linux/kernel/sched.c
<<
>>
Prefs
   1/*
   2 *  kernel/sched.c
   3 *
   4 *  Kernel scheduler and related syscalls
   5 *
   6 *  Copyright (C) 1991-2002  Linus Torvalds
   7 *
   8 *  1996-12-23  Modified by Dave Grothe to fix bugs in semaphores and
   9 *              make semaphores SMP safe
  10 *  1998-11-19  Implemented schedule_timeout() and related stuff
  11 *              by Andrea Arcangeli
  12 *  2002-01-04  New ultra-scalable O(1) scheduler by Ingo Molnar:
  13 *              hybrid priority-list and round-robin design with
  14 *              an array-switch method of distributing timeslices
  15 *              and per-CPU runqueues.  Cleanups and useful suggestions
  16 *              by Davide Libenzi, preemptible kernel bits by Robert Love.
  17 *  2003-09-03  Interactivity tuning by Con Kolivas.
  18 *  2004-04-02  Scheduler domains code by Nick Piggin
  19 *  2007-04-15  Work begun on replacing all interactivity tuning with a
  20 *              fair scheduling design by Con Kolivas.
  21 *  2007-05-05  Load balancing (smp-nice) and other improvements
  22 *              by Peter Williams
  23 *  2007-05-06  Interactivity improvements to CFS by Mike Galbraith
  24 *  2007-07-01  Group scheduling enhancements by Srivatsa Vaddagiri
  25 *  2007-11-29  RT balancing improvements by Steven Rostedt, Gregory Haskins,
  26 *              Thomas Gleixner, Mike Kravetz
  27 */
  28
  29#include <linux/mm.h>
  30#include <linux/module.h>
  31#include <linux/nmi.h>
  32#include <linux/init.h>
  33#include <linux/uaccess.h>
  34#include <linux/highmem.h>
  35#include <asm/mmu_context.h>
  36#include <linux/interrupt.h>
  37#include <linux/capability.h>
  38#include <linux/completion.h>
  39#include <linux/kernel_stat.h>
  40#include <linux/debug_locks.h>
  41#include <linux/perf_event.h>
  42#include <linux/security.h>
  43#include <linux/notifier.h>
  44#include <linux/profile.h>
  45#include <linux/freezer.h>
  46#include <linux/vmalloc.h>
  47#include <linux/blkdev.h>
  48#include <linux/delay.h>
  49#include <linux/pid_namespace.h>
  50#include <linux/smp.h>
  51#include <linux/threads.h>
  52#include <linux/timer.h>
  53#include <linux/rcupdate.h>
  54#include <linux/cpu.h>
  55#include <linux/cpuset.h>
  56#include <linux/percpu.h>
  57#include <linux/proc_fs.h>
  58#include <linux/seq_file.h>
  59#include <linux/stop_machine.h>
  60#include <linux/sysctl.h>
  61#include <linux/syscalls.h>
  62#include <linux/times.h>
  63#include <linux/tsacct_kern.h>
  64#include <linux/kprobes.h>
  65#include <linux/delayacct.h>
  66#include <linux/unistd.h>
  67#include <linux/pagemap.h>
  68#include <linux/hrtimer.h>
  69#include <linux/tick.h>
  70#include <linux/debugfs.h>
  71#include <linux/ctype.h>
  72#include <linux/ftrace.h>
  73#include <linux/slab.h>
  74
  75#include <asm/tlb.h>
  76#include <asm/irq_regs.h>
  77#include <asm/mutex.h>
  78
  79#include "sched_cpupri.h"
  80#include "workqueue_sched.h"
  81#include "sched_autogroup.h"
  82
  83#define CREATE_TRACE_POINTS
  84#include <trace/events/sched.h>
  85
  86/*
  87 * Convert user-nice values [ -20 ... 0 ... 19 ]
  88 * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ],
  89 * and back.
  90 */
  91#define NICE_TO_PRIO(nice)      (MAX_RT_PRIO + (nice) + 20)
  92#define PRIO_TO_NICE(prio)      ((prio) - MAX_RT_PRIO - 20)
  93#define TASK_NICE(p)            PRIO_TO_NICE((p)->static_prio)
  94
  95/*
  96 * 'User priority' is the nice value converted to something we
  97 * can work with better when scaling various scheduler parameters,
  98 * it's a [ 0 ... 39 ] range.
  99 */
 100#define USER_PRIO(p)            ((p)-MAX_RT_PRIO)
 101#define TASK_USER_PRIO(p)       USER_PRIO((p)->static_prio)
 102#define MAX_USER_PRIO           (USER_PRIO(MAX_PRIO))
 103
 104/*
 105 * Helpers for converting nanosecond timing to jiffy resolution
 106 */
 107#define NS_TO_JIFFIES(TIME)     ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
 108
 109#define NICE_0_LOAD             SCHED_LOAD_SCALE
 110#define NICE_0_SHIFT            SCHED_LOAD_SHIFT
 111
 112/*
 113 * These are the 'tuning knobs' of the scheduler:
 114 *
 115 * default timeslice is 100 msecs (used only for SCHED_RR tasks).
 116 * Timeslices get refilled after they expire.
 117 */
 118#define DEF_TIMESLICE           (100 * HZ / 1000)
 119
 120/*
 121 * single value that denotes runtime == period, ie unlimited time.
 122 */
 123#define RUNTIME_INF     ((u64)~0ULL)
 124
 125static inline int rt_policy(int policy)
 126{
 127        if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR))
 128                return 1;
 129        return 0;
 130}
 131
 132static inline int task_has_rt_policy(struct task_struct *p)
 133{
 134        return rt_policy(p->policy);
 135}
 136
 137/*
 138 * This is the priority-queue data structure of the RT scheduling class:
 139 */
 140struct rt_prio_array {
 141        DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
 142        struct list_head queue[MAX_RT_PRIO];
 143};
 144
 145struct rt_bandwidth {
 146        /* nests inside the rq lock: */
 147        raw_spinlock_t          rt_runtime_lock;
 148        ktime_t                 rt_period;
 149        u64                     rt_runtime;
 150        struct hrtimer          rt_period_timer;
 151};
 152
 153static struct rt_bandwidth def_rt_bandwidth;
 154
 155static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun);
 156
 157static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer)
 158{
 159        struct rt_bandwidth *rt_b =
 160                container_of(timer, struct rt_bandwidth, rt_period_timer);
 161        ktime_t now;
 162        int overrun;
 163        int idle = 0;
 164
 165        for (;;) {
 166                now = hrtimer_cb_get_time(timer);
 167                overrun = hrtimer_forward(timer, now, rt_b->rt_period);
 168
 169                if (!overrun)
 170                        break;
 171
 172                idle = do_sched_rt_period_timer(rt_b, overrun);
 173        }
 174
 175        return idle ? HRTIMER_NORESTART : HRTIMER_RESTART;
 176}
 177
 178static
 179void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime)
 180{
 181        rt_b->rt_period = ns_to_ktime(period);
 182        rt_b->rt_runtime = runtime;
 183
 184        raw_spin_lock_init(&rt_b->rt_runtime_lock);
 185
 186        hrtimer_init(&rt_b->rt_period_timer,
 187                        CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 188        rt_b->rt_period_timer.function = sched_rt_period_timer;
 189}
 190
 191static inline int rt_bandwidth_enabled(void)
 192{
 193        return sysctl_sched_rt_runtime >= 0;
 194}
 195
 196static void start_rt_bandwidth(struct rt_bandwidth *rt_b)
 197{
 198        ktime_t now;
 199
 200        if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF)
 201                return;
 202
 203        if (hrtimer_active(&rt_b->rt_period_timer))
 204                return;
 205
 206        raw_spin_lock(&rt_b->rt_runtime_lock);
 207        for (;;) {
 208                unsigned long delta;
 209                ktime_t soft, hard;
 210
 211                if (hrtimer_active(&rt_b->rt_period_timer))
 212                        break;
 213
 214                now = hrtimer_cb_get_time(&rt_b->rt_period_timer);
 215                hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period);
 216
 217                soft = hrtimer_get_softexpires(&rt_b->rt_period_timer);
 218                hard = hrtimer_get_expires(&rt_b->rt_period_timer);
 219                delta = ktime_to_ns(ktime_sub(hard, soft));
 220                __hrtimer_start_range_ns(&rt_b->rt_period_timer, soft, delta,
 221                                HRTIMER_MODE_ABS_PINNED, 0);
 222        }
 223        raw_spin_unlock(&rt_b->rt_runtime_lock);
 224}
 225
 226#ifdef CONFIG_RT_GROUP_SCHED
 227static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b)
 228{
 229        hrtimer_cancel(&rt_b->rt_period_timer);
 230}
 231#endif
 232
 233/*
 234 * sched_domains_mutex serializes calls to arch_init_sched_domains,
 235 * detach_destroy_domains and partition_sched_domains.
 236 */
 237static DEFINE_MUTEX(sched_domains_mutex);
 238
 239#ifdef CONFIG_CGROUP_SCHED
 240
 241#include <linux/cgroup.h>
 242
 243struct cfs_rq;
 244
 245static LIST_HEAD(task_groups);
 246
 247/* task group related information */
 248struct task_group {
 249        struct cgroup_subsys_state css;
 250
 251#ifdef CONFIG_FAIR_GROUP_SCHED
 252        /* schedulable entities of this group on each cpu */
 253        struct sched_entity **se;
 254        /* runqueue "owned" by this group on each cpu */
 255        struct cfs_rq **cfs_rq;
 256        unsigned long shares;
 257
 258        atomic_t load_weight;
 259#endif
 260
 261#ifdef CONFIG_RT_GROUP_SCHED
 262        struct sched_rt_entity **rt_se;
 263        struct rt_rq **rt_rq;
 264
 265        struct rt_bandwidth rt_bandwidth;
 266#endif
 267
 268        struct rcu_head rcu;
 269        struct list_head list;
 270
 271        struct task_group *parent;
 272        struct list_head siblings;
 273        struct list_head children;
 274
 275#ifdef CONFIG_SCHED_AUTOGROUP
 276        struct autogroup *autogroup;
 277#endif
 278};
 279
 280/* task_group_lock serializes the addition/removal of task groups */
 281static DEFINE_SPINLOCK(task_group_lock);
 282
 283#ifdef CONFIG_FAIR_GROUP_SCHED
 284
 285# define ROOT_TASK_GROUP_LOAD   NICE_0_LOAD
 286
 287/*
 288 * A weight of 0 or 1 can cause arithmetics problems.
 289 * A weight of a cfs_rq is the sum of weights of which entities
 290 * are queued on this cfs_rq, so a weight of a entity should not be
 291 * too large, so as the shares value of a task group.
 292 * (The default weight is 1024 - so there's no practical
 293 *  limitation from this.)
 294 */
 295#define MIN_SHARES      2
 296#define MAX_SHARES      (1UL << 18)
 297
 298static int root_task_group_load = ROOT_TASK_GROUP_LOAD;
 299#endif
 300
 301/* Default task group.
 302 *      Every task in system belong to this group at bootup.
 303 */
 304struct task_group root_task_group;
 305
 306#endif  /* CONFIG_CGROUP_SCHED */
 307
 308/* CFS-related fields in a runqueue */
 309struct cfs_rq {
 310        struct load_weight load;
 311        unsigned long nr_running;
 312
 313        u64 exec_clock;
 314        u64 min_vruntime;
 315
 316        struct rb_root tasks_timeline;
 317        struct rb_node *rb_leftmost;
 318
 319        struct list_head tasks;
 320        struct list_head *balance_iterator;
 321
 322        /*
 323         * 'curr' points to currently running entity on this cfs_rq.
 324         * It is set to NULL otherwise (i.e when none are currently running).
 325         */
 326        struct sched_entity *curr, *next, *last, *skip;
 327
 328        unsigned int nr_spread_over;
 329
 330#ifdef CONFIG_FAIR_GROUP_SCHED
 331        struct rq *rq;  /* cpu runqueue to which this cfs_rq is attached */
 332
 333        /*
 334         * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
 335         * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
 336         * (like users, containers etc.)
 337         *
 338         * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
 339         * list is used during load balance.
 340         */
 341        int on_list;
 342        struct list_head leaf_cfs_rq_list;
 343        struct task_group *tg;  /* group that "owns" this runqueue */
 344
 345#ifdef CONFIG_SMP
 346        /*
 347         * the part of load.weight contributed by tasks
 348         */
 349        unsigned long task_weight;
 350
 351        /*
 352         *   h_load = weight * f(tg)
 353         *
 354         * Where f(tg) is the recursive weight fraction assigned to
 355         * this group.
 356         */
 357        unsigned long h_load;
 358
 359        /*
 360         * Maintaining per-cpu shares distribution for group scheduling
 361         *
 362         * load_stamp is the last time we updated the load average
 363         * load_last is the last time we updated the load average and saw load
 364         * load_unacc_exec_time is currently unaccounted execution time
 365         */
 366        u64 load_avg;
 367        u64 load_period;
 368        u64 load_stamp, load_last, load_unacc_exec_time;
 369
 370        unsigned long load_contribution;
 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        struct cpupri cpupri;
 430};
 431
 432/*
 433 * By default the system creates a single root-domain with all cpus as
 434 * members (mimicking the global state we have today).
 435 */
 436static struct root_domain def_root_domain;
 437
 438#endif /* CONFIG_SMP */
 439
 440/*
 441 * This is the main, per-CPU runqueue data structure.
 442 *
 443 * Locking rule: those places that want to lock multiple runqueues
 444 * (such as the load balancing or the thread migration code), lock
 445 * acquire operations must be ordered by ascending &runqueue.
 446 */
 447struct rq {
 448        /* runqueue lock: */
 449        raw_spinlock_t lock;
 450
 451        /*
 452         * nr_running and cpu_load should be in the same cacheline because
 453         * remote CPUs use both these fields when doing load calculation.
 454         */
 455        unsigned long nr_running;
 456        #define CPU_LOAD_IDX_MAX 5
 457        unsigned long cpu_load[CPU_LOAD_IDX_MAX];
 458        unsigned long last_load_update_tick;
 459#ifdef CONFIG_NO_HZ
 460        u64 nohz_stamp;
 461        unsigned char nohz_balance_kick;
 462#endif
 463        unsigned int skip_clock_update;
 464
 465        /* capture load from *all* tasks on this cpu: */
 466        struct load_weight load;
 467        unsigned long nr_load_updates;
 468        u64 nr_switches;
 469
 470        struct cfs_rq cfs;
 471        struct rt_rq rt;
 472
 473#ifdef CONFIG_FAIR_GROUP_SCHED
 474        /* list of leaf cfs_rq on this cpu: */
 475        struct list_head leaf_cfs_rq_list;
 476#endif
 477#ifdef CONFIG_RT_GROUP_SCHED
 478        struct list_head leaf_rt_rq_list;
 479#endif
 480
 481        /*
 482         * This is part of a global counter where only the total sum
 483         * over all CPUs matters. A task can increase this counter on
 484         * one CPU and if it got migrated afterwards it may decrease
 485         * it on another CPU. Always updated under the runqueue lock:
 486         */
 487        unsigned long nr_uninterruptible;
 488
 489        struct task_struct *curr, *idle, *stop;
 490        unsigned long next_balance;
 491        struct mm_struct *prev_mm;
 492
 493        u64 clock;
 494        u64 clock_task;
 495
 496        atomic_t nr_iowait;
 497
 498#ifdef CONFIG_SMP
 499        struct root_domain *rd;
 500        struct sched_domain *sd;
 501
 502        unsigned long cpu_power;
 503
 504        unsigned char idle_at_tick;
 505        /* For active balancing */
 506        int post_schedule;
 507        int active_balance;
 508        int push_cpu;
 509        struct cpu_stop_work active_balance_work;
 510        /* cpu of this runqueue: */
 511        int cpu;
 512        int online;
 513
 514        unsigned long avg_load_per_task;
 515
 516        u64 rt_avg;
 517        u64 age_stamp;
 518        u64 idle_stamp;
 519        u64 avg_idle;
 520#endif
 521
 522#ifdef CONFIG_IRQ_TIME_ACCOUNTING
 523        u64 prev_irq_time;
 524#endif
 525
 526        /* calc_load related fields */
 527        unsigned long calc_load_update;
 528        long calc_load_active;
 529
 530#ifdef CONFIG_SCHED_HRTICK
 531#ifdef CONFIG_SMP
 532        int hrtick_csd_pending;
 533        struct call_single_data hrtick_csd;
 534#endif
 535        struct hrtimer hrtick_timer;
 536#endif
 537
 538#ifdef CONFIG_SCHEDSTATS
 539        /* latency stats */
 540        struct sched_info rq_sched_info;
 541        unsigned long long rq_cpu_time;
 542        /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
 543
 544        /* sys_sched_yield() stats */
 545        unsigned int yld_count;
 546
 547        /* schedule() stats */
 548        unsigned int sched_switch;
 549        unsigned int sched_count;
 550        unsigned int sched_goidle;
 551
 552        /* try_to_wake_up() stats */
 553        unsigned int ttwu_count;
 554        unsigned int ttwu_local;
 555#endif
 556};
 557
 558static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
 559
 560
 561static void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
 562
 563static inline int cpu_of(struct rq *rq)
 564{
 565#ifdef CONFIG_SMP
 566        return rq->cpu;
 567#else
 568        return 0;
 569#endif
 570}
 571
 572#define rcu_dereference_check_sched_domain(p) \
 573        rcu_dereference_check((p), \
 574                              rcu_read_lock_sched_held() || \
 575                              lockdep_is_held(&sched_domains_mutex))
 576
 577/*
 578 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
 579 * See detach_destroy_domains: synchronize_sched for details.
 580 *
 581 * The domain tree of any CPU may only be accessed from within
 582 * preempt-disabled sections.
 583 */
 584#define for_each_domain(cpu, __sd) \
 585        for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent)
 586
 587#define cpu_rq(cpu)             (&per_cpu(runqueues, (cpu)))
 588#define this_rq()               (&__get_cpu_var(runqueues))
 589#define task_rq(p)              cpu_rq(task_cpu(p))
 590#define cpu_curr(cpu)           (cpu_rq(cpu)->curr)
 591#define raw_rq()                (&__raw_get_cpu_var(runqueues))
 592
 593#ifdef CONFIG_CGROUP_SCHED
 594
 595/*
 596 * Return the group to which this tasks belongs.
 597 *
 598 * We use task_subsys_state_check() and extend the RCU verification
 599 * with lockdep_is_held(&task_rq(p)->lock) because cpu_cgroup_attach()
 600 * holds that lock for each task it moves into the cgroup. Therefore
 601 * by holding that lock, we pin the task to the current cgroup.
 602 */
 603static inline struct task_group *task_group(struct task_struct *p)
 604{
 605        struct task_group *tg;
 606        struct cgroup_subsys_state *css;
 607
 608        css = task_subsys_state_check(p, cpu_cgroup_subsys_id,
 609                        lockdep_is_held(&task_rq(p)->lock));
 610        tg = container_of(css, struct task_group, css);
 611
 612        return autogroup_task_group(p, tg);
 613}
 614
 615/* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
 616static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
 617{
 618#ifdef CONFIG_FAIR_GROUP_SCHED
 619        p->se.cfs_rq = task_group(p)->cfs_rq[cpu];
 620        p->se.parent = task_group(p)->se[cpu];
 621#endif
 622
 623#ifdef CONFIG_RT_GROUP_SCHED
 624        p->rt.rt_rq  = task_group(p)->rt_rq[cpu];
 625        p->rt.parent = task_group(p)->rt_se[cpu];
 626#endif
 627}
 628
 629#else /* CONFIG_CGROUP_SCHED */
 630
 631static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
 632static inline struct task_group *task_group(struct task_struct *p)
 633{
 634        return NULL;
 635}
 636
 637#endif /* CONFIG_CGROUP_SCHED */
 638
 639static void update_rq_clock_task(struct rq *rq, s64 delta);
 640
 641static void update_rq_clock(struct rq *rq)
 642{
 643        s64 delta;
 644
 645        if (rq->skip_clock_update)
 646                return;
 647
 648        delta = sched_clock_cpu(cpu_of(rq)) - rq->clock;
 649        rq->clock += delta;
 650        update_rq_clock_task(rq, delta);
 651}
 652
 653/*
 654 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
 655 */
 656#ifdef CONFIG_SCHED_DEBUG
 657# define const_debug __read_mostly
 658#else
 659# define const_debug static const
 660#endif
 661
 662/**
 663 * runqueue_is_locked - Returns true if the current cpu runqueue is locked
 664 * @cpu: the processor in question.
