kernel/sched/sched.h

   1
   2 #include <linux/sched.h>
   3 #include <linux/sched/sysctl.h>
   4 #include <linux/sched/rt.h>
   5 #include <linux/mutex.h>
   6 #include <linux/spinlock.h>
   7 #include <linux/stop_machine.h>
   8 #include <linux/tick.h>
   9
  10 #include "cpupri.h"
  11 #include "cpuacct.h"
  12
  13 extern __read_mostly int scheduler_running;
  14
  15 /*
  16  * Convert user-nice values [ -20 ... 0 ... 19 ]
  17  * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ],
  18  * and back.
  19  */
  20 #define NICE_TO_PRIO(nice)      (MAX_RT_PRIO + (nice) + 20)
  21 #define PRIO_TO_NICE(prio)      ((prio) - MAX_RT_PRIO - 20)
  22 #define TASK_NICE(p)            PRIO_TO_NICE((p)->static_prio)
  23
  24 /*
  25  * 'User priority' is the nice value converted to something we
  26  * can work with better when scaling various scheduler parameters,
  27  * it's a [ 0 ... 39 ] range.
  28  */
  29 #define USER_PRIO(p)            ((p)-MAX_RT_PRIO)
  30 #define TASK_USER_PRIO(p)       USER_PRIO((p)->static_prio)
  31 #define MAX_USER_PRIO           (USER_PRIO(MAX_PRIO))
  32
  33 /*
  34  * Helpers for converting nanosecond timing to jiffy resolution
  35  */
  36 #define NS_TO_JIFFIES(TIME)     ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
  37
  38 /*
  39  * Increase resolution of nice-level calculations for 64-bit architectures.
  40  * The extra resolution improves shares distribution and load balancing of
  41  * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
  42  * hierarchies, especially on larger systems. This is not a user-visible change
  43  * and does not change the user-interface for setting shares/weights.
  44  *
  45  * We increase resolution only if we have enough bits to allow this increased
  46  * resolution (i.e. BITS_PER_LONG > 32). The costs for increasing resolution
  47  * when BITS_PER_LONG <= 32 are pretty high and the returns do not justify the
  48  * increased costs.
  49  */
  50 #if 0 /* BITS_PER_LONG > 32 -- currently broken: it increases power usage under light load  */
  51 # define SCHED_LOAD_RESOLUTION  10
  52 # define scale_load(w)          ((w) << SCHED_LOAD_RESOLUTION)
  53 # define scale_load_down(w)     ((w) >> SCHED_LOAD_RESOLUTION)
  54 #else
  55 # define SCHED_LOAD_RESOLUTION  0
  56 # define scale_load(w)          (w)
  57 # define scale_load_down(w)     (w)
  58 #endif
  59
  60 #define SCHED_LOAD_SHIFT        (10 + SCHED_LOAD_RESOLUTION)
  61 #define SCHED_LOAD_SCALE        (1L << SCHED_LOAD_SHIFT)
  62
  63 #define NICE_0_LOAD             SCHED_LOAD_SCALE
  64 #define NICE_0_SHIFT            SCHED_LOAD_SHIFT
  65
  66 /*
  67  * These are the 'tuning knobs' of the scheduler:
  68  */
  69
  70 /*
  71  * single value that denotes runtime == period, ie unlimited time.
  72  */
  73 #define RUNTIME_INF     ((u64)~0ULL)
  74
  75 static inline int rt_policy(int policy)
  76 {
  77         if (policy == SCHED_FIFO || policy == SCHED_RR)
  78                 return 1;
  79         return 0;
  80 }
  81
  82 static inline int task_has_rt_policy(struct task_struct *p)
  83 {
  84         return rt_policy(p->policy);
  85 }
  86
  87 /*
  88  * This is the priority-queue data structure of the RT scheduling class:
  89  */
  90 struct rt_prio_array {
  91         DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
  92         struct list_head queue[MAX_RT_PRIO];
  93 };
  94
  95 struct rt_bandwidth {
  96         /* nests inside the rq lock: */
  97         raw_spinlock_t          rt_runtime_lock;
  98         ktime_t                 rt_period;
  99         u64                     rt_runtime;
 100         struct hrtimer          rt_period_timer;
 101 };
 102
 103 extern struct mutex sched_domains_mutex;
 104
 105 #ifdef CONFIG_CGROUP_SCHED
 106
 107 #include <linux/cgroup.h>
 108
 109 struct cfs_rq;
 110 struct rt_rq;
 111
 112 extern struct list_head task_groups;
 113
 114 struct cfs_bandwidth {
 115 #ifdef CONFIG_CFS_BANDWIDTH
 116         raw_spinlock_t lock;
 117         ktime_t period;
 118         u64 quota, runtime;
 119         s64 hierarchal_quota;
 120         u64 runtime_expires;
 121
 122         int idle, timer_active;
 123         struct hrtimer period_timer, slack_timer;
 124         struct list_head throttled_cfs_rq;
 125
 126         /* statistics */
 127         int nr_periods, nr_throttled;
 128         u64 throttled_time;
 129 #endif
 130 };
 131
 132 /* task group related information */
 133 struct task_group {
 134         struct cgroup_subsys_state css;
 135
 136 #ifdef CONFIG_FAIR_GROUP_SCHED
 137         /* schedulable entities of this group on each cpu */
 138         struct sched_entity **se;
 139         /* runqueue "owned" by this group on each cpu */
 140         struct cfs_rq **cfs_rq;
 141         unsigned long shares;
 142
 143         atomic_t load_weight;
 144         atomic64_t load_avg;
 145         atomic_t runnable_avg;
 146 #endif
 147
 148 #ifdef CONFIG_RT_GROUP_SCHED
 149         struct sched_rt_entity **rt_se;
 150         struct rt_rq **rt_rq;
 151
 152         struct rt_bandwidth rt_bandwidth;
 153 #endif
 154
 155         struct rcu_head rcu;
 156         struct list_head list;
 157
 158         struct task_group *parent;
 159         struct list_head siblings;
 160         struct list_head children;
 161
 162 #ifdef CONFIG_SCHED_AUTOGROUP
 163         struct autogroup *autogroup;
 164 #endif
 165
 166         struct cfs_bandwidth cfs_bandwidth;
 167 };
 168
 169 #ifdef CONFIG_FAIR_GROUP_SCHED
 170 #define ROOT_TASK_GROUP_LOAD    NICE_0_LOAD
 171
 172 /*
 173  * A weight of 0 or 1 can cause arithmetics problems.
