Kernel

Kernel Git Source Tree

Root/kernel/sched.c

1	/*
2	* kernel/sched.c
3	*
4	* Kernel scheduler and related syscalls
5	*
6	* Copyright (C) 1991-2002 Linus Torvalds
7	*
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
11	* by Andrea Arcangeli
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
19	* 2007-04-15 Work begun on replacing all interactivity tuning with a
20	* fair scheduling design by Con Kolivas.
21	* 2007-05-05 Load balancing (smp-nice) and other improvements
22	* by Peter Williams
23	* 2007-05-06 Interactivity improvements to CFS by Mike Galbraith
24	* 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri
25	* 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins,
26	* Thomas Gleixner, Mike Kravetz
27	*/
28
29	#include <linux/mm.h>
30	#include <linux/module.h>
31	#include <linux/nmi.h>
32	#include <linux/init.h>
33	#include <linux/uaccess.h>
34	#include <linux/highmem.h>
35	#include <linux/smp_lock.h>
36	#include <asm/mmu_context.h>
37	#include <linux/interrupt.h>
38	#include <linux/capability.h>
39	#include <linux/completion.h>
40	#include <linux/kernel_stat.h>
41	#include <linux/debug_locks.h>
42	#include <linux/perf_event.h>
43	#include <linux/security.h>
44	#include <linux/notifier.h>
45	#include <linux/profile.h>
46	#include <linux/freezer.h>
47	#include <linux/vmalloc.h>
48	#include <linux/blkdev.h>
49	#include <linux/delay.h>
50	#include <linux/pid_namespace.h>
51	#include <linux/smp.h>
52	#include <linux/threads.h>
53	#include <linux/timer.h>
54	#include <linux/rcupdate.h>
55	#include <linux/cpu.h>
56	#include <linux/cpuset.h>
57	#include <linux/percpu.h>
58	#include <linux/kthread.h>
59	#include <linux/proc_fs.h>
60	#include <linux/seq_file.h>
61	#include <linux/sysctl.h>
62	#include <linux/syscalls.h>
63	#include <linux/times.h>
64	#include <linux/tsacct_kern.h>
65	#include <linux/kprobes.h>
66	#include <linux/delayacct.h>
67	#include <linux/unistd.h>
68	#include <linux/pagemap.h>
69	#include <linux/hrtimer.h>
70	#include <linux/tick.h>
71	#include <linux/debugfs.h>
72	#include <linux/ctype.h>
73	#include <linux/ftrace.h>
74	#include <linux/slab.h>
75
76	#include <asm/tlb.h>
77	#include <asm/irq_regs.h>
78
79	#include "sched_cpupri.h"
80
81	#define CREATE_TRACE_POINTS
82	#include <trace/events/sched.h>
83
84	/*
85	* Convert user-nice values [ -20 ... 0 ... 19 ]
86	* to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ],
87	* and back.
88	*/
89	#define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20)
90	#define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20)
91	#define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio)
92
93	/*
94	* 'User priority' is the nice value converted to something we
95	* can work with better when scaling various scheduler parameters,
96	* it's a [ 0 ... 39 ] range.
97	*/
98	#define USER_PRIO(p) ((p)-MAX_RT_PRIO)
99	#define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio)
100	#define MAX_USER_PRIO (USER_PRIO(MAX_PRIO))
101
102	/*
103	* Helpers for converting nanosecond timing to jiffy resolution
104	*/
105	#define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
106
107	#define NICE_0_LOAD SCHED_LOAD_SCALE
108	#define NICE_0_SHIFT SCHED_LOAD_SHIFT
109
110	/*
111	* These are the 'tuning knobs' of the scheduler:
112	*
113	* default timeslice is 100 msecs (used only for SCHED_RR tasks).
114	* Timeslices get refilled after they expire.
115	*/
116	#define DEF_TIMESLICE (100 * HZ / 1000)
117
118	/*
119	* single value that denotes runtime == period, ie unlimited time.
120	*/
121	#define RUNTIME_INF ((u64)~0ULL)
122
123	static inline int rt_policy(int policy)
124	{
125	if (unlikely(policy == SCHED_FIFO \|\| policy == SCHED_RR))
126	return 1;
127	return 0;
128	}
129
130	static inline int task_has_rt_policy(struct task_struct *p)
131	{
132	return rt_policy(p->policy);
133	}
134
135	/*
136	* This is the priority-queue data structure of the RT scheduling class:
137	*/
138	struct rt_prio_array {
139	DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
140	struct list_head queue[MAX_RT_PRIO];
141	};
142
143	struct rt_bandwidth {
144	/* nests inside the rq lock: */
145	raw_spinlock_t rt_runtime_lock;
146	ktime_t rt_period;
147	u64 rt_runtime;
148	struct hrtimer rt_period_timer;
149	};
150
151	static struct rt_bandwidth def_rt_bandwidth;
152
153	static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun);
154
155	static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer)
156	{
157	struct rt_bandwidth *rt_b =
158	container_of(timer, struct rt_bandwidth, rt_period_timer);
159	ktime_t now;
160	int overrun;
161	int idle = 0;
162
163	for (;;) {
164	now = hrtimer_cb_get_time(timer);
165	overrun = hrtimer_forward(timer, now, rt_b->rt_period);
166
167	if (!overrun)
168	break;
169
170	idle = do_sched_rt_period_timer(rt_b, overrun);
171	}
172
173	return idle ? HRTIMER_NORESTART : HRTIMER_RESTART;
174	}
175
176	static
177	void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime)
178	{
179	rt_b->rt_period = ns_to_ktime(period);
180	rt_b->rt_runtime = runtime;
181
182	raw_spin_lock_init(&rt_b->rt_runtime_lock);
183
184	hrtimer_init(&rt_b->rt_period_timer,
185	CLOCK_MONOTONIC, HRTIMER_MODE_REL);
186	rt_b->rt_period_timer.function = sched_rt_period_timer;
187	}
188
189	static inline int rt_bandwidth_enabled(void)
190	{
191	return sysctl_sched_rt_runtime >= 0;
192	}
193
194	static void start_rt_bandwidth(struct rt_bandwidth *rt_b)
195	{
196	ktime_t now;
197
198	if (!rt_bandwidth_enabled() \|\| rt_b->rt_runtime == RUNTIME_INF)
199	return;
200
201	if (hrtimer_active(&rt_b->rt_period_timer))
202	return;
203
204	raw_spin_lock(&rt_b->rt_runtime_lock);
205	for (;;) {
206	unsigned long delta;
207	ktime_t soft, hard;
208
209	if (hrtimer_active(&rt_b->rt_period_timer))
210	break;
211
212	now = hrtimer_cb_get_time(&rt_b->rt_period_timer);
213	hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period);
214
215	soft = hrtimer_get_softexpires(&rt_b->rt_period_timer);
216	hard = hrtimer_get_expires(&rt_b->rt_period_timer);
217	delta = ktime_to_ns(ktime_sub(hard, soft));
218	__hrtimer_start_range_ns(&rt_b->rt_period_timer, soft, delta,
219	HRTIMER_MODE_ABS_PINNED, 0);
220	}
221	raw_spin_unlock(&rt_b->rt_runtime_lock);
222	}
223
224	#ifdef CONFIG_RT_GROUP_SCHED
225	static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b)
226	{
227	hrtimer_cancel(&rt_b->rt_period_timer);
228	}
229	#endif
230
231	/*
232	* sched_domains_mutex serializes calls to arch_init_sched_domains,
233	* detach_destroy_domains and partition_sched_domains.
234	*/
235	static DEFINE_MUTEX(sched_domains_mutex);
236
237	#ifdef CONFIG_CGROUP_SCHED
238
239	#include <linux/cgroup.h>
240
241	struct cfs_rq;
242
243	static LIST_HEAD(task_groups);
244
245	/* task group related information */
246	struct task_group {
247	struct cgroup_subsys_state css;
248
249	#ifdef CONFIG_FAIR_GROUP_SCHED
250	/* schedulable entities of this group on each cpu */
251	struct sched_entity **se;
252	/* runqueue "owned" by this group on each cpu */
253	struct cfs_rq **cfs_rq;
254	unsigned long shares;
255	#endif
256
257	#ifdef CONFIG_RT_GROUP_SCHED
258	struct sched_rt_entity **rt_se;
259	struct rt_rq **rt_rq;
260
261	struct rt_bandwidth rt_bandwidth;
262	#endif
263
264	struct rcu_head rcu;
265	struct list_head list;
266
267	struct task_group *parent;
268	struct list_head siblings;
269	struct list_head children;
270	};
271
272	#define root_task_group init_task_group
273
274	/* task_group_lock serializes add/remove of task groups and also changes to
275	* a task group's cpu shares.
276	*/
277	static DEFINE_SPINLOCK(task_group_lock);
278
279	#ifdef CONFIG_FAIR_GROUP_SCHED
280
281	#ifdef CONFIG_SMP
282	static int root_task_group_empty(void)
283	{
284	return list_empty(&root_task_group.children);
285	}
286	#endif
287
288	# define INIT_TASK_GROUP_LOAD NICE_0_LOAD
289
290	/*
291	* A weight of 0 or 1 can cause arithmetics problems.
292	* A weight of a cfs_rq is the sum of weights of which entities
293	* are queued on this cfs_rq, so a weight of a entity should not be
294	* too large, so as the shares value of a task group.
295	* (The default weight is 1024 - so there's no practical
296	* limitation from this.)
297	*/
298	#define MIN_SHARES 2
299	#define MAX_SHARES (1UL << 18)
300
301	static int init_task_group_load = INIT_TASK_GROUP_LOAD;
302	#endif
303
304	/* Default task group.
305	* Every task in system belong to this group at bootup.
306	*/
307	struct task_group init_task_group;
308
309	/* return group to which a task belongs */
310	static inline struct task_group task_group(struct task_struct p)
311	{
312	struct task_group *tg;
313
314	#ifdef CONFIG_CGROUP_SCHED
315	tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id),
316	struct task_group, css);
317	#else
318	tg = &init_task_group;
319	#endif
320	return tg;
321	}
322
323	/* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
324	static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
325	{
326	#ifdef CONFIG_FAIR_GROUP_SCHED
327	p->se.cfs_rq = task_group(p)->cfs_rq[cpu];
328	p->se.parent = task_group(p)->se[cpu];
329	#endif
330
331	#ifdef CONFIG_RT_GROUP_SCHED
332	p->rt.rt_rq = task_group(p)->rt_rq[cpu];
333	p->rt.parent = task_group(p)->rt_se[cpu];
334	#endif
335	}
336
337	#else
338
339	static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
340	static inline struct task_group task_group(struct task_struct p)
341	{
342	return NULL;
343	}
344
345	#endif /* CONFIG_CGROUP_SCHED */
346
347	/* CFS-related fields in a runqueue */
348	struct cfs_rq {
349	struct load_weight load;
350	unsigned long nr_running;
351
352	u64 exec_clock;
353	u64 min_vruntime;
354
355	struct rb_root tasks_timeline;
356	struct rb_node *rb_leftmost;
357
358	struct list_head tasks;
359	struct list_head *balance_iterator;
360
361	/*
362	* 'curr' points to currently running entity on this cfs_rq.
363	* It is set to NULL otherwise (i.e when none are currently running).
364	*/
365	struct sched_entity curr, next, *last;
366
367	unsigned int nr_spread_over;
368
369	#ifdef CONFIG_FAIR_GROUP_SCHED
370	struct rq rq; / cpu runqueue to which this cfs_rq is attached */
371
372	/*
373	* leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
374	* a hierarchy). Non-leaf lrqs hold other higher schedulable entities
375	* (like users, containers etc.)
376	*
377	* leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
378	* list is used during load balance.
379	*/
380	struct list_head leaf_cfs_rq_list;
381	struct task_group tg; / group that "owns" this runqueue */
382
383	#ifdef CONFIG_SMP
384	/*
385	* the part of load.weight contributed by tasks
386	*/
387	unsigned long task_weight;
388
389	/*
390	* h_load = weight * f(tg)
391	*
392	* Where f(tg) is the recursive weight fraction assigned to
393	* this group.
394	*/
395	unsigned long h_load;
396
397	/*
398	* this cpu's part of tg->shares
399	*/
400	unsigned long shares;
401
402	/*
403	* load.weight at the time we set shares
404	*/
405	unsigned long rq_weight;
406	#endif
407	#endif
408	};
409
410	/* Real-Time classes' related field in a runqueue: */
411	struct rt_rq {
412	struct rt_prio_array active;
413	unsigned long rt_nr_running;
414	#if defined CONFIG_SMP \|\| defined CONFIG_RT_GROUP_SCHED
415	struct {
416	int curr; /* highest queued rt task prio */
417	#ifdef CONFIG_SMP
418	int next; /* next highest */
419	#endif
420	} highest_prio;
421	#endif
422	#ifdef CONFIG_SMP
423	unsigned long rt_nr_migratory;
424	unsigned long rt_nr_total;
425	int overloaded;
426	struct plist_head pushable_tasks;
427	#endif
428	int rt_throttled;
429	u64 rt_time;
430	u64 rt_runtime;
431	/* Nests inside the rq lock: */
432	raw_spinlock_t rt_runtime_lock;
433
434	#ifdef CONFIG_RT_GROUP_SCHED
435	unsigned long rt_nr_boosted;
436
437	struct rq *rq;
438	struct list_head leaf_rt_rq_list;
439	struct task_group *tg;
440	#endif
441	};
442
443	#ifdef CONFIG_SMP
444
445	/*
446	* We add the notion of a root-domain which will be used to define per-domain
447	* variables. Each exclusive cpuset essentially defines an island domain by
448	* fully partitioning the member cpus from any other cpuset. Whenever a new
449	* exclusive cpuset is created, we also create and attach a new root-domain
450	* object.
451	*
452	*/
453	struct root_domain {
454	atomic_t refcount;
455	cpumask_var_t span;
456	cpumask_var_t online;
457
458	/*
459	* The "RT overload" flag: it gets set if a CPU has more than
460	* one runnable RT task.
461	*/
462	cpumask_var_t rto_mask;
463	atomic_t rto_count;
464	#ifdef CONFIG_SMP
465	struct cpupri cpupri;
466	#endif
467	};
468
469	/*
470	* By default the system creates a single root-domain with all cpus as
471	* members (mimicking the global state we have today).
472	*/
473	static struct root_domain def_root_domain;
474
475	#endif
476
477	/*
478	* This is the main, per-CPU runqueue data structure.
479	*
480	* Locking rule: those places that want to lock multiple runqueues
481	* (such as the load balancing or the thread migration code), lock
482	* acquire operations must be ordered by ascending &runqueue.
483	*/
484	struct rq {
485	/* runqueue lock: */
486	raw_spinlock_t lock;
487
488	/*
489	* nr_running and cpu_load should be in the same cacheline because
490	* remote CPUs use both these fields when doing load calculation.
491	*/
492	unsigned long nr_running;
493	#define CPU_LOAD_IDX_MAX 5
494	unsigned long cpu_load[CPU_LOAD_IDX_MAX];
495	#ifdef CONFIG_NO_HZ
496	unsigned char in_nohz_recently;
497	#endif
498	/* capture load from all tasks on this cpu: */
499	struct load_weight load;
500	unsigned long nr_load_updates;
501	u64 nr_switches;
502
503	struct cfs_rq cfs;
504	struct rt_rq rt;
505
506	#ifdef CONFIG_FAIR_GROUP_SCHED
507	/* list of leaf cfs_rq on this cpu: */
508	struct list_head leaf_cfs_rq_list;
509	#endif
510	#ifdef CONFIG_RT_GROUP_SCHED
511	struct list_head leaf_rt_rq_list;
512	#endif
513
514	/*
515	* This is part of a global counter where only the total sum
516	* over all CPUs matters. A task can increase this counter on
517	* one CPU and if it got migrated afterwards it may decrease
518	* it on another CPU. Always updated under the runqueue lock:
519	*/
520	unsigned long nr_uninterruptible;
521
522	struct task_struct curr, idle;
523	unsigned long next_balance;
524	struct mm_struct *prev_mm;
525
526	u64 clock;
527
528	atomic_t nr_iowait;
529
530	#ifdef CONFIG_SMP
531	struct root_domain *rd;
532	struct sched_domain *sd;
533
534	unsigned char idle_at_tick;
535	/* For active balancing */
536	int post_schedule;
537	int active_balance;
538	int push_cpu;
539	/* cpu of this runqueue: */
540	int cpu;
541	int online;
542
543	unsigned long avg_load_per_task;
544
545	struct task_struct *migration_thread;
546	struct list_head migration_queue;
547
548	u64 rt_avg;
549	u64 age_stamp;
550	u64 idle_stamp;
551	u64 avg_idle;
552	#endif
553
554	/* calc_load related fields */
555	unsigned long calc_load_update;
556	long calc_load_active;
557
558	#ifdef CONFIG_SCHED_HRTICK
559	#ifdef CONFIG_SMP
560	int hrtick_csd_pending;
561	struct call_single_data hrtick_csd;
562	#endif
563	struct hrtimer hrtick_timer;
564	#endif
565
566	#ifdef CONFIG_SCHEDSTATS
567	/* latency stats */
568	struct sched_info rq_sched_info;
569	unsigned long long rq_cpu_time;
570	/* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
571
572	/* sys_sched_yield() stats */
573	unsigned int yld_count;
574
575	/* schedule() stats */
576	unsigned int sched_switch;
577	unsigned int sched_count;
578	unsigned int sched_goidle;
579
580	/* try_to_wake_up() stats */
581	unsigned int ttwu_count;
582	unsigned int ttwu_local;
583
584	/* BKL stats */
585	unsigned int bkl_count;
586	#endif
587	};
588
589	static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
590
591	static inline
592	void check_preempt_curr(struct rq rq, struct task_struct p, int flags)
593	{
594	rq->curr->sched_class->check_preempt_curr(rq, p, flags);
595	}
596
597	static inline int cpu_of(struct rq *rq)
598	{
599	#ifdef CONFIG_SMP
600	return rq->cpu;
601	#else
602	return 0;
603	#endif
604	}
605
606	#define rcu_dereference_check_sched_domain(p) \
607	rcu_dereference_check((p), \
608	rcu_read_lock_sched_held() \|\| \
609	lockdep_is_held(&sched_domains_mutex))
610
611	/*
612	* The domain tree (rq->sd) is protected by RCU's quiescent state transition.
613	* See detach_destroy_domains: synchronize_sched for details.
614	*
615	* The domain tree of any CPU may only be accessed from within
616	* preempt-disabled sections.
617	*/
618	#define for_each_domain(cpu, __sd) \
619	for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent)
620
621	#define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
622	#define this_rq() (&__get_cpu_var(runqueues))
623	#define task_rq(p) cpu_rq(task_cpu(p))
624	#define cpu_curr(cpu) (cpu_rq(cpu)->curr)
625	#define raw_rq() (&__raw_get_cpu_var(runqueues))
626
627	inline void update_rq_clock(struct rq *rq)
628	{
629	rq->clock = sched_clock_cpu(cpu_of(rq));
630	}
631
632	/*
633	* Tunables that become constants when CONFIG_SCHED_DEBUG is off:
634	*/
635	#ifdef CONFIG_SCHED_DEBUG
636	# define const_debug __read_mostly
637	#else
638	# define const_debug static const
639	#endif
640
641	/**
642	* runqueue_is_locked
643	* @cpu: the processor in question.
644	*
645	* Returns true if the current cpu runqueue is locked.
646	* This interface allows printk to be called with the runqueue lock
647	* held and know whether or not it is OK to wake up the klogd.
648	*/
649	int runqueue_is_locked(int cpu)
650	{
651	return raw_spin_is_locked(&cpu_rq(cpu)->lock);
652	}
653
654	/*
655	* Debugging: various feature bits
656	*/
657
658	#define SCHED_FEAT(name, enabled) \
659	__SCHED_FEAT_##name ,
660
661	enum {
662	#include "sched_features.h"
663	};
664
665	#undef SCHED_FEAT
666
667	#define SCHED_FEAT(name, enabled) \
668	(1UL << __SCHED_FEAT_##name) * enabled \|
669
670	const_debug unsigned int sysctl_sched_features =
671	#include "sched_features.h"
672	0;
673
674	#undef SCHED_FEAT
675
676	#ifdef CONFIG_SCHED_DEBUG
677	#define SCHED_FEAT(name, enabled) \
678	#name ,
679
680	static __read_mostly char *sched_feat_names[] = {
681	#include "sched_features.h"
682	NULL
683	};
684
685	#undef SCHED_FEAT
686
687	static int sched_feat_show(struct seq_file m, void v)
688	{
689	int i;
690
691	for (i = 0; sched_feat_names[i]; i++) {
692	if (!(sysctl_sched_features & (1UL << i)))
693	seq_puts(m, "NO_");
694	seq_printf(m, "%s ", sched_feat_names[i]);
695	}
696	seq_puts(m, "\n");
697
698	return 0;
699	}
700
701	static ssize_t
702	sched_feat_write(struct file filp, const char __user ubuf,
703	size_t cnt, loff_t *ppos)
704	{
705	char buf[64];
706	char *cmp = buf;
707	int neg = 0;
708	int i;
709
710	if (cnt > 63)
711	cnt = 63;
712
713	if (copy_from_user(&buf, ubuf, cnt))
714	return -EFAULT;
715
716	buf[cnt] = 0;
717
718	if (strncmp(buf, "NO_", 3) == 0) {
719	neg = 1;
720	cmp += 3;
721	}
722
723	for (i = 0; sched_feat_names[i]; i++) {
724	int len = strlen(sched_feat_names[i]);
725
726	if (strncmp(cmp, sched_feat_names[i], len) == 0) {
727	if (neg)
728	sysctl_sched_features &= ~(1UL << i);
729	else
730	sysctl_sched_features \|= (1UL << i);
731	break;
732	}
733	}
734
735	if (!sched_feat_names[i])
736	return -EINVAL;
737
738	*ppos += cnt;
739
740	return cnt;
741	}
742
743	static int sched_feat_open(struct inode inode, struct file filp)
744	{
745	return single_open(filp, sched_feat_show, NULL);
746	}
747
748	static const struct file_operations sched_feat_fops = {
749	.open = sched_feat_open,
750	.write = sched_feat_write,
751	.read = seq_read,
752	.llseek = seq_lseek,
753	.release = single_release,
754	};
755
756	static __init int sched_init_debug(void)
757	{
758	debugfs_create_file("sched_features", 0644, NULL, NULL,
759	&sched_feat_fops);
760
761	return 0;
762	}
763	late_initcall(sched_init_debug);
764
765	#endif
766
767	#define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
768
769	/*
770	* Number of tasks to iterate in a single balance run.
771	* Limited because this is done with IRQs disabled.
772	*/
773	const_debug unsigned int sysctl_sched_nr_migrate = 32;
774
775	/*
776	* ratelimit for updating the group shares.
777	* default: 0.25ms
778	*/
779	unsigned int sysctl_sched_shares_ratelimit = 250000;
780	unsigned int normalized_sysctl_sched_shares_ratelimit = 250000;
781
782	/*
783	* Inject some fuzzyness into changing the per-cpu group shares
784	* this avoids remote rq-locks at the expense of fairness.
785	* default: 4
786	*/
787	unsigned int sysctl_sched_shares_thresh = 4;
788
789	/*
790	* period over which we average the RT time consumption, measured
791	* in ms.
792	*
793	* default: 1s
794	*/
795	const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC;
796
797	/*
798	* period over which we measure -rt task cpu usage in us.
799	* default: 1s
800	*/
801	unsigned int sysctl_sched_rt_period = 1000000;
802
803	static __read_mostly int scheduler_running;
804
805	/*
806	* part of the period that we allow rt tasks to run in us.
807	* default: 0.95s
808	*/
809	int sysctl_sched_rt_runtime = 950000;
810
811	static inline u64 global_rt_period(void)
812	{
813	return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
814	}
815
816	static inline u64 global_rt_runtime(void)
817	{
818	if (sysctl_sched_rt_runtime < 0)
819	return RUNTIME_INF;
820
821	return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
822	}
823
824	#ifndef prepare_arch_switch
825	# define prepare_arch_switch(next) do { } while (0)
826	#endif
827	#ifndef finish_arch_switch
828	# define finish_arch_switch(prev) do { } while (0)
829	#endif
830
831	static inline int task_current(struct rq rq, struct task_struct p)
832	{
833	return rq->curr == p;
834	}
835
836	#ifndef __ARCH_WANT_UNLOCKED_CTXSW
837	static inline int task_running(struct rq rq, struct task_struct p)
838	{
839	return task_current(rq, p);
840	}
841
842	static inline void prepare_lock_switch(struct rq rq, struct task_struct next)
843	{
844	}
845
846	static inline void finish_lock_switch(struct rq rq, struct task_struct prev)
847	{
848	#ifdef CONFIG_DEBUG_SPINLOCK
849	/* this is a valid case when another task releases the spinlock */
850	rq->lock.owner = current;
851	#endif
852	/*
853	* If we are tracking spinlock dependencies then we have to
854	* fix up the runqueue lock - which gets 'carried over' from
855	* prev into current:
856	*/
857	spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
858
859	raw_spin_unlock_irq(&rq->lock);
860	}
861
862	#else /* __ARCH_WANT_UNLOCKED_CTXSW */
863	static inline int task_running(struct rq rq, struct task_struct p)
864	{
865	#ifdef CONFIG_SMP
866	return p->oncpu;
867	#else
868	return task_current(rq, p);
869	#endif
870	}
871
872	static inline void prepare_lock_switch(struct rq rq, struct task_struct next)
873	{
874	#ifdef CONFIG_SMP
875	/*
876	* We can optimise this out completely for !SMP, because the
877	* SMP rebalancing from interrupt is the only thing that cares
878	* here.
