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
22 #include <linux/module.h>
23 #include <linux/nmi.h>
24 #include <linux/init.h>
25 #include <asm/uaccess.h>
26 #include <linux/highmem.h>
27 #include <linux/smp_lock.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/profile.h>
35 #include <linux/suspend.h>
36 #include <linux/blkdev.h>
37 #include <linux/delay.h>
38 #include <linux/smp.h>
39 #include <linux/threads.h>
40 #include <linux/timer.h>
41 #include <linux/rcupdate.h>
42 #include <linux/cpu.h>
43 #include <linux/cpuset.h>
44 #include <linux/percpu.h>
45 #include <linux/kthread.h>
46 #include <linux/seq_file.h>
47 #include <linux/syscalls.h>
48 #include <linux/times.h>
49 #include <linux/acct.h>
52 #include <asm/unistd.h>
55 * Convert user-nice values [ -20 ... 0 ... 19 ]
56 * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ],
59 #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20)
60 #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20)
61 #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio)
64 * 'User priority' is the nice value converted to something we
65 * can work with better when scaling various scheduler parameters,
66 * it's a [ 0 ... 39 ] range.
68 #define USER_PRIO(p) ((p)-MAX_RT_PRIO)
69 #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio)
70 #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO))
73 * Some helpers for converting nanosecond timing to jiffy resolution
75 #define NS_TO_JIFFIES(TIME) ((TIME) / (1000000000 / HZ))
76 #define JIFFIES_TO_NS(TIME) ((TIME) * (1000000000 / HZ))
79 * These are the 'tuning knobs' of the scheduler:
81 * Minimum timeslice is 5 msecs (or 1 jiffy, whichever is larger),
82 * default timeslice is 100 msecs, maximum timeslice is 800 msecs.
83 * Timeslices get refilled after they expire.
85 #define MIN_TIMESLICE max(5 * HZ / 1000, 1)
86 #define DEF_TIMESLICE (100 * HZ / 1000)
87 #define ON_RUNQUEUE_WEIGHT 30
88 #define CHILD_PENALTY 95
89 #define PARENT_PENALTY 100
91 #define PRIO_BONUS_RATIO 25
92 #define MAX_BONUS (MAX_USER_PRIO * PRIO_BONUS_RATIO / 100)
93 #define INTERACTIVE_DELTA 2
94 #define MAX_SLEEP_AVG (DEF_TIMESLICE * MAX_BONUS)
95 #define STARVATION_LIMIT (MAX_SLEEP_AVG)
96 #define NS_MAX_SLEEP_AVG (JIFFIES_TO_NS(MAX_SLEEP_AVG))
99 * If a task is 'interactive' then we reinsert it in the active
100 * array after it has expired its current timeslice. (it will not
101 * continue to run immediately, it will still roundrobin with
102 * other interactive tasks.)
104 * This part scales the interactivity limit depending on niceness.
106 * We scale it linearly, offset by the INTERACTIVE_DELTA delta.
107 * Here are a few examples of different nice levels:
109 * TASK_INTERACTIVE(-20): [1,1,1,1,1,1,1,1,1,0,0]
110 * TASK_INTERACTIVE(-10): [1,1,1,1,1,1,1,0,0,0,0]
111 * TASK_INTERACTIVE( 0): [1,1,1,1,0,0,0,0,0,0,0]
112 * TASK_INTERACTIVE( 10): [1,1,0,0,0,0,0,0,0,0,0]
113 * TASK_INTERACTIVE( 19): [0,0,0,0,0,0,0,0,0,0,0]
115 * (the X axis represents the possible -5 ... 0 ... +5 dynamic
116 * priority range a task can explore, a value of '1' means the
117 * task is rated interactive.)
119 * Ie. nice +19 tasks can never get 'interactive' enough to be
120 * reinserted into the active array. And only heavily CPU-hog nice -20
121 * tasks will be expired. Default nice 0 tasks are somewhere between,
122 * it takes some effort for them to get interactive, but it's not
126 #define CURRENT_BONUS(p) \
127 (NS_TO_JIFFIES((p)->sleep_avg) * MAX_BONUS / \
130 #define GRANULARITY (10 * HZ / 1000 ? : 1)
133 #define TIMESLICE_GRANULARITY(p) (GRANULARITY * \
134 (1 << (((MAX_BONUS - CURRENT_BONUS(p)) ? : 1) - 1)) * \
137 #define TIMESLICE_GRANULARITY(p) (GRANULARITY * \
138 (1 << (((MAX_BONUS - CURRENT_BONUS(p)) ? : 1) - 1)))
141 #define SCALE(v1,v1_max,v2_max) \
142 (v1) * (v2_max) / (v1_max)
145 (SCALE(TASK_NICE(p), 40, MAX_BONUS) + INTERACTIVE_DELTA)
147 #define TASK_INTERACTIVE(p) \
148 ((p)->prio <= (p)->static_prio - DELTA(p))
150 #define INTERACTIVE_SLEEP(p) \
151 (JIFFIES_TO_NS(MAX_SLEEP_AVG * \
152 (MAX_BONUS / 2 + DELTA((p)) + 1) / MAX_BONUS - 1))
154 #define TASK_PREEMPTS_CURR(p, rq) \
155 ((p)->prio < (rq)->curr->prio)
158 * task_timeslice() scales user-nice values [ -20 ... 0 ... 19 ]
159 * to time slice values: [800ms ... 100ms ... 5ms]
161 * The higher a thread's priority, the bigger timeslices
162 * it gets during one round of execution. But even the lowest
163 * priority thread gets MIN_TIMESLICE worth of execution time.
166 #define SCALE_PRIO(x, prio) \
167 max(x * (MAX_PRIO - prio) / (MAX_USER_PRIO/2), MIN_TIMESLICE)
169 static inline unsigned int task_timeslice(task_t
*p
)
171 if (p
->static_prio
< NICE_TO_PRIO(0))
172 return SCALE_PRIO(DEF_TIMESLICE
*4, p
->static_prio
);
174 return SCALE_PRIO(DEF_TIMESLICE
, p
->static_prio
);
176 #define task_hot(p, now, sd) ((long long) ((now) - (p)->last_ran) \
177 < (long long) (sd)->cache_hot_time)
180 * These are the runqueue data structures:
183 #define BITMAP_SIZE ((((MAX_PRIO+1+7)/8)+sizeof(long)-1)/sizeof(long))
185 typedef struct runqueue runqueue_t
;
188 unsigned int nr_active
;
189 unsigned long bitmap
[BITMAP_SIZE
];
190 struct list_head queue
[MAX_PRIO
];
194 * This is the main, per-CPU runqueue data structure.
196 * Locking rule: those places that want to lock multiple runqueues
197 * (such as the load balancing or the thread migration code), lock
198 * acquire operations must be ordered by ascending &runqueue.
204 * nr_running and cpu_load should be in the same cacheline because
205 * remote CPUs use both these fields when doing load calculation.
207 unsigned long nr_running
;
209 unsigned long cpu_load
[3];
211 unsigned long long nr_switches
;
214 * This is part of a global counter where only the total sum
215 * over all CPUs matters. A task can increase this counter on
216 * one CPU and if it got migrated afterwards it may decrease
217 * it on another CPU. Always updated under the runqueue lock:
219 unsigned long nr_uninterruptible
;
221 unsigned long expired_timestamp
;
222 unsigned long long timestamp_last_tick
;
224 struct mm_struct
*prev_mm
;
225 prio_array_t
*active
, *expired
, arrays
[2];
226 int best_expired_prio
;
230 struct sched_domain
*sd
;
232 /* For active balancing */
236 task_t
*migration_thread
;
237 struct list_head migration_queue
;
240 #ifdef CONFIG_SCHEDSTATS
242 struct sched_info rq_sched_info
;
244 /* sys_sched_yield() stats */
245 unsigned long yld_exp_empty
;
246 unsigned long yld_act_empty
;
247 unsigned long yld_both_empty
;
248 unsigned long yld_cnt
;
250 /* schedule() stats */
251 unsigned long sched_switch
;
252 unsigned long sched_cnt
;
253 unsigned long sched_goidle
;
255 /* try_to_wake_up() stats */
256 unsigned long ttwu_cnt
;
257 unsigned long ttwu_local
;
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))
281 * task_rq_lock - lock the runqueue a given task resides on and disable
282 * interrupts. Note the ordering: we can safely lookup the task_rq without
283 * explicitly disabling preemption.
285 static inline runqueue_t
*task_rq_lock(task_t
*p
, unsigned long *flags
)
291 local_irq_save(*flags
);
293 spin_lock(&rq
->lock
);
294 if (unlikely(rq
!= task_rq(p
))) {
295 spin_unlock_irqrestore(&rq
->lock
, *flags
);
296 goto repeat_lock_task
;
301 static inline void task_rq_unlock(runqueue_t
*rq
, unsigned long *flags
)
304 spin_unlock_irqrestore(&rq
->lock
, *flags
);
307 #ifdef CONFIG_SCHEDSTATS
309 * bump this up when changing the output format or the meaning of an existing
310 * format, so that tools can adapt (or abort)
312 #define SCHEDSTAT_VERSION 11
314 static int show_schedstat(struct seq_file
*seq
, void *v
)
318 seq_printf(seq
, "version %d\n", SCHEDSTAT_VERSION
);
319 seq_printf(seq
, "timestamp %lu\n", jiffies
);
320 for_each_online_cpu(cpu
) {
321 runqueue_t
*rq
= cpu_rq(cpu
);
323 struct sched_domain
*sd
;
327 /* runqueue-specific stats */
329 "cpu%d %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu",
330 cpu
, rq
->yld_both_empty
,
331 rq
->yld_act_empty
, rq
->yld_exp_empty
, rq
->yld_cnt
,
332 rq
->sched_switch
, rq
->sched_cnt
, rq
->sched_goidle
,
333 rq
->ttwu_cnt
, rq
->ttwu_local
,
334 rq
->rq_sched_info
.cpu_time
,
335 rq
->rq_sched_info
.run_delay
, rq
->rq_sched_info
.pcnt
);
337 seq_printf(seq
, "\n");
340 /* domain-specific stats */
341 for_each_domain(cpu
, sd
) {
342 enum idle_type itype
;
343 char mask_str
[NR_CPUS
];
345 cpumask_scnprintf(mask_str
, NR_CPUS
, sd
->span
);
346 seq_printf(seq
, "domain%d %s", dcnt
++, mask_str
);
347 for (itype
= SCHED_IDLE
; itype
< MAX_IDLE_TYPES
;
349 seq_printf(seq
, " %lu %lu %lu %lu %lu %lu %lu %lu",
351 sd
->lb_balanced
[itype
],
352 sd
->lb_failed
[itype
],
353 sd
->lb_imbalance
[itype
],
354 sd
->lb_gained
[itype
],
355 sd
->lb_hot_gained
[itype
],
356 sd
->lb_nobusyq
[itype
],
357 sd
->lb_nobusyg
[itype
]);
359 seq_printf(seq
, " %lu %lu %lu %lu %lu %lu %lu %lu\n",
360 sd
->alb_cnt
, sd
->alb_failed
, sd
->alb_pushed
,
361 sd
->sbe_pushed
, sd
->sbe_attempts
,
362 sd
->ttwu_wake_remote
, sd
->ttwu_move_affine
, sd
->ttwu_move_balance
);
369 static int schedstat_open(struct inode
*inode
, struct file
*file
)
371 unsigned int size
= PAGE_SIZE
* (1 + num_online_cpus() / 32);
372 char *buf
= kmalloc(size
, GFP_KERNEL
);
378 res
= single_open(file
, show_schedstat
, NULL
);
380 m
= file
->private_data
;
388 struct file_operations proc_schedstat_operations
= {
389 .open
= schedstat_open
,
392 .release
= single_release
,
395 # define schedstat_inc(rq, field) do { (rq)->field++; } while (0)
396 # define schedstat_add(rq, field, amt) do { (rq)->field += (amt); } while (0)
397 #else /* !CONFIG_SCHEDSTATS */
398 # define schedstat_inc(rq, field) do { } while (0)
399 # define schedstat_add(rq, field, amt) do { } while (0)
403 * rq_lock - lock a given runqueue and disable interrupts.
405 static inline runqueue_t
*this_rq_lock(void)
412 spin_lock(&rq
->lock
);
417 #ifdef CONFIG_SCHED_SMT
418 static int cpu_and_siblings_are_idle(int cpu
)
421 for_each_cpu_mask(sib
, cpu_sibling_map
[cpu
]) {
430 #define cpu_and_siblings_are_idle(A) idle_cpu(A)
433 #ifdef CONFIG_SCHEDSTATS
435 * Called when a process is dequeued from the active array and given
436 * the cpu. We should note that with the exception of interactive
437 * tasks, the expired queue will become the active queue after the active
438 * queue is empty, without explicitly dequeuing and requeuing tasks in the
439 * expired queue. (Interactive tasks may be requeued directly to the
440 * active queue, thus delaying tasks in the expired queue from running;
441 * see scheduler_tick()).
443 * This function is only called from sched_info_arrive(), rather than
444 * dequeue_task(). Even though a task may be queued and dequeued multiple
445 * times as it is shuffled about, we're really interested in knowing how
446 * long it was from the *first* time it was queued to the time that it
449 static inline void sched_info_dequeued(task_t
*t
)
451 t
->sched_info
.last_queued
= 0;
455 * Called when a task finally hits the cpu. We can now calculate how
456 * long it was waiting to run. We also note when it began so that we
457 * can keep stats on how long its timeslice is.
459 static inline void sched_info_arrive(task_t
*t
)
461 unsigned long now
= jiffies
, diff
= 0;
462 struct runqueue
*rq
= task_rq(t
);
464 if (t
->sched_info
.last_queued
)
465 diff
= now
- t
->sched_info
.last_queued
;
466 sched_info_dequeued(t
);
467 t
->sched_info
.run_delay
+= diff
;
468 t
->sched_info
.last_arrival
= now
;
469 t
->sched_info
.pcnt
++;
474 rq
->rq_sched_info
.run_delay
+= diff
;
475 rq
->rq_sched_info
.pcnt
++;
479 * Called when a process is queued into either the active or expired
480 * array. The time is noted and later used to determine how long we
481 * had to wait for us to reach the cpu. Since the expired queue will
482 * become the active queue after active queue is empty, without dequeuing
483 * and requeuing any tasks, we are interested in queuing to either. It
484 * is unusual but not impossible for tasks to be dequeued and immediately
485 * requeued in the same or another array: this can happen in sched_yield(),
486 * set_user_nice(), and even load_balance() as it moves tasks from runqueue
489 * This function is only called from enqueue_task(), but also only updates
490 * the timestamp if it is already not set. It's assumed that
491 * sched_info_dequeued() will clear that stamp when appropriate.
493 static inline void sched_info_queued(task_t
*t
)
495 if (!t
->sched_info
.last_queued
)
496 t
->sched_info
.last_queued
= jiffies
;
500 * Called when a process ceases being the active-running process, either
501 * voluntarily or involuntarily. Now we can calculate how long we ran.
503 static inline void sched_info_depart(task_t
*t
)
505 struct runqueue
*rq
= task_rq(t
);
506 unsigned long diff
= jiffies
- t
->sched_info
.last_arrival
;
508 t
->sched_info
.cpu_time
+= diff
;
511 rq
->rq_sched_info
.cpu_time
+= diff
;
515 * Called when tasks are switched involuntarily due, typically, to expiring
516 * their time slice. (This may also be called when switching to or from
517 * the idle task.) We are only called when prev != next.
519 static inline void sched_info_switch(task_t
*prev
, task_t
*next
)
521 struct runqueue
*rq
= task_rq(prev
);
524 * prev now departs the cpu. It's not interesting to record
525 * stats about how efficient we were at scheduling the idle
528 if (prev
!= rq
->idle
)
529 sched_info_depart(prev
);
531 if (next
!= rq
->idle
)
532 sched_info_arrive(next
);
535 #define sched_info_queued(t) do { } while (0)
536 #define sched_info_switch(t, next) do { } while (0)
537 #endif /* CONFIG_SCHEDSTATS */
540 * Adding/removing a task to/from a priority array:
542 static void dequeue_task(struct task_struct
*p
, prio_array_t
*array
)
545 list_del(&p
->run_list
);
546 if (list_empty(array
->queue
+ p
->prio
))
547 __clear_bit(p
->prio
, array
->bitmap
);
550 static void enqueue_task(struct task_struct
*p
, prio_array_t
*array
)
552 sched_info_queued(p
);
553 list_add_tail(&p
->run_list
, array
->queue
+ p
->prio
);
554 __set_bit(p
->prio
, array
->bitmap
);
560 * Put task to the end of the run list without the overhead of dequeue
561 * followed by enqueue.
563 static void requeue_task(struct task_struct
*p
, prio_array_t
*array
)
565 list_move_tail(&p
->run_list
, array
->queue
+ p
->prio
);
568 static inline void enqueue_task_head(struct task_struct
*p
, prio_array_t
*array
)
570 list_add(&p
->run_list
, array
->queue
+ p
->prio
);
571 __set_bit(p
->prio
, array
->bitmap
);
577 * effective_prio - return the priority that is based on the static
578 * priority but is modified by bonuses/penalties.
580 * We scale the actual sleep average [0 .... MAX_SLEEP_AVG]
581 * into the -5 ... 0 ... +5 bonus/penalty range.
583 * We use 25% of the full 0...39 priority range so that:
585 * 1) nice +19 interactive tasks do not preempt nice 0 CPU hogs.
586 * 2) nice -20 CPU hogs do not get preempted by nice 0 tasks.
588 * Both properties are important to certain workloads.
590 static int effective_prio(task_t
*p
)
597 bonus
= CURRENT_BONUS(p
) - MAX_BONUS
/ 2;
599 prio
= p
->static_prio
- bonus
;
600 if (prio
< MAX_RT_PRIO
)
602 if (prio
> MAX_PRIO
-1)
608 * __activate_task - move a task to the runqueue.
610 static inline void __activate_task(task_t
*p
, runqueue_t
*rq
)
612 enqueue_task(p
, rq
->active
);
617 * __activate_idle_task - move idle task to the _front_ of runqueue.
