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>
43 #include <linux/vserver/sched.h>
44 #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 BITMAP_SIZE ((((MAX_PRIO+1+7)/8)+sizeof(long)-1)/sizeof(long))
204 unsigned int nr_active;
205 unsigned long bitmap[BITMAP_SIZE];
206 struct list_head queue[MAX_PRIO];
208 #define rq_active(p,rq) (rq->active)
209 #define rq_expired(p,rq) (rq->expired)
210 #define ckrm_rebalance_tick(j,this_cpu) do {} while (0)
211 #define TASK_PREEMPTS_CURR(p, rq) \
212 ((p)->prio < (rq)->curr->prio)
216 * This is the main, per-CPU runqueue data structure.
218 * Locking rule: those places that want to lock multiple runqueues
219 * (such as the load balancing or the thread migration code), lock
220 * acquire operations must be ordered by ascending &runqueue.
226 * nr_running and cpu_load should be in the same cacheline because
227 * remote CPUs use both these fields when doing load calculation.
229 unsigned long nr_running;
230 #if defined(CONFIG_SMP)
231 unsigned long cpu_load;
233 unsigned long long nr_switches, nr_preempt;
234 unsigned long expired_timestamp, nr_uninterruptible;
235 unsigned long long timestamp_last_tick;
237 struct mm_struct *prev_mm;
238 #ifdef CONFIG_CKRM_CPU_SCHEDULE
239 unsigned long ckrm_cpu_load;
240 struct classqueue_struct classqueue;
242 prio_array_t *active, *expired, arrays[2];
244 int best_expired_prio;
248 struct sched_domain *sd;
250 /* For active balancing */
254 task_t *migration_thread;
255 struct list_head migration_queue;
257 struct list_head hold_queue;
261 static DEFINE_PER_CPU(struct runqueue, runqueues);
263 #define for_each_domain(cpu, domain) \
264 for (domain = cpu_rq(cpu)->sd; domain; domain = domain->parent)
266 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
267 #define this_rq() (&__get_cpu_var(runqueues))
268 #define task_rq(p) cpu_rq(task_cpu(p))
269 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
272 * Default context-switch locking:
274 #ifndef prepare_arch_switch
275 # define prepare_arch_switch(rq, next) do { } while (0)
276 # define finish_arch_switch(rq, next) spin_unlock_irq(&(rq)->lock)
277 # define task_running(rq, p) ((rq)->curr == (p))
280 #ifdef CONFIG_CKRM_CPU_SCHEDULE
281 #include <linux/ckrm_sched.h>
282 spinlock_t cvt_lock = SPIN_LOCK_UNLOCKED;
283 rwlock_t class_list_lock = RW_LOCK_UNLOCKED;
284 LIST_HEAD(active_cpu_classes); // list of active cpu classes; anchor
285 struct ckrm_cpu_class default_cpu_class_obj;
288 * the minimum CVT allowed is the base_cvt
289 * otherwise, it will starve others
291 CVT_t get_min_cvt(int cpu)
294 struct ckrm_local_runqueue * lrq;
297 node = classqueue_get_head(bpt_queue(cpu));
298 lrq = (node) ? class_list_entry(node) : NULL;
301 min_cvt = lrq->local_cvt;
309 * update the classueue base for all the runqueues
310 * TODO: we can only update half of the min_base to solve the movebackward issue
312 static inline void check_update_class_base(int this_cpu) {
313 unsigned long min_base = 0xFFFFFFFF;
317 if (! cpu_online(this_cpu)) return;
320 * find the min_base across all the processors
322 for_each_online_cpu(i) {
324 * I should change it to directly use bpt->base
326 node = classqueue_get_head(bpt_queue(i));
327 if (node && node->prio < min_base) {
328 min_base = node->prio;
331 if (min_base != 0xFFFFFFFF)
332 classqueue_update_base(bpt_queue(this_cpu),min_base);
335 static inline void ckrm_rebalance_tick(int j,int this_cpu)
337 #ifdef CONFIG_CKRM_CPU_SCHEDULE
338 read_lock(&class_list_lock);
339 if (!(j % CVT_UPDATE_TICK))
340 update_global_cvts(this_cpu);
342 #define CKRM_BASE_UPDATE_RATE 400
343 if (! (jiffies % CKRM_BASE_UPDATE_RATE))
344 check_update_class_base(this_cpu);
346 read_unlock(&class_list_lock);
350 static inline struct ckrm_local_runqueue *rq_get_next_class(struct runqueue *rq)
352 cq_node_t *node = classqueue_get_head(&rq->classqueue);
353 return ((node) ? class_list_entry(node) : NULL);
356 static inline struct task_struct * rq_get_next_task(struct runqueue* rq)
359 struct task_struct *next;
360 struct ckrm_local_runqueue *queue;
361 int cpu = smp_processor_id();
365 if ((queue = rq_get_next_class(rq))) {
366 array = queue->active;
367 //check switch active/expired queue
368 if (unlikely(!queue->active->nr_active)) {
371 array = queue->active;
372 queue->active = queue->expired;
373 queue->expired = array;
374 queue->expired_timestamp = 0;
376 if (queue->active->nr_active)
377 set_top_priority(queue,
378 find_first_bit(queue->active->bitmap, MAX_PRIO));
380 classqueue_dequeue(queue->classqueue,
381 &queue->classqueue_linkobj);
382 cpu_demand_event(get_rq_local_stat(queue,cpu),CPU_DEMAND_DEQUEUE,0);
385 goto retry_next_class;
387 BUG_ON(!queue->active->nr_active);
388 next = task_list_entry(array->queue[queue->top_priority].next);
393 static inline void rq_load_inc(runqueue_t *rq, struct task_struct *p) { rq->ckrm_cpu_load += cpu_class_weight(p->cpu_class); }
394 static inline void rq_load_dec(runqueue_t *rq, struct task_struct *p) { rq->ckrm_cpu_load -= cpu_class_weight(p->cpu_class); }
396 #else /*CONFIG_CKRM_CPU_SCHEDULE*/
398 static inline struct task_struct * rq_get_next_task(struct runqueue* rq)
401 struct list_head *queue;
405 if (unlikely(!array->nr_active)) {
407 * Switch the active and expired arrays.
409 rq->active = rq->expired;
412 rq->expired_timestamp = 0;
413 rq->best_expired_prio = MAX_PRIO;
416 idx = sched_find_first_bit(array->bitmap);
417 queue = array->queue + idx;
418 return list_entry(queue->next, task_t, run_list);
421 static inline void class_enqueue_task(struct task_struct* p, prio_array_t *array) { }
422 static inline void class_dequeue_task(struct task_struct* p, prio_array_t *array) { }
423 static inline void init_cpu_classes(void) { }
424 static inline void rq_load_inc(runqueue_t *rq, struct task_struct *p) { }
425 static inline void rq_load_dec(runqueue_t *rq, struct task_struct *p) { }
426 #endif /* CONFIG_CKRM_CPU_SCHEDULE */
430 * task_rq_lock - lock the runqueue a given task resides on and disable
431 * interrupts. Note the ordering: we can safely lookup the task_rq without
432 * explicitly disabling preemption.
434 runqueue_t *task_rq_lock(task_t *p, unsigned long *flags)
439 local_irq_save(*flags);
441 spin_lock(&rq->lock);
442 if (unlikely(rq != task_rq(p))) {
443 spin_unlock_irqrestore(&rq->lock, *flags);
444 goto repeat_lock_task;
449 void task_rq_unlock(runqueue_t *rq, unsigned long *flags)
451 spin_unlock_irqrestore(&rq->lock, *flags);
455 * rq_lock - lock a given runqueue and disable interrupts.
457 static runqueue_t *this_rq_lock(void)
463 spin_lock(&rq->lock);
468 static inline void rq_unlock(runqueue_t *rq)
470 spin_unlock_irq(&rq->lock);
474 * Adding/removing a task to/from a priority array:
476 void dequeue_task(struct task_struct *p, prio_array_t *array)
480 list_del(&p->run_list);
481 if (list_empty(array->queue + p->prio))
482 __clear_bit(p->prio, array->bitmap);
483 class_dequeue_task(p,array);
486 void enqueue_task(struct task_struct *p, prio_array_t *array)
488 list_add_tail(&p->run_list, array->queue + p->prio);
489 __set_bit(p->prio, array->bitmap);
492 class_enqueue_task(p,array);
496 * Used by the migration code - we pull tasks from the head of the
497 * remote queue so we want these tasks to show up at the head of the
500 static inline void enqueue_task_head(struct task_struct *p, prio_array_t *array)
502 list_add(&p->run_list, array->queue + p->prio);
503 __set_bit(p->prio, array->bitmap);
506 class_enqueue_task(p,array);
510 * effective_prio - return the priority that is based on the static
511 * priority but is modified by bonuses/penalties.
513 * We scale the actual sleep average [0 .... MAX_SLEEP_AVG]
514 * into the -5 ... 0 ... +5 bonus/penalty range.
516 * We use 25% of the full 0...39 priority range so that:
518 * 1) nice +19 interactive tasks do not preempt nice 0 CPU hogs.
519 * 2) nice -20 CPU hogs do not get preempted by nice 0 tasks.
521 * Both properties are important to certain workloads.
523 static int effective_prio(task_t *p)
530 bonus = CURRENT_BONUS(p) - MAX_BONUS / 2;
532 prio = p->static_prio - bonus;
533 if (__vx_task_flags(p, VXF_SCHED_PRIO, 0))
534 prio += effective_vavavoom(p, MAX_USER_PRIO);
536 if (prio < MAX_RT_PRIO)
538 if (prio > MAX_PRIO-1)
544 * __activate_task - move a task to the runqueue.
546 static inline void __activate_task(task_t *p, runqueue_t *rq)
548 enqueue_task(p, rq_active(p,rq));
554 * __activate_idle_task - move idle task to the _front_ of runqueue.
556 static inline void __activate_idle_task(task_t *p, runqueue_t *rq)
558 enqueue_task_head(p, rq_active(p,rq));
563 static void recalc_task_prio(task_t *p, unsigned long long now)
565 unsigned long long __sleep_time = now - p->timestamp;
566 unsigned long sleep_time;
568 if (__sleep_time > NS_MAX_SLEEP_AVG)
569 sleep_time = NS_MAX_SLEEP_AVG;
571 sleep_time = (unsigned long)__sleep_time;
573 if (likely(sleep_time > 0)) {
575 * User tasks that sleep a long time are categorised as
576 * idle and will get just interactive status to stay active &
577 * prevent them suddenly becoming cpu hogs and starving
580 if (p->mm && p->activated != -1 &&
581 sleep_time > INTERACTIVE_SLEEP(p)) {
582 p->sleep_avg = JIFFIES_TO_NS(MAX_SLEEP_AVG -
585 p->interactive_credit++;
588 * The lower the sleep avg a task has the more
589 * rapidly it will rise with sleep time.
591 sleep_time *= (MAX_BONUS - CURRENT_BONUS(p)) ? : 1;
594 * Tasks with low interactive_credit are limited to
595 * one timeslice worth of sleep avg bonus.
598 sleep_time > JIFFIES_TO_NS(task_timeslice(p)))
599 sleep_time = JIFFIES_TO_NS(task_timeslice(p));
602 * Non high_credit tasks waking from uninterruptible
603 * sleep are limited in their sleep_avg rise as they
604 * are likely to be cpu hogs waiting on I/O
606 if (p->activated == -1 && !HIGH_CREDIT(p) && p->mm) {
607 if (p->sleep_avg >= INTERACTIVE_SLEEP(p))
609 else if (p->sleep_avg + sleep_time >=
610 INTERACTIVE_SLEEP(p)) {
611 p->sleep_avg = INTERACTIVE_SLEEP(p);
617 * This code gives a bonus to interactive tasks.
619 * The boost works by updating the 'average sleep time'
620 * value here, based on ->timestamp. The more time a
621 * task spends sleeping, the higher the average gets -
622 * and the higher the priority boost gets as well.
624 p->sleep_avg += sleep_time;
626 if (p->sleep_avg > NS_MAX_SLEEP_AVG) {
627 p->sleep_avg = NS_MAX_SLEEP_AVG;
629 p->interactive_credit++;
634 p->prio = effective_prio(p);
638 * activate_task - move a task to the runqueue and do priority recalculation
640 * Update all the scheduling statistics stuff. (sleep average
641 * calculation, priority modifiers, etc.)
643 static void activate_task(task_t *p, runqueue_t *rq, int local)
645 unsigned long long now;
650 /* Compensate for drifting sched_clock */
651 runqueue_t *this_rq = this_rq();
652 now = (now - this_rq->timestamp_last_tick)
653 + rq->timestamp_last_tick;
657 recalc_task_prio(p, now);
660 * This checks to make sure it's not an uninterruptible task
661 * that is now waking up.
665 * Tasks which were woken up by interrupts (ie. hw events)
666 * are most likely of interactive nature. So we give them
667 * the credit of extending their sleep time to the period
668 * of time they spend on the runqueue, waiting for execution
669 * on a CPU, first time around:
675 * Normal first-time wakeups get a credit too for
676 * on-runqueue time, but it will be weighted down:
683 __activate_task(p, rq);
687 * deactivate_task - remove a task from the runqueue.
689 static void deactivate_task(struct task_struct *p, runqueue_t *rq)
693 if (p->state == TASK_UNINTERRUPTIBLE)
694 rq->nr_uninterruptible++;
695 dequeue_task(p, p->array);
700 * resched_task - mark a task 'to be rescheduled now'.
702 * On UP this means the setting of the need_resched flag, on SMP it
703 * might also involve a cross-CPU call to trigger the scheduler on
707 static void resched_task(task_t *p)
709 int need_resched, nrpolling;
712 /* minimise the chance of sending an interrupt to poll_idle() */
713 nrpolling = test_tsk_thread_flag(p,TIF_POLLING_NRFLAG);
714 need_resched = test_and_set_tsk_thread_flag(p,TIF_NEED_RESCHED);
715 nrpolling |= test_tsk_thread_flag(p,TIF_POLLING_NRFLAG);
717 if (!need_resched && !nrpolling && (task_cpu(p) != smp_processor_id()))
718 smp_send_reschedule(task_cpu(p));
722 static inline void resched_task(task_t *p)
724 set_tsk_need_resched(p);
729 * task_curr - is this task currently executing on a CPU?
730 * @p: the task in question.
732 inline int task_curr(const task_t *p)
734 return cpu_curr(task_cpu(p)) == p;
744 struct list_head list;
745 enum request_type type;
747 /* For REQ_MOVE_TASK */
751 /* For REQ_SET_DOMAIN */
752 struct sched_domain *sd;
754 struct completion done;
758 * The task's runqueue lock must be held.
