1 2#include <linux/sched.h> 3#include <linux/sched/sysctl.h> 4#include <linux/sched/rt.h> 5#include <linux/sched/deadline.h> 6#include <linux/mutex.h> 7#include <linux/spinlock.h> 8#include <linux/stop_machine.h> 9#include <linux/irq_work.h> 10#include <linux/tick.h> 11#include <linux/slab.h> 12 13#include "cpupri.h" 14#include "cpudeadline.h" 15#include "cpuacct.h" 16 17struct rq; 18struct cpuidle_state; 19 20/* task_struct::on_rq states: */ 21#define TASK_ON_RQ_QUEUED 1 22#define TASK_ON_RQ_MIGRATING 2 23 24extern __read_mostly int scheduler_running; 25 26extern unsigned long calc_load_update; 27extern atomic_long_t calc_load_tasks; 28 29extern long calc_load_fold_active(struct rq *this_rq); 30extern void update_cpu_load_active(struct rq *this_rq); 31 32/* 33 * Helpers for converting nanosecond timing to jiffy resolution 34 */ 35#define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) 36 37/* 38 * Increase resolution of nice-level calculations for 64-bit architectures. 39 * The extra resolution improves shares distribution and load balancing of 40 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup 41 * hierarchies, especially on larger systems. This is not a user-visible change 42 * and does not change the user-interface for setting shares/weights. 43 * 44 * We increase resolution only if we have enough bits to allow this increased 45 * resolution (i.e. BITS_PER_LONG > 32). The costs for increasing resolution 46 * when BITS_PER_LONG <= 32 are pretty high and the returns do not justify the 47 * increased costs. 48 */ 49#if 0 /* BITS_PER_LONG > 32 -- currently broken: it increases power usage under light load */ 50# define SCHED_LOAD_RESOLUTION 10 51# define scale_load(w) ((w) << SCHED_LOAD_RESOLUTION) 52# define scale_load_down(w) ((w) >> SCHED_LOAD_RESOLUTION) 53#else 54# define SCHED_LOAD_RESOLUTION 0 55# define scale_load(w) (w) 56# define scale_load_down(w) (w) 57#endif 58 59#define SCHED_LOAD_SHIFT (10 + SCHED_LOAD_RESOLUTION) 60#define SCHED_LOAD_SCALE (1L << SCHED_LOAD_SHIFT) 61 62#define NICE_0_LOAD SCHED_LOAD_SCALE 63#define NICE_0_SHIFT SCHED_LOAD_SHIFT 64 65/* 66 * Single value that decides SCHED_DEADLINE internal math precision. 67 * 10 -> just above 1us 68 * 9 -> just above 0.5us 69 */ 70#define DL_SCALE (10) 71 72/* 73 * These are the 'tuning knobs' of the scheduler: 74 */ 75 76/* 77 * single value that denotes runtime == period, ie unlimited time. 78 */ 79#define RUNTIME_INF ((u64)~0ULL) 80 81static inline int fair_policy(int policy) 82{ 83 return policy == SCHED_NORMAL || policy == SCHED_BATCH; 84} 85 86static inline int rt_policy(int policy) 87{ 88 return policy == SCHED_FIFO || policy == SCHED_RR; 89} 90 91static inline int dl_policy(int policy) 92{ 93 return policy == SCHED_DEADLINE; 94} 95 96static inline int task_has_rt_policy(struct task_struct *p) 97{ 98 return rt_policy(p->policy); 99} 100 101static inline int task_has_dl_policy(struct task_struct *p) 102{ 103 return dl_policy(p->policy); 104} 105 106static inline bool dl_time_before(u64 a, u64 b) 107{ 108 return (s64)(a - b) < 0; 109} 110 111/* 112 * Tells if entity @a should preempt entity @b. 113 */ 114static inline bool 115dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b) 116{ 117 return dl_time_before(a->deadline, b->deadline); 118} 119 120/* 121 * This is the priority-queue data structure of the RT scheduling class: 122 */ 123struct rt_prio_array { 124 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ 125 struct list_head queue[MAX_RT_PRIO]; 126}; 127 128struct rt_bandwidth { 129 /* nests inside the rq lock: */ 130 raw_spinlock_t rt_runtime_lock; 131 ktime_t rt_period; 132 u64 rt_runtime; 133 struct hrtimer rt_period_timer; 134}; 135 136void __dl_clear_params(struct task_struct *p); 137 138/* 139 * To keep the bandwidth of -deadline tasks and groups under control 140 * we need some place where: 141 * - store the maximum -deadline bandwidth of the system (the group); 142 * - cache the fraction of that bandwidth that is currently allocated. 143 * 144 * This is all done in the data structure below. It is similar to the 145 * one used for RT-throttling (rt_bandwidth), with the main difference 146 * that, since here we are only interested in admission control, we 147 * do not decrease any runtime while the group "executes", neither we 148 * need a timer to replenish it. 149 * 150 * With respect to SMP, the bandwidth is given on a per-CPU basis, 151 * meaning that: 152 * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU; 153 * - dl_total_bw array contains, in the i-eth element, the currently 154 * allocated bandwidth on the i-eth CPU. 155 * Moreover, groups consume bandwidth on each CPU, while tasks only 156 * consume bandwidth on the CPU they're running on. 157 * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw 158 * that will be shown the next time the proc or cgroup controls will 159 * be red. It on its turn can be changed by writing on its own 160 * control. 161 */ 162struct dl_bandwidth { 163 raw_spinlock_t dl_runtime_lock; 164 u64 dl_runtime; 165 u64 dl_period; 166}; 167 168static inline int dl_bandwidth_enabled(void) 169{ 170 return sysctl_sched_rt_runtime >= 0; 171} 172 173extern struct dl_bw *dl_bw_of(int i); 174 175struct dl_bw { 176 raw_spinlock_t lock; 177 u64 bw, total_bw; 178}; 179 180static inline 181void __dl_clear(struct dl_bw *dl_b, u64 tsk_bw) 182{ 183 dl_b->total_bw -= tsk_bw; 184} 185 186static inline 187void __dl_add(struct dl_bw *dl_b, u64 tsk_bw) 188{ 189 dl_b->total_bw += tsk_bw; 190} 191 192static inline 193bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw) 194{ 195 return dl_b->bw != -1 && 196 dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw; 197} 198 199extern struct mutex sched_domains_mutex; 200 201#ifdef CONFIG_CGROUP_SCHED 202 203#include <linux/cgroup.h> 204 205struct cfs_rq; 206struct rt_rq; 207 208extern struct list_head task_groups; 209 210struct cfs_bandwidth { 211#ifdef CONFIG_CFS_BANDWIDTH 212 raw_spinlock_t lock; 213 ktime_t period; 214 u64 quota, runtime; 215 s64 hierarchical_quota; 216 u64 runtime_expires; 217 218 int idle, timer_active; 219 struct hrtimer period_timer, slack_timer; 220 struct list_head throttled_cfs_rq; 221 222 /* statistics */ 223 