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 <asm/mmu_context.h>
36#include <linux/interrupt.h>
37#include <linux/capability.h>
38#include <linux/completion.h>
39#include <linux/kernel_stat.h>
40#include <linux/debug_locks.h>
41#include <linux/perf_event.h>
42#include <linux/security.h>
43#include <linux/notifier.h>
44#include <linux/profile.h>
45#include <linux/freezer.h>
46#include <linux/vmalloc.h>
47#include <linux/blkdev.h>
48#include <linux/delay.h>
49#include <linux/pid_namespace.h>
50#include <linux/smp.h>
51#include <linux/threads.h>
52#include <linux/timer.h>
53#include <linux/rcupdate.h>
54#include <linux/cpu.h>
55#include <linux/cpuset.h>
56#include <linux/percpu.h>
57#include <linux/proc_fs.h>
58#include <linux/seq_file.h>
59#include <linux/stop_machine.h>
60#include <linux/sysctl.h>
61#include <linux/syscalls.h>
62#include <linux/times.h>
63#include <linux/tsacct_kern.h>
64#include <linux/kprobes.h>
65#include <linux/delayacct.h>
66#include <linux/unistd.h>
67#include <linux/pagemap.h>
68#include <linux/hrtimer.h>
69#include <linux/tick.h>
70#include <linux/debugfs.h>
71#include <linux/ctype.h>
72#include <linux/ftrace.h>
73#include <linux/slab.h>
74
75#include <asm/tlb.h>
76#include <asm/irq_regs.h>
77#include <asm/mutex.h>
78
79#include "sched_cpupri.h"
80#include "workqueue_sched.h"
81#include "sched_autogroup.h"
82
83#define CREATE_TRACE_POINTS
84#include <trace/events/sched.h>
85
86/*
87 * Convert user-nice values [ -20 ... 0 ... 19 ]
88 * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ],
89 * and back.
90 */
91#define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20)
92#define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20)
93#define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio)
94
95/*
96 * 'User priority' is the nice value converted to something we
97 * can work with better when scaling various scheduler parameters,
98 * it's a [ 0 ... 39 ] range.
99 */
100#define USER_PRIO(p) ((p)-MAX_RT_PRIO)
101#define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio)
102#define MAX_USER_PRIO (USER_PRIO(MAX_PRIO))
103
104/*
105 * Helpers for converting nanosecond timing to jiffy resolution
106 */
107#define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
108
109#define NICE_0_LOAD SCHED_LOAD_SCALE
110#define NICE_0_SHIFT SCHED_LOAD_SHIFT
111
112/*
113 * These are the 'tuning knobs' of the scheduler:
114 *
115 * default timeslice is 100 msecs (used only for SCHED_RR tasks).
116 * Timeslices get refilled after they expire.
117 */
118#define DEF_TIMESLICE (100 * HZ / 1000)
119
120/*
121 * single value that denotes runtime == period, ie unlimited time.
122 */
123#define RUNTIME_INF ((u64)~0ULL)
124
125static inline int rt_policy(int policy)
126{
127    if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR))
128        return 1;
129    return 0;
130}
131
132static inline int task_has_rt_policy(struct task_struct *p)
133{
134    return rt_policy(p->policy);
135}
136
137/*
138 * This is the priority-queue data structure of the RT scheduling class:
139 */
140struct rt_prio_array {
141    DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
142    struct list_head queue[MAX_RT_PRIO];
143};
144
145struct rt_bandwidth {
146    /* nests inside the rq lock: */
147    raw_spinlock_t rt_runtime_lock;
148    ktime_t rt_period;
149    u64 rt_runtime;
150    struct hrtimer rt_period_timer;
151};
152
153static struct rt_bandwidth def_rt_bandwidth;
154
155static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun);
156
157static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer)
158{
159    struct rt_bandwidth *rt_b =
160        container_of(timer, struct rt_bandwidth, rt_period_timer);
161    ktime_t now;
162    int overrun;
163    int idle = 0;
164
165    for (;;) {
166        now = hrtimer_cb_get_time(timer);
167        overrun = hrtimer_forward(timer, now, rt_b->rt_period);
168
169        if (!overrun)
170            break;
171
172        idle = do_sched_rt_period_timer(rt_b, overrun);
173    }
174
175    return idle ? HRTIMER_NORESTART : HRTIMER_RESTART;
176}
177
178static
179void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime)
180{
181    rt_b->rt_period = ns_to_ktime(period);
182    rt_b->rt_runtime = runtime;
183
184    raw_spin_lock_init(&rt_b->rt_runtime_lock);
185
186    hrtimer_init(&rt_b->rt_period_timer,
187            CLOCK_MONOTONIC, HRTIMER_MODE_REL);
188    rt_b->rt_period_timer.function = sched_rt_period_timer;
189}
190
191static inline int rt_bandwidth_enabled(void)
192{
193    return sysctl_sched_rt_runtime >= 0;
194}
195
196static void start_rt_bandwidth(struct rt_bandwidth *rt_b)
197{
198    ktime_t now;
199
200    if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF)
201        return;
202
203    if (hrtimer_active(&rt_b->rt_period_timer))
204        return;
205
206    raw_spin_lock(&rt_b->rt_runtime_lock);
207    for (;;) {
208        unsigned long delta;
209        ktime_t soft, hard;
210
211        if (hrtimer_active(&rt_b->rt_period_timer))
212            break;
213
214        now = hrtimer_cb_get_time(&rt_b->rt_period_timer);
215        hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period);
216
217        soft = hrtimer_get_softexpires(&rt_b->rt_period_timer);
218        hard = hrtimer_get_expires(&rt_b->rt_period_timer);
219        delta = ktime_to_ns(ktime_sub(hard, soft));
220        __hrtimer_start_range_ns(&rt_b->rt_period_timer, soft, delta,
221                HRTIMER_MODE_ABS_PINNED, 0);
222    }
223    raw_spin_unlock(&rt_b->rt_runtime_lock);
224}
225
226#ifdef CONFIG_RT_GROUP_SCHED
227static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b)
228{
229    hrtimer_cancel(&rt_b->rt_period_timer);
230}
231#endif
232
233/*
234 * sched_domains_mutex serializes calls to init_sched_domains,
235 * detach_destroy_domains and partition_sched_domains.
236 */
237static DEFINE_MUTEX(sched_domains_mutex);
238
239#ifdef CONFIG_CGROUP_SCHED
240
241#include <linux/cgroup.h>
242
243struct cfs_rq;
244
245static LIST_HEAD(task_groups);
246
247/* task group related information */
248struct task_group {
249    struct cgroup_subsys_state css;
250
251#ifdef CONFIG_FAIR_GROUP_SCHED
252    /* schedulable entities of this group on each cpu */
253    struct sched_entity **se;
254    /* runqueue "owned" by this group on each cpu */
255    struct cfs_rq **cfs_rq;
256    unsigned long shares;
257
258    atomic_t load_weight;
259#endif
260
261#ifdef CONFIG_RT_GROUP_SCHED
262    struct sched_rt_entity **rt_se;
263    struct rt_rq **rt_rq;
264
265    struct rt_bandwidth rt_bandwidth;
266#endif
267
268    struct rcu_head rcu;
269    struct list_head list;
270
271    struct task_group *parent;
272    struct list_head siblings;
273    struct list_head children;
274
275#ifdef CONFIG_SCHED_AUTOGROUP
276    struct autogroup *autogroup;
277#endif
278};
279
280/* task_group_lock serializes the addition/removal of task groups */
281static DEFINE_SPINLOCK(task_group_lock);
282
283#ifdef CONFIG_FAIR_GROUP_SCHED
284
285# define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
286
287/*
288 * A weight of 0 or 1 can cause arithmetics problems.
289 * A weight of a cfs_rq is the sum of weights of which entities
290 * are queued on this cfs_rq, so a weight of a entity should not be
291 * too large, so as the shares value of a task group.
292 * (The default weight is 1024 - so there's no practical
293 * limitation from this.)
294 */
295#define MIN_SHARES (1UL << 1)
296#define MAX_SHARES (1UL << 18)
297
298static int root_task_group_load = ROOT_TASK_GROUP_LOAD;
299#endif
300
301/* Default task group.
302 * Every task in system belong to this group at bootup.
303 */
304struct task_group root_task_group;
305
306#endif /* CONFIG_CGROUP_SCHED */
307
308/* CFS-related fields in a runqueue */
309struct cfs_rq {
310    struct load_weight load;
311    unsigned long nr_running;
312
313    u64 exec_clock;
314    u64 min_vruntime;
315#ifndef CONFIG_64BIT
316    u64 min_vruntime_copy;
317#endif
318
319    struct rb_root tasks_timeline;
320    struct rb_node *rb_leftmost;
321
322    struct list_head tasks;
323    struct list_head *balance_iterator;
324
325    /*
326     * 'curr' points to currently running entity on this cfs_rq.
327     * It is set to NULL otherwise (i.e when none are currently running).
328     */
329    struct sched_entity *curr, *next, *last, *skip;
330
331#ifdef CONFIG_SCHED_DEBUG
332    unsigned int nr_spread_over;
333#endif
334
335#ifdef CONFIG_FAIR_GROUP_SCHED
336    struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */
337
338    /*
339     * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
340     * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
341     * (like users, containers etc.)
342     *
343     * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
344     * list is used during load balance.
345     */
346    int on_list;
347    struct list_head leaf_cfs_rq_list;
348    struct task_group *tg; /* group that "owns" this runqueue */
349
350#ifdef CONFIG_SMP
351    /*
352     * the part of load.weight contributed by tasks
353     */
354    unsigned long task_weight;
355
356    /*
357     * h_load = weight * f(tg)
358     *
359     * Where f(tg) is the recursive weight fraction assigned to
360     * this group.
361     */
362    unsigned long h_load;
363
364    /*
365     * Maintaining per-cpu shares distribution for group scheduling
366     *
367     * load_stamp is the last time we updated the load average
368     * load_last is the last time we updated the load average and saw load
369     * load_unacc_exec_time is currently unaccounted execution time
370     */
371    u64 load_avg;
372    u64 load_period;
373    u64 load_stamp, load_last, load_unacc_exec_time;
374
375    unsigned long load_contribution;
376#endif
377#endif
378};
379
380/* Real-Time classes' related field in a runqueue: */
381struct rt_rq {
382    struct rt_prio_array active;
383    unsigned long rt_nr_running;
384#if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
385    struct {
386        int curr; /* highest queued rt task prio */
387#ifdef CONFIG_SMP
388        int next; /* next highest */
389#endif
390    } highest_prio;
391#endif
392#ifdef CONFIG_SMP
393    unsigned long rt_nr_migratory;
394    unsigned long rt_nr_total;
395    int overloaded;
396    struct plist_head pushable_tasks;
397#endif
398    int rt_throttled;
399    u64 rt_time;
400    u64 rt_runtime;
401    /* Nests inside the rq lock: */
402    raw_spinlock_t rt_runtime_lock;
403
404#ifdef CONFIG_RT_GROUP_SCHED
405    unsigned long rt_nr_boosted;
406
407    struct rq *rq;
408    struct list_head leaf_rt_rq_list;
409    struct task_group *tg;
410#endif
411};
412
413#ifdef CONFIG_SMP
414
415/*
416 * We add the notion of a root-domain which will be used to define per-domain
417 * variables. Each exclusive cpuset essentially defines an island domain by
418 * fully partitioning the member cpus from any other cpuset. Whenever a new
419 * exclusive cpuset is created, we also create and attach a new root-domain
420 * object.
421 *
422 */
423struct root_domain {
424    atomic_t refcount;
425    struct rcu_head rcu;
426    cpumask_var_t span;
427    cpumask_var_t online;
428
429    /*
430     * The "RT overload" flag: it gets set if a CPU has more than
431     * one runnable RT task.
432     */
433    cpumask_var_t rto_mask;
434    atomic_t rto_count;
435    struct cpupri cpupri;
436};
437
438/*
439 * By default the system creates a single root-domain with all cpus as
440 * members (mimicking the global state we have today).
441 */
442static struct root_domain def_root_domain;
443
444#endif /* CONFIG_SMP */
445
446/*
447 * This is the main, per-CPU runqueue data structure.
448 *
449 * Locking rule: those places that want to lock multiple runqueues
450 * (such as the load balancing or the thread migration code), lock
451 * acquire operations must be ordered by ascending &runqueue.
452 */
453struct rq {
454    /* runqueue lock: */
455    raw_spinlock_t lock;
456
457    /*
458     * nr_running and cpu_load should be in the same cacheline because
459     * remote CPUs use both these fields when doing load calculation.
460     */
461    unsigned long nr_running;
462    #define CPU_LOAD_IDX_MAX 5
463    unsigned long cpu_load[CPU_LOAD_IDX_MAX];
464    unsigned long last_load_update_tick;
465#ifdef CONFIG_NO_HZ
466    u64 nohz_stamp;
467    unsigned char nohz_balance_kick;
468#endif
469    int skip_clock_update;
470
471    /* capture load from *all* tasks on this cpu: */
472    struct load_weight load;
473    unsigned long nr_load_updates;
474    u64 nr_switches;
475
476    struct cfs_rq cfs;
477    struct rt_rq rt;
478
479#ifdef CONFIG_FAIR_GROUP_SCHED
480    /* list of leaf cfs_rq on this cpu: */
481    struct list_head leaf_cfs_rq_list;
482#endif
483#ifdef CONFIG_RT_GROUP_SCHED
484    struct list_head leaf_rt_rq_list;
485#endif
486
487    /*
488     * This is part of a global counter where only the total sum
489     * over all CPUs matters. A task can increase this counter on
490     * one CPU and if it got migrated afterwards it may decrease
491     * it on another CPU. Always updated under the runqueue lock:
492     */
493    unsigned long nr_uninterruptible;
494
495    struct task_struct *curr, *idle, *stop;
496    unsigned long next_balance;
497    struct mm_struct *prev_mm;
498
499    u64 clock;
500    u64 clock_task;
501
502    atomic_t nr_iowait;
503
504#ifdef CONFIG_SMP
505    struct root_domain *rd;
506    struct sched_domain *sd;
507
508    unsigned long cpu_power;
509
510    unsigned char idle_at_tick;
511    /* For active balancing */
512    int post_schedule;
513    int active_balance;
514    int push_cpu;
515    struct cpu_stop_work active_balance_work;
516    /* cpu of this runqueue: */
517    int cpu;
518    int online;
519
520    unsigned long avg_load_per_task;
521
522    u64 rt_avg;
523    u64 age_stamp;
524    u64 idle_stamp;
525    u64 avg_idle;
526#endif
527
528#ifdef CONFIG_IRQ_TIME_ACCOUNTING
529    u64 prev_irq_time;
530#endif
531
532    /* calc_load related fields */
533    unsigned long calc_load_update;
534    long calc_load_active;
535
536#ifdef CONFIG_SCHED_HRTICK
537#ifdef CONFIG_SMP
538    int hrtick_csd_pending;
539    struct call_single_data hrtick_csd;
540#endif
541    struct hrtimer hrtick_timer;
542#endif
543
544#ifdef CONFIG_SCHEDSTATS
545    /* latency stats */
546    struct sched_info rq_sched_info;
547    unsigned long long rq_cpu_time;
548    /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
549
550    /* sys_sched_yield() stats */
551    unsigned int yld_count;
552
553    /* schedule() stats */
554    unsigned int sched_switch;
555    unsigned int sched_count;
556    unsigned int sched_goidle;
557
558    /* try_to_wake_up() stats */
559    unsigned int ttwu_count;
560    unsigned int ttwu_local;
561#endif
562
563#ifdef CONFIG_SMP
564    struct task_struct *wake_list;
565#endif
566};
567
568static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
569
570
571static void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
572
573static inline int cpu_of(struct rq *rq)
574{
575#ifdef CONFIG_SMP
576    return rq->cpu;
577#else
578    return 0;
579#endif
580}
581
582#define rcu_dereference_check_sched_domain(p) \
583    rcu_dereference_check((p), \
584                  rcu_read_lock_held() || \
585                  lockdep_is_held(&sched_domains_mutex))
586
587/*
588 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
589 * See detach_destroy_domains: synchronize_sched for details.
590 *
591 * The domain tree of any CPU may only be accessed from within
592 * preempt-disabled sections.
593 */
594#define for_each_domain(cpu, __sd) \
595    for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent)
596
597#define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
598#define this_rq() (&__get_cpu_var(runqueues))
599#define task_rq(p) cpu_rq(task_cpu(p))
600#define cpu_curr(cpu) (cpu_rq(cpu)->curr)
601#define raw_rq() (&__raw_get_cpu_var(runqueues))
602
603#ifdef CONFIG_CGROUP_SCHED
604
605/*
606 * Return the group to which this tasks belongs.
607 *
608 * We use task_subsys_state_check() and extend the RCU verification with
609 * pi->lock and rq->lock because cpu_cgroup_attach() holds those locks for each
610 * task it moves into the cgroup. Therefore by holding either of those locks,
611 * we pin the task to the current cgroup.
612 */
613static inline struct task_group *task_group(struct task_struct *p)
614{
615    struct task_group *tg;
616    struct cgroup_subsys_state *css;
617
618    css = task_subsys_state_check(p, cpu_cgroup_subsys_id,
619            lockdep_is_held(&p->pi_lock) ||
620            lockdep_is_held(&task_rq(p)->lock));
621    tg = container_of(css, struct task_group, css);
622
623    return autogroup_task_group(p, tg);
624}
625
626/* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
627static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
628{
629#ifdef CONFIG_FAIR_GROUP_SCHED
630    p->se.cfs_rq = task_group(p)->cfs_rq[cpu];
631    p->se.parent = task_group(p)->se[cpu];
632#endif
633
634#ifdef CONFIG_RT_GROUP_SCHED
635    p->rt.rt_rq = task_group(p)->rt_rq[cpu];
636    p->rt.parent = task_group(p)->rt_se[cpu];
637#endif
638}
639
640#else /* CONFIG_CGROUP_SCHED */
641
642static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
643static inline struct task_group *task_group(struct task_struct *p)
644{
645    return NULL;
646}
647
648#endif /* CONFIG_CGROUP_SCHED */
649
650static void update_rq_clock_task(struct rq *rq, s64 delta);
651
652static void update_rq_clock(struct rq *rq)
653{
654    s64 delta;
655
656    if (rq->skip_clock_update > 0)
657        return;
658
659    delta = sched_clock_cpu(cpu_of(rq)) - rq->clock;
660    rq->clock += delta;
661    update_rq_clock_task(rq, delta);
662}
663
664/*
665 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
666 */
667#ifdef CONFIG_SCHED_DEBUG
668# define const_debug __read_mostly
669#else
670# define const_debug static const
671#endif
672
673/**
674 * runqueue_is_locked - Returns true if the current cpu runqueue is locked
675 * @cpu: the processor in question.
676 *
677 * This interface allows printk to be called with the runqueue lock
678 * held and know whether or not it is OK to wake up the klogd.
679 */
680int runqueue_is_locked(int cpu)
681{
682    return raw_spin_is_locked(&cpu_rq(cpu)->lock);
683}
684
685/*
686 * Debugging: various feature bits
687 */
688
689#define SCHED_FEAT(name, enabled) \
690    __SCHED_FEAT_##name ,
691
692enum {
693#include "sched_features.h"
694};
695
696#undef SCHED_FEAT
697
698#define SCHED_FEAT(name, enabled) \
699    (1UL << __SCHED_FEAT_##name) * enabled |
700
701const_debug unsigned int sysctl_sched_features =
702#include "sched_features.h"
703    0;
704
705#undef SCHED_FEAT
706
707#ifdef CONFIG_SCHED_DEBUG
708#define SCHED_FEAT(name, enabled) \
709    #name ,
710
711static __read_mostly char *sched_feat_names[] = {
712#include "sched_features.h"
713    NULL
714};
715
716#undef SCHED_FEAT
717
718static int sched_feat_show(struct seq_file *m, void *v)
719{
720    int i;
721
722    for (i = 0; sched_feat_names[i]; i++) {
723        if (!(sysctl_sched_features & (1UL << i)))
724            seq_puts(m, "NO_");
725        seq_printf(m, "%s ", sched_feat_names[i]);
726    }
727    seq_puts(m, "\n");
728
729    return 0;
730}
731
732static ssize_t
733sched_feat_write(struct file *filp, const char __user *ubuf,
734        size_t cnt, loff_t *ppos)
735{
736    char buf[64];
737    char *cmp;
738    int neg = 0;
739    int i;
740
741    if (cnt > 63)
742        cnt = 63;
743
744    if (copy_from_user(&buf, ubuf, cnt))
745        return -EFAULT;
746
747    buf[cnt] = 0;
748    cmp = strstrip(buf);
749
750    if (strncmp(cmp, "NO_", 3) == 0) {
751        neg = 1;
752        cmp += 3;
753    }
754
755    for (i = 0; sched_feat_names[i]; i++) {
756        if (strcmp(cmp, sched_feat_names[i]) == 0) {
757            if (neg)
758                sysctl_sched_features &= ~(1UL << i);
759            else
760                sysctl_sched_features |= (1UL << i);
761            break;
762        }
763    }
764
765    if (!sched_feat_names[i])
766        return -EINVAL;
767
768    *ppos += cnt;
769
770    return cnt;
771}
772
773static int sched_feat_open(struct inode *inode, struct file *filp)
774{
775    return single_open(filp, sched_feat_show, NULL);
776}
777
778static const struct file_operations sched_feat_fops = {
779    .open = sched_feat_open,
780    .write = sched_feat_write,
781    .read = seq_read,
782    .llseek = seq_lseek,
783    .release = single_release,
784};
785
786static __init int sched_init_debug(void)
787{
788    debugfs_create_file("sched_features", 0644, NULL, NULL,
789            &sched_feat_fops);
790
791    return 0;
792}
793late_initcall(sched_init_debug);
794
795#endif
796
797#define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
798
799/*
800 * Number of tasks to iterate in a single balance run.
801 * Limited because this is done with IRQs disabled.
802 */
803const_debug unsigned int sysctl_sched_nr_migrate = 32;
804
805/*
806 * period over which we average the RT time consumption, measured
807 * in ms.
808 *
809 * default: 1s
810 */
811const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC;
812
813/*
814 * period over which we measure -rt task cpu usage in us.
815 * default: 1s
816 */
817unsigned int sysctl_sched_rt_period = 1000000;
818
819static __read_mostly int scheduler_running;
820
821/*
822 * part of the period that we allow rt tasks to run in us.
823 * default: 0.95s
824 */
825int sysctl_sched_rt_runtime = 950000;
826
827static inline u64 global_rt_period(void)
828{
829    return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
830}
831
832static inline u64 global_rt_runtime(void)
833{
834    if (sysctl_sched_rt_runtime < 0)
835        return RUNTIME_INF;
836
837    return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
838}
839
840#ifndef prepare_arch_switch
841# define prepare_arch_switch(next) do { } while (0)
842#endif
843#ifndef finish_arch_switch
844# define finish_arch_switch(prev) do { } while (0)
845#endif
846
847static inline int task_current(struct rq *rq, struct task_struct *p)
848{
849    return rq->curr == p;
850}
851
852static inline int task_running(struct rq *rq, struct task_struct *p)
853{
854#ifdef CONFIG_SMP
855    return p->on_cpu;
856#else
857    return task_current(rq, p);
858#endif
859}
860
861#ifndef __ARCH_WANT_UNLOCKED_CTXSW
862static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
863{
864#ifdef CONFIG_SMP
865    /*
866     * We can optimise this out completely for !SMP, because the
867     * SMP rebalancing from interrupt is the only thing that cares
868     * here.
869     */
870    next->on_cpu = 1;
871#endif
872}
873
874static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
875{
876#ifdef CONFIG_SMP
877    /*
878     * After ->on_cpu is cleared, the task can be moved to a different CPU.
879     * We must ensure this doesn't happen until the switch is completely
880     * finished.
881     */
882    smp_wmb();
883    prev->on_cpu = 0;
884#endif
885#ifdef CONFIG_DEBUG_SPINLOCK
886    /* this is a valid case when another task releases the spinlock */
887    rq->lock.owner = current;
888#endif
889    /*
890     * If we are tracking spinlock dependencies then we have to
891     * fix up the runqueue lock - which gets 'carried over' from
892     * prev into current:
893     */
894    spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
895
896    raw_spin_unlock_irq(&rq->lock);
897}
898
899#else /* __ARCH_WANT_UNLOCKED_CTXSW */
900static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
901{
902#ifdef CONFIG_SMP
903    /*
904     * We can optimise this out completely for !SMP, because the
905     * SMP rebalancing from interrupt is the only thing that cares
906     * here.
907     */
908    next->on_cpu = 1;
909#endif
910#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
911    raw_spin_unlock_irq(&rq->lock);
912#else
913    raw_spin_unlock(&rq->lock);
914#endif
915}
916
917static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
918{
919#ifdef CONFIG_SMP
920    /*
921     * After ->on_cpu is cleared, the task can be moved to a different CPU.