759 * Returns true if you have to wait for migration thread.
761 static int migrate_task(task_t *p, int dest_cpu, migration_req_t *req)
763 runqueue_t *rq = task_rq(p);
766 * If the task is not on a runqueue (and not running), then
767 * it is sufficient to simply update the task's cpu field.
769 if (!p->array && !task_running(rq, p)) {
770 set_task_cpu(p, dest_cpu);
774 init_completion(&req->done);
775 req->type = REQ_MOVE_TASK;
777 req->dest_cpu = dest_cpu;
778 list_add(&req->list, &rq->migration_queue);
783 * wait_task_inactive - wait for a thread to unschedule.
785 * The caller must ensure that the task *will* unschedule sometime soon,
786 * else this function might spin for a *long* time. This function can't
787 * be called with interrupts off, or it may introduce deadlock with
788 * smp_call_function() if an IPI is sent by the same process we are
789 * waiting to become inactive.
791 void wait_task_inactive(task_t * p)
798 rq = task_rq_lock(p, &flags);
799 /* Must be off runqueue entirely, not preempted. */
800 if (unlikely(p->array)) {
801 /* If it's preempted, we yield. It could be a while. */
802 preempted = !task_running(rq, p);
803 task_rq_unlock(rq, &flags);
809 task_rq_unlock(rq, &flags);
813 * kick_process - kick a running thread to enter/exit the kernel
814 * @p: the to-be-kicked thread
816 * Cause a process which is running on another CPU to enter
817 * kernel-mode, without any delay. (to get signals handled.)
819 void kick_process(task_t *p)
825 if ((cpu != smp_processor_id()) && task_curr(p))
826 smp_send_reschedule(cpu);
830 EXPORT_SYMBOL_GPL(kick_process);
833 * Return a low guess at the load of a migration-source cpu.
835 * We want to under-estimate the load of migration sources, to
836 * balance conservatively.
838 static inline unsigned long source_load(int cpu)
840 runqueue_t *rq = cpu_rq(cpu);
841 unsigned long load_now = rq->nr_running * SCHED_LOAD_SCALE;
843 return min(rq->cpu_load, load_now);
847 * Return a high guess at the load of a migration-target cpu
849 static inline unsigned long target_load(int cpu)
851 runqueue_t *rq = cpu_rq(cpu);
852 unsigned long load_now = rq->nr_running * SCHED_LOAD_SCALE;
854 return max(rq->cpu_load, load_now);
860 * wake_idle() is useful especially on SMT architectures to wake a
861 * task onto an idle sibling if we would otherwise wake it onto a
864 * Returns the CPU we should wake onto.
866 #if defined(ARCH_HAS_SCHED_WAKE_IDLE)
867 static int wake_idle(int cpu, task_t *p)
870 runqueue_t *rq = cpu_rq(cpu);
871 struct sched_domain *sd;
878 if (!(sd->flags & SD_WAKE_IDLE))
881 cpus_and(tmp, sd->span, cpu_online_map);
882 cpus_and(tmp, tmp, p->cpus_allowed);
884 for_each_cpu_mask(i, tmp) {
892 static inline int wake_idle(int cpu, task_t *p)
899 * try_to_wake_up - wake up a thread
900 * @p: the to-be-woken-up thread
901 * @state: the mask of task states that can be woken
902 * @sync: do a synchronous wakeup?
904 * Put it on the run-queue if it's not already there. The "current"
905 * thread is always on the run-queue (except when the actual
906 * re-schedule is in progress), and as such you're allowed to do
907 * the simpler "current->state = TASK_RUNNING" to mark yourself
908 * runnable without the overhead of this.
910 * returns failure only if the task is already active.
912 static int try_to_wake_up(task_t * p, unsigned int state, int sync)
914 int cpu, this_cpu, success = 0;
919 unsigned long load, this_load;
920 struct sched_domain *sd;
924 rq = task_rq_lock(p, &flags);
925 old_state = p->state;
926 if (!(old_state & state))
933 this_cpu = smp_processor_id();
936 if (unlikely(task_running(rq, p)))
941 if (cpu == this_cpu || unlikely(!cpu_isset(this_cpu, p->cpus_allowed)))
944 load = source_load(cpu);
945 this_load = target_load(this_cpu);
948 * If sync wakeup then subtract the (maximum possible) effect of
949 * the currently running task from the load of the current CPU:
952 this_load -= SCHED_LOAD_SCALE;
954 /* Don't pull the task off an idle CPU to a busy one */
955 if (load < SCHED_LOAD_SCALE/2 && this_load > SCHED_LOAD_SCALE/2)
958 new_cpu = this_cpu; /* Wake to this CPU if we can */
961 * Scan domains for affine wakeup and passive balancing
964 for_each_domain(this_cpu, sd) {
965 unsigned int imbalance;
967 * Start passive balancing when half the imbalance_pct
970 imbalance = sd->imbalance_pct + (sd->imbalance_pct - 100) / 2;
972 if ( ((sd->flags & SD_WAKE_AFFINE) &&
973 !task_hot(p, rq->timestamp_last_tick, sd))
974 || ((sd->flags & SD_WAKE_BALANCE) &&
975 imbalance*this_load <= 100*load) ) {
977 * Now sd has SD_WAKE_AFFINE and p is cache cold in sd
978 * or sd has SD_WAKE_BALANCE and there is an imbalance
980 if (cpu_isset(cpu, sd->span))
985 new_cpu = cpu; /* Could not wake to this_cpu. Wake to cpu instead */
987 new_cpu = wake_idle(new_cpu, p);
988 if (new_cpu != cpu && cpu_isset(new_cpu, p->cpus_allowed)) {
989 set_task_cpu(p, new_cpu);
990 task_rq_unlock(rq, &flags);
991 /* might preempt at this point */
992 rq = task_rq_lock(p, &flags);
993 old_state = p->state;
994 if (!(old_state & state))
999 this_cpu = smp_processor_id();
1004 #endif /* CONFIG_SMP */
1005 if (old_state == TASK_UNINTERRUPTIBLE) {
1006 rq->nr_uninterruptible--;
1008 * Tasks on involuntary sleep don't earn
1009 * sleep_avg beyond just interactive state.
1015 * Sync wakeups (i.e. those types of wakeups where the waker
1016 * has indicated that it will leave the CPU in short order)
1017 * don't trigger a preemption, if the woken up task will run on
1018 * this cpu. (in this case the 'I will reschedule' promise of
1019 * the waker guarantees that the freshly woken up task is going
1020 * to be considered on this CPU.)
1022 activate_task(p, rq, cpu == this_cpu);
1023 if (!sync || cpu != this_cpu) {
1024 if (TASK_PREEMPTS_CURR(p, rq))
1025 resched_task(rq->curr);
1030 p->state = TASK_RUNNING;
1032 task_rq_unlock(rq, &flags);
1037 int fastcall wake_up_process(task_t * p)
1039 return try_to_wake_up(p, TASK_STOPPED |
1040 TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE, 0);
1043 EXPORT_SYMBOL(wake_up_process);
1045 int fastcall wake_up_state(task_t *p, unsigned int state)
1047 return try_to_wake_up(p, state, 0);
1051 * Perform scheduler related setup for a newly forked process p.
1052 * p is forked by current.
1054 void fastcall sched_fork(task_t *p)
1057 * We mark the process as running here, but have not actually
1058 * inserted it onto the runqueue yet. This guarantees that
1059 * nobody will actually run it, and a signal or other external
1060 * event cannot wake it up and insert it on the runqueue either.
1062 p->state = TASK_RUNNING;
1063 INIT_LIST_HEAD(&p->run_list);
1065 spin_lock_init(&p->switch_lock);
1066 #ifdef CONFIG_PREEMPT
1068 * During context-switch we hold precisely one spinlock, which
1069 * schedule_tail drops. (in the common case it's this_rq()->lock,
1070 * but it also can be p->switch_lock.) So we compensate with a count
1071 * of 1. Also, we want to start with kernel preemption disabled.
1073 p->thread_info->preempt_count = 1;
1076 * Share the timeslice between parent and child, thus the
1077 * total amount of pending timeslices in the system doesn't change,
1078 * resulting in more scheduling fairness.
1080 local_irq_disable();
1081 p->time_slice = (current->time_slice + 1) >> 1;
1083 * The remainder of the first timeslice might be recovered by
1084 * the parent if the child exits early enough.
1086 p->first_time_slice = 1;
1087 current->time_slice >>= 1;
1088 p->timestamp = sched_clock();
1089 if (!current->time_slice) {
1091 * This case is rare, it happens when the parent has only
1092 * a single jiffy left from its timeslice. Taking the
1093 * runqueue lock is not a problem.
1095 current->time_slice = 1;
1097 scheduler_tick(0, 0);
1105 * wake_up_forked_process - wake up a freshly forked process.
1107 * This function will do some initial scheduler statistics housekeeping
1108 * that must be done for every newly created process.
1110 void fastcall wake_up_forked_process(task_t * p)
1112 unsigned long flags;
1113 runqueue_t *rq = task_rq_lock(current, &flags);
1115 BUG_ON(p->state != TASK_RUNNING);
1118 * We decrease the sleep average of forking parents
1119 * and children as well, to keep max-interactive tasks
1120 * from forking tasks that are max-interactive.
1122 current->sleep_avg = JIFFIES_TO_NS(CURRENT_BONUS(current) *
1123 PARENT_PENALTY / 100 * MAX_SLEEP_AVG / MAX_BONUS);
1125 p->sleep_avg = JIFFIES_TO_NS(CURRENT_BONUS(p) *
1126 CHILD_PENALTY / 100 * MAX_SLEEP_AVG / MAX_BONUS);
1128 p->interactive_credit = 0;
1130 p->prio = effective_prio(p);
1131 set_task_cpu(p, smp_processor_id());
1133 if (unlikely(!current->array))
1134 __activate_task(p, rq);
1136 p->prio = current->prio;
1137 list_add_tail(&p->run_list, ¤t->run_list);
1138 p->array = current->array;
1139 p->array->nr_active++;
1143 task_rq_unlock(rq, &flags);
1147 * Potentially available exiting-child timeslices are
1148 * retrieved here - this way the parent does not get
1149 * penalized for creating too many threads.
1151 * (this cannot be used to 'generate' timeslices
1152 * artificially, because any timeslice recovered here
1153 * was given away by the parent in the first place.)
1155 void fastcall sched_exit(task_t * p)
1157 unsigned long flags;
1160 local_irq_save(flags);
1161 if (p->first_time_slice) {
1162 p->parent->time_slice += p->time_slice;
1163 if (unlikely(p->parent->time_slice > MAX_TIMESLICE))
1164 p->parent->time_slice = MAX_TIMESLICE;
1166 local_irq_restore(flags);
1168 * If the child was a (relative-) CPU hog then decrease
1169 * the sleep_avg of the parent as well.
1171 rq = task_rq_lock(p->parent, &flags);
1172 if (p->sleep_avg < p->parent->sleep_avg)
1173 p->parent->sleep_avg = p->parent->sleep_avg /
1174 (EXIT_WEIGHT + 1) * EXIT_WEIGHT + p->sleep_avg /
1176 task_rq_unlock(rq, &flags);
1180 * finish_task_switch - clean up after a task-switch
1181 * @prev: the thread we just switched away from.
1183 * We enter this with the runqueue still locked, and finish_arch_switch()
1184 * will unlock it along with doing any other architecture-specific cleanup
1187 * Note that we may have delayed dropping an mm in context_switch(). If
1188 * so, we finish that here outside of the runqueue lock. (Doing it
1189 * with the lock held can cause deadlocks; see schedule() for
1192 static void finish_task_switch(task_t *prev)
1194 runqueue_t *rq = this_rq();
1195 struct mm_struct *mm = rq->prev_mm;
1196 unsigned long prev_task_flags;
1201 * A task struct has one reference for the use as "current".
1202 * If a task dies, then it sets TASK_ZOMBIE in tsk->state and calls
1203 * schedule one last time. The schedule call will never return,
1204 * and the scheduled task must drop that reference.
1205 * The test for TASK_ZOMBIE must occur while the runqueue locks are
1206 * still held, otherwise prev could be scheduled on another cpu, die
1207 * there before we look at prev->state, and then the reference would
1209 * Manfred Spraul <manfred@colorfullife.com>
1211 prev_task_flags = prev->flags;
1212 finish_arch_switch(rq, prev);
1215 if (unlikely(prev_task_flags & PF_DEAD))
1216 put_task_struct(prev);
1220 * schedule_tail - first thing a freshly forked thread must call.
1221 * @prev: the thread we just switched away from.
1223 asmlinkage void schedule_tail(task_t *prev)
1225 finish_task_switch(prev);
1227 if (current->set_child_tid)
1228 put_user(current->pid, current->set_child_tid);
1232 * context_switch - switch to the new MM and the new
1233 * thread's register state.
1236 task_t * context_switch(runqueue_t *rq, task_t *prev, task_t *next)
1238 struct mm_struct *mm = next->mm;
1239 struct mm_struct *oldmm = prev->active_mm;
1241 if (unlikely(!mm)) {
1242 next->active_mm = oldmm;
1243 atomic_inc(&oldmm->mm_count);
1244 enter_lazy_tlb(oldmm, next);
1246 switch_mm(oldmm, mm, next);
1248 if (unlikely(!prev->mm)) {
1249 prev->active_mm = NULL;
1250 WARN_ON(rq->prev_mm);
1251 rq->prev_mm = oldmm;
1254 /* Here we just switch the register state and the stack. */
1255 switch_to(prev, next, prev);
1261 * nr_running, nr_uninterruptible and nr_context_switches:
1263 * externally visible scheduler statistics: current number of runnable
1264 * threads, current number of uninterruptible-sleeping threads, total
1265 * number of context switches performed since bootup.
1267 unsigned long nr_running(void)
1269 unsigned long i, sum = 0;
1272 sum += cpu_rq(i)->nr_running;
1277 unsigned long nr_uninterruptible(void)
1279 unsigned long i, sum = 0;
1281 for_each_online_cpu(i)
1282 sum += cpu_rq(i)->nr_uninterruptible;
1287 unsigned long long nr_context_switches(void)
1289 unsigned long long i, sum = 0;
1291 for_each_online_cpu(i)
1292 sum += cpu_rq(i)->nr_switches;
1297 unsigned long nr_iowait(void)
1299 unsigned long i, sum = 0;
1301 for_each_online_cpu(i)
1302 sum += atomic_read(&cpu_rq(i)->nr_iowait);
1308 * double_rq_lock - safely lock two runqueues
1310 * Note this does not disable interrupts like task_rq_lock,
1311 * you need to do so manually before calling.