int nr_periods, nr_throttled; 224 u64 throttled_time; 225#endif 226}; 227 228/* task group related information */ 229struct task_group { 230 struct cgroup_subsys_state css; 231 232#ifdef CONFIG_FAIR_GROUP_SCHED 233 /* schedulable entities of this group on each cpu */ 234 struct sched_entity **se; 235 /* runqueue "owned" by this group on each cpu */ 236 struct cfs_rq **cfs_rq; 237 unsigned long shares; 238 239#ifdef CONFIG_SMP 240 atomic_long_t load_avg; 241 atomic_t runnable_avg; 242#endif 243#endif 244 245#ifdef CONFIG_RT_GROUP_SCHED 246 struct sched_rt_entity **rt_se; 247 struct rt_rq **rt_rq; 248 249 struct rt_bandwidth rt_bandwidth; 250#endif 251 252 struct rcu_head rcu; 253 struct list_head list; 254 255 struct task_group *parent; 256 struct list_head siblings; 257 struct list_head children; 258 259#ifdef CONFIG_SCHED_AUTOGROUP 260 struct autogroup *autogroup; 261#endif 262 263 struct cfs_bandwidth cfs_bandwidth; 264}; 265 266#ifdef CONFIG_FAIR_GROUP_SCHED 267#define ROOT_TASK_GROUP_LOAD NICE_0_LOAD 268 269/* 270 * A weight of 0 or 1 can cause arithmetics problems. 271 * A weight of a cfs_rq is the sum of weights of which entities 272 * are queued on this cfs_rq, so a weight of a entity should not be 273 * too large, so as the shares value of a task group. 274 * (The default weight is 1024 - so there's no practical 275 * limitation from this.) 276 */ 277#define MIN_SHARES (1UL << 1) 278#define MAX_SHARES (1UL << 18) 279#endif 280 281typedef int (*tg_visitor)(struct task_group *, void *); 282 283extern int walk_tg_tree_from(struct task_group *from, 284 tg_visitor down, tg_visitor up, void *data); 285 286/* 287 * Iterate the full tree, calling @down when first entering a node and @up when 288 * leaving it for the final time. 289 * 290 * Caller must hold rcu_lock or sufficient equivalent. 291 */ 292static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) 293{ 294 return walk_tg_tree_from(&root_task_group, down, up, data); 295} 296 297extern int tg_nop(struct task_group *tg, void *data); 298 299extern void free_fair_sched_group(struct task_group *tg); 300extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent); 301extern void unregister_fair_sched_group(struct task_group *tg, int cpu); 302extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, 303 struct sched_entity *se, int cpu, 304 struct sched_entity *parent); 305extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b); 306extern int sched_group_set_shares(struct task_group *tg, unsigned long shares); 307 308extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b); 309extern void __start_cfs_bandwidth(struct cfs_bandwidth *cfs_b, bool force); 310extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq); 311 312extern void free_rt_sched_group(struct task_group *tg); 313extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent); 314extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, 315 struct sched_rt_entity *rt_se, int cpu, 316 struct sched_rt_entity *parent); 317 318extern struct task_group *sched_create_group(struct task_group *parent); 319extern void sched_online_group(struct task_group *tg, 320 struct task_group *parent); 321extern void sched_destroy_group(struct task_group *tg); 322extern void sched_offline_group(struct task_group *tg); 323 324extern void sched_move_task(struct task_struct *tsk); 325 326#ifdef CONFIG_FAIR_GROUP_SCHED 327extern int sched_group_set_shares(struct task_group *tg, unsigned long shares); 328#endif 329 330#else /* CONFIG_CGROUP_SCHED */ 331 332struct cfs_bandwidth { }; 333 334#endif /* CONFIG_CGROUP_SCHED */ 335 336/* CFS-related fields in a runqueue */ 337struct cfs_rq { 338 struct load_weight load; 339 unsigned int nr_running, h_nr_running; 340 341 u64 exec_clock; 342 u64 min_vruntime; 343#ifndef CONFIG_64BIT 344 u64 min_vruntime_copy; 345#endif 346 347 struct rb_root tasks_timeline; 348 struct rb_node *rb_leftmost; 349 350 /* 351 * 'curr' points to currently running entity on this cfs_rq. 352 * It is set to NULL otherwise (i.e when none are currently running). 353 */ 354 struct sched_entity *curr, *next, *last, *skip; 355 356#ifdef CONFIG_SCHED_DEBUG 357 unsigned int nr_spread_over; 358#endif 359 360#ifdef CONFIG_SMP 361 /* 362 * CFS Load tracking 363 * Under CFS, load is tracked on a per-entity basis and aggregated up. 364 * This allows for the description of both thread and group usage (in 365 * the FAIR_GROUP_SCHED case). 366 * runnable_load_avg is the sum of the load_avg_contrib of the 367 * sched_entities on the rq. 368 * blocked_load_avg is similar to runnable_load_avg except that its 369 * the blocked sched_entities on the rq. 370 * utilization_load_avg is the sum of the average running time of the 371 * sched_entities on the rq. 372 */ 373 unsigned long runnable_load_avg, blocked_load_avg, utilization_load_avg; 374 atomic64_t decay_counter; 375 u64 last_decay; 376 atomic_long_t removed_load; 377 378#ifdef CONFIG_FAIR_GROUP_SCHED 379 /* Required to track per-cpu representation of a task_group */ 380 u32 tg_runnable_contrib; 381 unsigned long tg_load_contrib; 382 383 /* 384 * h_load = weight * f(tg) 385 * 386 * Where f(tg) is the recursive weight fraction assigned to 387 * this group. 388 */ 389 unsigned long h_load; 390 u64 last_h_load_update; 391 struct sched_entity *h_load_next; 392#endif /* CONFIG_FAIR_GROUP_SCHED */ 393#endif /* CONFIG_SMP */ 394 395#ifdef CONFIG_FAIR_GROUP_SCHED 396 struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ 397 398 /* 399 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in 400 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities 401 * (like users, containers etc.) 402 * 403 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This 404 * list is used during load balance. 