922     * We must ensure this doesn't happen until the switch is completely
923     * finished.
924     */
925    smp_wmb();
926    prev->on_cpu = 0;
927#endif
928#ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW
929    local_irq_enable();
930#endif
931}
932#endif /* __ARCH_WANT_UNLOCKED_CTXSW */
933
934/*
935 * __task_rq_lock - lock the rq @p resides on.
936 */
937static inline struct rq *__task_rq_lock(struct task_struct *p)
938    __acquires(rq->lock)
939{
940    struct rq *rq;
941
942    lockdep_assert_held(&p->pi_lock);
943
944    for (;;) {
945        rq = task_rq(p);
946        raw_spin_lock(&rq->lock);
947        if (likely(rq == task_rq(p)))
948            return rq;
949        raw_spin_unlock(&rq->lock);
950    }
951}
952
953/*
954 * task_rq_lock - lock p->pi_lock and lock the rq @p resides on.
955 */
956static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags)
957    __acquires(p->pi_lock)
958    __acquires(rq->lock)
959{
960    struct rq *rq;
961
962    for (;;) {
963        raw_spin_lock_irqsave(&p->pi_lock, *flags);
964        rq = task_rq(p);
965        raw_spin_lock(&rq->lock);
966        if (likely(rq == task_rq(p)))
967            return rq;
968        raw_spin_unlock(&rq->lock);
969        raw_spin_unlock_irqrestore(&p->pi_lock, *flags);
970    }
971}
972
973static void __task_rq_unlock(struct rq *rq)
974    __releases(rq->lock)
975{
976    raw_spin_unlock(&rq->lock);
977}
978
979static inline void
980task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags)
981    __releases(rq->lock)
982    __releases(p->pi_lock)
983{
984    raw_spin_unlock(&rq->lock);
985    raw_spin_unlock_irqrestore(&p->pi_lock, *flags);
986}
987
988/*
989 * this_rq_lock - lock this runqueue and disable interrupts.
990 */
991static struct rq *this_rq_lock(void)
992    __acquires(rq->lock)
993{
994    struct rq *rq;
995
996    local_irq_disable();
997    rq = this_rq();
998    raw_spin_lock(&rq->lock);
999
1000    return rq;
1001}
1002
1003#ifdef CONFIG_SCHED_HRTICK
1004/*
1005 * Use HR-timers to deliver accurate preemption points.
1006 *
1007 * Its all a bit involved since we cannot program an hrt while holding the
1008 * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a
1009 * reschedule event.
1010 *
1011 * When we get rescheduled we reprogram the hrtick_timer outside of the
1012 * rq->lock.
1013 */
1014
1015/*
1016 * Use hrtick when:
1017 * - enabled by features
1018 * - hrtimer is actually high res
1019 */
1020static inline int hrtick_enabled(struct rq *rq)
1021{
1022    if (!sched_feat(HRTICK))
1023        return 0;
1024    if (!cpu_active(cpu_of(rq)))
1025        return 0;
1026    return hrtimer_is_hres_active(&rq->hrtick_timer);
1027}
1028
1029static void hrtick_clear(struct rq *rq)
1030{
1031    if (hrtimer_active(&rq->hrtick_timer))
1032        hrtimer_cancel(&rq->hrtick_timer);
1033}
1034
1035/*
1036 * High-resolution timer tick.
1037 * Runs from hardirq context with interrupts disabled.
1038 */
1039static enum hrtimer_restart hrtick(struct hrtimer *timer)
1040{
1041    struct rq *rq = container_of(timer, struct rq, hrtick_timer);
1042
1043    WARN_ON_ONCE(cpu_of(rq) != smp_processor_id());
1044
1045    raw_spin_lock(&rq->lock);
1046    update_rq_clock(rq);
1047    rq->curr->sched_class->task_tick(rq, rq->curr, 1);
1048    raw_spin_unlock(&rq->lock);
1049
1050    return HRTIMER_NORESTART;
1051}
1052
1053#ifdef CONFIG_SMP
1054/*
1055 * called from hardirq (IPI) context
1056 */
1057static void __hrtick_start(void *arg)
1058{
1059    struct rq *rq = arg;
1060
1061    raw_spin_lock(&rq->lock);
1062    hrtimer_restart(&rq->hrtick_timer);
1063    rq->hrtick_csd_pending = 0;
1064    raw_spin_unlock(&rq->lock);
1065}
1066
1067/*
1068 * Called to set the hrtick timer state.
1069 *
1070 * called with rq->lock held and irqs disabled
1071 */
1072static void hrtick_start(struct rq *rq, u64 delay)
1073{
1074    struct hrtimer *timer = &rq->hrtick_timer;
1075    ktime_t time = ktime_add_ns(timer->base->get_time(), delay);
1076
1077    hrtimer_set_expires(timer, time);
1078
1079    if (rq == this_rq()) {
1080        hrtimer_restart(timer);
1081    } else if (!rq->hrtick_csd_pending) {
1082        __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0);
1083        rq->hrtick_csd_pending = 1;
1084    }
1085}
1086
1087static int
1088hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu)
1089{
1090    int cpu = (int)(long)hcpu;
1091
1092    switch (action) {
1093    case CPU_UP_CANCELED:
1094    case CPU_UP_CANCELED_FROZEN:
1095    case CPU_DOWN_PREPARE:
1096    case CPU_DOWN_PREPARE_FROZEN:
1097    case CPU_DEAD:
1098    case CPU_DEAD_FROZEN:
1099        hrtick_clear(cpu_rq(cpu));
1100        return NOTIFY_OK;
1101    }
1102
1103    return NOTIFY_DONE;
1104}
1105
1106static __init void init_hrtick(void)
1107{
1108    hotcpu_notifier(hotplug_hrtick, 0);
1109}
1110#else
1111/*
1112 * Called to set the hrtick timer state.
1113 *
1114 * called with rq->lock held and irqs disabled
1115 */
1116static void hrtick_start(struct rq *rq, u64 delay)
1117{
1118    __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0,
1119            HRTIMER_MODE_REL_PINNED, 0);
1120}
1121
1122static inline void init_hrtick(void)
1123{
1124}
1125#endif /* CONFIG_SMP */
1126
1127static void init_rq_hrtick(struct rq *rq)
1128{
1129#ifdef CONFIG_SMP
1130    rq->hrtick_csd_pending = 0;
1131
1132    rq->hrtick_csd.flags = 0;
1133    rq->hrtick_csd.func = __hrtick_start;
1134    rq->hrtick_csd.info = rq;
1135#endif
1136
1137    hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1138    rq->hrtick_timer.function = hrtick;
1139}
1140#else /* CONFIG_SCHED_HRTICK */
1141static inline void hrtick_clear(struct rq *rq)
1142{
1143}
1144
1145static inline void init_rq_hrtick(struct rq *rq)
1146{
1147}
1148
1149static inline void init_hrtick(void)
1150{
1151}
1152#endif /* CONFIG_SCHED_HRTICK */
1153
1154/*
1155 * resched_task - mark a task 'to be rescheduled now'.
1156 *
1157 * On UP this means the setting of the need_resched flag, on SMP it
1158 * might also involve a cross-CPU call to trigger the scheduler on
1159 * the target CPU.
1160 */
1161#ifdef CONFIG_SMP
1162
1163#ifndef tsk_is_polling
1164#define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG)
1165#endif
1166
1167static void resched_task(struct task_struct *p)
1168{
1169    int cpu;
1170
1171    assert_raw_spin_locked(&task_rq(p)->lock);
1172
1173    if (test_tsk_need_resched(p))
1174        return;
1175
1176    set_tsk_need_resched(p);
1177
1178    cpu = task_cpu(p);
1179    if (cpu == smp_processor_id())
1180        return;
1181
1182    /* NEED_RESCHED must be visible before we test polling */
1183    smp_mb();
1184    if (!tsk_is_polling(p))
1185        smp_send_reschedule(cpu);
1186}
1187
1188static void resched_cpu(int cpu)
1189{
1190    struct rq *rq = cpu_rq(cpu);
1191    unsigned long flags;
1192
1193    if (!raw_spin_trylock_irqsave(&rq->lock, flags))
1194        return;
1195    resched_task(cpu_curr(cpu));
1196    raw_spin_unlock_irqrestore(&rq->lock, flags);
1197}
1198
1199#ifdef CONFIG_NO_HZ
1200/*
1201 * In the semi idle case, use the nearest busy cpu for migrating timers
1202 * from an idle cpu. This is good for power-savings.
1203 *
1204 * We don't do similar optimization for completely idle system, as
1205 * selecting an idle cpu will add more delays to the timers than intended
1206 * (as that cpu's timer base may not be uptodate wrt jiffies etc).
1207 */
1208int get_nohz_timer_target(void)
1209{
1210    int cpu = smp_processor_id();
1211    int i;
1212    struct sched_domain *sd;
1213
1214    rcu_read_lock();
1215    for_each_domain(cpu, sd) {
1216        for_each_cpu(i, sched_domain_span(sd)) {
1217            if (!idle_cpu(i)) {
1218                cpu = i;
1219                goto unlock;
1220            }
1221        }
1222    }
1223unlock:
1224    rcu_read_unlock();
1225    return cpu;
1226}
1227/*
1228 * When add_timer_on() enqueues a timer into the timer wheel of an
1229 * idle CPU then this timer might expire before the next timer event
1230 * which is scheduled to wake up that CPU. In case of a completely
1231 * idle system the next event might even be infinite time into the
1232 * future. wake_up_idle_cpu() ensures that the CPU is woken up and
1233 * leaves the inner idle loop so the newly added timer is taken into
1234 * account when the CPU goes back to idle and evaluates the timer
1235 * wheel for the next timer event.
1236 */
1237void wake_up_idle_cpu(int cpu)
1238{
1239    struct rq *rq = cpu_rq(cpu);
1240
1241    if (cpu == smp_processor_id())
1242        return;
1243
1244    /*
1245     * This is safe, as this function is called with the timer
1246     * wheel base lock of (cpu) held. When the CPU is on the way
1247     * to idle and has not yet set rq->curr to idle then it will
1248     * be serialized on the timer wheel base lock and take the new
1249     * timer into account automatically.
1250     */
1251    if (rq->curr != rq->idle)
1252        return;
1253
1254    /*
1255     * We can set TIF_RESCHED on the idle task of the other CPU
1256     * lockless. The worst case is that the other CPU runs the
1257     * idle task through an additional NOOP schedule()
1258     */
1259    set_tsk_need_resched(rq->idle);
1260
1261    /* NEED_RESCHED must be visible before we test polling */
1262    smp_mb();
1263    if (!tsk_is_polling(rq->idle))
1264        smp_send_reschedule(cpu);
1265}
1266
1267#endif /* CONFIG_NO_HZ */
1268
1269static u64 sched_avg_period(void)
1270{
1271    return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2;
1272}
1273
1274static void sched_avg_update(struct rq *rq)
1275{
1276    s64 period = sched_avg_period();
1277
1278    while ((s64)(rq->clock - rq->age_stamp) > period) {
1279        /*
1280         * Inline assembly required to prevent the compiler
1281         * optimising this loop into a divmod call.
1282         * See __iter_div_u64_rem() for another example of this.
1283         */
1284        asm("" : "+rm" (rq->age_stamp));
1285        rq->age_stamp += period;
1286        rq->rt_avg /= 2;
1287    }
1288}
1289
1290static void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1291{
1292    rq->rt_avg += rt_delta;
1293    sched_avg_update(rq);
1294}
1295
1296#else /* !CONFIG_SMP */
1297static void resched_task(struct task_struct *p)
1298{
1299    assert_raw_spin_locked(&task_rq(p)->lock);
1300    set_tsk_need_resched(p);
1301}
1302
1303static void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1304{
1305}
1306
1307static void sched_avg_update(struct rq *rq)
1308{
1309}
1310#endif /* CONFIG_SMP */
1311
1312#if BITS_PER_LONG == 32
1313# define WMULT_CONST (~0UL)
1314#else
1315# define WMULT_CONST (1UL << 32)
1316#endif
1317
1318#define WMULT_SHIFT 32
1319
1320/*
1321 * Shift right and round:
1322 */
1323#define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y))
1324
1325/*
1326 * delta *= weight / lw
1327 */
1328static unsigned long
1329calc_delta_mine(unsigned long delta_exec, unsigned long weight,
1330        struct load_weight *lw)
1331{
1332    u64 tmp;
1333
1334    /*
1335     * weight can be less than 2^SCHED_LOAD_RESOLUTION for task group sched
1336     * entities since MIN_SHARES = 2. Treat weight as 1 if less than
1337     * 2^SCHED_LOAD_RESOLUTION.
1338     */
1339    if (likely(weight > (1UL << SCHED_LOAD_RESOLUTION)))
1340        tmp = (u64)delta_exec * scale_load_down(weight);
1341    else
1342        tmp = (u64)delta_exec;
1343
1344    if (!lw->inv_weight) {
1345        unsigned long w = scale_load_down(lw->weight);
1346
1347        if (BITS_PER_LONG > 32 && unlikely(w >= WMULT_CONST))
1348            lw->inv_weight = 1;
1349        else if (unlikely(!w))
1350            lw->inv_weight = WMULT_CONST;
1351        else
1352            lw->inv_weight = WMULT_CONST / w;
1353    }
1354
1355    /*
1356     * Check whether we'd overflow the 64-bit multiplication:
1357     */
1358    if (unlikely(tmp > WMULT_CONST))
1359        tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight,
1360            WMULT_SHIFT/2);
1361    else
1362        tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT);
1363
1364    return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX);
1365}
1366
1367static inline void update_load_add(struct load_weight *lw, unsigned long inc)
1368{
1369    lw->weight += inc;
1370    lw->inv_weight = 0;
1371}
1372
1373static inline void update_load_sub(struct load_weight *lw, unsigned long dec)
1374{
1375    lw->weight -= dec;
1376    lw->inv_weight = 0;
1377}
1378
1379static inline void update_load_set(struct load_weight *lw, unsigned long w)
1380{
1381    lw->weight = w;
1382    lw->inv_weight = 0;
1383}
1384
1385/*
1386 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1387 * of tasks with abnormal "nice" values across CPUs the contribution that
1388 * each task makes to its run queue's load is weighted according to its
1389 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1390 * scaled version of the new time slice allocation that they receive on time
1391 * slice expiry etc.
1392 */
1393
1394#define WEIGHT_IDLEPRIO 3
1395#define WMULT_IDLEPRIO 1431655765
1396
1397/*
1398 * Nice levels are multiplicative, with a gentle 10% change for every
1399 * nice level changed. I.e. when a CPU-bound task goes from nice 0 to
1400 * nice 1, it will get ~10% less CPU time than another CPU-bound task
1401 * that remained on nice 0.
1402 *
1403 * The "10% effect" is relative and cumulative: from _any_ nice level,
1404 * if you go up 1 level, it's -10% CPU usage, if you go down 1 level
1405 * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25.
1406 * If a task goes up by ~10% and another task goes down by ~10% then
1407 * the relative distance between them is ~25%.)
1408 */
1409static const int prio_to_weight[40] = {
1410 /* -20 */ 88761, 71755, 56483, 46273, 36291,
1411 /* -15 */ 29154, 23254, 18705, 14949, 11916,
1412 /* -10 */ 9548, 7620, 6100, 4904, 3906,
1413 /* -5 */ 3121, 2501, 1991, 1586, 1277,
1414 /* 0 */ 1024, 820, 655, 526, 423,
1415 /* 5 */ 335, 272, 215, 172, 137,
1416 /* 10 */ 110, 87, 70, 56, 45,
1417 /* 15 */ 36, 29, 23, 18, 15,
1418};
1419
1420/*
1421 * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated.
1422 *
1423 * In cases where the weight does not change often, we can use the
1424 * precalculated inverse to speed up arithmetics by turning divisions
1425 * into multiplications:
1426 */
1427static const u32 prio_to_wmult[40] = {
1428 /* -20 */ 48388, 59856, 76040, 92818, 118348,
1429 /* -15 */ 147320, 184698, 229616, 287308, 360437,
1430 /* -10 */ 449829, 563644, 704093, 875809, 1099582,
1431 /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326,
1432 /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587,
1433 /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126,
1434 /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717,
1435 /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153,
1436};
1437
1438/* Time spent by the tasks of the cpu accounting group executing in ... */
1439enum cpuacct_stat_index {
1440    CPUACCT_STAT_USER, /* ... user mode */
1441    CPUACCT_STAT_SYSTEM, /* ... kernel mode */
1442
1443    CPUACCT_STAT_NSTATS,
1444};
1445
1446#ifdef CONFIG_CGROUP_CPUACCT
1447static void cpuacct_charge(struct task_struct *tsk, u64 cputime);
1448static void cpuacct_update_stats(struct task_struct *tsk,
1449        enum cpuacct_stat_index idx, cputime_t val);
1450#else
1451static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {}
1452static inline void cpuacct_update_stats(struct task_struct *tsk,
1453        enum cpuacct_stat_index idx, cputime_t val) {}
1454#endif
1455
1456static inline void inc_cpu_load(struct rq *rq, unsigned long load)
1457{
1458    update_load_add(&rq->load, load);
1459}
1460
1461static inline void dec_cpu_load(struct rq *rq, unsigned long load)
1462{
1463    update_load_sub(&rq->load, load);
1464}
1465
1466#if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED)
1467typedef int (*tg_visitor)(struct task_group *, void *);
1468
1469/*
1470 * Iterate the full tree, calling @down when first entering a node and @up when
1471 * leaving it for the final time.
1472 */
1473static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
1474{
1475    struct task_group *parent, *child;
1476    int ret;
1477
1478    rcu_read_lock();
1479    parent = &root_task_group;
1480down:
1481    ret = (*down)(parent, data);
1482    if (ret)
1483        goto out_unlock;
1484    list_for_each_entry_rcu(child, &parent->children, siblings) {
1485        parent = child;
1486        goto down;
1487
1488up:
1489        continue;
1490    }
1491    ret = (*up)(parent, data);
1492    if (ret)
1493        goto out_unlock;
1494
1495    child = parent;
1496    parent = parent->parent;
1497    if (parent)
1498        goto up;
1499out_unlock:
1500    rcu_read_unlock();
1501
1502    return ret;
1503}
1504
1505static int tg_nop(struct task_group *tg, void *data)
1506{
1507    return 0;
1508}
1509#endif
1510
1511#ifdef CONFIG_SMP
1512/* Used instead of source_load when we know the type == 0 */
1513static unsigned long weighted_cpuload(const int cpu)
1514{
1515    return cpu_rq(cpu)->load.weight;
1516}
1517
1518/*
1519 * Return a low guess at the load of a migration-source cpu weighted
1520 * according to the scheduling class and "nice" value.
1521 *
1522 * We want to under-estimate the load of migration sources, to
1523 * balance conservatively.
1524 */
1525static unsigned long source_load(int cpu, int type)
1526{
1527    struct rq *rq = cpu_rq(cpu);
1528    unsigned long total = weighted_cpuload(cpu);
1529
1530    if (type == 0 || !sched_feat(LB_BIAS))
1531        return total;
1532
1533    return min(rq->cpu_load[type-1], total);
1534}
1535
1536/*
1537 * Return a high guess at the load of a migration-target cpu weighted
1538 * according to the scheduling class and "nice" value.
1539 */
1540static unsigned long target_load(int cpu, int type)
1541{
1542    struct rq *rq = cpu_rq(cpu);
1543    unsigned long total = weighted_cpuload(cpu);
1544
1545    if (type == 0 || !sched_feat(LB_BIAS))
1546        return total;
1547
1548    return max(rq->cpu_load[type-1], total);
1549}
1550
1551static unsigned long power_of(int cpu)
1552{
1553    return cpu_rq(cpu)->cpu_power;
1554}
1555
1556static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd);
1557
1558static unsigned long cpu_avg_load_per_task(int cpu)
1559{
1560    struct rq *rq = cpu_rq(cpu);
1561    unsigned long nr_running = ACCESS_ONCE(rq->nr_running);
1562
1563    if (nr_running)
1564        rq->avg_load_per_task = rq->load.weight / nr_running;
1565    else
1566        rq->avg_load_per_task = 0;
1567
1568    return rq->avg_load_per_task;
1569}
1570
1571#ifdef CONFIG_FAIR_GROUP_SCHED
1572
1573/*
1574 * Compute the cpu's hierarchical load factor for each task group.
1575 * This needs to be done in a top-down fashion because the load of a child
1576 * group is a fraction of its parents load.
1577 */
1578static int tg_load_down(struct task_group *tg, void *data)
1579{
1580    unsigned long load;
1581    long cpu = (long)data;
1582
1583    if (!tg->parent) {
1584        load = cpu_rq(cpu)->load.weight;
1585    } else {
1586        load = tg->parent->cfs_rq[cpu]->h_load;
1587        load *= tg->se[cpu]->load.weight;
1588        load /= tg->parent->cfs_rq[cpu]->load.weight + 1;
1589    }
1590
1591    tg->cfs_rq[cpu]->h_load = load;
1592
1593    return 0;
1594}
1595
1596static void update_h_load(long cpu)
1597{
1598    walk_tg_tree(tg_load_down, tg_nop, (void *)cpu);
1599}
1600
1601#endif
1602
1603#ifdef CONFIG_PREEMPT
1604
1605static void double_rq_lock(struct rq *rq1, struct rq *rq2);
1606
1607/*
1608 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1609 * way at the expense of forcing extra atomic operations in all
1610 * invocations. This assures that the double_lock is acquired using the
1611 * same underlying policy as the spinlock_t on this architecture, which
1612 * reduces latency compared to the unfair variant below. However, it
1613 * also adds more overhead and therefore may reduce throughput.
1614 */
1615static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1616    __releases(this_rq->lock)
1617    __acquires(busiest->lock)
1618    __acquires(this_rq->lock)
1619{
1620    raw_spin_unlock(&this_rq->lock);
1621    double_rq_lock(this_rq, busiest);
1622
1623    return 1;
1624}
1625
1626#else
1627/*
1628 * Unfair double_lock_balance: Optimizes throughput at the expense of
1629 * latency by eliminating extra atomic operations when the locks are
1630 * already in proper order on entry. This favors lower cpu-ids and will
1631 * grant the double lock to lower cpus over higher ids under contention,
1632 * regardless of entry order into the function.
1633 */
1634static int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1635    __releases(this_rq->lock)
1636    __acquires(busiest->lock)
1637    __acquires(this_rq->lock)
1638{
1639    int ret = 0;
1640
1641    if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1642        if (busiest < this_rq) {
1643            raw_spin_unlock(&this_rq->lock);
1644            raw_spin_lock(&busiest->lock);
1645            raw_spin_lock_nested(&this_rq->lock,
1646                          SINGLE_DEPTH_NESTING);
1647            ret = 1;
1648        } else
1649            raw_spin_lock_nested(&busiest->lock,
1650                          SINGLE_DEPTH_NESTING);
1651    }
1652    return ret;
1653}
1654
1655#endif /* CONFIG_PREEMPT */
1656
1657/*
1658 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1659 */
1660static int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1661{
1662    if (unlikely(!irqs_disabled())) {
1663        /* printk() doesn't work good under rq->lock */
1664        raw_spin_unlock(&this_rq->lock);
1665        BUG_ON(1);
1666    }
1667
1668    return _double_lock_balance(this_rq, busiest);
1669}
1670
1671static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1672    __releases(busiest->lock)
1673{
1674    raw_spin_unlock(&busiest->lock);
1675    lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1676}
1677
1678/*
1679 * double_rq_lock - safely lock two runqueues
1680 *
1681 * Note this does not disable interrupts like task_rq_lock,
1682 * you need to do so manually before calling.
1683 */
1684static void double_rq_lock(struct rq *rq1, struct rq *rq2)
1685    __acquires(rq1->lock)
1686    __acquires(rq2->lock)
1687{
1688    BUG_ON(!irqs_disabled());
1689    if (rq1 == rq2) {
1690        raw_spin_lock(&rq1->lock);
1691        __acquire(rq2->lock); /* Fake it out ;) */
1692    } else {
1693        if (rq1 < rq2) {
1694            raw_spin_lock(&rq1->lock);
1695            raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1696        } else {
1697            raw_spin_lock(&rq2->lock);
1698            raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1699        }
1700    }
1701}
1702
1703/*
1704 * double_rq_unlock - safely unlock two runqueues
1705 *
1706 * Note this does not restore interrupts like task_rq_unlock,
1707 * you need to do so manually after calling.