 174  * A weight of a cfs_rq is the sum of weights of which entities
 175  * are queued on this cfs_rq, so a weight of a entity should not be
 176  * too large, so as the shares value of a task group.
 177  * (The default weight is 1024 - so there's no practical
 178  *  limitation from this.)
 179  */
 180 #define MIN_SHARES      (1UL <<  1)
 181 #define MAX_SHARES      (1UL << 18)
 182 #endif
 183
 184 typedef int (*tg_visitor)(struct task_group *, void *);
 185
 186 extern int walk_tg_tree_from(struct task_group *from,
 187                              tg_visitor down, tg_visitor up, void *data);
 188
 189 /*
 190  * Iterate the full tree, calling @down when first entering a node and @up when
 191  * leaving it for the final time.
 192  *
 193  * Caller must hold rcu_lock or sufficient equivalent.
 194  */
 195 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
 196 {
 197         return walk_tg_tree_from(&root_task_group, down, up, data);
 198 }
 199
 200 extern int tg_nop(struct task_group *tg, void *data);
 201
 202 extern void free_fair_sched_group(struct task_group *tg);
 203 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
 204 extern void unregister_fair_sched_group(struct task_group *tg, int cpu);
 205 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
 206                         struct sched_entity *se, int cpu,
 207                         struct sched_entity *parent);
 208 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
 209 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
 210
 211 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
 212 extern void __start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
 213 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
 214
 215 extern void free_rt_sched_group(struct task_group *tg);
 216 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
 217 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
 218                 struct sched_rt_entity *rt_se, int cpu,
 219                 struct sched_rt_entity *parent);
 220
 221 extern struct task_group *sched_create_group(struct task_group *parent);
 222 extern void sched_online_group(struct task_group *tg,
 223                                struct task_group *parent);
 224 extern void sched_destroy_group(struct task_group *tg);
 225 extern void sched_offline_group(struct task_group *tg);
 226
 227 extern void sched_move_task(struct task_struct *tsk);
 228
 229 #ifdef CONFIG_FAIR_GROUP_SCHED
 230 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
 231 #endif
 232
 233 #else /* CONFIG_CGROUP_SCHED */
 234
 235 struct cfs_bandwidth { };
 236
 237 #endif  /* CONFIG_CGROUP_SCHED */
 238
 239 /* CFS-related fields in a runqueue */
 240 struct cfs_rq {
 241         struct load_weight load;
 242         unsigned int nr_running, h_nr_running;
 243
 244         u64 exec_clock;
 245         u64 min_vruntime;
 246 #ifndef CONFIG_64BIT
 247         u64 min_vruntime_copy;
 248 #endif
 249
 250         struct rb_root tasks_timeline;
 251         struct rb_node *rb_leftmost;
 252
 253         /*
 254          * 'curr' points to currently running entity on this cfs_rq.
 255          * It is set to NULL otherwise (i.e when none are currently running).
 256          */
 257         struct sched_entity *curr, *next, *last, *skip;
 258
 259 #ifdef  CONFIG_SCHED_DEBUG
 260         unsigned int nr_spread_over;
 261 #endif
 262
 263 #ifdef CONFIG_SMP
 264 /*
 265  * Load-tracking only depends on SMP, FAIR_GROUP_SCHED dependency below may be
 266  * removed when useful for applications beyond shares distribution (e.g.
 267  * load-balance).
 268  */
 269 #ifdef CONFIG_FAIR_GROUP_SCHED
 270         /*
 271          * CFS Load tracking
 272          * Under CFS, load is tracked on a per-entity basis and aggregated up.
 273          * This allows for the description of both thread and group usage (in
 274          * the FAIR_GROUP_SCHED case).
 275          */
 276         u64 runnable_load_avg, blocked_load_avg;
 277         atomic64_t decay_counter, removed_load;
 278         u64 last_decay;
 279 #endif /* CONFIG_FAIR_GROUP_SCHED */
 280 /* These always depend on CONFIG_FAIR_GROUP_SCHED */
 281 #ifdef CONFIG_FAIR_GROUP_SCHED
 282         u32 tg_runnable_contrib;
 283         u64 tg_load_contrib;
 284 #endif /* CONFIG_FAIR_GROUP_SCHED */
 285
 286         /*
 287          *   h_load = weight * f(tg)
 288          *
 289          * Where f(tg) is the recursive weight fraction assigned to
 290          * this group.
 291          */
 292         unsigned long h_load;
 293 #endif /* CONFIG_SMP */
 294
 295 #ifdef CONFIG_FAIR_GROUP_SCHED
 296         struct rq *rq;  /* cpu runqueue to which this cfs_rq is attached */
 297
 298         /*
 299          * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
 300          * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
 301          * (like users, containers etc.)
 302          *
 303          * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
 304          * list is used during load balance.
