Root/kernel/sched.c

1/*
2 * kernel/sched.c
3 *
4 * Kernel scheduler and related syscalls
5 *
6 * Copyright (C) 1991-2002 Linus Torvalds
7 *
8 * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and
9 * make semaphores SMP safe
10 * 1998-11-19 Implemented schedule_timeout() and related stuff
11 * by Andrea Arcangeli
12 * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar:
13 * hybrid priority-list and round-robin design with
14 * an array-switch method of distributing timeslices
15 * and per-CPU runqueues. Cleanups and useful suggestions
16 * by Davide Libenzi, preemptible kernel bits by Robert Love.
17 * 2003-09-03 Interactivity tuning by Con Kolivas.
18 * 2004-04-02 Scheduler domains code by Nick Piggin
19 * 2007-04-15 Work begun on replacing all interactivity tuning with a
20 * fair scheduling design by Con Kolivas.
21 * 2007-05-05 Load balancing (smp-nice) and other improvements
22 * by Peter Williams
23 * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith
24 * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri
25 * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins,
26 * Thomas Gleixner, Mike Kravetz
27 */
28
29#include <linux/mm.h>
30#include <linux/module.h>
31#include <linux/nmi.h>
32#include <linux/init.h>
33#include <linux/uaccess.h>
34#include <linux/highmem.h>
35#include <linux/smp_lock.h>
36#include <asm/mmu_context.h>
37#include <linux/interrupt.h>
38#include <linux/capability.h>
39#include <linux/completion.h>
40#include <linux/kernel_stat.h>
41#include <linux/debug_locks.h>
42#include <linux/perf_event.h>
43#include <linux/security.h>
44#include <linux/notifier.h>
45#include <linux/profile.h>
46#include <linux/freezer.h>
47#include <linux/vmalloc.h>
48#include <linux/blkdev.h>
49#include <linux/delay.h>
50#include <linux/pid_namespace.h>
51#include <linux/smp.h>
52#include <linux/threads.h>
53#include <linux/timer.h>
54#include <linux/rcupdate.h>
55#include <linux/cpu.h>
56#include <linux/cpuset.h>
57#include <linux/percpu.h>
58#include <linux/proc_fs.h>
59#include <linux/seq_file.h>
60#include <linux/stop_machine.h>
61#include <linux/sysctl.h>
62#include <linux/syscalls.h>
63#include <linux/times.h>
64#include <linux/tsacct_kern.h>
65#include <linux/kprobes.h>
66#include <linux/delayacct.h>
67#include <linux/unistd.h>
68#include <linux/pagemap.h>
69#include <linux/hrtimer.h>
70#include <linux/tick.h>
71#include <linux/debugfs.h>
72#include <linux/ctype.h>
73#include <linux/ftrace.h>
74#include <linux/slab.h>
75
76#include <asm/tlb.h>
77#include <asm/irq_regs.h>
78
79#include "sched_cpupri.h"
80#include "workqueue_sched.h"
81
82#define CREATE_TRACE_POINTS
83#include <trace/events/sched.h>
84
85/*
86 * Convert user-nice values [ -20 ... 0 ... 19 ]
87 * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ],
88 * and back.
89 */
90#define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20)
91#define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20)
92#define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio)
93
94/*
95 * 'User priority' is the nice value converted to something we
96 * can work with better when scaling various scheduler parameters,
97 * it's a [ 0 ... 39 ] range.
98 */
99#define USER_PRIO(p) ((p)-MAX_RT_PRIO)
100#define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio)
101#define MAX_USER_PRIO (USER_PRIO(MAX_PRIO))
102
103/*
104 * Helpers for converting nanosecond timing to jiffy resolution
105 */
106#define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
107
108#define NICE_0_LOAD SCHED_LOAD_SCALE
109#define NICE_0_SHIFT SCHED_LOAD_SHIFT
110
111/*
112 * These are the 'tuning knobs' of the scheduler:
113 *
114 * default timeslice is 100 msecs (used only for SCHED_RR tasks).
115 * Timeslices get refilled after they expire.
116 */
117#define DEF_TIMESLICE (100 * HZ / 1000)
118
119/*
120 * single value that denotes runtime == period, ie unlimited time.
121 */
122#define RUNTIME_INF ((u64)~0ULL)
123
124static inline int rt_policy(int policy)
125{
126    if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR))
127        return 1;
128    return 0;
129}
130
131static inline int task_has_rt_policy(struct task_struct *p)
132{
133    return rt_policy(p->policy);
134}
135
136/*
137 * This is the priority-queue data structure of the RT scheduling class:
138 */
139struct rt_prio_array {
140    DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
141    struct list_head queue[MAX_RT_PRIO];
142};
143
144struct rt_bandwidth {
145    /* nests inside the rq lock: */
146    raw_spinlock_t rt_runtime_lock;
147    ktime_t rt_period;
148    u64 rt_runtime;
149    struct hrtimer rt_period_timer;
150};
151
152static struct rt_bandwidth def_rt_bandwidth;
153
154static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun);
155
156static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer)
157{
158    struct rt_bandwidth *rt_b =
159        container_of(timer, struct rt_bandwidth, rt_period_timer);
160    ktime_t now;
161    int overrun;
162    int idle = 0;
163
164    for (;;) {
165        now = hrtimer_cb_get_time(timer);
166        overrun = hrtimer_forward(timer, now, rt_b->rt_period);
167
168        if (!overrun)
169            break;
170
171        idle = do_sched_rt_period_timer(rt_b, overrun);
172    }
173
174    return idle ? HRTIMER_NORESTART : HRTIMER_RESTART;
175}
176
177static
178void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime)
179{
180    rt_b->rt_period = ns_to_ktime(period);
181    rt_b->rt_runtime = runtime;
182
183    raw_spin_lock_init(&rt_b->rt_runtime_lock);
184
185    hrtimer_init(&rt_b->rt_period_timer,
186            CLOCK_MONOTONIC, HRTIMER_MODE_REL);
187    rt_b->rt_period_timer.function = sched_rt_period_timer;
188}
189
190static inline int rt_bandwidth_enabled(void)
191{
192    return sysctl_sched_rt_runtime >= 0;
193}
194
195static void start_rt_bandwidth(struct rt_bandwidth *rt_b)
196{
197    ktime_t now;
198
199    if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF)
200        return;
201
202    if (hrtimer_active(&rt_b->rt_period_timer))
203        return;
204
205    raw_spin_lock(&rt_b->rt_runtime_lock);
206    for (;;) {
207        unsigned long delta;
208        ktime_t soft, hard;
209
210        if (hrtimer_active(&rt_b->rt_period_timer))
211            break;
212
213        now = hrtimer_cb_get_time(&rt_b->rt_period_timer);
214        hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period);
215
216        soft = hrtimer_get_softexpires(&rt_b->rt_period_timer);
217        hard = hrtimer_get_expires(&rt_b->rt_period_timer);
218        delta = ktime_to_ns(ktime_sub(hard, soft));
219        __hrtimer_start_range_ns(&rt_b->rt_period_timer, soft, delta,
220                HRTIMER_MODE_ABS_PINNED, 0);
221    }
222    raw_spin_unlock(&rt_b->rt_runtime_lock);
223}
224
225#ifdef CONFIG_RT_GROUP_SCHED
226static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b)
227{
228    hrtimer_cancel(&rt_b->rt_period_timer);
229}
230#endif
231
232/*
233 * sched_domains_mutex serializes calls to arch_init_sched_domains,
234 * detach_destroy_domains and partition_sched_domains.
235 */
236static DEFINE_MUTEX(sched_domains_mutex);
237
238#ifdef CONFIG_CGROUP_SCHED
239
240#include <linux/cgroup.h>
241
242struct cfs_rq;
243
244static LIST_HEAD(task_groups);
245
246/* task group related information */
247struct task_group {
248    struct cgroup_subsys_state css;
249
250#ifdef CONFIG_FAIR_GROUP_SCHED
251    /* schedulable entities of this group on each cpu */
252    struct sched_entity **se;
253    /* runqueue "owned" by this group on each cpu */
254    struct cfs_rq **cfs_rq;
255    unsigned long shares;
256#endif
257
258#ifdef CONFIG_RT_GROUP_SCHED
259    struct sched_rt_entity **rt_se;
260    struct rt_rq **rt_rq;
261
262    struct rt_bandwidth rt_bandwidth;
263#endif
264
265    struct rcu_head rcu;
266    struct list_head list;
267
268    struct task_group *parent;
269    struct list_head siblings;
270    struct list_head children;
271};
272
273#define root_task_group init_task_group
274
275/* task_group_lock serializes add/remove of task groups and also changes to
276 * a task group's cpu shares.
277 */
278static DEFINE_SPINLOCK(task_group_lock);
279
280#ifdef CONFIG_FAIR_GROUP_SCHED
281
282#ifdef CONFIG_SMP
283static int root_task_group_empty(void)
284{
285    return list_empty(&root_task_group.children);
286}
287#endif
288
289# define INIT_TASK_GROUP_LOAD NICE_0_LOAD
290
291/*
292 * A weight of 0 or 1 can cause arithmetics problems.
293 * A weight of a cfs_rq is the sum of weights of which entities
294 * are queued on this cfs_rq, so a weight of a entity should not be
295 * too large, so as the shares value of a task group.
296 * (The default weight is 1024 - so there's no practical
297 * limitation from this.)
298 */
299#define MIN_SHARES 2
300#define MAX_SHARES (1UL << 18)
301
302static int init_task_group_load = INIT_TASK_GROUP_LOAD;
303#endif
304
305/* Default task group.
306 * Every task in system belong to this group at bootup.
307 */
308struct task_group init_task_group;
309
310#endif /* CONFIG_CGROUP_SCHED */
311
312/* CFS-related fields in a runqueue */
313struct cfs_rq {
314    struct load_weight load;
315    unsigned long nr_running;
316
317    u64 exec_clock;
318    u64 min_vruntime;
319
320    struct rb_root tasks_timeline;
321    struct rb_node *rb_leftmost;
322
323    struct list_head tasks;
324    struct list_head *balance_iterator;
325
326    /*
327     * 'curr' points to currently running entity on this cfs_rq.
328     * It is set to NULL otherwise (i.e when none are currently running).
329     */
330    struct sched_entity *curr, *next, *last;
331
332    unsigned int nr_spread_over;
333
334#ifdef CONFIG_FAIR_GROUP_SCHED
335    struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */
336
337    /*
338     * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
339     * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
340     * (like users, containers etc.)
341     *
342     * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
343     * list is used during load balance.
344     */
345    struct list_head leaf_cfs_rq_list;
346    struct task_group *tg; /* group that "owns" this runqueue */
347
348#ifdef CONFIG_SMP
349    /*
350     * the part of load.weight contributed by tasks
351     */
352    unsigned long task_weight;
353
354    /*
355     * h_load = weight * f(tg)
356     *
357     * Where f(tg) is the recursive weight fraction assigned to
358     * this group.
359     */
360    unsigned long h_load;
361
362    /*
363     * this cpu's part of tg->shares
364     */
365    unsigned long shares;
366
367    /*
368     * load.weight at the time we set shares
369     */
370    unsigned long rq_weight;
371#endif
372#endif
373};
374
375/* Real-Time classes' related field in a runqueue: */
376struct rt_rq {
377    struct rt_prio_array active;
378    unsigned long rt_nr_running;
379#if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
380    struct {
381        int curr; /* highest queued rt task prio */
382#ifdef CONFIG_SMP
383        int next; /* next highest */
384#endif
385    } highest_prio;
386#endif
387#ifdef CONFIG_SMP
388    unsigned long rt_nr_migratory;
389    unsigned long rt_nr_total;
390    int overloaded;
391    struct plist_head pushable_tasks;
392#endif
393    int rt_throttled;
394    u64 rt_time;
395    u64 rt_runtime;
396    /* Nests inside the rq lock: */
397    raw_spinlock_t rt_runtime_lock;
398
399#ifdef CONFIG_RT_GROUP_SCHED
400    unsigned long rt_nr_boosted;
401
402    struct rq *rq;
403    struct list_head leaf_rt_rq_list;
404    struct task_group *tg;
405#endif
406};
407
408#ifdef CONFIG_SMP
409
410/*
411 * We add the notion of a root-domain which will be used to define per-domain
412 * variables. Each exclusive cpuset essentially defines an island domain by
413 * fully partitioning the member cpus from any other cpuset. Whenever a new
414 * exclusive cpuset is created, we also create and attach a new root-domain
415 * object.
416 *
417 */
418struct root_domain {
419    atomic_t refcount;
420    cpumask_var_t span;
421    cpumask_var_t online;
422
423    /*
424     * The "RT overload" flag: it gets set if a CPU has more than
425     * one runnable RT task.
426     */
427    cpumask_var_t rto_mask;
428    atomic_t rto_count;
429#ifdef CONFIG_SMP
430    struct cpupri cpupri;
431#endif
432};
433
434/*
435 * By default the system creates a single root-domain with all cpus as
436 * members (mimicking the global state we have today).
437 */
438static struct root_domain def_root_domain;
439
440#endif
441
442/*
443 * This is the main, per-CPU runqueue data structure.
444 *
445 * Locking rule: those places that want to lock multiple runqueues
446 * (such as the load balancing or the thread migration code), lock
447 * acquire operations must be ordered by ascending &runqueue.
448 */
449struct rq {
450    /* runqueue lock: */
451    raw_spinlock_t lock;
452
453    /*
454     * nr_running and cpu_load should be in the same cacheline because
455     * remote CPUs use both these fields when doing load calculation.
456     */
457    unsigned long nr_running;
458    #define CPU_LOAD_IDX_MAX 5
459    unsigned long cpu_load[CPU_LOAD_IDX_MAX];
460    unsigned long last_load_update_tick;
461#ifdef CONFIG_NO_HZ
462    u64 nohz_stamp;
463    unsigned char nohz_balance_kick;
464#endif
465    unsigned int skip_clock_update;
466
467    /* capture load from *all* tasks on this cpu: */
468    struct load_weight load;
469    unsigned long nr_load_updates;
470    u64 nr_switches;
471
472    struct cfs_rq cfs;
473    struct rt_rq rt;
474
475#ifdef CONFIG_FAIR_GROUP_SCHED
476    /* list of leaf cfs_rq on this cpu: */
477    struct list_head leaf_cfs_rq_list;
478#endif
479#ifdef CONFIG_RT_GROUP_SCHED
480    struct list_head leaf_rt_rq_list;
481#endif
482
483    /*
484     * This is part of a global counter where only the total sum
485     * over all CPUs matters. A task can increase this counter on
486     * one CPU and if it got migrated afterwards it may decrease
487     * it on another CPU. Always updated under the runqueue lock:
488     */
489    unsigned long nr_uninterruptible;
490
491    struct task_struct *curr, *idle;
492    unsigned long next_balance;
493    struct mm_struct *prev_mm;
494
495    u64 clock;
496
497    atomic_t nr_iowait;
498
499#ifdef CONFIG_SMP
500    struct root_domain *rd;
501    struct sched_domain *sd;
502
503    unsigned long cpu_power;
504
505    unsigned char idle_at_tick;
506    /* For active balancing */
507    int post_schedule;
508    int active_balance;
509    int push_cpu;
510    struct cpu_stop_work active_balance_work;
511    /* cpu of this runqueue: */
512    int cpu;
513    int online;
514
515    unsigned long avg_load_per_task;
516
517    u64 rt_avg;
518    u64 age_stamp;
519    u64 idle_stamp;
520    u64 avg_idle;
521#endif
522
523    /* calc_load related fields */
524    unsigned long calc_load_update;
525    long calc_load_active;
526
527#ifdef CONFIG_SCHED_HRTICK
528#ifdef CONFIG_SMP
529    int hrtick_csd_pending;
530    struct call_single_data hrtick_csd;
531#endif
532    struct hrtimer hrtick_timer;
533#endif
534
535#ifdef CONFIG_SCHEDSTATS
536    /* latency stats */
537    struct sched_info rq_sched_info;
538    unsigned long long rq_cpu_time;
539    /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
540
541    /* sys_sched_yield() stats */
542    unsigned int yld_count;
543
544    /* schedule() stats */
545    unsigned int sched_switch;
546    unsigned int sched_count;
547    unsigned int sched_goidle;
548
549    /* try_to_wake_up() stats */
550    unsigned int ttwu_count;
551    unsigned int ttwu_local;
552
553    /* BKL stats */
554    unsigned int bkl_count;
555#endif
556};
557
558static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
559
560static inline
561void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags)
562{
563    rq->curr->sched_class->check_preempt_curr(rq, p, flags);
564
565    /*
566     * A queue event has occurred, and we're going to schedule. In
567     * this case, we can save a useless back to back clock update.
568     */
569    if (test_tsk_need_resched(p))
570        rq->skip_clock_update = 1;
571}
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_sched_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
609 * with lockdep_is_held(&task_rq(p)->lock) because cpu_cgroup_attach()
610 * holds that lock for each task it moves into the cgroup. Therefore
611 * by holding that lock, we pin the task to the current cgroup.
612 */
613static inline struct task_group *task_group(struct task_struct *p)
614{
615    struct cgroup_subsys_state *css;
616
617    css = task_subsys_state_check(p, cpu_cgroup_subsys_id,
618            lockdep_is_held(&task_rq(p)->lock));
619    return container_of(css, struct task_group, css);
620}
621
622/* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
623static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
624{
625#ifdef CONFIG_FAIR_GROUP_SCHED
626    p->se.cfs_rq = task_group(p)->cfs_rq[cpu];
627    p->se.parent = task_group(p)->se[cpu];
628#endif
629
630#ifdef CONFIG_RT_GROUP_SCHED
631    p->rt.rt_rq = task_group(p)->rt_rq[cpu];
632    p->rt.parent = task_group(p)->rt_se[cpu];
633#endif
634}
635
636#else /* CONFIG_CGROUP_SCHED */
637
638static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
639static inline struct task_group *task_group(struct task_struct *p)
640{
641    return NULL;
642}
643
644#endif /* CONFIG_CGROUP_SCHED */
645
646inline void update_rq_clock(struct rq *rq)
647{
648    if (!rq->skip_clock_update)
649        rq->clock = sched_clock_cpu(cpu_of(rq));
650}
651
652/*
653 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
654 */
655#ifdef CONFIG_SCHED_DEBUG
656# define const_debug __read_mostly
657#else
658# define const_debug static const
659#endif
660
661/**
662 * runqueue_is_locked
663 * @cpu: the processor in question.
664 *
665 * Returns true if the current cpu runqueue is locked.
666 * This interface allows printk to be called with the runqueue lock
667 * held and know whether or not it is OK to wake up the klogd.
668 */
669int runqueue_is_locked(int cpu)
670{
671    return raw_spin_is_locked(&cpu_rq(cpu)->lock);
672}
673
674/*
675 * Debugging: various feature bits
676 */
677
678#define SCHED_FEAT(name, enabled) \
679    __SCHED_FEAT_##name ,
680
681enum {
682#include "sched_features.h"
683};
684
685#undef SCHED_FEAT
686
687#define SCHED_FEAT(name, enabled) \
688    (1UL << __SCHED_FEAT_##name) * enabled |
689
690const_debug unsigned int sysctl_sched_features =
691#include "sched_features.h"
692    0;
693
694#undef SCHED_FEAT
695
696#ifdef CONFIG_SCHED_DEBUG
697#define SCHED_FEAT(name, enabled) \
698    #name ,
699
700static __read_mostly char *sched_feat_names[] = {
701#include "sched_features.h"
702    NULL
703};
704
705#undef SCHED_FEAT
706
707static int sched_feat_show(struct seq_file *m, void *v)
708{
709    int i;
710
711    for (i = 0; sched_feat_names[i]; i++) {
712        if (!(sysctl_sched_features & (1UL << i)))
713            seq_puts(m, "NO_");
714        seq_printf(m, "%s ", sched_feat_names[i]);
715    }
716    seq_puts(m, "\n");
717
718    return 0;
719}
720
721static ssize_t
722sched_feat_write(struct file *filp, const char __user *ubuf,
723        size_t cnt, loff_t *ppos)
724{
725    char buf[64];
726    char *cmp = buf;
727    int neg = 0;
728    int i;
729
730    if (cnt > 63)
731        cnt = 63;
732
733    if (copy_from_user(&buf, ubuf, cnt))
734        return -EFAULT;
735
736    buf[cnt] = 0;
737
738    if (strncmp(buf, "NO_", 3) == 0) {
739        neg = 1;
740        cmp += 3;
741    }
742
743    for (i = 0; sched_feat_names[i]; i++) {
744        int len = strlen(sched_feat_names[i]);
745
746        if (strncmp(cmp, sched_feat_names[i], len) == 0) {
747            if (neg)
748                sysctl_sched_features &= ~(1UL << i);
749            else
750                sysctl_sched_features |= (1UL << i);
751            break;
752        }
753    }
754
755    if (!sched_feat_names[i])
756        return -EINVAL;
757
758    *ppos += cnt;
759
760    return cnt;
761}
762
763static int sched_feat_open(struct inode *inode, struct file *filp)
764{
765    return single_open(filp, sched_feat_show, NULL);
766}
767
768static const struct file_operations sched_feat_fops = {
769    .open = sched_feat_open,
770    .write = sched_feat_write,
771    .read = seq_read,
772    .llseek = seq_lseek,
773    .release = single_release,
774};
775
776static __init int sched_init_debug(void)
777{
778    debugfs_create_file("sched_features", 0644, NULL, NULL,
779            &sched_feat_fops);
780
781    return 0;
782}
783late_initcall(sched_init_debug);
784
785#endif
786
787#define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
788
789/*
790 * Number of tasks to iterate in a single balance run.
791 * Limited because this is done with IRQs disabled.
792 */
793const_debug unsigned int sysctl_sched_nr_migrate = 32;
794
795/*
796 * ratelimit for updating the group shares.
797 * default: 0.25ms
798 */
799unsigned int sysctl_sched_shares_ratelimit = 250000;
800unsigned int normalized_sysctl_sched_shares_ratelimit = 250000;
801
802/*
803 * Inject some fuzzyness into changing the per-cpu group shares
804 * this avoids remote rq-locks at the expense of fairness.
805 * default: 4
806 */
807unsigned int sysctl_sched_shares_thresh = 4;
808
809/*
810 * period over which we average the RT time consumption, measured
811 * in ms.
812 *
813 * default: 1s
814 */
815const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC;
816
817/*
818 * period over which we measure -rt task cpu usage in us.
819 * default: 1s
820 */
821unsigned int sysctl_sched_rt_period = 1000000;
822
823static __read_mostly int scheduler_running;
824
825/*
826 * part of the period that we allow rt tasks to run in us.
827 * default: 0.95s
828 */
829int sysctl_sched_rt_runtime = 950000;
830
831static inline u64 global_rt_period(void)
832{
833    return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
834}
835
836static inline u64 global_rt_runtime(void)
837{
838    if (sysctl_sched_rt_runtime < 0)
839        return RUNTIME_INF;
840
841    return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
842}
843
844#ifndef prepare_arch_switch
845# define prepare_arch_switch(next) do { } while (0)
846#endif
847#ifndef finish_arch_switch
848# define finish_arch_switch(prev) do { } while (0)
849#endif
850
851static inline int task_current(struct rq *rq, struct task_struct *p)
852{
853    return rq->curr == p;
854}
855
856#ifndef __ARCH_WANT_UNLOCKED_CTXSW
857static inline int task_running(struct rq *rq, struct task_struct *p)
858{
859    return task_current(rq, p);
860}
861
862static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
863{
864}
865
866static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
867{
868#ifdef CONFIG_DEBUG_SPINLOCK
869    /* this is a valid case when another task releases the spinlock */
870    rq->lock.owner = current;
871#endif
872    /*
873     * If we are tracking spinlock dependencies then we have to
874     * fix up the runqueue lock - which gets 'carried over' from
875     * prev into current:
876     */
877    spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
878
879    raw_spin_unlock_irq(&rq->lock);
880}
881
882#else /* __ARCH_WANT_UNLOCKED_CTXSW */
883static inline int task_running(struct rq *rq, struct task_struct *p)
884{
885#ifdef CONFIG_SMP
886    return p->oncpu;
887#else
888    return task_current(rq, p);
889#endif
890}
891
892static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
893{
894#ifdef CONFIG_SMP
895    /*
896     * We can optimise this out completely for !SMP, because the
897     * SMP rebalancing from interrupt is the only thing that cares
898     * here.
