1 /* SPDX-License-Identifier: GPL-2.0 */ 2 /* 3 * Scheduler internal types and methods: 4 */ 5 #include <linux/sched.h> 6 7 #include <linux/sched/autogroup.h> 8 #include <linux/sched/clock.h> 9 #include <linux/sched/coredump.h> 10 #include <linux/sched/cpufreq.h> 11 #include <linux/sched/cputime.h> 12 #include <linux/sched/deadline.h> 13 #include <linux/sched/debug.h> 14 #include <linux/sched/hotplug.h> 15 #include <linux/sched/idle.h> 16 #include <linux/sched/init.h> 17 #include <linux/sched/isolation.h> 18 #include <linux/sched/jobctl.h> 19 #include <linux/sched/loadavg.h> 20 #include <linux/sched/mm.h> 21 #include <linux/sched/nohz.h> 22 #include <linux/sched/numa_balancing.h> 23 #include <linux/sched/prio.h> 24 #include <linux/sched/rt.h> 25 #include <linux/sched/signal.h> 26 #include <linux/sched/stat.h> 27 #include <linux/sched/sysctl.h> 28 #include <linux/sched/task.h> 29 #include <linux/sched/task_stack.h> 30 #include <linux/sched/topology.h> 31 #include <linux/sched/user.h> 32 #include <linux/sched/wake_q.h> 33 #include <linux/sched/xacct.h> 34 35 #include <uapi/linux/sched/types.h> 36 37 #include <linux/binfmts.h> 38 #include <linux/blkdev.h> 39 #include <linux/compat.h> 40 #include <linux/context_tracking.h> 41 #include <linux/cpufreq.h> 42 #include <linux/cpuidle.h> 43 #include <linux/cpuset.h> 44 #include <linux/ctype.h> 45 #include <linux/debugfs.h> 46 #include <linux/delayacct.h> 47 #include <linux/init_task.h> 48 #include <linux/kprobes.h> 49 #include <linux/kthread.h> 50 #include <linux/membarrier.h> 51 #include <linux/migrate.h> 52 #include <linux/mmu_context.h> 53 #include <linux/nmi.h> 54 #include <linux/proc_fs.h> 55 #include <linux/prefetch.h> 56 #include <linux/profile.h> 57 #include <linux/rcupdate_wait.h> 58 #include <linux/security.h> 59 #include <linux/stackprotector.h> 60 #include <linux/stop_machine.h> 61 #include <linux/suspend.h> 62 #include <linux/swait.h> 63 #include <linux/syscalls.h> 64 #include <linux/task_work.h> 65 #include <linux/tsacct_kern.h> 66 67 #include <asm/tlb.h> 68 69 #ifdef CONFIG_PARAVIRT 70 # include <asm/paravirt.h> 71 #endif 72 73 #include "cpupri.h" 74 #include "cpudeadline.h" 75 76 #ifdef CONFIG_SCHED_DEBUG 77 # define SCHED_WARN_ON(x) WARN_ONCE(x, #x) 78 #else 79 # define SCHED_WARN_ON(x) ({ (void)(x), 0; }) 80 #endif 81 82 struct rq; 83 struct cpuidle_state; 84 85 /* task_struct::on_rq states: */ 86 #define TASK_ON_RQ_QUEUED 1 87 #define TASK_ON_RQ_MIGRATING 2 88 89 extern __read_mostly int scheduler_running; 90 91 extern unsigned long calc_load_update; 92 extern atomic_long_t calc_load_tasks; 93 94 extern void calc_global_load_tick(struct rq *this_rq); 95 extern long calc_load_fold_active(struct rq *this_rq, long adjust); 96 97 #ifdef CONFIG_SMP 98 extern void cpu_load_update_active(struct rq *this_rq); 99 #else 100 static inline void cpu_load_update_active(struct rq *this_rq) { } 101 #endif 102 103 /* 104 * Helpers for converting nanosecond timing to jiffy resolution 105 */ 106 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) 107 108 /* 109 * Increase resolution of nice-level calculations for 64-bit architectures. 110 * The extra resolution improves shares distribution and load balancing of 111 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup 112 * hierarchies, especially on larger systems. This is not a user-visible change 113 * and does not change the user-interface for setting shares/weights. 114 * 115 * We increase resolution only if we have enough bits to allow this increased 116 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit 117 * are pretty high and the returns do not justify the increased costs. 118 * 119 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to 120 * increase coverage and consistency always enable it on 64-bit platforms. 121 */ 122 #ifdef CONFIG_64BIT 123 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT) 124 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT) 125 # define scale_load_down(w) ((w) >> SCHED_FIXEDPOINT_SHIFT) 126 #else 127 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT) 128 # define scale_load(w) (w) 129 # define scale_load_down(w) (w) 130 #endif 131 132 /* 133 * Task weight (visible to users) and its load (invisible to users) have 134 * independent resolution, but they should be well calibrated. We use 135 * scale_load() and scale_load_down(w) to convert between them. The 136 * following must be true: 137 * 138 * scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD 139 * 140 */ 141 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT) 142 143 /* 144 * Single value that decides SCHED_DEADLINE internal math precision. 145 * 10 -> just above 1us 146 * 9 -> just above 0.5us 147 */ 148 #define DL_SCALE 10 149 150 /* 151 * Single value that denotes runtime == period, ie unlimited time. 152 */ 153 #define RUNTIME_INF ((u64)~0ULL) 154 155 static inline int idle_policy(int policy) 156 { 157 return policy == SCHED_IDLE; 158 } 159 static inline int fair_policy(int policy) 160 { 161 return policy == SCHED_NORMAL || policy == SCHED_BATCH; 162 } 163 164 static inline int rt_policy(int policy) 165 { 166 return policy == SCHED_FIFO || policy == SCHED_RR; 167 } 168 169 static inline int dl_policy(int policy) 170 { 171 return policy == SCHED_DEADLINE; 172 } 173 static inline bool valid_policy(int policy) 174 { 175 return idle_policy(policy) || fair_policy(policy) || 176 rt_policy(policy) || dl_policy(policy); 177 } 178 179 static inline int task_has_rt_policy(struct task_struct *p) 180 { 181 return rt_policy(p->policy); 182 } 183 184 static inline int task_has_dl_policy(struct task_struct *p) 185 { 186 return dl_policy(p->policy); 187 } 188 189 #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT) 190 191 /* 192 * !! For sched_setattr_nocheck() (kernel) only !! 