 305          */
 306         int on_list;
 307         struct list_head leaf_cfs_rq_list;
 308         struct task_group *tg;  /* group that "owns" this runqueue */
 309
 310 #ifdef CONFIG_CFS_BANDWIDTH
 311         int runtime_enabled;
 312         u64 runtime_expires;
 313         s64 runtime_remaining;
 314
 315         u64 throttled_clock, throttled_clock_task;
 316         u64 throttled_clock_task_time;
 317         int throttled, throttle_count;
 318         struct list_head throttled_list;
 319 #endif /* CONFIG_CFS_BANDWIDTH */
 320 #endif /* CONFIG_FAIR_GROUP_SCHED */
 321 };
 322
 323 static inline int rt_bandwidth_enabled(void)
 324 {
 325         return sysctl_sched_rt_runtime >= 0;
 326 }
 327
 328 /* Real-Time classes' related field in a runqueue: */
 329 struct rt_rq {
 330         struct rt_prio_array active;
 331         unsigned int rt_nr_running;
 332 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
 333         struct {
 334                 int curr; /* highest queued rt task prio */
 335 #ifdef CONFIG_SMP
 336                 int next; /* next highest */
 337 #endif
 338         } highest_prio;
 339 #endif
 340 #ifdef CONFIG_SMP
 341         unsigned long rt_nr_migratory;
 342         unsigned long rt_nr_total;
 343         int overloaded;
 344         struct plist_head pushable_tasks;
 345 #endif
 346         int rt_throttled;
 347         u64 rt_time;
 348         u64 rt_runtime;
 349         /* Nests inside the rq lock: */
 350         raw_spinlock_t rt_runtime_lock;
 351
 352 #ifdef CONFIG_RT_GROUP_SCHED
 353         unsigned long rt_nr_boosted;
 354
 355         struct rq *rq;
 356         struct list_head leaf_rt_rq_list;
 357         struct task_group *tg;
 358 #endif
 359 };
 360
 361 #ifdef CONFIG_SMP
 362
 363 /*
 364  * We add the notion of a root-domain which will be used to define per-domain
 365  * variables. Each exclusive cpuset essentially defines an island domain by
 366  * fully partitioning the member cpus from any other cpuset. Whenever a new
 367  * exclusive cpuset is created, we also create and attach a new root-domain
 368  * object.
 369  *
 370  */
 371 struct root_domain {
 372         atomic_t refcount;
 373         atomic_t rto_count;
 374         struct rcu_head rcu;
 375         cpumask_var_t span;
 376         cpumask_var_t online;
 377
 378         /*
 379          * The "RT overload" flag: it gets set if a CPU has more than
 380          * one runnable RT task.
 381          */
 382         cpumask_var_t rto_mask;
 383         struct cpupri cpupri;
 384 };
 385
 386 extern struct root_domain def_root_domain;
 387
 388 #endif /* CONFIG_SMP */
 389
 390 /*
 391  * This is the main, per-CPU runqueue data structure.
 392  *
 393  * Locking rule: those places that want to lock multiple runqueues
 394  * (such as the load balancing or the thread migration code), lock
 395  * acquire operations must be ordered by ascending &runqueue.
 396  */
 397 struct rq {
 398         /* runqueue lock: */
 399         raw_spinlock_t lock;
 400
 401         /*
 402          * nr_running and cpu_load should be in the same cacheline because
 403          * remote CPUs use both these fields when doing load calculation.
 404          */
 405         unsigned int nr_running;
 406         #define CPU_LOAD_IDX_MAX 5
 407         unsigned long cpu_load[CPU_LOAD_IDX_MAX];
 408         unsigned long last_load_update_tick;
 409 #ifdef CONFIG_NO_HZ_COMMON
 410         u64 nohz_stamp;
 411         unsigned long nohz_flags;
 412 #endif
 413 #ifdef CONFIG_NO_HZ_FULL
 414         unsigned long last_sched_tick;
 415 #endif
 416         int skip_clock_update;
 417
 418         /* capture load from *all* tasks on this cpu: */
 419         struct load_weight load;
 420         unsigned long nr_load_updates;
 421         u64 nr_switches;
 422
 423         struct cfs_rq cfs;
 424         struct rt_rq rt;
 425
 426 #ifdef CONFIG_FAIR_GROUP_SCHED
 427         /* list of leaf cfs_rq on this cpu: */
 428         struct list_head leaf_cfs_rq_list;
 429 #ifdef CONFIG_SMP
 430         unsigned long h_load_throttle;
 431 #endif /* CONFIG_SMP */
 432 #endif /* CONFIG_FAIR_GROUP_SCHED */
 433
 434 #ifdef CONFIG_RT_GROUP_SCHED
 435         struct list_head leaf_rt_rq_list;
 436 #endif
 437
 438         /*
 439          * This is part of a global counter where only the total sum
 440          * over all CPUs matters. A task can increase this counter on
 441          * one CPU and if it got migrated afterwards it may decrease
 442          * it on another CPU. Always updated under the runqueue lock:
 443          */
 444         unsigned long nr_uninterruptible;
 445
 446         struct task_struct *curr, *idle, *stop;
 447         unsigned long next_balance;
 448         struct mm_struct *prev_mm;
 449
 450         u64 clock;
 451         u64 clock_task;
 452
 453         atomic_t nr_iowait;
 454
 455 #ifdef CONFIG_SMP
 456         struct root_domain *rd;
 457         struct sched_domain *sd;
 458
 459         unsigned long cpu_power;
 460
 461         unsigned char idle_balance;
 462         /* For active balancing */
 463         int post_schedule;
 464         int active_balance;
 465         int push_cpu;
 466         struct cpu_stop_work active_balance_work;
 467         /* cpu of this runqueue: */
 468         int cpu;
 469         int online;
 470
 471         struct list_head cfs_tasks;
 472
 473         u64 rt_avg;
 474         u64 age_stamp;
 475         u64 idle_stamp;
 476         u64 avg_idle;
 477 #endif
 478
 479 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
 480         u64 prev_irq_time;
 481 #endif
 482 #ifdef CONFIG_PARAVIRT
 483         u64 prev_steal_time;
 484 #endif