1708 */
1709static void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1710    __releases(rq1->lock)
1711    __releases(rq2->lock)
1712{
1713    raw_spin_unlock(&rq1->lock);
1714    if (rq1 != rq2)
1715        raw_spin_unlock(&rq2->lock);
1716    else
1717        __release(rq2->lock);
1718}
1719
1720#else /* CONFIG_SMP */
1721
1722/*
1723 * double_rq_lock - safely lock two runqueues
1724 *
1725 * Note this does not disable interrupts like task_rq_lock,
1726 * you need to do so manually before calling.
1727 */
1728static void double_rq_lock(struct rq *rq1, struct rq *rq2)
1729    __acquires(rq1->lock)
1730    __acquires(rq2->lock)
1731{
1732    BUG_ON(!irqs_disabled());
1733    BUG_ON(rq1 != rq2);
1734    raw_spin_lock(&rq1->lock);
1735    __acquire(rq2->lock); /* Fake it out ;) */
1736}
1737
1738/*
1739 * double_rq_unlock - safely unlock two runqueues
1740 *
1741 * Note this does not restore interrupts like task_rq_unlock,
1742 * you need to do so manually after calling.
1743 */
1744static void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1745    __releases(rq1->lock)
1746    __releases(rq2->lock)
1747{
1748    BUG_ON(rq1 != rq2);
1749    raw_spin_unlock(&rq1->lock);
1750    __release(rq2->lock);
1751}
1752
1753#endif
1754
1755static void calc_load_account_idle(struct rq *this_rq);
1756static void update_sysctl(void);
1757static int get_update_sysctl_factor(void);
1758static void update_cpu_load(struct rq *this_rq);
1759
1760static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1761{
1762    set_task_rq(p, cpu);
1763#ifdef CONFIG_SMP
1764    /*
1765     * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1766     * successfuly executed on another CPU. We must ensure that updates of
1767     * per-task data have been completed by this moment.
1768     */
1769    smp_wmb();
1770    task_thread_info(p)->cpu = cpu;
1771#endif
1772}
1773
1774static const struct sched_class rt_sched_class;
1775
1776#define sched_class_highest (&stop_sched_class)
1777#define for_each_class(class) \
1778   for (class = sched_class_highest; class; class = class->next)
1779
1780#include "sched_stats.h"
1781
1782static void inc_nr_running(struct rq *rq)
1783{
1784    rq->nr_running++;
1785}
1786
1787static void dec_nr_running(struct rq *rq)
1788{
1789    rq->nr_running--;
1790}
1791
1792static void set_load_weight(struct task_struct *p)
1793{
1794    int prio = p->static_prio - MAX_RT_PRIO;
1795    struct load_weight *load = &p->se.load;
1796
1797    /*
1798     * SCHED_IDLE tasks get minimal weight:
1799     */
1800    if (p->policy == SCHED_IDLE) {
1801        load->weight = scale_load(WEIGHT_IDLEPRIO);
1802        load->inv_weight = WMULT_IDLEPRIO;
1803        return;
1804    }
1805
1806    load->weight = scale_load(prio_to_weight[prio]);
1807    load->inv_weight = prio_to_wmult[prio];
1808}
1809
1810static void enqueue_task(struct rq *rq, struct task_struct *p, int flags)
1811{
1812    update_rq_clock(rq);
1813    sched_info_queued(p);
1814    p->sched_class->enqueue_task(rq, p, flags);
1815}
1816
1817static void dequeue_task(struct rq *rq, struct task_struct *p, int flags)
1818{
1819    update_rq_clock(rq);
1820    sched_info_dequeued(p);
1821    p->sched_class->dequeue_task(rq, p, flags);
1822}
1823
1824/*
1825 * activate_task - move a task to the runqueue.
1826 */
1827static void activate_task(struct rq *rq, struct task_struct *p, int flags)
1828{
1829    if (task_contributes_to_load(p))
1830        rq->nr_uninterruptible--;
1831
1832    enqueue_task(rq, p, flags);
1833    inc_nr_running(rq);
1834}
1835
1836/*
1837 * deactivate_task - remove a task from the runqueue.
1838 */
1839static void deactivate_task(struct rq *rq, struct task_struct *p, int flags)
1840{
1841    if (task_contributes_to_load(p))
1842        rq->nr_uninterruptible++;
1843
1844    dequeue_task(rq, p, flags);
1845    dec_nr_running(rq);
1846}
1847
1848#ifdef CONFIG_IRQ_TIME_ACCOUNTING
1849
1850/*
1851 * There are no locks covering percpu hardirq/softirq time.
1852 * They are only modified in account_system_vtime, on corresponding CPU
1853 * with interrupts disabled. So, writes are safe.
1854 * They are read and saved off onto struct rq in update_rq_clock().
1855 * This may result in other CPU reading this CPU's irq time and can
1856 * race with irq/account_system_vtime on this CPU. We would either get old
1857 * or new value with a side effect of accounting a slice of irq time to wrong
1858 * task when irq is in progress while we read rq->clock. That is a worthy
1859 * compromise in place of having locks on each irq in account_system_time.
1860 */
1861static DEFINE_PER_CPU(u64, cpu_hardirq_time);
1862static DEFINE_PER_CPU(u64, cpu_softirq_time);
1863
1864static DEFINE_PER_CPU(u64, irq_start_time);
1865static int sched_clock_irqtime;
1866
1867void enable_sched_clock_irqtime(void)
1868{
1869    sched_clock_irqtime = 1;
1870}
1871
1872void disable_sched_clock_irqtime(void)
1873{
1874    sched_clock_irqtime = 0;
1875}
1876
1877#ifndef CONFIG_64BIT
1878static DEFINE_PER_CPU(seqcount_t, irq_time_seq);
1879
1880static inline void irq_time_write_begin(void)
1881{
1882    __this_cpu_inc(irq_time_seq.sequence);
1883    smp_wmb();
1884}
1885
1886static inline void irq_time_write_end(void)
1887{
1888    smp_wmb();
1889    __this_cpu_inc(irq_time_seq.sequence);
1890}
1891
1892static inline u64 irq_time_read(int cpu)
1893{
1894    u64 irq_time;
1895    unsigned seq;
1896
1897    do {
1898        seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu));
1899        irq_time = per_cpu(cpu_softirq_time, cpu) +
1900               per_cpu(cpu_hardirq_time, cpu);
1901    } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq));
1902
1903    return irq_time;
1904}
1905#else /* CONFIG_64BIT */
1906static inline void irq_time_write_begin(void)
1907{
1908}
1909
1910static inline void irq_time_write_end(void)
1911{
1912}
1913
1914static inline u64 irq_time_read(int cpu)
1915{
1916    return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu);
1917}
1918#endif /* CONFIG_64BIT */
1919
1920/*
1921 * Called before incrementing preempt_count on {soft,}irq_enter
1922 * and before decrementing preempt_count on {soft,}irq_exit.
1923 */
1924void account_system_vtime(struct task_struct *curr)
1925{
1926    unsigned long flags;
1927    s64 delta;
1928    int cpu;
1929
1930    if (!sched_clock_irqtime)
1931        return;
1932
1933    local_irq_save(flags);
1934
1935    cpu = smp_processor_id();
1936    delta = sched_clock_cpu(cpu) - __this_cpu_read(irq_start_time);
1937    __this_cpu_add(irq_start_time, delta);
1938
1939    irq_time_write_begin();
1940    /*
1941     * We do not account for softirq time from ksoftirqd here.
1942     * We want to continue accounting softirq time to ksoftirqd thread
1943     * in that case, so as not to confuse scheduler with a special task
1944     * that do not consume any time, but still wants to run.
1945     */
1946    if (hardirq_count())
1947        __this_cpu_add(cpu_hardirq_time, delta);
1948    else if (in_serving_softirq() && curr != this_cpu_ksoftirqd())
1949        __this_cpu_add(cpu_softirq_time, delta);
1950
1951    irq_time_write_end();
1952    local_irq_restore(flags);
1953}
1954EXPORT_SYMBOL_GPL(account_system_vtime);
1955
1956static void update_rq_clock_task(struct rq *rq, s64 delta)
1957{
1958    s64 irq_delta;
1959
1960    irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time;
1961
1962    /*
1963     * Since irq_time is only updated on {soft,}irq_exit, we might run into
1964     * this case when a previous update_rq_clock() happened inside a
1965     * {soft,}irq region.
1966     *
1967     * When this happens, we stop ->clock_task and only update the
1968     * prev_irq_time stamp to account for the part that fit, so that a next
1969     * update will consume the rest. This ensures ->clock_task is
1970     * monotonic.
1971     *
1972     * It does however cause some slight miss-attribution of {soft,}irq
1973     * time, a more accurate solution would be to update the irq_time using
1974     * the current rq->clock timestamp, except that would require using
1975     * atomic ops.
1976     */
1977    if (irq_delta > delta)
1978        irq_delta = delta;
1979
1980    rq->prev_irq_time += irq_delta;
1981    delta -= irq_delta;
1982    rq->clock_task += delta;
1983
1984    if (irq_delta && sched_feat(NONIRQ_POWER))
1985        sched_rt_avg_update(rq, irq_delta);
1986}
1987
1988static int irqtime_account_hi_update(void)
1989{
1990    struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
1991    unsigned long flags;
1992    u64 latest_ns;
1993    int ret = 0;
1994
1995    local_irq_save(flags);
1996    latest_ns = this_cpu_read(cpu_hardirq_time);
1997    if (cputime64_gt(nsecs_to_cputime64(latest_ns), cpustat->irq))
1998        ret = 1;
1999    local_irq_restore(flags);
2000    return ret;
2001}
2002
2003static int irqtime_account_si_update(void)
2004{
2005    struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
2006    unsigned long flags;
2007    u64 latest_ns;
2008    int ret = 0;
2009
2010    local_irq_save(flags);
2011    latest_ns = this_cpu_read(cpu_softirq_time);
2012    if (cputime64_gt(nsecs_to_cputime64(latest_ns), cpustat->softirq))
2013        ret = 1;
2014    local_irq_restore(flags);
2015    return ret;
2016}
2017
2018#else /* CONFIG_IRQ_TIME_ACCOUNTING */
2019
2020#define sched_clock_irqtime (0)
2021
2022static void update_rq_clock_task(struct rq *rq, s64 delta)
2023{
2024    rq->clock_task += delta;
2025}
2026
2027#endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2028
2029#include "sched_idletask.c"
2030#include "sched_fair.c"
2031#include "sched_rt.c"
2032#include "sched_autogroup.c"
2033#include "sched_stoptask.c"
2034#ifdef CONFIG_SCHED_DEBUG
2035# include "sched_debug.c"
2036#endif
2037
2038void sched_set_stop_task(int cpu, struct task_struct *stop)
2039{
2040    struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 };
2041    struct task_struct *old_stop = cpu_rq(cpu)->stop;
2042
2043    if (stop) {
2044        /*
2045         * Make it appear like a SCHED_FIFO task, its something
2046         * userspace knows about and won't get confused about.
2047         *
2048         * Also, it will make PI more or less work without too
2049         * much confusion -- but then, stop work should not
2050         * rely on PI working anyway.
2051         */
2052        sched_setscheduler_nocheck(stop, SCHED_FIFO, &param);
2053
2054        stop->sched_class = &stop_sched_class;
2055    }
2056
2057    cpu_rq(cpu)->stop = stop;
2058
2059    if (old_stop) {
2060        /*
2061         * Reset it back to a normal scheduling class so that
2062         * it can die in pieces.
2063         */
2064        old_stop->sched_class = &rt_sched_class;
2065    }
2066}
2067
2068/*
2069 * __normal_prio - return the priority that is based on the static prio
2070 */
2071static inline int __normal_prio(struct task_struct *p)
2072{
2073    return p->static_prio;
2074}
2075
2076/*
2077 * Calculate the expected normal priority: i.e. priority
2078 * without taking RT-inheritance into account. Might be
2079 * boosted by interactivity modifiers. Changes upon fork,
2080 * setprio syscalls, and whenever the interactivity
2081 * estimator recalculates.
2082 */
2083static inline int normal_prio(struct task_struct *p)
2084{
2085    int prio;
2086
2087    if (task_has_rt_policy(p))
2088        prio = MAX_RT_PRIO-1 - p->rt_priority;
2089    else
2090        prio = __normal_prio(p);
2091    return prio;
2092}
2093
2094/*
2095 * Calculate the current priority, i.e. the priority
2096 * taken into account by the scheduler. This value might
2097 * be boosted by RT tasks, or might be boosted by
2098 * interactivity modifiers. Will be RT if the task got
2099 * RT-boosted. If not then it returns p->normal_prio.
2100 */
2101static int effective_prio(struct task_struct *p)
2102{
2103    p->normal_prio = normal_prio(p);
2104    /*
2105     * If we are RT tasks or we were boosted to RT priority,
2106     * keep the priority unchanged. Otherwise, update priority
2107     * to the normal priority:
2108     */
2109    if (!rt_prio(p->prio))
2110        return p->normal_prio;
2111    return p->prio;
2112}
2113
2114/**
2115 * task_curr - is this task currently executing on a CPU?
2116 * @p: the task in question.
2117 */
2118inline int task_curr(const struct task_struct *p)
2119{
2120    return cpu_curr(task_cpu(p)) == p;
2121}
2122
2123static inline void check_class_changed(struct rq *rq, struct task_struct *p,
2124                       const struct sched_class *prev_class,
2125                       int oldprio)
2126{
2127    if (prev_class != p->sched_class) {
2128        if (prev_class->switched_from)
2129            prev_class->switched_from(rq, p);
2130        p->sched_class->switched_to(rq, p);
2131    } else if (oldprio != p->prio)
2132        p->sched_class->prio_changed(rq, p, oldprio);
2133}
2134
2135static void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags)
2136{
2137    const struct sched_class *class;
2138
2139    if (p->sched_class == rq->curr->sched_class) {
2140        rq->curr->sched_class->check_preempt_curr(rq, p, flags);
2141    } else {
2142        for_each_class(class) {
2143            if (class == rq->curr->sched_class)
2144                break;
2145            if (class == p->sched_class) {
2146                resched_task(rq->curr);
2147                break;
2148            }
2149        }
2150    }
2151
2152    /*
2153     * A queue event has occurred, and we're going to schedule. In
2154     * this case, we can save a useless back to back clock update.
2155     */
2156    if (rq->curr->on_rq && test_tsk_need_resched(rq->curr))
2157        rq->skip_clock_update = 1;
2158}
2159
2160#ifdef CONFIG_SMP
2161/*
2162 * Is this task likely cache-hot:
2163 */
2164static int
2165task_hot(struct task_struct *p, u64 now, struct sched_domain *sd)
2166{
2167    s64 delta;
2168
2169    if (p->sched_class != &fair_sched_class)
2170        return 0;
2171
2172    if (unlikely(p->policy == SCHED_IDLE))
2173        return 0;
2174
2175    /*
2176     * Buddy candidates are cache hot:
2177     */
2178    if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running &&
2179            (&p->se == cfs_rq_of(&p->se)->next ||
2180             &p->se == cfs_rq_of(&p->se)->last))
2181        return 1;
2182
2183    if (sysctl_sched_migration_cost == -1)
2184        return 1;
2185    if (sysctl_sched_migration_cost == 0)
2186        return 0;
2187
2188    delta = now - p->se.exec_start;
2189
2190    return delta < (s64)sysctl_sched_migration_cost;
2191}
2192
2193void set_task_cpu(struct task_struct *p, unsigned int new_cpu)
2194{
2195#ifdef CONFIG_SCHED_DEBUG
2196    /*
2197     * We should never call set_task_cpu() on a blocked task,
2198     * ttwu() will sort out the placement.
2199     */
2200    WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING &&
2201            !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE));
2202
2203#ifdef CONFIG_LOCKDEP
2204    /*
2205     * The caller should hold either p->pi_lock or rq->lock, when changing
2206     * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks.
2207     *
2208     * sched_move_task() holds both and thus holding either pins the cgroup,
2209     * see set_task_rq().
2210     *
2211     * Furthermore, all task_rq users should acquire both locks, see
2212     * task_rq_lock().
2213     */
2214    WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) ||
2215                      lockdep_is_held(&task_rq(p)->lock)));
2216#endif
2217#endif
2218
2219    trace_sched_migrate_task(p, new_cpu);
2220
2221    if (task_cpu(p) != new_cpu) {
2222        p->se.nr_migrations++;
2223        perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 1, NULL, 0);
2224    }
2225
2226    __set_task_cpu(p, new_cpu);
2227}
2228
2229struct migration_arg {
2230    struct task_struct *task;
2231    int dest_cpu;
2232};
2233
2234static int migration_cpu_stop(void *data);
2235
2236/*
2237 * wait_task_inactive - wait for a thread to unschedule.
2238 *
2239 * If @match_state is nonzero, it's the @p->state value just checked and
2240 * not expected to change. If it changes, i.e. @p might have woken up,
2241 * then return zero. When we succeed in waiting for @p to be off its CPU,
2242 * we return a positive number (its total switch count). If a second call
2243 * a short while later returns the same number, the caller can be sure that
2244 * @p has remained unscheduled the whole time.
2245 *
2246 * The caller must ensure that the task *will* unschedule sometime soon,
2247 * else this function might spin for a *long* time. This function can't
2248 * be called with interrupts off, or it may introduce deadlock with
2249 * smp_call_function() if an IPI is sent by the same process we are
2250 * waiting to become inactive.
2251 */
2252unsigned long wait_task_inactive(struct task_struct *p, long match_state)
2253{
2254    unsigned long flags;
2255    int running, on_rq;
2256    unsigned long ncsw;
2257    struct rq *rq;
2258
2259    for (;;) {
2260        /*
2261         * We do the initial early heuristics without holding
2262         * any task-queue locks at all. We'll only try to get
2263         * the runqueue lock when things look like they will
2264         * work out!
2265         */
2266        rq = task_rq(p);
2267
2268        /*
2269         * If the task is actively running on another CPU
2270         * still, just relax and busy-wait without holding
2271         * any locks.
2272         *
2273         * NOTE! Since we don't hold any locks, it's not
2274         * even sure that "rq" stays as the right runqueue!
2275         * But we don't care, since "task_running()" will
2276         * return false if the runqueue has changed and p
2277         * is actually now running somewhere else!
2278         */
2279        while (task_running(rq, p)) {
2280            if (match_state && unlikely(p->state != match_state))
2281                return 0;
2282            cpu_relax();
2283        }
2284
2285        /*
2286         * Ok, time to look more closely! We need the rq
2287         * lock now, to be *sure*. If we're wrong, we'll
2288         * just go back and repeat.
2289         */
2290        rq = task_rq_lock(p, &flags);
2291        trace_sched_wait_task(p);
2292        running = task_running(rq, p);
2293        on_rq = p->on_rq;
2294        ncsw = 0;
2295        if (!match_state || p->state == match_state)
2296            ncsw = p->nvcsw | LONG_MIN; /* sets MSB */
2297        task_rq_unlock(rq, p, &flags);
2298
2299        /*
2300         * If it changed from the expected state, bail out now.
2301         */
2302        if (unlikely(!ncsw))
2303            break;
2304
2305        /*
2306         * Was it really running after all now that we
2307         * checked with the proper locks actually held?
2308         *
2309         * Oops. Go back and try again..
2310         */
2311        if (unlikely(running)) {
2312            cpu_relax();
2313            continue;
2314        }
2315
2316        /*
2317         * It's not enough that it's not actively running,
2318         * it must be off the runqueue _entirely_, and not
2319         * preempted!
2320         *
2321         * So if it was still runnable (but just not actively
2322         * running right now), it's preempted, and we should
2323         * yield - it could be a while.
2324         */
2325        if (unlikely(on_rq)) {
2326            ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ);
2327
2328            set_current_state(TASK_UNINTERRUPTIBLE);
2329            schedule_hrtimeout(&to, HRTIMER_MODE_REL);
2330            continue;
2331        }
2332
2333        /*
2334         * Ahh, all good. It wasn't running, and it wasn't
2335         * runnable, which means that it will never become
2336         * running in the future either. We're all done!
2337         */
2338        break;
2339    }
2340
2341    return ncsw;
2342}
2343
2344/***
2345 * kick_process - kick a running thread to enter/exit the kernel
2346 * @p: the to-be-kicked thread
2347 *
2348 * Cause a process which is running on another CPU to enter
2349 * kernel-mode, without any delay. (to get signals handled.)
2350 *
2351 * NOTE: this function doesn't have to take the runqueue lock,
2352 * because all it wants to ensure is that the remote task enters
2353 * the kernel. If the IPI races and the task has been migrated
2354 * to another CPU then no harm is done and the purpose has been
2355 * achieved as well.
2356 */
2357void kick_process(struct task_struct *p)
2358{
2359    int cpu;
2360
2361    preempt_disable();
2362    cpu = task_cpu(p);
2363    if ((cpu != smp_processor_id()) && task_curr(p))
2364        smp_send_reschedule(cpu);
2365    preempt_enable();
2366}
2367EXPORT_SYMBOL_GPL(kick_process);
2368#endif /* CONFIG_SMP */
2369
2370#ifdef CONFIG_SMP
2371/*
2372 * ->cpus_allowed is protected by both rq->lock and p->pi_lock
2373 */
2374static int select_fallback_rq(int cpu, struct task_struct *p)
2375{
2376    int dest_cpu;
2377    const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu));
2378
2379    /* Look for allowed, online CPU in same node. */
2380    for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask)
2381        if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed))
2382            return dest_cpu;
2383
2384    /* Any allowed, online CPU? */
2385    dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_active_mask);
2386    if (dest_cpu < nr_cpu_ids)
2387        return dest_cpu;
2388
2389    /* No more Mr. Nice Guy. */
2390    dest_cpu = cpuset_cpus_allowed_fallback(p);
2391    /*
2392     * Don't tell them about moving exiting tasks or
2393     * kernel threads (both mm NULL), since they never
2394     * leave kernel.
2395     */
2396    if (p->mm && printk_ratelimit()) {
2397        printk(KERN_INFO "process %d (%s) no longer affine to cpu%d\n",
2398                task_pid_nr(p), p->comm, cpu);
2399    }
2400
2401    return dest_cpu;
2402}
2403
2404/*
2405 * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable.
2406 */
2407static inline
2408int select_task_rq(struct task_struct *p, int sd_flags, int wake_flags)
2409{
2410    int cpu = p->sched_class->select_task_rq(p, sd_flags, wake_flags);
2411
2412    /*
2413     * In order not to call set_task_cpu() on a blocking task we need
2414     * to rely on ttwu() to place the task on a valid ->cpus_allowed
2415     * cpu.
2416     *
2417     * Since this is common to all placement strategies, this lives here.
2418     *
2419     * [ this allows ->select_task() to simply return task_cpu(p) and
2420     * not worry about this generic constraint ]
2421     */
2422    if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) ||
2423             !cpu_online(cpu)))
2424        cpu = select_fallback_rq(task_cpu(p), p);
2425
2426    return cpu;
2427}
2428
2429static void update_avg(u64 *avg, u64 sample)
2430{
2431    s64 diff = sample - *avg;
2432    *avg += diff >> 3;
2433}
2434#endif
2435
2436static void
2437ttwu_stat(struct task_struct *p, int cpu, int wake_flags)
2438{
2439#ifdef CONFIG_SCHEDSTATS
2440    struct rq *rq = this_rq();
2441
2442#ifdef CONFIG_SMP
2443    int this_cpu = smp_processor_id();
2444
2445    if (cpu == this_cpu) {
2446        schedstat_inc(rq, ttwu_local);
2447        schedstat_inc(p, se.statistics.nr_wakeups_local);
2448    } else {
2449        struct sched_domain *sd;
2450
2451        schedstat_inc(p, se.statistics.nr_wakeups_remote);
2452        rcu_read_lock();
2453        for_each_domain(this_cpu, sd) {
2454            if (cpumask_test_cpu(cpu, sched_domain_span(sd))) {
2455                schedstat_inc(sd, ttwu_wake_remote);
2456                break;
2457            }
2458        }
2459        rcu_read_unlock();
2460    }
2461
2462    if (wake_flags & WF_MIGRATED)
2463        schedstat_inc(p, se.statistics.nr_wakeups_migrate);
2464
2465#endif /* CONFIG_SMP */
2466
2467    schedstat_inc(rq, ttwu_count);
2468    schedstat_inc(p, se.statistics.nr_wakeups);
2469
2470    if (wake_flags & WF_SYNC)
2471        schedstat_inc(p, se.statistics.nr_wakeups_sync);
2472
2473#endif /* CONFIG_SCHEDSTATS */
2474}
2475
2476static void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags)
2477{
2478    activate_task(rq, p, en_flags);
2479    p->on_rq = 1;
2480
2481    /* if a worker is waking up, notify workqueue */
2482    if (p->flags & PF_WQ_WORKER)
2483        wq_worker_waking_up(p, cpu_of(rq));
2484}
2485
2486/*
2487 * Mark the task runnable and perform wakeup-preemption.