899     */
900    next->oncpu = 1;
901#endif
902#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
903    raw_spin_unlock_irq(&rq->lock);
904#else
905    raw_spin_unlock(&rq->lock);
906#endif
907}
908
909static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
910{
911#ifdef CONFIG_SMP
912    /*
913     * After ->oncpu is cleared, the task can be moved to a different CPU.
914     * We must ensure this doesn't happen until the switch is completely
915     * finished.
916     */
917    smp_wmb();
918    prev->oncpu = 0;
919#endif
920#ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW
921    local_irq_enable();
922#endif
923}
924#endif /* __ARCH_WANT_UNLOCKED_CTXSW */
925
926/*
927 * Check whether the task is waking, we use this to synchronize ->cpus_allowed
928 * against ttwu().
929 */
930static inline int task_is_waking(struct task_struct *p)
931{
932    return unlikely(p->state == TASK_WAKING);
933}
934
935/*
936 * __task_rq_lock - lock the runqueue a given task resides on.
937 * Must be called interrupts disabled.
938 */
939static inline struct rq *__task_rq_lock(struct task_struct *p)
940    __acquires(rq->lock)
941{
942    struct rq *rq;
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 the runqueue a given task resides on and disable
955 * interrupts. Note the ordering: we can safely lookup the task_rq without
956 * explicitly disabling preemption.
957 */
958static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags)
959    __acquires(rq->lock)
960{
961    struct rq *rq;
962
963    for (;;) {
964        local_irq_save(*flags);
965        rq = task_rq(p);
966        raw_spin_lock(&rq->lock);
967        if (likely(rq == task_rq(p)))
968            return rq;
969        raw_spin_unlock_irqrestore(&rq->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 task_rq_unlock(struct rq *rq, unsigned long *flags)
980    __releases(rq->lock)
981{
982    raw_spin_unlock_irqrestore(&rq->lock, *flags);
983}
984
985/*
986 * this_rq_lock - lock this runqueue and disable interrupts.
987 */
988static struct rq *this_rq_lock(void)
989    __acquires(rq->lock)
990{
991    struct rq *rq;
992
993    local_irq_disable();
994    rq = this_rq();
995    raw_spin_lock(&rq->lock);
996
997    return rq;
998}
999
1000#ifdef CONFIG_SCHED_HRTICK
1001/*
1002 * Use HR-timers to deliver accurate preemption points.
1003 *
1004 * Its all a bit involved since we cannot program an hrt while holding the
1005 * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a
1006 * reschedule event.
1007 *
1008 * When we get rescheduled we reprogram the hrtick_timer outside of the
1009 * rq->lock.
1010 */
1011
1012/*
1013 * Use hrtick when:
1014 * - enabled by features
1015 * - hrtimer is actually high res
1016 */
1017static inline int hrtick_enabled(struct rq *rq)
1018{
1019    if (!sched_feat(HRTICK))
1020        return 0;
1021    if (!cpu_active(cpu_of(rq)))
1022        return 0;
1023    return hrtimer_is_hres_active(&rq->hrtick_timer);
1024}
1025
1026static void hrtick_clear(struct rq *rq)
1027{
1028    if (hrtimer_active(&rq->hrtick_timer))
1029        hrtimer_cancel(&rq->hrtick_timer);
1030}
1031
1032/*
1033 * High-resolution timer tick.
1034 * Runs from hardirq context with interrupts disabled.
1035 */
1036static enum hrtimer_restart hrtick(struct hrtimer *timer)
1037{
1038    struct rq *rq = container_of(timer, struct rq, hrtick_timer);
1039
1040    WARN_ON_ONCE(cpu_of(rq) != smp_processor_id());
1041
1042    raw_spin_lock(&rq->lock);
1043    update_rq_clock(rq);
1044    rq->curr->sched_class->task_tick(rq, rq->curr, 1);
1045    raw_spin_unlock(&rq->lock);
1046
1047    return HRTIMER_NORESTART;
1048}
1049
1050#ifdef CONFIG_SMP
1051/*
1052 * called from hardirq (IPI) context
1053 */
1054static void __hrtick_start(void *arg)
1055{
1056    struct rq *rq = arg;
1057
1058    raw_spin_lock(&rq->lock);
1059    hrtimer_restart(&rq->hrtick_timer);
1060    rq->hrtick_csd_pending = 0;
1061    raw_spin_unlock(&rq->lock);
1062}
1063
1064/*
1065 * Called to set the hrtick timer state.
1066 *
1067 * called with rq->lock held and irqs disabled
1068 */
1069static void hrtick_start(struct rq *rq, u64 delay)
1070{
1071    struct hrtimer *timer = &rq->hrtick_timer;
1072    ktime_t time = ktime_add_ns(timer->base->get_time(), delay);
1073
1074    hrtimer_set_expires(timer, time);
1075
1076    if (rq == this_rq()) {
1077        hrtimer_restart(timer);
1078    } else if (!rq->hrtick_csd_pending) {
1079        __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0);
1080        rq->hrtick_csd_pending = 1;
1081    }
1082}
1083
1084static int
1085hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu)
1086{
1087    int cpu = (int)(long)hcpu;
1088
1089    switch (action) {
1090    case CPU_UP_CANCELED:
1091    case CPU_UP_CANCELED_FROZEN:
1092    case CPU_DOWN_PREPARE:
1093    case CPU_DOWN_PREPARE_FROZEN:
1094    case CPU_DEAD:
1095    case CPU_DEAD_FROZEN:
1096        hrtick_clear(cpu_rq(cpu));
1097        return NOTIFY_OK;
1098    }
1099
1100    return NOTIFY_DONE;
1101}
1102
1103static __init void init_hrtick(void)
1104{
1105    hotcpu_notifier(hotplug_hrtick, 0);
1106}
1107#else
1108/*
1109 * Called to set the hrtick timer state.
1110 *
1111 * called with rq->lock held and irqs disabled
1112 */
1113static void hrtick_start(struct rq *rq, u64 delay)
1114{
1115    __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0,
1116            HRTIMER_MODE_REL_PINNED, 0);
1117}
1118
1119static inline void init_hrtick(void)
1120{
1121}
1122#endif /* CONFIG_SMP */
1123
1124static void init_rq_hrtick(struct rq *rq)
1125{
1126#ifdef CONFIG_SMP
1127    rq->hrtick_csd_pending = 0;
1128
1129    rq->hrtick_csd.flags = 0;
1130    rq->hrtick_csd.func = __hrtick_start;
1131    rq->hrtick_csd.info = rq;
1132#endif
1133
1134    hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1135    rq->hrtick_timer.function = hrtick;
1136}
1137#else /* CONFIG_SCHED_HRTICK */
1138static inline void hrtick_clear(struct rq *rq)
1139{
1140}
1141
1142static inline void init_rq_hrtick(struct rq *rq)
1143{
1144}
1145
1146static inline void init_hrtick(void)
1147{
1148}
1149#endif /* CONFIG_SCHED_HRTICK */
1150
1151/*
1152 * resched_task - mark a task 'to be rescheduled now'.
1153 *
1154 * On UP this means the setting of the need_resched flag, on SMP it
1155 * might also involve a cross-CPU call to trigger the scheduler on
1156 * the target CPU.
1157 */
1158#ifdef CONFIG_SMP
1159
1160#ifndef tsk_is_polling
1161#define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG)
1162#endif
1163
1164static void resched_task(struct task_struct *p)
1165{
1166    int cpu;
1167
1168    assert_raw_spin_locked(&task_rq(p)->lock);
1169
1170    if (test_tsk_need_resched(p))
1171        return;
1172
1173    set_tsk_need_resched(p);
1174
1175    cpu = task_cpu(p);
1176    if (cpu == smp_processor_id())
1177        return;
1178
1179    /* NEED_RESCHED must be visible before we test polling */
1180    smp_mb();
1181    if (!tsk_is_polling(p))
1182        smp_send_reschedule(cpu);
1183}
1184
1185static void resched_cpu(int cpu)
1186{
1187    struct rq *rq = cpu_rq(cpu);
1188    unsigned long flags;
1189
1190    if (!raw_spin_trylock_irqsave(&rq->lock, flags))
1191        return;
1192    resched_task(cpu_curr(cpu));
1193    raw_spin_unlock_irqrestore(&rq->lock, flags);
1194}
1195
1196#ifdef CONFIG_NO_HZ
1197/*
1198 * In the semi idle case, use the nearest busy cpu for migrating timers
1199 * from an idle cpu. This is good for power-savings.
1200 *
1201 * We don't do similar optimization for completely idle system, as
1202 * selecting an idle cpu will add more delays to the timers than intended
1203 * (as that cpu's timer base may not be uptodate wrt jiffies etc).
1204 */
1205int get_nohz_timer_target(void)
1206{
1207    int cpu = smp_processor_id();
1208    int i;
1209    struct sched_domain *sd;
1210
1211    for_each_domain(cpu, sd) {
1212        for_each_cpu(i, sched_domain_span(sd))
1213            if (!idle_cpu(i))
1214                return i;
1215    }
1216    return cpu;
1217}
1218/*
1219 * When add_timer_on() enqueues a timer into the timer wheel of an
1220 * idle CPU then this timer might expire before the next timer event
1221 * which is scheduled to wake up that CPU. In case of a completely
1222 * idle system the next event might even be infinite time into the
1223 * future. wake_up_idle_cpu() ensures that the CPU is woken up and
1224 * leaves the inner idle loop so the newly added timer is taken into
1225 * account when the CPU goes back to idle and evaluates the timer
1226 * wheel for the next timer event.
1227 */
1228void wake_up_idle_cpu(int cpu)
1229{
1230    struct rq *rq = cpu_rq(cpu);
1231
1232    if (cpu == smp_processor_id())
1233        return;
1234
1235    /*
1236     * This is safe, as this function is called with the timer
1237     * wheel base lock of (cpu) held. When the CPU is on the way
1238     * to idle and has not yet set rq->curr to idle then it will
1239     * be serialized on the timer wheel base lock and take the new
1240     * timer into account automatically.
1241     */
1242    if (rq->curr != rq->idle)
1243        return;
1244
1245    /*
1246     * We can set TIF_RESCHED on the idle task of the other CPU
1247     * lockless. The worst case is that the other CPU runs the
1248     * idle task through an additional NOOP schedule()
1249     */
1250    set_tsk_need_resched(rq->idle);
1251
1252    /* NEED_RESCHED must be visible before we test polling */
1253    smp_mb();
1254    if (!tsk_is_polling(rq->idle))
1255        smp_send_reschedule(cpu);
1256}
1257
1258#endif /* CONFIG_NO_HZ */
1259
1260static u64 sched_avg_period(void)
1261{
1262    return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2;
1263}
1264
1265static void sched_avg_update(struct rq *rq)
1266{
1267    s64 period = sched_avg_period();
1268
1269    while ((s64)(rq->clock - rq->age_stamp) > period) {
1270        /*
1271         * Inline assembly required to prevent the compiler
1272         * optimising this loop into a divmod call.
1273         * See __iter_div_u64_rem() for another example of this.
1274         */
1275        asm("" : "+rm" (rq->age_stamp));
1276        rq->age_stamp += period;
1277        rq->rt_avg /= 2;
1278    }
1279}
1280
1281static void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1282{
1283    rq->rt_avg += rt_delta;
1284    sched_avg_update(rq);
1285}
1286
1287#else /* !CONFIG_SMP */
1288static void resched_task(struct task_struct *p)
1289{
1290    assert_raw_spin_locked(&task_rq(p)->lock);
1291    set_tsk_need_resched(p);
1292}
1293
1294static void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1295{
1296}
1297
1298static void sched_avg_update(struct rq *rq)
1299{
1300}
1301#endif /* CONFIG_SMP */
1302
1303#if BITS_PER_LONG == 32
1304# define WMULT_CONST (~0UL)
1305#else
1306# define WMULT_CONST (1UL << 32)
1307#endif
1308
1309#define WMULT_SHIFT 32
1310
1311/*
1312 * Shift right and round:
1313 */
1314#define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y))
1315
1316/*
1317 * delta *= weight / lw
1318 */
1319static unsigned long
1320calc_delta_mine(unsigned long delta_exec, unsigned long weight,
1321        struct load_weight *lw)
1322{
1323    u64 tmp;
1324
1325    if (!lw->inv_weight) {
1326        if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST))
1327            lw->inv_weight = 1;
1328        else
1329            lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2)
1330                / (lw->weight+1);
1331    }
1332
1333    tmp = (u64)delta_exec * weight;
1334    /*
1335     * Check whether we'd overflow the 64-bit multiplication:
1336     */
1337    if (unlikely(tmp > WMULT_CONST))
1338        tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight,
1339            WMULT_SHIFT/2);
1340    else
1341        tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT);
1342
1343    return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX);
1344}
1345
1346static inline void update_load_add(struct load_weight *lw, unsigned long inc)
1347{
1348    lw->weight += inc;
1349    lw->inv_weight = 0;
1350}
1351
1352static inline void update_load_sub(struct load_weight *lw, unsigned long dec)
1353{
1354    lw->weight -= dec;
1355    lw->inv_weight = 0;
1356}
1357
1358/*
1359 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1360 * of tasks with abnormal "nice" values across CPUs the contribution that
1361 * each task makes to its run queue's load is weighted according to its
1362 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1363 * scaled version of the new time slice allocation that they receive on time
1364 * slice expiry etc.
1365 */
1366
1367#define WEIGHT_IDLEPRIO 3
1368#define WMULT_IDLEPRIO 1431655765
1369
1370/*
1371 * Nice levels are multiplicative, with a gentle 10% change for every
1372 * nice level changed. I.e. when a CPU-bound task goes from nice 0 to
1373 * nice 1, it will get ~10% less CPU time than another CPU-bound task
1374 * that remained on nice 0.
1375 *
1376 * The "10% effect" is relative and cumulative: from _any_ nice level,
1377 * if you go up 1 level, it's -10% CPU usage, if you go down 1 level
1378 * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25.
1379 * If a task goes up by ~10% and another task goes down by ~10% then
1380 * the relative distance between them is ~25%.)
1381 */
1382static const int prio_to_weight[40] = {
1383 /* -20 */ 88761, 71755, 56483, 46273, 36291,
1384 /* -15 */ 29154, 23254, 18705, 14949, 11916,
1385 /* -10 */ 9548, 7620, 6100, 4904, 3906,
1386 /* -5 */ 3121, 2501, 1991, 1586, 1277,
1387 /* 0 */ 1024, 820, 655, 526, 423,
1388 /* 5 */ 335, 272, 215, 172, 137,
1389 /* 10 */ 110, 87, 70, 56, 45,
1390 /* 15 */ 36, 29, 23, 18, 15,
1391};
1392
1393/*
1394 * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated.
1395 *
1396 * In cases where the weight does not change often, we can use the
1397 * precalculated inverse to speed up arithmetics by turning divisions
1398 * into multiplications:
1399 */
1400static const u32 prio_to_wmult[40] = {
1401 /* -20 */ 48388, 59856, 76040, 92818, 118348,
1402 /* -15 */ 147320, 184698, 229616, 287308, 360437,
1403 /* -10 */ 449829, 563644, 704093, 875809, 1099582,
1404 /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326,
1405 /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587,
1406 /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126,
1407 /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717,
1408 /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153,
1409};
1410
1411/* Time spent by the tasks of the cpu accounting group executing in ... */
1412enum cpuacct_stat_index {
1413    CPUACCT_STAT_USER, /* ... user mode */
1414    CPUACCT_STAT_SYSTEM, /* ... kernel mode */
1415
1416    CPUACCT_STAT_NSTATS,
1417};
1418
1419#ifdef CONFIG_CGROUP_CPUACCT
1420static void cpuacct_charge(struct task_struct *tsk, u64 cputime);
1421static void cpuacct_update_stats(struct task_struct *tsk,
1422        enum cpuacct_stat_index idx, cputime_t val);
1423#else
1424static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {}
1425static inline void cpuacct_update_stats(struct task_struct *tsk,
1426        enum cpuacct_stat_index idx, cputime_t val) {}
1427#endif
1428
1429static inline void inc_cpu_load(struct rq *rq, unsigned long load)
1430{
1431    update_load_add(&rq->load, load);
1432}
1433
1434static inline void dec_cpu_load(struct rq *rq, unsigned long load)
1435{
1436    update_load_sub(&rq->load, load);
1437}
1438
1439#if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED)
1440typedef int (*tg_visitor)(struct task_group *, void *);
1441
1442/*
1443 * Iterate the full tree, calling @down when first entering a node and @up when
1444 * leaving it for the final time.
1445 */
1446static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
1447{
1448    struct task_group *parent, *child;
1449    int ret;
1450
1451    rcu_read_lock();
1452    parent = &root_task_group;
1453down:
1454    ret = (*down)(parent, data);
1455    if (ret)
1456        goto out_unlock;
1457    list_for_each_entry_rcu(child, &parent->children, siblings) {
1458        parent = child;
1459        goto down;
1460
1461up:
1462        continue;
1463    }
1464    ret = (*up)(parent, data);
1465    if (ret)
1466        goto out_unlock;
1467
1468    child = parent;
1469    parent = parent->parent;
1470    if (parent)
1471        goto up;
1472out_unlock:
1473    rcu_read_unlock();
1474
1475    return ret;
1476}
1477
1478static int tg_nop(struct task_group *tg, void *data)
1479{
1480    return 0;
1481}
1482#endif
1483
1484#ifdef CONFIG_SMP
1485/* Used instead of source_load when we know the type == 0 */
1486static unsigned long weighted_cpuload(const int cpu)
1487{
1488    return cpu_rq(cpu)->load.weight;
1489}
1490
1491/*
1492 * Return a low guess at the load of a migration-source cpu weighted
1493 * according to the scheduling class and "nice" value.
1494 *
1495 * We want to under-estimate the load of migration sources, to
1496 * balance conservatively.
1497 */
1498static unsigned long source_load(int cpu, int type)
1499{
1500    struct rq *rq = cpu_rq(cpu);
1501    unsigned long total = weighted_cpuload(cpu);
1502
1503    if (type == 0 || !sched_feat(LB_BIAS))
1504        return total;
1505
1506    return min(rq->cpu_load[type-1], total);
1507}
1508
1509/*
1510 * Return a high guess at the load of a migration-target cpu weighted
1511 * according to the scheduling class and "nice" value.
1512 */
1513static unsigned long target_load(int cpu, int type)
1514{
1515    struct rq *rq = cpu_rq(cpu);
1516    unsigned long total = weighted_cpuload(cpu);
1517
1518    if (type == 0 || !sched_feat(LB_BIAS))
1519        return total;
1520
1521    return max(rq->cpu_load[type-1], total);
1522}
1523
1524static unsigned long power_of(int cpu)
1525{
1526    return cpu_rq(cpu)->cpu_power;
1527}
1528
1529static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd);
1530
1531static unsigned long cpu_avg_load_per_task(int cpu)
1532{
1533    struct rq *rq = cpu_rq(cpu);
1534    unsigned long nr_running = ACCESS_ONCE(rq->nr_running);
1535
1536    if (nr_running)
1537        rq->avg_load_per_task = rq->load.weight / nr_running;
1538    else
1539        rq->avg_load_per_task = 0;
1540
1541    return rq->avg_load_per_task;
1542}
1543
1544#ifdef CONFIG_FAIR_GROUP_SCHED
1545
1546static __read_mostly unsigned long __percpu *update_shares_data;
1547
1548static void __set_se_shares(struct sched_entity *se, unsigned long shares);
1549
1550/*
1551 * Calculate and set the cpu's group shares.
1552 */
1553static void update_group_shares_cpu(struct task_group *tg, int cpu,
1554                    unsigned long sd_shares,
1555                    unsigned long sd_rq_weight,
1556                    unsigned long *usd_rq_weight)
1557{
1558    unsigned long shares, rq_weight;
1559    int boost = 0;
1560
1561    rq_weight = usd_rq_weight[cpu];
1562    if (!rq_weight) {
1563        boost = 1;
1564        rq_weight = NICE_0_LOAD;
1565    }
1566
1567    /*
1568     * \Sum_j shares_j * rq_weight_i
1569     * shares_i = -----------------------------
1570     * \Sum_j rq_weight_j
1571     */
1572    shares = (sd_shares * rq_weight) / sd_rq_weight;
1573    shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES);
1574
1575    if (abs(shares - tg->se[cpu]->load.weight) >
1576            sysctl_sched_shares_thresh) {
1577        struct rq *rq = cpu_rq(cpu);
1578        unsigned long flags;
1579
1580        raw_spin_lock_irqsave(&rq->lock, flags);
1581        tg->cfs_rq[cpu]->rq_weight = boost ? 0 : rq_weight;
1582        tg->cfs_rq[cpu]->shares = boost ? 0 : shares;
1583        __set_se_shares(tg->se[cpu], shares);
1584        raw_spin_unlock_irqrestore(&rq->lock, flags);
1585    }
1586}
1587
1588/*
1589 * Re-compute the task group their per cpu shares over the given domain.
1590 * This needs to be done in a bottom-up fashion because the rq weight of a
1591 * parent group depends on the shares of its child groups.
1592 */
1593static int tg_shares_up(struct task_group *tg, void *data)
1594{
1595    unsigned long weight, rq_weight = 0, sum_weight = 0, shares = 0;
1596    unsigned long *usd_rq_weight;
1597    struct sched_domain *sd = data;
1598    unsigned long flags;
1599    int i;
1600
1601    if (!tg->se[0])
1602        return 0;
1603
1604    local_irq_save(flags);
1605    usd_rq_weight = per_cpu_ptr(update_shares_data, smp_processor_id());
1606
1607    for_each_cpu(i, sched_domain_span(sd)) {
1608        weight = tg->cfs_rq[i]->load.weight;
1609        usd_rq_weight[i] = weight;
1610
1611        rq_weight += weight;
1612        /*
1613         * If there are currently no tasks on the cpu pretend there
1614         * is one of average load so that when a new task gets to
1615         * run here it will not get delayed by group starvation.
1616         */
1617        if (!weight)
1618            weight = NICE_0_LOAD;
1619
1620        sum_weight += weight;
1621        shares += tg->cfs_rq[i]->shares;
1622    }
1623
1624    if (!rq_weight)
1625        rq_weight = sum_weight;
1626
1627    if ((!shares && rq_weight) || shares > tg->shares)
1628        shares = tg->shares;
1629
1630    if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE))
1631        shares = tg->shares;
1632
1633    for_each_cpu(i, sched_domain_span(sd))
1634        update_group_shares_cpu(tg, i, shares, rq_weight, usd_rq_weight);
1635
1636    local_irq_restore(flags);
1637
1638    return 0;
1639}
1640
1641/*
1642 * Compute the cpu's hierarchical load factor for each task group.
1643 * This needs to be done in a top-down fashion because the load of a child
1644 * group is a fraction of its parents load.
1645 */
1646static int tg_load_down(struct task_group *tg, void *data)
1647{
1648    unsigned long load;
1649    long cpu = (long)data;
1650
1651    if (!tg->parent) {
1652        load = cpu_rq(cpu)->load.weight;
1653    } else {
1654        load = tg->parent->cfs_rq[cpu]->h_load;
1655        load *= tg->cfs_rq[cpu]->shares;
1656        load /= tg->parent->cfs_rq[cpu]->load.weight + 1;
1657    }
1658
1659    tg->cfs_rq[cpu]->h_load = load;
1660
1661    return 0;
1662}
1663
1664static void update_shares(struct sched_domain *sd)
1665{
1666    s64 elapsed;
1667    u64 now;
1668
1669    if (root_task_group_empty())
1670        return;
1671
1672    now = local_clock();
1673    elapsed = now - sd->last_update;
1674
1675    if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) {
1676        sd->last_update = now;
1677        walk_tg_tree(tg_nop, tg_shares_up, sd);
1678    }
1679}
1680
1681static void update_h_load(long cpu)
1682{
1683    walk_tg_tree(tg_load_down, tg_nop, (void *)cpu);
1684}
1685
1686#else
1687
1688static inline void update_shares(struct sched_domain *sd)
1689{
1690}
1691
1692#endif
1693
1694#ifdef CONFIG_PREEMPT
1695
1696static void double_rq_lock(struct rq *rq1, struct rq *rq2);
1697
1698/*
1699 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1700 * way at the expense of forcing extra atomic operations in all
1701 * invocations. This assures that the double_lock is acquired using the
1702 * same underlying policy as the spinlock_t on this architecture, which
1703 * reduces latency compared to the unfair variant below. However, it
1704 * also adds more overhead and therefore may reduce throughput.