193 * 194 * This is actually gross. :( 195 * 196 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE 197 * tasks, but still be able to sleep. We need this on platforms that cannot 198 * atomically change clock frequency. Remove once fast switching will be 199 * available on such platforms. 200 * 201 * SUGOV stands for SchedUtil GOVernor. 202 */ 203 #define SCHED_FLAG_SUGOV 0x10000000 204 205 static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se) 206 { 207 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL 208 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV); 209 #else 210 return false; 211 #endif 212 } 213 214 /* 215 * Tells if entity @a should preempt entity @b. 216 */ 217 static inline bool 218 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b) 219 { 220 return dl_entity_is_special(a) || 221 dl_time_before(a->deadline, b->deadline); 222 } 223 224 /* 225 * This is the priority-queue data structure of the RT scheduling class: 226 */ 227 struct rt_prio_array { 228 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ 229 struct list_head queue[MAX_RT_PRIO]; 230 }; 231 232 struct rt_bandwidth { 233 /* nests inside the rq lock: */ 234 raw_spinlock_t rt_runtime_lock; 235 ktime_t rt_period; 236 u64 rt_runtime; 237 struct hrtimer rt_period_timer; 238 unsigned int rt_period_active; 239 }; 240 241 void __dl_clear_params(struct task_struct *p); 242 243 /* 244 * To keep the bandwidth of -deadline tasks and groups under control 245 * we need some place where: 246 * - store the maximum -deadline bandwidth of the system (the group); 247 * - cache the fraction of that bandwidth that is currently allocated. 248 * 249 * This is all done in the data structure below. It is similar to the 250 * one used for RT-throttling (rt_bandwidth), with the main difference 251 * that, since here we are only interested in admission control, we 252 * do not decrease any runtime while the group "executes", neither we 253 * need a timer to replenish it. 254 * 255 * With respect to SMP, the bandwidth is given on a per-CPU basis, 256 * meaning that: 257 * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU; 258 * - dl_total_bw array contains, in the i-eth element, the currently 259 * allocated bandwidth on the i-eth CPU. 260 * Moreover, groups consume bandwidth on each CPU, while tasks only 261 * consume bandwidth on the CPU they're running on. 262 * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw 263 * that will be shown the next time the proc or cgroup controls will 264 * be red. It on its turn can be changed by writing on its own 265 * control. 266 */ 267 struct dl_bandwidth { 268 raw_spinlock_t dl_runtime_lock; 269 u64 dl_runtime; 270 u64 dl_period; 271 }; 272 273 static inline int dl_bandwidth_enabled(void) 274 { 275 return sysctl_sched_rt_runtime >= 0; 276 } 277 278 struct dl_bw { 279 raw_spinlock_t lock; 280 u64 bw; 281 u64 total_bw; 282 }; 283 284 static inline void __dl_update(struct dl_bw *dl_b, s64 bw); 285 286 static inline 287 void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus) 288 { 289 dl_b->total_bw -= tsk_bw; 290 __dl_update(dl_b, (s32)tsk_bw / cpus); 291 } 292 293 static inline 294 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus) 295 { 296 dl_b->total_bw += tsk_bw; 297 __dl_update(dl_b, -((s32)tsk_bw / cpus)); 298 } 299 300 static inline 301 bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw) 302 { 303 return dl_b->bw != -1 && 304 dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw; 305 } 306 307 extern void dl_change_utilization(struct task_struct *p, u64 new_bw); 308 extern void init_dl_bw(struct dl_bw *dl_b); 309 extern int sched_dl_global_validate(void); 310 extern void sched_dl_do_global(void); 311 extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr); 312 extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr); 313 extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr); 314 extern bool __checkparam_dl(const struct sched_attr *attr); 315 extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr); 316 extern int dl_task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed); 317 extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial); 318 extern bool dl_cpu_busy(unsigned int cpu); 319 320 #ifdef CONFIG_CGROUP_SCHED 321 322 #include <linux/cgroup.h> 323 324 struct cfs_rq; 325 struct rt_rq; 326 327 extern struct list_head task_groups; 328 329 struct cfs_bandwidth { 330 #ifdef CONFIG_CFS_BANDWIDTH 331 raw_spinlock_t lock; 332 ktime_t period; 333 u64 quota; 334 u64 runtime; 335 s64 hierarchical_quota; 336 u64 runtime_expires; 337 int expires_seq; 338 339 short idle; 340 short period_active; 341 struct hrtimer period_timer; 342 struct hrtimer slack_timer; 343 struct list_head throttled_cfs_rq; 344 345 /* Statistics: */ 346 int nr_periods; 347 int nr_throttled; 348 u64 throttled_time; 349 #endif 350 }; 351 352 /* Task group related information */ 353 struct task_group { 354 struct cgroup_subsys_state css; 355 356 #ifdef CONFIG_FAIR_GROUP_SCHED 357 /* schedulable entities of this group on each CPU */ 358 struct sched_entity **se; 359 /* runqueue "owned" by this group on each CPU */ 360 struct cfs_rq **cfs_rq; 361 unsigned long shares; 362 363 #ifdef CONFIG_SMP 364 /* 365 * load_avg can be heavily contended at clock tick time, so put 366 * it in its own cacheline separated from the fields above which 367 * will also be accessed at each tick. 368 */ 369 atomic_long_t load_avg ____cacheline_aligned; 370 #endif 371 #endif 372 373 #ifdef CONFIG_RT_GROUP_SCHED 374 struct sched_rt_entity **rt_se; 375 struct rt_rq **rt_rq; 376 377 struct rt_bandwidth rt_bandwidth; 378 #endif 379 380 struct rcu_head rcu; 381 struct list_head list; 382 383 struct task_group *parent; 384 struct list_head siblings; 385 struct list_head children; 386 387 #ifdef CONFIG_SCHED_AUTOGROUP 388 struct autogroup *autogroup; 389 #endif 390 391 struct cfs_bandwidth cfs_bandwidth; 392 }; 393 394 #ifdef CONFIG_FAIR_GROUP_SCHED 395 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD 396 397 /* 398 * A weight of 0 or 1 can cause arithmetics problems. 