879	*/
880	next->oncpu = 1;
881	#endif
882	#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
883	raw_spin_unlock_irq(&rq->lock);
884	#else
885	raw_spin_unlock(&rq->lock);
886	#endif
887	}
888
889	static inline void finish_lock_switch(struct rq rq, struct task_struct prev)
890	{
891	#ifdef CONFIG_SMP
892	/*
893	* After ->oncpu is cleared, the task can be moved to a different CPU.
894	* We must ensure this doesn't happen until the switch is completely
895	* finished.
896	*/
897	smp_wmb();
898	prev->oncpu = 0;
899	#endif
900	#ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW
901	local_irq_enable();
902	#endif
903	}
904	#endif /* __ARCH_WANT_UNLOCKED_CTXSW */
905
906	/*
907	* Check whether the task is waking, we use this to synchronize against
908	* ttwu() so that task_cpu() reports a stable number.
909	*
910	* We need to make an exception for PF_STARTING tasks because the fork
911	* path might require task_rq_lock() to work, eg. it can call
912	* set_cpus_allowed_ptr() from the cpuset clone_ns code.
913	*/
914	static inline int task_is_waking(struct task_struct *p)
915	{
916	return unlikely((p->state == TASK_WAKING) && !(p->flags & PF_STARTING));
917	}
918
919	/*
920	* __task_rq_lock - lock the runqueue a given task resides on.
921	* Must be called interrupts disabled.
922	*/
923	static inline struct rq __task_rq_lock(struct task_struct p)
924	__acquires(rq->lock)
925	{
926	struct rq *rq;
927
928	for (;;) {
929	while (task_is_waking(p))
930	cpu_relax();
931	rq = task_rq(p);
932	raw_spin_lock(&rq->lock);
933	if (likely(rq == task_rq(p) && !task_is_waking(p)))
934	return rq;
935	raw_spin_unlock(&rq->lock);
936	}
937	}
938
939	/*
940	* task_rq_lock - lock the runqueue a given task resides on and disable
941	* interrupts. Note the ordering: we can safely lookup the task_rq without
942	* explicitly disabling preemption.
943	*/
944	static struct rq task_rq_lock(struct task_struct p, unsigned long *flags)
945	__acquires(rq->lock)
946	{
947	struct rq *rq;
948
949	for (;;) {
950	while (task_is_waking(p))
951	cpu_relax();
952	local_irq_save(*flags);
953	rq = task_rq(p);
954	raw_spin_lock(&rq->lock);
955	if (likely(rq == task_rq(p) && !task_is_waking(p)))
956	return rq;
957	raw_spin_unlock_irqrestore(&rq->lock, *flags);
958	}
959	}
960
961	void task_rq_unlock_wait(struct task_struct *p)
962	{
963	struct rq *rq = task_rq(p);
964
965	smp_mb(); /* spin-unlock-wait is not a full memory barrier */
966	raw_spin_unlock_wait(&rq->lock);
967	}
968
969	static void __task_rq_unlock(struct rq *rq)
970	__releases(rq->lock)
971	{
972	raw_spin_unlock(&rq->lock);
973	}
974
975	static inline void task_rq_unlock(struct rq rq, unsigned long flags)
976	__releases(rq->lock)
977	{
978	raw_spin_unlock_irqrestore(&rq->lock, *flags);
979	}
980
981	/*
982	* this_rq_lock - lock this runqueue and disable interrupts.
983	*/
984	static struct rq *this_rq_lock(void)
985	__acquires(rq->lock)
986	{
987	struct rq *rq;
988
989	local_irq_disable();
990	rq = this_rq();
991	raw_spin_lock(&rq->lock);
992
993	return rq;
994	}
995
996	#ifdef CONFIG_SCHED_HRTICK
997	/*
998	* Use HR-timers to deliver accurate preemption points.
999	*
1000	* Its all a bit involved since we cannot program an hrt while holding the
1001	* rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a
1002	* reschedule event.
1003	*
1004	* When we get rescheduled we reprogram the hrtick_timer outside of the
1005	* rq->lock.
1006	*/
1007
1008	/*
1009	* Use hrtick when:
1010	* - enabled by features
1011	* - hrtimer is actually high res
1012	*/
1013	static inline int hrtick_enabled(struct rq *rq)
1014	{
1015	if (!sched_feat(HRTICK))
1016	return 0;
1017	if (!cpu_active(cpu_of(rq)))
1018	return 0;
1019	return hrtimer_is_hres_active(&rq->hrtick_timer);
1020	}
1021
1022	static void hrtick_clear(struct rq *rq)
1023	{
1024	if (hrtimer_active(&rq->hrtick_timer))
1025	hrtimer_cancel(&rq->hrtick_timer);
1026	}
1027
1028	/*
1029	* High-resolution timer tick.
1030	* Runs from hardirq context with interrupts disabled.
1031	*/
1032	static enum hrtimer_restart hrtick(struct hrtimer *timer)
1033	{
1034	struct rq *rq = container_of(timer, struct rq, hrtick_timer);
1035
1036	WARN_ON_ONCE(cpu_of(rq) != smp_processor_id());
1037
1038	raw_spin_lock(&rq->lock);
1039	update_rq_clock(rq);
1040	rq->curr->sched_class->task_tick(rq, rq->curr, 1);
1041	raw_spin_unlock(&rq->lock);
1042
1043	return HRTIMER_NORESTART;
1044	}
1045
1046	#ifdef CONFIG_SMP
1047	/*
1048	* called from hardirq (IPI) context
1049	*/
1050	static void __hrtick_start(void *arg)
1051	{
1052	struct rq *rq = arg;
1053
1054	raw_spin_lock(&rq->lock);
1055	hrtimer_restart(&rq->hrtick_timer);
1056	rq->hrtick_csd_pending = 0;
1057	raw_spin_unlock(&rq->lock);
1058	}
1059
1060	/*
1061	* Called to set the hrtick timer state.
1062	*
1063	* called with rq->lock held and irqs disabled
1064	*/
1065	static void hrtick_start(struct rq *rq, u64 delay)
1066	{
1067	struct hrtimer *timer = &rq->hrtick_timer;
1068	ktime_t time = ktime_add_ns(timer->base->get_time(), delay);
1069
1070	hrtimer_set_expires(timer, time);
1071
1072	if (rq == this_rq()) {
1073	hrtimer_restart(timer);
1074	} else if (!rq->hrtick_csd_pending) {
1075	__smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0);
1076	rq->hrtick_csd_pending = 1;
1077	}
1078	}
1079
1080	static int
1081	hotplug_hrtick(struct notifier_block nfb, unsigned long action, void hcpu)
1082	{
1083	int cpu = (int)(long)hcpu;
1084
1085	switch (action) {
1086	case CPU_UP_CANCELED:
1087	case CPU_UP_CANCELED_FROZEN:
1088	case CPU_DOWN_PREPARE:
1089	case CPU_DOWN_PREPARE_FROZEN:
1090	case CPU_DEAD:
1091	case CPU_DEAD_FROZEN:
1092	hrtick_clear(cpu_rq(cpu));
1093	return NOTIFY_OK;
1094	}
1095
1096	return NOTIFY_DONE;
1097	}
1098
1099	static __init void init_hrtick(void)
1100	{
1101	hotcpu_notifier(hotplug_hrtick, 0);
1102	}
1103	#else
1104	/*
1105	* Called to set the hrtick timer state.
1106	*
1107	* called with rq->lock held and irqs disabled
1108	*/
1109	static void hrtick_start(struct rq *rq, u64 delay)
1110	{
1111	__hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0,
1112	HRTIMER_MODE_REL_PINNED, 0);
1113	}
1114
1115	static inline void init_hrtick(void)
1116	{
1117	}
1118	#endif /* CONFIG_SMP */
1119
1120	static void init_rq_hrtick(struct rq *rq)
1121	{
1122	#ifdef CONFIG_SMP
1123	rq->hrtick_csd_pending = 0;
1124
1125	rq->hrtick_csd.flags = 0;
1126	rq->hrtick_csd.func = __hrtick_start;
1127	rq->hrtick_csd.info = rq;
1128	#endif
1129
1130	hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1131	rq->hrtick_timer.function = hrtick;
1132	}
1133	#else /* CONFIG_SCHED_HRTICK */
1134	static inline void hrtick_clear(struct rq *rq)
1135	{
1136	}
1137
1138	static inline void init_rq_hrtick(struct rq *rq)
1139	{
1140	}
1141
1142	static inline void init_hrtick(void)
1143	{
1144	}
1145	#endif /* CONFIG_SCHED_HRTICK */
1146
1147	/*
1148	* resched_task - mark a task 'to be rescheduled now'.
1149	*
1150	* On UP this means the setting of the need_resched flag, on SMP it
1151	* might also involve a cross-CPU call to trigger the scheduler on
1152	* the target CPU.
1153	*/
1154	#ifdef CONFIG_SMP
1155
1156	#ifndef tsk_is_polling
1157	#define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG)
1158	#endif
1159
1160	static void resched_task(struct task_struct *p)
1161	{
1162	int cpu;
1163
1164	assert_raw_spin_locked(&task_rq(p)->lock);
1165
1166	if (test_tsk_need_resched(p))
1167	return;
1168
1169	set_tsk_need_resched(p);
1170
1171	cpu = task_cpu(p);
1172	if (cpu == smp_processor_id())
1173	return;
1174
1175	/* NEED_RESCHED must be visible before we test polling */
1176	smp_mb();
1177	if (!tsk_is_polling(p))
1178	smp_send_reschedule(cpu);
1179	}
1180
1181	static void resched_cpu(int cpu)
1182	{
1183	struct rq *rq = cpu_rq(cpu);
1184	unsigned long flags;
1185
1186	if (!raw_spin_trylock_irqsave(&rq->lock, flags))
1187	return;
1188	resched_task(cpu_curr(cpu));
1189	raw_spin_unlock_irqrestore(&rq->lock, flags);
1190	}
1191
1192	#ifdef CONFIG_NO_HZ
1193	/*
1194	* When add_timer_on() enqueues a timer into the timer wheel of an
1195	* idle CPU then this timer might expire before the next timer event
1196	* which is scheduled to wake up that CPU. In case of a completely
1197	* idle system the next event might even be infinite time into the
1198	* future. wake_up_idle_cpu() ensures that the CPU is woken up and
1199	* leaves the inner idle loop so the newly added timer is taken into
1200	* account when the CPU goes back to idle and evaluates the timer
1201	* wheel for the next timer event.
1202	*/
1203	void wake_up_idle_cpu(int cpu)
1204	{
1205	struct rq *rq = cpu_rq(cpu);
1206
1207	if (cpu == smp_processor_id())
1208	return;
1209
1210	/*
1211	* This is safe, as this function is called with the timer
1212	* wheel base lock of (cpu) held. When the CPU is on the way
1213	* to idle and has not yet set rq->curr to idle then it will
1214	* be serialized on the timer wheel base lock and take the new
1215	* timer into account automatically.
1216	*/
1217	if (rq->curr != rq->idle)
1218	return;
1219
1220	/*
1221	* We can set TIF_RESCHED on the idle task of the other CPU
1222	* lockless. The worst case is that the other CPU runs the
1223	* idle task through an additional NOOP schedule()
1224	*/
1225	set_tsk_need_resched(rq->idle);
1226
1227	/* NEED_RESCHED must be visible before we test polling */
1228	smp_mb();
1229	if (!tsk_is_polling(rq->idle))
1230	smp_send_reschedule(cpu);
1231	}
1232	#endif /* CONFIG_NO_HZ */
1233
1234	static u64 sched_avg_period(void)
1235	{
1236	return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2;
1237	}
1238
1239	static void sched_avg_update(struct rq *rq)
1240	{
1241	s64 period = sched_avg_period();
1242
1243	while ((s64)(rq->clock - rq->age_stamp) > period) {
1244	rq->age_stamp += period;
1245	rq->rt_avg /= 2;
1246	}
1247	}
1248
1249	static void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1250	{
1251	rq->rt_avg += rt_delta;
1252	sched_avg_update(rq);
1253	}
1254
1255	#else /* !CONFIG_SMP */
1256	static void resched_task(struct task_struct *p)
1257	{
1258	assert_raw_spin_locked(&task_rq(p)->lock);
1259	set_tsk_need_resched(p);
1260	}
1261
1262	static void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1263	{
1264	}
1265	#endif /* CONFIG_SMP */
1266
1267	#if BITS_PER_LONG == 32
1268	# define WMULT_CONST (~0UL)
1269	#else
1270	# define WMULT_CONST (1UL << 32)
1271	#endif
1272
1273	#define WMULT_SHIFT 32
1274
1275	/*
1276	* Shift right and round:
1277	*/
1278	#define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y))
1279
1280	/*
1281	* delta *= weight / lw
1282	*/
1283	static unsigned long
1284	calc_delta_mine(unsigned long delta_exec, unsigned long weight,
1285	struct load_weight *lw)
1286	{
1287	u64 tmp;
1288
1289	if (!lw->inv_weight) {
1290	if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST))
1291	lw->inv_weight = 1;
1292	else
1293	lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2)
1294	/ (lw->weight+1);
1295	}
1296
1297	tmp = (u64)delta_exec * weight;
1298	/*
1299	* Check whether we'd overflow the 64-bit multiplication:
1300	*/
1301	if (unlikely(tmp > WMULT_CONST))
1302	tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight,
1303	WMULT_SHIFT/2);
1304	else
1305	tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT);
1306
1307	return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX);
1308	}
1309
1310	static inline void update_load_add(struct load_weight *lw, unsigned long inc)
1311	{
1312	lw->weight += inc;
1313	lw->inv_weight = 0;
1314	}
1315
1316	static inline void update_load_sub(struct load_weight *lw, unsigned long dec)
1317	{
1318	lw->weight -= dec;
1319	lw->inv_weight = 0;
1320	}
1321
1322	/*
1323	* To aid in avoiding the subversion of "niceness" due to uneven distribution
1324	* of tasks with abnormal "nice" values across CPUs the contribution that
1325	* each task makes to its run queue's load is weighted according to its
1326	* scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1327	* scaled version of the new time slice allocation that they receive on time
1328	* slice expiry etc.
1329	*/
1330
1331	#define WEIGHT_IDLEPRIO 3
1332	#define WMULT_IDLEPRIO 1431655765
1333
1334	/*
1335	* Nice levels are multiplicative, with a gentle 10% change for every
1336	* nice level changed. I.e. when a CPU-bound task goes from nice 0 to
1337	* nice 1, it will get ~10% less CPU time than another CPU-bound task
1338	* that remained on nice 0.
1339	*
1340	* The "10% effect" is relative and cumulative: from _any_ nice level,
1341	* if you go up 1 level, it's -10% CPU usage, if you go down 1 level
1342	* it's +10% CPU usage. (to achieve that we use a multiplier of 1.25.
1343	* If a task goes up by ~10% and another task goes down by ~10% then
1344	* the relative distance between them is ~25%.)
1345	*/
1346	static const int prio_to_weight[40] = {
1347	/* -20 */ 88761, 71755, 56483, 46273, 36291,
1348	/* -15 */ 29154, 23254, 18705, 14949, 11916,
1349	/* -10 */ 9548, 7620, 6100, 4904, 3906,
1350	/* -5 */ 3121, 2501, 1991, 1586, 1277,
1351	/* 0 */ 1024, 820, 655, 526, 423,
1352	/* 5 */ 335, 272, 215, 172, 137,
1353	/* 10 */ 110, 87, 70, 56, 45,
1354	/* 15 */ 36, 29, 23, 18, 15,
1355	};
1356
1357	/*
1358	* Inverse (2^32/x) values of the prio_to_weight[] array, precalculated.
1359	*
1360	* In cases where the weight does not change often, we can use the
1361	* precalculated inverse to speed up arithmetics by turning divisions
1362	* into multiplications:
1363	*/
1364	static const u32 prio_to_wmult[40] = {
1365	/* -20 */ 48388, 59856, 76040, 92818, 118348,
1366	/* -15 */ 147320, 184698, 229616, 287308, 360437,
1367	/* -10 */ 449829, 563644, 704093, 875809, 1099582,
1368	/* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326,
1369	/* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587,
1370	/* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126,
1371	/* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717,
1372	/* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153,
1373	};
1374
1375	/* Time spent by the tasks of the cpu accounting group executing in ... */
1376	enum cpuacct_stat_index {
1377	CPUACCT_STAT_USER, /* ... user mode */
1378	CPUACCT_STAT_SYSTEM, /* ... kernel mode */
1379
1380	CPUACCT_STAT_NSTATS,
1381	};
1382
1383	#ifdef CONFIG_CGROUP_CPUACCT
1384	static void cpuacct_charge(struct task_struct *tsk, u64 cputime);
1385	static void cpuacct_update_stats(struct task_struct *tsk,
1386	enum cpuacct_stat_index idx, cputime_t val);
1387	#else
1388	static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {}
1389	static inline void cpuacct_update_stats(struct task_struct *tsk,
1390	enum cpuacct_stat_index idx, cputime_t val) {}
1391	#endif
1392
1393	static inline void inc_cpu_load(struct rq *rq, unsigned long load)
1394	{
1395	update_load_add(&rq->load, load);
1396	}
1397
1398	static inline void dec_cpu_load(struct rq *rq, unsigned long load)
1399	{
1400	update_load_sub(&rq->load, load);
1401	}
1402
1403	#if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) \|\| defined(CONFIG_RT_GROUP_SCHED)
1404	typedef int (tg_visitor)(struct task_group , void *);
1405
1406	/*
1407	* Iterate the full tree, calling @down when first entering a node and @up when
1408	* leaving it for the final time.
1409	*/
1410	static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
1411	{
1412	struct task_group parent, child;
1413	int ret;
1414
1415	rcu_read_lock();
1416	parent = &root_task_group;
1417	down:
1418	ret = (*down)(parent, data);
1419	if (ret)
1420	goto out_unlock;
1421	list_for_each_entry_rcu(child, &parent->children, siblings) {
1422	parent = child;
1423	goto down;
1424
1425	up:
1426	continue;
1427	}
1428	ret = (*up)(parent, data);
1429	if (ret)
1430	goto out_unlock;
1431
1432	child = parent;
1433	parent = parent->parent;
1434	if (parent)
1435	goto up;
1436	out_unlock:
1437	rcu_read_unlock();
1438
1439	return ret;
1440	}
1441
1442	static int tg_nop(struct task_group tg, void data)
1443	{
1444	return 0;
1445	}
1446	#endif
1447
1448	#ifdef CONFIG_SMP
1449	/* Used instead of source_load when we know the type == 0 */
1450	static unsigned long weighted_cpuload(const int cpu)
1451	{
1452	return cpu_rq(cpu)->load.weight;
1453	}
1454
1455	/*
1456	* Return a low guess at the load of a migration-source cpu weighted
1457	* according to the scheduling class and "nice" value.
1458	*
1459	* We want to under-estimate the load of migration sources, to
1460	* balance conservatively.
1461	*/
1462	static unsigned long source_load(int cpu, int type)
1463	{
1464	struct rq *rq = cpu_rq(cpu);
1465	unsigned long total = weighted_cpuload(cpu);
1466
1467	if (type == 0 \|\| !sched_feat(LB_BIAS))
1468	return total;
1469
1470	return min(rq->cpu_load[type-1], total);
1471	}
1472
1473	/*
1474	* Return a high guess at the load of a migration-target cpu weighted
1475	* according to the scheduling class and "nice" value.
1476	*/
1477	static unsigned long target_load(int cpu, int type)
1478	{
1479	struct rq *rq = cpu_rq(cpu);
1480	unsigned long total = weighted_cpuload(cpu);
1481
1482	if (type == 0 \|\| !sched_feat(LB_BIAS))
1483	return total;
1484
1485	return max(rq->cpu_load[type-1], total);
1486	}
1487
1488	static struct sched_group *group_of(int cpu)
1489	{
1490	struct sched_domain *sd = rcu_dereference_sched(cpu_rq(cpu)->sd);
1491
1492	if (!sd)
1493	return NULL;
1494
1495	return sd->groups;
1496	}
1497
1498	static unsigned long power_of(int cpu)
1499	{
1500	struct sched_group *group = group_of(cpu);
1501
1502	if (!group)
1503	return SCHED_LOAD_SCALE;
1504
1505	return group->cpu_power;
1506	}
1507
1508	static int task_hot(struct task_struct p, u64 now, struct sched_domain sd);
1509
1510	static unsigned long cpu_avg_load_per_task(int cpu)
1511	{
1512	struct rq *rq = cpu_rq(cpu);
1513	unsigned long nr_running = ACCESS_ONCE(rq->nr_running);
1514
1515	if (nr_running)
1516	rq->avg_load_per_task = rq->load.weight / nr_running;
1517	else
1518	rq->avg_load_per_task = 0;
1519
1520	return rq->avg_load_per_task;
1521	}
1522
1523	#ifdef CONFIG_FAIR_GROUP_SCHED
1524
1525	static __read_mostly unsigned long __percpu *update_shares_data;
1526
1527	static void __set_se_shares(struct sched_entity *se, unsigned long shares);
1528
1529	/*
1530	* Calculate and set the cpu's group shares.
1531	*/
1532	static void update_group_shares_cpu(struct task_group *tg, int cpu,
1533	unsigned long sd_shares,
1534	unsigned long sd_rq_weight,
1535	unsigned long *usd_rq_weight)
1536	{
1537	unsigned long shares, rq_weight;
1538	int boost = 0;
1539
1540	rq_weight = usd_rq_weight[cpu];
1541	if (!rq_weight) {
1542	boost = 1;
1543	rq_weight = NICE_0_LOAD;
1544	}
1545
1546	/*
1547	* \Sum_j shares_j * rq_weight_i
1548	* shares_i = -----------------------------
1549	* \Sum_j rq_weight_j
1550	*/
1551	shares = (sd_shares * rq_weight) / sd_rq_weight;
1552	shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES);
1553
1554	if (abs(shares - tg->se[cpu]->load.weight) >
1555	sysctl_sched_shares_thresh) {
1556	struct rq *rq = cpu_rq(cpu);
1557	unsigned long flags;
1558
1559	raw_spin_lock_irqsave(&rq->lock, flags);
1560	tg->cfs_rq[cpu]->rq_weight = boost ? 0 : rq_weight;
1561	tg->cfs_rq[cpu]->shares = boost ? 0 : shares;
1562	__set_se_shares(tg->se[cpu], shares);
1563	raw_spin_unlock_irqrestore(&rq->lock, flags);
1564	}
1565	}
1566
1567	/*
1568	* Re-compute the task group their per cpu shares over the given domain.
1569	* This needs to be done in a bottom-up fashion because the rq weight of a
1570	* parent group depends on the shares of its child groups.
1571	*/
1572	static int tg_shares_up(struct task_group tg, void data)
1573	{
1574	unsigned long weight, rq_weight = 0, sum_weight = 0, shares = 0;
1575	unsigned long *usd_rq_weight;
1576	struct sched_domain *sd = data;
1577	unsigned long flags;
1578	int i;
1579
1580	if (!tg->se[0])
1581	return 0;
1582
1583	local_irq_save(flags);
1584	usd_rq_weight = per_cpu_ptr(update_shares_data, smp_processor_id());
1585
1586	for_each_cpu(i, sched_domain_span(sd)) {
1587	weight = tg->cfs_rq[i]->load.weight;
1588	usd_rq_weight[i] = weight;
1589
1590	rq_weight += weight;
1591	/*
1592	* If there are currently no tasks on the cpu pretend there
1593	* is one of average load so that when a new task gets to
1594	* run here it will not get delayed by group starvation.
1595	*/
1596	if (!weight)
1597	weight = NICE_0_LOAD;
1598
1599	sum_weight += weight;
1600	shares += tg->cfs_rq[i]->shares;
1601	}
1602
1603	if (!rq_weight)
1604	rq_weight = sum_weight;
1605
1606	if ((!shares && rq_weight) \|\| shares > tg->shares)
1607	shares = tg->shares;
1608
1609	if (!sd->parent \|\| !(sd->parent->flags & SD_LOAD_BALANCE))
1610	shares = tg->shares;
1611
1612	for_each_cpu(i, sched_domain_span(sd))
1613	update_group_shares_cpu(tg, i, shares, rq_weight, usd_rq_weight);
1614
1615	local_irq_restore(flags);
1616
1617	return 0;
1618	}
1619
1620	/*
1621	* Compute the cpu's hierarchical load factor for each task group.
1622	* This needs to be done in a top-down fashion because the load of a child
1623	* group is a fraction of its parents load.