619 static inline void __activate_idle_task(task_t
*p
, runqueue_t
*rq
)
621 enqueue_task_head(p
, rq
->active
);
625 static void recalc_task_prio(task_t
*p
, unsigned long long now
)
627 /* Caller must always ensure 'now >= p->timestamp' */
628 unsigned long long __sleep_time
= now
- p
->timestamp
;
629 unsigned long sleep_time
;
631 if (__sleep_time
> NS_MAX_SLEEP_AVG
)
632 sleep_time
= NS_MAX_SLEEP_AVG
;
634 sleep_time
= (unsigned long)__sleep_time
;
636 if (likely(sleep_time
> 0)) {
638 * User tasks that sleep a long time are categorised as
639 * idle and will get just interactive status to stay active &
640 * prevent them suddenly becoming cpu hogs and starving
643 if (p
->mm
&& p
->activated
!= -1 &&
644 sleep_time
> INTERACTIVE_SLEEP(p
)) {
645 p
->sleep_avg
= JIFFIES_TO_NS(MAX_SLEEP_AVG
-
649 * The lower the sleep avg a task has the more
650 * rapidly it will rise with sleep time.
652 sleep_time
*= (MAX_BONUS
- CURRENT_BONUS(p
)) ? : 1;
655 * Tasks waking from uninterruptible sleep are
656 * limited in their sleep_avg rise as they
657 * are likely to be waiting on I/O
659 if (p
->activated
== -1 && p
->mm
) {
660 if (p
->sleep_avg
>= INTERACTIVE_SLEEP(p
))
662 else if (p
->sleep_avg
+ sleep_time
>=
663 INTERACTIVE_SLEEP(p
)) {
664 p
->sleep_avg
= INTERACTIVE_SLEEP(p
);
670 * This code gives a bonus to interactive tasks.
672 * The boost works by updating the 'average sleep time'
673 * value here, based on ->timestamp. The more time a
674 * task spends sleeping, the higher the average gets -
675 * and the higher the priority boost gets as well.
677 p
->sleep_avg
+= sleep_time
;
679 if (p
->sleep_avg
> NS_MAX_SLEEP_AVG
)
680 p
->sleep_avg
= NS_MAX_SLEEP_AVG
;
684 p
->prio
= effective_prio(p
);
688 * activate_task - move a task to the runqueue and do priority recalculation
690 * Update all the scheduling statistics stuff. (sleep average
691 * calculation, priority modifiers, etc.)
693 static void activate_task(task_t
*p
, runqueue_t
*rq
, int local
)
695 unsigned long long now
;
700 /* Compensate for drifting sched_clock */
701 runqueue_t
*this_rq
= this_rq();
702 now
= (now
- this_rq
->timestamp_last_tick
)
703 + rq
->timestamp_last_tick
;
707 recalc_task_prio(p
, now
);
710 * This checks to make sure it's not an uninterruptible task
711 * that is now waking up.
715 * Tasks which were woken up by interrupts (ie. hw events)
716 * are most likely of interactive nature. So we give them
717 * the credit of extending their sleep time to the period
718 * of time they spend on the runqueue, waiting for execution
719 * on a CPU, first time around:
725 * Normal first-time wakeups get a credit too for
726 * on-runqueue time, but it will be weighted down:
733 __activate_task(p
, rq
);
737 * deactivate_task - remove a task from the runqueue.
739 static void deactivate_task(struct task_struct
*p
, runqueue_t
*rq
)
742 dequeue_task(p
, p
->array
);
747 * resched_task - mark a task 'to be rescheduled now'.
749 * On UP this means the setting of the need_resched flag, on SMP it
750 * might also involve a cross-CPU call to trigger the scheduler on
754 static void resched_task(task_t
*p
)
756 int need_resched
, nrpolling
;
758 assert_spin_locked(&task_rq(p
)->lock
);
760 /* minimise the chance of sending an interrupt to poll_idle() */
761 nrpolling
= test_tsk_thread_flag(p
,TIF_POLLING_NRFLAG
);
762 need_resched
= test_and_set_tsk_thread_flag(p
,TIF_NEED_RESCHED
);
763 nrpolling
|= test_tsk_thread_flag(p
,TIF_POLLING_NRFLAG
);
765 if (!need_resched
&& !nrpolling
&& (task_cpu(p
) != smp_processor_id()))
766 smp_send_reschedule(task_cpu(p
));
769 static inline void resched_task(task_t
*p
)
771 set_tsk_need_resched(p
);
776 * task_curr - is this task currently executing on a CPU?
777 * @p: the task in question.
779 inline int task_curr(const task_t
*p
)
781 return cpu_curr(task_cpu(p
)) == p
;
791 struct list_head list
;
792 enum request_type type
;
794 /* For REQ_MOVE_TASK */
798 /* For REQ_SET_DOMAIN */
799 struct sched_domain
*sd
;
801 struct completion done
;
805 * The task's runqueue lock must be held.
806 * Returns true if you have to wait for migration thread.
808 static int migrate_task(task_t
*p
, int dest_cpu
, migration_req_t
*req
)
810 runqueue_t
*rq
= task_rq(p
);
813 * If the task is not on a runqueue (and not running), then
814 * it is sufficient to simply update the task's cpu field.
816 if (!p
->array
&& !task_running(rq
, p
)) {
817 set_task_cpu(p
, dest_cpu
);
821 init_completion(&req
->done
);
822 req
->type
= REQ_MOVE_TASK
;
824 req
->dest_cpu
= dest_cpu
;
825 list_add(&req
->list
, &rq
->migration_queue
);
830 * wait_task_inactive - wait for a thread to unschedule.
832 * The caller must ensure that the task *will* unschedule sometime soon,
833 * else this function might spin for a *long* time. This function can't
834 * be called with interrupts off, or it may introduce deadlock with
835 * smp_call_function() if an IPI is sent by the same process we are
836 * waiting to become inactive.
838 void wait_task_inactive(task_t
* p
)
845 rq
= task_rq_lock(p
, &flags
);
846 /* Must be off runqueue entirely, not preempted. */
847 if (unlikely(p
->array
|| task_running(rq
, p
))) {
848 /* If it's preempted, we yield. It could be a while. */
849 preempted
= !task_running(rq
, p
);
850 task_rq_unlock(rq
, &flags
);
856 task_rq_unlock(rq
, &flags
);
860 * kick_process - kick a running thread to enter/exit the kernel
861 * @p: the to-be-kicked thread
863 * Cause a process which is running on another CPU to enter
864 * kernel-mode, without any delay. (to get signals handled.)
866 * NOTE: this function doesnt have to take the runqueue lock,
867 * because all it wants to ensure is that the remote task enters
868 * the kernel. If the IPI races and the task has been migrated
869 * to another CPU then no harm is done and the purpose has been
872 void kick_process(task_t
*p
)
878 if ((cpu
!= smp_processor_id()) && task_curr(p
))
879 smp_send_reschedule(cpu
);
884 * Return a low guess at the load of a migration-source cpu.
886 * We want to under-estimate the load of migration sources, to
887 * balance conservatively.
889 static inline unsigned long source_load(int cpu
, int type
)
891 runqueue_t
*rq
= cpu_rq(cpu
);
892 unsigned long load_now
= rq
->nr_running
* SCHED_LOAD_SCALE
;
896 return min(rq
->cpu_load
[type
-1], load_now
);
900 * Return a high guess at the load of a migration-target cpu
902 static inline unsigned long target_load(int cpu
, int type
)
904 runqueue_t
*rq
= cpu_rq(cpu
);
905 unsigned long load_now
= rq
->nr_running
* SCHED_LOAD_SCALE
;
909 return max(rq
->cpu_load
[type
-1], load_now
);
915 * wake_idle() will wake a task on an idle cpu if task->cpu is
916 * not idle and an idle cpu is available. The span of cpus to
917 * search starts with cpus closest then further out as needed,
918 * so we always favor a closer, idle cpu.
920 * Returns the CPU we should wake onto.
922 #if defined(ARCH_HAS_SCHED_WAKE_IDLE)
923 static int wake_idle(int cpu
, task_t
*p
)
926 struct sched_domain
*sd
;
932 for_each_domain(cpu
, sd
) {
933 if (sd
->flags
& SD_WAKE_IDLE
) {
934 cpus_and(tmp
, sd
->span
, p
->cpus_allowed
);
935 for_each_cpu_mask(i
, tmp
) {
946 static inline int wake_idle(int cpu
, task_t
*p
)
953 * try_to_wake_up - wake up a thread
954 * @p: the to-be-woken-up thread
955 * @state: the mask of task states that can be woken
956 * @sync: do a synchronous wakeup?
958 * Put it on the run-queue if it's not already there. The "current"
959 * thread is always on the run-queue (except when the actual
960 * re-schedule is in progress), and as such you're allowed to do
961 * the simpler "current->state = TASK_RUNNING" to mark yourself
962 * runnable without the overhead of this.
964 * returns failure only if the task is already active.
966 static int try_to_wake_up(task_t
* p
, unsigned int state
, int sync
)
968 int cpu
, this_cpu
, success
= 0;
973 unsigned long load
, this_load
;
974 struct sched_domain
*sd
, *this_sd
= NULL
;
978 rq
= task_rq_lock(p
, &flags
);
979 old_state
= p
->state
;
980 if (!(old_state
& state
))
987 this_cpu
= smp_processor_id();
990 if (unlikely(task_running(rq
, p
)))
995 schedstat_inc(rq
, ttwu_cnt
);
996 if (cpu
== this_cpu
) {
997 schedstat_inc(rq
, ttwu_local
);
1001 for_each_domain(this_cpu
, sd
) {
1002 if (cpu_isset(cpu
, sd
->span
)) {
1003 schedstat_inc(sd
, ttwu_wake_remote
);
1009 if (unlikely(!cpu_isset(this_cpu
, p
->cpus_allowed
)))
1013 * Check for affine wakeup and passive balancing possibilities.
1016 int idx
= this_sd
->wake_idx
;
1017 unsigned int imbalance
;
1019 imbalance
= 100 + (this_sd
->imbalance_pct
- 100) / 2;
1021 load
= source_load(cpu
, idx
);
1022 this_load
= target_load(this_cpu
, idx
);
1024 new_cpu
= this_cpu
; /* Wake to this CPU if we can */
1026 if (this_sd
->flags
& SD_WAKE_AFFINE
) {
1027 unsigned long tl
= this_load
;
1029 * If sync wakeup then subtract the (maximum possible)
1030 * effect of the currently running task from the load
1031 * of the current CPU:
1034 tl
-= SCHED_LOAD_SCALE
;
1037 tl
+ target_load(cpu
, idx
) <= SCHED_LOAD_SCALE
) ||
1038 100*(tl
+ SCHED_LOAD_SCALE
) <= imbalance
*load
) {
1040 * This domain has SD_WAKE_AFFINE and
1041 * p is cache cold in this domain, and
1042 * there is no bad imbalance.
1044 schedstat_inc(this_sd
, ttwu_move_affine
);
1050 * Start passive balancing when half the imbalance_pct
1053 if (this_sd
->flags
& SD_WAKE_BALANCE
) {
1054 if (imbalance
*this_load
<= 100*load
) {
1055 schedstat_inc(this_sd
, ttwu_move_balance
);
1061 new_cpu
= cpu
; /* Could not wake to this_cpu. Wake to cpu instead */
1063 new_cpu
= wake_idle(new_cpu
, p
);
1064 if (new_cpu
!= cpu
) {
1065 set_task_cpu(p
, new_cpu
);
1066 task_rq_unlock(rq
, &flags
);
1067 /* might preempt at this point */
1068 rq
= task_rq_lock(p
, &flags
);
1069 old_state
= p
->state
;
1070 if (!(old_state
& state
))
1075 this_cpu
= smp_processor_id();
1080 #endif /* CONFIG_SMP */
1081 if (old_state
== TASK_UNINTERRUPTIBLE
) {
1082 rq
->nr_uninterruptible
--;
1084 * Tasks on involuntary sleep don't earn
1085 * sleep_avg beyond just interactive state.
1091 * Sync wakeups (i.e. those types of wakeups where the waker
1092 * has indicated that it will leave the CPU in short order)
1093 * don't trigger a preemption, if the woken up task will run on
1094 * this cpu. (in this case the 'I will reschedule' promise of
1095 * the waker guarantees that the freshly woken up task is going
1096 * to be considered on this CPU.)
1098 activate_task(p
, rq
, cpu
== this_cpu
);
1099 if (!sync
|| cpu
!= this_cpu
) {
1100 if (TASK_PREEMPTS_CURR(p
, rq
))
1101 resched_task(rq
->curr
);
1106 p
->state
= TASK_RUNNING
;
1108 task_rq_unlock(rq
, &flags
);
1113 int fastcall
wake_up_process(task_t
* p
)
1115 return try_to_wake_up(p
, TASK_STOPPED
| TASK_TRACED
|
1116 TASK_INTERRUPTIBLE
| TASK_UNINTERRUPTIBLE
, 0);
1119 EXPORT_SYMBOL(wake_up_process
);
1121 int fastcall
wake_up_state(task_t
*p
, unsigned int state
)
1123 return try_to_wake_up(p
, state
, 0);
1127 static int find_idlest_cpu(struct task_struct
*p
, int this_cpu
,
1128 struct sched_domain
*sd
);
1132 * Perform scheduler related setup for a newly forked process p.
1133 * p is forked by current.
1135 void fastcall
sched_fork(task_t
*p
)
1138 * We mark the process as running here, but have not actually
1139 * inserted it onto the runqueue yet. This guarantees that
1140 * nobody will actually run it, and a signal or other external
1141 * event cannot wake it up and insert it on the runqueue either.
1143 p
->state
= TASK_RUNNING
;
1144 INIT_LIST_HEAD(&p
->run_list
);
1146 spin_lock_init(&p
->switch_lock
);
1147 #ifdef CONFIG_SCHEDSTATS
1148 memset(&p
->sched_info
, 0, sizeof(p
->sched_info
));
1150 #ifdef CONFIG_PREEMPT
1152 * During context-switch we hold precisely one spinlock, which
1153 * schedule_tail drops. (in the common case it's this_rq()->lock,
1154 * but it also can be p->switch_lock.) So we compensate with a count
1155 * of 1. Also, we want to start with kernel preemption disabled.
1157 p
->thread_info
->preempt_count
= 1;
1160 * Share the timeslice between parent and child, thus the
1161 * total amount of pending timeslices in the system doesn't change,
1162 * resulting in more scheduling fairness.
1164 local_irq_disable();
1165 p
->time_slice
= (current
->time_slice
+ 1) >> 1;
1167 * The remainder of the first timeslice might be recovered by
1168 * the parent if the child exits early enough.
1170 p
->first_time_slice
= 1;
1171 current
->time_slice
>>= 1;
1172 p
->timestamp
= sched_clock();
1173 if (unlikely(!current
->time_slice
)) {
1175 * This case is rare, it happens when the parent has only
1176 * a single jiffy left from its timeslice. Taking the
1177 * runqueue lock is not a problem.
1179 current
->time_slice
= 1;
1189 * wake_up_new_task - wake up a newly created task for the first time.
1191 * This function will do some initial scheduler statistics housekeeping
1192 * that must be done for every newly created context, then puts the task
1193 * on the runqueue and wakes it.
1195 void fastcall
wake_up_new_task(task_t
* p
, unsigned long clone_flags
)
1197 unsigned long flags
;
1199 runqueue_t
*rq
, *this_rq
;
1201 rq
= task_rq_lock(p
, &flags
);
1203 this_cpu
= smp_processor_id();
1205 BUG_ON(p
->state
!= TASK_RUNNING
);
1208 * We decrease the sleep average of forking parents
1209 * and children as well, to keep max-interactive tasks
1210 * from forking tasks that are max-interactive. The parent
1211 * (current) is done further down, under its lock.
1213 p
->sleep_avg
= JIFFIES_TO_NS(CURRENT_BONUS(p
) *
1214 CHILD_PENALTY
/ 100 * MAX_SLEEP_AVG
/ MAX_BONUS
);
1216 p
->prio
= effective_prio(p
);
1218 if (likely(cpu
== this_cpu
)) {
1219 if (!(clone_flags
& CLONE_VM
)) {
1221 * The VM isn't cloned, so we're in a good position to
1222 * do child-runs-first in anticipation of an exec. This
1223 * usually avoids a lot of COW overhead.
1225 if (unlikely(!current
->array
))
1226 __activate_task(p
, rq
);
1228 p
->prio
= current
->prio
;
1229 list_add_tail(&p
->run_list
, ¤t
->run_list
);
1230 p
->array
= current
->array
;
1231 p
->array
->nr_active
++;
1236 /* Run child last */
1237 __activate_task(p
, rq
);
1239 * We skip the following code due to cpu == this_cpu
1241 * task_rq_unlock(rq, &flags);
1242 * this_rq = task_rq_lock(current, &flags);
1246 this_rq
= cpu_rq(this_cpu
);
1249 * Not the local CPU - must adjust timestamp. This should
1250 * get optimised away in the !CONFIG_SMP case.
1252 p
->timestamp
= (p
->timestamp
- this_rq
->timestamp_last_tick
)
1253 + rq
->timestamp_last_tick
;
1254 __activate_task(p
, rq
);
1255 if (TASK_PREEMPTS_CURR(p
, rq
))
1256 resched_task(rq
->curr
);
1259 * Parent and child are on different CPUs, now get the
1260 * parent runqueue to update the parent's ->sleep_avg:
1262 task_rq_unlock(rq
, &flags
);
1263 this_rq
= task_rq_lock(current
, &flags
);
1265 current
->sleep_avg
= JIFFIES_TO_NS(CURRENT_BONUS(current
) *
1266 PARENT_PENALTY
/ 100 * MAX_SLEEP_AVG
/ MAX_BONUS
);
1267 task_rq_unlock(this_rq
, &flags
);
1271 * Potentially available exiting-child timeslices are
1272 * retrieved here - this way the parent does not get
1273 * penalized for creating too many threads.
1275 * (this cannot be used to 'generate' timeslices
1276 * artificially, because any timeslice recovered here
1277 * was given away by the parent in the first place.)
1279 void fastcall
sched_exit(task_t
* p
)
1281 unsigned long flags
;
1285 * If the child was a (relative-) CPU hog then decrease
1286 * the sleep_avg of the parent as well.
1288 rq
= task_rq_lock(p
->parent
, &flags
);
1289 if (p
->first_time_slice
) {
1290 p
->parent
->time_slice
+= p
->time_slice
;
1291 if (unlikely(p
->parent
->time_slice
> task_timeslice(p
)))
1292 p
->parent
->time_slice
= task_timeslice(p
);
1294 if (p
->sleep_avg
< p
->parent
->sleep_avg
)
1295 p
->parent
->sleep_avg
= p
->parent
->sleep_avg
/
1296 (EXIT_WEIGHT
+ 1) * EXIT_WEIGHT
+ p
->sleep_avg
/
1298 task_rq_unlock(rq
, &flags
);
1302 * finish_task_switch - clean up after a task-switch
1303 * @prev: the thread we just switched away from.