399 * A weight of a cfs_rq is the sum of weights of which entities 400 * are queued on this cfs_rq, so a weight of a entity should not be 401 * too large, so as the shares value of a task group. 402 * (The default weight is 1024 - so there's no practical 403 * limitation from this.) 404 */ 405 #define MIN_SHARES (1UL << 1) 406 #define MAX_SHARES (1UL << 18) 407 #endif 408 409 typedef int (*tg_visitor)(struct task_group *, void *); 410 411 extern int walk_tg_tree_from(struct task_group *from, 412 tg_visitor down, tg_visitor up, void *data); 413 414 /* 415 * Iterate the full tree, calling @down when first entering a node and @up when 416 * leaving it for the final time. 417 * 418 * Caller must hold rcu_lock or sufficient equivalent. 419 */ 420 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) 421 { 422 return walk_tg_tree_from(&root_task_group, down, up, data); 423 } 424 425 extern int tg_nop(struct task_group *tg, void *data); 426 427 extern void free_fair_sched_group(struct task_group *tg); 428 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent); 429 extern void online_fair_sched_group(struct task_group *tg); 430 extern void unregister_fair_sched_group(struct task_group *tg); 431 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, 432 struct sched_entity *se, int cpu, 433 struct sched_entity *parent); 434 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b); 435 436 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b); 437 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b); 438 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq); 439 440 extern void free_rt_sched_group(struct task_group *tg); 441 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent); 442 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, 443 struct sched_rt_entity *rt_se, int cpu, 444 struct sched_rt_entity *parent); 445 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us); 446 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us); 447 extern long sched_group_rt_runtime(struct task_group *tg); 448 extern long sched_group_rt_period(struct task_group *tg); 449 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk); 450 451 extern struct task_group *sched_create_group(struct task_group *parent); 452 extern void sched_online_group(struct task_group *tg, 453 struct task_group *parent); 454 extern void sched_destroy_group(struct task_group *tg); 455 extern void sched_offline_group(struct task_group *tg); 456 457 extern void sched_move_task(struct task_struct *tsk); 458 459 #ifdef CONFIG_FAIR_GROUP_SCHED 460 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares); 461 462 #ifdef CONFIG_SMP 463 extern void set_task_rq_fair(struct sched_entity *se, 464 struct cfs_rq *prev, struct cfs_rq *next); 465 #else /* !CONFIG_SMP */ 466 static inline void set_task_rq_fair(struct sched_entity *se, 467 struct cfs_rq *prev, struct cfs_rq *next) { } 468 #endif /* CONFIG_SMP */ 469 #endif /* CONFIG_FAIR_GROUP_SCHED */ 470 471 #else /* CONFIG_CGROUP_SCHED */ 472 473 struct cfs_bandwidth { }; 474 475 #endif /* CONFIG_CGROUP_SCHED */ 476 477 /* CFS-related fields in a runqueue */ 478 struct cfs_rq { 479 struct load_weight load; 480 unsigned long runnable_weight; 481 unsigned int nr_running; 482 unsigned int h_nr_running; 483 484 u64 exec_clock; 485 u64 min_vruntime; 486 #ifndef CONFIG_64BIT 487 u64 min_vruntime_copy; 488 #endif 489 490 struct rb_root_cached tasks_timeline; 491 492 /* 493 * 'curr' points to currently running entity on this cfs_rq. 494 * It is set to NULL otherwise (i.e when none are currently running). 495 */ 496 struct sched_entity *curr; 497 struct sched_entity *next; 498 struct sched_entity *last; 499 struct sched_entity *skip; 500 501 #ifdef CONFIG_SCHED_DEBUG 502 unsigned int nr_spread_over; 503 #endif 504 505 #ifdef CONFIG_SMP 506 /* 507 * CFS load tracking 508 */ 509 struct sched_avg avg; 510 #ifndef CONFIG_64BIT 511 u64 load_last_update_time_copy; 512 #endif 513 struct { 514 raw_spinlock_t lock ____cacheline_aligned; 515 int nr; 516 unsigned long load_avg; 517 unsigned long util_avg; 518 unsigned long runnable_sum; 519 } removed; 520 521 #ifdef CONFIG_FAIR_GROUP_SCHED 522 unsigned long tg_load_avg_contrib; 523 long propagate; 524 long prop_runnable_sum; 525 526 /* 527 * h_load = weight * f(tg) 528 * 529 * Where f(tg) is the recursive weight fraction assigned to 530 * this group. 531 */ 532 unsigned long h_load; 533 u64 last_h_load_update; 534 struct sched_entity *h_load_next; 535 #endif /* CONFIG_FAIR_GROUP_SCHED */ 536 #endif /* CONFIG_SMP */ 537 538 #ifdef CONFIG_FAIR_GROUP_SCHED 539 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */ 540 541 /* 542 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in 543 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities 544 * (like users, containers etc.) 545 * 546 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU. 547 * This list is used during load balance. 