2488 */
2489static void
2490ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags)
2491{
2492    trace_sched_wakeup(p, true);
2493    check_preempt_curr(rq, p, wake_flags);
2494
2495    p->state = TASK_RUNNING;
2496#ifdef CONFIG_SMP
2497    if (p->sched_class->task_woken)
2498        p->sched_class->task_woken(rq, p);
2499
2500    if (unlikely(rq->idle_stamp)) {
2501        u64 delta = rq->clock - rq->idle_stamp;
2502        u64 max = 2*sysctl_sched_migration_cost;
2503
2504        if (delta > max)
2505            rq->avg_idle = max;
2506        else
2507            update_avg(&rq->avg_idle, delta);
2508        rq->idle_stamp = 0;
2509    }
2510#endif
2511}
2512
2513static void
2514ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags)
2515{
2516#ifdef CONFIG_SMP
2517    if (p->sched_contributes_to_load)
2518        rq->nr_uninterruptible--;
2519#endif
2520
2521    ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING);
2522    ttwu_do_wakeup(rq, p, wake_flags);
2523}
2524
2525/*
2526 * Called in case the task @p isn't fully descheduled from its runqueue,
2527 * in this case we must do a remote wakeup. Its a 'light' wakeup though,
2528 * since all we need to do is flip p->state to TASK_RUNNING, since
2529 * the task is still ->on_rq.
2530 */
2531static int ttwu_remote(struct task_struct *p, int wake_flags)
2532{
2533    struct rq *rq;
2534    int ret = 0;
2535
2536    rq = __task_rq_lock(p);
2537    if (p->on_rq) {
2538        ttwu_do_wakeup(rq, p, wake_flags);
2539        ret = 1;
2540    }
2541    __task_rq_unlock(rq);
2542
2543    return ret;
2544}
2545
2546#ifdef CONFIG_SMP
2547static void sched_ttwu_do_pending(struct task_struct *list)
2548{
2549    struct rq *rq = this_rq();
2550
2551    raw_spin_lock(&rq->lock);
2552
2553    while (list) {
2554        struct task_struct *p = list;
2555        list = list->wake_entry;
2556        ttwu_do_activate(rq, p, 0);
2557    }
2558
2559    raw_spin_unlock(&rq->lock);
2560}
2561
2562#ifdef CONFIG_HOTPLUG_CPU
2563
2564static void sched_ttwu_pending(void)
2565{
2566    struct rq *rq = this_rq();
2567    struct task_struct *list = xchg(&rq->wake_list, NULL);
2568
2569    if (!list)
2570        return;
2571
2572    sched_ttwu_do_pending(list);
2573}
2574
2575#endif /* CONFIG_HOTPLUG_CPU */
2576
2577void scheduler_ipi(void)
2578{
2579    struct rq *rq = this_rq();
2580    struct task_struct *list = xchg(&rq->wake_list, NULL);
2581
2582    if (!list)
2583        return;
2584
2585    /*
2586     * Not all reschedule IPI handlers call irq_enter/irq_exit, since
2587     * traditionally all their work was done from the interrupt return
2588     * path. Now that we actually do some work, we need to make sure
2589     * we do call them.
2590     *
2591     * Some archs already do call them, luckily irq_enter/exit nest
2592     * properly.
2593     *
2594     * Arguably we should visit all archs and update all handlers,
2595     * however a fair share of IPIs are still resched only so this would
2596     * somewhat pessimize the simple resched case.
2597     */
2598    irq_enter();
2599    sched_ttwu_do_pending(list);
2600    irq_exit();
2601}
2602
2603static void ttwu_queue_remote(struct task_struct *p, int cpu)
2604{
2605    struct rq *rq = cpu_rq(cpu);
2606    struct task_struct *next = rq->wake_list;
2607
2608    for (;;) {
2609        struct task_struct *old = next;
2610
2611        p->wake_entry = next;
2612        next = cmpxchg(&rq->wake_list, old, p);
2613        if (next == old)
2614            break;
2615    }
2616
2617    if (!next)
2618        smp_send_reschedule(cpu);
2619}
2620
2621#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
2622static int ttwu_activate_remote(struct task_struct *p, int wake_flags)
2623{
2624    struct rq *rq;
2625    int ret = 0;
2626
2627    rq = __task_rq_lock(p);
2628    if (p->on_cpu) {
2629        ttwu_activate(rq, p, ENQUEUE_WAKEUP);
2630        ttwu_do_wakeup(rq, p, wake_flags);
2631        ret = 1;
2632    }
2633    __task_rq_unlock(rq);
2634
2635    return ret;
2636
2637}
2638#endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */
2639#endif /* CONFIG_SMP */
2640
2641static void ttwu_queue(struct task_struct *p, int cpu)
2642{
2643    struct rq *rq = cpu_rq(cpu);
2644
2645#if defined(CONFIG_SMP)
2646    if (sched_feat(TTWU_QUEUE) && cpu != smp_processor_id()) {
2647        sched_clock_cpu(cpu); /* sync clocks x-cpu */
2648        ttwu_queue_remote(p, cpu);
2649        return;
2650    }
2651#endif
2652
2653    raw_spin_lock(&rq->lock);
2654    ttwu_do_activate(rq, p, 0);
2655    raw_spin_unlock(&rq->lock);
2656}
2657
2658/**
2659 * try_to_wake_up - wake up a thread
2660 * @p: the thread to be awakened
2661 * @state: the mask of task states that can be woken
2662 * @wake_flags: wake modifier flags (WF_*)
2663 *
2664 * Put it on the run-queue if it's not already there. The "current"
2665 * thread is always on the run-queue (except when the actual
2666 * re-schedule is in progress), and as such you're allowed to do
2667 * the simpler "current->state = TASK_RUNNING" to mark yourself
2668 * runnable without the overhead of this.
2669 *
2670 * Returns %true if @p was woken up, %false if it was already running
2671 * or @state didn't match @p's state.
2672 */
2673static int
2674try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags)
2675{
2676    unsigned long flags;
2677    int cpu, success = 0;
2678
2679    smp_wmb();
2680    raw_spin_lock_irqsave(&p->pi_lock, flags);
2681    if (!(p->state & state))
2682        goto out;
2683
2684    success = 1; /* we're going to change ->state */
2685    cpu = task_cpu(p);
2686
2687    if (p->on_rq && ttwu_remote(p, wake_flags))
2688        goto stat;
2689
2690#ifdef CONFIG_SMP
2691    /*
2692     * If the owning (remote) cpu is still in the middle of schedule() with
2693     * this task as prev, wait until its done referencing the task.
2694     */
2695    while (p->on_cpu) {
2696#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
2697        /*
2698         * In case the architecture enables interrupts in
2699         * context_switch(), we cannot busy wait, since that
2700         * would lead to deadlocks when an interrupt hits and
2701         * tries to wake up @prev. So bail and do a complete
2702         * remote wakeup.
2703         */
2704        if (ttwu_activate_remote(p, wake_flags))
2705            goto stat;
2706#else
2707        cpu_relax();
2708#endif
2709    }
2710    /*
2711     * Pairs with the smp_wmb() in finish_lock_switch().
2712     */
2713    smp_rmb();
2714
2715    p->sched_contributes_to_load = !!task_contributes_to_load(p);
2716    p->state = TASK_WAKING;
2717
2718    if (p->sched_class->task_waking)
2719        p->sched_class->task_waking(p);
2720
2721    cpu = select_task_rq(p, SD_BALANCE_WAKE, wake_flags);
2722    if (task_cpu(p) != cpu) {
2723        wake_flags |= WF_MIGRATED;
2724        set_task_cpu(p, cpu);
2725    }
2726#endif /* CONFIG_SMP */
2727
2728    ttwu_queue(p, cpu);
2729stat:
2730    ttwu_stat(p, cpu, wake_flags);
2731out:
2732    raw_spin_unlock_irqrestore(&p->pi_lock, flags);
2733
2734    return success;
2735}
2736
2737/**
2738 * try_to_wake_up_local - try to wake up a local task with rq lock held
2739 * @p: the thread to be awakened
2740 *
2741 * Put @p on the run-queue if it's not already there. The caller must
2742 * ensure that this_rq() is locked, @p is bound to this_rq() and not
2743 * the current task.
2744 */
2745static void try_to_wake_up_local(struct task_struct *p)
2746{
2747    struct rq *rq = task_rq(p);
2748
2749    BUG_ON(rq != this_rq());
2750    BUG_ON(p == current);
2751    lockdep_assert_held(&rq->lock);
2752
2753    if (!raw_spin_trylock(&p->pi_lock)) {
2754        raw_spin_unlock(&rq->lock);
2755        raw_spin_lock(&p->pi_lock);
2756        raw_spin_lock(&rq->lock);
2757    }
2758
2759    if (!(p->state & TASK_NORMAL))
2760        goto out;
2761
2762    if (!p->on_rq)
2763        ttwu_activate(rq, p, ENQUEUE_WAKEUP);
2764
2765    ttwu_do_wakeup(rq, p, 0);
2766    ttwu_stat(p, smp_processor_id(), 0);
2767out:
2768    raw_spin_unlock(&p->pi_lock);
2769}
2770
2771/**
2772 * wake_up_process - Wake up a specific process
2773 * @p: The process to be woken up.
2774 *
2775 * Attempt to wake up the nominated process and move it to the set of runnable
2776 * processes. Returns 1 if the process was woken up, 0 if it was already
2777 * running.
2778 *
2779 * It may be assumed that this function implies a write memory barrier before
2780 * changing the task state if and only if any tasks are woken up.
2781 */
2782int wake_up_process(struct task_struct *p)
2783{
2784    return try_to_wake_up(p, TASK_ALL, 0);
2785}
2786EXPORT_SYMBOL(wake_up_process);
2787
2788int wake_up_state(struct task_struct *p, unsigned int state)
2789{
2790    return try_to_wake_up(p, state, 0);
2791}
2792
2793/*
2794 * Perform scheduler related setup for a newly forked process p.
2795 * p is forked by current.
2796 *
2797 * __sched_fork() is basic setup used by init_idle() too:
2798 */
2799static void __sched_fork(struct task_struct *p)
2800{
2801    p->on_rq = 0;
2802
2803    p->se.on_rq = 0;
2804    p->se.exec_start = 0;
2805    p->se.sum_exec_runtime = 0;
2806    p->se.prev_sum_exec_runtime = 0;
2807    p->se.nr_migrations = 0;
2808    p->se.vruntime = 0;
2809    INIT_LIST_HEAD(&p->se.group_node);
2810
2811#ifdef CONFIG_SCHEDSTATS
2812    memset(&p->se.statistics, 0, sizeof(p->se.statistics));
2813#endif
2814
2815    INIT_LIST_HEAD(&p->rt.run_list);
2816
2817#ifdef CONFIG_PREEMPT_NOTIFIERS
2818    INIT_HLIST_HEAD(&p->preempt_notifiers);
2819#endif
2820}
2821
2822/*
2823 * fork()/clone()-time setup:
2824 */
2825void sched_fork(struct task_struct *p)
2826{
2827    unsigned long flags;
2828    int cpu = get_cpu();
2829
2830    __sched_fork(p);
2831    /*
2832     * We mark the process as running here. This guarantees that
2833     * nobody will actually run it, and a signal or other external
2834     * event cannot wake it up and insert it on the runqueue either.
2835     */
2836    p->state = TASK_RUNNING;
2837
2838    /*
2839     * Revert to default priority/policy on fork if requested.
2840     */
2841    if (unlikely(p->sched_reset_on_fork)) {
2842        if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) {
2843            p->policy = SCHED_NORMAL;
2844            p->normal_prio = p->static_prio;
2845        }
2846
2847        if (PRIO_TO_NICE(p->static_prio) < 0) {
2848            p->static_prio = NICE_TO_PRIO(0);
2849            p->normal_prio = p->static_prio;
2850            set_load_weight(p);
2851        }
2852
2853        /*
2854         * We don't need the reset flag anymore after the fork. It has
2855         * fulfilled its duty:
2856         */
2857        p->sched_reset_on_fork = 0;
2858    }
2859
2860    /*
2861     * Make sure we do not leak PI boosting priority to the child.
2862     */
2863    p->prio = current->normal_prio;
2864
2865    if (!rt_prio(p->prio))
2866        p->sched_class = &fair_sched_class;
2867
2868    if (p->sched_class->task_fork)
2869        p->sched_class->task_fork(p);
2870
2871    /*
2872     * The child is not yet in the pid-hash so no cgroup attach races,
2873     * and the cgroup is pinned to this child due to cgroup_fork()
2874     * is ran before sched_fork().
2875     *
2876     * Silence PROVE_RCU.
2877     */
2878    raw_spin_lock_irqsave(&p->pi_lock, flags);
2879    set_task_cpu(p, cpu);
2880    raw_spin_unlock_irqrestore(&p->pi_lock, flags);
2881
2882#if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT)
2883    if (likely(sched_info_on()))
2884        memset(&p->sched_info, 0, sizeof(p->sched_info));
2885#endif
2886#if defined(CONFIG_SMP)
2887    p->on_cpu = 0;
2888#endif
2889#ifdef CONFIG_PREEMPT
2890    /* Want to start with kernel preemption disabled. */
2891    task_thread_info(p)->preempt_count = 1;
2892#endif
2893#ifdef CONFIG_SMP
2894    plist_node_init(&p->pushable_tasks, MAX_PRIO);
2895#endif
2896
2897    put_cpu();
2898}
2899
2900/*
2901 * wake_up_new_task - wake up a newly created task for the first time.
2902 *
2903 * This function will do some initial scheduler statistics housekeeping
2904 * that must be done for every newly created context, then puts the task
2905 * on the runqueue and wakes it.
2906 */
2907void wake_up_new_task(struct task_struct *p)
2908{
2909    unsigned long flags;
2910    struct rq *rq;
2911
2912    raw_spin_lock_irqsave(&p->pi_lock, flags);
2913#ifdef CONFIG_SMP
2914    /*
2915     * Fork balancing, do it here and not earlier because:
2916     * - cpus_allowed can change in the fork path
2917     * - any previously selected cpu might disappear through hotplug
2918     */
2919    set_task_cpu(p, select_task_rq(p, SD_BALANCE_FORK, 0));
2920#endif
2921
2922    rq = __task_rq_lock(p);
2923    activate_task(rq, p, 0);
2924    p->on_rq = 1;
2925    trace_sched_wakeup_new(p, true);
2926    check_preempt_curr(rq, p, WF_FORK);
2927#ifdef CONFIG_SMP
2928    if (p->sched_class->task_woken)
2929        p->sched_class->task_woken(rq, p);
2930#endif
2931    task_rq_unlock(rq, p, &flags);
2932}
2933
2934#ifdef CONFIG_PREEMPT_NOTIFIERS
2935
2936/**
2937 * preempt_notifier_register - tell me when current is being preempted & rescheduled
2938 * @notifier: notifier struct to register
2939 */
2940void preempt_notifier_register(struct preempt_notifier *notifier)
2941{
2942    hlist_add_head(&notifier->link, &current->preempt_notifiers);
2943}
2944EXPORT_SYMBOL_GPL(preempt_notifier_register);
2945
2946/**
2947 * preempt_notifier_unregister - no longer interested in preemption notifications
2948 * @notifier: notifier struct to unregister
2949 *
2950 * This is safe to call from within a preemption notifier.
2951 */
2952void preempt_notifier_unregister(struct preempt_notifier *notifier)
2953{
2954    hlist_del(&notifier->link);
2955}
2956EXPORT_SYMBOL_GPL(preempt_notifier_unregister);
2957
2958static void fire_sched_in_preempt_notifiers(struct task_struct *curr)
2959{
2960    struct preempt_notifier *notifier;
2961    struct hlist_node *node;
2962
2963    hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link)
2964        notifier->ops->sched_in(notifier, raw_smp_processor_id());
2965}
2966
2967static void
2968fire_sched_out_preempt_notifiers(struct task_struct *curr,
2969                 struct task_struct *next)
2970{
2971    struct preempt_notifier *notifier;
2972    struct hlist_node *node;
2973
2974    hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link)
2975        notifier->ops->sched_out(notifier, next);
2976}
2977
2978#else /* !CONFIG_PREEMPT_NOTIFIERS */
2979
2980static void fire_sched_in_preempt_notifiers(struct task_struct *curr)
2981{
2982}
2983
2984static void
2985fire_sched_out_preempt_notifiers(struct task_struct *curr,
2986                 struct task_struct *next)
2987{
2988}
2989
2990#endif /* CONFIG_PREEMPT_NOTIFIERS */
2991
2992/**
2993 * prepare_task_switch - prepare to switch tasks
2994 * @rq: the runqueue preparing to switch
2995 * @prev: the current task that is being switched out
2996 * @next: the task we are going to switch to.
2997 *
2998 * This is called with the rq lock held and interrupts off. It must
2999 * be paired with a subsequent finish_task_switch after the context
3000 * switch.
3001 *
3002 * prepare_task_switch sets up locking and calls architecture specific
3003 * hooks.
3004 */
3005static inline void
3006prepare_task_switch(struct rq *rq, struct task_struct *prev,
3007            struct task_struct *next)
3008{
3009    sched_info_switch(prev, next);
3010    perf_event_task_sched_out(prev, next);
3011    fire_sched_out_preempt_notifiers(prev, next);
3012    prepare_lock_switch(rq, next);
3013    prepare_arch_switch(next);
3014    trace_sched_switch(prev, next);
3015}
3016
3017/**
3018 * finish_task_switch - clean up after a task-switch
3019 * @rq: runqueue associated with task-switch
3020 * @prev: the thread we just switched away from.
3021 *
3022 * finish_task_switch must be called after the context switch, paired
3023 * with a prepare_task_switch call before the context switch.
3024 * finish_task_switch will reconcile locking set up by prepare_task_switch,
3025 * and do any other architecture-specific cleanup actions.
3026 *
3027 * Note that we may have delayed dropping an mm in context_switch(). If
3028 * so, we finish that here outside of the runqueue lock. (Doing it
3029 * with the lock held can cause deadlocks; see schedule() for
3030 * details.)
3031 */
3032static void finish_task_switch(struct rq *rq, struct task_struct *prev)
3033    __releases(rq->lock)
3034{
3035    struct mm_struct *mm = rq->prev_mm;
3036    long prev_state;
3037
3038    rq->prev_mm = NULL;
3039
3040    /*
3041     * A task struct has one reference for the use as "current".
3042     * If a task dies, then it sets TASK_DEAD in tsk->state and calls
3043     * schedule one last time. The schedule call will never return, and
3044     * the scheduled task must drop that reference.
3045     * The test for TASK_DEAD must occur while the runqueue locks are
3046     * still held, otherwise prev could be scheduled on another cpu, die
3047     * there before we look at prev->state, and then the reference would
3048     * be dropped twice.
3049     * Manfred Spraul <manfred@colorfullife.com>
3050     */
3051    prev_state = prev->state;
3052    finish_arch_switch(prev);
3053#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
3054    local_irq_disable();
3055#endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */
3056    perf_event_task_sched_in(current);
3057#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
3058    local_irq_enable();
3059#endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */
3060    finish_lock_switch(rq, prev);
3061
3062    fire_sched_in_preempt_notifiers(current);
3063    if (mm)
3064        mmdrop(mm);
3065    if (unlikely(prev_state == TASK_DEAD)) {
3066        /*
3067         * Remove function-return probe instances associated with this
3068         * task and put them back on the free list.
3069         */
3070        kprobe_flush_task(prev);
3071        put_task_struct(prev);
3072    }
3073}
3074
3075#ifdef CONFIG_SMP
3076
3077/* assumes rq->lock is held */
3078static inline void pre_schedule(struct rq *rq, struct task_struct *prev)
3079{
3080    if (prev->sched_class->pre_schedule)
3081        prev->sched_class->pre_schedule(rq, prev);
3082}
3083
3084/* rq->lock is NOT held, but preemption is disabled */
3085static inline void post_schedule(struct rq *rq)
3086{
3087    if (rq->post_schedule) {
3088        unsigned long flags;
3089
3090        raw_spin_lock_irqsave(&rq->lock, flags);
3091        if (rq->curr->sched_class->post_schedule)
3092            rq->curr->sched_class->post_schedule(rq);
3093        raw_spin_unlock_irqrestore(&rq->lock, flags);
3094
3095        rq->post_schedule = 0;
3096    }
3097}
3098
3099#else
3100
3101static inline void pre_schedule(struct rq *rq, struct task_struct *p)
3102{
3103}
3104
3105static inline void post_schedule(struct rq *rq)
3106{
3107}
3108
3109#endif
3110
3111/**
3112 * schedule_tail - first thing a freshly forked thread must call.
3113 * @prev: the thread we just switched away from.
3114 */
3115asmlinkage void schedule_tail(struct task_struct *prev)
3116    __releases(rq->lock)
3117{
3118    struct rq *rq = this_rq();
3119
3120    finish_task_switch(rq, prev);
3121
3122    /*
3123     * FIXME: do we need to worry about rq being invalidated by the
3124     * task_switch?
3125     */
3126    post_schedule(rq);
3127
3128#ifdef __ARCH_WANT_UNLOCKED_CTXSW
3129    /* In this case, finish_task_switch does not reenable preemption */
3130    preempt_enable();
3131#endif
3132    if (current->set_child_tid)
3133        put_user(task_pid_vnr(current), current->set_child_tid);
3134}
3135
3136/*
3137 * context_switch - switch to the new MM and the new
3138 * thread's register state.
3139 */
3140static inline void
3141context_switch(struct rq *rq, struct task_struct *prev,
3142           struct task_struct *next)
3143{
3144    struct mm_struct *mm, *oldmm;
3145
3146    prepare_task_switch(rq, prev, next);
3147
3148    mm = next->mm;
3149    oldmm = prev->active_mm;
3150    /*
3151     * For paravirt, this is coupled with an exit in switch_to to
3152     * combine the page table reload and the switch backend into
3153     * one hypercall.
3154     */
3155    arch_start_context_switch(prev);
3156
3157    if (!mm) {
3158        next->active_mm = oldmm;
3159        atomic_inc(&oldmm->mm_count);
3160        enter_lazy_tlb(oldmm, next);
3161    } else
3162        switch_mm(oldmm, mm, next);
3163
3164    if (!prev->mm) {
3165        prev->active_mm = NULL;
3166        rq->prev_mm = oldmm;
3167    }
3168    /*
3169     * Since the runqueue lock will be released by the next
3170     * task (which is an invalid locking op but in the case
3171     * of the scheduler it's an obvious special-case), so we
3172     * do an early lockdep release here:
3173     */
3174#ifndef __ARCH_WANT_UNLOCKED_CTXSW
3175    spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
3176#endif
3177
3178    /* Here we just switch the register state and the stack. */
3179    switch_to(prev, next, prev);
3180
3181    barrier();
3182    /*
3183     * this_rq must be evaluated again because prev may have moved
3184     * CPUs since it called schedule(), thus the 'rq' on its stack
3185     * frame will be invalid.
3186     */
3187    finish_task_switch(this_rq(), prev);
3188}
3189
3190/*
3191 * nr_running, nr_uninterruptible and nr_context_switches:
3192 *
3193 * externally visible scheduler statistics: current number of runnable
3194 * threads, current number of uninterruptible-sleeping threads, total
3195 * number of context switches performed since bootup.
3196 */
3197unsigned long nr_running(void)
3198{
3199    unsigned long i, sum = 0;
3200
3201    for_each_online_cpu(i)
3202        sum += cpu_rq(i)->nr_running;
3203
3204    return sum;
3205}
3206
3207unsigned long nr_uninterruptible(void)
3208{
3209    unsigned long i, sum = 0;
3210
3211    for_each_possible_cpu(i)
3212        sum += cpu_rq(i)->nr_uninterruptible;
3213
3214    /*
3215     * Since we read the counters lockless, it might be slightly
3216     * inaccurate. Do not allow it to go below zero though:
3217     */
3218    if (unlikely((long)sum < 0))
3219        sum = 0;
3220
3221    return sum;
3222}
3223
3224unsigned long long nr_context_switches(void)
3225{
3226    int i;
3227    unsigned long long sum = 0;
3228
3229    for_each_possible_cpu(i)
3230        sum += cpu_rq(i)->nr_switches;
3231
3232    return sum;
3233}
3234
3235unsigned long nr_iowait(void)
3236{
3237    unsigned long i, sum = 0;
3238
3239    for_each_possible_cpu(i)
3240        sum += atomic_read(&cpu_rq(i)->nr_iowait);
3241
3242    return sum;
3243}
3244
3245unsigned long nr_iowait_cpu(int cpu)
3246{
3247    struct rq *this = cpu_rq(cpu);
3248    return atomic_read(&this->nr_iowait);
3249}
3250
3251unsigned long this_cpu_load(void)
3252{
3253    struct rq *this = this_rq();
3254    return this->cpu_load[0];
3255}
3256
3257
3258/* Variables and functions for calc_load */
3259static atomic_long_t calc_load_tasks;
3260static unsigned long calc_load_update;
3261unsigned long avenrun[3];
3262EXPORT_SYMBOL(avenrun);
3263
3264static long calc_load_fold_active(struct rq *this_rq)
3265{
3266    long nr_active, delta = 0;
3267
3268    nr_active = this_rq->nr_running;
3269    nr_active += (long) this_rq->nr_uninterruptible;
3270
3271    if (nr_active != this_rq->calc_load_active) {
3272        delta = nr_active - this_rq->calc_load_active;
3273        this_rq->calc_load_active = nr_active;
3274    }
3275
3276    return delta;
3277}
3278
3279static unsigned long
3280calc_load(unsigned long load, unsigned long exp, unsigned long active)
3281{
3282    load *= exp;
3283    load += active * (FIXED_1 - exp);
3284    load += 1UL << (FSHIFT - 1);
3285    return load >> FSHIFT;
3286}
3287
3288#ifdef CONFIG_NO_HZ
3289/*
3290 * For NO_HZ we delay the active fold to the next LOAD_FREQ update.