1705 */
1706static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1707    __releases(this_rq->lock)
1708    __acquires(busiest->lock)
1709    __acquires(this_rq->lock)
1710{
1711    raw_spin_unlock(&this_rq->lock);
1712    double_rq_lock(this_rq, busiest);
1713
1714    return 1;
1715}
1716
1717#else
1718/*
1719 * Unfair double_lock_balance: Optimizes throughput at the expense of
1720 * latency by eliminating extra atomic operations when the locks are
1721 * already in proper order on entry. This favors lower cpu-ids and will
1722 * grant the double lock to lower cpus over higher ids under contention,
1723 * regardless of entry order into the function.
1724 */
1725static int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1726    __releases(this_rq->lock)
1727    __acquires(busiest->lock)
1728    __acquires(this_rq->lock)
1729{
1730    int ret = 0;
1731
1732    if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1733        if (busiest < this_rq) {
1734            raw_spin_unlock(&this_rq->lock);
1735            raw_spin_lock(&busiest->lock);
1736            raw_spin_lock_nested(&this_rq->lock,
1737                          SINGLE_DEPTH_NESTING);
1738            ret = 1;
1739        } else
1740            raw_spin_lock_nested(&busiest->lock,
1741                          SINGLE_DEPTH_NESTING);
1742    }
1743    return ret;
1744}
1745
1746#endif /* CONFIG_PREEMPT */
1747
1748/*
1749 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1750 */
1751static int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1752{
1753    if (unlikely(!irqs_disabled())) {
1754        /* printk() doesn't work good under rq->lock */
1755        raw_spin_unlock(&this_rq->lock);
1756        BUG_ON(1);
1757    }
1758
1759    return _double_lock_balance(this_rq, busiest);
1760}
1761
1762static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1763    __releases(busiest->lock)
1764{
1765    raw_spin_unlock(&busiest->lock);
1766    lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1767}
1768
1769/*
1770 * double_rq_lock - safely lock two runqueues
1771 *
1772 * Note this does not disable interrupts like task_rq_lock,
1773 * you need to do so manually before calling.
1774 */
1775static void double_rq_lock(struct rq *rq1, struct rq *rq2)
1776    __acquires(rq1->lock)
1777    __acquires(rq2->lock)
1778{
1779    BUG_ON(!irqs_disabled());
1780    if (rq1 == rq2) {
1781        raw_spin_lock(&rq1->lock);
1782        __acquire(rq2->lock); /* Fake it out ;) */
1783    } else {
1784        if (rq1 < rq2) {
1785            raw_spin_lock(&rq1->lock);
1786            raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1787        } else {
1788            raw_spin_lock(&rq2->lock);
1789            raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1790        }
1791    }
1792}
1793
1794/*
1795 * double_rq_unlock - safely unlock two runqueues
1796 *
1797 * Note this does not restore interrupts like task_rq_unlock,
1798 * you need to do so manually after calling.
1799 */
1800static void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1801    __releases(rq1->lock)
1802    __releases(rq2->lock)
1803{
1804    raw_spin_unlock(&rq1->lock);
1805    if (rq1 != rq2)
1806        raw_spin_unlock(&rq2->lock);
1807    else
1808        __release(rq2->lock);
1809}
1810
1811#endif
1812
1813#ifdef CONFIG_FAIR_GROUP_SCHED
1814static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares)
1815{
1816#ifdef CONFIG_SMP
1817    cfs_rq->shares = shares;
1818#endif
1819}
1820#endif
1821
1822static void calc_load_account_idle(struct rq *this_rq);
1823static void update_sysctl(void);
1824static int get_update_sysctl_factor(void);
1825static void update_cpu_load(struct rq *this_rq);
1826
1827static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1828{
1829    set_task_rq(p, cpu);
1830#ifdef CONFIG_SMP
1831    /*
1832     * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1833     * successfuly executed on another CPU. We must ensure that updates of
1834     * per-task data have been completed by this moment.
1835     */
1836    smp_wmb();
1837    task_thread_info(p)->cpu = cpu;
1838#endif
1839}
1840
1841static const struct sched_class rt_sched_class;
1842
1843#define sched_class_highest (&rt_sched_class)
1844#define for_each_class(class) \
1845   for (class = sched_class_highest; class; class = class->next)
1846
1847#include "sched_stats.h"
1848
1849static void inc_nr_running(struct rq *rq)
1850{
1851    rq->nr_running++;
1852}
1853
1854static void dec_nr_running(struct rq *rq)
1855{
1856    rq->nr_running--;
1857}
1858
1859static void set_load_weight(struct task_struct *p)
1860{
1861    if (task_has_rt_policy(p)) {
1862        p->se.load.weight = 0;
1863        p->se.load.inv_weight = WMULT_CONST;
1864        return;
1865    }
1866
1867    /*
1868     * SCHED_IDLE tasks get minimal weight:
1869     */
1870    if (p->policy == SCHED_IDLE) {
1871        p->se.load.weight = WEIGHT_IDLEPRIO;
1872        p->se.load.inv_weight = WMULT_IDLEPRIO;
1873        return;
1874    }
1875
1876    p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO];
1877    p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO];
1878}
1879
1880static void enqueue_task(struct rq *rq, struct task_struct *p, int flags)
1881{
1882    update_rq_clock(rq);
1883    sched_info_queued(p);
1884    p->sched_class->enqueue_task(rq, p, flags);
1885    p->se.on_rq = 1;
1886}
1887
1888static void dequeue_task(struct rq *rq, struct task_struct *p, int flags)
1889{
1890    update_rq_clock(rq);
1891    sched_info_dequeued(p);
1892    p->sched_class->dequeue_task(rq, p, flags);
1893    p->se.on_rq = 0;
1894}
1895
1896/*
1897 * activate_task - move a task to the runqueue.
1898 */
1899static void activate_task(struct rq *rq, struct task_struct *p, int flags)
1900{
1901    if (task_contributes_to_load(p))
1902        rq->nr_uninterruptible--;
1903
1904    enqueue_task(rq, p, flags);
1905    inc_nr_running(rq);
1906}
1907
1908/*
1909 * deactivate_task - remove a task from the runqueue.
1910 */
1911static void deactivate_task(struct rq *rq, struct task_struct *p, int flags)
1912{
1913    if (task_contributes_to_load(p))
1914        rq->nr_uninterruptible++;
1915
1916    dequeue_task(rq, p, flags);
1917    dec_nr_running(rq);
1918}
1919
1920#include "sched_idletask.c"
1921#include "sched_fair.c"
1922#include "sched_rt.c"
1923#ifdef CONFIG_SCHED_DEBUG
1924# include "sched_debug.c"
1925#endif
1926
1927/*
1928 * __normal_prio - return the priority that is based on the static prio
1929 */
1930static inline int __normal_prio(struct task_struct *p)
1931{
1932    return p->static_prio;
1933}
1934
1935/*
1936 * Calculate the expected normal priority: i.e. priority
1937 * without taking RT-inheritance into account. Might be
1938 * boosted by interactivity modifiers. Changes upon fork,
1939 * setprio syscalls, and whenever the interactivity
1940 * estimator recalculates.
1941 */
1942static inline int normal_prio(struct task_struct *p)
1943{
1944    int prio;
1945
1946    if (task_has_rt_policy(p))
1947        prio = MAX_RT_PRIO-1 - p->rt_priority;
1948    else
1949        prio = __normal_prio(p);
1950    return prio;
1951}
1952
1953/*
1954 * Calculate the current priority, i.e. the priority
1955 * taken into account by the scheduler. This value might
1956 * be boosted by RT tasks, or might be boosted by
1957 * interactivity modifiers. Will be RT if the task got
1958 * RT-boosted. If not then it returns p->normal_prio.
1959 */
1960static int effective_prio(struct task_struct *p)
1961{
1962    p->normal_prio = normal_prio(p);
1963    /*
1964     * If we are RT tasks or we were boosted to RT priority,
1965     * keep the priority unchanged. Otherwise, update priority
1966     * to the normal priority:
1967     */
1968    if (!rt_prio(p->prio))
1969        return p->normal_prio;
1970    return p->prio;
1971}
1972
1973/**
1974 * task_curr - is this task currently executing on a CPU?
1975 * @p: the task in question.
1976 */
1977inline int task_curr(const struct task_struct *p)
1978{
1979    return cpu_curr(task_cpu(p)) == p;
1980}
1981
1982static inline void check_class_changed(struct rq *rq, struct task_struct *p,
1983                       const struct sched_class *prev_class,
1984                       int oldprio, int running)
1985{
1986    if (prev_class != p->sched_class) {
1987        if (prev_class->switched_from)
1988            prev_class->switched_from(rq, p, running);
1989        p->sched_class->switched_to(rq, p, running);
1990    } else
1991        p->sched_class->prio_changed(rq, p, oldprio, running);
1992}
1993
1994#ifdef CONFIG_SMP
1995/*
1996 * Is this task likely cache-hot:
1997 */
1998static int
1999task_hot(struct task_struct *p, u64 now, struct sched_domain *sd)
2000{
2001    s64 delta;
2002
2003    if (p->sched_class != &fair_sched_class)
2004        return 0;
2005
2006    /*
2007     * Buddy candidates are cache hot:
2008     */
2009    if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running &&
2010            (&p->se == cfs_rq_of(&p->se)->next ||
2011             &p->se == cfs_rq_of(&p->se)->last))
2012        return 1;
2013
2014    if (sysctl_sched_migration_cost == -1)
2015        return 1;
2016    if (sysctl_sched_migration_cost == 0)
2017        return 0;
2018
2019    delta = now - p->se.exec_start;
2020
2021    return delta < (s64)sysctl_sched_migration_cost;
2022}
2023
2024void set_task_cpu(struct task_struct *p, unsigned int new_cpu)
2025{
2026#ifdef CONFIG_SCHED_DEBUG
2027    /*
2028     * We should never call set_task_cpu() on a blocked task,
2029     * ttwu() will sort out the placement.
2030     */
2031    WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING &&
2032            !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE));
2033#endif
2034
2035    trace_sched_migrate_task(p, new_cpu);
2036
2037    if (task_cpu(p) != new_cpu) {
2038        p->se.nr_migrations++;
2039        perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 1, NULL, 0);
2040    }
2041
2042    __set_task_cpu(p, new_cpu);
2043}
2044
2045struct migration_arg {
2046    struct task_struct *task;
2047    int dest_cpu;
2048};
2049
2050static int migration_cpu_stop(void *data);
2051
2052/*
2053 * The task's runqueue lock must be held.
2054 * Returns true if you have to wait for migration thread.
2055 */
2056static bool migrate_task(struct task_struct *p, int dest_cpu)
2057{
2058    struct rq *rq = task_rq(p);
2059
2060    /*
2061     * If the task is not on a runqueue (and not running), then
2062     * the next wake-up will properly place the task.
2063     */
2064    return p->se.on_rq || task_running(rq, p);
2065}
2066
2067/*
2068 * wait_task_inactive - wait for a thread to unschedule.
2069 *
2070 * If @match_state is nonzero, it's the @p->state value just checked and
2071 * not expected to change. If it changes, i.e. @p might have woken up,
2072 * then return zero. When we succeed in waiting for @p to be off its CPU,
2073 * we return a positive number (its total switch count). If a second call
2074 * a short while later returns the same number, the caller can be sure that
2075 * @p has remained unscheduled the whole time.
2076 *
2077 * The caller must ensure that the task *will* unschedule sometime soon,
2078 * else this function might spin for a *long* time. This function can't
2079 * be called with interrupts off, or it may introduce deadlock with
2080 * smp_call_function() if an IPI is sent by the same process we are
2081 * waiting to become inactive.
2082 */
2083unsigned long wait_task_inactive(struct task_struct *p, long match_state)
2084{
2085    unsigned long flags;
2086    int running, on_rq;
2087    unsigned long ncsw;
2088    struct rq *rq;
2089
2090    for (;;) {
2091        /*
2092         * We do the initial early heuristics without holding
2093         * any task-queue locks at all. We'll only try to get
2094         * the runqueue lock when things look like they will
2095         * work out!
2096         */
2097        rq = task_rq(p);
2098
2099        /*
2100         * If the task is actively running on another CPU
2101         * still, just relax and busy-wait without holding
2102         * any locks.
2103         *
2104         * NOTE! Since we don't hold any locks, it's not
2105         * even sure that "rq" stays as the right runqueue!
2106         * But we don't care, since "task_running()" will
2107         * return false if the runqueue has changed and p
2108         * is actually now running somewhere else!
2109         */
2110        while (task_running(rq, p)) {
2111            if (match_state && unlikely(p->state != match_state))
2112                return 0;
2113            cpu_relax();
2114        }
2115
2116        /*
2117         * Ok, time to look more closely! We need the rq
2118         * lock now, to be *sure*. If we're wrong, we'll
2119         * just go back and repeat.
2120         */
2121        rq = task_rq_lock(p, &flags);
2122        trace_sched_wait_task(p);
2123        running = task_running(rq, p);
2124        on_rq = p->se.on_rq;
2125        ncsw = 0;
2126        if (!match_state || p->state == match_state)
2127            ncsw = p->nvcsw | LONG_MIN; /* sets MSB */
2128        task_rq_unlock(rq, &flags);
2129
2130        /*
2131         * If it changed from the expected state, bail out now.
2132         */
2133        if (unlikely(!ncsw))
2134            break;
2135
2136        /*
2137         * Was it really running after all now that we
2138         * checked with the proper locks actually held?
2139         *
2140         * Oops. Go back and try again..
2141         */
2142        if (unlikely(running)) {
2143            cpu_relax();
2144            continue;
2145        }
2146
2147        /*
2148         * It's not enough that it's not actively running,
2149         * it must be off the runqueue _entirely_, and not
2150         * preempted!
2151         *
2152         * So if it was still runnable (but just not actively
2153         * running right now), it's preempted, and we should
2154         * yield - it could be a while.
2155         */
2156        if (unlikely(on_rq)) {
2157            schedule_timeout_uninterruptible(1);
2158            continue;
2159        }
2160
2161        /*
2162         * Ahh, all good. It wasn't running, and it wasn't
2163         * runnable, which means that it will never become
2164         * running in the future either. We're all done!
2165         */
2166        break;
2167    }
2168
2169    return ncsw;
2170}
2171
2172/***
2173 * kick_process - kick a running thread to enter/exit the kernel
2174 * @p: the to-be-kicked thread
2175 *
2176 * Cause a process which is running on another CPU to enter
2177 * kernel-mode, without any delay. (to get signals handled.)
2178 *
2179 * NOTE: this function doesnt have to take the runqueue lock,
2180 * because all it wants to ensure is that the remote task enters
2181 * the kernel. If the IPI races and the task has been migrated
2182 * to another CPU then no harm is done and the purpose has been
2183 * achieved as well.
2184 */
2185void kick_process(struct task_struct *p)
2186{
2187    int cpu;
2188
2189    preempt_disable();
2190    cpu = task_cpu(p);
2191    if ((cpu != smp_processor_id()) && task_curr(p))
2192        smp_send_reschedule(cpu);
2193    preempt_enable();
2194}
2195EXPORT_SYMBOL_GPL(kick_process);
2196#endif /* CONFIG_SMP */
2197
2198/**
2199 * task_oncpu_function_call - call a function on the cpu on which a task runs
2200 * @p: the task to evaluate
2201 * @func: the function to be called
2202 * @info: the function call argument
2203 *
2204 * Calls the function @func when the task is currently running. This might
2205 * be on the current CPU, which just calls the function directly
2206 */
2207void task_oncpu_function_call(struct task_struct *p,
2208                  void (*func) (void *info), void *info)
2209{
2210    int cpu;
2211
2212    preempt_disable();
2213    cpu = task_cpu(p);
2214    if (task_curr(p))
2215        smp_call_function_single(cpu, func, info, 1);
2216    preempt_enable();
2217}
2218
2219#ifdef CONFIG_SMP
2220/*
2221 * ->cpus_allowed is protected by either TASK_WAKING or rq->lock held.
2222 */
2223static int select_fallback_rq(int cpu, struct task_struct *p)
2224{
2225    int dest_cpu;
2226    const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu));
2227
2228    /* Look for allowed, online CPU in same node. */
2229    for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask)
2230        if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed))
2231            return dest_cpu;
2232
2233    /* Any allowed, online CPU? */
2234    dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_active_mask);
2235    if (dest_cpu < nr_cpu_ids)
2236        return dest_cpu;
2237
2238    /* No more Mr. Nice Guy. */
2239    if (unlikely(dest_cpu >= nr_cpu_ids)) {
2240        dest_cpu = cpuset_cpus_allowed_fallback(p);
2241        /*
2242         * Don't tell them about moving exiting tasks or
2243         * kernel threads (both mm NULL), since they never
2244         * leave kernel.
2245         */
2246        if (p->mm && printk_ratelimit()) {
2247            printk(KERN_INFO "process %d (%s) no "
2248                   "longer affine to cpu%d\n",
2249                   task_pid_nr(p), p->comm, cpu);
2250        }
2251    }
2252
2253    return dest_cpu;
2254}
2255
2256/*
2257 * The caller (fork, wakeup) owns TASK_WAKING, ->cpus_allowed is stable.
2258 */
2259static inline
2260int select_task_rq(struct rq *rq, struct task_struct *p, int sd_flags, int wake_flags)
2261{
2262    int cpu = p->sched_class->select_task_rq(rq, p, sd_flags, wake_flags);
2263
2264    /*
2265     * In order not to call set_task_cpu() on a blocking task we need
2266     * to rely on ttwu() to place the task on a valid ->cpus_allowed
2267     * cpu.
2268     *
2269     * Since this is common to all placement strategies, this lives here.
2270     *
2271     * [ this allows ->select_task() to simply return task_cpu(p) and
2272     * not worry about this generic constraint ]
2273     */
2274    if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) ||
2275             !cpu_online(cpu)))
2276        cpu = select_fallback_rq(task_cpu(p), p);
2277
2278    return cpu;
2279}
2280
2281static void update_avg(u64 *avg, u64 sample)
2282{
2283    s64 diff = sample - *avg;
2284    *avg += diff >> 3;
2285}
2286#endif
2287
2288static inline void ttwu_activate(struct task_struct *p, struct rq *rq,
2289                 bool is_sync, bool is_migrate, bool is_local,
2290                 unsigned long en_flags)
2291{
2292    schedstat_inc(p, se.statistics.nr_wakeups);
2293    if (is_sync)
2294        schedstat_inc(p, se.statistics.nr_wakeups_sync);
2295    if (is_migrate)
2296        schedstat_inc(p, se.statistics.nr_wakeups_migrate);
2297    if (is_local)
2298        schedstat_inc(p, se.statistics.nr_wakeups_local);
2299    else
2300        schedstat_inc(p, se.statistics.nr_wakeups_remote);
2301
2302    activate_task(rq, p, en_flags);
2303}
2304
2305static inline void ttwu_post_activation(struct task_struct *p, struct rq *rq,
2306                    int wake_flags, bool success)
2307{
2308    trace_sched_wakeup(p, success);
2309    check_preempt_curr(rq, p, wake_flags);
2310
2311    p->state = TASK_RUNNING;
2312#ifdef CONFIG_SMP
2313    if (p->sched_class->task_woken)
2314        p->sched_class->task_woken(rq, p);
2315
2316    if (unlikely(rq->idle_stamp)) {
2317        u64 delta = rq->clock - rq->idle_stamp;
2318        u64 max = 2*sysctl_sched_migration_cost;
2319
2320        if (delta > max)
2321            rq->avg_idle = max;
2322        else
2323            update_avg(&rq->avg_idle, delta);
2324        rq->idle_stamp = 0;
2325    }
2326#endif
2327    /* if a worker is waking up, notify workqueue */
2328    if ((p->flags & PF_WQ_WORKER) && success)
2329        wq_worker_waking_up(p, cpu_of(rq));
2330}
2331
2332/**
2333 * try_to_wake_up - wake up a thread
2334 * @p: the thread to be awakened
2335 * @state: the mask of task states that can be woken
2336 * @wake_flags: wake modifier flags (WF_*)
2337 *
2338 * Put it on the run-queue if it's not already there. The "current"
2339 * thread is always on the run-queue (except when the actual
2340 * re-schedule is in progress), and as such you're allowed to do
2341 * the simpler "current->state = TASK_RUNNING" to mark yourself
2342 * runnable without the overhead of this.
2343 *
2344 * Returns %true if @p was woken up, %false if it was already running
2345 * or @state didn't match @p's state.
2346 */
2347static int try_to_wake_up(struct task_struct *p, unsigned int state,
2348              int wake_flags)
2349{
2350    int cpu, orig_cpu, this_cpu, success = 0;
2351    unsigned long flags;
2352    unsigned long en_flags = ENQUEUE_WAKEUP;
2353    struct rq *rq;
2354
2355    this_cpu = get_cpu();
2356
2357    smp_wmb();
2358    rq = task_rq_lock(p, &flags);
2359    if (!(p->state & state))
2360        goto out;
2361
2362    if (p->se.on_rq)
2363        goto out_running;
2364
2365    cpu = task_cpu(p);
2366    orig_cpu = cpu;
2367
2368#ifdef CONFIG_SMP
2369    if (unlikely(task_running(rq, p)))
2370        goto out_activate;
2371
2372    /*
2373     * In order to handle concurrent wakeups and release the rq->lock
2374     * we put the task in TASK_WAKING state.
2375     *
2376     * First fix up the nr_uninterruptible count:
2377     */
2378    if (task_contributes_to_load(p)) {
2379        if (likely(cpu_online(orig_cpu)))
2380            rq->nr_uninterruptible--;
2381        else
2382            this_rq()->nr_uninterruptible--;
2383    }
2384    p->state = TASK_WAKING;
2385
2386    if (p->sched_class->task_waking) {
2387        p->sched_class->task_waking(rq, p);
2388        en_flags |= ENQUEUE_WAKING;
2389    }
2390
2391    cpu = select_task_rq(rq, p, SD_BALANCE_WAKE, wake_flags);
2392    if (cpu != orig_cpu)
2393        set_task_cpu(p, cpu);
2394    __task_rq_unlock(rq);
2395
2396    rq = cpu_rq(cpu);
2397    raw_spin_lock(&rq->lock);
2398
2399    /*
2400     * We migrated the task without holding either rq->lock, however
2401     * since the task is not on the task list itself, nobody else
2402     * will try and migrate the task, hence the rq should match the
2403     * cpu we just moved it to.
2404     */
2405    WARN_ON(task_cpu(p) != cpu);
2406    WARN_ON(p->state != TASK_WAKING);
2407
2408#ifdef CONFIG_SCHEDSTATS
2409    schedstat_inc(rq, ttwu_count);
2410    if (cpu == this_cpu)
2411        schedstat_inc(rq, ttwu_local);
2412    else {
2413        struct sched_domain *sd;
2414        for_each_domain(this_cpu, sd) {
2415            if (cpumask_test_cpu(cpu, sched_domain_span(sd))) {
2416                schedstat_inc(sd, ttwu_wake_remote);
2417                break;
2418            }
2419        }
2420    }
2421#endif /* CONFIG_SCHEDSTATS */
2422
2423out_activate:
2424#endif /* CONFIG_SMP */
2425    ttwu_activate(p, rq, wake_flags & WF_SYNC, orig_cpu != cpu,
2426              cpu == this_cpu, en_flags);
2427    success = 1;
2428out_running:
2429    ttwu_post_activation(p, rq, wake_flags, success);
2430out:
2431    task_rq_unlock(rq, &flags);
2432    put_cpu();
2433
2434    return success;
2435}
2436
2437/**
2438 * try_to_wake_up_local - try to wake up a local task with rq lock held
2439 * @p: the thread to be awakened
2440 *
2441 * Put @p on the run-queue if it's not alredy there. The caller must
2442 * ensure that this_rq() is locked, @p is bound to this_rq() and not
2443 * the current task. this_rq() stays locked over invocation.
2444 */
2445static void try_to_wake_up_local(struct task_struct *p)
2446{
2447    struct rq *rq = task_rq(p);
2448    bool success = false;
2449
2450    BUG_ON(rq != this_rq());
2451    BUG_ON(p == current);
2452    lockdep_assert_held(&rq->lock);
2453
2454    if (!(p->state & TASK_NORMAL))
2455        return;
2456
2457    if (!p->se.on_rq) {
2458        if (likely(!task_running(rq, p))) {
2459            schedstat_inc(rq, ttwu_count);
2460            schedstat_inc(rq, ttwu_local);
2461        }
2462        ttwu_activate(p, rq, false, false, true, ENQUEUE_WAKEUP);
2463        success = true;
2464    }
2465    ttwu_post_activation(p, rq, 0, success);
2466}
2467
2468/**
2469 * wake_up_process - Wake up a specific process
2470 * @p: The process to be woken up.