405 */ 406 int on_list; 407 struct list_head leaf_cfs_rq_list; 408 struct task_group *tg; /* group that "owns" this runqueue */ 409 410#ifdef CONFIG_CFS_BANDWIDTH 411 int runtime_enabled; 412 u64 runtime_expires; 413 s64 runtime_remaining; 414 415 u64 throttled_clock, throttled_clock_task; 416 u64 throttled_clock_task_time; 417 int throttled, throttle_count; 418 struct list_head throttled_list; 419#endif /* CONFIG_CFS_BANDWIDTH */ 420#endif /* CONFIG_FAIR_GROUP_SCHED */ 421}; 422 423static inline int rt_bandwidth_enabled(void) 424{ 425 return sysctl_sched_rt_runtime >= 0; 426} 427 428/* RT IPI pull logic requires IRQ_WORK */ 429#ifdef CONFIG_IRQ_WORK 430# define HAVE_RT_PUSH_IPI 431#endif 432 433/* Real-Time classes' related field in a runqueue: */ 434struct rt_rq { 435 struct rt_prio_array active; 436 unsigned int rt_nr_running; 437#if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED 438 struct { 439 int curr; /* highest queued rt task prio */ 440#ifdef CONFIG_SMP 441 int next; /* next highest */ 442#endif 443 } highest_prio; 444#endif 445#ifdef CONFIG_SMP 446 unsigned long rt_nr_migratory; 447 unsigned long rt_nr_total; 448 int overloaded; 449 struct plist_head pushable_tasks; 450#ifdef HAVE_RT_PUSH_IPI 451 int push_flags; 452 int push_cpu; 453 struct irq_work push_work; 454 raw_spinlock_t push_lock; 455#endif 456#endif /* CONFIG_SMP */ 457 int rt_queued; 458 459 int rt_throttled; 460 u64 rt_time; 461 u64 rt_runtime; 462 /* Nests inside the rq lock: */ 463 raw_spinlock_t rt_runtime_lock; 464 465#ifdef CONFIG_RT_GROUP_SCHED 466 unsigned long rt_nr_boosted; 467 468 struct rq *rq; 469 struct task_group *tg; 470#endif 471}; 472 473/* Deadline class' related fields in a runqueue */ 474struct dl_rq { 475 /* runqueue is an rbtree, ordered by deadline */ 476 struct rb_root rb_root; 477 struct rb_node *rb_leftmost; 478 479 unsigned long dl_nr_running; 480 481#ifdef CONFIG_SMP 482 /* 483 * Deadline values of the currently executing and the 484 * earliest ready task on this rq. Caching these facilitates 485 * the decision wether or not a ready but not running task 486 * should migrate somewhere else. 487 */ 488 struct { 489 u64 curr; 490 u64 next; 491 } earliest_dl; 492 493 unsigned long dl_nr_migratory; 494 int overloaded; 495 496 /* 497 * Tasks on this rq that can be pushed away. They are kept in 498 * an rb-tree, ordered by tasks' deadlines, with caching 499 * of the leftmost (earliest deadline) element. 500 */ 501 struct rb_root pushable_dl_tasks_root; 502 struct rb_node *pushable_dl_tasks_leftmost; 503#else 504 struct dl_bw dl_bw; 505#endif 506}; 507 508#ifdef CONFIG_SMP 509 510/* 511 * We add the notion of a root-domain which will be used to define per-domain 512 * variables. Each exclusive cpuset essentially defines an island domain by 513 * fully partitioning the member cpus from any other cpuset. Whenever a new 514 * exclusive cpuset is created, we also create and attach a new root-domain 515 * object. 516 * 517 */ 518struct root_domain { 519 atomic_t refcount; 520 atomic_t rto_count; 521 struct rcu_head rcu; 522 cpumask_var_t span; 523 cpumask_var_t online; 524 525 /* Indicate more than one runnable task for any CPU */ 526 bool overload; 527 528 /* 529 * The bit corresponding to a CPU gets set here if such CPU has more 530 * than one runnable -deadline task (as it is below for RT tasks). 531 */ 532 cpumask_var_t dlo_mask; 533 atomic_t dlo_count; 534 struct dl_bw dl_bw; 535 struct cpudl cpudl; 536 537 /* 538 * The "RT overload" flag: it gets set if a CPU has more than 539 * one runnable RT task. 540 */ 541 cpumask_var_t rto_mask; 542 struct cpupri cpupri; 543}; 544 545extern struct root_domain def_root_domain; 546 547#endif /* CONFIG_SMP */ 548 549/* 550 * This is the main, per-CPU runqueue data structure. 551 * 552 * Locking rule: those places that want to lock multiple runqueues 553 * (such as the load balancing or the thread migration code), lock 554 * acquire operations must be ordered by ascending &runqueue. 555 */ 556struct rq { 557 /* runqueue lock: */ 558 raw_spinlock_t lock; 559 560 /* 561 * nr_running and cpu_load should be in the same cacheline because 562 * remote CPUs use both these fields when doing load calculation. 563 */ 564 unsigned int nr_running; 565#ifdef CONFIG_NUMA_BALANCING 566 unsigned int nr_numa_running; 567 unsigned int nr_preferred_running; 568#endif 569 #define CPU_LOAD_IDX_MAX 5 570 unsigned long cpu_load[CPU_LOAD_IDX_MAX]; 571 unsigned long last_load_update_tick; 572#ifdef CONFIG_NO_HZ_COMMON 573 u64 nohz_stamp; 574 unsigned long nohz_flags; 575#endif 576#ifdef CONFIG_NO_HZ_FULL 577 unsigned long last_sched_tick; 578#endif 579 /* capture load from *all* tasks on this cpu: */ 580 struct load_weight load; 581 unsigned long nr_load_updates; 582 u64 nr_switches; 583 584 struct cfs_rq cfs; 585 struct rt_rq rt; 586 struct dl_rq dl; 587 588#ifdef CONFIG_FAIR_GROUP_SCHED 589 /* list of leaf cfs_rq on this cpu: */ 590 struct list_head leaf_cfs_rq_list; 591 592 struct sched_avg avg; 593#endif /* CONFIG_FAIR_GROUP_SCHED */ 594 595 /* 596 * This is part of a global counter where only the total sum 597 * over all CPUs matters. A task can increase this counter on 598 * one CPU and if it got migrated afterwards it may decrease 599 * it on another CPU. Always updated under the runqueue lock: 600 */ 601 unsigned long nr_uninterruptible; 602 603 struct task_struct *curr, *idle, *stop; 604 unsigned long next_balance; 605 struct mm_struct *prev_mm; 606 607 unsigned int clock_skip_update; 608 u64 clock; 609 u64 clock_task; 610 611 atomic_t nr_iowait; 612 613#ifdef CONFIG_SMP 614 struct root_domain *rd; 615 struct sched_domain *sd; 616 617 unsigned long cpu_capacity; 618 unsigned long cpu_capacity_orig; 619 620 unsigned char idle_balance; 621 /* For active balancing */ 622 int post_schedule; 623 int active_balance; 624 int push_cpu; 625 struct cpu_stop_work active_balance_work; 626 /* cpu of this runqueue: */ 627 int cpu; 628 int online; 629 630 struct list_head cfs_tasks; 631 632 u64 rt_avg; 633 u64 age_stamp; 634 u64 idle_stamp; 635 u64 avg_idle; 636 637 /* This is used to determine avg_idle's max value */ 638 u64 max_idle_balance_cost; 639#endif 640 641#ifdef CONFIG_IRQ_TIME_ACCOUNTING 642 u64 prev_irq_time; 643#endif 644#ifdef CONFIG_PARAVIRT 645 u64 prev_steal_time; 646#endif 647#ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING 648 u64 prev_steal_time_rq; 649#endif 650 651 /* calc_load related fields */ 652 unsigned long calc_load_update; 653 long calc_load_active; 654 655#ifdef CONFIG_SCHED_HRTICK 656#ifdef CONFIG_SMP 657 int hrtick_csd_pending; 658 struct call_single_data hrtick_csd; 659#endif 660 struct hrtimer hrtick_timer; 661#endif 662 663#ifdef