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/binfmts.h> 7 #include <linux/mutex.h> 8 #include <linux/spinlock.h> 9 #include <linux/stop_machine.h> 10 #include <linux/irq_work.h> 11 #include <linux/tick.h> 12 #include <linux/slab.h> 13 14 #include "cpupri.h" 15 #include "cpudeadline.h" 16 #include "cpuacct.h" 17 18 struct rq; 19 struct cpuidle_state; 20 21 /* task_struct::on_rq states: */ 22 #define TASK_ON_RQ_QUEUED 1 23 #define TASK_ON_RQ_MIGRATING 2 24 25 extern __read_mostly int scheduler_running; 26 27 extern unsigned long calc_load_update; 28 extern atomic_long_t calc_load_tasks; 29 30 extern void calc_global_load_tick(struct rq *this_rq); 31 extern long calc_load_fold_active(struct rq *this_rq); 32 33 #ifdef CONFIG_SMP 34 extern void cpu_load_update_active(struct rq *this_rq); 35 #else 36 static inline void cpu_load_update_active(struct rq *this_rq) { } 37 #endif 38 39 /* 40 * Helpers for converting nanosecond timing to jiffy resolution 41 */ 42 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) 43 44 /* 45 * Increase resolution of nice-level calculations for 64-bit architectures. 46 * The extra resolution improves shares distribution and load balancing of 47 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup 48 * hierarchies, especially on larger systems. This is not a user-visible change 49 * and does not change the user-interface for setting shares/weights. 50 * 51 * We increase resolution only if we have enough bits to allow this increased 52 * resolution (i.e. 64bit). The costs for increasing resolution when 32bit are 53 * pretty high and the returns do not justify the increased costs. 54 * 55 * Really only required when CONFIG_FAIR_GROUP_SCHED is also set, but to 56 * increase coverage and consistency always enable it on 64bit platforms. 57 */ 58 #ifdef CONFIG_64BIT 59 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT) 60 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT) 61 # define scale_load_down(w) ((w) >> SCHED_FIXEDPOINT_SHIFT) 62 #else 63 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT) 64 # define scale_load(w) (w) 65 # define scale_load_down(w) (w) 66 #endif 67 68 /* 69 * Task weight (visible to users) and its load (invisible to users) have 70 * independent resolution, but they should be well calibrated. We use 71 * scale_load() and scale_load_down(w) to convert between them. The 72 * following must be true: 73 * 74 * scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD 75 * 76 */ 77 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT) 78 79 /* 80 * Single value that decides SCHED_DEADLINE internal math precision. 81 * 10 -> just above 1us 82 * 9 -> just above 0.5us 83 */ 84 #define DL_SCALE (10) 85 86 /* 87 * These are the 'tuning knobs' of the scheduler: 88 */ 89 90 /* 91 * single value that denotes runtime == period, ie unlimited time. 92 */ 93 #define RUNTIME_INF ((u64)~0ULL) 94 95 static inline int idle_policy(int policy) 96 { 97 return policy == SCHED_IDLE; 98 } 99 static inline int fair_policy(int policy) 100 { 101 return policy == SCHED_NORMAL || policy == SCHED_BATCH; 102 } 103 104 static inline int rt_policy(int policy) 105 { 106 return policy == SCHED_FIFO || policy == SCHED_RR; 107 } 108 109 static inline int dl_policy(int policy) 110 { 111 return policy == SCHED_DEADLINE; 112 } 113 static inline bool valid_policy(int policy) 114 { 115 return idle_policy(policy) || fair_policy(policy) || 116 rt_policy(policy) || dl_policy(policy); 117 } 118 119 static inline int task_has_rt_policy(struct task_struct *p) 120 { 121 return rt_policy(p->policy); 122 } 123 124 static inline int task_has_dl_policy(struct task_struct *p) 125 { 126 return dl_policy(p->policy); 127 } 128 129 /* 130 * Tells if entity @a should preempt entity @b. 131 */ 132 static inline bool 133 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b) 134 { 135 return dl_time_before(a->deadline, b->deadline); 136 } 137 138 /* 139 * This is the priority-queue data structure of the RT scheduling class: 140 */ 141 struct rt_prio_array { 142 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ 143 struct list_head queue[MAX_RT_PRIO]; 144 }; 145 146 struct rt_bandwidth { 147 /* nests inside the rq lock: */ 148 raw_spinlock_t rt_runtime_lock; 149 ktime_t rt_period; 150 u64 rt_runtime; 151 struct hrtimer rt_period_timer; 152 unsigned int rt_period_active; 153 }; 154 155 void __dl_clear_params(struct task_struct *p); 156 157 /* 158 * To keep the bandwidth of -deadline tasks and groups under control 159 * we need some place where: 160 * - store the maximum -deadline bandwidth of the system (the group); 161 * - cache the fraction of that bandwidth that is currently allocated. 162 * 163 * This is all done in the data structure below. It is similar to the 164 * one used for RT-throttling (rt_bandwidth), with the main difference 165 * that, since here we are only interested in admission control, we 166 * do not decrease any runtime while the group "executes", neither we 167 * need a timer to replenish it. 168 * 169 * With respect to SMP, the bandwidth is given on a per-CPU basis, 170 * meaning that: 171 * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU; 172 * - dl_total_bw array contains, in the i-eth element, the currently 173 * allocated bandwidth on the i-eth CPU. 174 * Moreover, groups consume bandwidth on each CPU, while tasks only 175 * consume bandwidth on the CPU they're running on. 176 * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw 177 * that will be shown the next time the proc or cgroup controls will 178 * be red. It on its turn can be changed by writing on its own 179 * control. 