1305 * We enter this with the runqueue still locked, and finish_arch_switch()
1306 * will unlock it along with doing any other architecture-specific cleanup
1309 * Note that we may have delayed dropping an mm in context_switch(). If
1310 * so, we finish that here outside of the runqueue lock. (Doing it
1311 * with the lock held can cause deadlocks; see schedule() for
1314 static inline void finish_task_switch(task_t
*prev
)
1315 __releases(rq
->lock
)
1317 runqueue_t
*rq
= this_rq();
1318 struct mm_struct
*mm
= rq
->prev_mm
;
1319 unsigned long prev_task_flags
;
1324 * A task struct has one reference for the use as "current".
1325 * If a task dies, then it sets EXIT_ZOMBIE in tsk->exit_state and
1326 * calls schedule one last time. The schedule call will never return,
1327 * and the scheduled task must drop that reference.
1328 * The test for EXIT_ZOMBIE must occur while the runqueue locks are
1329 * still held, otherwise prev could be scheduled on another cpu, die
1330 * there before we look at prev->state, and then the reference would
1332 * Manfred Spraul <manfred@colorfullife.com>
1334 prev_task_flags
= prev
->flags
;
1335 finish_arch_switch(rq
, prev
);
1338 if (unlikely(prev_task_flags
& PF_DEAD
))
1339 put_task_struct(prev
);
1343 * schedule_tail - first thing a freshly forked thread must call.
1344 * @prev: the thread we just switched away from.
1346 asmlinkage
void schedule_tail(task_t
*prev
)
1347 __releases(rq
->lock
)
1349 finish_task_switch(prev
);
1351 if (current
->set_child_tid
)
1352 put_user(current
->pid
, current
->set_child_tid
);
1356 * context_switch - switch to the new MM and the new
1357 * thread's register state.
1360 task_t
* context_switch(runqueue_t
*rq
, task_t
*prev
, task_t
*next
)
1362 struct mm_struct
*mm
= next
->mm
;
1363 struct mm_struct
*oldmm
= prev
->active_mm
;
1365 if (unlikely(!mm
)) {
1366 next
->active_mm
= oldmm
;
1367 atomic_inc(&oldmm
->mm_count
);
1368 enter_lazy_tlb(oldmm
, next
);
1370 switch_mm(oldmm
, mm
, next
);
1372 if (unlikely(!prev
->mm
)) {
1373 prev
->active_mm
= NULL
;
1374 WARN_ON(rq
->prev_mm
);
1375 rq
->prev_mm
= oldmm
;
1378 /* Here we just switch the register state and the stack. */
1379 switch_to(prev
, next
, prev
);
1385 * nr_running, nr_uninterruptible and nr_context_switches:
1387 * externally visible scheduler statistics: current number of runnable
1388 * threads, current number of uninterruptible-sleeping threads, total
1389 * number of context switches performed since bootup.
1391 unsigned long nr_running(void)
1393 unsigned long i
, sum
= 0;
1395 for_each_online_cpu(i
)
1396 sum
+= cpu_rq(i
)->nr_running
;
1401 unsigned long nr_uninterruptible(void)
1403 unsigned long i
, sum
= 0;
1406 sum
+= cpu_rq(i
)->nr_uninterruptible
;
1409 * Since we read the counters lockless, it might be slightly
1410 * inaccurate. Do not allow it to go below zero though:
1412 if (unlikely((long)sum
< 0))
1418 unsigned long long nr_context_switches(void)
1420 unsigned long long i
, sum
= 0;
1423 sum
+= cpu_rq(i
)->nr_switches
;
1428 unsigned long nr_iowait(void)
1430 unsigned long i
, sum
= 0;
1433 sum
+= atomic_read(&cpu_rq(i
)->nr_iowait
);
1441 * double_rq_lock - safely lock two runqueues
1443 * Note this does not disable interrupts like task_rq_lock,
1444 * you need to do so manually before calling.
1446 static void double_rq_lock(runqueue_t
*rq1
, runqueue_t
*rq2
)
1447 __acquires(rq1
->lock
)
1448 __acquires(rq2
->lock
)
1451 spin_lock(&rq1
->lock
);
1452 __acquire(rq2
->lock
); /* Fake it out ;) */
1455 spin_lock(&rq1
->lock
);
1456 spin_lock(&rq2
->lock
);
1458 spin_lock(&rq2
->lock
);
1459 spin_lock(&rq1
->lock
);
1465 * double_rq_unlock - safely unlock two runqueues
1467 * Note this does not restore interrupts like task_rq_unlock,
1468 * you need to do so manually after calling.
1470 static void double_rq_unlock(runqueue_t
*rq1
, runqueue_t
*rq2
)
1471 __releases(rq1
->lock
)
1472 __releases(rq2
->lock
)
1474 spin_unlock(&rq1
->lock
);
1476 spin_unlock(&rq2
->lock
);
1478 __release(rq2
->lock
);
1482 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1484 static void double_lock_balance(runqueue_t
*this_rq
, runqueue_t
*busiest
)
1485 __releases(this_rq
->lock
)
1486 __acquires(busiest
->lock
)
1487 __acquires(this_rq
->lock
)
1489 if (unlikely(!spin_trylock(&busiest
->lock
))) {
1490 if (busiest
< this_rq
) {
1491 spin_unlock(&this_rq
->lock
);
1492 spin_lock(&busiest
->lock
);
1493 spin_lock(&this_rq
->lock
);
1495 spin_lock(&busiest
->lock
);
1500 * find_idlest_cpu - find the least busy runqueue.
1502 static int find_idlest_cpu(struct task_struct
*p
, int this_cpu
,
1503 struct sched_domain
*sd
)
1505 unsigned long load
, min_load
, this_load
;
1510 min_load
= ULONG_MAX
;
1512 cpus_and(mask
, sd
->span
, p
->cpus_allowed
);
1514 for_each_cpu_mask(i
, mask
) {
1515 load
= target_load(i
, sd
->wake_idx
);
1517 if (load
< min_load
) {
1521 /* break out early on an idle CPU: */
1527 /* add +1 to account for the new task */
1528 this_load
= source_load(this_cpu
, sd
->wake_idx
) + SCHED_LOAD_SCALE
;
1531 * Would with the addition of the new task to the
1532 * current CPU there be an imbalance between this
1533 * CPU and the idlest CPU?
1535 * Use half of the balancing threshold - new-context is
1536 * a good opportunity to balance.
1538 if (min_load
*(100 + (sd
->imbalance_pct
-100)/2) < this_load
*100)
1545 * If dest_cpu is allowed for this process, migrate the task to it.
1546 * This is accomplished by forcing the cpu_allowed mask to only
1547 * allow dest_cpu, which will force the cpu onto dest_cpu. Then
1548 * the cpu_allowed mask is restored.
1550 static void sched_migrate_task(task_t
*p
, int dest_cpu
)
1552 migration_req_t req
;
1554 unsigned long flags
;
1556 rq
= task_rq_lock(p
, &flags
);
1557 if (!cpu_isset(dest_cpu
, p
->cpus_allowed
)
1558 || unlikely(cpu_is_offline(dest_cpu
)))
1561 /* force the process onto the specified CPU */
1562 if (migrate_task(p
, dest_cpu
, &req
)) {
1563 /* Need to wait for migration thread (might exit: take ref). */
1564 struct task_struct
*mt
= rq
->migration_thread
;
1565 get_task_struct(mt
);
1566 task_rq_unlock(rq
, &flags
);
1567 wake_up_process(mt
);
1568 put_task_struct(mt
);
1569 wait_for_completion(&req
.done
);
1573 task_rq_unlock(rq
, &flags
);
1577 * sched_exec(): find the highest-level, exec-balance-capable
1578 * domain and try to migrate the task to the least loaded CPU.
1580 * execve() is a valuable balancing opportunity, because at this point
1581 * the task has the smallest effective memory and cache footprint.
1583 void sched_exec(void)
1585 struct sched_domain
*tmp
, *sd
= NULL
;
1586 int new_cpu
, this_cpu
= get_cpu();
1588 /* Prefer the current CPU if there's only this task running */
1589 if (this_rq()->nr_running
<= 1)
1592 for_each_domain(this_cpu
, tmp
)
1593 if (tmp
->flags
& SD_BALANCE_EXEC
)
1597 schedstat_inc(sd
, sbe_attempts
);
1598 new_cpu
= find_idlest_cpu(current
, this_cpu
, sd
);
1599 if (new_cpu
!= this_cpu
) {
1600 schedstat_inc(sd
, sbe_pushed
);
1602 sched_migrate_task(current
, new_cpu
);
1611 * pull_task - move a task from a remote runqueue to the local runqueue.
1612 * Both runqueues must be locked.
1615 void pull_task(runqueue_t
*src_rq
, prio_array_t
*src_array
, task_t
*p
,
1616 runqueue_t
*this_rq
, prio_array_t
*this_array
, int this_cpu
)
1618 dequeue_task(p
, src_array
);
1619 src_rq
->nr_running
--;
1620 set_task_cpu(p
, this_cpu
);
1621 this_rq
->nr_running
++;
1622 enqueue_task(p
, this_array
);
1623 p
->timestamp
= (p
->timestamp
- src_rq
->timestamp_last_tick
)
1624 + this_rq
->timestamp_last_tick
;
1626 * Note that idle threads have a prio of MAX_PRIO, for this test
1627 * to be always true for them.
1629 if (TASK_PREEMPTS_CURR(p
, this_rq
))
1630 resched_task(this_rq
->curr
);
1634 * can_migrate_task - may task p from runqueue rq be migrated to this_cpu?
1637 int can_migrate_task(task_t
*p
, runqueue_t
*rq
, int this_cpu
,
1638 struct sched_domain
*sd
, enum idle_type idle
, int *all_pinned
)
1641 * We do not migrate tasks that are:
1642 * 1) running (obviously), or
1643 * 2) cannot be migrated to this CPU due to cpus_allowed, or
1644 * 3) are cache-hot on their current CPU.
1646 if (!cpu_isset(this_cpu
, p
->cpus_allowed
))
1650 if (task_running(rq
, p
))
1654 * Aggressive migration if:
1655 * 1) the [whole] cpu is idle, or
1656 * 2) too many balance attempts have failed.
1659 if (cpu_and_siblings_are_idle(this_cpu
) || \
1660 sd
->nr_balance_failed
> sd
->cache_nice_tries
)
1663 if (task_hot(p
, rq
->timestamp_last_tick
, sd
))
1669 * move_tasks tries to move up to max_nr_move tasks from busiest to this_rq,
1670 * as part of a balancing operation within "domain". Returns the number of
1673 * Called with both runqueues locked.
1675 static int move_tasks(runqueue_t
*this_rq
, int this_cpu
, runqueue_t
*busiest
,
1676 unsigned long max_nr_move
, struct sched_domain
*sd
,
1677 enum idle_type idle
, int *all_pinned
)
1679 prio_array_t
*array
, *dst_array
;
1680 struct list_head
*head
, *curr
;
1681 int idx
, pulled
= 0, pinned
= 0;
1684 if (max_nr_move
== 0)
1690 * We first consider expired tasks. Those will likely not be
1691 * executed in the near future, and they are most likely to
1692 * be cache-cold, thus switching CPUs has the least effect
1695 if (busiest
->expired
->nr_active
) {
1696 array
= busiest
->expired
;
1697 dst_array
= this_rq
->expired
;
1699 array
= busiest
->active
;
1700 dst_array
= this_rq
->active
;
1704 /* Start searching at priority 0: */
1708 idx
= sched_find_first_bit(array
->bitmap
);
1710 idx
= find_next_bit(array
->bitmap
, MAX_PRIO
, idx
);
1711 if (idx
>= MAX_PRIO
) {
1712 if (array
== busiest
->expired
&& busiest
->active
->nr_active
) {
1713 array
= busiest
->active
;
1714 dst_array
= this_rq
->active
;
1720 head
= array
->queue
+ idx
;
1723 tmp
= list_entry(curr
, task_t
, run_list
);
1727 if (!can_migrate_task(tmp
, busiest
, this_cpu
, sd
, idle
, &pinned
)) {
1734 #ifdef CONFIG_SCHEDSTATS
1735 if (task_hot(tmp
, busiest
->timestamp_last_tick
, sd
))
1736 schedstat_inc(sd
, lb_hot_gained
[idle
]);
1739 pull_task(busiest
, array
, tmp
, this_rq
, dst_array
, this_cpu
);
1742 /* We only want to steal up to the prescribed number of tasks. */
1743 if (pulled
< max_nr_move
) {
1751 * Right now, this is the only place pull_task() is called,
1752 * so we can safely collect pull_task() stats here rather than
1753 * inside pull_task().
1755 schedstat_add(sd
, lb_gained
[idle
], pulled
);
1758 *all_pinned
= pinned
;
1763 * find_busiest_group finds and returns the busiest CPU group within the
1764 * domain. It calculates and returns the number of tasks which should be
1765 * moved to restore balance via the imbalance parameter.
1767 static struct sched_group
*
1768 find_busiest_group(struct sched_domain
*sd
, int this_cpu
,
1769 unsigned long *imbalance
, enum idle_type idle
)
1771 struct sched_group
*busiest
= NULL
, *this = NULL
, *group
= sd
->groups
;
1772 unsigned long max_load
, avg_load
, total_load
, this_load
, total_pwr
;
1775 max_load
= this_load
= total_load
= total_pwr
= 0;
1776 if (idle
== NOT_IDLE
)
1777 load_idx
= sd
->busy_idx
;
1778 else if (idle
== NEWLY_IDLE
)
1779 load_idx
= sd
->newidle_idx
;
1781 load_idx
= sd
->idle_idx
;
1788 local_group
= cpu_isset(this_cpu
, group
->cpumask
);
1790 /* Tally up the load of all CPUs in the group */
1793 for_each_cpu_mask(i
, group
->cpumask
) {
1794 /* Bias balancing toward cpus of our domain */
1796 load
= target_load(i
, load_idx
);
1798 load
= source_load(i
, load_idx
);
1803 total_load
+= avg_load
;
1804 total_pwr
+= group
->cpu_power
;
1806 /* Adjust by relative CPU power of the group */
1807 avg_load
= (avg_load
* SCHED_LOAD_SCALE
) / group
->cpu_power
;
1810 this_load
= avg_load
;
1813 } else if (avg_load
> max_load
) {
1814 max_load
= avg_load
;
1818 group
= group
->next
;
1819 } while (group
!= sd
->groups
);
1821 if (!busiest
|| this_load
>= max_load
)
1824 avg_load
= (SCHED_LOAD_SCALE
* total_load
) / total_pwr
;
1826 if (this_load
>= avg_load
||
1827 100*max_load
<= sd
->imbalance_pct
*this_load
)
1831 * We're trying to get all the cpus to the average_load, so we don't
1832 * want to push ourselves above the average load, nor do we wish to
1833 * reduce the max loaded cpu below the average load, as either of these
1834 * actions would just result in more rebalancing later, and ping-pong
1835 * tasks around. Thus we look for the minimum possible imbalance.
1836 * Negative imbalances (*we* are more loaded than anyone else) will
1837 * be counted as no imbalance for these purposes -- we can't fix that
1838 * by pulling tasks to us. Be careful of negative numbers as they'll
1839 * appear as very large values with unsigned longs.
1841 /* How much load to actually move to equalise the imbalance */
1842 *imbalance
= min((max_load
- avg_load
) * busiest
->cpu_power
,
1843 (avg_load
- this_load
) * this->cpu_power
)
1846 if (*imbalance
< SCHED_LOAD_SCALE
) {
1847 unsigned long pwr_now
= 0, pwr_move
= 0;
1850 if (max_load
- this_load
>= SCHED_LOAD_SCALE
*2) {
1856 * OK, we don't have enough imbalance to justify moving tasks,
1857 * however we may be able to increase total CPU power used by
1861 pwr_now
+= busiest
->cpu_power
*min(SCHED_LOAD_SCALE
, max_load
);
1862 pwr_now
+= this->cpu_power
*min(SCHED_LOAD_SCALE
, this_load
);
1863 pwr_now
/= SCHED_LOAD_SCALE
;
1865 /* Amount of load we'd subtract */
1866 tmp
= SCHED_LOAD_SCALE
*SCHED_LOAD_SCALE
/busiest
->cpu_power
;
1868 pwr_move
+= busiest
->cpu_power
*min(SCHED_LOAD_SCALE
,
1871 /* Amount of load we'd add */
1872 if (max_load
*busiest
->cpu_power
<
1873 SCHED_LOAD_SCALE
*SCHED_LOAD_SCALE
)
1874 tmp
= max_load
*busiest
->cpu_power
/this->cpu_power
;
1876 tmp
= SCHED_LOAD_SCALE
*SCHED_LOAD_SCALE
/this->cpu_power
;
1877 pwr_move
+= this->cpu_power
*min(SCHED_LOAD_SCALE
, this_load
+ tmp
);
1878 pwr_move
/= SCHED_LOAD_SCALE
;
1880 /* Move if we gain throughput */
1881 if (pwr_move
<= pwr_now
)
1888 /* Get rid of the scaling factor, rounding down as we divide */
1889 *imbalance
= *imbalance
/ SCHED_LOAD_SCALE
;
1899 * find_busiest_queue - find the busiest runqueue among the cpus in group.
1901 static runqueue_t
*find_busiest_queue(struct sched_group
*group
)
1903 unsigned long load
, max_load
= 0;
1904 runqueue_t
*busiest
= NULL
;
1907 for_each_cpu_mask(i
, group
->cpumask
) {
1908 load
= source_load(i
, 0);
1910 if (load
> max_load
) {
1912 busiest
= cpu_rq(i
);
1920 * Check this_cpu to ensure it is balanced within domain. Attempt to move
1921 * tasks if there is an imbalance.
1923 * Called with this_rq unlocked.
1925 static int load_balance(int this_cpu
, runqueue_t
*this_rq
,
1926 struct sched_domain
*sd
, enum idle_type idle
)
1928 struct sched_group
*group
;
1929 runqueue_t
*busiest
;
1930 unsigned long imbalance
;
1931 int nr_moved
, all_pinned
;
1932 int active_balance
= 0;
1934 spin_lock(&this_rq
->lock
);
1935 schedstat_inc(sd
, lb_cnt
[idle
]);
1937 group
= find_busiest_group(sd
, this_cpu
, &imbalance
, idle
);
1939 schedstat_inc(sd
, lb_nobusyg
[idle
]);
1943 busiest
= find_busiest_queue(group
);
1945 schedstat_inc(sd
, lb_nobusyq
[idle
]);
1949 BUG_ON(busiest
== this_rq
);
1951 schedstat_add(sd
, lb_imbalance
[idle
], imbalance
);
1954 if (busiest
->nr_running
> 1) {
1956 * Attempt to move tasks. If find_busiest_group has found
1957 * an imbalance but busiest->nr_running <= 1, the group is
1958 * still unbalanced. nr_moved simply stays zero, so it is
1959 * correctly treated as an imbalance.