3291 *
3292 * When making the ILB scale, we should try to pull this in as well.
3293 */
3294static atomic_long_t calc_load_tasks_idle;
3295
3296static void calc_load_account_idle(struct rq *this_rq)
3297{
3298    long delta;
3299
3300    delta = calc_load_fold_active(this_rq);
3301    if (delta)
3302        atomic_long_add(delta, &calc_load_tasks_idle);
3303}
3304
3305static long calc_load_fold_idle(void)
3306{
3307    long delta = 0;
3308
3309    /*
3310     * Its got a race, we don't care...
3311     */
3312    if (atomic_long_read(&calc_load_tasks_idle))
3313        delta = atomic_long_xchg(&calc_load_tasks_idle, 0);
3314
3315    return delta;
3316}
3317
3318/**
3319 * fixed_power_int - compute: x^n, in O(log n) time
3320 *
3321 * @x: base of the power
3322 * @frac_bits: fractional bits of @x
3323 * @n: power to raise @x to.
3324 *
3325 * By exploiting the relation between the definition of the natural power
3326 * function: x^n := x*x*...*x (x multiplied by itself for n times), and
3327 * the binary encoding of numbers used by computers: n := \Sum n_i * 2^i,
3328 * (where: n_i \elem {0, 1}, the binary vector representing n),
3329 * we find: x^n := x^(\Sum n_i * 2^i) := \Prod x^(n_i * 2^i), which is
3330 * of course trivially computable in O(log_2 n), the length of our binary
3331 * vector.
3332 */
3333static unsigned long
3334fixed_power_int(unsigned long x, unsigned int frac_bits, unsigned int n)
3335{
3336    unsigned long result = 1UL << frac_bits;
3337
3338    if (n) for (;;) {
3339        if (n & 1) {
3340            result *= x;
3341            result += 1UL << (frac_bits - 1);
3342            result >>= frac_bits;
3343        }
3344        n >>= 1;
3345        if (!n)
3346            break;
3347        x *= x;
3348        x += 1UL << (frac_bits - 1);
3349        x >>= frac_bits;
3350    }
3351
3352    return result;
3353}
3354
3355/*
3356 * a1 = a0 * e + a * (1 - e)
3357 *
3358 * a2 = a1 * e + a * (1 - e)
3359 * = (a0 * e + a * (1 - e)) * e + a * (1 - e)
3360 * = a0 * e^2 + a * (1 - e) * (1 + e)
3361 *
3362 * a3 = a2 * e + a * (1 - e)
3363 * = (a0 * e^2 + a * (1 - e) * (1 + e)) * e + a * (1 - e)
3364 * = a0 * e^3 + a * (1 - e) * (1 + e + e^2)
3365 *
3366 * ...
3367 *
3368 * an = a0 * e^n + a * (1 - e) * (1 + e + ... + e^n-1) [1]
3369 * = a0 * e^n + a * (1 - e) * (1 - e^n)/(1 - e)
3370 * = a0 * e^n + a * (1 - e^n)
3371 *
3372 * [1] application of the geometric series:
3373 *
3374 * n 1 - x^(n+1)
3375 * S_n := \Sum x^i = -------------
3376 * i=0 1 - x
3377 */
3378static unsigned long
3379calc_load_n(unsigned long load, unsigned long exp,
3380        unsigned long active, unsigned int n)
3381{
3382
3383    return calc_load(load, fixed_power_int(exp, FSHIFT, n), active);
3384}
3385
3386/*
3387 * NO_HZ can leave us missing all per-cpu ticks calling
3388 * calc_load_account_active(), but since an idle CPU folds its delta into
3389 * calc_load_tasks_idle per calc_load_account_idle(), all we need to do is fold
3390 * in the pending idle delta if our idle period crossed a load cycle boundary.
3391 *
3392 * Once we've updated the global active value, we need to apply the exponential
3393 * weights adjusted to the number of cycles missed.
3394 */
3395static void calc_global_nohz(unsigned long ticks)
3396{
3397    long delta, active, n;
3398
3399    if (time_before(jiffies, calc_load_update))
3400        return;
3401
3402    /*
3403     * If we crossed a calc_load_update boundary, make sure to fold
3404     * any pending idle changes, the respective CPUs might have
3405     * missed the tick driven calc_load_account_active() update
3406     * due to NO_HZ.
3407     */
3408    delta = calc_load_fold_idle();
3409    if (delta)
3410        atomic_long_add(delta, &calc_load_tasks);
3411
3412    /*
3413     * If we were idle for multiple load cycles, apply them.
3414     */
3415    if (ticks >= LOAD_FREQ) {
3416        n = ticks / LOAD_FREQ;
3417
3418        active = atomic_long_read(&calc_load_tasks);
3419        active = active > 0 ? active * FIXED_1 : 0;
3420
3421        avenrun[0] = calc_load_n(avenrun[0], EXP_1, active, n);
3422        avenrun[1] = calc_load_n(avenrun[1], EXP_5, active, n);
3423        avenrun[2] = calc_load_n(avenrun[2], EXP_15, active, n);
3424
3425        calc_load_update += n * LOAD_FREQ;
3426    }
3427
3428    /*
3429     * Its possible the remainder of the above division also crosses
3430     * a LOAD_FREQ period, the regular check in calc_global_load()
3431     * which comes after this will take care of that.
3432     *
3433     * Consider us being 11 ticks before a cycle completion, and us
3434     * sleeping for 4*LOAD_FREQ + 22 ticks, then the above code will
3435     * age us 4 cycles, and the test in calc_global_load() will
3436     * pick up the final one.
3437     */
3438}
3439#else
3440static void calc_load_account_idle(struct rq *this_rq)
3441{
3442}
3443
3444static inline long calc_load_fold_idle(void)
3445{
3446    return 0;
3447}
3448
3449static void calc_global_nohz(unsigned long ticks)
3450{
3451}
3452#endif
3453
3454/**
3455 * get_avenrun - get the load average array
3456 * @loads: pointer to dest load array
3457 * @offset: offset to add
3458 * @shift: shift count to shift the result left
3459 *
3460 * These values are estimates at best, so no need for locking.
3461 */
3462void get_avenrun(unsigned long *loads, unsigned long offset, int shift)
3463{
3464    loads[0] = (avenrun[0] + offset) << shift;
3465    loads[1] = (avenrun[1] + offset) << shift;
3466    loads[2] = (avenrun[2] + offset) << shift;
3467}
3468
3469/*
3470 * calc_load - update the avenrun load estimates 10 ticks after the
3471 * CPUs have updated calc_load_tasks.
3472 */
3473void calc_global_load(unsigned long ticks)
3474{
3475    long active;
3476
3477    calc_global_nohz(ticks);
3478
3479    if (time_before(jiffies, calc_load_update + 10))
3480        return;
3481
3482    active = atomic_long_read(&calc_load_tasks);
3483    active = active > 0 ? active * FIXED_1 : 0;
3484
3485    avenrun[0] = calc_load(avenrun[0], EXP_1, active);
3486    avenrun[1] = calc_load(avenrun[1], EXP_5, active);
3487    avenrun[2] = calc_load(avenrun[2], EXP_15, active);
3488
3489    calc_load_update += LOAD_FREQ;
3490}
3491
3492/*
3493 * Called from update_cpu_load() to periodically update this CPU's
3494 * active count.
3495 */
3496static void calc_load_account_active(struct rq *this_rq)
3497{
3498    long delta;
3499
3500    if (time_before(jiffies, this_rq->calc_load_update))
3501        return;
3502
3503    delta = calc_load_fold_active(this_rq);
3504    delta += calc_load_fold_idle();
3505    if (delta)
3506        atomic_long_add(delta, &calc_load_tasks);
3507
3508    this_rq->calc_load_update += LOAD_FREQ;
3509}
3510
3511/*
3512 * The exact cpuload at various idx values, calculated at every tick would be
3513 * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load
3514 *
3515 * If a cpu misses updates for n-1 ticks (as it was idle) and update gets called
3516 * on nth tick when cpu may be busy, then we have:
3517 * load = ((2^idx - 1) / 2^idx)^(n-1) * load
3518 * load = (2^idx - 1) / 2^idx) * load + 1 / 2^idx * cur_load
3519 *
3520 * decay_load_missed() below does efficient calculation of
3521 * load = ((2^idx - 1) / 2^idx)^(n-1) * load
3522 * avoiding 0..n-1 loop doing load = ((2^idx - 1) / 2^idx) * load
3523 *
3524 * The calculation is approximated on a 128 point scale.
3525 * degrade_zero_ticks is the number of ticks after which load at any
3526 * particular idx is approximated to be zero.
3527 * degrade_factor is a precomputed table, a row for each load idx.
3528 * Each column corresponds to degradation factor for a power of two ticks,
3529 * based on 128 point scale.
3530 * Example:
3531 * row 2, col 3 (=12) says that the degradation at load idx 2 after
3532 * 8 ticks is 12/128 (which is an approximation of exact factor 3^8/4^8).
3533 *
3534 * With this power of 2 load factors, we can degrade the load n times
3535 * by looking at 1 bits in n and doing as many mult/shift instead of
3536 * n mult/shifts needed by the exact degradation.
3537 */
3538#define DEGRADE_SHIFT 7
3539static const unsigned char
3540        degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128};
3541static const unsigned char
3542        degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = {
3543                    {0, 0, 0, 0, 0, 0, 0, 0},
3544                    {64, 32, 8, 0, 0, 0, 0, 0},
3545                    {96, 72, 40, 12, 1, 0, 0},
3546                    {112, 98, 75, 43, 15, 1, 0},
3547                    {120, 112, 98, 76, 45, 16, 2} };
3548
3549/*
3550 * Update cpu_load for any missed ticks, due to tickless idle. The backlog
3551 * would be when CPU is idle and so we just decay the old load without
3552 * adding any new load.
3553 */
3554static unsigned long
3555decay_load_missed(unsigned long load, unsigned long missed_updates, int idx)
3556{
3557    int j = 0;
3558
3559    if (!missed_updates)
3560        return load;
3561
3562    if (missed_updates >= degrade_zero_ticks[idx])
3563        return 0;
3564
3565    if (idx == 1)
3566        return load >> missed_updates;
3567
3568    while (missed_updates) {
3569        if (missed_updates % 2)
3570            load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT;
3571
3572        missed_updates >>= 1;
3573        j++;
3574    }
3575    return load;
3576}
3577
3578/*
3579 * Update rq->cpu_load[] statistics. This function is usually called every
3580 * scheduler tick (TICK_NSEC). With tickless idle this will not be called
3581 * every tick. We fix it up based on jiffies.
3582 */
3583static void update_cpu_load(struct rq *this_rq)
3584{
3585    unsigned long this_load = this_rq->load.weight;
3586    unsigned long curr_jiffies = jiffies;
3587    unsigned long pending_updates;
3588    int i, scale;
3589
3590    this_rq->nr_load_updates++;
3591
3592    /* Avoid repeated calls on same jiffy, when moving in and out of idle */
3593    if (curr_jiffies == this_rq->last_load_update_tick)
3594        return;
3595
3596    pending_updates = curr_jiffies - this_rq->last_load_update_tick;
3597    this_rq->last_load_update_tick = curr_jiffies;
3598
3599    /* Update our load: */
3600    this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */
3601    for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) {
3602        unsigned long old_load, new_load;
3603
3604        /* scale is effectively 1 << i now, and >> i divides by scale */
3605
3606        old_load = this_rq->cpu_load[i];
3607        old_load = decay_load_missed(old_load, pending_updates - 1, i);
3608        new_load = this_load;
3609        /*
3610         * Round up the averaging division if load is increasing. This
3611         * prevents us from getting stuck on 9 if the load is 10, for
3612         * example.
3613         */
3614        if (new_load > old_load)
3615            new_load += scale - 1;
3616
3617        this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i;
3618    }
3619
3620    sched_avg_update(this_rq);
3621}
3622
3623static void update_cpu_load_active(struct rq *this_rq)
3624{
3625    update_cpu_load(this_rq);
3626
3627    calc_load_account_active(this_rq);
3628}
3629
3630#ifdef CONFIG_SMP
3631
3632/*
3633 * sched_exec - execve() is a valuable balancing opportunity, because at
3634 * this point the task has the smallest effective memory and cache footprint.
3635 */
3636void sched_exec(void)
3637{
3638    struct task_struct *p = current;
3639    unsigned long flags;
3640    int dest_cpu;
3641
3642    raw_spin_lock_irqsave(&p->pi_lock, flags);
3643    dest_cpu = p->sched_class->select_task_rq(p, SD_BALANCE_EXEC, 0);
3644    if (dest_cpu == smp_processor_id())
3645        goto unlock;
3646
3647    if (likely(cpu_active(dest_cpu))) {
3648        struct migration_arg arg = { p, dest_cpu };
3649
3650        raw_spin_unlock_irqrestore(&p->pi_lock, flags);
3651        stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg);
3652        return;
3653    }
3654unlock:
3655    raw_spin_unlock_irqrestore(&p->pi_lock, flags);
3656}
3657
3658#endif
3659
3660DEFINE_PER_CPU(struct kernel_stat, kstat);
3661
3662EXPORT_PER_CPU_SYMBOL(kstat);
3663
3664/*
3665 * Return any ns on the sched_clock that have not yet been accounted in
3666 * @p in case that task is currently running.
3667 *
3668 * Called with task_rq_lock() held on @rq.
3669 */
3670static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq)
3671{
3672    u64 ns = 0;
3673
3674    if (task_current(rq, p)) {
3675        update_rq_clock(rq);
3676        ns = rq->clock_task - p->se.exec_start;
3677        if ((s64)ns < 0)
3678            ns = 0;
3679    }
3680
3681    return ns;
3682}
3683
3684unsigned long long task_delta_exec(struct task_struct *p)
3685{
3686    unsigned long flags;
3687    struct rq *rq;
3688    u64 ns = 0;
3689
3690    rq = task_rq_lock(p, &flags);
3691    ns = do_task_delta_exec(p, rq);
3692    task_rq_unlock(rq, p, &flags);
3693
3694    return ns;
3695}
3696
3697/*
3698 * Return accounted runtime for the task.
3699 * In case the task is currently running, return the runtime plus current's
3700 * pending runtime that have not been accounted yet.
3701 */
3702unsigned long long task_sched_runtime(struct task_struct *p)
3703{
3704    unsigned long flags;
3705    struct rq *rq;
3706    u64 ns = 0;
3707
3708    rq = task_rq_lock(p, &flags);
3709    ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq);
3710    task_rq_unlock(rq, p, &flags);
3711
3712    return ns;
3713}
3714
3715/*
3716 * Return sum_exec_runtime for the thread group.
3717 * In case the task is currently running, return the sum plus current's
3718 * pending runtime that have not been accounted yet.
3719 *
3720 * Note that the thread group might have other running tasks as well,
3721 * so the return value not includes other pending runtime that other
3722 * running tasks might have.
3723 */
3724unsigned long long thread_group_sched_runtime(struct task_struct *p)
3725{
3726    struct task_cputime totals;
3727    unsigned long flags;
3728    struct rq *rq;
3729    u64 ns;
3730
3731    rq = task_rq_lock(p, &flags);
3732    thread_group_cputime(p, &totals);
3733    ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq);
3734    task_rq_unlock(rq, p, &flags);
3735
3736    return ns;
3737}
3738
3739/*
3740 * Account user cpu time to a process.
3741 * @p: the process that the cpu time gets accounted to
3742 * @cputime: the cpu time spent in user space since the last update
3743 * @cputime_scaled: cputime scaled by cpu frequency
3744 */
3745void account_user_time(struct task_struct *p, cputime_t cputime,
3746               cputime_t cputime_scaled)
3747{
3748    struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3749    cputime64_t tmp;
3750
3751    /* Add user time to process. */
3752    p->utime = cputime_add(p->utime, cputime);
3753    p->utimescaled = cputime_add(p->utimescaled, cputime_scaled);
3754    account_group_user_time(p, cputime);
3755
3756    /* Add user time to cpustat. */
3757    tmp = cputime_to_cputime64(cputime);
3758    if (TASK_NICE(p) > 0)
3759        cpustat->nice = cputime64_add(cpustat->nice, tmp);
3760    else
3761        cpustat->user = cputime64_add(cpustat->user, tmp);
3762
3763    cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime);
3764    /* Account for user time used */
3765    acct_update_integrals(p);
3766}
3767
3768/*
3769 * Account guest cpu time to a process.
3770 * @p: the process that the cpu time gets accounted to
3771 * @cputime: the cpu time spent in virtual machine since the last update
3772 * @cputime_scaled: cputime scaled by cpu frequency
3773 */
3774static void account_guest_time(struct task_struct *p, cputime_t cputime,
3775                   cputime_t cputime_scaled)
3776{
3777    cputime64_t tmp;
3778    struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3779
3780    tmp = cputime_to_cputime64(cputime);
3781
3782    /* Add guest time to process. */
3783    p->utime = cputime_add(p->utime, cputime);
3784    p->utimescaled = cputime_add(p->utimescaled, cputime_scaled);
3785    account_group_user_time(p, cputime);
3786    p->gtime = cputime_add(p->gtime, cputime);
3787
3788    /* Add guest time to cpustat. */
3789    if (TASK_NICE(p) > 0) {
3790        cpustat->nice = cputime64_add(cpustat->nice, tmp);
3791        cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp);
3792    } else {
3793        cpustat->user = cputime64_add(cpustat->user, tmp);
3794        cpustat->guest = cputime64_add(cpustat->guest, tmp);
3795    }
3796}
3797
3798/*
3799 * Account system cpu time to a process and desired cpustat field
3800 * @p: the process that the cpu time gets accounted to
3801 * @cputime: the cpu time spent in kernel space since the last update
3802 * @cputime_scaled: cputime scaled by cpu frequency
3803 * @target_cputime64: pointer to cpustat field that has to be updated
3804 */
3805static inline
3806void __account_system_time(struct task_struct *p, cputime_t cputime,
3807            cputime_t cputime_scaled, cputime64_t *target_cputime64)
3808{
3809    cputime64_t tmp = cputime_to_cputime64(cputime);
3810
3811    /* Add system time to process. */
3812    p->stime = cputime_add(p->stime, cputime);
3813    p->stimescaled = cputime_add(p->stimescaled, cputime_scaled);
3814    account_group_system_time(p, cputime);
3815
3816    /* Add system time to cpustat. */
3817    *target_cputime64 = cputime64_add(*target_cputime64, tmp);
3818    cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime);
3819
3820    /* Account for system time used */
3821    acct_update_integrals(p);
3822}
3823
3824/*
3825 * Account system cpu time to a process.
3826 * @p: the process that the cpu time gets accounted to
3827 * @hardirq_offset: the offset to subtract from hardirq_count()
3828 * @cputime: the cpu time spent in kernel space since the last update
3829 * @cputime_scaled: cputime scaled by cpu frequency
3830 */
3831void account_system_time(struct task_struct *p, int hardirq_offset,
3832             cputime_t cputime, cputime_t cputime_scaled)
3833{
3834    struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3835    cputime64_t *target_cputime64;
3836
3837    if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) {
3838        account_guest_time(p, cputime, cputime_scaled);
3839        return;
3840    }
3841
3842    if (hardirq_count() - hardirq_offset)
3843        target_cputime64 = &cpustat->irq;
3844    else if (in_serving_softirq())
3845        target_cputime64 = &cpustat->softirq;
3846    else
3847        target_cputime64 = &cpustat->system;
3848
3849    __account_system_time(p, cputime, cputime_scaled, target_cputime64);
3850}
3851
3852/*
3853 * Account for involuntary wait time.
3854 * @cputime: the cpu time spent in involuntary wait
3855 */
3856void account_steal_time(cputime_t cputime)
3857{
3858    struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3859    cputime64_t cputime64 = cputime_to_cputime64(cputime);
3860
3861    cpustat->steal = cputime64_add(cpustat->steal, cputime64);
3862}
3863
3864/*
3865 * Account for idle time.
3866 * @cputime: the cpu time spent in idle wait
3867 */
3868void account_idle_time(cputime_t cputime)
3869{
3870    struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3871    cputime64_t cputime64 = cputime_to_cputime64(cputime);
3872    struct rq *rq = this_rq();
3873
3874    if (atomic_read(&rq->nr_iowait) > 0)
3875        cpustat->iowait = cputime64_add(cpustat->iowait, cputime64);
3876    else
3877        cpustat->idle = cputime64_add(cpustat->idle, cputime64);
3878}
3879
3880#ifndef CONFIG_VIRT_CPU_ACCOUNTING
3881
3882#ifdef CONFIG_IRQ_TIME_ACCOUNTING
3883/*
3884 * Account a tick to a process and cpustat
3885 * @p: the process that the cpu time gets accounted to
3886 * @user_tick: is the tick from userspace
3887 * @rq: the pointer to rq
3888 *
3889 * Tick demultiplexing follows the order
3890 * - pending hardirq update
3891 * - pending softirq update
3892 * - user_time
3893 * - idle_time
3894 * - system time
3895 * - check for guest_time
3896 * - else account as system_time
3897 *
3898 * Check for hardirq is done both for system and user time as there is
3899 * no timer going off while we are on hardirq and hence we may never get an
3900 * opportunity to update it solely in system time.
3901 * p->stime and friends are only updated on system time and not on irq
3902 * softirq as those do not count in task exec_runtime any more.
3903 */
3904static void irqtime_account_process_tick(struct task_struct *p, int user_tick,
3905                        struct rq *rq)
3906{
3907    cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy);
3908    cputime64_t tmp = cputime_to_cputime64(cputime_one_jiffy);
3909    struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3910
3911    if (irqtime_account_hi_update()) {
3912        cpustat->irq = cputime64_add(cpustat->irq, tmp);
3913    } else if (irqtime_account_si_update()) {
3914        cpustat->softirq = cputime64_add(cpustat->softirq, tmp);
3915    } else if (this_cpu_ksoftirqd() == p) {
3916        /*
3917         * ksoftirqd time do not get accounted in cpu_softirq_time.
3918         * So, we have to handle it separately here.
3919         * Also, p->stime needs to be updated for ksoftirqd.
3920         */
3921        __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled,
3922                    &cpustat->softirq);
3923    } else if (user_tick) {
3924        account_user_time(p, cputime_one_jiffy, one_jiffy_scaled);
3925    } else if (p == rq->idle) {
3926        account_idle_time(cputime_one_jiffy);
3927    } else if (p->flags & PF_VCPU) { /* System time or guest time */
3928        account_guest_time(p, cputime_one_jiffy, one_jiffy_scaled);
3929    } else {
3930        __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled,
3931                    &cpustat->system);
3932    }
3933}
3934
3935static void irqtime_account_idle_ticks(int ticks)
3936{
3937    int i;
3938    struct rq *rq = this_rq();
3939
3940    for (i = 0; i < ticks; i++)
3941        irqtime_account_process_tick(current, 0, rq);
3942}
3943#else /* CONFIG_IRQ_TIME_ACCOUNTING */
3944static void irqtime_account_idle_ticks(int ticks) {}
3945static void irqtime_account_process_tick(struct task_struct *p, int user_tick,
3946                        struct rq *rq) {}
3947#endif /* CONFIG_IRQ_TIME_ACCOUNTING */
3948
3949/*
3950 * Account a single tick of cpu time.
3951 * @p: the process that the cpu time gets accounted to
3952 * @user_tick: indicates if the tick is a user or a system tick
3953 */
3954void account_process_tick(struct task_struct *p, int user_tick)
3955{
3956    cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy);
3957    struct rq *rq = this_rq();
3958
3959    if (sched_clock_irqtime) {
3960        irqtime_account_process_tick(p, user_tick, rq);
3961        return;
3962    }
3963
3964    if (user_tick)
3965        account_user_time(p, cputime_one_jiffy, one_jiffy_scaled);
3966    else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET))
3967        account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy,
3968                    one_jiffy_scaled);
3969    else
3970        account_idle_time(cputime_one_jiffy);
3971}
3972
3973/*
3974 * Account multiple ticks of steal time.