180 */ 181 struct dl_bandwidth { 182 raw_spinlock_t dl_runtime_lock; 183 u64 dl_runtime; 184 u64 dl_period; 185 }; 186 187 static inline int dl_bandwidth_enabled(void) 188 { 189 return sysctl_sched_rt_runtime >= 0; 190 } 191 192 extern struct dl_bw *dl_bw_of(int i); 193 194 struct dl_bw { 195 raw_spinlock_t lock; 196 u64 bw, total_bw; 197 }; 198 199 static inline 200 void __dl_clear(struct dl_bw *dl_b, u64 tsk_bw) 201 { 202 dl_b->total_bw -= tsk_bw; 203 } 204 205 static inline 206 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw) 207 { 208 dl_b->total_bw += tsk_bw; 209 } 210 211 static inline 212 bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw) 213 { 214 return dl_b->bw != -1 && 215 dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw; 216 } 217 218 extern struct mutex sched_domains_mutex; 219 220 #ifdef CONFIG_CGROUP_SCHED 221 222 #include <linux/cgroup.h> 223 224 struct cfs_rq; 225 struct rt_rq; 226 227 extern struct list_head task_groups; 228 229 struct cfs_bandwidth { 230 #ifdef CONFIG_CFS_BANDWIDTH 231 raw_spinlock_t lock; 232 ktime_t period; 233 u64 quota, runtime; 234 s64 hierarchical_quota; 235 u64 runtime_expires; 236 237 int idle, period_active; 238 struct hrtimer period_timer, slack_timer; 239 struct list_head throttled_cfs_rq; 240 241 /* statistics */ 242 int nr_periods, nr_throttled; 243 u64 throttled_time; 244 #endif 245 }; 246 247 /* task group related information */ 248 struct task_group { 249 struct cgroup_subsys_state css; 250 251 #ifdef CONFIG_FAIR_GROUP_SCHED 252 /* schedulable entities of this group on each cpu */ 253 struct sched_entity **se; 254 /* runqueue "owned" by this group on each cpu */ 255 struct cfs_rq **cfs_rq; 256 unsigned long shares; 257 258 #ifdef CONFIG_SMP 259 /* 260 * load_avg can be heavily contended at clock tick time, so put 261 * it in its own cacheline separated from the fields above which 262 * will also be accessed at each tick. 263 */ 264 atomic_long_t load_avg ____cacheline_aligned; 265 #endif 266 #endif 267 268 #ifdef CONFIG_RT_GROUP_SCHED 269 struct sched_rt_entity **rt_se; 270 struct rt_rq **rt_rq; 271 272 struct rt_bandwidth rt_bandwidth; 273 #endif 274 275 struct rcu_head rcu; 276 struct list_head list; 277 278 struct task_group *parent; 279 struct list_head siblings; 280 struct list_head children; 281 282 #ifdef CONFIG_SCHED_AUTOGROUP 283 struct autogroup *autogroup; 284 #endif 285 286 struct cfs_bandwidth cfs_bandwidth; 287 }; 288 289 #ifdef CONFIG_FAIR_GROUP_SCHED 290 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD 291 292 /* 293 * A weight of 0 or 1 can cause arithmetics problems. 294 * A weight of a cfs_rq is the sum of weights of which entities 295 * are queued on this cfs_rq, so a weight of a entity should not be 296 * too large, so as the shares value of a task group. 297 * (The default weight is 1024 - so there's no practical 298 * limitation from this.) 299 */ 300 #define MIN_SHARES (1UL << 1) 301 #define MAX_SHARES (1UL << 18) 302 #endif 303 304 typedef int (*tg_visitor)(struct task_group *, void *); 305 306 extern int walk_tg_tree_from(struct task_group *from, 307 tg_visitor down, tg_visitor up, void *data); 308 309 /* 310 * Iterate the full tree, calling @down when first entering a node and @up when 311 * leaving it for the final time. 312 * 313 * Caller must hold rcu_lock or sufficient equivalent. 314 */ 315 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) 316 { 317 return walk_tg_tree_from(&root_task_group, down, up, data); 318 } 319 320 extern int tg_nop(struct task_group *tg, void *data); 321 322 extern void free_fair_sched_group(struct task_group *tg); 323 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent); 324 extern void unregister_fair_sched_group(struct task_group *tg); 325 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, 326 struct sched_entity *se, int cpu, 327 struct sched_entity *parent); 328 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b); 329 330 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b); 331 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b); 332 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq); 333 334 extern void free_rt_sched_group(struct task_group *tg); 335 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent); 336 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, 337 struct sched_rt_entity *rt_se, int cpu, 338 struct sched_rt_entity *parent); 339 340 extern struct task_group *sched_create_group(struct task_group *parent); 341 extern void sched_online_group(struct task_group *tg, 342 struct task_group *parent); 343 extern void sched_destroy_group(struct task_group *tg); 344 extern void sched_offline_group(struct task_group *tg); 345 346 extern void sched_move_task(struct task_struct *tsk); 347 348 #ifdef CONFIG_FAIR_GROUP_SCHED 349 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares); 350 351 #ifdef CONFIG_SMP 352 extern void set_task_rq_fair(struct sched_entity *se, 353 struct cfs_rq *prev, struct cfs_rq *next); 354 #else /* !CONFIG_SMP */ 355 static inline void set_task_rq_fair(struct sched_entity *se, 356 struct cfs_rq *prev, struct cfs_rq *next) { } 357 #endif /* CONFIG_SMP */ 358 #endif /* CONFIG_FAIR_GROUP_SCHED */ 359 360 #else /* CONFIG_CGROUP_SCHED */ 361 362 struct cfs_bandwidth { }; 363 364 #endif /* CONFIG_CGROUP_SCHED */ 365 366 /* CFS-related fields in a runqueue */ 367 struct cfs_rq { 368 struct load_weight load; 369 unsigned int nr_running, h_nr_running; 370 371 u64 exec_clock; 372 u64 min_vruntime; 373 #ifndef CONFIG_64BIT 374 u64 min_vruntime_copy; 375 #endif 376 377 struct rb_root tasks_timeline; 378 struct rb_node *rb_leftmost; 379 380 /* 381 * 'curr' points to currently running entity on this cfs_rq. 382 * It is set to NULL otherwise (i.e when none are currently running). 383 */ 384 struct sched_entity *curr, *next, *last, *skip; 385 386 #ifdef CONFIG_SCHED_DEBUG 387 unsigned int nr_spread_over; 388 #endif 389 390 #ifdef CONFIG_SMP 391 /* 392 * CFS load tracking 393 */ 394 struct sched_avg avg; 395 u64 runnable_load_sum; 396 unsigned long runnable_load_avg; 397 #ifdef CONFIG_FAIR_GROUP_SCHED 398 unsigned long tg_load_avg_contrib; 399 #endif 400 atomic_long_t removed_load_avg, removed_util_avg; 401 #ifndef CONFIG_64BIT 402 u64 load_last_update_time_copy; 403 #endif 404 405 #ifdef CONFIG_FAIR_GROUP_SCHED 406 /* 407 * h_load = weight * f(tg) 408 * 409 * Where f(tg) is the recursive weight fraction assigned to 410 * this group. 411 */ 412 unsigned long h_load; 413 u64 last_h_load_update; 414 struct sched_entity *h_load_next; 415 #endif /* CONFIG_FAIR_GROUP_SCHED */ 416 #endif /* CONFIG_SMP */ 417 418 #ifdef CONFIG_FAIR_GROUP_SCHED 419 struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ 420 421 /* 422 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in 423 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities 424 * (like users, containers etc.) 425 * 426 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This 427 * list is used during load balance. 