1313 static void double_rq_lock(runqueue_t *rq1, runqueue_t *rq2)
1316 spin_lock(&rq1->lock);
1319 spin_lock(&rq1->lock);
1320 spin_lock(&rq2->lock);
1322 spin_lock(&rq2->lock);
1323 spin_lock(&rq1->lock);
1329 * double_rq_unlock - safely unlock two runqueues
1331 * Note this does not restore interrupts like task_rq_unlock,
1332 * you need to do so manually after calling.
1334 static void double_rq_unlock(runqueue_t *rq1, runqueue_t *rq2)
1336 spin_unlock(&rq1->lock);
1338 spin_unlock(&rq2->lock);
1341 unsigned long long nr_preempt(void)
1343 unsigned long long i, sum = 0;
1345 for_each_online_cpu(i)
1346 sum += cpu_rq(i)->nr_preempt;
1361 * find_idlest_cpu - find the least busy runqueue.
1363 static int find_idlest_cpu(struct task_struct *p, int this_cpu,
1364 struct sched_domain *sd)
1366 unsigned long load, min_load, this_load;
1371 min_load = ULONG_MAX;
1373 cpus_and(mask, sd->span, cpu_online_map);
1374 cpus_and(mask, mask, p->cpus_allowed);
1376 for_each_cpu_mask(i, mask) {
1377 load = target_load(i);
1379 if (load < min_load) {
1383 /* break out early on an idle CPU: */
1389 /* add +1 to account for the new task */
1390 this_load = source_load(this_cpu) + SCHED_LOAD_SCALE;
1393 * Would with the addition of the new task to the
1394 * current CPU there be an imbalance between this
1395 * CPU and the idlest CPU?
1397 * Use half of the balancing threshold - new-context is
1398 * a good opportunity to balance.
1400 if (min_load*(100 + (sd->imbalance_pct-100)/2) < this_load*100)
1407 * wake_up_forked_thread - wake up a freshly forked thread.
1409 * This function will do some initial scheduler statistics housekeeping
1410 * that must be done for every newly created context, and it also does
1411 * runqueue balancing.
1413 void fastcall wake_up_forked_thread(task_t * p)
1415 unsigned long flags;
1416 int this_cpu = get_cpu(), cpu;
1417 struct sched_domain *tmp, *sd = NULL;
1418 runqueue_t *this_rq = cpu_rq(this_cpu), *rq;
1421 * Find the largest domain that this CPU is part of that
1422 * is willing to balance on clone:
1424 for_each_domain(this_cpu, tmp)
1425 if (tmp->flags & SD_BALANCE_CLONE)
1428 cpu = find_idlest_cpu(p, this_cpu, sd);
1432 local_irq_save(flags);
1435 double_rq_lock(this_rq, rq);
1437 BUG_ON(p->state != TASK_RUNNING);
1440 * We did find_idlest_cpu() unlocked, so in theory
1441 * the mask could have changed - just dont migrate
1444 if (unlikely(!cpu_isset(cpu, p->cpus_allowed))) {
1446 double_rq_unlock(this_rq, rq);
1450 * We decrease the sleep average of forking parents
1451 * and children as well, to keep max-interactive tasks
1452 * from forking tasks that are max-interactive.
1454 current->sleep_avg = JIFFIES_TO_NS(CURRENT_BONUS(current) *
1455 PARENT_PENALTY / 100 * MAX_SLEEP_AVG / MAX_BONUS);
1457 p->sleep_avg = JIFFIES_TO_NS(CURRENT_BONUS(p) *
1458 CHILD_PENALTY / 100 * MAX_SLEEP_AVG / MAX_BONUS);
1460 p->interactive_credit = 0;
1462 p->prio = effective_prio(p);
1463 set_task_cpu(p, cpu);
1465 if (cpu == this_cpu) {
1466 if (unlikely(!current->array))
1467 __activate_task(p, rq);
1469 p->prio = current->prio;
1470 list_add_tail(&p->run_list, ¤t->run_list);
1471 p->array = current->array;
1472 p->array->nr_active++;
1477 /* Not the local CPU - must adjust timestamp */
1478 p->timestamp = (p->timestamp - this_rq->timestamp_last_tick)
1479 + rq->timestamp_last_tick;
1480 __activate_task(p, rq);
1481 if (TASK_PREEMPTS_CURR(p, rq))
1482 resched_task(rq->curr);
1485 double_rq_unlock(this_rq, rq);
1486 local_irq_restore(flags);
1491 * If dest_cpu is allowed for this process, migrate the task to it.
1492 * This is accomplished by forcing the cpu_allowed mask to only
1493 * allow dest_cpu, which will force the cpu onto dest_cpu. Then
1494 * the cpu_allowed mask is restored.
1496 static void sched_migrate_task(task_t *p, int dest_cpu)
1498 migration_req_t req;
1500 unsigned long flags;
1502 rq = task_rq_lock(p, &flags);
1503 if (!cpu_isset(dest_cpu, p->cpus_allowed)
1504 || unlikely(cpu_is_offline(dest_cpu)))
1507 /* force the process onto the specified CPU */
1508 if (migrate_task(p, dest_cpu, &req)) {
1509 /* Need to wait for migration thread (might exit: take ref). */
1510 struct task_struct *mt = rq->migration_thread;
1511 get_task_struct(mt);
1512 task_rq_unlock(rq, &flags);
1513 wake_up_process(mt);
1514 put_task_struct(mt);
1515 wait_for_completion(&req.done);
1519 task_rq_unlock(rq, &flags);
1523 * sched_balance_exec(): find the highest-level, exec-balance-capable
1524 * domain and try to migrate the task to the least loaded CPU.
1526 * execve() is a valuable balancing opportunity, because at this point
1527 * the task has the smallest effective memory and cache footprint.
1529 void sched_balance_exec(void)
1531 struct sched_domain *tmp, *sd = NULL;
1532 int new_cpu, this_cpu = get_cpu();
1534 /* Prefer the current CPU if there's only this task running */
1535 if (this_rq()->nr_running <= 1)
1538 for_each_domain(this_cpu, tmp)
1539 if (tmp->flags & SD_BALANCE_EXEC)
1543 new_cpu = find_idlest_cpu(current, this_cpu, sd);
1544 if (new_cpu != this_cpu) {
1546 sched_migrate_task(current, new_cpu);
1555 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1557 static void double_lock_balance(runqueue_t *this_rq, runqueue_t *busiest)
1559 if (unlikely(!spin_trylock(&busiest->lock))) {
1560 if (busiest < this_rq) {
1561 spin_unlock(&this_rq->lock);
1562 spin_lock(&busiest->lock);
1563 spin_lock(&this_rq->lock);
1565 spin_lock(&busiest->lock);
1570 * pull_task - move a task from a remote runqueue to the local runqueue.
1571 * Both runqueues must be locked.
1574 void pull_task(runqueue_t *src_rq, prio_array_t *src_array, task_t *p,
1575 runqueue_t *this_rq, prio_array_t *this_array, int this_cpu)
1577 dequeue_task(p, src_array);
1578 src_rq->nr_running--;
1579 rq_load_dec(src_rq,p);
1581 set_task_cpu(p, this_cpu);
1582 this_rq->nr_running++;
1583 rq_load_inc(this_rq,p);
1584 enqueue_task(p, this_array);
1586 p->timestamp = (p->timestamp - src_rq->timestamp_last_tick)
1587 + this_rq->timestamp_last_tick;
1589 * Note that idle threads have a prio of MAX_PRIO, for this test
1590 * to be always true for them.
1592 if (TASK_PREEMPTS_CURR(p, this_rq))
1593 resched_task(this_rq->curr);
1597 * can_migrate_task - may task p from runqueue rq be migrated to this_cpu?
1600 int can_migrate_task(task_t *p, runqueue_t *rq, int this_cpu,
1601 struct sched_domain *sd, enum idle_type idle)
1604 * We do not migrate tasks that are:
1605 * 1) running (obviously), or
1606 * 2) cannot be migrated to this CPU due to cpus_allowed, or
1607 * 3) are cache-hot on their current CPU.
1609 if (task_running(rq, p))
1611 if (!cpu_isset(this_cpu, p->cpus_allowed))
1614 /* Aggressive migration if we've failed balancing */
1615 if (idle == NEWLY_IDLE ||
1616 sd->nr_balance_failed < sd->cache_nice_tries) {
1617 if (task_hot(p, rq->timestamp_last_tick, sd))
1624 #ifdef CONFIG_CKRM_CPU_SCHEDULE
1626 struct ckrm_cpu_class *find_unbalanced_class(int busiest_cpu, int this_cpu, unsigned long *cls_imbalance)
1628 struct ckrm_cpu_class *most_unbalanced_class = NULL;
1629 struct ckrm_cpu_class *clsptr;
1630 int max_unbalance = 0;
1632 list_for_each_entry(clsptr,&active_cpu_classes,links) {
1633 struct ckrm_local_runqueue *this_lrq = get_ckrm_local_runqueue(clsptr,this_cpu);
1634 struct ckrm_local_runqueue *busiest_lrq = get_ckrm_local_runqueue(clsptr,busiest_cpu);
1635 int unbalance_degree;
1637 unbalance_degree = (local_queue_nr_running(busiest_lrq) - local_queue_nr_running(this_lrq)) * cpu_class_weight(clsptr);
1638 if (unbalance_degree >= *cls_imbalance)
1639 continue; // already looked at this class
1641 if (unbalance_degree > max_unbalance) {
1642 max_unbalance = unbalance_degree;
1643 most_unbalanced_class = clsptr;
1646 *cls_imbalance = max_unbalance;
1647 return most_unbalanced_class;
1652 * find_busiest_queue - find the busiest runqueue among the cpus in cpumask.
1654 static int find_busiest_cpu(runqueue_t *this_rq, int this_cpu, int idle,
1657 int cpu_load, load, max_load, i, busiest_cpu;
1658 runqueue_t *busiest, *rq_src;
1661 /*Hubertus ... the concept of nr_running is replace with cpu_load */
1662 cpu_load = this_rq->ckrm_cpu_load;
1668 for_each_online_cpu(i) {
1670 load = rq_src->ckrm_cpu_load;
1672 if ((load > max_load) && (rq_src != this_rq)) {
1679 if (likely(!busiest))
1682 *imbalance = max_load - cpu_load;
1684 /* It needs an at least ~25% imbalance to trigger balancing. */
1685 if (!idle && ((*imbalance)*4 < max_load)) {
1690 double_lock_balance(this_rq, busiest);
1692 * Make sure nothing changed since we checked the
1695 if (busiest->ckrm_cpu_load <= cpu_load) {
1696 spin_unlock(&busiest->lock);
1700 return (busiest ? busiest_cpu : -1);
1703 static int load_balance(int this_cpu, runqueue_t *this_rq,
1704 struct sched_domain *sd, enum idle_type idle)
1708 runqueue_t *busiest;
1709 prio_array_t *array;
1710 struct list_head *head, *curr;
1712 struct ckrm_local_runqueue * busiest_local_queue;
1713 struct ckrm_cpu_class *clsptr;
1715 unsigned long cls_imbalance; // so we can retry other classes
1717 // need to update global CVT based on local accumulated CVTs
1718 read_lock(&class_list_lock);
1719 busiest_cpu = find_busiest_cpu(this_rq, this_cpu, idle, &imbalance);
1720 if (busiest_cpu == -1)
1723 busiest = cpu_rq(busiest_cpu);
1726 * We only want to steal a number of tasks equal to 1/2 the imbalance,
1727 * otherwise we'll just shift the imbalance to the new queue:
1731 /* now find class on that runqueue with largest inbalance */
1732 cls_imbalance = 0xFFFFFFFF;
1735 clsptr = find_unbalanced_class(busiest_cpu, this_cpu, &cls_imbalance);
1739 busiest_local_queue = get_ckrm_local_runqueue(clsptr,busiest_cpu);
1740 weight = cpu_class_weight(clsptr);
1743 * We first consider expired tasks. Those will likely not be
1744 * executed in the near future, and they are most likely to
1745 * be cache-cold, thus switching CPUs has the least effect
1748 if (busiest_local_queue->expired->nr_active)
1749 array = busiest_local_queue->expired;
1751 array = busiest_local_queue->active;
1754 /* Start searching at priority 0: */
1758 idx = sched_find_first_bit(array->bitmap);
1760 idx = find_next_bit(array->bitmap, MAX_PRIO, idx);
1761 if (idx >= MAX_PRIO) {
1762 if (array == busiest_local_queue->expired && busiest_local_queue->active->nr_active) {
1763 array = busiest_local_queue->active;
1766 goto retry_other_class;
1769 head = array->queue + idx;
1772 tmp = list_entry(curr, task_t, run_list);
1776 if (!can_migrate_task(tmp, busiest, this_cpu, sd,idle)) {
1782 pull_task(busiest, array, tmp, this_rq, rq_active(tmp,this_rq),this_cpu);
1784 * tmp BUG FIX: hzheng
1785 * load balancing can make the busiest local queue empty
1786 * thus it should be removed from bpt
1788 if (! local_queue_nr_running(busiest_local_queue)) {
1789 classqueue_dequeue(busiest_local_queue->classqueue,&busiest_local_queue->classqueue_linkobj);
1790 cpu_demand_event(get_rq_local_stat(busiest_local_queue,busiest_cpu),CPU_DEMAND_DEQUEUE,0);
1793 imbalance -= weight;
1794 if (!idle && (imbalance>0)) {
1801 spin_unlock(&busiest->lock);
1803 read_unlock(&class_list_lock);
1808 static inline void idle_balance(int this_cpu, runqueue_t *this_rq)
1811 #else /* CONFIG_CKRM_CPU_SCHEDULE */
1813 * move_tasks tries to move up to max_nr_move tasks from busiest to this_rq,
1814 * as part of a balancing operation within "domain". Returns the number of
1817 * Called with both runqueues locked.