3975 * @p: the process from which the cpu time has been stolen
3976 * @ticks: number of stolen ticks
3977 */
3978void account_steal_ticks(unsigned long ticks)
3979{
3980    account_steal_time(jiffies_to_cputime(ticks));
3981}
3982
3983/*
3984 * Account multiple ticks of idle time.
3985 * @ticks: number of stolen ticks
3986 */
3987void account_idle_ticks(unsigned long ticks)
3988{
3989
3990    if (sched_clock_irqtime) {
3991        irqtime_account_idle_ticks(ticks);
3992        return;
3993    }
3994
3995    account_idle_time(jiffies_to_cputime(ticks));
3996}
3997
3998#endif
3999
4000/*
4001 * Use precise platform statistics if available:
4002 */
4003#ifdef CONFIG_VIRT_CPU_ACCOUNTING
4004void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
4005{
4006    *ut = p->utime;
4007    *st = p->stime;
4008}
4009
4010void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
4011{
4012    struct task_cputime cputime;
4013
4014    thread_group_cputime(p, &cputime);
4015
4016    *ut = cputime.utime;
4017    *st = cputime.stime;
4018}
4019#else
4020
4021#ifndef nsecs_to_cputime
4022# define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs)
4023#endif
4024
4025void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
4026{
4027    cputime_t rtime, utime = p->utime, total = cputime_add(utime, p->stime);
4028
4029    /*
4030     * Use CFS's precise accounting:
4031     */
4032    rtime = nsecs_to_cputime(p->se.sum_exec_runtime);
4033
4034    if (total) {
4035        u64 temp = rtime;
4036
4037        temp *= utime;
4038        do_div(temp, total);
4039        utime = (cputime_t)temp;
4040    } else
4041        utime = rtime;
4042
4043    /*
4044     * Compare with previous values, to keep monotonicity:
4045     */
4046    p->prev_utime = max(p->prev_utime, utime);
4047    p->prev_stime = max(p->prev_stime, cputime_sub(rtime, p->prev_utime));
4048
4049    *ut = p->prev_utime;
4050    *st = p->prev_stime;
4051}
4052
4053/*
4054 * Must be called with siglock held.
4055 */
4056void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
4057{
4058    struct signal_struct *sig = p->signal;
4059    struct task_cputime cputime;
4060    cputime_t rtime, utime, total;
4061
4062    thread_group_cputime(p, &cputime);
4063
4064    total = cputime_add(cputime.utime, cputime.stime);
4065    rtime = nsecs_to_cputime(cputime.sum_exec_runtime);
4066
4067    if (total) {
4068        u64 temp = rtime;
4069
4070        temp *= cputime.utime;
4071        do_div(temp, total);
4072        utime = (cputime_t)temp;
4073    } else
4074        utime = rtime;
4075
4076    sig->prev_utime = max(sig->prev_utime, utime);
4077    sig->prev_stime = max(sig->prev_stime,
4078                  cputime_sub(rtime, sig->prev_utime));
4079
4080    *ut = sig->prev_utime;
4081    *st = sig->prev_stime;
4082}
4083#endif
4084
4085/*
4086 * This function gets called by the timer code, with HZ frequency.
4087 * We call it with interrupts disabled.
4088 */
4089void scheduler_tick(void)
4090{
4091    int cpu = smp_processor_id();
4092    struct rq *rq = cpu_rq(cpu);
4093    struct task_struct *curr = rq->curr;
4094
4095    sched_clock_tick();
4096
4097    raw_spin_lock(&rq->lock);
4098    update_rq_clock(rq);
4099    update_cpu_load_active(rq);
4100    curr->sched_class->task_tick(rq, curr, 0);
4101    raw_spin_unlock(&rq->lock);
4102
4103    perf_event_task_tick();
4104
4105#ifdef CONFIG_SMP
4106    rq->idle_at_tick = idle_cpu(cpu);
4107    trigger_load_balance(rq, cpu);
4108#endif
4109}
4110
4111notrace unsigned long get_parent_ip(unsigned long addr)
4112{
4113    if (in_lock_functions(addr)) {
4114        addr = CALLER_ADDR2;
4115        if (in_lock_functions(addr))
4116            addr = CALLER_ADDR3;
4117    }
4118    return addr;
4119}
4120
4121#if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \
4122                defined(CONFIG_PREEMPT_TRACER))
4123
4124void __kprobes add_preempt_count(int val)
4125{
4126#ifdef CONFIG_DEBUG_PREEMPT
4127    /*
4128     * Underflow?
4129     */
4130    if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0)))
4131        return;
4132#endif
4133    preempt_count() += val;
4134#ifdef CONFIG_DEBUG_PREEMPT
4135    /*
4136     * Spinlock count overflowing soon?
4137     */
4138    DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >=
4139                PREEMPT_MASK - 10);
4140#endif
4141    if (preempt_count() == val)
4142        trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1));
4143}
4144EXPORT_SYMBOL(add_preempt_count);
4145
4146void __kprobes sub_preempt_count(int val)
4147{
4148#ifdef CONFIG_DEBUG_PREEMPT
4149    /*
4150     * Underflow?
4151     */
4152    if (DEBUG_LOCKS_WARN_ON(val > preempt_count()))
4153        return;
4154    /*
4155     * Is the spinlock portion underflowing?
4156     */
4157    if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) &&
4158            !(preempt_count() & PREEMPT_MASK)))
4159        return;
4160#endif
4161
4162    if (preempt_count() == val)
4163        trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1));
4164    preempt_count() -= val;
4165}
4166EXPORT_SYMBOL(sub_preempt_count);
4167
4168#endif
4169
4170/*
4171 * Print scheduling while atomic bug:
4172 */
4173static noinline void __schedule_bug(struct task_struct *prev)
4174{
4175    struct pt_regs *regs = get_irq_regs();
4176
4177    printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n",
4178        prev->comm, prev->pid, preempt_count());
4179
4180    debug_show_held_locks(prev);
4181    print_modules();
4182    if (irqs_disabled())
4183        print_irqtrace_events(prev);
4184
4185    if (regs)
4186        show_regs(regs);
4187    else
4188        dump_stack();
4189}
4190
4191/*
4192 * Various schedule()-time debugging checks and statistics:
4193 */
4194static inline void schedule_debug(struct task_struct *prev)
4195{
4196    /*
4197     * Test if we are atomic. Since do_exit() needs to call into
4198     * schedule() atomically, we ignore that path for now.
4199     * Otherwise, whine if we are scheduling when we should not be.
4200     */
4201    if (unlikely(in_atomic_preempt_off() && !prev->exit_state))
4202        __schedule_bug(prev);
4203
4204    profile_hit(SCHED_PROFILING, __builtin_return_address(0));
4205
4206    schedstat_inc(this_rq(), sched_count);
4207}
4208
4209static void put_prev_task(struct rq *rq, struct task_struct *prev)
4210{
4211    if (prev->on_rq || rq->skip_clock_update < 0)
4212        update_rq_clock(rq);
4213    prev->sched_class->put_prev_task(rq, prev);
4214}
4215
4216/*
4217 * Pick up the highest-prio task:
4218 */
4219static inline struct task_struct *
4220pick_next_task(struct rq *rq)
4221{
4222    const struct sched_class *class;
4223    struct task_struct *p;
4224
4225    /*
4226     * Optimization: we know that if all tasks are in
4227     * the fair class we can call that function directly:
4228     */
4229    if (likely(rq->nr_running == rq->cfs.nr_running)) {
4230        p = fair_sched_class.pick_next_task(rq);
4231        if (likely(p))
4232            return p;
4233    }
4234
4235    for_each_class(class) {
4236        p = class->pick_next_task(rq);
4237        if (p)
4238            return p;
4239    }
4240
4241    BUG(); /* the idle class will always have a runnable task */
4242}
4243
4244/*
4245 * schedule() is the main scheduler function.
4246 */
4247asmlinkage void __sched schedule(void)
4248{
4249    struct task_struct *prev, *next;
4250    unsigned long *switch_count;
4251    struct rq *rq;
4252    int cpu;
4253
4254need_resched:
4255    preempt_disable();
4256    cpu = smp_processor_id();
4257    rq = cpu_rq(cpu);
4258    rcu_note_context_switch(cpu);
4259    prev = rq->curr;
4260
4261    schedule_debug(prev);
4262
4263    if (sched_feat(HRTICK))
4264        hrtick_clear(rq);
4265
4266    raw_spin_lock_irq(&rq->lock);
4267
4268    switch_count = &prev->nivcsw;
4269    if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) {
4270        if (unlikely(signal_pending_state(prev->state, prev))) {
4271            prev->state = TASK_RUNNING;
4272        } else {
4273            deactivate_task(rq, prev, DEQUEUE_SLEEP);
4274            prev->on_rq = 0;
4275
4276            /*
4277             * If a worker went to sleep, notify and ask workqueue
4278             * whether it wants to wake up a task to maintain
4279             * concurrency.
4280             */
4281            if (prev->flags & PF_WQ_WORKER) {
4282                struct task_struct *to_wakeup;
4283
4284                to_wakeup = wq_worker_sleeping(prev, cpu);
4285                if (to_wakeup)
4286                    try_to_wake_up_local(to_wakeup);
4287            }
4288
4289            /*
4290             * If we are going to sleep and we have plugged IO
4291             * queued, make sure to submit it to avoid deadlocks.
4292             */
4293            if (blk_needs_flush_plug(prev)) {
4294                raw_spin_unlock(&rq->lock);
4295                blk_schedule_flush_plug(prev);
4296                raw_spin_lock(&rq->lock);
4297            }
4298        }
4299        switch_count = &prev->nvcsw;
4300    }
4301
4302    pre_schedule(rq, prev);
4303
4304    if (unlikely(!rq->nr_running))
4305        idle_balance(cpu, rq);
4306
4307    put_prev_task(rq, prev);
4308    next = pick_next_task(rq);
4309    clear_tsk_need_resched(prev);
4310    rq->skip_clock_update = 0;
4311
4312    if (likely(prev != next)) {
4313        rq->nr_switches++;
4314        rq->curr = next;
4315        ++*switch_count;
4316
4317        context_switch(rq, prev, next); /* unlocks the rq */
4318        /*
4319         * The context switch have flipped the stack from under us
4320         * and restored the local variables which were saved when
4321         * this task called schedule() in the past. prev == current
4322         * is still correct, but it can be moved to another cpu/rq.
4323         */
4324        cpu = smp_processor_id();
4325        rq = cpu_rq(cpu);
4326    } else
4327        raw_spin_unlock_irq(&rq->lock);
4328
4329    post_schedule(rq);
4330
4331    preempt_enable_no_resched();
4332    if (need_resched())
4333        goto need_resched;
4334}
4335EXPORT_SYMBOL(schedule);
4336
4337#ifdef CONFIG_MUTEX_SPIN_ON_OWNER
4338
4339static inline bool owner_running(struct mutex *lock, struct task_struct *owner)
4340{
4341    bool ret = false;
4342
4343    rcu_read_lock();
4344    if (lock->owner != owner)
4345        goto fail;
4346
4347    /*
4348     * Ensure we emit the owner->on_cpu, dereference _after_ checking
4349     * lock->owner still matches owner, if that fails, owner might
4350     * point to free()d memory, if it still matches, the rcu_read_lock()
4351     * ensures the memory stays valid.
4352     */
4353    barrier();
4354
4355    ret = owner->on_cpu;
4356fail:
4357    rcu_read_unlock();
4358
4359    return ret;
4360}
4361
4362/*
4363 * Look out! "owner" is an entirely speculative pointer
4364 * access and not reliable.
4365 */
4366int mutex_spin_on_owner(struct mutex *lock, struct task_struct *owner)
4367{
4368    if (!sched_feat(OWNER_SPIN))
4369        return 0;
4370
4371    while (owner_running(lock, owner)) {
4372        if (need_resched())
4373            return 0;
4374
4375        arch_mutex_cpu_relax();
4376    }
4377
4378    /*
4379     * If the owner changed to another task there is likely
4380     * heavy contention, stop spinning.
4381     */
4382    if (lock->owner)
4383        return 0;
4384
4385    return 1;
4386}
4387#endif
4388
4389#ifdef CONFIG_PREEMPT
4390/*
4391 * this is the entry point to schedule() from in-kernel preemption
4392 * off of preempt_enable. Kernel preemptions off return from interrupt
4393 * occur there and call schedule directly.
4394 */
4395asmlinkage void __sched notrace preempt_schedule(void)
4396{
4397    struct thread_info *ti = current_thread_info();
4398
4399    /*
4400     * If there is a non-zero preempt_count or interrupts are disabled,
4401     * we do not want to preempt the current task. Just return..
4402     */
4403    if (likely(ti->preempt_count || irqs_disabled()))
4404        return;
4405
4406    do {
4407        add_preempt_count_notrace(PREEMPT_ACTIVE);
4408        schedule();
4409        sub_preempt_count_notrace(PREEMPT_ACTIVE);
4410
4411        /*
4412         * Check again in case we missed a preemption opportunity
4413         * between schedule and now.
4414         */
4415        barrier();
4416    } while (need_resched());
4417}
4418EXPORT_SYMBOL(preempt_schedule);
4419
4420/*
4421 * this is the entry point to schedule() from kernel preemption
4422 * off of irq context.
4423 * Note, that this is called and return with irqs disabled. This will
4424 * protect us against recursive calling from irq.
4425 */
4426asmlinkage void __sched preempt_schedule_irq(void)
4427{
4428    struct thread_info *ti = current_thread_info();
4429
4430    /* Catch callers which need to be fixed */
4431    BUG_ON(ti->preempt_count || !irqs_disabled());
4432
4433    do {
4434        add_preempt_count(PREEMPT_ACTIVE);
4435        local_irq_enable();
4436        schedule();
4437        local_irq_disable();
4438        sub_preempt_count(PREEMPT_ACTIVE);
4439
4440        /*
4441         * Check again in case we missed a preemption opportunity
4442         * between schedule and now.
4443         */
4444        barrier();
4445    } while (need_resched());
4446}
4447
4448#endif /* CONFIG_PREEMPT */
4449
4450int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags,
4451              void *key)
4452{
4453    return try_to_wake_up(curr->private, mode, wake_flags);
4454}
4455EXPORT_SYMBOL(default_wake_function);
4456
4457/*
4458 * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just
4459 * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve
4460 * number) then we wake all the non-exclusive tasks and one exclusive task.
4461 *
4462 * There are circumstances in which we can try to wake a task which has already
4463 * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns
4464 * zero in this (rare) case, and we handle it by continuing to scan the queue.
4465 */
4466static void __wake_up_common(wait_queue_head_t *q, unsigned int mode,
4467            int nr_exclusive, int wake_flags, void *key)
4468{
4469    wait_queue_t *curr, *next;
4470
4471    list_for_each_entry_safe(curr, next, &q->task_list, task_list) {
4472        unsigned flags = curr->flags;
4473
4474        if (curr->func(curr, mode, wake_flags, key) &&
4475                (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive)
4476            break;
4477    }
4478}
4479
4480/**
4481 * __wake_up - wake up threads blocked on a waitqueue.
4482 * @q: the waitqueue
4483 * @mode: which threads
4484 * @nr_exclusive: how many wake-one or wake-many threads to wake up
4485 * @key: is directly passed to the wakeup function
4486 *
4487 * It may be assumed that this function implies a write memory barrier before
4488 * changing the task state if and only if any tasks are woken up.
4489 */
4490void __wake_up(wait_queue_head_t *q, unsigned int mode,
4491            int nr_exclusive, void *key)
4492{
4493    unsigned long flags;
4494
4495    spin_lock_irqsave(&q->lock, flags);
4496    __wake_up_common(q, mode, nr_exclusive, 0, key);
4497    spin_unlock_irqrestore(&q->lock, flags);
4498}
4499EXPORT_SYMBOL(__wake_up);
4500
4501/*
4502 * Same as __wake_up but called with the spinlock in wait_queue_head_t held.
4503 */
4504void __wake_up_locked(wait_queue_head_t *q, unsigned int mode)
4505{
4506    __wake_up_common(q, mode, 1, 0, NULL);
4507}
4508EXPORT_SYMBOL_GPL(__wake_up_locked);
4509
4510void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key)
4511{
4512    __wake_up_common(q, mode, 1, 0, key);
4513}
4514EXPORT_SYMBOL_GPL(__wake_up_locked_key);
4515
4516/**
4517 * __wake_up_sync_key - wake up threads blocked on a waitqueue.
4518 * @q: the waitqueue
4519 * @mode: which threads
4520 * @nr_exclusive: how many wake-one or wake-many threads to wake up
4521 * @key: opaque value to be passed to wakeup targets
4522 *
4523 * The sync wakeup differs that the waker knows that it will schedule
4524 * away soon, so while the target thread will be woken up, it will not
4525 * be migrated to another CPU - ie. the two threads are 'synchronized'
4526 * with each other. This can prevent needless bouncing between CPUs.
4527 *
4528 * On UP it can prevent extra preemption.
4529 *
4530 * It may be assumed that this function implies a write memory barrier before
4531 * changing the task state if and only if any tasks are woken up.
4532 */
4533void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode,
4534            int nr_exclusive, void *key)
4535{
4536    unsigned long flags;
4537    int wake_flags = WF_SYNC;
4538
4539    if (unlikely(!q))
4540        return;
4541
4542    if (unlikely(!nr_exclusive))
4543        wake_flags = 0;
4544
4545    spin_lock_irqsave(&q->lock, flags);
4546    __wake_up_common(q, mode, nr_exclusive, wake_flags, key);
4547    spin_unlock_irqrestore(&q->lock, flags);
4548}
4549EXPORT_SYMBOL_GPL(__wake_up_sync_key);
4550
4551/*
4552 * __wake_up_sync - see __wake_up_sync_key()
4553 */
4554void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive)
4555{
4556    __wake_up_sync_key(q, mode, nr_exclusive, NULL);
4557}
4558EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */
4559
4560/**
4561 * complete: - signals a single thread waiting on this completion
4562 * @x: holds the state of this particular completion
4563 *
4564 * This will wake up a single thread waiting on this completion. Threads will be
4565 * awakened in the same order in which they were queued.
4566 *
4567 * See also complete_all(), wait_for_completion() and related routines.
4568 *
4569 * It may be assumed that this function implies a write memory barrier before
4570 * changing the task state if and only if any tasks are woken up.
4571 */
4572void complete(struct completion *x)
4573{
4574    unsigned long flags;
4575
4576    spin_lock_irqsave(&x->wait.lock, flags);
4577    x->done++;
4578    __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL);
4579    spin_unlock_irqrestore(&x->wait.lock, flags);
4580}
4581EXPORT_SYMBOL(complete);
4582
4583/**
4584 * complete_all: - signals all threads waiting on this completion
4585 * @x: holds the state of this particular completion
4586 *
4587 * This will wake up all threads waiting on this particular completion event.
4588 *
4589 * It may be assumed that this function implies a write memory barrier before
4590 * changing the task state if and only if any tasks are woken up.
4591 */
4592void complete_all(struct completion *x)
4593{
4594    unsigned long flags;
4595
4596    spin_lock_irqsave(&x->wait.lock, flags);
4597    x->done += UINT_MAX/2;
4598    __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL);
4599    spin_unlock_irqrestore(&x->wait.lock, flags);
4600}
4601EXPORT_SYMBOL(complete_all);
4602
4603static inline long __sched
4604do_wait_for_common(struct completion *x, long timeout, int state)
4605{
4606    if (!x->done) {
4607        DECLARE_WAITQUEUE(wait, current);
4608
4609        __add_wait_queue_tail_exclusive(&x->wait, &wait);
4610        do {
4611            if (signal_pending_state(state, current)) {
4612                timeout = -ERESTARTSYS;
4613                break;
4614            }
4615            __set_current_state(state);
4616            spin_unlock_irq(&x->wait.lock);
4617            timeout = schedule_timeout(timeout);
4618            spin_lock_irq(&x->wait.lock);
4619        } while (!x->done && timeout);
4620        __remove_wait_queue(&x->wait, &wait);
4621        if (!x->done)
4622            return timeout;
4623    }
4624    x->done--;
4625    return timeout ?: 1;
4626}
4627
4628static long __sched
4629wait_for_common(struct completion *x, long timeout, int state)
4630{
4631    might_sleep();
4632
4633    spin_lock_irq(&x->wait.lock);
4634    timeout = do_wait_for_common(x, timeout, state);
4635    spin_unlock_irq(&x->wait.lock);
4636    return timeout;
4637}
4638
4639/**
4640 * wait_for_completion: - waits for completion of a task
4641 * @x: holds the state of this particular completion
4642 *
4643 * This waits to be signaled for completion of a specific task. It is NOT
4644 * interruptible and there is no timeout.
4645 *
4646 * See also similar routines (i.e. wait_for_completion_timeout()) with timeout
4647 * and interrupt capability. Also see complete().
4648 */
4649void __sched wait_for_completion(struct completion *x)
4650{
4651    wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE);
4652}
4653EXPORT_SYMBOL(wait_for_completion);
4654
4655/**
4656 * wait_for_completion_timeout: - waits for completion of a task (w/timeout)
4657 * @x: holds the state of this particular completion
4658 * @timeout: timeout value in jiffies
4659 *
4660 * This waits for either a completion of a specific task to be signaled or for a
4661 * specified timeout to expire. The timeout is in jiffies. It is not
4662 * interruptible.
4663 */
4664unsigned long __sched
4665wait_for_completion_timeout(struct completion *x, unsigned long timeout)
4666{
4667    return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE);
4668}
4669EXPORT_SYMBOL(wait_for_completion_timeout);
4670
4671/**
4672 * wait_for_completion_interruptible: - waits for completion of a task (w/intr)
4673 * @x: holds the state of this particular completion
4674 *
4675 * This waits for completion of a specific task to be signaled. It is
4676 * interruptible.
4677 */
4678int __sched wait_for_completion_interruptible(struct completion *x)
4679{
4680    long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE);
4681    if (t == -ERESTARTSYS)
4682        return t;
4683    return 0;
4684}
4685EXPORT_SYMBOL(wait_for_completion_interruptible);
4686
4687/**
4688 * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr))
4689 * @x: holds the state of this particular completion
4690 * @timeout: timeout value in jiffies
4691 *
4692 * This waits for either a completion of a specific task to be signaled or for a
4693 * specified timeout to expire. It is interruptible. The timeout is in jiffies.
4694 */
4695long __sched
4696wait_for_completion_interruptible_timeout(struct completion *x,
4697                      unsigned long timeout)
4698{
4699    return wait_for_common(x, timeout, TASK_INTERRUPTIBLE);
4700}
4701EXPORT_SYMBOL(wait_for_completion_interruptible_timeout);
4702
4703/**
4704 * wait_for_completion_killable: - waits for completion of a task (killable)
4705 * @x: holds the state of this particular completion
4706 *
4707 * This waits to be signaled for completion of a specific task. It can be
4708 * interrupted by a kill signal.
4709 */
4710int __sched wait_for_completion_killable(struct completion *x)
4711{
4712    long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE);
4713    if (t == -ERESTARTSYS)
4714        return t;
4715    return 0;
4716}
4717EXPORT_SYMBOL(wait_for_completion_killable);
4718
4719/**
4720 * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable))
4721 * @x: holds the state of this particular completion
4722 * @timeout: timeout value in jiffies
4723 *
4724 * This waits for either a completion of a specific task to be
4725 * signaled or for a specified timeout to expire. It can be
4726 * interrupted by a kill signal. The timeout is in jiffies.
4727 */
4728long __sched
4729wait_for_completion_killable_timeout(struct completion *x,
4730                     unsigned long timeout)
4731{
4732    return wait_for_common(x, timeout, TASK_KILLABLE);
4733}
4734EXPORT_SYMBOL(wait_for_completion_killable_timeout);
4735
4736/**
4737 * try_wait_for_completion - try to decrement a completion without blocking
4738 * @x: completion structure
4739 *
4740 * Returns: 0 if a decrement cannot be done without blocking
4741 * 1 if a decrement succeeded.
4742 *
4743 * If a completion is being used as a counting completion,
4744 * attempt to decrement the counter without blocking. This
4745 * enables us to avoid waiting if the resource the completion
4746 * is protecting is not available.
4747 */
4748bool try_wait_for_completion(struct completion *x)
4749{
4750    unsigned long flags;
4751    int ret = 1;
4752
4753    spin_lock_irqsave(&x->wait.lock, flags);
4754    if (!x->done)
4755        ret = 0;
4756    else
4757        x->done--;
4758    spin_unlock_irqrestore(&x->wait.lock, flags);
4759    return ret;
4760}
4761EXPORT_SYMBOL(try_wait_for_completion);
4762
4763/**
4764 * completion_done - Test to see if a completion has any waiters
4765 * @x: completion structure
4766 *
4767 * Returns: 0 if there are waiters (wait_for_completion() in progress)
4768 * 1 if there are no waiters.