1961 double_lock_balance(this_rq
, busiest
);
1962 nr_moved
= move_tasks(this_rq
, this_cpu
, busiest
,
1963 imbalance
, sd
, idle
,
1965 spin_unlock(&busiest
->lock
);
1967 /* All tasks on this runqueue were pinned by CPU affinity */
1968 if (unlikely(all_pinned
))
1972 spin_unlock(&this_rq
->lock
);
1975 schedstat_inc(sd
, lb_failed
[idle
]);
1976 sd
->nr_balance_failed
++;
1978 if (unlikely(sd
->nr_balance_failed
> sd
->cache_nice_tries
+2)) {
1980 spin_lock(&busiest
->lock
);
1981 if (!busiest
->active_balance
) {
1982 busiest
->active_balance
= 1;
1983 busiest
->push_cpu
= this_cpu
;
1986 spin_unlock(&busiest
->lock
);
1988 wake_up_process(busiest
->migration_thread
);
1991 * We've kicked active balancing, reset the failure
1994 sd
->nr_balance_failed
= sd
->cache_nice_tries
+1;
1997 sd
->nr_balance_failed
= 0;
1999 if (likely(!active_balance
)) {
2000 /* We were unbalanced, so reset the balancing interval */
2001 sd
->balance_interval
= sd
->min_interval
;
2004 * If we've begun active balancing, start to back off. This
2005 * case may not be covered by the all_pinned logic if there
2006 * is only 1 task on the busy runqueue (because we don't call
2009 if (sd
->balance_interval
< sd
->max_interval
)
2010 sd
->balance_interval
*= 2;
2016 spin_unlock(&this_rq
->lock
);
2018 schedstat_inc(sd
, lb_balanced
[idle
]);
2020 sd
->nr_balance_failed
= 0;
2021 /* tune up the balancing interval */
2022 if (sd
->balance_interval
< sd
->max_interval
)
2023 sd
->balance_interval
*= 2;
2029 * Check this_cpu to ensure it is balanced within domain. Attempt to move
2030 * tasks if there is an imbalance.
2032 * Called from schedule when this_rq is about to become idle (NEWLY_IDLE).
2033 * this_rq is locked.
2035 static int load_balance_newidle(int this_cpu
, runqueue_t
*this_rq
,
2036 struct sched_domain
*sd
)
2038 struct sched_group
*group
;
2039 runqueue_t
*busiest
= NULL
;
2040 unsigned long imbalance
;
2043 schedstat_inc(sd
, lb_cnt
[NEWLY_IDLE
]);
2044 group
= find_busiest_group(sd
, this_cpu
, &imbalance
, NEWLY_IDLE
);
2046 schedstat_inc(sd
, lb_nobusyg
[NEWLY_IDLE
]);
2050 busiest
= find_busiest_queue(group
);
2052 schedstat_inc(sd
, lb_nobusyq
[NEWLY_IDLE
]);
2056 BUG_ON(busiest
== this_rq
);
2058 /* Attempt to move tasks */
2059 double_lock_balance(this_rq
, busiest
);
2061 schedstat_add(sd
, lb_imbalance
[NEWLY_IDLE
], imbalance
);
2062 nr_moved
= move_tasks(this_rq
, this_cpu
, busiest
,
2063 imbalance
, sd
, NEWLY_IDLE
, NULL
);
2065 schedstat_inc(sd
, lb_failed
[NEWLY_IDLE
]);
2067 sd
->nr_balance_failed
= 0;
2069 spin_unlock(&busiest
->lock
);
2073 schedstat_inc(sd
, lb_balanced
[NEWLY_IDLE
]);
2074 sd
->nr_balance_failed
= 0;
2079 * idle_balance is called by schedule() if this_cpu is about to become
2080 * idle. Attempts to pull tasks from other CPUs.
2082 static inline void idle_balance(int this_cpu
, runqueue_t
*this_rq
)
2084 struct sched_domain
*sd
;
2086 for_each_domain(this_cpu
, sd
) {
2087 if (sd
->flags
& SD_BALANCE_NEWIDLE
) {
2088 if (load_balance_newidle(this_cpu
, this_rq
, sd
)) {
2089 /* We've pulled tasks over so stop searching */
2097 * active_load_balance is run by migration threads. It pushes running tasks
2098 * off the busiest CPU onto idle CPUs. It requires at least 1 task to be
2099 * running on each physical CPU where possible, and avoids physical /
2100 * logical imbalances.
2102 * Called with busiest_rq locked.
2104 static void active_load_balance(runqueue_t
*busiest_rq
, int busiest_cpu
)
2106 struct sched_domain
*sd
;
2107 runqueue_t
*target_rq
;
2108 int target_cpu
= busiest_rq
->push_cpu
;
2110 if (busiest_rq
->nr_running
<= 1)
2111 /* no task to move */
2114 target_rq
= cpu_rq(target_cpu
);
2117 * This condition is "impossible", if it occurs
2118 * we need to fix it. Originally reported by
2119 * Bjorn Helgaas on a 128-cpu setup.
2121 BUG_ON(busiest_rq
== target_rq
);
2123 /* move a task from busiest_rq to target_rq */
2124 double_lock_balance(busiest_rq
, target_rq
);
2126 /* Search for an sd spanning us and the target CPU. */
2127 for_each_domain(target_cpu
, sd
)
2128 if ((sd
->flags
& SD_LOAD_BALANCE
) &&
2129 cpu_isset(busiest_cpu
, sd
->span
))
2132 if (unlikely(sd
== NULL
))
2135 schedstat_inc(sd
, alb_cnt
);
2137 if (move_tasks(target_rq
, target_cpu
, busiest_rq
, 1, sd
, SCHED_IDLE
, NULL
))
2138 schedstat_inc(sd
, alb_pushed
);
2140 schedstat_inc(sd
, alb_failed
);
2142 spin_unlock(&target_rq
->lock
);
2146 * rebalance_tick will get called every timer tick, on every CPU.
2148 * It checks each scheduling domain to see if it is due to be balanced,
2149 * and initiates a balancing operation if so.
2151 * Balancing parameters are set up in arch_init_sched_domains.
2154 /* Don't have all balancing operations going off at once */
2155 #define CPU_OFFSET(cpu) (HZ * cpu / NR_CPUS)
2157 static void rebalance_tick(int this_cpu
, runqueue_t
*this_rq
,
2158 enum idle_type idle
)
2160 unsigned long old_load
, this_load
;
2161 unsigned long j
= jiffies
+ CPU_OFFSET(this_cpu
);
2162 struct sched_domain
*sd
;
2165 this_load
= this_rq
->nr_running
* SCHED_LOAD_SCALE
;
2166 /* Update our load */
2167 for (i
= 0; i
< 3; i
++) {
2168 unsigned long new_load
= this_load
;
2170 old_load
= this_rq
->cpu_load
[i
];
2172 * Round up the averaging division if load is increasing. This
2173 * prevents us from getting stuck on 9 if the load is 10, for
2176 if (new_load
> old_load
)
2177 new_load
+= scale
-1;
2178 this_rq
->cpu_load
[i
] = (old_load
*(scale
-1) + new_load
) / scale
;
2181 for_each_domain(this_cpu
, sd
) {
2182 unsigned long interval
;
2184 if (!(sd
->flags
& SD_LOAD_BALANCE
))
2187 interval
= sd
->balance_interval
;
2188 if (idle
!= SCHED_IDLE
)
2189 interval
*= sd
->busy_factor
;
2191 /* scale ms to jiffies */
2192 interval
= msecs_to_jiffies(interval
);
2193 if (unlikely(!interval
))
2196 if (j
- sd
->last_balance
>= interval
) {
2197 if (load_balance(this_cpu
, this_rq
, sd
, idle
)) {
2198 /* We've pulled tasks over so no longer idle */
2201 sd
->last_balance
+= interval
;
2207 * on UP we do not need to balance between CPUs:
2209 static inline void rebalance_tick(int cpu
, runqueue_t
*rq
, enum idle_type idle
)
2212 static inline void idle_balance(int cpu
, runqueue_t
*rq
)
2217 static inline int wake_priority_sleeper(runqueue_t
*rq
)
2220 #ifdef CONFIG_SCHED_SMT
2221 spin_lock(&rq
->lock
);
2223 * If an SMT sibling task has been put to sleep for priority
2224 * reasons reschedule the idle task to see if it can now run.
2226 if (rq
->nr_running
) {
2227 resched_task(rq
->idle
);
2230 spin_unlock(&rq
->lock
);
2235 DEFINE_PER_CPU(struct kernel_stat
, kstat
);
2237 EXPORT_PER_CPU_SYMBOL(kstat
);
2240 * This is called on clock ticks and on context switches.
2241 * Bank in p->sched_time the ns elapsed since the last tick or switch.
2243 static inline void update_cpu_clock(task_t
*p
, runqueue_t
*rq
,
2244 unsigned long long now
)
2246 unsigned long long last
= max(p
->timestamp
, rq
->timestamp_last_tick
);
2247 p
->sched_time
+= now
- last
;
2251 * Return current->sched_time plus any more ns on the sched_clock
2252 * that have not yet been banked.
2254 unsigned long long current_sched_time(const task_t
*tsk
)
2256 unsigned long long ns
;
2257 unsigned long flags
;
2258 local_irq_save(flags
);
2259 ns
= max(tsk
->timestamp
, task_rq(tsk
)->timestamp_last_tick
);
2260 ns
= tsk
->sched_time
+ (sched_clock() - ns
);
2261 local_irq_restore(flags
);
2266 * We place interactive tasks back into the active array, if possible.
2268 * To guarantee that this does not starve expired tasks we ignore the
2269 * interactivity of a task if the first expired task had to wait more
2270 * than a 'reasonable' amount of time. This deadline timeout is
2271 * load-dependent, as the frequency of array switched decreases with
2272 * increasing number of running tasks. We also ignore the interactivity
2273 * if a better static_prio task has expired:
2275 #define EXPIRED_STARVING(rq) \
2276 ((STARVATION_LIMIT && ((rq)->expired_timestamp && \
2277 (jiffies - (rq)->expired_timestamp >= \
2278 STARVATION_LIMIT * ((rq)->nr_running) + 1))) || \
2279 ((rq)->curr->static_prio > (rq)->best_expired_prio))
2282 * Account user cpu time to a process.
2283 * @p: the process that the cpu time gets accounted to
2284 * @hardirq_offset: the offset to subtract from hardirq_count()
2285 * @cputime: the cpu time spent in user space since the last update
2287 void account_user_time(struct task_struct
*p
, cputime_t cputime
)
2289 struct cpu_usage_stat
*cpustat
= &kstat_this_cpu
.cpustat
;
2292 p
->utime
= cputime_add(p
->utime
, cputime
);
2294 /* Add user time to cpustat. */
2295 tmp
= cputime_to_cputime64(cputime
);
2296 if (TASK_NICE(p
) > 0)
2297 cpustat
->nice
= cputime64_add(cpustat
->nice
, tmp
);
2299 cpustat
->user
= cputime64_add(cpustat
->user
, tmp
);
2303 * Account system cpu time to a process.
2304 * @p: the process that the cpu time gets accounted to
2305 * @hardirq_offset: the offset to subtract from hardirq_count()
2306 * @cputime: the cpu time spent in kernel space since the last update
2308 void account_system_time(struct task_struct
*p
, int hardirq_offset
,
2311 struct cpu_usage_stat
*cpustat
= &kstat_this_cpu
.cpustat
;
2312 runqueue_t
*rq
= this_rq();
2315 p
->stime
= cputime_add(p
->stime
, cputime
);
2317 /* Add system time to cpustat. */
2318 tmp
= cputime_to_cputime64(cputime
);
2319 if (hardirq_count() - hardirq_offset
)
2320 cpustat
->irq
= cputime64_add(cpustat
->irq
, tmp
);
2321 else if (softirq_count())
2322 cpustat
->softirq
= cputime64_add(cpustat
->softirq
, tmp
);
2323 else if (p
!= rq
->idle
)
2324 cpustat
->system
= cputime64_add(cpustat
->system
, tmp
);
2325 else if (atomic_read(&rq
->nr_iowait
) > 0)
2326 cpustat
->iowait
= cputime64_add(cpustat
->iowait
, tmp
);
2328 cpustat
->idle
= cputime64_add(cpustat
->idle
, tmp
);
2329 /* Account for system time used */
2330 acct_update_integrals(p
);
2331 /* Update rss highwater mark */
2332 update_mem_hiwater(p
);
2336 * Account for involuntary wait time.
2337 * @p: the process from which the cpu time has been stolen
2338 * @steal: the cpu time spent in involuntary wait
2340 void account_steal_time(struct task_struct
*p
, cputime_t steal
)
2342 struct cpu_usage_stat
*cpustat
= &kstat_this_cpu
.cpustat
;
2343 cputime64_t tmp
= cputime_to_cputime64(steal
);
2344 runqueue_t
*rq
= this_rq();
2346 if (p
== rq
->idle
) {
2347 p
->stime
= cputime_add(p
->stime
, steal
);
2348 if (atomic_read(&rq
->nr_iowait
) > 0)
2349 cpustat
->iowait
= cputime64_add(cpustat
->iowait
, tmp
);
2351 cpustat
->idle
= cputime64_add(cpustat
->idle
, tmp
);
2353 cpustat
->steal
= cputime64_add(cpustat
->steal
, tmp
);
2357 * This function gets called by the timer code, with HZ frequency.
2358 * We call it with interrupts disabled.
2360 * It also gets called by the fork code, when changing the parent's
2363 void scheduler_tick(void)
2365 int cpu
= smp_processor_id();
2366 runqueue_t
*rq
= this_rq();
2367 task_t
*p
= current
;
2368 unsigned long long now
= sched_clock();
2370 update_cpu_clock(p
, rq
, now
);
2372 rq
->timestamp_last_tick
= now
;
2374 if (p
== rq
->idle
) {
2375 if (wake_priority_sleeper(rq
))
2377 rebalance_tick(cpu
, rq
, SCHED_IDLE
);
2381 /* Task might have expired already, but not scheduled off yet */
2382 if (p
->array
!= rq
->active
) {
2383 set_tsk_need_resched(p
);
2386 spin_lock(&rq
->lock
);
2388 * The task was running during this tick - update the
2389 * time slice counter. Note: we do not update a thread's
2390 * priority until it either goes to sleep or uses up its
2391 * timeslice. This makes it possible for interactive tasks
2392 * to use up their timeslices at their highest priority levels.
2396 * RR tasks need a special form of timeslice management.
2397 * FIFO tasks have no timeslices.
2399 if ((p
->policy
== SCHED_RR
) && !--p
->time_slice
) {
2400 p
->time_slice
= task_timeslice(p
);
2401 p
->first_time_slice
= 0;
2402 set_tsk_need_resched(p
);
2404 /* put it at the end of the queue: */
2405 requeue_task(p
, rq
->active
);
2409 if (!--p
->time_slice
) {
2410 dequeue_task(p
, rq
->active
);
2411 set_tsk_need_resched(p
);
2412 p
->prio
= effective_prio(p
);
2413 p
->time_slice
= task_timeslice(p
);
2414 p
->first_time_slice
= 0;
2416 if (!rq
->expired_timestamp
)
2417 rq
->expired_timestamp
= jiffies
;
2418 if (!TASK_INTERACTIVE(p
) || EXPIRED_STARVING(rq
)) {
2419 enqueue_task(p
, rq
->expired
);
2420 if (p
->static_prio
< rq
->best_expired_prio
)
2421 rq
->best_expired_prio
= p
->static_prio
;
2423 enqueue_task(p
, rq
->active
);
2426 * Prevent a too long timeslice allowing a task to monopolize
2427 * the CPU. We do this by splitting up the timeslice into
2430 * Note: this does not mean the task's timeslices expire or
2431 * get lost in any way, they just might be preempted by
2432 * another task of equal priority. (one with higher
2433 * priority would have preempted this task already.) We
2434 * requeue this task to the end of the list on this priority
2435 * level, which is in essence a round-robin of tasks with
2438 * This only applies to tasks in the interactive
2439 * delta range with at least TIMESLICE_GRANULARITY to requeue.
2441 if (TASK_INTERACTIVE(p
) && !((task_timeslice(p
) -
2442 p
->time_slice
) % TIMESLICE_GRANULARITY(p
)) &&
2443 (p
->time_slice
>= TIMESLICE_GRANULARITY(p
)) &&
2444 (p
->array
== rq
->active
)) {
2446 requeue_task(p
, rq
->active
);
2447 set_tsk_need_resched(p
);
2451 spin_unlock(&rq
->lock
);
2453 rebalance_tick(cpu
, rq
, NOT_IDLE
);
2456 #ifdef CONFIG_SCHED_SMT
2457 static inline void wake_sleeping_dependent(int this_cpu
, runqueue_t
*this_rq
)
2459 struct sched_domain
*sd
= this_rq
->sd
;
2460 cpumask_t sibling_map
;
2463 if (!(sd
->flags
& SD_SHARE_CPUPOWER
))
2467 * Unlock the current runqueue because we have to lock in
2468 * CPU order to avoid deadlocks. Caller knows that we might
2469 * unlock. We keep IRQs disabled.
2471 spin_unlock(&this_rq
->lock
);
2473 sibling_map
= sd
->span
;
2475 for_each_cpu_mask(i
, sibling_map
)
2476 spin_lock(&cpu_rq(i
)->lock
);
2478 * We clear this CPU from the mask. This both simplifies the
2479 * inner loop and keps this_rq locked when we exit:
2481 cpu_clear(this_cpu
, sibling_map
);
2483 for_each_cpu_mask(i
, sibling_map
) {
2484 runqueue_t
*smt_rq
= cpu_rq(i
);
2487 * If an SMT sibling task is sleeping due to priority
2488 * reasons wake it up now.