 485 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
 486         u64 prev_steal_time_rq;
 487 #endif
 488
 489         /* calc_load related fields */
 490         unsigned long calc_load_update;
 491         long calc_load_active;
 492
 493 #ifdef CONFIG_SCHED_HRTICK
 494 #ifdef CONFIG_SMP
 495         int hrtick_csd_pending;
 496         struct call_single_data hrtick_csd;
 497 #endif
 498         struct hrtimer hrtick_timer;
 499 #endif
 500
 501 #ifdef CONFIG_SCHEDSTATS
 502         /* latency stats */
 503         struct sched_info rq_sched_info;
 504         unsigned long long rq_cpu_time;
 505         /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
 506
 507         /* sys_sched_yield() stats */
 508         unsigned int yld_count;
 509
 510         /* schedule() stats */
 511         unsigned int sched_count;
 512         unsigned int sched_goidle;
 513
 514         /* try_to_wake_up() stats */
 515         unsigned int ttwu_count;
 516         unsigned int ttwu_local;
 517 #endif
 518
 519 #ifdef CONFIG_SMP
 520         struct llist_head wake_list;
 521 #endif
 522
 523         struct sched_avg avg;
 524 };
 525
 526 static inline int cpu_of(struct rq *rq)
 527 {
 528 #ifdef CONFIG_SMP
 529         return rq->cpu;
 530 #else
 531         return 0;
 532 #endif
 533 }
 534
 535 DECLARE_PER_CPU(struct rq, runqueues);
 536
 537 #define cpu_rq(cpu)             (&per_cpu(runqueues, (cpu)))
 538 #define this_rq()               (&__get_cpu_var(runqueues))
 539 #define task_rq(p)              cpu_rq(task_cpu(p))
 540 #define cpu_curr(cpu)           (cpu_rq(cpu)->curr)
 541 #define raw_rq()                (&__raw_get_cpu_var(runqueues))
 542
 543 #ifdef CONFIG_SMP
 544
 545 #define rcu_dereference_check_sched_domain(p) \
 546         rcu_dereference_check((p), \
 547                               lockdep_is_held(&sched_domains_mutex))
 548
 549 /*
 550  * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
 551  * See detach_destroy_domains: synchronize_sched for details.
 552  *
 553  * The domain tree of any CPU may only be accessed from within
 554  * preempt-disabled sections.
 555  */
 556 #define for_each_domain(cpu, __sd) \
 557         for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
 558                         __sd; __sd = __sd->parent)
 559
 560 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
 561
 562 /**
 563  * highest_flag_domain - Return highest sched_domain containing flag.
 564  * @cpu:        The cpu whose highest level of sched domain is to
 565  *              be returned.
 566  * @flag:       The flag to check for the highest sched_domain
 567  *              for the given cpu.
 568  *
 569  * Returns the highest sched_domain of a cpu which contains the given flag.
 570  */
 571 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
 572 {
 573         struct sched_domain *sd, *hsd = NULL;
 574
 575         for_each_domain(cpu, sd) {
 576                 if (!(sd->flags & flag))
 577                         break;
 578                 hsd = sd;
 579         }
 580
 581         return hsd;
 582 }
 583
 584 DECLARE_PER_CPU(struct sched_domain *, sd_llc);
 585 DECLARE_PER_CPU(int, sd_llc_id);
 586
 587 struct sched_group_power {
 588         atomic_t ref;
 589         /*
 590          * CPU power of this group, SCHED_LOAD_SCALE being max power for a
 591          * single CPU.
 592          */
 593         unsigned int power, power_orig;
 594         unsigned long next_update;
 595         /*
 596          * Number of busy cpus in this group.
 597          */
 598         atomic_t nr_busy_cpus;
 599
 600         unsigned long cpumask[0]; /* iteration mask */
 601 };
 602
 603 struct sched_group {
 604         struct sched_group *next;       /* Must be a circular list */
 605         atomic_t ref;
 606
 607         unsigned int group_weight;
 608         struct sched_group_power *sgp;
 609
 610         /*
 611          * The CPUs this group covers.
 612          *
 613          * NOTE: this field is variable length. (Allocated dynamically
 614          * by attaching extra space to the end of the structure,
 615          * depending on how many CPUs the kernel has booted up with)
 616          */
 617         unsigned long cpumask[0];
 618 };
 619
 620 static inline struct cpumask *sched_group_cpus(struct sched_group *sg)
 621 {
 622         return to_cpumask(sg->cpumask);
 623 }
 624
 625 /*
 626  * cpumask masking which cpus in the group are allowed to iterate up the domain
 627  * tree.
 628  */
 629 static inline struct cpumask *sched_group_mask(struct sched_group *sg)
 630 {
 631         return to_cpumask(sg->sgp->cpumask);
 632 }
 633
 634 /**
 635  * group_first_cpu - Returns the first cpu in the cpumask of a sched_group.
 636  * @group: The group whose first cpu is to be returned.
 637  */
 638 static inline unsigned int group_first_cpu(struct sched_group *group)
 639 {
 640         return cpumask_first(sched_group_cpus(group));
 641 }
 642
 643 extern int group_balance_cpu(struct sched_group *sg);
 644
 645 #endif /* CONFIG_SMP */
 646
 647 #include "stats.h"
 648 #include "auto_group.h"
 649
 650 #ifdef CONFIG_CGROUP_SCHED
 651
 652 /*
 653  * Return the group to which this tasks belongs.
 654  *
 655  * We cannot use task_subsys_state() and friends because the cgroup
 656  * subsystem changes that value before the cgroup_subsys::attach() method
 657  * is called, therefore we cannot pin it and might observe the wrong value.
 658  *
 659  * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
 660  * core changes this before calling sched_move_task().
 661  *
 662  * Instead we use a 'copy' which is updated from sched_move_task() while
 663  * holding both task_struct::pi_lock and rq::lock.