428 */ 429 int on_list; 430 struct list_head leaf_cfs_rq_list; 431 struct task_group *tg; /* group that "owns" this runqueue */ 432 433 #ifdef CONFIG_CFS_BANDWIDTH 434 int runtime_enabled; 435 u64 runtime_expires; 436 s64 runtime_remaining; 437 438 u64 throttled_clock, throttled_clock_task; 439 u64 throttled_clock_task_time; 440 int throttled, throttle_count; 441 struct list_head throttled_list; 442 #endif /* CONFIG_CFS_BANDWIDTH */ 443 #endif /* CONFIG_FAIR_GROUP_SCHED */ 444 }; 445 446 static inline int rt_bandwidth_enabled(void) 447 { 448 return sysctl_sched_rt_runtime >= 0; 449 } 450 451 /* RT IPI pull logic requires IRQ_WORK */ 452 #ifdef CONFIG_IRQ_WORK 453 # define HAVE_RT_PUSH_IPI 454 #endif 455 456 /* Real-Time classes' related field in a runqueue: */ 457 struct rt_rq { 458 struct rt_prio_array active; 459 unsigned int rt_nr_running; 460 unsigned int rr_nr_running; 461 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED 462 struct { 463 int curr; /* highest queued rt task prio */ 464 #ifdef CONFIG_SMP 465 int next; /* next highest */ 466 #endif 467 } highest_prio; 468 #endif 469 #ifdef CONFIG_SMP 470 unsigned long rt_nr_migratory; 471 unsigned long rt_nr_total; 472 int overloaded; 473 struct plist_head pushable_tasks; 474 #ifdef HAVE_RT_PUSH_IPI 475 int push_flags; 476 int push_cpu; 477 struct irq_work push_work; 478 raw_spinlock_t push_lock; 479 #endif 480 #endif /* CONFIG_SMP */ 481 int rt_queued; 482 483 int rt_throttled; 484 u64 rt_time; 485 u64 rt_runtime; 486 /* Nests inside the rq lock: */ 487 raw_spinlock_t rt_runtime_lock; 488 489 #ifdef CONFIG_RT_GROUP_SCHED 490 unsigned long rt_nr_boosted; 491 492 struct rq *rq; 493 struct task_group *tg; 494 #endif 495 }; 496 497 /* Deadline class' related fields in a runqueue */ 498 struct dl_rq { 499 /* runqueue is an rbtree, ordered by deadline */ 500 struct rb_root rb_root; 501 struct rb_node *rb_leftmost; 502 503 unsigned long dl_nr_running; 504 505 #ifdef CONFIG_SMP 506 /* 507 * Deadline values of the currently executing and the 508 * earliest ready task on this rq. Caching these facilitates 509 * the decision wether or not a ready but not running task 510 * should migrate somewhere else. 511 */ 512 struct { 513 u64 curr; 514 u64 next; 515 } earliest_dl; 516 517 unsigned long dl_nr_migratory; 518 int overloaded; 519 520 /* 521 * Tasks on this rq that can be pushed away. They are kept in 522 * an rb-tree, ordered by tasks' deadlines, with caching 523 * of the leftmost (earliest deadline) element. 524 */ 525 struct rb_root pushable_dl_tasks_root; 526 struct rb_node *pushable_dl_tasks_leftmost; 527 #else 528 struct dl_bw dl_bw; 529 #endif 530 }; 531 532 #ifdef CONFIG_SMP 533 534 /* 535 * We add the notion of a root-domain which will be used to define per-domain 536 * variables. Each exclusive cpuset essentially defines an island domain by 537 * fully partitioning the member cpus from any other cpuset. Whenever a new 538 * exclusive cpuset is created, we also create and attach a new root-domain 539 * object. 540 * 541 */ 542 struct root_domain { 543 atomic_t refcount; 544 atomic_t rto_count; 545 struct rcu_head rcu; 546 cpumask_var_t span; 547 cpumask_var_t online; 548 549 /* Indicate more than one runnable task for any CPU */ 550 bool overload; 551 552 /* 553 * The bit corresponding to a CPU gets set here if such CPU has more 554 * than one runnable -deadline task (as it is below for RT tasks). 555 */ 556 cpumask_var_t dlo_mask; 557 atomic_t dlo_count; 558 struct dl_bw dl_bw; 559 struct cpudl cpudl; 560 561 /* 562 * The "RT overload" flag: it gets set if a CPU has more than 563 * one runnable RT task. 564 */ 565 cpumask_var_t rto_mask; 566 struct cpupri cpupri; 567 }; 568 569 extern struct root_domain def_root_domain; 570 571 #endif /* CONFIG_SMP */ 572 573 /* 574 * This is the main, per-CPU runqueue data structure. 575 * 576 * Locking rule: those places that want to lock multiple runqueues 577 * (such as the load balancing or the thread migration code), lock 578 * acquire operations must be ordered by ascending &runqueue. 579 */ 580 struct rq { 581 /* runqueue lock: */ 582 raw_spinlock_t lock; 583 584 /* 585 * nr_running and cpu_load should be in the same cacheline because 586 * remote CPUs use both these fields when doing load calculation. 587 */ 588 unsigned int nr_running; 589 #ifdef CONFIG_NUMA_BALANCING 590 unsigned int nr_numa_running; 591 unsigned int nr_preferred_running; 592 #endif 593 #define CPU_LOAD_IDX_MAX 5 594 unsigned long cpu_load[CPU_LOAD_IDX_MAX]; 595 #ifdef CONFIG_NO_HZ_COMMON 596 #ifdef CONFIG_SMP 597 unsigned long last_load_update_tick; 598 #endif /* CONFIG_SMP */ 599 u64 nohz_stamp; 600 unsigned long nohz_flags; 601 #endif /* CONFIG_NO_HZ_COMMON */ 602 #ifdef CONFIG_NO_HZ_FULL 603 unsigned long last_sched_tick; 604 #endif 605 /* capture load from *all* tasks on this cpu: */ 606 struct load_weight load; 607 unsigned long nr_load_updates; 608 u64 nr_switches; 609 610 struct cfs_rq cfs; 611 struct rt_rq rt; 612 struct dl_rq dl; 613 614 #ifdef CONFIG_FAIR_GROUP_SCHED 615 /* list of leaf cfs_rq on this cpu: */ 616 struct list_head leaf_cfs_rq_list; 617 #endif /* CONFIG_FAIR_GROUP_SCHED */ 618 619 /* 620 * This is part of a global counter where only the total sum 621 * over all CPUs matters. A task can increase this counter on 622 * one CPU and if it got migrated afterwards it may decrease 623 * it on another CPU. Always updated under the runqueue lock: 624 */ 625 unsigned long nr_uninterruptible; 626 627 struct task_struct *curr, *idle, *stop; 628 unsigned long next_balance; 629 struct mm_struct *prev_mm; 630 631 unsigned int clock_skip_update; 632 u64 clock; 633 u64 clock_task; 634 635 atomic_t nr_iowait; 636 637 #ifdef CONFIG_SMP 638 struct root_domain *rd; 639 struct sched_domain *sd; 640 641 unsigned long cpu_capacity; 642 unsigned long cpu_capacity_orig; 643 644 struct callback_head *balance_callback; 645 646 unsigned char idle_balance; 647 /* For active balancing */ 648 int active_balance; 649 int push_cpu; 650 struct cpu_stop_work active_balance_work; 651 /* cpu of this runqueue: */ 652 int cpu; 653 int online; 654 655 struct list_head cfs_tasks; 656 657 u64 rt_avg; 658 u64 age_stamp; 659 u64 idle_stamp; 660 u64 avg_idle; 661 662 /* This is used to determine avg_idle's max value */ 663 u64 max_idle_balance_cost; 664 #endif 665 666 #ifdef CONFIG_IRQ_TIME_ACCOUNTING 667 u64 prev_irq_time; 668 #endif 669 #ifdef CONFIG_PARAVIRT 670 u64 prev_steal_time; 671 #endif 672 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING 673 u64 prev_steal_time_rq; 674 #endif 675 676 /* calc_load related fields */ 677 unsigned long calc_load_update; 678 long calc_load_active; 679 680 #ifdef CONFIG_SCHED_HRTICK 681 #ifdef CONFIG_SMP 682 int hrtick_csd_pending; 683 struct call_single_data hrtick_csd; 684 #endif 685 struct hrtimer hrtick_timer; 686 #endif 687 688 #ifdef CONFIG_SCHEDSTATS 689 /* latency stats */ 690 struct sched_info rq_sched_info; 691 unsigned long long rq_cpu_time; 692 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ 693 694 /* sys_sched_yield() stats */ 695 unsigned int yld_count; 696 697 /* schedule() stats */ 698 unsigned int sched_count; 699 unsigned int sched_goidle; 700 701 /* try_to_wake_up() stats */ 702 unsigned int ttwu_count; 703 unsigned int ttwu_local; 704 #endif 705 706 #ifdef CONFIG_SMP 707 struct llist_head wake_list; 708 #endif 709 710 #ifdef CONFIG_CPU_IDLE 711 /* Must be inspected within a rcu lock section */ 712 struct cpuidle_state *idle_state; 713 #endif 714 }; 715 716 static inline int cpu_of(struct rq *rq) 717 { 718 #ifdef CONFIG_SMP 719 return rq->cpu; 720 #else 721 return 0; 722 #endif 723 } 724 725 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); 726 727 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) 728 #define this_rq() this_cpu_ptr(&runqueues) 729 #define task_rq(p) cpu_rq(task_cpu(p)) 730 #define cpu_curr(cpu) (cpu_rq(cpu)->curr) 731 #define raw_rq() raw_cpu_ptr(&runqueues) 732 733 static inline u64 __rq_clock_broken(struct rq *rq) 734 { 735 return READ_ONCE(rq->clock); 736 } 737 738 static inline u64 rq_clock(struct rq *rq) 739 { 740 lockdep_assert_held(&rq->lock); 741 return rq->clock; 742 } 743 744 static inline u64 rq_clock_task(struct rq *rq) 745 { 746 lockdep_assert_held(&rq->lock); 747 return rq->clock_task; 748 } 749 750 #define RQCF_REQ_SKIP 0x01 751 #define RQCF_ACT_SKIP 0x02 752 753 static inline void rq_clock_skip_update(struct rq *rq, bool skip) 754 { 755 lockdep_assert_held(&rq->lock); 756 if (skip) 757 rq->clock_skip_update |= RQCF_REQ_SKIP; 758 else 759 rq->clock_skip_update &= ~RQCF_REQ_SKIP; 760 } 761 762 #ifdef CONFIG_NUMA 763 enum numa_topology_type { 764 NUMA_DIRECT, 765 NUMA_GLUELESS_MESH, 766 NUMA_BACKPLANE, 767 }; 768 extern enum numa_topology_type sched_numa_topology_type; 769 extern int sched_max_numa_distance; 770 extern bool find_numa_distance(int distance); 771 #endif 772 773 #ifdef CONFIG_NUMA_BALANCING 774 /* The regions in numa_faults array from task_struct */ 775 enum numa_faults_stats { 776 NUMA_MEM = 0, 777 NUMA_CPU, 778 NUMA_MEMBUF, 779 NUMA_CPUBUF 780 }; 781 extern void sched_setnuma(struct task_struct *p, int node); 782 extern int migrate_task_to(struct task_struct *p, int cpu); 783 extern int migrate_swap(struct task_struct *, struct task_struct *); 784 #endif /* CONFIG_NUMA_BALANCING */ 785 786 #ifdef CONFIG_SMP 787 788 static inline void 789 queue_balance_callback(struct rq *rq, 790 struct callback_head *head, 791 void (*func)(struct rq *rq)) 792 { 793 lockdep_assert_held(&rq->lock); 794 795 if (unlikely(head->next)) 796 return; 797 798 head->func = (void (*)(struct callback_head *))func; 799 head->next = rq->balance_callback; 800 rq->balance_callback = head; 801 } 802 803 extern void sched_ttwu_pending(void); 804 805 #define rcu_dereference_check_sched_domain(p) \ 806 rcu_dereference_check((p), \ 807 lockdep_is_held(&sched_domains_mutex)) 808 809 /* 810 * The domain tree (rq->sd) is protected by RCU's quiescent state transition. 811 * See detach_destroy_domains: synchronize_sched for details. 812 * 813 * The domain tree of any CPU may only be accessed from within 814 * preempt-disabled sections. 815 */ 816 #define for_each_domain(cpu, __sd) \ 817 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \ 818 __sd; __sd = __sd->parent) 819 820 #define for_each_lower_domain(sd) for (; sd; sd = sd->child) 821 822 /** 823 * highest_flag_domain - Return highest sched_domain containing flag. 824 * @cpu: The cpu whose highest level of sched domain is to 825 * be returned. 826 * @flag: The flag to check for the highest sched_domain 827 * for the given cpu. 828 * 829 * Returns the highest sched_domain of a cpu which contains the given flag. 830 */ 831 static inline struct sched_domain *highest_flag_domain(int cpu, int flag) 832 { 833 struct sched_domain *sd, *hsd = NULL; 834 835 for_each_domain(cpu, sd) { 836 if (!(sd->flags & flag)) 837 break; 838 hsd = sd; 839 } 840 841 return hsd; 842 } 843 844 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag) 845 { 846 struct sched_domain *sd; 847 848 for_each_domain(cpu, sd) { 849 if (sd->flags & flag) 850 break; 851 } 852 853 return sd; 854 } 855 856 DECLARE_PER_CPU(struct sched_domain *, sd_llc); 857 DECLARE_PER_CPU(int, sd_llc_size); 858 DECLARE_PER_CPU(int, sd_llc_id); 859 DECLARE_PER_CPU(struct sched_domain *, sd_numa); 860 DECLARE_PER_CPU(struct sched_domain *, sd_busy); 861 DECLARE_PER_CPU(struct sched_domain *, sd_asym); 862 863 struct sched_group_capacity { 864 atomic_t ref; 865 /* 866 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity 867 * for a single CPU. 868 */ 869 unsigned int capacity; 870 unsigned long next_update; 871 int imbalance; /* XXX unrelated to capacity but shared group state */ 872 /* 873 * Number of busy cpus in this group. 874 */ 875 atomic_t nr_busy_cpus; 876 877 unsigned long cpumask[0]; /* iteration mask */ 878 }; 879 880 struct sched_group { 881 struct sched_group *next; /* Must be a circular list */ 882 atomic_t ref; 883 884 unsigned int group_weight; 885 struct sched_group_capacity *sgc; 886 887 /* 888 * The CPUs this group covers. 889 * 890 * NOTE: this field is variable length. (Allocated dynamically 891 * by attaching extra space to the end of the structure, 892 * depending on how many CPUs the kernel has booted up with) 893 */ 894 unsigned long cpumask[0]; 895 }; 896 897 static inline struct cpumask *sched_group_cpus(struct sched_group *sg) 898 { 899 return to_cpumask(sg->cpumask); 900 } 901 902 /* 903 * cpumask masking which cpus in the group are allowed to iterate up the domain 904 * tree. 905 */ 906 static inline struct cpumask *sched_group_mask(struct sched_group *sg) 907 { 908 return to_cpumask(sg->sgc->cpumask); 909 } 910 911 /** 912 * group_first_cpu - Returns the first cpu in the cpumask of a sched_group. 913 * @group: The group whose first cpu is to be returned. 914 */ 915 static inline unsigned int group_first_cpu(struct sched_group *group) 916 { 917 return cpumask_first(sched_group_cpus(group)); 918 } 919 920 extern int group_balance_cpu(struct sched_group *sg); 921 922 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) 923 void register_sched_domain_sysctl(void); 924 void unregister_sched_domain_sysctl(void); 925 #else 926 static inline void register_sched_domain_sysctl(void) 927 { 928 } 929 static inline void unregister_sched_domain_sysctl(void) 930 { 931 } 932 #endif 933 934 #else 935 936 static inline void sched_ttwu_pending(void) { } 937 938 #endif /* CONFIG_SMP */ 939 940 #include "stats.h" 941 #include "auto_group.h" 942 943 #ifdef CONFIG_CGROUP_SCHED 944 945 /* 946 * Return the group to which this tasks belongs. 947 * 948 * We cannot use task_css() and friends because the cgroup subsystem 949 * changes that value before the cgroup_subsys::attach() method is called, 950 * therefore we cannot pin it and might observe the wrong value. 