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/capability.h>
31 #include <linux/completion.h>
32 #include <linux/kernel_stat.h>
33 #include <linux/debug_locks.h>
34 #include <linux/security.h>
35 #include <linux/notifier.h>
36 #include <linux/profile.h>
37 #include <linux/suspend.h>
38 #include <linux/vmalloc.h>
39 #include <linux/blkdev.h>
40 #include <linux/delay.h>
41 #include <linux/smp.h>
42 #include <linux/threads.h>
43 #include <linux/timer.h>
44 #include <linux/rcupdate.h>
45 #include <linux/cpu.h>
46 #include <linux/cpuset.h>
47 #include <linux/percpu.h>
48 #include <linux/kthread.h>
49 #include <linux/seq_file.h>
50 #include <linux/syscalls.h>
51 #include <linux/times.h>
52 #include <linux/acct.h>
53 #include <linux/kprobes.h>
56 #include <asm/unistd.h>
59 * Convert user-nice values [ -20 ... 0 ... 19 ]
60 * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ],
63 #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20)
64 #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20)
65 #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio)
68 * 'User priority' is the nice value converted to something we
69 * can work with better when scaling various scheduler parameters,
70 * it's a [ 0 ... 39 ] range.
72 #define USER_PRIO(p) ((p)-MAX_RT_PRIO)
73 #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio)
74 #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO))
77 * Some helpers for converting nanosecond timing to jiffy resolution
79 #define NS_TO_JIFFIES(TIME) ((TIME) / (1000000000 / HZ))
80 #define JIFFIES_TO_NS(TIME) ((TIME) * (1000000000 / HZ))
83 * These are the 'tuning knobs' of the scheduler:
85 * Minimum timeslice is 5 msecs (or 1 jiffy, whichever is larger),
86 * default timeslice is 100 msecs, maximum timeslice is 800 msecs.
87 * Timeslices get refilled after they expire.
89 #define MIN_TIMESLICE max(5 * HZ / 1000, 1)
90 #define DEF_TIMESLICE (100 * HZ / 1000)
91 #define ON_RUNQUEUE_WEIGHT 30
92 #define CHILD_PENALTY 95
93 #define PARENT_PENALTY 100
95 #define PRIO_BONUS_RATIO 25
96 #define MAX_BONUS (MAX_USER_PRIO * PRIO_BONUS_RATIO / 100)
97 #define INTERACTIVE_DELTA 2
98 #define MAX_SLEEP_AVG (DEF_TIMESLICE * MAX_BONUS)
99 #define STARVATION_LIMIT (MAX_SLEEP_AVG)
100 #define NS_MAX_SLEEP_AVG (JIFFIES_TO_NS(MAX_SLEEP_AVG))
103 * If a task is 'interactive' then we reinsert it in the active
104 * array after it has expired its current timeslice. (it will not
105 * continue to run immediately, it will still roundrobin with
106 * other interactive tasks.)
108 * This part scales the interactivity limit depending on niceness.
110 * We scale it linearly, offset by the INTERACTIVE_DELTA delta.
111 * Here are a few examples of different nice levels:
113 * TASK_INTERACTIVE(-20): [1,1,1,1,1,1,1,1,1,0,0]
114 * TASK_INTERACTIVE(-10): [1,1,1,1,1,1,1,0,0,0,0]
115 * TASK_INTERACTIVE( 0): [1,1,1,1,0,0,0,0,0,0,0]
116 * TASK_INTERACTIVE( 10): [1,1,0,0,0,0,0,0,0,0,0]
117 * TASK_INTERACTIVE( 19): [0,0,0,0,0,0,0,0,0,0,0]
119 * (the X axis represents the possible -5 ... 0 ... +5 dynamic
120 * priority range a task can explore, a value of '1' means the
121 * task is rated interactive.)
123 * Ie. nice +19 tasks can never get 'interactive' enough to be
124 * reinserted into the active array. And only heavily CPU-hog nice -20
125 * tasks will be expired. Default nice 0 tasks are somewhere between,
126 * it takes some effort for them to get interactive, but it's not
130 #define CURRENT_BONUS(p) \
131 (NS_TO_JIFFIES((p)->sleep_avg) * MAX_BONUS / \
134 #define GRANULARITY (10 * HZ / 1000 ? : 1)
137 #define TIMESLICE_GRANULARITY(p) (GRANULARITY * \
138 (1 << (((MAX_BONUS - CURRENT_BONUS(p)) ? : 1) - 1)) * \
141 #define TIMESLICE_GRANULARITY(p) (GRANULARITY * \
142 (1 << (((MAX_BONUS - CURRENT_BONUS(p)) ? : 1) - 1)))
145 #define SCALE(v1,v1_max,v2_max) \
146 (v1) * (v2_max) / (v1_max)
149 (SCALE(TASK_NICE(p) + 20, 40, MAX_BONUS) - 20 * MAX_BONUS / 40 + \
152 #define TASK_INTERACTIVE(p) \
153 ((p)->prio <= (p)->static_prio - DELTA(p))
155 #define INTERACTIVE_SLEEP(p) \
156 (JIFFIES_TO_NS(MAX_SLEEP_AVG * \
157 (MAX_BONUS / 2 + DELTA((p)) + 1) / MAX_BONUS - 1))
159 #define TASK_PREEMPTS_CURR(p, rq) \
160 ((p)->prio < (rq)->curr->prio)
163 * task_timeslice() scales user-nice values [ -20 ... 0 ... 19 ]
164 * to time slice values: [800ms ... 100ms ... 5ms]
166 * The higher a thread's priority, the bigger timeslices
167 * it gets during one round of execution. But even the lowest
168 * priority thread gets MIN_TIMESLICE worth of execution time.
171 #define SCALE_PRIO(x, prio) \
172 max(x * (MAX_PRIO - prio) / (MAX_USER_PRIO / 2), MIN_TIMESLICE)
174 static unsigned int static_prio_timeslice(int static_prio
)
176 if (static_prio
< NICE_TO_PRIO(0))
177 return SCALE_PRIO(DEF_TIMESLICE
* 4, static_prio
);
179 return SCALE_PRIO(DEF_TIMESLICE
, static_prio
);
182 static inline unsigned int task_timeslice(task_t
*p
)
184 return static_prio_timeslice(p
->static_prio
);
187 #define task_hot(p, now, sd) ((long long) ((now) - (p)->last_ran) \
188 < (long long) (sd)->cache_hot_time)
191 * These are the runqueue data structures:
194 typedef struct runqueue runqueue_t
;
197 unsigned int nr_active
;
198 DECLARE_BITMAP(bitmap
, MAX_PRIO
+1); /* include 1 bit for delimiter */
199 struct list_head queue
[MAX_PRIO
];
203 * This is the main, per-CPU runqueue data structure.
205 * Locking rule: those places that want to lock multiple runqueues
206 * (such as the load balancing or the thread migration code), lock
207 * acquire operations must be ordered by ascending &runqueue.
213 * nr_running and cpu_load should be in the same cacheline because
214 * remote CPUs use both these fields when doing load calculation.
216 unsigned long nr_running
;
217 unsigned long raw_weighted_load
;
219 unsigned long cpu_load
[3];
221 unsigned long long nr_switches
;
224 * This is part of a global counter where only the total sum
225 * over all CPUs matters. A task can increase this counter on
226 * one CPU and if it got migrated afterwards it may decrease
227 * it on another CPU. Always updated under the runqueue lock:
229 unsigned long nr_uninterruptible
;
231 unsigned long expired_timestamp
;
232 unsigned long long timestamp_last_tick
;
234 struct mm_struct
*prev_mm
;
235 prio_array_t
*active
, *expired
, arrays
[2];
236 int best_expired_prio
;
240 struct sched_domain
*sd
;
242 /* For active balancing */
246 task_t
*migration_thread
;
247 struct list_head migration_queue
;
250 #ifdef CONFIG_SCHEDSTATS
252 struct sched_info rq_sched_info
;
254 /* sys_sched_yield() stats */
255 unsigned long yld_exp_empty
;
256 unsigned long yld_act_empty
;
257 unsigned long yld_both_empty
;
258 unsigned long yld_cnt
;
260 /* schedule() stats */
261 unsigned long sched_switch
;
262 unsigned long sched_cnt
;
263 unsigned long sched_goidle
;
265 /* try_to_wake_up() stats */
266 unsigned long ttwu_cnt
;
267 unsigned long ttwu_local
;
271 static DEFINE_PER_CPU(struct runqueue
, runqueues
);
274 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
275 * See detach_destroy_domains: synchronize_sched for details.
277 * The domain tree of any CPU may only be accessed from within
278 * preempt-disabled sections.
280 #define for_each_domain(cpu, domain) \
281 for (domain = rcu_dereference(cpu_rq(cpu)->sd); domain; domain = domain->parent)
283 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
284 #define this_rq() (&__get_cpu_var(runqueues))
285 #define task_rq(p) cpu_rq(task_cpu(p))
286 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
288 #ifndef prepare_arch_switch
289 # define prepare_arch_switch(next) do { } while (0)
291 #ifndef finish_arch_switch
292 # define finish_arch_switch(prev) do { } while (0)
295 #ifndef __ARCH_WANT_UNLOCKED_CTXSW
296 static inline int task_running(runqueue_t
*rq
, task_t
*p
)
298 return rq
->curr
== p
;
301 static inline void prepare_lock_switch(runqueue_t
*rq
, task_t
*next
)
305 static inline void finish_lock_switch(runqueue_t
*rq
, task_t
*prev
)
307 #ifdef CONFIG_DEBUG_SPINLOCK
308 /* this is a valid case when another task releases the spinlock */
309 rq
->lock
.owner
= current
;
311 spin_unlock_irq(&rq
->lock
);
314 #else /* __ARCH_WANT_UNLOCKED_CTXSW */
315 static inline int task_running(runqueue_t
*rq
, task_t
*p
)
320 return rq
->curr
== p
;
324 static inline void prepare_lock_switch(runqueue_t
*rq
, task_t
*next
)
328 * We can optimise this out completely for !SMP, because the
329 * SMP rebalancing from interrupt is the only thing that cares
334 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
335 spin_unlock_irq(&rq
->lock
);
337 spin_unlock(&rq
->lock
);
341 static inline void finish_lock_switch(runqueue_t
*rq
, task_t
*prev
)
345 * After ->oncpu is cleared, the task can be moved to a different CPU.
346 * We must ensure this doesn't happen until the switch is completely
352 #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW
356 #endif /* __ARCH_WANT_UNLOCKED_CTXSW */
359 * __task_rq_lock - lock the runqueue a given task resides on.
360 * Must be called interrupts disabled.
362 static inline runqueue_t
*__task_rq_lock(task_t
*p
)
369 spin_lock(&rq
->lock
);
370 if (unlikely(rq
!= task_rq(p
))) {
371 spin_unlock(&rq
->lock
);
372 goto repeat_lock_task
;
378 * task_rq_lock - lock the runqueue a given task resides on and disable
379 * interrupts. Note the ordering: we can safely lookup the task_rq without
380 * explicitly disabling preemption.
382 static runqueue_t
*task_rq_lock(task_t
*p
, unsigned long *flags
)
388 local_irq_save(*flags
);
390 spin_lock(&rq
->lock
);
391 if (unlikely(rq
!= task_rq(p
))) {
392 spin_unlock_irqrestore(&rq
->lock
, *flags
);
393 goto repeat_lock_task
;
398 static inline void __task_rq_unlock(runqueue_t
*rq
)
401 spin_unlock(&rq
->lock
);
404 static inline void task_rq_unlock(runqueue_t
*rq
, unsigned long *flags
)
407 spin_unlock_irqrestore(&rq
->lock
, *flags
);
410 #ifdef CONFIG_SCHEDSTATS
412 * bump this up when changing the output format or the meaning of an existing
413 * format, so that tools can adapt (or abort)
415 #define SCHEDSTAT_VERSION 12
417 static int show_schedstat(struct seq_file
*seq
, void *v
)
421 seq_printf(seq
, "version %d\n", SCHEDSTAT_VERSION
);
422 seq_printf(seq
, "timestamp %lu\n", jiffies
);
423 for_each_online_cpu(cpu
) {
424 runqueue_t
*rq
= cpu_rq(cpu
);
426 struct sched_domain
*sd
;
430 /* runqueue-specific stats */
432 "cpu%d %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu",
433 cpu
, rq
->yld_both_empty
,
434 rq
->yld_act_empty
, rq
->yld_exp_empty
, rq
->yld_cnt
,
435 rq
->sched_switch
, rq
->sched_cnt
, rq
->sched_goidle
,
436 rq
->ttwu_cnt
, rq
->ttwu_local
,
437 rq
->rq_sched_info
.cpu_time
,
438 rq
->rq_sched_info
.run_delay
, rq
->rq_sched_info
.pcnt
);
440 seq_printf(seq
, "\n");
443 /* domain-specific stats */
445 for_each_domain(cpu
, sd
) {
446 enum idle_type itype
;
447 char mask_str
[NR_CPUS
];
449 cpumask_scnprintf(mask_str
, NR_CPUS
, sd
->span
);
450 seq_printf(seq
, "domain%d %s", dcnt
++, mask_str
);
451 for (itype
= SCHED_IDLE
; itype
< MAX_IDLE_TYPES
;
453 seq_printf(seq
, " %lu %lu %lu %lu %lu %lu %lu %lu",
455 sd
->lb_balanced
[itype
],
456 sd
->lb_failed
[itype
],
457 sd
->lb_imbalance
[itype
],
458 sd
->lb_gained
[itype
],
459 sd
->lb_hot_gained
[itype
],
460 sd
->lb_nobusyq
[itype
],
461 sd
->lb_nobusyg
[itype
]);
463 seq_printf(seq
, " %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu\n",
464 sd
->alb_cnt
, sd
->alb_failed
, sd
->alb_pushed
,
465 sd
->sbe_cnt
, sd
->sbe_balanced
, sd
->sbe_pushed
,
466 sd
->sbf_cnt
, sd
->sbf_balanced
, sd
->sbf_pushed
,
467 sd
->ttwu_wake_remote
, sd
->ttwu_move_affine
, sd
->ttwu_move_balance
);
475 static int schedstat_open(struct inode
*inode
, struct file
*file
)
477 unsigned int size
= PAGE_SIZE
* (1 + num_online_cpus() / 32);
478 char *buf
= kmalloc(size
, GFP_KERNEL
);
484 res
= single_open(file
, show_schedstat
, NULL
);
486 m
= file
->private_data
;
494 struct file_operations proc_schedstat_operations
= {
495 .open
= schedstat_open
,
498 .release
= single_release
,
501 # define schedstat_inc(rq, field) do { (rq)->field++; } while (0)
502 # define schedstat_add(rq, field, amt) do { (rq)->field += (amt); } while (0)
503 #else /* !CONFIG_SCHEDSTATS */
504 # define schedstat_inc(rq, field) do { } while (0)
505 # define schedstat_add(rq, field, amt) do { } while (0)
509 * rq_lock - lock a given runqueue and disable interrupts.
511 static inline runqueue_t
*this_rq_lock(void)
518 spin_lock(&rq
->lock
);
523 #ifdef CONFIG_SCHEDSTATS
525 * Called when a process is dequeued from the active array and given
526 * the cpu. We should note that with the exception of interactive
527 * tasks, the expired queue will become the active queue after the active
528 * queue is empty, without explicitly dequeuing and requeuing tasks in the
529 * expired queue. (Interactive tasks may be requeued directly to the
530 * active queue, thus delaying tasks in the expired queue from running;
531 * see scheduler_tick()).
533 * This function is only called from sched_info_arrive(), rather than
534 * dequeue_task(). Even though a task may be queued and dequeued multiple
535 * times as it is shuffled about, we're really interested in knowing how
536 * long it was from the *first* time it was queued to the time that it
539 static inline void sched_info_dequeued(task_t
*t
)
541 t
->sched_info
.last_queued
= 0;
545 * Called when a task finally hits the cpu. We can now calculate how
546 * long it was waiting to run. We also note when it began so that we
547 * can keep stats on how long its timeslice is.
549 static void sched_info_arrive(task_t
*t
)
551 unsigned long now
= jiffies
, diff
= 0;
552 struct runqueue
*rq
= task_rq(t
);
554 if (t
->sched_info
.last_queued
)
555 diff
= now
- t
->sched_info
.last_queued
;
556 sched_info_dequeued(t
);
557 t
->sched_info
.run_delay
+= diff
;
558 t
->sched_info
.last_arrival
= now
;
559 t
->sched_info
.pcnt
++;
564 rq
->rq_sched_info
.run_delay
+= diff
;
565 rq
->rq_sched_info
.pcnt
++;
569 * Called when a process is queued into either the active or expired
570 * array. The time is noted and later used to determine how long we
571 * had to wait for us to reach the cpu. Since the expired queue will
572 * become the active queue after active queue is empty, without dequeuing
573 * and requeuing any tasks, we are interested in queuing to either. It
574 * is unusual but not impossible for tasks to be dequeued and immediately
575 * requeued in the same or another array: this can happen in sched_yield(),
576 * set_user_nice(), and even load_balance() as it moves tasks from runqueue
579 * This function is only called from enqueue_task(), but also only updates
580 * the timestamp if it is already not set. It's assumed that
581 * sched_info_dequeued() will clear that stamp when appropriate.
583 static inline void sched_info_queued(task_t
*t
)
585 if (!t
->sched_info
.last_queued
)
586 t
->sched_info
.last_queued
= jiffies
;
590 * Called when a process ceases being the active-running process, either
591 * voluntarily or involuntarily. Now we can calculate how long we ran.
593 static inline void sched_info_depart(task_t
*t
)
595 struct runqueue
*rq
= task_rq(t
);
596 unsigned long diff
= jiffies
- t
->sched_info
.last_arrival
;
598 t
->sched_info
.cpu_time
+= diff
;
601 rq
->rq_sched_info
.cpu_time
+= diff
;
605 * Called when tasks are switched involuntarily due, typically, to expiring
606 * their time slice. (This may also be called when switching to or from
607 * the idle task.) We are only called when prev != next.
609 static inline void sched_info_switch(task_t
*prev
, task_t
*next
)
611 struct runqueue
*rq
= task_rq(prev
);
614 * prev now departs the cpu. It's not interesting to record
615 * stats about how efficient we were at scheduling the idle
618 if (prev
!= rq
->idle
)
619 sched_info_depart(prev
);
621 if (next
!= rq
->idle
)
622 sched_info_arrive(next
);
625 #define sched_info_queued(t) do { } while (0)
626 #define sched_info_switch(t, next) do { } while (0)
627 #endif /* CONFIG_SCHEDSTATS */
630 * Adding/removing a task to/from a priority array:
632 static void dequeue_task(struct task_struct
*p
, prio_array_t
*array
)
635 list_del(&p
->run_list
);
636 if (list_empty(array
->queue
+ p
->prio
))
637 __clear_bit(p
->prio
, array
->bitmap
);
640 static void enqueue_task(struct task_struct
*p
, prio_array_t
*array
)
642 sched_info_queued(p
);
643 list_add_tail(&p
->run_list
, array
->queue
+ p
->prio
);
644 __set_bit(p
->prio
, array
->bitmap
);
650 * Put task to the end of the run list without the overhead of dequeue
651 * followed by enqueue.
653 static void requeue_task(struct task_struct
*p
, prio_array_t
*array
)
655 list_move_tail(&p
->run_list
, array
->queue
+ p
->prio
);
658 static inline void enqueue_task_head(struct task_struct
*p
, prio_array_t
*array
)
660 list_add(&p
->run_list
, array
->queue
+ p
->prio
);
661 __set_bit(p
->prio
, array
->bitmap
);
667 * __normal_prio - return the priority that is based on the static
668 * priority but is modified by bonuses/penalties.
670 * We scale the actual sleep average [0 .... MAX_SLEEP_AVG]
671 * into the -5 ... 0 ... +5 bonus/penalty range.
673 * We use 25% of the full 0...39 priority range so that:
675 * 1) nice +19 interactive tasks do not preempt nice 0 CPU hogs.
676 * 2) nice -20 CPU hogs do not get preempted by nice 0 tasks.
678 * Both properties are important to certain workloads.
681 static inline int __normal_prio(task_t
*p
)
685 bonus
= CURRENT_BONUS(p
) - MAX_BONUS
/ 2;
687 prio
= p
->static_prio
- bonus
;
688 if (prio
< MAX_RT_PRIO
)
690 if (prio
> MAX_PRIO
-1)
696 * To aid in avoiding the subversion of "niceness" due to uneven distribution
697 * of tasks with abnormal "nice" values across CPUs the contribution that
698 * each task makes to its run queue's load is weighted according to its
699 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
700 * scaled version of the new time slice allocation that they receive on time
705 * Assume: static_prio_timeslice(NICE_TO_PRIO(0)) == DEF_TIMESLICE
706 * If static_prio_timeslice() is ever changed to break this assumption then
707 * this code will need modification
709 #define TIME_SLICE_NICE_ZERO DEF_TIMESLICE
710 #define LOAD_WEIGHT(lp) \
711 (((lp) * SCHED_LOAD_SCALE) / TIME_SLICE_NICE_ZERO)
712 #define PRIO_TO_LOAD_WEIGHT(prio) \
713 LOAD_WEIGHT(static_prio_timeslice(prio))
714 #define RTPRIO_TO_LOAD_WEIGHT(rp) \
715 (PRIO_TO_LOAD_WEIGHT(MAX_RT_PRIO) + LOAD_WEIGHT(rp))
717 static void set_load_weight(task_t
*p
)
719 if (has_rt_policy(p
)) {
721 if (p
== task_rq(p
)->migration_thread
)
723 * The migration thread does the actual balancing.
724 * Giving its load any weight will skew balancing
730 p
->load_weight
= RTPRIO_TO_LOAD_WEIGHT(p
->rt_priority
);
732 p
->load_weight
= PRIO_TO_LOAD_WEIGHT(p
->static_prio
);
735 static inline void inc_raw_weighted_load(runqueue_t
*rq
, const task_t
*p
)
737 rq
->raw_weighted_load
+= p
->load_weight
;
740 static inline void dec_raw_weighted_load(runqueue_t
*rq
, const task_t
*p
)
742 rq
->raw_weighted_load
-= p
->load_weight
;
745 static inline void inc_nr_running(task_t
*p
, runqueue_t
*rq
)
748 inc_raw_weighted_load(rq
, p
);
751 static inline void dec_nr_running(task_t
*p
, runqueue_t
*rq
)
754 dec_raw_weighted_load(rq
, p
);
758 * Calculate the expected normal priority: i.e. priority
759 * without taking RT-inheritance into account. Might be
760 * boosted by interactivity modifiers. Changes upon fork,
761 * setprio syscalls, and whenever the interactivity
762 * estimator recalculates.
764 static inline int normal_prio(task_t
*p
)
768 if (has_rt_policy(p
))
769 prio
= MAX_RT_PRIO
-1 - p
->rt_priority
;
771 prio
= __normal_prio(p
);
776 * Calculate the current priority, i.e. the priority
777 * taken into account by the scheduler. This value might
778 * be boosted by RT tasks, or might be boosted by
779 * interactivity modifiers. Will be RT if the task got
780 * RT-boosted. If not then it returns p->normal_prio.
782 static int effective_prio(task_t
*p
)
784 p
->normal_prio
= normal_prio(p
);
786 * If we are RT tasks or we were boosted to RT priority,
787 * keep the priority unchanged. Otherwise, update priority
788 * to the normal priority:
790 if (!rt_prio(p
->prio
))
791 return p
->normal_prio
;
796 * __activate_task - move a task to the runqueue.
798 static void __activate_task(task_t
*p
, runqueue_t
*rq
)
800 prio_array_t
*target
= rq
->active
;
803 target
= rq
->expired
;
804 enqueue_task(p
, target
);
805 inc_nr_running(p
, rq
);
809 * __activate_idle_task - move idle task to the _front_ of runqueue.
811 static inline void __activate_idle_task(task_t
*p
, runqueue_t
*rq
)
813 enqueue_task_head(p
, rq
->active
);
814 inc_nr_running(p
, rq
);
818 * Recalculate p->normal_prio and p->prio after having slept,
819 * updating the sleep-average too:
821 static int recalc_task_prio(task_t
*p
, unsigned long long now
)
823 /* Caller must always ensure 'now >= p->timestamp' */
824 unsigned long sleep_time
= now
- p
->timestamp
;
829 if (likely(sleep_time
> 0)) {
831 * This ceiling is set to the lowest priority that would allow
832 * a task to be reinserted into the active array on timeslice
835 unsigned long ceiling
= INTERACTIVE_SLEEP(p
);
837 if (p
->mm
&& sleep_time
> ceiling
&& p
->sleep_avg
< ceiling
) {
839 * Prevents user tasks from achieving best priority
840 * with one single large enough sleep.
842 p
->sleep_avg
= ceiling
;
844 * Using INTERACTIVE_SLEEP() as a ceiling places a
845 * nice(0) task 1ms sleep away from promotion, and
846 * gives it 700ms to round-robin with no chance of
847 * being demoted. This is more than generous, so
848 * mark this sleep as non-interactive to prevent the
849 * on-runqueue bonus logic from intervening should
850 * this task not receive cpu immediately.