 664  */
 665 static inline struct task_group *task_group(struct task_struct *p)
 666 {
 667         return p->sched_task_group;
 668 }
 669
 670 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
 671 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
 672 {
 673 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
 674         struct task_group *tg = task_group(p);
 675 #endif
 676
 677 #ifdef CONFIG_FAIR_GROUP_SCHED
 678         p->se.cfs_rq = tg->cfs_rq[cpu];
 679         p->se.parent = tg->se[cpu];
 680 #endif
 681
 682 #ifdef CONFIG_RT_GROUP_SCHED
 683         p->rt.rt_rq  = tg->rt_rq[cpu];
 684         p->rt.parent = tg->rt_se[cpu];
 685 #endif
 686 }
 687
 688 #else /* CONFIG_CGROUP_SCHED */
 689
 690 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
 691 static inline struct task_group *task_group(struct task_struct *p)
 692 {
 693         return NULL;
 694 }
 695
 696 #endif /* CONFIG_CGROUP_SCHED */
 697
 698 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
 699 {
 700         set_task_rq(p, cpu);
 701 #ifdef CONFIG_SMP
 702         /*
 703          * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
 704          * successfuly executed on another CPU. We must ensure that updates of
 705          * per-task data have been completed by this moment.
 706          */
 707         smp_wmb();
 708         task_thread_info(p)->cpu = cpu;
 709 #endif
 710 }
 711
 712 /*
 713  * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
 714  */
 715 #ifdef CONFIG_SCHED_DEBUG
 716 # include <linux/static_key.h>
 717 # define const_debug __read_mostly
 718 #else
 719 # define const_debug const
 720 #endif
 721
 722 extern const_debug unsigned int sysctl_sched_features;
 723
 724 #define SCHED_FEAT(name, enabled)       \
 725         __SCHED_FEAT_##name ,
 726
 727 enum {
 728 #include "features.h"
 729         __SCHED_FEAT_NR,
 730 };
 731
 732 #undef SCHED_FEAT
 733
 734 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
 735 static __always_inline bool static_branch__true(struct static_key *key)
 736 {
 737         return static_key_true(key); /* Not out of line branch. */
 738 }
 739
 740 static __always_inline bool static_branch__false(struct static_key *key)
 741 {
 742         return static_key_false(key); /* Out of line branch. */
 743 }
 744
 745 #define SCHED_FEAT(name, enabled)                                       \
 746 static __always_inline bool static_branch_##name(struct static_key *key) \
 747 {                                                                       \
 748         return static_branch__##enabled(key);                           \
 749 }
 750
 751 #include "features.h"
 752
 753 #undef SCHED_FEAT
 754
 755 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
 756 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
 757 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
 758 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
 759 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
 760
 761 #ifdef CONFIG_NUMA_BALANCING
 762 #define sched_feat_numa(x) sched_feat(x)
 763 #ifdef CONFIG_SCHED_DEBUG
 764 #define numabalancing_enabled sched_feat_numa(NUMA)
 765 #else
 766 extern bool numabalancing_enabled;
 767 #endif /* CONFIG_SCHED_DEBUG */
 768 #else
 769 #define sched_feat_numa(x) (0)
 770 #define numabalancing_enabled (0)
 771 #endif /* CONFIG_NUMA_BALANCING */
 772
 773 static inline u64 global_rt_period(void)
 774 {
 775         return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
 776 }
 777
 778 static inline u64 global_rt_runtime(void)
 779 {
 780         if (sysctl_sched_rt_runtime < 0)
 781                 return RUNTIME_INF;
 782
 783         return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
 784 }
 785
 786
 787
 788 static inline int task_current(struct rq *rq, struct task_struct *p)
 789 {
 790         return rq->curr == p;
 791 }
 792
 793 static inline int task_running(struct rq *rq, struct task_struct *p)
 794 {
 795 #ifdef CONFIG_SMP
 796         return p->on_cpu;
 797 #else
 798         return task_current(rq, p);
 799 #endif
 800 }
 801
 802
 803 #ifndef prepare_arch_switch
 804 # define prepare_arch_switch(next)      do { } while (0)
 805 #endif
 806 #ifndef finish_arch_switch
 807 # define finish_arch_switch(prev)       do { } while (0)
 808 #endif
 809 #ifndef finish_arch_post_lock_switch
 810 # define finish_arch_post_lock_switch() do { } while (0)
 811 #endif
 812
 813 #ifndef __ARCH_WANT_UNLOCKED_CTXSW
 814 static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
 815 {
 816 #ifdef CONFIG_SMP
 817         /*
 818          * We can optimise this out completely for !SMP, because the
 819          * SMP rebalancing from interrupt is the only thing that cares
 820          * here.
 821          */
 822         next->on_cpu = 1;
 823 #endif
 824 }
 825
 826 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
 827 {
 828 #ifdef CONFIG_SMP
 829         /*
 830          * After ->on_cpu is cleared, the task can be moved to a different CPU.
 831          * We must ensure this doesn't happen until the switch is completely
 832          * finished.
 833          */
 834         smp_wmb();
 835         prev->on_cpu = 0;
 836 #endif
 837 #ifdef CONFIG_DEBUG_SPINLOCK
 838         /* this is a valid case when another task releases the spinlock */
 839         rq->lock.owner = current;
 840 #endif
 841         /*
 842          * If we are tracking spinlock dependencies then we have to
 843          * fix up the runqueue lock - which gets 'carried over' from
 844          * prev into current:
 845          */
 846         spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
 847
 848         raw_spin_unlock_irq(&rq->lock);
 849 }
 850
 851 #else /* __ARCH_WANT_UNLOCKED_CTXSW */
 852 static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
 853 {
 854 #ifdef CONFIG_SMP
 855         /*
 856          * We can optimise this out completely for !SMP, because the
 857          * SMP rebalancing from interrupt is the only thing that cares
 858          * here.
 859          */
 860         next->on_cpu = 1;
 861 #endif
 862         raw_spin_unlock(&rq->lock);
 863 }
 864
 865 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
 866 {
 867 #ifdef CONFIG_SMP
 868         /*
 869          * After ->on_cpu is cleared, the task can be moved to a different CPU.
 870          * We must ensure this doesn't happen until the switch is completely
 871          * finished.