CONFIG_SCHEDSTATS 664 /* latency stats */ 665 struct sched_info rq_sched_info; 666 unsigned long long rq_cpu_time; 667 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ 668 669 /* sys_sched_yield() stats */ 670 unsigned int yld_count; 671 672 /* schedule() stats */ 673 unsigned int sched_count; 674 unsigned int sched_goidle; 675 676 /* try_to_wake_up() stats */ 677 unsigned int ttwu_count; 678 unsigned int ttwu_local; 679#endif 680 681#ifdef CONFIG_SMP 682 struct llist_head wake_list; 683#endif 684 685#ifdef CONFIG_CPU_IDLE 686 /* Must be inspected within a rcu lock section */ 687 struct cpuidle_state *idle_state; 688#endif 689}; 690 691static inline int cpu_of(struct rq *rq) 692{ 693#ifdef CONFIG_SMP 694 return rq->cpu; 695#else 696 return 0; 697#endif 698} 699 700DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); 701 702#define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) 703#define this_rq() this_cpu_ptr(&runqueues) 704#define task_rq(p) cpu_rq(task_cpu(p)) 705#define cpu_curr(cpu) (cpu_rq(cpu)->curr) 706#define raw_rq() raw_cpu_ptr(&runqueues) 707 708static inline u64 __rq_clock_broken(struct rq *rq) 709{ 710 return ACCESS_ONCE(rq->clock); 711} 712 713static inline u64 rq_clock(struct rq *rq) 714{ 715 lockdep_assert_held(&rq->lock); 716 return rq->clock; 717} 718 719static inline u64 rq_clock_task(struct rq *rq) 720{ 721 lockdep_assert_held(&rq->lock); 722 return rq->clock_task; 723} 724 725#define RQCF_REQ_SKIP 0x01 726#define RQCF_ACT_SKIP 0x02 727 728static inline void rq_clock_skip_update(struct rq *rq, bool skip) 729{ 730 lockdep_assert_held(&rq->lock); 731 if (skip) 732 rq->clock_skip_update |= RQCF_REQ_SKIP; 733 else 734 rq->clock_skip_update &= ~RQCF_REQ_SKIP; 735} 736 737#ifdef CONFIG_NUMA 738enum numa_topology_type { 739 NUMA_DIRECT, 740 NUMA_GLUELESS_MESH, 741 NUMA_BACKPLANE, 742}; 743extern enum numa_topology_type sched_numa_topology_type; 744extern int sched_max_numa_distance; 745extern bool find_numa_distance(int distance); 746#endif 747 748#ifdef CONFIG_NUMA_BALANCING 749/* The regions in numa_faults array from task_struct */ 750enum numa_faults_stats { 751 NUMA_MEM = 0, 752 NUMA_CPU, 753 NUMA_MEMBUF, 754 NUMA_CPUBUF 755}; 756extern void sched_setnuma(struct task_struct *p, int node); 757extern int migrate_task_to(struct task_struct *p, int cpu); 758extern int migrate_swap(struct task_struct *, struct task_struct *); 759#endif /* CONFIG_NUMA_BALANCING */ 760 761#ifdef CONFIG_SMP 762 763extern void sched_ttwu_pending(void); 764 765#define rcu_dereference_check_sched_domain(p) \ 766 rcu_dereference_check((p), \ 767 lockdep_is_held(&sched_domains_mutex)) 768 769/* 770 * The domain tree (rq->sd) is protected by RCU's quiescent state transition. 771 * See detach_destroy_domains: synchronize_sched for details. 772 * 773 * The domain tree of any CPU may only be accessed from within 774 * preempt-disabled sections. 775 */ 776#define for_each_domain(cpu, __sd) \ 777 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \ 778 __sd; __sd = __sd->parent) 779 780#define for_each_lower_domain(sd) for (; sd; sd = sd->child) 781 782/** 783 * highest_flag_domain - Return highest sched_domain containing flag. 784 * @cpu: The cpu whose highest level of sched domain is to 785 * be returned. 786 * @flag: The flag to check for the highest sched_domain 787 * for the given cpu. 788 * 789 * Returns the highest sched_domain of a cpu which contains the given flag. 790 */ 791static inline struct sched_domain *highest_flag_domain(int cpu, int flag) 792{ 793 struct sched_domain *sd, *hsd = NULL; 794 795 for_each_domain(cpu, sd) { 796 if (!(sd->flags & flag)) 797 break; 798 hsd = sd; 799 } 800 801 return hsd; 802} 803 804static inline struct sched_domain *lowest_flag_domain(int cpu, int flag) 805{ 806 struct sched_domain *sd; 807 808 for_each_domain(cpu, sd) { 809 if (sd->flags & flag) 810 break; 811 } 812 813 return sd; 814} 815 816DECLARE_PER_CPU(struct sched_domain *, sd_llc); 817DECLARE_PER_CPU(int, sd_llc_size); 818DECLARE_PER_CPU(int, sd_llc_id); 819DECLARE_PER_CPU(struct sched_domain *, sd_numa); 820DECLARE_PER_CPU(struct sched_domain *, sd_busy); 821DECLARE_PER_CPU(struct sched_domain *, sd_asym); 822 823struct sched_group_capacity { 824 atomic_t ref; 825 /* 826 * CPU capacity of this group, SCHED_LOAD_SCALE being max capacity 827 * for a single CPU. 828 */ 829 unsigned int capacity; 830 unsigned long next_update; 831 int imbalance; /* XXX unrelated to capacity but shared group state */ 832 /* 833 * Number of busy cpus in this group. 834 */ 835 atomic_t nr_busy_cpus; 836 837 unsigned long cpumask[0]; /* iteration mask */ 838}; 839 840struct sched_group { 841 struct sched_group *next; /* Must be a circular list */ 842 atomic_t ref; 843 844 unsigned int group_weight; 845 struct sched_group_capacity *sgc; 846 847 /* 848 * The CPUs this group covers. 849 * 850 * NOTE: this field is variable length. (Allocated dynamically 851 * by attaching extra space to the end of the structure, 852 * depending on how many CPUs the kernel has booted up with) 853 */ 854 unsigned long cpumask[0]; 855}; 856 857static inline struct cpumask *sched_group_cpus(struct sched_group *sg) 858{ 859 return to_cpumask(sg->cpumask); 860} 861 862/* 863 * cpumask masking which cpus in the group are allowed to iterate up the domain 864 * tree. 865 */ 866static inline struct cpumask *sched_group_mask(struct sched_group *sg) 867{ 868 return to_cpumask(sg->sgc->cpumask); 869} 870 871/** 872 * group_first_cpu - Returns the first cpu in the cpumask of a sched_group. 873 * @group: The group whose first cpu is to be returned. 874 */ 875static inline unsigned int group_first_cpu(struct sched_group *group) 876{ 877 return cpumask_first(sched_group_cpus(group)); 878} 879 880extern int group_balance_cpu(struct sched_group *sg); 881 882#else 883 884static inline void sched_ttwu_pending(void) { } 885 886#endif /* CONFIG_SMP */ 887 888#include "stats.h" 889#include "auto_group.h" 890 891#ifdef CONFIG_CGROUP_SCHED 892 893/* 894 * Return the group to which this tasks belongs. 895 * 896 * We cannot use task_css() and friends because the cgroup subsystem 897 * changes that value before the cgroup_subsys::attach() method is called, 898 * therefore we cannot pin it and might observe the wrong value. 