 665 *
 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;
 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        cmp = strstrip(buf);
 738
 739        if (strncmp(cmp, "NO_", 3) == 0) {
 740                neg = 1;
 741                cmp += 3;
 742        }
 743
 744        for (i = 0; sched_feat_names[i]; i++) {
 745                if (strcmp(cmp, sched_feat_names[i]) == 0) {
 746                        if (neg)
 747                                sysctl_sched_features &= ~(1UL << i);
 748                        else
 749                                sysctl_sched_features |= (1UL << i);
 750                        break;
 751                }
 752        }
 753
 754        if (!sched_feat_names[i])
 755                return -EINVAL;
 756
 757        *ppos += cnt;
 758
 759        return cnt;
 760}
 761
 762static int sched_feat_open(struct inode *inode, struct file *filp)
 763{
 764        return single_open(filp, sched_feat_show, NULL);
 765}
 766
 767static const struct file_operations sched_feat_fops = {
 768        .open           = sched_feat_open,
 769        .write          = sched_feat_write,
 770        .read           = seq_read,
 771        .llseek         = seq_lseek,
 772        .release        = single_release,
 773};
 774
 775static __init int sched_init_debug(void)
 776{
 777        debugfs_create_file("sched_features", 0644, NULL, NULL,
 778                        &sched_feat_fops);
 779
 780        return 0;
 781}
 782late_initcall(sched_init_debug);
 783
 784#endif
 785
 786#define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
 787
 788/*
 789 * Number of tasks to iterate in a single balance run.
 790 * Limited because this is done with IRQs disabled.
 791 */
 792const_debug unsigned int sysctl_sched_nr_migrate = 32;
 793
 794/*
 795 * period over which we average the RT time consumption, measured
 796 * in ms.
 797 *
 798 * default: 1s
 799 */
 800const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC;
 801
 802/*
 803 * period over which we measure -rt task cpu usage in us.
 804 * default: 1s
 805 */
 806unsigned int sysctl_sched_rt_period = 1000000;
 807
 808static __read_mostly int scheduler_running;
 809
 810/*
 811 * part of the period that we allow rt tasks to run in us.
 812 * default: 0.95s
 813 */
 814int sysctl_sched_rt_runtime = 950000;
 815
 816static inline u64 global_rt_period(void)
 817{
 818        return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
 819}
 820
 821static inline u64 global_rt_runtime(void)
 822{
 823        if (sysctl_sched_rt_runtime < 0)
 824                return RUNTIME_INF;
 825
 826        return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
 827}
 828
 829#ifndef prepare_arch_switch
 830# define prepare_arch_switch(next)      do { } while (0)
 831#endif
 832#ifndef finish_arch_switch
 833# define finish_arch_switch(prev)       do { } while (0)
 834#endif
 835
 836static inline int task_current(struct rq *rq, struct task_struct *p)
 837{
 838        return rq->curr == p;
 839}
 840
 841#ifndef __ARCH_WANT_UNLOCKED_CTXSW
 842static inline int task_running(struct rq *rq, struct task_struct *p)
 843{
 844        return task_current(rq, p);
 845}
 846
 847static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
 848{
 849}
 850
 851static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
 852{
 853#ifdef CONFIG_DEBUG_SPINLOCK
 854        /* this is a valid case when another task releases the spinlock */
 855        rq->lock.owner = current;
 856#endif
 857        /*
 858         * If we are tracking spinlock dependencies then we have to
 859         * fix up the runqueue lock - which gets 'carried over' from
 860         * prev into current:
 861         */
 862        spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
 863
 864        raw_spin_unlock_irq(&rq->lock);
 865}
 866
 867#else /* __ARCH_WANT_UNLOCKED_CTXSW */
 868static inline int task_running(struct rq *rq, struct task_struct *p)
 869{
 870#ifdef CONFIG_SMP
 871        return p->oncpu;
 872#else
 873        return task_current(rq, p);
 874#endif
 875}
 876
 877static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
 878{
 879#ifdef CONFIG_SMP
 880        /*
 881         * We can optimise this out completely for !SMP, because the
 882         * SMP rebalancing from interrupt is the only thing that cares
 883         * here.
 884         */
 885        next->oncpu = 1;
 886#endif
 887#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
 888        raw_spin_unlock_irq(&rq->lock);
 889#else
 890        raw_spin_unlock(&rq->lock);
 891#endif
 892}
 893
 894static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
 895{
 896#ifdef CONFIG_SMP
 897        /*
 898         * After ->oncpu is cleared, the task can be moved to a different CPU.
 899         * We must ensure this doesn't happen until the switch is completely
 900         * finished.
 901         */
 902        smp_wmb();
 903        prev->oncpu = 0;
 904#endif
 905#ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW
 906        local_irq_enable();
 907#endif
 908}
 909#endif /* __ARCH_WANT_UNLOCKED_CTXSW */
 910
 911/*
 912 * Check whether the task is waking, we use this to synchronize ->cpus_allowed
 913 * against ttwu().
 914 */
 915static inline int task_is_waking(struct task_struct *p)
 916{
 917        return unlikely(p->state == TASK_WAKING);
 918}
 919
 920/*
 921 * __task_rq_lock - lock the runqueue a given task resides on.
 922 * Must be called interrupts disabled.
 923 */
 924static inline struct rq *__task_rq_lock(struct task_struct *p)
 925        __acquires(rq->lock)
 926{
 927        struct rq *rq;
 928
 929        for (;;) {
 930                rq = task_rq(p);
 931                raw_spin_lock(&rq->lock);
 932                if (likely(rq == task_rq(p)))
 933                        return rq;
 934                raw_spin_unlock(&rq->lock);
 935        }
 936}
 937
 938/*
 939 * task_rq_lock - lock the runqueue a given task resides on and disable
 940 * interrupts. Note the ordering: we can safely lookup the task_rq without
 941 * explicitly disabling preemption.
 942 */
 943static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags)
 944        __acquires(rq->lock)
 945{
 946        struct rq *rq;
 947
 948        for (;;) {
 949                local_irq_save(*flags);
 950                rq = task_rq(p);
 951                raw_spin_lock(&rq->lock);
 952                if (likely(rq == task_rq(p)))
 953                        return rq;
 954                raw_spin_unlock_irqrestore(&rq->lock, *flags);
 955        }
 956}
 957
 958static void __task_rq_unlock(struct rq *rq)
 959        __releases(rq->lock)
 960{
 961        raw_spin_unlock(&rq->lock);
 962}
 963
 964static inline void task_rq_unlock(struct rq *rq, unsigned long *flags)
 965        __releases(rq->lock)
 966{
 967        raw_spin_unlock_irqrestore(&rq->lock, *flags);
 968}
 969
 970/*
 971 * this_rq_lock - lock this runqueue and disable interrupts.
 972 */
 973static struct rq *this_rq_lock(void)
 974        __acquires(rq->lock)
 975{
 976        struct rq *rq;
 977
 978        local_irq_disable();
 979        rq = this_rq();
 980        raw_spin_lock(&rq->lock);
 981
 982        return rq;
 983}
 984
 985#ifdef CONFIG_SCHED_HRTICK
 986/*
 987 * Use HR-timers to deliver accurate preemption points.
 988 *
 989 * Its all a bit involved since we cannot program an hrt while holding the
 990 * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a
 991 * reschedule event.
 992 *
 993 * When we get rescheduled we reprogram the hrtick_timer outside of the
 994 * rq->lock.
 995 */
 996
 997/*
 998 * Use hrtick when:
 999 *  - enabled by features
1000 *  - hrtimer is actually high res
1001 */
1002static inline int hrtick_enabled(struct rq *rq)
1003{
1004        if (!sched_feat(HRTICK))
1005                return 0;
1006        if (!cpu_active(cpu_of(rq)))
1007                return 0;
1008        return hrtimer_is_hres_active(&rq->hrtick_timer);
1009}
1010
1011static void hrtick_clear(struct rq *rq)
1012{
1013        if (hrtimer_active(&rq->hrtick_timer))
1014                hrtimer_cancel(&rq->hrtick_timer);
1015}
1016
1017/*
1018 * High-resolution timer tick.
1019 * Runs from hardirq context with interrupts disabled.
1020 */
1021static enum hrtimer_restart hrtick(struct hrtimer *timer)
1022{
1023        struct rq *rq = container_of(timer, struct rq, hrtick_timer);
1024
1025        WARN_ON_ONCE(cpu_of(rq) != smp_processor_id());
1026
1027        raw_spin_lock(&rq->lock);
1028        update_rq_clock(rq);
1029        rq->curr->sched_class->task_tick(rq, rq->curr, 1);
1030        raw_spin_unlock(&rq->lock);
1031
1032        return HRTIMER_NORESTART;
1033}
1034
1035#ifdef CONFIG_SMP
1036/*
1037 * called from hardirq (IPI) context
1038 */
1039static void __hrtick_start(void *arg)
1040{
1041        struct rq *rq = arg;
1042
1043        raw_spin_lock(&rq->lock);
1044        hrtimer_restart(&rq->hrtick_timer);
1045        rq->hrtick_csd_pending = 0;
1046        raw_spin_unlock(&rq->lock);
1047}
1048
1049/*
1050 * Called to set the hrtick timer state.
1051 *
1052 * called with rq->lock held and irqs disabled
1053 */
1054static void hrtick_start(struct rq *rq, u64 delay)
1055{
1056        struct hrtimer *timer = &rq->hrtick_timer;
1057        ktime_t time = ktime_add_ns(timer->base->get_time(), delay);
1058
1059        hrtimer_set_expires(timer, time);
1060
1061        if (rq == this_rq()) {
1062                hrtimer_restart(timer);
1063        } else if (!rq->hrtick_csd_pending) {
1064                __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0);
1065                rq->hrtick_csd_pending = 1;
1066        }
1067}
1068
1069static int
1070hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu)
1071{
1072        int cpu = (int)(long)hcpu;
1073
1074        switch (action) {
1075        case CPU_UP_CANCELED:
1076        case CPU_UP_CANCELED_FROZEN:
1077        case CPU_DOWN_PREPARE:
1078        case CPU_DOWN_PREPARE_FROZEN:
1079        case CPU_DEAD:
1080        case CPU_DEAD_FROZEN:
1081                hrtick_clear(cpu_rq(cpu));
1082                return NOTIFY_OK;
1083        }
1084
1085        return NOTIFY_DONE;
1086}
1087
1088static __init void init_hrtick(void)
1089{
1090        hotcpu_notifier(hotplug_hrtick, 0);
1091}
1092#else
1093/*
1094 * Called to set the hrtick timer state.
1095 *
1096 * called with rq->lock held and irqs disabled
1097 */
1098static void hrtick_start(struct rq *rq, u64 delay)
1099{
1100        __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0,
1101                        HRTIMER_MODE_REL_PINNED, 0);
1102}
1103
1104static inline void init_hrtick(void)
1105{
1106}
1107#endif /* CONFIG_SMP */
1108
1109static void init_rq_hrtick(struct rq *rq)
1110{
1111#ifdef CONFIG_SMP
1112        rq->hrtick_csd_pending = 0;
1113
1114        rq->hrtick_csd.flags = 0;
1115        rq->hrtick_csd.func = __hrtick_start;
1116        rq->hrtick_csd.info = rq;
1117#endif
1118
1119        hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1120        rq->hrtick_timer.function = hrtick;
1121}
1122#else   /* CONFIG_SCHED_HRTICK */
1123static inline void hrtick_clear(struct rq *rq)
1124{
1125}
1126
1127static inline void init_rq_hrtick(struct rq *rq)
1128{
1129}
1130
1131static inline void init_hrtick(void)
1132{
1133}
1134#endif  /* CONFIG_SCHED_HRTICK */
1135
1136/*
1137 * resched_task - mark a task 'to be rescheduled now'.
1138 *
1139 * On UP this means the setting of the need_resched flag, on SMP it
1140 * might also involve a cross-CPU call to trigger the scheduler on
1141 * the target CPU.
1142 */
1143#ifdef CONFIG_SMP
1144
1145#ifndef tsk_is_polling
1146#define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG)
1147#endif
1148
1149static void resched_task(struct task_struct *p)
1150{
1151        int cpu;
1152
1153        assert_raw_spin_locked(&task_rq(p)->lock);
1154
1155        if (test_tsk_need_resched(p))
1156                return;
1157
1158        set_tsk_need_resched(p);
1159
1160        cpu = task_cpu(p);
1161        if (cpu == smp_processor_id())
1162                return;
1163
1164        /* NEED_RESCHED must be visible before we test polling */
1165        smp_mb();
1166        if (!tsk_is_polling(p))
1167                smp_send_reschedule(cpu);
1168}
1169
1170static void resched_cpu(int cpu)
1171{
1172        struct rq *rq = cpu_rq(cpu);
1173        unsigned long flags;
1174
1175        if (!raw_spin_trylock_irqsave(&rq->lock, flags))
1176                return;
1177        resched_task(cpu_curr(cpu));
1178        raw_spin_unlock_irqrestore(&rq->lock, flags);
1179}
1180
1181#ifdef CONFIG_NO_HZ
1182/*
1183 * In the semi idle case, use the nearest busy cpu for migrating timers
1184 * from an idle cpu.  This is good for power-savings.
1185 *
1186 * We don't do similar optimization for completely idle system, as
1187 * selecting an idle cpu will add more delays to the timers than intended
1188 * (as that cpu's timer base may not be uptodate wrt jiffies etc).
1189 */
1190int get_nohz_timer_target(void)
1191{
1192        int cpu = smp_processor_id();
1193        int i;
1194        struct sched_domain *sd;
1195
1196        for_each_domain(cpu, sd) {
1197                for_each_cpu(i, sched_domain_span(sd))
1198                        if (!idle_cpu(i))
1199                                return i;
1200        }
1201        return cpu;
1202}
1203/*
1204 * When add_timer_on() enqueues a timer into the timer wheel of an
1205 * idle CPU then this timer might expire before the next timer event
1206 * which is scheduled to wake up that CPU. In case of a completely
1207 * idle system the next event might even be infinite time into the
1208 * future. wake_up_idle_cpu() ensures that the CPU is woken up and
1209 * leaves the inner idle loop so the newly added timer is taken into
1210 * account when the CPU goes back to idle and evaluates the timer
1211 * wheel for the next timer event.
1212 */
1213void wake_up_idle_cpu(int cpu)
1214{
1215        struct rq *rq = cpu_rq(cpu);
1216
1217        if (cpu == smp_processor_id())
1218                return;
1219
1220        /*
1221         * This is safe, as this function is called with the timer
1222         * wheel base lock of (cpu) held. When the CPU is on the way
1223         * to idle and has not yet set rq->curr to idle then it will
1224         * be serialized on the timer wheel base lock and take the new
1225         * timer into account automatically.
1226         */
1227        if (rq->curr != rq->idle)
1228                return;
1229
1230        /*
1231         * We can set TIF_RESCHED on the idle task of the other CPU
1232         * lockless. The worst case is that the other CPU runs the
1233         * idle task through an additional NOOP schedule()
1234         */
1235        set_tsk_need_resched(rq->idle);
1236
1237        /* NEED_RESCHED must be visible before we test polling */
1238        smp_mb();
1239        if (!tsk_is_polling(rq->idle))
1240                smp_send_reschedule(cpu);
1241}
1242
1243#endif /* CONFIG_NO_HZ */
1244
1245static u64 sched_avg_period(void)
1246{
1247        return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2;
1248}
1249
1250static void sched_avg_update(struct rq *rq)
1251{
1252        s64 period = sched_avg_period();
1253
1254        while ((s64)(rq->clock - rq->age_stamp) > period) {
1255                /*
1256                 * Inline assembly required to prevent the compiler
1257                 * optimising this loop into a divmod call.
1258                 * See __iter_div_u64_rem() for another example of this.
1259                 */
1260                asm("" : "+rm" (rq->age_stamp));
1261                rq->age_stamp += period;
1262                rq->rt_avg /= 2;
1263        }
1264}
1265
1266static void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1267{
1268        rq->rt_avg += rt_delta;
1269        sched_avg_update(rq);
1270}
1271
1272#else /* !CONFIG_SMP */
1273static void resched_task(struct task_struct *p)
1274{
1275        assert_raw_spin_locked(&task_rq(p)->lock);
1276        set_tsk_need_resched(p);
1277}
1278
1279static void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1280{
1281}
1282
1283static void sched_avg_update(struct rq *rq)
1284{
1285}
1286#endif /* CONFIG_SMP */
1287
1288#if BITS_PER_LONG == 32
1289# define WMULT_CONST    (~0UL)
1290#else
1291# define WMULT_CONST    (1UL << 32)
1292#endif
1293
1294#define WMULT_SHIFT     32
1295
1296/*
1297 * Shift right and round:
1298 */
1299#define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y))
1300
1301/*
1302 * delta *= weight / lw
1303 */
1304static unsigned long
1305calc_delta_mine(unsigned long delta_exec, unsigned long weight,
1306                struct load_weight *lw)
1307{
1308        u64 tmp;
1309
1310        if (!lw->inv_weight) {
1311                if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST))
1312                        lw->inv_weight = 1;
1313                else
1314                        lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2)
1315                                / (lw->weight+1);
1316        }
1317
1318        tmp = (u64)delta_exec * weight;
1319        /*
1320         * Check whether we'd overflow the 64-bit multiplication:
1321         */
1322        if (unlikely(tmp > WMULT_CONST))
1323                tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight,
1324                        WMULT_SHIFT/2);
1325        else
1326                tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT);
1327
1328        return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX);
1329}
1330
1331static inline void update_load_add(struct load_weight *lw, unsigned long inc)
1332{
1333        lw->weight += inc;
1334        lw->inv_weight = 0;
1335}
1336
1337static inline void update_load_sub(struct load_weight *lw, unsigned long dec)
1338{
1339        lw->weight -= dec;
1340        lw->inv_weight = 0;
1341}
1342
1343static inline void update_load_set(struct load_weight *lw, unsigned long w)
1344{
1345        lw->weight = w;
1346        lw->inv_weight = 0;
1347}
1348
1349/*
1350 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1351 * of tasks with abnormal "nice" values across CPUs the contribution that
1352 * each task makes to its run queue's load is weighted according to its
1353 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1354 * scaled version of the new time slice allocation that they receive on time
1355 * slice expiry etc.
1356 */
1357
1358#define WEIGHT_IDLEPRIO                3
1359#define WMULT_IDLEPRIO         1431655765
1360
1361/*
1362 * Nice levels are multiplicative, with a gentle 10% change for every
1363 * nice level changed. I.e. when a CPU-bound task goes from nice 0 to
1364 * nice 1, it will get ~10% less CPU time than another CPU-bound task
1365 * that remained on nice 0.
1366 *
1367 * The "10% effect" is relative and cumulative: from _any_ nice level,
1368 * if you go up 1 level, it's -10% CPU usage, if you go down 1 level
1369 * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25.
1370 * If a task goes up by ~10% and another task goes down by ~10% then
1371 * the relative distance between them is ~25%.)
1372 */
1373static const int prio_to_weight[40] = {
1374 /* -20 */     88761,     71755,     56483,     46273,     36291,
1375 /* -15 */     29154,     23254,     18705,     14949,     11916,
1376 /* -10 */      9548,      7620,      6100,      4904,      3906,
1377 /*  -5 */      3121,      2501,      1991,      1586,      1277,
1378 /*   0 */      1024,       820,       655,       526,       423,
1379 /*   5 */       335,       272,       215,       172,       137,
1380 /*  10 */       110,        87,        70,        56,        45,
1381 /*  15 */        36,        29,        23,        18,        15,
1382};
1383
1384/*
1385 * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated.
1386 *
1387 * In cases where the weight does not change often, we can use the
1388 * precalculated inverse to speed up arithmetics by turning divisions
1389 * into multiplications:
1390 */
1391static const u32 prio_to_wmult[40] = {
1392 /* -20 */     48388,     59856,     76040,     92818,    118348,
1393 /* -15 */    147320,    184698,    229616,    287308,    360437,
1394 /* -10 */    449829,    563644,    704093,    875809,   1099582,
1395 /*  -5 */   1376151,   1717300,   2157191,   2708050,   3363326,
1396 /*   0 */   4194304,   5237765,   6557202,   8165337,  10153587,
1397 /*   5 */  12820798,  15790321,  19976592,  24970740,  31350126,
1398 /*  10 */  39045157,  49367440,  61356676,  76695844,  95443717,
1399 /*  15 */ 119304647, 148102320, 186737708, 238609294, 286331153,
1400};
1401
1402/* Time spent by the tasks of the cpu accounting group executing in ... */
1403enum cpuacct_stat_index {
1404        CPUACCT_STAT_USER,      /* ... user mode */
1405        CPUACCT_STAT_SYSTEM,    /* ... kernel mode */
1406
1407        CPUACCT_STAT_NSTATS,
1408};
1409
1410#ifdef CONFIG_CGROUP_CPUACCT
1411static void cpuacct_charge(struct task_struct *tsk, u64 cputime);
1412static void cpuacct_update_stats(struct task_struct *tsk,
1413                enum cpuacct_stat_index idx, cputime_t val);
1414#else
1415static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {}
1416static inline void cpuacct_update_stats(struct task_struct *tsk,
1417                enum cpuacct_stat_index idx, cputime_t val) {}
1418#endif
1419
1420static inline void inc_cpu_load(struct rq *rq, unsigned long load)
1421{
1422        update_load_add(&rq->load, load);
1423}
1424
1425static inline void dec_cpu_load(struct rq *rq, unsigned long load)
1426{
1427        update_load_sub(&rq->load, load);
1428}
1429
1430#if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED)
1431typedef int (*tg_visitor)(struct task_group *, void *);
1432
1433/*
1434 * Iterate the full tree, calling @down when first entering a node and @up when
1435 * leaving it for the final time.