852 p
->sleep_type
= SLEEP_NONINTERACTIVE
;
855 * Tasks waking from uninterruptible sleep are
856 * limited in their sleep_avg rise as they
857 * are likely to be waiting on I/O
859 if (p
->sleep_type
== SLEEP_NONINTERACTIVE
&& p
->mm
) {
860 if (p
->sleep_avg
>= ceiling
)
862 else if (p
->sleep_avg
+ sleep_time
>=
864 p
->sleep_avg
= ceiling
;
870 * This code gives a bonus to interactive tasks.
872 * The boost works by updating the 'average sleep time'
873 * value here, based on ->timestamp. The more time a
874 * task spends sleeping, the higher the average gets -
875 * and the higher the priority boost gets as well.
877 p
->sleep_avg
+= sleep_time
;
880 if (p
->sleep_avg
> NS_MAX_SLEEP_AVG
)
881 p
->sleep_avg
= NS_MAX_SLEEP_AVG
;
884 return effective_prio(p
);
888 * activate_task - move a task to the runqueue and do priority recalculation
890 * Update all the scheduling statistics stuff. (sleep average
891 * calculation, priority modifiers, etc.)
893 static void activate_task(task_t
*p
, runqueue_t
*rq
, int local
)
895 unsigned long long now
;
900 /* Compensate for drifting sched_clock */
901 runqueue_t
*this_rq
= this_rq();
902 now
= (now
- this_rq
->timestamp_last_tick
)
903 + rq
->timestamp_last_tick
;
908 p
->prio
= recalc_task_prio(p
, now
);
911 * This checks to make sure it's not an uninterruptible task
912 * that is now waking up.
914 if (p
->sleep_type
== SLEEP_NORMAL
) {
916 * Tasks which were woken up by interrupts (ie. hw events)
917 * are most likely of interactive nature. So we give them
918 * the credit of extending their sleep time to the period
919 * of time they spend on the runqueue, waiting for execution
920 * on a CPU, first time around:
923 p
->sleep_type
= SLEEP_INTERRUPTED
;
926 * Normal first-time wakeups get a credit too for
927 * on-runqueue time, but it will be weighted down:
929 p
->sleep_type
= SLEEP_INTERACTIVE
;
934 __activate_task(p
, rq
);
938 * deactivate_task - remove a task from the runqueue.
940 static void deactivate_task(struct task_struct
*p
, runqueue_t
*rq
)
942 dec_nr_running(p
, rq
);
943 dequeue_task(p
, p
->array
);
948 * resched_task - mark a task 'to be rescheduled now'.
950 * On UP this means the setting of the need_resched flag, on SMP it
951 * might also involve a cross-CPU call to trigger the scheduler on
956 #ifndef tsk_is_polling
957 #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG)
960 static void resched_task(task_t
*p
)
964 assert_spin_locked(&task_rq(p
)->lock
);
966 if (unlikely(test_tsk_thread_flag(p
, TIF_NEED_RESCHED
)))
969 set_tsk_thread_flag(p
, TIF_NEED_RESCHED
);
972 if (cpu
== smp_processor_id())
975 /* NEED_RESCHED must be visible before we test polling */
977 if (!tsk_is_polling(p
))
978 smp_send_reschedule(cpu
);
981 static inline void resched_task(task_t
*p
)
983 assert_spin_locked(&task_rq(p
)->lock
);
984 set_tsk_need_resched(p
);
989 * task_curr - is this task currently executing on a CPU?
990 * @p: the task in question.
992 inline int task_curr(const task_t
*p
)
994 return cpu_curr(task_cpu(p
)) == p
;
997 /* Used instead of source_load when we know the type == 0 */
998 unsigned long weighted_cpuload(const int cpu
)
1000 return cpu_rq(cpu
)->raw_weighted_load
;
1005 struct list_head list
;
1010 struct completion done
;
1014 * The task's runqueue lock must be held.
1015 * Returns true if you have to wait for migration thread.
1017 static int migrate_task(task_t
*p
, int dest_cpu
, migration_req_t
*req
)
1019 runqueue_t
*rq
= task_rq(p
);
1022 * If the task is not on a runqueue (and not running), then
1023 * it is sufficient to simply update the task's cpu field.
1025 if (!p
->array
&& !task_running(rq
, p
)) {
1026 set_task_cpu(p
, dest_cpu
);
1030 init_completion(&req
->done
);
1032 req
->dest_cpu
= dest_cpu
;
1033 list_add(&req
->list
, &rq
->migration_queue
);
1038 * wait_task_inactive - wait for a thread to unschedule.
1040 * The caller must ensure that the task *will* unschedule sometime soon,
1041 * else this function might spin for a *long* time. This function can't
1042 * be called with interrupts off, or it may introduce deadlock with
1043 * smp_call_function() if an IPI is sent by the same process we are
1044 * waiting to become inactive.
1046 void wait_task_inactive(task_t
*p
)
1048 unsigned long flags
;
1053 rq
= task_rq_lock(p
, &flags
);
1054 /* Must be off runqueue entirely, not preempted. */
1055 if (unlikely(p
->array
|| task_running(rq
, p
))) {
1056 /* If it's preempted, we yield. It could be a while. */
1057 preempted
= !task_running(rq
, p
);
1058 task_rq_unlock(rq
, &flags
);
1064 task_rq_unlock(rq
, &flags
);
1068 * kick_process - kick a running thread to enter/exit the kernel
1069 * @p: the to-be-kicked thread
1071 * Cause a process which is running on another CPU to enter
1072 * kernel-mode, without any delay. (to get signals handled.)
1074 * NOTE: this function doesnt have to take the runqueue lock,
1075 * because all it wants to ensure is that the remote task enters
1076 * the kernel. If the IPI races and the task has been migrated
1077 * to another CPU then no harm is done and the purpose has been
1080 void kick_process(task_t
*p
)
1086 if ((cpu
!= smp_processor_id()) && task_curr(p
))
1087 smp_send_reschedule(cpu
);
1092 * Return a low guess at the load of a migration-source cpu weighted
1093 * according to the scheduling class and "nice" value.
1095 * We want to under-estimate the load of migration sources, to
1096 * balance conservatively.
1098 static inline unsigned long source_load(int cpu
, int type
)
1100 runqueue_t
*rq
= cpu_rq(cpu
);
1103 return rq
->raw_weighted_load
;
1105 return min(rq
->cpu_load
[type
-1], rq
->raw_weighted_load
);
1109 * Return a high guess at the load of a migration-target cpu weighted
1110 * according to the scheduling class and "nice" value.
1112 static inline unsigned long target_load(int cpu
, int type
)
1114 runqueue_t
*rq
= cpu_rq(cpu
);
1117 return rq
->raw_weighted_load
;
1119 return max(rq
->cpu_load
[type
-1], rq
->raw_weighted_load
);
1123 * Return the average load per task on the cpu's run queue
1125 static inline unsigned long cpu_avg_load_per_task(int cpu
)
1127 runqueue_t
*rq
= cpu_rq(cpu
);
1128 unsigned long n
= rq
->nr_running
;
1130 return n
? rq
->raw_weighted_load
/ n
: SCHED_LOAD_SCALE
;
1134 * find_idlest_group finds and returns the least busy CPU group within the
1137 static struct sched_group
*
1138 find_idlest_group(struct sched_domain
*sd
, struct task_struct
*p
, int this_cpu
)
1140 struct sched_group
*idlest
= NULL
, *this = NULL
, *group
= sd
->groups
;
1141 unsigned long min_load
= ULONG_MAX
, this_load
= 0;
1142 int load_idx
= sd
->forkexec_idx
;
1143 int imbalance
= 100 + (sd
->imbalance_pct
-100)/2;
1146 unsigned long load
, avg_load
;
1150 /* Skip over this group if it has no CPUs allowed */
1151 if (!cpus_intersects(group
->cpumask
, p
->cpus_allowed
))
1154 local_group
= cpu_isset(this_cpu
, group
->cpumask
);
1156 /* Tally up the load of all CPUs in the group */
1159 for_each_cpu_mask(i
, group
->cpumask
) {
1160 /* Bias balancing toward cpus of our domain */
1162 load
= source_load(i
, load_idx
);
1164 load
= target_load(i
, load_idx
);
1169 /* Adjust by relative CPU power of the group */
1170 avg_load
= (avg_load
* SCHED_LOAD_SCALE
) / group
->cpu_power
;
1173 this_load
= avg_load
;
1175 } else if (avg_load
< min_load
) {
1176 min_load
= avg_load
;
1180 group
= group
->next
;
1181 } while (group
!= sd
->groups
);
1183 if (!idlest
|| 100*this_load
< imbalance
*min_load
)
1189 * find_idlest_queue - find the idlest runqueue among the cpus in group.
1192 find_idlest_cpu(struct sched_group
*group
, struct task_struct
*p
, int this_cpu
)
1195 unsigned long load
, min_load
= ULONG_MAX
;
1199 /* Traverse only the allowed CPUs */
1200 cpus_and(tmp
, group
->cpumask
, p
->cpus_allowed
);
1202 for_each_cpu_mask(i
, tmp
) {
1203 load
= weighted_cpuload(i
);
1205 if (load
< min_load
|| (load
== min_load
&& i
== this_cpu
)) {
1215 * sched_balance_self: balance the current task (running on cpu) in domains
1216 * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and
1219 * Balance, ie. select the least loaded group.
1221 * Returns the target CPU number, or the same CPU if no balancing is needed.
1223 * preempt must be disabled.
1225 static int sched_balance_self(int cpu
, int flag
)
1227 struct task_struct
*t
= current
;
1228 struct sched_domain
*tmp
, *sd
= NULL
;
1230 for_each_domain(cpu
, tmp
) {
1232 * If power savings logic is enabled for a domain, stop there.
1234 if (tmp
->flags
& SD_POWERSAVINGS_BALANCE
)
1236 if (tmp
->flags
& flag
)
1242 struct sched_group
*group
;
1247 group
= find_idlest_group(sd
, t
, cpu
);
1251 new_cpu
= find_idlest_cpu(group
, t
, cpu
);
1252 if (new_cpu
== -1 || new_cpu
== cpu
)
1255 /* Now try balancing at a lower domain level */
1259 weight
= cpus_weight(span
);
1260 for_each_domain(cpu
, tmp
) {
1261 if (weight
<= cpus_weight(tmp
->span
))
1263 if (tmp
->flags
& flag
)
1266 /* while loop will break here if sd == NULL */
1272 #endif /* CONFIG_SMP */
1275 * wake_idle() will wake a task on an idle cpu if task->cpu is
1276 * not idle and an idle cpu is available. The span of cpus to
1277 * search starts with cpus closest then further out as needed,
1278 * so we always favor a closer, idle cpu.
1280 * Returns the CPU we should wake onto.
1282 #if defined(ARCH_HAS_SCHED_WAKE_IDLE)
1283 static int wake_idle(int cpu
, task_t
*p
)
1286 struct sched_domain
*sd
;
1292 for_each_domain(cpu
, sd
) {
1293 if (sd
->flags
& SD_WAKE_IDLE
) {
1294 cpus_and(tmp
, sd
->span
, p
->cpus_allowed
);
1295 for_each_cpu_mask(i
, tmp
) {
1306 static inline int wake_idle(int cpu
, task_t
*p
)
1313 * try_to_wake_up - wake up a thread
1314 * @p: the to-be-woken-up thread
1315 * @state: the mask of task states that can be woken
1316 * @sync: do a synchronous wakeup?
1318 * Put it on the run-queue if it's not already there. The "current"
1319 * thread is always on the run-queue (except when the actual
1320 * re-schedule is in progress), and as such you're allowed to do
1321 * the simpler "current->state = TASK_RUNNING" to mark yourself
1322 * runnable without the overhead of this.
1324 * returns failure only if the task is already active.
1326 static int try_to_wake_up(task_t
*p
, unsigned int state
, int sync
)
1328 int cpu
, this_cpu
, success
= 0;
1329 unsigned long flags
;
1333 unsigned long load
, this_load
;
1334 struct sched_domain
*sd
, *this_sd
= NULL
;
1338 rq
= task_rq_lock(p
, &flags
);
1339 old_state
= p
->state
;
1340 if (!(old_state
& state
))
1347 this_cpu
= smp_processor_id();
1350 if (unlikely(task_running(rq
, p
)))
1355 schedstat_inc(rq
, ttwu_cnt
);
1356 if (cpu
== this_cpu
) {
1357 schedstat_inc(rq
, ttwu_local
);
1361 for_each_domain(this_cpu
, sd
) {
1362 if (cpu_isset(cpu
, sd
->span
)) {
1363 schedstat_inc(sd
, ttwu_wake_remote
);
1369 if (unlikely(!cpu_isset(this_cpu
, p
->cpus_allowed
)))
1373 * Check for affine wakeup and passive balancing possibilities.
1376 int idx
= this_sd
->wake_idx
;
1377 unsigned int imbalance
;
1379 imbalance
= 100 + (this_sd
->imbalance_pct
- 100) / 2;
1381 load
= source_load(cpu
, idx
);
1382 this_load
= target_load(this_cpu
, idx
);
1384 new_cpu
= this_cpu
; /* Wake to this CPU if we can */
1386 if (this_sd
->flags
& SD_WAKE_AFFINE
) {
1387 unsigned long tl
= this_load
;
1388 unsigned long tl_per_task
= cpu_avg_load_per_task(this_cpu
);
1391 * If sync wakeup then subtract the (maximum possible)
1392 * effect of the currently running task from the load
1393 * of the current CPU:
1396 tl
-= current
->load_weight
;
1399 tl
+ target_load(cpu
, idx
) <= tl_per_task
) ||
1400 100*(tl
+ p
->load_weight
) <= imbalance
*load
) {
1402 * This domain has SD_WAKE_AFFINE and
1403 * p is cache cold in this domain, and
1404 * there is no bad imbalance.
1406 schedstat_inc(this_sd
, ttwu_move_affine
);
1412 * Start passive balancing when half the imbalance_pct
1415 if (this_sd
->flags
& SD_WAKE_BALANCE
) {
1416 if (imbalance
*this_load
<= 100*load
) {
1417 schedstat_inc(this_sd
, ttwu_move_balance
);
1423 new_cpu
= cpu
; /* Could not wake to this_cpu. Wake to cpu instead */
1425 new_cpu
= wake_idle(new_cpu
, p
);
1426 if (new_cpu
!= cpu
) {
1427 set_task_cpu(p
, new_cpu
);
1428 task_rq_unlock(rq
, &flags
);
1429 /* might preempt at this point */
1430 rq
= task_rq_lock(p
, &flags
);
1431 old_state
= p
->state
;
1432 if (!(old_state
& state
))
1437 this_cpu
= smp_processor_id();
1442 #endif /* CONFIG_SMP */
1443 if (old_state
== TASK_UNINTERRUPTIBLE
) {
1444 rq
->nr_uninterruptible
--;
1446 * Tasks on involuntary sleep don't earn
1447 * sleep_avg beyond just interactive state.
1449 p
->sleep_type
= SLEEP_NONINTERACTIVE
;
1453 * Tasks that have marked their sleep as noninteractive get
1454 * woken up with their sleep average not weighted in an
1457 if (old_state
& TASK_NONINTERACTIVE
)
1458 p
->sleep_type
= SLEEP_NONINTERACTIVE
;
1461 activate_task(p
, rq
, cpu
== this_cpu
);
1463 * Sync wakeups (i.e. those types of wakeups where the waker
1464 * has indicated that it will leave the CPU in short order)
1465 * don't trigger a preemption, if the woken up task will run on
1466 * this cpu. (in this case the 'I will reschedule' promise of
1467 * the waker guarantees that the freshly woken up task is going
1468 * to be considered on this CPU.)
1470 if (!sync
|| cpu
!= this_cpu
) {
1471 if (TASK_PREEMPTS_CURR(p
, rq
))
1472 resched_task(rq
->curr
);
1477 p
->state
= TASK_RUNNING
;
1479 task_rq_unlock(rq
, &flags
);
1484 int fastcall
wake_up_process(task_t
*p
)
1486 return try_to_wake_up(p
, TASK_STOPPED
| TASK_TRACED
|
1487 TASK_INTERRUPTIBLE
| TASK_UNINTERRUPTIBLE
, 0);
1490 EXPORT_SYMBOL(wake_up_process
);
1492 int fastcall
wake_up_state(task_t
*p
, unsigned int state
)
1494 return try_to_wake_up(p
, state
, 0);
1498 * Perform scheduler related setup for a newly forked process p.
1499 * p is forked by current.
1501 void fastcall
sched_fork(task_t
*p
, int clone_flags
)
1503 int cpu
= get_cpu();
1506 cpu
= sched_balance_self(cpu
, SD_BALANCE_FORK
);
1508 set_task_cpu(p
, cpu
);
1511 * We mark the process as running here, but have not actually
1512 * inserted it onto the runqueue yet. This guarantees that
1513 * nobody will actually run it, and a signal or other external
1514 * event cannot wake it up and insert it on the runqueue either.
1516 p
->state
= TASK_RUNNING
;
1519 * Make sure we do not leak PI boosting priority to the child:
1521 p
->prio
= current
->normal_prio
;
1523 INIT_LIST_HEAD(&p
->run_list
);
1525 #ifdef CONFIG_SCHEDSTATS
1526 memset(&p
->sched_info
, 0, sizeof(p
->sched_info
));
1528 #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW)
1531 #ifdef CONFIG_PREEMPT
1532 /* Want to start with kernel preemption disabled. */
1533 task_thread_info(p
)->preempt_count
= 1;
1536 * Share the timeslice between parent and child, thus the
1537 * total amount of pending timeslices in the system doesn't change,
1538 * resulting in more scheduling fairness.
1540 local_irq_disable();
1541 p
->time_slice
= (current
->time_slice
+ 1) >> 1;
1543 * The remainder of the first timeslice might be recovered by
1544 * the parent if the child exits early enough.
1546 p
->first_time_slice
= 1;
1547 current
->time_slice
>>= 1;
1548 p
->timestamp
= sched_clock();
1549 if (unlikely(!current
->time_slice
)) {
1551 * This case is rare, it happens when the parent has only
1552 * a single jiffy left from its timeslice. Taking the
1553 * runqueue lock is not a problem.
1555 current
->time_slice
= 1;
1563 * wake_up_new_task - wake up a newly created task for the first time.
1565 * This function will do some initial scheduler statistics housekeeping
1566 * that must be done for every newly created context, then puts the task
1567 * on the runqueue and wakes it.
1569 void fastcall
wake_up_new_task(task_t
*p
, unsigned long clone_flags
)
1571 unsigned long flags
;
1573 runqueue_t
*rq
, *this_rq
;
1575 rq
= task_rq_lock(p
, &flags
);
1576 BUG_ON(p
->state
!= TASK_RUNNING
);
1577 this_cpu
= smp_processor_id();
1581 * We decrease the sleep average of forking parents
1582 * and children as well, to keep max-interactive tasks
1583 * from forking tasks that are max-interactive. The parent
1584 * (current) is done further down, under its lock.
1586 p
->sleep_avg
= JIFFIES_TO_NS(CURRENT_BONUS(p
) *
1587 CHILD_PENALTY
/ 100 * MAX_SLEEP_AVG
/ MAX_BONUS
);
1589 p
->prio
= effective_prio(p
);
1591 if (likely(cpu
== this_cpu
)) {
1592 if (!(clone_flags
& CLONE_VM
)) {
1594 * The VM isn't cloned, so we're in a good position to
1595 * do child-runs-first in anticipation of an exec. This
1596 * usually avoids a lot of COW overhead.
1598 if (unlikely(!current
->array
))
1599 __activate_task(p
, rq
);
1601 p
->prio
= current
->prio
;
1602 p
->normal_prio
= current
->normal_prio
;
1603 list_add_tail(&p
->run_list
, ¤t
->run_list
);
1604 p
->array
= current
->array
;
1605 p
->array
->nr_active
++;
1606 inc_nr_running(p
, rq
);
1610 /* Run child last */
1611 __activate_task(p
, rq
);
1613 * We skip the following code due to cpu == this_cpu
1615 * task_rq_unlock(rq, &flags);
1616 * this_rq = task_rq_lock(current, &flags);
1620 this_rq
= cpu_rq(this_cpu
);
1623 * Not the local CPU - must adjust timestamp. This should
1624 * get optimised away in the !CONFIG_SMP case.
1626 p
->timestamp
= (p
->timestamp
- this_rq
->timestamp_last_tick
)
1627 + rq
->timestamp_last_tick
;
1628 __activate_task(p
, rq
);
1629 if (TASK_PREEMPTS_CURR(p
, rq
))
1630 resched_task(rq
->curr
);
1633 * Parent and child are on different CPUs, now get the
1634 * parent runqueue to update the parent's ->sleep_avg:
1636 task_rq_unlock(rq
, &flags
);
1637 this_rq
= task_rq_lock(current
, &flags
);
1639 current
->sleep_avg
= JIFFIES_TO_NS(CURRENT_BONUS(current
) *
1640 PARENT_PENALTY
/ 100 * MAX_SLEEP_AVG
/ MAX_BONUS
);
1641 task_rq_unlock(this_rq
, &flags
);
1645 * Potentially available exiting-child timeslices are
1646 * retrieved here - this way the parent does not get
1647 * penalized for creating too many threads.
1649 * (this cannot be used to 'generate' timeslices
1650 * artificially, because any timeslice recovered here
1651 * was given away by the parent in the first place.)
1653 void fastcall
sched_exit(task_t
*p
)
1655 unsigned long flags
;
1659 * If the child was a (relative-) CPU hog then decrease
1660 * the sleep_avg of the parent as well.
1662 rq
= task_rq_lock(p
->parent
, &flags
);
1663 if (p
->first_time_slice
&& task_cpu(p
) == task_cpu(p
->parent
)) {
1664 p
->parent
->time_slice
+= p
->time_slice
;
1665 if (unlikely(p
->parent
->time_slice
> task_timeslice(p
)))
1666 p
->parent
->time_slice
= task_timeslice(p
);
1668 if (p
->sleep_avg
< p
->parent
->sleep_avg
)
1669 p
->parent
->sleep_avg
= p
->parent
->sleep_avg
/
1670 (EXIT_WEIGHT
+ 1) * EXIT_WEIGHT
+ p
->sleep_avg
/
1672 task_rq_unlock(rq
, &flags
);
1676 * prepare_task_switch - prepare to switch tasks
1677 * @rq: the runqueue preparing to switch
1678 * @next: the task we are going to switch to.
1680 * This is called with the rq lock held and interrupts off. It must
1681 * be paired with a subsequent finish_task_switch after the context
1684 * prepare_task_switch sets up locking and calls architecture specific
1687 static inline void prepare_task_switch(runqueue_t
*rq
, task_t
*next
)
1689 prepare_lock_switch(rq
, next
);
1690 prepare_arch_switch(next
);
1694 * finish_task_switch - clean up after a task-switch
1695 * @rq: runqueue associated with task-switch
1696 * @prev: the thread we just switched away from.
1698 * finish_task_switch must be called after the context switch, paired
1699 * with a prepare_task_switch call before the context switch.
1700 * finish_task_switch will reconcile locking set up by prepare_task_switch,
1701 * and do any other architecture-specific cleanup actions.
1703 * Note that we may have delayed dropping an mm in context_switch(). If
1704 * so, we finish that here outside of the runqueue lock. (Doing it
1705 * with the lock held can cause deadlocks; see schedule() for
1708 static inline void finish_task_switch(runqueue_t
*rq
, task_t
*prev
)
1709 __releases(rq
->lock
)
1711 struct mm_struct
*mm
= rq
->prev_mm
;
1712 unsigned long prev_task_flags
;
1717 * A task struct has one reference for the use as "current".
1718 * If a task dies, then it sets EXIT_ZOMBIE in tsk->exit_state and
1719 * calls schedule one last time. The schedule call will never return,
1720 * and the scheduled task must drop that reference.
1721 * The test for EXIT_ZOMBIE must occur while the runqueue locks are
1722 * still held, otherwise prev could be scheduled on another cpu, die
1723 * there before we look at prev->state, and then the reference would
1725 * Manfred Spraul <manfred@colorfullife.com>
1727 prev_task_flags
= prev
->flags
;
1728 finish_arch_switch(prev
);
1729 finish_lock_switch(rq
, prev
);
1732 if (unlikely(prev_task_flags
& PF_DEAD
)) {
1734 * Remove function-return probe instances associated with this
1735 * task and put them back on the free list.
1737 kprobe_flush_task(prev
);
1738 put_task_struct(prev
);
1743 * schedule_tail - first thing a freshly forked thread must call.
1744 * @prev: the thread we just switched away from.