2471 *
2472 * Attempt to wake up the nominated process and move it to the set of runnable
2473 * processes. Returns 1 if the process was woken up, 0 if it was already
2474 * running.
2475 *
2476 * It may be assumed that this function implies a write memory barrier before
2477 * changing the task state if and only if any tasks are woken up.
2478 */
2479int wake_up_process(struct task_struct *p)
2480{
2481    return try_to_wake_up(p, TASK_ALL, 0);
2482}
2483EXPORT_SYMBOL(wake_up_process);
2484
2485int wake_up_state(struct task_struct *p, unsigned int state)
2486{
2487    return try_to_wake_up(p, state, 0);
2488}
2489
2490/*
2491 * Perform scheduler related setup for a newly forked process p.
2492 * p is forked by current.
2493 *
2494 * __sched_fork() is basic setup used by init_idle() too:
2495 */
2496static void __sched_fork(struct task_struct *p)
2497{
2498    p->se.exec_start = 0;
2499    p->se.sum_exec_runtime = 0;
2500    p->se.prev_sum_exec_runtime = 0;
2501    p->se.nr_migrations = 0;
2502
2503#ifdef CONFIG_SCHEDSTATS
2504    memset(&p->se.statistics, 0, sizeof(p->se.statistics));
2505#endif
2506
2507    INIT_LIST_HEAD(&p->rt.run_list);
2508    p->se.on_rq = 0;
2509    INIT_LIST_HEAD(&p->se.group_node);
2510
2511#ifdef CONFIG_PREEMPT_NOTIFIERS
2512    INIT_HLIST_HEAD(&p->preempt_notifiers);
2513#endif
2514}
2515
2516/*
2517 * fork()/clone()-time setup:
2518 */
2519void sched_fork(struct task_struct *p, int clone_flags)
2520{
2521    int cpu = get_cpu();
2522
2523    __sched_fork(p);
2524    /*
2525     * We mark the process as running here. This guarantees that
2526     * nobody will actually run it, and a signal or other external
2527     * event cannot wake it up and insert it on the runqueue either.
2528     */
2529    p->state = TASK_RUNNING;
2530
2531    /*
2532     * Revert to default priority/policy on fork if requested.
2533     */
2534    if (unlikely(p->sched_reset_on_fork)) {
2535        if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) {
2536            p->policy = SCHED_NORMAL;
2537            p->normal_prio = p->static_prio;
2538        }
2539
2540        if (PRIO_TO_NICE(p->static_prio) < 0) {
2541            p->static_prio = NICE_TO_PRIO(0);
2542            p->normal_prio = p->static_prio;
2543            set_load_weight(p);
2544        }
2545
2546        /*
2547         * We don't need the reset flag anymore after the fork. It has
2548         * fulfilled its duty:
2549         */
2550        p->sched_reset_on_fork = 0;
2551    }
2552
2553    /*
2554     * Make sure we do not leak PI boosting priority to the child.
2555     */
2556    p->prio = current->normal_prio;
2557
2558    if (!rt_prio(p->prio))
2559        p->sched_class = &fair_sched_class;
2560
2561    if (p->sched_class->task_fork)
2562        p->sched_class->task_fork(p);
2563
2564    /*
2565     * The child is not yet in the pid-hash so no cgroup attach races,
2566     * and the cgroup is pinned to this child due to cgroup_fork()
2567     * is ran before sched_fork().
2568     *
2569     * Silence PROVE_RCU.
2570     */
2571    rcu_read_lock();
2572    set_task_cpu(p, cpu);
2573    rcu_read_unlock();
2574
2575#if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT)
2576    if (likely(sched_info_on()))
2577        memset(&p->sched_info, 0, sizeof(p->sched_info));
2578#endif
2579#if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW)
2580    p->oncpu = 0;
2581#endif
2582#ifdef CONFIG_PREEMPT
2583    /* Want to start with kernel preemption disabled. */
2584    task_thread_info(p)->preempt_count = 1;
2585#endif
2586    plist_node_init(&p->pushable_tasks, MAX_PRIO);
2587
2588    put_cpu();
2589}
2590
2591/*
2592 * wake_up_new_task - wake up a newly created task for the first time.
2593 *
2594 * This function will do some initial scheduler statistics housekeeping
2595 * that must be done for every newly created context, then puts the task
2596 * on the runqueue and wakes it.
2597 */
2598void wake_up_new_task(struct task_struct *p, unsigned long clone_flags)
2599{
2600    unsigned long flags;
2601    struct rq *rq;
2602    int cpu __maybe_unused = get_cpu();
2603
2604#ifdef CONFIG_SMP
2605    rq = task_rq_lock(p, &flags);
2606    p->state = TASK_WAKING;
2607
2608    /*
2609     * Fork balancing, do it here and not earlier because:
2610     * - cpus_allowed can change in the fork path
2611     * - any previously selected cpu might disappear through hotplug
2612     *
2613     * We set TASK_WAKING so that select_task_rq() can drop rq->lock
2614     * without people poking at ->cpus_allowed.
2615     */
2616    cpu = select_task_rq(rq, p, SD_BALANCE_FORK, 0);
2617    set_task_cpu(p, cpu);
2618
2619    p->state = TASK_RUNNING;
2620    task_rq_unlock(rq, &flags);
2621#endif
2622
2623    rq = task_rq_lock(p, &flags);
2624    activate_task(rq, p, 0);
2625    trace_sched_wakeup_new(p, 1);
2626    check_preempt_curr(rq, p, WF_FORK);
2627#ifdef CONFIG_SMP
2628    if (p->sched_class->task_woken)
2629        p->sched_class->task_woken(rq, p);
2630#endif
2631    task_rq_unlock(rq, &flags);
2632    put_cpu();
2633}
2634
2635#ifdef CONFIG_PREEMPT_NOTIFIERS
2636
2637/**
2638 * preempt_notifier_register - tell me when current is being preempted & rescheduled
2639 * @notifier: notifier struct to register
2640 */
2641void preempt_notifier_register(struct preempt_notifier *notifier)
2642{
2643    hlist_add_head(&notifier->link, &current->preempt_notifiers);
2644}
2645EXPORT_SYMBOL_GPL(preempt_notifier_register);
2646
2647/**
2648 * preempt_notifier_unregister - no longer interested in preemption notifications
2649 * @notifier: notifier struct to unregister
2650 *
2651 * This is safe to call from within a preemption notifier.
2652 */
2653void preempt_notifier_unregister(struct preempt_notifier *notifier)
2654{
2655    hlist_del(&notifier->link);
2656}
2657EXPORT_SYMBOL_GPL(preempt_notifier_unregister);
2658
2659static void fire_sched_in_preempt_notifiers(struct task_struct *curr)
2660{
2661    struct preempt_notifier *notifier;
2662    struct hlist_node *node;
2663
2664    hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link)
2665        notifier->ops->sched_in(notifier, raw_smp_processor_id());
2666}
2667
2668static void
2669fire_sched_out_preempt_notifiers(struct task_struct *curr,
2670                 struct task_struct *next)
2671{
2672    struct preempt_notifier *notifier;
2673    struct hlist_node *node;
2674
2675    hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link)
2676        notifier->ops->sched_out(notifier, next);
2677}
2678
2679#else /* !CONFIG_PREEMPT_NOTIFIERS */
2680
2681static void fire_sched_in_preempt_notifiers(struct task_struct *curr)
2682{
2683}
2684
2685static void
2686fire_sched_out_preempt_notifiers(struct task_struct *curr,
2687                 struct task_struct *next)
2688{
2689}
2690
2691#endif /* CONFIG_PREEMPT_NOTIFIERS */
2692
2693/**
2694 * prepare_task_switch - prepare to switch tasks
2695 * @rq: the runqueue preparing to switch
2696 * @prev: the current task that is being switched out
2697 * @next: the task we are going to switch to.
2698 *
2699 * This is called with the rq lock held and interrupts off. It must
2700 * be paired with a subsequent finish_task_switch after the context
2701 * switch.
2702 *
2703 * prepare_task_switch sets up locking and calls architecture specific
2704 * hooks.
2705 */
2706static inline void
2707prepare_task_switch(struct rq *rq, struct task_struct *prev,
2708            struct task_struct *next)
2709{
2710    fire_sched_out_preempt_notifiers(prev, next);
2711    prepare_lock_switch(rq, next);
2712    prepare_arch_switch(next);
2713}
2714
2715/**
2716 * finish_task_switch - clean up after a task-switch
2717 * @rq: runqueue associated with task-switch
2718 * @prev: the thread we just switched away from.
2719 *
2720 * finish_task_switch must be called after the context switch, paired
2721 * with a prepare_task_switch call before the context switch.
2722 * finish_task_switch will reconcile locking set up by prepare_task_switch,
2723 * and do any other architecture-specific cleanup actions.
2724 *
2725 * Note that we may have delayed dropping an mm in context_switch(). If
2726 * so, we finish that here outside of the runqueue lock. (Doing it
2727 * with the lock held can cause deadlocks; see schedule() for
2728 * details.)
2729 */
2730static void finish_task_switch(struct rq *rq, struct task_struct *prev)
2731    __releases(rq->lock)
2732{
2733    struct mm_struct *mm = rq->prev_mm;
2734    long prev_state;
2735
2736    rq->prev_mm = NULL;
2737
2738    /*
2739     * A task struct has one reference for the use as "current".
2740     * If a task dies, then it sets TASK_DEAD in tsk->state and calls
2741     * schedule one last time. The schedule call will never return, and
2742     * the scheduled task must drop that reference.
2743     * The test for TASK_DEAD must occur while the runqueue locks are
2744     * still held, otherwise prev could be scheduled on another cpu, die
2745     * there before we look at prev->state, and then the reference would
2746     * be dropped twice.
2747     * Manfred Spraul <manfred@colorfullife.com>
2748     */
2749    prev_state = prev->state;
2750    finish_arch_switch(prev);
2751#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
2752    local_irq_disable();
2753#endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */
2754    perf_event_task_sched_in(current);
2755#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
2756    local_irq_enable();
2757#endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */
2758    finish_lock_switch(rq, prev);
2759
2760    fire_sched_in_preempt_notifiers(current);
2761    if (mm)
2762        mmdrop(mm);
2763    if (unlikely(prev_state == TASK_DEAD)) {
2764        /*
2765         * Remove function-return probe instances associated with this
2766         * task and put them back on the free list.
2767         */
2768        kprobe_flush_task(prev);
2769        put_task_struct(prev);
2770    }
2771}
2772
2773#ifdef CONFIG_SMP
2774
2775/* assumes rq->lock is held */
2776static inline void pre_schedule(struct rq *rq, struct task_struct *prev)
2777{
2778    if (prev->sched_class->pre_schedule)
2779        prev->sched_class->pre_schedule(rq, prev);
2780}
2781
2782/* rq->lock is NOT held, but preemption is disabled */
2783static inline void post_schedule(struct rq *rq)
2784{
2785    if (rq->post_schedule) {
2786        unsigned long flags;
2787
2788        raw_spin_lock_irqsave(&rq->lock, flags);
2789        if (rq->curr->sched_class->post_schedule)
2790            rq->curr->sched_class->post_schedule(rq);
2791        raw_spin_unlock_irqrestore(&rq->lock, flags);
2792
2793        rq->post_schedule = 0;
2794    }
2795}
2796
2797#else
2798
2799static inline void pre_schedule(struct rq *rq, struct task_struct *p)
2800{
2801}
2802
2803static inline void post_schedule(struct rq *rq)
2804{
2805}
2806
2807#endif
2808
2809/**
2810 * schedule_tail - first thing a freshly forked thread must call.
2811 * @prev: the thread we just switched away from.
2812 */
2813asmlinkage void schedule_tail(struct task_struct *prev)
2814    __releases(rq->lock)
2815{
2816    struct rq *rq = this_rq();
2817
2818    finish_task_switch(rq, prev);
2819
2820    /*
2821     * FIXME: do we need to worry about rq being invalidated by the
2822     * task_switch?
2823     */
2824    post_schedule(rq);
2825
2826#ifdef __ARCH_WANT_UNLOCKED_CTXSW
2827    /* In this case, finish_task_switch does not reenable preemption */
2828    preempt_enable();
2829#endif
2830    if (current->set_child_tid)
2831        put_user(task_pid_vnr(current), current->set_child_tid);
2832}
2833
2834/*
2835 * context_switch - switch to the new MM and the new
2836 * thread's register state.
2837 */
2838static inline void
2839context_switch(struct rq *rq, struct task_struct *prev,
2840           struct task_struct *next)
2841{
2842    struct mm_struct *mm, *oldmm;
2843
2844    prepare_task_switch(rq, prev, next);
2845    trace_sched_switch(prev, next);
2846    mm = next->mm;
2847    oldmm = prev->active_mm;
2848    /*
2849     * For paravirt, this is coupled with an exit in switch_to to
2850     * combine the page table reload and the switch backend into
2851     * one hypercall.
2852     */
2853    arch_start_context_switch(prev);
2854
2855    if (likely(!mm)) {
2856        next->active_mm = oldmm;
2857        atomic_inc(&oldmm->mm_count);
2858        enter_lazy_tlb(oldmm, next);
2859    } else
2860        switch_mm(oldmm, mm, next);
2861
2862    if (likely(!prev->mm)) {
2863        prev->active_mm = NULL;
2864        rq->prev_mm = oldmm;
2865    }
2866    /*
2867     * Since the runqueue lock will be released by the next
2868     * task (which is an invalid locking op but in the case
2869     * of the scheduler it's an obvious special-case), so we
2870     * do an early lockdep release here:
2871     */
2872#ifndef __ARCH_WANT_UNLOCKED_CTXSW
2873    spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
2874#endif
2875
2876    /* Here we just switch the register state and the stack. */
2877    switch_to(prev, next, prev);
2878
2879    barrier();
2880    /*
2881     * this_rq must be evaluated again because prev may have moved
2882     * CPUs since it called schedule(), thus the 'rq' on its stack
2883     * frame will be invalid.
2884     */
2885    finish_task_switch(this_rq(), prev);
2886}
2887
2888/*
2889 * nr_running, nr_uninterruptible and nr_context_switches:
2890 *
2891 * externally visible scheduler statistics: current number of runnable
2892 * threads, current number of uninterruptible-sleeping threads, total
2893 * number of context switches performed since bootup.
2894 */
2895unsigned long nr_running(void)
2896{
2897    unsigned long i, sum = 0;
2898
2899    for_each_online_cpu(i)
2900        sum += cpu_rq(i)->nr_running;
2901
2902    return sum;
2903}
2904
2905unsigned long nr_uninterruptible(void)
2906{
2907    unsigned long i, sum = 0;
2908
2909    for_each_possible_cpu(i)
2910        sum += cpu_rq(i)->nr_uninterruptible;
2911
2912    /*
2913     * Since we read the counters lockless, it might be slightly
2914     * inaccurate. Do not allow it to go below zero though:
2915     */
2916    if (unlikely((long)sum < 0))
2917        sum = 0;
2918
2919    return sum;
2920}
2921
2922unsigned long long nr_context_switches(void)
2923{
2924    int i;
2925    unsigned long long sum = 0;
2926
2927    for_each_possible_cpu(i)
2928        sum += cpu_rq(i)->nr_switches;
2929
2930    return sum;
2931}
2932
2933unsigned long nr_iowait(void)
2934{
2935    unsigned long i, sum = 0;
2936
2937    for_each_possible_cpu(i)
2938        sum += atomic_read(&cpu_rq(i)->nr_iowait);
2939
2940    return sum;
2941}
2942
2943unsigned long nr_iowait_cpu(int cpu)
2944{
2945    struct rq *this = cpu_rq(cpu);
2946    return atomic_read(&this->nr_iowait);
2947}
2948
2949unsigned long this_cpu_load(void)
2950{
2951    struct rq *this = this_rq();
2952    return this->cpu_load[0];
2953}
2954
2955
2956/* Variables and functions for calc_load */
2957static atomic_long_t calc_load_tasks;
2958static unsigned long calc_load_update;
2959unsigned long avenrun[3];
2960EXPORT_SYMBOL(avenrun);
2961
2962static long calc_load_fold_active(struct rq *this_rq)
2963{
2964    long nr_active, delta = 0;
2965
2966    nr_active = this_rq->nr_running;
2967    nr_active += (long) this_rq->nr_uninterruptible;
2968
2969    if (nr_active != this_rq->calc_load_active) {
2970        delta = nr_active - this_rq->calc_load_active;
2971        this_rq->calc_load_active = nr_active;
2972    }
2973
2974    return delta;
2975}
2976
2977#ifdef CONFIG_NO_HZ
2978/*
2979 * For NO_HZ we delay the active fold to the next LOAD_FREQ update.
2980 *
2981 * When making the ILB scale, we should try to pull this in as well.
2982 */
2983static atomic_long_t calc_load_tasks_idle;
2984
2985static void calc_load_account_idle(struct rq *this_rq)
2986{
2987    long delta;
2988
2989    delta = calc_load_fold_active(this_rq);
2990    if (delta)
2991        atomic_long_add(delta, &calc_load_tasks_idle);
2992}
2993
2994static long calc_load_fold_idle(void)
2995{
2996    long delta = 0;
2997
2998    /*
2999     * Its got a race, we don't care...
3000     */
3001    if (atomic_long_read(&calc_load_tasks_idle))
3002        delta = atomic_long_xchg(&calc_load_tasks_idle, 0);
3003
3004    return delta;
3005}
3006#else
3007static void calc_load_account_idle(struct rq *this_rq)
3008{
3009}
3010
3011static inline long calc_load_fold_idle(void)
3012{
3013    return 0;
3014}
3015#endif
3016
3017/**
3018 * get_avenrun - get the load average array
3019 * @loads: pointer to dest load array
3020 * @offset: offset to add
3021 * @shift: shift count to shift the result left
3022 *
3023 * These values are estimates at best, so no need for locking.
3024 */
3025void get_avenrun(unsigned long *loads, unsigned long offset, int shift)
3026{
3027    loads[0] = (avenrun[0] + offset) << shift;
3028    loads[1] = (avenrun[1] + offset) << shift;
3029    loads[2] = (avenrun[2] + offset) << shift;
3030}
3031
3032static unsigned long
3033calc_load(unsigned long load, unsigned long exp, unsigned long active)
3034{
3035    load *= exp;
3036    load += active * (FIXED_1 - exp);
3037    return load >> FSHIFT;
3038}
3039
3040/*
3041 * calc_load - update the avenrun load estimates 10 ticks after the
3042 * CPUs have updated calc_load_tasks.
3043 */
3044void calc_global_load(void)
3045{
3046    unsigned long upd = calc_load_update + 10;
3047    long active;
3048
3049    if (time_before(jiffies, upd))
3050        return;
3051
3052    active = atomic_long_read(&calc_load_tasks);
3053    active = active > 0 ? active * FIXED_1 : 0;
3054
3055    avenrun[0] = calc_load(avenrun[0], EXP_1, active);
3056    avenrun[1] = calc_load(avenrun[1], EXP_5, active);
3057    avenrun[2] = calc_load(avenrun[2], EXP_15, active);
3058
3059    calc_load_update += LOAD_FREQ;
3060}
3061
3062/*
3063 * Called from update_cpu_load() to periodically update this CPU's
3064 * active count.
3065 */
3066static void calc_load_account_active(struct rq *this_rq)
3067{
3068    long delta;
3069
3070    if (time_before(jiffies, this_rq->calc_load_update))
3071        return;
3072
3073    delta = calc_load_fold_active(this_rq);
3074    delta += calc_load_fold_idle();
3075    if (delta)
3076        atomic_long_add(delta, &calc_load_tasks);
3077
3078    this_rq->calc_load_update += LOAD_FREQ;
3079}
3080
3081/*
3082 * The exact cpuload at various idx values, calculated at every tick would be
3083 * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load
3084 *
3085 * If a cpu misses updates for n-1 ticks (as it was idle) and update gets called
3086 * on nth tick when cpu may be busy, then we have:
3087 * load = ((2^idx - 1) / 2^idx)^(n-1) * load
3088 * load = (2^idx - 1) / 2^idx) * load + 1 / 2^idx * cur_load
3089 *
3090 * decay_load_missed() below does efficient calculation of
3091 * load = ((2^idx - 1) / 2^idx)^(n-1) * load
3092 * avoiding 0..n-1 loop doing load = ((2^idx - 1) / 2^idx) * load
3093 *
3094 * The calculation is approximated on a 128 point scale.
3095 * degrade_zero_ticks is the number of ticks after which load at any
3096 * particular idx is approximated to be zero.
3097 * degrade_factor is a precomputed table, a row for each load idx.
3098 * Each column corresponds to degradation factor for a power of two ticks,
3099 * based on 128 point scale.
3100 * Example:
3101 * row 2, col 3 (=12) says that the degradation at load idx 2 after
3102 * 8 ticks is 12/128 (which is an approximation of exact factor 3^8/4^8).
3103 *
3104 * With this power of 2 load factors, we can degrade the load n times
3105 * by looking at 1 bits in n and doing as many mult/shift instead of
3106 * n mult/shifts needed by the exact degradation.
3107 */
3108#define DEGRADE_SHIFT 7
3109static const unsigned char
3110        degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128};
3111static const unsigned char
3112        degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = {
3113                    {0, 0, 0, 0, 0, 0, 0, 0},
3114                    {64, 32, 8, 0, 0, 0, 0, 0},
3115                    {96, 72, 40, 12, 1, 0, 0},
3116                    {112, 98, 75, 43, 15, 1, 0},
3117                    {120, 112, 98, 76, 45, 16, 2} };
3118
3119/*
3120 * Update cpu_load for any missed ticks, due to tickless idle. The backlog
3121 * would be when CPU is idle and so we just decay the old load without
3122 * adding any new load.
3123 */
3124static unsigned long
3125decay_load_missed(unsigned long load, unsigned long missed_updates, int idx)
3126{
3127    int j = 0;
3128
3129    if (!missed_updates)
3130        return load;
3131
3132    if (missed_updates >= degrade_zero_ticks[idx])
3133        return 0;
3134
3135    if (idx == 1)
3136        return load >> missed_updates;
3137
3138    while (missed_updates) {
3139        if (missed_updates % 2)
3140            load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT;
3141
3142        missed_updates >>= 1;
3143        j++;
3144    }
3145    return load;
3146}
3147
3148/*
3149 * Update rq->cpu_load[] statistics. This function is usually called every
3150 * scheduler tick (TICK_NSEC). With tickless idle this will not be called
3151 * every tick. We fix it up based on jiffies.
3152 */
3153static void update_cpu_load(struct rq *this_rq)
3154{
3155    unsigned long this_load = this_rq->load.weight;
3156    unsigned long curr_jiffies = jiffies;
3157    unsigned long pending_updates;
3158    int i, scale;
3159
3160    this_rq->nr_load_updates++;
3161
3162    /* Avoid repeated calls on same jiffy, when moving in and out of idle */
3163    if (curr_jiffies == this_rq->last_load_update_tick)
3164        return;
3165
3166    pending_updates = curr_jiffies - this_rq->last_load_update_tick;
3167    this_rq->last_load_update_tick = curr_jiffies;
3168
3169    /* Update our load: */
3170    this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */
3171    for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) {
3172        unsigned long old_load, new_load;
3173
3174        /* scale is effectively 1 << i now, and >> i divides by scale */
3175
3176        old_load = this_rq->cpu_load[i];
3177        old_load = decay_load_missed(old_load, pending_updates - 1, i);
3178        new_load = this_load;
3179        /*
3180         * Round up the averaging division if load is increasing. This
3181         * prevents us from getting stuck on 9 if the load is 10, for
3182         * example.
3183         */
3184        if (new_load > old_load)
3185            new_load += scale - 1;
3186
3187        this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i;
3188    }
3189
3190    sched_avg_update(this_rq);
3191}
3192
3193static void update_cpu_load_active(struct rq *this_rq)
3194{
3195    update_cpu_load(this_rq);
3196
3197    calc_load_account_active(this_rq);
3198}
3199
3200#ifdef CONFIG_SMP
3201
3202/*
3203 * sched_exec - execve() is a valuable balancing opportunity, because at
3204 * this point the task has the smallest effective memory and cache footprint.