1436 */
1437static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
1438{
1439        struct task_group *parent, *child;
1440        int ret;
1441
1442        rcu_read_lock();
1443        parent = &root_task_group;
1444down:
1445        ret = (*down)(parent, data);
1446        if (ret)
1447                goto out_unlock;
1448        list_for_each_entry_rcu(child, &parent->children, siblings) {
1449                parent = child;
1450                goto down;
1451
1452up:
1453                continue;
1454        }
1455        ret = (*up)(parent, data);
1456        if (ret)
1457                goto out_unlock;
1458
1459        child = parent;
1460        parent = parent->parent;
1461        if (parent)
1462                goto up;
1463out_unlock:
1464        rcu_read_unlock();
1465
1466        return ret;
1467}
1468
1469static int tg_nop(struct task_group *tg, void *data)
1470{
1471        return 0;
1472}
1473#endif
1474
1475#ifdef CONFIG_SMP
1476/* Used instead of source_load when we know the type == 0 */
1477static unsigned long weighted_cpuload(const int cpu)
1478{
1479        return cpu_rq(cpu)->load.weight;
1480}
1481
1482/*
1483 * Return a low guess at the load of a migration-source cpu weighted
1484 * according to the scheduling class and "nice" value.
1485 *
1486 * We want to under-estimate the load of migration sources, to
1487 * balance conservatively.
1488 */
1489static unsigned long source_load(int cpu, int type)
1490{
1491        struct rq *rq = cpu_rq(cpu);
1492        unsigned long total = weighted_cpuload(cpu);
1493
1494        if (type == 0 || !sched_feat(LB_BIAS))
1495                return total;
1496
1497        return min(rq->cpu_load[type-1], total);
1498}
1499
1500/*
1501 * Return a high guess at the load of a migration-target cpu weighted
1502 * according to the scheduling class and "nice" value.
1503 */
1504static unsigned long target_load(int cpu, int type)
1505{
1506        struct rq *rq = cpu_rq(cpu);
1507        unsigned long total = weighted_cpuload(cpu);
1508
1509        if (type == 0 || !sched_feat(LB_BIAS))
1510                return total;
1511
1512        return max(rq->cpu_load[type-1], total);
1513}
1514
1515static unsigned long power_of(int cpu)
1516{
1517        return cpu_rq(cpu)->cpu_power;
1518}
1519
1520static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd);
1521
1522static unsigned long cpu_avg_load_per_task(int cpu)
1523{
1524        struct rq *rq = cpu_rq(cpu);
1525        unsigned long nr_running = ACCESS_ONCE(rq->nr_running);
1526
1527        if (nr_running)
1528                rq->avg_load_per_task = rq->load.weight / nr_running;
1529        else
1530                rq->avg_load_per_task = 0;
1531
1532        return rq->avg_load_per_task;
1533}
1534
1535#ifdef CONFIG_FAIR_GROUP_SCHED
1536
1537/*
1538 * Compute the cpu's hierarchical load factor for each task group.
1539 * This needs to be done in a top-down fashion because the load of a child
1540 * group is a fraction of its parents load.
1541 */
1542static int tg_load_down(struct task_group *tg, void *data)
1543{
1544        unsigned long load;
1545        long cpu = (long)data;
1546
1547        if (!tg->parent) {
1548                load = cpu_rq(cpu)->load.weight;
1549        } else {
1550                load = tg->parent->cfs_rq[cpu]->h_load;
1551                load *= tg->se[cpu]->load.weight;
1552                load /= tg->parent->cfs_rq[cpu]->load.weight + 1;
1553        }
1554
1555        tg->cfs_rq[cpu]->h_load = load;
1556
1557        return 0;
1558}
1559
1560static void update_h_load(long cpu)
1561{
1562        walk_tg_tree(tg_load_down, tg_nop, (void *)cpu);
1563}
1564
1565#endif
1566
1567#ifdef CONFIG_PREEMPT
1568
1569static void double_rq_lock(struct rq *rq1, struct rq *rq2);
1570
1571/*
1572 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1573 * way at the expense of forcing extra atomic operations in all
1574 * invocations.  This assures that the double_lock is acquired using the
1575 * same underlying policy as the spinlock_t on this architecture, which
1576 * reduces latency compared to the unfair variant below.  However, it
1577 * also adds more overhead and therefore may reduce throughput.
1578 */
1579static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1580        __releases(this_rq->lock)
1581        __acquires(busiest->lock)
1582        __acquires(this_rq->lock)
1583{
1584        raw_spin_unlock(&this_rq->lock);
1585        double_rq_lock(this_rq, busiest);
1586
1587        return 1;
1588}
1589
1590#else
1591/*
1592 * Unfair double_lock_balance: Optimizes throughput at the expense of
1593 * latency by eliminating extra atomic operations when the locks are
1594 * already in proper order on entry.  This favors lower cpu-ids and will
1595 * grant the double lock to lower cpus over higher ids under contention,
1596 * regardless of entry order into the function.
1597 */
1598static int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1599        __releases(this_rq->lock)
1600        __acquires(busiest->lock)
1601        __acquires(this_rq->lock)
1602{
1603        int ret = 0;
1604
1605        if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1606                if (busiest < this_rq) {
1607                        raw_spin_unlock(&this_rq->lock);
1608                        raw_spin_lock(&busiest->lock);
1609                        raw_spin_lock_nested(&this_rq->lock,
1610                                              SINGLE_DEPTH_NESTING);
1611                        ret = 1;
1612                } else
1613                        raw_spin_lock_nested(&busiest->lock,
1614                                              SINGLE_DEPTH_NESTING);
1615        }
1616        return ret;
1617}
1618
1619#endif /* CONFIG_PREEMPT */
1620
1621/*
1622 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1623 */
1624static int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1625{
1626        if (unlikely(!irqs_disabled())) {
1627                /* printk() doesn't work good under rq->lock */
1628                raw_spin_unlock(&this_rq->lock);
1629                BUG_ON(1);
1630        }
1631
1632        return _double_lock_balance(this_rq, busiest);
1633}
1634
1635static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1636        __releases(busiest->lock)
1637{
1638        raw_spin_unlock(&busiest->lock);
1639        lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1640}
1641
1642/*
1643 * double_rq_lock - safely lock two runqueues
1644 *
1645 * Note this does not disable interrupts like task_rq_lock,
1646 * you need to do so manually before calling.
1647 */
1648static void double_rq_lock(struct rq *rq1, struct rq *rq2)
1649        __acquires(rq1->lock)
1650        __acquires(rq2->lock)
1651{
1652        BUG_ON(!irqs_disabled());
1653        if (rq1 == rq2) {
1654                raw_spin_lock(&rq1->lock);
1655                __acquire(rq2->lock);   /* Fake it out ;) */
1656        } else {
1657                if (rq1 < rq2) {
1658                        raw_spin_lock(&rq1->lock);
1659                        raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1660                } else {
1661                        raw_spin_lock(&rq2->lock);
1662                        raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1663                }
1664        }
1665}
1666
1667/*
1668 * double_rq_unlock - safely unlock two runqueues
1669 *
1670 * Note this does not restore interrupts like task_rq_unlock,
1671 * you need to do so manually after calling.
1672 */
1673static void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1674        __releases(rq1->lock)
1675        __releases(rq2->lock)
1676{
1677        raw_spin_unlock(&rq1->lock);
1678        if (rq1 != rq2)
1679                raw_spin_unlock(&rq2->lock);
1680        else
1681                __release(rq2->lock);
1682}
1683
1684#else /* CONFIG_SMP */
1685
1686/*
1687 * double_rq_lock - safely lock two runqueues
1688 *
1689 * Note this does not disable interrupts like task_rq_lock,
1690 * you need to do so manually before calling.
1691 */
1692static void double_rq_lock(struct rq *rq1, struct rq *rq2)
1693        __acquires(rq1->lock)
1694        __acquires(rq2->lock)
1695{
1696        BUG_ON(!irqs_disabled());
1697        BUG_ON(rq1 != rq2);
1698        raw_spin_lock(&rq1->lock);
1699        __acquire(rq2->lock);   /* Fake it out ;) */
1700}
1701
1702/*
1703 * double_rq_unlock - safely unlock two runqueues
1704 *
1705 * Note this does not restore interrupts like task_rq_unlock,
1706 * you need to do so manually after calling.
1707 */
1708static void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1709        __releases(rq1->lock)
1710        __releases(rq2->lock)
1711{
1712        BUG_ON(rq1 != rq2);
1713        raw_spin_unlock(&rq1->lock);
1714        __release(rq2->lock);
1715}
1716
1717#endif
1718
1719static void calc_load_account_idle(struct rq *this_rq);
1720static void update_sysctl(void);
1721static int get_update_sysctl_factor(void);
1722static void update_cpu_load(struct rq *this_rq);
1723
1724static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1725{
1726        set_task_rq(p, cpu);
1727#ifdef CONFIG_SMP
1728        /*
1729         * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1730         * successfuly executed on another CPU. We must ensure that updates of
1731         * per-task data have been completed by this moment.
1732         */
1733        smp_wmb();
1734        task_thread_info(p)->cpu = cpu;
1735#endif
1736}
1737
1738static const struct sched_class rt_sched_class;
1739
1740#define sched_class_highest (&stop_sched_class)
1741#define for_each_class(class) \
1742   for (class = sched_class_highest; class; class = class->next)
1743
1744#include "sched_stats.h"
1745
1746static void inc_nr_running(struct rq *rq)
1747{
1748        rq->nr_running++;
1749}
1750
1751static void dec_nr_running(struct rq *rq)
1752{
1753        rq->nr_running--;
1754}
1755
1756static void set_load_weight(struct task_struct *p)
1757{
1758        /*
1759         * SCHED_IDLE tasks get minimal weight:
1760         */
1761        if (p->policy == SCHED_IDLE) {
1762                p->se.load.weight = WEIGHT_IDLEPRIO;
1763                p->se.load.inv_weight = WMULT_IDLEPRIO;
1764                return;
1765        }
1766
1767        p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO];
1768        p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO];
1769}
1770
1771static void enqueue_task(struct rq *rq, struct task_struct *p, int flags)
1772{
1773        update_rq_clock(rq);
1774        sched_info_queued(p);
1775        p->sched_class->enqueue_task(rq, p, flags);
1776        p->se.on_rq = 1;
1777}
1778
1779static void dequeue_task(struct rq *rq, struct task_struct *p, int flags)
1780{
1781        update_rq_clock(rq);
1782        sched_info_dequeued(p);
1783        p->sched_class->dequeue_task(rq, p, flags);
1784        p->se.on_rq = 0;
1785}
1786
1787/*
1788 * activate_task - move a task to the runqueue.
1789 */
1790static void activate_task(struct rq *rq, struct task_struct *p, int flags)
1791{
1792        if (task_contributes_to_load(p))
1793                rq->nr_uninterruptible--;
1794
1795        enqueue_task(rq, p, flags);
1796        inc_nr_running(rq);
1797}
1798
1799/*
1800 * deactivate_task - remove a task from the runqueue.
1801 */
1802static void deactivate_task(struct rq *rq, struct task_struct *p, int flags)
1803{
1804        if (task_contributes_to_load(p))
1805                rq->nr_uninterruptible++;
1806
1807        dequeue_task(rq, p, flags);
1808        dec_nr_running(rq);
1809}
1810
1811#ifdef CONFIG_IRQ_TIME_ACCOUNTING
1812
1813/*
1814 * There are no locks covering percpu hardirq/softirq time.
1815 * They are only modified in account_system_vtime, on corresponding CPU
1816 * with interrupts disabled. So, writes are safe.
1817 * They are read and saved off onto struct rq in update_rq_clock().
1818 * This may result in other CPU reading this CPU's irq time and can
1819 * race with irq/account_system_vtime on this CPU. We would either get old
1820 * or new value with a side effect of accounting a slice of irq time to wrong
1821 * task when irq is in progress while we read rq->clock. That is a worthy
1822 * compromise in place of having locks on each irq in account_system_time.
1823 */
1824static DEFINE_PER_CPU(u64, cpu_hardirq_time);
1825static DEFINE_PER_CPU(u64, cpu_softirq_time);
1826
1827static DEFINE_PER_CPU(u64, irq_start_time);
1828static int sched_clock_irqtime;
1829
1830void enable_sched_clock_irqtime(void)
1831{
1832        sched_clock_irqtime = 1;
1833}
1834
1835void disable_sched_clock_irqtime(void)
1836{
1837        sched_clock_irqtime = 0;
1838}
1839
1840#ifndef CONFIG_64BIT
1841static DEFINE_PER_CPU(seqcount_t, irq_time_seq);
1842
1843static inline void irq_time_write_begin(void)
1844{
1845        __this_cpu_inc(irq_time_seq.sequence);
1846        smp_wmb();
1847}
1848
1849static inline void irq_time_write_end(void)
1850{
1851        smp_wmb();
1852        __this_cpu_inc(irq_time_seq.sequence);
1853}
1854
1855static inline u64 irq_time_read(int cpu)
1856{
1857        u64 irq_time;
1858        unsigned seq;
1859
1860        do {
1861                seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu));
1862                irq_time = per_cpu(cpu_softirq_time, cpu) +
1863                           per_cpu(cpu_hardirq_time, cpu);
1864        } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq));
1865
1866        return irq_time;
1867}
1868#else /* CONFIG_64BIT */
1869static inline void irq_time_write_begin(void)
1870{
1871}
1872
1873static inline void irq_time_write_end(void)
1874{
1875}
1876
1877static inline u64 irq_time_read(int cpu)
1878{
1879        return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu);
1880}
1881#endif /* CONFIG_64BIT */
1882
1883/*
1884 * Called before incrementing preempt_count on {soft,}irq_enter
1885 * and before decrementing preempt_count on {soft,}irq_exit.
1886 */
1887void account_system_vtime(struct task_struct *curr)
1888{
1889        unsigned long flags;
1890        s64 delta;
1891        int cpu;
1892
1893        if (!sched_clock_irqtime)
1894                return;
1895
1896        local_irq_save(flags);
1897
1898        cpu = smp_processor_id();
1899        delta = sched_clock_cpu(cpu) - __this_cpu_read(irq_start_time);
1900        __this_cpu_add(irq_start_time, delta);
1901
1902        irq_time_write_begin();
1903        /*
1904         * We do not account for softirq time from ksoftirqd here.
1905         * We want to continue accounting softirq time to ksoftirqd thread
1906         * in that case, so as not to confuse scheduler with a special task
1907         * that do not consume any time, but still wants to run.
1908         */
1909        if (hardirq_count())
1910                __this_cpu_add(cpu_hardirq_time, delta);
1911        else if (in_serving_softirq() && curr != this_cpu_ksoftirqd())
1912                __this_cpu_add(cpu_softirq_time, delta);
1913
1914        irq_time_write_end();
1915        local_irq_restore(flags);
1916}
1917EXPORT_SYMBOL_GPL(account_system_vtime);
1918
1919static void update_rq_clock_task(struct rq *rq, s64 delta)
1920{
1921        s64 irq_delta;
1922
1923        irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time;
1924
1925        /*
1926         * Since irq_time is only updated on {soft,}irq_exit, we might run into
1927         * this case when a previous update_rq_clock() happened inside a
1928         * {soft,}irq region.
1929         *
1930         * When this happens, we stop ->clock_task and only update the
1931         * prev_irq_time stamp to account for the part that fit, so that a next
1932         * update will consume the rest. This ensures ->clock_task is
1933         * monotonic.
1934         *
1935         * It does however cause some slight miss-attribution of {soft,}irq
1936         * time, a more accurate solution would be to update the irq_time using
1937         * the current rq->clock timestamp, except that would require using
1938         * atomic ops.
1939         */
1940        if (irq_delta > delta)
1941                irq_delta = delta;
1942
1943        rq->prev_irq_time += irq_delta;
1944        delta -= irq_delta;
1945        rq->clock_task += delta;
1946
1947        if (irq_delta && sched_feat(NONIRQ_POWER))
1948                sched_rt_avg_update(rq, irq_delta);
1949}
1950
1951static int irqtime_account_hi_update(void)
1952{
1953        struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
1954        unsigned long flags;
1955        u64 latest_ns;
1956        int ret = 0;
1957
1958        local_irq_save(flags);
1959        latest_ns = this_cpu_read(cpu_hardirq_time);
1960        if (cputime64_gt(nsecs_to_cputime64(latest_ns), cpustat->irq))
1961                ret = 1;
1962        local_irq_restore(flags);
1963        return ret;
1964}
1965
1966static int irqtime_account_si_update(void)
1967{
1968        struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
1969        unsigned long flags;
1970        u64 latest_ns;
1971        int ret = 0;
1972
1973        local_irq_save(flags);
1974        latest_ns = this_cpu_read(cpu_softirq_time);
1975        if (cputime64_gt(nsecs_to_cputime64(latest_ns), cpustat->softirq))
1976                ret = 1;
1977        local_irq_restore(flags);
1978        return ret;
1979}
1980
1981#else /* CONFIG_IRQ_TIME_ACCOUNTING */
1982
1983#define sched_clock_irqtime     (0)
1984
1985static void update_rq_clock_task(struct rq *rq, s64 delta)
1986{
1987        rq->clock_task += delta;
1988}
1989
1990#endif /* CONFIG_IRQ_TIME_ACCOUNTING */
1991
1992#include "sched_idletask.c"
1993#include "sched_fair.c"
1994#include "sched_rt.c"
1995#include "sched_autogroup.c"
1996#include "sched_stoptask.c"
1997#ifdef CONFIG_SCHED_DEBUG
1998# include "sched_debug.c"
1999#endif
2000
2001void sched_set_stop_task(int cpu, struct task_struct *stop)
2002{
2003        struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 };
2004        struct task_struct *old_stop = cpu_rq(cpu)->stop;
2005
2006        if (stop) {
2007                /*
2008                 * Make it appear like a SCHED_FIFO task, its something
2009                 * userspace knows about and won't get confused about.
2010                 *
2011                 * Also, it will make PI more or less work without too
2012                 * much confusion -- but then, stop work should not
2013                 * rely on PI working anyway.
2014                 */
2015                sched_setscheduler_nocheck(stop, SCHED_FIFO, &param);
2016
2017                stop->sched_class = &stop_sched_class;
2018        }
2019
2020        cpu_rq(cpu)->stop = stop;
2021
2022        if (old_stop) {
2023                /*
2024                 * Reset it back to a normal scheduling class so that
2025                 * it can die in pieces.
2026                 */
2027                old_stop->sched_class = &rt_sched_class;
2028        }
2029}
2030
2031/*
2032 * __normal_prio - return the priority that is based on the static prio
2033 */
2034static inline int __normal_prio(struct task_struct *p)
2035{
2036        return p->static_prio;
2037}
2038
2039/*
2040 * Calculate the expected normal priority: i.e. priority
2041 * without taking RT-inheritance into account. Might be
2042 * boosted by interactivity modifiers. Changes upon fork,
2043 * setprio syscalls, and whenever the interactivity
2044 * estimator recalculates.
2045 */
2046static inline int normal_prio(struct task_struct *p)
2047{
2048        int prio;
2049
2050        if (task_has_rt_policy(p))
2051                prio = MAX_RT_PRIO-1 - p->rt_priority;
2052        else
2053                prio = __normal_prio(p);
2054        return prio;
2055}
2056
2057/*
2058 * Calculate the current priority, i.e. the priority
2059 * taken into account by the scheduler. This value might
2060 * be boosted by RT tasks, or might be boosted by
2061 * interactivity modifiers. Will be RT if the task got
2062 * RT-boosted. If not then it returns p->normal_prio.
2063 */
2064static int effective_prio(struct task_struct *p)
2065{
2066        p->normal_prio = normal_prio(p);
2067        /*
2068         * If we are RT tasks or we were boosted to RT priority,
2069         * keep the priority unchanged. Otherwise, update priority
2070         * to the normal priority:
2071         */
2072        if (!rt_prio(p->prio))
2073                return p->normal_prio;
2074        return p->prio;
2075}
2076
2077/**
2078 * task_curr - is this task currently executing on a CPU?
2079 * @p: the task in question.