1624	*/
1625	static int tg_load_down(struct task_group tg, void data)
1626	{
1627	unsigned long load;
1628	long cpu = (long)data;
1629
1630	if (!tg->parent) {
1631	load = cpu_rq(cpu)->load.weight;
1632	} else {
1633	load = tg->parent->cfs_rq[cpu]->h_load;
1634	load *= tg->cfs_rq[cpu]->shares;
1635	load /= tg->parent->cfs_rq[cpu]->load.weight + 1;
1636	}
1637
1638	tg->cfs_rq[cpu]->h_load = load;
1639
1640	return 0;
1641	}
1642
1643	static void update_shares(struct sched_domain *sd)
1644	{
1645	s64 elapsed;
1646	u64 now;
1647
1648	if (root_task_group_empty())
1649	return;
1650
1651	now = cpu_clock(raw_smp_processor_id());
1652	elapsed = now - sd->last_update;
1653
1654	if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) {
1655	sd->last_update = now;
1656	walk_tg_tree(tg_nop, tg_shares_up, sd);
1657	}
1658	}
1659
1660	static void update_h_load(long cpu)
1661	{
1662	if (root_task_group_empty())
1663	return;
1664
1665	walk_tg_tree(tg_load_down, tg_nop, (void *)cpu);
1666	}
1667
1668	#else
1669
1670	static inline void update_shares(struct sched_domain *sd)
1671	{
1672	}
1673
1674	#endif
1675
1676	#ifdef CONFIG_PREEMPT
1677
1678	static void double_rq_lock(struct rq rq1, struct rq rq2);
1679
1680	/*
1681	* fair double_lock_balance: Safely acquires both rq->locks in a fair
1682	* way at the expense of forcing extra atomic operations in all
1683	* invocations. This assures that the double_lock is acquired using the
1684	* same underlying policy as the spinlock_t on this architecture, which
1685	* reduces latency compared to the unfair variant below. However, it
1686	* also adds more overhead and therefore may reduce throughput.
1687	*/
1688	static inline int _double_lock_balance(struct rq this_rq, struct rq busiest)
1689	__releases(this_rq->lock)
1690	__acquires(busiest->lock)
1691	__acquires(this_rq->lock)
1692	{
1693	raw_spin_unlock(&this_rq->lock);
1694	double_rq_lock(this_rq, busiest);
1695
1696	return 1;
1697	}
1698
1699	#else
1700	/*
1701	* Unfair double_lock_balance: Optimizes throughput at the expense of
1702	* latency by eliminating extra atomic operations when the locks are
1703	* already in proper order on entry. This favors lower cpu-ids and will
1704	* grant the double lock to lower cpus over higher ids under contention,
1705	* regardless of entry order into the function.
1706	*/
1707	static int _double_lock_balance(struct rq this_rq, struct rq busiest)
1708	__releases(this_rq->lock)
1709	__acquires(busiest->lock)
1710	__acquires(this_rq->lock)
1711	{
1712	int ret = 0;
1713
1714	if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1715	if (busiest < this_rq) {
1716	raw_spin_unlock(&this_rq->lock);
1717	raw_spin_lock(&busiest->lock);
1718	raw_spin_lock_nested(&this_rq->lock,
1719	SINGLE_DEPTH_NESTING);
1720	ret = 1;
1721	} else
1722	raw_spin_lock_nested(&busiest->lock,
1723	SINGLE_DEPTH_NESTING);
1724	}
1725	return ret;
1726	}
1727
1728	#endif /* CONFIG_PREEMPT */
1729
1730	/*
1731	* double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1732	*/
1733	static int double_lock_balance(struct rq this_rq, struct rq busiest)
1734	{
1735	if (unlikely(!irqs_disabled())) {
1736	/* printk() doesn't work good under rq->lock */
1737	raw_spin_unlock(&this_rq->lock);
1738	BUG_ON(1);
1739	}
1740
1741	return _double_lock_balance(this_rq, busiest);
1742	}
1743
1744	static inline void double_unlock_balance(struct rq this_rq, struct rq busiest)
1745	__releases(busiest->lock)
1746	{
1747	raw_spin_unlock(&busiest->lock);
1748	lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1749	}
1750
1751	/*
1752	* double_rq_lock - safely lock two runqueues
1753	*
1754	* Note this does not disable interrupts like task_rq_lock,
1755	* you need to do so manually before calling.
1756	*/
1757	static void double_rq_lock(struct rq rq1, struct rq rq2)
1758	__acquires(rq1->lock)
1759	__acquires(rq2->lock)
1760	{
1761	BUG_ON(!irqs_disabled());
1762	if (rq1 == rq2) {
1763	raw_spin_lock(&rq1->lock);
1764	__acquire(rq2->lock); /* Fake it out ;) */
1765	} else {
1766	if (rq1 < rq2) {
1767	raw_spin_lock(&rq1->lock);
1768	raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1769	} else {
1770	raw_spin_lock(&rq2->lock);
1771	raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1772	}
1773	}
1774	update_rq_clock(rq1);
1775	update_rq_clock(rq2);
1776	}
1777
1778	/*
1779	* double_rq_unlock - safely unlock two runqueues
1780	*
1781	* Note this does not restore interrupts like task_rq_unlock,
1782	* you need to do so manually after calling.
1783	*/
1784	static void double_rq_unlock(struct rq rq1, struct rq rq2)
1785	__releases(rq1->lock)
1786	__releases(rq2->lock)
1787	{
1788	raw_spin_unlock(&rq1->lock);
1789	if (rq1 != rq2)
1790	raw_spin_unlock(&rq2->lock);
1791	else
1792	__release(rq2->lock);
1793	}
1794
1795	#endif
1796
1797	#ifdef CONFIG_FAIR_GROUP_SCHED
1798	static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares)
1799	{
1800	#ifdef CONFIG_SMP
1801	cfs_rq->shares = shares;
1802	#endif
1803	}
1804	#endif
1805
1806	static void calc_load_account_active(struct rq *this_rq);
1807	static void update_sysctl(void);
1808	static int get_update_sysctl_factor(void);
1809
1810	static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1811	{
1812	set_task_rq(p, cpu);
1813	#ifdef CONFIG_SMP
1814	/*
1815	* After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1816	* successfuly executed on another CPU. We must ensure that updates of
1817	* per-task data have been completed by this moment.
1818	*/
1819	smp_wmb();
1820	task_thread_info(p)->cpu = cpu;
1821	#endif
1822	}
1823
1824	static const struct sched_class rt_sched_class;
1825
1826	#define sched_class_highest (&rt_sched_class)
1827	#define for_each_class(class) \
1828	for (class = sched_class_highest; class; class = class->next)
1829
1830	#include "sched_stats.h"
1831
1832	static void inc_nr_running(struct rq *rq)
1833	{
1834	rq->nr_running++;
1835	}
1836
1837	static void dec_nr_running(struct rq *rq)
1838	{
1839	rq->nr_running--;
1840	}
1841
1842	static void set_load_weight(struct task_struct *p)
1843	{
1844	if (task_has_rt_policy(p)) {
1845	p->se.load.weight = prio_to_weight[0] * 2;
1846	p->se.load.inv_weight = prio_to_wmult[0] >> 1;
1847	return;
1848	}
1849
1850	/*
1851	* SCHED_IDLE tasks get minimal weight:
1852	*/
1853	if (p->policy == SCHED_IDLE) {
1854	p->se.load.weight = WEIGHT_IDLEPRIO;
1855	p->se.load.inv_weight = WMULT_IDLEPRIO;
1856	return;
1857	}
1858
1859	p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO];
1860	p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO];
1861	}
1862
1863	static void update_avg(u64 *avg, u64 sample)
1864	{
1865	s64 diff = sample - *avg;
1866	*avg += diff >> 3;
1867	}
1868
1869	static void
1870	enqueue_task(struct rq rq, struct task_struct p, int wakeup, bool head)
1871	{
1872	if (wakeup)
1873	p->se.start_runtime = p->se.sum_exec_runtime;
1874
1875	sched_info_queued(p);
1876	p->sched_class->enqueue_task(rq, p, wakeup, head);
1877	p->se.on_rq = 1;
1878	}
1879
1880	static void dequeue_task(struct rq rq, struct task_struct p, int sleep)
1881	{
1882	if (sleep) {
1883	if (p->se.last_wakeup) {
1884	update_avg(&p->se.avg_overlap,
1885	p->se.sum_exec_runtime - p->se.last_wakeup);
1886	p->se.last_wakeup = 0;
1887	} else {
1888	update_avg(&p->se.avg_wakeup,
1889	sysctl_sched_wakeup_granularity);
1890	}
1891	}
1892
1893	sched_info_dequeued(p);
1894	p->sched_class->dequeue_task(rq, p, sleep);
1895	p->se.on_rq = 0;
1896	}
1897
1898	/*
1899	* activate_task - move a task to the runqueue.
1900	*/
1901	static void activate_task(struct rq rq, struct task_struct p, int wakeup)
1902	{
1903	if (task_contributes_to_load(p))
1904	rq->nr_uninterruptible--;
1905
1906	enqueue_task(rq, p, wakeup, false);
1907	inc_nr_running(rq);
1908	}
1909
1910	/*
1911	* deactivate_task - remove a task from the runqueue.
1912	*/
1913	static void deactivate_task(struct rq rq, struct task_struct p, int sleep)
1914	{
1915	if (task_contributes_to_load(p))
1916	rq->nr_uninterruptible++;
1917
1918	dequeue_task(rq, p, sleep);
1919	dec_nr_running(rq);
1920	}
1921
1922	#include "sched_idletask.c"
1923	#include "sched_fair.c"
1924	#include "sched_rt.c"
1925	#ifdef CONFIG_SCHED_DEBUG
1926	# include "sched_debug.c"
1927	#endif
1928
1929	/*
1930	* __normal_prio - return the priority that is based on the static prio
1931	*/
1932	static inline int __normal_prio(struct task_struct *p)
1933	{
1934	return p->static_prio;
1935	}
1936
1937	/*
1938	* Calculate the expected normal priority: i.e. priority
1939	* without taking RT-inheritance into account. Might be
1940	* boosted by interactivity modifiers. Changes upon fork,
1941	* setprio syscalls, and whenever the interactivity
1942	* estimator recalculates.
1943	*/
1944	static inline int normal_prio(struct task_struct *p)
1945	{
1946	int prio;
1947
1948	if (task_has_rt_policy(p))
1949	prio = MAX_RT_PRIO-1 - p->rt_priority;
1950	else
1951	prio = __normal_prio(p);
1952	return prio;
1953	}
1954
1955	/*
1956	* Calculate the current priority, i.e. the priority
1957	* taken into account by the scheduler. This value might
1958	* be boosted by RT tasks, or might be boosted by
1959	* interactivity modifiers. Will be RT if the task got
1960	* RT-boosted. If not then it returns p->normal_prio.
1961	*/
1962	static int effective_prio(struct task_struct *p)
1963	{
1964	p->normal_prio = normal_prio(p);
1965	/*
1966	* If we are RT tasks or we were boosted to RT priority,
1967	* keep the priority unchanged. Otherwise, update priority
1968	* to the normal priority:
1969	*/
1970	if (!rt_prio(p->prio))
1971	return p->normal_prio;
1972	return p->prio;
1973	}
1974
1975	/**
1976	* task_curr - is this task currently executing on a CPU?
1977	* @p: the task in question.
1978	*/
1979	inline int task_curr(const struct task_struct *p)
1980	{
1981	return cpu_curr(task_cpu(p)) == p;
1982	}
1983
1984	static inline void check_class_changed(struct rq rq, struct task_struct p,
1985	const struct sched_class *prev_class,
1986	int oldprio, int running)
1987	{
1988	if (prev_class != p->sched_class) {
1989	if (prev_class->switched_from)
1990	prev_class->switched_from(rq, p, running);
1991	p->sched_class->switched_to(rq, p, running);
1992	} else
1993	p->sched_class->prio_changed(rq, p, oldprio, running);
1994	}
1995
1996	#ifdef CONFIG_SMP
1997	/*
1998	* Is this task likely cache-hot:
1999	*/
2000	static int
2001	task_hot(struct task_struct p, u64 now, struct sched_domain sd)
2002	{
2003	s64 delta;
2004
2005	if (p->sched_class != &fair_sched_class)
2006	return 0;
2007
2008	/*
2009	* Buddy candidates are cache hot:
2010	*/
2011	if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running &&
2012	(&p->se == cfs_rq_of(&p->se)->next \|\|
2013	&p->se == cfs_rq_of(&p->se)->last))
2014	return 1;
2015
2016	if (sysctl_sched_migration_cost == -1)
2017	return 1;
2018	if (sysctl_sched_migration_cost == 0)
2019	return 0;
2020
2021	delta = now - p->se.exec_start;
2022
2023	return delta < (s64)sysctl_sched_migration_cost;
2024	}
2025
2026	void set_task_cpu(struct task_struct *p, unsigned int new_cpu)
2027	{
2028	#ifdef CONFIG_SCHED_DEBUG
2029	/*
2030	* We should never call set_task_cpu() on a blocked task,
2031	* ttwu() will sort out the placement.
2032	*/
2033	WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING &&
2034	!(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE));
2035	#endif
2036
2037	trace_sched_migrate_task(p, new_cpu);
2038
2039	if (task_cpu(p) != new_cpu) {
2040	p->se.nr_migrations++;
2041	perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 1, NULL, 0);
2042	}
2043
2044	__set_task_cpu(p, new_cpu);
2045	}
2046
2047	struct migration_req {
2048	struct list_head list;
2049
2050	struct task_struct *task;
2051	int dest_cpu;
2052
2053	struct completion done;
2054	};
2055
2056	/*
2057	* The task's runqueue lock must be held.
2058	* Returns true if you have to wait for migration thread.
2059	*/
2060	static int
2061	migrate_task(struct task_struct p, int dest_cpu, struct migration_req req)
2062	{
2063	struct rq *rq = task_rq(p);
2064
2065	/*
2066	* If the task is not on a runqueue (and not running), then
2067	* the next wake-up will properly place the task.
2068	*/
2069	if (!p->se.on_rq && !task_running(rq, p))
2070	return 0;
2071
2072	init_completion(&req->done);
2073	req->task = p;
2074	req->dest_cpu = dest_cpu;
2075	list_add(&req->list, &rq->migration_queue);
2076
2077	return 1;
2078	}
2079
2080	/*
2081	* wait_task_context_switch - wait for a thread to complete at least one
2082	* context switch.
2083	*
2084	* @p must not be current.
2085	*/
2086	void wait_task_context_switch(struct task_struct *p)
2087	{
2088	unsigned long nvcsw, nivcsw, flags;
2089	int running;
2090	struct rq *rq;
2091
2092	nvcsw = p->nvcsw;
2093	nivcsw = p->nivcsw;
2094	for (;;) {
2095	/*
2096	* The runqueue is assigned before the actual context
2097	* switch. We need to take the runqueue lock.
2098	*
2099	* We could check initially without the lock but it is
2100	* very likely that we need to take the lock in every
2101	* iteration.
2102	*/
2103	rq = task_rq_lock(p, &flags);
2104	running = task_running(rq, p);
2105	task_rq_unlock(rq, &flags);
2106
2107	if (likely(!running))
2108	break;
2109	/*
2110	* The switch count is incremented before the actual
2111	* context switch. We thus wait for two switches to be
2112	* sure at least one completed.
2113	*/
2114	if ((p->nvcsw - nvcsw) > 1)
2115	break;
2116	if ((p->nivcsw - nivcsw) > 1)
2117	break;
2118
2119	cpu_relax();
2120	}
2121	}
2122
2123	/*
2124	* wait_task_inactive - wait for a thread to unschedule.
2125	*
2126	* If @match_state is nonzero, it's the @p->state value just checked and
2127	* not expected to change. If it changes, i.e. @p might have woken up,
2128	* then return zero. When we succeed in waiting for @p to be off its CPU,
2129	* we return a positive number (its total switch count). If a second call
2130	* a short while later returns the same number, the caller can be sure that
2131	* @p has remained unscheduled the whole time.
2132	*
2133	* The caller must ensure that the task will unschedule sometime soon,
2134	* else this function might spin for a long time. This function can't
2135	* be called with interrupts off, or it may introduce deadlock with
2136	* smp_call_function() if an IPI is sent by the same process we are
2137	* waiting to become inactive.
2138	*/
2139	unsigned long wait_task_inactive(struct task_struct *p, long match_state)
2140	{
2141	unsigned long flags;
2142	int running, on_rq;
2143	unsigned long ncsw;
2144	struct rq *rq;
2145
2146	for (;;) {
2147	/*
2148	* We do the initial early heuristics without holding
2149	* any task-queue locks at all. We'll only try to get
2150	* the runqueue lock when things look like they will
2151	* work out!
2152	*/
2153	rq = task_rq(p);
2154
2155	/*
2156	* If the task is actively running on another CPU
2157	* still, just relax and busy-wait without holding
2158	* any locks.
2159	*
2160	* NOTE! Since we don't hold any locks, it's not
2161	* even sure that "rq" stays as the right runqueue!
2162	* But we don't care, since "task_running()" will
2163	* return false if the runqueue has changed and p
2164	* is actually now running somewhere else!
2165	*/
2166	while (task_running(rq, p)) {
2167	if (match_state && unlikely(p->state != match_state))
2168	return 0;
2169	cpu_relax();
2170	}
2171
2172	/*
2173	* Ok, time to look more closely! We need the rq
2174	* lock now, to be sure. If we're wrong, we'll
2175	* just go back and repeat.
2176	*/
2177	rq = task_rq_lock(p, &flags);
2178	trace_sched_wait_task(rq, p);
2179	running = task_running(rq, p);
2180	on_rq = p->se.on_rq;
2181	ncsw = 0;
2182	if (!match_state \|\| p->state == match_state)
2183	ncsw = p->nvcsw \| LONG_MIN; /* sets MSB */
2184	task_rq_unlock(rq, &flags);
2185
2186	/*
2187	* If it changed from the expected state, bail out now.
2188	*/
2189	if (unlikely(!ncsw))
2190	break;
2191
2192	/*
2193	* Was it really running after all now that we
2194	* checked with the proper locks actually held?
2195	*
2196	* Oops. Go back and try again..
2197	*/
2198	if (unlikely(running)) {
2199	cpu_relax();
2200	continue;
2201	}
2202
2203	/*
2204	* It's not enough that it's not actively running,
2205	* it must be off the runqueue _entirely_, and not
2206	* preempted!
2207	*
2208	* So if it was still runnable (but just not actively
2209	* running right now), it's preempted, and we should
2210	* yield - it could be a while.
2211	*/
2212	if (unlikely(on_rq)) {
2213	schedule_timeout_uninterruptible(1);
2214	continue;
2215	}
2216
2217	/*
2218	* Ahh, all good. It wasn't running, and it wasn't
2219	* runnable, which means that it will never become
2220	* running in the future either. We're all done!
2221	*/
2222	break;
2223	}
2224
2225	return ncsw;
2226	}
2227
2228	/***
2229	* kick_process - kick a running thread to enter/exit the kernel
2230	* @p: the to-be-kicked thread
2231	*
2232	* Cause a process which is running on another CPU to enter
2233	* kernel-mode, without any delay. (to get signals handled.)
2234	*
2235	* NOTE: this function doesnt have to take the runqueue lock,
2236	* because all it wants to ensure is that the remote task enters
2237	* the kernel. If the IPI races and the task has been migrated
2238	* to another CPU then no harm is done and the purpose has been
2239	* achieved as well.
2240	*/
2241	void kick_process(struct task_struct *p)
2242	{
2243	int cpu;
2244
2245	preempt_disable();
2246	cpu = task_cpu(p);
2247	if ((cpu != smp_processor_id()) && task_curr(p))
2248	smp_send_reschedule(cpu);
2249	preempt_enable();
2250	}
2251	EXPORT_SYMBOL_GPL(kick_process);
2252	#endif /* CONFIG_SMP */
2253
2254	/**
2255	* task_oncpu_function_call - call a function on the cpu on which a task runs
2256	* @p: the task to evaluate
2257	* @func: the function to be called
2258	* @info: the function call argument
2259	*
2260	* Calls the function @func when the task is currently running. This might
2261	* be on the current CPU, which just calls the function directly
2262	*/
2263	void task_oncpu_function_call(struct task_struct *p,
2264	void (func) (void info), void *info)
2265	{
2266	int cpu;
2267
2268	preempt_disable();
2269	cpu = task_cpu(p);
2270	if (task_curr(p))
2271	smp_call_function_single(cpu, func, info, 1);
2272	preempt_enable();
2273	}
2274
2275	#ifdef CONFIG_SMP
2276	static int select_fallback_rq(int cpu, struct task_struct *p)
2277	{
2278	int dest_cpu;
2279	const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu));
2280
2281	/* Look for allowed, online CPU in same node. */
2282	for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask)
2283	if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed))
2284	return dest_cpu;
2285
2286	/* Any allowed, online CPU? */
2287	dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_active_mask);
2288	if (dest_cpu < nr_cpu_ids)
2289	return dest_cpu;
2290
2291	/* No more Mr. Nice Guy. */
2292	if (dest_cpu >= nr_cpu_ids) {
2293	rcu_read_lock();
2294	cpuset_cpus_allowed_locked(p, &p->cpus_allowed);
2295	rcu_read_unlock();
2296	dest_cpu = cpumask_any_and(cpu_active_mask, &p->cpus_allowed);
2297
2298	/*
2299	* Don't tell them about moving exiting tasks or
2300	* kernel threads (both mm NULL), since they never
2301	* leave kernel.
2302	*/
2303	if (p->mm && printk_ratelimit()) {
2304	printk(KERN_INFO "process %d (%s) no "
2305	"longer affine to cpu%d\n",
2306	task_pid_nr(p), p->comm, cpu);
2307	}
2308	}
2309
2310	return dest_cpu;
2311	}
2312
2313	/*
2314	* Gets called from 3 sites (exec, fork, wakeup), since it is called without
2315	* holding rq->lock we need to ensure ->cpus_allowed is stable, this is done
2316	* by:
2317	*
2318	* exec: is unstable, retry loop
2319	* fork & wake-up: serialize ->cpus_allowed against TASK_WAKING
2320	*/
2321	static inline
2322	int select_task_rq(struct task_struct *p, int sd_flags, int wake_flags)
2323	{
2324	int cpu = p->sched_class->select_task_rq(p, sd_flags, wake_flags);
2325
2326	/*
2327	* In order not to call set_task_cpu() on a blocking task we need
2328	* to rely on ttwu() to place the task on a valid ->cpus_allowed
2329	* cpu.
2330	*
2331	* Since this is common to all placement strategies, this lives here.
2332	*
2333	* [ this allows ->select_task() to simply return task_cpu(p) and
2334	* not worry about this generic constraint ]
2335	*/
2336	if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) \|\|
2337	!cpu_online(cpu)))
2338	cpu = select_fallback_rq(task_cpu(p), p);
2339
2340	return cpu;
2341	}
2342	#endif
2343
2344	/***
2345	* try_to_wake_up - wake up a thread
2346	* @p: the to-be-woken-up thread
2347	* @state: the mask of task states that can be woken
2348	* @sync: do a synchronous wakeup?
2349	*
2350	* Put it on the run-queue if it's not already there. The "current"
2351	* thread is always on the run-queue (except when the actual
2352	* re-schedule is in progress), and as such you're allowed to do
2353	* the simpler "current->state = TASK_RUNNING" to mark yourself
2354	* runnable without the overhead of this.
2355	*
2356	* returns failure only if the task is already active.
2357	*/
2358	static int try_to_wake_up(struct task_struct *p, unsigned int state,
2359	int wake_flags)
2360	{
2361	int cpu, orig_cpu, this_cpu, success = 0;
2362	unsigned long flags;
2363	struct rq *rq;
2364
2365	if (!sched_feat(SYNC_WAKEUPS))
2366	wake_flags &= ~WF_SYNC;
2367
2368	this_cpu = get_cpu();
2369
2370	smp_wmb();
2371	rq = task_rq_lock(p, &flags);
2372	update_rq_clock(rq);
2373	if (!(p->state & state))
2374	goto out;
2375
2376	if (p->se.on_rq)
2377	goto out_running;
2378
2379	cpu = task_cpu(p);
2380	orig_cpu = cpu;
2381
2382	#ifdef CONFIG_SMP
2383	if (unlikely(task_running(rq, p)))
2384	goto out_activate;
2385
2386	/*
2387	* In order to handle concurrent wakeups and release the rq->lock
2388	* we put the task in TASK_WAKING state.
2389	*
2390	* First fix up the nr_uninterruptible count:
2391	*/
2392	if (task_contributes_to_load(p))
2393	rq->nr_uninterruptible--;
2394	p->state = TASK_WAKING;
2395
2396	if (p->sched_class->task_waking)
2397	p->sched_class->task_waking(rq, p);
2398
2399	__task_rq_unlock(rq);
2400
2401	cpu = select_task_rq(p, SD_BALANCE_WAKE, wake_flags);
2402	if (cpu != orig_cpu) {
2403	/*
2404	* Since we migrate the task without holding any rq->lock,
2405	* we need to be careful with task_rq_lock(), since that
2406	* might end up locking an invalid rq.
2407	*/
2408	set_task_cpu(p, cpu);
2409	}
2410
2411	rq = cpu_rq(cpu);
2412	raw_spin_lock(&rq->lock);
2413	update_rq_clock(rq);
2414
2415	/*
2416	* We migrated the task without holding either rq->lock, however
2417	* since the task is not on the task list itself, nobody else
2418	* will try and migrate the task, hence the rq should match the
2419	* cpu we just moved it to.