4769 *
4770 */
4771bool completion_done(struct completion *x)
4772{
4773    unsigned long flags;
4774    int ret = 1;
4775
4776    spin_lock_irqsave(&x->wait.lock, flags);
4777    if (!x->done)
4778        ret = 0;
4779    spin_unlock_irqrestore(&x->wait.lock, flags);
4780    return ret;
4781}
4782EXPORT_SYMBOL(completion_done);
4783
4784static long __sched
4785sleep_on_common(wait_queue_head_t *q, int state, long timeout)
4786{
4787    unsigned long flags;
4788    wait_queue_t wait;
4789
4790    init_waitqueue_entry(&wait, current);
4791
4792    __set_current_state(state);
4793
4794    spin_lock_irqsave(&q->lock, flags);
4795    __add_wait_queue(q, &wait);
4796    spin_unlock(&q->lock);
4797    timeout = schedule_timeout(timeout);
4798    spin_lock_irq(&q->lock);
4799    __remove_wait_queue(q, &wait);
4800    spin_unlock_irqrestore(&q->lock, flags);
4801
4802    return timeout;
4803}
4804
4805void __sched interruptible_sleep_on(wait_queue_head_t *q)
4806{
4807    sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
4808}
4809EXPORT_SYMBOL(interruptible_sleep_on);
4810
4811long __sched
4812interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout)
4813{
4814    return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout);
4815}
4816EXPORT_SYMBOL(interruptible_sleep_on_timeout);
4817
4818void __sched sleep_on(wait_queue_head_t *q)
4819{
4820    sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
4821}
4822EXPORT_SYMBOL(sleep_on);
4823
4824long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout)
4825{
4826    return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout);
4827}
4828EXPORT_SYMBOL(sleep_on_timeout);
4829
4830#ifdef CONFIG_RT_MUTEXES
4831
4832/*
4833 * rt_mutex_setprio - set the current priority of a task
4834 * @p: task
4835 * @prio: prio value (kernel-internal form)
4836 *
4837 * This function changes the 'effective' priority of a task. It does
4838 * not touch ->normal_prio like __setscheduler().
4839 *
4840 * Used by the rt_mutex code to implement priority inheritance logic.
4841 */
4842void rt_mutex_setprio(struct task_struct *p, int prio)
4843{
4844    int oldprio, on_rq, running;
4845    struct rq *rq;
4846    const struct sched_class *prev_class;
4847
4848    BUG_ON(prio < 0 || prio > MAX_PRIO);
4849
4850    rq = __task_rq_lock(p);
4851
4852    trace_sched_pi_setprio(p, prio);
4853    oldprio = p->prio;
4854    prev_class = p->sched_class;
4855    on_rq = p->on_rq;
4856    running = task_current(rq, p);
4857    if (on_rq)
4858        dequeue_task(rq, p, 0);
4859    if (running)
4860        p->sched_class->put_prev_task(rq, p);
4861
4862    if (rt_prio(prio))
4863        p->sched_class = &rt_sched_class;
4864    else
4865        p->sched_class = &fair_sched_class;
4866
4867    p->prio = prio;
4868
4869    if (running)
4870        p->sched_class->set_curr_task(rq);
4871    if (on_rq)
4872        enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0);
4873
4874    check_class_changed(rq, p, prev_class, oldprio);
4875    __task_rq_unlock(rq);
4876}
4877
4878#endif
4879
4880void set_user_nice(struct task_struct *p, long nice)
4881{
4882    int old_prio, delta, on_rq;
4883    unsigned long flags;
4884    struct rq *rq;
4885
4886    if (TASK_NICE(p) == nice || nice < -20 || nice > 19)
4887        return;
4888    /*
4889     * We have to be careful, if called from sys_setpriority(),
4890     * the task might be in the middle of scheduling on another CPU.
4891     */
4892    rq = task_rq_lock(p, &flags);
4893    /*
4894     * The RT priorities are set via sched_setscheduler(), but we still
4895     * allow the 'normal' nice value to be set - but as expected
4896     * it wont have any effect on scheduling until the task is
4897     * SCHED_FIFO/SCHED_RR:
4898     */
4899    if (task_has_rt_policy(p)) {
4900        p->static_prio = NICE_TO_PRIO(nice);
4901        goto out_unlock;
4902    }
4903    on_rq = p->on_rq;
4904    if (on_rq)
4905        dequeue_task(rq, p, 0);
4906
4907    p->static_prio = NICE_TO_PRIO(nice);
4908    set_load_weight(p);
4909    old_prio = p->prio;
4910    p->prio = effective_prio(p);
4911    delta = p->prio - old_prio;
4912
4913    if (on_rq) {
4914        enqueue_task(rq, p, 0);
4915        /*
4916         * If the task increased its priority or is running and
4917         * lowered its priority, then reschedule its CPU:
4918         */
4919        if (delta < 0 || (delta > 0 && task_running(rq, p)))
4920            resched_task(rq->curr);
4921    }
4922out_unlock:
4923    task_rq_unlock(rq, p, &flags);
4924}
4925EXPORT_SYMBOL(set_user_nice);
4926
4927/*
4928 * can_nice - check if a task can reduce its nice value
4929 * @p: task
4930 * @nice: nice value
4931 */
4932int can_nice(const struct task_struct *p, const int nice)
4933{
4934    /* convert nice value [19,-20] to rlimit style value [1,40] */
4935    int nice_rlim = 20 - nice;
4936
4937    return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) ||
4938        capable(CAP_SYS_NICE));
4939}
4940
4941#ifdef __ARCH_WANT_SYS_NICE
4942
4943/*
4944 * sys_nice - change the priority of the current process.
4945 * @increment: priority increment
4946 *
4947 * sys_setpriority is a more generic, but much slower function that
4948 * does similar things.
4949 */
4950SYSCALL_DEFINE1(nice, int, increment)
4951{
4952    long nice, retval;
4953
4954    /*
4955     * Setpriority might change our priority at the same moment.
4956     * We don't have to worry. Conceptually one call occurs first
4957     * and we have a single winner.
4958     */
4959    if (increment < -40)
4960        increment = -40;
4961    if (increment > 40)
4962        increment = 40;
4963
4964    nice = TASK_NICE(current) + increment;
4965    if (nice < -20)
4966        nice = -20;
4967    if (nice > 19)
4968        nice = 19;
4969
4970    if (increment < 0 && !can_nice(current, nice))
4971        return -EPERM;
4972
4973    retval = security_task_setnice(current, nice);
4974    if (retval)
4975        return retval;
4976
4977    set_user_nice(current, nice);
4978    return 0;
4979}
4980
4981#endif
4982
4983/**
4984 * task_prio - return the priority value of a given task.
4985 * @p: the task in question.
4986 *
4987 * This is the priority value as seen by users in /proc.
4988 * RT tasks are offset by -200. Normal tasks are centered
4989 * around 0, value goes from -16 to +15.
4990 */
4991int task_prio(const struct task_struct *p)
4992{
4993    return p->prio - MAX_RT_PRIO;
4994}
4995
4996/**
4997 * task_nice - return the nice value of a given task.
4998 * @p: the task in question.
4999 */
5000int task_nice(const struct task_struct *p)
5001{
5002    return TASK_NICE(p);
5003}
5004EXPORT_SYMBOL(task_nice);
5005
5006/**
5007 * idle_cpu - is a given cpu idle currently?
5008 * @cpu: the processor in question.
5009 */
5010int idle_cpu(int cpu)
5011{
5012    return cpu_curr(cpu) == cpu_rq(cpu)->idle;
5013}
5014
5015/**
5016 * idle_task - return the idle task for a given cpu.
5017 * @cpu: the processor in question.
5018 */
5019struct task_struct *idle_task(int cpu)
5020{
5021    return cpu_rq(cpu)->idle;
5022}
5023
5024/**
5025 * find_process_by_pid - find a process with a matching PID value.
5026 * @pid: the pid in question.
5027 */
5028static struct task_struct *find_process_by_pid(pid_t pid)
5029{
5030    return pid ? find_task_by_vpid(pid) : current;
5031}
5032
5033/* Actually do priority change: must hold rq lock. */
5034static void
5035__setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio)
5036{
5037    p->policy = policy;
5038    p->rt_priority = prio;
5039    p->normal_prio = normal_prio(p);
5040    /* we are holding p->pi_lock already */
5041    p->prio = rt_mutex_getprio(p);
5042    if (rt_prio(p->prio))
5043        p->sched_class = &rt_sched_class;
5044    else
5045        p->sched_class = &fair_sched_class;
5046    set_load_weight(p);
5047}
5048
5049/*
5050 * check the target process has a UID that matches the current process's
5051 */
5052static bool check_same_owner(struct task_struct *p)
5053{
5054    const struct cred *cred = current_cred(), *pcred;
5055    bool match;
5056
5057    rcu_read_lock();
5058    pcred = __task_cred(p);
5059    if (cred->user->user_ns == pcred->user->user_ns)
5060        match = (cred->euid == pcred->euid ||
5061             cred->euid == pcred->uid);
5062    else
5063        match = false;
5064    rcu_read_unlock();
5065    return match;
5066}
5067
5068static int __sched_setscheduler(struct task_struct *p, int policy,
5069                const struct sched_param *param, bool user)
5070{
5071    int retval, oldprio, oldpolicy = -1, on_rq, running;
5072    unsigned long flags;
5073    const struct sched_class *prev_class;
5074    struct rq *rq;
5075    int reset_on_fork;
5076
5077    /* may grab non-irq protected spin_locks */
5078    BUG_ON(in_interrupt());
5079recheck:
5080    /* double check policy once rq lock held */
5081    if (policy < 0) {
5082        reset_on_fork = p->sched_reset_on_fork;
5083        policy = oldpolicy = p->policy;
5084    } else {
5085        reset_on_fork = !!(policy & SCHED_RESET_ON_FORK);
5086        policy &= ~SCHED_RESET_ON_FORK;
5087
5088        if (policy != SCHED_FIFO && policy != SCHED_RR &&
5089                policy != SCHED_NORMAL && policy != SCHED_BATCH &&
5090                policy != SCHED_IDLE)
5091            return -EINVAL;
5092    }
5093
5094    /*
5095     * Valid priorities for SCHED_FIFO and SCHED_RR are
5096     * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL,
5097     * SCHED_BATCH and SCHED_IDLE is 0.
5098     */
5099    if (param->sched_priority < 0 ||
5100        (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) ||
5101        (!p->mm && param->sched_priority > MAX_RT_PRIO-1))
5102        return -EINVAL;
5103    if (rt_policy(policy) != (param->sched_priority != 0))
5104        return -EINVAL;
5105
5106    /*
5107     * Allow unprivileged RT tasks to decrease priority:
5108     */
5109    if (user && !capable(CAP_SYS_NICE)) {
5110        if (rt_policy(policy)) {
5111            unsigned long rlim_rtprio =
5112                    task_rlimit(p, RLIMIT_RTPRIO);
5113
5114            /* can't set/change the rt policy */
5115            if (policy != p->policy && !rlim_rtprio)
5116                return -EPERM;
5117
5118            /* can't increase priority */
5119            if (param->sched_priority > p->rt_priority &&
5120                param->sched_priority > rlim_rtprio)
5121                return -EPERM;
5122        }
5123
5124        /*
5125         * Treat SCHED_IDLE as nice 20. Only allow a switch to
5126         * SCHED_NORMAL if the RLIMIT_NICE would normally permit it.
5127         */
5128        if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) {
5129            if (!can_nice(p, TASK_NICE(p)))
5130                return -EPERM;
5131        }
5132
5133        /* can't change other user's priorities */
5134        if (!check_same_owner(p))
5135            return -EPERM;
5136
5137        /* Normal users shall not reset the sched_reset_on_fork flag */
5138        if (p->sched_reset_on_fork && !reset_on_fork)
5139            return -EPERM;
5140    }
5141
5142    if (user) {
5143        retval = security_task_setscheduler(p);
5144        if (retval)
5145            return retval;
5146    }
5147
5148    /*
5149     * make sure no PI-waiters arrive (or leave) while we are
5150     * changing the priority of the task:
5151     *
5152     * To be able to change p->policy safely, the appropriate
5153     * runqueue lock must be held.
5154     */
5155    rq = task_rq_lock(p, &flags);
5156
5157    /*
5158     * Changing the policy of the stop threads its a very bad idea
5159     */
5160    if (p == rq->stop) {
5161        task_rq_unlock(rq, p, &flags);
5162        return -EINVAL;
5163    }
5164
5165    /*
5166     * If not changing anything there's no need to proceed further:
5167     */
5168    if (unlikely(policy == p->policy && (!rt_policy(policy) ||
5169            param->sched_priority == p->rt_priority))) {
5170
5171        __task_rq_unlock(rq);
5172        raw_spin_unlock_irqrestore(&p->pi_lock, flags);
5173        return 0;
5174    }
5175
5176#ifdef CONFIG_RT_GROUP_SCHED
5177    if (user) {
5178        /*
5179         * Do not allow realtime tasks into groups that have no runtime
5180         * assigned.
5181         */
5182        if (rt_bandwidth_enabled() && rt_policy(policy) &&
5183                task_group(p)->rt_bandwidth.rt_runtime == 0 &&
5184                !task_group_is_autogroup(task_group(p))) {
5185            task_rq_unlock(rq, p, &flags);
5186            return -EPERM;
5187        }
5188    }
5189#endif
5190
5191    /* recheck policy now with rq lock held */
5192    if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) {
5193        policy = oldpolicy = -1;
5194        task_rq_unlock(rq, p, &flags);
5195        goto recheck;
5196    }
5197    on_rq = p->on_rq;
5198    running = task_current(rq, p);
5199    if (on_rq)
5200        deactivate_task(rq, p, 0);
5201    if (running)
5202        p->sched_class->put_prev_task(rq, p);
5203
5204    p->sched_reset_on_fork = reset_on_fork;
5205
5206    oldprio = p->prio;
5207    prev_class = p->sched_class;
5208    __setscheduler(rq, p, policy, param->sched_priority);
5209
5210    if (running)
5211        p->sched_class->set_curr_task(rq);
5212    if (on_rq)
5213        activate_task(rq, p, 0);
5214
5215    check_class_changed(rq, p, prev_class, oldprio);
5216    task_rq_unlock(rq, p, &flags);
5217
5218    rt_mutex_adjust_pi(p);
5219
5220    return 0;
5221}
5222
5223/**
5224 * sched_setscheduler - change the scheduling policy and/or RT priority of a thread.
5225 * @p: the task in question.
5226 * @policy: new policy.
5227 * @param: structure containing the new RT priority.
5228 *
5229 * NOTE that the task may be already dead.
5230 */
5231int sched_setscheduler(struct task_struct *p, int policy,
5232               const struct sched_param *param)
5233{
5234    return __sched_setscheduler(p, policy, param, true);
5235}
5236EXPORT_SYMBOL_GPL(sched_setscheduler);
5237
5238/**
5239 * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace.
5240 * @p: the task in question.
5241 * @policy: new policy.
5242 * @param: structure containing the new RT priority.
5243 *
5244 * Just like sched_setscheduler, only don't bother checking if the
5245 * current context has permission. For example, this is needed in
5246 * stop_machine(): we create temporary high priority worker threads,
5247 * but our caller might not have that capability.
5248 */
5249int sched_setscheduler_nocheck(struct task_struct *p, int policy,
5250                   const struct sched_param *param)
5251{
5252    return __sched_setscheduler(p, policy, param, false);
5253}
5254
5255static int
5256do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param)
5257{
5258    struct sched_param lparam;
5259    struct task_struct *p;
5260    int retval;
5261
5262    if (!param || pid < 0)
5263        return -EINVAL;
5264    if (copy_from_user(&lparam, param, sizeof(struct sched_param)))
5265        return -EFAULT;
5266
5267    rcu_read_lock();
5268    retval = -ESRCH;
5269    p = find_process_by_pid(pid);
5270    if (p != NULL)
5271        retval = sched_setscheduler(p, policy, &lparam);
5272    rcu_read_unlock();
5273
5274    return retval;
5275}
5276
5277/**
5278 * sys_sched_setscheduler - set/change the scheduler policy and RT priority
5279 * @pid: the pid in question.
5280 * @policy: new policy.
5281 * @param: structure containing the new RT priority.
5282 */
5283SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy,
5284        struct sched_param __user *, param)
5285{
5286    /* negative values for policy are not valid */
5287    if (policy < 0)
5288        return -EINVAL;
5289
5290    return do_sched_setscheduler(pid, policy, param);
5291}
5292
5293/**
5294 * sys_sched_setparam - set/change the RT priority of a thread
5295 * @pid: the pid in question.
5296 * @param: structure containing the new RT priority.
5297 */
5298SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param)
5299{
5300    return do_sched_setscheduler(pid, -1, param);
5301}
5302
5303/**
5304 * sys_sched_getscheduler - get the policy (scheduling class) of a thread
5305 * @pid: the pid in question.
5306 */
5307SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid)
5308{
5309    struct task_struct *p;
5310    int retval;
5311
5312    if (pid < 0)
5313        return -EINVAL;
5314
5315    retval = -ESRCH;
5316    rcu_read_lock();
5317    p = find_process_by_pid(pid);
5318    if (p) {
5319        retval = security_task_getscheduler(p);
5320        if (!retval)
5321            retval = p->policy
5322                | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0);
5323    }
5324    rcu_read_unlock();
5325    return retval;
5326}
5327
5328/**
5329 * sys_sched_getparam - get the RT priority of a thread
5330 * @pid: the pid in question.
5331 * @param: structure containing the RT priority.
5332 */
5333SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param)
5334{
5335    struct sched_param lp;
5336    struct task_struct *p;
5337    int retval;
5338
5339    if (!param || pid < 0)
5340        return -EINVAL;
5341
5342    rcu_read_lock();
5343    p = find_process_by_pid(pid);
5344    retval = -ESRCH;
5345    if (!p)
5346        goto out_unlock;
5347
5348    retval = security_task_getscheduler(p);
5349    if (retval)
5350        goto out_unlock;
5351
5352    lp.sched_priority = p->rt_priority;
5353    rcu_read_unlock();
5354
5355    /*
5356     * This one might sleep, we cannot do it with a spinlock held ...
5357     */
5358    retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0;
5359
5360    return retval;
5361
5362out_unlock:
5363    rcu_read_unlock();
5364    return retval;
5365}
5366
5367long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)
5368{
5369    cpumask_var_t cpus_allowed, new_mask;
5370    struct task_struct *p;
5371    int retval;
5372
5373    get_online_cpus();
5374    rcu_read_lock();
5375
5376    p = find_process_by_pid(pid);
5377    if (!p) {
5378        rcu_read_unlock();
5379        put_online_cpus();
5380        return -ESRCH;
5381    }
5382
5383    /* Prevent p going away */
5384    get_task_struct(p);
5385    rcu_read_unlock();
5386
5387    if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) {
5388        retval = -ENOMEM;
5389        goto out_put_task;
5390    }
5391    if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) {
5392        retval = -ENOMEM;
5393        goto out_free_cpus_allowed;
5394    }
5395    retval = -EPERM;
5396    if (!check_same_owner(p) && !task_ns_capable(p, CAP_SYS_NICE))
5397        goto out_unlock;
5398
5399    retval = security_task_setscheduler(p);
5400    if (retval)
5401        goto out_unlock;
5402
5403    cpuset_cpus_allowed(p, cpus_allowed);
5404    cpumask_and(new_mask, in_mask, cpus_allowed);
5405again:
5406    retval = set_cpus_allowed_ptr(p, new_mask);
5407
5408    if (!retval) {
5409        cpuset_cpus_allowed(p, cpus_allowed);
5410        if (!cpumask_subset(new_mask, cpus_allowed)) {
5411            /*
5412             * We must have raced with a concurrent cpuset
5413             * update. Just reset the cpus_allowed to the
5414             * cpuset's cpus_allowed
5415             */
5416            cpumask_copy(new_mask, cpus_allowed);
5417            goto again;
5418        }
5419    }
5420out_unlock:
5421    free_cpumask_var(new_mask);
5422out_free_cpus_allowed:
5423    free_cpumask_var(cpus_allowed);
5424out_put_task:
5425    put_task_struct(p);
5426    put_online_cpus();
5427    return retval;
5428}
5429
5430static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len,
5431                 struct cpumask *new_mask)
5432{
5433    if (len < cpumask_size())
5434        cpumask_clear(new_mask);
5435    else if (len > cpumask_size())
5436        len = cpumask_size();
5437
5438    return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0;
5439}
5440
5441/**
5442 * sys_sched_setaffinity - set the cpu affinity of a process
5443 * @pid: pid of the process
5444 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
5445 * @user_mask_ptr: user-space pointer to the new cpu mask
5446 */
5447SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len,
5448        unsigned long __user *, user_mask_ptr)
5449{
5450    cpumask_var_t new_mask;
5451    int retval;
5452
5453    if (!alloc_cpumask_var(&new_mask, GFP_KERNEL))
5454        return -ENOMEM;
5455
5456    retval = get_user_cpu_mask(user_mask_ptr, len, new_mask);
5457    if (retval == 0)
5458        retval = sched_setaffinity(pid, new_mask);
5459    free_cpumask_var(new_mask);
5460    return retval;
5461}
5462
5463long sched_getaffinity(pid_t pid, struct cpumask *mask)
5464{
5465    struct task_struct *p;
5466    unsigned long flags;
5467    int retval;
5468
5469    get_online_cpus();
5470    rcu_read_lock();
5471
5472    retval = -ESRCH;
5473    p = find_process_by_pid(pid);
5474    if (!p)
5475        goto out_unlock;
5476
5477    retval = security_task_getscheduler(p);
5478    if (retval)
5479        goto out_unlock;
5480
5481    raw_spin_lock_irqsave(&p->pi_lock, flags);
5482    cpumask_and(mask, &p->cpus_allowed, cpu_online_mask);
5483    raw_spin_unlock_irqrestore(&p->pi_lock, flags);
5484
5485out_unlock:
5486    rcu_read_unlock();
5487    put_online_cpus();
5488
5489    return retval;
5490}
5491
5492/**
5493 * sys_sched_getaffinity - get the cpu affinity of a process
5494 * @pid: pid of the process
5495 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
5496 * @user_mask_ptr: user-space pointer to hold the current cpu mask
5497 */
5498SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len,
5499        unsigned long __user *, user_mask_ptr)
5500{
5501    int ret;
5502    cpumask_var_t mask;
5503
5504    if ((len * BITS_PER_BYTE) < nr_cpu_ids)
5505        return -EINVAL;
5506    if (len & (sizeof(unsigned long)-1))
5507        return -EINVAL;
5508
5509    if (!alloc_cpumask_var(&mask, GFP_KERNEL))
5510        return -ENOMEM;
5511
5512    ret = sched_getaffinity(pid, mask);
5513    if (ret == 0) {
5514        size_t retlen = min_t(size_t, len, cpumask_size());
5515
5516        if (copy_to_user(user_mask_ptr, mask, retlen))
5517            ret = -EFAULT;
5518        else
5519            ret = retlen;
5520    }
5521    free_cpumask_var(mask);
5522
5523    return ret;
5524}
5525
5526/**
5527 * sys_sched_yield - yield the current processor to other threads.
5528 *
5529 * This function yields the current CPU to other tasks. If there are no
5530 * other threads running on this CPU then this function will return.
5531 */
5532SYSCALL_DEFINE0(sched_yield)
5533{
5534    struct rq *rq = this_rq_lock();
5535
5536    schedstat_inc(rq, yld_count);
5537    current->sched_class->yield_task(rq);
5538
5539    /*
5540     * Since we are going to call schedule() anyway, there's
5541     * no need to preempt or enable interrupts:
5542     */
5543    __release(rq->lock);
5544    spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
5545    do_raw_spin_unlock(&rq->lock);
5546    preempt_enable_no_resched();
5547
5548    schedule();
5549
5550    return 0;
5551}
5552
5553static inline int should_resched(void)
5554{
5555    return need_resched() && !(preempt_count() & PREEMPT_ACTIVE);
5556}
5557
5558static void __cond_resched(void)
5559{
5560    add_preempt_count(PREEMPT_ACTIVE);
5561    schedule();
5562    sub_preempt_count(PREEMPT_ACTIVE);
5563}
5564
5565int __sched _cond_resched(void)
5566{
5567    if (should_resched()) {
5568        __cond_resched();
5569        return 1;
5570    }
5571    return 0;
5572}
5573EXPORT_SYMBOL(_cond_resched);
5574
5575/*
5576 * __cond_resched_lock() - if a reschedule is pending, drop the given lock,
5577 * call schedule, and on return reacquire the lock.