2080 */
2081inline int task_curr(const struct task_struct *p)
2082{
2083        return cpu_curr(task_cpu(p)) == p;
2084}
2085
2086static inline void check_class_changed(struct rq *rq, struct task_struct *p,
2087                                       const struct sched_class *prev_class,
2088                                       int oldprio)
2089{
2090        if (prev_class != p->sched_class) {
2091                if (prev_class->switched_from)
2092                        prev_class->switched_from(rq, p);
2093                p->sched_class->switched_to(rq, p);
2094        } else if (oldprio != p->prio)
2095                p->sched_class->prio_changed(rq, p, oldprio);
2096}
2097
2098static void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags)
2099{
2100        const struct sched_class *class;
2101
2102        if (p->sched_class == rq->curr->sched_class) {
2103                rq->curr->sched_class->check_preempt_curr(rq, p, flags);
2104        } else {
2105                for_each_class(class) {
2106                        if (class == rq->curr->sched_class)
2107                                break;
2108                        if (class == p->sched_class) {
2109                                resched_task(rq->curr);
2110                                break;
2111                        }
2112                }
2113        }
2114
2115        /*
2116         * A queue event has occurred, and we're going to schedule.  In
2117         * this case, we can save a useless back to back clock update.
2118         */
2119        if (rq->curr->se.on_rq && test_tsk_need_resched(rq->curr))
2120                rq->skip_clock_update = 1;
2121}
2122
2123#ifdef CONFIG_SMP
2124/*
2125 * Is this task likely cache-hot:
2126 */
2127static int
2128task_hot(struct task_struct *p, u64 now, struct sched_domain *sd)
2129{
2130        s64 delta;
2131
2132        if (p->sched_class != &fair_sched_class)
2133                return 0;
2134
2135        if (unlikely(p->policy == SCHED_IDLE))
2136                return 0;
2137
2138        /*
2139         * Buddy candidates are cache hot:
2140         */
2141        if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running &&
2142                        (&p->se == cfs_rq_of(&p->se)->next ||
2143                         &p->se == cfs_rq_of(&p->se)->last))
2144                return 1;
2145
2146        if (sysctl_sched_migration_cost == -1)
2147                return 1;
2148        if (sysctl_sched_migration_cost == 0)
2149                return 0;
2150
2151        delta = now - p->se.exec_start;
2152
2153        return delta < (s64)sysctl_sched_migration_cost;
2154}
2155
2156void set_task_cpu(struct task_struct *p, unsigned int new_cpu)
2157{
2158#ifdef CONFIG_SCHED_DEBUG
2159        /*
2160         * We should never call set_task_cpu() on a blocked task,
2161         * ttwu() will sort out the placement.
2162         */
2163        WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING &&
2164                        !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE));
2165#endif
2166
2167        trace_sched_migrate_task(p, new_cpu);
2168
2169        if (task_cpu(p) != new_cpu) {
2170                p->se.nr_migrations++;
2171                perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 1, NULL, 0);
2172        }
2173
2174        __set_task_cpu(p, new_cpu);
2175}
2176
2177struct migration_arg {
2178        struct task_struct *task;
2179        int dest_cpu;
2180};
2181
2182static int migration_cpu_stop(void *data);
2183
2184/*
2185 * The task's runqueue lock must be held.
2186 * Returns true if you have to wait for migration thread.
2187 */
2188static bool migrate_task(struct task_struct *p, struct rq *rq)
2189{
2190        /*
2191         * If the task is not on a runqueue (and not running), then
2192         * the next wake-up will properly place the task.
2193         */
2194        return p->se.on_rq || task_running(rq, p);
2195}
2196
2197/*
2198 * wait_task_inactive - wait for a thread to unschedule.
2199 *
2200 * If @match_state is nonzero, it's the @p->state value just checked and
2201 * not expected to change.  If it changes, i.e. @p might have woken up,
2202 * then return zero.  When we succeed in waiting for @p to be off its CPU,
2203 * we return a positive number (its total switch count).  If a second call
2204 * a short while later returns the same number, the caller can be sure that
2205 * @p has remained unscheduled the whole time.
2206 *
2207 * The caller must ensure that the task *will* unschedule sometime soon,
2208 * else this function might spin for a *long* time. This function can't
2209 * be called with interrupts off, or it may introduce deadlock with
2210 * smp_call_function() if an IPI is sent by the same process we are
2211 * waiting to become inactive.
2212 */
2213unsigned long wait_task_inactive(struct task_struct *p, long match_state)
2214{
2215        unsigned long flags;
2216        int running, on_rq;
2217        unsigned long ncsw;
2218        struct rq *rq;
2219
2220        for (;;) {
2221                /*
2222                 * We do the initial early heuristics without holding
2223                 * any task-queue locks at all. We'll only try to get
2224                 * the runqueue lock when things look like they will
2225                 * work out!
2226                 */
2227                rq = task_rq(p);
2228
2229                /*
2230                 * If the task is actively running on another CPU
2231                 * still, just relax and busy-wait without holding
2232                 * any locks.
2233                 *
2234                 * NOTE! Since we don't hold any locks, it's not
2235                 * even sure that "rq" stays as the right runqueue!
2236                 * But we don't care, since "task_running()" will
2237                 * return false if the runqueue has changed and p
2238                 * is actually now running somewhere else!
2239                 */
2240                while (task_running(rq, p)) {
2241                        if (match_state && unlikely(p->state != match_state))
2242                                return 0;
2243                        cpu_relax();
2244                }
2245
2246                /*
2247                 * Ok, time to look more closely! We need the rq
2248                 * lock now, to be *sure*. If we're wrong, we'll
2249                 * just go back and repeat.
2250                 */
2251                rq = task_rq_lock(p, &flags);
2252                trace_sched_wait_task(p);
2253                running = task_running(rq, p);
2254                on_rq = p->se.on_rq;
2255                ncsw = 0;
2256                if (!match_state || p->state == match_state)
2257                        ncsw = p->nvcsw | LONG_MIN; /* sets MSB */
2258                task_rq_unlock(rq, &flags);
2259
2260                /*
2261                 * If it changed from the expected state, bail out now.
2262                 */
2263                if (unlikely(!ncsw))
2264                        break;
2265
2266                /*
2267                 * Was it really running after all now that we
2268                 * checked with the proper locks actually held?
2269                 *
2270                 * Oops. Go back and try again..
2271                 */
2272                if (unlikely(running)) {
2273                        cpu_relax();
2274                        continue;
2275                }
2276
2277                /*
2278                 * It's not enough that it's not actively running,
2279                 * it must be off the runqueue _entirely_, and not
2280                 * preempted!
2281                 *
2282                 * So if it was still runnable (but just not actively
2283                 * running right now), it's preempted, and we should
2284                 * yield - it could be a while.
2285                 */
2286                if (unlikely(on_rq)) {
2287                        ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ);
2288
2289                        set_current_state(TASK_UNINTERRUPTIBLE);
2290                        schedule_hrtimeout(&to, HRTIMER_MODE_REL);
2291                        continue;
2292                }
2293
2294                /*
2295                 * Ahh, all good. It wasn't running, and it wasn't
2296                 * runnable, which means that it will never become
2297                 * running in the future either. We're all done!
2298                 */
2299                break;
2300        }
2301
2302        return ncsw;
2303}
2304
2305/***
2306 * kick_process - kick a running thread to enter/exit the kernel
2307 * @p: the to-be-kicked thread
2308 *
2309 * Cause a process which is running on another CPU to enter
2310 * kernel-mode, without any delay. (to get signals handled.)
2311 *
2312 * NOTE: this function doesn't have to take the runqueue lock,
2313 * because all it wants to ensure is that the remote task enters
2314 * the kernel. If the IPI races and the task has been migrated
2315 * to another CPU then no harm is done and the purpose has been
2316 * achieved as well.
2317 */
2318void kick_process(struct task_struct *p)
2319{
2320        int cpu;
2321
2322        preempt_disable();
2323        cpu = task_cpu(p);
2324        if ((cpu != smp_processor_id()) && task_curr(p))
2325                smp_send_reschedule(cpu);
2326        preempt_enable();
2327}
2328EXPORT_SYMBOL_GPL(kick_process);
2329#endif /* CONFIG_SMP */
2330
2331#ifdef CONFIG_SMP
2332/*
2333 * ->cpus_allowed is protected by either TASK_WAKING or rq->lock held.
2334 */
2335static int select_fallback_rq(int cpu, struct task_struct *p)
2336{
2337        int dest_cpu;
2338        const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu));
2339
2340        /* Look for allowed, online CPU in same node. */
2341        for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask)
2342                if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed))
2343                        return dest_cpu;
2344
2345        /* Any allowed, online CPU? */
2346        dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_active_mask);
2347        if (dest_cpu < nr_cpu_ids)
2348                return dest_cpu;
2349
2350        /* No more Mr. Nice Guy. */
2351        dest_cpu = cpuset_cpus_allowed_fallback(p);
2352        /*
2353         * Don't tell them about moving exiting tasks or
2354         * kernel threads (both mm NULL), since they never
2355         * leave kernel.
2356         */
2357        if (p->mm && printk_ratelimit()) {
2358                printk(KERN_INFO "process %d (%s) no longer affine to cpu%d\n",
2359                                task_pid_nr(p), p->comm, cpu);
2360        }
2361
2362        return dest_cpu;
2363}
2364
2365/*
2366 * The caller (fork, wakeup) owns TASK_WAKING, ->cpus_allowed is stable.
2367 */
2368static inline
2369int select_task_rq(struct rq *rq, struct task_struct *p, int sd_flags, int wake_flags)
2370{
2371        int cpu = p->sched_class->select_task_rq(rq, p, sd_flags, wake_flags);
2372
2373        /*
2374         * In order not to call set_task_cpu() on a blocking task we need
2375         * to rely on ttwu() to place the task on a valid ->cpus_allowed
2376         * cpu.
2377         *
2378         * Since this is common to all placement strategies, this lives here.
2379         *
2380         * [ this allows ->select_task() to simply return task_cpu(p) and
2381         *   not worry about this generic constraint ]
2382         */
2383        if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) ||
2384                     !cpu_online(cpu)))
2385                cpu = select_fallback_rq(task_cpu(p), p);
2386
2387        return cpu;
2388}
2389
2390static void update_avg(u64 *avg, u64 sample)
2391{
2392        s64 diff = sample - *avg;
2393        *avg += diff >> 3;
2394}
2395#endif
2396
2397static inline void ttwu_activate(struct task_struct *p, struct rq *rq,
2398                                 bool is_sync, bool is_migrate, bool is_local,
2399                                 unsigned long en_flags)
2400{
2401        schedstat_inc(p, se.statistics.nr_wakeups);
2402        if (is_sync)
2403                schedstat_inc(p, se.statistics.nr_wakeups_sync);
2404        if (is_migrate)
2405                schedstat_inc(p, se.statistics.nr_wakeups_migrate);
2406        if (is_local)
2407                schedstat_inc(p, se.statistics.nr_wakeups_local);
2408        else
2409                schedstat_inc(p, se.statistics.nr_wakeups_remote);
2410
2411        activate_task(rq, p, en_flags);
2412}
2413
2414static inline void ttwu_post_activation(struct task_struct *p, struct rq *rq,
2415                                        int wake_flags, bool success)
2416{
2417        trace_sched_wakeup(p, success);
2418        check_preempt_curr(rq, p, wake_flags);
2419
2420        p->state = TASK_RUNNING;
2421#ifdef CONFIG_SMP
2422        if (p->sched_class->task_woken)
2423                p->sched_class->task_woken(rq, p);
2424
2425        if (unlikely(rq->idle_stamp)) {
2426                u64 delta = rq->clock - rq->idle_stamp;
2427                u64 max = 2*sysctl_sched_migration_cost;
2428
2429                if (delta > max)
2430                        rq->avg_idle = max;
2431                else
2432                        update_avg(&rq->avg_idle, delta);
2433                rq->idle_stamp = 0;
2434        }
2435#endif
2436        /* if a worker is waking up, notify workqueue */
2437        if ((p->flags & PF_WQ_WORKER) && success)
2438                wq_worker_waking_up(p, cpu_of(rq));
2439}
2440
2441/**
2442 * try_to_wake_up - wake up a thread
2443 * @p: the thread to be awakened
2444 * @state: the mask of task states that can be woken
2445 * @wake_flags: wake modifier flags (WF_*)
2446 *
2447 * Put it on the run-queue if it's not already there. The "current"
2448 * thread is always on the run-queue (except when the actual
2449 * re-schedule is in progress), and as such you're allowed to do
2450 * the simpler "current->state = TASK_RUNNING" to mark yourself
2451 * runnable without the overhead of this.
2452 *
2453 * Returns %true if @p was woken up, %false if it was already running
2454 * or @state didn't match @p's state.
2455 */
2456static int try_to_wake_up(struct task_struct *p, unsigned int state,
2457                          int wake_flags)
2458{
2459        int cpu, orig_cpu, this_cpu, success = 0;
2460        unsigned long flags;
2461        unsigned long en_flags = ENQUEUE_WAKEUP;
2462        struct rq *rq;
2463
2464        this_cpu = get_cpu();
2465
2466        smp_wmb();
2467        rq = task_rq_lock(p, &flags);
2468        if (!(p->state & state))
2469                goto out;
2470
2471        if (p->se.on_rq)
2472                goto out_running;
2473
2474        cpu = task_cpu(p);
2475        orig_cpu = cpu;
2476
2477#ifdef CONFIG_SMP
2478        if (unlikely(task_running(rq, p)))
2479                goto out_activate;
2480
2481        /*
2482         * In order to handle concurrent wakeups and release the rq->lock
2483         * we put the task in TASK_WAKING state.
2484         *
2485         * First fix up the nr_uninterruptible count:
2486         */
2487        if (task_contributes_to_load(p)) {
2488                if (likely(cpu_online(orig_cpu)))
2489                        rq->nr_uninterruptible--;
2490                else
2491                        this_rq()->nr_uninterruptible--;
2492        }
2493        p->state = TASK_WAKING;
2494
2495        if (p->sched_class->task_waking) {
2496                p->sched_class->task_waking(rq, p);
2497                en_flags |= ENQUEUE_WAKING;
2498        }
2499
2500        cpu = select_task_rq(rq, p, SD_BALANCE_WAKE, wake_flags);
2501        if (cpu != orig_cpu)
2502                set_task_cpu(p, cpu);
2503        __task_rq_unlock(rq);
2504
2505        rq = cpu_rq(cpu);
2506        raw_spin_lock(&rq->lock);
2507
2508        /*
2509         * We migrated the task without holding either rq->lock, however
2510         * since the task is not on the task list itself, nobody else
2511         * will try and migrate the task, hence the rq should match the
2512         * cpu we just moved it to.
2513         */
2514        WARN_ON(task_cpu(p) != cpu);
2515        WARN_ON(p->state != TASK_WAKING);
2516
2517#ifdef CONFIG_SCHEDSTATS
2518        schedstat_inc(rq, ttwu_count);
2519        if (cpu == this_cpu)
2520                schedstat_inc(rq, ttwu_local);
2521        else {
2522                struct sched_domain *sd;
2523                for_each_domain(this_cpu, sd) {
2524                        if (cpumask_test_cpu(cpu, sched_domain_span(sd))) {
2525                                schedstat_inc(sd, ttwu_wake_remote);
2526                                break;
2527                        }
2528                }
2529        }
2530#endif /* CONFIG_SCHEDSTATS */
2531
2532out_activate:
2533#endif /* CONFIG_SMP */
2534        ttwu_activate(p, rq, wake_flags & WF_SYNC, orig_cpu != cpu,
2535                      cpu == this_cpu, en_flags);
2536        success = 1;
2537out_running:
2538        ttwu_post_activation(p, rq, wake_flags, success);
2539out:
2540        task_rq_unlock(rq, &flags);
2541        put_cpu();
2542
2543        return success;
2544}
2545
2546/**
2547 * try_to_wake_up_local - try to wake up a local task with rq lock held
2548 * @p: the thread to be awakened
2549 *
2550 * Put @p on the run-queue if it's not already there.  The caller must
2551 * ensure that this_rq() is locked, @p is bound to this_rq() and not
2552 * the current task.  this_rq() stays locked over invocation.
2553 */
2554static void try_to_wake_up_local(struct task_struct *p)
2555{
2556        struct rq *rq = task_rq(p);
2557        bool success = false;
2558
2559        BUG_ON(rq != this_rq());
2560        BUG_ON(p == current);
2561        lockdep_assert_held(&rq->lock);
2562
2563        if (!(p->state & TASK_NORMAL))
2564                return;
2565
2566        if (!p->se.on_rq) {
2567                if (likely(!task_running(rq, p))) {
2568                        schedstat_inc(rq, ttwu_count);
2569                        schedstat_inc(rq, ttwu_local);
2570                }
2571                ttwu_activate(p, rq, false, false, true, ENQUEUE_WAKEUP);
2572                success = true;
2573        }
2574        ttwu_post_activation(p, rq, 0, success);
2575}
2576
2577/**
2578 * wake_up_process - Wake up a specific process
2579 * @p: The process to be woken up.
2580 *
2581 * Attempt to wake up the nominated process and move it to the set of runnable
2582 * processes.  Returns 1 if the process was woken up, 0 if it was already
2583 * running.
2584 *
2585 * It may be assumed that this function implies a write memory barrier before
2586 * changing the task state if and only if any tasks are woken up.
2587 */
2588int wake_up_process(struct task_struct *p)
2589{
2590        return try_to_wake_up(p, TASK_ALL, 0);
2591}
2592EXPORT_SYMBOL(wake_up_process);
2593
2594int wake_up_state(struct task_struct *p, unsigned int state)
2595{
2596        return try_to_wake_up(p, state, 0);
2597}
2598
2599/*
2600 * Perform scheduler related setup for a newly forked process p.
2601 * p is forked by current.
2602 *
2603 * __sched_fork() is basic setup used by init_idle() too:
2604 */
2605static void __sched_fork(struct task_struct *p)
2606{
2607        p->se.exec_start                = 0;
2608        p->se.sum_exec_runtime          = 0;
2609        p->se.prev_sum_exec_runtime     = 0;
2610        p->se.nr_migrations             = 0;
2611        p->se.vruntime                  = 0;
2612
2613#ifdef CONFIG_SCHEDSTATS
2614        memset(&p->se.statistics, 0, sizeof(p->se.statistics));
2615#endif
2616
2617        INIT_LIST_HEAD(&p->rt.run_list);
2618        p->se.on_rq = 0;
2619        INIT_LIST_HEAD(&p->se.group_node);
2620
2621#ifdef CONFIG_PREEMPT_NOTIFIERS
2622        INIT_HLIST_HEAD(&p->preempt_notifiers);
2623#endif
2624}
2625
2626/*
2627 * fork()/clone()-time setup:
2628 */
2629void sched_fork(struct task_struct *p, int clone_flags)
2630{
2631        int cpu = get_cpu();
2632
2633        __sched_fork(p);
2634        /*
2635         * We mark the process as running here. This guarantees that
2636         * nobody will actually run it, and a signal or other external
2637         * event cannot wake it up and insert it on the runqueue either.
2638         */
2639        p->state = TASK_RUNNING;
2640
2641        /*
2642         * Revert to default priority/policy on fork if requested.
2643         */
2644        if (unlikely(p->sched_reset_on_fork)) {
2645                if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) {
2646                        p->policy = SCHED_NORMAL;
2647                        p->normal_prio = p->static_prio;
2648                }
2649
2650                if (PRIO_TO_NICE(p->static_prio) < 0) {
2651                        p->static_prio = NICE_TO_PRIO(0);
2652                        p->normal_prio = p->static_prio;
2653                        set_load_weight(p);
2654                }
2655
2656                /*
2657                 * We don't need the reset flag anymore after the fork. It has
2658                 * fulfilled its duty:
2659                 */
2660                p->sched_reset_on_fork = 0;
2661        }
2662
2663        /*
2664         * Make sure we do not leak PI boosting priority to the child.
2665         */
2666        p->prio = current->normal_prio;
2667
2668        if (!rt_prio(p->prio))
2669                p->sched_class = &fair_sched_class;
2670
2671        if (p->sched_class->task_fork)
2672                p->sched_class->task_fork(p);
2673
2674        /*
2675         * The child is not yet in the pid-hash so no cgroup attach races,
2676         * and the cgroup is pinned to this child due to cgroup_fork()
2677         * is ran before sched_fork().
2678         *
2679         * Silence PROVE_RCU.
2680         */
2681        rcu_read_lock();
2682        set_task_cpu(p, cpu);
2683        rcu_read_unlock();
2684
2685#if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT)
2686        if (likely(sched_info_on()))
2687                memset(&p->sched_info, 0, sizeof(p->sched_info));
2688#endif
2689#if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW)
2690        p->oncpu = 0;
2691#endif
2692#ifdef CONFIG_PREEMPT
2693        /* Want to start with kernel preemption disabled. */
2694        task_thread_info(p)->preempt_count = 1;
2695#endif
2696#ifdef CONFIG_SMP
2697        plist_node_init(&p->pushable_tasks, MAX_PRIO);
2698#endif
2699
2700        put_cpu();
2701}
2702
2703/*
2704 * wake_up_new_task - wake up a newly created task for the first time.