2420	*/
2421	WARN_ON(task_cpu(p) != cpu);
2422	WARN_ON(p->state != TASK_WAKING);
2423
2424	#ifdef CONFIG_SCHEDSTATS
2425	schedstat_inc(rq, ttwu_count);
2426	if (cpu == this_cpu)
2427	schedstat_inc(rq, ttwu_local);
2428	else {
2429	struct sched_domain *sd;
2430	for_each_domain(this_cpu, sd) {
2431	if (cpumask_test_cpu(cpu, sched_domain_span(sd))) {
2432	schedstat_inc(sd, ttwu_wake_remote);
2433	break;
2434	}
2435	}
2436	}
2437	#endif /* CONFIG_SCHEDSTATS */
2438
2439	out_activate:
2440	#endif /* CONFIG_SMP */
2441	schedstat_inc(p, se.nr_wakeups);
2442	if (wake_flags & WF_SYNC)
2443	schedstat_inc(p, se.nr_wakeups_sync);
2444	if (orig_cpu != cpu)
2445	schedstat_inc(p, se.nr_wakeups_migrate);
2446	if (cpu == this_cpu)
2447	schedstat_inc(p, se.nr_wakeups_local);
2448	else
2449	schedstat_inc(p, se.nr_wakeups_remote);
2450	activate_task(rq, p, 1);
2451	success = 1;
2452
2453	/*
2454	* Only attribute actual wakeups done by this task.
2455	*/
2456	if (!in_interrupt()) {
2457	struct sched_entity *se = &current->se;
2458	u64 sample = se->sum_exec_runtime;
2459
2460	if (se->last_wakeup)
2461	sample -= se->last_wakeup;
2462	else
2463	sample -= se->start_runtime;
2464	update_avg(&se->avg_wakeup, sample);
2465
2466	se->last_wakeup = se->sum_exec_runtime;
2467	}
2468
2469	out_running:
2470	trace_sched_wakeup(rq, p, success);
2471	check_preempt_curr(rq, p, wake_flags);
2472
2473	p->state = TASK_RUNNING;
2474	#ifdef CONFIG_SMP
2475	if (p->sched_class->task_woken)
2476	p->sched_class->task_woken(rq, p);
2477
2478	if (unlikely(rq->idle_stamp)) {
2479	u64 delta = rq->clock - rq->idle_stamp;
2480	u64 max = 2*sysctl_sched_migration_cost;
2481
2482	if (delta > max)
2483	rq->avg_idle = max;
2484	else
2485	update_avg(&rq->avg_idle, delta);
2486	rq->idle_stamp = 0;
2487	}
2488	#endif
2489	out:
2490	task_rq_unlock(rq, &flags);
2491	put_cpu();
2492
2493	return success;
2494	}
2495
2496	/**
2497	* wake_up_process - Wake up a specific process
2498	* @p: The process to be woken up.
2499	*
2500	* Attempt to wake up the nominated process and move it to the set of runnable
2501	* processes. Returns 1 if the process was woken up, 0 if it was already
2502	* running.
2503	*
2504	* It may be assumed that this function implies a write memory barrier before
2505	* changing the task state if and only if any tasks are woken up.
2506	*/
2507	int wake_up_process(struct task_struct *p)
2508	{
2509	return try_to_wake_up(p, TASK_ALL, 0);
2510	}
2511	EXPORT_SYMBOL(wake_up_process);
2512
2513	int wake_up_state(struct task_struct *p, unsigned int state)
2514	{
2515	return try_to_wake_up(p, state, 0);
2516	}
2517
2518	/*
2519	* Perform scheduler related setup for a newly forked process p.
2520	* p is forked by current.
2521	*
2522	* __sched_fork() is basic setup used by init_idle() too:
2523	*/
2524	static void __sched_fork(struct task_struct *p)
2525	{
2526	p->se.exec_start = 0;
2527	p->se.sum_exec_runtime = 0;
2528	p->se.prev_sum_exec_runtime = 0;
2529	p->se.nr_migrations = 0;
2530	p->se.last_wakeup = 0;
2531	p->se.avg_overlap = 0;
2532	p->se.start_runtime = 0;
2533	p->se.avg_wakeup = sysctl_sched_wakeup_granularity;
2534
2535	#ifdef CONFIG_SCHEDSTATS
2536	p->se.wait_start = 0;
2537	p->se.wait_max = 0;
2538	p->se.wait_count = 0;
2539	p->se.wait_sum = 0;
2540
2541	p->se.sleep_start = 0;
2542	p->se.sleep_max = 0;
2543	p->se.sum_sleep_runtime = 0;
2544
2545	p->se.block_start = 0;
2546	p->se.block_max = 0;
2547	p->se.exec_max = 0;
2548	p->se.slice_max = 0;
2549
2550	p->se.nr_migrations_cold = 0;
2551	p->se.nr_failed_migrations_affine = 0;
2552	p->se.nr_failed_migrations_running = 0;
2553	p->se.nr_failed_migrations_hot = 0;
2554	p->se.nr_forced_migrations = 0;
2555
2556	p->se.nr_wakeups = 0;
2557	p->se.nr_wakeups_sync = 0;
2558	p->se.nr_wakeups_migrate = 0;
2559	p->se.nr_wakeups_local = 0;
2560	p->se.nr_wakeups_remote = 0;
2561	p->se.nr_wakeups_affine = 0;
2562	p->se.nr_wakeups_affine_attempts = 0;
2563	p->se.nr_wakeups_passive = 0;
2564	p->se.nr_wakeups_idle = 0;
2565
2566	#endif
2567
2568	INIT_LIST_HEAD(&p->rt.run_list);
2569	p->se.on_rq = 0;
2570	INIT_LIST_HEAD(&p->se.group_node);
2571
2572	#ifdef CONFIG_PREEMPT_NOTIFIERS
2573	INIT_HLIST_HEAD(&p->preempt_notifiers);
2574	#endif
2575	}
2576
2577	/*
2578	* fork()/clone()-time setup:
2579	*/
2580	void sched_fork(struct task_struct *p, int clone_flags)
2581	{
2582	int cpu = get_cpu();
2583
2584	__sched_fork(p);
2585	/*
2586	* We mark the process as waking here. This guarantees that
2587	* nobody will actually run it, and a signal or other external
2588	* event cannot wake it up and insert it on the runqueue either.
2589	*/
2590	p->state = TASK_WAKING;
2591
2592	/*
2593	* Revert to default priority/policy on fork if requested.
2594	*/
2595	if (unlikely(p->sched_reset_on_fork)) {
2596	if (p->policy == SCHED_FIFO \|\| p->policy == SCHED_RR) {
2597	p->policy = SCHED_NORMAL;
2598	p->normal_prio = p->static_prio;
2599	}
2600
2601	if (PRIO_TO_NICE(p->static_prio) < 0) {
2602	p->static_prio = NICE_TO_PRIO(0);
2603	p->normal_prio = p->static_prio;
2604	set_load_weight(p);
2605	}
2606
2607	/*
2608	* We don't need the reset flag anymore after the fork. It has
2609	* fulfilled its duty:
2610	*/
2611	p->sched_reset_on_fork = 0;
2612	}
2613
2614	/*
2615	* Make sure we do not leak PI boosting priority to the child.
2616	*/
2617	p->prio = current->normal_prio;
2618
2619	if (!rt_prio(p->prio))
2620	p->sched_class = &fair_sched_class;
2621
2622	if (p->sched_class->task_fork)
2623	p->sched_class->task_fork(p);
2624
2625	set_task_cpu(p, cpu);
2626
2627	#if defined(CONFIG_SCHEDSTATS) \|\| defined(CONFIG_TASK_DELAY_ACCT)
2628	if (likely(sched_info_on()))
2629	memset(&p->sched_info, 0, sizeof(p->sched_info));
2630	#endif
2631	#if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW)
2632	p->oncpu = 0;
2633	#endif
2634	#ifdef CONFIG_PREEMPT
2635	/* Want to start with kernel preemption disabled. */
2636	task_thread_info(p)->preempt_count = 1;
2637	#endif
2638	plist_node_init(&p->pushable_tasks, MAX_PRIO);
2639
2640	put_cpu();
2641	}
2642
2643	/*
2644	* wake_up_new_task - wake up a newly created task for the first time.
2645	*
2646	* This function will do some initial scheduler statistics housekeeping
2647	* that must be done for every newly created context, then puts the task
2648	* on the runqueue and wakes it.
2649	*/
2650	void wake_up_new_task(struct task_struct *p, unsigned long clone_flags)
2651	{
2652	unsigned long flags;
2653	struct rq *rq;
2654	int cpu __maybe_unused = get_cpu();
2655
2656	#ifdef CONFIG_SMP
2657	/*
2658	* Fork balancing, do it here and not earlier because:
2659	* - cpus_allowed can change in the fork path
2660	* - any previously selected cpu might disappear through hotplug
2661	*
2662	* We still have TASK_WAKING but PF_STARTING is gone now, meaning
2663	* ->cpus_allowed is stable, we have preemption disabled, meaning
2664	* cpu_online_mask is stable.
2665	*/
2666	cpu = select_task_rq(p, SD_BALANCE_FORK, 0);
2667	set_task_cpu(p, cpu);
2668	#endif
2669
2670	/*
2671	* Since the task is not on the rq and we still have TASK_WAKING set
2672	* nobody else will migrate this task.
2673	*/
2674	rq = cpu_rq(cpu);
2675	raw_spin_lock_irqsave(&rq->lock, flags);
2676
2677	BUG_ON(p->state != TASK_WAKING);
2678	p->state = TASK_RUNNING;
2679	update_rq_clock(rq);
2680	activate_task(rq, p, 0);
2681	trace_sched_wakeup_new(rq, p, 1);
2682	check_preempt_curr(rq, p, WF_FORK);
2683	#ifdef CONFIG_SMP
2684	if (p->sched_class->task_woken)
2685	p->sched_class->task_woken(rq, p);
2686	#endif
2687	task_rq_unlock(rq, &flags);
2688	put_cpu();
2689	}
2690
2691	#ifdef CONFIG_PREEMPT_NOTIFIERS
2692
2693	/**
2694	* preempt_notifier_register - tell me when current is being preempted & rescheduled
2695	* @notifier: notifier struct to register
2696	*/
2697	void preempt_notifier_register(struct preempt_notifier *notifier)
2698	{
2699	hlist_add_head(&notifier->link, &current->preempt_notifiers);
2700	}
2701	EXPORT_SYMBOL_GPL(preempt_notifier_register);
2702
2703	/**
2704	* preempt_notifier_unregister - no longer interested in preemption notifications
2705	* @notifier: notifier struct to unregister
2706	*
2707	* This is safe to call from within a preemption notifier.
2708	*/
2709	void preempt_notifier_unregister(struct preempt_notifier *notifier)
2710	{
2711	hlist_del(&notifier->link);
2712	}
2713	EXPORT_SYMBOL_GPL(preempt_notifier_unregister);
2714
2715	static void fire_sched_in_preempt_notifiers(struct task_struct *curr)
2716	{
2717	struct preempt_notifier *notifier;
2718	struct hlist_node *node;
2719
2720	hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link)
2721	notifier->ops->sched_in(notifier, raw_smp_processor_id());
2722	}
2723
2724	static void
2725	fire_sched_out_preempt_notifiers(struct task_struct *curr,
2726	struct task_struct *next)
2727	{
2728	struct preempt_notifier *notifier;
2729	struct hlist_node *node;
2730
2731	hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link)
2732	notifier->ops->sched_out(notifier, next);
2733	}
2734
2735	#else /* !CONFIG_PREEMPT_NOTIFIERS */
2736
2737	static void fire_sched_in_preempt_notifiers(struct task_struct *curr)
2738	{
2739	}
2740
2741	static void
2742	fire_sched_out_preempt_notifiers(struct task_struct *curr,
2743	struct task_struct *next)
2744	{
2745	}
2746
2747	#endif /* CONFIG_PREEMPT_NOTIFIERS */
2748
2749	/**
2750	* prepare_task_switch - prepare to switch tasks
2751	* @rq: the runqueue preparing to switch
2752	* @prev: the current task that is being switched out
2753	* @next: the task we are going to switch to.
2754	*
2755	* This is called with the rq lock held and interrupts off. It must
2756	* be paired with a subsequent finish_task_switch after the context
2757	* switch.
2758	*
2759	* prepare_task_switch sets up locking and calls architecture specific
2760	* hooks.
2761	*/
2762	static inline void
2763	prepare_task_switch(struct rq rq, struct task_struct prev,
2764	struct task_struct *next)
2765	{
2766	fire_sched_out_preempt_notifiers(prev, next);
2767	prepare_lock_switch(rq, next);
2768	prepare_arch_switch(next);
2769	}
2770
2771	/**
2772	* finish_task_switch - clean up after a task-switch
2773	* @rq: runqueue associated with task-switch
2774	* @prev: the thread we just switched away from.
2775	*
2776	* finish_task_switch must be called after the context switch, paired
2777	* with a prepare_task_switch call before the context switch.
2778	* finish_task_switch will reconcile locking set up by prepare_task_switch,
2779	* and do any other architecture-specific cleanup actions.
2780	*
2781	* Note that we may have delayed dropping an mm in context_switch(). If
2782	* so, we finish that here outside of the runqueue lock. (Doing it
2783	* with the lock held can cause deadlocks; see schedule() for
2784	* details.)
2785	*/
2786	static void finish_task_switch(struct rq rq, struct task_struct prev)
2787	__releases(rq->lock)
2788	{
2789	struct mm_struct *mm = rq->prev_mm;
2790	long prev_state;
2791
2792	rq->prev_mm = NULL;
2793
2794	/*
2795	* A task struct has one reference for the use as "current".
2796	* If a task dies, then it sets TASK_DEAD in tsk->state and calls
2797	* schedule one last time. The schedule call will never return, and
2798	* the scheduled task must drop that reference.
2799	* The test for TASK_DEAD must occur while the runqueue locks are
2800	* still held, otherwise prev could be scheduled on another cpu, die
2801	* there before we look at prev->state, and then the reference would
2802	* be dropped twice.
2803	* Manfred Spraul <manfred@colorfullife.com>
2804	*/
2805	prev_state = prev->state;
2806	finish_arch_switch(prev);
2807	#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
2808	local_irq_disable();
2809	#endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */
2810	perf_event_task_sched_in(current);
2811	#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
2812	local_irq_enable();
2813	#endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */
2814	finish_lock_switch(rq, prev);
2815
2816	fire_sched_in_preempt_notifiers(current);
2817	if (mm)
2818	mmdrop(mm);
2819	if (unlikely(prev_state == TASK_DEAD)) {
2820	/*
2821	* Remove function-return probe instances associated with this
2822	* task and put them back on the free list.
2823	*/
2824	kprobe_flush_task(prev);
2825	put_task_struct(prev);
2826	}
2827	}
2828
2829	#ifdef CONFIG_SMP
2830
2831	/* assumes rq->lock is held */
2832	static inline void pre_schedule(struct rq rq, struct task_struct prev)
2833	{
2834	if (prev->sched_class->pre_schedule)
2835	prev->sched_class->pre_schedule(rq, prev);
2836	}
2837
2838	/* rq->lock is NOT held, but preemption is disabled */
2839	static inline void post_schedule(struct rq *rq)
2840	{
2841	if (rq->post_schedule) {
2842	unsigned long flags;
2843
2844	raw_spin_lock_irqsave(&rq->lock, flags);
2845	if (rq->curr->sched_class->post_schedule)
2846	rq->curr->sched_class->post_schedule(rq);
2847	raw_spin_unlock_irqrestore(&rq->lock, flags);
2848
2849	rq->post_schedule = 0;
2850	}
2851	}
2852
2853	#else
2854
2855	static inline void pre_schedule(struct rq rq, struct task_struct p)
2856	{
2857	}
2858
2859	static inline void post_schedule(struct rq *rq)
2860	{
2861	}
2862
2863	#endif
2864
2865	/**
2866	* schedule_tail - first thing a freshly forked thread must call.
2867	* @prev: the thread we just switched away from.
2868	*/
2869	asmlinkage void schedule_tail(struct task_struct *prev)
2870	__releases(rq->lock)
2871	{
2872	struct rq *rq = this_rq();
2873
2874	finish_task_switch(rq, prev);
2875
2876	/*
2877	* FIXME: do we need to worry about rq being invalidated by the
2878	* task_switch?
2879	*/
2880	post_schedule(rq);
2881
2882	#ifdef __ARCH_WANT_UNLOCKED_CTXSW
2883	/* In this case, finish_task_switch does not reenable preemption */
2884	preempt_enable();
2885	#endif
2886	if (current->set_child_tid)
2887	put_user(task_pid_vnr(current), current->set_child_tid);
2888	}
2889
2890	/*
2891	* context_switch - switch to the new MM and the new
2892	* thread's register state.
2893	*/
2894	static inline void
2895	context_switch(struct rq rq, struct task_struct prev,
2896	struct task_struct *next)
2897	{
2898	struct mm_struct mm, oldmm;
2899
2900	prepare_task_switch(rq, prev, next);
2901	trace_sched_switch(rq, prev, next);
2902	mm = next->mm;
2903	oldmm = prev->active_mm;
2904	/*
2905	* For paravirt, this is coupled with an exit in switch_to to
2906	* combine the page table reload and the switch backend into
2907	* one hypercall.
2908	*/
2909	arch_start_context_switch(prev);
2910
2911	if (likely(!mm)) {
2912	next->active_mm = oldmm;
2913	atomic_inc(&oldmm->mm_count);
2914	enter_lazy_tlb(oldmm, next);
2915	} else
2916	switch_mm(oldmm, mm, next);
2917
2918	if (likely(!prev->mm)) {
2919	prev->active_mm = NULL;
2920	rq->prev_mm = oldmm;
2921	}
2922	/*
2923	* Since the runqueue lock will be released by the next
2924	* task (which is an invalid locking op but in the case
2925	* of the scheduler it's an obvious special-case), so we
2926	* do an early lockdep release here:
2927	*/
2928	#ifndef __ARCH_WANT_UNLOCKED_CTXSW
2929	spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
2930	#endif
2931
2932	/* Here we just switch the register state and the stack. */
2933	switch_to(prev, next, prev);
2934
2935	barrier();
2936	/*
2937	* this_rq must be evaluated again because prev may have moved
2938	* CPUs since it called schedule(), thus the 'rq' on its stack
2939	* frame will be invalid.
2940	*/
2941	finish_task_switch(this_rq(), prev);
2942	}
2943
2944	/*
2945	* nr_running, nr_uninterruptible and nr_context_switches:
2946	*
2947	* externally visible scheduler statistics: current number of runnable
2948	* threads, current number of uninterruptible-sleeping threads, total
2949	* number of context switches performed since bootup.
2950	*/
2951	unsigned long nr_running(void)
2952	{
2953	unsigned long i, sum = 0;
2954
2955	for_each_online_cpu(i)
2956	sum += cpu_rq(i)->nr_running;
2957
2958	return sum;
2959	}
2960
2961	unsigned long nr_uninterruptible(void)
2962	{
2963	unsigned long i, sum = 0;
2964
2965	for_each_possible_cpu(i)
2966	sum += cpu_rq(i)->nr_uninterruptible;
2967
2968	/*
2969	* Since we read the counters lockless, it might be slightly
2970	* inaccurate. Do not allow it to go below zero though:
2971	*/
2972	if (unlikely((long)sum < 0))
2973	sum = 0;
2974
2975	return sum;
2976	}
2977
2978	unsigned long long nr_context_switches(void)
2979	{
2980	int i;
2981	unsigned long long sum = 0;
2982
2983	for_each_possible_cpu(i)
2984	sum += cpu_rq(i)->nr_switches;
2985
2986	return sum;
2987	}
2988
2989	unsigned long nr_iowait(void)
2990	{
2991	unsigned long i, sum = 0;
2992
2993	for_each_possible_cpu(i)
2994	sum += atomic_read(&cpu_rq(i)->nr_iowait);
2995
2996	return sum;
2997	}
2998
2999	unsigned long nr_iowait_cpu(void)
3000	{
3001	struct rq *this = this_rq();
3002	return atomic_read(&this->nr_iowait);
3003	}
3004
3005	unsigned long this_cpu_load(void)
3006	{
3007	struct rq *this = this_rq();
3008	return this->cpu_load[0];
3009	}
3010
3011
3012	/* Variables and functions for calc_load */
3013	static atomic_long_t calc_load_tasks;
3014	static unsigned long calc_load_update;
3015	unsigned long avenrun[3];
3016	EXPORT_SYMBOL(avenrun);
3017
3018	/**
3019	* get_avenrun - get the load average array
3020	* @loads: pointer to dest load array
3021	* @offset: offset to add
3022	* @shift: shift count to shift the result left
3023	*
3024	* These values are estimates at best, so no need for locking.
3025	*/
3026	void get_avenrun(unsigned long *loads, unsigned long offset, int shift)
3027	{
3028	loads[0] = (avenrun[0] + offset) << shift;
3029	loads[1] = (avenrun[1] + offset) << shift;
3030	loads[2] = (avenrun[2] + offset) << shift;
3031	}
3032
3033	static unsigned long
3034	calc_load(unsigned long load, unsigned long exp, unsigned long active)
3035	{
3036	load *= exp;
3037	load += active * (FIXED_1 - exp);
3038	return load >> FSHIFT;
3039	}
3040
3041	/*
3042	* calc_load - update the avenrun load estimates 10 ticks after the
3043	* CPUs have updated calc_load_tasks.
3044	*/
3045	void calc_global_load(void)
3046	{
3047	unsigned long upd = calc_load_update + 10;
3048	long active;
3049
3050	if (time_before(jiffies, upd))
3051	return;
3052
3053	active = atomic_long_read(&calc_load_tasks);
3054	active = active > 0 ? active * FIXED_1 : 0;
3055
3056	avenrun[0] = calc_load(avenrun[0], EXP_1, active);
3057	avenrun[1] = calc_load(avenrun[1], EXP_5, active);
3058	avenrun[2] = calc_load(avenrun[2], EXP_15, active);
3059
3060	calc_load_update += LOAD_FREQ;
3061	}
3062
3063	/*
3064	* Either called from update_cpu_load() or from a cpu going idle
3065	*/
3066	static void calc_load_account_active(struct rq *this_rq)
3067	{
3068	long nr_active, delta;
3069
3070	nr_active = this_rq->nr_running;
3071	nr_active += (long) this_rq->nr_uninterruptible;
3072
3073	if (nr_active != this_rq->calc_load_active) {
3074	delta = nr_active - this_rq->calc_load_active;
3075	this_rq->calc_load_active = nr_active;
3076	atomic_long_add(delta, &calc_load_tasks);
3077	}
3078	}
3079
3080	/*
3081	* Update rq->cpu_load[] statistics. This function is usually called every
3082	* scheduler tick (TICK_NSEC).
3083	*/
3084	static void update_cpu_load(struct rq *this_rq)
3085	{
3086	unsigned long this_load = this_rq->load.weight;
3087	int i, scale;
3088
3089	this_rq->nr_load_updates++;
3090
3091	/* Update our load: */
3092	for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) {
3093	unsigned long old_load, new_load;
3094
3095	/* scale is effectively 1 << i now, and >> i divides by scale */
3096
3097	old_load = this_rq->cpu_load[i];
3098	new_load = this_load;
3099	/*
3100	* Round up the averaging division if load is increasing. This
3101	* prevents us from getting stuck on 9 if the load is 10, for
3102	* example.
3103	*/
3104	if (new_load > old_load)
3105	new_load += scale-1;
3106	this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i;
3107	}
3108
3109	if (time_after_eq(jiffies, this_rq->calc_load_update)) {
3110	this_rq->calc_load_update += LOAD_FREQ;
3111	calc_load_account_active(this_rq);
3112	}
3113	}
3114
3115	#ifdef CONFIG_SMP
3116
3117	/*
3118	* sched_exec - execve() is a valuable balancing opportunity, because at
3119	* this point the task has the smallest effective memory and cache footprint.
3120	*/
3121	void sched_exec(void)
3122	{
3123	struct task_struct *p = current;
3124	struct migration_req req;
3125	int dest_cpu, this_cpu;
3126	unsigned long flags;
3127	struct rq *rq;
3128
3129	again:
3130	this_cpu = get_cpu();
3131	dest_cpu = select_task_rq(p, SD_BALANCE_EXEC, 0);
3132	if (dest_cpu == this_cpu) {
3133	put_cpu();
3134	return;
3135	}
3136
3137	rq = task_rq_lock(p, &flags);
3138	put_cpu();
3139
3140	/*
3141	* select_task_rq() can race against ->cpus_allowed
3142	*/
3143	if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)
3144	\|\| unlikely(!cpu_active(dest_cpu))) {
3145	task_rq_unlock(rq, &flags);
3146	goto again;
3147	}
3148
3149	/* force the process onto the specified CPU */
3150	if (migrate_task(p, dest_cpu, &req)) {
3151	/* Need to wait for migration thread (might exit: take ref). */
3152	struct task_struct *mt = rq->migration_thread;
3153
3154	get_task_struct(mt);
3155	task_rq_unlock(rq, &flags);
3156	wake_up_process(mt);
3157	put_task_struct(mt);
3158	wait_for_completion(&req.done);
3159
3160	return;
3161	}
3162	task_rq_unlock(rq, &flags);
3163	}
3164
3165	#endif
3166
3167	DEFINE_PER_CPU(struct kernel_stat, kstat);
3168
3169	EXPORT_PER_CPU_SYMBOL(kstat);
3170
3171	/*
3172	* Return any ns on the sched_clock that have not yet been accounted in
3173	* @p in case that task is currently running.
3174	*
3175	* Called with task_rq_lock() held on @rq.