1746 asmlinkage
void schedule_tail(task_t
*prev
)
1747 __releases(rq
->lock
)
1749 runqueue_t
*rq
= this_rq();
1750 finish_task_switch(rq
, prev
);
1751 #ifdef __ARCH_WANT_UNLOCKED_CTXSW
1752 /* In this case, finish_task_switch does not reenable preemption */
1755 if (current
->set_child_tid
)
1756 put_user(current
->pid
, current
->set_child_tid
);
1760 * context_switch - switch to the new MM and the new
1761 * thread's register state.
1764 task_t
* context_switch(runqueue_t
*rq
, task_t
*prev
, task_t
*next
)
1766 struct mm_struct
*mm
= next
->mm
;
1767 struct mm_struct
*oldmm
= prev
->active_mm
;
1769 if (unlikely(!mm
)) {
1770 next
->active_mm
= oldmm
;
1771 atomic_inc(&oldmm
->mm_count
);
1772 enter_lazy_tlb(oldmm
, next
);
1774 switch_mm(oldmm
, mm
, next
);
1776 if (unlikely(!prev
->mm
)) {
1777 prev
->active_mm
= NULL
;
1778 WARN_ON(rq
->prev_mm
);
1779 rq
->prev_mm
= oldmm
;
1782 /* Here we just switch the register state and the stack. */
1783 switch_to(prev
, next
, prev
);
1789 * nr_running, nr_uninterruptible and nr_context_switches:
1791 * externally visible scheduler statistics: current number of runnable
1792 * threads, current number of uninterruptible-sleeping threads, total
1793 * number of context switches performed since bootup.
1795 unsigned long nr_running(void)
1797 unsigned long i
, sum
= 0;
1799 for_each_online_cpu(i
)
1800 sum
+= cpu_rq(i
)->nr_running
;
1805 unsigned long nr_uninterruptible(void)
1807 unsigned long i
, sum
= 0;
1809 for_each_possible_cpu(i
)
1810 sum
+= cpu_rq(i
)->nr_uninterruptible
;
1813 * Since we read the counters lockless, it might be slightly
1814 * inaccurate. Do not allow it to go below zero though:
1816 if (unlikely((long)sum
< 0))
1822 unsigned long long nr_context_switches(void)
1825 unsigned long long sum
= 0;
1827 for_each_possible_cpu(i
)
1828 sum
+= cpu_rq(i
)->nr_switches
;
1833 unsigned long nr_iowait(void)
1835 unsigned long i
, sum
= 0;
1837 for_each_possible_cpu(i
)
1838 sum
+= atomic_read(&cpu_rq(i
)->nr_iowait
);
1843 unsigned long nr_active(void)
1845 unsigned long i
, running
= 0, uninterruptible
= 0;
1847 for_each_online_cpu(i
) {
1848 running
+= cpu_rq(i
)->nr_running
;
1849 uninterruptible
+= cpu_rq(i
)->nr_uninterruptible
;
1852 if (unlikely((long)uninterruptible
< 0))
1853 uninterruptible
= 0;
1855 return running
+ uninterruptible
;
1861 * double_rq_lock - safely lock two runqueues
1863 * Note this does not disable interrupts like task_rq_lock,
1864 * you need to do so manually before calling.
1866 static void double_rq_lock(runqueue_t
*rq1
, runqueue_t
*rq2
)
1867 __acquires(rq1
->lock
)
1868 __acquires(rq2
->lock
)
1871 spin_lock(&rq1
->lock
);
1872 __acquire(rq2
->lock
); /* Fake it out ;) */
1875 spin_lock(&rq1
->lock
);
1876 spin_lock(&rq2
->lock
);
1878 spin_lock(&rq2
->lock
);
1879 spin_lock(&rq1
->lock
);
1885 * double_rq_unlock - safely unlock two runqueues
1887 * Note this does not restore interrupts like task_rq_unlock,
1888 * you need to do so manually after calling.
1890 static void double_rq_unlock(runqueue_t
*rq1
, runqueue_t
*rq2
)
1891 __releases(rq1
->lock
)
1892 __releases(rq2
->lock
)
1894 spin_unlock(&rq1
->lock
);
1896 spin_unlock(&rq2
->lock
);
1898 __release(rq2
->lock
);
1902 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1904 static void double_lock_balance(runqueue_t
*this_rq
, runqueue_t
*busiest
)
1905 __releases(this_rq
->lock
)
1906 __acquires(busiest
->lock
)
1907 __acquires(this_rq
->lock
)
1909 if (unlikely(!spin_trylock(&busiest
->lock
))) {
1910 if (busiest
< this_rq
) {
1911 spin_unlock(&this_rq
->lock
);
1912 spin_lock(&busiest
->lock
);
1913 spin_lock(&this_rq
->lock
);
1915 spin_lock(&busiest
->lock
);
1920 * If dest_cpu is allowed for this process, migrate the task to it.
1921 * This is accomplished by forcing the cpu_allowed mask to only
1922 * allow dest_cpu, which will force the cpu onto dest_cpu. Then
1923 * the cpu_allowed mask is restored.
1925 static void sched_migrate_task(task_t
*p
, int dest_cpu
)
1927 migration_req_t req
;
1929 unsigned long flags
;
1931 rq
= task_rq_lock(p
, &flags
);
1932 if (!cpu_isset(dest_cpu
, p
->cpus_allowed
)
1933 || unlikely(cpu_is_offline(dest_cpu
)))
1936 /* force the process onto the specified CPU */
1937 if (migrate_task(p
, dest_cpu
, &req
)) {
1938 /* Need to wait for migration thread (might exit: take ref). */
1939 struct task_struct
*mt
= rq
->migration_thread
;
1940 get_task_struct(mt
);
1941 task_rq_unlock(rq
, &flags
);
1942 wake_up_process(mt
);
1943 put_task_struct(mt
);
1944 wait_for_completion(&req
.done
);
1948 task_rq_unlock(rq
, &flags
);
1952 * sched_exec - execve() is a valuable balancing opportunity, because at
1953 * this point the task has the smallest effective memory and cache footprint.
1955 void sched_exec(void)
1957 int new_cpu
, this_cpu
= get_cpu();
1958 new_cpu
= sched_balance_self(this_cpu
, SD_BALANCE_EXEC
);
1960 if (new_cpu
!= this_cpu
)
1961 sched_migrate_task(current
, new_cpu
);
1965 * pull_task - move a task from a remote runqueue to the local runqueue.
1966 * Both runqueues must be locked.
1969 void pull_task(runqueue_t
*src_rq
, prio_array_t
*src_array
, task_t
*p
,
1970 runqueue_t
*this_rq
, prio_array_t
*this_array
, int this_cpu
)
1972 dequeue_task(p
, src_array
);
1973 dec_nr_running(p
, src_rq
);
1974 set_task_cpu(p
, this_cpu
);
1975 inc_nr_running(p
, this_rq
);
1976 enqueue_task(p
, this_array
);
1977 p
->timestamp
= (p
->timestamp
- src_rq
->timestamp_last_tick
)
1978 + this_rq
->timestamp_last_tick
;
1980 * Note that idle threads have a prio of MAX_PRIO, for this test
1981 * to be always true for them.
1983 if (TASK_PREEMPTS_CURR(p
, this_rq
))
1984 resched_task(this_rq
->curr
);
1988 * can_migrate_task - may task p from runqueue rq be migrated to this_cpu?
1991 int can_migrate_task(task_t
*p
, runqueue_t
*rq
, int this_cpu
,
1992 struct sched_domain
*sd
, enum idle_type idle
,
1996 * We do not migrate tasks that are:
1997 * 1) running (obviously), or
1998 * 2) cannot be migrated to this CPU due to cpus_allowed, or
1999 * 3) are cache-hot on their current CPU.
2001 if (!cpu_isset(this_cpu
, p
->cpus_allowed
))
2005 if (task_running(rq
, p
))
2009 * Aggressive migration if:
2010 * 1) task is cache cold, or
2011 * 2) too many balance attempts have failed.
2014 if (sd
->nr_balance_failed
> sd
->cache_nice_tries
)
2017 if (task_hot(p
, rq
->timestamp_last_tick
, sd
))
2022 #define rq_best_prio(rq) min((rq)->curr->prio, (rq)->best_expired_prio)
2024 * move_tasks tries to move up to max_nr_move tasks and max_load_move weighted
2025 * load from busiest to this_rq, as part of a balancing operation within
2026 * "domain". Returns the number of tasks moved.
2028 * Called with both runqueues locked.
2030 static int move_tasks(runqueue_t
*this_rq
, int this_cpu
, runqueue_t
*busiest
,
2031 unsigned long max_nr_move
, unsigned long max_load_move
,
2032 struct sched_domain
*sd
, enum idle_type idle
,
2035 prio_array_t
*array
, *dst_array
;
2036 struct list_head
*head
, *curr
;
2037 int idx
, pulled
= 0, pinned
= 0, this_best_prio
, busiest_best_prio
;
2038 int busiest_best_prio_seen
;
2039 int skip_for_load
; /* skip the task based on weighted load issues */
2043 if (max_nr_move
== 0 || max_load_move
== 0)
2046 rem_load_move
= max_load_move
;
2048 this_best_prio
= rq_best_prio(this_rq
);
2049 busiest_best_prio
= rq_best_prio(busiest
);
2051 * Enable handling of the case where there is more than one task
2052 * with the best priority. If the current running task is one
2053 * of those with prio==busiest_best_prio we know it won't be moved
2054 * and therefore it's safe to override the skip (based on load) of
2055 * any task we find with that prio.
2057 busiest_best_prio_seen
= busiest_best_prio
== busiest
->curr
->prio
;
2060 * We first consider expired tasks. Those will likely not be
2061 * executed in the near future, and they are most likely to
2062 * be cache-cold, thus switching CPUs has the least effect
2065 if (busiest
->expired
->nr_active
) {
2066 array
= busiest
->expired
;
2067 dst_array
= this_rq
->expired
;
2069 array
= busiest
->active
;
2070 dst_array
= this_rq
->active
;
2074 /* Start searching at priority 0: */
2078 idx
= sched_find_first_bit(array
->bitmap
);
2080 idx
= find_next_bit(array
->bitmap
, MAX_PRIO
, idx
);
2081 if (idx
>= MAX_PRIO
) {
2082 if (array
== busiest
->expired
&& busiest
->active
->nr_active
) {
2083 array
= busiest
->active
;
2084 dst_array
= this_rq
->active
;
2090 head
= array
->queue
+ idx
;
2093 tmp
= list_entry(curr
, task_t
, run_list
);
2098 * To help distribute high priority tasks accross CPUs we don't
2099 * skip a task if it will be the highest priority task (i.e. smallest
2100 * prio value) on its new queue regardless of its load weight
2102 skip_for_load
= tmp
->load_weight
> rem_load_move
;
2103 if (skip_for_load
&& idx
< this_best_prio
)
2104 skip_for_load
= !busiest_best_prio_seen
&& idx
== busiest_best_prio
;
2105 if (skip_for_load
||
2106 !can_migrate_task(tmp
, busiest
, this_cpu
, sd
, idle
, &pinned
)) {
2107 busiest_best_prio_seen
|= idx
== busiest_best_prio
;
2114 #ifdef CONFIG_SCHEDSTATS
2115 if (task_hot(tmp
, busiest
->timestamp_last_tick
, sd
))
2116 schedstat_inc(sd
, lb_hot_gained
[idle
]);
2119 pull_task(busiest
, array
, tmp
, this_rq
, dst_array
, this_cpu
);
2121 rem_load_move
-= tmp
->load_weight
;
2124 * We only want to steal up to the prescribed number of tasks
2125 * and the prescribed amount of weighted load.
2127 if (pulled
< max_nr_move
&& rem_load_move
> 0) {
2128 if (idx
< this_best_prio
)
2129 this_best_prio
= idx
;
2137 * Right now, this is the only place pull_task() is called,
2138 * so we can safely collect pull_task() stats here rather than
2139 * inside pull_task().
2141 schedstat_add(sd
, lb_gained
[idle
], pulled
);
2144 *all_pinned
= pinned
;
2149 * find_busiest_group finds and returns the busiest CPU group within the
2150 * domain. It calculates and returns the amount of weighted load which should be
2151 * moved to restore balance via the imbalance parameter.
2153 static struct sched_group
*
2154 find_busiest_group(struct sched_domain
*sd
, int this_cpu
,
2155 unsigned long *imbalance
, enum idle_type idle
, int *sd_idle
)
2157 struct sched_group
*busiest
= NULL
, *this = NULL
, *group
= sd
->groups
;
2158 unsigned long max_load
, avg_load
, total_load
, this_load
, total_pwr
;
2159 unsigned long max_pull
;
2160 unsigned long busiest_load_per_task
, busiest_nr_running
;
2161 unsigned long this_load_per_task
, this_nr_running
;
2163 #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT)
2164 int power_savings_balance
= 1;
2165 unsigned long leader_nr_running
= 0, min_load_per_task
= 0;
2166 unsigned long min_nr_running
= ULONG_MAX
;
2167 struct sched_group
*group_min
= NULL
, *group_leader
= NULL
;
2170 max_load
= this_load
= total_load
= total_pwr
= 0;
2171 busiest_load_per_task
= busiest_nr_running
= 0;
2172 this_load_per_task
= this_nr_running
= 0;
2173 if (idle
== NOT_IDLE
)
2174 load_idx
= sd
->busy_idx
;
2175 else if (idle
== NEWLY_IDLE
)
2176 load_idx
= sd
->newidle_idx
;
2178 load_idx
= sd
->idle_idx
;
2181 unsigned long load
, group_capacity
;
2184 unsigned long sum_nr_running
, sum_weighted_load
;
2186 local_group
= cpu_isset(this_cpu
, group
->cpumask
);
2188 /* Tally up the load of all CPUs in the group */
2189 sum_weighted_load
= sum_nr_running
= avg_load
= 0;
2191 for_each_cpu_mask(i
, group
->cpumask
) {
2192 runqueue_t
*rq
= cpu_rq(i
);
2194 if (*sd_idle
&& !idle_cpu(i
))
2197 /* Bias balancing toward cpus of our domain */
2199 load
= target_load(i
, load_idx
);
2201 load
= source_load(i
, load_idx
);
2204 sum_nr_running
+= rq
->nr_running
;
2205 sum_weighted_load
+= rq
->raw_weighted_load
;
2208 total_load
+= avg_load
;
2209 total_pwr
+= group
->cpu_power
;
2211 /* Adjust by relative CPU power of the group */
2212 avg_load
= (avg_load
* SCHED_LOAD_SCALE
) / group
->cpu_power
;
2214 group_capacity
= group
->cpu_power
/ SCHED_LOAD_SCALE
;
2217 this_load
= avg_load
;
2219 this_nr_running
= sum_nr_running
;
2220 this_load_per_task
= sum_weighted_load
;
2221 } else if (avg_load
> max_load
&&
2222 sum_nr_running
> group_capacity
) {
2223 max_load
= avg_load
;
2225 busiest_nr_running
= sum_nr_running
;
2226 busiest_load_per_task
= sum_weighted_load
;
2229 #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT)
2231 * Busy processors will not participate in power savings
2234 if (idle
== NOT_IDLE
|| !(sd
->flags
& SD_POWERSAVINGS_BALANCE
))
2238 * If the local group is idle or completely loaded
2239 * no need to do power savings balance at this domain
2241 if (local_group
&& (this_nr_running
>= group_capacity
||
2243 power_savings_balance
= 0;
2246 * If a group is already running at full capacity or idle,
2247 * don't include that group in power savings calculations
2249 if (!power_savings_balance
|| sum_nr_running
>= group_capacity
2254 * Calculate the group which has the least non-idle load.
2255 * This is the group from where we need to pick up the load
2258 if ((sum_nr_running
< min_nr_running
) ||
2259 (sum_nr_running
== min_nr_running
&&
2260 first_cpu(group
->cpumask
) <
2261 first_cpu(group_min
->cpumask
))) {
2263 min_nr_running
= sum_nr_running
;
2264 min_load_per_task
= sum_weighted_load
/
2269 * Calculate the group which is almost near its
2270 * capacity but still has some space to pick up some load
2271 * from other group and save more power
2273 if (sum_nr_running
<= group_capacity
- 1)
2274 if (sum_nr_running
> leader_nr_running
||
2275 (sum_nr_running
== leader_nr_running
&&
2276 first_cpu(group
->cpumask
) >
2277 first_cpu(group_leader
->cpumask
))) {
2278 group_leader
= group
;
2279 leader_nr_running
= sum_nr_running
;
2284 group
= group
->next
;
2285 } while (group
!= sd
->groups
);
2287 if (!busiest
|| this_load
>= max_load
|| busiest_nr_running
== 0)
2290 avg_load
= (SCHED_LOAD_SCALE
* total_load
) / total_pwr
;
2292 if (this_load
>= avg_load
||
2293 100*max_load
<= sd
->imbalance_pct
*this_load
)
2296 busiest_load_per_task
/= busiest_nr_running
;
2298 * We're trying to get all the cpus to the average_load, so we don't
2299 * want to push ourselves above the average load, nor do we wish to
2300 * reduce the max loaded cpu below the average load, as either of these
2301 * actions would just result in more rebalancing later, and ping-pong
2302 * tasks around. Thus we look for the minimum possible imbalance.
2303 * Negative imbalances (*we* are more loaded than anyone else) will
2304 * be counted as no imbalance for these purposes -- we can't fix that
2305 * by pulling tasks to us. Be careful of negative numbers as they'll
2306 * appear as very large values with unsigned longs.
2308 if (max_load
<= busiest_load_per_task
)
2312 * In the presence of smp nice balancing, certain scenarios can have
2313 * max load less than avg load(as we skip the groups at or below
2314 * its cpu_power, while calculating max_load..)
2316 if (max_load
< avg_load
) {
2318 goto small_imbalance
;
2321 /* Don't want to pull so many tasks that a group would go idle */
2322 max_pull
= min(max_load
- avg_load
, max_load
- busiest_load_per_task
);
2324 /* How much load to actually move to equalise the imbalance */
2325 *imbalance
= min(max_pull
* busiest
->cpu_power
,
2326 (avg_load
- this_load
) * this->cpu_power
)
2330 * if *imbalance is less than the average load per runnable task
2331 * there is no gaurantee that any tasks will be moved so we'll have
2332 * a think about bumping its value to force at least one task to be
2335 if (*imbalance
< busiest_load_per_task
) {
2336 unsigned long pwr_now
, pwr_move
;
2341 pwr_move
= pwr_now
= 0;
2343 if (this_nr_running
) {
2344 this_load_per_task
/= this_nr_running
;
2345 if (busiest_load_per_task
> this_load_per_task
)
2348 this_load_per_task
= SCHED_LOAD_SCALE
;
2350 if (max_load
- this_load
>= busiest_load_per_task
* imbn
) {
2351 *imbalance
= busiest_load_per_task
;
2356 * OK, we don't have enough imbalance to justify moving tasks,
2357 * however we may be able to increase total CPU power used by
2361 pwr_now
+= busiest
->cpu_power
*
2362 min(busiest_load_per_task
, max_load
);
2363 pwr_now
+= this->cpu_power
*
2364 min(this_load_per_task
, this_load
);
2365 pwr_now
/= SCHED_LOAD_SCALE
;
2367 /* Amount of load we'd subtract */
2368 tmp
= busiest_load_per_task
*SCHED_LOAD_SCALE
/busiest
->cpu_power
;
2370 pwr_move
+= busiest
->cpu_power
*
2371 min(busiest_load_per_task
, max_load
- tmp
);
2373 /* Amount of load we'd add */
2374 if (max_load
*busiest
->cpu_power
<
2375 busiest_load_per_task
*SCHED_LOAD_SCALE
)
2376 tmp
= max_load
*busiest
->cpu_power
/this->cpu_power
;
2378 tmp
= busiest_load_per_task
*SCHED_LOAD_SCALE
/this->cpu_power
;
2379 pwr_move
+= this->cpu_power
*min(this_load_per_task
, this_load
+ tmp
);
2380 pwr_move
/= SCHED_LOAD_SCALE
;
2382 /* Move if we gain throughput */
2383 if (pwr_move
<= pwr_now
)
2386 *imbalance
= busiest_load_per_task
;
2392 #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT)
2393 if (idle
== NOT_IDLE
|| !(sd
->flags
& SD_POWERSAVINGS_BALANCE
))
2396 if (this == group_leader
&& group_leader
!= group_min
) {
2397 *imbalance
= min_load_per_task
;
2407 * find_busiest_queue - find the busiest runqueue among the cpus in group.
2409 static runqueue_t
*find_busiest_queue(struct sched_group
*group
,
2410 enum idle_type idle
, unsigned long imbalance
)
2412 unsigned long max_load
= 0;
2413 runqueue_t
*busiest
= NULL
, *rqi
;
2416 for_each_cpu_mask(i
, group
->cpumask
) {
2419 if (rqi
->nr_running
== 1 && rqi
->raw_weighted_load
> imbalance
)
2422 if (rqi
->raw_weighted_load
> max_load
) {
2423 max_load
= rqi
->raw_weighted_load
;
2432 * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but
2433 * so long as it is large enough.
2435 #define MAX_PINNED_INTERVAL 512
2437 #define minus_1_or_zero(n) ((n) > 0 ? (n) - 1 : 0)
2439 * Check this_cpu to ensure it is balanced within domain. Attempt to move
2440 * tasks if there is an imbalance.
2442 * Called with this_rq unlocked.
2444 static int load_balance(int this_cpu
, runqueue_t
*this_rq
,
2445 struct sched_domain
*sd
, enum idle_type idle
)
2447 struct sched_group
*group
;
2448 runqueue_t
*busiest
;
2449 unsigned long imbalance
;
2450 int nr_moved
, all_pinned
= 0;
2451 int active_balance
= 0;
2454 if (idle
!= NOT_IDLE
&& sd
->flags
& SD_SHARE_CPUPOWER
&&
2455 !sched_smt_power_savings
)
2458 schedstat_inc(sd
, lb_cnt
[idle
]);
2460 group
= find_busiest_group(sd
, this_cpu
, &imbalance
, idle
, &sd_idle
);
2462 schedstat_inc(sd
, lb_nobusyg
[idle
]);
2466 busiest
= find_busiest_queue(group
, idle
, imbalance
);
2468 schedstat_inc(sd
, lb_nobusyq
[idle
]);
2472 BUG_ON(busiest
== this_rq
);
2474 schedstat_add(sd
, lb_imbalance
[idle
], imbalance
);
2477 if (busiest
->nr_running
> 1) {
2479 * Attempt to move tasks. If find_busiest_group has found
2480 * an imbalance but busiest->nr_running <= 1, the group is
2481 * still unbalanced. nr_moved simply stays zero, so it is
2482 * correctly treated as an imbalance.
2484 double_rq_lock(this_rq
, busiest
);
2485 nr_moved
= move_tasks(this_rq
, this_cpu
, busiest
,
2486 minus_1_or_zero(busiest
->nr_running
),
2487 imbalance
, sd
, idle
, &all_pinned
);
2488 double_rq_unlock(this_rq
, busiest
);
2490 /* All tasks on this runqueue were pinned by CPU affinity */
2491 if (unlikely(all_pinned
))
2496 schedstat_inc(sd
, lb_failed
[idle
]);
2497 sd
->nr_balance_failed
++;
2499 if (unlikely(sd
->nr_balance_failed
> sd
->cache_nice_tries
+2)) {
2501 spin_lock(&busiest
->lock
);
2503 /* don't kick the migration_thread, if the curr
2504 * task on busiest cpu can't be moved to this_cpu
2506 if (!cpu_isset(this_cpu
, busiest
->curr
->cpus_allowed
)) {
2507 spin_unlock(&busiest
->lock
);
2509 goto out_one_pinned
;
2512 if (!busiest
->active_balance
) {
2513 busiest
->active_balance
= 1;
2514 busiest
->push_cpu
= this_cpu
;
2517 spin_unlock(&busiest
->lock
);
2519 wake_up_process(busiest
->migration_thread
);
2522 * We've kicked active balancing, reset the failure
2525 sd
->nr_balance_failed
= sd
->cache_nice_tries
+1;
2528 sd
->nr_balance_failed
= 0;
2530 if (likely(!active_balance
)) {
2531 /* We were unbalanced, so reset the balancing interval */
2532 sd
->balance_interval
= sd
->min_interval
;
2535 * If we've begun active balancing, start to back off. This
2536 * case may not be covered by the all_pinned logic if there
2537 * is only 1 task on the busy runqueue (because we don't call
2540 if (sd
->balance_interval
< sd
->max_interval
)
2541 sd
->balance_interval
*= 2;
2544 if (!nr_moved
&& !sd_idle
&& sd
->flags
& SD_SHARE_CPUPOWER
&&
2545 !sched_smt_power_savings
)
2550 schedstat_inc(sd
, lb_balanced
[idle
]);
2552 sd
->nr_balance_failed
= 0;
2555 /* tune up the balancing interval */
2556 if ((all_pinned
&& sd
->balance_interval
< MAX_PINNED_INTERVAL
) ||
2557 (sd
->balance_interval
< sd
->max_interval
))
2558 sd
->balance_interval
*= 2;
2560 if (!sd_idle
&& sd
->flags
& SD_SHARE_CPUPOWER
&& !sched_smt_power_savings
)
2566 * Check this_cpu to ensure it is balanced within domain. Attempt to move
2567 * tasks if there is an imbalance.