548 */ 549 int on_list; 550 struct list_head leaf_cfs_rq_list; 551 struct task_group *tg; /* group that "owns" this runqueue */ 552 553 #ifdef CONFIG_CFS_BANDWIDTH 554 int runtime_enabled; 555 int expires_seq; 556 u64 runtime_expires; 557 s64 runtime_remaining; 558 559 u64 throttled_clock; 560 u64 throttled_clock_task; 561 u64 throttled_clock_task_time; 562 int throttled; 563 int throttle_count; 564 struct list_head throttled_list; 565 #endif /* CONFIG_CFS_BANDWIDTH */ 566 #endif /* CONFIG_FAIR_GROUP_SCHED */ 567 }; 568 569 static inline int rt_bandwidth_enabled(void) 570 { 571 return sysctl_sched_rt_runtime >= 0; 572 } 573 574 /* RT IPI pull logic requires IRQ_WORK */ 575 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP) 576 # define HAVE_RT_PUSH_IPI 577 #endif 578 579 /* Real-Time classes' related field in a runqueue: */ 580 struct rt_rq { 581 struct rt_prio_array active; 582 unsigned int rt_nr_running; 583 unsigned int rr_nr_running; 584 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED 585 struct { 586 int curr; /* highest queued rt task prio */ 587 #ifdef CONFIG_SMP 588 int next; /* next highest */ 589 #endif 590 } highest_prio; 591 #endif 592 #ifdef CONFIG_SMP 593 unsigned long rt_nr_migratory; 594 unsigned long rt_nr_total; 595 int overloaded; 596 struct plist_head pushable_tasks; 597 598 #endif /* CONFIG_SMP */ 599 int rt_queued; 600 601 int rt_throttled; 602 u64 rt_time; 603 u64 rt_runtime; 604 /* Nests inside the rq lock: */ 605 raw_spinlock_t rt_runtime_lock; 606 607 #ifdef CONFIG_RT_GROUP_SCHED 608 unsigned long rt_nr_boosted; 609 610 struct rq *rq; 611 struct task_group *tg; 612 #endif 613 }; 614 615 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq) 616 { 617 return rt_rq->rt_queued && rt_rq->rt_nr_running; 618 } 619 620 /* Deadline class' related fields in a runqueue */ 621 struct dl_rq { 622 /* runqueue is an rbtree, ordered by deadline */ 623 struct rb_root_cached root; 624 625 unsigned long dl_nr_running; 626 627 #ifdef CONFIG_SMP 628 /* 629 * Deadline values of the currently executing and the 630 * earliest ready task on this rq. Caching these facilitates 631 * the decision wether or not a ready but not running task 632 * should migrate somewhere else. 633 */ 634 struct { 635 u64 curr; 636 u64 next; 637 } earliest_dl; 638 639 unsigned long dl_nr_migratory; 640 int overloaded; 641 642 /* 643 * Tasks on this rq that can be pushed away. They are kept in 644 * an rb-tree, ordered by tasks' deadlines, with caching 645 * of the leftmost (earliest deadline) element. 646 */ 647 struct rb_root_cached pushable_dl_tasks_root; 648 #else 649 struct dl_bw dl_bw; 650 #endif 651 /* 652 * "Active utilization" for this runqueue: increased when a 653 * task wakes up (becomes TASK_RUNNING) and decreased when a 654 * task blocks 655 */ 656 u64 running_bw; 657 658 /* 659 * Utilization of the tasks "assigned" to this runqueue (including 660 * the tasks that are in runqueue and the tasks that executed on this 661 * CPU and blocked). Increased when a task moves to this runqueue, and 662 * decreased when the task moves away (migrates, changes scheduling 663 * policy, or terminates). 664 * This is needed to compute the "inactive utilization" for the 665 * runqueue (inactive utilization = this_bw - running_bw). 666 */ 667 u64 this_bw; 668 u64 extra_bw; 669 670 /* 671 * Inverse of the fraction of CPU utilization that can be reclaimed 672 * by the GRUB algorithm. 673 */ 674 u64 bw_ratio; 675 }; 676 677 #ifdef CONFIG_FAIR_GROUP_SCHED 678 /* An entity is a task if it doesn't "own" a runqueue */ 679 #define entity_is_task(se) (!se->my_q) 680 #else 681 #define entity_is_task(se) 1 682 #endif 683 684 #ifdef CONFIG_SMP 685 /* 686 * XXX we want to get rid of these helpers and use the full load resolution. 687 */ 688 static inline long se_weight(struct sched_entity *se) 689 { 690 return scale_load_down(se->load.weight); 691 } 692 693 static inline long se_runnable(struct sched_entity *se) 694 { 695 return scale_load_down(se->runnable_weight); 696 } 697 698 static inline bool sched_asym_prefer(int a, int b) 699 { 700 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b); 701 } 702 703 /* 704 * We add the notion of a root-domain which will be used to define per-domain 705 * variables. Each exclusive cpuset essentially defines an island domain by 706 * fully partitioning the member CPUs from any other cpuset. Whenever a new 707 * exclusive cpuset is created, we also create and attach a new root-domain 708 * object. 709 * 710 */ 711 struct root_domain { 712 atomic_t refcount; 713 atomic_t rto_count; 714 struct rcu_head rcu; 715 cpumask_var_t span; 716 cpumask_var_t online; 717 718 /* Indicate more than one runnable task for any CPU */ 719 bool overload; 720 721 /* 722 * The bit corresponding to a CPU gets set here if such CPU has more 723 * than one runnable -deadline task (as it is below for RT tasks). 724 */ 725 cpumask_var_t dlo_mask; 726 atomic_t dlo_count; 727 struct dl_bw dl_bw; 728 struct cpudl cpudl; 729 730 #ifdef HAVE_RT_PUSH_IPI 731 /* 732 * For IPI pull requests, loop across the rto_mask. 733 */ 734 struct irq_work rto_push_work; 735 raw_spinlock_t rto_lock; 736 /* These are only updated and read within rto_lock */ 737 int rto_loop; 738 int rto_cpu; 739 /* These atomics are updated outside of a lock */ 740 atomic_t rto_loop_next; 741 atomic_t rto_loop_start; 742 #endif 743 /* 744 * The "RT overload" flag: it gets set if a CPU has more than 745 * one runnable RT task. 746 */ 747 cpumask_var_t rto_mask; 748 struct cpupri cpupri; 749 750 unsigned long max_cpu_capacity; 751 }; 752 753 extern struct root_domain def_root_domain; 754 extern struct mutex sched_domains_mutex; 755 756 extern void init_defrootdomain(void); 757 extern int sched_init_domains(const struct cpumask *cpu_map); 758 extern void rq_attach_root(struct rq *rq, struct root_domain *rd); 759 extern void sched_get_rd(struct root_domain *rd); 760 extern void sched_put_rd(struct root_domain *rd); 761 762 #ifdef HAVE_RT_PUSH_IPI 763 extern void rto_push_irq_work_func(struct irq_work *work); 764 #endif 765 #endif /* CONFIG_SMP */ 766 767 /* 768 * This is the main, per-CPU runqueue data structure. 