3176	*/
3177	static u64 do_task_delta_exec(struct task_struct p, struct rq rq)
3178	{
3179	u64 ns = 0;
3180
3181	if (task_current(rq, p)) {
3182	update_rq_clock(rq);
3183	ns = rq->clock - p->se.exec_start;
3184	if ((s64)ns < 0)
3185	ns = 0;
3186	}
3187
3188	return ns;
3189	}
3190
3191	unsigned long long task_delta_exec(struct task_struct *p)
3192	{
3193	unsigned long flags;
3194	struct rq *rq;
3195	u64 ns = 0;
3196
3197	rq = task_rq_lock(p, &flags);
3198	ns = do_task_delta_exec(p, rq);
3199	task_rq_unlock(rq, &flags);
3200
3201	return ns;
3202	}
3203
3204	/*
3205	* Return accounted runtime for the task.
3206	* In case the task is currently running, return the runtime plus current's
3207	* pending runtime that have not been accounted yet.
3208	*/
3209	unsigned long long task_sched_runtime(struct task_struct *p)
3210	{
3211	unsigned long flags;
3212	struct rq *rq;
3213	u64 ns = 0;
3214
3215	rq = task_rq_lock(p, &flags);
3216	ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq);
3217	task_rq_unlock(rq, &flags);
3218
3219	return ns;
3220	}
3221
3222	/*
3223	* Return sum_exec_runtime for the thread group.
3224	* In case the task is currently running, return the sum plus current's
3225	* pending runtime that have not been accounted yet.
3226	*
3227	* Note that the thread group might have other running tasks as well,
3228	* so the return value not includes other pending runtime that other
3229	* running tasks might have.
3230	*/
3231	unsigned long long thread_group_sched_runtime(struct task_struct *p)
3232	{
3233	struct task_cputime totals;
3234	unsigned long flags;
3235	struct rq *rq;
3236	u64 ns;
3237
3238	rq = task_rq_lock(p, &flags);
3239	thread_group_cputime(p, &totals);
3240	ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq);
3241	task_rq_unlock(rq, &flags);
3242
3243	return ns;
3244	}
3245
3246	/*
3247	* Account user cpu time to a process.
3248	* @p: the process that the cpu time gets accounted to
3249	* @cputime: the cpu time spent in user space since the last update
3250	* @cputime_scaled: cputime scaled by cpu frequency
3251	*/
3252	void account_user_time(struct task_struct *p, cputime_t cputime,
3253	cputime_t cputime_scaled)
3254	{
3255	struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3256	cputime64_t tmp;
3257
3258	/* Add user time to process. */
3259	p->utime = cputime_add(p->utime, cputime);
3260	p->utimescaled = cputime_add(p->utimescaled, cputime_scaled);
3261	account_group_user_time(p, cputime);
3262
3263	/* Add user time to cpustat. */
3264	tmp = cputime_to_cputime64(cputime);
3265	if (TASK_NICE(p) > 0)
3266	cpustat->nice = cputime64_add(cpustat->nice, tmp);
3267	else
3268	cpustat->user = cputime64_add(cpustat->user, tmp);
3269
3270	cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime);
3271	/* Account for user time used */
3272	acct_update_integrals(p);
3273	}
3274
3275	/*
3276	* Account guest cpu time to a process.
3277	* @p: the process that the cpu time gets accounted to
3278	* @cputime: the cpu time spent in virtual machine since the last update
3279	* @cputime_scaled: cputime scaled by cpu frequency
3280	*/
3281	static void account_guest_time(struct task_struct *p, cputime_t cputime,
3282	cputime_t cputime_scaled)
3283	{
3284	cputime64_t tmp;
3285	struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3286
3287	tmp = cputime_to_cputime64(cputime);
3288
3289	/* Add guest time to process. */
3290	p->utime = cputime_add(p->utime, cputime);
3291	p->utimescaled = cputime_add(p->utimescaled, cputime_scaled);
3292	account_group_user_time(p, cputime);
3293	p->gtime = cputime_add(p->gtime, cputime);
3294
3295	/* Add guest time to cpustat. */
3296	if (TASK_NICE(p) > 0) {
3297	cpustat->nice = cputime64_add(cpustat->nice, tmp);
3298	cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp);
3299	} else {
3300	cpustat->user = cputime64_add(cpustat->user, tmp);
3301	cpustat->guest = cputime64_add(cpustat->guest, tmp);
3302	}
3303	}
3304
3305	/*
3306	* Account system cpu time to a process.
3307	* @p: the process that the cpu time gets accounted to
3308	* @hardirq_offset: the offset to subtract from hardirq_count()
3309	* @cputime: the cpu time spent in kernel space since the last update
3310	* @cputime_scaled: cputime scaled by cpu frequency
3311	*/
3312	void account_system_time(struct task_struct *p, int hardirq_offset,
3313	cputime_t cputime, cputime_t cputime_scaled)
3314	{
3315	struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3316	cputime64_t tmp;
3317
3318	if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) {
3319	account_guest_time(p, cputime, cputime_scaled);
3320	return;
3321	}
3322
3323	/* Add system time to process. */
3324	p->stime = cputime_add(p->stime, cputime);
3325	p->stimescaled = cputime_add(p->stimescaled, cputime_scaled);
3326	account_group_system_time(p, cputime);
3327
3328	/* Add system time to cpustat. */
3329	tmp = cputime_to_cputime64(cputime);
3330	if (hardirq_count() - hardirq_offset)
3331	cpustat->irq = cputime64_add(cpustat->irq, tmp);
3332	else if (softirq_count())
3333	cpustat->softirq = cputime64_add(cpustat->softirq, tmp);
3334	else
3335	cpustat->system = cputime64_add(cpustat->system, tmp);
3336
3337	cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime);
3338
3339	/* Account for system time used */
3340	acct_update_integrals(p);
3341	}
3342
3343	/*
3344	* Account for involuntary wait time.
3345	* @steal: the cpu time spent in involuntary wait
3346	*/
3347	void account_steal_time(cputime_t cputime)
3348	{
3349	struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3350	cputime64_t cputime64 = cputime_to_cputime64(cputime);
3351
3352	cpustat->steal = cputime64_add(cpustat->steal, cputime64);
3353	}
3354
3355	/*
3356	* Account for idle time.
3357	* @cputime: the cpu time spent in idle wait
3358	*/
3359	void account_idle_time(cputime_t cputime)
3360	{
3361	struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3362	cputime64_t cputime64 = cputime_to_cputime64(cputime);
3363	struct rq *rq = this_rq();
3364
3365	if (atomic_read(&rq->nr_iowait) > 0)
3366	cpustat->iowait = cputime64_add(cpustat->iowait, cputime64);
3367	else
3368	cpustat->idle = cputime64_add(cpustat->idle, cputime64);
3369	}
3370
3371	#ifndef CONFIG_VIRT_CPU_ACCOUNTING
3372
3373	/*
3374	* Account a single tick of cpu time.
3375	* @p: the process that the cpu time gets accounted to
3376	* @user_tick: indicates if the tick is a user or a system tick
3377	*/
3378	void account_process_tick(struct task_struct *p, int user_tick)
3379	{
3380	cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy);
3381	struct rq *rq = this_rq();
3382
3383	if (user_tick)
3384	account_user_time(p, cputime_one_jiffy, one_jiffy_scaled);
3385	else if ((p != rq->idle) \|\| (irq_count() != HARDIRQ_OFFSET))
3386	account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy,
3387	one_jiffy_scaled);
3388	else
3389	account_idle_time(cputime_one_jiffy);
3390	}
3391
3392	/*
3393	* Account multiple ticks of steal time.
3394	* @p: the process from which the cpu time has been stolen
3395	* @ticks: number of stolen ticks
3396	*/
3397	void account_steal_ticks(unsigned long ticks)
3398	{
3399	account_steal_time(jiffies_to_cputime(ticks));
3400	}
3401
3402	/*
3403	* Account multiple ticks of idle time.
3404	* @ticks: number of stolen ticks
3405	*/
3406	void account_idle_ticks(unsigned long ticks)
3407	{
3408	account_idle_time(jiffies_to_cputime(ticks));
3409	}
3410
3411	#endif
3412
3413	/*
3414	* Use precise platform statistics if available:
3415	*/
3416	#ifdef CONFIG_VIRT_CPU_ACCOUNTING
3417	void task_times(struct task_struct p, cputime_t ut, cputime_t *st)
3418	{
3419	*ut = p->utime;
3420	*st = p->stime;
3421	}
3422
3423	void thread_group_times(struct task_struct p, cputime_t ut, cputime_t *st)
3424	{
3425	struct task_cputime cputime;
3426
3427	thread_group_cputime(p, &cputime);
3428
3429	*ut = cputime.utime;
3430	*st = cputime.stime;
3431	}
3432	#else
3433
3434	#ifndef nsecs_to_cputime
3435	# define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs)
3436	#endif
3437
3438	void task_times(struct task_struct p, cputime_t ut, cputime_t *st)
3439	{
3440	cputime_t rtime, utime = p->utime, total = cputime_add(utime, p->stime);
3441
3442	/*
3443	* Use CFS's precise accounting:
3444	*/
3445	rtime = nsecs_to_cputime(p->se.sum_exec_runtime);
3446
3447	if (total) {
3448	u64 temp;
3449
3450	temp = (u64)(rtime * utime);
3451	do_div(temp, total);
3452	utime = (cputime_t)temp;
3453	} else
3454	utime = rtime;
3455
3456	/*
3457	* Compare with previous values, to keep monotonicity:
3458	*/
3459	p->prev_utime = max(p->prev_utime, utime);
3460	p->prev_stime = max(p->prev_stime, cputime_sub(rtime, p->prev_utime));
3461
3462	*ut = p->prev_utime;
3463	*st = p->prev_stime;
3464	}
3465
3466	/*
3467	* Must be called with siglock held.
3468	*/
3469	void thread_group_times(struct task_struct p, cputime_t ut, cputime_t *st)
3470	{
3471	struct signal_struct *sig = p->signal;
3472	struct task_cputime cputime;
3473	cputime_t rtime, utime, total;
3474
3475	thread_group_cputime(p, &cputime);
3476
3477	total = cputime_add(cputime.utime, cputime.stime);
3478	rtime = nsecs_to_cputime(cputime.sum_exec_runtime);
3479
3480	if (total) {
3481	u64 temp;
3482
3483	temp = (u64)(rtime * cputime.utime);
3484	do_div(temp, total);
3485	utime = (cputime_t)temp;
3486	} else
3487	utime = rtime;
3488
3489	sig->prev_utime = max(sig->prev_utime, utime);
3490	sig->prev_stime = max(sig->prev_stime,
3491	cputime_sub(rtime, sig->prev_utime));
3492
3493	*ut = sig->prev_utime;
3494	*st = sig->prev_stime;
3495	}
3496	#endif
3497
3498	/*
3499	* This function gets called by the timer code, with HZ frequency.
3500	* We call it with interrupts disabled.
3501	*
3502	* It also gets called by the fork code, when changing the parent's
3503	* timeslices.
3504	*/
3505	void scheduler_tick(void)
3506	{
3507	int cpu = smp_processor_id();
3508	struct rq *rq = cpu_rq(cpu);
3509	struct task_struct *curr = rq->curr;
3510
3511	sched_clock_tick();
3512
3513	raw_spin_lock(&rq->lock);
3514	update_rq_clock(rq);
3515	update_cpu_load(rq);
3516	curr->sched_class->task_tick(rq, curr, 0);
3517	raw_spin_unlock(&rq->lock);
3518
3519	perf_event_task_tick(curr);
3520
3521	#ifdef CONFIG_SMP
3522	rq->idle_at_tick = idle_cpu(cpu);
3523	trigger_load_balance(rq, cpu);
3524	#endif
3525	}
3526
3527	notrace unsigned long get_parent_ip(unsigned long addr)
3528	{
3529	if (in_lock_functions(addr)) {
3530	addr = CALLER_ADDR2;
3531	if (in_lock_functions(addr))
3532	addr = CALLER_ADDR3;
3533	}
3534	return addr;
3535	}
3536
3537	#if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) \|\| \
3538	defined(CONFIG_PREEMPT_TRACER))
3539
3540	void __kprobes add_preempt_count(int val)
3541	{
3542	#ifdef CONFIG_DEBUG_PREEMPT
3543	/*
3544	* Underflow?
3545	*/
3546	if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0)))
3547	return;
3548	#endif
3549	preempt_count() += val;
3550	#ifdef CONFIG_DEBUG_PREEMPT
3551	/*
3552	* Spinlock count overflowing soon?
3553	*/
3554	DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >=
3555	PREEMPT_MASK - 10);
3556	#endif
3557	if (preempt_count() == val)
3558	trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1));
3559	}
3560	EXPORT_SYMBOL(add_preempt_count);
3561
3562	void __kprobes sub_preempt_count(int val)
3563	{
3564	#ifdef CONFIG_DEBUG_PREEMPT
3565	/*
3566	* Underflow?
3567	*/
3568	if (DEBUG_LOCKS_WARN_ON(val > preempt_count()))
3569	return;
3570	/*
3571	* Is the spinlock portion underflowing?
3572	*/
3573	if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) &&
3574	!(preempt_count() & PREEMPT_MASK)))
3575	return;
3576	#endif
3577
3578	if (preempt_count() == val)
3579	trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1));
3580	preempt_count() -= val;
3581	}
3582	EXPORT_SYMBOL(sub_preempt_count);
3583
3584	#endif
3585
3586	/*
3587	* Print scheduling while atomic bug:
3588	*/
3589	static noinline void __schedule_bug(struct task_struct *prev)
3590	{
3591	struct pt_regs *regs = get_irq_regs();
3592
3593	printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n",
3594	prev->comm, prev->pid, preempt_count());
3595
3596	debug_show_held_locks(prev);
3597	print_modules();
3598	if (irqs_disabled())
3599	print_irqtrace_events(prev);
3600
3601	if (regs)
3602	show_regs(regs);
3603	else
3604	dump_stack();
3605	}
3606
3607	/*
3608	* Various schedule()-time debugging checks and statistics:
3609	*/
3610	static inline void schedule_debug(struct task_struct *prev)
3611	{
3612	/*
3613	* Test if we are atomic. Since do_exit() needs to call into
3614	* schedule() atomically, we ignore that path for now.
3615	* Otherwise, whine if we are scheduling when we should not be.
3616	*/
3617	if (unlikely(in_atomic_preempt_off() && !prev->exit_state))
3618	__schedule_bug(prev);
3619
3620	profile_hit(SCHED_PROFILING, __builtin_return_address(0));
3621
3622	schedstat_inc(this_rq(), sched_count);
3623	#ifdef CONFIG_SCHEDSTATS
3624	if (unlikely(prev->lock_depth >= 0)) {
3625	schedstat_inc(this_rq(), bkl_count);
3626	schedstat_inc(prev, sched_info.bkl_count);
3627	}
3628	#endif
3629	}
3630
3631	static void put_prev_task(struct rq rq, struct task_struct prev)
3632	{
3633	if (prev->state == TASK_RUNNING) {
3634	u64 runtime = prev->se.sum_exec_runtime;
3635
3636	runtime -= prev->se.prev_sum_exec_runtime;
3637	runtime = min_t(u64, runtime, 2*sysctl_sched_migration_cost);
3638
3639	/*
3640	* In order to avoid avg_overlap growing stale when we are
3641	* indeed overlapping and hence not getting put to sleep, grow
3642	* the avg_overlap on preemption.
3643	*
3644	* We use the average preemption runtime because that
3645	* correlates to the amount of cache footprint a task can
3646	* build up.
3647	*/
3648	update_avg(&prev->se.avg_overlap, runtime);
3649	}
3650	prev->sched_class->put_prev_task(rq, prev);
3651	}
3652
3653	/*
3654	* Pick up the highest-prio task:
3655	*/
3656	static inline struct task_struct *
3657	pick_next_task(struct rq *rq)
3658	{
3659	const struct sched_class *class;
3660	struct task_struct *p;
3661
3662	/*
3663	* Optimization: we know that if all tasks are in
3664	* the fair class we can call that function directly:
3665	*/
3666	if (likely(rq->nr_running == rq->cfs.nr_running)) {
3667	p = fair_sched_class.pick_next_task(rq);
3668	if (likely(p))
3669	return p;
3670	}
3671
3672	class = sched_class_highest;
3673	for ( ; ; ) {
3674	p = class->pick_next_task(rq);
3675	if (p)
3676	return p;
3677	/*
3678	* Will never be NULL as the idle class always
3679	* returns a non-NULL p:
3680	*/
3681	class = class->next;
3682	}
3683	}
3684
3685	/*
3686	* schedule() is the main scheduler function.
3687	*/
3688	asmlinkage void __sched schedule(void)
3689	{
3690	struct task_struct prev, next;
3691	unsigned long *switch_count;
3692	struct rq *rq;
3693	int cpu;
3694
3695	need_resched:
3696	preempt_disable();
3697	cpu = smp_processor_id();
3698	rq = cpu_rq(cpu);
3699	rcu_sched_qs(cpu);
3700	prev = rq->curr;
3701	switch_count = &prev->nivcsw;
3702
3703	release_kernel_lock(prev);
3704	need_resched_nonpreemptible:
3705
3706	schedule_debug(prev);
3707
3708	if (sched_feat(HRTICK))
3709	hrtick_clear(rq);
3710
3711	raw_spin_lock_irq(&rq->lock);
3712	update_rq_clock(rq);
3713	clear_tsk_need_resched(prev);
3714
3715	if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) {
3716	if (unlikely(signal_pending_state(prev->state, prev)))
3717	prev->state = TASK_RUNNING;
3718	else
3719	deactivate_task(rq, prev, 1);
3720	switch_count = &prev->nvcsw;
3721	}
3722
3723	pre_schedule(rq, prev);
3724
3725	if (unlikely(!rq->nr_running))
3726	idle_balance(cpu, rq);
3727
3728	put_prev_task(rq, prev);
3729	next = pick_next_task(rq);
3730
3731	if (likely(prev != next)) {
3732	sched_info_switch(prev, next);
3733	perf_event_task_sched_out(prev, next);
3734
3735	rq->nr_switches++;
3736	rq->curr = next;
3737	++*switch_count;
3738
3739	context_switch(rq, prev, next); /* unlocks the rq */
3740	/*
3741	* the context switch might have flipped the stack from under
3742	* us, hence refresh the local variables.
3743	*/
3744	cpu = smp_processor_id();
3745	rq = cpu_rq(cpu);
3746	} else
3747	raw_spin_unlock_irq(&rq->lock);
3748
3749	post_schedule(rq);
3750
3751	if (unlikely(reacquire_kernel_lock(current) < 0)) {
3752	prev = rq->curr;
3753	switch_count = &prev->nivcsw;
3754	goto need_resched_nonpreemptible;
3755	}
3756
3757	preempt_enable_no_resched();
3758	if (need_resched())
3759	goto need_resched;
3760	}
3761	EXPORT_SYMBOL(schedule);
3762
3763	#ifdef CONFIG_MUTEX_SPIN_ON_OWNER
3764	/*
3765	* Look out! "owner" is an entirely speculative pointer
3766	* access and not reliable.
3767	*/
3768	int mutex_spin_on_owner(struct mutex lock, struct thread_info owner)
3769	{
3770	unsigned int cpu;
3771	struct rq *rq;
3772
3773	if (!sched_feat(OWNER_SPIN))
3774	return 0;
3775
3776	#ifdef CONFIG_DEBUG_PAGEALLOC
3777	/*
3778	* Need to access the cpu field knowing that
3779	* DEBUG_PAGEALLOC could have unmapped it if
3780	* the mutex owner just released it and exited.
3781	*/
3782	if (probe_kernel_address(&owner->cpu, cpu))
3783	goto out;
3784	#else
3785	cpu = owner->cpu;
3786	#endif
3787
3788	/*
3789	* Even if the access succeeded (likely case),
3790	* the cpu field may no longer be valid.
3791	*/
3792	if (cpu >= nr_cpumask_bits)
3793	goto out;
3794
3795	/*
3796	* We need to validate that we can do a
3797	* get_cpu() and that we have the percpu area.
3798	*/
3799	if (!cpu_online(cpu))
3800	goto out;
3801
3802	rq = cpu_rq(cpu);
3803
3804	for (;;) {
3805	/*
3806	* Owner changed, break to re-assess state.
3807	*/
3808	if (lock->owner != owner)
3809	break;
3810
3811	/*
3812	* Is that owner really running on that cpu?
3813	*/
3814	if (task_thread_info(rq->curr) != owner \|\| need_resched())
3815	return 0;
3816
3817	cpu_relax();
3818	}
3819	out:
3820	return 1;
3821	}
3822	#endif
3823
3824	#ifdef CONFIG_PREEMPT
3825	/*
3826	* this is the entry point to schedule() from in-kernel preemption
3827	* off of preempt_enable. Kernel preemptions off return from interrupt
3828	* occur there and call schedule directly.
3829	*/
3830	asmlinkage void __sched preempt_schedule(void)
3831	{
3832	struct thread_info *ti = current_thread_info();
3833
3834	/*
3835	* If there is a non-zero preempt_count or interrupts are disabled,
3836	* we do not want to preempt the current task. Just return..
3837	*/
3838	if (likely(ti->preempt_count \|\| irqs_disabled()))
3839	return;
3840
3841	do {
3842	add_preempt_count(PREEMPT_ACTIVE);
3843	schedule();
3844	sub_preempt_count(PREEMPT_ACTIVE);
3845
3846	/*
3847	* Check again in case we missed a preemption opportunity
3848	* between schedule and now.
3849	*/
3850	barrier();
3851	} while (need_resched());
3852	}
3853	EXPORT_SYMBOL(preempt_schedule);
3854
3855	/*
3856	* this is the entry point to schedule() from kernel preemption
3857	* off of irq context.
3858	* Note, that this is called and return with irqs disabled. This will
3859	* protect us against recursive calling from irq.
3860	*/
3861	asmlinkage void __sched preempt_schedule_irq(void)
3862	{
3863	struct thread_info *ti = current_thread_info();
3864
3865	/* Catch callers which need to be fixed */
3866	BUG_ON(ti->preempt_count \|\| !irqs_disabled());
3867
3868	do {
3869	add_preempt_count(PREEMPT_ACTIVE);
3870	local_irq_enable();
3871	schedule();
3872	local_irq_disable();
3873	sub_preempt_count(PREEMPT_ACTIVE);
3874
3875	/*
3876	* Check again in case we missed a preemption opportunity
3877	* between schedule and now.
3878	*/
3879	barrier();
3880	} while (need_resched());
3881	}
3882
3883	#endif /* CONFIG_PREEMPT */
3884
3885	int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags,
3886	void *key)
3887	{
3888	return try_to_wake_up(curr->private, mode, wake_flags);
3889	}
3890	EXPORT_SYMBOL(default_wake_function);
3891
3892	/*
3893	* The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just
3894	* wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve
3895	* number) then we wake all the non-exclusive tasks and one exclusive task.
3896	*
3897	* There are circumstances in which we can try to wake a task which has already
3898	* started to run but is not in state TASK_RUNNING. try_to_wake_up() returns
3899	* zero in this (rare) case, and we handle it by continuing to scan the queue.
3900	*/
3901	static void __wake_up_common(wait_queue_head_t *q, unsigned int mode,
3902	int nr_exclusive, int wake_flags, void *key)
3903	{
3904	wait_queue_t curr, next;
3905
3906	list_for_each_entry_safe(curr, next, &q->task_list, task_list) {
3907	unsigned flags = curr->flags;
3908
3909	if (curr->func(curr, mode, wake_flags, key) &&
3910	(flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive)
3911	break;
3912	}
3913	}
3914
3915	/**
3916	* __wake_up - wake up threads blocked on a waitqueue.
3917	* @q: the waitqueue
3918	* @mode: which threads
3919	* @nr_exclusive: how many wake-one or wake-many threads to wake up
3920	* @key: is directly passed to the wakeup function
3921	*
3922	* It may be assumed that this function implies a write memory barrier before
3923	* changing the task state if and only if any tasks are woken up.
3924	*/
3925	void __wake_up(wait_queue_head_t *q, unsigned int mode,
3926	int nr_exclusive, void *key)
3927	{
3928	unsigned long flags;
3929
3930	spin_lock_irqsave(&q->lock, flags);
3931	__wake_up_common(q, mode, nr_exclusive, 0, key);
3932	spin_unlock_irqrestore(&q->lock, flags);
3933	}
3934	EXPORT_SYMBOL(__wake_up);
3935
3936	/*
3937	* Same as __wake_up but called with the spinlock in wait_queue_head_t held.
3938	*/
3939	void __wake_up_locked(wait_queue_head_t *q, unsigned int mode)
3940	{
3941	__wake_up_common(q, mode, 1, 0, NULL);
3942	}
3943
3944	void __wake_up_locked_key(wait_queue_head_t q, unsigned int mode, void key)
3945	{
3946	__wake_up_common(q, mode, 1, 0, key);
3947	}
3948
3949	/**
3950	* __wake_up_sync_key - wake up threads blocked on a waitqueue.
3951	* @q: the waitqueue
3952	* @mode: which threads
3953	* @nr_exclusive: how many wake-one or wake-many threads to wake up
3954	* @key: opaque value to be passed to wakeup targets
3955	*
3956	* The sync wakeup differs that the waker knows that it will schedule
3957	* away soon, so while the target thread will be woken up, it will not
3958	* be migrated to another CPU - ie. the two threads are 'synchronized'
3959	* with each other. This can prevent needless bouncing between CPUs.
3960	*
3961	* On UP it can prevent extra preemption.
3962	*
3963	* It may be assumed that this function implies a write memory barrier before
3964	* changing the task state if and only if any tasks are woken up.
