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