2569 * Called from schedule when this_rq is about to become idle (NEWLY_IDLE).
2570 * this_rq is locked.
2572 static int load_balance_newidle(int this_cpu
, runqueue_t
*this_rq
,
2573 struct sched_domain
*sd
)
2575 struct sched_group
*group
;
2576 runqueue_t
*busiest
= NULL
;
2577 unsigned long imbalance
;
2581 if (sd
->flags
& SD_SHARE_CPUPOWER
&& !sched_smt_power_savings
)
2584 schedstat_inc(sd
, lb_cnt
[NEWLY_IDLE
]);
2585 group
= find_busiest_group(sd
, this_cpu
, &imbalance
, NEWLY_IDLE
, &sd_idle
);
2587 schedstat_inc(sd
, lb_nobusyg
[NEWLY_IDLE
]);
2591 busiest
= find_busiest_queue(group
, NEWLY_IDLE
, imbalance
);
2593 schedstat_inc(sd
, lb_nobusyq
[NEWLY_IDLE
]);
2597 BUG_ON(busiest
== this_rq
);
2599 schedstat_add(sd
, lb_imbalance
[NEWLY_IDLE
], imbalance
);
2602 if (busiest
->nr_running
> 1) {
2603 /* Attempt to move tasks */
2604 double_lock_balance(this_rq
, busiest
);
2605 nr_moved
= move_tasks(this_rq
, this_cpu
, busiest
,
2606 minus_1_or_zero(busiest
->nr_running
),
2607 imbalance
, sd
, NEWLY_IDLE
, NULL
);
2608 spin_unlock(&busiest
->lock
);
2612 schedstat_inc(sd
, lb_failed
[NEWLY_IDLE
]);
2613 if (!sd_idle
&& sd
->flags
& SD_SHARE_CPUPOWER
)
2616 sd
->nr_balance_failed
= 0;
2621 schedstat_inc(sd
, lb_balanced
[NEWLY_IDLE
]);
2622 if (!sd_idle
&& sd
->flags
& SD_SHARE_CPUPOWER
&& !sched_smt_power_savings
)
2624 sd
->nr_balance_failed
= 0;
2629 * idle_balance is called by schedule() if this_cpu is about to become
2630 * idle. Attempts to pull tasks from other CPUs.
2632 static void idle_balance(int this_cpu
, runqueue_t
*this_rq
)
2634 struct sched_domain
*sd
;
2636 for_each_domain(this_cpu
, sd
) {
2637 if (sd
->flags
& SD_BALANCE_NEWIDLE
) {
2638 if (load_balance_newidle(this_cpu
, this_rq
, sd
)) {
2639 /* We've pulled tasks over so stop searching */
2647 * active_load_balance is run by migration threads. It pushes running tasks
2648 * off the busiest CPU onto idle CPUs. It requires at least 1 task to be
2649 * running on each physical CPU where possible, and avoids physical /
2650 * logical imbalances.
2652 * Called with busiest_rq locked.
2654 static void active_load_balance(runqueue_t
*busiest_rq
, int busiest_cpu
)
2656 struct sched_domain
*sd
;
2657 runqueue_t
*target_rq
;
2658 int target_cpu
= busiest_rq
->push_cpu
;
2660 if (busiest_rq
->nr_running
<= 1)
2661 /* no task to move */
2664 target_rq
= cpu_rq(target_cpu
);
2667 * This condition is "impossible", if it occurs
2668 * we need to fix it. Originally reported by
2669 * Bjorn Helgaas on a 128-cpu setup.
2671 BUG_ON(busiest_rq
== target_rq
);
2673 /* move a task from busiest_rq to target_rq */
2674 double_lock_balance(busiest_rq
, target_rq
);
2676 /* Search for an sd spanning us and the target CPU. */
2677 for_each_domain(target_cpu
, sd
) {
2678 if ((sd
->flags
& SD_LOAD_BALANCE
) &&
2679 cpu_isset(busiest_cpu
, sd
->span
))
2683 if (unlikely(sd
== NULL
))
2686 schedstat_inc(sd
, alb_cnt
);
2688 if (move_tasks(target_rq
, target_cpu
, busiest_rq
, 1,
2689 RTPRIO_TO_LOAD_WEIGHT(100), sd
, SCHED_IDLE
, NULL
))
2690 schedstat_inc(sd
, alb_pushed
);
2692 schedstat_inc(sd
, alb_failed
);
2694 spin_unlock(&target_rq
->lock
);
2698 * rebalance_tick will get called every timer tick, on every CPU.
2700 * It checks each scheduling domain to see if it is due to be balanced,
2701 * and initiates a balancing operation if so.
2703 * Balancing parameters are set up in arch_init_sched_domains.
2706 /* Don't have all balancing operations going off at once */
2707 #define CPU_OFFSET(cpu) (HZ * cpu / NR_CPUS)
2709 static void rebalance_tick(int this_cpu
, runqueue_t
*this_rq
,
2710 enum idle_type idle
)
2712 unsigned long old_load
, this_load
;
2713 unsigned long j
= jiffies
+ CPU_OFFSET(this_cpu
);
2714 struct sched_domain
*sd
;
2717 this_load
= this_rq
->raw_weighted_load
;
2718 /* Update our load */
2719 for (i
= 0; i
< 3; i
++) {
2720 unsigned long new_load
= this_load
;
2722 old_load
= this_rq
->cpu_load
[i
];
2724 * Round up the averaging division if load is increasing. This
2725 * prevents us from getting stuck on 9 if the load is 10, for
2728 if (new_load
> old_load
)
2729 new_load
+= scale
-1;
2730 this_rq
->cpu_load
[i
] = (old_load
*(scale
-1) + new_load
) / scale
;
2733 for_each_domain(this_cpu
, sd
) {
2734 unsigned long interval
;
2736 if (!(sd
->flags
& SD_LOAD_BALANCE
))
2739 interval
= sd
->balance_interval
;
2740 if (idle
!= SCHED_IDLE
)
2741 interval
*= sd
->busy_factor
;
2743 /* scale ms to jiffies */
2744 interval
= msecs_to_jiffies(interval
);
2745 if (unlikely(!interval
))
2748 if (j
- sd
->last_balance
>= interval
) {
2749 if (load_balance(this_cpu
, this_rq
, sd
, idle
)) {
2751 * We've pulled tasks over so either we're no
2752 * longer idle, or one of our SMT siblings is
2757 sd
->last_balance
+= interval
;
2763 * on UP we do not need to balance between CPUs:
2765 static inline void rebalance_tick(int cpu
, runqueue_t
*rq
, enum idle_type idle
)
2768 static inline void idle_balance(int cpu
, runqueue_t
*rq
)
2773 static inline int wake_priority_sleeper(runqueue_t
*rq
)
2776 #ifdef CONFIG_SCHED_SMT
2777 spin_lock(&rq
->lock
);
2779 * If an SMT sibling task has been put to sleep for priority
2780 * reasons reschedule the idle task to see if it can now run.
2782 if (rq
->nr_running
) {
2783 resched_task(rq
->idle
);
2786 spin_unlock(&rq
->lock
);
2791 DEFINE_PER_CPU(struct kernel_stat
, kstat
);
2793 EXPORT_PER_CPU_SYMBOL(kstat
);
2796 * This is called on clock ticks and on context switches.
2797 * Bank in p->sched_time the ns elapsed since the last tick or switch.
2799 static inline void update_cpu_clock(task_t
*p
, runqueue_t
*rq
,
2800 unsigned long long now
)
2802 unsigned long long last
= max(p
->timestamp
, rq
->timestamp_last_tick
);
2803 p
->sched_time
+= now
- last
;
2807 * Return current->sched_time plus any more ns on the sched_clock
2808 * that have not yet been banked.
2810 unsigned long long current_sched_time(const task_t
*tsk
)
2812 unsigned long long ns
;
2813 unsigned long flags
;
2814 local_irq_save(flags
);
2815 ns
= max(tsk
->timestamp
, task_rq(tsk
)->timestamp_last_tick
);
2816 ns
= tsk
->sched_time
+ (sched_clock() - ns
);
2817 local_irq_restore(flags
);
2822 * We place interactive tasks back into the active array, if possible.
2824 * To guarantee that this does not starve expired tasks we ignore the
2825 * interactivity of a task if the first expired task had to wait more
2826 * than a 'reasonable' amount of time. This deadline timeout is
2827 * load-dependent, as the frequency of array switched decreases with
2828 * increasing number of running tasks. We also ignore the interactivity
2829 * if a better static_prio task has expired:
2831 #define EXPIRED_STARVING(rq) \
2832 ((STARVATION_LIMIT && ((rq)->expired_timestamp && \
2833 (jiffies - (rq)->expired_timestamp >= \
2834 STARVATION_LIMIT * ((rq)->nr_running) + 1))) || \
2835 ((rq)->curr->static_prio > (rq)->best_expired_prio))
2838 * Account user cpu time to a process.
2839 * @p: the process that the cpu time gets accounted to
2840 * @hardirq_offset: the offset to subtract from hardirq_count()
2841 * @cputime: the cpu time spent in user space since the last update
2843 void account_user_time(struct task_struct
*p
, cputime_t cputime
)
2845 struct cpu_usage_stat
*cpustat
= &kstat_this_cpu
.cpustat
;
2848 p
->utime
= cputime_add(p
->utime
, cputime
);
2850 /* Add user time to cpustat. */
2851 tmp
= cputime_to_cputime64(cputime
);
2852 if (TASK_NICE(p
) > 0)
2853 cpustat
->nice
= cputime64_add(cpustat
->nice
, tmp
);
2855 cpustat
->user
= cputime64_add(cpustat
->user
, tmp
);
2859 * Account system cpu time to a process.
2860 * @p: the process that the cpu time gets accounted to
2861 * @hardirq_offset: the offset to subtract from hardirq_count()
2862 * @cputime: the cpu time spent in kernel space since the last update
2864 void account_system_time(struct task_struct
*p
, int hardirq_offset
,
2867 struct cpu_usage_stat
*cpustat
= &kstat_this_cpu
.cpustat
;
2868 runqueue_t
*rq
= this_rq();
2871 p
->stime
= cputime_add(p
->stime
, cputime
);
2873 /* Add system time to cpustat. */
2874 tmp
= cputime_to_cputime64(cputime
);
2875 if (hardirq_count() - hardirq_offset
)
2876 cpustat
->irq
= cputime64_add(cpustat
->irq
, tmp
);
2877 else if (softirq_count())
2878 cpustat
->softirq
= cputime64_add(cpustat
->softirq
, tmp
);
2879 else if (p
!= rq
->idle
)
2880 cpustat
->system
= cputime64_add(cpustat
->system
, tmp
);
2881 else if (atomic_read(&rq
->nr_iowait
) > 0)
2882 cpustat
->iowait
= cputime64_add(cpustat
->iowait
, tmp
);
2884 cpustat
->idle
= cputime64_add(cpustat
->idle
, tmp
);
2885 /* Account for system time used */
2886 acct_update_integrals(p
);
2890 * Account for involuntary wait time.
2891 * @p: the process from which the cpu time has been stolen
2892 * @steal: the cpu time spent in involuntary wait
2894 void account_steal_time(struct task_struct
*p
, cputime_t steal
)
2896 struct cpu_usage_stat
*cpustat
= &kstat_this_cpu
.cpustat
;
2897 cputime64_t tmp
= cputime_to_cputime64(steal
);
2898 runqueue_t
*rq
= this_rq();
2900 if (p
== rq
->idle
) {
2901 p
->stime
= cputime_add(p
->stime
, steal
);
2902 if (atomic_read(&rq
->nr_iowait
) > 0)
2903 cpustat
->iowait
= cputime64_add(cpustat
->iowait
, tmp
);
2905 cpustat
->idle
= cputime64_add(cpustat
->idle
, tmp
);
2907 cpustat
->steal
= cputime64_add(cpustat
->steal
, tmp
);
2911 * This function gets called by the timer code, with HZ frequency.
2912 * We call it with interrupts disabled.
2914 * It also gets called by the fork code, when changing the parent's
2917 void scheduler_tick(void)
2919 int cpu
= smp_processor_id();
2920 runqueue_t
*rq
= this_rq();
2921 task_t
*p
= current
;
2922 unsigned long long now
= sched_clock();
2924 update_cpu_clock(p
, rq
, now
);
2926 rq
->timestamp_last_tick
= now
;
2928 if (p
== rq
->idle
) {
2929 if (wake_priority_sleeper(rq
))
2931 rebalance_tick(cpu
, rq
, SCHED_IDLE
);
2935 /* Task might have expired already, but not scheduled off yet */
2936 if (p
->array
!= rq
->active
) {
2937 set_tsk_need_resched(p
);
2940 spin_lock(&rq
->lock
);
2942 * The task was running during this tick - update the
2943 * time slice counter. Note: we do not update a thread's
2944 * priority until it either goes to sleep or uses up its
2945 * timeslice. This makes it possible for interactive tasks
2946 * to use up their timeslices at their highest priority levels.
2950 * RR tasks need a special form of timeslice management.
2951 * FIFO tasks have no timeslices.
2953 if ((p
->policy
== SCHED_RR
) && !--p
->time_slice
) {
2954 p
->time_slice
= task_timeslice(p
);
2955 p
->first_time_slice
= 0;
2956 set_tsk_need_resched(p
);
2958 /* put it at the end of the queue: */
2959 requeue_task(p
, rq
->active
);
2963 if (!--p
->time_slice
) {
2964 dequeue_task(p
, rq
->active
);
2965 set_tsk_need_resched(p
);
2966 p
->prio
= effective_prio(p
);
2967 p
->time_slice
= task_timeslice(p
);
2968 p
->first_time_slice
= 0;
2970 if (!rq
->expired_timestamp
)
2971 rq
->expired_timestamp
= jiffies
;
2972 if (!TASK_INTERACTIVE(p
) || EXPIRED_STARVING(rq
)) {
2973 enqueue_task(p
, rq
->expired
);
2974 if (p
->static_prio
< rq
->best_expired_prio
)
2975 rq
->best_expired_prio
= p
->static_prio
;
2977 enqueue_task(p
, rq
->active
);
2980 * Prevent a too long timeslice allowing a task to monopolize
2981 * the CPU. We do this by splitting up the timeslice into
2984 * Note: this does not mean the task's timeslices expire or
2985 * get lost in any way, they just might be preempted by
2986 * another task of equal priority. (one with higher
2987 * priority would have preempted this task already.) We
2988 * requeue this task to the end of the list on this priority
2989 * level, which is in essence a round-robin of tasks with
2992 * This only applies to tasks in the interactive
2993 * delta range with at least TIMESLICE_GRANULARITY to requeue.
2995 if (TASK_INTERACTIVE(p
) && !((task_timeslice(p
) -
2996 p
->time_slice
) % TIMESLICE_GRANULARITY(p
)) &&
2997 (p
->time_slice
>= TIMESLICE_GRANULARITY(p
)) &&
2998 (p
->array
== rq
->active
)) {
3000 requeue_task(p
, rq
->active
);
3001 set_tsk_need_resched(p
);
3005 spin_unlock(&rq
->lock
);
3007 rebalance_tick(cpu
, rq
, NOT_IDLE
);
3010 #ifdef CONFIG_SCHED_SMT
3011 static inline void wakeup_busy_runqueue(runqueue_t
*rq
)
3013 /* If an SMT runqueue is sleeping due to priority reasons wake it up */
3014 if (rq
->curr
== rq
->idle
&& rq
->nr_running
)
3015 resched_task(rq
->idle
);
3019 * Called with interrupt disabled and this_rq's runqueue locked.
3021 static void wake_sleeping_dependent(int this_cpu
)
3023 struct sched_domain
*tmp
, *sd
= NULL
;
3026 for_each_domain(this_cpu
, tmp
) {
3027 if (tmp
->flags
& SD_SHARE_CPUPOWER
) {
3036 for_each_cpu_mask(i
, sd
->span
) {
3037 runqueue_t
*smt_rq
= cpu_rq(i
);
3041 if (unlikely(!spin_trylock(&smt_rq
->lock
)))
3044 wakeup_busy_runqueue(smt_rq
);
3045 spin_unlock(&smt_rq
->lock
);
3050 * number of 'lost' timeslices this task wont be able to fully
3051 * utilize, if another task runs on a sibling. This models the
3052 * slowdown effect of other tasks running on siblings:
3054 static inline unsigned long smt_slice(task_t
*p
, struct sched_domain
*sd
)
3056 return p
->time_slice
* (100 - sd
->per_cpu_gain
) / 100;
3060 * To minimise lock contention and not have to drop this_rq's runlock we only
3061 * trylock the sibling runqueues and bypass those runqueues if we fail to
3062 * acquire their lock. As we only trylock the normal locking order does not
3063 * need to be obeyed.
3065 static int dependent_sleeper(int this_cpu
, runqueue_t
*this_rq
, task_t
*p
)
3067 struct sched_domain
*tmp
, *sd
= NULL
;
3070 /* kernel/rt threads do not participate in dependent sleeping */
3071 if (!p
->mm
|| rt_task(p
))
3074 for_each_domain(this_cpu
, tmp
) {
3075 if (tmp
->flags
& SD_SHARE_CPUPOWER
) {
3084 for_each_cpu_mask(i
, sd
->span
) {
3092 if (unlikely(!spin_trylock(&smt_rq
->lock
)))
3095 smt_curr
= smt_rq
->curr
;
3101 * If a user task with lower static priority than the
3102 * running task on the SMT sibling is trying to schedule,
3103 * delay it till there is proportionately less timeslice
3104 * left of the sibling task to prevent a lower priority
3105 * task from using an unfair proportion of the
3106 * physical cpu's resources. -ck
3108 if (rt_task(smt_curr
)) {
3110 * With real time tasks we run non-rt tasks only
3111 * per_cpu_gain% of the time.
3113 if ((jiffies
% DEF_TIMESLICE
) >
3114 (sd
->per_cpu_gain
* DEF_TIMESLICE
/ 100))
3117 if (smt_curr
->static_prio
< p
->static_prio
&&
3118 !TASK_PREEMPTS_CURR(p
, smt_rq
) &&
3119 smt_slice(smt_curr
, sd
) > task_timeslice(p
))
3123 spin_unlock(&smt_rq
->lock
);
3128 static inline void wake_sleeping_dependent(int this_cpu
)
3132 static inline int dependent_sleeper(int this_cpu
, runqueue_t
*this_rq
,
3139 #if defined(CONFIG_PREEMPT) && defined(CONFIG_DEBUG_PREEMPT)
3141 void fastcall
add_preempt_count(int val
)
3146 if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0)))
3148 preempt_count() += val
;
3150 * Spinlock count overflowing soon?
3152 DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK
) >= PREEMPT_MASK
-10);
3154 EXPORT_SYMBOL(add_preempt_count
);
3156 void fastcall
sub_preempt_count(int val
)
3161 if (DEBUG_LOCKS_WARN_ON(val
> preempt_count()))
3164 * Is the spinlock portion underflowing?
3166 if (DEBUG_LOCKS_WARN_ON((val
< PREEMPT_MASK
) &&
3167 !(preempt_count() & PREEMPT_MASK
)))
3170 preempt_count() -= val
;
3172 EXPORT_SYMBOL(sub_preempt_count
);
3176 static inline int interactive_sleep(enum sleep_type sleep_type
)
3178 return (sleep_type
== SLEEP_INTERACTIVE
||
3179 sleep_type
== SLEEP_INTERRUPTED
);
3183 * schedule() is the main scheduler function.
3185 asmlinkage
void __sched
schedule(void)
3188 task_t
*prev
, *next
;
3190 prio_array_t
*array
;
3191 struct list_head
*queue
;
3192 unsigned long long now
;
3193 unsigned long run_time
;
3194 int cpu
, idx
, new_prio
;
3197 * Test if we are atomic. Since do_exit() needs to call into
3198 * schedule() atomically, we ignore that path for now.
3199 * Otherwise, whine if we are scheduling when we should not be.
3201 if (unlikely(in_atomic() && !current
->exit_state
)) {
3202 printk(KERN_ERR
"BUG: scheduling while atomic: "
3204 current
->comm
, preempt_count(), current
->pid
);
3207 profile_hit(SCHED_PROFILING
, __builtin_return_address(0));
3212 release_kernel_lock(prev
);
3213 need_resched_nonpreemptible
:
3217 * The idle thread is not allowed to schedule!
3218 * Remove this check after it has been exercised a bit.
3220 if (unlikely(prev
== rq
->idle
) && prev
->state
!= TASK_RUNNING
) {
3221 printk(KERN_ERR
"bad: scheduling from the idle thread!\n");
3225 schedstat_inc(rq
, sched_cnt
);
3226 now
= sched_clock();
3227 if (likely((long long)(now
- prev
->timestamp
) < NS_MAX_SLEEP_AVG
)) {
3228 run_time
= now
- prev
->timestamp
;
3229 if (unlikely((long long)(now
- prev
->timestamp
) < 0))
3232 run_time
= NS_MAX_SLEEP_AVG
;
3235 * Tasks charged proportionately less run_time at high sleep_avg to
3236 * delay them losing their interactive status
3238 run_time
/= (CURRENT_BONUS(prev
) ? : 1);
3240 spin_lock_irq(&rq
->lock
);
3242 if (unlikely(prev
->flags
& PF_DEAD
))
3243 prev
->state
= EXIT_DEAD
;
3245 switch_count
= &prev
->nivcsw
;
3246 if (prev
->state
&& !(preempt_count() & PREEMPT_ACTIVE
)) {
3247 switch_count
= &prev
->nvcsw
;
3248 if (unlikely((prev
->state
& TASK_INTERRUPTIBLE
) &&
3249 unlikely(signal_pending(prev
))))
3250 prev
->state
= TASK_RUNNING
;
3252 if (prev
->state
== TASK_UNINTERRUPTIBLE
)
3253 rq
->nr_uninterruptible
++;
3254 deactivate_task(prev
, rq
);
3258 cpu
= smp_processor_id();
3259 if (unlikely(!rq
->nr_running
)) {
3260 idle_balance(cpu
, rq
);
3261 if (!rq
->nr_running
) {
3263 rq
->expired_timestamp
= 0;
3264 wake_sleeping_dependent(cpu
);
3270 if (unlikely(!array
->nr_active
)) {
3272 * Switch the active and expired arrays.
3274 schedstat_inc(rq
, sched_switch
);
3275 rq
->active
= rq
->expired
;
3276 rq
->expired
= array
;
3278 rq
->expired_timestamp
= 0;
3279 rq
->best_expired_prio
= MAX_PRIO
;
3282 idx
= sched_find_first_bit(array
->bitmap
);
3283 queue
= array
->queue
+ idx
;
3284 next
= list_entry(queue
->next
, task_t
, run_list
);
3286 if (!rt_task(next
) && interactive_sleep(next
->sleep_type
)) {
3287 unsigned long long delta
= now
- next
->timestamp
;
3288 if (unlikely((long long)(now
- next
->timestamp
) < 0))
3291 if (next
->sleep_type
== SLEEP_INTERACTIVE
)
3292 delta
= delta
* (ON_RUNQUEUE_WEIGHT
* 128 / 100) / 128;
3294 array
= next
->array
;
3295 new_prio
= recalc_task_prio(next
, next
->timestamp
+ delta
);
3297 if (unlikely(next
->prio
!= new_prio
)) {
3298 dequeue_task(next
, array
);
3299 next
->prio
= new_prio
;
3300 enqueue_task(next
, array
);
3303 next
->sleep_type
= SLEEP_NORMAL
;
3304 if (dependent_sleeper(cpu
, rq
, next
))
3307 if (next
== rq
->idle
)
3308 schedstat_inc(rq
, sched_goidle
);
3310 prefetch_stack(next
);
3311 clear_tsk_need_resched(prev
);
3312 rcu_qsctr_inc(task_cpu(prev
));
3314 update_cpu_clock(prev
, rq
, now
);
3316 prev
->sleep_avg
-= run_time
;
3317 if ((long)prev
->sleep_avg
<= 0)
3318 prev
->sleep_avg
= 0;
3319 prev
->timestamp
= prev
->last_ran
= now
;
3321 sched_info_switch(prev
, next
);
3322 if (likely(prev
!= next
)) {
3323 next
->timestamp
= now
;
3328 prepare_task_switch(rq
, next
);
3329 prev
= context_switch(rq
, prev
, next
);
3332 * this_rq must be evaluated again because prev may have moved
3333 * CPUs since it called schedule(), thus the 'rq' on its stack
3334 * frame will be invalid.
3336 finish_task_switch(this_rq(), prev
);
3338 spin_unlock_irq(&rq
->lock
);
3341 if (unlikely(reacquire_kernel_lock(prev
) < 0))
3342 goto need_resched_nonpreemptible
;
3343 preempt_enable_no_resched();
3344 if (unlikely(test_thread_flag(TIF_NEED_RESCHED
)))
3348 EXPORT_SYMBOL(schedule
);
3350 #ifdef CONFIG_PREEMPT
3352 * this is is the entry point to schedule() from in-kernel preemption
3353 * off of preempt_enable. Kernel preemptions off return from interrupt
3354 * occur there and call schedule directly.