3965	*/
3966	void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode,
3967	int nr_exclusive, void *key)
3968	{
3969	unsigned long flags;
3970	int wake_flags = WF_SYNC;
3971
3972	if (unlikely(!q))
3973	return;
3974
3975	if (unlikely(!nr_exclusive))
3976	wake_flags = 0;
3977
3978	spin_lock_irqsave(&q->lock, flags);
3979	__wake_up_common(q, mode, nr_exclusive, wake_flags, key);
3980	spin_unlock_irqrestore(&q->lock, flags);
3981	}
3982	EXPORT_SYMBOL_GPL(__wake_up_sync_key);
3983
3984	/*
3985	* __wake_up_sync - see __wake_up_sync_key()
3986	*/
3987	void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive)
3988	{
3989	__wake_up_sync_key(q, mode, nr_exclusive, NULL);
3990	}
3991	EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */
3992
3993	/**
3994	* complete: - signals a single thread waiting on this completion
3995	* @x: holds the state of this particular completion
3996	*
3997	* This will wake up a single thread waiting on this completion. Threads will be
3998	* awakened in the same order in which they were queued.
3999	*
4000	* See also complete_all(), wait_for_completion() and related routines.
4001	*
4002	* It may be assumed that this function implies a write memory barrier before
4003	* changing the task state if and only if any tasks are woken up.
4004	*/
4005	void complete(struct completion *x)
4006	{
4007	unsigned long flags;
4008
4009	spin_lock_irqsave(&x->wait.lock, flags);
4010	x->done++;
4011	__wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL);
4012	spin_unlock_irqrestore(&x->wait.lock, flags);
4013	}
4014	EXPORT_SYMBOL(complete);
4015
4016	/**
4017	* complete_all: - signals all threads waiting on this completion
4018	* @x: holds the state of this particular completion
4019	*
4020	* This will wake up all threads waiting on this particular completion event.
4021	*
4022	* It may be assumed that this function implies a write memory barrier before
4023	* changing the task state if and only if any tasks are woken up.
4024	*/
4025	void complete_all(struct completion *x)
4026	{
4027	unsigned long flags;
4028
4029	spin_lock_irqsave(&x->wait.lock, flags);
4030	x->done += UINT_MAX/2;
4031	__wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL);
4032	spin_unlock_irqrestore(&x->wait.lock, flags);
4033	}
4034	EXPORT_SYMBOL(complete_all);
4035
4036	static inline long __sched
4037	do_wait_for_common(struct completion *x, long timeout, int state)
4038	{
4039	if (!x->done) {
4040	DECLARE_WAITQUEUE(wait, current);
4041
4042	wait.flags \|= WQ_FLAG_EXCLUSIVE;
4043	__add_wait_queue_tail(&x->wait, &wait);
4044	do {
4045	if (signal_pending_state(state, current)) {
4046	timeout = -ERESTARTSYS;
4047	break;
4048	}
4049	__set_current_state(state);
4050	spin_unlock_irq(&x->wait.lock);
4051	timeout = schedule_timeout(timeout);
4052	spin_lock_irq(&x->wait.lock);
4053	} while (!x->done && timeout);
4054	__remove_wait_queue(&x->wait, &wait);
4055	if (!x->done)
4056	return timeout;
4057	}
4058	x->done--;
4059	return timeout ?: 1;
4060	}
4061
4062	static long __sched
4063	wait_for_common(struct completion *x, long timeout, int state)
4064	{
4065	might_sleep();
4066
4067	spin_lock_irq(&x->wait.lock);
4068	timeout = do_wait_for_common(x, timeout, state);
4069	spin_unlock_irq(&x->wait.lock);
4070	return timeout;
4071	}
4072
4073	/**
4074	* wait_for_completion: - waits for completion of a task
4075	* @x: holds the state of this particular completion
4076	*
4077	* This waits to be signaled for completion of a specific task. It is NOT
4078	* interruptible and there is no timeout.
4079	*
4080	* See also similar routines (i.e. wait_for_completion_timeout()) with timeout
4081	* and interrupt capability. Also see complete().
4082	*/
4083	void __sched wait_for_completion(struct completion *x)
4084	{
4085	wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE);
4086	}
4087	EXPORT_SYMBOL(wait_for_completion);
4088
4089	/**
4090	* wait_for_completion_timeout: - waits for completion of a task (w/timeout)
4091	* @x: holds the state of this particular completion
4092	* @timeout: timeout value in jiffies
4093	*
4094	* This waits for either a completion of a specific task to be signaled or for a
4095	* specified timeout to expire. The timeout is in jiffies. It is not
4096	* interruptible.
4097	*/
4098	unsigned long __sched
4099	wait_for_completion_timeout(struct completion *x, unsigned long timeout)
4100	{
4101	return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE);
4102	}
4103	EXPORT_SYMBOL(wait_for_completion_timeout);
4104
4105	/**
4106	* wait_for_completion_interruptible: - waits for completion of a task (w/intr)
4107	* @x: holds the state of this particular completion
4108	*
4109	* This waits for completion of a specific task to be signaled. It is
4110	* interruptible.
4111	*/
4112	int __sched wait_for_completion_interruptible(struct completion *x)
4113	{
4114	long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE);
4115	if (t == -ERESTARTSYS)
4116	return t;
4117	return 0;
4118	}
4119	EXPORT_SYMBOL(wait_for_completion_interruptible);
4120
4121	/**
4122	* wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr))
4123	* @x: holds the state of this particular completion
4124	* @timeout: timeout value in jiffies
4125	*
4126	* This waits for either a completion of a specific task to be signaled or for a
4127	* specified timeout to expire. It is interruptible. The timeout is in jiffies.
4128	*/
4129	unsigned long __sched
4130	wait_for_completion_interruptible_timeout(struct completion *x,
4131	unsigned long timeout)
4132	{
4133	return wait_for_common(x, timeout, TASK_INTERRUPTIBLE);
4134	}
4135	EXPORT_SYMBOL(wait_for_completion_interruptible_timeout);
4136
4137	/**
4138	* wait_for_completion_killable: - waits for completion of a task (killable)
4139	* @x: holds the state of this particular completion
4140	*
4141	* This waits to be signaled for completion of a specific task. It can be
4142	* interrupted by a kill signal.
4143	*/
4144	int __sched wait_for_completion_killable(struct completion *x)
4145	{
4146	long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE);
4147	if (t == -ERESTARTSYS)
4148	return t;
4149	return 0;
4150	}
4151	EXPORT_SYMBOL(wait_for_completion_killable);
4152
4153	/**
4154	* try_wait_for_completion - try to decrement a completion without blocking
4155	* @x: completion structure
4156	*
4157	* Returns: 0 if a decrement cannot be done without blocking
4158	* 1 if a decrement succeeded.
4159	*
4160	* If a completion is being used as a counting completion,
4161	* attempt to decrement the counter without blocking. This
4162	* enables us to avoid waiting if the resource the completion
4163	* is protecting is not available.
4164	*/
4165	bool try_wait_for_completion(struct completion *x)
4166	{
4167	unsigned long flags;
4168	int ret = 1;
4169
4170	spin_lock_irqsave(&x->wait.lock, flags);
4171	if (!x->done)
4172	ret = 0;
4173	else
4174	x->done--;
4175	spin_unlock_irqrestore(&x->wait.lock, flags);
4176	return ret;
4177	}
4178	EXPORT_SYMBOL(try_wait_for_completion);
4179
4180	/**
4181	* completion_done - Test to see if a completion has any waiters
4182	* @x: completion structure
4183	*
4184	* Returns: 0 if there are waiters (wait_for_completion() in progress)
4185	* 1 if there are no waiters.
4186	*
4187	*/
4188	bool completion_done(struct completion *x)
4189	{
4190	unsigned long flags;
4191	int ret = 1;
4192
4193	spin_lock_irqsave(&x->wait.lock, flags);
4194	if (!x->done)
4195	ret = 0;
4196	spin_unlock_irqrestore(&x->wait.lock, flags);
4197	return ret;
4198	}
4199	EXPORT_SYMBOL(completion_done);
4200
4201	static long __sched
4202	sleep_on_common(wait_queue_head_t *q, int state, long timeout)
4203	{
4204	unsigned long flags;
4205	wait_queue_t wait;
4206
4207	init_waitqueue_entry(&wait, current);
4208
4209	__set_current_state(state);
4210
4211	spin_lock_irqsave(&q->lock, flags);
4212	__add_wait_queue(q, &wait);
4213	spin_unlock(&q->lock);
4214	timeout = schedule_timeout(timeout);
4215	spin_lock_irq(&q->lock);
4216	__remove_wait_queue(q, &wait);
4217	spin_unlock_irqrestore(&q->lock, flags);
4218
4219	return timeout;
4220	}
4221
4222	void __sched interruptible_sleep_on(wait_queue_head_t *q)
4223	{
4224	sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
4225	}
4226	EXPORT_SYMBOL(interruptible_sleep_on);
4227
4228	long __sched
4229	interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout)
4230	{
4231	return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout);
4232	}
4233	EXPORT_SYMBOL(interruptible_sleep_on_timeout);
4234
4235	void __sched sleep_on(wait_queue_head_t *q)
4236	{
4237	sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
4238	}
4239	EXPORT_SYMBOL(sleep_on);
4240
4241	long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout)
4242	{
4243	return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout);
4244	}
4245	EXPORT_SYMBOL(sleep_on_timeout);
4246
4247	#ifdef CONFIG_RT_MUTEXES
4248
4249	/*
4250	* rt_mutex_setprio - set the current priority of a task
4251	* @p: task
4252	* @prio: prio value (kernel-internal form)
4253	*
4254	* This function changes the 'effective' priority of a task. It does
4255	* not touch ->normal_prio like __setscheduler().
4256	*
4257	* Used by the rt_mutex code to implement priority inheritance logic.
4258	*/
4259	void rt_mutex_setprio(struct task_struct *p, int prio)
4260	{
4261	unsigned long flags;
4262	int oldprio, on_rq, running;
4263	struct rq *rq;
4264	const struct sched_class *prev_class;
4265
4266	BUG_ON(prio < 0 \|\| prio > MAX_PRIO);
4267
4268	rq = task_rq_lock(p, &flags);
4269	update_rq_clock(rq);
4270
4271	oldprio = p->prio;
4272	prev_class = p->sched_class;
4273	on_rq = p->se.on_rq;
4274	running = task_current(rq, p);
4275	if (on_rq)
4276	dequeue_task(rq, p, 0);
4277	if (running)
4278	p->sched_class->put_prev_task(rq, p);
4279
4280	if (rt_prio(prio))
4281	p->sched_class = &rt_sched_class;
4282	else
4283	p->sched_class = &fair_sched_class;
4284
4285	p->prio = prio;
4286
4287	if (running)
4288	p->sched_class->set_curr_task(rq);
4289	if (on_rq) {
4290	enqueue_task(rq, p, 0, oldprio < prio);
4291
4292	check_class_changed(rq, p, prev_class, oldprio, running);
4293	}
4294	task_rq_unlock(rq, &flags);
4295	}
4296
4297	#endif
4298
4299	void set_user_nice(struct task_struct *p, long nice)
4300	{
4301	int old_prio, delta, on_rq;
4302	unsigned long flags;
4303	struct rq *rq;
4304
4305	if (TASK_NICE(p) == nice \|\| nice < -20 \|\| nice > 19)
4306	return;
4307	/*
4308	* We have to be careful, if called from sys_setpriority(),
4309	* the task might be in the middle of scheduling on another CPU.
4310	*/
4311	rq = task_rq_lock(p, &flags);
4312	update_rq_clock(rq);
4313	/*
4314	* The RT priorities are set via sched_setscheduler(), but we still
4315	* allow the 'normal' nice value to be set - but as expected
4316	* it wont have any effect on scheduling until the task is
4317	* SCHED_FIFO/SCHED_RR:
4318	*/
4319	if (task_has_rt_policy(p)) {
4320	p->static_prio = NICE_TO_PRIO(nice);
4321	goto out_unlock;
4322	}
4323	on_rq = p->se.on_rq;
4324	if (on_rq)
4325	dequeue_task(rq, p, 0);
4326
4327	p->static_prio = NICE_TO_PRIO(nice);
4328	set_load_weight(p);
4329	old_prio = p->prio;
4330	p->prio = effective_prio(p);
4331	delta = p->prio - old_prio;
4332
4333	if (on_rq) {
4334	enqueue_task(rq, p, 0, false);
4335	/*
4336	* If the task increased its priority or is running and
4337	* lowered its priority, then reschedule its CPU:
4338	*/
4339	if (delta < 0 \|\| (delta > 0 && task_running(rq, p)))
4340	resched_task(rq->curr);
4341	}
4342	out_unlock:
4343	task_rq_unlock(rq, &flags);
4344	}
4345	EXPORT_SYMBOL(set_user_nice);
4346
4347	/*
4348	* can_nice - check if a task can reduce its nice value
4349	* @p: task
4350	* @nice: nice value
4351	*/
4352	int can_nice(const struct task_struct *p, const int nice)
4353	{
4354	/* convert nice value [19,-20] to rlimit style value [1,40] */
4355	int nice_rlim = 20 - nice;
4356
4357	return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) \|\|
4358	capable(CAP_SYS_NICE));
4359	}
4360
4361	#ifdef __ARCH_WANT_SYS_NICE
4362
4363	/*
4364	* sys_nice - change the priority of the current process.
4365	* @increment: priority increment
4366	*
4367	* sys_setpriority is a more generic, but much slower function that
4368	* does similar things.
4369	*/
4370	SYSCALL_DEFINE1(nice, int, increment)
4371	{
4372	long nice, retval;
4373
4374	/*
4375	* Setpriority might change our priority at the same moment.
4376	* We don't have to worry. Conceptually one call occurs first
4377	* and we have a single winner.
4378	*/
4379	if (increment < -40)
4380	increment = -40;
4381	if (increment > 40)
4382	increment = 40;
4383
4384	nice = TASK_NICE(current) + increment;
4385	if (nice < -20)
4386	nice = -20;
4387	if (nice > 19)
4388	nice = 19;
4389
4390	if (increment < 0 && !can_nice(current, nice))
4391	return -EPERM;
4392
4393	retval = security_task_setnice(current, nice);
4394	if (retval)
4395	return retval;
4396
4397	set_user_nice(current, nice);
4398	return 0;
4399	}
4400
4401	#endif
4402
4403	/**
4404	* task_prio - return the priority value of a given task.
4405	* @p: the task in question.
4406	*
4407	* This is the priority value as seen by users in /proc.
4408	* RT tasks are offset by -200. Normal tasks are centered
4409	* around 0, value goes from -16 to +15.
4410	*/
4411	int task_prio(const struct task_struct *p)
4412	{
4413	return p->prio - MAX_RT_PRIO;
4414	}
4415
4416	/**
4417	* task_nice - return the nice value of a given task.
4418	* @p: the task in question.
4419	*/
4420	int task_nice(const struct task_struct *p)
4421	{
4422	return TASK_NICE(p);
4423	}
4424	EXPORT_SYMBOL(task_nice);
4425
4426	/**
4427	* idle_cpu - is a given cpu idle currently?
4428	* @cpu: the processor in question.
4429	*/
4430	int idle_cpu(int cpu)
4431	{
4432	return cpu_curr(cpu) == cpu_rq(cpu)->idle;
4433	}
4434
4435	/**
4436	* idle_task - return the idle task for a given cpu.
4437	* @cpu: the processor in question.
4438	*/
4439	struct task_struct *idle_task(int cpu)
4440	{
4441	return cpu_rq(cpu)->idle;
4442	}
4443
4444	/**
4445	* find_process_by_pid - find a process with a matching PID value.
4446	* @pid: the pid in question.
4447	*/
4448	static struct task_struct *find_process_by_pid(pid_t pid)
4449	{
4450	return pid ? find_task_by_vpid(pid) : current;
4451	}
4452
4453	/* Actually do priority change: must hold rq lock. */
4454	static void
4455	__setscheduler(struct rq rq, struct task_struct p, int policy, int prio)
4456	{
4457	BUG_ON(p->se.on_rq);
4458
4459	p->policy = policy;
4460	p->rt_priority = prio;
4461	p->normal_prio = normal_prio(p);
4462	/* we are holding p->pi_lock already */
4463	p->prio = rt_mutex_getprio(p);
4464	if (rt_prio(p->prio))
4465	p->sched_class = &rt_sched_class;
4466	else
4467	p->sched_class = &fair_sched_class;
4468	set_load_weight(p);
4469	}
4470
4471	/*
4472	* check the target process has a UID that matches the current process's
4473	*/
4474	static bool check_same_owner(struct task_struct *p)
4475	{
4476	const struct cred cred = current_cred(), pcred;
4477	bool match;
4478
4479	rcu_read_lock();
4480	pcred = __task_cred(p);
4481	match = (cred->euid == pcred->euid \|\|
4482	cred->euid == pcred->uid);
4483	rcu_read_unlock();
4484	return match;
4485	}
4486
4487	static int __sched_setscheduler(struct task_struct *p, int policy,
4488	struct sched_param *param, bool user)
4489	{
4490	int retval, oldprio, oldpolicy = -1, on_rq, running;
4491	unsigned long flags;
4492	const struct sched_class *prev_class;
4493	struct rq *rq;
4494	int reset_on_fork;
4495
4496	/* may grab non-irq protected spin_locks */
4497	BUG_ON(in_interrupt());
4498	recheck:
4499	/* double check policy once rq lock held */
4500	if (policy < 0) {
4501	reset_on_fork = p->sched_reset_on_fork;
4502	policy = oldpolicy = p->policy;
4503	} else {
4504	reset_on_fork = !!(policy & SCHED_RESET_ON_FORK);
4505	policy &= ~SCHED_RESET_ON_FORK;
4506
4507	if (policy != SCHED_FIFO && policy != SCHED_RR &&
4508	policy != SCHED_NORMAL && policy != SCHED_BATCH &&
4509	policy != SCHED_IDLE)
4510	return -EINVAL;
4511	}
4512
4513	/*
4514	* Valid priorities for SCHED_FIFO and SCHED_RR are
4515	* 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL,
4516	* SCHED_BATCH and SCHED_IDLE is 0.
4517	*/
4518	if (param->sched_priority < 0 \|\|
4519	(p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) \|\|
4520	(!p->mm && param->sched_priority > MAX_RT_PRIO-1))
4521	return -EINVAL;
4522	if (rt_policy(policy) != (param->sched_priority != 0))
4523	return -EINVAL;
4524
4525	/*
4526	* Allow unprivileged RT tasks to decrease priority:
4527	*/
4528	if (user && !capable(CAP_SYS_NICE)) {
4529	if (rt_policy(policy)) {
4530	unsigned long rlim_rtprio;
4531
4532	if (!lock_task_sighand(p, &flags))
4533	return -ESRCH;
4534	rlim_rtprio = task_rlimit(p, RLIMIT_RTPRIO);
4535	unlock_task_sighand(p, &flags);
4536
4537	/* can't set/change the rt policy */
4538	if (policy != p->policy && !rlim_rtprio)
4539	return -EPERM;
4540
4541	/* can't increase priority */
4542	if (param->sched_priority > p->rt_priority &&
4543	param->sched_priority > rlim_rtprio)
4544	return -EPERM;
4545	}
4546	/*
4547	* Like positive nice levels, dont allow tasks to
4548	* move out of SCHED_IDLE either:
4549	*/
4550	if (p->policy == SCHED_IDLE && policy != SCHED_IDLE)
4551	return -EPERM;
4552
4553	/* can't change other user's priorities */
4554	if (!check_same_owner(p))
4555	return -EPERM;
4556
4557	/* Normal users shall not reset the sched_reset_on_fork flag */
4558	if (p->sched_reset_on_fork && !reset_on_fork)
4559	return -EPERM;
4560	}
4561
4562	if (user) {
4563	#ifdef CONFIG_RT_GROUP_SCHED
4564	/*
4565	* Do not allow realtime tasks into groups that have no runtime
4566	* assigned.
4567	*/
4568	if (rt_bandwidth_enabled() && rt_policy(policy) &&
4569	task_group(p)->rt_bandwidth.rt_runtime == 0)
4570	return -EPERM;
4571	#endif
4572
4573	retval = security_task_setscheduler(p, policy, param);
4574	if (retval)
4575	return retval;
4576	}
4577
4578	/*
4579	* make sure no PI-waiters arrive (or leave) while we are
4580	* changing the priority of the task:
4581	*/
4582	raw_spin_lock_irqsave(&p->pi_lock, flags);
4583	/*
4584	* To be able to change p->policy safely, the apropriate
4585	* runqueue lock must be held.
4586	*/
4587	rq = __task_rq_lock(p);
4588	/* recheck policy now with rq lock held */
4589	if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) {
4590	policy = oldpolicy = -1;
4591	__task_rq_unlock(rq);
4592	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
4593	goto recheck;
4594	}
4595	update_rq_clock(rq);
4596	on_rq = p->se.on_rq;
4597	running = task_current(rq, p);
4598	if (on_rq)
4599	deactivate_task(rq, p, 0);
4600	if (running)
4601	p->sched_class->put_prev_task(rq, p);
4602
4603	p->sched_reset_on_fork = reset_on_fork;
4604
4605	oldprio = p->prio;
4606	prev_class = p->sched_class;
4607	__setscheduler(rq, p, policy, param->sched_priority);
4608
4609	if (running)
4610	p->sched_class->set_curr_task(rq);
4611	if (on_rq) {
4612	activate_task(rq, p, 0);
4613
4614	check_class_changed(rq, p, prev_class, oldprio, running);
4615	}
4616	__task_rq_unlock(rq);
4617	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
4618
4619	rt_mutex_adjust_pi(p);
4620
4621	return 0;
4622	}
4623
4624	/**
4625	* sched_setscheduler - change the scheduling policy and/or RT priority of a thread.
4626	* @p: the task in question.
4627	* @policy: new policy.
4628	* @param: structure containing the new RT priority.
4629	*
4630	* NOTE that the task may be already dead.
4631	*/
4632	int sched_setscheduler(struct task_struct *p, int policy,
4633	struct sched_param *param)
4634	{
4635	return __sched_setscheduler(p, policy, param, true);
4636	}
4637	EXPORT_SYMBOL_GPL(sched_setscheduler);
4638
4639	/**
4640	* sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace.
4641	* @p: the task in question.
4642	* @policy: new policy.
4643	* @param: structure containing the new RT priority.
4644	*
4645	* Just like sched_setscheduler, only don't bother checking if the
4646	* current context has permission. For example, this is needed in
4647	* stop_machine(): we create temporary high priority worker threads,
4648	* but our caller might not have that capability.
4649	*/
4650	int sched_setscheduler_nocheck(struct task_struct *p, int policy,
4651	struct sched_param *param)
4652	{
4653	return __sched_setscheduler(p, policy, param, false);
4654	}
4655
4656	static int
4657	do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param)
4658	{
4659	struct sched_param lparam;
4660	struct task_struct *p;
4661	int retval;
4662
4663	if (!param \|\| pid < 0)
4664	return -EINVAL;
4665	if (copy_from_user(&lparam, param, sizeof(struct sched_param)))
4666	return -EFAULT;
4667
4668	rcu_read_lock();
4669	retval = -ESRCH;
4670	p = find_process_by_pid(pid);
4671	if (p != NULL)
4672	retval = sched_setscheduler(p, policy, &lparam);
4673	rcu_read_unlock();
4674
4675	return retval;
4676	}
4677
4678	/**
4679	* sys_sched_setscheduler - set/change the scheduler policy and RT priority
4680	* @pid: the pid in question.
4681	* @policy: new policy.
4682	* @param: structure containing the new RT priority.
4683	*/
4684	SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy,
4685	struct sched_param __user *, param)
4686	{
4687	/* negative values for policy are not valid */
4688	if (policy < 0)
4689	return -EINVAL;
4690
4691	return do_sched_setscheduler(pid, policy, param);
4692	}
4693
4694	/**
4695	* sys_sched_setparam - set/change the RT priority of a thread
4696	* @pid: the pid in question.
4697	* @param: structure containing the new RT priority.
4698	*/
4699	SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param)
4700	{
4701	return do_sched_setscheduler(pid, -1, param);
4702	}
4703
4704	/**
4705	* sys_sched_getscheduler - get the policy (scheduling class) of a thread
4706	* @pid: the pid in question.
4707	*/
4708	SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid)
4709	{
4710	struct task_struct *p;
4711	int retval;
4712
4713	if (pid < 0)
4714	return -EINVAL;
4715
4716	retval = -ESRCH;
4717	rcu_read_lock();
4718	p = find_process_by_pid(pid);
4719	if (p) {
4720	retval = security_task_getscheduler(p);
4721	if (!retval)
4722	retval = p->policy
4723	\| (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0);
4724	}
4725	rcu_read_unlock();
4726	return retval;
4727	}
4728
4729	/**
4730	* sys_sched_getparam - get the RT priority of a thread
4731	* @pid: the pid in question.
4732	* @param: structure containing the RT priority.
4733	*/
4734	SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param)
4735	{
4736	struct sched_param lp;
4737	struct task_struct *p;
4738	int retval;
4739
4740	if (!param \|\| pid < 0)
4741	return -EINVAL;
4742
4743	rcu_read_lock();
4744	p = find_process_by_pid(pid);
4745	retval = -ESRCH;
4746	if (!p)
4747	goto out_unlock;
4748
4749	retval = security_task_getscheduler(p);
4750	if (retval)
4751	goto out_unlock;
4752
4753	lp.sched_priority = p->rt_priority;
4754	rcu_read_unlock();
4755
4756	/*
4757	* This one might sleep, we cannot do it with a spinlock held ...