2705 *
2706 * This function will do some initial scheduler statistics housekeeping
2707 * that must be done for every newly created context, then puts the task
2708 * on the runqueue and wakes it.
2709 */
2710void wake_up_new_task(struct task_struct *p, unsigned long clone_flags)
2711{
2712        unsigned long flags;
2713        struct rq *rq;
2714        int cpu __maybe_unused = get_cpu();
2715
2716#ifdef CONFIG_SMP
2717        rq = task_rq_lock(p, &flags);
2718        p->state = TASK_WAKING;
2719
2720        /*
2721         * Fork balancing, do it here and not earlier because:
2722         *  - cpus_allowed can change in the fork path
2723         *  - any previously selected cpu might disappear through hotplug
2724         *
2725         * We set TASK_WAKING so that select_task_rq() can drop rq->lock
2726         * without people poking at ->cpus_allowed.
2727         */
2728        cpu = select_task_rq(rq, p, SD_BALANCE_FORK, 0);
2729        set_task_cpu(p, cpu);
2730
2731        p->state = TASK_RUNNING;
2732        task_rq_unlock(rq, &flags);
2733#endif
2734
2735        rq = task_rq_lock(p, &flags);
2736        activate_task(rq, p, 0);
2737        trace_sched_wakeup_new(p, 1);
2738        check_preempt_curr(rq, p, WF_FORK);
2739#ifdef CONFIG_SMP
2740        if (p->sched_class->task_woken)
2741                p->sched_class->task_woken(rq, p);
2742#endif
2743        task_rq_unlock(rq, &flags);
2744        put_cpu();
2745}
2746
2747#ifdef CONFIG_PREEMPT_NOTIFIERS
2748
2749/**
2750 * preempt_notifier_register - tell me when current is being preempted & rescheduled
2751 * @notifier: notifier struct to register
2752 */
2753void preempt_notifier_register(struct preempt_notifier *notifier)
2754{
2755        hlist_add_head(&notifier->link, &current->preempt_notifiers);
2756}
2757EXPORT_SYMBOL_GPL(preempt_notifier_register);
2758
2759/**
2760 * preempt_notifier_unregister - no longer interested in preemption notifications
2761 * @notifier: notifier struct to unregister
2762 *
2763 * This is safe to call from within a preemption notifier.
2764 */
2765void preempt_notifier_unregister(struct preempt_notifier *notifier)
2766{
2767        hlist_del(&notifier->link);
2768}
2769EXPORT_SYMBOL_GPL(preempt_notifier_unregister);
2770
2771static void fire_sched_in_preempt_notifiers(struct task_struct *curr)
2772{
2773        struct preempt_notifier *notifier;
2774        struct hlist_node *node;
2775
2776        hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link)
2777                notifier->ops->sched_in(notifier, raw_smp_processor_id());
2778}
2779
2780static void
2781fire_sched_out_preempt_notifiers(struct task_struct *curr,
2782                                 struct task_struct *next)
2783{
2784        struct preempt_notifier *notifier;
2785        struct hlist_node *node;
2786
2787        hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link)
2788                notifier->ops->sched_out(notifier, next);
2789}
2790
2791#else /* !CONFIG_PREEMPT_NOTIFIERS */
2792
2793static void fire_sched_in_preempt_notifiers(struct task_struct *curr)
2794{
2795}
2796
2797static void
2798fire_sched_out_preempt_notifiers(struct task_struct *curr,
2799                                 struct task_struct *next)
2800{
2801}
2802
2803#endif /* CONFIG_PREEMPT_NOTIFIERS */
2804
2805/**
2806 * prepare_task_switch - prepare to switch tasks
2807 * @rq: the runqueue preparing to switch
2808 * @prev: the current task that is being switched out
2809 * @next: the task we are going to switch to.
2810 *
2811 * This is called with the rq lock held and interrupts off. It must
2812 * be paired with a subsequent finish_task_switch after the context
2813 * switch.
2814 *
2815 * prepare_task_switch sets up locking and calls architecture specific
2816 * hooks.
2817 */
2818static inline void
2819prepare_task_switch(struct rq *rq, struct task_struct *prev,
2820                    struct task_struct *next)
2821{
2822        sched_info_switch(prev, next);
2823        perf_event_task_sched_out(prev, next);
2824        fire_sched_out_preempt_notifiers(prev, next);
2825        prepare_lock_switch(rq, next);
2826        prepare_arch_switch(next);
2827        trace_sched_switch(prev, next);
2828}
2829
2830/**
2831 * finish_task_switch - clean up after a task-switch
2832 * @rq: runqueue associated with task-switch
2833 * @prev: the thread we just switched away from.
2834 *
2835 * finish_task_switch must be called after the context switch, paired
2836 * with a prepare_task_switch call before the context switch.
2837 * finish_task_switch will reconcile locking set up by prepare_task_switch,
2838 * and do any other architecture-specific cleanup actions.
2839 *
2840 * Note that we may have delayed dropping an mm in context_switch(). If
2841 * so, we finish that here outside of the runqueue lock. (Doing it
2842 * with the lock held can cause deadlocks; see schedule() for
2843 * details.)
2844 */
2845static void finish_task_switch(struct rq *rq, struct task_struct *prev)
2846        __releases(rq->lock)
2847{
2848        struct mm_struct *mm = rq->prev_mm;
2849        long prev_state;
2850
2851        rq->prev_mm = NULL;
2852
2853        /*
2854         * A task struct has one reference for the use as "current".
2855         * If a task dies, then it sets TASK_DEAD in tsk->state and calls
2856         * schedule one last time. The schedule call will never return, and
2857         * the scheduled task must drop that reference.
2858         * The test for TASK_DEAD must occur while the runqueue locks are
2859         * still held, otherwise prev could be scheduled on another cpu, die
2860         * there before we look at prev->state, and then the reference would
2861         * be dropped twice.
2862         *              Manfred Spraul <manfred@colorfullife.com>
2863         */
2864        prev_state = prev->state;
2865        finish_arch_switch(prev);
2866#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
2867        local_irq_disable();
2868#endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */
2869        perf_event_task_sched_in(current);
2870#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
2871        local_irq_enable();
2872#endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */
2873        finish_lock_switch(rq, prev);
2874
2875        fire_sched_in_preempt_notifiers(current);
2876        if (mm)
2877                mmdrop(mm);
2878        if (unlikely(prev_state == TASK_DEAD)) {
2879                /*
2880                 * Remove function-return probe instances associated with this
2881                 * task and put them back on the free list.
2882                 */
2883                kprobe_flush_task(prev);
2884                put_task_struct(prev);
2885        }
2886}
2887
2888#ifdef CONFIG_SMP
2889
2890/* assumes rq->lock is held */
2891static inline void pre_schedule(struct rq *rq, struct task_struct *prev)
2892{
2893        if (prev->sched_class->pre_schedule)
2894                prev->sched_class->pre_schedule(rq, prev);
2895}
2896
2897/* rq->lock is NOT held, but preemption is disabled */
2898static inline void post_schedule(struct rq *rq)
2899{
2900        if (rq->post_schedule) {
2901                unsigned long flags;
2902
2903                raw_spin_lock_irqsave(&rq->lock, flags);
2904                if (rq->curr->sched_class->post_schedule)
2905                        rq->curr->sched_class->post_schedule(rq);
2906                raw_spin_unlock_irqrestore(&rq->lock, flags);
2907
2908                rq->post_schedule = 0;
2909        }
2910}
2911
2912#else
2913
2914static inline void pre_schedule(struct rq *rq, struct task_struct *p)
2915{
2916}
2917
2918static inline void post_schedule(struct rq *rq)
2919{
2920}
2921
2922#endif
2923
2924/**
2925 * schedule_tail - first thing a freshly forked thread must call.
2926 * @prev: the thread we just switched away from.
2927 */
2928asmlinkage void schedule_tail(struct task_struct *prev)
2929        __releases(rq->lock)
2930{
2931        struct rq *rq = this_rq();
2932
2933        finish_task_switch(rq, prev);
2934
2935        /*
2936         * FIXME: do we need to worry about rq being invalidated by the
2937         * task_switch?
2938         */
2939        post_schedule(rq);
2940
2941#ifdef __ARCH_WANT_UNLOCKED_CTXSW
2942        /* In this case, finish_task_switch does not reenable preemption */
2943        preempt_enable();
2944#endif
2945        if (current->set_child_tid)
2946                put_user(task_pid_vnr(current), current->set_child_tid);
2947}
2948
2949/*
2950 * context_switch - switch to the new MM and the new
2951 * thread's register state.
2952 */
2953static inline void
2954context_switch(struct rq *rq, struct task_struct *prev,
2955               struct task_struct *next)
2956{
2957        struct mm_struct *mm, *oldmm;
2958
2959        prepare_task_switch(rq, prev, next);
2960
2961        mm = next->mm;
2962        oldmm = prev->active_mm;
2963        /*
2964         * For paravirt, this is coupled with an exit in switch_to to
2965         * combine the page table reload and the switch backend into
2966         * one hypercall.
2967         */
2968        arch_start_context_switch(prev);
2969
2970        if (!mm) {
2971                next->active_mm = oldmm;
2972                atomic_inc(&oldmm->mm_count);
2973                enter_lazy_tlb(oldmm, next);
2974        } else
2975                switch_mm(oldmm, mm, next);
2976
2977        if (!prev->mm) {
2978                prev->active_mm = NULL;
2979                rq->prev_mm = oldmm;
2980        }
2981        /*
2982         * Since the runqueue lock will be released by the next
2983         * task (which is an invalid locking op but in the case
2984         * of the scheduler it's an obvious special-case), so we
2985         * do an early lockdep release here:
2986         */
2987#ifndef __ARCH_WANT_UNLOCKED_CTXSW
2988        spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
2989#endif
2990
2991        /* Here we just switch the register state and the stack. */
2992        switch_to(prev, next, prev);
2993
2994        barrier();
2995        /*
2996         * this_rq must be evaluated again because prev may have moved
2997         * CPUs since it called schedule(), thus the 'rq' on its stack
2998         * frame will be invalid.
2999         */
3000        finish_task_switch(this_rq(), prev);
3001}
3002
3003/*
3004 * nr_running, nr_uninterruptible and nr_context_switches:
3005 *
3006 * externally visible scheduler statistics: current number of runnable
3007 * threads, current number of uninterruptible-sleeping threads, total
3008 * number of context switches performed since bootup.
3009 */
3010unsigned long nr_running(void)
3011{
3012        unsigned long i, sum = 0;
3013
3014        for_each_online_cpu(i)
3015                sum += cpu_rq(i)->nr_running;
3016
3017        return sum;
3018}
3019
3020unsigned long nr_uninterruptible(void)
3021{
3022        unsigned long i, sum = 0;
3023
3024        for_each_possible_cpu(i)
3025                sum += cpu_rq(i)->nr_uninterruptible;
3026
3027        /*
3028         * Since we read the counters lockless, it might be slightly
3029         * inaccurate. Do not allow it to go below zero though:
3030         */
3031        if (unlikely((long)sum < 0))
3032                sum = 0;
3033
3034        return sum;
3035}
3036
3037unsigned long long nr_context_switches(void)
3038{
3039        int i;
3040        unsigned long long sum = 0;
3041
3042        for_each_possible_cpu(i)
3043                sum += cpu_rq(i)->nr_switches;
3044
3045        return sum;
3046}
3047
3048unsigned long nr_iowait(void)
3049{
3050        unsigned long i, sum = 0;
3051
3052        for_each_possible_cpu(i)
3053                sum += atomic_read(&cpu_rq(i)->nr_iowait);
3054
3055        return sum;
3056}
3057
3058unsigned long nr_iowait_cpu(int cpu)
3059{
3060        struct rq *this = cpu_rq(cpu);
3061        return atomic_read(&this->nr_iowait);
3062}
3063
3064unsigned long this_cpu_load(void)
3065{
3066        struct rq *this = this_rq();
3067        return this->cpu_load[0];
3068}
3069
3070
3071/* Variables and functions for calc_load */
3072static atomic_long_t calc_load_tasks;
3073static unsigned long calc_load_update;
3074unsigned long avenrun[3];
3075EXPORT_SYMBOL(avenrun);
3076
3077static long calc_load_fold_active(struct rq *this_rq)
3078{
3079        long nr_active, delta = 0;
3080
3081        nr_active = this_rq->nr_running;
3082        nr_active += (long) this_rq->nr_uninterruptible;
3083
3084        if (nr_active != this_rq->calc_load_active) {
3085                delta = nr_active - this_rq->calc_load_active;
3086                this_rq->calc_load_active = nr_active;
3087        }
3088
3089        return delta;
3090}
3091
3092static unsigned long
3093calc_load(unsigned long load, unsigned long exp, unsigned long active)
3094{
3095        load *= exp;
3096        load += active * (FIXED_1 - exp);
3097        load += 1UL << (FSHIFT - 1);
3098        return load >> FSHIFT;
3099}
3100
3101#ifdef CONFIG_NO_HZ
3102/*
3103 * For NO_HZ we delay the active fold to the next LOAD_FREQ update.
3104 *
3105 * When making the ILB scale, we should try to pull this in as well.
3106 */
3107static atomic_long_t calc_load_tasks_idle;
3108
3109static void calc_load_account_idle(struct rq *this_rq)
3110{
3111        long delta;
3112
3113        delta = calc_load_fold_active(this_rq);
3114        if (delta)
3115                atomic_long_add(delta, &calc_load_tasks_idle);
3116}
3117
3118static long calc_load_fold_idle(void)
3119{
3120        long delta = 0;
3121
3122        /*
3123         * Its got a race, we don't care...
3124         */
3125        if (atomic_long_read(&calc_load_tasks_idle))
3126                delta = atomic_long_xchg(&calc_load_tasks_idle, 0);
3127
3128        return delta;
3129}
3130
3131/**
3132 * fixed_power_int - compute: x^n, in O(log n) time
3133 *
3134 * @x:         base of the power
3135 * @frac_bits: fractional bits of @x
3136 * @n:         power to raise @x to.
3137 *
3138 * By exploiting the relation between the definition of the natural power
3139 * function: x^n := x*x*...*x (x multiplied by itself for n times), and
3140 * the binary encoding of numbers used by computers: n := \Sum n_i * 2^i,
3141 * (where: n_i \elem {0, 1}, the binary vector representing n),
3142 * we find: x^n := x^(\Sum n_i * 2^i) := \Prod x^(n_i * 2^i), which is
3143 * of course trivially computable in O(log_2 n), the length of our binary
3144 * vector.
3145 */
3146static unsigned long
3147fixed_power_int(unsigned long x, unsigned int frac_bits, unsigned int n)
3148{
3149        unsigned long result = 1UL << frac_bits;
3150
3151        if (n) for (;;) {
3152                if (n & 1) {
3153                        result *= x;
3154                        result += 1UL << (frac_bits - 1);
3155                        result >>= frac_bits;
3156                }
3157                n >>= 1;
3158                if (!n)
3159                        break;
3160                x *= x;
3161                x += 1UL << (frac_bits - 1);
3162                x >>= frac_bits;
3163        }
3164
3165        return result;
3166}
3167
3168/*
3169 * a1 = a0 * e + a * (1 - e)
3170 *
3171 * a2 = a1 * e + a * (1 - e)
3172 *    = (a0 * e + a * (1 - e)) * e + a * (1 - e)
3173 *    = a0 * e^2 + a * (1 - e) * (1 + e)
3174 *
3175 * a3 = a2 * e + a * (1 - e)
3176 *    = (a0 * e^2 + a * (1 - e) * (1 + e)) * e + a * (1 - e)
3177 *    = a0 * e^3 + a * (1 - e) * (1 + e + e^2)
3178 *
3179 *  ...
3180 *
3181 * an = a0 * e^n + a * (1 - e) * (1 + e + ... + e^n-1) [1]
3182 *    = a0 * e^n + a * (1 - e) * (1 - e^n)/(1 - e)
3183 *    = a0 * e^n + a * (1 - e^n)
3184 *
3185 * [1] application of the geometric series:
3186 *
3187 *              n         1 - x^(n+1)
3188 *     S_n := \Sum x^i = -------------
3189 *             i=0          1 - x
3190 */
3191static unsigned long
3192calc_load_n(unsigned long load, unsigned long exp,
3193            unsigned long active, unsigned int n)
3194{
3195
3196        return calc_load(load, fixed_power_int(exp, FSHIFT, n), active);
3197}
3198
3199/*
3200 * NO_HZ can leave us missing all per-cpu ticks calling
3201 * calc_load_account_active(), but since an idle CPU folds its delta into
3202 * calc_load_tasks_idle per calc_load_account_idle(), all we need to do is fold
3203 * in the pending idle delta if our idle period crossed a load cycle boundary.
3204 *
3205 * Once we've updated the global active value, we need to apply the exponential
3206 * weights adjusted to the number of cycles missed.
3207 */
3208static void calc_global_nohz(unsigned long ticks)
3209{
3210        long delta, active, n;
3211
3212        if (time_before(jiffies, calc_load_update))
3213                return;
3214
3215        /*
3216         * If we crossed a calc_load_update boundary, make sure to fold
3217         * any pending idle changes, the respective CPUs might have
3218         * missed the tick driven calc_load_account_active() update
3219         * due to NO_HZ.
3220         */
3221        delta = calc_load_fold_idle();
3222        if (delta)
3223                atomic_long_add(delta, &calc_load_tasks);
3224
3225        /*
3226         * If we were idle for multiple load cycles, apply them.
3227         */
3228        if (ticks >= LOAD_FREQ) {
3229                n = ticks / LOAD_FREQ;
3230
3231                active = atomic_long_read(&calc_load_tasks);
3232                active = active > 0 ? active * FIXED_1 : 0;
3233
3234                avenrun[0] = calc_load_n(avenrun[0], EXP_1, active, n);
3235                avenrun[1] = calc_load_n(avenrun[1], EXP_5, active, n);
3236                avenrun[2] = calc_load_n(avenrun[2], EXP_15, active, n);
3237
3238                calc_load_update += n * LOAD_FREQ;
3239        }
3240
3241        /*
3242         * Its possible the remainder of the above division also crosses
3243         * a LOAD_FREQ period, the regular check in calc_global_load()
3244         * which comes after this will take care of that.
3245         *
3246         * Consider us being 11 ticks before a cycle completion, and us
3247         * sleeping for 4*LOAD_FREQ + 22 ticks, then the above code will
3248         * age us 4 cycles, and the test in calc_global_load() will
3249         * pick up the final one.
3250         */
3251}
3252#else
3253static void calc_load_account_idle(struct rq *this_rq)
3254{
3255}
3256
3257static inline long calc_load_fold_idle(void)
3258{
3259        return 0;
3260}
3261
3262static void calc_global_nohz(unsigned long ticks)
3263{
3264}
3265#endif
3266
3267/**
3268 * get_avenrun - get the load average array
3269 * @loads:      pointer to dest load array
3270 * @offset:     offset to add
3271 * @shift:      shift count to shift the result left
3272 *
3273 * These values are estimates at best, so no need for locking.
3274 */
3275void get_avenrun(unsigned long *loads, unsigned long offset, int shift)
3276{
3277        loads[0] = (avenrun[0] + offset) << shift;
3278        loads[1] = (avenrun[1] + offset) << shift;
3279        loads[2] = (avenrun[2] + offset) << shift;
3280}
3281
3282/*
3283 * calc_load - update the avenrun load estimates 10 ticks after the
3284 * CPUs have updated calc_load_tasks.
3285 */
3286void calc_global_load(unsigned long ticks)
3287{
3288        long active;
3289
3290        calc_global_nohz(ticks);
3291
3292        if (time_before(jiffies, calc_load_update + 10))
3293                return;
3294
3295        active = atomic_long_read(&calc_load_tasks);
3296        active = active > 0 ? active * FIXED_1 : 0;
3297
3298        avenrun[0] = calc_load(avenrun[0], EXP_1, active);
3299        avenrun[1] = calc_load(avenrun[1], EXP_5, active);
3300        avenrun[2] = calc_load(avenrun[2], EXP_15, active);
3301
3302        calc_load_update += LOAD_FREQ;
3303}
3304
3305/*
3306 * Called from update_cpu_load() to periodically update this CPU's
3307 * active count.