3205 */
3206void sched_exec(void)
3207{
3208    struct task_struct *p = current;
3209    unsigned long flags;
3210    struct rq *rq;
3211    int dest_cpu;
3212
3213    rq = task_rq_lock(p, &flags);
3214    dest_cpu = p->sched_class->select_task_rq(rq, p, SD_BALANCE_EXEC, 0);
3215    if (dest_cpu == smp_processor_id())
3216        goto unlock;
3217
3218    /*
3219     * select_task_rq() can race against ->cpus_allowed
3220     */
3221    if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed) &&
3222        likely(cpu_active(dest_cpu)) && migrate_task(p, dest_cpu)) {
3223        struct migration_arg arg = { p, dest_cpu };
3224
3225        task_rq_unlock(rq, &flags);
3226        stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg);
3227        return;
3228    }
3229unlock:
3230    task_rq_unlock(rq, &flags);
3231}
3232
3233#endif
3234
3235DEFINE_PER_CPU(struct kernel_stat, kstat);
3236
3237EXPORT_PER_CPU_SYMBOL(kstat);
3238
3239/*
3240 * Return any ns on the sched_clock that have not yet been accounted in
3241 * @p in case that task is currently running.
3242 *
3243 * Called with task_rq_lock() held on @rq.
3244 */
3245static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq)
3246{
3247    u64 ns = 0;
3248
3249    if (task_current(rq, p)) {
3250        update_rq_clock(rq);
3251        ns = rq->clock - p->se.exec_start;
3252        if ((s64)ns < 0)
3253            ns = 0;
3254    }
3255
3256    return ns;
3257}
3258
3259unsigned long long task_delta_exec(struct task_struct *p)
3260{
3261    unsigned long flags;
3262    struct rq *rq;
3263    u64 ns = 0;
3264
3265    rq = task_rq_lock(p, &flags);
3266    ns = do_task_delta_exec(p, rq);
3267    task_rq_unlock(rq, &flags);
3268
3269    return ns;
3270}
3271
3272/*
3273 * Return accounted runtime for the task.
3274 * In case the task is currently running, return the runtime plus current's
3275 * pending runtime that have not been accounted yet.
3276 */
3277unsigned long long task_sched_runtime(struct task_struct *p)
3278{
3279    unsigned long flags;
3280    struct rq *rq;
3281    u64 ns = 0;
3282
3283    rq = task_rq_lock(p, &flags);
3284    ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq);
3285    task_rq_unlock(rq, &flags);
3286
3287    return ns;
3288}
3289
3290/*
3291 * Return sum_exec_runtime for the thread group.
3292 * In case the task is currently running, return the sum plus current's
3293 * pending runtime that have not been accounted yet.
3294 *
3295 * Note that the thread group might have other running tasks as well,
3296 * so the return value not includes other pending runtime that other
3297 * running tasks might have.
3298 */
3299unsigned long long thread_group_sched_runtime(struct task_struct *p)
3300{
3301    struct task_cputime totals;
3302    unsigned long flags;
3303    struct rq *rq;
3304    u64 ns;
3305
3306    rq = task_rq_lock(p, &flags);
3307    thread_group_cputime(p, &totals);
3308    ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq);
3309    task_rq_unlock(rq, &flags);
3310
3311    return ns;
3312}
3313
3314/*
3315 * Account user cpu time to a process.
3316 * @p: the process that the cpu time gets accounted to
3317 * @cputime: the cpu time spent in user space since the last update
3318 * @cputime_scaled: cputime scaled by cpu frequency
3319 */
3320void account_user_time(struct task_struct *p, cputime_t cputime,
3321               cputime_t cputime_scaled)
3322{
3323    struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3324    cputime64_t tmp;
3325
3326    /* Add user time to process. */
3327    p->utime = cputime_add(p->utime, cputime);
3328    p->utimescaled = cputime_add(p->utimescaled, cputime_scaled);
3329    account_group_user_time(p, cputime);
3330
3331    /* Add user time to cpustat. */
3332    tmp = cputime_to_cputime64(cputime);
3333    if (TASK_NICE(p) > 0)
3334        cpustat->nice = cputime64_add(cpustat->nice, tmp);
3335    else
3336        cpustat->user = cputime64_add(cpustat->user, tmp);
3337
3338    cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime);
3339    /* Account for user time used */
3340    acct_update_integrals(p);
3341}
3342
3343/*
3344 * Account guest cpu time to a process.
3345 * @p: the process that the cpu time gets accounted to
3346 * @cputime: the cpu time spent in virtual machine since the last update
3347 * @cputime_scaled: cputime scaled by cpu frequency
3348 */
3349static void account_guest_time(struct task_struct *p, cputime_t cputime,
3350                   cputime_t cputime_scaled)
3351{
3352    cputime64_t tmp;
3353    struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3354
3355    tmp = cputime_to_cputime64(cputime);
3356
3357    /* Add guest time to process. */
3358    p->utime = cputime_add(p->utime, cputime);
3359    p->utimescaled = cputime_add(p->utimescaled, cputime_scaled);
3360    account_group_user_time(p, cputime);
3361    p->gtime = cputime_add(p->gtime, cputime);
3362
3363    /* Add guest time to cpustat. */
3364    if (TASK_NICE(p) > 0) {
3365        cpustat->nice = cputime64_add(cpustat->nice, tmp);
3366        cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp);
3367    } else {
3368        cpustat->user = cputime64_add(cpustat->user, tmp);
3369        cpustat->guest = cputime64_add(cpustat->guest, tmp);
3370    }
3371}
3372
3373/*
3374 * Account system cpu time to a process.
3375 * @p: the process that the cpu time gets accounted to
3376 * @hardirq_offset: the offset to subtract from hardirq_count()
3377 * @cputime: the cpu time spent in kernel space since the last update
3378 * @cputime_scaled: cputime scaled by cpu frequency
3379 */
3380void account_system_time(struct task_struct *p, int hardirq_offset,
3381             cputime_t cputime, cputime_t cputime_scaled)
3382{
3383    struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3384    cputime64_t tmp;
3385
3386    if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) {
3387        account_guest_time(p, cputime, cputime_scaled);
3388        return;
3389    }
3390
3391    /* Add system time to process. */
3392    p->stime = cputime_add(p->stime, cputime);
3393    p->stimescaled = cputime_add(p->stimescaled, cputime_scaled);
3394    account_group_system_time(p, cputime);
3395
3396    /* Add system time to cpustat. */
3397    tmp = cputime_to_cputime64(cputime);
3398    if (hardirq_count() - hardirq_offset)
3399        cpustat->irq = cputime64_add(cpustat->irq, tmp);
3400    else if (softirq_count())
3401        cpustat->softirq = cputime64_add(cpustat->softirq, tmp);
3402    else
3403        cpustat->system = cputime64_add(cpustat->system, tmp);
3404
3405    cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime);
3406
3407    /* Account for system time used */
3408    acct_update_integrals(p);
3409}
3410
3411/*
3412 * Account for involuntary wait time.
3413 * @steal: the cpu time spent in involuntary wait
3414 */
3415void account_steal_time(cputime_t cputime)
3416{
3417    struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3418    cputime64_t cputime64 = cputime_to_cputime64(cputime);
3419
3420    cpustat->steal = cputime64_add(cpustat->steal, cputime64);
3421}
3422
3423/*
3424 * Account for idle time.
3425 * @cputime: the cpu time spent in idle wait
3426 */
3427void account_idle_time(cputime_t cputime)
3428{
3429    struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3430    cputime64_t cputime64 = cputime_to_cputime64(cputime);
3431    struct rq *rq = this_rq();
3432
3433    if (atomic_read(&rq->nr_iowait) > 0)
3434        cpustat->iowait = cputime64_add(cpustat->iowait, cputime64);
3435    else
3436        cpustat->idle = cputime64_add(cpustat->idle, cputime64);
3437}
3438
3439#ifndef CONFIG_VIRT_CPU_ACCOUNTING
3440
3441/*
3442 * Account a single tick of cpu time.
3443 * @p: the process that the cpu time gets accounted to
3444 * @user_tick: indicates if the tick is a user or a system tick
3445 */
3446void account_process_tick(struct task_struct *p, int user_tick)
3447{
3448    cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy);
3449    struct rq *rq = this_rq();
3450
3451    if (user_tick)
3452        account_user_time(p, cputime_one_jiffy, one_jiffy_scaled);
3453    else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET))
3454        account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy,
3455                    one_jiffy_scaled);
3456    else
3457        account_idle_time(cputime_one_jiffy);
3458}
3459
3460/*
3461 * Account multiple ticks of steal time.
3462 * @p: the process from which the cpu time has been stolen
3463 * @ticks: number of stolen ticks
3464 */
3465void account_steal_ticks(unsigned long ticks)
3466{
3467    account_steal_time(jiffies_to_cputime(ticks));
3468}
3469
3470/*
3471 * Account multiple ticks of idle time.
3472 * @ticks: number of stolen ticks
3473 */
3474void account_idle_ticks(unsigned long ticks)
3475{
3476    account_idle_time(jiffies_to_cputime(ticks));
3477}
3478
3479#endif
3480
3481/*
3482 * Use precise platform statistics if available:
3483 */
3484#ifdef CONFIG_VIRT_CPU_ACCOUNTING
3485void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
3486{
3487    *ut = p->utime;
3488    *st = p->stime;
3489}
3490
3491void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
3492{
3493    struct task_cputime cputime;
3494
3495    thread_group_cputime(p, &cputime);
3496
3497    *ut = cputime.utime;
3498    *st = cputime.stime;
3499}
3500#else
3501
3502#ifndef nsecs_to_cputime
3503# define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs)
3504#endif
3505
3506void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
3507{
3508    cputime_t rtime, utime = p->utime, total = cputime_add(utime, p->stime);
3509
3510    /*
3511     * Use CFS's precise accounting:
3512     */
3513    rtime = nsecs_to_cputime(p->se.sum_exec_runtime);
3514
3515    if (total) {
3516        u64 temp = rtime;
3517
3518        temp *= utime;
3519        do_div(temp, total);
3520        utime = (cputime_t)temp;
3521    } else
3522        utime = rtime;
3523
3524    /*
3525     * Compare with previous values, to keep monotonicity:
3526     */
3527    p->prev_utime = max(p->prev_utime, utime);
3528    p->prev_stime = max(p->prev_stime, cputime_sub(rtime, p->prev_utime));
3529
3530    *ut = p->prev_utime;
3531    *st = p->prev_stime;
3532}
3533
3534/*
3535 * Must be called with siglock held.
3536 */
3537void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
3538{
3539    struct signal_struct *sig = p->signal;
3540    struct task_cputime cputime;
3541    cputime_t rtime, utime, total;
3542
3543    thread_group_cputime(p, &cputime);
3544
3545    total = cputime_add(cputime.utime, cputime.stime);
3546    rtime = nsecs_to_cputime(cputime.sum_exec_runtime);
3547
3548    if (total) {
3549        u64 temp = rtime;
3550
3551        temp *= cputime.utime;
3552        do_div(temp, total);
3553        utime = (cputime_t)temp;
3554    } else
3555        utime = rtime;
3556
3557    sig->prev_utime = max(sig->prev_utime, utime);
3558    sig->prev_stime = max(sig->prev_stime,
3559                  cputime_sub(rtime, sig->prev_utime));
3560
3561    *ut = sig->prev_utime;
3562    *st = sig->prev_stime;
3563}
3564#endif
3565
3566/*
3567 * This function gets called by the timer code, with HZ frequency.
3568 * We call it with interrupts disabled.
3569 *
3570 * It also gets called by the fork code, when changing the parent's
3571 * timeslices.
3572 */
3573void scheduler_tick(void)
3574{
3575    int cpu = smp_processor_id();
3576    struct rq *rq = cpu_rq(cpu);
3577    struct task_struct *curr = rq->curr;
3578
3579    sched_clock_tick();
3580
3581    raw_spin_lock(&rq->lock);
3582    update_rq_clock(rq);
3583    update_cpu_load_active(rq);
3584    curr->sched_class->task_tick(rq, curr, 0);
3585    raw_spin_unlock(&rq->lock);
3586
3587    perf_event_task_tick(curr);
3588
3589#ifdef CONFIG_SMP
3590    rq->idle_at_tick = idle_cpu(cpu);
3591    trigger_load_balance(rq, cpu);
3592#endif
3593}
3594
3595notrace unsigned long get_parent_ip(unsigned long addr)
3596{
3597    if (in_lock_functions(addr)) {
3598        addr = CALLER_ADDR2;
3599        if (in_lock_functions(addr))
3600            addr = CALLER_ADDR3;
3601    }
3602    return addr;
3603}
3604
3605#if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \
3606                defined(CONFIG_PREEMPT_TRACER))
3607
3608void __kprobes add_preempt_count(int val)
3609{
3610#ifdef CONFIG_DEBUG_PREEMPT
3611    /*
3612     * Underflow?
3613     */
3614    if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0)))
3615        return;
3616#endif
3617    preempt_count() += val;
3618#ifdef CONFIG_DEBUG_PREEMPT
3619    /*
3620     * Spinlock count overflowing soon?
3621     */
3622    DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >=
3623                PREEMPT_MASK - 10);
3624#endif
3625    if (preempt_count() == val)
3626        trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1));
3627}
3628EXPORT_SYMBOL(add_preempt_count);
3629
3630void __kprobes sub_preempt_count(int val)
3631{
3632#ifdef CONFIG_DEBUG_PREEMPT
3633    /*
3634     * Underflow?
3635     */
3636    if (DEBUG_LOCKS_WARN_ON(val > preempt_count()))
3637        return;
3638    /*
3639     * Is the spinlock portion underflowing?
3640     */
3641    if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) &&
3642            !(preempt_count() & PREEMPT_MASK)))
3643        return;
3644#endif
3645
3646    if (preempt_count() == val)
3647        trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1));
3648    preempt_count() -= val;
3649}
3650EXPORT_SYMBOL(sub_preempt_count);
3651
3652#endif
3653
3654/*
3655 * Print scheduling while atomic bug:
3656 */
3657static noinline void __schedule_bug(struct task_struct *prev)
3658{
3659    struct pt_regs *regs = get_irq_regs();
3660
3661    printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n",
3662        prev->comm, prev->pid, preempt_count());
3663
3664    debug_show_held_locks(prev);
3665    print_modules();
3666    if (irqs_disabled())
3667        print_irqtrace_events(prev);
3668
3669    if (regs)
3670        show_regs(regs);
3671    else
3672        dump_stack();
3673}
3674
3675/*
3676 * Various schedule()-time debugging checks and statistics:
3677 */
3678static inline void schedule_debug(struct task_struct *prev)
3679{
3680    /*
3681     * Test if we are atomic. Since do_exit() needs to call into
3682     * schedule() atomically, we ignore that path for now.
3683     * Otherwise, whine if we are scheduling when we should not be.
3684     */
3685    if (unlikely(in_atomic_preempt_off() && !prev->exit_state))
3686        __schedule_bug(prev);
3687
3688    profile_hit(SCHED_PROFILING, __builtin_return_address(0));
3689
3690    schedstat_inc(this_rq(), sched_count);
3691#ifdef CONFIG_SCHEDSTATS
3692    if (unlikely(prev->lock_depth >= 0)) {
3693        schedstat_inc(this_rq(), bkl_count);
3694        schedstat_inc(prev, sched_info.bkl_count);
3695    }
3696#endif
3697}
3698
3699static void put_prev_task(struct rq *rq, struct task_struct *prev)
3700{
3701    if (prev->se.on_rq)
3702        update_rq_clock(rq);
3703    rq->skip_clock_update = 0;
3704    prev->sched_class->put_prev_task(rq, prev);
3705}
3706
3707/*
3708 * Pick up the highest-prio task:
3709 */
3710static inline struct task_struct *
3711pick_next_task(struct rq *rq)
3712{
3713    const struct sched_class *class;
3714    struct task_struct *p;
3715
3716    /*
3717     * Optimization: we know that if all tasks are in
3718     * the fair class we can call that function directly:
3719     */
3720    if (likely(rq->nr_running == rq->cfs.nr_running)) {
3721        p = fair_sched_class.pick_next_task(rq);
3722        if (likely(p))
3723            return p;
3724    }
3725
3726    class = sched_class_highest;
3727    for ( ; ; ) {
3728        p = class->pick_next_task(rq);
3729        if (p)
3730            return p;
3731        /*
3732         * Will never be NULL as the idle class always
3733         * returns a non-NULL p:
3734         */
3735        class = class->next;
3736    }
3737}
3738
3739/*
3740 * schedule() is the main scheduler function.
3741 */
3742asmlinkage void __sched schedule(void)
3743{
3744    struct task_struct *prev, *next;
3745    unsigned long *switch_count;
3746    struct rq *rq;
3747    int cpu;
3748
3749need_resched:
3750    preempt_disable();
3751    cpu = smp_processor_id();
3752    rq = cpu_rq(cpu);
3753    rcu_note_context_switch(cpu);
3754    prev = rq->curr;
3755
3756    release_kernel_lock(prev);
3757need_resched_nonpreemptible:
3758
3759    schedule_debug(prev);
3760
3761    if (sched_feat(HRTICK))
3762        hrtick_clear(rq);
3763
3764    raw_spin_lock_irq(&rq->lock);
3765    clear_tsk_need_resched(prev);
3766
3767    switch_count = &prev->nivcsw;
3768    if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) {
3769        if (unlikely(signal_pending_state(prev->state, prev))) {
3770            prev->state = TASK_RUNNING;
3771        } else {
3772            /*
3773             * If a worker is going to sleep, notify and
3774             * ask workqueue whether it wants to wake up a
3775             * task to maintain concurrency. If so, wake
3776             * up the task.
3777             */
3778            if (prev->flags & PF_WQ_WORKER) {
3779                struct task_struct *to_wakeup;
3780
3781                to_wakeup = wq_worker_sleeping(prev, cpu);
3782                if (to_wakeup)
3783                    try_to_wake_up_local(to_wakeup);
3784            }
3785            deactivate_task(rq, prev, DEQUEUE_SLEEP);
3786        }
3787        switch_count = &prev->nvcsw;
3788    }
3789
3790    pre_schedule(rq, prev);
3791
3792    if (unlikely(!rq->nr_running))
3793        idle_balance(cpu, rq);
3794
3795    put_prev_task(rq, prev);
3796    next = pick_next_task(rq);
3797
3798    if (likely(prev != next)) {
3799        sched_info_switch(prev, next);
3800        perf_event_task_sched_out(prev, next);
3801
3802        rq->nr_switches++;
3803        rq->curr = next;
3804        ++*switch_count;
3805
3806        context_switch(rq, prev, next); /* unlocks the rq */
3807        /*
3808         * The context switch have flipped the stack from under us
3809         * and restored the local variables which were saved when
3810         * this task called schedule() in the past. prev == current
3811         * is still correct, but it can be moved to another cpu/rq.
3812         */
3813        cpu = smp_processor_id();
3814        rq = cpu_rq(cpu);
3815    } else
3816        raw_spin_unlock_irq(&rq->lock);
3817
3818    post_schedule(rq);
3819
3820    if (unlikely(reacquire_kernel_lock(prev)))
3821        goto need_resched_nonpreemptible;
3822
3823    preempt_enable_no_resched();
3824    if (need_resched())
3825        goto need_resched;
3826}
3827EXPORT_SYMBOL(schedule);
3828
3829#ifdef CONFIG_MUTEX_SPIN_ON_OWNER
3830/*
3831 * Look out! "owner" is an entirely speculative pointer
3832 * access and not reliable.
3833 */
3834int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner)
3835{
3836    unsigned int cpu;
3837    struct rq *rq;
3838
3839    if (!sched_feat(OWNER_SPIN))
3840        return 0;
3841
3842#ifdef CONFIG_DEBUG_PAGEALLOC
3843    /*
3844     * Need to access the cpu field knowing that
3845     * DEBUG_PAGEALLOC could have unmapped it if
3846     * the mutex owner just released it and exited.
3847     */
3848    if (probe_kernel_address(&owner->cpu, cpu))
3849        return 0;
3850#else
3851    cpu = owner->cpu;
3852#endif
3853
3854    /*
3855     * Even if the access succeeded (likely case),
3856     * the cpu field may no longer be valid.
3857     */
3858    if (cpu >= nr_cpumask_bits)
3859        return 0;
3860
3861    /*
3862     * We need to validate that we can do a
3863     * get_cpu() and that we have the percpu area.
3864     */
3865    if (!cpu_online(cpu))
3866        return 0;
3867
3868    rq = cpu_rq(cpu);
3869
3870    for (;;) {
3871        /*
3872         * Owner changed, break to re-assess state.
3873         */
3874        if (lock->owner != owner) {
3875            /*
3876             * If the lock has switched to a different owner,
3877             * we likely have heavy contention. Return 0 to quit
3878             * optimistic spinning and not contend further:
3879             */
3880            if (lock->owner)
3881                return 0;
3882            break;
3883        }
3884
3885        /*
3886         * Is that owner really running on that cpu?
3887         */
3888        if (task_thread_info(rq->curr) != owner || need_resched())
3889            return 0;
3890
3891        cpu_relax();
3892    }
3893
3894    return 1;
3895}
3896#endif
3897
3898#ifdef CONFIG_PREEMPT
3899/*
3900 * this is the entry point to schedule() from in-kernel preemption
3901 * off of preempt_enable. Kernel preemptions off return from interrupt
3902 * occur there and call schedule directly.
3903 */
3904asmlinkage void __sched notrace preempt_schedule(void)
3905{
3906    struct thread_info *ti = current_thread_info();
3907
3908    /*
3909     * If there is a non-zero preempt_count or interrupts are disabled,
3910     * we do not want to preempt the current task. Just return..
3911     */
3912    if (likely(ti->preempt_count || irqs_disabled()))
3913        return;
3914
3915    do {
3916        add_preempt_count_notrace(PREEMPT_ACTIVE);
3917        schedule();
3918        sub_preempt_count_notrace(PREEMPT_ACTIVE);
3919
3920        /*
3921         * Check again in case we missed a preemption opportunity
3922         * between schedule and now.
3923         */
3924        barrier();
3925    } while (need_resched());
3926}
3927EXPORT_SYMBOL(preempt_schedule);
3928
3929/*
3930 * this is the entry point to schedule() from kernel preemption
3931 * off of irq context.
3932 * Note, that this is called and return with irqs disabled. This will
3933 * protect us against recursive calling from irq.
3934 */
3935asmlinkage void __sched preempt_schedule_irq(void)
3936{
3937    struct thread_info *ti = current_thread_info();
3938
3939    /* Catch callers which need to be fixed */
3940    BUG_ON(ti->preempt_count || !irqs_disabled());
3941
3942    do {
3943        add_preempt_count(PREEMPT_ACTIVE);
3944        local_irq_enable();
3945        schedule();
3946        local_irq_disable();
3947        sub_preempt_count(PREEMPT_ACTIVE);
3948
3949        /*
3950         * Check again in case we missed a preemption opportunity
3951         * between schedule and now.
3952         */
3953        barrier();
3954    } while (need_resched());
3955}
3956
3957#endif /* CONFIG_PREEMPT */
3958
3959int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags,
3960              void *key)
3961{
3962    return try_to_wake_up(curr->private, mode, wake_flags);
3963}
3964EXPORT_SYMBOL(default_wake_function);
3965
3966/*
3967 * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just
3968 * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve
3969 * number) then we wake all the non-exclusive tasks and one exclusive task.
3970 *
3971 * There are circumstances in which we can try to wake a task which has already
3972 * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns
3973 * zero in this (rare) case, and we handle it by continuing to scan the queue.
3974 */
3975static void __wake_up_common(wait_queue_head_t *q, unsigned int mode,
3976            int nr_exclusive, int wake_flags, void *key)
3977{
3978    wait_queue_t *curr, *next;
3979
3980    list_for_each_entry_safe(curr, next, &q->task_list, task_list) {
3981        unsigned flags = curr->flags;
3982
3983        if (curr->func(curr, mode, wake_flags, key) &&
3984                (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive)
3985            break;
3986    }
3987}
3988
3989/**
3990 * __wake_up - wake up threads blocked on a waitqueue.
3991 * @q: the waitqueue
3992 * @mode: which threads
3993 * @nr_exclusive: how many wake-one or wake-many threads to wake up
3994 * @key: is directly passed to the wakeup function
3995 *
3996 * It may be assumed that this function implies a write memory barrier before
3997 * changing the task state if and only if any tasks are woken up.
3998 */
3999void __wake_up(wait_queue_head_t *q, unsigned int mode,
4000            int nr_exclusive, void *key)
4001{
4002    unsigned long flags;
4003
4004    spin_lock_irqsave(&q->lock, flags);
4005    __wake_up_common(q, mode, nr_exclusive, 0, key);
4006    spin_unlock_irqrestore(&q->lock, flags);
4007}
4008EXPORT_SYMBOL(__wake_up);
4009
4010/*
4011 * Same as __wake_up but called with the spinlock in wait_queue_head_t held.