1819 static int move_tasks(runqueue_t *this_rq, int this_cpu, runqueue_t *busiest,
1820 unsigned long max_nr_move, struct sched_domain *sd,
1821 enum idle_type idle)
1823 prio_array_t *array, *dst_array;
1824 struct list_head *head, *curr;
1825 int idx, pulled = 0;
1828 if (max_nr_move <= 0 || busiest->nr_running <= 1)
1832 * We first consider expired tasks. Those will likely not be
1833 * executed in the near future, and they are most likely to
1834 * be cache-cold, thus switching CPUs has the least effect
1837 if (busiest->expired->nr_active) {
1838 array = busiest->expired;
1839 dst_array = this_rq->expired;
1841 array = busiest->active;
1842 dst_array = this_rq->active;
1846 /* Start searching at priority 0: */
1850 idx = sched_find_first_bit(array->bitmap);
1852 idx = find_next_bit(array->bitmap, MAX_PRIO, idx);
1853 if (idx >= MAX_PRIO) {
1854 if (array == busiest->expired && busiest->active->nr_active) {
1855 array = busiest->active;
1856 dst_array = this_rq->active;
1862 head = array->queue + idx;
1865 tmp = list_entry(curr, task_t, run_list);
1869 if (!can_migrate_task(tmp, busiest, this_cpu, sd, idle)) {
1875 pull_task(busiest, array, tmp, this_rq, dst_array, this_cpu);
1878 /* We only want to steal up to the prescribed number of tasks. */
1879 if (pulled < max_nr_move) {
1890 * find_busiest_group finds and returns the busiest CPU group within the
1891 * domain. It calculates and returns the number of tasks which should be
1892 * moved to restore balance via the imbalance parameter.
1894 static struct sched_group *
1895 find_busiest_group(struct sched_domain *sd, int this_cpu,
1896 unsigned long *imbalance, enum idle_type idle)
1898 struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups;
1899 unsigned long max_load, avg_load, total_load, this_load, total_pwr;
1901 max_load = this_load = total_load = total_pwr = 0;
1909 local_group = cpu_isset(this_cpu, group->cpumask);
1911 /* Tally up the load of all CPUs in the group */
1913 cpus_and(tmp, group->cpumask, cpu_online_map);
1914 if (unlikely(cpus_empty(tmp)))
1917 for_each_cpu_mask(i, tmp) {
1918 /* Bias balancing toward cpus of our domain */
1920 load = target_load(i);
1922 load = source_load(i);
1931 total_load += avg_load;
1932 total_pwr += group->cpu_power;
1934 /* Adjust by relative CPU power of the group */
1935 avg_load = (avg_load * SCHED_LOAD_SCALE) / group->cpu_power;
1938 this_load = avg_load;
1941 } else if (avg_load > max_load) {
1942 max_load = avg_load;
1946 group = group->next;
1947 } while (group != sd->groups);
1949 if (!busiest || this_load >= max_load)
1952 avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr;
1954 if (this_load >= avg_load ||
1955 100*max_load <= sd->imbalance_pct*this_load)
1959 * We're trying to get all the cpus to the average_load, so we don't
1960 * want to push ourselves above the average load, nor do we wish to
1961 * reduce the max loaded cpu below the average load, as either of these
1962 * actions would just result in more rebalancing later, and ping-pong
1963 * tasks around. Thus we look for the minimum possible imbalance.
1964 * Negative imbalances (*we* are more loaded than anyone else) will
1965 * be counted as no imbalance for these purposes -- we can't fix that
1966 * by pulling tasks to us. Be careful of negative numbers as they'll
1967 * appear as very large values with unsigned longs.
1969 *imbalance = min(max_load - avg_load, avg_load - this_load);
1971 /* How much load to actually move to equalise the imbalance */
1972 *imbalance = (*imbalance * min(busiest->cpu_power, this->cpu_power))
1975 if (*imbalance < SCHED_LOAD_SCALE - 1) {
1976 unsigned long pwr_now = 0, pwr_move = 0;
1979 if (max_load - this_load >= SCHED_LOAD_SCALE*2) {
1985 * OK, we don't have enough imbalance to justify moving tasks,
1986 * however we may be able to increase total CPU power used by
1990 pwr_now += busiest->cpu_power*min(SCHED_LOAD_SCALE, max_load);
1991 pwr_now += this->cpu_power*min(SCHED_LOAD_SCALE, this_load);
1992 pwr_now /= SCHED_LOAD_SCALE;
1994 /* Amount of load we'd subtract */
1995 tmp = SCHED_LOAD_SCALE*SCHED_LOAD_SCALE/busiest->cpu_power;
1997 pwr_move += busiest->cpu_power*min(SCHED_LOAD_SCALE,
2000 /* Amount of load we'd add */
2001 tmp = SCHED_LOAD_SCALE*SCHED_LOAD_SCALE/this->cpu_power;
2004 pwr_move += this->cpu_power*min(SCHED_LOAD_SCALE, this_load + tmp);
2005 pwr_move /= SCHED_LOAD_SCALE;
2007 /* Move if we gain another 8th of a CPU worth of throughput */
2008 if (pwr_move < pwr_now + SCHED_LOAD_SCALE / 8)
2015 /* Get rid of the scaling factor, rounding down as we divide */
2016 *imbalance = (*imbalance + 1) / SCHED_LOAD_SCALE;
2021 if (busiest && (idle == NEWLY_IDLE ||
2022 (idle == IDLE && max_load > SCHED_LOAD_SCALE)) ) {
2032 * find_busiest_queue - find the busiest runqueue among the cpus in group.
2034 static runqueue_t *find_busiest_queue(struct sched_group *group)
2037 unsigned long load, max_load = 0;
2038 runqueue_t *busiest = NULL;
2041 cpus_and(tmp, group->cpumask, cpu_online_map);
2042 for_each_cpu_mask(i, tmp) {
2043 load = source_load(i);
2045 if (load > max_load) {
2047 busiest = cpu_rq(i);
2055 * Check this_cpu to ensure it is balanced within domain. Attempt to move
2056 * tasks if there is an imbalance.
2058 * Called with this_rq unlocked.
2060 static int load_balance(int this_cpu, runqueue_t *this_rq,
2061 struct sched_domain *sd, enum idle_type idle)
2063 struct sched_group *group;
2064 runqueue_t *busiest;
2065 unsigned long imbalance;
2068 spin_lock(&this_rq->lock);
2070 group = find_busiest_group(sd, this_cpu, &imbalance, idle);
2074 busiest = find_busiest_queue(group);
2078 * This should be "impossible", but since load
2079 * balancing is inherently racy and statistical,
2080 * it could happen in theory.
2082 if (unlikely(busiest == this_rq)) {
2088 if (busiest->nr_running > 1) {
2090 * Attempt to move tasks. If find_busiest_group has found
2091 * an imbalance but busiest->nr_running <= 1, the group is
2092 * still unbalanced. nr_moved simply stays zero, so it is
2093 * correctly treated as an imbalance.
2095 double_lock_balance(this_rq, busiest);
2096 nr_moved = move_tasks(this_rq, this_cpu, busiest,
2097 imbalance, sd, idle);
2098 spin_unlock(&busiest->lock);
2100 spin_unlock(&this_rq->lock);
2103 sd->nr_balance_failed++;
2105 if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) {
2108 spin_lock(&busiest->lock);
2109 if (!busiest->active_balance) {
2110 busiest->active_balance = 1;
2111 busiest->push_cpu = this_cpu;
2114 spin_unlock(&busiest->lock);
2116 wake_up_process(busiest->migration_thread);
2119 * We've kicked active balancing, reset the failure
2122 sd->nr_balance_failed = sd->cache_nice_tries;
2125 sd->nr_balance_failed = 0;
2127 /* We were unbalanced, so reset the balancing interval */
2128 sd->balance_interval = sd->min_interval;
2133 spin_unlock(&this_rq->lock);
2135 /* tune up the balancing interval */
2136 if (sd->balance_interval < sd->max_interval)
2137 sd->balance_interval *= 2;
2143 * Check this_cpu to ensure it is balanced within domain. Attempt to move
2144 * tasks if there is an imbalance.
2146 * Called from schedule when this_rq is about to become idle (NEWLY_IDLE).
2147 * this_rq is locked.
2149 static int load_balance_newidle(int this_cpu, runqueue_t *this_rq,
2150 struct sched_domain *sd)
2152 struct sched_group *group;
2153 runqueue_t *busiest = NULL;
2154 unsigned long imbalance;
2157 group = find_busiest_group(sd, this_cpu, &imbalance, NEWLY_IDLE);
2161 busiest = find_busiest_queue(group);
2162 if (!busiest || busiest == this_rq)
2165 /* Attempt to move tasks */
2166 double_lock_balance(this_rq, busiest);
2168 nr_moved = move_tasks(this_rq, this_cpu, busiest,
2169 imbalance, sd, NEWLY_IDLE);
2171 spin_unlock(&busiest->lock);
2178 * idle_balance is called by schedule() if this_cpu is about to become
2179 * idle. Attempts to pull tasks from other CPUs.
2181 static inline void idle_balance(int this_cpu, runqueue_t *this_rq)
2183 struct sched_domain *sd;
2185 for_each_domain(this_cpu, sd) {
2186 if (sd->flags & SD_BALANCE_NEWIDLE) {
2187 if (load_balance_newidle(this_cpu, this_rq, sd)) {
2188 /* We've pulled tasks over so stop searching */
2196 * active_load_balance is run by migration threads. It pushes a running
2197 * task off the cpu. It can be required to correctly have at least 1 task
2198 * running on each physical CPU where possible, and not have a physical /
2199 * logical imbalance.
2201 * Called with busiest locked.
2203 static void active_load_balance(runqueue_t *busiest, int busiest_cpu)
2205 struct sched_domain *sd;
2206 struct sched_group *group, *busy_group;
2209 if (busiest->nr_running <= 1)
2212 for_each_domain(busiest_cpu, sd)
2213 if (cpu_isset(busiest->push_cpu, sd->span))
2221 while (!cpu_isset(busiest_cpu, group->cpumask))
2222 group = group->next;
2231 if (group == busy_group)
2234 cpus_and(tmp, group->cpumask, cpu_online_map);
2235 if (!cpus_weight(tmp))
2238 for_each_cpu_mask(i, tmp) {
2244 rq = cpu_rq(push_cpu);
2247 * This condition is "impossible", but since load
2248 * balancing is inherently a bit racy and statistical,
2249 * it can trigger.. Reported by Bjorn Helgaas on a
2252 if (unlikely(busiest == rq))
2254 double_lock_balance(busiest, rq);
2255 move_tasks(rq, push_cpu, busiest, 1, sd, IDLE);
2256 spin_unlock(&rq->lock);
2258 group = group->next;
2259 } while (group != sd->groups);
2261 #endif /* CONFIG_CKRM_CPU_SCHEDULE*/
2264 * rebalance_tick will get called every timer tick, on every CPU.
2266 * It checks each scheduling domain to see if it is due to be balanced,
2267 * and initiates a balancing operation if so.
2269 * Balancing parameters are set up in arch_init_sched_domains.
2272 /* Don't have all balancing operations going off at once */
2273 #define CPU_OFFSET(cpu) (HZ * cpu / NR_CPUS)
2275 static void rebalance_tick(int this_cpu, runqueue_t *this_rq,
2276 enum idle_type idle)
2278 unsigned long old_load, this_load;
2279 unsigned long j = jiffies + CPU_OFFSET(this_cpu);
2280 struct sched_domain *sd;
2282 ckrm_rebalance_tick(j,this_cpu);
2284 /* Update our load */
2285 old_load = this_rq->cpu_load;
2286 this_load = this_rq->nr_running * SCHED_LOAD_SCALE;
2288 * Round up the averaging division if load is increasing. This
2289 * prevents us from getting stuck on 9 if the load is 10, for
2292 if (this_load > old_load)
2294 this_rq->cpu_load = (old_load + this_load) / 2;
2296 for_each_domain(this_cpu, sd) {
2297 unsigned long interval = sd->balance_interval;
2300 interval *= sd->busy_factor;
2302 /* scale ms to jiffies */
2303 interval = msecs_to_jiffies(interval);
2304 if (unlikely(!interval))
2307 if (j - sd->last_balance >= interval) {
2308 if (load_balance(this_cpu, this_rq, sd, idle)) {
2309 /* We've pulled tasks over so no longer idle */
2312 sd->last_balance += interval;
2318 * on UP we do not need to balance between CPUs:
2320 static inline void rebalance_tick(int cpu, runqueue_t *rq, enum idle_type idle)
2322 ckrm_rebalance_tick(jiffies,cpu);
2325 static inline void idle_balance(int cpu, runqueue_t *rq)
2330 static inline int wake_priority_sleeper(runqueue_t *rq)
2332 #ifdef CONFIG_SCHED_SMT
2334 * If an SMT sibling task has been put to sleep for priority
2335 * reasons reschedule the idle task to see if it can now run.
2337 if (rq->nr_running) {
2338 resched_task(rq->idle);
2345 DEFINE_PER_CPU(struct kernel_stat, kstat) = { { 0 } };
2347 EXPORT_PER_CPU_SYMBOL(kstat);
2350 * We place interactive tasks back into the active array, if possible.
2352 * To guarantee that this does not starve expired tasks we ignore the
2353 * interactivity of a task if the first expired task had to wait more
2354 * than a 'reasonable' amount of time. This deadline timeout is
2355 * load-dependent, as the frequency of array switched decreases with
2356 * increasing number of running tasks. We also ignore the interactivity
2357 * if a better static_prio task has expired:
2360 #ifndef CONFIG_CKRM_CPU_SCHEDULE
2361 #define EXPIRED_STARVING(rq) \
2362 ((STARVATION_LIMIT && ((rq)->expired_timestamp && \
2363 (jiffies - (rq)->expired_timestamp >= \
2364 STARVATION_LIMIT * ((rq)->nr_running) + 1))) || \
2365 ((rq)->curr->static_prio > (rq)->best_expired_prio))
2367 #define EXPIRED_STARVING(rq) \
2368 (STARVATION_LIMIT && ((rq)->expired_timestamp && \
2369 (jiffies - (rq)->expired_timestamp >= \
2370 STARVATION_LIMIT * (local_queue_nr_running(rq)) + 1)))
2374 * This function gets called by the timer code, with HZ frequency.
2375 * We call it with interrupts disabled.