899 * 900 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup 901 * core changes this before calling sched_move_task(). 902 * 903 * Instead we use a 'copy' which is updated from sched_move_task() while 904 * holding both task_struct::pi_lock and rq::lock. 905 */ 906static inline struct task_group *task_group(struct task_struct *p) 907{ 908 return p->sched_task_group; 909} 910 911/* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ 912static inline void set_task_rq(struct task_struct *p, unsigned int cpu) 913{ 914#if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED) 915 struct task_group *tg = task_group(p); 916#endif 917 918#ifdef CONFIG_FAIR_GROUP_SCHED 919 p->se.cfs_rq = tg->cfs_rq[cpu]; 920 p->se.parent = tg->se[cpu]; 921#endif 922 923#ifdef CONFIG_RT_GROUP_SCHED 924 p->rt.rt_rq = tg->rt_rq[cpu]; 925 p->rt.parent = tg->rt_se[cpu]; 926#endif 927} 928 929#else /* CONFIG_CGROUP_SCHED */ 930 931static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } 932static inline struct task_group *task_group(struct task_struct *p) 933{ 934 return NULL; 935} 936 937#endif /* CONFIG_CGROUP_SCHED */ 938 939static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) 940{ 941 set_task_rq(p, cpu); 942#ifdef CONFIG_SMP 943 /* 944 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be 945 * successfuly executed on another CPU. We must ensure that updates of 946 * per-task data have been completed by this moment. 947 */ 948 smp_wmb(); 949 task_thread_info(p)->cpu = cpu; 950 p->wake_cpu = cpu; 951#endif 952} 953 954/* 955 * Tunables that become constants when CONFIG_SCHED_DEBUG is off: 956 */ 957#ifdef CONFIG_SCHED_DEBUG 958# include <linux/static_key.h> 959# define const_debug __read_mostly 960#else 961# define const_debug const 962#endif 963 964extern const_debug unsigned int sysctl_sched_features; 965 966#define SCHED_FEAT(name, enabled) \ 967 __SCHED_FEAT_##name , 968 969enum { 970#include "features.h" 971 __SCHED_FEAT_NR, 972}; 973 974#undef SCHED_FEAT 975 976#if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL) 977#define SCHED_FEAT(name, enabled) \ 978static __always_inline bool static_branch_##name(struct static_key *key) \ 979{ \ 980 return static_key_##enabled(key); \ 981} 982 983#include "features.h" 984 985#undef SCHED_FEAT 986 987extern struct static_key sched_feat_keys[__SCHED_FEAT_NR]; 988#define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x])) 989#else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */ 990#define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) 991#endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */ 992 993#ifdef CONFIG_NUMA_BALANCING 994#define sched_feat_numa(x) sched_feat(x) 995#ifdef CONFIG_SCHED_DEBUG 996#define numabalancing_enabled sched_feat_numa(NUMA) 997#else 998extern bool numabalancing_enabled; 999#endif /* CONFIG_SCHED_DEBUG */ 1000#else 1001#define sched_feat_numa(x) (0) 1002#define numabalancing_enabled (0) 1003#endif /* CONFIG_NUMA_BALANCING */ 1004 1005static inline u64 global_rt_period(void) 1006{ 1007 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; 1008} 1009 1010static inline u64 global_rt_runtime(void) 1011{ 1012 if (sysctl_sched_rt_runtime < 0) 1013 return RUNTIME_INF; 1014 1015 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; 1016} 1017 1018static inline int task_current(struct rq *rq, struct task_struct *p) 1019{ 1020 return rq->curr == p; 1021} 1022 1023static inline int task_running(struct rq *rq, struct task_struct *p) 1024{ 1025#ifdef CONFIG_SMP 1026 return p->on_cpu; 1027#else 1028 return task_current(rq, p); 1029#endif 1030} 1031 1032static inline int task_on_rq_queued(struct task_struct *p) 1033{ 1034 return p->on_rq == TASK_ON_RQ_QUEUED; 1035} 1036 1037static inline int task_on_rq_migrating(struct task_struct *p) 1038{ 1039 return p->on_rq == TASK_ON_RQ_MIGRATING; 1040} 1041 1042#ifndef prepare_arch_switch 1043# define prepare_arch_switch(next) do { } while (0) 1044#endif 1045#ifndef finish_arch_switch 1046# define finish_arch_switch(prev) do { } while (0) 1047#endif 1048#ifndef finish_arch_post_lock_switch 1049# define finish_arch_post_lock_switch() do { } while (0) 1050#endif 1051 1052static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) 1053{ 1054#ifdef CONFIG_SMP 1055 /* 1056 * We can optimise this out completely for !SMP, because the 1057 * SMP rebalancing from interrupt is the only thing that cares 1058 * here. 1059 */ 1060 next->on_cpu = 1; 1061#endif 1062} 1063 1064static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) 1065{ 1066#ifdef CONFIG_SMP 1067 /* 1068 * After ->on_cpu is cleared, the task can be moved to a different CPU. 1069 * We must ensure this doesn't happen until the switch is completely 1070 * finished. 1071 * 1072 * Pairs with the control dependency and rmb in try_to_wake_up(). 1073 */ 1074 smp_store_release(&prev->on_cpu, 0); 1075#endif 1076#ifdef CONFIG_DEBUG_SPINLOCK 1077 /* this is a valid case when another task releases the spinlock */ 1078 rq->lock.owner = current; 1079#endif 1080 /* 1081 * If we are tracking spinlock dependencies then we have to 1082 * fix up the runqueue lock - which gets 'carried over' from 1083 * prev into current: 1084 */ 1085 spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); 1086 1087 raw_spin_unlock_irq(&rq->lock); 1088} 1089 1090/* 1091 * wake flags 1092 */ 1093#define WF_SYNC 0x01 /* waker goes to sleep after wakeup */ 1094#define WF_FORK 0x02 /* child wakeup after fork */ 1095#define WF_MIGRATED 0x4 /* internal use, task got migrated */ 1096 1097/* 1098 * To aid in avoiding the subversion of "niceness" due to uneven distribution 1099 * of tasks with abnormal "nice" values across CPUs the contribution that 1100 * each task makes to its run queue's load is weighted according to its 1101 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a 1102 * scaled version of the new time slice allocation that they receive on time 1103 * slice expiry etc. 1104 */ 1105 1106#define WEIGHT_IDLEPRIO 3 1107#define WMULT_IDLEPRIO 1431655765 1108 1109/* 1110 * Nice levels are multiplicative, with a gentle 10% change for every 1111 * nice level changed. I.e. when a CPU-bound task goes from nice 0 to 1112 * nice 1, it will get ~10% less CPU time than another CPU-bound task 1113 * that remained on nice 0. 