769 * 770 * Locking rule: those places that want to lock multiple runqueues 771 * (such as the load balancing or the thread migration code), lock 772 * acquire operations must be ordered by ascending &runqueue. 773 */ 774 struct rq { 775 /* runqueue lock: */ 776 raw_spinlock_t lock; 777 778 /* 779 * nr_running and cpu_load should be in the same cacheline because 780 * remote CPUs use both these fields when doing load calculation. 781 */ 782 unsigned int nr_running; 783 #ifdef CONFIG_NUMA_BALANCING 784 unsigned int nr_numa_running; 785 unsigned int nr_preferred_running; 786 unsigned int numa_migrate_on; 787 #endif 788 #define CPU_LOAD_IDX_MAX 5 789 unsigned long cpu_load[CPU_LOAD_IDX_MAX]; 790 #ifdef CONFIG_NO_HZ_COMMON 791 #ifdef CONFIG_SMP 792 unsigned long last_load_update_tick; 793 unsigned long last_blocked_load_update_tick; 794 unsigned int has_blocked_load; 795 #endif /* CONFIG_SMP */ 796 unsigned int nohz_tick_stopped; 797 atomic_t nohz_flags; 798 #endif /* CONFIG_NO_HZ_COMMON */ 799 800 /* capture load from *all* tasks on this CPU: */ 801 struct load_weight load; 802 unsigned long nr_load_updates; 803 u64 nr_switches; 804 805 struct cfs_rq cfs; 806 struct rt_rq rt; 807 struct dl_rq dl; 808 809 #ifdef CONFIG_FAIR_GROUP_SCHED 810 /* list of leaf cfs_rq on this CPU: */ 811 struct list_head leaf_cfs_rq_list; 812 struct list_head *tmp_alone_branch; 813 #endif /* CONFIG_FAIR_GROUP_SCHED */ 814 815 /* 816 * This is part of a global counter where only the total sum 817 * over all CPUs matters. A task can increase this counter on 818 * one CPU and if it got migrated afterwards it may decrease 819 * it on another CPU. Always updated under the runqueue lock: 820 */ 821 unsigned long nr_uninterruptible; 822 823 struct task_struct *curr; 824 struct task_struct *idle; 825 struct task_struct *stop; 826 unsigned long next_balance; 827 struct mm_struct *prev_mm; 828 829 unsigned int clock_update_flags; 830 u64 clock; 831 u64 clock_task; 832 833 atomic_t nr_iowait; 834 835 #ifdef CONFIG_SMP 836 struct root_domain *rd; 837 struct sched_domain *sd; 838 839 unsigned long cpu_capacity; 840 unsigned long cpu_capacity_orig; 841 842 struct callback_head *balance_callback; 843 844 unsigned char idle_balance; 845 846 /* For active balancing */ 847 int active_balance; 848 int push_cpu; 849 struct cpu_stop_work active_balance_work; 850 851 /* CPU of this runqueue: */ 852 int cpu; 853 int online; 854 855 struct list_head cfs_tasks; 856 857 struct sched_avg avg_rt; 858 struct sched_avg avg_dl; 859 #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) 860 #define HAVE_SCHED_AVG_IRQ 861 struct sched_avg avg_irq; 862 #endif 863 u64 idle_stamp; 864 u64 avg_idle; 865 866 /* This is used to determine avg_idle's max value */ 867 u64 max_idle_balance_cost; 868 #endif 869 870 #ifdef CONFIG_IRQ_TIME_ACCOUNTING 871 u64 prev_irq_time; 872 #endif 873 #ifdef CONFIG_PARAVIRT 874 u64 prev_steal_time; 875 #endif 876 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING 877 u64 prev_steal_time_rq; 878 #endif 879 880 /* calc_load related fields */ 881 unsigned long calc_load_update; 882 long calc_load_active; 883 884 #ifdef CONFIG_SCHED_HRTICK 885 #ifdef CONFIG_SMP 886 int hrtick_csd_pending; 887 call_single_data_t hrtick_csd; 888 #endif 889 struct hrtimer hrtick_timer; 890 #endif 891 892 #ifdef CONFIG_SCHEDSTATS 893 /* latency stats */ 894 struct sched_info rq_sched_info; 895 unsigned long long rq_cpu_time; 896 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ 897 898 /* sys_sched_yield() stats */ 899 unsigned int yld_count; 900 901 /* schedule() stats */ 902 unsigned int sched_count; 903 unsigned int sched_goidle; 904 905 /* try_to_wake_up() stats */ 906 unsigned int ttwu_count; 907 unsigned int ttwu_local; 908 #endif 909 910 #ifdef CONFIG_SMP 911 struct llist_head wake_list; 912 #endif 913 914 #ifdef CONFIG_CPU_IDLE 915 /* Must be inspected within a rcu lock section */ 916 struct cpuidle_state *idle_state; 917 #endif 918 }; 919 920 static inline int cpu_of(struct rq *rq) 921 { 922 #ifdef CONFIG_SMP 923 return rq->cpu; 924 #else 925 return 0; 926 #endif 927 } 928 929 930 #ifdef CONFIG_SCHED_SMT 931 932 extern struct static_key_false sched_smt_present; 933 934 extern void __update_idle_core(struct rq *rq); 935 936 static inline void update_idle_core(struct rq *rq) 937 { 938 if (static_branch_unlikely(&sched_smt_present)) 939 __update_idle_core(rq); 940 } 941 942 #else 943 static inline void update_idle_core(struct rq *rq) { } 944 #endif 945 946 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); 947 948 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) 949 #define this_rq() this_cpu_ptr(&runqueues) 950 #define task_rq(p) cpu_rq(task_cpu(p)) 951 #define cpu_curr(cpu) (cpu_rq(cpu)->curr) 952 #define raw_rq() raw_cpu_ptr(&runqueues) 953 954 static inline u64 __rq_clock_broken(struct rq *rq) 955 { 956 return READ_ONCE(rq->clock); 957 } 958 959 /* 960 * rq::clock_update_flags bits 961 * 962 * %RQCF_REQ_SKIP - will request skipping of clock update on the next 963 * call to __schedule(). This is an optimisation to avoid 964 * neighbouring rq clock updates. 965 * 966 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is 967 * in effect and calls to update_rq_clock() are being ignored. 968 * 969 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been 970 * made to update_rq_clock() since the last time rq::lock was pinned. 