3356 asmlinkage
void __sched
preempt_schedule(void)
3358 struct thread_info
*ti
= current_thread_info();
3359 #ifdef CONFIG_PREEMPT_BKL
3360 struct task_struct
*task
= current
;
3361 int saved_lock_depth
;
3364 * If there is a non-zero preempt_count or interrupts are disabled,
3365 * we do not want to preempt the current task. Just return..
3367 if (unlikely(ti
->preempt_count
|| irqs_disabled()))
3371 add_preempt_count(PREEMPT_ACTIVE
);
3373 * We keep the big kernel semaphore locked, but we
3374 * clear ->lock_depth so that schedule() doesnt
3375 * auto-release the semaphore:
3377 #ifdef CONFIG_PREEMPT_BKL
3378 saved_lock_depth
= task
->lock_depth
;
3379 task
->lock_depth
= -1;
3382 #ifdef CONFIG_PREEMPT_BKL
3383 task
->lock_depth
= saved_lock_depth
;
3385 sub_preempt_count(PREEMPT_ACTIVE
);
3387 /* we could miss a preemption opportunity between schedule and now */
3389 if (unlikely(test_thread_flag(TIF_NEED_RESCHED
)))
3393 EXPORT_SYMBOL(preempt_schedule
);
3396 * this is is the entry point to schedule() from kernel preemption
3397 * off of irq context.
3398 * Note, that this is called and return with irqs disabled. This will
3399 * protect us against recursive calling from irq.
3401 asmlinkage
void __sched
preempt_schedule_irq(void)
3403 struct thread_info
*ti
= current_thread_info();
3404 #ifdef CONFIG_PREEMPT_BKL
3405 struct task_struct
*task
= current
;
3406 int saved_lock_depth
;
3408 /* Catch callers which need to be fixed*/
3409 BUG_ON(ti
->preempt_count
|| !irqs_disabled());
3412 add_preempt_count(PREEMPT_ACTIVE
);
3414 * We keep the big kernel semaphore locked, but we
3415 * clear ->lock_depth so that schedule() doesnt
3416 * auto-release the semaphore:
3418 #ifdef CONFIG_PREEMPT_BKL
3419 saved_lock_depth
= task
->lock_depth
;
3420 task
->lock_depth
= -1;
3424 local_irq_disable();
3425 #ifdef CONFIG_PREEMPT_BKL
3426 task
->lock_depth
= saved_lock_depth
;
3428 sub_preempt_count(PREEMPT_ACTIVE
);
3430 /* we could miss a preemption opportunity between schedule and now */
3432 if (unlikely(test_thread_flag(TIF_NEED_RESCHED
)))
3436 #endif /* CONFIG_PREEMPT */
3438 int default_wake_function(wait_queue_t
*curr
, unsigned mode
, int sync
,
3441 task_t
*p
= curr
->private;
3442 return try_to_wake_up(p
, mode
, sync
);
3445 EXPORT_SYMBOL(default_wake_function
);
3448 * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just
3449 * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve
3450 * number) then we wake all the non-exclusive tasks and one exclusive task.
3452 * There are circumstances in which we can try to wake a task which has already
3453 * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns
3454 * zero in this (rare) case, and we handle it by continuing to scan the queue.
3456 static void __wake_up_common(wait_queue_head_t
*q
, unsigned int mode
,
3457 int nr_exclusive
, int sync
, void *key
)
3459 struct list_head
*tmp
, *next
;
3461 list_for_each_safe(tmp
, next
, &q
->task_list
) {
3464 curr
= list_entry(tmp
, wait_queue_t
, task_list
);
3465 flags
= curr
->flags
;
3466 if (curr
->func(curr
, mode
, sync
, key
) &&
3467 (flags
& WQ_FLAG_EXCLUSIVE
) &&
3474 * __wake_up - wake up threads blocked on a waitqueue.
3476 * @mode: which threads
3477 * @nr_exclusive: how many wake-one or wake-many threads to wake up
3478 * @key: is directly passed to the wakeup function
3480 void fastcall
__wake_up(wait_queue_head_t
*q
, unsigned int mode
,
3481 int nr_exclusive
, void *key
)
3483 unsigned long flags
;
3485 spin_lock_irqsave(&q
->lock
, flags
);
3486 __wake_up_common(q
, mode
, nr_exclusive
, 0, key
);
3487 spin_unlock_irqrestore(&q
->lock
, flags
);
3490 EXPORT_SYMBOL(__wake_up
);
3493 * Same as __wake_up but called with the spinlock in wait_queue_head_t held.
3495 void fastcall
__wake_up_locked(wait_queue_head_t
*q
, unsigned int mode
)
3497 __wake_up_common(q
, mode
, 1, 0, NULL
);
3501 * __wake_up_sync - wake up threads blocked on a waitqueue.
3503 * @mode: which threads
3504 * @nr_exclusive: how many wake-one or wake-many threads to wake up
3506 * The sync wakeup differs that the waker knows that it will schedule
3507 * away soon, so while the target thread will be woken up, it will not
3508 * be migrated to another CPU - ie. the two threads are 'synchronized'
3509 * with each other. This can prevent needless bouncing between CPUs.
3511 * On UP it can prevent extra preemption.
3514 __wake_up_sync(wait_queue_head_t
*q
, unsigned int mode
, int nr_exclusive
)
3516 unsigned long flags
;
3522 if (unlikely(!nr_exclusive
))
3525 spin_lock_irqsave(&q
->lock
, flags
);
3526 __wake_up_common(q
, mode
, nr_exclusive
, sync
, NULL
);
3527 spin_unlock_irqrestore(&q
->lock
, flags
);
3529 EXPORT_SYMBOL_GPL(__wake_up_sync
); /* For internal use only */
3531 void fastcall
complete(struct completion
*x
)
3533 unsigned long flags
;
3535 spin_lock_irqsave(&x
->wait
.lock
, flags
);
3537 __wake_up_common(&x
->wait
, TASK_UNINTERRUPTIBLE
| TASK_INTERRUPTIBLE
,
3539 spin_unlock_irqrestore(&x
->wait
.lock
, flags
);
3541 EXPORT_SYMBOL(complete
);
3543 void fastcall
complete_all(struct completion
*x
)
3545 unsigned long flags
;
3547 spin_lock_irqsave(&x
->wait
.lock
, flags
);
3548 x
->done
+= UINT_MAX
/2;
3549 __wake_up_common(&x
->wait
, TASK_UNINTERRUPTIBLE
| TASK_INTERRUPTIBLE
,
3551 spin_unlock_irqrestore(&x
->wait
.lock
, flags
);
3553 EXPORT_SYMBOL(complete_all
);
3555 void fastcall __sched
wait_for_completion(struct completion
*x
)
3558 spin_lock_irq(&x
->wait
.lock
);
3560 DECLARE_WAITQUEUE(wait
, current
);
3562 wait
.flags
|= WQ_FLAG_EXCLUSIVE
;
3563 __add_wait_queue_tail(&x
->wait
, &wait
);
3565 __set_current_state(TASK_UNINTERRUPTIBLE
);
3566 spin_unlock_irq(&x
->wait
.lock
);
3568 spin_lock_irq(&x
->wait
.lock
);
3570 __remove_wait_queue(&x
->wait
, &wait
);
3573 spin_unlock_irq(&x
->wait
.lock
);
3575 EXPORT_SYMBOL(wait_for_completion
);
3577 unsigned long fastcall __sched
3578 wait_for_completion_timeout(struct completion
*x
, unsigned long timeout
)
3582 spin_lock_irq(&x
->wait
.lock
);
3584 DECLARE_WAITQUEUE(wait
, current
);
3586 wait
.flags
|= WQ_FLAG_EXCLUSIVE
;
3587 __add_wait_queue_tail(&x
->wait
, &wait
);
3589 __set_current_state(TASK_UNINTERRUPTIBLE
);
3590 spin_unlock_irq(&x
->wait
.lock
);
3591 timeout
= schedule_timeout(timeout
);
3592 spin_lock_irq(&x
->wait
.lock
);
3594 __remove_wait_queue(&x
->wait
, &wait
);
3598 __remove_wait_queue(&x
->wait
, &wait
);
3602 spin_unlock_irq(&x
->wait
.lock
);
3605 EXPORT_SYMBOL(wait_for_completion_timeout
);
3607 int fastcall __sched
wait_for_completion_interruptible(struct completion
*x
)
3613 spin_lock_irq(&x
->wait
.lock
);
3615 DECLARE_WAITQUEUE(wait
, current
);
3617 wait
.flags
|= WQ_FLAG_EXCLUSIVE
;
3618 __add_wait_queue_tail(&x
->wait
, &wait
);
3620 if (signal_pending(current
)) {
3622 __remove_wait_queue(&x
->wait
, &wait
);
3625 __set_current_state(TASK_INTERRUPTIBLE
);
3626 spin_unlock_irq(&x
->wait
.lock
);
3628 spin_lock_irq(&x
->wait
.lock
);
3630 __remove_wait_queue(&x
->wait
, &wait
);
3634 spin_unlock_irq(&x
->wait
.lock
);
3638 EXPORT_SYMBOL(wait_for_completion_interruptible
);
3640 unsigned long fastcall __sched
3641 wait_for_completion_interruptible_timeout(struct completion
*x
,
3642 unsigned long timeout
)
3646 spin_lock_irq(&x
->wait
.lock
);
3648 DECLARE_WAITQUEUE(wait
, current
);
3650 wait
.flags
|= WQ_FLAG_EXCLUSIVE
;
3651 __add_wait_queue_tail(&x
->wait
, &wait
);
3653 if (signal_pending(current
)) {
3654 timeout
= -ERESTARTSYS
;
3655 __remove_wait_queue(&x
->wait
, &wait
);
3658 __set_current_state(TASK_INTERRUPTIBLE
);
3659 spin_unlock_irq(&x
->wait
.lock
);
3660 timeout
= schedule_timeout(timeout
);
3661 spin_lock_irq(&x
->wait
.lock
);
3663 __remove_wait_queue(&x
->wait
, &wait
);
3667 __remove_wait_queue(&x
->wait
, &wait
);
3671 spin_unlock_irq(&x
->wait
.lock
);
3674 EXPORT_SYMBOL(wait_for_completion_interruptible_timeout
);
3677 #define SLEEP_ON_VAR \
3678 unsigned long flags; \
3679 wait_queue_t wait; \
3680 init_waitqueue_entry(&wait, current);
3682 #define SLEEP_ON_HEAD \
3683 spin_lock_irqsave(&q->lock,flags); \
3684 __add_wait_queue(q, &wait); \
3685 spin_unlock(&q->lock);
3687 #define SLEEP_ON_TAIL \
3688 spin_lock_irq(&q->lock); \
3689 __remove_wait_queue(q, &wait); \
3690 spin_unlock_irqrestore(&q->lock, flags);
3692 void fastcall __sched
interruptible_sleep_on(wait_queue_head_t
*q
)
3696 current
->state
= TASK_INTERRUPTIBLE
;
3703 EXPORT_SYMBOL(interruptible_sleep_on
);
3705 long fastcall __sched
3706 interruptible_sleep_on_timeout(wait_queue_head_t
*q
, long timeout
)
3710 current
->state
= TASK_INTERRUPTIBLE
;
3713 timeout
= schedule_timeout(timeout
);
3719 EXPORT_SYMBOL(interruptible_sleep_on_timeout
);
3721 void fastcall __sched
sleep_on(wait_queue_head_t
*q
)
3725 current
->state
= TASK_UNINTERRUPTIBLE
;
3732 EXPORT_SYMBOL(sleep_on
);
3734 long fastcall __sched
sleep_on_timeout(wait_queue_head_t
*q
, long timeout
)
3738 current
->state
= TASK_UNINTERRUPTIBLE
;
3741 timeout
= schedule_timeout(timeout
);
3747 EXPORT_SYMBOL(sleep_on_timeout
);
3749 #ifdef CONFIG_RT_MUTEXES
3752 * rt_mutex_setprio - set the current priority of a task
3754 * @prio: prio value (kernel-internal form)
3756 * This function changes the 'effective' priority of a task. It does
3757 * not touch ->normal_prio like __setscheduler().
3759 * Used by the rt_mutex code to implement priority inheritance logic.
3761 void rt_mutex_setprio(task_t
*p
, int prio
)
3763 unsigned long flags
;
3764 prio_array_t
*array
;
3768 BUG_ON(prio
< 0 || prio
> MAX_PRIO
);
3770 rq
= task_rq_lock(p
, &flags
);
3775 dequeue_task(p
, array
);
3780 * If changing to an RT priority then queue it
3781 * in the active array!
3785 enqueue_task(p
, array
);
3787 * Reschedule if we are currently running on this runqueue and
3788 * our priority decreased, or if we are not currently running on
3789 * this runqueue and our priority is higher than the current's
3791 if (task_running(rq
, p
)) {
3792 if (p
->prio
> oldprio
)
3793 resched_task(rq
->curr
);
3794 } else if (TASK_PREEMPTS_CURR(p
, rq
))
3795 resched_task(rq
->curr
);
3797 task_rq_unlock(rq
, &flags
);
3802 void set_user_nice(task_t
*p
, long nice
)
3804 unsigned long flags
;
3805 prio_array_t
*array
;
3807 int old_prio
, delta
;
3809 if (TASK_NICE(p
) == nice
|| nice
< -20 || nice
> 19)
3812 * We have to be careful, if called from sys_setpriority(),
3813 * the task might be in the middle of scheduling on another CPU.
3815 rq
= task_rq_lock(p
, &flags
);
3817 * The RT priorities are set via sched_setscheduler(), but we still
3818 * allow the 'normal' nice value to be set - but as expected
3819 * it wont have any effect on scheduling until the task is
3820 * not SCHED_NORMAL/SCHED_BATCH:
3822 if (has_rt_policy(p
)) {
3823 p
->static_prio
= NICE_TO_PRIO(nice
);
3828 dequeue_task(p
, array
);
3829 dec_raw_weighted_load(rq
, p
);
3832 p
->static_prio
= NICE_TO_PRIO(nice
);
3835 p
->prio
= effective_prio(p
);
3836 delta
= p
->prio
- old_prio
;
3839 enqueue_task(p
, array
);
3840 inc_raw_weighted_load(rq
, p
);
3842 * If the task increased its priority or is running and
3843 * lowered its priority, then reschedule its CPU:
3845 if (delta
< 0 || (delta
> 0 && task_running(rq
, p
)))
3846 resched_task(rq
->curr
);
3849 task_rq_unlock(rq
, &flags
);
3851 EXPORT_SYMBOL(set_user_nice
);
3854 * can_nice - check if a task can reduce its nice value
3858 int can_nice(const task_t
*p
, const int nice
)
3860 /* convert nice value [19,-20] to rlimit style value [1,40] */
3861 int nice_rlim
= 20 - nice
;
3862 return (nice_rlim
<= p
->signal
->rlim
[RLIMIT_NICE
].rlim_cur
||
3863 capable(CAP_SYS_NICE
));
3866 #ifdef __ARCH_WANT_SYS_NICE
3869 * sys_nice - change the priority of the current process.
3870 * @increment: priority increment
3872 * sys_setpriority is a more generic, but much slower function that
3873 * does similar things.
3875 asmlinkage
long sys_nice(int increment
)
3881 * Setpriority might change our priority at the same moment.
3882 * We don't have to worry. Conceptually one call occurs first
3883 * and we have a single winner.
3885 if (increment
< -40)
3890 nice
= PRIO_TO_NICE(current
->static_prio
) + increment
;
3896 if (increment
< 0 && !can_nice(current
, nice
))
3899 retval
= security_task_setnice(current
, nice
);
3903 set_user_nice(current
, nice
);
3910 * task_prio - return the priority value of a given task.
3911 * @p: the task in question.
3913 * This is the priority value as seen by users in /proc.
3914 * RT tasks are offset by -200. Normal tasks are centered
3915 * around 0, value goes from -16 to +15.
3917 int task_prio(const task_t
*p
)
3919 return p
->prio
- MAX_RT_PRIO
;
3923 * task_nice - return the nice value of a given task.
3924 * @p: the task in question.
3926 int task_nice(const task_t
*p
)
3928 return TASK_NICE(p
);
3930 EXPORT_SYMBOL_GPL(task_nice
);
3933 * idle_cpu - is a given cpu idle currently?
3934 * @cpu: the processor in question.
3936 int idle_cpu(int cpu
)
3938 return cpu_curr(cpu
) == cpu_rq(cpu
)->idle
;
3942 * idle_task - return the idle task for a given cpu.
3943 * @cpu: the processor in question.
3945 task_t
*idle_task(int cpu
)
3947 return cpu_rq(cpu
)->idle
;
3951 * find_process_by_pid - find a process with a matching PID value.
3952 * @pid: the pid in question.
3954 static inline task_t
*find_process_by_pid(pid_t pid
)
3956 return pid
? find_task_by_pid(pid
) : current
;
3959 /* Actually do priority change: must hold rq lock. */
3960 static void __setscheduler(struct task_struct
*p
, int policy
, int prio
)
3964 p
->rt_priority
= prio
;
3965 p
->normal_prio
= normal_prio(p
);
3966 /* we are holding p->pi_lock already */
3967 p
->prio
= rt_mutex_getprio(p
);
3969 * SCHED_BATCH tasks are treated as perpetual CPU hogs:
3971 if (policy
== SCHED_BATCH
)
3977 * sched_setscheduler - change the scheduling policy and/or RT priority of
3979 * @p: the task in question.
3980 * @policy: new policy.
3981 * @param: structure containing the new RT priority.
3983 int sched_setscheduler(struct task_struct
*p
, int policy
,
3984 struct sched_param
*param
)
3987 int oldprio
, oldpolicy
= -1;
3988 prio_array_t
*array
;
3989 unsigned long flags
;
3992 /* may grab non-irq protected spin_locks */
3993 BUG_ON(in_interrupt());
3995 /* double check policy once rq lock held */
3997 policy
= oldpolicy
= p
->policy
;
3998 else if (policy
!= SCHED_FIFO
&& policy
!= SCHED_RR
&&
3999 policy
!= SCHED_NORMAL
&& policy
!= SCHED_BATCH
)
4002 * Valid priorities for SCHED_FIFO and SCHED_RR are
4003 * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL and
4006 if (param
->sched_priority
< 0 ||
4007 (p
->mm
&& param
->sched_priority
> MAX_USER_RT_PRIO
-1) ||
4008 (!p
->mm
&& param
->sched_priority
> MAX_RT_PRIO
-1))
4010 if ((policy
== SCHED_NORMAL
|| policy
== SCHED_BATCH
)
4011 != (param
->sched_priority
== 0))
4015 * Allow unprivileged RT tasks to decrease priority:
4017 if (!capable(CAP_SYS_NICE
)) {
4019 * can't change policy, except between SCHED_NORMAL
4022 if (((policy
!= SCHED_NORMAL
&& p
->policy
!= SCHED_BATCH
) &&
4023 (policy
!= SCHED_BATCH
&& p
->policy
!= SCHED_NORMAL
)) &&
4024 !p
->signal
->rlim
[RLIMIT_RTPRIO
].rlim_cur
)
4026 /* can't increase priority */
4027 if ((policy
!= SCHED_NORMAL
&& policy
!= SCHED_BATCH
) &&
4028 param
->sched_priority
> p
->rt_priority
&&
4029 param
->sched_priority
>
4030 p
->signal
->rlim
[RLIMIT_RTPRIO
].rlim_cur
)
4032 /* can't change other user's priorities */
4033 if ((current
->euid
!= p
->euid
) &&
4034 (current
->euid
!= p
->uid
))
4038 retval
= security_task_setscheduler(p
, policy
, param
);
4042 * make sure no PI-waiters arrive (or leave) while we are
4043 * changing the priority of the task:
4045 spin_lock_irqsave(&p
->pi_lock
, flags
);
4047 * To be able to change p->policy safely, the apropriate
4048 * runqueue lock must be held.
4050 rq
= __task_rq_lock(p
);
4051 /* recheck policy now with rq lock held */
4052 if (unlikely(oldpolicy
!= -1 && oldpolicy
!= p
->policy
)) {
4053 policy
= oldpolicy
= -1;
4054 __task_rq_unlock(rq
);
4055 spin_unlock_irqrestore(&p
->pi_lock
, flags
);
4060 deactivate_task(p
, rq
);
4062 __setscheduler(p
, policy
, param
->sched_priority
);
4064 __activate_task(p
, rq
);
4066 * Reschedule if we are currently running on this runqueue and
4067 * our priority decreased, or if we are not currently running on
4068 * this runqueue and our priority is higher than the current's
4070 if (task_running(rq
, p
)) {
4071 if (p
->prio
> oldprio
)
4072 resched_task(rq
->curr
);
4073 } else if (TASK_PREEMPTS_CURR(p
, rq
))
4074 resched_task(rq
->curr
);
4076 __task_rq_unlock(rq
);
4077 spin_unlock_irqrestore(&p
->pi_lock
, flags
);
4079 rt_mutex_adjust_pi(p
);
4083 EXPORT_SYMBOL_GPL(sched_setscheduler
);
4086 do_sched_setscheduler(pid_t pid
, int policy
, struct sched_param __user
*param
)
4089 struct sched_param lparam
;
4090 struct task_struct
*p
;
4092 if (!param
|| pid
< 0)
4094 if (copy_from_user(&lparam
, param
, sizeof(struct sched_param
)))
4096 read_lock_irq(&tasklist_lock
);
4097 p
= find_process_by_pid(pid
);
4099 read_unlock_irq(&tasklist_lock
);
4103 read_unlock_irq(&tasklist_lock
);
4104 retval
= sched_setscheduler(p
, policy
, &lparam
);
4110 * sys_sched_setscheduler - set/change the scheduler policy and RT priority
4111 * @pid: the pid in question.
4112 * @policy: new policy.
4113 * @param: structure containing the new RT priority.
4115 asmlinkage
long sys_sched_setscheduler(pid_t pid
, int policy
,
4116 struct sched_param __user
*param
)
4118 /* negative values for policy are not valid */
4122 return do_sched_setscheduler(pid
, policy
, param
);
4126 * sys_sched_setparam - set/change the RT priority of a thread
4127 * @pid: the pid in question.
4128 * @param: structure containing the new RT priority.
4130 asmlinkage
long sys_sched_setparam(pid_t pid
, struct sched_param __user
*param
)
4132 return do_sched_setscheduler(pid
, -1, param
);
4136 * sys_sched_getscheduler - get the policy (scheduling class) of a thread
4137 * @pid: the pid in question.
4139 asmlinkage
long sys_sched_getscheduler(pid_t pid
)
4141 int retval
= -EINVAL
;
4148 read_lock(&tasklist_lock
);
4149 p
= find_process_by_pid(pid
);
4151 retval
= security_task_getscheduler(p
);
4155 read_unlock(&tasklist_lock
);
4162 * sys_sched_getscheduler - get the RT priority of a thread
4163 * @pid: the pid in question.
4164 * @param: structure containing the RT priority.
4166 asmlinkage
long sys_sched_getparam(pid_t pid
, struct sched_param __user
*param
)
4168 struct sched_param lp
;
4169 int retval
= -EINVAL
;
4172 if (!param
|| pid
< 0)
4175 read_lock(&tasklist_lock
);
4176 p
= find_process_by_pid(pid
);
4181 retval
= security_task_getscheduler(p
);
4185 lp
.sched_priority
= p
->rt_priority
;
4186 read_unlock(&tasklist_lock
);
4189 * This one might sleep, we cannot do it with a spinlock held ...
4191 retval
= copy_to_user(param
, &lp
, sizeof(*param
)) ? -EFAULT
: 0;
4197 read_unlock(&tasklist_lock
);
4201 long sched_setaffinity(pid_t pid
, cpumask_t new_mask
)
4205 cpumask_t cpus_allowed
;
4208 read_lock(&tasklist_lock
);
4210 p
= find_process_by_pid(pid
);
4212 read_unlock(&tasklist_lock
);
4213 unlock_cpu_hotplug();
4218 * It is not safe to call set_cpus_allowed with the
4219 * tasklist_lock held. We will bump the task_struct's
4220 * usage count and then drop tasklist_lock.