2377 * It also gets called by the fork code, when changing the parent's
2380 void scheduler_tick(int user_ticks, int sys_ticks)
2382 int cpu = smp_processor_id();
2383 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
2384 runqueue_t *rq = this_rq();
2385 task_t *p = current;
2387 rq->timestamp_last_tick = sched_clock();
2389 if (rcu_pending(cpu))
2390 rcu_check_callbacks(cpu, user_ticks);
2392 /* note: this timer irq context must be accounted for as well */
2393 if (hardirq_count() - HARDIRQ_OFFSET) {
2394 cpustat->irq += sys_ticks;
2396 } else if (softirq_count()) {
2397 cpustat->softirq += sys_ticks;
2401 if (p == rq->idle) {
2402 if (!--rq->idle_tokens && !list_empty(&rq->hold_queue))
2405 if (atomic_read(&rq->nr_iowait) > 0)
2406 cpustat->iowait += sys_ticks;
2408 cpustat->idle += sys_ticks;
2409 if (wake_priority_sleeper(rq))
2411 rebalance_tick(cpu, rq, IDLE);
2414 if (TASK_NICE(p) > 0)
2415 cpustat->nice += user_ticks;
2417 cpustat->user += user_ticks;
2418 cpustat->system += sys_ticks;
2420 /* Task might have expired already, but not scheduled off yet */
2421 if (p->array != rq_active(p,rq)) {
2422 set_tsk_need_resched(p);
2425 spin_lock(&rq->lock);
2427 * The task was running during this tick - update the
2428 * time slice counter. Note: we do not update a thread's
2429 * priority until it either goes to sleep or uses up its
2430 * timeslice. This makes it possible for interactive tasks
2431 * to use up their timeslices at their highest priority levels.
2433 if (unlikely(rt_task(p))) {
2435 * RR tasks need a special form of timeslice management.
2436 * FIFO tasks have no timeslices.
2438 if ((p->policy == SCHED_RR) && !--p->time_slice) {
2439 p->time_slice = task_timeslice(p);
2440 p->first_time_slice = 0;
2441 set_tsk_need_resched(p);
2443 /* put it at the end of the queue: */
2444 dequeue_task(p, rq_active(p,rq));
2445 enqueue_task(p, rq_active(p,rq));
2449 #warning MEF PLANETLAB: "if (vx_need_resched(p)) was if (!--p->time_slice) */"
2450 if (vx_need_resched(p)) {
2451 #ifdef CONFIG_CKRM_CPU_SCHEDULE
2452 /* Hubertus ... we can abstract this out */
2453 struct ckrm_local_runqueue* rq = get_task_class_queue(p);
2455 dequeue_task(p, rq->active);
2456 set_tsk_need_resched(p);
2457 p->prio = effective_prio(p);
2458 p->time_slice = task_timeslice(p);
2459 p->first_time_slice = 0;
2461 if (!rq->expired_timestamp)
2462 rq->expired_timestamp = jiffies;
2463 if (!TASK_INTERACTIVE(p) || EXPIRED_STARVING(rq)) {
2464 enqueue_task(p, rq->expired);
2465 if (p->static_prio < this_rq()->best_expired_prio)
2466 this_rq()->best_expired_prio = p->static_prio;
2468 enqueue_task(p, rq->active);
2471 * Prevent a too long timeslice allowing a task to monopolize
2472 * the CPU. We do this by splitting up the timeslice into
2475 * Note: this does not mean the task's timeslices expire or
2476 * get lost in any way, they just might be preempted by
2477 * another task of equal priority. (one with higher
2478 * priority would have preempted this task already.) We
2479 * requeue this task to the end of the list on this priority
2480 * level, which is in essence a round-robin of tasks with
2483 * This only applies to tasks in the interactive
2484 * delta range with at least TIMESLICE_GRANULARITY to requeue.
2486 if (TASK_INTERACTIVE(p) && !((task_timeslice(p) -
2487 p->time_slice) % TIMESLICE_GRANULARITY(p)) &&
2488 (p->time_slice >= TIMESLICE_GRANULARITY(p)) &&
2489 (p->array == rq_active(p,rq))) {
2491 dequeue_task(p, rq_active(p,rq));
2492 set_tsk_need_resched(p);
2493 p->prio = effective_prio(p);
2494 enqueue_task(p, rq_active(p,rq));
2498 spin_unlock(&rq->lock);
2500 rebalance_tick(cpu, rq, NOT_IDLE);
2503 #ifdef CONFIG_SCHED_SMT
2504 static inline void wake_sleeping_dependent(int cpu, runqueue_t *rq)
2507 struct sched_domain *sd = rq->sd;
2508 cpumask_t sibling_map;
2510 if (!(sd->flags & SD_SHARE_CPUPOWER))
2513 cpus_and(sibling_map, sd->span, cpu_online_map);
2514 for_each_cpu_mask(i, sibling_map) {
2523 * If an SMT sibling task is sleeping due to priority
2524 * reasons wake it up now.
2526 if (smt_rq->curr == smt_rq->idle && smt_rq->nr_running)
2527 resched_task(smt_rq->idle);
2531 static inline int dependent_sleeper(int cpu, runqueue_t *rq, task_t *p)
2533 struct sched_domain *sd = rq->sd;
2534 cpumask_t sibling_map;
2537 if (!(sd->flags & SD_SHARE_CPUPOWER))
2540 cpus_and(sibling_map, sd->span, cpu_online_map);
2541 for_each_cpu_mask(i, sibling_map) {
2549 smt_curr = smt_rq->curr;
2552 * If a user task with lower static priority than the
2553 * running task on the SMT sibling is trying to schedule,
2554 * delay it till there is proportionately less timeslice
2555 * left of the sibling task to prevent a lower priority
2556 * task from using an unfair proportion of the
2557 * physical cpu's resources. -ck
2559 if (((smt_curr->time_slice * (100 - sd->per_cpu_gain) / 100) >
2560 task_timeslice(p) || rt_task(smt_curr)) &&
2561 p->mm && smt_curr->mm && !rt_task(p))
2565 * Reschedule a lower priority task on the SMT sibling,
2566 * or wake it up if it has been put to sleep for priority
2569 if ((((p->time_slice * (100 - sd->per_cpu_gain) / 100) >
2570 task_timeslice(smt_curr) || rt_task(p)) &&
2571 smt_curr->mm && p->mm && !rt_task(smt_curr)) ||
2572 (smt_curr == smt_rq->idle && smt_rq->nr_running))
2573 resched_task(smt_curr);
2578 static inline void wake_sleeping_dependent(int cpu, runqueue_t *rq)
2582 static inline int dependent_sleeper(int cpu, runqueue_t *rq, task_t *p)
2589 * schedule() is the main scheduler function.
2591 asmlinkage void __sched schedule(void)
2594 task_t *prev, *next;
2596 prio_array_t *array;
2597 unsigned long long now;
2598 unsigned long run_time;
2600 #ifdef CONFIG_VSERVER_HARDCPU
2601 struct vx_info *vxi;
2605 //WARN_ON(system_state == SYSTEM_BOOTING);
2607 * Test if we are atomic. Since do_exit() needs to call into
2608 * schedule() atomically, we ignore that path for now.
2609 * Otherwise, whine if we are scheduling when we should not be.
2611 if (likely(!(current->state & (TASK_DEAD | TASK_ZOMBIE)))) {
2612 if (unlikely(in_atomic())) {
2613 printk(KERN_ERR "bad: scheduling while atomic!\n");
2623 release_kernel_lock(prev);
2624 now = sched_clock();
2625 if (likely(now - prev->timestamp < NS_MAX_SLEEP_AVG))
2626 run_time = now - prev->timestamp;
2628 run_time = NS_MAX_SLEEP_AVG;
2631 * Tasks with interactive credits get charged less run_time
2632 * at high sleep_avg to delay them losing their interactive
2635 if (HIGH_CREDIT(prev))
2636 run_time /= (CURRENT_BONUS(prev) ? : 1);
2638 spin_lock_irq(&rq->lock);
2641 * if entering off of a kernel preemption go straight
2642 * to picking the next task.
2644 switch_count = &prev->nivcsw;
2645 if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) {
2646 switch_count = &prev->nvcsw;
2647 if (unlikely((prev->state & TASK_INTERRUPTIBLE) &&
2648 unlikely(signal_pending(prev))))
2649 prev->state = TASK_RUNNING;
2651 deactivate_task(prev, rq);
2654 cpu = smp_processor_id();
2655 #ifdef CONFIG_VSERVER_HARDCPU
2656 if (!list_empty(&rq->hold_queue)) {
2657 struct list_head *l, *n;
2661 list_for_each_safe(l, n, &rq->hold_queue) {
2662 next = list_entry(l, task_t, run_list);
2663 if (vxi == next->vx_info)
2666 vxi = next->vx_info;
2667 ret = vx_tokens_recalc(vxi);
2668 // tokens = vx_tokens_avail(next);
2671 list_del(&next->run_list);
2672 next->state &= ~TASK_ONHOLD;
2673 recalc_task_prio(next, now);
2674 __activate_task(next, rq);
2675 // printk("··· unhold %p\n", next);
2678 if ((ret < 0) && (maxidle < ret))
2682 rq->idle_tokens = -maxidle;
2686 if (unlikely(!rq->nr_running)) {
2687 idle_balance(cpu, rq);
2688 if (!rq->nr_running) {
2690 rq->expired_timestamp = 0;
2691 wake_sleeping_dependent(cpu, rq);
2696 next = rq_get_next_task(rq);
2697 if (next == rq->idle)
2700 if (dependent_sleeper(cpu, rq, next)) {
2705 #ifdef CONFIG_VSERVER_HARDCPU
2706 vxi = next->vx_info;
2707 if (vxi && __vx_flags(vxi->vx_flags,
2708 VXF_SCHED_PAUSE|VXF_SCHED_HARD, 0)) {
2709 int ret = vx_tokens_recalc(vxi);
2711 if (unlikely(ret <= 0)) {
2712 if (ret && (rq->idle_tokens > -ret))
2713 rq->idle_tokens = -ret;
2714 deactivate_task(next, rq);
2715 list_add_tail(&next->run_list, &rq->hold_queue);
2716 next->state |= TASK_ONHOLD;
2722 if (!rt_task(next) && next->activated > 0) {
2723 unsigned long long delta = now - next->timestamp;
2725 if (next->activated == 1)
2726 delta = delta * (ON_RUNQUEUE_WEIGHT * 128 / 100) / 128;
2728 array = next->array;
2729 dequeue_task(next, array);
2730 recalc_task_prio(next, next->timestamp + delta);
2731 enqueue_task(next, array);
2733 next->activated = 0;
2736 if (test_and_clear_tsk_thread_flag(prev,TIF_NEED_RESCHED))
2738 RCU_qsctr(task_cpu(prev))++;
2740 #ifdef CONFIG_CKRM_CPU_SCHEDULE
2741 if (prev != rq->idle) {
2742 unsigned long long run = now - prev->timestamp;
2743 cpu_demand_event(get_task_local_stat(prev),CPU_DEMAND_DESCHEDULE,run);
2744 update_local_cvt(prev, run);
2748 prev->sleep_avg -= run_time;
2749 if ((long)prev->sleep_avg <= 0) {
2750 prev->sleep_avg = 0;
2751 if (!(HIGH_CREDIT(prev) || LOW_CREDIT(prev)))
2752 prev->interactive_credit--;
2754 add_delay_ts(prev,runcpu_total,prev->timestamp,now);
2755 prev->timestamp = now;
2757 if (likely(prev != next)) {
2758 add_delay_ts(next,waitcpu_total,next->timestamp,now);
2759 inc_delay(next,runs);
2760 next->timestamp = now;
2765 prepare_arch_switch(rq, next);
2766 prev = context_switch(rq, prev, next);
2769 finish_task_switch(prev);
2771 spin_unlock_irq(&rq->lock);
2773 reacquire_kernel_lock(current);
2774 preempt_enable_no_resched();
2775 if (test_thread_flag(TIF_NEED_RESCHED))
2779 EXPORT_SYMBOL(schedule);
2781 #ifdef CONFIG_PREEMPT
2783 * this is is the entry point to schedule() from in-kernel preemption
2784 * off of preempt_enable. Kernel preemptions off return from interrupt
2785 * occur there and call schedule directly.
2787 asmlinkage void __sched preempt_schedule(void)
2789 struct thread_info *ti = current_thread_info();
2792 * If there is a non-zero preempt_count or interrupts are disabled,
2793 * we do not want to preempt the current task. Just return..
2795 if (unlikely(ti->preempt_count || irqs_disabled()))
2799 ti->preempt_count = PREEMPT_ACTIVE;
2801 ti->preempt_count = 0;
2803 /* we could miss a preemption opportunity between schedule and now */
2805 if (unlikely(test_thread_flag(TIF_NEED_RESCHED)))
2809 EXPORT_SYMBOL(preempt_schedule);
2810 #endif /* CONFIG_PREEMPT */
2812 int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, void *key)
2814 task_t *p = curr->task;
2815 return try_to_wake_up(p, mode, sync);
2818 EXPORT_SYMBOL(default_wake_function);
2821 * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just
2822 * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve
2823 * number) then we wake all the non-exclusive tasks and one exclusive task.
2825 * There are circumstances in which we can try to wake a task which has already
2826 * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns
2827 * zero in this (rare) case, and we handle it by continuing to scan the queue.
2829 static void __wake_up_common(wait_queue_head_t *q, unsigned int mode,
2830 int nr_exclusive, int sync, void *key)
2832 struct list_head *tmp, *next;
2834 list_for_each_safe(tmp, next, &q->task_list) {
2837 curr = list_entry(tmp, wait_queue_t, task_list);
2838 flags = curr->flags;
2839 if (curr->func(curr, mode, sync, key) &&
2840 (flags & WQ_FLAG_EXCLUSIVE) &&
2847 * __wake_up - wake up threads blocked on a waitqueue.
2849 * @mode: which threads
2850 * @nr_exclusive: how many wake-one or wake-many threads to wake up
2852 void fastcall __wake_up(wait_queue_head_t *q, unsigned int mode,
2853 int nr_exclusive, void *key)
2855 unsigned long flags;
2857 spin_lock_irqsave(&q->lock, flags);
2858 __wake_up_common(q, mode, nr_exclusive, 0, key);
2859 spin_unlock_irqrestore(&q->lock, flags);
2862 EXPORT_SYMBOL(__wake_up);
2865 * Same as __wake_up but called with the spinlock in wait_queue_head_t held.
2867 void fastcall __wake_up_locked(wait_queue_head_t *q, unsigned int mode)
2869 __wake_up_common(q, mode, 1, 0, NULL);
2873 * __wake_up - sync- wake up threads blocked on a waitqueue.
2875 * @mode: which threads
2876 * @nr_exclusive: how many wake-one or wake-many threads to wake up
2878 * The sync wakeup differs that the waker knows that it will schedule
2879 * away soon, so while the target thread will be woken up, it will not
2880 * be migrated to another CPU - ie. the two threads are 'synchronized'
2881 * with each other. This can prevent needless bouncing between CPUs.
2883 * On UP it can prevent extra preemption.