971 * 972 * If inside of __schedule(), clock_update_flags will have been 973 * shifted left (a left shift is a cheap operation for the fast path 974 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use, 975 * 976 * if (rq-clock_update_flags >= RQCF_UPDATED) 977 * 978 * to check if %RQCF_UPADTED is set. It'll never be shifted more than 979 * one position though, because the next rq_unpin_lock() will shift it 980 * back. 981 */ 982 #define RQCF_REQ_SKIP 0x01 983 #define RQCF_ACT_SKIP 0x02 984 #define RQCF_UPDATED 0x04 985 986 static inline void assert_clock_updated(struct rq *rq) 987 { 988 /* 989 * The only reason for not seeing a clock update since the 990 * last rq_pin_lock() is if we're currently skipping updates. 991 */ 992 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP); 993 } 994 995 static inline u64 rq_clock(struct rq *rq) 996 { 997 lockdep_assert_held(&rq->lock); 998 assert_clock_updated(rq); 999 1000 return rq->clock; 1001 } 1002 1003 static inline u64 rq_clock_task(struct rq *rq) 1004 { 1005 lockdep_assert_held(&rq->lock); 1006 assert_clock_updated(rq); 1007 1008 return rq->clock_task; 1009 } 1010 1011 static inline void rq_clock_skip_update(struct rq *rq) 1012 { 1013 lockdep_assert_held(&rq->lock); 1014 rq->clock_update_flags |= RQCF_REQ_SKIP; 1015 } 1016 1017 /* 1018 * See rt task throttling, which is the only time a skip 1019 * request is cancelled. 1020 */ 1021 static inline void rq_clock_cancel_skipupdate(struct rq *rq) 1022 { 1023 lockdep_assert_held(&rq->lock); 1024 rq->clock_update_flags &= ~RQCF_REQ_SKIP; 1025 } 1026 1027 struct rq_flags { 1028 unsigned long flags; 1029 struct pin_cookie cookie; 1030 #ifdef CONFIG_SCHED_DEBUG 1031 /* 1032 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the 1033 * current pin context is stashed here in case it needs to be 1034 * restored in rq_repin_lock(). 1035 */ 1036 unsigned int clock_update_flags; 1037 #endif 1038 }; 1039 1040 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf) 1041 { 1042 rf->cookie = lockdep_pin_lock(&rq->lock); 1043 1044 #ifdef CONFIG_SCHED_DEBUG 1045 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP); 1046 rf->clock_update_flags = 0; 1047 #endif 1048 } 1049 1050 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf) 1051 { 1052 #ifdef CONFIG_SCHED_DEBUG 1053 if (rq->clock_update_flags > RQCF_ACT_SKIP) 1054 rf->clock_update_flags = RQCF_UPDATED; 1055 #endif 1056 1057 lockdep_unpin_lock(&rq->lock, rf->cookie); 1058 } 1059 1060 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf) 1061 { 1062 lockdep_repin_lock(&rq->lock, rf->cookie); 1063 1064 #ifdef CONFIG_SCHED_DEBUG 1065 /* 1066 * Restore the value we stashed in @rf for this pin context. 1067 */ 1068 rq->clock_update_flags |= rf->clock_update_flags; 1069 #endif 1070 } 1071 1072 #ifdef CONFIG_NUMA 1073 enum numa_topology_type { 1074 NUMA_DIRECT, 1075 NUMA_GLUELESS_MESH, 1076 NUMA_BACKPLANE, 1077 }; 1078 extern enum numa_topology_type sched_numa_topology_type; 1079 extern int sched_max_numa_distance; 1080 extern bool find_numa_distance(int distance); 1081 #endif 1082 1083 #ifdef CONFIG_NUMA 1084 extern void sched_init_numa(void); 1085 extern void sched_domains_numa_masks_set(unsigned int cpu); 1086 extern void sched_domains_numa_masks_clear(unsigned int cpu); 1087 #else 1088 static inline void sched_init_numa(void) { } 1089 static inline void sched_domains_numa_masks_set(unsigned int cpu) { } 1090 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { } 1091 #endif 1092 1093 #ifdef CONFIG_NUMA_BALANCING 1094 /* The regions in numa_faults array from task_struct */ 1095 enum numa_faults_stats { 1096 NUMA_MEM = 0, 1097 NUMA_CPU, 1098 NUMA_MEMBUF, 1099 NUMA_CPUBUF 1100 }; 1101 extern void sched_setnuma(struct task_struct *p, int node); 1102 extern int migrate_task_to(struct task_struct *p, int cpu); 1103 extern int migrate_swap(struct task_struct *p, struct task_struct *t, 1104 int cpu, int scpu); 1105 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p); 1106 #else 1107 static inline void 1108 init_numa_balancing(unsigned long clone_flags, struct task_struct *p) 1109 { 1110 } 1111 #endif /* CONFIG_NUMA_BALANCING */ 1112 1113 #ifdef CONFIG_SMP 1114 1115 static inline void 1116 queue_balance_callback(struct rq *rq, 1117 struct callback_head *head, 1118 void (*func)(struct rq *rq)) 1119 { 1120 lockdep_assert_held(&rq->lock); 1121 1122 if (unlikely(head->next)) 1123 return; 1124 1125 head->func = (void (*)(struct callback_head *))func; 1126 head->next = rq->balance_callback; 1127 rq->balance_callback = head; 1128 } 1129 1130 extern void sched_ttwu_pending(void); 1131 1132 #define rcu_dereference_check_sched_domain(p) \ 1133 rcu_dereference_check((p), \ 1134 lockdep_is_held(&sched_domains_mutex)) 1135 1136 /* 1137 * The domain tree (rq->sd) is protected by RCU's quiescent state transition. 1138 * See detach_destroy_domains: synchronize_sched for details. 1139 * 1140 * The domain tree of any CPU may only be accessed from within 1141 * preempt-disabled sections. 1142 */ 1143 #define for_each_domain(cpu, __sd) \ 1144 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \ 1145 __sd; __sd = __sd->parent) 1146 1147 #define for_each_lower_domain(sd) for (; sd; sd = sd->child) 1148 1149 /** 1150 * highest_flag_domain - Return highest sched_domain containing flag. 1151 * @cpu: The CPU whose highest level of sched domain is to 1152 * be returned. 1153 * @flag: The flag to check for the highest sched_domain 1154 * for the given CPU. 1155 * 1156 * Returns the highest sched_domain of a CPU which contains the given flag. 1157 */ 1158 static inline struct sched_domain *highest_flag_domain(int cpu, int flag) 1159 { 1160 struct sched_domain *sd, *hsd = NULL; 1161 1162 for_each_domain(cpu, sd) { 1163 if (!