4758	*/
4759	retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0;
4760
4761	return retval;
4762
4763	out_unlock:
4764	rcu_read_unlock();
4765	return retval;
4766	}
4767
4768	long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)
4769	{
4770	cpumask_var_t cpus_allowed, new_mask;
4771	struct task_struct *p;
4772	int retval;
4773
4774	get_online_cpus();
4775	rcu_read_lock();
4776
4777	p = find_process_by_pid(pid);
4778	if (!p) {
4779	rcu_read_unlock();
4780	put_online_cpus();
4781	return -ESRCH;
4782	}
4783
4784	/* Prevent p going away */
4785	get_task_struct(p);
4786	rcu_read_unlock();
4787
4788	if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) {
4789	retval = -ENOMEM;
4790	goto out_put_task;
4791	}
4792	if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) {
4793	retval = -ENOMEM;
4794	goto out_free_cpus_allowed;
4795	}
4796	retval = -EPERM;
4797	if (!check_same_owner(p) && !capable(CAP_SYS_NICE))
4798	goto out_unlock;
4799
4800	retval = security_task_setscheduler(p, 0, NULL);
4801	if (retval)
4802	goto out_unlock;
4803
4804	cpuset_cpus_allowed(p, cpus_allowed);
4805	cpumask_and(new_mask, in_mask, cpus_allowed);
4806	again:
4807	retval = set_cpus_allowed_ptr(p, new_mask);
4808
4809	if (!retval) {
4810	cpuset_cpus_allowed(p, cpus_allowed);
4811	if (!cpumask_subset(new_mask, cpus_allowed)) {
4812	/*
4813	* We must have raced with a concurrent cpuset
4814	* update. Just reset the cpus_allowed to the
4815	* cpuset's cpus_allowed
4816	*/
4817	cpumask_copy(new_mask, cpus_allowed);
4818	goto again;
4819	}
4820	}
4821	out_unlock:
4822	free_cpumask_var(new_mask);
4823	out_free_cpus_allowed:
4824	free_cpumask_var(cpus_allowed);
4825	out_put_task:
4826	put_task_struct(p);
4827	put_online_cpus();
4828	return retval;
4829	}
4830
4831	static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len,
4832	struct cpumask *new_mask)
4833	{
4834	if (len < cpumask_size())
4835	cpumask_clear(new_mask);
4836	else if (len > cpumask_size())
4837	len = cpumask_size();
4838
4839	return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0;
4840	}
4841
4842	/**
4843	* sys_sched_setaffinity - set the cpu affinity of a process
4844	* @pid: pid of the process
4845	* @len: length in bytes of the bitmask pointed to by user_mask_ptr
4846	* @user_mask_ptr: user-space pointer to the new cpu mask
4847	*/
4848	SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len,
4849	unsigned long __user *, user_mask_ptr)
4850	{
4851	cpumask_var_t new_mask;
4852	int retval;
4853
4854	if (!alloc_cpumask_var(&new_mask, GFP_KERNEL))
4855	return -ENOMEM;
4856
4857	retval = get_user_cpu_mask(user_mask_ptr, len, new_mask);
4858	if (retval == 0)
4859	retval = sched_setaffinity(pid, new_mask);
4860	free_cpumask_var(new_mask);
4861	return retval;
4862	}
4863
4864	long sched_getaffinity(pid_t pid, struct cpumask *mask)
4865	{
4866	struct task_struct *p;
4867	unsigned long flags;
4868	struct rq *rq;
4869	int retval;
4870
4871	get_online_cpus();
4872	rcu_read_lock();
4873
4874	retval = -ESRCH;
4875	p = find_process_by_pid(pid);
4876	if (!p)
4877	goto out_unlock;
4878
4879	retval = security_task_getscheduler(p);
4880	if (retval)
4881	goto out_unlock;
4882
4883	rq = task_rq_lock(p, &flags);
4884	cpumask_and(mask, &p->cpus_allowed, cpu_online_mask);
4885	task_rq_unlock(rq, &flags);
4886
4887	out_unlock:
4888	rcu_read_unlock();
4889	put_online_cpus();
4890
4891	return retval;
4892	}
4893
4894	/**
4895	* sys_sched_getaffinity - get the cpu affinity of a process
4896	* @pid: pid of the process
4897	* @len: length in bytes of the bitmask pointed to by user_mask_ptr
4898	* @user_mask_ptr: user-space pointer to hold the current cpu mask
4899	*/
4900	SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len,
4901	unsigned long __user *, user_mask_ptr)
4902	{
4903	int ret;
4904	cpumask_var_t mask;
4905
4906	if ((len * BITS_PER_BYTE) < nr_cpu_ids)
4907	return -EINVAL;
4908	if (len & (sizeof(unsigned long)-1))
4909	return -EINVAL;
4910
4911	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
4912	return -ENOMEM;
4913
4914	ret = sched_getaffinity(pid, mask);
4915	if (ret == 0) {
4916	size_t retlen = min_t(size_t, len, cpumask_size());
4917
4918	if (copy_to_user(user_mask_ptr, mask, retlen))
4919	ret = -EFAULT;
4920	else
4921	ret = retlen;
4922	}
4923	free_cpumask_var(mask);
4924
4925	return ret;
4926	}
4927
4928	/**
4929	* sys_sched_yield - yield the current processor to other threads.
4930	*
4931	* This function yields the current CPU to other tasks. If there are no
4932	* other threads running on this CPU then this function will return.
4933	*/
4934	SYSCALL_DEFINE0(sched_yield)
4935	{
4936	struct rq *rq = this_rq_lock();
4937
4938	schedstat_inc(rq, yld_count);
4939	current->sched_class->yield_task(rq);
4940
4941	/*
4942	* Since we are going to call schedule() anyway, there's
4943	* no need to preempt or enable interrupts:
4944	*/
4945	__release(rq->lock);
4946	spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
4947	do_raw_spin_unlock(&rq->lock);
4948	preempt_enable_no_resched();
4949
4950	schedule();
4951
4952	return 0;
4953	}
4954
4955	static inline int should_resched(void)
4956	{
4957	return need_resched() && !(preempt_count() & PREEMPT_ACTIVE);
4958	}
4959
4960	static void __cond_resched(void)
4961	{
4962	add_preempt_count(PREEMPT_ACTIVE);
4963	schedule();
4964	sub_preempt_count(PREEMPT_ACTIVE);
4965	}
4966
4967	int __sched _cond_resched(void)
4968	{
4969	if (should_resched()) {
4970	__cond_resched();
4971	return 1;
4972	}
4973	return 0;
4974	}
4975	EXPORT_SYMBOL(_cond_resched);
4976
4977	/*
4978	* __cond_resched_lock() - if a reschedule is pending, drop the given lock,
4979	* call schedule, and on return reacquire the lock.
4980	*
4981	* This works OK both with and without CONFIG_PREEMPT. We do strange low-level
4982	* operations here to prevent schedule() from being called twice (once via
4983	* spin_unlock(), once by hand).
4984	*/
4985	int __cond_resched_lock(spinlock_t *lock)
4986	{
4987	int resched = should_resched();
4988	int ret = 0;
4989
4990	lockdep_assert_held(lock);
4991
4992	if (spin_needbreak(lock) \|\| resched) {
4993	spin_unlock(lock);
4994	if (resched)
4995	__cond_resched();
4996	else
4997	cpu_relax();
4998	ret = 1;
4999	spin_lock(lock);
5000	}
5001	return ret;
5002	}
5003	EXPORT_SYMBOL(__cond_resched_lock);
5004
5005	int __sched __cond_resched_softirq(void)
5006	{
5007	BUG_ON(!in_softirq());
5008
5009	if (should_resched()) {
5010	local_bh_enable();
5011	__cond_resched();
5012	local_bh_disable();
5013	return 1;
5014	}
5015	return 0;
5016	}
5017	EXPORT_SYMBOL(__cond_resched_softirq);
5018
5019	/**
5020	* yield - yield the current processor to other threads.
5021	*
5022	* This is a shortcut for kernel-space yielding - it marks the
5023	* thread runnable and calls sys_sched_yield().
5024	*/
5025	void __sched yield(void)
5026	{
5027	set_current_state(TASK_RUNNING);
5028	sys_sched_yield();
5029	}
5030	EXPORT_SYMBOL(yield);
5031
5032	/*
5033	* This task is about to go to sleep on IO. Increment rq->nr_iowait so
5034	* that process accounting knows that this is a task in IO wait state.
5035	*/
5036	void __sched io_schedule(void)
5037	{
5038	struct rq *rq = raw_rq();
5039
5040	delayacct_blkio_start();
5041	atomic_inc(&rq->nr_iowait);
5042	current->in_iowait = 1;
5043	schedule();
5044	current->in_iowait = 0;
5045	atomic_dec(&rq->nr_iowait);
5046	delayacct_blkio_end();
5047	}
5048	EXPORT_SYMBOL(io_schedule);
5049
5050	long __sched io_schedule_timeout(long timeout)
5051	{
5052	struct rq *rq = raw_rq();
5053	long ret;
5054
5055	delayacct_blkio_start();
5056	atomic_inc(&rq->nr_iowait);
5057	current->in_iowait = 1;
5058	ret = schedule_timeout(timeout);
5059	current->in_iowait = 0;
5060	atomic_dec(&rq->nr_iowait);
5061	delayacct_blkio_end();
5062	return ret;
5063	}
5064
5065	/**
5066	* sys_sched_get_priority_max - return maximum RT priority.
5067	* @policy: scheduling class.
5068	*
5069	* this syscall returns the maximum rt_priority that can be used
5070	* by a given scheduling class.
5071	*/
5072	SYSCALL_DEFINE1(sched_get_priority_max, int, policy)
5073	{
5074	int ret = -EINVAL;
5075
5076	switch (policy) {
5077	case SCHED_FIFO:
5078	case SCHED_RR:
5079	ret = MAX_USER_RT_PRIO-1;
5080	break;
5081	case SCHED_NORMAL:
5082	case SCHED_BATCH:
5083	case SCHED_IDLE:
5084	ret = 0;
5085	break;
5086	}
5087	return ret;
5088	}
5089
5090	/**
5091	* sys_sched_get_priority_min - return minimum RT priority.
5092	* @policy: scheduling class.
5093	*
5094	* this syscall returns the minimum rt_priority that can be used
5095	* by a given scheduling class.
5096	*/
5097	SYSCALL_DEFINE1(sched_get_priority_min, int, policy)
5098	{
5099	int ret = -EINVAL;
5100
5101	switch (policy) {
5102	case SCHED_FIFO:
5103	case SCHED_RR:
5104	ret = 1;
5105	break;
5106	case SCHED_NORMAL:
5107	case SCHED_BATCH:
5108	case SCHED_IDLE:
5109	ret = 0;
5110	}
5111	return ret;
5112	}
5113
5114	/**
5115	* sys_sched_rr_get_interval - return the default timeslice of a process.
5116	* @pid: pid of the process.
5117	* @interval: userspace pointer to the timeslice value.
5118	*
5119	* this syscall writes the default timeslice value of a given process
5120	* into the user-space timespec buffer. A value of '0' means infinity.
5121	*/
5122	SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid,
5123	struct timespec __user *, interval)
5124	{
5125	struct task_struct *p;
5126	unsigned int time_slice;
5127	unsigned long flags;
5128	struct rq *rq;
5129	int retval;
5130	struct timespec t;
5131
5132	if (pid < 0)
5133	return -EINVAL;
5134
5135	retval = -ESRCH;
5136	rcu_read_lock();
5137	p = find_process_by_pid(pid);
5138	if (!p)
5139	goto out_unlock;
5140
5141	retval = security_task_getscheduler(p);
5142	if (retval)
5143	goto out_unlock;
5144
5145	rq = task_rq_lock(p, &flags);
5146	time_slice = p->sched_class->get_rr_interval(rq, p);
5147	task_rq_unlock(rq, &flags);
5148
5149	rcu_read_unlock();
5150	jiffies_to_timespec(time_slice, &t);
5151	retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0;
5152	return retval;
5153
5154	out_unlock:
5155	rcu_read_unlock();
5156	return retval;
5157	}
5158
5159	static const char stat_nam[] = TASK_STATE_TO_CHAR_STR;
5160
5161	void sched_show_task(struct task_struct *p)
5162	{
5163	unsigned long free = 0;
5164	unsigned state;
5165
5166	state = p->state ? __ffs(p->state) + 1 : 0;
5167	printk(KERN_INFO "%-13.13s %c", p->comm,
5168	state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?');
5169	#if BITS_PER_LONG == 32
5170	if (state == TASK_RUNNING)
5171	printk(KERN_CONT " running ");
5172	else
5173	printk(KERN_CONT " %08lx ", thread_saved_pc(p));
5174	#else
5175	if (state == TASK_RUNNING)
5176	printk(KERN_CONT " running task ");
5177	else
5178	printk(KERN_CONT " %016lx ", thread_saved_pc(p));
5179	#endif
5180	#ifdef CONFIG_DEBUG_STACK_USAGE
5181	free = stack_not_used(p);
5182	#endif
5183	printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free,
5184	task_pid_nr(p), task_pid_nr(p->real_parent),
5185	(unsigned long)task_thread_info(p)->flags);
5186
5187	show_stack(p, NULL);
5188	}
5189
5190	void show_state_filter(unsigned long state_filter)
5191	{
5192	struct task_struct g, p;
5193
5194	#if BITS_PER_LONG == 32
5195	printk(KERN_INFO
5196	" task PC stack pid father\n");
5197	#else
5198	printk(KERN_INFO
5199	" task PC stack pid father\n");
5200	#endif
5201	read_lock(&tasklist_lock);
5202	do_each_thread(g, p) {
5203	/*
5204	* reset the NMI-timeout, listing all files on a slow
5205	* console might take alot of time:
5206	*/
5207	touch_nmi_watchdog();
5208	if (!state_filter \|\| (p->state & state_filter))
5209	sched_show_task(p);
5210	} while_each_thread(g, p);
5211
5212	touch_all_softlockup_watchdogs();
5213
5214	#ifdef CONFIG_SCHED_DEBUG
5215	sysrq_sched_debug_show();
5216	#endif
5217	read_unlock(&tasklist_lock);
5218	/*
5219	* Only show locks if all tasks are dumped:
5220	*/
5221	if (!state_filter)
5222	debug_show_all_locks();
5223	}
5224
5225	void __cpuinit init_idle_bootup_task(struct task_struct *idle)
5226	{
5227	idle->sched_class = &idle_sched_class;
5228	}
5229
5230	/**
5231	* init_idle - set up an idle thread for a given CPU
5232	* @idle: task in question
5233	* @cpu: cpu the idle task belongs to
5234	*
5235	* NOTE: this function does not set the idle thread's NEED_RESCHED
5236	* flag, to make booting more robust.
5237	*/
5238	void __cpuinit init_idle(struct task_struct *idle, int cpu)
5239	{
5240	struct rq *rq = cpu_rq(cpu);
5241	unsigned long flags;
5242
5243	raw_spin_lock_irqsave(&rq->lock, flags);
5244
5245	__sched_fork(idle);
5246	idle->state = TASK_RUNNING;
5247	idle->se.exec_start = sched_clock();
5248
5249	cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu));
5250	__set_task_cpu(idle, cpu);
5251
5252	rq->curr = rq->idle = idle;
5253	#if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW)
5254	idle->oncpu = 1;
5255	#endif
5256	raw_spin_unlock_irqrestore(&rq->lock, flags);
5257
5258	/* Set the preempt count _outside_ the spinlocks! */
5259	#if defined(CONFIG_PREEMPT)
5260	task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0);
5261	#else
5262	task_thread_info(idle)->preempt_count = 0;
5263	#endif
5264	/*
5265	* The idle tasks have their own, simple scheduling class:
5266	*/
5267	idle->sched_class = &idle_sched_class;
5268	ftrace_graph_init_task(idle);
5269	}
5270
5271	/*
5272	* In a system that switches off the HZ timer nohz_cpu_mask
5273	* indicates which cpus entered this state. This is used
5274	* in the rcu update to wait only for active cpus. For system
5275	* which do not switch off the HZ timer nohz_cpu_mask should
5276	* always be CPU_BITS_NONE.
5277	*/
5278	cpumask_var_t nohz_cpu_mask;
5279
5280	/*
5281	* Increase the granularity value when there are more CPUs,
5282	* because with more CPUs the 'effective latency' as visible
5283	* to users decreases. But the relationship is not linear,
5284	* so pick a second-best guess by going with the log2 of the
5285	* number of CPUs.
5286	*
5287	* This idea comes from the SD scheduler of Con Kolivas:
5288	*/
5289	static int get_update_sysctl_factor(void)
5290	{
5291	unsigned int cpus = min_t(int, num_online_cpus(), 8);
5292	unsigned int factor;
5293
5294	switch (sysctl_sched_tunable_scaling) {
5295	case SCHED_TUNABLESCALING_NONE:
5296	factor = 1;
5297	break;
5298	case SCHED_TUNABLESCALING_LINEAR:
5299	factor = cpus;
5300	break;
5301	case SCHED_TUNABLESCALING_LOG:
5302	default:
5303	factor = 1 + ilog2(cpus);
5304	break;
5305	}
5306
5307	return factor;
5308	}
5309
5310	static void update_sysctl(void)
5311	{
5312	unsigned int factor = get_update_sysctl_factor();
5313
5314	#define SET_SYSCTL(name) \
5315	(sysctl_##name = (factor) * normalized_sysctl_##name)
5316	SET_SYSCTL(sched_min_granularity);
5317	SET_SYSCTL(sched_latency);
5318	SET_SYSCTL(sched_wakeup_granularity);
5319	SET_SYSCTL(sched_shares_ratelimit);
5320	#undef SET_SYSCTL
5321	}
5322
5323	static inline void sched_init_granularity(void)
5324	{
5325	update_sysctl();
5326	}
5327
5328	#ifdef CONFIG_SMP
5329	/*
5330	* This is how migration works:
5331	*
5332	* 1) we queue a struct migration_req structure in the source CPU's
5333	* runqueue and wake up that CPU's migration thread.
5334	* 2) we down() the locked semaphore => thread blocks.
5335	* 3) migration thread wakes up (implicitly it forces the migrated
5336	* thread off the CPU)
5337	* 4) it gets the migration request and checks whether the migrated
5338	* task is still in the wrong runqueue.
5339	* 5) if it's in the wrong runqueue then the migration thread removes
5340	* it and puts it into the right queue.
5341	* 6) migration thread up()s the semaphore.
5342	* 7) we wake up and the migration is done.
5343	*/
5344
5345	/*
5346	* Change a given task's CPU affinity. Migrate the thread to a
5347	* proper CPU and schedule it away if the CPU it's executing on
5348	* is removed from the allowed bitmask.
5349	*
5350	* NOTE: the caller must have a valid reference to the task, the
5351	* task must not exit() & deallocate itself prematurely. The
5352	* call is not atomic; no spinlocks may be held.
5353	*/
5354	int set_cpus_allowed_ptr(struct task_struct p, const struct cpumask new_mask)
5355	{
5356	struct migration_req req;
5357	unsigned long flags;
5358	struct rq *rq;
5359	int ret = 0;
5360
5361	rq = task_rq_lock(p, &flags);
5362
5363	if (!cpumask_intersects(new_mask, cpu_active_mask)) {
5364	ret = -EINVAL;
5365	goto out;
5366	}
5367
5368	if (unlikely((p->flags & PF_THREAD_BOUND) && p != current &&
5369	!cpumask_equal(&p->cpus_allowed, new_mask))) {
5370	ret = -EINVAL;
5371	goto out;
5372	}
5373
5374	if (p->sched_class->set_cpus_allowed)
5375	p->sched_class->set_cpus_allowed(p, new_mask);
5376	else {
5377	cpumask_copy(&p->cpus_allowed, new_mask);
5378	p->rt.nr_cpus_allowed = cpumask_weight(new_mask);
5379	}
5380
5381	/* Can the task run on the task's current CPU? If so, we're done */
5382	if (cpumask_test_cpu(task_cpu(p), new_mask))
5383	goto out;
5384
5385	if (migrate_task(p, cpumask_any_and(cpu_active_mask, new_mask), &req)) {
5386	/* Need help from migration thread: drop lock and wait. */
5387	struct task_struct *mt = rq->migration_thread;
5388
5389	get_task_struct(mt);
5390	task_rq_unlock(rq, &flags);
5391	wake_up_process(mt);
5392	put_task_struct(mt);
5393	wait_for_completion(&req.done);
5394	tlb_migrate_finish(p->mm);
5395	return 0;
5396	}
5397	out:
5398	task_rq_unlock(rq, &flags);
5399
5400	return ret;
5401	}
5402	EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr);
5403
5404	/*
5405	* Move (not current) task off this cpu, onto dest cpu. We're doing
5406	* this because either it can't run here any more (set_cpus_allowed()
5407	* away from this CPU, or CPU going down), or because we're
5408	* attempting to rebalance this task on exec (sched_exec).
5409	*
5410	* So we race with normal scheduler movements, but that's OK, as long
5411	* as the task is no longer on this CPU.
5412	*
5413	* Returns non-zero if task was successfully migrated.
5414	*/
5415	static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu)
5416	{
5417	struct rq rq_dest, rq_src;
5418	int ret = 0;
5419
5420	if (unlikely(!cpu_active(dest_cpu)))
5421	return ret;
5422
5423	rq_src = cpu_rq(src_cpu);
5424	rq_dest = cpu_rq(dest_cpu);
5425
5426	double_rq_lock(rq_src, rq_dest);
5427	/* Already moved. */
5428	if (task_cpu(p) != src_cpu)
5429	goto done;
5430	/* Affinity changed (again). */
5431	if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed))
5432	goto fail;
5433
5434	/*
5435	* If we're not on a rq, the next wake-up will ensure we're
5436	* placed properly.
5437	*/
5438	if (p->se.on_rq) {
5439	deactivate_task(rq_src, p, 0);
5440	set_task_cpu(p, dest_cpu);
5441	activate_task(rq_dest, p, 0);
5442	check_preempt_curr(rq_dest, p, 0);
5443	}
5444	done:
5445	ret = 1;
5446	fail:
5447	double_rq_unlock(rq_src, rq_dest);
5448	return ret;
5449	}
5450
5451	#define RCU_MIGRATION_IDLE 0
5452	#define RCU_MIGRATION_NEED_QS 1
5453	#define RCU_MIGRATION_GOT_QS 2
5454	#define RCU_MIGRATION_MUST_SYNC 3
5455
5456	/*
5457	* migration_thread - this is a highprio system thread that performs
5458	* thread migration by bumping thread off CPU then 'pushing' onto
5459	* another runqueue.
5460	*/
5461	static int migration_thread(void *data)
5462	{
5463	int badcpu;
5464	int cpu = (long)data;
5465	struct rq *rq;
5466
5467	rq = cpu_rq(cpu);
5468	BUG_ON(rq->migration_thread != current);
5469
5470	set_current_state(TASK_INTERRUPTIBLE);
5471	while (!kthread_should_stop()) {
5472	struct migration_req *req;
5473	struct list_head *head;
5474
5475	raw_spin_lock_irq(&rq->lock);
5476
5477	if (cpu_is_offline(cpu)) {
5478	raw_spin_unlock_irq(&rq->lock);
5479	break;
5480	}
5481
5482	if (rq->active_balance) {
5483	active_load_balance(rq, cpu);
5484	rq->active_balance = 0;
5485	}
5486
5487	head = &rq->migration_queue;
5488
5489	if (list_empty(head)) {
5490	raw_spin_unlock_irq(&rq->lock);
5491	schedule();
5492	set_current_state(TASK_INTERRUPTIBLE);
5493	continue;
5494	}
5495	req = list_entry(head->next, struct migration_req, list);
5496	list_del_init(head->next);
5497
5498	if (req->task != NULL) {
5499	raw_spin_unlock(&rq->lock);
5500	__migrate_task(req->task, cpu, req->dest_cpu);
5501	} else if (likely(cpu == (badcpu = smp_processor_id()))) {
5502	req->dest_cpu = RCU_MIGRATION_GOT_QS;
5503	raw_spin_unlock(&rq->lock);
5504	} else {
5505	req->dest_cpu = RCU_MIGRATION_MUST_SYNC;
5506	raw_spin_unlock(&rq->lock);
5507	WARN_ONCE(1, "migration_thread() on CPU %d, expected %d\n", badcpu, cpu);
5508	}
5509	local_irq_enable();
5510
5511	complete(&req->done);
5512	}
5513	__set_current_state(TASK_RUNNING);
5514
5515	return 0;
5516	}
5517
5518	#ifdef CONFIG_HOTPLUG_CPU
5519
5520	static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu)
5521	{
5522	int ret;
5523
5524	local_irq_disable();
5525	ret = __migrate_task(p, src_cpu, dest_cpu);
5526	local_irq_enable();
5527	return ret;
5528	}
5529
5530	/*
5531	* Figure out where task on dead CPU should go, use force if necessary.