 872          */
 873         smp_wmb();
 874         prev->on_cpu = 0;
 875 #endif
 876         local_irq_enable();
 877 }
 878 #endif /* __ARCH_WANT_UNLOCKED_CTXSW */
 879
 880 /*
 881  * wake flags
 882  */
 883 #define WF_SYNC         0x01            /* waker goes to sleep after wakeup */
 884 #define WF_FORK         0x02            /* child wakeup after fork */
 885 #define WF_MIGRATED     0x4             /* internal use, task got migrated */
 886
 887 static inline void update_load_add(struct load_weight *lw, unsigned long inc)
 888 {
 889         lw->weight += inc;
 890         lw->inv_weight = 0;
 891 }
 892
 893 static inline void update_load_sub(struct load_weight *lw, unsigned long dec)
 894 {
 895         lw->weight -= dec;
 896         lw->inv_weight = 0;
 897 }
 898
 899 static inline void update_load_set(struct load_weight *lw, unsigned long w)
 900 {
 901         lw->weight = w;
 902         lw->inv_weight = 0;
 903 }
 904
 905 /*
 906  * To aid in avoiding the subversion of "niceness" due to uneven distribution
 907  * of tasks with abnormal "nice" values across CPUs the contribution that
 908  * each task makes to its run queue's load is weighted according to its
 909  * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
 910  * scaled version of the new time slice allocation that they receive on time
 911  * slice expiry etc.
 912  */
 913
 914 #define WEIGHT_IDLEPRIO                3
 915 #define WMULT_IDLEPRIO         1431655765
 916
 917 /*
 918  * Nice levels are multiplicative, with a gentle 10% change for every
 919  * nice level changed. I.e. when a CPU-bound task goes from nice 0 to
 920  * nice 1, it will get ~10% less CPU time than another CPU-bound task
 921  * that remained on nice 0.
 922  *
 923  * The "10% effect" is relative and cumulative: from _any_ nice level,
 924  * if you go up 1 level, it's -10% CPU usage, if you go down 1 level
 925  * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25.
 926  * If a task goes up by ~10% and another task goes down by ~10% then
 927  * the relative distance between them is ~25%.)
 928  */
 929 static const int prio_to_weight[40] = {
 930  /* -20 */     88761,     71755,     56483,     46273,     36291,
 931  /* -15 */     29154,     23254,     18705,     14949,     11916,
 932  /* -10 */      9548,      7620,      6100,      4904,      3906,
 933  /*  -5 */      3121,      2501,      1991,      1586,      1277,
 934  /*   0 */      1024,       820,       655,       526,       423,
 935  /*   5 */       335,       272,       215,       172,       137,
 936  /*  10 */       110,        87,        70,        56,        45,
 937  /*  15 */        36,        29,        23,        18,        15,
 938 };
 939
 940 /*
 941  * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated.
 942  *
 943  * In cases where the weight does not change often, we can use the
 944  * precalculated inverse to speed up arithmetics by turning divisions
 945  * into multiplications:
 946  */
 947 static const u32 prio_to_wmult[40] = {
 948  /* -20 */     48388,     59856,     76040,     92818,    118348,
 949  /* -15 */    147320,    184698,    229616,    287308,    360437,
 950  /* -10 */    449829,    563644,    704093,    875809,   1099582,
 951  /*  -5 */   1376151,   1717300,   2157191,   2708050,   3363326,
 952  /*   0 */   4194304,   5237765,   6557202,   8165337,  10153587,
 953  /*   5 */  12820798,  15790321,  19976592,  24970740,  31350126,
 954  /*  10 */  39045157,  49367440,  61356676,  76695844,  95443717,
 955  /*  15 */ 119304647, 148102320, 186737708, 238609294, 286331153,
 956 };
 957
 958 #define ENQUEUE_WAKEUP          1
 959 #define ENQUEUE_HEAD            2
 960 #ifdef CONFIG_SMP
 961 #define ENQUEUE_WAKING          4       /* sched_class::task_waking was called */
 962 #else
 963 #define ENQUEUE_WAKING          0
 964 #endif
 965
 966 #define DEQUEUE_SLEEP           1
 967
 968 struct sched_class {
 969         const struct sched_class *next;
 970
 971         void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
 972         void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
 973         void (*yield_task) (struct rq *rq);
 974         bool (*yield_to_task) (struct rq *rq, struct task_struct *p, bool preempt);
 975
 976         void (*check_preempt_curr) (struct rq *rq, struct task_struct *p, int flags);
 977
 978         struct task_struct * (*pick_next_task) (struct rq *rq);
 979         void (*put_prev_task) (struct rq *rq, struct task_struct *p);
 980
 981 #ifdef CONFIG_SMP
 982         int  (*select_task_rq)(struct task_struct *p, int sd_flag, int flags);
 983         void (*migrate_task_rq)(struct task_struct *p, int next_cpu);
 984
 985         void (*pre_schedule) (struct rq *this_rq, struct task_struct *task);
 986         void (*post_schedule) (struct rq *this_rq);
 987         void (*task_waking) (struct task_struct *task);
 988         void (*task_woken) (struct rq *this_rq, struct task_struct *task);
 989
 990         void (*set_cpus_allowed)(struct task_struct *p,
 991                                  const struct cpumask *newmask);
 992
 993         void (*rq_online)(struct rq *rq);
 994         void (*rq_offline)(struct rq *rq);
 995 #endif
 996
 997         void (*set_curr_task) (struct rq *rq);
 998         void (*task_tick) (struct rq *rq, struct task_struct *p, int queued);
 999         void (*task_fork) (struct task_struct *p);
1000
1001         void (*switched_from) (struct rq *this_rq, struct task_struct *task);
1002         void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1003         void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1004                              int oldprio);
1005
1006         unsigned int (*get_rr_interval) (struct rq *rq,
1007                                          struct task_struct *task);
1008