2885 void fastcall __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive)
2887 unsigned long flags;
2893 if (unlikely(!nr_exclusive))
2896 spin_lock_irqsave(&q->lock, flags);
2897 __wake_up_common(q, mode, nr_exclusive, sync, NULL);
2898 spin_unlock_irqrestore(&q->lock, flags);
2900 EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */
2902 void fastcall complete(struct completion *x)
2904 unsigned long flags;
2906 spin_lock_irqsave(&x->wait.lock, flags);
2908 __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2910 spin_unlock_irqrestore(&x->wait.lock, flags);
2912 EXPORT_SYMBOL(complete);
2914 void fastcall complete_all(struct completion *x)
2916 unsigned long flags;
2918 spin_lock_irqsave(&x->wait.lock, flags);
2919 x->done += UINT_MAX/2;
2920 __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2922 spin_unlock_irqrestore(&x->wait.lock, flags);
2924 EXPORT_SYMBOL(complete_all);
2926 void fastcall __sched wait_for_completion(struct completion *x)
2929 spin_lock_irq(&x->wait.lock);
2931 DECLARE_WAITQUEUE(wait, current);
2933 wait.flags |= WQ_FLAG_EXCLUSIVE;
2934 __add_wait_queue_tail(&x->wait, &wait);
2936 __set_current_state(TASK_UNINTERRUPTIBLE);
2937 spin_unlock_irq(&x->wait.lock);
2939 spin_lock_irq(&x->wait.lock);
2941 __remove_wait_queue(&x->wait, &wait);
2944 spin_unlock_irq(&x->wait.lock);
2946 EXPORT_SYMBOL(wait_for_completion);
2948 #define SLEEP_ON_VAR \
2949 unsigned long flags; \
2950 wait_queue_t wait; \
2951 init_waitqueue_entry(&wait, current);
2953 #define SLEEP_ON_HEAD \
2954 spin_lock_irqsave(&q->lock,flags); \
2955 __add_wait_queue(q, &wait); \
2956 spin_unlock(&q->lock);
2958 #define SLEEP_ON_TAIL \
2959 spin_lock_irq(&q->lock); \
2960 __remove_wait_queue(q, &wait); \
2961 spin_unlock_irqrestore(&q->lock, flags);
2963 #define SLEEP_ON_BKLCHECK \
2964 if (unlikely(!kernel_locked()) && \
2965 sleep_on_bkl_warnings < 10) { \
2966 sleep_on_bkl_warnings++; \
2970 static int sleep_on_bkl_warnings;
2972 void fastcall __sched interruptible_sleep_on(wait_queue_head_t *q)
2978 current->state = TASK_INTERRUPTIBLE;
2985 EXPORT_SYMBOL(interruptible_sleep_on);
2987 long fastcall __sched interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout)
2993 current->state = TASK_INTERRUPTIBLE;
2996 timeout = schedule_timeout(timeout);
3002 EXPORT_SYMBOL(interruptible_sleep_on_timeout);
3004 long fastcall __sched sleep_on_timeout(wait_queue_head_t *q, long timeout)
3010 current->state = TASK_UNINTERRUPTIBLE;
3013 timeout = schedule_timeout(timeout);
3019 EXPORT_SYMBOL(sleep_on_timeout);
3021 void set_user_nice(task_t *p, long nice)
3023 unsigned long flags;
3024 prio_array_t *array;
3026 int old_prio, new_prio, delta;
3028 if (TASK_NICE(p) == nice || nice < -20 || nice > 19)
3031 * We have to be careful, if called from sys_setpriority(),
3032 * the task might be in the middle of scheduling on another CPU.
3034 rq = task_rq_lock(p, &flags);
3036 * The RT priorities are set via setscheduler(), but we still
3037 * allow the 'normal' nice value to be set - but as expected
3038 * it wont have any effect on scheduling until the task is
3042 p->static_prio = NICE_TO_PRIO(nice);
3047 dequeue_task(p, array);
3050 new_prio = NICE_TO_PRIO(nice);
3051 delta = new_prio - old_prio;
3052 p->static_prio = NICE_TO_PRIO(nice);
3056 enqueue_task(p, array);
3058 * If the task increased its priority or is running and
3059 * lowered its priority, then reschedule its CPU:
3061 if (delta < 0 || (delta > 0 && task_running(rq, p)))
3062 resched_task(rq->curr);
3065 task_rq_unlock(rq, &flags);
3068 EXPORT_SYMBOL(set_user_nice);
3070 #ifdef __ARCH_WANT_SYS_NICE
3073 * sys_nice - change the priority of the current process.
3074 * @increment: priority increment
3076 * sys_setpriority is a more generic, but much slower function that
3077 * does similar things.
3079 asmlinkage long sys_nice(int increment)
3085 * Setpriority might change our priority at the same moment.
3086 * We don't have to worry. Conceptually one call occurs first
3087 * and we have a single winner.
3089 if (increment < 0) {
3090 if (!capable(CAP_SYS_NICE))
3092 if (increment < -40)
3098 nice = PRIO_TO_NICE(current->static_prio) + increment;
3104 retval = security_task_setnice(current, nice);
3108 set_user_nice(current, nice);
3115 * task_prio - return the priority value of a given task.
3116 * @p: the task in question.
3118 * This is the priority value as seen by users in /proc.
3119 * RT tasks are offset by -200. Normal tasks are centered
3120 * around 0, value goes from -16 to +15.
3122 int task_prio(const task_t *p)
3124 return p->prio - MAX_RT_PRIO;
3128 * task_nice - return the nice value of a given task.
3129 * @p: the task in question.
3131 int task_nice(const task_t *p)
3133 return TASK_NICE(p);
3136 EXPORT_SYMBOL(task_nice);
3139 * idle_cpu - is a given cpu idle currently?
3140 * @cpu: the processor in question.
3142 int idle_cpu(int cpu)
3144 return cpu_curr(cpu) == cpu_rq(cpu)->idle;
3147 EXPORT_SYMBOL_GPL(idle_cpu);
3150 * find_process_by_pid - find a process with a matching PID value.
3151 * @pid: the pid in question.
3153 static inline task_t *find_process_by_pid(pid_t pid)
3155 return pid ? find_task_by_pid(pid) : current;
3158 /* Actually do priority change: must hold rq lock. */
3159 static void __setscheduler(struct task_struct *p, int policy, int prio)
3163 p->rt_priority = prio;
3164 if (policy != SCHED_NORMAL)
3165 p->prio = MAX_USER_RT_PRIO-1 - p->rt_priority;
3167 p->prio = p->static_prio;
3171 * setscheduler - change the scheduling policy and/or RT priority of a thread.
3173 static int setscheduler(pid_t pid, int policy, struct sched_param __user *param)
3175 struct sched_param lp;
3176 int retval = -EINVAL;
3178 prio_array_t *array;
3179 unsigned long flags;
3183 if (!param || pid < 0)
3187 if (copy_from_user(&lp, param, sizeof(struct sched_param)))
3191 * We play safe to avoid deadlocks.
3193 read_lock_irq(&tasklist_lock);
3195 p = find_process_by_pid(pid);
3199 goto out_unlock_tasklist;
3202 * To be able to change p->policy safely, the apropriate
3203 * runqueue lock must be held.
3205 rq = task_rq_lock(p, &flags);
3211 if (policy != SCHED_FIFO && policy != SCHED_RR &&
3212 policy != SCHED_NORMAL)
3217 * Valid priorities for SCHED_FIFO and SCHED_RR are
3218 * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL is 0.
3221 if (lp.sched_priority < 0 || lp.sched_priority > MAX_USER_RT_PRIO-1)
3223 if ((policy == SCHED_NORMAL) != (lp.sched_priority == 0))
3227 if ((policy == SCHED_FIFO || policy == SCHED_RR) &&
3228 !capable(CAP_SYS_NICE))
3230 if ((current->euid != p->euid) && (current->euid != p->uid) &&
3231 !capable(CAP_SYS_NICE))
3234 retval = security_task_setscheduler(p, policy, &lp);
3240 deactivate_task(p, task_rq(p));
3243 __setscheduler(p, policy, lp.sched_priority);
3245 __activate_task(p, task_rq(p));
3247 * Reschedule if we are currently running on this runqueue and
3248 * our priority decreased, or if we are not currently running on
3249 * this runqueue and our priority is higher than the current's
3251 if (task_running(rq, p)) {
3252 if (p->prio > oldprio)
3253 resched_task(rq->curr);
3254 } else if (TASK_PREEMPTS_CURR(p, rq))
3255 resched_task(rq->curr);
3259 task_rq_unlock(rq, &flags);
3260 out_unlock_tasklist:
3261 read_unlock_irq(&tasklist_lock);
3268 * sys_sched_setscheduler - set/change the scheduler policy and RT priority
3269 * @pid: the pid in question.
3270 * @policy: new policy
3271 * @param: structure containing the new RT priority.
3273 asmlinkage long sys_sched_setscheduler(pid_t pid, int policy,
3274 struct sched_param __user *param)
3276 return setscheduler(pid, policy, param);
3280 * sys_sched_setparam - set/change the RT priority of a thread
3281 * @pid: the pid in question.
3282 * @param: structure containing the new RT priority.
3284 asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param)
3286 return setscheduler(pid, -1, param);
3290 * sys_sched_getscheduler - get the policy (scheduling class) of a thread
3291 * @pid: the pid in question.
3293 asmlinkage long sys_sched_getscheduler(pid_t pid)
3295 int retval = -EINVAL;
3302 read_lock(&tasklist_lock);
3303 p = find_process_by_pid(pid);
3305 retval = security_task_getscheduler(p);
3309 read_unlock(&tasklist_lock);
3316 * sys_sched_getscheduler - get the RT priority of a thread
3317 * @pid: the pid in question.
3318 * @param: structure containing the RT priority.
3320 asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param)
3322 struct sched_param lp;
3323 int retval = -EINVAL;
3326 if (!param || pid < 0)
3329 read_lock(&tasklist_lock);
3330 p = find_process_by_pid(pid);
3335 retval = security_task_getscheduler(p);
3339 lp.sched_priority = p->rt_priority;
3340 read_unlock(&tasklist_lock);
3343 * This one might sleep, we cannot do it with a spinlock held ...
3345 retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0;
3351 read_unlock(&tasklist_lock);
3356 * sys_sched_setaffinity - set the cpu affinity of a process
3357 * @pid: pid of the process
3358 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
3359 * @user_mask_ptr: user-space pointer to the new cpu mask
3361 asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len,
3362 unsigned long __user *user_mask_ptr)
3368 if (len < sizeof(new_mask))
3371 if (copy_from_user(&new_mask, user_mask_ptr, sizeof(new_mask)))
3375 read_lock(&tasklist_lock);
3377 p = find_process_by_pid(pid);
3379 read_unlock(&tasklist_lock);
3380 unlock_cpu_hotplug();
3385 * It is not safe to call set_cpus_allowed with the
3386 * tasklist_lock held. We will bump the task_struct's
3387 * usage count and then drop tasklist_lock.
3390 read_unlock(&tasklist_lock);
3393 if ((current->euid != p->euid) && (current->euid != p->uid) &&
3394 !capable(CAP_SYS_NICE))
3397 retval = set_cpus_allowed(p, new_mask);
3401 unlock_cpu_hotplug();
3406 * Represents all cpu's present in the system
3407 * In systems capable of hotplug, this map could dynamically grow
3408 * as new cpu's are detected in the system via any platform specific
3409 * method, such as ACPI for e.g.
3412 cpumask_t cpu_present_map;
3413 EXPORT_SYMBOL(cpu_present_map);
3416 cpumask_t cpu_online_map = CPU_MASK_ALL;
3417 cpumask_t cpu_possible_map = CPU_MASK_ALL;
3421 * sys_sched_getaffinity - get the cpu affinity of a process
3422 * @pid: pid of the process
3423 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
3424 * @user_mask_ptr: user-space pointer to hold the current cpu mask
3426 asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len,
3427 unsigned long __user *user_mask_ptr)
3429 unsigned int real_len;
3434 real_len = sizeof(mask);
3439 read_lock(&tasklist_lock);
3442 p = find_process_by_pid(pid);
3447 cpus_and(mask, p->cpus_allowed, cpu_possible_map);
3450 read_unlock(&tasklist_lock);
3451 unlock_cpu_hotplug();
3454 if (copy_to_user(user_mask_ptr, &mask, real_len))
3460 * sys_sched_yield - yield the current processor to other threads.
3462 * this function yields the current CPU by moving the calling thread
3463 * to the expired array. If there are no other threads running on this
3464 * CPU then this function will return.
3466 asmlinkage long sys_sched_yield(void)
3468 runqueue_t *rq = this_rq_lock();
3469 prio_array_t *array = current->array;
3470 prio_array_t *target = rq_expired(current,rq);
3473 * We implement yielding by moving the task into the expired
3476 * (special rule: RT tasks will just roundrobin in the active
3479 if (unlikely(rt_task(current)))
3480 target = rq_active(current,rq);
3482 dequeue_task(current, array);
3483 enqueue_task(current, target);
3486 * Since we are going to call schedule() anyway, there's
3487 * no need to preempt or enable interrupts:
3489 _raw_spin_unlock(&rq->lock);
3490 preempt_enable_no_resched();
3497 void __sched __cond_resched(void)
3499 #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP
3500 __might_sleep(__FILE__, __LINE__, 0);
3503 * The system_state check is somewhat ugly but we might be
3504 * called during early boot when we are not yet ready to reschedule.
3506 if (need_resched() && system_state >= SYSTEM_BOOTING_SCHEDULER_OK) {
3507 set_current_state(TASK_RUNNING);
3512 EXPORT_SYMBOL(__cond_resched);
3514 void __sched __cond_resched_lock(spinlock_t * lock)
3516 if (need_resched()) {
3517 _raw_spin_unlock(lock);
3518 preempt_enable_no_resched();
3519 set_current_state(TASK_RUNNING);
3525 EXPORT_SYMBOL(__cond_resched_lock);
3528 * yield - yield the current processor to other threads.
3530 * this is a shortcut for kernel-space yielding - it marks the
3531 * thread runnable and calls sys_sched_yield().
3533 void __sched yield(void)
3535 set_current_state(TASK_RUNNING);
3539 EXPORT_SYMBOL(yield);
3542 * This task is about to go to sleep on IO. Increment rq->nr_iowait so
3543 * that process accounting knows that this is a task in IO wait state.