4223 read_unlock(&tasklist_lock
);
4226 if ((current
->euid
!= p
->euid
) && (current
->euid
!= p
->uid
) &&
4227 !capable(CAP_SYS_NICE
))
4230 retval
= security_task_setscheduler(p
, 0, NULL
);
4234 cpus_allowed
= cpuset_cpus_allowed(p
);
4235 cpus_and(new_mask
, new_mask
, cpus_allowed
);
4236 retval
= set_cpus_allowed(p
, new_mask
);
4240 unlock_cpu_hotplug();
4244 static int get_user_cpu_mask(unsigned long __user
*user_mask_ptr
, unsigned len
,
4245 cpumask_t
*new_mask
)
4247 if (len
< sizeof(cpumask_t
)) {
4248 memset(new_mask
, 0, sizeof(cpumask_t
));
4249 } else if (len
> sizeof(cpumask_t
)) {
4250 len
= sizeof(cpumask_t
);
4252 return copy_from_user(new_mask
, user_mask_ptr
, len
) ? -EFAULT
: 0;
4256 * sys_sched_setaffinity - set the cpu affinity of a process
4257 * @pid: pid of the process
4258 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
4259 * @user_mask_ptr: user-space pointer to the new cpu mask
4261 asmlinkage
long sys_sched_setaffinity(pid_t pid
, unsigned int len
,
4262 unsigned long __user
*user_mask_ptr
)
4267 retval
= get_user_cpu_mask(user_mask_ptr
, len
, &new_mask
);
4271 return sched_setaffinity(pid
, new_mask
);
4275 * Represents all cpu's present in the system
4276 * In systems capable of hotplug, this map could dynamically grow
4277 * as new cpu's are detected in the system via any platform specific
4278 * method, such as ACPI for e.g.
4281 cpumask_t cpu_present_map __read_mostly
;
4282 EXPORT_SYMBOL(cpu_present_map
);
4285 cpumask_t cpu_online_map __read_mostly
= CPU_MASK_ALL
;
4286 cpumask_t cpu_possible_map __read_mostly
= CPU_MASK_ALL
;
4289 long sched_getaffinity(pid_t pid
, cpumask_t
*mask
)
4295 read_lock(&tasklist_lock
);
4298 p
= find_process_by_pid(pid
);
4302 retval
= security_task_getscheduler(p
);
4306 cpus_and(*mask
, p
->cpus_allowed
, cpu_online_map
);
4309 read_unlock(&tasklist_lock
);
4310 unlock_cpu_hotplug();
4318 * sys_sched_getaffinity - get the cpu affinity of a process
4319 * @pid: pid of the process
4320 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
4321 * @user_mask_ptr: user-space pointer to hold the current cpu mask
4323 asmlinkage
long sys_sched_getaffinity(pid_t pid
, unsigned int len
,
4324 unsigned long __user
*user_mask_ptr
)
4329 if (len
< sizeof(cpumask_t
))
4332 ret
= sched_getaffinity(pid
, &mask
);
4336 if (copy_to_user(user_mask_ptr
, &mask
, sizeof(cpumask_t
)))
4339 return sizeof(cpumask_t
);
4343 * sys_sched_yield - yield the current processor to other threads.
4345 * this function yields the current CPU by moving the calling thread
4346 * to the expired array. If there are no other threads running on this
4347 * CPU then this function will return.
4349 asmlinkage
long sys_sched_yield(void)
4351 runqueue_t
*rq
= this_rq_lock();
4352 prio_array_t
*array
= current
->array
;
4353 prio_array_t
*target
= rq
->expired
;
4355 schedstat_inc(rq
, yld_cnt
);
4357 * We implement yielding by moving the task into the expired
4360 * (special rule: RT tasks will just roundrobin in the active
4363 if (rt_task(current
))
4364 target
= rq
->active
;
4366 if (array
->nr_active
== 1) {
4367 schedstat_inc(rq
, yld_act_empty
);
4368 if (!rq
->expired
->nr_active
)
4369 schedstat_inc(rq
, yld_both_empty
);
4370 } else if (!rq
->expired
->nr_active
)
4371 schedstat_inc(rq
, yld_exp_empty
);
4373 if (array
!= target
) {
4374 dequeue_task(current
, array
);
4375 enqueue_task(current
, target
);
4378 * requeue_task is cheaper so perform that if possible.
4380 requeue_task(current
, array
);
4383 * Since we are going to call schedule() anyway, there's
4384 * no need to preempt or enable interrupts:
4386 __release(rq
->lock
);
4387 _raw_spin_unlock(&rq
->lock
);
4388 preempt_enable_no_resched();
4395 static inline int __resched_legal(void)
4397 if (unlikely(preempt_count()))
4399 if (unlikely(system_state
!= SYSTEM_RUNNING
))
4404 static void __cond_resched(void)
4406 #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP
4407 __might_sleep(__FILE__
, __LINE__
);
4410 * The BKS might be reacquired before we have dropped
4411 * PREEMPT_ACTIVE, which could trigger a second
4412 * cond_resched() call.
4415 add_preempt_count(PREEMPT_ACTIVE
);
4417 sub_preempt_count(PREEMPT_ACTIVE
);
4418 } while (need_resched());
4421 int __sched
cond_resched(void)
4423 if (need_resched() && __resched_legal()) {
4429 EXPORT_SYMBOL(cond_resched
);
4432 * cond_resched_lock() - if a reschedule is pending, drop the given lock,
4433 * call schedule, and on return reacquire the lock.
4435 * This works OK both with and without CONFIG_PREEMPT. We do strange low-level
4436 * operations here to prevent schedule() from being called twice (once via
4437 * spin_unlock(), once by hand).
4439 int cond_resched_lock(spinlock_t
*lock
)
4443 if (need_lockbreak(lock
)) {
4449 if (need_resched() && __resched_legal()) {
4450 _raw_spin_unlock(lock
);
4451 preempt_enable_no_resched();
4458 EXPORT_SYMBOL(cond_resched_lock
);
4460 int __sched
cond_resched_softirq(void)
4462 BUG_ON(!in_softirq());
4464 if (need_resched() && __resched_legal()) {
4465 __local_bh_enable();
4472 EXPORT_SYMBOL(cond_resched_softirq
);
4475 * yield - yield the current processor to other threads.
4477 * this is a shortcut for kernel-space yielding - it marks the
4478 * thread runnable and calls sys_sched_yield().
4480 void __sched
yield(void)
4482 set_current_state(TASK_RUNNING
);
4486 EXPORT_SYMBOL(yield
);
4489 * This task is about to go to sleep on IO. Increment rq->nr_iowait so
4490 * that process accounting knows that this is a task in IO wait state.
4492 * But don't do that if it is a deliberate, throttling IO wait (this task
4493 * has set its backing_dev_info: the queue against which it should throttle)
4495 void __sched
io_schedule(void)
4497 struct runqueue
*rq
= &__raw_get_cpu_var(runqueues
);
4499 atomic_inc(&rq
->nr_iowait
);
4501 atomic_dec(&rq
->nr_iowait
);
4504 EXPORT_SYMBOL(io_schedule
);
4506 long __sched
io_schedule_timeout(long timeout
)
4508 struct runqueue
*rq
= &__raw_get_cpu_var(runqueues
);
4511 atomic_inc(&rq
->nr_iowait
);
4512 ret
= schedule_timeout(timeout
);
4513 atomic_dec(&rq
->nr_iowait
);
4518 * sys_sched_get_priority_max - return maximum RT priority.
4519 * @policy: scheduling class.
4521 * this syscall returns the maximum rt_priority that can be used
4522 * by a given scheduling class.
4524 asmlinkage
long sys_sched_get_priority_max(int policy
)
4531 ret
= MAX_USER_RT_PRIO
-1;
4542 * sys_sched_get_priority_min - return minimum RT priority.
4543 * @policy: scheduling class.
4545 * this syscall returns the minimum rt_priority that can be used
4546 * by a given scheduling class.
4548 asmlinkage
long sys_sched_get_priority_min(int policy
)
4565 * sys_sched_rr_get_interval - return the default timeslice of a process.
4566 * @pid: pid of the process.
4567 * @interval: userspace pointer to the timeslice value.
4569 * this syscall writes the default timeslice value of a given process
4570 * into the user-space timespec buffer. A value of '0' means infinity.
4573 long sys_sched_rr_get_interval(pid_t pid
, struct timespec __user
*interval
)
4575 int retval
= -EINVAL
;
4583 read_lock(&tasklist_lock
);
4584 p
= find_process_by_pid(pid
);
4588 retval
= security_task_getscheduler(p
);
4592 jiffies_to_timespec(p
->policy
== SCHED_FIFO
?
4593 0 : task_timeslice(p
), &t
);
4594 read_unlock(&tasklist_lock
);
4595 retval
= copy_to_user(interval
, &t
, sizeof(t
)) ? -EFAULT
: 0;
4599 read_unlock(&tasklist_lock
);
4603 static inline struct task_struct
*eldest_child(struct task_struct
*p
)
4605 if (list_empty(&p
->children
)) return NULL
;
4606 return list_entry(p
->children
.next
,struct task_struct
,sibling
);
4609 static inline struct task_struct
*older_sibling(struct task_struct
*p
)
4611 if (p
->sibling
.prev
==&p
->parent
->children
) return NULL
;
4612 return list_entry(p
->sibling
.prev
,struct task_struct
,sibling
);
4615 static inline struct task_struct
*younger_sibling(struct task_struct
*p
)
4617 if (p
->sibling
.next
==&p
->parent
->children
) return NULL
;
4618 return list_entry(p
->sibling
.next
,struct task_struct
,sibling
);
4621 static void show_task(task_t
*p
)
4625 unsigned long free
= 0;
4626 static const char *stat_nam
[] = { "R", "S", "D", "T", "t", "Z", "X" };
4628 printk("%-13.13s ", p
->comm
);
4629 state
= p
->state
? __ffs(p
->state
) + 1 : 0;
4630 if (state
< ARRAY_SIZE(stat_nam
))
4631 printk(stat_nam
[state
]);
4634 #if (BITS_PER_LONG == 32)
4635 if (state
== TASK_RUNNING
)
4636 printk(" running ");
4638 printk(" %08lX ", thread_saved_pc(p
));
4640 if (state
== TASK_RUNNING
)
4641 printk(" running task ");
4643 printk(" %016lx ", thread_saved_pc(p
));
4645 #ifdef CONFIG_DEBUG_STACK_USAGE
4647 unsigned long *n
= end_of_stack(p
);
4650 free
= (unsigned long)n
- (unsigned long)end_of_stack(p
);
4653 printk("%5lu %5d %6d ", free
, p
->pid
, p
->parent
->pid
);
4654 if ((relative
= eldest_child(p
)))
4655 printk("%5d ", relative
->pid
);
4658 if ((relative
= younger_sibling(p
)))
4659 printk("%7d", relative
->pid
);
4662 if ((relative
= older_sibling(p
)))
4663 printk(" %5d", relative
->pid
);
4667 printk(" (L-TLB)\n");
4669 printk(" (NOTLB)\n");
4671 if (state
!= TASK_RUNNING
)
4672 show_stack(p
, NULL
);
4675 void show_state(void)
4679 #if (BITS_PER_LONG == 32)
4682 printk(" task PC pid father child younger older\n");
4686 printk(" task PC pid father child younger older\n");
4688 read_lock(&tasklist_lock
);
4689 do_each_thread(g
, p
) {
4691 * reset the NMI-timeout, listing all files on a slow
4692 * console might take alot of time:
4694 touch_nmi_watchdog();
4696 } while_each_thread(g
, p
);
4698 read_unlock(&tasklist_lock
);
4699 debug_show_all_locks();
4703 * init_idle - set up an idle thread for a given CPU
4704 * @idle: task in question
4705 * @cpu: cpu the idle task belongs to
4707 * NOTE: this function does not set the idle thread's NEED_RESCHED
4708 * flag, to make booting more robust.
4710 void __devinit
init_idle(task_t
*idle
, int cpu
)
4712 runqueue_t
*rq
= cpu_rq(cpu
);
4713 unsigned long flags
;
4715 idle
->timestamp
= sched_clock();
4716 idle
->sleep_avg
= 0;
4718 idle
->prio
= idle
->normal_prio
= MAX_PRIO
;
4719 idle
->state
= TASK_RUNNING
;
4720 idle
->cpus_allowed
= cpumask_of_cpu(cpu
);
4721 set_task_cpu(idle
, cpu
);
4723 spin_lock_irqsave(&rq
->lock
, flags
);
4724 rq
->curr
= rq
->idle
= idle
;
4725 #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW)
4728 spin_unlock_irqrestore(&rq
->lock
, flags
);
4730 /* Set the preempt count _outside_ the spinlocks! */
4731 #if defined(CONFIG_PREEMPT) && !defined(CONFIG_PREEMPT_BKL)
4732 task_thread_info(idle
)->preempt_count
= (idle
->lock_depth
>= 0);
4734 task_thread_info(idle
)->preempt_count
= 0;
4739 * In a system that switches off the HZ timer nohz_cpu_mask
4740 * indicates which cpus entered this state. This is used
4741 * in the rcu update to wait only for active cpus. For system
4742 * which do not switch off the HZ timer nohz_cpu_mask should
4743 * always be CPU_MASK_NONE.
4745 cpumask_t nohz_cpu_mask
= CPU_MASK_NONE
;
4749 * This is how migration works:
4751 * 1) we queue a migration_req_t structure in the source CPU's
4752 * runqueue and wake up that CPU's migration thread.
4753 * 2) we down() the locked semaphore => thread blocks.
4754 * 3) migration thread wakes up (implicitly it forces the migrated
4755 * thread off the CPU)
4756 * 4) it gets the migration request and checks whether the migrated
4757 * task is still in the wrong runqueue.
4758 * 5) if it's in the wrong runqueue then the migration thread removes
4759 * it and puts it into the right queue.
4760 * 6) migration thread up()s the semaphore.
4761 * 7) we wake up and the migration is done.
4765 * Change a given task's CPU affinity. Migrate the thread to a
4766 * proper CPU and schedule it away if the CPU it's executing on
4767 * is removed from the allowed bitmask.
4769 * NOTE: the caller must have a valid reference to the task, the
4770 * task must not exit() & deallocate itself prematurely. The
4771 * call is not atomic; no spinlocks may be held.
4773 int set_cpus_allowed(task_t
*p
, cpumask_t new_mask
)
4775 unsigned long flags
;
4777 migration_req_t req
;
4780 rq
= task_rq_lock(p
, &flags
);
4781 if (!cpus_intersects(new_mask
, cpu_online_map
)) {
4786 p
->cpus_allowed
= new_mask
;
4787 /* Can the task run on the task's current CPU? If so, we're done */
4788 if (cpu_isset(task_cpu(p
), new_mask
))
4791 if (migrate_task(p
, any_online_cpu(new_mask
), &req
)) {
4792 /* Need help from migration thread: drop lock and wait. */
4793 task_rq_unlock(rq
, &flags
);
4794 wake_up_process(rq
->migration_thread
);
4795 wait_for_completion(&req
.done
);
4796 tlb_migrate_finish(p
->mm
);
4800 task_rq_unlock(rq
, &flags
);
4804 EXPORT_SYMBOL_GPL(set_cpus_allowed
);
4807 * Move (not current) task off this cpu, onto dest cpu. We're doing
4808 * this because either it can't run here any more (set_cpus_allowed()
4809 * away from this CPU, or CPU going down), or because we're
4810 * attempting to rebalance this task on exec (sched_exec).
4812 * So we race with normal scheduler movements, but that's OK, as long
4813 * as the task is no longer on this CPU.
4815 * Returns non-zero if task was successfully migrated.
4817 static int __migrate_task(struct task_struct
*p
, int src_cpu
, int dest_cpu
)
4819 runqueue_t
*rq_dest
, *rq_src
;
4822 if (unlikely(cpu_is_offline(dest_cpu
)))
4825 rq_src
= cpu_rq(src_cpu
);
4826 rq_dest
= cpu_rq(dest_cpu
);
4828 double_rq_lock(rq_src
, rq_dest
);
4829 /* Already moved. */
4830 if (task_cpu(p
) != src_cpu
)
4832 /* Affinity changed (again). */
4833 if (!cpu_isset(dest_cpu
, p
->cpus_allowed
))
4836 set_task_cpu(p
, dest_cpu
);
4839 * Sync timestamp with rq_dest's before activating.
4840 * The same thing could be achieved by doing this step
4841 * afterwards, and pretending it was a local activate.
4842 * This way is cleaner and logically correct.
4844 p
->timestamp
= p
->timestamp
- rq_src
->timestamp_last_tick
4845 + rq_dest
->timestamp_last_tick
;
4846 deactivate_task(p
, rq_src
);
4847 activate_task(p
, rq_dest
, 0);
4848 if (TASK_PREEMPTS_CURR(p
, rq_dest
))
4849 resched_task(rq_dest
->curr
);
4853 double_rq_unlock(rq_src
, rq_dest
);
4858 * migration_thread - this is a highprio system thread that performs
4859 * thread migration by bumping thread off CPU then 'pushing' onto
4862 static int migration_thread(void *data
)
4865 int cpu
= (long)data
;
4868 BUG_ON(rq
->migration_thread
!= current
);
4870 set_current_state(TASK_INTERRUPTIBLE
);
4871 while (!kthread_should_stop()) {
4872 struct list_head
*head
;
4873 migration_req_t
*req
;
4877 spin_lock_irq(&rq
->lock
);
4879 if (cpu_is_offline(cpu
)) {
4880 spin_unlock_irq(&rq
->lock
);
4884 if (rq
->active_balance
) {
4885 active_load_balance(rq
, cpu
);
4886 rq
->active_balance
= 0;
4889 head
= &rq
->migration_queue
;
4891 if (list_empty(head
)) {
4892 spin_unlock_irq(&rq
->lock
);
4894 set_current_state(TASK_INTERRUPTIBLE
);
4897 req
= list_entry(head
->next
, migration_req_t
, list
);
4898 list_del_init(head
->next
);
4900 spin_unlock(&rq
->lock
);
4901 __migrate_task(req
->task
, cpu
, req
->dest_cpu
);
4904 complete(&req
->done
);
4906 __set_current_state(TASK_RUNNING
);
4910 /* Wait for kthread_stop */
4911 set_current_state(TASK_INTERRUPTIBLE
);
4912 while (!kthread_should_stop()) {
4914 set_current_state(TASK_INTERRUPTIBLE
);
4916 __set_current_state(TASK_RUNNING
);
4920 #ifdef CONFIG_HOTPLUG_CPU
4921 /* Figure out where task on dead CPU should go, use force if neccessary. */
4922 static void move_task_off_dead_cpu(int dead_cpu
, struct task_struct
*tsk
)
4925 unsigned long flags
;
4931 mask
= node_to_cpumask(cpu_to_node(dead_cpu
));
4932 cpus_and(mask
, mask
, tsk
->cpus_allowed
);
4933 dest_cpu
= any_online_cpu(mask
);
4935 /* On any allowed CPU? */
4936 if (dest_cpu
== NR_CPUS
)
4937 dest_cpu
= any_online_cpu(tsk
->cpus_allowed
);
4939 /* No more Mr. Nice Guy. */
4940 if (dest_cpu
== NR_CPUS
) {
4941 rq
= task_rq_lock(tsk
, &flags
);
4942 cpus_setall(tsk
->cpus_allowed
);
4943 dest_cpu
= any_online_cpu(tsk
->cpus_allowed
);
4944 task_rq_unlock(rq
, &flags
);
4947 * Don't tell them about moving exiting tasks or
4948 * kernel threads (both mm NULL), since they never
4951 if (tsk
->mm
&& printk_ratelimit())
4952 printk(KERN_INFO
"process %d (%s) no "
4953 "longer affine to cpu%d\n",
4954 tsk
->pid
, tsk
->comm
, dead_cpu
);
4956 if (!__migrate_task(tsk
, dead_cpu
, dest_cpu
))
4961 * While a dead CPU has no uninterruptible tasks queued at this point,
4962 * it might still have a nonzero ->nr_uninterruptible counter, because
4963 * for performance reasons the counter is not stricly tracking tasks to
4964 * their home CPUs. So we just add the counter to another CPU's counter,
4965 * to keep the global sum constant after CPU-down:
4967 static void migrate_nr_uninterruptible(runqueue_t
*rq_src
)
4969 runqueue_t
*rq_dest
= cpu_rq(any_online_cpu(CPU_MASK_ALL
));
4970 unsigned long flags
;
4972 local_irq_save(flags
);
4973 double_rq_lock(rq_src
, rq_dest
);
4974 rq_dest
->nr_uninterruptible
+= rq_src
->nr_uninterruptible
;
4975 rq_src
->nr_uninterruptible
= 0;
4976 double_rq_unlock(rq_src
, rq_dest
);
4977 local_irq_restore(flags
);
4980 /* Run through task list and migrate tasks from the dead cpu. */
4981 static void migrate_live_tasks(int src_cpu
)
4983 struct task_struct
*tsk
, *t
;
4985 write_lock_irq(&tasklist_lock
);
4987 do_each_thread(t
, tsk
) {
4991 if (task_cpu(tsk
) == src_cpu
)
4992 move_task_off_dead_cpu(src_cpu
, tsk
);
4993 } while_each_thread(t
, tsk
);
4995 write_unlock_irq(&tasklist_lock
);
4998 /* Schedules idle task to be the next runnable task on current CPU.
4999 * It does so by boosting its priority to highest possible and adding it to
5000 * the _front_ of runqueue. Used by CPU offline code.
5002 void sched_idle_next(void)
5004 int cpu
= smp_processor_id();
5005 runqueue_t
*rq
= this_rq();
5006 struct task_struct
*p
= rq
->idle
;
5007 unsigned long flags
;
5009 /* cpu has to be offline */
5010 BUG_ON(cpu_online(cpu
));
5012 /* Strictly not necessary since rest of the CPUs are stopped by now
5013 * and interrupts disabled on current cpu.
5015 spin_lock_irqsave(&rq
->lock
, flags
);
5017 __setscheduler(p
, SCHED_FIFO
, MAX_RT_PRIO
-1);
5018 /* Add idle task to _front_ of it's priority queue */
5019 __activate_idle_task(p
, rq
);
5021 spin_unlock_irqrestore(&rq
->lock
, flags
);
5024 /* Ensures that the idle task is using init_mm right before its cpu goes
5027 void idle_task_exit(void)
5029 struct mm_struct
*mm
= current
->active_mm
;
5031 BUG_ON(cpu_online(smp_processor_id()));
5034 switch_mm(mm
, &init_mm
, current
);
5038 static void migrate_dead(unsigned int dead_cpu
, task_t
*tsk
)
5040 struct runqueue
*rq
= cpu_rq(dead_cpu
);
5042 /* Must be exiting, otherwise would be on tasklist. */
5043 BUG_ON(tsk
->exit_state
!= EXIT_ZOMBIE
&& tsk
->exit_state
!= EXIT_DEAD
);
5045 /* Cannot have done final schedule yet: would have vanished. */
5046 BUG_ON(tsk
->flags
& PF_DEAD
);
5048 get_task_struct(tsk
);
5051 * Drop lock around migration; if someone else moves it,
5052 * that's OK. No task can be added to this CPU, so iteration is
5055 spin_unlock_irq(&rq
->lock
);
5056 move_task_off_dead_cpu(dead_cpu
, tsk
);
5057 spin_lock_irq(&rq
->lock
);
5059 put_task_struct(tsk
);
5062 /* release_task() removes task from tasklist, so we won't find dead tasks. */
5063 static void migrate_dead_tasks(unsigned int dead_cpu
)
5066 struct runqueue
*rq
= cpu_rq(dead_cpu
);
5068 for (arr
= 0; arr
< 2; arr
++) {
5069 for (i
= 0; i
< MAX_PRIO
; i
++) {
5070 struct list_head
*list
= &rq
->arrays
[arr
].queue
[i
];
5071 while (!list_empty(list
))
5072 migrate_dead(dead_cpu
,
5073 list_entry(list
->next
, task_t
,
5078 #endif /* CONFIG_HOTPLUG_CPU */
5081 * migration_call - callback that gets triggered when a CPU is added.
5082 * Here we can start up the necessary migration thread for the new CPU.