1009 #ifdef CONFIG_FAIR_GROUP_SCHED
1010         void (*task_move_group) (struct task_struct *p, int on_rq);
1011 #endif
1012 };
1013
1014 #define sched_class_highest (&stop_sched_class)
1015 #define for_each_class(class) \
1016    for (class = sched_class_highest; class; class = class->next)
1017
1018 extern const struct sched_class stop_sched_class;
1019 extern const struct sched_class rt_sched_class;
1020 extern const struct sched_class fair_sched_class;
1021 extern const struct sched_class idle_sched_class;
1022
1023
1024 #ifdef CONFIG_SMP
1025
1026 extern void update_group_power(struct sched_domain *sd, int cpu);
1027
1028 extern void trigger_load_balance(struct rq *rq, int cpu);
1029 extern void idle_balance(int this_cpu, struct rq *this_rq);
1030
1031 /*
1032  * Only depends on SMP, FAIR_GROUP_SCHED may be removed when runnable_avg
1033  * becomes useful in lb
1034  */
1035 #if defined(CONFIG_FAIR_GROUP_SCHED)
1036 extern void idle_enter_fair(struct rq *this_rq);
1037 extern void idle_exit_fair(struct rq *this_rq);
1038 #else
1039 static inline void idle_enter_fair(struct rq *this_rq) {}
1040 static inline void idle_exit_fair(struct rq *this_rq) {}
1041 #endif
1042
1043 #else   /* CONFIG_SMP */
1044
1045 static inline void idle_balance(int cpu, struct rq *rq)
1046 {
1047 }
1048
1049 #endif
1050
1051 extern void sysrq_sched_debug_show(void);
1052 extern void sched_init_granularity(void);
1053 extern void update_max_interval(void);
1054 extern int update_runtime(struct notifier_block *nfb, unsigned long action, void *hcpu);
1055 extern void init_sched_rt_class(void);
1056 extern void init_sched_fair_class(void);
1057
1058 extern void resched_task(struct task_struct *p);
1059 extern void resched_cpu(int cpu);
1060
1061 extern struct rt_bandwidth def_rt_bandwidth;
1062 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1063
1064 extern void update_idle_cpu_load(struct rq *this_rq);
1065
1066 #ifdef CONFIG_PARAVIRT
1067 static inline u64 steal_ticks(u64 steal)
1068 {
1069         if (unlikely(steal > NSEC_PER_SEC))
1070                 return div_u64(steal, TICK_NSEC);
1071
1072         return __iter_div_u64_rem(steal, TICK_NSEC, &steal);
1073 }
1074 #endif
1075
1076 static inline void inc_nr_running(struct rq *rq)
1077 {
1078         rq->nr_running++;
1079
1080 #ifdef CONFIG_NO_HZ_FULL
1081         if (rq->nr_running == 2) {
1082                 if (tick_nohz_full_cpu(rq->cpu)) {
1083                         /* Order rq->nr_running write against the IPI */
1084                         smp_wmb();
1085                         smp_send_reschedule(rq->cpu);
1086                 }
1087        }
1088 #endif
1089 }
1090
1091 static inline void dec_nr_running(struct rq *rq)
1092 {
1093         rq->nr_running--;
1094 }
1095
1096 static inline void rq_last_tick_reset(struct rq *rq)
1097 {
1098 #ifdef CONFIG_NO_HZ_FULL
1099         rq->last_sched_tick = jiffies;
1100 #endif
1101 }
1102
1103 extern void update_rq_clock(struct rq *rq);
1104
1105 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1106 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1107
1108 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1109
1110 extern const_debug unsigned int sysctl_sched_time_avg;
1111 extern const_debug unsigned int sysctl_sched_nr_migrate;
1112 extern const_debug unsigned int sysctl_sched_migration_cost;
1113
1114 static inline u64 sched_avg_period(void)
1115 {
1116         return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2;
1117 }
1118
1119 #ifdef CONFIG_SCHED_HRTICK
1120
1121 /*
1122  * Use hrtick when:
1123  *  - enabled by features
1124  *  - hrtimer is actually high res
1125  */
1126 static inline int hrtick_enabled(struct rq *rq)
1127 {
1128         if (!sched_feat(HRTICK))
1129                 return 0;
1130         if (!cpu_active(cpu_of(rq)))
1131                 return 0;
1132         return hrtimer_is_hres_active(&rq->hrtick_timer);
1133 }
1134
1135 void hrtick_start(struct rq *rq, u64 delay);
1136
1137 #else
1138
1139 static inline int hrtick_enabled(struct rq *rq)
1140 {
1141         return 0;
1142 }
1143
1144 #endif /* CONFIG_SCHED_HRTICK */
1145
1146 #ifdef CONFIG_SMP
1147 extern void sched_avg_update(struct rq *rq);
1148 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1149 {
1150         rq->rt_avg += rt_delta;
1151         sched_avg_update(rq);
1152 }
1153 #else
1154 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { }
1155 static inline void sched_avg_update(struct rq *rq) { }
1156 #endif
1157
1158 extern void start_bandwidth_timer(struct hrtimer *period_timer, ktime_t period);
1159
1160 #ifdef CONFIG_SMP
1161 #ifdef CONFIG_PREEMPT
1162
1163 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1164
1165 /*
1166  * fair double_lock_balance: Safely acquires both rq->locks in a fair
1167  * way at the expense of forcing extra atomic operations in all
1168  * invocations.  This assures that the double_lock is acquired using the
1169  * same underlying policy as the spinlock_t on this architecture, which
1170  * reduces latency compared to the unfair variant below.  However, it
1171  * also adds more overhead and therefore may reduce throughput.
1172  */
1173 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1174         __releases(this_rq->lock)
1175         __acquires(busiest->lock)
1176         __acquires(this_rq->lock)
1177 {
1178         raw_spin_unlock(&this_rq->lock);
1179         double_rq_lock(this_rq, busiest);
1180
1181         return 1;
1182 }
1183
1184 #else
1185 /*
1186  * Unfair double_lock_balance: Optimizes throughput at the expense of
1187  * latency by eliminating extra atomic operations when the locks are
1188  * already in proper order on entry.  This favors lower cpu-ids and will
1189  * grant the double lock to lower cpus over higher ids under contention,
1190  * regardless of entry order into the function.