5578 *
5579 * This works OK both with and without CONFIG_PREEMPT. We do strange low-level
5580 * operations here to prevent schedule() from being called twice (once via
5581 * spin_unlock(), once by hand).
5582 */
5583int __cond_resched_lock(spinlock_t *lock)
5584{
5585    int resched = should_resched();
5586    int ret = 0;
5587
5588    lockdep_assert_held(lock);
5589
5590    if (spin_needbreak(lock) || resched) {
5591        spin_unlock(lock);
5592        if (resched)
5593            __cond_resched();
5594        else
5595            cpu_relax();
5596        ret = 1;
5597        spin_lock(lock);
5598    }
5599    return ret;
5600}
5601EXPORT_SYMBOL(__cond_resched_lock);
5602
5603int __sched __cond_resched_softirq(void)
5604{
5605    BUG_ON(!in_softirq());
5606
5607    if (should_resched()) {
5608        local_bh_enable();
5609        __cond_resched();
5610        local_bh_disable();
5611        return 1;
5612    }
5613    return 0;
5614}
5615EXPORT_SYMBOL(__cond_resched_softirq);
5616
5617/**
5618 * yield - yield the current processor to other threads.
5619 *
5620 * This is a shortcut for kernel-space yielding - it marks the
5621 * thread runnable and calls sys_sched_yield().
5622 */
5623void __sched yield(void)
5624{
5625    set_current_state(TASK_RUNNING);
5626    sys_sched_yield();
5627}
5628EXPORT_SYMBOL(yield);
5629
5630/**
5631 * yield_to - yield the current processor to another thread in
5632 * your thread group, or accelerate that thread toward the
5633 * processor it's on.
5634 * @p: target task
5635 * @preempt: whether task preemption is allowed or not
5636 *
5637 * It's the caller's job to ensure that the target task struct
5638 * can't go away on us before we can do any checks.
5639 *
5640 * Returns true if we indeed boosted the target task.
5641 */
5642bool __sched yield_to(struct task_struct *p, bool preempt)
5643{
5644    struct task_struct *curr = current;
5645    struct rq *rq, *p_rq;
5646    unsigned long flags;
5647    bool yielded = 0;
5648
5649    local_irq_save(flags);
5650    rq = this_rq();
5651
5652again:
5653    p_rq = task_rq(p);
5654    double_rq_lock(rq, p_rq);
5655    while (task_rq(p) != p_rq) {
5656        double_rq_unlock(rq, p_rq);
5657        goto again;
5658    }
5659
5660    if (!curr->sched_class->yield_to_task)
5661        goto out;
5662
5663    if (curr->sched_class != p->sched_class)
5664        goto out;
5665
5666    if (task_running(p_rq, p) || p->state)
5667        goto out;
5668
5669    yielded = curr->sched_class->yield_to_task(rq, p, preempt);
5670    if (yielded) {
5671        schedstat_inc(rq, yld_count);
5672        /*
5673         * Make p's CPU reschedule; pick_next_entity takes care of
5674         * fairness.
5675         */
5676        if (preempt && rq != p_rq)
5677            resched_task(p_rq->curr);
5678    }
5679
5680out:
5681    double_rq_unlock(rq, p_rq);
5682    local_irq_restore(flags);
5683
5684    if (yielded)
5685        schedule();
5686
5687    return yielded;
5688}
5689EXPORT_SYMBOL_GPL(yield_to);
5690
5691/*
5692 * This task is about to go to sleep on IO. Increment rq->nr_iowait so
5693 * that process accounting knows that this is a task in IO wait state.
5694 */
5695void __sched io_schedule(void)
5696{
5697    struct rq *rq = raw_rq();
5698
5699    delayacct_blkio_start();
5700    atomic_inc(&rq->nr_iowait);
5701    blk_flush_plug(current);
5702    current->in_iowait = 1;
5703    schedule();
5704    current->in_iowait = 0;
5705    atomic_dec(&rq->nr_iowait);
5706    delayacct_blkio_end();
5707}
5708EXPORT_SYMBOL(io_schedule);
5709
5710long __sched io_schedule_timeout(long timeout)
5711{
5712    struct rq *rq = raw_rq();
5713    long ret;
5714
5715    delayacct_blkio_start();
5716    atomic_inc(&rq->nr_iowait);
5717    blk_flush_plug(current);
5718    current->in_iowait = 1;
5719    ret = schedule_timeout(timeout);
5720    current->in_iowait = 0;
5721    atomic_dec(&rq->nr_iowait);
5722    delayacct_blkio_end();
5723    return ret;
5724}
5725
5726/**
5727 * sys_sched_get_priority_max - return maximum RT priority.
5728 * @policy: scheduling class.
5729 *
5730 * this syscall returns the maximum rt_priority that can be used
5731 * by a given scheduling class.
5732 */
5733SYSCALL_DEFINE1(sched_get_priority_max, int, policy)
5734{
5735    int ret = -EINVAL;
5736
5737    switch (policy) {
5738    case SCHED_FIFO:
5739    case SCHED_RR:
5740        ret = MAX_USER_RT_PRIO-1;
5741        break;
5742    case SCHED_NORMAL:
5743    case SCHED_BATCH:
5744    case SCHED_IDLE:
5745        ret = 0;
5746        break;
5747    }
5748    return ret;
5749}
5750
5751/**
5752 * sys_sched_get_priority_min - return minimum RT priority.
5753 * @policy: scheduling class.
5754 *
5755 * this syscall returns the minimum rt_priority that can be used
5756 * by a given scheduling class.
5757 */
5758SYSCALL_DEFINE1(sched_get_priority_min, int, policy)
5759{
5760    int ret = -EINVAL;
5761
5762    switch (policy) {
5763    case SCHED_FIFO:
5764    case SCHED_RR:
5765        ret = 1;
5766        break;
5767    case SCHED_NORMAL:
5768    case SCHED_BATCH:
5769    case SCHED_IDLE:
5770        ret = 0;
5771    }
5772    return ret;
5773}
5774
5775/**
5776 * sys_sched_rr_get_interval - return the default timeslice of a process.
5777 * @pid: pid of the process.
5778 * @interval: userspace pointer to the timeslice value.
5779 *
5780 * this syscall writes the default timeslice value of a given process
5781 * into the user-space timespec buffer. A value of '0' means infinity.
5782 */
5783SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid,
5784        struct timespec __user *, interval)
5785{
5786    struct task_struct *p;
5787    unsigned int time_slice;
5788    unsigned long flags;
5789    struct rq *rq;
5790    int retval;
5791    struct timespec t;
5792
5793    if (pid < 0)
5794        return -EINVAL;
5795
5796    retval = -ESRCH;
5797    rcu_read_lock();
5798    p = find_process_by_pid(pid);
5799    if (!p)
5800        goto out_unlock;
5801
5802    retval = security_task_getscheduler(p);
5803    if (retval)
5804        goto out_unlock;
5805
5806    rq = task_rq_lock(p, &flags);
5807    time_slice = p->sched_class->get_rr_interval(rq, p);
5808    task_rq_unlock(rq, p, &flags);
5809
5810    rcu_read_unlock();
5811    jiffies_to_timespec(time_slice, &t);
5812    retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0;
5813    return retval;
5814
5815out_unlock:
5816    rcu_read_unlock();
5817    return retval;
5818}
5819
5820static const char stat_nam[] = TASK_STATE_TO_CHAR_STR;
5821
5822void sched_show_task(struct task_struct *p)
5823{
5824    unsigned long free = 0;
5825    unsigned state;
5826
5827    state = p->state ? __ffs(p->state) + 1 : 0;
5828    printk(KERN_INFO "%-15.15s %c", p->comm,
5829        state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?');
5830#if BITS_PER_LONG == 32
5831    if (state == TASK_RUNNING)
5832        printk(KERN_CONT " running ");
5833    else
5834        printk(KERN_CONT " %08lx ", thread_saved_pc(p));
5835#else
5836    if (state == TASK_RUNNING)
5837        printk(KERN_CONT " running task ");
5838    else
5839        printk(KERN_CONT " %016lx ", thread_saved_pc(p));
5840#endif
5841#ifdef CONFIG_DEBUG_STACK_USAGE
5842    free = stack_not_used(p);
5843#endif
5844    printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free,
5845        task_pid_nr(p), task_pid_nr(p->real_parent),
5846        (unsigned long)task_thread_info(p)->flags);
5847
5848    show_stack(p, NULL);
5849}
5850
5851void show_state_filter(unsigned long state_filter)
5852{
5853    struct task_struct *g, *p;
5854
5855#if BITS_PER_LONG == 32
5856    printk(KERN_INFO
5857        " task PC stack pid father\n");
5858#else
5859    printk(KERN_INFO
5860        " task PC stack pid father\n");
5861#endif
5862    read_lock(&tasklist_lock);
5863    do_each_thread(g, p) {
5864        /*
5865         * reset the NMI-timeout, listing all files on a slow
5866         * console might take a lot of time:
5867         */
5868        touch_nmi_watchdog();
5869        if (!state_filter || (p->state & state_filter))
5870            sched_show_task(p);
5871    } while_each_thread(g, p);
5872
5873    touch_all_softlockup_watchdogs();
5874
5875#ifdef CONFIG_SCHED_DEBUG
5876    sysrq_sched_debug_show();
5877#endif
5878    read_unlock(&tasklist_lock);
5879    /*
5880     * Only show locks if all tasks are dumped:
5881     */
5882    if (!state_filter)
5883        debug_show_all_locks();
5884}
5885
5886void __cpuinit init_idle_bootup_task(struct task_struct *idle)
5887{
5888    idle->sched_class = &idle_sched_class;
5889}
5890
5891/**
5892 * init_idle - set up an idle thread for a given CPU
5893 * @idle: task in question
5894 * @cpu: cpu the idle task belongs to
5895 *
5896 * NOTE: this function does not set the idle thread's NEED_RESCHED
5897 * flag, to make booting more robust.
5898 */
5899void __cpuinit init_idle(struct task_struct *idle, int cpu)
5900{
5901    struct rq *rq = cpu_rq(cpu);
5902    unsigned long flags;
5903
5904    raw_spin_lock_irqsave(&rq->lock, flags);
5905
5906    __sched_fork(idle);
5907    idle->state = TASK_RUNNING;
5908    idle->se.exec_start = sched_clock();
5909
5910    do_set_cpus_allowed(idle, cpumask_of(cpu));
5911    /*
5912     * We're having a chicken and egg problem, even though we are
5913     * holding rq->lock, the cpu isn't yet set to this cpu so the
5914     * lockdep check in task_group() will fail.
5915     *
5916     * Similar case to sched_fork(). / Alternatively we could
5917     * use task_rq_lock() here and obtain the other rq->lock.
5918     *
5919     * Silence PROVE_RCU
5920     */
5921    rcu_read_lock();
5922    __set_task_cpu(idle, cpu);
5923    rcu_read_unlock();
5924
5925    rq->curr = rq->idle = idle;
5926#if defined(CONFIG_SMP)
5927    idle->on_cpu = 1;
5928#endif
5929    raw_spin_unlock_irqrestore(&rq->lock, flags);
5930
5931    /* Set the preempt count _outside_ the spinlocks! */
5932    task_thread_info(idle)->preempt_count = 0;
5933
5934    /*
5935     * The idle tasks have their own, simple scheduling class:
5936     */
5937    idle->sched_class = &idle_sched_class;
5938    ftrace_graph_init_idle_task(idle, cpu);
5939}
5940
5941/*
5942 * In a system that switches off the HZ timer nohz_cpu_mask
5943 * indicates which cpus entered this state. This is used
5944 * in the rcu update to wait only for active cpus. For system
5945 * which do not switch off the HZ timer nohz_cpu_mask should
5946 * always be CPU_BITS_NONE.
5947 */
5948cpumask_var_t nohz_cpu_mask;
5949
5950/*
5951 * Increase the granularity value when there are more CPUs,
5952 * because with more CPUs the 'effective latency' as visible
5953 * to users decreases. But the relationship is not linear,
5954 * so pick a second-best guess by going with the log2 of the
5955 * number of CPUs.
5956 *
5957 * This idea comes from the SD scheduler of Con Kolivas:
5958 */
5959static int get_update_sysctl_factor(void)
5960{
5961    unsigned int cpus = min_t(int, num_online_cpus(), 8);
5962    unsigned int factor;
5963
5964    switch (sysctl_sched_tunable_scaling) {
5965    case SCHED_TUNABLESCALING_NONE:
5966        factor = 1;
5967        break;
5968    case SCHED_TUNABLESCALING_LINEAR:
5969        factor = cpus;
5970        break;
5971    case SCHED_TUNABLESCALING_LOG:
5972    default:
5973        factor = 1 + ilog2(cpus);
5974        break;
5975    }
5976
5977    return factor;
5978}
5979
5980static void update_sysctl(void)
5981{
5982    unsigned int factor = get_update_sysctl_factor();
5983
5984#define SET_SYSCTL(name) \
5985    (sysctl_##name = (factor) * normalized_sysctl_##name)
5986    SET_SYSCTL(sched_min_granularity);
5987    SET_SYSCTL(sched_latency);
5988    SET_SYSCTL(sched_wakeup_granularity);
5989#undef SET_SYSCTL
5990}
5991
5992static inline void sched_init_granularity(void)
5993{
5994    update_sysctl();
5995}
5996
5997#ifdef CONFIG_SMP
5998void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask)
5999{
6000    if (p->sched_class && p->sched_class->set_cpus_allowed)
6001        p->sched_class->set_cpus_allowed(p, new_mask);
6002    else {
6003        cpumask_copy(&p->cpus_allowed, new_mask);
6004        p->rt.nr_cpus_allowed = cpumask_weight(new_mask);
6005    }
6006}
6007
6008/*
6009 * This is how migration works:
6010 *
6011 * 1) we invoke migration_cpu_stop() on the target CPU using
6012 * stop_one_cpu().
6013 * 2) stopper starts to run (implicitly forcing the migrated thread
6014 * off the CPU)
6015 * 3) it checks whether the migrated task is still in the wrong runqueue.
6016 * 4) if it's in the wrong runqueue then the migration thread removes
6017 * it and puts it into the right queue.
6018 * 5) stopper completes and stop_one_cpu() returns and the migration
6019 * is done.
6020 */
6021
6022/*
6023 * Change a given task's CPU affinity. Migrate the thread to a
6024 * proper CPU and schedule it away if the CPU it's executing on
6025 * is removed from the allowed bitmask.
6026 *
6027 * NOTE: the caller must have a valid reference to the task, the
6028 * task must not exit() & deallocate itself prematurely. The
6029 * call is not atomic; no spinlocks may be held.
6030 */
6031int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask)
6032{
6033    unsigned long flags;
6034    struct rq *rq;
6035    unsigned int dest_cpu;
6036    int ret = 0;
6037
6038    rq = task_rq_lock(p, &flags);
6039
6040    if (cpumask_equal(&p->cpus_allowed, new_mask))
6041        goto out;
6042
6043    if (!cpumask_intersects(new_mask, cpu_active_mask)) {
6044        ret = -EINVAL;
6045        goto out;
6046    }
6047
6048    if (unlikely((p->flags & PF_THREAD_BOUND) && p != current)) {
6049        ret = -EINVAL;
6050        goto out;
6051    }
6052
6053    do_set_cpus_allowed(p, new_mask);
6054
6055    /* Can the task run on the task's current CPU? If so, we're done */
6056    if (cpumask_test_cpu(task_cpu(p), new_mask))
6057        goto out;
6058
6059    dest_cpu = cpumask_any_and(cpu_active_mask, new_mask);
6060    if (p->on_rq) {
6061        struct migration_arg arg = { p, dest_cpu };
6062        /* Need help from migration thread: drop lock and wait. */
6063        task_rq_unlock(rq, p, &flags);
6064        stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg);
6065        tlb_migrate_finish(p->mm);
6066        return 0;
6067    }
6068out:
6069    task_rq_unlock(rq, p, &flags);
6070
6071    return ret;
6072}
6073EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr);
6074
6075/*
6076 * Move (not current) task off this cpu, onto dest cpu. We're doing
6077 * this because either it can't run here any more (set_cpus_allowed()
6078 * away from this CPU, or CPU going down), or because we're
6079 * attempting to rebalance this task on exec (sched_exec).
6080 *
6081 * So we race with normal scheduler movements, but that's OK, as long
6082 * as the task is no longer on this CPU.
6083 *
6084 * Returns non-zero if task was successfully migrated.
6085 */
6086static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu)
6087{
6088    struct rq *rq_dest, *rq_src;
6089    int ret = 0;
6090
6091    if (unlikely(!cpu_active(dest_cpu)))
6092        return ret;
6093
6094    rq_src = cpu_rq(src_cpu);
6095    rq_dest = cpu_rq(dest_cpu);
6096
6097    raw_spin_lock(&p->pi_lock);
6098    double_rq_lock(rq_src, rq_dest);
6099    /* Already moved. */
6100    if (task_cpu(p) != src_cpu)
6101        goto done;
6102    /* Affinity changed (again). */
6103    if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed))
6104        goto fail;
6105
6106    /*
6107     * If we're not on a rq, the next wake-up will ensure we're
6108     * placed properly.
6109     */
6110    if (p->on_rq) {
6111        deactivate_task(rq_src, p, 0);
6112        set_task_cpu(p, dest_cpu);
6113        activate_task(rq_dest, p, 0);
6114        check_preempt_curr(rq_dest, p, 0);
6115    }
6116done:
6117    ret = 1;
6118fail:
6119    double_rq_unlock(rq_src, rq_dest);
6120    raw_spin_unlock(&p->pi_lock);
6121    return ret;
6122}
6123
6124/*
6125 * migration_cpu_stop - this will be executed by a highprio stopper thread
6126 * and performs thread migration by bumping thread off CPU then
6127 * 'pushing' onto another runqueue.
6128 */
6129static int migration_cpu_stop(void *data)
6130{
6131    struct migration_arg *arg = data;
6132
6133    /*
6134     * The original target cpu might have gone down and we might
6135     * be on another cpu but it doesn't matter.
6136     */
6137    local_irq_disable();
6138    __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu);
6139    local_irq_enable();
6140    return 0;
6141}
6142
6143#ifdef CONFIG_HOTPLUG_CPU
6144
6145/*
6146 * Ensures that the idle task is using init_mm right before its cpu goes
6147 * offline.
6148 */
6149void idle_task_exit(void)
6150{
6151    struct mm_struct *mm = current->active_mm;
6152
6153    BUG_ON(cpu_online(smp_processor_id()));
6154
6155    if (mm != &init_mm)
6156        switch_mm(mm, &init_mm, current);
6157    mmdrop(mm);
6158}
6159
6160/*
6161 * While a dead CPU has no uninterruptible tasks queued at this point,
6162 * it might still have a nonzero ->nr_uninterruptible counter, because
6163 * for performance reasons the counter is not stricly tracking tasks to
6164 * their home CPUs. So we just add the counter to another CPU's counter,
6165 * to keep the global sum constant after CPU-down:
6166 */
6167static void migrate_nr_uninterruptible(struct rq *rq_src)
6168{
6169    struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask));
6170
6171    rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible;
6172    rq_src->nr_uninterruptible = 0;
6173}
6174
6175/*
6176 * remove the tasks which were accounted by rq from calc_load_tasks.
6177 */
6178static void calc_global_load_remove(struct rq *rq)
6179{
6180    atomic_long_sub(rq->calc_load_active, &calc_load_tasks);
6181    rq->calc_load_active = 0;
6182}
6183
6184/*
6185 * Migrate all tasks from the rq, sleeping tasks will be migrated by
6186 * try_to_wake_up()->select_task_rq().
6187 *
6188 * Called with rq->lock held even though we'er in stop_machine() and
6189 * there's no concurrency possible, we hold the required locks anyway
6190 * because of lock validation efforts.
6191 */
6192static void migrate_tasks(unsigned int dead_cpu)
6193{
6194    struct rq *rq = cpu_rq(dead_cpu);
6195    struct task_struct *next, *stop = rq->stop;
6196    int dest_cpu;
6197
6198    /*
6199     * Fudge the rq selection such that the below task selection loop
6200     * doesn't get stuck on the currently eligible stop task.
6201     *
6202     * We're currently inside stop_machine() and the rq is either stuck
6203     * in the stop_machine_cpu_stop() loop, or we're executing this code,
6204     * either way we should never end up calling schedule() until we're
6205     * done here.
6206     */
6207    rq->stop = NULL;
6208
6209    for ( ; ; ) {
6210        /*
6211         * There's this thread running, bail when that's the only
6212         * remaining thread.
6213         */
6214        if (rq->nr_running == 1)
6215            break;
6216
6217        next = pick_next_task(rq);
6218        BUG_ON(!next);
6219        next->sched_class->put_prev_task(rq, next);
6220
6221        /* Find suitable destination for @next, with force if needed. */
6222        dest_cpu = select_fallback_rq(dead_cpu, next);
6223        raw_spin_unlock(&rq->lock);
6224
6225        __migrate_task(next, dead_cpu, dest_cpu);
6226
6227        raw_spin_lock(&rq->lock);
6228    }
6229
6230    rq->stop = stop;
6231}
6232
6233#endif /* CONFIG_HOTPLUG_CPU */
6234
6235#if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
6236
6237static struct ctl_table sd_ctl_dir[] = {
6238    {
6239        .procname = "sched_domain",
6240        .mode = 0555,
6241    },
6242    {}
6243};
6244
6245static struct ctl_table sd_ctl_root[] = {
6246    {
6247        .procname = "kernel",
6248        .mode = 0555,
6249        .child = sd_ctl_dir,
6250    },
6251    {}
6252};
6253
6254static struct ctl_table *sd_alloc_ctl_entry(int n)
6255{
6256    struct ctl_table *entry =
6257        kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL);
6258
6259    return entry;
6260}
6261
6262static void sd_free_ctl_entry(struct ctl_table **tablep)
6263{
6264    struct ctl_table *entry;
6265
6266    /*
6267     * In the intermediate directories, both the child directory and
6268     * procname are dynamically allocated and could fail but the mode
6269     * will always be set. In the lowest directory the names are
6270     * static strings and all have proc handlers.
6271     */
6272    for (entry = *tablep; entry->mode; entry++) {
6273        if (entry->child)
6274            sd_free_ctl_entry(&entry->child);
6275        if (entry->proc_handler == NULL)
6276            kfree(entry->procname);
6277    }
6278
6279    kfree(*tablep);
6280    *tablep = NULL;
6281}
6282
6283static void
6284set_table_entry(struct ctl_table *entry,
6285        const char *procname, void *data, int maxlen,
6286        mode_t mode, proc_handler *proc_handler)
6287{
6288    entry->procname = procname;
6289    entry->data = data;
6290    entry->maxlen = maxlen;
6291    entry->mode = mode;
6292    entry->proc_handler = proc_handler;
6293}
6294
6295static struct ctl_table *
6296sd_alloc_ctl_domain_table(struct sched_domain *sd)
6297{
6298    struct ctl_table *table = sd_alloc_ctl_entry(13);
6299
6300    if (table == NULL)
6301        return NULL;
6302
6303    set_table_entry(&table[0], "min_interval", &sd->min_interval,
6304        sizeof(long), 0644, proc_doulongvec_minmax);
6305    set_table_entry(&table[1], "max_interval", &sd->max_interval,
6306        sizeof(long), 0644, proc_doulongvec_minmax);
6307    set_table_entry(&table[2], "busy_idx", &sd->busy_idx,
6308        sizeof(int), 0644, proc_dointvec_minmax);
6309    set_table_entry(&table[3], "idle_idx", &sd->idle_idx,
6310        sizeof(int), 0644, proc_dointvec_minmax);
6311    set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx,
6312        sizeof(int), 0644, proc_dointvec_minmax);
6313    set_table_entry(&table[5], "wake_idx", &sd->wake_idx,
6314        sizeof(int), 0644, proc_dointvec_minmax);
6315    set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx,
6316        sizeof(int), 0644, proc_dointvec_minmax);
6317    set_table_entry(&table[7], "busy_factor", &sd->busy_factor,
6318        sizeof(int), 0644, proc_dointvec_minmax);
6319    set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct,
6320        sizeof(int), 0644, proc_dointvec_minmax);
6321    set_table_entry(&table[9], "cache_nice_tries",
6322        &sd->cache_nice_tries,
6323        sizeof(int), 0644, proc_dointvec_minmax);
6324    set_table_entry(&table[10], "flags", &sd->flags,
6325        sizeof(int), 0644, proc_dointvec_minmax);
6326    set_table_entry(&table[11], "name", sd->name,
6327        CORENAME_MAX_SIZE, 0444, proc_dostring);
6328    /* &table[12] is terminator */
6329
6330    return table;
6331}
6332
6333static ctl_table *sd_alloc_ctl_cpu_table(int cpu)
6334{