2490 if (smt_rq
->curr
== smt_rq
->idle
&& smt_rq
->nr_running
)
2491 resched_task(smt_rq
->idle
);
2494 for_each_cpu_mask(i
, sibling_map
)
2495 spin_unlock(&cpu_rq(i
)->lock
);
2497 * We exit with this_cpu's rq still held and IRQs
2502 static inline int dependent_sleeper(int this_cpu
, runqueue_t
*this_rq
)
2504 struct sched_domain
*sd
= this_rq
->sd
;
2505 cpumask_t sibling_map
;
2506 prio_array_t
*array
;
2510 if (!(sd
->flags
& SD_SHARE_CPUPOWER
))
2514 * The same locking rules and details apply as for
2515 * wake_sleeping_dependent():
2517 spin_unlock(&this_rq
->lock
);
2518 sibling_map
= sd
->span
;
2519 for_each_cpu_mask(i
, sibling_map
)
2520 spin_lock(&cpu_rq(i
)->lock
);
2521 cpu_clear(this_cpu
, sibling_map
);
2524 * Establish next task to be run - it might have gone away because
2525 * we released the runqueue lock above:
2527 if (!this_rq
->nr_running
)
2529 array
= this_rq
->active
;
2530 if (!array
->nr_active
)
2531 array
= this_rq
->expired
;
2532 BUG_ON(!array
->nr_active
);
2534 p
= list_entry(array
->queue
[sched_find_first_bit(array
->bitmap
)].next
,
2537 for_each_cpu_mask(i
, sibling_map
) {
2538 runqueue_t
*smt_rq
= cpu_rq(i
);
2539 task_t
*smt_curr
= smt_rq
->curr
;
2542 * If a user task with lower static priority than the
2543 * running task on the SMT sibling is trying to schedule,
2544 * delay it till there is proportionately less timeslice
2545 * left of the sibling task to prevent a lower priority
2546 * task from using an unfair proportion of the
2547 * physical cpu's resources. -ck
2549 if (((smt_curr
->time_slice
* (100 - sd
->per_cpu_gain
) / 100) >
2550 task_timeslice(p
) || rt_task(smt_curr
)) &&
2551 p
->mm
&& smt_curr
->mm
&& !rt_task(p
))
2555 * Reschedule a lower priority task on the SMT sibling,
2556 * or wake it up if it has been put to sleep for priority
2559 if ((((p
->time_slice
* (100 - sd
->per_cpu_gain
) / 100) >
2560 task_timeslice(smt_curr
) || rt_task(p
)) &&
2561 smt_curr
->mm
&& p
->mm
&& !rt_task(smt_curr
)) ||
2562 (smt_curr
== smt_rq
->idle
&& smt_rq
->nr_running
))
2563 resched_task(smt_curr
);
2566 for_each_cpu_mask(i
, sibling_map
)
2567 spin_unlock(&cpu_rq(i
)->lock
);
2571 static inline void wake_sleeping_dependent(int this_cpu
, runqueue_t
*this_rq
)
2575 static inline int dependent_sleeper(int this_cpu
, runqueue_t
*this_rq
)
2581 #if defined(CONFIG_PREEMPT) && defined(CONFIG_DEBUG_PREEMPT)
2583 void fastcall
add_preempt_count(int val
)
2588 BUG_ON((preempt_count() < 0));
2589 preempt_count() += val
;
2591 * Spinlock count overflowing soon?
2593 BUG_ON((preempt_count() & PREEMPT_MASK
) >= PREEMPT_MASK
-10);
2595 EXPORT_SYMBOL(add_preempt_count
);
2597 void fastcall
sub_preempt_count(int val
)
2602 BUG_ON(val
> preempt_count());
2604 * Is the spinlock portion underflowing?
2606 BUG_ON((val
< PREEMPT_MASK
) && !(preempt_count() & PREEMPT_MASK
));
2607 preempt_count() -= val
;
2609 EXPORT_SYMBOL(sub_preempt_count
);
2614 * schedule() is the main scheduler function.
2616 asmlinkage
void __sched
schedule(void)
2619 task_t
*prev
, *next
;
2621 prio_array_t
*array
;
2622 struct list_head
*queue
;
2623 unsigned long long now
;
2624 unsigned long run_time
;
2628 * Test if we are atomic. Since do_exit() needs to call into
2629 * schedule() atomically, we ignore that path for now.
2630 * Otherwise, whine if we are scheduling when we should not be.
2632 if (likely(!current
->exit_state
)) {
2633 if (unlikely(in_atomic())) {
2634 printk(KERN_ERR
"scheduling while atomic: "
2636 current
->comm
, preempt_count(), current
->pid
);
2640 profile_hit(SCHED_PROFILING
, __builtin_return_address(0));
2645 release_kernel_lock(prev
);
2646 need_resched_nonpreemptible
:
2650 * The idle thread is not allowed to schedule!
2651 * Remove this check after it has been exercised a bit.
2653 if (unlikely(prev
== rq
->idle
) && prev
->state
!= TASK_RUNNING
) {
2654 printk(KERN_ERR
"bad: scheduling from the idle thread!\n");
2658 schedstat_inc(rq
, sched_cnt
);
2659 now
= sched_clock();
2660 if (likely((long long)(now
- prev
->timestamp
) < NS_MAX_SLEEP_AVG
)) {
2661 run_time
= now
- prev
->timestamp
;
2662 if (unlikely((long long)(now
- prev
->timestamp
) < 0))
2665 run_time
= NS_MAX_SLEEP_AVG
;
2668 * Tasks charged proportionately less run_time at high sleep_avg to
2669 * delay them losing their interactive status
2671 run_time
/= (CURRENT_BONUS(prev
) ? : 1);
2673 spin_lock_irq(&rq
->lock
);
2675 if (unlikely(prev
->flags
& PF_DEAD
))
2676 prev
->state
= EXIT_DEAD
;
2678 switch_count
= &prev
->nivcsw
;
2679 if (prev
->state
&& !(preempt_count() & PREEMPT_ACTIVE
)) {
2680 switch_count
= &prev
->nvcsw
;
2681 if (unlikely((prev
->state
& TASK_INTERRUPTIBLE
) &&
2682 unlikely(signal_pending(prev
))))
2683 prev
->state
= TASK_RUNNING
;
2685 if (prev
->state
== TASK_UNINTERRUPTIBLE
)
2686 rq
->nr_uninterruptible
++;
2687 deactivate_task(prev
, rq
);
2691 cpu
= smp_processor_id();
2692 if (unlikely(!rq
->nr_running
)) {
2694 idle_balance(cpu
, rq
);
2695 if (!rq
->nr_running
) {
2697 rq
->expired_timestamp
= 0;
2698 wake_sleeping_dependent(cpu
, rq
);
2700 * wake_sleeping_dependent() might have released
2701 * the runqueue, so break out if we got new
2704 if (!rq
->nr_running
)
2708 if (dependent_sleeper(cpu
, rq
)) {
2713 * dependent_sleeper() releases and reacquires the runqueue
2714 * lock, hence go into the idle loop if the rq went
2717 if (unlikely(!rq
->nr_running
))
2722 if (unlikely(!array
->nr_active
)) {
2724 * Switch the active and expired arrays.
2726 schedstat_inc(rq
, sched_switch
);
2727 rq
->active
= rq
->expired
;
2728 rq
->expired
= array
;
2730 rq
->expired_timestamp
= 0;
2731 rq
->best_expired_prio
= MAX_PRIO
;
2734 idx
= sched_find_first_bit(array
->bitmap
);
2735 queue
= array
->queue
+ idx
;
2736 next
= list_entry(queue
->next
, task_t
, run_list
);
2738 if (!rt_task(next
) && next
->activated
> 0) {
2739 unsigned long long delta
= now
- next
->timestamp
;
2740 if (unlikely((long long)(now
- next
->timestamp
) < 0))
2743 if (next
->activated
== 1)
2744 delta
= delta
* (ON_RUNQUEUE_WEIGHT
* 128 / 100) / 128;
2746 array
= next
->array
;
2747 dequeue_task(next
, array
);
2748 recalc_task_prio(next
, next
->timestamp
+ delta
);
2749 enqueue_task(next
, array
);
2751 next
->activated
= 0;
2753 if (next
== rq
->idle
)
2754 schedstat_inc(rq
, sched_goidle
);
2756 clear_tsk_need_resched(prev
);
2757 rcu_qsctr_inc(task_cpu(prev
));
2759 update_cpu_clock(prev
, rq
, now
);
2761 prev
->sleep_avg
-= run_time
;
2762 if ((long)prev
->sleep_avg
<= 0)
2763 prev
->sleep_avg
= 0;
2764 prev
->timestamp
= prev
->last_ran
= now
;
2766 sched_info_switch(prev
, next
);
2767 if (likely(prev
!= next
)) {
2768 next
->timestamp
= now
;
2773 prepare_arch_switch(rq
, next
);
2774 prev
= context_switch(rq
, prev
, next
);
2777 finish_task_switch(prev
);
2779 spin_unlock_irq(&rq
->lock
);
2782 if (unlikely(reacquire_kernel_lock(prev
) < 0))
2783 goto need_resched_nonpreemptible
;
2784 preempt_enable_no_resched();
2785 if (unlikely(test_thread_flag(TIF_NEED_RESCHED
)))
2789 EXPORT_SYMBOL(schedule
);
2791 #ifdef CONFIG_PREEMPT
2793 * this is is the entry point to schedule() from in-kernel preemption
2794 * off of preempt_enable. Kernel preemptions off return from interrupt
2795 * occur there and call schedule directly.
2797 asmlinkage
void __sched
preempt_schedule(void)
2799 struct thread_info
*ti
= current_thread_info();
2800 #ifdef CONFIG_PREEMPT_BKL
2801 struct task_struct
*task
= current
;
2802 int saved_lock_depth
;
2805 * If there is a non-zero preempt_count or interrupts are disabled,
2806 * we do not want to preempt the current task. Just return..
2808 if (unlikely(ti
->preempt_count
|| irqs_disabled()))
2812 add_preempt_count(PREEMPT_ACTIVE
);
2814 * We keep the big kernel semaphore locked, but we
2815 * clear ->lock_depth so that schedule() doesnt
2816 * auto-release the semaphore:
2818 #ifdef CONFIG_PREEMPT_BKL
2819 saved_lock_depth
= task
->lock_depth
;
2820 task
->lock_depth
= -1;
2823 #ifdef CONFIG_PREEMPT_BKL
2824 task
->lock_depth
= saved_lock_depth
;
2826 sub_preempt_count(PREEMPT_ACTIVE
);
2828 /* we could miss a preemption opportunity between schedule and now */
2830 if (unlikely(test_thread_flag(TIF_NEED_RESCHED
)))
2834 EXPORT_SYMBOL(preempt_schedule
);
2837 * this is is the entry point to schedule() from kernel preemption
2838 * off of irq context.
2839 * Note, that this is called and return with irqs disabled. This will
2840 * protect us against recursive calling from irq.
2842 asmlinkage
void __sched
preempt_schedule_irq(void)
2844 struct thread_info
*ti
= current_thread_info();
2845 #ifdef CONFIG_PREEMPT_BKL
2846 struct task_struct
*task
= current
;
2847 int saved_lock_depth
;
2849 /* Catch callers which need to be fixed*/
2850 BUG_ON(ti
->preempt_count
|| !irqs_disabled());
2853 add_preempt_count(PREEMPT_ACTIVE
);
2855 * We keep the big kernel semaphore locked, but we
2856 * clear ->lock_depth so that schedule() doesnt
2857 * auto-release the semaphore:
2859 #ifdef CONFIG_PREEMPT_BKL
2860 saved_lock_depth
= task
->lock_depth
;
2861 task
->lock_depth
= -1;
2865 local_irq_disable();
2866 #ifdef CONFIG_PREEMPT_BKL
2867 task
->lock_depth
= saved_lock_depth
;
2869 sub_preempt_count(PREEMPT_ACTIVE
);
2871 /* we could miss a preemption opportunity between schedule and now */
2873 if (unlikely(test_thread_flag(TIF_NEED_RESCHED
)))
2877 #endif /* CONFIG_PREEMPT */
2879 int default_wake_function(wait_queue_t
*curr
, unsigned mode
, int sync
, void *key
)
2881 task_t
*p
= curr
->private;
2882 return try_to_wake_up(p
, mode
, sync
);
2885 EXPORT_SYMBOL(default_wake_function
);
2888 * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just
2889 * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve
2890 * number) then we wake all the non-exclusive tasks and one exclusive task.
2892 * There are circumstances in which we can try to wake a task which has already
2893 * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns
2894 * zero in this (rare) case, and we handle it by continuing to scan the queue.
2896 static void __wake_up_common(wait_queue_head_t
*q
, unsigned int mode
,
2897 int nr_exclusive
, int sync
, void *key
)
2899 struct list_head
*tmp
, *next
;
2901 list_for_each_safe(tmp
, next
, &q
->task_list
) {
2904 curr
= list_entry(tmp
, wait_queue_t
, task_list
);
2905 flags
= curr
->flags
;
2906 if (curr
->func(curr
, mode
, sync
, key
) &&
2907 (flags
& WQ_FLAG_EXCLUSIVE
) &&
2914 * __wake_up - wake up threads blocked on a waitqueue.
2916 * @mode: which threads
2917 * @nr_exclusive: how many wake-one or wake-many threads to wake up
2918 * @key: is directly passed to the wakeup function
2920 void fastcall
__wake_up(wait_queue_head_t
*q
, unsigned int mode
,
2921 int nr_exclusive
, void *key
)
2923 unsigned long flags
;
2925 spin_lock_irqsave(&q
->lock
, flags
);
2926 __wake_up_common(q
, mode
, nr_exclusive
, 0, key
);
2927 spin_unlock_irqrestore(&q
->lock
, flags
);
2930 EXPORT_SYMBOL(__wake_up
);
2933 * Same as __wake_up but called with the spinlock in wait_queue_head_t held.
2935 void fastcall
__wake_up_locked(wait_queue_head_t
*q
, unsigned int mode
)
2937 __wake_up_common(q
, mode
, 1, 0, NULL
);
2941 * __wake_up_sync - wake up threads blocked on a waitqueue.
2943 * @mode: which threads
2944 * @nr_exclusive: how many wake-one or wake-many threads to wake up
2946 * The sync wakeup differs that the waker knows that it will schedule
2947 * away soon, so while the target thread will be woken up, it will not
2948 * be migrated to another CPU - ie. the two threads are 'synchronized'
2949 * with each other. This can prevent needless bouncing between CPUs.
2951 * On UP it can prevent extra preemption.
2953 void fastcall
__wake_up_sync(wait_queue_head_t
*q
, unsigned int mode
, int nr_exclusive
)
2955 unsigned long flags
;
2961 if (unlikely(!nr_exclusive
))
2964 spin_lock_irqsave(&q
->lock
, flags
);
2965 __wake_up_common(q
, mode
, nr_exclusive
, sync
, NULL
);
2966 spin_unlock_irqrestore(&q
->lock
, flags
);
2968 EXPORT_SYMBOL_GPL(__wake_up_sync
); /* For internal use only */
2970 void fastcall
complete(struct completion
*x
)
2972 unsigned long flags
;
2974 spin_lock_irqsave(&x
->wait
.lock
, flags
);
2976 __wake_up_common(&x
->wait
, TASK_UNINTERRUPTIBLE
| TASK_INTERRUPTIBLE
,
2978 spin_unlock_irqrestore(&x
->wait
.lock
, flags
);
2980 EXPORT_SYMBOL(complete
);
2982 void fastcall
complete_all(struct completion
*x
)
2984 unsigned long flags
;
2986 spin_lock_irqsave(&x
->wait
.lock
, flags
);
2987 x
->done
+= UINT_MAX
/2;
2988 __wake_up_common(&x
->wait
, TASK_UNINTERRUPTIBLE
| TASK_INTERRUPTIBLE
,
2990 spin_unlock_irqrestore(&x
->wait
.lock
, flags
);
2992 EXPORT_SYMBOL(complete_all
);
2994 void fastcall __sched
wait_for_completion(struct completion
*x
)
2997 spin_lock_irq(&x
->wait
.lock
);
2999 DECLARE_WAITQUEUE(wait
, current
);
3001 wait
.flags
|= WQ_FLAG_EXCLUSIVE
;
3002 __add_wait_queue_tail(&x
->wait
, &wait
);
3004 __set_current_state(TASK_UNINTERRUPTIBLE
);
3005 spin_unlock_irq(&x
->wait
.lock
);
3007 spin_lock_irq(&x
->wait
.lock
);
3009 __remove_wait_queue(&x
->wait
, &wait
);
3012 spin_unlock_irq(&x
->wait
.lock
);
3014 EXPORT_SYMBOL(wait_for_completion
);
3016 unsigned long fastcall __sched
3017 wait_for_completion_timeout(struct completion
*x
, unsigned long timeout
)
3021 spin_lock_irq(&x
->wait
.lock
);
3023 DECLARE_WAITQUEUE(wait
, current
);
3025 wait
.flags
|= WQ_FLAG_EXCLUSIVE
;
3026 __add_wait_queue_tail(&x
->wait
, &wait
);
3028 __set_current_state(TASK_UNINTERRUPTIBLE
);
3029 spin_unlock_irq(&x
->wait
.lock
);
3030 timeout
= schedule_timeout(timeout
);
3031 spin_lock_irq(&x
->wait
.lock
);
3033 __remove_wait_queue(&x
->wait
, &wait
);
3037 __remove_wait_queue(&x
->wait
, &wait
);
3041 spin_unlock_irq(&x
->wait
.lock
);
3044 EXPORT_SYMBOL(wait_for_completion_timeout
);
3046 int fastcall __sched
wait_for_completion_interruptible(struct completion
*x
)
3052 spin_lock_irq(&x
->wait
.lock
);
3054 DECLARE_WAITQUEUE(wait
, current
);
3056 wait
.flags
|= WQ_FLAG_EXCLUSIVE
;
3057 __add_wait_queue_tail(&x
->wait
, &wait
);
3059 if (signal_pending(current
)) {
3061 __remove_wait_queue(&x
->wait
, &wait
);
3064 __set_current_state(TASK_INTERRUPTIBLE
);
3065 spin_unlock_irq(&x
->wait
.lock
);
3067 spin_lock_irq(&x
->wait
.lock
);
3069 __remove_wait_queue(&x
->wait
, &wait
);
3073 spin_unlock_irq(&x
->wait
.lock
);
3077 EXPORT_SYMBOL(wait_for_completion_interruptible
);
3079 unsigned long fastcall __sched
3080 wait_for_completion_interruptible_timeout(struct completion
*x
,
3081 unsigned long timeout
)
3085 spin_lock_irq(&x
->wait
.lock
);
3087 DECLARE_WAITQUEUE(wait
, current
);
3089 wait
.flags
|= WQ_FLAG_EXCLUSIVE
;
3090 __add_wait_queue_tail(&x
->wait
, &wait
);
3092 if (signal_pending(current
)) {
3093 timeout
= -ERESTARTSYS
;
3094 __remove_wait_queue(&x
->wait
, &wait
);
3097 __set_current_state(TASK_INTERRUPTIBLE
);
3098 spin_unlock_irq(&x
->wait
.lock
);
3099 timeout
= schedule_timeout(timeout
);
3100 spin_lock_irq(&x
->wait
.lock
);
3102 __remove_wait_queue(&x
->wait
, &wait
);
3106 __remove_wait_queue(&x
->wait
, &wait
);
3110 spin_unlock_irq(&x
->wait
.lock
);
3113 EXPORT_SYMBOL(wait_for_completion_interruptible_timeout
);
3116 #define SLEEP_ON_VAR \
3117 unsigned long flags; \
3118 wait_queue_t wait; \
3119 init_waitqueue_entry(&wait, current);
3121 #define SLEEP_ON_HEAD \
3122 spin_lock_irqsave(&q->lock,flags); \
3123 __add_wait_queue(q, &wait); \
3124 spin_unlock(&q->lock);
3126 #define SLEEP_ON_TAIL \
3127 spin_lock_irq(&q->lock); \
3128 __remove_wait_queue(q, &wait); \
3129 spin_unlock_irqrestore(&q->lock, flags);
3131 void fastcall __sched
interruptible_sleep_on(wait_queue_head_t
*q
)
3135 current
->state
= TASK_INTERRUPTIBLE
;
3142 EXPORT_SYMBOL(interruptible_sleep_on
);
3144 long fastcall __sched
interruptible_sleep_on_timeout(wait_queue_head_t
*q
, long timeout
)
3148 current
->state
= TASK_INTERRUPTIBLE
;
3151 timeout
= schedule_timeout(timeout
);
3157 EXPORT_SYMBOL(interruptible_sleep_on_timeout
);
3159 void fastcall __sched
sleep_on(wait_queue_head_t
*q
)
3163 current
->state
= TASK_UNINTERRUPTIBLE
;
3170 EXPORT_SYMBOL(sleep_on
);
3172 long fastcall __sched
sleep_on_timeout(wait_queue_head_t
*q
, long timeout
)
3176 current
->state
= TASK_UNINTERRUPTIBLE
;
3179 timeout
= schedule_timeout(timeout
);
3185 EXPORT_SYMBOL(sleep_on_timeout
);
3187 void set_user_nice(task_t
*p
, long nice
)
3189 unsigned long flags
;
3190 prio_array_t
*array
;
3192 int old_prio
, new_prio
, delta
;
3194 if (TASK_NICE(p
) == nice
|| nice
< -20 || nice
> 19)
3197 * We have to be careful, if called from sys_setpriority(),
3198 * the task might be in the middle of scheduling on another CPU.