951 * 952 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup 953 * core changes this before calling sched_move_task(). 954 * 955 * Instead we use a 'copy' which is updated from sched_move_task() while 956 * holding both task_struct::pi_lock and rq::lock. 957 */ 958 static inline struct task_group *task_group(struct task_struct *p) 959 { 960 return p->sched_task_group; 961 } 962 963 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ 964 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) 965 { 966 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED) 967 struct task_group *tg = task_group(p); 968 #endif 969 970 #ifdef CONFIG_FAIR_GROUP_SCHED 971 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]); 972 p->se.cfs_rq = tg->cfs_rq[cpu]; 973 p->se.parent = tg->se[cpu]; 974 #endif 975 976 #ifdef CONFIG_RT_GROUP_SCHED 977 p->rt.rt_rq = tg->rt_rq[cpu]; 978 p->rt.parent = tg->rt_se[cpu]; 979 #endif 980 } 981 982 #else /* CONFIG_CGROUP_SCHED */ 983 984 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } 985 static inline struct task_group *task_group(struct task_struct *p) 986 { 987 return NULL; 988 } 989 990 #endif /* CONFIG_CGROUP_SCHED */ 991 992 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) 993 { 994 set_task_rq(p, cpu); 995 #ifdef CONFIG_SMP 996 /* 997 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be 998 * successfuly executed on another CPU. We must ensure that updates of 999 * per-task data have been completed by this moment. 1000 */ 1001 smp_wmb(); 1002 task_thread_info(p)->cpu = cpu; 1003 p->wake_cpu = cpu; 1004 #endif 1005 } 1006 1007 /* 1008 * Tunables that become constants when CONFIG_SCHED_DEBUG is off: 1009 */ 1010 #ifdef CONFIG_SCHED_DEBUG 1011 # include <linux/static_key.h> 1012 # define const_debug __read_mostly 1013 #else 1014 # define const_debug const 1015 #endif 1016 1017 extern const_debug unsigned int sysctl_sched_features; 1018 1019 #define SCHED_FEAT(name, enabled) \ 1020 __SCHED_FEAT_##name , 1021 1022 enum { 1023 #include "features.h" 1024 __SCHED_FEAT_NR, 1025 }; 1026 1027 #undef SCHED_FEAT 1028 1029 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL) 1030 #define SCHED_FEAT(name, enabled) \ 1031 static __always_inline bool static_branch_##name(struct static_key *key) \ 1032 { \ 1033 return static_key_##enabled(key); \ 1034 } 1035 1036 #include "features.h" 1037 1038 #undef SCHED_FEAT 1039 1040 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR]; 1041 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x])) 1042 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */ 1043 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) 1044 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */ 1045 1046 extern struct static_key_false sched_numa_balancing; 1047 extern struct static_key_false sched_schedstats; 1048 1049 static inline u64 global_rt_period(void) 1050 { 1051 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; 1052 } 1053 1054 static inline u64 global_rt_runtime(void) 1055 { 1056 if (sysctl_sched_rt_runtime < 0) 1057 return RUNTIME_INF; 1058 1059 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; 1060 } 1061 1062 static inline int task_current(struct rq *rq, struct task_struct *p) 1063 { 1064 return rq->curr == p; 1065 } 1066 1067 static inline int task_running(struct rq *rq, struct task_struct *p) 1068 { 1069 #ifdef CONFIG_SMP 1070 return p->on_cpu; 1071 #else 1072 return task_current(rq, p); 1073 #endif 1074 } 1075 1076 static inline int task_on_rq_queued(struct task_struct *p) 1077 { 1078 return p->on_rq == TASK_ON_RQ_QUEUED; 1079 } 1080 1081 static inline int task_on_rq_migrating(struct task_struct *p) 1082 { 1083 return p->on_rq == TASK_ON_RQ_MIGRATING; 1084 } 1085 1086 #ifndef prepare_arch_switch 1087 # define prepare_arch_switch(next) do { } while (0) 1088 #endif 1089 #ifndef finish_arch_post_lock_switch 1090 # define finish_arch_post_lock_switch() do { } while (0) 1091 #endif 1092 1093 static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) 1094 { 1095 #ifdef CONFIG_SMP 1096 /* 1097 * We can optimise this out completely for !SMP, because the 1098 * SMP rebalancing from interrupt is the only thing that cares 1099 * here. 1100 */ 1101 next->on_cpu = 1; 1102 #endif 1103 } 1104 1105 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) 1106 { 1107 #ifdef CONFIG_SMP 1108 /* 1109 * After ->on_cpu is cleared, the task can be moved to a different CPU. 1110 * We must ensure this doesn't happen until the switch is completely 1111 * finished. 1112 * 1113 * In particular, the load of prev->state in finish_task_switch() must 1114 * happen before this. 1115 * 1116 * Pairs with the smp_cond_acquire() in try_to_wake_up(). 1117 */ 1118 smp_store_release(&prev->on_cpu, 0); 1119 #endif 1120 #ifdef CONFIG_DEBUG_SPINLOCK 1121 /* this is a valid case when another task releases the spinlock */ 1122 rq->lock.owner = current; 1123 #endif 1124 /* 1125 * If we are tracking spinlock dependencies then we have to 1126 * fix up the runqueue lock - which gets 'carried over' from 1127 * prev into current: 1128 */ 1129 spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); 1130 1131 raw_spin_unlock_irq(&rq->lock); 1132 } 1133 1134 /* 1135 * wake flags 1136 */ 1137 #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */ 1138 #define WF_FORK 0x02 /* child wakeup after fork */ 1139 #define WF_MIGRATED 0x4 /* internal use, task got migrated */ 1140 1141 /* 1142 * To aid in avoiding the subversion of "niceness" due to uneven distribution 1143 * of tasks with abnormal "nice" values across CPUs the contribution that 1144 * each task makes to its run queue's load is weighted according to its 1145 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a 1146 * scaled version of the new time slice allocation that they receive on time 1147 * slice expiry etc. 1148 */ 1149 1150 #define WEIGHT_IDLEPRIO 3 1151 #define WMULT_IDLEPRIO 1431655765 1152 1153 extern const int sched_prio_to_weight[40]; 1154 extern const u32 sched_prio_to_wmult[40]; 1155 1156 /* 1157 * {de,en}queue flags: 1158 * 1159 * DEQUEUE_SLEEP - task is no longer runnable 1160 * ENQUEUE_WAKEUP - task just became runnable 1161 * 1162 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks 1163 * are in a known state which allows modification. Such pairs 1164 * should preserve as much state as possible. 1165 * 1166 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location 1167 * in the runqueue. 