(sd->flags & flag)) 1164 break; 1165 hsd = sd; 1166 } 1167 1168 return hsd; 1169 } 1170 1171 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag) 1172 { 1173 struct sched_domain *sd; 1174 1175 for_each_domain(cpu, sd) { 1176 if (sd->flags & flag) 1177 break; 1178 } 1179 1180 return sd; 1181 } 1182 1183 DECLARE_PER_CPU(struct sched_domain *, sd_llc); 1184 DECLARE_PER_CPU(int, sd_llc_size); 1185 DECLARE_PER_CPU(int, sd_llc_id); 1186 DECLARE_PER_CPU(struct sched_domain_shared *, sd_llc_shared); 1187 DECLARE_PER_CPU(struct sched_domain *, sd_numa); 1188 DECLARE_PER_CPU(struct sched_domain *, sd_asym); 1189 1190 struct sched_group_capacity { 1191 atomic_t ref; 1192 /* 1193 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity 1194 * for a single CPU. 1195 */ 1196 unsigned long capacity; 1197 unsigned long min_capacity; /* Min per-CPU capacity in group */ 1198 unsigned long next_update; 1199 int imbalance; /* XXX unrelated to capacity but shared group state */ 1200 1201 #ifdef CONFIG_SCHED_DEBUG 1202 int id; 1203 #endif 1204 1205 unsigned long cpumask[0]; /* Balance mask */ 1206 }; 1207 1208 struct sched_group { 1209 struct sched_group *next; /* Must be a circular list */ 1210 atomic_t ref; 1211 1212 unsigned int group_weight; 1213 struct sched_group_capacity *sgc; 1214 int asym_prefer_cpu; /* CPU of highest priority in group */ 1215 1216 /* 1217 * The CPUs this group covers. 1218 * 1219 * NOTE: this field is variable length. (Allocated dynamically 1220 * by attaching extra space to the end of the structure, 1221 * depending on how many CPUs the kernel has booted up with) 1222 */ 1223 unsigned long cpumask[0]; 1224 }; 1225 1226 static inline struct cpumask *sched_group_span(struct sched_group *sg) 1227 { 1228 return to_cpumask(sg->cpumask); 1229 } 1230 1231 /* 1232 * See build_balance_mask(). 1233 */ 1234 static inline struct cpumask *group_balance_mask(struct sched_group *sg) 1235 { 1236 return to_cpumask(sg->sgc->cpumask); 1237 } 1238 1239 /** 1240 * group_first_cpu - Returns the first CPU in the cpumask of a sched_group. 1241 * @group: The group whose first CPU is to be returned. 1242 */ 1243 static inline unsigned int group_first_cpu(struct sched_group *group) 1244 { 1245 return cpumask_first(sched_group_span(group)); 1246 } 1247 1248 extern int group_balance_cpu(struct sched_group *sg); 1249 1250 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) 1251 void register_sched_domain_sysctl(void); 1252 void dirty_sched_domain_sysctl(int cpu); 1253 void unregister_sched_domain_sysctl(void); 1254 #else 1255 static inline void register_sched_domain_sysctl(void) 1256 { 1257 } 1258 static inline void dirty_sched_domain_sysctl(int cpu) 1259 { 1260 } 1261 static inline void unregister_sched_domain_sysctl(void) 1262 { 1263 } 1264 #endif 1265 1266 #else 1267 1268 static inline void sched_ttwu_pending(void) { } 1269 1270 #endif /* CONFIG_SMP */ 1271 1272 #include "stats.h" 1273 #include "autogroup.h" 1274 1275 #ifdef CONFIG_CGROUP_SCHED 1276 1277 /* 1278 * Return the group to which this tasks belongs. 1279 * 1280 * We cannot use task_css() and friends because the cgroup subsystem 1281 * changes that value before the cgroup_subsys::attach() method is called, 1282 * therefore we cannot pin it and might observe the wrong value. 1283 * 1284 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup 1285 * core changes this before calling sched_move_task(). 1286 * 1287 * Instead we use a 'copy' which is updated from sched_move_task() while 1288 * holding both task_struct::pi_lock and rq::lock. 1289 */ 1290 static inline struct task_group *task_group(struct task_struct *p) 1291 { 1292 return p->sched_task_group; 1293 } 1294 1295 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ 1296 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) 1297 { 1298 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED) 1299 struct task_group *tg = task_group(p); 1300 #endif 1301 1302 #ifdef CONFIG_FAIR_GROUP_SCHED 1303 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]); 1304 p->se.cfs_rq = tg->cfs_rq[cpu]; 1305 p->se.parent = tg->se[cpu]; 1306 #endif 1307 1308 #ifdef CONFIG_RT_GROUP_SCHED 1309 p->rt.rt_rq = tg->rt_rq[cpu]; 1310 p->rt.parent = tg->rt_se[cpu]; 1311 #endif 1312 } 1313 1314 #else /* CONFIG_CGROUP_SCHED */ 1315 1316 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } 1317 static inline struct task_group *task_group(struct task_struct *p) 1318 { 1319 return NULL; 1320 } 1321 1322 #endif /* CONFIG_CGROUP_SCHED */ 1323 1324 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) 1325 { 1326 set_task_rq(p, cpu); 1327 #ifdef CONFIG_SMP 1328 /* 1329 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be 1330 * successfuly executed on another CPU. We must ensure that updates of 1331 * per-task data have been completed by this moment. 1332 */ 1333 smp_wmb(); 1334 #ifdef CONFIG_THREAD_INFO_IN_TASK 1335 p->cpu = cpu; 1336 #else 1337 task_thread_info(p)->cpu = cpu; 1338 #endif 1339 p->wake_cpu = cpu; 1340 #endif 1341 } 1342 1343 /* 1344 * Tunables that become constants when CONFIG_SCHED_DEBUG is off: 1345 */ 1346 #ifdef CONFIG_SCHED_DEBUG 1347 # include <linux/static_key.h> 1348 # define const_debug __read_mostly 1349 #else 1350 # define const_debug const 1351 #endif 1352 1353 #define SCHED_FEAT(name, enabled) \ 1354 __SCHED_FEAT_##name , 1355 1356 enum { 1357 #include "features.h" 1358 __SCHED_FEAT_NR, 1359 }; 1360 1361 #undef SCHED_FEAT 1362 1363 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL) 1364 1365 /* 1366 * To support run-time toggling of sched features, all the translation units 1367 * (but core.c) reference the sysctl_sched_features defined in core.c. 1368 */ 1369 extern const_debug unsigned int sysctl_sched_features; 1370 1371 #define SCHED_FEAT(name, enabled) \ 1372 static __always_inline bool static_branch_##name(struct static_key *key) \ 1373 { \ 1374 return static_key_##enabled(key); \ 1375 } 1376 1377 #include "features.h" 1378 #undef SCHED_FEAT 1379 1380 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR]; 1381 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x])) 1382 1383 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */ 1384 1385 /* 1386 * Each translation unit has its own copy of sysctl_sched_features to allow 1387 * constants propagation at compile time and compiler optimization based on 1388 * features default. 