4012 */
4013void __wake_up_locked(wait_queue_head_t *q, unsigned int mode)
4014{
4015    __wake_up_common(q, mode, 1, 0, NULL);
4016}
4017EXPORT_SYMBOL_GPL(__wake_up_locked);
4018
4019void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key)
4020{
4021    __wake_up_common(q, mode, 1, 0, key);
4022}
4023
4024/**
4025 * __wake_up_sync_key - wake up threads blocked on a waitqueue.
4026 * @q: the waitqueue
4027 * @mode: which threads
4028 * @nr_exclusive: how many wake-one or wake-many threads to wake up
4029 * @key: opaque value to be passed to wakeup targets
4030 *
4031 * The sync wakeup differs that the waker knows that it will schedule
4032 * away soon, so while the target thread will be woken up, it will not
4033 * be migrated to another CPU - ie. the two threads are 'synchronized'
4034 * with each other. This can prevent needless bouncing between CPUs.
4035 *
4036 * On UP it can prevent extra preemption.
4037 *
4038 * It may be assumed that this function implies a write memory barrier before
4039 * changing the task state if and only if any tasks are woken up.
4040 */
4041void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode,
4042            int nr_exclusive, void *key)
4043{
4044    unsigned long flags;
4045    int wake_flags = WF_SYNC;
4046
4047    if (unlikely(!q))
4048        return;
4049
4050    if (unlikely(!nr_exclusive))
4051        wake_flags = 0;
4052
4053    spin_lock_irqsave(&q->lock, flags);
4054    __wake_up_common(q, mode, nr_exclusive, wake_flags, key);
4055    spin_unlock_irqrestore(&q->lock, flags);
4056}
4057EXPORT_SYMBOL_GPL(__wake_up_sync_key);
4058
4059/*
4060 * __wake_up_sync - see __wake_up_sync_key()
4061 */
4062void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive)
4063{
4064    __wake_up_sync_key(q, mode, nr_exclusive, NULL);
4065}
4066EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */
4067
4068/**
4069 * complete: - signals a single thread waiting on this completion
4070 * @x: holds the state of this particular completion
4071 *
4072 * This will wake up a single thread waiting on this completion. Threads will be
4073 * awakened in the same order in which they were queued.
4074 *
4075 * See also complete_all(), wait_for_completion() and related routines.
4076 *
4077 * It may be assumed that this function implies a write memory barrier before
4078 * changing the task state if and only if any tasks are woken up.
4079 */
4080void complete(struct completion *x)
4081{
4082    unsigned long flags;
4083
4084    spin_lock_irqsave(&x->wait.lock, flags);
4085    x->done++;
4086    __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL);
4087    spin_unlock_irqrestore(&x->wait.lock, flags);
4088}
4089EXPORT_SYMBOL(complete);
4090
4091/**
4092 * complete_all: - signals all threads waiting on this completion
4093 * @x: holds the state of this particular completion
4094 *
4095 * This will wake up all threads waiting on this particular completion event.
4096 *
4097 * It may be assumed that this function implies a write memory barrier before
4098 * changing the task state if and only if any tasks are woken up.
4099 */
4100void complete_all(struct completion *x)
4101{
4102    unsigned long flags;
4103
4104    spin_lock_irqsave(&x->wait.lock, flags);
4105    x->done += UINT_MAX/2;
4106    __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL);
4107    spin_unlock_irqrestore(&x->wait.lock, flags);
4108}
4109EXPORT_SYMBOL(complete_all);
4110
4111static inline long __sched
4112do_wait_for_common(struct completion *x, long timeout, int state)
4113{
4114    if (!x->done) {
4115        DECLARE_WAITQUEUE(wait, current);
4116
4117        __add_wait_queue_tail_exclusive(&x->wait, &wait);
4118        do {
4119            if (signal_pending_state(state, current)) {
4120                timeout = -ERESTARTSYS;
4121                break;
4122            }
4123            __set_current_state(state);
4124            spin_unlock_irq(&x->wait.lock);
4125            timeout = schedule_timeout(timeout);
4126            spin_lock_irq(&x->wait.lock);
4127        } while (!x->done && timeout);
4128        __remove_wait_queue(&x->wait, &wait);
4129        if (!x->done)
4130            return timeout;
4131    }
4132    x->done--;
4133    return timeout ?: 1;
4134}
4135
4136static long __sched
4137wait_for_common(struct completion *x, long timeout, int state)
4138{
4139    might_sleep();
4140
4141    spin_lock_irq(&x->wait.lock);
4142    timeout = do_wait_for_common(x, timeout, state);
4143    spin_unlock_irq(&x->wait.lock);
4144    return timeout;
4145}
4146
4147/**
4148 * wait_for_completion: - waits for completion of a task
4149 * @x: holds the state of this particular completion
4150 *
4151 * This waits to be signaled for completion of a specific task. It is NOT
4152 * interruptible and there is no timeout.
4153 *
4154 * See also similar routines (i.e. wait_for_completion_timeout()) with timeout
4155 * and interrupt capability. Also see complete().
4156 */
4157void __sched wait_for_completion(struct completion *x)
4158{
4159    wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE);
4160}
4161EXPORT_SYMBOL(wait_for_completion);
4162
4163/**
4164 * wait_for_completion_timeout: - waits for completion of a task (w/timeout)
4165 * @x: holds the state of this particular completion
4166 * @timeout: timeout value in jiffies
4167 *
4168 * This waits for either a completion of a specific task to be signaled or for a
4169 * specified timeout to expire. The timeout is in jiffies. It is not
4170 * interruptible.
4171 */
4172unsigned long __sched
4173wait_for_completion_timeout(struct completion *x, unsigned long timeout)
4174{
4175    return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE);
4176}
4177EXPORT_SYMBOL(wait_for_completion_timeout);
4178
4179/**
4180 * wait_for_completion_interruptible: - waits for completion of a task (w/intr)
4181 * @x: holds the state of this particular completion
4182 *
4183 * This waits for completion of a specific task to be signaled. It is
4184 * interruptible.
4185 */
4186int __sched wait_for_completion_interruptible(struct completion *x)
4187{
4188    long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE);
4189    if (t == -ERESTARTSYS)
4190        return t;
4191    return 0;
4192}
4193EXPORT_SYMBOL(wait_for_completion_interruptible);
4194
4195/**
4196 * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr))
4197 * @x: holds the state of this particular completion
4198 * @timeout: timeout value in jiffies
4199 *
4200 * This waits for either a completion of a specific task to be signaled or for a
4201 * specified timeout to expire. It is interruptible. The timeout is in jiffies.
4202 */
4203unsigned long __sched
4204wait_for_completion_interruptible_timeout(struct completion *x,
4205                      unsigned long timeout)
4206{
4207    return wait_for_common(x, timeout, TASK_INTERRUPTIBLE);
4208}
4209EXPORT_SYMBOL(wait_for_completion_interruptible_timeout);
4210
4211/**
4212 * wait_for_completion_killable: - waits for completion of a task (killable)
4213 * @x: holds the state of this particular completion
4214 *
4215 * This waits to be signaled for completion of a specific task. It can be
4216 * interrupted by a kill signal.
4217 */
4218int __sched wait_for_completion_killable(struct completion *x)
4219{
4220    long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE);
4221    if (t == -ERESTARTSYS)
4222        return t;
4223    return 0;
4224}
4225EXPORT_SYMBOL(wait_for_completion_killable);
4226
4227/**
4228 * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable))
4229 * @x: holds the state of this particular completion
4230 * @timeout: timeout value in jiffies
4231 *
4232 * This waits for either a completion of a specific task to be
4233 * signaled or for a specified timeout to expire. It can be
4234 * interrupted by a kill signal. The timeout is in jiffies.
4235 */
4236unsigned long __sched
4237wait_for_completion_killable_timeout(struct completion *x,
4238                     unsigned long timeout)
4239{
4240    return wait_for_common(x, timeout, TASK_KILLABLE);
4241}
4242EXPORT_SYMBOL(wait_for_completion_killable_timeout);
4243
4244/**
4245 * try_wait_for_completion - try to decrement a completion without blocking
4246 * @x: completion structure
4247 *
4248 * Returns: 0 if a decrement cannot be done without blocking
4249 * 1 if a decrement succeeded.
4250 *
4251 * If a completion is being used as a counting completion,
4252 * attempt to decrement the counter without blocking. This
4253 * enables us to avoid waiting if the resource the completion
4254 * is protecting is not available.
4255 */
4256bool try_wait_for_completion(struct completion *x)
4257{
4258    unsigned long flags;
4259    int ret = 1;
4260
4261    spin_lock_irqsave(&x->wait.lock, flags);
4262    if (!x->done)
4263        ret = 0;
4264    else
4265        x->done--;
4266    spin_unlock_irqrestore(&x->wait.lock, flags);
4267    return ret;
4268}
4269EXPORT_SYMBOL(try_wait_for_completion);
4270
4271/**
4272 * completion_done - Test to see if a completion has any waiters
4273 * @x: completion structure
4274 *
4275 * Returns: 0 if there are waiters (wait_for_completion() in progress)
4276 * 1 if there are no waiters.
4277 *
4278 */
4279bool completion_done(struct completion *x)
4280{
4281    unsigned long flags;
4282    int ret = 1;
4283
4284    spin_lock_irqsave(&x->wait.lock, flags);
4285    if (!x->done)
4286        ret = 0;
4287    spin_unlock_irqrestore(&x->wait.lock, flags);
4288    return ret;
4289}
4290EXPORT_SYMBOL(completion_done);
4291
4292static long __sched
4293sleep_on_common(wait_queue_head_t *q, int state, long timeout)
4294{
4295    unsigned long flags;
4296    wait_queue_t wait;
4297
4298    init_waitqueue_entry(&wait, current);
4299
4300    __set_current_state(state);
4301
4302    spin_lock_irqsave(&q->lock, flags);
4303    __add_wait_queue(q, &wait);
4304    spin_unlock(&q->lock);
4305    timeout = schedule_timeout(timeout);
4306    spin_lock_irq(&q->lock);
4307    __remove_wait_queue(q, &wait);
4308    spin_unlock_irqrestore(&q->lock, flags);
4309
4310    return timeout;
4311}
4312
4313void __sched interruptible_sleep_on(wait_queue_head_t *q)
4314{
4315    sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
4316}
4317EXPORT_SYMBOL(interruptible_sleep_on);
4318
4319long __sched
4320interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout)
4321{
4322    return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout);
4323}
4324EXPORT_SYMBOL(interruptible_sleep_on_timeout);
4325
4326void __sched sleep_on(wait_queue_head_t *q)
4327{
4328    sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
4329}
4330EXPORT_SYMBOL(sleep_on);
4331
4332long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout)
4333{
4334    return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout);
4335}
4336EXPORT_SYMBOL(sleep_on_timeout);
4337
4338#ifdef CONFIG_RT_MUTEXES
4339
4340/*
4341 * rt_mutex_setprio - set the current priority of a task
4342 * @p: task
4343 * @prio: prio value (kernel-internal form)
4344 *
4345 * This function changes the 'effective' priority of a task. It does
4346 * not touch ->normal_prio like __setscheduler().
4347 *
4348 * Used by the rt_mutex code to implement priority inheritance logic.
4349 */
4350void rt_mutex_setprio(struct task_struct *p, int prio)
4351{
4352    unsigned long flags;
4353    int oldprio, on_rq, running;
4354    struct rq *rq;
4355    const struct sched_class *prev_class;
4356
4357    BUG_ON(prio < 0 || prio > MAX_PRIO);
4358
4359    rq = task_rq_lock(p, &flags);
4360
4361    oldprio = p->prio;
4362    prev_class = p->sched_class;
4363    on_rq = p->se.on_rq;
4364    running = task_current(rq, p);
4365    if (on_rq)
4366        dequeue_task(rq, p, 0);
4367    if (running)
4368        p->sched_class->put_prev_task(rq, p);
4369
4370    if (rt_prio(prio))
4371        p->sched_class = &rt_sched_class;
4372    else
4373        p->sched_class = &fair_sched_class;
4374
4375    p->prio = prio;
4376
4377    if (running)
4378        p->sched_class->set_curr_task(rq);
4379    if (on_rq) {
4380        enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0);
4381
4382        check_class_changed(rq, p, prev_class, oldprio, running);
4383    }
4384    task_rq_unlock(rq, &flags);
4385}
4386
4387#endif
4388
4389void set_user_nice(struct task_struct *p, long nice)
4390{
4391    int old_prio, delta, on_rq;
4392    unsigned long flags;
4393    struct rq *rq;
4394
4395    if (TASK_NICE(p) == nice || nice < -20 || nice > 19)
4396        return;
4397    /*
4398     * We have to be careful, if called from sys_setpriority(),
4399     * the task might be in the middle of scheduling on another CPU.
4400     */
4401    rq = task_rq_lock(p, &flags);
4402    /*
4403     * The RT priorities are set via sched_setscheduler(), but we still
4404     * allow the 'normal' nice value to be set - but as expected
4405     * it wont have any effect on scheduling until the task is
4406     * SCHED_FIFO/SCHED_RR:
4407     */
4408    if (task_has_rt_policy(p)) {
4409        p->static_prio = NICE_TO_PRIO(nice);
4410        goto out_unlock;
4411    }
4412    on_rq = p->se.on_rq;
4413    if (on_rq)
4414        dequeue_task(rq, p, 0);
4415
4416    p->static_prio = NICE_TO_PRIO(nice);
4417    set_load_weight(p);
4418    old_prio = p->prio;
4419    p->prio = effective_prio(p);
4420    delta = p->prio - old_prio;
4421
4422    if (on_rq) {
4423        enqueue_task(rq, p, 0);
4424        /*
4425         * If the task increased its priority or is running and
4426         * lowered its priority, then reschedule its CPU:
4427         */
4428        if (delta < 0 || (delta > 0 && task_running(rq, p)))
4429            resched_task(rq->curr);
4430    }
4431out_unlock:
4432    task_rq_unlock(rq, &flags);
4433}
4434EXPORT_SYMBOL(set_user_nice);
4435
4436/*
4437 * can_nice - check if a task can reduce its nice value
4438 * @p: task
4439 * @nice: nice value
4440 */
4441int can_nice(const struct task_struct *p, const int nice)
4442{
4443    /* convert nice value [19,-20] to rlimit style value [1,40] */
4444    int nice_rlim = 20 - nice;
4445
4446    return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) ||
4447        capable(CAP_SYS_NICE));
4448}
4449
4450#ifdef __ARCH_WANT_SYS_NICE
4451
4452/*
4453 * sys_nice - change the priority of the current process.
4454 * @increment: priority increment
4455 *
4456 * sys_setpriority is a more generic, but much slower function that
4457 * does similar things.
4458 */
4459SYSCALL_DEFINE1(nice, int, increment)
4460{
4461    long nice, retval;
4462
4463    /*
4464     * Setpriority might change our priority at the same moment.
4465     * We don't have to worry. Conceptually one call occurs first
4466     * and we have a single winner.
4467     */
4468    if (increment < -40)
4469        increment = -40;
4470    if (increment > 40)
4471        increment = 40;
4472
4473    nice = TASK_NICE(current) + increment;
4474    if (nice < -20)
4475        nice = -20;
4476    if (nice > 19)
4477        nice = 19;
4478
4479    if (increment < 0 && !can_nice(current, nice))
4480        return -EPERM;
4481
4482    retval = security_task_setnice(current, nice);
4483    if (retval)
4484        return retval;
4485
4486    set_user_nice(current, nice);
4487    return 0;
4488}
4489
4490#endif
4491
4492/**
4493 * task_prio - return the priority value of a given task.
4494 * @p: the task in question.
4495 *
4496 * This is the priority value as seen by users in /proc.
4497 * RT tasks are offset by -200. Normal tasks are centered
4498 * around 0, value goes from -16 to +15.
4499 */
4500int task_prio(const struct task_struct *p)
4501{
4502    return p->prio - MAX_RT_PRIO;
4503}
4504
4505/**
4506 * task_nice - return the nice value of a given task.
4507 * @p: the task in question.
4508 */
4509int task_nice(const struct task_struct *p)
4510{
4511    return TASK_NICE(p);
4512}
4513EXPORT_SYMBOL(task_nice);
4514
4515/**
4516 * idle_cpu - is a given cpu idle currently?
4517 * @cpu: the processor in question.
4518 */
4519int idle_cpu(int cpu)
4520{
4521    return cpu_curr(cpu) == cpu_rq(cpu)->idle;
4522}
4523
4524/**
4525 * idle_task - return the idle task for a given cpu.
4526 * @cpu: the processor in question.
4527 */
4528struct task_struct *idle_task(int cpu)
4529{
4530    return cpu_rq(cpu)->idle;
4531}
4532
4533/**
4534 * find_process_by_pid - find a process with a matching PID value.
4535 * @pid: the pid in question.
4536 */
4537static struct task_struct *find_process_by_pid(pid_t pid)
4538{
4539    return pid ? find_task_by_vpid(pid) : current;
4540}
4541
4542/* Actually do priority change: must hold rq lock. */
4543static void
4544__setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio)
4545{
4546    BUG_ON(p->se.on_rq);
4547
4548    p->policy = policy;
4549    p->rt_priority = prio;
4550    p->normal_prio = normal_prio(p);
4551    /* we are holding p->pi_lock already */
4552    p->prio = rt_mutex_getprio(p);
4553    if (rt_prio(p->prio))
4554        p->sched_class = &rt_sched_class;
4555    else
4556        p->sched_class = &fair_sched_class;
4557    set_load_weight(p);
4558}
4559
4560/*
4561 * check the target process has a UID that matches the current process's
4562 */
4563static bool check_same_owner(struct task_struct *p)
4564{
4565    const struct cred *cred = current_cred(), *pcred;
4566    bool match;
4567
4568    rcu_read_lock();
4569    pcred = __task_cred(p);
4570    match = (cred->euid == pcred->euid ||
4571         cred->euid == pcred->uid);
4572    rcu_read_unlock();
4573    return match;
4574}
4575
4576static int __sched_setscheduler(struct task_struct *p, int policy,
4577                struct sched_param *param, bool user)
4578{
4579    int retval, oldprio, oldpolicy = -1, on_rq, running;
4580    unsigned long flags;
4581    const struct sched_class *prev_class;
4582    struct rq *rq;
4583    int reset_on_fork;
4584
4585    /* may grab non-irq protected spin_locks */
4586    BUG_ON(in_interrupt());
4587recheck:
4588    /* double check policy once rq lock held */
4589    if (policy < 0) {
4590        reset_on_fork = p->sched_reset_on_fork;
4591        policy = oldpolicy = p->policy;
4592    } else {
4593        reset_on_fork = !!(policy & SCHED_RESET_ON_FORK);
4594        policy &= ~SCHED_RESET_ON_FORK;
4595
4596        if (policy != SCHED_FIFO && policy != SCHED_RR &&
4597                policy != SCHED_NORMAL && policy != SCHED_BATCH &&
4598                policy != SCHED_IDLE)
4599            return -EINVAL;
4600    }
4601
4602    /*
4603     * Valid priorities for SCHED_FIFO and SCHED_RR are
4604     * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL,
4605     * SCHED_BATCH and SCHED_IDLE is 0.
4606     */
4607    if (param->sched_priority < 0 ||
4608        (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) ||
4609        (!p->mm && param->sched_priority > MAX_RT_PRIO-1))
4610        return -EINVAL;
4611    if (rt_policy(policy) != (param->sched_priority != 0))
4612        return -EINVAL;
4613
4614    /*
4615     * Allow unprivileged RT tasks to decrease priority:
4616     */
4617    if (user && !capable(CAP_SYS_NICE)) {
4618        if (rt_policy(policy)) {
4619            unsigned long rlim_rtprio =
4620                    task_rlimit(p, RLIMIT_RTPRIO);
4621
4622            /* can't set/change the rt policy */
4623            if (policy != p->policy && !rlim_rtprio)
4624                return -EPERM;
4625
4626            /* can't increase priority */
4627            if (param->sched_priority > p->rt_priority &&
4628                param->sched_priority > rlim_rtprio)
4629                return -EPERM;
4630        }
4631        /*
4632         * Like positive nice levels, dont allow tasks to
4633         * move out of SCHED_IDLE either:
4634         */
4635        if (p->policy == SCHED_IDLE && policy != SCHED_IDLE)
4636            return -EPERM;
4637
4638        /* can't change other user's priorities */
4639        if (!check_same_owner(p))
4640            return -EPERM;
4641
4642        /* Normal users shall not reset the sched_reset_on_fork flag */
4643        if (p->sched_reset_on_fork && !reset_on_fork)
4644            return -EPERM;
4645    }
4646
4647    if (user) {
4648        retval = security_task_setscheduler(p, policy, param);
4649        if (retval)
4650            return retval;
4651    }
4652
4653    /*
4654     * make sure no PI-waiters arrive (or leave) while we are
4655     * changing the priority of the task:
4656     */
4657    raw_spin_lock_irqsave(&p->pi_lock, flags);
4658    /*
4659     * To be able to change p->policy safely, the apropriate
4660     * runqueue lock must be held.
4661     */
4662    rq = __task_rq_lock(p);
4663
4664#ifdef CONFIG_RT_GROUP_SCHED
4665    if (user) {
4666        /*
4667         * Do not allow realtime tasks into groups that have no runtime
4668         * assigned.
4669         */
4670        if (rt_bandwidth_enabled() && rt_policy(policy) &&
4671                task_group(p)->rt_bandwidth.rt_runtime == 0) {
4672            __task_rq_unlock(rq);
4673            raw_spin_unlock_irqrestore(&p->pi_lock, flags);
4674            return -EPERM;
4675        }
4676    }
4677#endif
4678
4679    /* recheck policy now with rq lock held */
4680    if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) {
4681        policy = oldpolicy = -1;
4682        __task_rq_unlock(rq);
4683        raw_spin_unlock_irqrestore(&p->pi_lock, flags);
4684        goto recheck;
4685    }
4686    on_rq = p->se.on_rq;
4687    running = task_current(rq, p);
4688    if (on_rq)
4689        deactivate_task(rq, p, 0);
4690    if (running)
4691        p->sched_class->put_prev_task(rq, p);
4692
4693    p->sched_reset_on_fork = reset_on_fork;
4694
4695    oldprio = p->prio;
4696    prev_class = p->sched_class;
4697    __setscheduler(rq, p, policy, param->sched_priority);
4698
4699    if (running)
4700        p->sched_class->set_curr_task(rq);
4701    if (on_rq) {
4702        activate_task(rq, p, 0);
4703
4704        check_class_changed(rq, p, prev_class, oldprio, running);
4705    }
4706    __task_rq_unlock(rq);
4707    raw_spin_unlock_irqrestore(&p->pi_lock, flags);
4708
4709    rt_mutex_adjust_pi(p);
4710
4711    return 0;
4712}
4713
4714/**
4715 * sched_setscheduler - change the scheduling policy and/or RT priority of a thread.
4716 * @p: the task in question.
4717 * @policy: new policy.
4718 * @param: structure containing the new RT priority.
4719 *
4720 * NOTE that the task may be already dead.
4721 */
4722int sched_setscheduler(struct task_struct *p, int policy,
4723               struct sched_param *param)
4724{
4725    return __sched_setscheduler(p, policy, param, true);
4726}
4727EXPORT_SYMBOL_GPL(sched_setscheduler);
4728
4729/**
4730 * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace.
4731 * @p: the task in question.
4732 * @policy: new policy.
4733 * @param: structure containing the new RT priority.
4734 *
4735 * Just like sched_setscheduler, only don't bother checking if the
4736 * current context has permission. For example, this is needed in
4737 * stop_machine(): we create temporary high priority worker threads,
4738 * but our caller might not have that capability.
4739 */
4740int sched_setscheduler_nocheck(struct task_struct *p, int policy,
4741                   struct sched_param *param)
4742{
4743    return __sched_setscheduler(p, policy, param, false);
4744}
4745
4746static int
4747do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param)
4748{
4749    struct sched_param lparam;
4750    struct task_struct *p;
4751    int retval;
4752
4753    if (!param || pid < 0)
4754        return -EINVAL;
4755    if (copy_from_user(&lparam, param, sizeof(struct sched_param)))
4756        return -EFAULT;
4757
4758    rcu_read_lock();
4759    retval = -ESRCH;
4760    p = find_process_by_pid(pid);
4761    if (p != NULL)
4762        retval = sched_setscheduler(p, policy, &lparam);
4763    rcu_read_unlock();
4764
4765    return retval;
4766}
4767
4768/**
4769 * sys_sched_setscheduler - set/change the scheduler policy and RT priority
4770 * @pid: the pid in question.