1168 * 1169 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified) 1170 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline) 1171 * ENQUEUE_MIGRATED - the task was migrated during wakeup 1172 * 1173 */ 1174 1175 #define DEQUEUE_SLEEP 0x01 1176 #define DEQUEUE_SAVE 0x02 /* matches ENQUEUE_RESTORE */ 1177 #define DEQUEUE_MOVE 0x04 /* matches ENQUEUE_MOVE */ 1178 1179 #define ENQUEUE_WAKEUP 0x01 1180 #define ENQUEUE_RESTORE 0x02 1181 #define ENQUEUE_MOVE 0x04 1182 1183 #define ENQUEUE_HEAD 0x08 1184 #define ENQUEUE_REPLENISH 0x10 1185 #ifdef CONFIG_SMP 1186 #define ENQUEUE_MIGRATED 0x20 1187 #else 1188 #define ENQUEUE_MIGRATED 0x00 1189 #endif 1190 1191 #define RETRY_TASK ((void *)-1UL) 1192 1193 struct sched_class { 1194 const struct sched_class *next; 1195 1196 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags); 1197 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags); 1198 void (*yield_task) (struct rq *rq); 1199 bool (*yield_to_task) (struct rq *rq, struct task_struct *p, bool preempt); 1200 1201 void (*check_preempt_curr) (struct rq *rq, struct task_struct *p, int flags); 1202 1203 /* 1204 * It is the responsibility of the pick_next_task() method that will 1205 * return the next task to call put_prev_task() on the @prev task or 1206 * something equivalent. 1207 * 1208 * May return RETRY_TASK when it finds a higher prio class has runnable 1209 * tasks. 1210 */ 1211 struct task_struct * (*pick_next_task) (struct rq *rq, 1212 struct task_struct *prev, 1213 struct pin_cookie cookie); 1214 void (*put_prev_task) (struct rq *rq, struct task_struct *p); 1215 1216 #ifdef CONFIG_SMP 1217 int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags); 1218 void (*migrate_task_rq)(struct task_struct *p); 1219 1220 void (*task_woken) (struct rq *this_rq, struct task_struct *task); 1221 1222 void (*set_cpus_allowed)(struct task_struct *p, 1223 const struct cpumask *newmask); 1224 1225 void (*rq_online)(struct rq *rq); 1226 void (*rq_offline)(struct rq *rq); 1227 #endif 1228 1229 void (*set_curr_task) (struct rq *rq); 1230 void (*task_tick) (struct rq *rq, struct task_struct *p, int queued); 1231 void (*task_fork) (struct task_struct *p); 1232 void (*task_dead) (struct task_struct *p); 1233 1234 /* 1235 * The switched_from() call is allowed to drop rq->lock, therefore we 1236 * cannot assume the switched_from/switched_to pair is serliazed by 1237 * rq->lock. They are however serialized by p->pi_lock. 1238 */ 1239 void (*switched_from) (struct rq *this_rq, struct task_struct *task); 1240 void (*switched_to) (struct rq *this_rq, struct task_struct *task); 1241 void (*prio_changed) (struct rq *this_rq, struct task_struct *task, 1242 int oldprio); 1243 1244 unsigned int (*get_rr_interval) (struct rq *rq, 1245 struct task_struct *task); 1246 1247 void (*update_curr) (struct rq *rq); 1248 1249 #ifdef CONFIG_FAIR_GROUP_SCHED 1250 void (*task_move_group) (struct task_struct *p); 1251 #endif 1252 }; 1253 1254 static inline void put_prev_task(struct rq *rq, struct task_struct *prev) 1255 { 1256 prev->sched_class->put_prev_task(rq, prev); 1257 } 1258 1259 #define sched_class_highest (&stop_sched_class) 1260 #define for_each_class(class) \ 1261 for (class = sched_class_highest; class; class = class->next) 1262 1263 extern const struct sched_class stop_sched_class; 1264 extern const struct sched_class dl_sched_class; 1265 extern const struct sched_class rt_sched_class; 1266 extern const struct sched_class fair_sched_class; 1267 extern const struct sched_class idle_sched_class; 1268 1269 1270 #ifdef CONFIG_SMP 1271 1272 extern void update_group_capacity(struct sched_domain *sd, int cpu); 1273 1274 extern void trigger_load_balance(struct rq *rq); 1275 1276 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask); 1277 1278 #endif 1279 1280 #ifdef CONFIG_CPU_IDLE 1281 static inline void idle_set_state(struct rq *rq, 1282 struct cpuidle_state *idle_state) 1283 { 1284 rq->idle_state = idle_state; 1285 } 1286 1287 static inline struct cpuidle_state *idle_get_state(struct rq *rq) 1288 { 1289 WARN_ON(!rcu_read_lock_held()); 1290 return rq->idle_state; 1291 } 1292 #else 1293 static inline void idle_set_state(struct rq *rq, 1294 struct cpuidle_state *idle_state) 1295 { 1296 } 1297 1298 static inline struct cpuidle_state *idle_get_state(struct rq *rq) 1299 { 1300 return NULL; 1301 } 1302 #endif 1303 1304 extern void sysrq_sched_debug_show(void); 1305 extern void sched_init_granularity(void); 1306 extern void update_max_interval(void); 1307 1308 extern void init_sched_dl_class(void); 1309 extern void init_sched_rt_class(void); 1310 extern void init_sched_fair_class(void); 1311 1312 extern void resched_curr(struct rq *rq); 1313 extern void resched_cpu(int cpu); 1314 1315 extern struct rt_bandwidth def_rt_bandwidth; 1316 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime); 1317 1318 extern struct dl_bandwidth def_dl_bandwidth; 1319 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime); 1320 extern void init_dl_task_timer(struct sched_dl_entity *dl_se); 1321 1322 unsigned long to_ratio(u64 period, u64 runtime); 1323 1324 extern void init_entity_runnable_average(struct sched_entity *se); 1325 extern void post_init_entity_util_avg(struct sched_entity *se); 1326 1327 #ifdef CONFIG_NO_HZ_FULL 1328 extern bool sched_can_stop_tick(struct rq *rq); 1329 1330 /* 1331 * Tick may be needed by tasks in the runqueue depending on their policy and 1332 * requirements. If tick is needed, lets send the target an IPI to kick it out of 1333 * nohz mode if necessary. 1334 */ 1335 static inline void sched_update_tick_dependency(struct rq *rq) 1336 { 1337 int cpu; 1338 1339 if (!tick_nohz_full_enabled()) 1340 return; 1341 1342 cpu = cpu_of(rq); 1343 1344 if (!tick_nohz_full_cpu(cpu)) 1345 return; 1346 1347 if (sched_can_stop_tick(rq)) 1348 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED); 1349 else 1350 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED); 1351 } 1352 #else 1353 static inline void sched_update_tick_dependency(struct rq *rq) { } 1354 #endif 1355 1356 static inline void add_nr_running(struct rq *rq, unsigned count) 1357 { 1358 unsigned prev_nr = rq->nr_running; 1359 1360 rq->nr_running = prev_nr + count; 1361 1362 if (prev_nr < 2 && rq->nr_running >= 2) { 1363 #ifdef CONFIG_SMP 1364 if (!rq->rd->overload) 1365 rq->rd->overload = true; 1366 #endif 1367 } 1368 1369 sched_update_tick_dependency(rq); 1370 } 1371 1372 static inline void sub_nr_running(struct rq *rq, unsigned count) 1373 { 1374 rq->nr_running -= count; 1375 /* Check if we still need preemption */ 1376 sched_update_tick_dependency(rq); 1377 } 1378 1379 static inline void rq_last_tick_reset(struct rq *rq) 1380 { 1381 #ifdef CONFIG_NO_HZ_FULL 1382 rq->last_sched_tick = jiffies; 1383 #endif 1384 } 1385 1386 extern void update_rq_clock(struct rq *rq); 1387 1388 extern void activate_task(struct rq *rq, struct task_struct *p, int flags); 1389 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags); 1390 1391 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags); 1392 1393 extern const_debug unsigned int sysctl_sched_time_avg; 1394 extern const_debug unsigned int sysctl_sched_nr_migrate; 1395 extern const_debug unsigned int sysctl_sched_migration_cost; 1396 1397 static inline u64 sched_avg_period(void) 1398 { 1399 return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; 1400 } 1401 1402 #ifdef CONFIG_SCHED_HRTICK 1403 1404 /* 1405 * Use hrtick when: 1406 * - enabled by features 1407 * - hrtimer is actually high res 1408 */ 1409 static inline int hrtick_enabled(struct rq *rq) 1410 { 1411 if (!sched_feat(HRTICK)) 1412 return 0; 1413 if (!cpu_active(cpu_of(rq))) 1414 return 0; 1415 return hrtimer_is_hres_active(&rq->hrtick_timer); 1416 } 1417 1418 void hrtick_start(struct rq *rq, u64 delay); 1419 1420 #else 1421 1422 static inline int hrtick_enabled(struct rq *rq) 1423 { 1424 return 0; 1425 } 1426 1427 #endif /* CONFIG_SCHED_HRTICK */ 1428 1429 #ifdef CONFIG_SMP 1430 extern void sched_avg_update(struct rq *rq); 1431 1432 #ifndef arch_scale_freq_capacity 1433 static __always_inline 1434 unsigned long arch_scale_freq_capacity(struct sched_domain *sd, int cpu) 1435 { 1436 return SCHED_CAPACITY_SCALE; 1437 } 1438 #endif 1439 1440 #ifndef arch_scale_cpu_capacity 1441 static __always_inline 1442 unsigned long arch_scale_cpu_capacity(struct sched_domain *sd, int cpu) 1443 { 1444 if (sd && (sd->flags & SD_SHARE_CPUCAPACITY) && (sd->span_weight > 1)) 1445 return sd->smt_gain / sd->span_weight; 1446 1447 return SCHED_CAPACITY_SCALE; 1448 } 1449 #endif 1450 1451 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) 1452 { 1453 rq->rt_avg += rt_delta * arch_scale_freq_capacity(NULL, cpu_of(rq)); 1454 sched_avg_update(rq); 1455 } 1456 #else 1457 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { } 1458 static inline void sched_avg_update(struct rq *rq) { } 1459 #endif 1460 1461 struct rq_flags { 1462 unsigned long flags; 1463 struct pin_cookie cookie; 1464 }; 1465 1466 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf) 1467 __acquires(rq->lock); 1468 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf) 1469 __acquires(p->pi_lock) 1470 __acquires(rq->lock); 1471 1472 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf) 1473 __releases(rq->lock) 1474 { 1475 lockdep_unpin_lock(&rq->lock, rf->cookie); 1476 raw_spin_unlock(&rq->lock); 1477 } 1478 1479 static inline void 1480 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf) 1481 __releases(rq->lock) 1482 __releases(p->pi_lock) 1483 { 1484 lockdep_unpin_lock(&rq->lock, rf->cookie); 1485 raw_spin_unlock(&rq->lock); 1486 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags); 1487 } 1488 1489 #ifdef CONFIG_SMP 1490 #ifdef CONFIG_PREEMPT 1491 1492 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2); 1493 1494 /* 1495 * fair double_lock_balance: Safely acquires both rq->locks in a fair 1496 * way at the expense of forcing extra atomic operations in all 1497 * invocations. This assures that the double_lock is acquired using the 1498 * same underlying policy as the spinlock_t on this architecture, which 1499 * reduces latency compared to the unfair variant below. However, it 1500 * also adds more overhead and therefore may reduce throughput. 1501 */ 1502 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) 1503 __releases(this_rq->lock) 1504 __acquires(busiest->lock) 1505 __acquires(this_rq->lock) 1506 { 1507 raw_spin_unlock(&this_rq->lock); 1508 double_rq_lock(this_rq, busiest); 1509 1510 return 1; 1511 } 1512 1513 #else 1514 /* 1515 * Unfair double_lock_balance: Optimizes throughput at the expense of 1516 * latency by eliminating extra atomic operations when the locks are 1517 * already in proper order on entry. This favors lower cpu-ids and will 1518 * grant the double lock to lower cpus over higher ids under contention, 1519 * regardless of entry order into the function. 1520 */ 1521 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) 1522 __releases(this_rq->lock) 1523 __acquires(busiest->lock) 1524 __acquires(this_rq->lock) 1525 { 1526 int ret = 0; 1527 1528 if (unlikely(!raw_spin_trylock(&busiest->lock))) { 1529 if (busiest < this_rq) { 1530 raw_spin_unlock(&this_rq->lock); 1531 raw_spin_lock(&busiest->lock); 1532 raw_spin_lock_nested(&this_rq->lock, 1533 SINGLE_DEPTH_NESTING); 1534 ret = 1; 1535 } else 1536 raw_spin_lock_nested(&busiest->lock, 1537 SINGLE_DEPTH_NESTING); 1538 } 1539 return ret; 1540 } 1541 1542 #endif /* CONFIG_PREEMPT */ 1543 1544 /* 1545 * double_lock_balance - lock the busiest runqueue, this_rq is locked already. 1546 */ 1547 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest) 1548 { 1549 if (unlikely(!irqs_disabled())) { 1550 /* printk() doesn't work good under rq->lock */ 1551 raw_spin_unlock(&this_rq->lock); 1552 BUG_ON(1); 1553 } 1554 1555 return _double_lock_balance(this_rq, busiest); 1556 } 1557 1558 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) 1559 __releases(busiest->lock) 1560 { 1561 raw_spin_unlock(&busiest->lock); 1562 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); 1563 } 1564 1565 static inline void double_lock(spinlock_t *l1, spinlock_t *l2) 1566 { 1567 if (l1 > l2) 1568 swap(l1, l2); 1569 1570 spin_lock(l1); 1571 spin_lock_nested(l2, SINGLE_DEPTH_NESTING); 1572 } 1573 1574 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2) 1575 { 1576 if (l1 > l2) 1577 swap(l1, l2); 1578 1579 spin_lock_irq(l1); 1580 spin_lock_nested(l2, SINGLE_DEPTH_NESTING); 1581 } 1582 1583 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2) 1584 { 1585 if (l1 > l2) 1586 swap(l1, l2); 1587 1588 raw_spin_lock(l1); 1589 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING); 1590 } 1591 1592 /* 1593 * double_rq_lock - safely lock two runqueues 1594 * 1595 * Note this does not disable interrupts like task_rq_lock, 1596 * you need to do so manually before calling. 1597 */ 1598 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) 1599 __acquires(rq1->lock) 1600 __acquires(rq2->lock) 1601 { 1602 BUG_ON(!irqs_disabled()); 1603 if (rq1 == rq2) { 1604 raw_spin_lock(&rq1->lock); 1605 __acquire(rq2->lock); /* Fake it out ;) */ 1606 } else { 1607 if (rq1 < rq2) { 1608 raw_spin_lock(&rq1->lock); 1609 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); 1610 } else { 1611 raw_spin_lock(&rq2->lock); 1612 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); 1613 } 1614 } 1615 } 1616 1617 /* 1618 * double_rq_unlock - safely unlock two runqueues 1619 * 1620 * Note this does not restore interrupts like task_rq_unlock, 1621 * you need to do so manually after calling. 