5084 static int __cpuinit
migration_call(struct notifier_block
*nfb
,
5085 unsigned long action
,
5088 int cpu
= (long)hcpu
;
5089 struct task_struct
*p
;
5090 struct runqueue
*rq
;
5091 unsigned long flags
;
5094 case CPU_UP_PREPARE
:
5095 p
= kthread_create(migration_thread
, hcpu
, "migration/%d",cpu
);
5098 p
->flags
|= PF_NOFREEZE
;
5099 kthread_bind(p
, cpu
);
5100 /* Must be high prio: stop_machine expects to yield to it. */
5101 rq
= task_rq_lock(p
, &flags
);
5102 __setscheduler(p
, SCHED_FIFO
, MAX_RT_PRIO
-1);
5103 task_rq_unlock(rq
, &flags
);
5104 cpu_rq(cpu
)->migration_thread
= p
;
5107 /* Strictly unneccessary, as first user will wake it. */
5108 wake_up_process(cpu_rq(cpu
)->migration_thread
);
5110 #ifdef CONFIG_HOTPLUG_CPU
5111 case CPU_UP_CANCELED
:
5112 if (!cpu_rq(cpu
)->migration_thread
)
5114 /* Unbind it from offline cpu so it can run. Fall thru. */
5115 kthread_bind(cpu_rq(cpu
)->migration_thread
,
5116 any_online_cpu(cpu_online_map
));
5117 kthread_stop(cpu_rq(cpu
)->migration_thread
);
5118 cpu_rq(cpu
)->migration_thread
= NULL
;
5121 migrate_live_tasks(cpu
);
5123 kthread_stop(rq
->migration_thread
);
5124 rq
->migration_thread
= NULL
;
5125 /* Idle task back to normal (off runqueue, low prio) */
5126 rq
= task_rq_lock(rq
->idle
, &flags
);
5127 deactivate_task(rq
->idle
, rq
);
5128 rq
->idle
->static_prio
= MAX_PRIO
;
5129 __setscheduler(rq
->idle
, SCHED_NORMAL
, 0);
5130 migrate_dead_tasks(cpu
);
5131 task_rq_unlock(rq
, &flags
);
5132 migrate_nr_uninterruptible(rq
);
5133 BUG_ON(rq
->nr_running
!= 0);
5135 /* No need to migrate the tasks: it was best-effort if
5136 * they didn't do lock_cpu_hotplug(). Just wake up
5137 * the requestors. */
5138 spin_lock_irq(&rq
->lock
);
5139 while (!list_empty(&rq
->migration_queue
)) {
5140 migration_req_t
*req
;
5141 req
= list_entry(rq
->migration_queue
.next
,
5142 migration_req_t
, list
);
5143 list_del_init(&req
->list
);
5144 complete(&req
->done
);
5146 spin_unlock_irq(&rq
->lock
);
5153 /* Register at highest priority so that task migration (migrate_all_tasks)
5154 * happens before everything else.
5156 static struct notifier_block __cpuinitdata migration_notifier
= {
5157 .notifier_call
= migration_call
,
5161 int __init
migration_init(void)
5163 void *cpu
= (void *)(long)smp_processor_id();
5164 /* Start one for boot CPU. */
5165 migration_call(&migration_notifier
, CPU_UP_PREPARE
, cpu
);
5166 migration_call(&migration_notifier
, CPU_ONLINE
, cpu
);
5167 register_cpu_notifier(&migration_notifier
);
5173 #undef SCHED_DOMAIN_DEBUG
5174 #ifdef SCHED_DOMAIN_DEBUG
5175 static void sched_domain_debug(struct sched_domain
*sd
, int cpu
)
5180 printk(KERN_DEBUG
"CPU%d attaching NULL sched-domain.\n", cpu
);
5184 printk(KERN_DEBUG
"CPU%d attaching sched-domain:\n", cpu
);
5189 struct sched_group
*group
= sd
->groups
;
5190 cpumask_t groupmask
;
5192 cpumask_scnprintf(str
, NR_CPUS
, sd
->span
);
5193 cpus_clear(groupmask
);
5196 for (i
= 0; i
< level
+ 1; i
++)
5198 printk("domain %d: ", level
);
5200 if (!(sd
->flags
& SD_LOAD_BALANCE
)) {
5201 printk("does not load-balance\n");
5203 printk(KERN_ERR
"ERROR: !SD_LOAD_BALANCE domain has parent");
5207 printk("span %s\n", str
);
5209 if (!cpu_isset(cpu
, sd
->span
))
5210 printk(KERN_ERR
"ERROR: domain->span does not contain CPU%d\n", cpu
);
5211 if (!cpu_isset(cpu
, group
->cpumask
))
5212 printk(KERN_ERR
"ERROR: domain->groups does not contain CPU%d\n", cpu
);
5215 for (i
= 0; i
< level
+ 2; i
++)
5221 printk(KERN_ERR
"ERROR: group is NULL\n");
5225 if (!group
->cpu_power
) {
5227 printk(KERN_ERR
"ERROR: domain->cpu_power not set\n");
5230 if (!cpus_weight(group
->cpumask
)) {
5232 printk(KERN_ERR
"ERROR: empty group\n");
5235 if (cpus_intersects(groupmask
, group
->cpumask
)) {
5237 printk(KERN_ERR
"ERROR: repeated CPUs\n");
5240 cpus_or(groupmask
, groupmask
, group
->cpumask
);
5242 cpumask_scnprintf(str
, NR_CPUS
, group
->cpumask
);
5245 group
= group
->next
;
5246 } while (group
!= sd
->groups
);
5249 if (!cpus_equal(sd
->span
, groupmask
))
5250 printk(KERN_ERR
"ERROR: groups don't span domain->span\n");
5256 if (!cpus_subset(groupmask
, sd
->span
))
5257 printk(KERN_ERR
"ERROR: parent span is not a superset of domain->span\n");
5263 #define sched_domain_debug(sd, cpu) {}
5266 static int sd_degenerate(struct sched_domain
*sd
)
5268 if (cpus_weight(sd
->span
) == 1)
5271 /* Following flags need at least 2 groups */
5272 if (sd
->flags
& (SD_LOAD_BALANCE
|
5273 SD_BALANCE_NEWIDLE
|
5276 if (sd
->groups
!= sd
->groups
->next
)
5280 /* Following flags don't use groups */
5281 if (sd
->flags
& (SD_WAKE_IDLE
|
5289 static int sd_parent_degenerate(struct sched_domain
*sd
,
5290 struct sched_domain
*parent
)
5292 unsigned long cflags
= sd
->flags
, pflags
= parent
->flags
;
5294 if (sd_degenerate(parent
))
5297 if (!cpus_equal(sd
->span
, parent
->span
))
5300 /* Does parent contain flags not in child? */
5301 /* WAKE_BALANCE is a subset of WAKE_AFFINE */
5302 if (cflags
& SD_WAKE_AFFINE
)
5303 pflags
&= ~SD_WAKE_BALANCE
;
5304 /* Flags needing groups don't count if only 1 group in parent */
5305 if (parent
->groups
== parent
->groups
->next
) {
5306 pflags
&= ~(SD_LOAD_BALANCE
|
5307 SD_BALANCE_NEWIDLE
|
5311 if (~cflags
& pflags
)
5318 * Attach the domain 'sd' to 'cpu' as its base domain. Callers must
5319 * hold the hotplug lock.
5321 static void cpu_attach_domain(struct sched_domain
*sd
, int cpu
)
5323 runqueue_t
*rq
= cpu_rq(cpu
);
5324 struct sched_domain
*tmp
;
5326 /* Remove the sched domains which do not contribute to scheduling. */
5327 for (tmp
= sd
; tmp
; tmp
= tmp
->parent
) {
5328 struct sched_domain
*parent
= tmp
->parent
;
5331 if (sd_parent_degenerate(tmp
, parent
))
5332 tmp
->parent
= parent
->parent
;
5335 if (sd
&& sd_degenerate(sd
))
5338 sched_domain_debug(sd
, cpu
);
5340 rcu_assign_pointer(rq
->sd
, sd
);
5343 /* cpus with isolated domains */
5344 static cpumask_t __devinitdata cpu_isolated_map
= CPU_MASK_NONE
;
5346 /* Setup the mask of cpus configured for isolated domains */
5347 static int __init
isolated_cpu_setup(char *str
)
5349 int ints
[NR_CPUS
], i
;
5351 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
5352 cpus_clear(cpu_isolated_map
);
5353 for (i
= 1; i
<= ints
[0]; i
++)
5354 if (ints
[i
] < NR_CPUS
)
5355 cpu_set(ints
[i
], cpu_isolated_map
);
5359 __setup ("isolcpus=", isolated_cpu_setup
);
5362 * init_sched_build_groups takes an array of groups, the cpumask we wish
5363 * to span, and a pointer to a function which identifies what group a CPU
5364 * belongs to. The return value of group_fn must be a valid index into the
5365 * groups[] array, and must be >= 0 and < NR_CPUS (due to the fact that we
5366 * keep track of groups covered with a cpumask_t).
5368 * init_sched_build_groups will build a circular linked list of the groups
5369 * covered by the given span, and will set each group's ->cpumask correctly,
5370 * and ->cpu_power to 0.
5372 static void init_sched_build_groups(struct sched_group groups
[], cpumask_t span
,
5373 int (*group_fn
)(int cpu
))
5375 struct sched_group
*first
= NULL
, *last
= NULL
;
5376 cpumask_t covered
= CPU_MASK_NONE
;
5379 for_each_cpu_mask(i
, span
) {
5380 int group
= group_fn(i
);
5381 struct sched_group
*sg
= &groups
[group
];
5384 if (cpu_isset(i
, covered
))
5387 sg
->cpumask
= CPU_MASK_NONE
;
5390 for_each_cpu_mask(j
, span
) {
5391 if (group_fn(j
) != group
)
5394 cpu_set(j
, covered
);
5395 cpu_set(j
, sg
->cpumask
);
5406 #define SD_NODES_PER_DOMAIN 16
5409 * Self-tuning task migration cost measurement between source and target CPUs.
5411 * This is done by measuring the cost of manipulating buffers of varying
5412 * sizes. For a given buffer-size here are the steps that are taken:
5414 * 1) the source CPU reads+dirties a shared buffer
5415 * 2) the target CPU reads+dirties the same shared buffer
5417 * We measure how long they take, in the following 4 scenarios:
5419 * - source: CPU1, target: CPU2 | cost1
5420 * - source: CPU2, target: CPU1 | cost2
5421 * - source: CPU1, target: CPU1 | cost3
5422 * - source: CPU2, target: CPU2 | cost4
5424 * We then calculate the cost3+cost4-cost1-cost2 difference - this is
5425 * the cost of migration.
5427 * We then start off from a small buffer-size and iterate up to larger
5428 * buffer sizes, in 5% steps - measuring each buffer-size separately, and
5429 * doing a maximum search for the cost. (The maximum cost for a migration
5430 * normally occurs when the working set size is around the effective cache
5433 #define SEARCH_SCOPE 2
5434 #define MIN_CACHE_SIZE (64*1024U)
5435 #define DEFAULT_CACHE_SIZE (5*1024*1024U)
5436 #define ITERATIONS 1
5437 #define SIZE_THRESH 130
5438 #define COST_THRESH 130
5441 * The migration cost is a function of 'domain distance'. Domain
5442 * distance is the number of steps a CPU has to iterate down its
5443 * domain tree to share a domain with the other CPU. The farther
5444 * two CPUs are from each other, the larger the distance gets.
5446 * Note that we use the distance only to cache measurement results,
5447 * the distance value is not used numerically otherwise. When two
5448 * CPUs have the same distance it is assumed that the migration
5449 * cost is the same. (this is a simplification but quite practical)
5451 #define MAX_DOMAIN_DISTANCE 32
5453 static unsigned long long migration_cost
[MAX_DOMAIN_DISTANCE
] =
5454 { [ 0 ... MAX_DOMAIN_DISTANCE
-1 ] =
5456 * Architectures may override the migration cost and thus avoid
5457 * boot-time calibration. Unit is nanoseconds. Mostly useful for
5458 * virtualized hardware:
5460 #ifdef CONFIG_DEFAULT_MIGRATION_COST
5461 CONFIG_DEFAULT_MIGRATION_COST
5468 * Allow override of migration cost - in units of microseconds.
5469 * E.g. migration_cost=1000,2000,3000 will set up a level-1 cost
5470 * of 1 msec, level-2 cost of 2 msecs and level3 cost of 3 msecs:
5472 static int __init
migration_cost_setup(char *str
)
5474 int ints
[MAX_DOMAIN_DISTANCE
+1], i
;
5476 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
5478 printk("#ints: %d\n", ints
[0]);
5479 for (i
= 1; i
<= ints
[0]; i
++) {
5480 migration_cost
[i
-1] = (unsigned long long)ints
[i
]*1000;
5481 printk("migration_cost[%d]: %Ld\n", i
-1, migration_cost
[i
-1]);
5486 __setup ("migration_cost=", migration_cost_setup
);
5489 * Global multiplier (divisor) for migration-cutoff values,
5490 * in percentiles. E.g. use a value of 150 to get 1.5 times
5491 * longer cache-hot cutoff times.
5493 * (We scale it from 100 to 128 to long long handling easier.)
5496 #define MIGRATION_FACTOR_SCALE 128
5498 static unsigned int migration_factor
= MIGRATION_FACTOR_SCALE
;
5500 static int __init
setup_migration_factor(char *str
)
5502 get_option(&str
, &migration_factor
);
5503 migration_factor
= migration_factor
* MIGRATION_FACTOR_SCALE
/ 100;
5507 __setup("migration_factor=", setup_migration_factor
);
5510 * Estimated distance of two CPUs, measured via the number of domains
5511 * we have to pass for the two CPUs to be in the same span:
5513 static unsigned long domain_distance(int cpu1
, int cpu2
)
5515 unsigned long distance
= 0;
5516 struct sched_domain
*sd
;
5518 for_each_domain(cpu1
, sd
) {
5519 WARN_ON(!cpu_isset(cpu1
, sd
->span
));
5520 if (cpu_isset(cpu2
, sd
->span
))
5524 if (distance
>= MAX_DOMAIN_DISTANCE
) {
5526 distance
= MAX_DOMAIN_DISTANCE
-1;
5532 static unsigned int migration_debug
;
5534 static int __init
setup_migration_debug(char *str
)
5536 get_option(&str
, &migration_debug
);
5540 __setup("migration_debug=", setup_migration_debug
);
5543 * Maximum cache-size that the scheduler should try to measure.
5544 * Architectures with larger caches should tune this up during
5545 * bootup. Gets used in the domain-setup code (i.e. during SMP
5548 unsigned int max_cache_size
;
5550 static int __init
setup_max_cache_size(char *str
)
5552 get_option(&str
, &max_cache_size
);
5556 __setup("max_cache_size=", setup_max_cache_size
);
5559 * Dirty a big buffer in a hard-to-predict (for the L2 cache) way. This
5560 * is the operation that is timed, so we try to generate unpredictable
5561 * cachemisses that still end up filling the L2 cache:
5563 static void touch_cache(void *__cache
, unsigned long __size
)
5565 unsigned long size
= __size
/sizeof(long), chunk1
= size
/3,
5567 unsigned long *cache
= __cache
;
5570 for (i
= 0; i
< size
/6; i
+= 8) {
5573 case 1: cache
[size
-1-i
]++;
5574 case 2: cache
[chunk1
-i
]++;
5575 case 3: cache
[chunk1
+i
]++;
5576 case 4: cache
[chunk2
-i
]++;
5577 case 5: cache
[chunk2
+i
]++;
5583 * Measure the cache-cost of one task migration. Returns in units of nsec.
5585 static unsigned long long measure_one(void *cache
, unsigned long size
,
5586 int source
, int target
)
5588 cpumask_t mask
, saved_mask
;
5589 unsigned long long t0
, t1
, t2
, t3
, cost
;
5591 saved_mask
= current
->cpus_allowed
;
5594 * Flush source caches to RAM and invalidate them:
5599 * Migrate to the source CPU:
5601 mask
= cpumask_of_cpu(source
);
5602 set_cpus_allowed(current
, mask
);
5603 WARN_ON(smp_processor_id() != source
);
5606 * Dirty the working set:
5609 touch_cache(cache
, size
);
5613 * Migrate to the target CPU, dirty the L2 cache and access
5614 * the shared buffer. (which represents the working set
5615 * of a migrated task.)
5617 mask
= cpumask_of_cpu(target
);
5618 set_cpus_allowed(current
, mask
);
5619 WARN_ON(smp_processor_id() != target
);
5622 touch_cache(cache
, size
);
5625 cost
= t1
-t0
+ t3
-t2
;
5627 if (migration_debug
>= 2)
5628 printk("[%d->%d]: %8Ld %8Ld %8Ld => %10Ld.\n",
5629 source
, target
, t1
-t0
, t1
-t0
, t3
-t2
, cost
);
5631 * Flush target caches to RAM and invalidate them:
5635 set_cpus_allowed(current
, saved_mask
);
5641 * Measure a series of task migrations and return the average
5642 * result. Since this code runs early during bootup the system
5643 * is 'undisturbed' and the average latency makes sense.
5645 * The algorithm in essence auto-detects the relevant cache-size,
5646 * so it will properly detect different cachesizes for different
5647 * cache-hierarchies, depending on how the CPUs are connected.
5649 * Architectures can prime the upper limit of the search range via
5650 * max_cache_size, otherwise the search range defaults to 20MB...64K.
5652 static unsigned long long
5653 measure_cost(int cpu1
, int cpu2
, void *cache
, unsigned int size
)
5655 unsigned long long cost1
, cost2
;
5659 * Measure the migration cost of 'size' bytes, over an
5660 * average of 10 runs:
5662 * (We perturb the cache size by a small (0..4k)
5663 * value to compensate size/alignment related artifacts.
5664 * We also subtract the cost of the operation done on
5670 * dry run, to make sure we start off cache-cold on cpu1,
5671 * and to get any vmalloc pagefaults in advance:
5673 measure_one(cache
, size
, cpu1
, cpu2
);
5674 for (i
= 0; i
< ITERATIONS
; i
++)
5675 cost1
+= measure_one(cache
, size
- i
*1024, cpu1
, cpu2
);
5677 measure_one(cache
, size
, cpu2
, cpu1
);
5678 for (i
= 0; i
< ITERATIONS
; i
++)
5679 cost1
+= measure_one(cache
, size
- i
*1024, cpu2
, cpu1
);
5682 * (We measure the non-migrating [cached] cost on both
5683 * cpu1 and cpu2, to handle CPUs with different speeds)
5687 measure_one(cache
, size
, cpu1
, cpu1
);
5688 for (i
= 0; i
< ITERATIONS
; i
++)
5689 cost2
+= measure_one(cache
, size
- i
*1024, cpu1
, cpu1
);
5691 measure_one(cache
, size
, cpu2
, cpu2
);
5692 for (i
= 0; i
< ITERATIONS
; i
++)
5693 cost2
+= measure_one(cache
, size
- i
*1024, cpu2
, cpu2
);
5696 * Get the per-iteration migration cost:
5698 do_div(cost1
, 2*ITERATIONS
);
5699 do_div(cost2
, 2*ITERATIONS
);
5701 return cost1
- cost2
;
5704 static unsigned long long measure_migration_cost(int cpu1
, int cpu2
)
5706 unsigned long long max_cost
= 0, fluct
= 0, avg_fluct
= 0;
5707 unsigned int max_size
, size
, size_found
= 0;
5708 long long cost
= 0, prev_cost
;
5712 * Search from max_cache_size*5 down to 64K - the real relevant
5713 * cachesize has to lie somewhere inbetween.
5715 if (max_cache_size
) {
5716 max_size
= max(max_cache_size
* SEARCH_SCOPE
, MIN_CACHE_SIZE
);
5717 size
= max(max_cache_size
/ SEARCH_SCOPE
, MIN_CACHE_SIZE
);
5720 * Since we have no estimation about the relevant
5723 max_size
= DEFAULT_CACHE_SIZE
* SEARCH_SCOPE
;
5724 size
= MIN_CACHE_SIZE
;
5727 if (!cpu_online(cpu1
) || !cpu_online(cpu2
)) {
5728 printk("cpu %d and %d not both online!\n", cpu1
, cpu2
);
5733 * Allocate the working set:
5735 cache
= vmalloc(max_size
);
5737 printk("could not vmalloc %d bytes for cache!\n", 2*max_size
);
5738 return 1000000; // return 1 msec on very small boxen
5741 while (size
<= max_size
) {
5743 cost
= measure_cost(cpu1
, cpu2
, cache
, size
);
5749 if (max_cost
< cost
) {
5755 * Calculate average fluctuation, we use this to prevent
5756 * noise from triggering an early break out of the loop:
5758 fluct
= abs(cost
- prev_cost
);
5759 avg_fluct
= (avg_fluct
+ fluct
)/2;
5761 if (migration_debug
)
5762 printk("-> [%d][%d][%7d] %3ld.%ld [%3ld.%ld] (%ld): (%8Ld %8Ld)\n",
5764 (long)cost
/ 1000000,
5765 ((long)cost
/ 100000) % 10,
5766 (long)max_cost
/ 1000000,
5767 ((long)max_cost
/ 100000) % 10,
5768 domain_distance(cpu1
, cpu2
),
5772 * If we iterated at least 20% past the previous maximum,
5773 * and the cost has dropped by more than 20% already,
5774 * (taking fluctuations into account) then we assume to
5775 * have found the maximum and break out of the loop early:
5777 if (size_found
&& (size
*100 > size_found
*SIZE_THRESH
))
5778 if (cost
+avg_fluct
<= 0 ||
5779 max_cost
*100 > (cost
+avg_fluct
)*COST_THRESH
) {
5781 if (migration_debug
)
5782 printk("-> found max.\n");
5786 * Increase the cachesize in 10% steps:
5788 size
= size
* 10 / 9;
5791 if (migration_debug
)
5792 printk("[%d][%d] working set size found: %d, cost: %Ld\n",
5793 cpu1
, cpu2
, size_found
, max_cost
);
5798 * A task is considered 'cache cold' if at least 2 times
5799 * the worst-case cost of migration has passed.
5801 * (this limit is only listened to if the load-balancing
5802 * situation is 'nice' - if there is a large imbalance we
5803 * ignore it for the sake of CPU utilization and
5804 * processing fairness.)
5806 return 2 * max_cost
* migration_factor
/ MIGRATION_FACTOR_SCALE
;
5809 static void calibrate_migration_costs(const cpumask_t
*cpu_map
)
5811 int cpu1
= -1, cpu2
= -1, cpu
, orig_cpu
= raw_smp_processor_id();
5812 unsigned long j0
, j1
, distance
, max_distance
= 0;
5813 struct sched_domain
*sd
;
5818 * First pass - calculate the cacheflush times:
5820 for_each_cpu_mask(cpu1
, *cpu_map
) {
5821 for_each_cpu_mask(cpu2
, *cpu_map
) {
5824 distance
= domain_distance(cpu1
, cpu2
);
5825 max_distance
= max(max_distance
, distance
);
5827 * No result cached yet?
5829 if (migration_cost
[distance
] == -1LL)
5830 migration_cost
[distance
] =
5831 measure_migration_cost(cpu1
, cpu2
);
5835 * Second pass - update the sched domain hierarchy with
5836 * the new cache-hot-time estimations:
5838 for_each_cpu_mask(cpu
, *cpu_map
) {
5840 for_each_domain(cpu
, sd
) {
5841 sd
->cache_hot_time
= migration_cost
[distance
];
5848 if (migration_debug
)
5849 printk("migration: max_cache_size: %d, cpu: %d MHz:\n",
5857 if (system_state
== SYSTEM_BOOTING
) {
5858 printk("migration_cost=");
5859 for (distance
= 0; distance
<= max_distance
; distance
++) {
5862 printk("%ld", (long)migration_cost
[distance
] / 1000);
5867 if (migration_debug
)
5868 printk("migration: %ld seconds\n", (j1
-j0
)/HZ
);
5871 * Move back to the original CPU. NUMA-Q gets confused
5872 * if we migrate to another quad during bootup.
5874 if (raw_smp_processor_id() != orig_cpu
) {
5875 cpumask_t mask
= cpumask_of_cpu(orig_cpu
),
5876 saved_mask
= current
->cpus_allowed
;
5878 set_cpus_allowed(current
, mask
);
5879 set_cpus_allowed(current
, saved_mask
);
5886 * find_next_best_node - find the next node to include in a sched_domain
5887 * @node: node whose sched_domain we're building
5888 * @used_nodes: nodes already in the sched_domain
5890 * Find the next node to include in a given scheduling domain. Simply
5891 * finds the closest node not already in the @used_nodes map.
5893 * Should use nodemask_t.
5895 static int find_next_best_node(int node
, unsigned long *used_nodes
)
5897 int i
, n
, val
, min_val
, best_node
= 0;
5901 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
5902 /* Start at @node */
5903 n
= (node
+ i
) % MAX_NUMNODES
;
5905 if (!nr_cpus_node(n
))
5908 /* Skip already used nodes */
5909 if (test_bit(n
, used_nodes
))
5912 /* Simple min distance search */
5913 val
= node_distance(node
, n
);
5915 if (val
< min_val
) {
5921 set_bit(best_node
, used_nodes
);
5926 * sched_domain_node_span - get a cpumask for a node's sched_domain
5927 * @node: node whose cpumask we're constructing
5928 * @size: number of nodes to include in this span
5930 * Given a node, construct a good cpumask for its sched_domain to span. It
5931 * should be one that prevents unnecessary balancing, but also spreads tasks
5934 static cpumask_t
sched_domain_node_span(int node
)
5937 cpumask_t span
, nodemask
;
5938 DECLARE_BITMAP(used_nodes
, MAX_NUMNODES
);
5941 bitmap_zero(used_nodes
, MAX_NUMNODES
);
5943 nodemask
= node_to_cpumask(node
);
5944 cpus_or(span
, span
, nodemask
);
5945 set_bit(node
, used_nodes
);
5947 for (i
= 1; i
< SD_NODES_PER_DOMAIN
; i
++) {
5948 int next_node
= find_next_best_node(node
, used_nodes
);
5949 nodemask
= node_to_cpumask(next_node
);
5950 cpus_or(span
, span
, nodemask
);
5957 int sched_smt_power_savings
= 0, sched_mc_power_savings
= 0;
5959 * At the moment, CONFIG_SCHED_SMT is never defined, but leave it in so we
5960 * can switch it on easily if needed.