3545 * But don't do that if it is a deliberate, throttling IO wait (this task
3546 * has set its backing_dev_info: the queue against which it should throttle)
3548 void __sched io_schedule(void)
3550 struct runqueue *rq = this_rq();
3551 def_delay_var(dstart);
3553 start_delay_set(dstart,PF_IOWAIT);
3554 atomic_inc(&rq->nr_iowait);
3556 atomic_dec(&rq->nr_iowait);
3557 add_io_delay(dstart);
3560 EXPORT_SYMBOL(io_schedule);
3562 long __sched io_schedule_timeout(long timeout)
3564 struct runqueue *rq = this_rq();
3566 def_delay_var(dstart);
3568 start_delay_set(dstart,PF_IOWAIT);
3569 atomic_inc(&rq->nr_iowait);
3570 ret = schedule_timeout(timeout);
3571 atomic_dec(&rq->nr_iowait);
3572 add_io_delay(dstart);
3577 * sys_sched_get_priority_max - return maximum RT priority.
3578 * @policy: scheduling class.
3580 * this syscall returns the maximum rt_priority that can be used
3581 * by a given scheduling class.
3583 asmlinkage long sys_sched_get_priority_max(int policy)
3590 ret = MAX_USER_RT_PRIO-1;
3600 * sys_sched_get_priority_min - return minimum RT priority.
3601 * @policy: scheduling class.
3603 * this syscall returns the minimum rt_priority that can be used
3604 * by a given scheduling class.
3606 asmlinkage long sys_sched_get_priority_min(int policy)
3622 * sys_sched_rr_get_interval - return the default timeslice of a process.
3623 * @pid: pid of the process.
3624 * @interval: userspace pointer to the timeslice value.
3626 * this syscall writes the default timeslice value of a given process
3627 * into the user-space timespec buffer. A value of '0' means infinity.
3630 long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval)
3632 int retval = -EINVAL;
3640 read_lock(&tasklist_lock);
3641 p = find_process_by_pid(pid);
3645 retval = security_task_getscheduler(p);
3649 jiffies_to_timespec(p->policy & SCHED_FIFO ?
3650 0 : task_timeslice(p), &t);
3651 read_unlock(&tasklist_lock);
3652 retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0;
3656 read_unlock(&tasklist_lock);
3660 static inline struct task_struct *eldest_child(struct task_struct *p)
3662 if (list_empty(&p->children)) return NULL;
3663 return list_entry(p->children.next,struct task_struct,sibling);
3666 static inline struct task_struct *older_sibling(struct task_struct *p)
3668 if (p->sibling.prev==&p->parent->children) return NULL;
3669 return list_entry(p->sibling.prev,struct task_struct,sibling);
3672 static inline struct task_struct *younger_sibling(struct task_struct *p)
3674 if (p->sibling.next==&p->parent->children) return NULL;
3675 return list_entry(p->sibling.next,struct task_struct,sibling);
3678 static void show_task(task_t * p)
3682 unsigned long free = 0;
3683 static const char *stat_nam[] = { "R", "S", "D", "T", "Z", "W" };
3685 printk("%-13.13s ", p->comm);
3686 state = p->state ? __ffs(p->state) + 1 : 0;
3687 if (state < ARRAY_SIZE(stat_nam))
3688 printk(stat_nam[state]);
3691 #if (BITS_PER_LONG == 32)
3692 if (state == TASK_RUNNING)
3693 printk(" running ");
3695 printk(" %08lX ", thread_saved_pc(p));
3697 if (state == TASK_RUNNING)
3698 printk(" running task ");
3700 printk(" %016lx ", thread_saved_pc(p));
3702 #ifdef CONFIG_DEBUG_STACK_USAGE
3704 unsigned long * n = (unsigned long *) (p->thread_info+1);
3707 free = (unsigned long) n - (unsigned long)(p->thread_info+1);
3710 printk("%5lu %5d %6d ", free, p->pid, p->parent->pid);
3711 if ((relative = eldest_child(p)))
3712 printk("%5d ", relative->pid);
3715 if ((relative = younger_sibling(p)))
3716 printk("%7d", relative->pid);
3719 if ((relative = older_sibling(p)))
3720 printk(" %5d", relative->pid);
3724 printk(" (L-TLB)\n");
3726 printk(" (NOTLB)\n");
3728 if (state != TASK_RUNNING)
3729 show_stack(p, NULL);
3732 void show_state(void)
3736 #if (BITS_PER_LONG == 32)
3739 printk(" task PC pid father child younger older\n");
3743 printk(" task PC pid father child younger older\n");
3745 read_lock(&tasklist_lock);
3746 do_each_thread(g, p) {
3748 * reset the NMI-timeout, listing all files on a slow
3749 * console might take alot of time:
3751 touch_nmi_watchdog();
3753 } while_each_thread(g, p);
3755 read_unlock(&tasklist_lock);
3758 EXPORT_SYMBOL_GPL(show_state);
3760 void __devinit init_idle(task_t *idle, int cpu)
3762 runqueue_t *idle_rq = cpu_rq(cpu), *rq = cpu_rq(task_cpu(idle));
3763 unsigned long flags;
3765 local_irq_save(flags);
3766 double_rq_lock(idle_rq, rq);
3768 idle_rq->curr = idle_rq->idle = idle;
3769 deactivate_task(idle, rq);
3771 idle->prio = MAX_PRIO;
3772 idle->state = TASK_RUNNING;
3773 set_task_cpu(idle, cpu);
3774 double_rq_unlock(idle_rq, rq);
3775 set_tsk_need_resched(idle);
3776 local_irq_restore(flags);
3778 /* Set the preempt count _outside_ the spinlocks! */
3779 #ifdef CONFIG_PREEMPT
3780 idle->thread_info->preempt_count = (idle->lock_depth >= 0);
3782 idle->thread_info->preempt_count = 0;
3787 * In a system that switches off the HZ timer nohz_cpu_mask
3788 * indicates which cpus entered this state. This is used
3789 * in the rcu update to wait only for active cpus. For system
3790 * which do not switch off the HZ timer nohz_cpu_mask should
3791 * always be CPU_MASK_NONE.
3793 cpumask_t nohz_cpu_mask = CPU_MASK_NONE;
3797 * This is how migration works:
3799 * 1) we queue a migration_req_t structure in the source CPU's
3800 * runqueue and wake up that CPU's migration thread.
3801 * 2) we down() the locked semaphore => thread blocks.
3802 * 3) migration thread wakes up (implicitly it forces the migrated
3803 * thread off the CPU)
3804 * 4) it gets the migration request and checks whether the migrated
3805 * task is still in the wrong runqueue.
3806 * 5) if it's in the wrong runqueue then the migration thread removes
3807 * it and puts it into the right queue.
3808 * 6) migration thread up()s the semaphore.
3809 * 7) we wake up and the migration is done.
3813 * Change a given task's CPU affinity. Migrate the thread to a
3814 * proper CPU and schedule it away if the CPU it's executing on
3815 * is removed from the allowed bitmask.
3817 * NOTE: the caller must have a valid reference to the task, the
3818 * task must not exit() & deallocate itself prematurely. The
3819 * call is not atomic; no spinlocks may be held.
3821 int set_cpus_allowed(task_t *p, cpumask_t new_mask)
3823 unsigned long flags;
3825 migration_req_t req;
3828 rq = task_rq_lock(p, &flags);
3829 if (!cpus_intersects(new_mask, cpu_online_map)) {
3834 p->cpus_allowed = new_mask;
3835 /* Can the task run on the task's current CPU? If so, we're done */
3836 if (cpu_isset(task_cpu(p), new_mask))
3839 if (migrate_task(p, any_online_cpu(new_mask), &req)) {
3840 /* Need help from migration thread: drop lock and wait. */
3841 task_rq_unlock(rq, &flags);
3842 wake_up_process(rq->migration_thread);
3843 wait_for_completion(&req.done);
3844 tlb_migrate_finish(p->mm);
3848 task_rq_unlock(rq, &flags);
3852 EXPORT_SYMBOL_GPL(set_cpus_allowed);
3855 * Move (not current) task off this cpu, onto dest cpu. We're doing
3856 * this because either it can't run here any more (set_cpus_allowed()
3857 * away from this CPU, or CPU going down), or because we're
3858 * attempting to rebalance this task on exec (sched_balance_exec).
3860 * So we race with normal scheduler movements, but that's OK, as long
3861 * as the task is no longer on this CPU.
3863 static void __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu)
3865 runqueue_t *rq_dest, *rq_src;
3867 if (unlikely(cpu_is_offline(dest_cpu)))
3870 rq_src = cpu_rq(src_cpu);
3871 rq_dest = cpu_rq(dest_cpu);
3873 double_rq_lock(rq_src, rq_dest);
3874 /* Already moved. */
3875 if (task_cpu(p) != src_cpu)
3877 /* Affinity changed (again). */
3878 if (!cpu_isset(dest_cpu, p->cpus_allowed))
3881 set_task_cpu(p, dest_cpu);
3884 * Sync timestamp with rq_dest's before activating.
3885 * The same thing could be achieved by doing this step
3886 * afterwards, and pretending it was a local activate.
3887 * This way is cleaner and logically correct.
3889 p->timestamp = p->timestamp - rq_src->timestamp_last_tick
3890 + rq_dest->timestamp_last_tick;
3891 deactivate_task(p, rq_src);
3892 activate_task(p, rq_dest, 0);
3893 if (TASK_PREEMPTS_CURR(p, rq_dest))
3894 resched_task(rq_dest->curr);
3898 double_rq_unlock(rq_src, rq_dest);
3902 * migration_thread - this is a highprio system thread that performs
3903 * thread migration by bumping thread off CPU then 'pushing' onto
3906 static int migration_thread(void * data)
3909 int cpu = (long)data;
3912 BUG_ON(rq->migration_thread != current);
3914 set_current_state(TASK_INTERRUPTIBLE);
3915 while (!kthread_should_stop()) {
3916 struct list_head *head;
3917 migration_req_t *req;
3919 if (current->flags & PF_FREEZE)
3920 refrigerator(PF_FREEZE);
3922 spin_lock_irq(&rq->lock);
3924 if (cpu_is_offline(cpu)) {
3925 spin_unlock_irq(&rq->lock);
3929 if (rq->active_balance) {
3930 #ifndef CONFIG_CKRM_CPU_SCHEDULE
3931 active_load_balance(rq, cpu);
3933 rq->active_balance = 0;
3936 head = &rq->migration_queue;
3938 if (list_empty(head)) {
3939 spin_unlock_irq(&rq->lock);
3941 set_current_state(TASK_INTERRUPTIBLE);
3944 req = list_entry(head->next, migration_req_t, list);
3945 list_del_init(head->next);
3947 if (req->type == REQ_MOVE_TASK) {
3948 spin_unlock(&rq->lock);
3949 __migrate_task(req->task, smp_processor_id(),
3952 } else if (req->type == REQ_SET_DOMAIN) {
3954 spin_unlock_irq(&rq->lock);
3956 spin_unlock_irq(&rq->lock);
3960 complete(&req->done);
3962 __set_current_state(TASK_RUNNING);
3966 /* Wait for kthread_stop */
3967 set_current_state(TASK_INTERRUPTIBLE);
3968 while (!kthread_should_stop()) {
3970 set_current_state(TASK_INTERRUPTIBLE);
3972 __set_current_state(TASK_RUNNING);
3976 #ifdef CONFIG_HOTPLUG_CPU
3977 /* migrate_all_tasks - function to migrate all tasks from the dead cpu. */
3978 static void migrate_all_tasks(int src_cpu)
3980 struct task_struct *tsk, *t;
3984 write_lock_irq(&tasklist_lock);
3986 /* watch out for per node tasks, let's stay on this node */
3987 node = cpu_to_node(src_cpu);
3989 do_each_thread(t, tsk) {
3994 if (task_cpu(tsk) != src_cpu)
3997 /* Figure out where this task should go (attempting to
3998 * keep it on-node), and check if it can be migrated
3999 * as-is. NOTE that kernel threads bound to more than
4000 * one online cpu will be migrated. */
4001 mask = node_to_cpumask(node);
4002 cpus_and(mask, mask, tsk->cpus_allowed);
4003 dest_cpu = any_online_cpu(mask);
4004 if (dest_cpu == NR_CPUS)
4005 dest_cpu = any_online_cpu(tsk->cpus_allowed);
4006 if (dest_cpu == NR_CPUS) {
4007 cpus_setall(tsk->cpus_allowed);
4008 dest_cpu = any_online_cpu(tsk->cpus_allowed);
4010 /* Don't tell them about moving exiting tasks
4011 or kernel threads (both mm NULL), since
4012 they never leave kernel. */
4013 if (tsk->mm && printk_ratelimit())
4014 printk(KERN_INFO "process %d (%s) no "
4015 "longer affine to cpu%d\n",
4016 tsk->pid, tsk->comm, src_cpu);
4019 __migrate_task(tsk, src_cpu, dest_cpu);
4020 } while_each_thread(t, tsk);
4022 write_unlock_irq(&tasklist_lock);
4025 /* Schedules idle task to be the next runnable task on current CPU.
4026 * It does so by boosting its priority to highest possible and adding it to
4027 * the _front_ of runqueue. Used by CPU offline code.
4029 void sched_idle_next(void)
4031 int cpu = smp_processor_id();
4032 runqueue_t *rq = this_rq();
4033 struct task_struct *p = rq->idle;
4034 unsigned long flags;
4036 /* cpu has to be offline */
4037 BUG_ON(cpu_online(cpu));
4039 /* Strictly not necessary since rest of the CPUs are stopped by now
4040 * and interrupts disabled on current cpu.
4042 spin_lock_irqsave(&rq->lock, flags);
4044 __setscheduler(p, SCHED_FIFO, MAX_RT_PRIO-1);
4045 /* Add idle task to _front_ of it's priority queue */
4046 __activate_idle_task(p, rq);
4048 spin_unlock_irqrestore(&rq->lock, flags);
4050 #endif /* CONFIG_HOTPLUG_CPU */
4053 * migration_call - callback that gets triggered when a CPU is added.
4054 * Here we can start up the necessary migration thread for the new CPU.