1114 * 1115 * The "10% effect" is relative and cumulative: from _any_ nice level, 1116 * if you go up 1 level, it's -10% CPU usage, if you go down 1 level 1117 * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. 1118 * If a task goes up by ~10% and another task goes down by ~10% then 1119 * the relative distance between them is ~25%.) 1120 */ 1121static const int prio_to_weight[40] = { 1122 /* -20 */ 88761, 71755, 56483, 46273, 36291, 1123 /* -15 */ 29154, 23254, 18705, 14949, 11916, 1124 /* -10 */ 9548, 7620, 6100, 4904, 3906, 1125 /* -5 */ 3121, 2501, 1991, 1586, 1277, 1126 /* 0 */ 1024, 820, 655, 526, 423, 1127 /* 5 */ 335, 272, 215, 172, 137, 1128 /* 10 */ 110, 87, 70, 56, 45, 1129 /* 15 */ 36, 29, 23, 18, 15, 1130}; 1131 1132/* 1133 * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. 1134 * 1135 * In cases where the weight does not change often, we can use the 1136 * precalculated inverse to speed up arithmetics by turning divisions 1137 * into multiplications: 1138 */ 1139static const u32 prio_to_wmult[40] = { 1140 /* -20 */ 48388, 59856, 76040, 92818, 118348, 1141 /* -15 */ 147320, 184698, 229616, 287308, 360437, 1142 /* -10 */ 449829, 563644, 704093, 875809, 1099582, 1143 /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, 1144 /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, 1145 /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, 1146 /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, 1147 /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, 1148}; 1149 1150#define ENQUEUE_WAKEUP 1 1151#define ENQUEUE_HEAD 2 1152#ifdef CONFIG_SMP 1153#define ENQUEUE_WAKING 4 /* sched_class::task_waking was called */ 1154#else 1155#define ENQUEUE_WAKING 0 1156#endif 1157#define ENQUEUE_REPLENISH 8 1158 1159#define DEQUEUE_SLEEP 1 1160 1161#define RETRY_TASK ((void *)-1UL) 1162 1163struct sched_class { 1164 const struct sched_class *next; 1165 1166 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags); 1167 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags); 1168 void (*yield_task) (struct rq *rq); 1169 bool (*yield_to_task) (struct rq *rq, struct task_struct *p, bool preempt); 1170 1171 void (*check_preempt_curr) (struct rq *rq, struct task_struct *p, int flags); 1172 1173 /* 1174 * It is the responsibility of the pick_next_task() method that will 1175 * return the next task to call put_prev_task() on the @prev task or 1176 * something equivalent. 1177 * 1178 * May return RETRY_TASK when it finds a higher prio class has runnable 1179 * tasks. 1180 */ 1181 struct task_struct * (*pick_next_task) (struct rq *rq, 1182 struct task_struct *prev); 1183 void (*put_prev_task) (struct rq *rq, struct task_struct *p); 1184 1185#ifdef CONFIG_SMP 1186 int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags); 1187 void (*migrate_task_rq)(struct task_struct *p, int next_cpu); 1188 1189 void (*post_schedule) (struct rq *this_rq); 1190 void (*task_waking) (struct task_struct *task); 1191 void (*task_woken) (struct rq *this_rq, struct task_struct *task); 1192 1193 void (*set_cpus_allowed)(struct task_struct *p, 1194 const struct cpumask *newmask); 1195 1196 void (*rq_online)(struct rq *rq); 1197 void (*rq_offline)(struct rq *rq); 1198#endif 1199 1200 void (*set_curr_task) (struct rq *rq); 1201 void (*task_tick) (struct rq *rq, struct task_struct *p, int queued); 1202 void (*task_fork) (struct task_struct *p); 1203 void (*task_dead) (struct task_struct *p); 1204 1205 /* 1206 * The switched_from() call is allowed to drop rq->lock, therefore we 1207 * cannot assume the switched_from/switched_to pair is serliazed by 1208 * rq->lock. They are however serialized by p->pi_lock. 1209 */ 1210 void (*switched_from) (struct rq *this_rq, struct task_struct *task); 1211 void (*switched_to) (struct rq *this_rq, struct task_struct *task); 1212 void (*prio_changed) (struct rq *this_rq, struct task_struct *task, 1213 int oldprio); 1214 1215 unsigned int (*get_rr_interval) (struct rq *rq, 1216 struct task_struct *task); 1217 1218 void (*update_curr) (struct rq *rq); 1219 1220#ifdef CONFIG_FAIR_GROUP_SCHED 1221 void (*task_move_group) (struct task_struct *p, int on_rq); 1222#endif 1223}; 1224 1225static inline void put_prev_task(struct rq *rq, struct task_struct *prev) 1226{ 1227 prev->sched_class->put_prev_task(rq, prev); 1228} 1229 1230#define sched_class_highest (&stop_sched_class) 1231#define for_each_class(class) \ 1232 for (class = sched_class_highest; class; class = class->next) 1233 1234extern const struct sched_class stop_sched_class; 1235extern const struct sched_class dl_sched_class; 1236extern const struct sched_class rt_sched_class; 1237extern const struct sched_class fair_sched_class; 1238extern const struct sched_class idle_sched_class; 1239 1240 1241#ifdef CONFIG_SMP 1242 1243extern void update_group_capacity(struct sched_domain *sd, int cpu); 1244 1245extern void trigger_load_balance(struct rq *rq); 1246 1247extern void idle_enter_fair(struct rq *this_rq); 1248extern void idle_exit_fair(struct rq *this_rq); 1249 1250#else 1251 1252static inline void idle_enter_fair(struct rq *rq) { } 1253static inline void idle_exit_fair(struct rq *rq) { } 1254 1255#endif 1256 1257#ifdef CONFIG_CPU_IDLE 1258static inline void idle_set_state(struct rq *rq, 1259 struct cpuidle_state *idle_state) 1260{ 1261 rq->idle_state = idle_state; 1262} 1263 1264static inline struct cpuidle_state *idle_get_state(struct rq *rq) 1265{ 1266 WARN_ON(!rcu_read_lock_held()); 1267 return rq->idle_state; 1268} 1269#else 1270static inline void idle_set_state(struct rq *rq, 1271 struct cpuidle_state *idle_state) 1272{ 1273} 1274 1275static inline struct cpuidle_state *idle_get_state(struct rq *rq) 1276{ 1277 return NULL; 1278} 1279#endif 1280 1281extern void sysrq_sched_debug_show(void); 1282extern void sched_init_granularity(void); 1283extern void update_max_interval(void); 1284 1285extern void init_sched_dl_class(void); 1286extern void init_sched_rt_class(void); 1287extern void init_sched_fair_class(void); 1288extern void init_sched_dl_class(void); 1289 1290extern void resched_curr(struct rq *rq); 1291extern void resched_cpu(int cpu); 1292 