5962 #ifdef CONFIG_SCHED_SMT
5963 static DEFINE_PER_CPU(struct sched_domain
, cpu_domains
);
5964 static struct sched_group sched_group_cpus
[NR_CPUS
];
5965 static int cpu_to_cpu_group(int cpu
)
5971 #ifdef CONFIG_SCHED_MC
5972 static DEFINE_PER_CPU(struct sched_domain
, core_domains
);
5973 static struct sched_group
*sched_group_core_bycpu
[NR_CPUS
];
5976 #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT)
5977 static int cpu_to_core_group(int cpu
)
5979 return first_cpu(cpu_sibling_map
[cpu
]);
5981 #elif defined(CONFIG_SCHED_MC)
5982 static int cpu_to_core_group(int cpu
)
5988 static DEFINE_PER_CPU(struct sched_domain
, phys_domains
);
5989 static struct sched_group
*sched_group_phys_bycpu
[NR_CPUS
];
5990 static int cpu_to_phys_group(int cpu
)
5992 #if defined(CONFIG_SCHED_MC)
5993 cpumask_t mask
= cpu_coregroup_map(cpu
);
5994 return first_cpu(mask
);
5995 #elif defined(CONFIG_SCHED_SMT)
5996 return first_cpu(cpu_sibling_map
[cpu
]);
6004 * The init_sched_build_groups can't handle what we want to do with node
6005 * groups, so roll our own. Now each node has its own list of groups which
6006 * gets dynamically allocated.
6008 static DEFINE_PER_CPU(struct sched_domain
, node_domains
);
6009 static struct sched_group
**sched_group_nodes_bycpu
[NR_CPUS
];
6011 static DEFINE_PER_CPU(struct sched_domain
, allnodes_domains
);
6012 static struct sched_group
*sched_group_allnodes_bycpu
[NR_CPUS
];
6014 static int cpu_to_allnodes_group(int cpu
)
6016 return cpu_to_node(cpu
);
6018 static void init_numa_sched_groups_power(struct sched_group
*group_head
)
6020 struct sched_group
*sg
= group_head
;
6026 for_each_cpu_mask(j
, sg
->cpumask
) {
6027 struct sched_domain
*sd
;
6029 sd
= &per_cpu(phys_domains
, j
);
6030 if (j
!= first_cpu(sd
->groups
->cpumask
)) {
6032 * Only add "power" once for each
6038 sg
->cpu_power
+= sd
->groups
->cpu_power
;
6041 if (sg
!= group_head
)
6046 /* Free memory allocated for various sched_group structures */
6047 static void free_sched_groups(const cpumask_t
*cpu_map
)
6053 for_each_cpu_mask(cpu
, *cpu_map
) {
6054 struct sched_group
*sched_group_allnodes
6055 = sched_group_allnodes_bycpu
[cpu
];
6056 struct sched_group
**sched_group_nodes
6057 = sched_group_nodes_bycpu
[cpu
];
6059 if (sched_group_allnodes
) {
6060 kfree(sched_group_allnodes
);
6061 sched_group_allnodes_bycpu
[cpu
] = NULL
;
6064 if (!sched_group_nodes
)
6067 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
6068 cpumask_t nodemask
= node_to_cpumask(i
);
6069 struct sched_group
*oldsg
, *sg
= sched_group_nodes
[i
];
6071 cpus_and(nodemask
, nodemask
, *cpu_map
);
6072 if (cpus_empty(nodemask
))
6082 if (oldsg
!= sched_group_nodes
[i
])
6085 kfree(sched_group_nodes
);
6086 sched_group_nodes_bycpu
[cpu
] = NULL
;
6089 for_each_cpu_mask(cpu
, *cpu_map
) {
6090 if (sched_group_phys_bycpu
[cpu
]) {
6091 kfree(sched_group_phys_bycpu
[cpu
]);
6092 sched_group_phys_bycpu
[cpu
] = NULL
;
6094 #ifdef CONFIG_SCHED_MC
6095 if (sched_group_core_bycpu
[cpu
]) {
6096 kfree(sched_group_core_bycpu
[cpu
]);
6097 sched_group_core_bycpu
[cpu
] = NULL
;
6104 * Build sched domains for a given set of cpus and attach the sched domains
6105 * to the individual cpus
6107 static int build_sched_domains(const cpumask_t
*cpu_map
)
6110 struct sched_group
*sched_group_phys
= NULL
;
6111 #ifdef CONFIG_SCHED_MC
6112 struct sched_group
*sched_group_core
= NULL
;
6115 struct sched_group
**sched_group_nodes
= NULL
;
6116 struct sched_group
*sched_group_allnodes
= NULL
;
6119 * Allocate the per-node list of sched groups
6121 sched_group_nodes
= kzalloc(sizeof(struct sched_group
*)*MAX_NUMNODES
,
6123 if (!sched_group_nodes
) {
6124 printk(KERN_WARNING
"Can not alloc sched group node list\n");
6127 sched_group_nodes_bycpu
[first_cpu(*cpu_map
)] = sched_group_nodes
;
6131 * Set up domains for cpus specified by the cpu_map.
6133 for_each_cpu_mask(i
, *cpu_map
) {
6135 struct sched_domain
*sd
= NULL
, *p
;
6136 cpumask_t nodemask
= node_to_cpumask(cpu_to_node(i
));
6138 cpus_and(nodemask
, nodemask
, *cpu_map
);
6141 if (cpus_weight(*cpu_map
)
6142 > SD_NODES_PER_DOMAIN
*cpus_weight(nodemask
)) {
6143 if (!sched_group_allnodes
) {
6144 sched_group_allnodes
6145 = kmalloc(sizeof(struct sched_group
)
6148 if (!sched_group_allnodes
) {
6150 "Can not alloc allnodes sched group\n");
6153 sched_group_allnodes_bycpu
[i
]
6154 = sched_group_allnodes
;
6156 sd
= &per_cpu(allnodes_domains
, i
);
6157 *sd
= SD_ALLNODES_INIT
;
6158 sd
->span
= *cpu_map
;
6159 group
= cpu_to_allnodes_group(i
);
6160 sd
->groups
= &sched_group_allnodes
[group
];
6165 sd
= &per_cpu(node_domains
, i
);
6167 sd
->span
= sched_domain_node_span(cpu_to_node(i
));
6169 cpus_and(sd
->span
, sd
->span
, *cpu_map
);
6172 if (!sched_group_phys
) {
6174 = kmalloc(sizeof(struct sched_group
) * NR_CPUS
,
6176 if (!sched_group_phys
) {
6177 printk (KERN_WARNING
"Can not alloc phys sched"
6181 sched_group_phys_bycpu
[i
] = sched_group_phys
;
6185 sd
= &per_cpu(phys_domains
, i
);
6186 group
= cpu_to_phys_group(i
);
6188 sd
->span
= nodemask
;
6190 sd
->groups
= &sched_group_phys
[group
];
6192 #ifdef CONFIG_SCHED_MC
6193 if (!sched_group_core
) {
6195 = kmalloc(sizeof(struct sched_group
) * NR_CPUS
,
6197 if (!sched_group_core
) {
6198 printk (KERN_WARNING
"Can not alloc core sched"
6202 sched_group_core_bycpu
[i
] = sched_group_core
;
6206 sd
= &per_cpu(core_domains
, i
);
6207 group
= cpu_to_core_group(i
);
6209 sd
->span
= cpu_coregroup_map(i
);
6210 cpus_and(sd
->span
, sd
->span
, *cpu_map
);
6212 sd
->groups
= &sched_group_core
[group
];
6215 #ifdef CONFIG_SCHED_SMT
6217 sd
= &per_cpu(cpu_domains
, i
);
6218 group
= cpu_to_cpu_group(i
);
6219 *sd
= SD_SIBLING_INIT
;
6220 sd
->span
= cpu_sibling_map
[i
];
6221 cpus_and(sd
->span
, sd
->span
, *cpu_map
);
6223 sd
->groups
= &sched_group_cpus
[group
];
6227 #ifdef CONFIG_SCHED_SMT
6228 /* Set up CPU (sibling) groups */
6229 for_each_cpu_mask(i
, *cpu_map
) {
6230 cpumask_t this_sibling_map
= cpu_sibling_map
[i
];
6231 cpus_and(this_sibling_map
, this_sibling_map
, *cpu_map
);
6232 if (i
!= first_cpu(this_sibling_map
))
6235 init_sched_build_groups(sched_group_cpus
, this_sibling_map
,
6240 #ifdef CONFIG_SCHED_MC
6241 /* Set up multi-core groups */
6242 for_each_cpu_mask(i
, *cpu_map
) {
6243 cpumask_t this_core_map
= cpu_coregroup_map(i
);
6244 cpus_and(this_core_map
, this_core_map
, *cpu_map
);
6245 if (i
!= first_cpu(this_core_map
))
6247 init_sched_build_groups(sched_group_core
, this_core_map
,
6248 &cpu_to_core_group
);
6253 /* Set up physical groups */
6254 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
6255 cpumask_t nodemask
= node_to_cpumask(i
);
6257 cpus_and(nodemask
, nodemask
, *cpu_map
);
6258 if (cpus_empty(nodemask
))
6261 init_sched_build_groups(sched_group_phys
, nodemask
,
6262 &cpu_to_phys_group
);
6266 /* Set up node groups */
6267 if (sched_group_allnodes
)
6268 init_sched_build_groups(sched_group_allnodes
, *cpu_map
,
6269 &cpu_to_allnodes_group
);
6271 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
6272 /* Set up node groups */
6273 struct sched_group
*sg
, *prev
;
6274 cpumask_t nodemask
= node_to_cpumask(i
);
6275 cpumask_t domainspan
;
6276 cpumask_t covered
= CPU_MASK_NONE
;
6279 cpus_and(nodemask
, nodemask
, *cpu_map
);
6280 if (cpus_empty(nodemask
)) {
6281 sched_group_nodes
[i
] = NULL
;
6285 domainspan
= sched_domain_node_span(i
);
6286 cpus_and(domainspan
, domainspan
, *cpu_map
);
6288 sg
= kmalloc_node(sizeof(struct sched_group
), GFP_KERNEL
, i
);
6290 printk(KERN_WARNING
"Can not alloc domain group for "
6294 sched_group_nodes
[i
] = sg
;
6295 for_each_cpu_mask(j
, nodemask
) {
6296 struct sched_domain
*sd
;
6297 sd
= &per_cpu(node_domains
, j
);
6301 sg
->cpumask
= nodemask
;
6303 cpus_or(covered
, covered
, nodemask
);
6306 for (j
= 0; j
< MAX_NUMNODES
; j
++) {
6307 cpumask_t tmp
, notcovered
;
6308 int n
= (i
+ j
) % MAX_NUMNODES
;
6310 cpus_complement(notcovered
, covered
);
6311 cpus_and(tmp
, notcovered
, *cpu_map
);
6312 cpus_and(tmp
, tmp
, domainspan
);
6313 if (cpus_empty(tmp
))
6316 nodemask
= node_to_cpumask(n
);
6317 cpus_and(tmp
, tmp
, nodemask
);
6318 if (cpus_empty(tmp
))
6321 sg
= kmalloc_node(sizeof(struct sched_group
),
6325 "Can not alloc domain group for node %d\n", j
);
6330 sg
->next
= prev
->next
;
6331 cpus_or(covered
, covered
, tmp
);
6338 /* Calculate CPU power for physical packages and nodes */
6339 #ifdef CONFIG_SCHED_SMT
6340 for_each_cpu_mask(i
, *cpu_map
) {
6341 struct sched_domain
*sd
;
6342 sd
= &per_cpu(cpu_domains
, i
);
6343 sd
->groups
->cpu_power
= SCHED_LOAD_SCALE
;
6346 #ifdef CONFIG_SCHED_MC
6347 for_each_cpu_mask(i
, *cpu_map
) {
6349 struct sched_domain
*sd
;
6350 sd
= &per_cpu(core_domains
, i
);
6351 if (sched_smt_power_savings
)
6352 power
= SCHED_LOAD_SCALE
* cpus_weight(sd
->groups
->cpumask
);
6354 power
= SCHED_LOAD_SCALE
+ (cpus_weight(sd
->groups
->cpumask
)-1)
6355 * SCHED_LOAD_SCALE
/ 10;
6356 sd
->groups
->cpu_power
= power
;
6360 for_each_cpu_mask(i
, *cpu_map
) {
6361 struct sched_domain
*sd
;
6362 #ifdef CONFIG_SCHED_MC
6363 sd
= &per_cpu(phys_domains
, i
);
6364 if (i
!= first_cpu(sd
->groups
->cpumask
))
6367 sd
->groups
->cpu_power
= 0;
6368 if (sched_mc_power_savings
|| sched_smt_power_savings
) {
6371 for_each_cpu_mask(j
, sd
->groups
->cpumask
) {
6372 struct sched_domain
*sd1
;
6373 sd1
= &per_cpu(core_domains
, j
);
6375 * for each core we will add once
6376 * to the group in physical domain
6378 if (j
!= first_cpu(sd1
->groups
->cpumask
))
6381 if (sched_smt_power_savings
)
6382 sd
->groups
->cpu_power
+= sd1
->groups
->cpu_power
;
6384 sd
->groups
->cpu_power
+= SCHED_LOAD_SCALE
;
6388 * This has to be < 2 * SCHED_LOAD_SCALE
6389 * Lets keep it SCHED_LOAD_SCALE, so that
6390 * while calculating NUMA group's cpu_power
6392 * numa_group->cpu_power += phys_group->cpu_power;
6394 * See "only add power once for each physical pkg"
6397 sd
->groups
->cpu_power
= SCHED_LOAD_SCALE
;
6400 sd
= &per_cpu(phys_domains
, i
);
6401 if (sched_smt_power_savings
)
6402 power
= SCHED_LOAD_SCALE
* cpus_weight(sd
->groups
->cpumask
);
6404 power
= SCHED_LOAD_SCALE
;
6405 sd
->groups
->cpu_power
= power
;
6410 for (i
= 0; i
< MAX_NUMNODES
; i
++)
6411 init_numa_sched_groups_power(sched_group_nodes
[i
]);
6413 init_numa_sched_groups_power(sched_group_allnodes
);
6416 /* Attach the domains */
6417 for_each_cpu_mask(i
, *cpu_map
) {
6418 struct sched_domain
*sd
;
6419 #ifdef CONFIG_SCHED_SMT
6420 sd
= &per_cpu(cpu_domains
, i
);
6421 #elif defined(CONFIG_SCHED_MC)
6422 sd
= &per_cpu(core_domains
, i
);
6424 sd
= &per_cpu(phys_domains
, i
);
6426 cpu_attach_domain(sd
, i
);
6429 * Tune cache-hot values:
6431 calibrate_migration_costs(cpu_map
);
6436 free_sched_groups(cpu_map
);
6440 * Set up scheduler domains and groups. Callers must hold the hotplug lock.
6442 static int arch_init_sched_domains(const cpumask_t
*cpu_map
)
6444 cpumask_t cpu_default_map
;
6448 * Setup mask for cpus without special case scheduling requirements.
6449 * For now this just excludes isolated cpus, but could be used to
6450 * exclude other special cases in the future.
6452 cpus_andnot(cpu_default_map
, *cpu_map
, cpu_isolated_map
);
6454 err
= build_sched_domains(&cpu_default_map
);
6459 static void arch_destroy_sched_domains(const cpumask_t
*cpu_map
)
6461 free_sched_groups(cpu_map
);
6465 * Detach sched domains from a group of cpus specified in cpu_map
6466 * These cpus will now be attached to the NULL domain
6468 static void detach_destroy_domains(const cpumask_t
*cpu_map
)
6472 for_each_cpu_mask(i
, *cpu_map
)
6473 cpu_attach_domain(NULL
, i
);
6474 synchronize_sched();
6475 arch_destroy_sched_domains(cpu_map
);
6479 * Partition sched domains as specified by the cpumasks below.
6480 * This attaches all cpus from the cpumasks to the NULL domain,
6481 * waits for a RCU quiescent period, recalculates sched
6482 * domain information and then attaches them back to the
6483 * correct sched domains
6484 * Call with hotplug lock held
6486 int partition_sched_domains(cpumask_t
*partition1
, cpumask_t
*partition2
)
6488 cpumask_t change_map
;
6491 cpus_and(*partition1
, *partition1
, cpu_online_map
);
6492 cpus_and(*partition2
, *partition2
, cpu_online_map
);
6493 cpus_or(change_map
, *partition1
, *partition2
);
6495 /* Detach sched domains from all of the affected cpus */
6496 detach_destroy_domains(&change_map
);
6497 if (!cpus_empty(*partition1
))
6498 err
= build_sched_domains(partition1
);
6499 if (!err
&& !cpus_empty(*partition2
))
6500 err
= build_sched_domains(partition2
);
6505 #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT)
6506 int arch_reinit_sched_domains(void)
6511 detach_destroy_domains(&cpu_online_map
);
6512 err
= arch_init_sched_domains(&cpu_online_map
);
6513 unlock_cpu_hotplug();
6518 static ssize_t
sched_power_savings_store(const char *buf
, size_t count
, int smt
)
6522 if (buf
[0] != '0' && buf
[0] != '1')
6526 sched_smt_power_savings
= (buf
[0] == '1');
6528 sched_mc_power_savings
= (buf
[0] == '1');
6530 ret
= arch_reinit_sched_domains();
6532 return ret
? ret
: count
;
6535 int sched_create_sysfs_power_savings_entries(struct sysdev_class
*cls
)
6538 #ifdef CONFIG_SCHED_SMT
6540 err
= sysfs_create_file(&cls
->kset
.kobj
,
6541 &attr_sched_smt_power_savings
.attr
);
6543 #ifdef CONFIG_SCHED_MC
6544 if (!err
&& mc_capable())
6545 err
= sysfs_create_file(&cls
->kset
.kobj
,
6546 &attr_sched_mc_power_savings
.attr
);
6552 #ifdef CONFIG_SCHED_MC
6553 static ssize_t
sched_mc_power_savings_show(struct sys_device
*dev
, char *page
)
6555 return sprintf(page
, "%u\n", sched_mc_power_savings
);
6557 static ssize_t
sched_mc_power_savings_store(struct sys_device
*dev
, const char *buf
, size_t count
)
6559 return sched_power_savings_store(buf
, count
, 0);
6561 SYSDEV_ATTR(sched_mc_power_savings
, 0644, sched_mc_power_savings_show
,
6562 sched_mc_power_savings_store
);
6565 #ifdef CONFIG_SCHED_SMT
6566 static ssize_t
sched_smt_power_savings_show(struct sys_device
*dev
, char *page
)
6568 return sprintf(page
, "%u\n", sched_smt_power_savings
);
6570 static ssize_t
sched_smt_power_savings_store(struct sys_device
*dev
, const char *buf
, size_t count
)
6572 return sched_power_savings_store(buf
, count
, 1);
6574 SYSDEV_ATTR(sched_smt_power_savings
, 0644, sched_smt_power_savings_show
,
6575 sched_smt_power_savings_store
);
6579 #ifdef CONFIG_HOTPLUG_CPU
6581 * Force a reinitialization of the sched domains hierarchy. The domains
6582 * and groups cannot be updated in place without racing with the balancing
6583 * code, so we temporarily attach all running cpus to the NULL domain
6584 * which will prevent rebalancing while the sched domains are recalculated.
6586 static int update_sched_domains(struct notifier_block
*nfb
,
6587 unsigned long action
, void *hcpu
)
6590 case CPU_UP_PREPARE
:
6591 case CPU_DOWN_PREPARE
:
6592 detach_destroy_domains(&cpu_online_map
);
6595 case CPU_UP_CANCELED
:
6596 case CPU_DOWN_FAILED
:
6600 * Fall through and re-initialise the domains.
6607 /* The hotplug lock is already held by cpu_up/cpu_down */
6608 arch_init_sched_domains(&cpu_online_map
);
6614 void __init
sched_init_smp(void)
6617 arch_init_sched_domains(&cpu_online_map
);
6618 unlock_cpu_hotplug();
6619 /* XXX: Theoretical race here - CPU may be hotplugged now */
6620 hotcpu_notifier(update_sched_domains
, 0);
6623 void __init
sched_init_smp(void)
6626 #endif /* CONFIG_SMP */
6628 int in_sched_functions(unsigned long addr
)
6630 /* Linker adds these: start and end of __sched functions */
6631 extern char __sched_text_start
[], __sched_text_end
[];
6632 return in_lock_functions(addr
) ||
6633 (addr
>= (unsigned long)__sched_text_start
6634 && addr
< (unsigned long)__sched_text_end
);
6637 void __init
sched_init(void)
6642 for_each_possible_cpu(i
) {
6643 prio_array_t
*array
;
6646 spin_lock_init(&rq
->lock
);
6648 rq
->active
= rq
->arrays
;
6649 rq
->expired
= rq
->arrays
+ 1;
6650 rq
->best_expired_prio
= MAX_PRIO
;
6654 for (j
= 1; j
< 3; j
++)
6655 rq
->cpu_load
[j
] = 0;
6656 rq
->active_balance
= 0;
6658 rq
->migration_thread
= NULL
;
6659 INIT_LIST_HEAD(&rq
->migration_queue
);
6661 atomic_set(&rq
->nr_iowait
, 0);
6663 for (j
= 0; j
< 2; j
++) {
6664 array
= rq
->arrays
+ j
;
6665 for (k
= 0; k
< MAX_PRIO
; k
++) {
6666 INIT_LIST_HEAD(array
->queue
+ k
);
6667 __clear_bit(k
, array
->bitmap
);
6669 // delimiter for bitsearch
6670 __set_bit(MAX_PRIO
, array
->bitmap
);
6674 set_load_weight(&init_task
);
6676 * The boot idle thread does lazy MMU switching as well:
6678 atomic_inc(&init_mm
.mm_count
);
6679 enter_lazy_tlb(&init_mm
, current
);
6682 * Make us the idle thread. Technically, schedule() should not be
6683 * called from this thread, however somewhere below it might be,
6684 * but because we are the idle thread, we just pick up running again
6685 * when this runqueue becomes "idle".
6687 init_idle(current
, smp_processor_id());
6690 #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP
6691 void __might_sleep(char *file
, int line
)
6693 #if defined(in_atomic)
6694 static unsigned long prev_jiffy
; /* ratelimiting */
6696 if ((in_atomic() || irqs_disabled()) &&
6697 system_state
== SYSTEM_RUNNING
&& !oops_in_progress
) {
6698 if (time_before(jiffies
, prev_jiffy
+ HZ
) && prev_jiffy
)
6700 prev_jiffy
= jiffies
;
6701 printk(KERN_ERR
"BUG: sleeping function called from invalid"
6702 " context at %s:%d\n", file
, line
);
6703 printk("in_atomic():%d, irqs_disabled():%d\n",
6704 in_atomic(), irqs_disabled());
6709 EXPORT_SYMBOL(__might_sleep
);
6712 #ifdef CONFIG_MAGIC_SYSRQ
6713 void normalize_rt_tasks(void)
6715 struct task_struct
*p
;
6716 prio_array_t
*array
;
6717 unsigned long flags
;
6720 read_lock_irq(&tasklist_lock
);
6721 for_each_process(p
) {
6725 spin_lock_irqsave(&p
->pi_lock
, flags
);
6726 rq
= __task_rq_lock(p
);
6730 deactivate_task(p
, task_rq(p
));
6731 __setscheduler(p
, SCHED_NORMAL
, 0);
6733 __activate_task(p
, task_rq(p
));
6734 resched_task(rq
->curr
);
6737 __task_rq_unlock(rq
);
6738 spin_unlock_irqrestore(&p
->pi_lock
, flags
);
6740 read_unlock_irq(&tasklist_lock
);
6743 #endif /* CONFIG_MAGIC_SYSRQ */
6747 * These functions are only useful for the IA64 MCA handling.
6749 * They can only be called when the whole system has been
6750 * stopped - every CPU needs to be quiescent, and no scheduling
6751 * activity can take place. Using them for anything else would
6752 * be a serious bug, and as a result, they aren't even visible
6753 * under any other configuration.
6757 * curr_task - return the current task for a given cpu.
6758 * @cpu: the processor in question.
6760 * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED!
6762 task_t
*curr_task(int cpu
)
6764 return cpu_curr(cpu
);
6768 * set_curr_task - set the current task for a given cpu.
6769 * @cpu: the processor in question.
6770 * @p: the task pointer to set.
6772 * Description: This function must only be used when non-maskable interrupts
6773 * are serviced on a separate stack. It allows the architecture to switch the
6774 * notion of the current task on a cpu in a non-blocking manner. This function
6775 * must be called with all CPU's synchronized, and interrupts disabled, the
6776 * and caller must save the original value of the current task (see
6777 * curr_task() above) and restore that value before reenabling interrupts and
6778 * re-starting the system.
6780 * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED!
6782 void set_curr_task(int cpu
, task_t
*p
)