3200 rq
= task_rq_lock(p
, &flags
);
3202 * The RT priorities are set via sched_setscheduler(), but we still
3203 * allow the 'normal' nice value to be set - but as expected
3204 * it wont have any effect on scheduling until the task is
3208 p
->static_prio
= NICE_TO_PRIO(nice
);
3213 dequeue_task(p
, array
);
3216 new_prio
= NICE_TO_PRIO(nice
);
3217 delta
= new_prio
- old_prio
;
3218 p
->static_prio
= NICE_TO_PRIO(nice
);
3222 enqueue_task(p
, array
);
3224 * If the task increased its priority or is running and
3225 * lowered its priority, then reschedule its CPU:
3227 if (delta
< 0 || (delta
> 0 && task_running(rq
, p
)))
3228 resched_task(rq
->curr
);
3231 task_rq_unlock(rq
, &flags
);
3234 EXPORT_SYMBOL(set_user_nice
);
3237 * can_nice - check if a task can reduce its nice value
3241 int can_nice(const task_t
*p
, const int nice
)
3243 /* convert nice value [19,-20] to rlimit style value [0,39] */
3244 int nice_rlim
= 19 - nice
;
3245 return (nice_rlim
<= p
->signal
->rlim
[RLIMIT_NICE
].rlim_cur
||
3246 capable(CAP_SYS_NICE
));
3249 #ifdef __ARCH_WANT_SYS_NICE
3252 * sys_nice - change the priority of the current process.
3253 * @increment: priority increment
3255 * sys_setpriority is a more generic, but much slower function that
3256 * does similar things.
3258 asmlinkage
long sys_nice(int increment
)
3264 * Setpriority might change our priority at the same moment.
3265 * We don't have to worry. Conceptually one call occurs first
3266 * and we have a single winner.
3268 if (increment
< -40)
3273 nice
= PRIO_TO_NICE(current
->static_prio
) + increment
;
3279 if (increment
< 0 && !can_nice(current
, nice
))
3282 retval
= security_task_setnice(current
, nice
);
3286 set_user_nice(current
, nice
);
3293 * task_prio - return the priority value of a given task.
3294 * @p: the task in question.
3296 * This is the priority value as seen by users in /proc.
3297 * RT tasks are offset by -200. Normal tasks are centered
3298 * around 0, value goes from -16 to +15.
3300 int task_prio(const task_t
*p
)
3302 return p
->prio
- MAX_RT_PRIO
;
3306 * task_nice - return the nice value of a given task.
3307 * @p: the task in question.
3309 int task_nice(const task_t
*p
)
3311 return TASK_NICE(p
);
3315 * The only users of task_nice are binfmt_elf and binfmt_elf32.
3316 * binfmt_elf is no longer modular, but binfmt_elf32 still is.
3317 * Therefore, task_nice is needed if there is a compat_mode.
3319 #ifdef CONFIG_COMPAT
3320 EXPORT_SYMBOL_GPL(task_nice
);
3324 * idle_cpu - is a given cpu idle currently?
3325 * @cpu: the processor in question.
3327 int idle_cpu(int cpu
)
3329 return cpu_curr(cpu
) == cpu_rq(cpu
)->idle
;
3332 EXPORT_SYMBOL_GPL(idle_cpu
);
3335 * idle_task - return the idle task for a given cpu.
3336 * @cpu: the processor in question.
3338 task_t
*idle_task(int cpu
)
3340 return cpu_rq(cpu
)->idle
;
3344 * find_process_by_pid - find a process with a matching PID value.
3345 * @pid: the pid in question.
3347 static inline task_t
*find_process_by_pid(pid_t pid
)
3349 return pid
? find_task_by_pid(pid
) : current
;
3352 /* Actually do priority change: must hold rq lock. */
3353 static void __setscheduler(struct task_struct
*p
, int policy
, int prio
)
3357 p
->rt_priority
= prio
;
3358 if (policy
!= SCHED_NORMAL
)
3359 p
->prio
= MAX_USER_RT_PRIO
-1 - p
->rt_priority
;
3361 p
->prio
= p
->static_prio
;
3365 * sched_setscheduler - change the scheduling policy and/or RT priority of
3367 * @p: the task in question.
3368 * @policy: new policy.
3369 * @param: structure containing the new RT priority.
3371 int sched_setscheduler(struct task_struct
*p
, int policy
, struct sched_param
*param
)
3374 int oldprio
, oldpolicy
= -1;
3375 prio_array_t
*array
;
3376 unsigned long flags
;
3380 /* double check policy once rq lock held */
3382 policy
= oldpolicy
= p
->policy
;
3383 else if (policy
!= SCHED_FIFO
&& policy
!= SCHED_RR
&&
3384 policy
!= SCHED_NORMAL
)
3387 * Valid priorities for SCHED_FIFO and SCHED_RR are
3388 * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL is 0.
3390 if (param
->sched_priority
< 0 ||
3391 param
->sched_priority
> MAX_USER_RT_PRIO
-1)
3393 if ((policy
== SCHED_NORMAL
) != (param
->sched_priority
== 0))
3396 if ((policy
== SCHED_FIFO
|| policy
== SCHED_RR
) &&
3397 param
->sched_priority
> p
->signal
->rlim
[RLIMIT_RTPRIO
].rlim_cur
&&
3398 !capable(CAP_SYS_NICE
))
3400 if ((current
->euid
!= p
->euid
) && (current
->euid
!= p
->uid
) &&
3401 !capable(CAP_SYS_NICE
))
3404 retval
= security_task_setscheduler(p
, policy
, param
);
3408 * To be able to change p->policy safely, the apropriate
3409 * runqueue lock must be held.
3411 rq
= task_rq_lock(p
, &flags
);
3412 /* recheck policy now with rq lock held */
3413 if (unlikely(oldpolicy
!= -1 && oldpolicy
!= p
->policy
)) {
3414 policy
= oldpolicy
= -1;
3415 task_rq_unlock(rq
, &flags
);
3420 deactivate_task(p
, rq
);
3422 __setscheduler(p
, policy
, param
->sched_priority
);
3424 __activate_task(p
, rq
);
3426 * Reschedule if we are currently running on this runqueue and
3427 * our priority decreased, or if we are not currently running on
3428 * this runqueue and our priority is higher than the current's
3430 if (task_running(rq
, p
)) {
3431 if (p
->prio
> oldprio
)
3432 resched_task(rq
->curr
);
3433 } else if (TASK_PREEMPTS_CURR(p
, rq
))
3434 resched_task(rq
->curr
);
3436 task_rq_unlock(rq
, &flags
);
3439 EXPORT_SYMBOL_GPL(sched_setscheduler
);
3441 static int do_sched_setscheduler(pid_t pid
, int policy
, struct sched_param __user
*param
)
3444 struct sched_param lparam
;
3445 struct task_struct
*p
;
3447 if (!param
|| pid
< 0)
3449 if (copy_from_user(&lparam
, param
, sizeof(struct sched_param
)))
3451 read_lock_irq(&tasklist_lock
);
3452 p
= find_process_by_pid(pid
);
3454 read_unlock_irq(&tasklist_lock
);
3457 retval
= sched_setscheduler(p
, policy
, &lparam
);
3458 read_unlock_irq(&tasklist_lock
);
3463 * sys_sched_setscheduler - set/change the scheduler policy and RT priority
3464 * @pid: the pid in question.
3465 * @policy: new policy.
3466 * @param: structure containing the new RT priority.
3468 asmlinkage
long sys_sched_setscheduler(pid_t pid
, int policy
,
3469 struct sched_param __user
*param
)
3471 return do_sched_setscheduler(pid
, policy
, param
);
3475 * sys_sched_setparam - set/change the RT priority of a thread
3476 * @pid: the pid in question.
3477 * @param: structure containing the new RT priority.
3479 asmlinkage
long sys_sched_setparam(pid_t pid
, struct sched_param __user
*param
)
3481 return do_sched_setscheduler(pid
, -1, param
);
3485 * sys_sched_getscheduler - get the policy (scheduling class) of a thread
3486 * @pid: the pid in question.
3488 asmlinkage
long sys_sched_getscheduler(pid_t pid
)
3490 int retval
= -EINVAL
;
3497 read_lock(&tasklist_lock
);
3498 p
= find_process_by_pid(pid
);
3500 retval
= security_task_getscheduler(p
);
3504 read_unlock(&tasklist_lock
);
3511 * sys_sched_getscheduler - get the RT priority of a thread
3512 * @pid: the pid in question.
3513 * @param: structure containing the RT priority.
3515 asmlinkage
long sys_sched_getparam(pid_t pid
, struct sched_param __user
*param
)
3517 struct sched_param lp
;
3518 int retval
= -EINVAL
;
3521 if (!param
|| pid
< 0)
3524 read_lock(&tasklist_lock
);
3525 p
= find_process_by_pid(pid
);
3530 retval
= security_task_getscheduler(p
);
3534 lp
.sched_priority
= p
->rt_priority
;
3535 read_unlock(&tasklist_lock
);
3538 * This one might sleep, we cannot do it with a spinlock held ...
3540 retval
= copy_to_user(param
, &lp
, sizeof(*param
)) ? -EFAULT
: 0;
3546 read_unlock(&tasklist_lock
);
3550 long sched_setaffinity(pid_t pid
, cpumask_t new_mask
)
3554 cpumask_t cpus_allowed
;
3557 read_lock(&tasklist_lock
);
3559 p
= find_process_by_pid(pid
);
3561 read_unlock(&tasklist_lock
);
3562 unlock_cpu_hotplug();
3567 * It is not safe to call set_cpus_allowed with the
3568 * tasklist_lock held. We will bump the task_struct's
3569 * usage count and then drop tasklist_lock.
3572 read_unlock(&tasklist_lock
);
3575 if ((current
->euid
!= p
->euid
) && (current
->euid
!= p
->uid
) &&
3576 !capable(CAP_SYS_NICE
))
3579 cpus_allowed
= cpuset_cpus_allowed(p
);
3580 cpus_and(new_mask
, new_mask
, cpus_allowed
);
3581 retval
= set_cpus_allowed(p
, new_mask
);
3585 unlock_cpu_hotplug();
3589 static int get_user_cpu_mask(unsigned long __user
*user_mask_ptr
, unsigned len
,
3590 cpumask_t
*new_mask
)
3592 if (len
< sizeof(cpumask_t
)) {
3593 memset(new_mask
, 0, sizeof(cpumask_t
));
3594 } else if (len
> sizeof(cpumask_t
)) {
3595 len
= sizeof(cpumask_t
);
3597 return copy_from_user(new_mask
, user_mask_ptr
, len
) ? -EFAULT
: 0;
3601 * sys_sched_setaffinity - set the cpu affinity of a process
3602 * @pid: pid of the process
3603 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
3604 * @user_mask_ptr: user-space pointer to the new cpu mask
3606 asmlinkage
long sys_sched_setaffinity(pid_t pid
, unsigned int len
,
3607 unsigned long __user
*user_mask_ptr
)
3612 retval
= get_user_cpu_mask(user_mask_ptr
, len
, &new_mask
);
3616 return sched_setaffinity(pid
, new_mask
);
3620 * Represents all cpu's present in the system
3621 * In systems capable of hotplug, this map could dynamically grow
3622 * as new cpu's are detected in the system via any platform specific
3623 * method, such as ACPI for e.g.
3626 cpumask_t cpu_present_map
;
3627 EXPORT_SYMBOL(cpu_present_map
);
3630 cpumask_t cpu_online_map
= CPU_MASK_ALL
;
3631 cpumask_t cpu_possible_map
= CPU_MASK_ALL
;
3634 long sched_getaffinity(pid_t pid
, cpumask_t
*mask
)
3640 read_lock(&tasklist_lock
);
3643 p
= find_process_by_pid(pid
);
3648 cpus_and(*mask
, p
->cpus_allowed
, cpu_possible_map
);
3651 read_unlock(&tasklist_lock
);
3652 unlock_cpu_hotplug();
3660 * sys_sched_getaffinity - get the cpu affinity of a process
3661 * @pid: pid of the process
3662 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
3663 * @user_mask_ptr: user-space pointer to hold the current cpu mask
3665 asmlinkage
long sys_sched_getaffinity(pid_t pid
, unsigned int len
,
3666 unsigned long __user
*user_mask_ptr
)
3671 if (len
< sizeof(cpumask_t
))
3674 ret
= sched_getaffinity(pid
, &mask
);
3678 if (copy_to_user(user_mask_ptr
, &mask
, sizeof(cpumask_t
)))
3681 return sizeof(cpumask_t
);
3685 * sys_sched_yield - yield the current processor to other threads.
3687 * this function yields the current CPU by moving the calling thread
3688 * to the expired array. If there are no other threads running on this
3689 * CPU then this function will return.
3691 asmlinkage
long sys_sched_yield(void)
3693 runqueue_t
*rq
= this_rq_lock();
3694 prio_array_t
*array
= current
->array
;
3695 prio_array_t
*target
= rq
->expired
;
3697 schedstat_inc(rq
, yld_cnt
);
3699 * We implement yielding by moving the task into the expired
3702 * (special rule: RT tasks will just roundrobin in the active
3705 if (rt_task(current
))
3706 target
= rq
->active
;
3708 if (current
->array
->nr_active
== 1) {
3709 schedstat_inc(rq
, yld_act_empty
);
3710 if (!rq
->expired
->nr_active
)
3711 schedstat_inc(rq
, yld_both_empty
);
3712 } else if (!rq
->expired
->nr_active
)
3713 schedstat_inc(rq
, yld_exp_empty
);
3715 if (array
!= target
) {
3716 dequeue_task(current
, array
);
3717 enqueue_task(current
, target
);
3720 * requeue_task is cheaper so perform that if possible.
3722 requeue_task(current
, array
);
3725 * Since we are going to call schedule() anyway, there's
3726 * no need to preempt or enable interrupts:
3728 __release(rq
->lock
);
3729 _raw_spin_unlock(&rq
->lock
);
3730 preempt_enable_no_resched();
3737 static inline void __cond_resched(void)
3740 add_preempt_count(PREEMPT_ACTIVE
);
3742 sub_preempt_count(PREEMPT_ACTIVE
);
3743 } while (need_resched());
3746 int __sched
cond_resched(void)
3748 if (need_resched()) {
3755 EXPORT_SYMBOL(cond_resched
);
3758 * cond_resched_lock() - if a reschedule is pending, drop the given lock,
3759 * call schedule, and on return reacquire the lock.
3761 * This works OK both with and without CONFIG_PREEMPT. We do strange low-level
3762 * operations here to prevent schedule() from being called twice (once via
3763 * spin_unlock(), once by hand).
3765 int cond_resched_lock(spinlock_t
* lock
)
3769 if (need_lockbreak(lock
)) {
3775 if (need_resched()) {
3776 _raw_spin_unlock(lock
);
3777 preempt_enable_no_resched();
3785 EXPORT_SYMBOL(cond_resched_lock
);
3787 int __sched
cond_resched_softirq(void)
3789 BUG_ON(!in_softirq());
3791 if (need_resched()) {
3792 __local_bh_enable();
3800 EXPORT_SYMBOL(cond_resched_softirq
);
3804 * yield - yield the current processor to other threads.
3806 * this is a shortcut for kernel-space yielding - it marks the
3807 * thread runnable and calls sys_sched_yield().
3809 void __sched
yield(void)
3811 set_current_state(TASK_RUNNING
);
3815 EXPORT_SYMBOL(yield
);
3818 * This task is about to go to sleep on IO. Increment rq->nr_iowait so
3819 * that process accounting knows that this is a task in IO wait state.