3308 */
3309static void calc_load_account_active(struct rq *this_rq)
3310{
3311        long delta;
3312
3313        if (time_before(jiffies, this_rq->calc_load_update))
3314                return;
3315
3316        delta  = calc_load_fold_active(this_rq);
3317        delta += calc_load_fold_idle();
3318        if (delta)
3319                atomic_long_add(delta, &calc_load_tasks);
3320
3321        this_rq->calc_load_update += LOAD_FREQ;
3322}
3323
3324/*
3325 * The exact cpuload at various idx values, calculated at every tick would be
3326 * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load
3327 *
3328 * If a cpu misses updates for n-1 ticks (as it was idle) and update gets called
3329 * on nth tick when cpu may be busy, then we have:
3330 * load = ((2^idx - 1) / 2^idx)^(n-1) * load
3331 * load = (2^idx - 1) / 2^idx) * load + 1 / 2^idx * cur_load
3332 *
3333 * decay_load_missed() below does efficient calculation of
3334 * load = ((2^idx - 1) / 2^idx)^(n-1) * load
3335 * avoiding 0..n-1 loop doing load = ((2^idx - 1) / 2^idx) * load
3336 *
3337 * The calculation is approximated on a 128 point scale.
3338 * degrade_zero_ticks is the number of ticks after which load at any
3339 * particular idx is approximated to be zero.
3340 * degrade_factor is a precomputed table, a row for each load idx.
3341 * Each column corresponds to degradation factor for a power of two ticks,
3342 * based on 128 point scale.
3343 * Example:
3344 * row 2, col 3 (=12) says that the degradation at load idx 2 after
3345 * 8 ticks is 12/128 (which is an approximation of exact factor 3^8/4^8).
3346 *
3347 * With this power of 2 load factors, we can degrade the load n times
3348 * by looking at 1 bits in n and doing as many mult/shift instead of
3349 * n mult/shifts needed by the exact degradation.
3350 */
3351#define DEGRADE_SHIFT           7
3352static const unsigned char
3353                degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128};
3354static const unsigned char
3355                degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = {
3356                                        {0, 0, 0, 0, 0, 0, 0, 0},
3357                                        {64, 32, 8, 0, 0, 0, 0, 0},
3358                                        {96, 72, 40, 12, 1, 0, 0},
3359                                        {112, 98, 75, 43, 15, 1, 0},
3360                                        {120, 112, 98, 76, 45, 16, 2} };
3361
3362/*
3363 * Update cpu_load for any missed ticks, due to tickless idle. The backlog
3364 * would be when CPU is idle and so we just decay the old load without
3365 * adding any new load.
3366 */
3367static unsigned long
3368decay_load_missed(unsigned long load, unsigned long missed_updates, int idx)
3369{
3370        int j = 0;
3371
3372        if (!missed_updates)
3373                return load;
3374
3375        if (missed_updates >= degrade_zero_ticks[idx])
3376                return 0;
3377
3378        if (idx == 1)
3379                return load >> missed_updates;
3380
3381        while (missed_updates) {
3382                if (missed_updates % 2)
3383                        load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT;
3384
3385                missed_updates >>= 1;
3386                j++;
3387        }
3388        return load;
3389}
3390
3391/*
3392 * Update rq->cpu_load[] statistics. This function is usually called every
3393 * scheduler tick (TICK_NSEC). With tickless idle this will not be called
3394 * every tick. We fix it up based on jiffies.
3395 */
3396static void update_cpu_load(struct rq *this_rq)
3397{
3398        unsigned long this_load = this_rq->load.weight;
3399        unsigned long curr_jiffies = jiffies;
3400        unsigned long pending_updates;
3401        int i, scale;
3402
3403        this_rq->nr_load_updates++;
3404
3405        /* Avoid repeated calls on same jiffy, when moving in and out of idle */
3406        if (curr_jiffies == this_rq->last_load_update_tick)
3407                return;
3408
3409        pending_updates = curr_jiffies - this_rq->last_load_update_tick;
3410        this_rq->last_load_update_tick = curr_jiffies;
3411
3412        /* Update our load: */
3413        this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */
3414        for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) {
3415                unsigned long old_load, new_load;
3416
3417                /* scale is effectively 1 << i now, and >> i divides by scale */
3418
3419                old_load = this_rq->cpu_load[i];
3420                old_load = decay_load_missed(old_load, pending_updates - 1, i);
3421                new_load = this_load;
3422                /*
3423                 * Round up the averaging division if load is increasing. This
3424                 * prevents us from getting stuck on 9 if the load is 10, for
3425                 * example.
3426                 */
3427                if (new_load > old_load)
3428                        new_load += scale - 1;
3429
3430                this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i;
3431        }
3432
3433        sched_avg_update(this_rq);
3434}
3435
3436static void update_cpu_load_active(struct rq *this_rq)
3437{
3438        update_cpu_load(this_rq);
3439
3440        calc_load_account_active(this_rq);
3441}
3442
3443#ifdef CONFIG_SMP
3444
3445/*
3446 * sched_exec - execve() is a valuable balancing opportunity, because at
3447 * this point the task has the smallest effective memory and cache footprint.
3448 */
3449void sched_exec(void)
3450{
3451        struct task_struct *p = current;
3452        unsigned long flags;
3453        struct rq *rq;
3454        int dest_cpu;
3455
3456        rq = task_rq_lock(p, &flags);
3457        dest_cpu = p->sched_class->select_task_rq(rq, p, SD_BALANCE_EXEC, 0);
3458        if (dest_cpu == smp_processor_id())
3459                goto unlock;
3460
3461        /*
3462         * select_task_rq() can race against ->cpus_allowed
3463         */
3464        if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed) &&
3465            likely(cpu_active(dest_cpu)) && migrate_task(p, rq)) {
3466                struct migration_arg arg = { p, dest_cpu };
3467
3468                task_rq_unlock(rq, &flags);
3469                stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg);
3470                return;
3471        }
3472unlock:
3473        task_rq_unlock(rq, &flags);
3474}
3475
3476#endif
3477
3478DEFINE_PER_CPU(struct kernel_stat, kstat);
3479
3480EXPORT_PER_CPU_SYMBOL(kstat);
3481
3482/*
3483 * Return any ns on the sched_clock that have not yet been accounted in
3484 * @p in case that task is currently running.
3485 *
3486 * Called with task_rq_lock() held on @rq.
3487 */
3488static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq)
3489{
3490        u64 ns = 0;
3491
3492        if (task_current(rq, p)) {
3493                update_rq_clock(rq);
3494                ns = rq->clock_task - p->se.exec_start;
3495                if ((s64)ns < 0)
3496                        ns = 0;
3497        }
3498
3499        return ns;
3500}
3501
3502unsigned long long task_delta_exec(struct task_struct *p)
3503{
3504        unsigned long flags;
3505        struct rq *rq;
3506        u64 ns = 0;
3507
3508        rq = task_rq_lock(p, &flags);
3509        ns = do_task_delta_exec(p, rq);
3510        task_rq_unlock(rq, &flags);
3511
3512        return ns;
3513}
3514
3515/*
3516 * Return accounted runtime for the task.
3517 * In case the task is currently running, return the runtime plus current's
3518 * pending runtime that have not been accounted yet.
3519 */
3520unsigned long long task_sched_runtime(struct task_struct *p)
3521{
3522        unsigned long flags;
3523        struct rq *rq;
3524        u64 ns = 0;
3525
3526        rq = task_rq_lock(p, &flags);
3527        ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq);
3528        task_rq_unlock(rq, &flags);
3529
3530        return ns;
3531}
3532
3533/*
3534 * Return sum_exec_runtime for the thread group.
3535 * In case the task is currently running, return the sum plus current's
3536 * pending runtime that have not been accounted yet.
3537 *
3538 * Note that the thread group might have other running tasks as well,
3539 * so the return value not includes other pending runtime that other
3540 * running tasks might have.
3541 */
3542unsigned long long thread_group_sched_runtime(struct task_struct *p)
3543{
3544        struct task_cputime totals;
3545        unsigned long flags;
3546        struct rq *rq;
3547        u64 ns;
3548
3549        rq = task_rq_lock(p, &flags);
3550        thread_group_cputime(p, &totals);
3551        ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq);
3552        task_rq_unlock(rq, &flags);
3553
3554        return ns;
3555}
3556
3557/*
3558 * Account user cpu time to a process.
3559 * @p: the process that the cpu time gets accounted to
3560 * @cputime: the cpu time spent in user space since the last update
3561 * @cputime_scaled: cputime scaled by cpu frequency
3562 */
3563void account_user_time(struct task_struct *p, cputime_t cputime,
3564                       cputime_t cputime_scaled)
3565{
3566        struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3567        cputime64_t tmp;
3568
3569        /* Add user time to process. */
3570        p->utime = cputime_add(p->utime, cputime);
3571        p->utimescaled = cputime_add(p->utimescaled, cputime_scaled);
3572        account_group_user_time(p, cputime);
3573
3574        /* Add user time to cpustat. */
3575        tmp = cputime_to_cputime64(cputime);
3576        if (TASK_NICE(p) > 0)
3577                cpustat->nice = cputime64_add(cpustat->nice, tmp);
3578        else
3579                cpustat->user = cputime64_add(cpustat->user, tmp);
3580
3581        cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime);
3582        /* Account for user time used */
3583        acct_update_integrals(p);
3584}
3585
3586/*
3587 * Account guest cpu time to a process.
3588 * @p: the process that the cpu time gets accounted to
3589 * @cputime: the cpu time spent in virtual machine since the last update
3590 * @cputime_scaled: cputime scaled by cpu frequency
3591 */
3592static void account_guest_time(struct task_struct *p, cputime_t cputime,
3593                               cputime_t cputime_scaled)
3594{
3595        cputime64_t tmp;
3596        struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3597
3598        tmp = cputime_to_cputime64(cputime);
3599
3600        /* Add guest time to process. */
3601        p->utime = cputime_add(p->utime, cputime);
3602        p->utimescaled = cputime_add(p->utimescaled, cputime_scaled);
3603        account_group_user_time(p, cputime);
3604        p->gtime = cputime_add(p->gtime, cputime);
3605
3606        /* Add guest time to cpustat. */
3607        if (TASK_NICE(p) > 0) {
3608                cpustat->nice = cputime64_add(cpustat->nice, tmp);
3609                cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp);
3610        } else {
3611                cpustat->user = cputime64_add(cpustat->user, tmp);
3612                cpustat->guest = cputime64_add(cpustat->guest, tmp);
3613        }
3614}
3615
3616/*
3617 * Account system cpu time to a process and desired cpustat field
3618 * @p: the process that the cpu time gets accounted to
3619 * @cputime: the cpu time spent in kernel space since the last update
3620 * @cputime_scaled: cputime scaled by cpu frequency
3621 * @target_cputime64: pointer to cpustat field that has to be updated
3622 */
3623static inline
3624void __account_system_time(struct task_struct *p, cputime_t cputime,
3625                        cputime_t cputime_scaled, cputime64_t *target_cputime64)
3626{
3627        cputime64_t tmp = cputime_to_cputime64(cputime);
3628
3629        /* Add system time to process. */
3630        p->stime = cputime_add(p->stime, cputime);
3631        p->stimescaled = cputime_add(p->stimescaled, cputime_scaled);
3632        account_group_system_time(p, cputime);
3633
3634        /* Add system time to cpustat. */
3635        *target_cputime64 = cputime64_add(*target_cputime64, tmp);
3636        cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime);
3637
3638        /* Account for system time used */
3639        acct_update_integrals(p);
3640}
3641
3642/*
3643 * Account system cpu time to a process.
3644 * @p: the process that the cpu time gets accounted to
3645 * @hardirq_offset: the offset to subtract from hardirq_count()
3646 * @cputime: the cpu time spent in kernel space since the last update
3647 * @cputime_scaled: cputime scaled by cpu frequency
3648 */
3649void account_system_time(struct task_struct *p, int hardirq_offset,
3650                         cputime_t cputime, cputime_t cputime_scaled)
3651{
3652        struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3653        cputime64_t *target_cputime64;
3654
3655        if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) {
3656                account_guest_time(p, cputime, cputime_scaled);
3657                return;
3658        }
3659
3660        if (hardirq_count() - hardirq_offset)
3661                target_cputime64 = &cpustat->irq;
3662        else if (in_serving_softirq())
3663                target_cputime64 = &cpustat->softirq;
3664        else
3665                target_cputime64 = &cpustat->system;
3666
3667        __account_system_time(p, cputime, cputime_scaled, target_cputime64);
3668}
3669
3670/*
3671 * Account for involuntary wait time.
3672 * @cputime: the cpu time spent in involuntary wait
3673 */
3674void account_steal_time(cputime_t cputime)
3675{
3676        struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3677        cputime64_t cputime64 = cputime_to_cputime64(cputime);
3678
3679        cpustat->steal = cputime64_add(cpustat->steal, cputime64);
3680}
3681
3682/*
3683 * Account for idle time.
3684 * @cputime: the cpu time spent in idle wait
3685 */
3686void account_idle_time(cputime_t cputime)
3687{
3688        struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3689        cputime64_t cputime64 = cputime_to_cputime64(cputime);
3690        struct rq *rq = this_rq();
3691
3692        if (atomic_read(&rq->nr_iowait) > 0)
3693                cpustat->iowait = cputime64_add(cpustat->iowait, cputime64);
3694        else
3695                cpustat->idle = cputime64_add(cpustat->idle, cputime64);
3696}
3697
3698#ifndef CONFIG_VIRT_CPU_ACCOUNTING
3699
3700#ifdef CONFIG_IRQ_TIME_ACCOUNTING
3701/*
3702 * Account a tick to a process and cpustat
3703 * @p: the process that the cpu time gets accounted to
3704 * @user_tick: is the tick from userspace
3705 * @rq: the pointer to rq
3706 *
3707 * Tick demultiplexing follows the order
3708 * - pending hardirq update
3709 * - pending softirq update
3710 * - user_time
3711 * - idle_time
3712 * - system time
3713 *   - check for guest_time
3714 *   - else account as system_time
3715 *
3716 * Check for hardirq is done both for system and user time as there is
3717 * no timer going off while we are on hardirq and hence we may never get an
3718 * opportunity to update it solely in system time.
3719 * p->stime and friends are only updated on system time and not on irq
3720 * softirq as those do not count in task exec_runtime any more.
3721 */
3722static void irqtime_account_process_tick(struct task_struct *p, int user_tick,
3723                                                struct rq *rq)
3724{
3725        cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy);
3726        cputime64_t tmp = cputime_to_cputime64(cputime_one_jiffy);
3727        struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3728
3729        if (irqtime_account_hi_update()) {
3730                cpustat->irq = cputime64_add(cpustat->irq, tmp);
3731        } else if (irqtime_account_si_update()) {
3732                cpustat->softirq = cputime64_add(cpustat->softirq, tmp);
3733        } else if (this_cpu_ksoftirqd() == p) {
3734                /*
3735                 * ksoftirqd time do not get accounted in cpu_softirq_time.
3736                 * So, we have to handle it separately here.
3737                 * Also, p->stime needs to be updated for ksoftirqd.
3738                 */
3739                __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled,
3740                                        &cpustat->softirq);
3741        } else if (user_tick) {
3742                account_user_time(p, cputime_one_jiffy, one_jiffy_scaled);
3743        } else if (p == rq->idle) {
3744                account_idle_time(cputime_one_jiffy);
3745        } else if (p->flags & PF_VCPU) { /* System time or guest time */
3746                account_guest_time(p, cputime_one_jiffy, one_jiffy_scaled);
3747        } else {
3748                __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled,
3749                                        &cpustat->system);
3750        }
3751}
3752
3753static void irqtime_account_idle_ticks(int ticks)
3754{
3755        int i;
3756        struct rq *rq = this_rq();
3757
3758        for (i = 0; i < ticks; i++)
3759                irqtime_account_process_tick(current, 0, rq);
3760}
3761#else /* CONFIG_IRQ_TIME_ACCOUNTING */
3762static void irqtime_account_idle_ticks(int ticks) {}
3763static void irqtime_account_process_tick(struct task_struct *p, int user_tick,
3764                                                struct rq *rq) {}
3765#endif /* CONFIG_IRQ_TIME_ACCOUNTING */
3766
3767/*
3768 * Account a single tick of cpu time.
3769 * @p: the process that the cpu time gets accounted to
3770 * @user_tick: indicates if the tick is a user or a system tick
3771 */
3772void account_process_tick(struct task_struct *p, int user_tick)
3773{
3774        cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy);
3775        struct rq *rq = this_rq();
3776
3777        if (sched_clock_irqtime) {
3778                irqtime_account_process_tick(p, user_tick, rq);
3779                return;
3780        }
3781
3782        if (user_tick)
3783                account_user_time(p, cputime_one_jiffy, one_jiffy_scaled);
3784        else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET))
3785                account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy,
3786                                    one_jiffy_scaled);
3787        else
3788                account_idle_time(cputime_one_jiffy);
3789}
3790
3791/*
3792 * Account multiple ticks of steal time.
3793 * @p: the process from which the cpu time has been stolen
3794 * @ticks: number of stolen ticks
3795 */
3796void account_steal_ticks(unsigned long ticks)
3797{
3798        account_steal_time(jiffies_to_cputime(ticks));
3799}
3800
3801/*
3802 * Account multiple ticks of idle time.
3803 * @ticks: number of stolen ticks
3804 */
3805void account_idle_ticks(unsigned long ticks)
3806{
3807
3808        if (sched_clock_irqtime) {
3809                irqtime_account_idle_ticks(ticks);
3810                return;
3811        }
3812
3813        account_idle_time(jiffies_to_cputime(ticks));
3814}
3815
3816#endif
3817
3818/*
3819 * Use precise platform statistics if available:
3820 */
3821#ifdef CONFIG_VIRT_CPU_ACCOUNTING
3822void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
3823{
3824        *ut = p->utime;
3825        *st = p->stime;
3826}
3827
3828void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
3829{
3830        struct task_cputime cputime;
3831
3832        thread_group_cputime(p, &cputime);
3833
3834        *ut = cputime.utime;
3835        *st = cputime.stime;
3836}
3837#else
3838
3839#ifndef nsecs_to_cputime
3840# define nsecs_to_cputime(__nsecs)      nsecs_to_jiffies(__nsecs)
3841#endif
3842
3843void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
3844{
3845        cputime_t rtime, utime = p->utime, total = cputime_add(utime, p->stime);
3846
3847        /*
3848         * Use CFS's precise accounting:
3849         */
3850        rtime = nsecs_to_cputime(p->se.sum_exec_runtime);
3851
3852        if (total) {
3853                u64 temp = rtime;
3854
3855                temp *= utime;
3856                do_div(temp, total);
3857                utime = (cputime_t)temp;
3858        } else
3859                utime = rtime;
3860
3861        /*
3862         * Compare with previous values, to keep monotonicity:
3863         */
3864        p->prev_utime = max(p->prev_utime, utime);
3865        p->prev_stime = max(p->prev_stime, cputime_sub(rtime, p->prev_utime));
3866
3867        *ut = p->prev_utime;
3868        *st = p->prev_stime;
3869}
3870
3871/*
3872 * Must be called with siglock held.
3873 */
3874void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
3875{
3876        struct signal_struct *sig = p->signal;
3877        struct task_cputime cputime;
3878        cputime_t rtime, utime, total;
3879
3880        thread_group_cputime(p, &cputime);
3881
3882        total = cputime_add(cputime.utime, cputime.stime);
3883        rtime = nsecs_to_cputime(cputime.sum_exec_runtime);
3884
3885        if (total) {
3886                u64 temp = rtime;
3887
3888                temp *= cputime.utime;
3889                do_div(temp, total);
3890                utime = (cputime_t)temp;
3891        } else
3892                utime = rtime;
3893
3894        sig->prev_utime = max(sig->prev_utime, utime);
3895        sig->prev_stime = max(sig->prev_stime,
3896                              cputime_sub(rtime, sig->prev_utime));
3897
3898        *ut = sig->prev_utime;
3899        *st = sig->prev_stime;
3900}
3901#endif
3902
3903/*
3904 * This function gets called by the timer code, with HZ frequency.
3905 * We call it with interrupts disabled.
3906 *
3907 * It also gets called by the fork code, when changing the parent's
3908 * timeslices.
3909 */
3910void scheduler_tick(void)
3911{
3912        int cpu = smp_processor_id();
3913        struct rq *rq = cpu_rq(cpu);
3914        struct task_struct *curr = rq->curr;
3915
3916        sched_clock_tick();
3917
3918        raw_spin_lock(&rq->lock);
3919        update_rq_clock(rq);
3920        update_cpu_load_active(rq);
3921        curr->sched_class->task_tick(rq, curr, 0);
3922        raw_spin_unlock(&rq->lock);
3923
3924        perf_event_task_tick();
3925
3926#ifdef CONFIG_SMP
3927        rq->idle_at_tick = idle_cpu(cpu);
3928        trigger_load_balance(rq, cpu);
3929#endif
3930}
3931
3932notrace unsigned long get_parent_ip(unsigned long addr)
3933{
3934        if (in_lock_functions(addr)) {
3935                addr = CALLER_ADDR2;
3936                if (in_lock_functions(addr))
3937                        addr = CALLER_ADDR3;
3938        }
3939        return addr;
3940}
3941
3942#if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \
3943                                defined(CONFIG_PREEMPT_TRACER))
3944
3945void __kprobes add_preempt_count(int val)
3946{
3947#ifdef CONFIG_DEBUG_PREEMPT
3948        /*
3949         * Underflow?