4771 * @policy: new policy.
4772 * @param: structure containing the new RT priority.
4773 */
4774SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy,
4775        struct sched_param __user *, param)
4776{
4777    /* negative values for policy are not valid */
4778    if (policy < 0)
4779        return -EINVAL;
4780
4781    return do_sched_setscheduler(pid, policy, param);
4782}
4783
4784/**
4785 * sys_sched_setparam - set/change the RT priority of a thread
4786 * @pid: the pid in question.
4787 * @param: structure containing the new RT priority.
4788 */
4789SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param)
4790{
4791    return do_sched_setscheduler(pid, -1, param);
4792}
4793
4794/**
4795 * sys_sched_getscheduler - get the policy (scheduling class) of a thread
4796 * @pid: the pid in question.
4797 */
4798SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid)
4799{
4800    struct task_struct *p;
4801    int retval;
4802
4803    if (pid < 0)
4804        return -EINVAL;
4805
4806    retval = -ESRCH;
4807    rcu_read_lock();
4808    p = find_process_by_pid(pid);
4809    if (p) {
4810        retval = security_task_getscheduler(p);
4811        if (!retval)
4812            retval = p->policy
4813                | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0);
4814    }
4815    rcu_read_unlock();
4816    return retval;
4817}
4818
4819/**
4820 * sys_sched_getparam - get the RT priority of a thread
4821 * @pid: the pid in question.
4822 * @param: structure containing the RT priority.
4823 */
4824SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param)
4825{
4826    struct sched_param lp;
4827    struct task_struct *p;
4828    int retval;
4829
4830    if (!param || pid < 0)
4831        return -EINVAL;
4832
4833    rcu_read_lock();
4834    p = find_process_by_pid(pid);
4835    retval = -ESRCH;
4836    if (!p)
4837        goto out_unlock;
4838
4839    retval = security_task_getscheduler(p);
4840    if (retval)
4841        goto out_unlock;
4842
4843    lp.sched_priority = p->rt_priority;
4844    rcu_read_unlock();
4845
4846    /*
4847     * This one might sleep, we cannot do it with a spinlock held ...
4848     */
4849    retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0;
4850
4851    return retval;
4852
4853out_unlock:
4854    rcu_read_unlock();
4855    return retval;
4856}
4857
4858long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)
4859{
4860    cpumask_var_t cpus_allowed, new_mask;
4861    struct task_struct *p;
4862    int retval;
4863
4864    get_online_cpus();
4865    rcu_read_lock();
4866
4867    p = find_process_by_pid(pid);
4868    if (!p) {
4869        rcu_read_unlock();
4870        put_online_cpus();
4871        return -ESRCH;
4872    }
4873
4874    /* Prevent p going away */
4875    get_task_struct(p);
4876    rcu_read_unlock();
4877
4878    if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) {
4879        retval = -ENOMEM;
4880        goto out_put_task;
4881    }
4882    if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) {
4883        retval = -ENOMEM;
4884        goto out_free_cpus_allowed;
4885    }
4886    retval = -EPERM;
4887    if (!check_same_owner(p) && !capable(CAP_SYS_NICE))
4888        goto out_unlock;
4889
4890    retval = security_task_setscheduler(p, 0, NULL);
4891    if (retval)
4892        goto out_unlock;
4893
4894    cpuset_cpus_allowed(p, cpus_allowed);
4895    cpumask_and(new_mask, in_mask, cpus_allowed);
4896 again:
4897    retval = set_cpus_allowed_ptr(p, new_mask);
4898
4899    if (!retval) {
4900        cpuset_cpus_allowed(p, cpus_allowed);
4901        if (!cpumask_subset(new_mask, cpus_allowed)) {
4902            /*
4903             * We must have raced with a concurrent cpuset
4904             * update. Just reset the cpus_allowed to the
4905             * cpuset's cpus_allowed
4906             */
4907            cpumask_copy(new_mask, cpus_allowed);
4908            goto again;
4909        }
4910    }
4911out_unlock:
4912    free_cpumask_var(new_mask);
4913out_free_cpus_allowed:
4914    free_cpumask_var(cpus_allowed);
4915out_put_task:
4916    put_task_struct(p);
4917    put_online_cpus();
4918    return retval;
4919}
4920
4921static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len,
4922                 struct cpumask *new_mask)
4923{
4924    if (len < cpumask_size())
4925        cpumask_clear(new_mask);
4926    else if (len > cpumask_size())
4927        len = cpumask_size();
4928
4929    return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0;
4930}
4931
4932/**
4933 * sys_sched_setaffinity - set the cpu affinity of a process
4934 * @pid: pid of the process
4935 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
4936 * @user_mask_ptr: user-space pointer to the new cpu mask
4937 */
4938SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len,
4939        unsigned long __user *, user_mask_ptr)
4940{
4941    cpumask_var_t new_mask;
4942    int retval;
4943
4944    if (!alloc_cpumask_var(&new_mask, GFP_KERNEL))
4945        return -ENOMEM;
4946
4947    retval = get_user_cpu_mask(user_mask_ptr, len, new_mask);
4948    if (retval == 0)
4949        retval = sched_setaffinity(pid, new_mask);
4950    free_cpumask_var(new_mask);
4951    return retval;
4952}
4953
4954long sched_getaffinity(pid_t pid, struct cpumask *mask)
4955{
4956    struct task_struct *p;
4957    unsigned long flags;
4958    struct rq *rq;
4959    int retval;
4960
4961    get_online_cpus();
4962    rcu_read_lock();
4963
4964    retval = -ESRCH;
4965    p = find_process_by_pid(pid);
4966    if (!p)
4967        goto out_unlock;
4968
4969    retval = security_task_getscheduler(p);
4970    if (retval)
4971        goto out_unlock;
4972
4973    rq = task_rq_lock(p, &flags);
4974    cpumask_and(mask, &p->cpus_allowed, cpu_online_mask);
4975    task_rq_unlock(rq, &flags);
4976
4977out_unlock:
4978    rcu_read_unlock();
4979    put_online_cpus();
4980
4981    return retval;
4982}
4983
4984/**
4985 * sys_sched_getaffinity - get the cpu affinity of a process
4986 * @pid: pid of the process
4987 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
4988 * @user_mask_ptr: user-space pointer to hold the current cpu mask
4989 */
4990SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len,
4991        unsigned long __user *, user_mask_ptr)
4992{
4993    int ret;
4994    cpumask_var_t mask;
4995
4996    if ((len * BITS_PER_BYTE) < nr_cpu_ids)
4997        return -EINVAL;
4998    if (len & (sizeof(unsigned long)-1))
4999        return -EINVAL;
5000
5001    if (!alloc_cpumask_var(&mask, GFP_KERNEL))
5002        return -ENOMEM;
5003
5004    ret = sched_getaffinity(pid, mask);
5005    if (ret == 0) {
5006        size_t retlen = min_t(size_t, len, cpumask_size());
5007
5008        if (copy_to_user(user_mask_ptr, mask, retlen))
5009            ret = -EFAULT;
5010        else
5011            ret = retlen;
5012    }
5013    free_cpumask_var(mask);
5014
5015    return ret;
5016}
5017
5018/**
5019 * sys_sched_yield - yield the current processor to other threads.
5020 *
5021 * This function yields the current CPU to other tasks. If there are no
5022 * other threads running on this CPU then this function will return.
5023 */
5024SYSCALL_DEFINE0(sched_yield)
5025{
5026    struct rq *rq = this_rq_lock();
5027
5028    schedstat_inc(rq, yld_count);
5029    current->sched_class->yield_task(rq);
5030
5031    /*
5032     * Since we are going to call schedule() anyway, there's
5033     * no need to preempt or enable interrupts:
5034     */
5035    __release(rq->lock);
5036    spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
5037    do_raw_spin_unlock(&rq->lock);
5038    preempt_enable_no_resched();
5039
5040    schedule();
5041
5042    return 0;
5043}
5044
5045static inline int should_resched(void)
5046{
5047    return need_resched() && !(preempt_count() & PREEMPT_ACTIVE);
5048}
5049
5050static void __cond_resched(void)
5051{
5052    add_preempt_count(PREEMPT_ACTIVE);
5053    schedule();
5054    sub_preempt_count(PREEMPT_ACTIVE);
5055}
5056
5057int __sched _cond_resched(void)
5058{
5059    if (should_resched()) {
5060        __cond_resched();
5061        return 1;
5062    }
5063    return 0;
5064}
5065EXPORT_SYMBOL(_cond_resched);
5066
5067/*
5068 * __cond_resched_lock() - if a reschedule is pending, drop the given lock,
5069 * call schedule, and on return reacquire the lock.
5070 *
5071 * This works OK both with and without CONFIG_PREEMPT. We do strange low-level
5072 * operations here to prevent schedule() from being called twice (once via
5073 * spin_unlock(), once by hand).
5074 */
5075int __cond_resched_lock(spinlock_t *lock)
5076{
5077    int resched = should_resched();
5078    int ret = 0;
5079
5080    lockdep_assert_held(lock);
5081
5082    if (spin_needbreak(lock) || resched) {
5083        spin_unlock(lock);
5084        if (resched)
5085            __cond_resched();
5086        else
5087            cpu_relax();
5088        ret = 1;
5089        spin_lock(lock);
5090    }
5091    return ret;
5092}
5093EXPORT_SYMBOL(__cond_resched_lock);
5094
5095int __sched __cond_resched_softirq(void)
5096{
5097    BUG_ON(!in_softirq());
5098
5099    if (should_resched()) {
5100        local_bh_enable();
5101        __cond_resched();
5102        local_bh_disable();
5103        return 1;
5104    }
5105    return 0;
5106}
5107EXPORT_SYMBOL(__cond_resched_softirq);
5108
5109/**
5110 * yield - yield the current processor to other threads.
5111 *
5112 * This is a shortcut for kernel-space yielding - it marks the
5113 * thread runnable and calls sys_sched_yield().
5114 */
5115void __sched yield(void)
5116{
5117    set_current_state(TASK_RUNNING);
5118    sys_sched_yield();
5119}
5120EXPORT_SYMBOL(yield);
5121
5122/*
5123 * This task is about to go to sleep on IO. Increment rq->nr_iowait so
5124 * that process accounting knows that this is a task in IO wait state.
5125 */
5126void __sched io_schedule(void)
5127{
5128    struct rq *rq = raw_rq();
5129
5130    delayacct_blkio_start();
5131    atomic_inc(&rq->nr_iowait);
5132    current->in_iowait = 1;
5133    schedule();
5134    current->in_iowait = 0;
5135    atomic_dec(&rq->nr_iowait);
5136    delayacct_blkio_end();
5137}
5138EXPORT_SYMBOL(io_schedule);
5139
5140long __sched io_schedule_timeout(long timeout)
5141{
5142    struct rq *rq = raw_rq();
5143    long ret;
5144
5145    delayacct_blkio_start();
5146    atomic_inc(&rq->nr_iowait);
5147    current->in_iowait = 1;
5148    ret = schedule_timeout(timeout);
5149    current->in_iowait = 0;
5150    atomic_dec(&rq->nr_iowait);
5151    delayacct_blkio_end();
5152    return ret;
5153}
5154
5155/**
5156 * sys_sched_get_priority_max - return maximum RT priority.
5157 * @policy: scheduling class.
5158 *
5159 * this syscall returns the maximum rt_priority that can be used
5160 * by a given scheduling class.
5161 */
5162SYSCALL_DEFINE1(sched_get_priority_max, int, policy)
5163{
5164    int ret = -EINVAL;
5165
5166    switch (policy) {
5167    case SCHED_FIFO:
5168    case SCHED_RR:
5169        ret = MAX_USER_RT_PRIO-1;
5170        break;
5171    case SCHED_NORMAL:
5172    case SCHED_BATCH:
5173    case SCHED_IDLE:
5174        ret = 0;
5175        break;
5176    }
5177    return ret;
5178}
5179
5180/**
5181 * sys_sched_get_priority_min - return minimum RT priority.
5182 * @policy: scheduling class.
5183 *
5184 * this syscall returns the minimum rt_priority that can be used
5185 * by a given scheduling class.
5186 */
5187SYSCALL_DEFINE1(sched_get_priority_min, int, policy)
5188{
5189    int ret = -EINVAL;
5190
5191    switch (policy) {
5192    case SCHED_FIFO:
5193    case SCHED_RR:
5194        ret = 1;
5195        break;
5196    case SCHED_NORMAL:
5197    case SCHED_BATCH:
5198    case SCHED_IDLE:
5199        ret = 0;
5200    }
5201    return ret;
5202}
5203
5204/**
5205 * sys_sched_rr_get_interval - return the default timeslice of a process.
5206 * @pid: pid of the process.
5207 * @interval: userspace pointer to the timeslice value.
5208 *
5209 * this syscall writes the default timeslice value of a given process
5210 * into the user-space timespec buffer. A value of '0' means infinity.
5211 */
5212SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid,
5213        struct timespec __user *, interval)
5214{
5215    struct task_struct *p;
5216    unsigned int time_slice;
5217    unsigned long flags;
5218    struct rq *rq;
5219    int retval;
5220    struct timespec t;
5221
5222    if (pid < 0)
5223        return -EINVAL;
5224
5225    retval = -ESRCH;
5226    rcu_read_lock();
5227    p = find_process_by_pid(pid);
5228    if (!p)
5229        goto out_unlock;
5230
5231    retval = security_task_getscheduler(p);
5232    if (retval)
5233        goto out_unlock;
5234
5235    rq = task_rq_lock(p, &flags);
5236    time_slice = p->sched_class->get_rr_interval(rq, p);
5237    task_rq_unlock(rq, &flags);
5238
5239    rcu_read_unlock();
5240    jiffies_to_timespec(time_slice, &t);
5241    retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0;
5242    return retval;
5243
5244out_unlock:
5245    rcu_read_unlock();
5246    return retval;
5247}
5248
5249static const char stat_nam[] = TASK_STATE_TO_CHAR_STR;
5250
5251void sched_show_task(struct task_struct *p)
5252{
5253    unsigned long free = 0;
5254    unsigned state;
5255
5256    state = p->state ? __ffs(p->state) + 1 : 0;
5257    printk(KERN_INFO "%-13.13s %c", p->comm,
5258        state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?');
5259#if BITS_PER_LONG == 32
5260    if (state == TASK_RUNNING)
5261        printk(KERN_CONT " running ");
5262    else
5263        printk(KERN_CONT " %08lx ", thread_saved_pc(p));
5264#else
5265    if (state == TASK_RUNNING)
5266        printk(KERN_CONT " running task ");
5267    else
5268        printk(KERN_CONT " %016lx ", thread_saved_pc(p));
5269#endif
5270#ifdef CONFIG_DEBUG_STACK_USAGE
5271    free = stack_not_used(p);
5272#endif
5273    printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free,
5274        task_pid_nr(p), task_pid_nr(p->real_parent),
5275        (unsigned long)task_thread_info(p)->flags);
5276
5277    show_stack(p, NULL);
5278}
5279
5280void show_state_filter(unsigned long state_filter)
5281{
5282    struct task_struct *g, *p;
5283
5284#if BITS_PER_LONG == 32
5285    printk(KERN_INFO
5286        " task PC stack pid father\n");
5287#else
5288    printk(KERN_INFO
5289        " task PC stack pid father\n");
5290#endif
5291    read_lock(&tasklist_lock);
5292    do_each_thread(g, p) {
5293        /*
5294         * reset the NMI-timeout, listing all files on a slow
5295         * console might take alot of time:
5296         */
5297        touch_nmi_watchdog();
5298        if (!state_filter || (p->state & state_filter))
5299            sched_show_task(p);
5300    } while_each_thread(g, p);
5301
5302    touch_all_softlockup_watchdogs();
5303
5304#ifdef CONFIG_SCHED_DEBUG
5305    sysrq_sched_debug_show();
5306#endif
5307    read_unlock(&tasklist_lock);
5308    /*
5309     * Only show locks if all tasks are dumped:
5310     */
5311    if (!state_filter)
5312        debug_show_all_locks();
5313}
5314
5315void __cpuinit init_idle_bootup_task(struct task_struct *idle)
5316{
5317    idle->sched_class = &idle_sched_class;
5318}
5319
5320/**
5321 * init_idle - set up an idle thread for a given CPU
5322 * @idle: task in question
5323 * @cpu: cpu the idle task belongs to
5324 *
5325 * NOTE: this function does not set the idle thread's NEED_RESCHED
5326 * flag, to make booting more robust.
5327 */
5328void __cpuinit init_idle(struct task_struct *idle, int cpu)
5329{
5330    struct rq *rq = cpu_rq(cpu);
5331    unsigned long flags;
5332
5333    raw_spin_lock_irqsave(&rq->lock, flags);
5334
5335    __sched_fork(idle);
5336    idle->state = TASK_RUNNING;
5337    idle->se.exec_start = sched_clock();
5338
5339    cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu));
5340    __set_task_cpu(idle, cpu);
5341
5342    rq->curr = rq->idle = idle;
5343#if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW)
5344    idle->oncpu = 1;
5345#endif
5346    raw_spin_unlock_irqrestore(&rq->lock, flags);
5347
5348    /* Set the preempt count _outside_ the spinlocks! */
5349#if defined(CONFIG_PREEMPT)
5350    task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0);
5351#else
5352    task_thread_info(idle)->preempt_count = 0;
5353#endif
5354    /*
5355     * The idle tasks have their own, simple scheduling class:
5356     */
5357    idle->sched_class = &idle_sched_class;
5358    ftrace_graph_init_task(idle);
5359}
5360
5361/*
5362 * In a system that switches off the HZ timer nohz_cpu_mask
5363 * indicates which cpus entered this state. This is used
5364 * in the rcu update to wait only for active cpus. For system
5365 * which do not switch off the HZ timer nohz_cpu_mask should
5366 * always be CPU_BITS_NONE.
5367 */
5368cpumask_var_t nohz_cpu_mask;
5369
5370/*
5371 * Increase the granularity value when there are more CPUs,
5372 * because with more CPUs the 'effective latency' as visible
5373 * to users decreases. But the relationship is not linear,
5374 * so pick a second-best guess by going with the log2 of the
5375 * number of CPUs.
5376 *
5377 * This idea comes from the SD scheduler of Con Kolivas:
5378 */
5379static int get_update_sysctl_factor(void)
5380{
5381    unsigned int cpus = min_t(int, num_online_cpus(), 8);
5382    unsigned int factor;
5383
5384    switch (sysctl_sched_tunable_scaling) {
5385    case SCHED_TUNABLESCALING_NONE:
5386        factor = 1;
5387        break;
5388    case SCHED_TUNABLESCALING_LINEAR:
5389        factor = cpus;
5390        break;
5391    case SCHED_TUNABLESCALING_LOG:
5392    default:
5393        factor = 1 + ilog2(cpus);
5394        break;
5395    }
5396
5397    return factor;
5398}
5399
5400static void update_sysctl(void)
5401{
5402    unsigned int factor = get_update_sysctl_factor();
5403
5404#define SET_SYSCTL(name) \
5405    (sysctl_##name = (factor) * normalized_sysctl_##name)
5406    SET_SYSCTL(sched_min_granularity);
5407    SET_SYSCTL(sched_latency);
5408    SET_SYSCTL(sched_wakeup_granularity);
5409    SET_SYSCTL(sched_shares_ratelimit);
5410#undef SET_SYSCTL
5411}
5412
5413static inline void sched_init_granularity(void)
5414{
5415    update_sysctl();
5416}
5417
5418#ifdef CONFIG_SMP
5419/*
5420 * This is how migration works:
5421 *
5422 * 1) we invoke migration_cpu_stop() on the target CPU using
5423 * stop_one_cpu().
5424 * 2) stopper starts to run (implicitly forcing the migrated thread
5425 * off the CPU)
5426 * 3) it checks whether the migrated task is still in the wrong runqueue.
5427 * 4) if it's in the wrong runqueue then the migration thread removes
5428 * it and puts it into the right queue.
5429 * 5) stopper completes and stop_one_cpu() returns and the migration
5430 * is done.
5431 */
5432
5433/*
5434 * Change a given task's CPU affinity. Migrate the thread to a
5435 * proper CPU and schedule it away if the CPU it's executing on
5436 * is removed from the allowed bitmask.
5437 *
5438 * NOTE: the caller must have a valid reference to the task, the
5439 * task must not exit() & deallocate itself prematurely. The
5440 * call is not atomic; no spinlocks may be held.
5441 */
5442int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask)
5443{
5444    unsigned long flags;
5445    struct rq *rq;
5446    unsigned int dest_cpu;
5447    int ret = 0;
5448
5449    /*
5450     * Serialize against TASK_WAKING so that ttwu() and wunt() can
5451     * drop the rq->lock and still rely on ->cpus_allowed.
5452     */
5453again:
5454    while (task_is_waking(p))
5455        cpu_relax();
5456    rq = task_rq_lock(p, &flags);
5457    if (task_is_waking(p)) {
5458        task_rq_unlock(rq, &flags);
5459        goto again;
5460    }
5461
5462    if (!cpumask_intersects(new_mask, cpu_active_mask)) {
5463        ret = -EINVAL;
5464        goto out;
5465    }
5466
5467    if (unlikely((p->flags & PF_THREAD_BOUND) && p != current &&
5468             !cpumask_equal(&p->cpus_allowed, new_mask))) {
5469        ret = -EINVAL;
5470        goto out;
5471    }
5472
5473    if (p->sched_class->set_cpus_allowed)
5474        p->sched_class->set_cpus_allowed(p, new_mask);
5475    else {
5476        cpumask_copy(&p->cpus_allowed, new_mask);
5477        p->rt.nr_cpus_allowed = cpumask_weight(new_mask);
5478    }
5479
5480    /* Can the task run on the task's current CPU? If so, we're done */
5481    if (cpumask_test_cpu(task_cpu(p), new_mask))
5482        goto out;
5483
5484    dest_cpu = cpumask_any_and(cpu_active_mask, new_mask);
5485    if (migrate_task(p, dest_cpu)) {
5486        struct migration_arg arg = { p, dest_cpu };
5487        /* Need help from migration thread: drop lock and wait. */
5488        task_rq_unlock(rq, &flags);
5489        stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg);
5490        tlb_migrate_finish(p->mm);
5491        return 0;
5492    }
5493out:
5494    task_rq_unlock(rq, &flags);
5495
5496    return ret;
5497}
5498EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr);
5499
5500/*
5501 * Move (not current) task off this cpu, onto dest cpu. We're doing
5502 * this because either it can't run here any more (set_cpus_allowed()
5503 * away from this CPU, or CPU going down), or because we're
5504 * attempting to rebalance this task on exec (sched_exec).
5505 *
5506 * So we race with normal scheduler movements, but that's OK, as long
5507 * as the task is no longer on this CPU.
5508 *
5509 * Returns non-zero if task was successfully migrated.
5510 */
5511static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu)
5512{
5513    struct rq *rq_dest, *rq_src;
5514    int ret = 0;
5515
5516    if (unlikely(!cpu_active(dest_cpu)))
5517        return ret;
5518
5519    rq_src = cpu_rq(src_cpu);
5520    rq_dest = cpu_rq(dest_cpu);
5521
5522    double_rq_lock(rq_src, rq_dest);
5523    /* Already moved. */
5524    if (task_cpu(p) != src_cpu)
5525        goto done;
5526    /* Affinity changed (again). */
5527    if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed))
5528        goto fail;
5529
5530    /*
5531     * If we're not on a rq, the next wake-up will ensure we're
5532     * placed properly.
5533     */
5534    if (p->se.on_rq) {
5535        deactivate_task(rq_src, p, 0);
5536        set_task_cpu(p, dest_cpu);
5537        activate_task(rq_dest, p, 0);
5538        check_preempt_curr(rq_dest, p, 0);
5539    }
5540done:
5541    ret = 1;
5542fail:
5543    double_rq_unlock(rq_src, rq_dest);
5544    return ret;
5545}
5546
5547/*
5548 * migration_cpu_stop - this will be executed by a highprio stopper thread
5549 * and performs thread migration by bumping thread off CPU then
5550 * 'pushing' onto another runqueue.
5551 */
5552static int migration_cpu_stop(void *data)
5553{
5554    struct migration_arg *arg = data;
5555
5556    /*
5557     * The original target cpu might have gone down and we might
5558     * be on another cpu but it doesn't matter.