4056 static int migration_call(struct notifier_block *nfb, unsigned long action,
4059 int cpu = (long)hcpu;
4060 struct task_struct *p;
4061 struct runqueue *rq;
4062 unsigned long flags;
4065 case CPU_UP_PREPARE:
4066 p = kthread_create(migration_thread, hcpu, "migration/%d",cpu);
4069 p->flags |= PF_NOFREEZE;
4070 kthread_bind(p, cpu);
4071 /* Must be high prio: stop_machine expects to yield to it. */
4072 rq = task_rq_lock(p, &flags);
4073 __setscheduler(p, SCHED_FIFO, MAX_RT_PRIO-1);
4074 task_rq_unlock(rq, &flags);
4075 cpu_rq(cpu)->migration_thread = p;
4078 /* Strictly unneccessary, as first user will wake it. */
4079 wake_up_process(cpu_rq(cpu)->migration_thread);
4081 #ifdef CONFIG_HOTPLUG_CPU
4082 case CPU_UP_CANCELED:
4083 /* Unbind it from offline cpu so it can run. Fall thru. */
4084 kthread_bind(cpu_rq(cpu)->migration_thread,smp_processor_id());
4085 kthread_stop(cpu_rq(cpu)->migration_thread);
4086 cpu_rq(cpu)->migration_thread = NULL;
4089 migrate_all_tasks(cpu);
4091 kthread_stop(rq->migration_thread);
4092 rq->migration_thread = NULL;
4093 /* Idle task back to normal (off runqueue, low prio) */
4094 rq = task_rq_lock(rq->idle, &flags);
4095 deactivate_task(rq->idle, rq);
4096 rq->idle->static_prio = MAX_PRIO;
4097 __setscheduler(rq->idle, SCHED_NORMAL, 0);
4098 task_rq_unlock(rq, &flags);
4099 BUG_ON(rq->nr_running != 0);
4101 /* No need to migrate the tasks: it was best-effort if
4102 * they didn't do lock_cpu_hotplug(). Just wake up
4103 * the requestors. */
4104 spin_lock_irq(&rq->lock);
4105 while (!list_empty(&rq->migration_queue)) {
4106 migration_req_t *req;
4107 req = list_entry(rq->migration_queue.next,
4108 migration_req_t, list);
4109 BUG_ON(req->type != REQ_MOVE_TASK);
4110 list_del_init(&req->list);
4111 complete(&req->done);
4113 spin_unlock_irq(&rq->lock);
4120 /* Register at highest priority so that task migration (migrate_all_tasks)
4121 * happens before everything else.
4123 static struct notifier_block __devinitdata migration_notifier = {
4124 .notifier_call = migration_call,
4128 int __init migration_init(void)
4130 void *cpu = (void *)(long)smp_processor_id();
4131 /* Start one for boot CPU. */
4132 migration_call(&migration_notifier, CPU_UP_PREPARE, cpu);
4133 migration_call(&migration_notifier, CPU_ONLINE, cpu);
4134 register_cpu_notifier(&migration_notifier);
4140 * The 'big kernel lock'
4142 * This spinlock is taken and released recursively by lock_kernel()
4143 * and unlock_kernel(). It is transparently dropped and reaquired
4144 * over schedule(). It is used to protect legacy code that hasn't
4145 * been migrated to a proper locking design yet.
4147 * Don't use in new code.
4149 * Note: spinlock debugging needs this even on !CONFIG_SMP.
4151 spinlock_t kernel_flag __cacheline_aligned_in_smp = SPIN_LOCK_UNLOCKED;
4152 EXPORT_SYMBOL(kernel_flag);
4155 /* Attach the domain 'sd' to 'cpu' as its base domain */
4156 void cpu_attach_domain(struct sched_domain *sd, int cpu)
4158 migration_req_t req;
4159 unsigned long flags;
4160 runqueue_t *rq = cpu_rq(cpu);
4165 spin_lock_irqsave(&rq->lock, flags);
4167 if (cpu == smp_processor_id() || !cpu_online(cpu)) {
4170 init_completion(&req.done);
4171 req.type = REQ_SET_DOMAIN;
4173 list_add(&req.list, &rq->migration_queue);
4177 spin_unlock_irqrestore(&rq->lock, flags);
4180 wake_up_process(rq->migration_thread);
4181 wait_for_completion(&req.done);
4184 unlock_cpu_hotplug();
4187 #ifdef ARCH_HAS_SCHED_DOMAIN
4188 extern void __init arch_init_sched_domains(void);
4190 static struct sched_group sched_group_cpus[NR_CPUS];
4191 static DEFINE_PER_CPU(struct sched_domain, cpu_domains);
4193 static struct sched_group sched_group_nodes[MAX_NUMNODES];
4194 static DEFINE_PER_CPU(struct sched_domain, node_domains);
4195 static void __init arch_init_sched_domains(void)
4198 struct sched_group *first_node = NULL, *last_node = NULL;
4200 /* Set up domains */
4202 int node = cpu_to_node(i);
4203 cpumask_t nodemask = node_to_cpumask(node);
4204 struct sched_domain *node_sd = &per_cpu(node_domains, i);
4205 struct sched_domain *cpu_sd = &per_cpu(cpu_domains, i);
4207 *node_sd = SD_NODE_INIT;
4208 node_sd->span = cpu_possible_map;
4209 node_sd->groups = &sched_group_nodes[cpu_to_node(i)];
4211 *cpu_sd = SD_CPU_INIT;
4212 cpus_and(cpu_sd->span, nodemask, cpu_possible_map);
4213 cpu_sd->groups = &sched_group_cpus[i];
4214 cpu_sd->parent = node_sd;
4218 for (i = 0; i < MAX_NUMNODES; i++) {
4219 cpumask_t tmp = node_to_cpumask(i);
4221 struct sched_group *first_cpu = NULL, *last_cpu = NULL;
4222 struct sched_group *node = &sched_group_nodes[i];
4225 cpus_and(nodemask, tmp, cpu_possible_map);
4227 if (cpus_empty(nodemask))
4230 node->cpumask = nodemask;
4231 node->cpu_power = SCHED_LOAD_SCALE * cpus_weight(node->cpumask);
4233 for_each_cpu_mask(j, node->cpumask) {
4234 struct sched_group *cpu = &sched_group_cpus[j];
4236 cpus_clear(cpu->cpumask);
4237 cpu_set(j, cpu->cpumask);
4238 cpu->cpu_power = SCHED_LOAD_SCALE;
4243 last_cpu->next = cpu;
4246 last_cpu->next = first_cpu;
4251 last_node->next = node;
4254 last_node->next = first_node;
4258 struct sched_domain *cpu_sd = &per_cpu(cpu_domains, i);
4259 cpu_attach_domain(cpu_sd, i);
4263 #else /* !CONFIG_NUMA */
4264 static void __init arch_init_sched_domains(void)
4267 struct sched_group *first_cpu = NULL, *last_cpu = NULL;
4269 /* Set up domains */
4271 struct sched_domain *cpu_sd = &per_cpu(cpu_domains, i);
4273 *cpu_sd = SD_CPU_INIT;
4274 cpu_sd->span = cpu_possible_map;
4275 cpu_sd->groups = &sched_group_cpus[i];
4278 /* Set up CPU groups */
4279 for_each_cpu_mask(i, cpu_possible_map) {
4280 struct sched_group *cpu = &sched_group_cpus[i];
4282 cpus_clear(cpu->cpumask);
4283 cpu_set(i, cpu->cpumask);
4284 cpu->cpu_power = SCHED_LOAD_SCALE;
4289 last_cpu->next = cpu;
4292 last_cpu->next = first_cpu;
4294 mb(); /* domains were modified outside the lock */
4296 struct sched_domain *cpu_sd = &per_cpu(cpu_domains, i);
4297 cpu_attach_domain(cpu_sd, i);
4301 #endif /* CONFIG_NUMA */
4302 #endif /* ARCH_HAS_SCHED_DOMAIN */
4304 #define SCHED_DOMAIN_DEBUG
4305 #ifdef SCHED_DOMAIN_DEBUG
4306 void sched_domain_debug(void)
4311 runqueue_t *rq = cpu_rq(i);
4312 struct sched_domain *sd;
4317 printk(KERN_DEBUG "CPU%d: %s\n",
4318 i, (cpu_online(i) ? " online" : "offline"));
4323 struct sched_group *group = sd->groups;
4324 cpumask_t groupmask;
4326 cpumask_scnprintf(str, NR_CPUS, sd->span);
4327 cpus_clear(groupmask);
4330 for (j = 0; j < level + 1; j++)
4332 printk("domain %d: span %s\n", level, str);
4334 if (!cpu_isset(i, sd->span))
4335 printk(KERN_DEBUG "ERROR domain->span does not contain CPU%d\n", i);
4336 if (!cpu_isset(i, group->cpumask))
4337 printk(KERN_DEBUG "ERROR domain->groups does not contain CPU%d\n", i);
4338 if (!group->cpu_power)
4339 printk(KERN_DEBUG "ERROR domain->cpu_power not set\n");
4342 for (j = 0; j < level + 2; j++)
4347 printk(" ERROR: NULL");
4351 if (!cpus_weight(group->cpumask))
4352 printk(" ERROR empty group:");
4354 if (cpus_intersects(groupmask, group->cpumask))
4355 printk(" ERROR repeated CPUs:");
4357 cpus_or(groupmask, groupmask, group->cpumask);
4359 cpumask_scnprintf(str, NR_CPUS, group->cpumask);
4362 group = group->next;
4363 } while (group != sd->groups);
4366 if (!cpus_equal(sd->span, groupmask))
4367 printk(KERN_DEBUG "ERROR groups don't span domain->span\n");
4373 if (!cpus_subset(groupmask, sd->span))
4374 printk(KERN_DEBUG "ERROR parent span is not a superset of domain->span\n");
4381 #define sched_domain_debug() {}
4384 void __init sched_init_smp(void)
4386 arch_init_sched_domains();
4387 sched_domain_debug();
4390 void __init sched_init_smp(void)
4393 #endif /* CONFIG_SMP */
4395 int in_sched_functions(unsigned long addr)
4397 /* Linker adds these: start and end of __sched functions */
4398 extern char __sched_text_start[], __sched_text_end[];
4399 return addr >= (unsigned long)__sched_text_start
4400 && addr < (unsigned long)__sched_text_end;
4403 void __init sched_init(void)
4407 #ifndef CONFIG_CKRM_CPU_SCHEDULE
4412 /* Set up an initial dummy domain for early boot */
4413 static struct sched_domain sched_domain_init;
4414 static struct sched_group sched_group_init;
4416 memset(&sched_domain_init, 0, sizeof(struct sched_domain));
4417 sched_domain_init.span = CPU_MASK_ALL;
4418 sched_domain_init.groups = &sched_group_init;
4419 sched_domain_init.last_balance = jiffies;
4420 sched_domain_init.balance_interval = INT_MAX; /* Don't balance */
4422 memset(&sched_group_init, 0, sizeof(struct sched_group));
4423 sched_group_init.cpumask = CPU_MASK_ALL;
4424 sched_group_init.next = &sched_group_init;
4425 sched_group_init.cpu_power = SCHED_LOAD_SCALE;
4430 for (i = 0; i < NR_CPUS; i++) {
4431 #ifndef CONFIG_CKRM_CPU_SCHEDULE
4432 prio_array_t *array;
4435 spin_lock_init(&rq->lock);
4437 #ifndef CONFIG_CKRM_CPU_SCHEDULE
4438 rq->active = rq->arrays;
4439 rq->expired = rq->arrays + 1;
4441 rq->ckrm_cpu_load = 0;
4443 rq->best_expired_prio = MAX_PRIO;
4446 rq->sd = &sched_domain_init;
4448 rq->active_balance = 0;
4450 rq->migration_thread = NULL;
4451 INIT_LIST_HEAD(&rq->migration_queue);
4453 INIT_LIST_HEAD(&rq->hold_queue);
4454 atomic_set(&rq->nr_iowait, 0);
4456 #ifndef CONFIG_CKRM_CPU_SCHEDULE
4457 for (j = 0; j < 2; j++) {
4458 array = rq->arrays + j;
4459 for (k = 0; k < MAX_PRIO; k++) {
4460 INIT_LIST_HEAD(array->queue + k);
4461 __clear_bit(k, array->bitmap);
4463 // delimiter for bitsearch
4464 __set_bit(MAX_PRIO, array->bitmap);
4470 * We have to do a little magic to get the first
4471 * thread right in SMP mode.
4476 set_task_cpu(current, smp_processor_id());
4477 #ifdef CONFIG_CKRM_CPU_SCHEDULE
4478 current->cpu_class = default_cpu_class;
4479 current->array = NULL;
4481 wake_up_forked_process(current);
4484 * The boot idle thread does lazy MMU switching as well:
4486 atomic_inc(&init_mm.mm_count);
4487 enter_lazy_tlb(&init_mm, current);
4490 #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP
4491 void __might_sleep(char *file, int line, int atomic_depth)
4493 #if defined(in_atomic)
4494 static unsigned long prev_jiffy; /* ratelimiting */
4496 #ifndef CONFIG_PREEMPT
4499 if (((in_atomic() != atomic_depth) || irqs_disabled()) &&
4500 system_state == SYSTEM_RUNNING) {
4501 if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy)
4503 prev_jiffy = jiffies;
4504 printk(KERN_ERR "Debug: sleeping function called from invalid"
4505 " context at %s:%d\n", file, line);
4506 printk("in_atomic():%d[expected: %d], irqs_disabled():%d\n",
4507 in_atomic(), atomic_depth, irqs_disabled());
4512 EXPORT_SYMBOL(__might_sleep);
4516 #if defined(CONFIG_SMP) && defined(CONFIG_PREEMPT)
4518 * This could be a long-held lock. If another CPU holds it for a long time,
4519 * and that CPU is not asked to reschedule then *this* CPU will spin on the
4520 * lock for a long time, even if *this* CPU is asked to reschedule.
4522 * So what we do here, in the slow (contended) path is to spin on the lock by
4523 * hand while permitting preemption.
4525 * Called inside preempt_disable().
4527 void __sched __preempt_spin_lock(spinlock_t *lock)
4529 if (preempt_count() > 1) {
4530 _raw_spin_lock(lock);
4535 while (spin_is_locked(lock))
4538 } while (!_raw_spin_trylock(lock));
4541 EXPORT_SYMBOL(__preempt_spin_lock);
4543 void __sched __preempt_write_lock(rwlock_t *lock)
4545 if (preempt_count() > 1) {
4546 _raw_write_lock(lock);
4552 while (rwlock_is_locked(lock))
4555 } while (!_raw_write_trylock(lock));
4558 EXPORT_SYMBOL(__preempt_write_lock);
4559 #endif /* defined(CONFIG_SMP) && defined(CONFIG_PREEMPT) */
4561 #ifdef CONFIG_DELAY_ACCT
4562 int task_running_sys(struct task_struct *p)
4564 return task_running(task_rq(p),p);
4566 EXPORT_SYMBOL(task_running_sys);
4569 #ifdef CONFIG_CKRM_CPU_SCHEDULE
4571 * return the classqueue object of a certain processor
4572 * Note: not supposed to be used in performance sensitive functions
4574 struct classqueue_struct * get_cpu_classqueue(int cpu)
4576 return (& (cpu_rq(cpu)->classqueue) );