1293extern struct rt_bandwidth def_rt_bandwidth; 1294extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime); 1295 1296extern struct dl_bandwidth def_dl_bandwidth; 1297extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime); 1298extern void init_dl_task_timer(struct sched_dl_entity *dl_se); 1299 1300unsigned long to_ratio(u64 period, u64 runtime); 1301 1302extern void update_idle_cpu_load(struct rq *this_rq); 1303 1304extern void init_task_runnable_average(struct task_struct *p); 1305 1306static inline void add_nr_running(struct rq *rq, unsigned count) 1307{ 1308 unsigned prev_nr = rq->nr_running; 1309 1310 rq->nr_running = prev_nr + count; 1311 1312 if (prev_nr < 2 && rq->nr_running >= 2) { 1313#ifdef CONFIG_SMP 1314 if (!rq->rd->overload) 1315 rq->rd->overload = true; 1316#endif 1317 1318#ifdef CONFIG_NO_HZ_FULL 1319 if (tick_nohz_full_cpu(rq->cpu)) { 1320 /* 1321 * Tick is needed if more than one task runs on a CPU. 1322 * Send the target an IPI to kick it out of nohz mode. 1323 * 1324 * We assume that IPI implies full memory barrier and the 1325 * new value of rq->nr_running is visible on reception 1326 * from the target. 1327 */ 1328 tick_nohz_full_kick_cpu(rq->cpu); 1329 } 1330#endif 1331 } 1332} 1333 1334static inline void sub_nr_running(struct rq *rq, unsigned count) 1335{ 1336 rq->nr_running -= count; 1337} 1338 1339static inline void rq_last_tick_reset(struct rq *rq) 1340{ 1341#ifdef CONFIG_NO_HZ_FULL 1342 rq->last_sched_tick = jiffies; 1343#endif 1344} 1345 1346extern void update_rq_clock(struct rq *rq); 1347 1348extern void activate_task(struct rq *rq, struct task_struct *p, int flags); 1349extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags); 1350 1351extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags); 1352 1353extern const_debug unsigned int sysctl_sched_time_avg; 1354extern const_debug unsigned int sysctl_sched_nr_migrate; 1355extern const_debug unsigned int sysctl_sched_migration_cost; 1356 1357static inline u64 sched_avg_period(void) 1358{ 1359 return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; 1360} 1361 1362#ifdef CONFIG_SCHED_HRTICK 1363 1364/* 1365 * Use hrtick when: 1366 * - enabled by features 1367 * - hrtimer is actually high res 1368 */ 1369static inline int hrtick_enabled(struct rq *rq) 1370{ 1371 if (!sched_feat(HRTICK)) 1372 return 0; 1373 if (!cpu_active(cpu_of(rq))) 1374 return 0; 1375 return hrtimer_is_hres_active(&rq->hrtick_timer); 1376} 1377 1378void hrtick_start(struct rq *rq, u64 delay); 1379 1380#else 1381 1382static inline int hrtick_enabled(struct rq *rq) 1383{ 1384 return 0; 1385} 1386 1387#endif /* CONFIG_SCHED_HRTICK */ 1388 1389#ifdef CONFIG_SMP 1390extern void sched_avg_update(struct rq *rq); 1391 1392#ifndef arch_scale_freq_capacity 1393static __always_inline 1394unsigned long arch_scale_freq_capacity(struct sched_domain *sd, int cpu) 1395{ 1396 return SCHED_CAPACITY_SCALE; 1397} 1398#endif 1399 1400static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) 1401{ 1402 rq->rt_avg += rt_delta * arch_scale_freq_capacity(NULL, cpu_of(rq)); 1403 sched_avg_update(rq); 1404} 1405#else 1406static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { } 1407static inline void sched_avg_update(struct rq *rq) { } 1408#endif 1409 1410extern void start_bandwidth_timer(struct hrtimer *period_timer, ktime_t period); 1411 1412/* 1413 * __task_rq_lock - lock the rq @p resides on. 1414 */ 1415static inline struct rq *__task_rq_lock(struct task_struct *p) 1416 __acquires(rq->lock) 1417{ 1418 struct rq *rq; 1419 1420 lockdep_assert_held(&p->pi_lock); 1421 1422 for (;;) { 1423 rq = task_rq(p); 1424 raw_spin_lock(&rq->lock); 1425 if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) 1426 return rq; 1427 raw_spin_unlock(&rq->lock); 1428 1429 while (unlikely(task_on_rq_migrating(p))) 1430 cpu_relax(); 1431 } 1432} 1433 1434/* 1435 * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. 1436 */ 1437static inline struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) 1438 __acquires(p->pi_lock) 1439 __acquires(rq->lock) 1440{ 1441 struct rq *rq; 1442 1443 for (;;) { 1444 raw_spin_lock_irqsave(&p->pi_lock, *flags); 1445 rq = task_rq(p); 1446 raw_spin_lock(&rq->lock); 1447 /* 1448 * move_queued_task() task_rq_lock() 1449 * 1450 * ACQUIRE (rq->lock) 1451 * [S] ->on_rq = MIGRATING [L] rq = task_rq() 1452 * WMB (__set_task_cpu()) ACQUIRE (rq->lock); 1453 * [S] ->cpu = new_cpu [L] task_rq() 1454 * [L] ->on_rq 1455 * RELEASE (rq->lock) 1456 * 1457 * If we observe the old cpu in task_rq_lock, the acquire of 1458 * the old rq->lock will fully serialize against the stores. 1459 * 1460 * If we observe the new cpu in task_rq_lock, the acquire will 1461 * pair with the WMB to ensure we must then also see migrating. 1462 */ 1463 if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) 1464 return rq; 1465 raw_spin_unlock(&rq->lock); 1466 raw_spin_unlock_irqrestore(&p->pi_lock, *flags); 1467 1468 while (unlikely(task_on_rq_migrating(p))) 1469 cpu_relax(); 1470 } 1471} 1472 1473static inline void __task_rq_unlock(struct rq *rq) 1474 __releases(rq->lock) 1475{ 1476 raw_spin_unlock(&rq->lock); 1477} 1478 1479static inline void 1480task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags) 1481 __releases(rq->lock) 1482 __releases(p->pi_lock) 1483{ 1484 raw_spin_unlock(&rq->lock); 1485 raw_spin_unlock_irqrestore(&p->pi_lock, *flags); 1486} 1487 1488#ifdef CONFIG_SMP 1489#ifdef CONFIG_PREEMPT 1490 1491static inline void double_rq_lock(struct rq *rq1, struct rq *rq2); 1492 1493/* 1494 * fair double_lock_balance: Safely acquires both rq->locks in a fair 1495 * way at the expense of forcing extra atomic operations in all 1496 * invocations. This assures that the double_lock is acquired using the 1497 * same underlying policy as the spinlock_t on this architecture, which 1498 * reduces latency compared to the unfair variant below. However, it 1499 * also adds more overhead and therefore may reduce throughput. 