1191  */
1192 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1193         __releases(this_rq->lock)
1194         __acquires(busiest->lock)
1195         __acquires(this_rq->lock)
1196 {
1197         int ret = 0;
1198
1199         if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1200                 if (busiest < this_rq) {
1201                         raw_spin_unlock(&this_rq->lock);
1202                         raw_spin_lock(&busiest->lock);
1203                         raw_spin_lock_nested(&this_rq->lock,
1204                                               SINGLE_DEPTH_NESTING);
1205                         ret = 1;
1206                 } else
1207                         raw_spin_lock_nested(&busiest->lock,
1208                                               SINGLE_DEPTH_NESTING);
1209         }
1210         return ret;
1211 }
1212
1213 #endif /* CONFIG_PREEMPT */
1214
1215 /*
1216  * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1217  */
1218 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1219 {
1220         if (unlikely(!irqs_disabled())) {
1221                 /* printk() doesn't work good under rq->lock */
1222                 raw_spin_unlock(&this_rq->lock);
1223                 BUG_ON(1);
1224         }
1225
1226         return _double_lock_balance(this_rq, busiest);
1227 }
1228
1229 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1230         __releases(busiest->lock)
1231 {
1232         raw_spin_unlock(&busiest->lock);
1233         lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1234 }
1235
1236 /*
1237  * double_rq_lock - safely lock two runqueues
1238  *
1239  * Note this does not disable interrupts like task_rq_lock,
1240  * you need to do so manually before calling.
1241  */
1242 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1243         __acquires(rq1->lock)
1244         __acquires(rq2->lock)
1245 {
1246         BUG_ON(!irqs_disabled());
1247         if (rq1 == rq2) {
1248                 raw_spin_lock(&rq1->lock);
1249                 __acquire(rq2->lock);   /* Fake it out ;) */
1250         } else {
1251                 if (rq1 < rq2) {
1252                         raw_spin_lock(&rq1->lock);
1253                         raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1254                 } else {
1255                         raw_spin_lock(&rq2->lock);
1256                         raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1257                 }
1258         }
1259 }
1260
1261 /*
1262  * double_rq_unlock - safely unlock two runqueues
1263  *
1264  * Note this does not restore interrupts like task_rq_unlock,
1265  * you need to do so manually after calling.
1266  */
1267 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1268         __releases(rq1->lock)
1269         __releases(rq2->lock)
1270 {
1271         raw_spin_unlock(&rq1->lock);
1272         if (rq1 != rq2)
1273                 raw_spin_unlock(&rq2->lock);
1274         else
1275                 __release(rq2->lock);
1276 }
1277
1278 #else /* CONFIG_SMP */
1279
1280 /*
1281  * double_rq_lock - safely lock two runqueues
1282  *
1283  * Note this does not disable interrupts like task_rq_lock,
1284  * you need to do so manually before calling.
1285  */
1286 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1287         __acquires(rq1->lock)
1288         __acquires(rq2->lock)
1289 {
1290         BUG_ON(!irqs_disabled());
1291         BUG_ON(rq1 != rq2);
1292         raw_spin_lock(&rq1->lock);
1293         __acquire(rq2->lock);   /* Fake it out ;) */
1294 }
1295
1296 /*
1297  * double_rq_unlock - safely unlock two runqueues
1298  *
1299  * Note this does not restore interrupts like task_rq_unlock,
1300  * you need to do so manually after calling.
1301  */
1302 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1303         __releases(rq1->lock)
1304         __releases(rq2->lock)
1305 {
1306         BUG_ON(rq1 != rq2);
1307         raw_spin_unlock(&rq1->lock);
1308         __release(rq2->lock);
1309 }
1310
1311 #endif
1312
1313 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
1314 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
1315 extern void print_cfs_stats(struct seq_file *m, int cpu);
1316 extern void print_rt_stats(struct seq_file *m, int cpu);
1317
1318 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
1319 extern void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq);
1320
1321 extern void account_cfs_bandwidth_used(int enabled, int was_enabled);
1322
1323 #ifdef CONFIG_NO_HZ_COMMON
1324 enum rq_nohz_flag_bits {
1325         NOHZ_TICK_STOPPED,
1326         NOHZ_BALANCE_KICK,
1327 };
1328
1329 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
1330 #endif
1331
1332 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1333
1334 DECLARE_PER_CPU(u64, cpu_hardirq_time);
1335 DECLARE_PER_CPU(u64, cpu_softirq_time);
1336
1337 #ifndef CONFIG_64BIT
1338 DECLARE_PER_CPU(seqcount_t, irq_time_seq);
1339
1340 static inline void irq_time_write_begin(void)
1341 {
1342         __this_cpu_inc(irq_time_seq.sequence);
1343         smp_wmb();
1344 }
1345
1346 static inline void irq_time_write_end(void)
1347 {
1348         smp_wmb();
1349         __this_cpu_inc(irq_time_seq.sequence);
1350 }
1351
1352 static inline u64 irq_time_read(int cpu)
1353 {
1354         u64 irq_time;
1355         unsigned seq;
1356
1357         do {
1358                 seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu));
1359                 irq_time = per_cpu(cpu_softirq_time, cpu) +
1360                            per_cpu(cpu_hardirq_time, cpu);
1361         } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq));
1362
1363         return irq_time;
1364 }
1365 #else /* CONFIG_64BIT */
1366 static inline void irq_time_write_begin(void)
1367 {
1368 }
1369
1370 static inline void irq_time_write_end(void)
1371 {
1372 }
1373
1374 static inline u64 irq_time_read(int cpu)
1375 {
1376         return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu);
1377 }
1378 #endif /* CONFIG_64BIT */
1379 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */