset(groupmask, sd->span))
4363 printk(KERN_DEBUG "ERROR parent span is not a superset of domain->span\n");
4370 #define sched_domain_debug() {}
4373 void __init sched_init_smp(void)
4375 arch_init_sched_domains();
4376 sched_domain_debug();
4379 void __init sched_init_smp(void)
4382 #endif /* CONFIG_SMP */
4384 int in_sched_functions(unsigned long addr)
4386 /* Linker adds these: start and end of __sched functions */
4387 extern char __sched_text_start[], __sched_text_end[];
4388 return addr >= (unsigned long)__sched_text_start
4389 && addr < (unsigned long)__sched_text_end;
4392 void __init sched_init(void)
4396 #ifndef CONFIG_CKRM_CPU_SCHEDULE
4401 /* Set up an initial dummy domain for early boot */
4402 static struct sched_domain sched_domain_init;
4403 static struct sched_group sched_group_init;
4405 memset(&sched_domain_init, 0, sizeof(struct sched_domain));
4406 sched_domain_init.span = CPU_MASK_ALL;
4407 sched_domain_init.groups = &sched_group_init;
4408 sched_domain_init.last_balance = jiffies;
4409 sched_domain_init.balance_interval = INT_MAX; /* Don't balance */
4411 memset(&sched_group_init, 0, sizeof(struct sched_group));
4412 sched_group_init.cpumask = CPU_MASK_ALL;
4413 sched_group_init.next = &sched_group_init;
4414 sched_group_init.cpu_power = SCHED_LOAD_SCALE;
4419 for (i = 0; i < NR_CPUS; i++) {
4420 #ifndef CONFIG_CKRM_CPU_SCHEDULE
4421 prio_array_t *array;
4424 spin_lock_init(&rq->lock);
4426 #ifndef CONFIG_CKRM_CPU_SCHEDULE
4427 rq->active = rq->arrays;
4428 rq->expired = rq->arrays + 1;
4430 rq->ckrm_cpu_load = 0;
4432 rq->best_expired_prio = MAX_PRIO;
4435 rq->sd = &sched_domain_init;
4437 rq->active_balance = 0;
4439 rq->migration_thread = NULL;
4440 INIT_LIST_HEAD(&rq->migration_queue);
4442 INIT_LIST_HEAD(&rq->hold_queue);
4443 atomic_set(&rq->nr_iowait, 0);
4445 #ifndef CONFIG_CKRM_CPU_SCHEDULE
4446 for (j = 0; j < 2; j++) {
4447 array = rq->arrays + j;
4448 for (k = 0; k < MAX_PRIO; k++) {
4449 INIT_LIST_HEAD(array->queue + k);
4450 __clear_bit(k, array->bitmap);
4452 // delimiter for bitsearch
4453 __set_bit(MAX_PRIO, array->bitmap);
4459 * We have to do a little magic to get the first
4460 * thread right in SMP mode.
4465 set_task_cpu(current, smp_processor_id());
4466 #ifdef CONFIG_CKRM_CPU_SCHEDULE
4467 current->cpu_class = default_cpu_class;
4468 current->array = NULL;
4470 wake_up_forked_process(current);
4473 * The boot idle thread does lazy MMU switching as well:
4475 atomic_inc(&init_mm.mm_count);
4476 enter_lazy_tlb(&init_mm, current);
4479 #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP
4480 void __might_sleep(char *file, int line, int atomic_depth)
4482 #if defined(in_atomic)
4483 static unsigned long prev_jiffy; /* ratelimiting */
4485 #ifndef CONFIG_PREEMPT
4488 if (((in_atomic() != atomic_depth) || irqs_disabled()) &&
4489 system_state == SYSTEM_RUNNING) {
4490 if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy)
4492 prev_jiffy = jiffies;
4493 printk(KERN_ERR "Debug: sleeping function called from invalid"
4494 " context at %s:%d\n", file, line);
4495 printk("in_atomic():%d[expected: %d], irqs_disabled():%d\n",
4496 in_atomic(), atomic_depth, irqs_disabled());
4501 EXPORT_SYMBOL(__might_sleep);
4505 #if defined(CONFIG_SMP) && defined(CONFIG_PREEMPT)
4507 * This could be a long-held lock. If another CPU holds it for a long time,
4508 * and that CPU is not asked to reschedule then *this* CPU will spin on the
4509 * lock for a long time, even if *this* CPU is asked to reschedule.
4511 * So what we do here, in the slow (contended) path is to spin on the lock by
4512 * hand while permitting preemption.
4514 * Called inside preempt_disable().
4516 void __sched __preempt_spin_lock(spinlock_t *lock)
4518 if (preempt_count() > 1) {
4519 _raw_spin_lock(lock);
4524 while (spin_is_locked(lock))
4527 } while (!_raw_spin_trylock(lock));
4530 EXPORT_SYMBOL(__preempt_spin_lock);
4532 void __sched __preempt_write_lock(rwlock_t *lock)
4534 if (preempt_count() > 1) {
4535 _raw_write_lock(lock);
4541 while (rwlock_is_locked(lock))
4544 } while (!_raw_write_trylock(lock));
4547 EXPORT_SYMBOL(__preempt_write_lock);
4548 #endif /* defined(CONFIG_SMP) && defined(CONFIG_PREEMPT) */
4550 #ifdef CONFIG_DELAY_ACCT
4551 int task_running_sys(struct task_struct *p)
4553 return task_running(task_rq(p),p);
4555 EXPORT_SYMBOL(task_running_sys);
4558 #ifdef CONFIG_CKRM_CPU_SCHEDULE
4560 * return the classqueue object of a certain processor
4561 * Note: not supposed to be used in performance sensitive functions
4563 struct classqueue_struct * get_cpu_classqueue(int cpu)
4565 return (& (cpu_rq(cpu)->classqueue) );