1500 */ 1501static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) 1502 __releases(this_rq->lock) 1503 __acquires(busiest->lock) 1504 __acquires(this_rq->lock) 1505{ 1506 raw_spin_unlock(&this_rq->lock); 1507 double_rq_lock(this_rq, busiest); 1508 1509 return 1; 1510} 1511 1512#else 1513/* 1514 * Unfair double_lock_balance: Optimizes throughput at the expense of 1515 * latency by eliminating extra atomic operations when the locks are 1516 * already in proper order on entry. This favors lower cpu-ids and will 1517 * grant the double lock to lower cpus over higher ids under contention, 1518 * regardless of entry order into the function. 1519 */ 1520static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) 1521 __releases(this_rq->lock) 1522 __acquires(busiest->lock) 1523 __acquires(this_rq->lock) 1524{ 1525 int ret = 0; 1526 1527 if (unlikely(!raw_spin_trylock(&busiest->lock))) { 1528 if (busiest < this_rq) { 1529 raw_spin_unlock(&this_rq->lock); 1530 raw_spin_lock(&busiest->lock); 1531 raw_spin_lock_nested(&this_rq->lock, 1532 SINGLE_DEPTH_NESTING); 1533 ret = 1; 1534 } else 1535 raw_spin_lock_nested(&busiest->lock, 1536 SINGLE_DEPTH_NESTING); 1537 } 1538 return ret; 1539} 1540 1541#endif /* CONFIG_PREEMPT */ 1542 1543/* 1544 * double_lock_balance - lock the busiest runqueue, this_rq is locked already. 1545 */ 1546static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest) 1547{ 1548 if (unlikely(!irqs_disabled())) { 1549 /* printk() doesn't work good under rq->lock */ 1550 raw_spin_unlock(&this_rq->lock); 1551 BUG_ON(1); 1552 } 1553 1554 return _double_lock_balance(this_rq, busiest); 1555} 1556 1557static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) 1558 __releases(busiest->lock) 1559{ 1560 raw_spin_unlock(&busiest->lock); 1561 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); 1562} 1563 1564static inline void double_lock(spinlock_t *l1, spinlock_t *l2) 1565{ 1566 if (l1 > l2) 1567 swap(l1, l2); 1568 1569 spin_lock(l1); 1570 spin_lock_nested(l2, SINGLE_DEPTH_NESTING); 1571} 1572 1573static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2) 1574{ 1575 if (l1 > l2) 1576 swap(l1, l2); 1577 1578 spin_lock_irq(l1); 1579 spin_lock_nested(l2, SINGLE_DEPTH_NESTING); 1580} 1581 1582static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2) 1583{ 1584 if (l1 > l2) 1585 swap(l1, l2); 1586 1587 raw_spin_lock(l1); 1588 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING); 1589} 1590 1591/* 1592 * double_rq_lock - safely lock two runqueues 1593 * 1594 * Note this does not disable interrupts like task_rq_lock, 1595 * you need to do so manually before calling. 1596 */ 1597static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) 1598 __acquires(rq1->lock) 1599 __acquires(rq2->lock) 1600{ 1601 BUG_ON(!irqs_disabled()); 1602 if (rq1 == rq2) { 1603 raw_spin_lock(&rq1->lock); 1604 __acquire(rq2->lock); /* Fake it out ;) */ 1605 } else { 1606 if (rq1 < rq2) { 1607 raw_spin_lock(&rq1->lock); 1608 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); 1609 } else { 1610 raw_spin_lock(&rq2->lock); 1611 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); 1612 } 1613 } 1614} 1615 1616/* 1617 * double_rq_unlock - safely unlock two runqueues 1618 * 1619 * Note this does not restore interrupts like task_rq_unlock, 1620 * you need to do so manually after calling. 1621 */ 1622static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) 1623 __releases(rq1->lock) 1624 __releases(rq2->lock) 1625{ 1626 raw_spin_unlock(&rq1->lock); 1627 if (rq1 != rq2) 1628 raw_spin_unlock(&rq2->lock); 1629 else 1630 __release(rq2->lock); 1631} 1632 1633#else /* CONFIG_SMP */ 1634 1635/* 1636 * double_rq_lock - safely lock two runqueues 1637 * 1638 * Note this does not disable interrupts like task_rq_lock, 1639 * you need to do so manually before calling. 1640 */ 1641static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) 1642 __acquires(rq1->lock) 1643 __acquires(rq2->lock) 1644{ 1645 BUG_ON(!irqs_disabled()); 1646 BUG_ON(rq1 != rq2); 1647 raw_spin_lock(&rq1->lock); 1648 __acquire(rq2->lock); /* Fake it out ;) */ 1649} 1650 1651/* 1652 * double_rq_unlock - safely unlock two runqueues 1653 * 1654 * Note this does not restore interrupts like task_rq_unlock, 1655 * you need to do so manually after calling. 1656 */ 1657static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) 1658 __releases(rq1->lock) 1659 __releases(rq2->lock) 1660{ 1661 BUG_ON(rq1 != rq2); 1662 raw_spin_unlock(&rq1->lock); 1663 __release(rq2->lock); 1664} 1665 1666#endif 1667 1668extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq); 1669extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq); 1670extern void print_cfs_stats(struct seq_file *m, int cpu); 1671extern void print_rt_stats(struct seq_file *m, int cpu); 1672extern void print_dl_stats(struct seq_file *m, int cpu); 1673 1674extern void init_cfs_rq(struct cfs_rq *cfs_rq); 1675extern void init_rt_rq(struct rt_rq *rt_rq); 1676extern void init_dl_rq(struct dl_rq *dl_rq); 1677 1678extern void cfs_bandwidth_usage_inc(void); 1679extern void cfs_bandwidth_usage_dec(void); 1680 1681#ifdef CONFIG_NO_HZ_COMMON 1682enum rq_nohz_flag_bits { 1683 NOHZ_TICK_STOPPED, 1684 NOHZ_BALANCE_KICK, 1685}; 1686 1687#define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags) 1688#endif 1689 1690#ifdef CONFIG_IRQ_TIME_ACCOUNTING 1691 1692DECLARE_PER_CPU(u64, cpu_hardirq_time); 1693DECLARE_PER_CPU(u64, cpu_softirq_time); 1694 1695#ifndef CONFIG_64BIT 1696DECLARE_PER_CPU(seqcount_t, irq_time_seq); 1697 1698static inline void irq_time_write_begin(void) 1699{ 1700 __this_cpu_inc(irq_time_seq.sequence); 1701 smp_wmb(); 1702} 1703 1704static inline void irq_time_write_end(void) 1705{ 1706 smp_wmb(); 1707 __this_cpu_inc(irq_time_seq.sequence); 1708} 1709 1710static inline u64 irq_time_read(int cpu) 1711{ 1712 u64 irq_time; 1713 unsigned seq; 1714 1715 do { 1716 seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu)); 1717 irq_time = per_cpu(cpu_softirq_time, cpu) + 1718 per_cpu(cpu_hardirq_time, cpu); 1719 } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq)); 1720 1721 return irq_time; 1722} 1723#else /* CONFIG_64BIT */ 1724static inline void irq_time_write_begin(void) 1725{ 1726} 1727 1728static inline void irq_time_write_end(void) 1729{ 1730} 1731 1732static inline u64 irq_time_read(int cpu) 1733{ 1734 return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu); 1735} 1736#endif /* CONFIG_64BIT */ 1737#endif /* CONFIG_IRQ_TIME_ACCOUNTING */ 1738