1622 */ 1623 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) 1624 __releases(rq1->lock) 1625 __releases(rq2->lock) 1626 { 1627 raw_spin_unlock(&rq1->lock); 1628 if (rq1 != rq2) 1629 raw_spin_unlock(&rq2->lock); 1630 else 1631 __release(rq2->lock); 1632 } 1633 1634 #else /* CONFIG_SMP */ 1635 1636 /* 1637 * double_rq_lock - safely lock two runqueues 1638 * 1639 * Note this does not disable interrupts like task_rq_lock, 1640 * you need to do so manually before calling. 1641 */ 1642 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) 1643 __acquires(rq1->lock) 1644 __acquires(rq2->lock) 1645 { 1646 BUG_ON(!irqs_disabled()); 1647 BUG_ON(rq1 != rq2); 1648 raw_spin_lock(&rq1->lock); 1649 __acquire(rq2->lock); /* Fake it out ;) */ 1650 } 1651 1652 /* 1653 * double_rq_unlock - safely unlock two runqueues 1654 * 1655 * Note this does not restore interrupts like task_rq_unlock, 1656 * you need to do so manually after calling. 1657 */ 1658 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) 1659 __releases(rq1->lock) 1660 __releases(rq2->lock) 1661 { 1662 BUG_ON(rq1 != rq2); 1663 raw_spin_unlock(&rq1->lock); 1664 __release(rq2->lock); 1665 } 1666 1667 #endif 1668 1669 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq); 1670 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq); 1671 1672 #ifdef CONFIG_SCHED_DEBUG 1673 extern void print_cfs_stats(struct seq_file *m, int cpu); 1674 extern void print_rt_stats(struct seq_file *m, int cpu); 1675 extern void print_dl_stats(struct seq_file *m, int cpu); 1676 extern void 1677 print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq); 1678 1679 #ifdef CONFIG_NUMA_BALANCING 1680 extern void 1681 show_numa_stats(struct task_struct *p, struct seq_file *m); 1682 extern void 1683 print_numa_stats(struct seq_file *m, int node, unsigned long tsf, 1684 unsigned long tpf, unsigned long gsf, unsigned long gpf); 1685 #endif /* CONFIG_NUMA_BALANCING */ 1686 #endif /* CONFIG_SCHED_DEBUG */ 1687 1688 extern void init_cfs_rq(struct cfs_rq *cfs_rq); 1689 extern void init_rt_rq(struct rt_rq *rt_rq); 1690 extern void init_dl_rq(struct dl_rq *dl_rq); 1691 1692 extern void cfs_bandwidth_usage_inc(void); 1693 extern void cfs_bandwidth_usage_dec(void); 1694 1695 #ifdef CONFIG_NO_HZ_COMMON 1696 enum rq_nohz_flag_bits { 1697 NOHZ_TICK_STOPPED, 1698 NOHZ_BALANCE_KICK, 1699 }; 1700 1701 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags) 1702 1703 extern void nohz_balance_exit_idle(unsigned int cpu); 1704 #else 1705 static inline void nohz_balance_exit_idle(unsigned int cpu) { } 1706 #endif 1707 1708 #ifdef CONFIG_IRQ_TIME_ACCOUNTING 1709 1710 DECLARE_PER_CPU(u64, cpu_hardirq_time); 1711 DECLARE_PER_CPU(u64, cpu_softirq_time); 1712 1713 #ifndef CONFIG_64BIT 1714 DECLARE_PER_CPU(seqcount_t, irq_time_seq); 1715 1716 static inline void irq_time_write_begin(void) 1717 { 1718 __this_cpu_inc(irq_time_seq.sequence); 1719 smp_wmb(); 1720 } 1721 1722 static inline void irq_time_write_end(void) 1723 { 1724 smp_wmb(); 1725 __this_cpu_inc(irq_time_seq.sequence); 1726 } 1727 1728 static inline u64 irq_time_read(int cpu) 1729 { 1730 u64 irq_time; 1731 unsigned seq; 1732 1733 do { 1734 seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu)); 1735 irq_time = per_cpu(cpu_softirq_time, cpu) + 1736 per_cpu(cpu_hardirq_time, cpu); 1737 } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq)); 1738 1739 return irq_time; 1740 } 1741 #else /* CONFIG_64BIT */ 1742 static inline void irq_time_write_begin(void) 1743 { 1744 } 1745 1746 static inline void irq_time_write_end(void) 1747 { 1748 } 1749 1750 static inline u64 irq_time_read(int cpu) 1751 { 1752 return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu); 1753 } 1754 #endif /* CONFIG_64BIT */ 1755 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ 1756 1757 #ifdef CONFIG_CPU_FREQ 1758 DECLARE_PER_CPU(struct update_util_data *, cpufreq_update_util_data); 1759 1760 /** 1761 * cpufreq_update_util - Take a note about CPU utilization changes. 1762 * @time: Current time. 1763 * @util: Current utilization. 1764 * @max: Utilization ceiling. 1765 * 1766 * This function is called by the scheduler on every invocation of 1767 * update_load_avg() on the CPU whose utilization is being updated. 1768 * 1769 * It can only be called from RCU-sched read-side critical sections. 1770 */ 1771 static inline void cpufreq_update_util(u64 time, unsigned long util, unsigned long max) 1772 { 1773 struct update_util_data *data; 1774 1775 data = rcu_dereference_sched(*this_cpu_ptr(&cpufreq_update_util_data)); 1776 if (data) 1777 data->func(data, time, util, max); 1778 } 1779 1780 /** 1781 * cpufreq_trigger_update - Trigger CPU performance state evaluation if needed. 1782 * @time: Current time. 1783 * 1784 * The way cpufreq is currently arranged requires it to evaluate the CPU 1785 * performance state (frequency/voltage) on a regular basis to prevent it from 1786 * being stuck in a completely inadequate performance level for too long. 1787 * That is not guaranteed to happen if the updates are only triggered from CFS, 1788 * though, because they may not be coming in if RT or deadline tasks are active 1789 * all the time (or there are RT and DL tasks only). 1790 * 1791 * As a workaround for that issue, this function is called by the RT and DL 1792 * sched classes to trigger extra cpufreq updates to prevent it from stalling, 1793 * but that really is a band-aid. Going forward it should be replaced with 1794 * solutions targeted more specifically at RT and DL tasks. 1795 */ 1796 static inline void cpufreq_trigger_update(u64 time) 1797 { 1798 cpufreq_update_util(time, ULONG_MAX, 0); 1799 } 1800 #else 1801 static inline void cpufreq_update_util(u64 time, unsigned long util, unsigned long max) {} 1802 static inline void cpufreq_trigger_update(u64 time) {} 1803 #endif /* CONFIG_CPU_FREQ */ 1804 1805 #ifdef arch_scale_freq_capacity 1806 #ifndef arch_scale_freq_invariant 1807 #define arch_scale_freq_invariant() (true) 1808 #endif 1809 #else /* arch_scale_freq_capacity */ 1810 #define arch_scale_freq_invariant() (false) 1811 #endif 1812 1813 static inline void account_reset_rq(struct rq *rq) 1814 { 1815 #ifdef CONFIG_IRQ_TIME_ACCOUNTING 1816 rq->prev_irq_time = 0; 1817 #endif 1818 #ifdef CONFIG_PARAVIRT 1819 rq->prev_steal_time = 0; 1820 #endif 1821 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING 1822 rq->prev_steal_time_rq = 0; 1823 #endif 1824 } 1825