3821 * But don't do that if it is a deliberate, throttling IO wait (this task
3822 * has set its backing_dev_info: the queue against which it should throttle)
3824 void __sched
io_schedule(void)
3826 struct runqueue
*rq
= &per_cpu(runqueues
, raw_smp_processor_id());
3828 atomic_inc(&rq
->nr_iowait
);
3830 atomic_dec(&rq
->nr_iowait
);
3833 EXPORT_SYMBOL(io_schedule
);
3835 long __sched
io_schedule_timeout(long timeout
)
3837 struct runqueue
*rq
= &per_cpu(runqueues
, raw_smp_processor_id());
3840 atomic_inc(&rq
->nr_iowait
);
3841 ret
= schedule_timeout(timeout
);
3842 atomic_dec(&rq
->nr_iowait
);
3847 * sys_sched_get_priority_max - return maximum RT priority.
3848 * @policy: scheduling class.
3850 * this syscall returns the maximum rt_priority that can be used
3851 * by a given scheduling class.
3853 asmlinkage
long sys_sched_get_priority_max(int policy
)
3860 ret
= MAX_USER_RT_PRIO
-1;
3870 * sys_sched_get_priority_min - return minimum RT priority.
3871 * @policy: scheduling class.
3873 * this syscall returns the minimum rt_priority that can be used
3874 * by a given scheduling class.
3876 asmlinkage
long sys_sched_get_priority_min(int policy
)
3892 * sys_sched_rr_get_interval - return the default timeslice of a process.
3893 * @pid: pid of the process.
3894 * @interval: userspace pointer to the timeslice value.
3896 * this syscall writes the default timeslice value of a given process
3897 * into the user-space timespec buffer. A value of '0' means infinity.
3900 long sys_sched_rr_get_interval(pid_t pid
, struct timespec __user
*interval
)
3902 int retval
= -EINVAL
;
3910 read_lock(&tasklist_lock
);
3911 p
= find_process_by_pid(pid
);
3915 retval
= security_task_getscheduler(p
);
3919 jiffies_to_timespec(p
->policy
& SCHED_FIFO
?
3920 0 : task_timeslice(p
), &t
);
3921 read_unlock(&tasklist_lock
);
3922 retval
= copy_to_user(interval
, &t
, sizeof(t
)) ? -EFAULT
: 0;
3926 read_unlock(&tasklist_lock
);
3930 static inline struct task_struct
*eldest_child(struct task_struct
*p
)
3932 if (list_empty(&p
->children
)) return NULL
;
3933 return list_entry(p
->children
.next
,struct task_struct
,sibling
);
3936 static inline struct task_struct
*older_sibling(struct task_struct
*p
)
3938 if (p
->sibling
.prev
==&p
->parent
->children
) return NULL
;
3939 return list_entry(p
->sibling
.prev
,struct task_struct
,sibling
);
3942 static inline struct task_struct
*younger_sibling(struct task_struct
*p
)
3944 if (p
->sibling
.next
==&p
->parent
->children
) return NULL
;
3945 return list_entry(p
->sibling
.next
,struct task_struct
,sibling
);
3948 static void show_task(task_t
* p
)
3952 unsigned long free
= 0;
3953 static const char *stat_nam
[] = { "R", "S", "D", "T", "t", "Z", "X" };
3955 printk("%-13.13s ", p
->comm
);
3956 state
= p
->state
? __ffs(p
->state
) + 1 : 0;
3957 if (state
< ARRAY_SIZE(stat_nam
))
3958 printk(stat_nam
[state
]);
3961 #if (BITS_PER_LONG == 32)
3962 if (state
== TASK_RUNNING
)
3963 printk(" running ");
3965 printk(" %08lX ", thread_saved_pc(p
));
3967 if (state
== TASK_RUNNING
)
3968 printk(" running task ");
3970 printk(" %016lx ", thread_saved_pc(p
));
3972 #ifdef CONFIG_DEBUG_STACK_USAGE
3974 unsigned long * n
= (unsigned long *) (p
->thread_info
+1);
3977 free
= (unsigned long) n
- (unsigned long)(p
->thread_info
+1);
3980 printk("%5lu %5d %6d ", free
, p
->pid
, p
->parent
->pid
);
3981 if ((relative
= eldest_child(p
)))
3982 printk("%5d ", relative
->pid
);
3985 if ((relative
= younger_sibling(p
)))
3986 printk("%7d", relative
->pid
);
3989 if ((relative
= older_sibling(p
)))
3990 printk(" %5d", relative
->pid
);
3994 printk(" (L-TLB)\n");
3996 printk(" (NOTLB)\n");
3998 if (state
!= TASK_RUNNING
)
3999 show_stack(p
, NULL
);
4002 void show_state(void)
4006 #if (BITS_PER_LONG == 32)
4009 printk(" task PC pid father child younger older\n");
4013 printk(" task PC pid father child younger older\n");
4015 read_lock(&tasklist_lock
);
4016 do_each_thread(g
, p
) {
4018 * reset the NMI-timeout, listing all files on a slow
4019 * console might take alot of time:
4021 touch_nmi_watchdog();
4023 } while_each_thread(g
, p
);
4025 read_unlock(&tasklist_lock
);
4028 void __devinit
init_idle(task_t
*idle
, int cpu
)
4030 runqueue_t
*rq
= cpu_rq(cpu
);
4031 unsigned long flags
;
4033 idle
->sleep_avg
= 0;
4035 idle
->prio
= MAX_PRIO
;
4036 idle
->state
= TASK_RUNNING
;
4037 idle
->cpus_allowed
= cpumask_of_cpu(cpu
);
4038 set_task_cpu(idle
, cpu
);
4040 spin_lock_irqsave(&rq
->lock
, flags
);
4041 rq
->curr
= rq
->idle
= idle
;
4042 set_tsk_need_resched(idle
);
4043 spin_unlock_irqrestore(&rq
->lock
, flags
);
4045 /* Set the preempt count _outside_ the spinlocks! */
4046 #if defined(CONFIG_PREEMPT) && !defined(CONFIG_PREEMPT_BKL)
4047 idle
->thread_info
->preempt_count
= (idle
->lock_depth
>= 0);
4049 idle
->thread_info
->preempt_count
= 0;
4054 * In a system that switches off the HZ timer nohz_cpu_mask
4055 * indicates which cpus entered this state. This is used
4056 * in the rcu update to wait only for active cpus. For system
4057 * which do not switch off the HZ timer nohz_cpu_mask should
4058 * always be CPU_MASK_NONE.
4060 cpumask_t nohz_cpu_mask
= CPU_MASK_NONE
;
4064 * This is how migration works:
4066 * 1) we queue a migration_req_t structure in the source CPU's
4067 * runqueue and wake up that CPU's migration thread.
4068 * 2) we down() the locked semaphore => thread blocks.
4069 * 3) migration thread wakes up (implicitly it forces the migrated
4070 * thread off the CPU)
4071 * 4) it gets the migration request and checks whether the migrated
4072 * task is still in the wrong runqueue.
4073 * 5) if it's in the wrong runqueue then the migration thread removes
4074 * it and puts it into the right queue.
4075 * 6) migration thread up()s the semaphore.
4076 * 7) we wake up and the migration is done.
4080 * Change a given task's CPU affinity. Migrate the thread to a
4081 * proper CPU and schedule it away if the CPU it's executing on
4082 * is removed from the allowed bitmask.
4084 * NOTE: the caller must have a valid reference to the task, the
4085 * task must not exit() & deallocate itself prematurely. The
4086 * call is not atomic; no spinlocks may be held.
4088 int set_cpus_allowed(task_t
*p
, cpumask_t new_mask
)
4090 unsigned long flags
;
4092 migration_req_t req
;
4095 rq
= task_rq_lock(p
, &flags
);
4096 if (!cpus_intersects(new_mask
, cpu_online_map
)) {
4101 p
->cpus_allowed
= new_mask
;
4102 /* Can the task run on the task's current CPU? If so, we're done */
4103 if (cpu_isset(task_cpu(p
), new_mask
))
4106 if (migrate_task(p
, any_online_cpu(new_mask
), &req
)) {
4107 /* Need help from migration thread: drop lock and wait. */
4108 task_rq_unlock(rq
, &flags
);
4109 wake_up_process(rq
->migration_thread
);
4110 wait_for_completion(&req
.done
);
4111 tlb_migrate_finish(p
->mm
);
4115 task_rq_unlock(rq
, &flags
);
4119 EXPORT_SYMBOL_GPL(set_cpus_allowed
);
4122 * Move (not current) task off this cpu, onto dest cpu. We're doing
4123 * this because either it can't run here any more (set_cpus_allowed()
4124 * away from this CPU, or CPU going down), or because we're
4125 * attempting to rebalance this task on exec (sched_exec).
4127 * So we race with normal scheduler movements, but that's OK, as long
4128 * as the task is no longer on this CPU.
4130 static void __migrate_task(struct task_struct
*p
, int src_cpu
, int dest_cpu
)
4132 runqueue_t
*rq_dest
, *rq_src
;
4134 if (unlikely(cpu_is_offline(dest_cpu
)))
4137 rq_src
= cpu_rq(src_cpu
);
4138 rq_dest
= cpu_rq(dest_cpu
);
4140 double_rq_lock(rq_src
, rq_dest
);
4141 /* Already moved. */
4142 if (task_cpu(p
) != src_cpu
)
4144 /* Affinity changed (again). */
4145 if (!cpu_isset(dest_cpu
, p
->cpus_allowed
))
4148 set_task_cpu(p
, dest_cpu
);
4151 * Sync timestamp with rq_dest's before activating.
4152 * The same thing could be achieved by doing this step
4153 * afterwards, and pretending it was a local activate.
4154 * This way is cleaner and logically correct.
4156 p
->timestamp
= p
->timestamp
- rq_src
->timestamp_last_tick
4157 + rq_dest
->timestamp_last_tick
;
4158 deactivate_task(p
, rq_src
);
4159 activate_task(p
, rq_dest
, 0);
4160 if (TASK_PREEMPTS_CURR(p
, rq_dest
))
4161 resched_task(rq_dest
->curr
);
4165 double_rq_unlock(rq_src
, rq_dest
);
4169 * migration_thread - this is a highprio system thread that performs
4170 * thread migration by bumping thread off CPU then 'pushing' onto
4173 static int migration_thread(void * data
)
4176 int cpu
= (long)data
;
4179 BUG_ON(rq
->migration_thread
!= current
);
4181 set_current_state(TASK_INTERRUPTIBLE
);
4182 while (!kthread_should_stop()) {
4183 struct list_head
*head
;
4184 migration_req_t
*req
;
4186 if (current
->flags
& PF_FREEZE
)
4187 refrigerator(PF_FREEZE
);
4189 spin_lock_irq(&rq
->lock
);
4191 if (cpu_is_offline(cpu
)) {
4192 spin_unlock_irq(&rq
->lock
);
4196 if (rq
->active_balance
) {
4197 active_load_balance(rq
, cpu
);
4198 rq
->active_balance
= 0;
4201 head
= &rq
->migration_queue
;
4203 if (list_empty(head
)) {
4204 spin_unlock_irq(&rq
->lock
);
4206 set_current_state(TASK_INTERRUPTIBLE
);
4209 req
= list_entry(head
->next
, migration_req_t
, list
);
4210 list_del_init(head
->next
);
4212 if (req
->type
== REQ_MOVE_TASK
) {
4213 spin_unlock(&rq
->lock
);
4214 __migrate_task(req
->task
, cpu
, req
->dest_cpu
);
4216 } else if (req
->type
== REQ_SET_DOMAIN
) {
4218 spin_unlock_irq(&rq
->lock
);
4220 spin_unlock_irq(&rq
->lock
);
4224 complete(&req
->done
);
4226 __set_current_state(TASK_RUNNING
);
4230 /* Wait for kthread_stop */
4231 set_current_state(TASK_INTERRUPTIBLE
);
4232 while (!kthread_should_stop()) {
4234 set_current_state(TASK_INTERRUPTIBLE
);
4236 __set_current_state(TASK_RUNNING
);
4240 #ifdef CONFIG_HOTPLUG_CPU
4241 /* Figure out where task on dead CPU should go, use force if neccessary. */
4242 static void move_task_off_dead_cpu(int dead_cpu
, struct task_struct
*tsk
)
4248 mask
= node_to_cpumask(cpu_to_node(dead_cpu
));
4249 cpus_and(mask
, mask
, tsk
->cpus_allowed
);
4250 dest_cpu
= any_online_cpu(mask
);
4252 /* On any allowed CPU? */
4253 if (dest_cpu
== NR_CPUS
)
4254 dest_cpu
= any_online_cpu(tsk
->cpus_allowed
);
4256 /* No more Mr. Nice Guy. */
4257 if (dest_cpu
== NR_CPUS
) {
4258 cpus_setall(tsk
->cpus_allowed
);
4259 dest_cpu
= any_online_cpu(tsk
->cpus_allowed
);
4262 * Don't tell them about moving exiting tasks or
4263 * kernel threads (both mm NULL), since they never
4266 if (tsk
->mm
&& printk_ratelimit())
4267 printk(KERN_INFO
"process %d (%s) no "
4268 "longer affine to cpu%d\n",
4269 tsk
->pid
, tsk
->comm
, dead_cpu
);
4271 __migrate_task(tsk
, dead_cpu
, dest_cpu
);
4275 * While a dead CPU has no uninterruptible tasks queued at this point,
4276 * it might still have a nonzero ->nr_uninterruptible counter, because
4277 * for performance reasons the counter is not stricly tracking tasks to
4278 * their home CPUs. So we just add the counter to another CPU's counter,
4279 * to keep the global sum constant after CPU-down:
4281 static void migrate_nr_uninterruptible(runqueue_t
*rq_src
)
4283 runqueue_t
*rq_dest
= cpu_rq(any_online_cpu(CPU_MASK_ALL
));
4284 unsigned long flags
;
4286 local_irq_save(flags
);
4287 double_rq_lock(rq_src
, rq_dest
);
4288 rq_dest
->nr_uninterruptible
+= rq_src
->nr_uninterruptible
;
4289 rq_src
->nr_uninterruptible
= 0;
4290 double_rq_unlock(rq_src
, rq_dest
);
4291 local_irq_restore(flags
);
4294 /* Run through task list and migrate tasks from the dead cpu. */
4295 static void migrate_live_tasks(int src_cpu
)
4297 struct task_struct
*tsk
, *t
;
4299 write_lock_irq(&tasklist_lock
);
4301 do_each_thread(t
, tsk
) {
4305 if (task_cpu(tsk
) == src_cpu
)
4306 move_task_off_dead_cpu(src_cpu
, tsk
);
4307 } while_each_thread(t
, tsk
);
4309 write_unlock_irq(&tasklist_lock
);
4312 /* Schedules idle task to be the next runnable task on current CPU.
4313 * It does so by boosting its priority to highest possible and adding it to
4314 * the _front_ of runqueue. Used by CPU offline code.
4316 void sched_idle_next(void)
4318 int cpu
= smp_processor_id();
4319 runqueue_t
*rq
= this_rq();
4320 struct task_struct
*p
= rq
->idle
;
4321 unsigned long flags
;
4323 /* cpu has to be offline */
4324 BUG_ON(cpu_online(cpu
));
4326 /* Strictly not necessary since rest of the CPUs are stopped by now
4327 * and interrupts disabled on current cpu.
4329 spin_lock_irqsave(&rq
->lock
, flags
);
4331 __setscheduler(p
, SCHED_FIFO
, MAX_RT_PRIO
-1);
4332 /* Add idle task to _front_ of it's priority queue */
4333 __activate_idle_task(p
, rq
);
4335 spin_unlock_irqrestore(&rq
->lock
, flags
);
4338 /* Ensures that the idle task is using init_mm right before its cpu goes
4341 void idle_task_exit(void)
4343 struct mm_struct
*mm
= current
->active_mm
;
4345 BUG_ON(cpu_online(smp_processor_id()));
4348 switch_mm(mm
, &init_mm
, current
);
4352 static void migrate_dead(unsigned int dead_cpu
, task_t
*tsk
)
4354 struct runqueue
*rq
= cpu_rq(dead_cpu
);
4356 /* Must be exiting, otherwise would be on tasklist. */
4357 BUG_ON(tsk
->exit_state
!= EXIT_ZOMBIE
&& tsk
->exit_state
!= EXIT_DEAD
);
4359 /* Cannot have done final schedule yet: would have vanished. */
4360 BUG_ON(tsk
->flags
& PF_DEAD
);
4362 get_task_struct(tsk
);
4365 * Drop lock around migration; if someone else moves it,
4366 * that's OK. No task can be added to this CPU, so iteration is
4369 spin_unlock_irq(&rq
->lock
);
4370 move_task_off_dead_cpu(dead_cpu
, tsk
);
4371 spin_lock_irq(&rq
->lock
);
4373 put_task_struct(tsk
);
4376 /* release_task() removes task from tasklist, so we won't find dead tasks. */
4377 static void migrate_dead_tasks(unsigned int dead_cpu
)
4380 struct runqueue
*rq
= cpu_rq(dead_cpu
);
4382 for (arr
= 0; arr
< 2; arr
++) {
4383 for (i
= 0; i
< MAX_PRIO
; i
++) {
4384 struct list_head
*list
= &rq
->arrays
[arr
].queue
[i
];
4385 while (!list_empty(list
))
4386 migrate_dead(dead_cpu
,
4387 list_entry(list
->next
, task_t
,
4392 #endif /* CONFIG_HOTPLUG_CPU */
4395 * migration_call - callback that gets triggered when a CPU is added.
4396 * Here we can start up the necessary migration thread for the new CPU.