5559     */
5560    local_irq_disable();
5561    __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu);
5562    local_irq_enable();
5563    return 0;
5564}
5565
5566#ifdef CONFIG_HOTPLUG_CPU
5567/*
5568 * Figure out where task on dead CPU should go, use force if necessary.
5569 */
5570void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p)
5571{
5572    struct rq *rq = cpu_rq(dead_cpu);
5573    int needs_cpu, uninitialized_var(dest_cpu);
5574    unsigned long flags;
5575
5576    local_irq_save(flags);
5577
5578    raw_spin_lock(&rq->lock);
5579    needs_cpu = (task_cpu(p) == dead_cpu) && (p->state != TASK_WAKING);
5580    if (needs_cpu)
5581        dest_cpu = select_fallback_rq(dead_cpu, p);
5582    raw_spin_unlock(&rq->lock);
5583    /*
5584     * It can only fail if we race with set_cpus_allowed(),
5585     * in the racer should migrate the task anyway.
5586     */
5587    if (needs_cpu)
5588        __migrate_task(p, dead_cpu, dest_cpu);
5589    local_irq_restore(flags);
5590}
5591
5592/*
5593 * While a dead CPU has no uninterruptible tasks queued at this point,
5594 * it might still have a nonzero ->nr_uninterruptible counter, because
5595 * for performance reasons the counter is not stricly tracking tasks to
5596 * their home CPUs. So we just add the counter to another CPU's counter,
5597 * to keep the global sum constant after CPU-down:
5598 */
5599static void migrate_nr_uninterruptible(struct rq *rq_src)
5600{
5601    struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask));
5602    unsigned long flags;
5603
5604    local_irq_save(flags);
5605    double_rq_lock(rq_src, rq_dest);
5606    rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible;
5607    rq_src->nr_uninterruptible = 0;
5608    double_rq_unlock(rq_src, rq_dest);
5609    local_irq_restore(flags);
5610}
5611
5612/* Run through task list and migrate tasks from the dead cpu. */
5613static void migrate_live_tasks(int src_cpu)
5614{
5615    struct task_struct *p, *t;
5616
5617    read_lock(&tasklist_lock);
5618
5619    do_each_thread(t, p) {
5620        if (p == current)
5621            continue;
5622
5623        if (task_cpu(p) == src_cpu)
5624            move_task_off_dead_cpu(src_cpu, p);
5625    } while_each_thread(t, p);
5626
5627    read_unlock(&tasklist_lock);
5628}
5629
5630/*
5631 * Schedules idle task to be the next runnable task on current CPU.
5632 * It does so by boosting its priority to highest possible.
5633 * Used by CPU offline code.
5634 */
5635void sched_idle_next(void)
5636{
5637    int this_cpu = smp_processor_id();
5638    struct rq *rq = cpu_rq(this_cpu);
5639    struct task_struct *p = rq->idle;
5640    unsigned long flags;
5641
5642    /* cpu has to be offline */
5643    BUG_ON(cpu_online(this_cpu));
5644
5645    /*
5646     * Strictly not necessary since rest of the CPUs are stopped by now
5647     * and interrupts disabled on the current cpu.
5648     */
5649    raw_spin_lock_irqsave(&rq->lock, flags);
5650
5651    __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1);
5652
5653    activate_task(rq, p, 0);
5654
5655    raw_spin_unlock_irqrestore(&rq->lock, flags);
5656}
5657
5658/*
5659 * Ensures that the idle task is using init_mm right before its cpu goes
5660 * offline.
5661 */
5662void idle_task_exit(void)
5663{
5664    struct mm_struct *mm = current->active_mm;
5665
5666    BUG_ON(cpu_online(smp_processor_id()));
5667
5668    if (mm != &init_mm)
5669        switch_mm(mm, &init_mm, current);
5670    mmdrop(mm);
5671}
5672
5673/* called under rq->lock with disabled interrupts */
5674static void migrate_dead(unsigned int dead_cpu, struct task_struct *p)
5675{
5676    struct rq *rq = cpu_rq(dead_cpu);
5677
5678    /* Must be exiting, otherwise would be on tasklist. */
5679    BUG_ON(!p->exit_state);
5680
5681    /* Cannot have done final schedule yet: would have vanished. */
5682    BUG_ON(p->state == TASK_DEAD);
5683
5684    get_task_struct(p);
5685
5686    /*
5687     * Drop lock around migration; if someone else moves it,
5688     * that's OK. No task can be added to this CPU, so iteration is
5689     * fine.
5690     */
5691    raw_spin_unlock_irq(&rq->lock);
5692    move_task_off_dead_cpu(dead_cpu, p);
5693    raw_spin_lock_irq(&rq->lock);
5694
5695    put_task_struct(p);
5696}
5697
5698/* release_task() removes task from tasklist, so we won't find dead tasks. */
5699static void migrate_dead_tasks(unsigned int dead_cpu)
5700{
5701    struct rq *rq = cpu_rq(dead_cpu);
5702    struct task_struct *next;
5703
5704    for ( ; ; ) {
5705        if (!rq->nr_running)
5706            break;
5707        next = pick_next_task(rq);
5708        if (!next)
5709            break;
5710        next->sched_class->put_prev_task(rq, next);
5711        migrate_dead(dead_cpu, next);
5712
5713    }
5714}
5715
5716/*
5717 * remove the tasks which were accounted by rq from calc_load_tasks.
5718 */
5719static void calc_global_load_remove(struct rq *rq)
5720{
5721    atomic_long_sub(rq->calc_load_active, &calc_load_tasks);
5722    rq->calc_load_active = 0;
5723}
5724#endif /* CONFIG_HOTPLUG_CPU */
5725
5726#if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
5727
5728static struct ctl_table sd_ctl_dir[] = {
5729    {
5730        .procname = "sched_domain",
5731        .mode = 0555,
5732    },
5733    {}
5734};
5735
5736static struct ctl_table sd_ctl_root[] = {
5737    {
5738        .procname = "kernel",
5739        .mode = 0555,
5740        .child = sd_ctl_dir,
5741    },
5742    {}
5743};
5744
5745static struct ctl_table *sd_alloc_ctl_entry(int n)
5746{
5747    struct ctl_table *entry =
5748        kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL);
5749
5750    return entry;
5751}
5752
5753static void sd_free_ctl_entry(struct ctl_table **tablep)
5754{
5755    struct ctl_table *entry;
5756
5757    /*
5758     * In the intermediate directories, both the child directory and
5759     * procname are dynamically allocated and could fail but the mode
5760     * will always be set. In the lowest directory the names are
5761     * static strings and all have proc handlers.
5762     */
5763    for (entry = *tablep; entry->mode; entry++) {
5764        if (entry->child)
5765            sd_free_ctl_entry(&entry->child);
5766        if (entry->proc_handler == NULL)
5767            kfree(entry->procname);
5768    }
5769
5770    kfree(*tablep);
5771    *tablep = NULL;
5772}
5773
5774static void
5775set_table_entry(struct ctl_table *entry,
5776        const char *procname, void *data, int maxlen,
5777        mode_t mode, proc_handler *proc_handler)
5778{
5779    entry->procname = procname;
5780    entry->data = data;
5781    entry->maxlen = maxlen;
5782    entry->mode = mode;
5783    entry->proc_handler = proc_handler;
5784}
5785
5786static struct ctl_table *
5787sd_alloc_ctl_domain_table(struct sched_domain *sd)
5788{
5789    struct ctl_table *table = sd_alloc_ctl_entry(13);
5790
5791    if (table == NULL)
5792        return NULL;
5793
5794    set_table_entry(&table[0], "min_interval", &sd->min_interval,
5795        sizeof(long), 0644, proc_doulongvec_minmax);
5796    set_table_entry(&table[1], "max_interval", &sd->max_interval,
5797        sizeof(long), 0644, proc_doulongvec_minmax);
5798    set_table_entry(&table[2], "busy_idx", &sd->busy_idx,
5799        sizeof(int), 0644, proc_dointvec_minmax);
5800    set_table_entry(&table[3], "idle_idx", &sd->idle_idx,
5801        sizeof(int), 0644, proc_dointvec_minmax);
5802    set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx,
5803        sizeof(int), 0644, proc_dointvec_minmax);
5804    set_table_entry(&table[5], "wake_idx", &sd->wake_idx,
5805        sizeof(int), 0644, proc_dointvec_minmax);
5806    set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx,
5807        sizeof(int), 0644, proc_dointvec_minmax);
5808    set_table_entry(&table[7], "busy_factor", &sd->busy_factor,
5809        sizeof(int), 0644, proc_dointvec_minmax);
5810    set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct,
5811        sizeof(int), 0644, proc_dointvec_minmax);
5812    set_table_entry(&table[9], "cache_nice_tries",
5813        &sd->cache_nice_tries,
5814        sizeof(int), 0644, proc_dointvec_minmax);
5815    set_table_entry(&table[10], "flags", &sd->flags,
5816        sizeof(int), 0644, proc_dointvec_minmax);
5817    set_table_entry(&table[11], "name", sd->name,
5818        CORENAME_MAX_SIZE, 0444, proc_dostring);
5819    /* &table[12] is terminator */
5820
5821    return table;
5822}
5823
5824static ctl_table *sd_alloc_ctl_cpu_table(int cpu)
5825{
5826    struct ctl_table *entry, *table;
5827    struct sched_domain *sd;
5828    int domain_num = 0, i;
5829    char buf[32];
5830
5831    for_each_domain(cpu, sd)
5832        domain_num++;
5833    entry = table = sd_alloc_ctl_entry(domain_num + 1);
5834    if (table == NULL)
5835        return NULL;
5836
5837    i = 0;
5838    for_each_domain(cpu, sd) {
5839        snprintf(buf, 32, "domain%d", i);
5840        entry->procname = kstrdup(buf, GFP_KERNEL);
5841        entry->mode = 0555;
5842        entry->child = sd_alloc_ctl_domain_table(sd);
5843        entry++;
5844        i++;
5845    }
5846    return table;
5847}
5848
5849static struct ctl_table_header *sd_sysctl_header;
5850static void register_sched_domain_sysctl(void)
5851{
5852    int i, cpu_num = num_possible_cpus();
5853    struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1);
5854    char buf[32];
5855
5856    WARN_ON(sd_ctl_dir[0].child);
5857    sd_ctl_dir[0].child = entry;
5858
5859    if (entry == NULL)
5860        return;
5861
5862    for_each_possible_cpu(i) {
5863        snprintf(buf, 32, "cpu%d", i);
5864        entry->procname = kstrdup(buf, GFP_KERNEL);
5865        entry->mode = 0555;
5866        entry->child = sd_alloc_ctl_cpu_table(i);
5867        entry++;
5868    }
5869
5870    WARN_ON(sd_sysctl_header);
5871    sd_sysctl_header = register_sysctl_table(sd_ctl_root);
5872}
5873
5874/* may be called multiple times per register */
5875static void unregister_sched_domain_sysctl(void)
5876{
5877    if (sd_sysctl_header)
5878        unregister_sysctl_table(sd_sysctl_header);
5879    sd_sysctl_header = NULL;
5880    if (sd_ctl_dir[0].child)
5881        sd_free_ctl_entry(&sd_ctl_dir[0].child);
5882}
5883#else
5884static void register_sched_domain_sysctl(void)
5885{
5886}
5887static void unregister_sched_domain_sysctl(void)
5888{
5889}
5890#endif
5891
5892static void set_rq_online(struct rq *rq)
5893{
5894    if (!rq->online) {
5895        const struct sched_class *class;
5896
5897        cpumask_set_cpu(rq->cpu, rq->rd->online);
5898        rq->online = 1;
5899
5900        for_each_class(class) {
5901            if (class->rq_online)
5902                class->rq_online(rq);
5903        }
5904    }
5905}
5906
5907static void set_rq_offline(struct rq *rq)
5908{
5909    if (rq->online) {
5910        const struct sched_class *class;
5911
5912        for_each_class(class) {
5913            if (class->rq_offline)
5914                class->rq_offline(rq);
5915        }
5916
5917        cpumask_clear_cpu(rq->cpu, rq->rd->online);
5918        rq->online = 0;
5919    }
5920}
5921
5922/*
5923 * migration_call - callback that gets triggered when a CPU is added.
5924 * Here we can start up the necessary migration thread for the new CPU.
5925 */
5926static int __cpuinit
5927migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu)
5928{
5929    int cpu = (long)hcpu;
5930    unsigned long flags;
5931    struct rq *rq = cpu_rq(cpu);
5932
5933    switch (action) {
5934
5935    case CPU_UP_PREPARE:
5936    case CPU_UP_PREPARE_FROZEN:
5937        rq->calc_load_update = calc_load_update;
5938        break;
5939
5940    case CPU_ONLINE:
5941    case CPU_ONLINE_FROZEN:
5942        /* Update our root-domain */
5943        raw_spin_lock_irqsave(&rq->lock, flags);
5944        if (rq->rd) {
5945            BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
5946
5947            set_rq_online(rq);
5948        }
5949        raw_spin_unlock_irqrestore(&rq->lock, flags);
5950        break;
5951
5952#ifdef CONFIG_HOTPLUG_CPU
5953    case CPU_DEAD:
5954    case CPU_DEAD_FROZEN:
5955        migrate_live_tasks(cpu);
5956        /* Idle task back to normal (off runqueue, low prio) */
5957        raw_spin_lock_irq(&rq->lock);
5958        deactivate_task(rq, rq->idle, 0);
5959        __setscheduler(rq, rq->idle, SCHED_NORMAL, 0);
5960        rq->idle->sched_class = &idle_sched_class;
5961        migrate_dead_tasks(cpu);
5962        raw_spin_unlock_irq(&rq->lock);
5963        migrate_nr_uninterruptible(rq);
5964        BUG_ON(rq->nr_running != 0);
5965        calc_global_load_remove(rq);
5966        break;
5967
5968    case CPU_DYING:
5969    case CPU_DYING_FROZEN:
5970        /* Update our root-domain */
5971        raw_spin_lock_irqsave(&rq->lock, flags);
5972        if (rq->rd) {
5973            BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
5974            set_rq_offline(rq);
5975        }
5976        raw_spin_unlock_irqrestore(&rq->lock, flags);
5977        break;
5978#endif
5979    }
5980    return NOTIFY_OK;
5981}
5982
5983/*
5984 * Register at high priority so that task migration (migrate_all_tasks)
5985 * happens before everything else. This has to be lower priority than
5986 * the notifier in the perf_event subsystem, though.
5987 */
5988static struct notifier_block __cpuinitdata migration_notifier = {
5989    .notifier_call = migration_call,
5990    .priority = CPU_PRI_MIGRATION,
5991};
5992
5993static int __cpuinit sched_cpu_active(struct notifier_block *nfb,
5994                      unsigned long action, void *hcpu)
5995{
5996    switch (action & ~CPU_TASKS_FROZEN) {
5997    case CPU_ONLINE:
5998    case CPU_DOWN_FAILED:
5999        set_cpu_active((long)hcpu, true);
6000        return NOTIFY_OK;
6001    default:
6002        return NOTIFY_DONE;
6003    }
6004}
6005
6006static int __cpuinit sched_cpu_inactive(struct notifier_block *nfb,
6007                    unsigned long action, void *hcpu)
6008{
6009    switch (action & ~CPU_TASKS_FROZEN) {
6010    case CPU_DOWN_PREPARE:
6011        set_cpu_active((long)hcpu, false);
6012        return NOTIFY_OK;
6013    default:
6014        return NOTIFY_DONE;
6015    }
6016}
6017
6018static int __init migration_init(void)
6019{
6020    void *cpu = (void *)(long)smp_processor_id();
6021    int err;
6022
6023    /* Initialize migration for the boot CPU */
6024    err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu);
6025    BUG_ON(err == NOTIFY_BAD);
6026    migration_call(&migration_notifier, CPU_ONLINE, cpu);
6027    register_cpu_notifier(&migration_notifier);
6028
6029    /* Register cpu active notifiers */
6030    cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE);
6031    cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE);
6032
6033    return 0;
6034}
6035early_initcall(migration_init);
6036#endif
6037
6038#ifdef CONFIG_SMP
6039
6040#ifdef CONFIG_SCHED_DEBUG
6041
6042static __read_mostly int sched_domain_debug_enabled;
6043
6044static int __init sched_domain_debug_setup(char *str)
6045{
6046    sched_domain_debug_enabled = 1;
6047
6048    return 0;
6049}
6050early_param("sched_debug", sched_domain_debug_setup);
6051
6052static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
6053                  struct cpumask *groupmask)
6054{
6055    struct sched_group *group = sd->groups;
6056    char str[256];
6057
6058    cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd));
6059    cpumask_clear(groupmask);
6060
6061    printk(KERN_DEBUG "%*s domain %d: ", level, "", level);
6062
6063    if (!(sd->flags & SD_LOAD_BALANCE)) {
6064        printk("does not load-balance\n");
6065        if (sd->parent)
6066            printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain"
6067                    " has parent");
6068        return -1;
6069    }
6070
6071    printk(KERN_CONT "span %s level %s\n", str, sd->name);
6072
6073    if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) {
6074        printk(KERN_ERR "ERROR: domain->span does not contain "
6075                "CPU%d\n", cpu);
6076    }
6077    if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) {
6078        printk(KERN_ERR "ERROR: domain->groups does not contain"
6079                " CPU%d\n", cpu);
6080    }
6081
6082    printk(KERN_DEBUG "%*s groups:", level + 1, "");
6083    do {
6084        if (!group) {
6085            printk("\n");
6086            printk(KERN_ERR "ERROR: group is NULL\n");
6087            break;
6088        }
6089
6090        if (!group->cpu_power) {
6091            printk(KERN_CONT "\n");
6092            printk(KERN_ERR "ERROR: domain->cpu_power not "
6093                    "set\n");
6094            break;
6095        }
6096
6097        if (!cpumask_weight(sched_group_cpus(group))) {
6098            printk(KERN_CONT "\n");
6099            printk(KERN_ERR "ERROR: empty group\n");
6100            break;
6101        }
6102
6103        if (cpumask_intersects(groupmask, sched_group_cpus(group))) {
6104            printk(KERN_CONT "\n");
6105            printk(KERN_ERR "ERROR: repeated CPUs\n");
6106            break;
6107        }
6108
6109        cpumask_or(groupmask, groupmask, sched_group_cpus(group));
6110
6111        cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group));
6112
6113        printk(KERN_CONT " %s", str);
6114        if (group->cpu_power != SCHED_LOAD_SCALE) {
6115            printk(KERN_CONT " (cpu_power = %d)",
6116                group->cpu_power);
6117        }
6118
6119        group = group->next;
6120    } while (group != sd->groups);
6121    printk(KERN_CONT "\n");
6122
6123    if (!cpumask_equal(sched_domain_span(sd), groupmask))
6124        printk(KERN_ERR "ERROR: groups don't span domain->span\n");
6125
6126    if (sd->parent &&
6127        !cpumask_subset(groupmask, sched_domain_span(sd->parent)))
6128        printk(KERN_ERR "ERROR: parent span is not a superset "
6129            "of domain->span\n");
6130    return 0;
6131}
6132
6133static void sched_domain_debug(struct sched_domain *sd, int cpu)
6134{
6135    cpumask_var_t groupmask;
6136    int level = 0;
6137
6138    if (!sched_domain_debug_enabled)
6139        return;
6140
6141    if (!sd) {
6142        printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu);
6143        return;
6144    }
6145
6146    printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu);
6147
6148    if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) {
6149        printk(KERN_DEBUG "Cannot load-balance (out of memory)\n");
6150        return;
6151    }
6152
6153    for (;;) {
6154        if (sched_domain_debug_one(sd, cpu, level, groupmask))
6155            break;
6156        level++;
6157        sd = sd->parent;
6158        if (!sd)
6159            break;
6160    }
6161    free_cpumask_var(groupmask);
6162}
6163#else /* !CONFIG_SCHED_DEBUG */
6164# define sched_domain_debug(sd, cpu) do { } while (0)
6165#endif /* CONFIG_SCHED_DEBUG */
6166
6167static int sd_degenerate(struct sched_domain *sd)
6168{
6169    if (cpumask_weight(sched_domain_span(sd)) == 1)
6170        return 1;
6171
6172    /* Following flags need at least 2 groups */
6173    if (sd->flags & (SD_LOAD_BALANCE |
6174             SD_BALANCE_NEWIDLE |
6175             SD_BALANCE_FORK |
6176             SD_BALANCE_EXEC |
6177             SD_SHARE_CPUPOWER |
6178             SD_SHARE_PKG_RESOURCES)) {
6179        if (sd->groups != sd->groups->next)
6180            return 0;
6181    }
6182
6183    /* Following flags don't use groups */
6184    if (sd->flags & (SD_WAKE_AFFINE))
6185        return 0;
6186
6187    return 1;
6188}
6189
6190static int
6191sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent)
6192{
6193    unsigned long cflags = sd->flags, pflags = parent->flags;
6194
6195    if (sd_degenerate(parent))
6196        return 1;
6197
6198    if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent)))
6199        return 0;
6200
6201    /* Flags needing groups don't count if only 1 group in parent */
6202    if (parent->groups == parent->groups->next) {
6203        pflags &= ~(SD_LOAD_BALANCE |
6204                SD_BALANCE_NEWIDLE |
6205                SD_BALANCE_FORK |
6206                SD_BALANCE_EXEC |
6207                SD_SHARE_CPUPOWER |
6208                SD_SHARE_PKG_RESOURCES);
6209        if (nr_node_ids == 1)
6210            pflags &= ~SD_SERIALIZE;
6211    }
6212    if (~cflags & pflags)
6213        return 0;
6214
6215    return 1;
6216}
6217
6218static void free_rootdomain(struct root_domain *rd)
6219{
6220    synchronize_sched();
6221
6222    cpupri_cleanup(&rd->cpupri);
6223
6224    free_cpumask_var(rd->rto_mask);
6225    free_cpumask_var(rd->online);
6226    free_cpumask_var(rd->span);
6227    kfree(rd);
6228}
6229
6230static void rq_attach_root(struct rq *rq, struct root_domain *rd)
6231{
6232    struct root_domain *old_rd = NULL;
6233    unsigned long flags;
6234
6235    raw_spin_lock_irqsave(&rq->lock, flags);
6236
6237    if (rq->rd) {
6238        old_rd = rq->rd;
6239
6240        if (cpumask_test_cpu(rq->cpu, old_rd->online))
6241            set_rq_offline(rq);
6242
6243        cpumask_clear_cpu(rq->cpu, old_rd->span);
6244
6245        /*
6246         * If we dont want to free the old_rt yet then
6247         * set old_rd to NULL to skip the freeing later
6248         * in this function:
6249         */
6250        if (!atomic_dec_and_test(&old_rd->refcount))
6251            old_rd = NULL;
6252    }
6253
6254    atomic_inc(&rd->refcount);
6255    rq->rd = rd;
6256
6257    cpumask_set_cpu(rq->cpu, rd->span);
6258    if (cpumask_test_cpu(rq->cpu, cpu_active_mask))
6259        set_rq_online(rq);
6260
6261    raw_spin_unlock_irqrestore(&rq->lock, flags);
6262
6263    if (old_rd)
6264        free_rootdomain(old_rd);
6265}
6266
6267static int init_rootdomain(struct root_domain *rd)
6268{
6269    memset(rd, 0, sizeof(*rd));
6270
6271    if (!alloc_cpumask_var(&rd->span, GFP_KERNEL))
6272        goto out;
6273    if (!alloc_cpumask_var(&rd->online, GFP_KERNEL))
6274        goto free_span;
6275    if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL))
6276        goto free_online;
6277
6278    if (cpupri_init(&rd->cpupri) != 0)
6279        goto free_rto_mask;
6280    return 0;
6281
6282free_rto_mask:
6283    free_cpumask_var(rd->rto_mask);
6284free_online:
6285    free_cpumask_var(rd->online);
6286free_span:
6287    free_cpumask_var(rd->span);
6288out:
6289    return -ENOMEM;
6290}
6291
6292static void init_defrootdomain(void)
6293{
6294    init_rootdomain(&def_root_domain);
6295
6296    atomic_set(&def_root_domain.refcount, 1);
6297}
6298
6299static struct root_domain *alloc_rootdomain(void)
6300{
6301    struct root_domain *rd;
6302
6303    rd = kmalloc(sizeof(*rd), GFP_KERNEL);
6304    if (!rd)
6305        return NULL;
6306
6307    if (init_rootdomain(rd) != 0) {
6308        kfree(rd);
6309        return NULL;
6310    }
6311
6312    return rd;
6313}
6314