3950         */
3951        if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0)))
3952                return;
3953#endif
3954        preempt_count() += val;
3955#ifdef CONFIG_DEBUG_PREEMPT
3956        /*
3957         * Spinlock count overflowing soon?
3958         */
3959        DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >=
3960                                PREEMPT_MASK - 10);
3961#endif
3962        if (preempt_count() == val)
3963                trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1));
3964}
3965EXPORT_SYMBOL(add_preempt_count);
3966
3967void __kprobes sub_preempt_count(int val)
3968{
3969#ifdef CONFIG_DEBUG_PREEMPT
3970        /*
3971         * Underflow?
3972         */
3973        if (DEBUG_LOCKS_WARN_ON(val > preempt_count()))
3974                return;
3975        /*
3976         * Is the spinlock portion underflowing?
3977         */
3978        if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) &&
3979                        !(preempt_count() & PREEMPT_MASK)))
3980                return;
3981#endif
3982
3983        if (preempt_count() == val)
3984                trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1));
3985        preempt_count() -= val;
3986}
3987EXPORT_SYMBOL(sub_preempt_count);
3988
3989#endif
3990
3991/*
3992 * Print scheduling while atomic bug:
3993 */
3994static noinline void __schedule_bug(struct task_struct *prev)
3995{
3996        struct pt_regs *regs = get_irq_regs();
3997
3998        printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n",
3999                prev->comm, prev->pid, preempt_count());
4000
4001        debug_show_held_locks(prev);
4002        print_modules();
4003        if (irqs_disabled())
4004                print_irqtrace_events(prev);
4005
4006        if (regs)
4007                show_regs(regs);
4008        else
4009                dump_stack();
4010}
4011
4012/*
4013 * Various schedule()-time debugging checks and statistics:
4014 */
4015static inline void schedule_debug(struct task_struct *prev)
4016{
4017        /*
4018         * Test if we are atomic. Since do_exit() needs to call into
4019         * schedule() atomically, we ignore that path for now.
4020         * Otherwise, whine if we are scheduling when we should not be.
4021         */
4022        if (unlikely(in_atomic_preempt_off() && !prev->exit_state))
4023                __schedule_bug(prev);
4024
4025        profile_hit(SCHED_PROFILING, __builtin_return_address(0));
4026
4027        schedstat_inc(this_rq(), sched_count);
4028#ifdef CONFIG_SCHEDSTATS
4029        if (unlikely(prev->lock_depth >= 0)) {
4030                schedstat_inc(this_rq(), rq_sched_info.bkl_count);
4031                schedstat_inc(prev, sched_info.bkl_count);
4032        }
4033#endif
4034}
4035
4036static void put_prev_task(struct rq *rq, struct task_struct *prev)
4037{
4038        if (prev->se.on_rq)
4039                update_rq_clock(rq);
4040        prev->sched_class->put_prev_task(rq, prev);
4041}
4042
4043/*
4044 * Pick up the highest-prio task:
4045 */
4046static inline struct task_struct *
4047pick_next_task(struct rq *rq)
4048{
4049        const struct sched_class *class;
4050        struct task_struct *p;
4051
4052        /*
4053         * Optimization: we know that if all tasks are in
4054         * the fair class we can call that function directly:
4055         */
4056        if (likely(rq->nr_running == rq->cfs.nr_running)) {
4057                p = fair_sched_class.pick_next_task(rq);
4058                if (likely(p))
4059                        return p;
4060        }
4061
4062        for_each_class(class) {
4063                p = class->pick_next_task(rq);
4064                if (p)
4065                        return p;
4066        }
4067
4068        BUG(); /* the idle class will always have a runnable task */
4069}
4070
4071/*
4072 * schedule() is the main scheduler function.
4073 */
4074asmlinkage void __sched schedule(void)
4075{
4076        struct task_struct *prev, *next;
4077        unsigned long *switch_count;
4078        struct rq *rq;
4079        int cpu;
4080
4081need_resched:
4082        preempt_disable();
4083        cpu = smp_processor_id();
4084        rq = cpu_rq(cpu);
4085        rcu_note_context_switch(cpu);
4086        prev = rq->curr;
4087
4088        schedule_debug(prev);
4089
4090        if (sched_feat(HRTICK))
4091                hrtick_clear(rq);
4092
4093        raw_spin_lock_irq(&rq->lock);
4094
4095        switch_count = &prev->nivcsw;
4096        if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) {
4097                if (unlikely(signal_pending_state(prev->state, prev))) {
4098                        prev->state = TASK_RUNNING;
4099                } else {
4100                        /*
4101                         * If a worker is going to sleep, notify and
4102                         * ask workqueue whether it wants to wake up a
4103                         * task to maintain concurrency.  If so, wake
4104                         * up the task.
4105                         */
4106                        if (prev->flags & PF_WQ_WORKER) {
4107                                struct task_struct *to_wakeup;
4108
4109                                to_wakeup = wq_worker_sleeping(prev, cpu);
4110                                if (to_wakeup)
4111                                        try_to_wake_up_local(to_wakeup);
4112                        }
4113                        deactivate_task(rq, prev, DEQUEUE_SLEEP);
4114
4115                        /*
4116                         * If we are going to sleep and we have plugged IO queued, make
4117                         * sure to submit it to avoid deadlocks.
4118                         */
4119                        if (blk_needs_flush_plug(prev)) {
4120                                raw_spin_unlock(&rq->lock);
4121                                blk_schedule_flush_plug(prev);
4122                                raw_spin_lock(&rq->lock);
4123                        }
4124                }
4125                switch_count = &prev->nvcsw;
4126        }
4127
4128        pre_schedule(rq, prev);
4129
4130        if (unlikely(!rq->nr_running))
4131                idle_balance(cpu, rq);
4132
4133        put_prev_task(rq, prev);
4134        next = pick_next_task(rq);
4135        clear_tsk_need_resched(prev);
4136        rq->skip_clock_update = 0;
4137
4138        if (likely(prev != next)) {
4139                rq->nr_switches++;
4140                rq->curr = next;
4141                ++*switch_count;
4142
4143                context_switch(rq, prev, next); /* unlocks the rq */
4144                /*
4145                 * The context switch have flipped the stack from under us
4146                 * and restored the local variables which were saved when
4147                 * this task called schedule() in the past. prev == current
4148                 * is still correct, but it can be moved to another cpu/rq.
4149                 */
4150                cpu = smp_processor_id();
4151                rq = cpu_rq(cpu);
4152        } else
4153                raw_spin_unlock_irq(&rq->lock);
4154
4155        post_schedule(rq);
4156
4157        preempt_enable_no_resched();
4158        if (need_resched())
4159                goto need_resched;
4160}
4161EXPORT_SYMBOL(schedule);
4162
4163#ifdef CONFIG_MUTEX_SPIN_ON_OWNER
4164/*
4165 * Look out! "owner" is an entirely speculative pointer
4166 * access and not reliable.
4167 */
4168int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner)
4169{
4170        unsigned int cpu;
4171        struct rq *rq;
4172
4173        if (!sched_feat(OWNER_SPIN))
4174                return 0;
4175
4176#ifdef CONFIG_DEBUG_PAGEALLOC
4177        /*
4178         * Need to access the cpu field knowing that
4179         * DEBUG_PAGEALLOC could have unmapped it if
4180         * the mutex owner just released it and exited.
4181         */
4182        if (probe_kernel_address(&owner->cpu, cpu))
4183                return 0;
4184#else
4185        cpu = owner->cpu;
4186#endif
4187
4188        /*
4189         * Even if the access succeeded (likely case),
4190         * the cpu field may no longer be valid.
4191         */
4192        if (cpu >= nr_cpumask_bits)
4193                return 0;
4194
4195        /*
4196         * We need to validate that we can do a
4197         * get_cpu() and that we have the percpu area.
4198         */
4199        if (!cpu_online(cpu))
4200                return 0;
4201
4202        rq = cpu_rq(cpu);
4203
4204        for (;;) {
4205                /*
4206                 * Owner changed, break to re-assess state.
4207                 */
4208                if (lock->owner != owner) {
4209                        /*
4210                         * If the lock has switched to a different owner,
4211                         * we likely have heavy contention. Return 0 to quit
4212                         * optimistic spinning and not contend further:
4213                         */
4214                        if (lock->owner)
4215                                return 0;
4216                        break;
4217                }
4218
4219                /*
4220                 * Is that owner really running on that cpu?
4221                 */
4222                if (task_thread_info(rq->curr) != owner || need_resched())
4223                        return 0;
4224
4225                arch_mutex_cpu_relax();
4226        }
4227
4228        return 1;
4229}
4230#endif
4231
4232#ifdef CONFIG_PREEMPT
4233/*
4234 * this is the entry point to schedule() from in-kernel preemption
4235 * off of preempt_enable. Kernel preemptions off return from interrupt
4236 * occur there and call schedule directly.
4237 */
4238asmlinkage void __sched notrace preempt_schedule(void)
4239{
4240        struct thread_info *ti = current_thread_info();
4241
4242        /*
4243         * If there is a non-zero preempt_count or interrupts are disabled,
4244         * we do not want to preempt the current task. Just return..
4245         */
4246        if (likely(ti->preempt_count || irqs_disabled()))
4247                return;
4248
4249        do {
4250                add_preempt_count_notrace(PREEMPT_ACTIVE);
4251                schedule();
4252                sub_preempt_count_notrace(PREEMPT_ACTIVE);
4253
4254                /*
4255                 * Check again in case we missed a preemption opportunity
4256                 * between schedule and now.
4257                 */
4258                barrier();
4259        } while (need_resched());
4260}
4261EXPORT_SYMBOL(preempt_schedule);
4262
4263/*
4264 * this is the entry point to schedule() from kernel preemption
4265 * off of irq context.
4266 * Note, that this is called and return with irqs disabled. This will
4267 * protect us against recursive calling from irq.
4268 */
4269asmlinkage void __sched preempt_schedule_irq(void)
4270{
4271        struct thread_info *ti = current_thread_info();
4272
4273        /* Catch callers which need to be fixed */
4274        BUG_ON(ti->preempt_count || !irqs_disabled());
4275
4276        do {
4277                add_preempt_count(PREEMPT_ACTIVE);
4278                local_irq_enable();
4279                schedule();
4280                local_irq_disable();
4281                sub_preempt_count(PREEMPT_ACTIVE);
4282
4283                /*
4284                 * Check again in case we missed a preemption opportunity
4285                 * between schedule and now.
4286                 */
4287                barrier();
4288        } while (need_resched());
4289}
4290
4291#endif /* CONFIG_PREEMPT */
4292
4293int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags,
4294                          void *key)
4295{
4296        return try_to_wake_up(curr->private, mode, wake_flags);
4297}
4298EXPORT_SYMBOL(default_wake_function);
4299
4300/*
4301 * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just
4302 * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve
4303 * number) then we wake all the non-exclusive tasks and one exclusive task.
4304 *
4305 * There are circumstances in which we can try to wake a task which has already
4306 * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns
4307 * zero in this (rare) case, and we handle it by continuing to scan the queue.
4308 */
4309static void __wake_up_common(wait_queue_head_t *q, unsigned int mode,
4310                        int nr_exclusive, int wake_flags, void *key)
4311{
4312        wait_queue_t *curr, *next;
4313
4314        list_for_each_entry_safe(curr, next, &q->task_list, task_list) {
4315                unsigned flags = curr->flags;
4316
4317                if (curr->func(curr, mode, wake_flags, key) &&
4318                                (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive)
4319                        break;
4320        }
4321}
4322
4323/**
4324 * __wake_up - wake up threads blocked on a waitqueue.
4325 * @q: the waitqueue
4326 * @mode: which threads
4327 * @nr_exclusive: how many wake-one or wake-many threads to wake up
4328 * @key: is directly passed to the wakeup function
4329 *
4330 * It may be assumed that this function implies a write memory barrier before
4331 * changing the task state if and only if any tasks are woken up.
4332 */
4333void __wake_up(wait_queue_head_t *q, unsigned int mode,
4334                        int nr_exclusive, void *key)
4335{
4336        unsigned long flags;
4337
4338        spin_lock_irqsave(&q->lock, flags);
4339        __wake_up_common(q, mode, nr_exclusive, 0, key);
4340        spin_unlock_irqrestore(&q->lock, flags);
4341}
4342EXPORT_SYMBOL(__wake_up);
4343
4344/*
4345 * Same as __wake_up but called with the spinlock in wait_queue_head_t held.
4346 */
4347void __wake_up_locked(wait_queue_head_t *q, unsigned int mode)
4348{
4349        __wake_up_common(q, mode, 1, 0, NULL);
4350}
4351EXPORT_SYMBOL_GPL(__wake_up_locked);
4352
4353void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key)
4354{
4355        __wake_up_common(q, mode, 1, 0, key);
4356}
4357EXPORT_SYMBOL_GPL(__wake_up_locked_key);
4358
4359/**
4360 * __wake_up_sync_key - wake up threads blocked on a waitqueue.
4361 * @q: the waitqueue
4362 * @mode: which threads
4363 * @nr_exclusive: how many wake-one or wake-many threads to wake up
4364 * @key: opaque value to be passed to wakeup targets
4365 *
4366 * The sync wakeup differs that the waker knows that it will schedule
4367 * away soon, so while the target thread will be woken up, it will not
4368 * be migrated to another CPU - ie. the two threads are 'synchronized'
4369 * with each other. This can prevent needless bouncing between CPUs.
4370 *
4371 * On UP it can prevent extra preemption.
4372 *
4373 * It may be assumed that this function implies a write memory barrier before
4374 * changing the task state if and only if any tasks are woken up.
4375 */
4376void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode,
4377                        int nr_exclusive, void *key)
4378{
4379        unsigned long flags;
4380        int wake_flags = WF_SYNC;
4381
4382        if (unlikely(!q))
4383                return;
4384
4385        if (unlikely(!nr_exclusive))
4386                wake_flags = 0;
4387
4388        spin_lock_irqsave(&q->lock, flags);
4389        __wake_up_common(q, mode, nr_exclusive, wake_flags, key);
4390        spin_unlock_irqrestore(&q->lock, flags);
4391}
4392EXPORT_SYMBOL_GPL(__wake_up_sync_key);
4393
4394/*
4395 * __wake_up_sync - see __wake_up_sync_key()
4396 */
4397void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive)
4398{
4399        __wake_up_sync_key(q, mode, nr_exclusive, NULL);
4400}
4401EXPORT_SYMBOL_GPL(__wake_up_sync);      /* For internal use only */
4402
4403/**
4404 * complete: - signals a single thread waiting on this completion
4405 * @x:  holds the state of this particular completion
4406 *
4407 * This will wake up a single thread waiting on this completion. Threads will be
4408 * awakened in the same order in which they were queued.
4409 *
4410 * See also complete_all(), wait_for_completion() and related routines.
4411 *
4412 * It may be assumed that this function implies a write memory barrier before
4413 * changing the task state if and only if any tasks are woken up.
4414 */
4415void complete(struct completion *x)
4416{
4417        unsigned long flags;
4418
4419        spin_lock_irqsave(&x->wait.lock, flags);
4420        x->done++;
4421        __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL);
4422        spin_unlock_irqrestore(&x->wait.lock, flags);
4423}
4424EXPORT_SYMBOL(complete);
4425
4426/**
4427 * complete_all: - signals all threads waiting on this completion
4428 * @x:  holds the state of this particular completion
4429 *
4430 * This will wake up all threads waiting on this particular completion event.
4431 *
4432 * It may be assumed that this function implies a write memory barrier before
4433 * changing the task state if and only if any tasks are woken up.
4434 */
4435void complete_all(struct completion *x)
4436{
4437        unsigned long flags;
4438
4439        spin_lock_irqsave(&x->wait.lock, flags);
4440        x->done += UINT_MAX/2;
4441        __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL);
4442        spin_unlock_irqrestore(&x->wait.lock, flags);
4443}
4444EXPORT_SYMBOL(complete_all);
4445
4446static inline long __sched
4447do_wait_for_common(struct completion *x, long timeout, int state)
4448{
4449        if (!x->done) {
4450                DECLARE_WAITQUEUE(wait, current);
4451
4452                __add_wait_queue_tail_exclusive(&x->wait, &wait);
4453                do {
4454                        if (signal_pending_state(state, current)) {
4455                                timeout = -ERESTARTSYS;
4456                                break;
4457                        }
4458                        __set_current_state(state);
4459                        spin_unlock_irq(&x->wait.lock);
4460                        timeout = schedule_timeout(timeout);
4461                        spin_lock_irq(&x->wait.lock);
4462                } while (!x->done && timeout);
4463                __remove_wait_queue(&x->wait, &wait);
4464                if (!x->done)
4465                        return timeout;
4466        }
4467        x->done--;
4468        return timeout ?: 1;
4469}
4470
4471static long __sched
4472wait_for_common(struct completion *x, long timeout, int state)
4473{
4474        might_sleep();
4475
4476        spin_lock_irq(&x->wait.lock);
4477        timeout = do_wait_for_common(x, timeout, state);
4478        spin_unlock_irq(&x->wait.lock);
4479        return timeout;
4480}
4481
4482/**
4483 * wait_for_completion: - waits for completion of a task
4484 * @x:  holds the state of this particular completion
4485 *
4486 * This waits to be signaled for completion of a specific task. It is NOT
4487 * interruptible and there is no timeout.
4488 *
4489 * See also similar routines (i.e. wait_for_completion_timeout()) with timeout
4490 * and interrupt capability. Also see complete().
4491 */
4492void __sched wait_for_completion(struct completion *x)
4493{
4494        wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE);
4495}
4496EXPORT_SYMBOL(wait_for_completion);
4497
4498/**
4499 * wait_for_completion_timeout: - waits for completion of a task (w/timeout)
4500 * @x:  holds the state of this particular completion
4501 * @timeout:  timeout value in jiffies
4502 *
4503 * This waits for either a completion of a specific task to be signaled or for a
4504 * specified timeout to expire. The timeout is in jiffies. It is not
4505 * interruptible.
4506 */
4507unsigned long __sched
4508wait_for_completion_timeout(struct completion *x, unsigned long timeout)
4509{
4510        return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE);
4511}
4512EXPORT_SYMBOL(wait_for_completion_timeout);
4513
4514/**
4515 * wait_for_completion_interruptible: - waits for completion of a task (w/intr)
4516 * @x:  holds the state of this particular completion
4517 *
4518 * This waits for completion of a specific task to be signaled. It is
4519 * interruptible.
4520 */
4521int __sched wait_for_completion_interruptible(struct completion *x)
4522{
4523        long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE);
4524        if (t == -ERESTARTSYS)
4525                return t;
4526        return 0;
4527}
4528EXPORT_SYMBOL(wait_for_completion_interruptible);
4529
4530/**
4531 * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr))
4532 * @x:  holds the state of this particular completion
4533 * @timeout:  timeout value in jiffies
4534 *
4535 * This waits for either a completion of a specific task to be signaled or for a
4536 * specified timeout to expire. It is interruptible. The timeout is in jiffies.
4537 */
4538long __sched
4539wait_for_completion_interruptible_timeout(struct completion *x,
4540                                          unsigned long timeout)
4541{
4542        return wait_for_common(x, timeout, TASK_INTERRUPTIBLE);
4543}
4544EXPORT_SYMBOL(wait_for_completion_interruptible_timeout);
4545
4546/**
4547 * wait_for_completion_killable: - waits for completion of a task (killable)
4548 * @x:  holds the state of this particular completion
4549 *
4550 * This waits to be signaled for completion of a specific task. It can be
4551 * interrupted by a kill signal.
4552 */
4553int __sched wait_for_completion_killable(struct completion *x)
4554{
4555        long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE);
4556        if (t == -ERESTARTSYS)
4557                return t;
4558        return 0;
4559}
4560EXPORT_SYMBOL(wait_for_completion_killable);
4561
4562/**
4563 * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable))
4564 * @x:  holds the state of this particular completion
4565 * @timeout:  timeout value in jiffies
4566 *
4567 * This waits for either a completion of a specific task to be
4568 * signaled or for a specified timeout to expire. It can be
4569 * interrupted by a kill signal. The timeout is in jiffies.