4398 static int migration_call(struct notifier_block
*nfb
, unsigned long action
,
4401 int cpu
= (long)hcpu
;
4402 struct task_struct
*p
;
4403 struct runqueue
*rq
;
4404 unsigned long flags
;
4407 case CPU_UP_PREPARE
:
4408 p
= kthread_create(migration_thread
, hcpu
, "migration/%d",cpu
);
4411 p
->flags
|= PF_NOFREEZE
;
4412 kthread_bind(p
, cpu
);
4413 /* Must be high prio: stop_machine expects to yield to it. */
4414 rq
= task_rq_lock(p
, &flags
);
4415 __setscheduler(p
, SCHED_FIFO
, MAX_RT_PRIO
-1);
4416 task_rq_unlock(rq
, &flags
);
4417 cpu_rq(cpu
)->migration_thread
= p
;
4420 /* Strictly unneccessary, as first user will wake it. */
4421 wake_up_process(cpu_rq(cpu
)->migration_thread
);
4423 #ifdef CONFIG_HOTPLUG_CPU
4424 case CPU_UP_CANCELED
:
4425 /* Unbind it from offline cpu so it can run. Fall thru. */
4426 kthread_bind(cpu_rq(cpu
)->migration_thread
,smp_processor_id());
4427 kthread_stop(cpu_rq(cpu
)->migration_thread
);
4428 cpu_rq(cpu
)->migration_thread
= NULL
;
4431 migrate_live_tasks(cpu
);
4433 kthread_stop(rq
->migration_thread
);
4434 rq
->migration_thread
= NULL
;
4435 /* Idle task back to normal (off runqueue, low prio) */
4436 rq
= task_rq_lock(rq
->idle
, &flags
);
4437 deactivate_task(rq
->idle
, rq
);
4438 rq
->idle
->static_prio
= MAX_PRIO
;
4439 __setscheduler(rq
->idle
, SCHED_NORMAL
, 0);
4440 migrate_dead_tasks(cpu
);
4441 task_rq_unlock(rq
, &flags
);
4442 migrate_nr_uninterruptible(rq
);
4443 BUG_ON(rq
->nr_running
!= 0);
4445 /* No need to migrate the tasks: it was best-effort if
4446 * they didn't do lock_cpu_hotplug(). Just wake up
4447 * the requestors. */
4448 spin_lock_irq(&rq
->lock
);
4449 while (!list_empty(&rq
->migration_queue
)) {
4450 migration_req_t
*req
;
4451 req
= list_entry(rq
->migration_queue
.next
,
4452 migration_req_t
, list
);
4453 BUG_ON(req
->type
!= REQ_MOVE_TASK
);
4454 list_del_init(&req
->list
);
4455 complete(&req
->done
);
4457 spin_unlock_irq(&rq
->lock
);
4464 /* Register at highest priority so that task migration (migrate_all_tasks)
4465 * happens before everything else.
4467 static struct notifier_block __devinitdata migration_notifier
= {
4468 .notifier_call
= migration_call
,
4472 int __init
migration_init(void)
4474 void *cpu
= (void *)(long)smp_processor_id();
4475 /* Start one for boot CPU. */
4476 migration_call(&migration_notifier
, CPU_UP_PREPARE
, cpu
);
4477 migration_call(&migration_notifier
, CPU_ONLINE
, cpu
);
4478 register_cpu_notifier(&migration_notifier
);
4484 #define SCHED_DOMAIN_DEBUG
4485 #ifdef SCHED_DOMAIN_DEBUG
4486 static void sched_domain_debug(struct sched_domain
*sd
, int cpu
)
4490 printk(KERN_DEBUG
"CPU%d attaching sched-domain:\n", cpu
);
4495 struct sched_group
*group
= sd
->groups
;
4496 cpumask_t groupmask
;
4498 cpumask_scnprintf(str
, NR_CPUS
, sd
->span
);
4499 cpus_clear(groupmask
);
4502 for (i
= 0; i
< level
+ 1; i
++)
4504 printk("domain %d: ", level
);
4506 if (!(sd
->flags
& SD_LOAD_BALANCE
)) {
4507 printk("does not load-balance\n");
4509 printk(KERN_ERR
"ERROR: !SD_LOAD_BALANCE domain has parent");
4513 printk("span %s\n", str
);
4515 if (!cpu_isset(cpu
, sd
->span
))
4516 printk(KERN_ERR
"ERROR: domain->span does not contain CPU%d\n", cpu
);
4517 if (!cpu_isset(cpu
, group
->cpumask
))
4518 printk(KERN_ERR
"ERROR: domain->groups does not contain CPU%d\n", cpu
);
4521 for (i
= 0; i
< level
+ 2; i
++)
4527 printk(KERN_ERR
"ERROR: group is NULL\n");
4531 if (!group
->cpu_power
) {
4533 printk(KERN_ERR
"ERROR: domain->cpu_power not set\n");
4536 if (!cpus_weight(group
->cpumask
)) {
4538 printk(KERN_ERR
"ERROR: empty group\n");
4541 if (cpus_intersects(groupmask
, group
->cpumask
)) {
4543 printk(KERN_ERR
"ERROR: repeated CPUs\n");
4546 cpus_or(groupmask
, groupmask
, group
->cpumask
);
4548 cpumask_scnprintf(str
, NR_CPUS
, group
->cpumask
);
4551 group
= group
->next
;
4552 } while (group
!= sd
->groups
);
4555 if (!cpus_equal(sd
->span
, groupmask
))
4556 printk(KERN_ERR
"ERROR: groups don't span domain->span\n");
4562 if (!cpus_subset(groupmask
, sd
->span
))
4563 printk(KERN_ERR
"ERROR: parent span is not a superset of domain->span\n");
4569 #define sched_domain_debug(sd, cpu) {}
4573 * Attach the domain 'sd' to 'cpu' as its base domain. Callers must
4574 * hold the hotplug lock.
4576 void __devinit
cpu_attach_domain(struct sched_domain
*sd
, int cpu
)
4578 migration_req_t req
;
4579 unsigned long flags
;
4580 runqueue_t
*rq
= cpu_rq(cpu
);
4583 sched_domain_debug(sd
, cpu
);
4585 spin_lock_irqsave(&rq
->lock
, flags
);
4587 if (cpu
== smp_processor_id() || !cpu_online(cpu
)) {
4590 init_completion(&req
.done
);
4591 req
.type
= REQ_SET_DOMAIN
;
4593 list_add(&req
.list
, &rq
->migration_queue
);
4597 spin_unlock_irqrestore(&rq
->lock
, flags
);
4600 wake_up_process(rq
->migration_thread
);
4601 wait_for_completion(&req
.done
);
4605 /* cpus with isolated domains */
4606 cpumask_t __devinitdata cpu_isolated_map
= CPU_MASK_NONE
;
4608 /* Setup the mask of cpus configured for isolated domains */
4609 static int __init
isolated_cpu_setup(char *str
)
4611 int ints
[NR_CPUS
], i
;
4613 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
4614 cpus_clear(cpu_isolated_map
);
4615 for (i
= 1; i
<= ints
[0]; i
++)
4616 if (ints
[i
] < NR_CPUS
)
4617 cpu_set(ints
[i
], cpu_isolated_map
);
4621 __setup ("isolcpus=", isolated_cpu_setup
);
4624 * init_sched_build_groups takes an array of groups, the cpumask we wish
4625 * to span, and a pointer to a function which identifies what group a CPU
4626 * belongs to. The return value of group_fn must be a valid index into the
4627 * groups[] array, and must be >= 0 and < NR_CPUS (due to the fact that we
4628 * keep track of groups covered with a cpumask_t).
4630 * init_sched_build_groups will build a circular linked list of the groups
4631 * covered by the given span, and will set each group's ->cpumask correctly,
4632 * and ->cpu_power to 0.
4634 void __devinit
init_sched_build_groups(struct sched_group groups
[],
4635 cpumask_t span
, int (*group_fn
)(int cpu
))
4637 struct sched_group
*first
= NULL
, *last
= NULL
;
4638 cpumask_t covered
= CPU_MASK_NONE
;
4641 for_each_cpu_mask(i
, span
) {
4642 int group
= group_fn(i
);
4643 struct sched_group
*sg
= &groups
[group
];
4646 if (cpu_isset(i
, covered
))
4649 sg
->cpumask
= CPU_MASK_NONE
;
4652 for_each_cpu_mask(j
, span
) {
4653 if (group_fn(j
) != group
)
4656 cpu_set(j
, covered
);
4657 cpu_set(j
, sg
->cpumask
);
4669 #ifdef ARCH_HAS_SCHED_DOMAIN
4670 extern void __devinit
arch_init_sched_domains(void);
4671 extern void __devinit
arch_destroy_sched_domains(void);
4673 #ifdef CONFIG_SCHED_SMT
4674 static DEFINE_PER_CPU(struct sched_domain
, cpu_domains
);
4675 static struct sched_group sched_group_cpus
[NR_CPUS
];
4676 static int __devinit
cpu_to_cpu_group(int cpu
)
4682 static DEFINE_PER_CPU(struct sched_domain
, phys_domains
);
4683 static struct sched_group sched_group_phys
[NR_CPUS
];
4684 static int __devinit
cpu_to_phys_group(int cpu
)
4686 #ifdef CONFIG_SCHED_SMT
4687 return first_cpu(cpu_sibling_map
[cpu
]);
4695 static DEFINE_PER_CPU(struct sched_domain
, node_domains
);
4696 static struct sched_group sched_group_nodes
[MAX_NUMNODES
];
4697 static int __devinit
cpu_to_node_group(int cpu
)
4699 return cpu_to_node(cpu
);
4703 #if defined(CONFIG_SCHED_SMT) && defined(CONFIG_NUMA)
4705 * The domains setup code relies on siblings not spanning
4706 * multiple nodes. Make sure the architecture has a proper
4709 static void check_sibling_maps(void)
4713 for_each_online_cpu(i
) {
4714 for_each_cpu_mask(j
, cpu_sibling_map
[i
]) {
4715 if (cpu_to_node(i
) != cpu_to_node(j
)) {
4716 printk(KERN_INFO
"warning: CPU %d siblings map "
4717 "to different node - isolating "
4719 cpu_sibling_map
[i
] = cpumask_of_cpu(i
);
4728 * Set up scheduler domains and groups. Callers must hold the hotplug lock.
4730 static void __devinit
arch_init_sched_domains(void)
4733 cpumask_t cpu_default_map
;
4735 #if defined(CONFIG_SCHED_SMT) && defined(CONFIG_NUMA)
4736 check_sibling_maps();
4739 * Setup mask for cpus without special case scheduling requirements.
4740 * For now this just excludes isolated cpus, but could be used to
4741 * exclude other special cases in the future.
4743 cpus_complement(cpu_default_map
, cpu_isolated_map
);
4744 cpus_and(cpu_default_map
, cpu_default_map
, cpu_online_map
);
4747 * Set up domains. Isolated domains just stay on the dummy domain.
4749 for_each_cpu_mask(i
, cpu_default_map
) {
4751 struct sched_domain
*sd
= NULL
, *p
;
4752 cpumask_t nodemask
= node_to_cpumask(cpu_to_node(i
));
4754 cpus_and(nodemask
, nodemask
, cpu_default_map
);
4757 sd
= &per_cpu(node_domains
, i
);
4758 group
= cpu_to_node_group(i
);
4760 sd
->span
= cpu_default_map
;
4761 sd
->groups
= &sched_group_nodes
[group
];
4765 sd
= &per_cpu(phys_domains
, i
);
4766 group
= cpu_to_phys_group(i
);
4768 sd
->span
= nodemask
;
4770 sd
->groups
= &sched_group_phys
[group
];
4772 #ifdef CONFIG_SCHED_SMT
4774 sd
= &per_cpu(cpu_domains
, i
);
4775 group
= cpu_to_cpu_group(i
);
4776 *sd
= SD_SIBLING_INIT
;
4777 sd
->span
= cpu_sibling_map
[i
];
4778 cpus_and(sd
->span
, sd
->span
, cpu_default_map
);
4780 sd
->groups
= &sched_group_cpus
[group
];
4784 #ifdef CONFIG_SCHED_SMT
4785 /* Set up CPU (sibling) groups */
4786 for_each_online_cpu(i
) {
4787 cpumask_t this_sibling_map
= cpu_sibling_map
[i
];
4788 cpus_and(this_sibling_map
, this_sibling_map
, cpu_default_map
);
4789 if (i
!= first_cpu(this_sibling_map
))
4792 init_sched_build_groups(sched_group_cpus
, this_sibling_map
,
4797 /* Set up physical groups */
4798 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
4799 cpumask_t nodemask
= node_to_cpumask(i
);
4801 cpus_and(nodemask
, nodemask
, cpu_default_map
);
4802 if (cpus_empty(nodemask
))
4805 init_sched_build_groups(sched_group_phys
, nodemask
,
4806 &cpu_to_phys_group
);
4810 /* Set up node groups */
4811 init_sched_build_groups(sched_group_nodes
, cpu_default_map
,
4812 &cpu_to_node_group
);
4815 /* Calculate CPU power for physical packages and nodes */
4816 for_each_cpu_mask(i
, cpu_default_map
) {
4818 struct sched_domain
*sd
;
4819 #ifdef CONFIG_SCHED_SMT
4820 sd
= &per_cpu(cpu_domains
, i
);
4821 power
= SCHED_LOAD_SCALE
;
4822 sd
->groups
->cpu_power
= power
;
4825 sd
= &per_cpu(phys_domains
, i
);
4826 power
= SCHED_LOAD_SCALE
+ SCHED_LOAD_SCALE
*
4827 (cpus_weight(sd
->groups
->cpumask
)-1) / 10;
4828 sd
->groups
->cpu_power
= power
;
4831 if (i
== first_cpu(sd
->groups
->cpumask
)) {
4832 /* Only add "power" once for each physical package. */
4833 sd
= &per_cpu(node_domains
, i
);
4834 sd
->groups
->cpu_power
+= power
;
4839 /* Attach the domains */
4840 for_each_online_cpu(i
) {
4841 struct sched_domain
*sd
;
4842 #ifdef CONFIG_SCHED_SMT
4843 sd
= &per_cpu(cpu_domains
, i
);
4845 sd
= &per_cpu(phys_domains
, i
);
4847 cpu_attach_domain(sd
, i
);
4851 #ifdef CONFIG_HOTPLUG_CPU
4852 static void __devinit
arch_destroy_sched_domains(void)
4854 /* Do nothing: everything is statically allocated. */
4858 #endif /* ARCH_HAS_SCHED_DOMAIN */
4861 * Initial dummy domain for early boot and for hotplug cpu. Being static,
4862 * it is initialized to zero, so all balancing flags are cleared which is
4865 static struct sched_domain sched_domain_dummy
;
4867 #ifdef CONFIG_HOTPLUG_CPU
4869 * Force a reinitialization of the sched domains hierarchy. The domains
4870 * and groups cannot be updated in place without racing with the balancing
4871 * code, so we temporarily attach all running cpus to a "dummy" domain
4872 * which will prevent rebalancing while the sched domains are recalculated.
4874 static int update_sched_domains(struct notifier_block
*nfb
,
4875 unsigned long action
, void *hcpu
)
4880 case CPU_UP_PREPARE
:
4881 case CPU_DOWN_PREPARE
:
4882 for_each_online_cpu(i
)
4883 cpu_attach_domain(&sched_domain_dummy
, i
);
4884 arch_destroy_sched_domains();
4887 case CPU_UP_CANCELED
:
4888 case CPU_DOWN_FAILED
:
4892 * Fall through and re-initialise the domains.
4899 /* The hotplug lock is already held by cpu_up/cpu_down */
4900 arch_init_sched_domains();
4906 void __init
sched_init_smp(void)
4909 arch_init_sched_domains();
4910 unlock_cpu_hotplug();
4911 /* XXX: Theoretical race here - CPU may be hotplugged now */
4912 hotcpu_notifier(update_sched_domains
, 0);
4915 void __init
sched_init_smp(void)
4918 #endif /* CONFIG_SMP */
4920 int in_sched_functions(unsigned long addr
)
4922 /* Linker adds these: start and end of __sched functions */
4923 extern char __sched_text_start
[], __sched_text_end
[];
4924 return in_lock_functions(addr
) ||
4925 (addr
>= (unsigned long)__sched_text_start
4926 && addr
< (unsigned long)__sched_text_end
);
4929 void __init
sched_init(void)
4934 for (i
= 0; i
< NR_CPUS
; i
++) {
4935 prio_array_t
*array
;
4938 spin_lock_init(&rq
->lock
);
4940 rq
->active
= rq
->arrays
;
4941 rq
->expired
= rq
->arrays
+ 1;
4942 rq
->best_expired_prio
= MAX_PRIO
;
4945 rq
->sd
= &sched_domain_dummy
;
4946 for (j
= 1; j
< 3; j
++)
4947 rq
->cpu_load
[j
] = 0;
4948 rq
->active_balance
= 0;
4950 rq
->migration_thread
= NULL
;
4951 INIT_LIST_HEAD(&rq
->migration_queue
);
4953 atomic_set(&rq
->nr_iowait
, 0);
4955 for (j
= 0; j
< 2; j
++) {
4956 array
= rq
->arrays
+ j
;
4957 for (k
= 0; k
< MAX_PRIO
; k
++) {
4958 INIT_LIST_HEAD(array
->queue
+ k
);
4959 __clear_bit(k
, array
->bitmap
);
4961 // delimiter for bitsearch
4962 __set_bit(MAX_PRIO
, array
->bitmap
);
4967 * The boot idle thread does lazy MMU switching as well:
4969 atomic_inc(&init_mm
.mm_count
);
4970 enter_lazy_tlb(&init_mm
, current
);
4973 * Make us the idle thread. Technically, schedule() should not be
4974 * called from this thread, however somewhere below it might be,
4975 * but because we are the idle thread, we just pick up running again
4976 * when this runqueue becomes "idle".
4978 init_idle(current
, smp_processor_id());
4981 #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP
4982 void __might_sleep(char *file
, int line
)
4984 #if defined(in_atomic)
4985 static unsigned long prev_jiffy
; /* ratelimiting */
4987 if ((in_atomic() || irqs_disabled()) &&
4988 system_state
== SYSTEM_RUNNING
&& !oops_in_progress
) {
4989 if (time_before(jiffies
, prev_jiffy
+ HZ
) && prev_jiffy
)
4991 prev_jiffy
= jiffies
;
4992 printk(KERN_ERR
"Debug: sleeping function called from invalid"
4993 " context at %s:%d\n", file
, line
);
4994 printk("in_atomic():%d, irqs_disabled():%d\n",
4995 in_atomic(), irqs_disabled());
5000 EXPORT_SYMBOL(__might_sleep
);
5003 #ifdef CONFIG_MAGIC_SYSRQ
5004 void normalize_rt_tasks(void)
5006 struct task_struct
*p
;
5007 prio_array_t
*array
;
5008 unsigned long flags
;
5011 read_lock_irq(&tasklist_lock
);
5012 for_each_process (p
) {
5016 rq
= task_rq_lock(p
, &flags
);
5020 deactivate_task(p
, task_rq(p
));
5021 __setscheduler(p
, SCHED_NORMAL
, 0);
5023 __activate_task(p
, task_rq(p
));
5024 resched_task(rq
->curr
);
5027 task_rq_unlock(rq
, &flags
);
5029 read_unlock_irq(&tasklist_lock
);
5032 #endif /* CONFIG_MAGIC_SYSRQ */