5532	*/
5533	static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p)
5534	{
5535	int dest_cpu;
5536
5537	again:
5538	dest_cpu = select_fallback_rq(dead_cpu, p);
5539
5540	/* It can have affinity changed while we were choosing. */
5541	if (unlikely(!__migrate_task_irq(p, dead_cpu, dest_cpu)))
5542	goto again;
5543	}
5544
5545	/*
5546	* While a dead CPU has no uninterruptible tasks queued at this point,
5547	* it might still have a nonzero ->nr_uninterruptible counter, because
5548	* for performance reasons the counter is not stricly tracking tasks to
5549	* their home CPUs. So we just add the counter to another CPU's counter,
5550	* to keep the global sum constant after CPU-down:
5551	*/
5552	static void migrate_nr_uninterruptible(struct rq *rq_src)
5553	{
5554	struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask));
5555	unsigned long flags;
5556
5557	local_irq_save(flags);
5558	double_rq_lock(rq_src, rq_dest);
5559	rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible;
5560	rq_src->nr_uninterruptible = 0;
5561	double_rq_unlock(rq_src, rq_dest);
5562	local_irq_restore(flags);
5563	}
5564
5565	/* Run through task list and migrate tasks from the dead cpu. */
5566	static void migrate_live_tasks(int src_cpu)
5567	{
5568	struct task_struct p, t;
5569
5570	read_lock(&tasklist_lock);
5571
5572	do_each_thread(t, p) {
5573	if (p == current)
5574	continue;
5575
5576	if (task_cpu(p) == src_cpu)
5577	move_task_off_dead_cpu(src_cpu, p);
5578	} while_each_thread(t, p);
5579
5580	read_unlock(&tasklist_lock);
5581	}
5582
5583	/*
5584	* Schedules idle task to be the next runnable task on current CPU.
5585	* It does so by boosting its priority to highest possible.
5586	* Used by CPU offline code.
5587	*/
5588	void sched_idle_next(void)
5589	{
5590	int this_cpu = smp_processor_id();
5591	struct rq *rq = cpu_rq(this_cpu);
5592	struct task_struct *p = rq->idle;
5593	unsigned long flags;
5594
5595	/* cpu has to be offline */
5596	BUG_ON(cpu_online(this_cpu));
5597
5598	/*
5599	* Strictly not necessary since rest of the CPUs are stopped by now
5600	* and interrupts disabled on the current cpu.
5601	*/
5602	raw_spin_lock_irqsave(&rq->lock, flags);
5603
5604	__setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1);
5605
5606	update_rq_clock(rq);
5607	activate_task(rq, p, 0);
5608
5609	raw_spin_unlock_irqrestore(&rq->lock, flags);
5610	}
5611
5612	/*
5613	* Ensures that the idle task is using init_mm right before its cpu goes
5614	* offline.
5615	*/
5616	void idle_task_exit(void)
5617	{
5618	struct mm_struct *mm = current->active_mm;
5619
5620	BUG_ON(cpu_online(smp_processor_id()));
5621
5622	if (mm != &init_mm)
5623	switch_mm(mm, &init_mm, current);
5624	mmdrop(mm);
5625	}
5626
5627	/* called under rq->lock with disabled interrupts */
5628	static void migrate_dead(unsigned int dead_cpu, struct task_struct *p)
5629	{
5630	struct rq *rq = cpu_rq(dead_cpu);
5631
5632	/* Must be exiting, otherwise would be on tasklist. */
5633	BUG_ON(!p->exit_state);
5634
5635	/* Cannot have done final schedule yet: would have vanished. */
5636	BUG_ON(p->state == TASK_DEAD);
5637
5638	get_task_struct(p);
5639
5640	/*
5641	* Drop lock around migration; if someone else moves it,
5642	* that's OK. No task can be added to this CPU, so iteration is
5643	* fine.
5644	*/
5645	raw_spin_unlock_irq(&rq->lock);
5646	move_task_off_dead_cpu(dead_cpu, p);
5647	raw_spin_lock_irq(&rq->lock);
5648
5649	put_task_struct(p);
5650	}
5651
5652	/* release_task() removes task from tasklist, so we won't find dead tasks. */
5653	static void migrate_dead_tasks(unsigned int dead_cpu)
5654	{
5655	struct rq *rq = cpu_rq(dead_cpu);
5656	struct task_struct *next;
5657
5658	for ( ; ; ) {
5659	if (!rq->nr_running)
5660	break;
5661	update_rq_clock(rq);
5662	next = pick_next_task(rq);
5663	if (!next)
5664	break;
5665	next->sched_class->put_prev_task(rq, next);
5666	migrate_dead(dead_cpu, next);
5667
5668	}
5669	}
5670
5671	/*
5672	* remove the tasks which were accounted by rq from calc_load_tasks.
5673	*/
5674	static void calc_global_load_remove(struct rq *rq)
5675	{
5676	atomic_long_sub(rq->calc_load_active, &calc_load_tasks);
5677	rq->calc_load_active = 0;
5678	}
5679	#endif /* CONFIG_HOTPLUG_CPU */
5680
5681	#if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
5682
5683	static struct ctl_table sd_ctl_dir[] = {
5684	{
5685	.procname = "sched_domain",
5686	.mode = 0555,
5687	},
5688	{}
5689	};
5690
5691	static struct ctl_table sd_ctl_root[] = {
5692	{
5693	.procname = "kernel",
5694	.mode = 0555,
5695	.child = sd_ctl_dir,
5696	},
5697	{}
5698	};
5699
5700	static struct ctl_table *sd_alloc_ctl_entry(int n)
5701	{
5702	struct ctl_table *entry =
5703	kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL);
5704
5705	return entry;
5706	}
5707
5708	static void sd_free_ctl_entry(struct ctl_table **tablep)
5709	{
5710	struct ctl_table *entry;
5711
5712	/*
5713	* In the intermediate directories, both the child directory and
5714	* procname are dynamically allocated and could fail but the mode
5715	* will always be set. In the lowest directory the names are
5716	* static strings and all have proc handlers.
5717	*/
5718	for (entry = *tablep; entry->mode; entry++) {
5719	if (entry->child)
5720	sd_free_ctl_entry(&entry->child);
5721	if (entry->proc_handler == NULL)
5722	kfree(entry->procname);
5723	}
5724
5725	kfree(*tablep);
5726	*tablep = NULL;
5727	}
5728
5729	static void
5730	set_table_entry(struct ctl_table *entry,
5731	const char procname, void data, int maxlen,
5732	mode_t mode, proc_handler *proc_handler)
5733	{
5734	entry->procname = procname;
5735	entry->data = data;
5736	entry->maxlen = maxlen;
5737	entry->mode = mode;
5738	entry->proc_handler = proc_handler;
5739	}
5740
5741	static struct ctl_table *
5742	sd_alloc_ctl_domain_table(struct sched_domain *sd)
5743	{
5744	struct ctl_table *table = sd_alloc_ctl_entry(13);
5745
5746	if (table == NULL)
5747	return NULL;
5748
5749	set_table_entry(&table[0], "min_interval", &sd->min_interval,
5750	sizeof(long), 0644, proc_doulongvec_minmax);
5751	set_table_entry(&table[1], "max_interval", &sd->max_interval,
5752	sizeof(long), 0644, proc_doulongvec_minmax);
5753	set_table_entry(&table[2], "busy_idx", &sd->busy_idx,
5754	sizeof(int), 0644, proc_dointvec_minmax);
5755	set_table_entry(&table[3], "idle_idx", &sd->idle_idx,
5756	sizeof(int), 0644, proc_dointvec_minmax);
5757	set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx,
5758	sizeof(int), 0644, proc_dointvec_minmax);
5759	set_table_entry(&table[5], "wake_idx", &sd->wake_idx,
5760	sizeof(int), 0644, proc_dointvec_minmax);
5761	set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx,
5762	sizeof(int), 0644, proc_dointvec_minmax);
5763	set_table_entry(&table[7], "busy_factor", &sd->busy_factor,
5764	sizeof(int), 0644, proc_dointvec_minmax);
5765	set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct,
5766	sizeof(int), 0644, proc_dointvec_minmax);
5767	set_table_entry(&table[9], "cache_nice_tries",
5768	&sd->cache_nice_tries,
5769	sizeof(int), 0644, proc_dointvec_minmax);
5770	set_table_entry(&table[10], "flags", &sd->flags,
5771	sizeof(int), 0644, proc_dointvec_minmax);
5772	set_table_entry(&table[11], "name", sd->name,
5773	CORENAME_MAX_SIZE, 0444, proc_dostring);
5774	/* &table[12] is terminator */
5775
5776	return table;
5777	}
5778
5779	static ctl_table *sd_alloc_ctl_cpu_table(int cpu)
5780	{
5781	struct ctl_table entry, table;
5782	struct sched_domain *sd;
5783	int domain_num = 0, i;
5784	char buf[32];
5785
5786	for_each_domain(cpu, sd)
5787	domain_num++;
5788	entry = table = sd_alloc_ctl_entry(domain_num + 1);
5789	if (table == NULL)
5790	return NULL;
5791
5792	i = 0;
5793	for_each_domain(cpu, sd) {
5794	snprintf(buf, 32, "domain%d", i);
5795	entry->procname = kstrdup(buf, GFP_KERNEL);
5796	entry->mode = 0555;
5797	entry->child = sd_alloc_ctl_domain_table(sd);
5798	entry++;
5799	i++;
5800	}
5801	return table;
5802	}
5803
5804	static struct ctl_table_header *sd_sysctl_header;
5805	static void register_sched_domain_sysctl(void)
5806	{
5807	int i, cpu_num = num_possible_cpus();
5808	struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1);
5809	char buf[32];
5810
5811	WARN_ON(sd_ctl_dir[0].child);
5812	sd_ctl_dir[0].child = entry;
5813
5814	if (entry == NULL)
5815	return;
5816
5817	for_each_possible_cpu(i) {
5818	snprintf(buf, 32, "cpu%d", i);
5819	entry->procname = kstrdup(buf, GFP_KERNEL);
5820	entry->mode = 0555;
5821	entry->child = sd_alloc_ctl_cpu_table(i);
5822	entry++;
5823	}
5824
5825	WARN_ON(sd_sysctl_header);
5826	sd_sysctl_header = register_sysctl_table(sd_ctl_root);
5827	}
5828
5829	/* may be called multiple times per register */
5830	static void unregister_sched_domain_sysctl(void)
5831	{
5832	if (sd_sysctl_header)
5833	unregister_sysctl_table(sd_sysctl_header);
5834	sd_sysctl_header = NULL;
5835	if (sd_ctl_dir[0].child)
5836	sd_free_ctl_entry(&sd_ctl_dir[0].child);
5837	}
5838	#else
5839	static void register_sched_domain_sysctl(void)
5840	{
5841	}
5842	static void unregister_sched_domain_sysctl(void)
5843	{
5844	}
5845	#endif
5846
5847	static void set_rq_online(struct rq *rq)
5848	{
5849	if (!rq->online) {
5850	const struct sched_class *class;
5851
5852	cpumask_set_cpu(rq->cpu, rq->rd->online);
5853	rq->online = 1;
5854
5855	for_each_class(class) {
5856	if (class->rq_online)
5857	class->rq_online(rq);
5858	}
5859	}
5860	}
5861
5862	static void set_rq_offline(struct rq *rq)
5863	{
5864	if (rq->online) {
5865	const struct sched_class *class;
5866
5867	for_each_class(class) {
5868	if (class->rq_offline)
5869	class->rq_offline(rq);
5870	}
5871
5872	cpumask_clear_cpu(rq->cpu, rq->rd->online);
5873	rq->online = 0;
5874	}
5875	}
5876
5877	/*
5878	* migration_call - callback that gets triggered when a CPU is added.
5879	* Here we can start up the necessary migration thread for the new CPU.
5880	*/
5881	static int __cpuinit
5882	migration_call(struct notifier_block nfb, unsigned long action, void hcpu)
5883	{
5884	struct task_struct *p;
5885	int cpu = (long)hcpu;
5886	unsigned long flags;
5887	struct rq *rq;
5888
5889	switch (action) {
5890
5891	case CPU_UP_PREPARE:
5892	case CPU_UP_PREPARE_FROZEN:
5893	p = kthread_create(migration_thread, hcpu, "migration/%d", cpu);
5894	if (IS_ERR(p))
5895	return NOTIFY_BAD;
5896	kthread_bind(p, cpu);
5897	/* Must be high prio: stop_machine expects to yield to it. */
5898	rq = task_rq_lock(p, &flags);
5899	__setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1);
5900	task_rq_unlock(rq, &flags);
5901	get_task_struct(p);
5902	cpu_rq(cpu)->migration_thread = p;
5903	rq->calc_load_update = calc_load_update;
5904	break;
5905
5906	case CPU_ONLINE:
5907	case CPU_ONLINE_FROZEN:
5908	/* Strictly unnecessary, as first user will wake it. */
5909	wake_up_process(cpu_rq(cpu)->migration_thread);
5910
5911	/* Update our root-domain */
5912	rq = cpu_rq(cpu);
5913	raw_spin_lock_irqsave(&rq->lock, flags);
5914	if (rq->rd) {
5915	BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
5916
5917	set_rq_online(rq);
5918	}
5919	raw_spin_unlock_irqrestore(&rq->lock, flags);
5920	break;
5921
5922	#ifdef CONFIG_HOTPLUG_CPU
5923	case CPU_UP_CANCELED:
5924	case CPU_UP_CANCELED_FROZEN:
5925	if (!cpu_rq(cpu)->migration_thread)
5926	break;
5927	/* Unbind it from offline cpu so it can run. Fall thru. */
5928	kthread_bind(cpu_rq(cpu)->migration_thread,
5929	cpumask_any(cpu_online_mask));
5930	kthread_stop(cpu_rq(cpu)->migration_thread);
5931	put_task_struct(cpu_rq(cpu)->migration_thread);
5932	cpu_rq(cpu)->migration_thread = NULL;
5933	break;
5934
5935	case CPU_DEAD:
5936	case CPU_DEAD_FROZEN:
5937	cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */
5938	migrate_live_tasks(cpu);
5939	rq = cpu_rq(cpu);
5940	kthread_stop(rq->migration_thread);
5941	put_task_struct(rq->migration_thread);
5942	rq->migration_thread = NULL;
5943	/* Idle task back to normal (off runqueue, low prio) */
5944	raw_spin_lock_irq(&rq->lock);
5945	update_rq_clock(rq);
5946	deactivate_task(rq, rq->idle, 0);
5947	__setscheduler(rq, rq->idle, SCHED_NORMAL, 0);
5948	rq->idle->sched_class = &idle_sched_class;
5949	migrate_dead_tasks(cpu);
5950	raw_spin_unlock_irq(&rq->lock);
5951	cpuset_unlock();
5952	migrate_nr_uninterruptible(rq);
5953	BUG_ON(rq->nr_running != 0);
5954	calc_global_load_remove(rq);
5955	/*
5956	* No need to migrate the tasks: it was best-effort if
5957	* they didn't take sched_hotcpu_mutex. Just wake up
5958	* the requestors.
5959	*/
5960	raw_spin_lock_irq(&rq->lock);
5961	while (!list_empty(&rq->migration_queue)) {
5962	struct migration_req *req;
5963
5964	req = list_entry(rq->migration_queue.next,
5965	struct migration_req, list);
5966	list_del_init(&req->list);
5967	raw_spin_unlock_irq(&rq->lock);
5968	complete(&req->done);
5969	raw_spin_lock_irq(&rq->lock);
5970	}
5971	raw_spin_unlock_irq(&rq->lock);
5972	break;
5973
5974	case CPU_DYING:
5975	case CPU_DYING_FROZEN:
5976	/* Update our root-domain */
5977	rq = cpu_rq(cpu);
5978	raw_spin_lock_irqsave(&rq->lock, flags);
5979	if (rq->rd) {
5980	BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
5981	set_rq_offline(rq);
5982	}
5983	raw_spin_unlock_irqrestore(&rq->lock, flags);
5984	break;
5985	#endif
5986	}
5987	return NOTIFY_OK;
5988	}
5989
5990	/*
5991	* Register at high priority so that task migration (migrate_all_tasks)
5992	* happens before everything else. This has to be lower priority than
5993	* the notifier in the perf_event subsystem, though.
5994	*/
5995	static struct notifier_block __cpuinitdata migration_notifier = {
5996	.notifier_call = migration_call,
5997	.priority = 10
5998	};
5999
6000	static int __init migration_init(void)
6001	{
6002	void cpu = (void )(long)smp_processor_id();
6003	int err;
6004
6005	/* Start one for the boot CPU: */
6006	err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu);
6007	BUG_ON(err == NOTIFY_BAD);
6008	migration_call(&migration_notifier, CPU_ONLINE, cpu);
6009	register_cpu_notifier(&migration_notifier);
6010
6011	return 0;
6012	}
6013	early_initcall(migration_init);
6014	#endif
6015
6016	#ifdef CONFIG_SMP
6017
6018	#ifdef CONFIG_SCHED_DEBUG
6019
6020	static __read_mostly int sched_domain_debug_enabled;
6021
6022	static int __init sched_domain_debug_setup(char *str)
6023	{
6024	sched_domain_debug_enabled = 1;
6025
6026	return 0;
6027	}
6028	early_param("sched_debug", sched_domain_debug_setup);
6029
6030	static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
6031	struct cpumask *groupmask)
6032	{
6033	struct sched_group *group = sd->groups;
6034	char str[256];
6035
6036	cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd));
6037	cpumask_clear(groupmask);
6038
6039	printk(KERN_DEBUG "%*s domain %d: ", level, "", level);
6040
6041	if (!(sd->flags & SD_LOAD_BALANCE)) {
6042	printk("does not load-balance\n");
6043	if (sd->parent)
6044	printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain"
6045	" has parent");
6046	return -1;
6047	}
6048
6049	printk(KERN_CONT "span %s level %s\n", str, sd->name);
6050
6051	if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) {
6052	printk(KERN_ERR "ERROR: domain->span does not contain "
6053	"CPU%d\n", cpu);
6054	}
6055	if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) {
6056	printk(KERN_ERR "ERROR: domain->groups does not contain"
6057	" CPU%d\n", cpu);
6058	}
6059
6060	printk(KERN_DEBUG "%*s groups:", level + 1, "");
6061	do {
6062	if (!group) {
6063	printk("\n");
6064	printk(KERN_ERR "ERROR: group is NULL\n");
6065	break;
6066	}
6067
6068	if (!group->cpu_power) {
6069	printk(KERN_CONT "\n");
6070	printk(KERN_ERR "ERROR: domain->cpu_power not "
6071	"set\n");
6072	break;
6073	}
6074
6075	if (!cpumask_weight(sched_group_cpus(group))) {
6076	printk(KERN_CONT "\n");
6077	printk(KERN_ERR "ERROR: empty group\n");
6078	break;
6079	}
6080
6081	if (cpumask_intersects(groupmask, sched_group_cpus(group))) {
6082	printk(KERN_CONT "\n");
6083	printk(KERN_ERR "ERROR: repeated CPUs\n");
6084	break;
6085	}
6086
6087	cpumask_or(groupmask, groupmask, sched_group_cpus(group));
6088
6089	cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group));
6090
6091	printk(KERN_CONT " %s", str);
6092	if (group->cpu_power != SCHED_LOAD_SCALE) {
6093	printk(KERN_CONT " (cpu_power = %d)",
6094	group->cpu_power);
6095	}
6096
6097	group = group->next;
6098	} while (group != sd->groups);
6099	printk(KERN_CONT "\n");
6100
6101	if (!cpumask_equal(sched_domain_span(sd), groupmask))
6102	printk(KERN_ERR "ERROR: groups don't span domain->span\n");
6103
6104	if (sd->parent &&
6105	!cpumask_subset(groupmask, sched_domain_span(sd->parent)))
6106	printk(KERN_ERR "ERROR: parent span is not a superset "
6107	"of domain->span\n");
6108	return 0;
6109	}
6110
6111	static void sched_domain_debug(struct sched_domain *sd, int cpu)
6112	{
6113	cpumask_var_t groupmask;
6114	int level = 0;
6115
6116	if (!sched_domain_debug_enabled)
6117	return;
6118
6119	if (!sd) {
6120	printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu);
6121	return;
6122	}
6123
6124	printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu);
6125
6126	if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) {
6127	printk(KERN_DEBUG "Cannot load-balance (out of memory)\n");
6128	return;
6129	}
6130
6131	for (;;) {
6132	if (sched_domain_debug_one(sd, cpu, level, groupmask))
6133	break;
6134	level++;
6135	sd = sd->parent;
6136	if (!sd)
6137	break;
6138	}
6139	free_cpumask_var(groupmask);
6140	}
6141	#else /* !CONFIG_SCHED_DEBUG */
6142	# define sched_domain_debug(sd, cpu) do { } while (0)
6143	#endif /* CONFIG_SCHED_DEBUG */
6144
6145	static int sd_degenerate(struct sched_domain *sd)
6146	{
6147	if (cpumask_weight(sched_domain_span(sd)) == 1)
6148	return 1;
6149
6150	/* Following flags need at least 2 groups */
6151	if (sd->flags & (SD_LOAD_BALANCE \|
6152	SD_BALANCE_NEWIDLE \|
6153	SD_BALANCE_FORK \|
6154	SD_BALANCE_EXEC \|
6155	SD_SHARE_CPUPOWER \|
6156	SD_SHARE_PKG_RESOURCES)) {
6157	if (sd->groups != sd->groups->next)
6158	return 0;
6159	}
6160
6161	/* Following flags don't use groups */
6162	if (sd->flags & (SD_WAKE_AFFINE))
6163	return 0;
6164
6165	return 1;
6166	}
6167
6168	static int
6169	sd_parent_degenerate(struct sched_domain sd, struct sched_domain parent)
6170	{
6171	unsigned long cflags = sd->flags, pflags = parent->flags;
6172
6173	if (sd_degenerate(parent))
6174	return 1;
6175
6176	if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent)))
6177	return 0;
6178
6179	/* Flags needing groups don't count if only 1 group in parent */
6180	if (parent->groups == parent->groups->next) {
6181	pflags &= ~(SD_LOAD_BALANCE \|
6182	SD_BALANCE_NEWIDLE \|
6183	SD_BALANCE_FORK \|
6184	SD_BALANCE_EXEC \|
6185	SD_SHARE_CPUPOWER \|
6186	SD_SHARE_PKG_RESOURCES);
6187	if (nr_node_ids == 1)
6188	pflags &= ~SD_SERIALIZE;
6189	}
6190	if (~cflags & pflags)
6191	return 0;
6192
6193	return 1;
6194	}
6195
6196	static void free_rootdomain(struct root_domain *rd)
6197	{
6198	synchronize_sched();
6199
6200	cpupri_cleanup(&rd->cpupri);
6201
6202	free_cpumask_var(rd->rto_mask);
6203	free_cpumask_var(rd->online);
6204	free_cpumask_var(rd->span);
6205	kfree(rd);
6206	}
6207
6208	static void rq_attach_root(struct rq rq, struct root_domain rd)
6209	{
6210	struct root_domain *old_rd = NULL;
6211	unsigned long flags;
6212
6213	raw_spin_lock_irqsave(&rq->lock, flags);
6214
6215	if (rq->rd) {
6216	old_rd = rq->rd;
6217
6218	if (cpumask_test_cpu(rq->cpu, old_rd->online))
6219	set_rq_offline(rq);
6220
6221	cpumask_clear_cpu(rq->cpu, old_rd->span);
6222
6223	/*
6224	* If we dont want to free the old_rt yet then
6225	* set old_rd to NULL to skip the freeing later
6226	* in this function:
6227	*/
6228	if (!atomic_dec_and_test(&old_rd->refcount))
6229	old_rd = NULL;
6230	}
6231
6232	atomic_inc(&rd->refcount);
6233	rq->rd = rd;
6234
6235	cpumask_set_cpu(rq->cpu, rd->span);
6236	if (cpumask_test_cpu(rq->cpu, cpu_active_mask))
6237	set_rq_online(rq);
6238
6239	raw_spin_unlock_irqrestore(&rq->lock, flags);
6240
6241	if (old_rd)
6242	free_rootdomain(old_rd);
6243	}
6244
6245	static int init_rootdomain(struct root_domain *rd, bool bootmem)
6246	{
6247	gfp_t gfp = GFP_KERNEL;
6248
6249	memset(rd, 0, sizeof(*rd));
6250
6251	if (bootmem)
6252	gfp = GFP_NOWAIT;
6253
6254	if (!alloc_cpumask_var(&rd->span, gfp))
6255	goto out;
6256	if (!alloc_cpumask_var(&rd->online, gfp))
6257	goto free_span;
6258	if (!alloc_cpumask_var(&rd->rto_mask, gfp))
6259	goto free_online;
6260
6261	if (cpupri_init(&rd->cpupri, bootmem) != 0)
6262	goto free_rto_mask;
6263	return 0;
6264
6265	free_rto_mask:
6266	free_cpumask_var(rd->rto_mask);
6267	free_online:
6268	free_cpumask_var(rd->online);
6269	free_span:
6270	free_cpumask_var(rd->span);
6271	out:
6272	return -ENOMEM;
6273	}
6274
6275	static void init_defrootdomain(void)
6276	{
6277	init_rootdomain(&def_root_domain, true);
6278
6279	atomic_set(&def_root_domain.refcount, 1);
6280	}
6281
6282	static struct root_domain *alloc_rootdomain(void)
6283	{
6284	struct root_domain *rd;
6285
6286	rd = kmalloc(sizeof(*rd), GFP_KERNEL);
6287	if (!rd)
6288	return NULL;
6289
6290	if (init_rootdomain(rd, false) != 0) {
6291	kfree(rd);
6292	return NULL;
6293	}
6294
6295	return rd;
6296	}
6297
6298	/*
6299	* Attach the domain 'sd' to 'cpu' as its base domain. Callers must