4570 */
4571long __sched
4572wait_for_completion_killable_timeout(struct completion *x,
4573                                     unsigned long timeout)
4574{
4575        return wait_for_common(x, timeout, TASK_KILLABLE);
4576}
4577EXPORT_SYMBOL(wait_for_completion_killable_timeout);
4578
4579/**
4580 *      try_wait_for_completion - try to decrement a completion without blocking
4581 *      @x:     completion structure
4582 *
4583 *      Returns: 0 if a decrement cannot be done without blocking
4584 *               1 if a decrement succeeded.
4585 *
4586 *      If a completion is being used as a counting completion,
4587 *      attempt to decrement the counter without blocking. This
4588 *      enables us to avoid waiting if the resource the completion
4589 *      is protecting is not available.
4590 */
4591bool try_wait_for_completion(struct completion *x)
4592{
4593        unsigned long flags;
4594        int ret = 1;
4595
4596        spin_lock_irqsave(&x->wait.lock, flags);
4597        if (!x->done)
4598                ret = 0;
4599        else
4600                x->done--;
4601        spin_unlock_irqrestore(&x->wait.lock, flags);
4602        return ret;
4603}
4604EXPORT_SYMBOL(try_wait_for_completion);
4605
4606/**
4607 *      completion_done - Test to see if a completion has any waiters
4608 *      @x:     completion structure
4609 *
4610 *      Returns: 0 if there are waiters (wait_for_completion() in progress)
4611 *               1 if there are no waiters.
4612 *
4613 */
4614bool completion_done(struct completion *x)
4615{
4616        unsigned long flags;
4617        int ret = 1;
4618
4619        spin_lock_irqsave(&x->wait.lock, flags);
4620        if (!x->done)
4621                ret = 0;
4622        spin_unlock_irqrestore(&x->wait.lock, flags);
4623        return ret;
4624}
4625EXPORT_SYMBOL(completion_done);
4626
4627static long __sched
4628sleep_on_common(wait_queue_head_t *q, int state, long timeout)
4629{
4630        unsigned long flags;
4631        wait_queue_t wait;
4632
4633        init_waitqueue_entry(&wait, current);
4634
4635        __set_current_state(state);
4636
4637        spin_lock_irqsave(&q->lock, flags);
4638        __add_wait_queue(q, &wait);
4639        spin_unlock(&q->lock);
4640        timeout = schedule_timeout(timeout);
4641        spin_lock_irq(&q->lock);
4642        __remove_wait_queue(q, &wait);
4643        spin_unlock_irqrestore(&q->lock, flags);
4644
4645        return timeout;
4646}
4647
4648void __sched interruptible_sleep_on(wait_queue_head_t *q)
4649{
4650        sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
4651}
4652EXPORT_SYMBOL(interruptible_sleep_on);
4653
4654long __sched
4655interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout)
4656{
4657        return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout);
4658}
4659EXPORT_SYMBOL(interruptible_sleep_on_timeout);
4660
4661void __sched sleep_on(wait_queue_head_t *q)
4662{
4663        sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
4664}
4665EXPORT_SYMBOL(sleep_on);
4666
4667long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout)
4668{
4669        return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout);
4670}
4671EXPORT_SYMBOL(sleep_on_timeout);
4672
4673#ifdef CONFIG_RT_MUTEXES
4674
4675/*
4676 * rt_mutex_setprio - set the current priority of a task
4677 * @p: task
4678 * @prio: prio value (kernel-internal form)
4679 *
4680 * This function changes the 'effective' priority of a task. It does
4681 * not touch ->normal_prio like __setscheduler().
4682 *
4683 * Used by the rt_mutex code to implement priority inheritance logic.
4684 */
4685void rt_mutex_setprio(struct task_struct *p, int prio)
4686{
4687        unsigned long flags;
4688        int oldprio, on_rq, running;
4689        struct rq *rq;
4690        const struct sched_class *prev_class;
4691
4692        BUG_ON(prio < 0 || prio > MAX_PRIO);
4693
4694        rq = task_rq_lock(p, &flags);
4695
4696        trace_sched_pi_setprio(p, prio);
4697        oldprio = p->prio;
4698        prev_class = p->sched_class;
4699        on_rq = p->se.on_rq;
4700        running = task_current(rq, p);
4701        if (on_rq)
4702                dequeue_task(rq, p, 0);
4703        if (running)
4704                p->sched_class->put_prev_task(rq, p);
4705
4706        if (rt_prio(prio))
4707                p->sched_class = &rt_sched_class;
4708        else
4709                p->sched_class = &fair_sched_class;
4710
4711        p->prio = prio;
4712
4713        if (running)
4714                p->sched_class->set_curr_task(rq);
4715        if (on_rq)
4716                enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0);
4717
4718        check_class_changed(rq, p, prev_class, oldprio);
4719        task_rq_unlock(rq, &flags);
4720}
4721
4722#endif
4723
4724void set_user_nice(struct task_struct *p, long nice)
4725{
4726        int old_prio, delta, on_rq;
4727        unsigned long flags;
4728        struct rq *rq;
4729
4730        if (TASK_NICE(p) == nice || nice < -20 || nice > 19)
4731                return;
4732        /*
4733         * We have to be careful, if called from sys_setpriority(),
4734         * the task might be in the middle of scheduling on another CPU.
4735         */
4736        rq = task_rq_lock(p, &flags);
4737        /*
4738         * The RT priorities are set via sched_setscheduler(), but we still
4739         * allow the 'normal' nice value to be set - but as expected
4740         * it wont have any effect on scheduling until the task is
4741         * SCHED_FIFO/SCHED_RR:
4742         */
4743        if (task_has_rt_policy(p)) {
4744                p->static_prio = NICE_TO_PRIO(nice);
4745                goto out_unlock;
4746        }
4747        on_rq = p->se.on_rq;
4748        if (on_rq)
4749                dequeue_task(rq, p, 0);
4750
4751        p->static_prio = NICE_TO_PRIO(nice);
4752        set_load_weight(p);
4753        old_prio = p->prio;
4754        p->prio = effective_prio(p);
4755        delta = p->prio - old_prio;
4756
4757        if (on_rq) {
4758                enqueue_task(rq, p, 0);
4759                /*
4760                 * If the task increased its priority or is running and
4761                 * lowered its priority, then reschedule its CPU:
4762                 */
4763                if (delta < 0 || (delta > 0 && task_running(rq, p)))
4764                        resched_task(rq->curr);
4765        }
4766out_unlock:
4767        task_rq_unlock(rq, &flags);
4768}
4769EXPORT_SYMBOL(set_user_nice);
4770
4771/*
4772 * can_nice - check if a task can reduce its nice value
4773 * @p: task
4774 * @nice: nice value
4775 */
4776int can_nice(const struct task_struct *p, const int nice)
4777{
4778        /* convert nice value [19,-20] to rlimit style value [1,40] */
4779        int nice_rlim = 20 - nice;
4780
4781        return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) ||
4782                capable(CAP_SYS_NICE));
4783}
4784
4785#ifdef __ARCH_WANT_SYS_NICE
4786
4787/*
4788 * sys_nice - change the priority of the current process.
4789 * @increment: priority increment
4790 *
4791 * sys_setpriority is a more generic, but much slower function that
4792 * does similar things.
4793 */
4794SYSCALL_DEFINE1(nice, int, increment)
4795{
4796        long nice, retval;
4797
4798        /*
4799         * Setpriority might change our priority at the same moment.
4800         * We don't have to worry. Conceptually one call occurs first
4801         * and we have a single winner.
4802         */
4803        if (increment < -40)
4804                increment = -40;
4805        if (increment > 40)
4806                increment = 40;
4807
4808        nice = TASK_NICE(current) + increment;
4809        if (nice < -20)
4810                nice = -20;
4811        if (nice > 19)
4812                nice = 19;
4813
4814        if (increment < 0 && !can_nice(current, nice))
4815                return -EPERM;
4816
4817        retval = security_task_setnice(current, nice);
4818        if (retval)
4819                return retval;
4820
4821        set_user_nice(current, nice);
4822        return 0;
4823}
4824
4825#endif
4826
4827/**
4828 * task_prio - return the priority value of a given task.
4829 * @p: the task in question.
4830 *
4831 * This is the priority value as seen by users in /proc.
4832 * RT tasks are offset by -200. Normal tasks are centered
4833 * around 0, value goes from -16 to +15.
4834 */
4835int task_prio(const struct task_struct *p)
4836{
4837        return p->prio - MAX_RT_PRIO;
4838}
4839
4840/**
4841 * task_nice - return the nice value of a given task.
4842 * @p: the task in question.
4843 */
4844int task_nice(const struct task_struct *p)
4845{
4846        return TASK_NICE(p);
4847}
4848EXPORT_SYMBOL(task_nice);
4849
4850/**
4851 * idle_cpu - is a given cpu idle currently?
4852 * @cpu: the processor in question.
4853 */
4854int idle_cpu(int cpu)
4855{
4856        return cpu_curr(cpu) == cpu_rq(cpu)->idle;
4857}
4858
4859/**
4860 * idle_task - return the idle task for a given cpu.
4861 * @cpu: the processor in question.
4862 */
4863struct task_struct *idle_task(int cpu)
4864{
4865        return cpu_rq(cpu)->idle;
4866}
4867
4868/**
4869 * find_process_by_pid - find a process with a matching PID value.
4870 * @pid: the pid in question.
4871 */
4872static struct task_struct *find_process_by_pid(pid_t pid)
4873{
4874        return pid ? find_task_by_vpid(pid) : current;
4875}
4876
4877/* Actually do priority change: must hold rq lock. */
4878static void
4879__setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio)
4880{
4881        BUG_ON(p->se.on_rq);
4882
4883        p->policy = policy;
4884        p->rt_priority = prio;
4885        p->normal_prio = normal_prio(p);
4886        /* we are holding p->pi_lock already */
4887        p->prio = rt_mutex_getprio(p);
4888        if (rt_prio(p->prio))
4889                p->sched_class = &rt_sched_class;
4890        else
4891                p->sched_class = &fair_sched_class;
4892        set_load_weight(p);
4893}
4894
4895/*
4896 * check the target process has a UID that matches the current process's
4897 */
4898static bool check_same_owner(struct task_struct *p)
4899{
4900        const struct cred *cred = current_cred(), *pcred;
4901        bool match;
4902
4903        rcu_read_lock();
4904        pcred = __task_cred(p);
4905        if (cred->user->user_ns == pcred->user->user_ns)
4906                match = (cred->euid == pcred->euid ||
4907                         cred->euid == pcred->uid);
4908        else
4909                match = false;
4910        rcu_read_unlock();
4911        return match;
4912}
4913
4914static int __sched_setscheduler(struct task_struct *p, int policy,
4915                                const struct sched_param *param, bool user)
4916{
4917        int retval, oldprio, oldpolicy = -1, on_rq, running;
4918        unsigned long flags;
4919        const struct sched_class *prev_class;
4920        struct rq *rq;
4921        int reset_on_fork;
4922
4923        /* may grab non-irq protected spin_locks */
4924        BUG_ON(in_interrupt());
4925recheck:
4926        /* double check policy once rq lock held */
4927        if (policy < 0) {
4928                reset_on_fork = p->sched_reset_on_fork;
4929                policy = oldpolicy = p->policy;
4930        } else {
4931                reset_on_fork = !!(policy & SCHED_RESET_ON_FORK);
4932                policy &= ~SCHED_RESET_ON_FORK;
4933
4934                if (policy != SCHED_FIFO && policy != SCHED_RR &&
4935                                policy != SCHED_NORMAL && policy != SCHED_BATCH &&
4936                                policy != SCHED_IDLE)
4937                        return -EINVAL;
4938        }
4939
4940        /*
4941         * Valid priorities for SCHED_FIFO and SCHED_RR are
4942         * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL,
4943         * SCHED_BATCH and SCHED_IDLE is 0.
4944         */
4945        if (param->sched_priority < 0 ||
4946            (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) ||
4947            (!p->mm && param->sched_priority > MAX_RT_PRIO-1))
4948                return -EINVAL;
4949        if (rt_policy(policy) != (param->sched_priority != 0))
4950                return -EINVAL;
4951
4952        /*
4953         * Allow unprivileged RT tasks to decrease priority:
4954         */
4955        if (user && !capable(CAP_SYS_NICE)) {
4956                if (rt_policy(policy)) {
4957                        unsigned long rlim_rtprio =
4958                                        task_rlimit(p, RLIMIT_RTPRIO);
4959
4960                        /* can't set/change the rt policy */
4961                        if (policy != p->policy && !rlim_rtprio)
4962                                return -EPERM;
4963
4964                        /* can't increase priority */
4965                        if (param->sched_priority > p->rt_priority &&
4966                            param->sched_priority > rlim_rtprio)
4967                                return -EPERM;
4968                }
4969
4970                /*
4971                 * Treat SCHED_IDLE as nice 20. Only allow a switch to
4972                 * SCHED_NORMAL if the RLIMIT_NICE would normally permit it.
4973                 */
4974                if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) {
4975                        if (!can_nice(p, TASK_NICE(p)))
4976                                return -EPERM;
4977                }
4978
4979                /* can't change other user's priorities */
4980                if (!check_same_owner(p))
4981                        return -EPERM;
4982
4983                /* Normal users shall not reset the sched_reset_on_fork flag */
4984                if (p->sched_reset_on_fork && !reset_on_fork)
4985                        return -EPERM;
4986        }
4987
4988        if (user) {
4989                retval = security_task_setscheduler(p);
4990                if (retval)
4991                        return retval;
4992        }
4993
4994        /*
4995         * make sure no PI-waiters arrive (or leave) while we are
4996         * changing the priority of the task:
4997         */
4998        raw_spin_lock_irqsave(&p->pi_lock, flags);
4999        /*
5000         * To be able to change p->policy safely, the appropriate
5001         * runqueue lock must be held.
5002         */
5003        rq = __task_rq_lock(p);
5004
5005        /*
5006         * Changing the policy of the stop threads its a very bad idea
5007         */
5008        if (p == rq->stop) {
5009                __task_rq_unlock(rq);
5010                raw_spin_unlock_irqrestore(&p->pi_lock, flags);
5011                return -EINVAL;
5012        }
5013
5014        /*
5015         * If not changing anything there's no need to proceed further:
5016         */
5017        if (unlikely(policy == p->policy && (!rt_policy(policy) ||
5018                        param->sched_priority == p->rt_priority))) {
5019
5020                __task_rq_unlock(rq);
5021                raw_spin_unlock_irqrestore(&p->pi_lock, flags);
5022                return 0;
5023        }
5024
5025#ifdef CONFIG_RT_GROUP_SCHED
5026        if (user) {
5027                /*
5028                 * Do not allow realtime tasks into groups that have no runtime
5029                 * assigned.
5030                 */
5031                if (rt_bandwidth_enabled() && rt_policy(policy) &&
5032                                task_group(p)->rt_bandwidth.rt_runtime == 0 &&
5033                                !task_group_is_autogroup(task_group(p))) {
5034                        __task_rq_unlock(rq);
5035                        raw_spin_unlock_irqrestore(&p->pi_lock, flags);
5036                        return -EPERM;
5037                }
5038        }
5039#endif
5040
5041        /* recheck policy now with rq lock held */
5042        if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) {
5043                policy = oldpolicy = -1;
5044                __task_rq_unlock(rq);
5045                raw_spin_unlock_irqrestore(&p->pi_lock, flags);
5046                goto recheck;
5047        }
5048        on_rq = p->se.on_rq;
5049        running = task_current(rq, p);
5050        if (on_rq)
5051                deactivate_task(rq, p, 0);
5052        if (running)
5053                p->sched_class->put_prev_task(rq, p);
5054
5055        p->sched_reset_on_fork = reset_on_fork;
5056
5057        oldprio = p->prio;
5058        prev_class = p->sched_class;
5059        __setscheduler(rq, p, policy, param->sched_priority);
5060
5061        if (running)
5062                p->sched_class->set_curr_task(rq);
5063        if (on_rq)
5064                activate_task(rq, p, 0);
5065
5066        check_class_changed(rq, p, prev_class, oldprio);
5067        __task_rq_unlock(rq);
5068        raw_spin_unlock_irqrestore(&p->pi_lock, flags);
5069
5070        rt_mutex_adjust_pi(p);
5071
5072        return 0;
5073}
5074
5075/**
5076 * sched_setscheduler - change the scheduling policy and/or RT priority of a thread.
5077 * @p: the task in question.
5078 * @policy: new policy.
5079 * @param: structure containing the new RT priority.
5080 *
5081 * NOTE that the task may be already dead.
5082 */
5083int sched_setscheduler(struct task_struct *p, int policy,
5084                       const struct sched_param *param)
5085{
5086        return __sched_setscheduler(p, policy, param, true);
5087}
5088EXPORT_SYMBOL_GPL(sched_setscheduler);
5089
5090/**
5091 * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace.
5092 * @p: the task in question.
5093 * @policy: new policy.
5094 * @param: structure containing the new RT priority.
5095 *
5096 * Just like sched_setscheduler, only don't bother checking if the
5097 * current context has permission.  For example, this is needed in
5098 * stop_machine(): we create temporary high priority worker threads,
5099 * but our caller might not have that capability.
5100 */
5101int sched_setscheduler_nocheck(struct task_struct *p, int policy,
5102                               const struct sched_param *param)
5103{
5104        return __sched_setscheduler(p, policy, param, false);
5105}
5106
5107static int
5108do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param)
5109{
5110        struct sched_param lparam;
5111        struct task_struct *p;
5112        int retval;
5113
5114        if (!param || pid < 0)
5115                return -EINVAL;
5116        if (copy_from_user(&lparam, param, sizeof(struct sched_param)))
5117                return -EFAULT;
5118
5119        rcu_read_lock();
5120        retval = -ESRCH;
5121        p = find_process_by_pid(pid);
5122        if (p != NULL)
5123                retval = sched_setscheduler(p, policy, &lparam);
5124        rcu_read_unlock();
5125
5126        return retval;
5127}
5128
5129/**
5130 * sys_sched_setscheduler - set/change the scheduler policy and RT priority
5131 * @pid: the pid in question.
5132 * @policy: new policy.
5133 * @param: structure containing the new RT priority.
5134 */
5135SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy,
5136                struct sched_param __user *, param)
5137{
5138        /* negative values for policy are not valid */
5139        if (policy < 0)
5140                return -EINVAL;
5141
5142        return do_sched_setscheduler(pid, policy, param);
5143}
5144
5145/**
5146 * sys_sched_setparam - set/change the RT priority of a thread
5147 * @pid: the pid in question.
5148 * @param: structure containing the new RT priority.
5149 */
5150SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param)
5151{
5152        return do_sched_setscheduler(pid, -1, param);
5153}
5154
5155/**
5156 * sys_sched_getscheduler - get the policy (scheduling class) of a thread
5157 * @pid: the pid in question.
5158 */
5159SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid)
5160{
5161        struct task_struct *p;
5162        int retval;
5163
5164        if (pid < 0)
5165                return -EINVAL;
5166
5167        retval = -ESRCH;
5168        rcu_read_lock();
5169        p = find_process_by_pid(pid);
5170        if (p) {
5171                retval = security_task_getscheduler(p);
5172                if (!retval)
5173                        retval = p->policy
5174                                | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0);
5175        }
5176        rcu_read_unlock();
5177        return retval;
5178}
5179
5180/**
5181 * sys_sched_getparam - get the RT priority of a thread
5182 * @pid: the pid in question.
5183 * @param: structure containing the RT priority.
5184 */
5185SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param)
5186{
5187        struct sched_param lp;
5188        struct task_struct *p;
5189        int retval;
5190
5191        if (!param || pid < 0)
5192                return -EINVAL;
5193
5194        rcu_read_lock();
5195        p = find_process_by_pid(pid);
5196        retval = -ESRCH;
5197        if (!p)
5198                goto out_unlock;
5199
5200        retval = security_task_getscheduler(p);
5201        if (retval)
5202                goto out_unlock;
5203
5204        lp.sched_priority = p->rt_priority;
5205        rcu_read_unlock();
5206
5207        /*
5208         * This one might sleep, we cannot do it with a spinlock held ...
5209         */
5210        retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0;
5211
5212        return retval;
5213
5214out_unlock:
5215        rcu_read_unlock();
5216        return retval;
5217}
5218
5219long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)
5220{
5221        cpumask_var_t cpus_allowed, new_mask;
5222        struct task_struct *p;
5223        int retval;
5224
5225        get_online_cpus();
5226        rcu_read_lock();
5227
5228        p = find_process_by_pid(pid);
5229        if (!p) {
5230                rcu_read_unlock();
5231                put_online_cpus();
5232                return -ESRCH;
5233        }
5234
5235        /* Prevent p going away */
5236        get_task_struct(p);
5237        rcu_read_unlock();
5238
5239        if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) {
5240                retval = -ENOMEM;
5241                goto out_put_task;
5242        }
5243        if (!alloc_cpumask_var(&new_mask, GFP_KERNE