1389 */ 1390 #define SCHED_FEAT(name, enabled) \ 1391 (1UL << __SCHED_FEAT_##name) * enabled | 1392 static const_debug __maybe_unused unsigned int sysctl_sched_features = 1393 #include "features.h" 1394 0; 1395 #undef SCHED_FEAT 1396 1397 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) 1398 1399 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */ 1400 1401 extern struct static_key_false sched_numa_balancing; 1402 extern struct static_key_false sched_schedstats; 1403 1404 static inline u64 global_rt_period(void) 1405 { 1406 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; 1407 } 1408 1409 static inline u64 global_rt_runtime(void) 1410 { 1411 if (sysctl_sched_rt_runtime < 0) 1412 return RUNTIME_INF; 1413 1414 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; 1415 } 1416 1417 static inline int task_current(struct rq *rq, struct task_struct *p) 1418 { 1419 return rq->curr == p; 1420 } 1421 1422 static inline int task_running(struct rq *rq, struct task_struct *p) 1423 { 1424 #ifdef CONFIG_SMP 1425 return p->on_cpu; 1426 #else 1427 return task_current(rq, p); 1428 #endif 1429 } 1430 1431 static inline int task_on_rq_queued(struct task_struct *p) 1432 { 1433 return p->on_rq == TASK_ON_RQ_QUEUED; 1434 } 1435 1436 static inline int task_on_rq_migrating(struct task_struct *p) 1437 { 1438 return p->on_rq == TASK_ON_RQ_MIGRATING; 1439 } 1440 1441 /* 1442 * wake flags 1443 */ 1444 #define WF_SYNC 0x01 /* Waker goes to sleep after wakeup */ 1445 #define WF_FORK 0x02 /* Child wakeup after fork */ 1446 #define WF_MIGRATED 0x4 /* Internal use, task got migrated */ 1447 1448 /* 1449 * To aid in avoiding the subversion of "niceness" due to uneven distribution 1450 * of tasks with abnormal "nice" values across CPUs the contribution that 1451 * each task makes to its run queue's load is weighted according to its 1452 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a 1453 * scaled version of the new time slice allocation that they receive on time 1454 * slice expiry etc. 1455 */ 1456 1457 #define WEIGHT_IDLEPRIO 3 1458 #define WMULT_IDLEPRIO 1431655765 1459 1460 extern const int sched_prio_to_weight[40]; 1461 extern const u32 sched_prio_to_wmult[40]; 1462 1463 /* 1464 * {de,en}queue flags: 1465 * 1466 * DEQUEUE_SLEEP - task is no longer runnable 1467 * ENQUEUE_WAKEUP - task just became runnable 1468 * 1469 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks 1470 * are in a known state which allows modification. Such pairs 1471 * should preserve as much state as possible. 1472 * 1473 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location 1474 * in the runqueue. 1475 * 1476 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified) 1477 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline) 1478 * ENQUEUE_MIGRATED - the task was migrated during wakeup 1479 * 1480 */ 1481 1482 #define DEQUEUE_SLEEP 0x01 1483 #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */ 1484 #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */ 1485 #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */ 1486 1487 #define ENQUEUE_WAKEUP 0x01 1488 #define ENQUEUE_RESTORE 0x02 1489 #define ENQUEUE_MOVE 0x04 1490 #define ENQUEUE_NOCLOCK 0x08 1491 1492 #define ENQUEUE_HEAD 0x10 1493 #define ENQUEUE_REPLENISH 0x20 1494 #ifdef CONFIG_SMP 1495 #define ENQUEUE_MIGRATED 0x40 1496 #else 1497 #define ENQUEUE_MIGRATED 0x00 1498 #endif 1499 1500 #define RETRY_TASK ((void *)-1UL) 1501 1502 struct sched_class { 1503 const struct sched_class *next; 1504 1505 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags); 1506 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags); 1507 void (*yield_task) (struct rq *rq); 1508 bool (*yield_to_task)(struct rq *rq, struct task_struct *p, bool preempt); 1509 1510 void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags); 1511 1512 /* 1513 * It is the responsibility of the pick_next_task() method that will 1514 * return the next task to call put_prev_task() on the @prev task or 1515 * something equivalent. 1516 * 1517 * May return RETRY_TASK when it finds a higher prio class has runnable 1518 * tasks. 1519 */ 1520 struct task_struct * (*pick_next_task)(struct rq *rq, 1521 struct task_struct *prev, 1522 struct rq_flags *rf); 1523 void (*put_prev_task)(struct rq *rq, struct task_struct *p); 1524 1525 #ifdef CONFIG_SMP 1526 int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags); 1527 void (*migrate_task_rq)(struct task_struct *p, int new_cpu); 1528 1529 void (*task_woken)(struct rq *this_rq, struct task_struct *task); 1530 1531 void (*set_cpus_allowed)(struct task_struct *p, 1532 const struct cpumask *newmask); 1533 1534 void (*rq_online)(struct rq *rq); 1535 void (*rq_offline)(struct rq *rq); 1536 #endif 1537 1538 void (*set_curr_task)(struct rq *rq); 1539 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued); 1540 void (*task_fork)(struct task_struct *p); 1541 void (*task_dead)(struct task_struct *p); 1542 1543 /* 1544 * The switched_from() call is allowed to drop rq->lock, therefore we 1545 * cannot assume the switched_from/switched_to pair is serliazed by 1546 * rq->lock. They are however serialized by p->pi_lock. 1547 */ 1548 void (*switched_from)(struct rq *this_rq, struct task_struct *task); 1549 void (*switched_to) (struct rq *this_rq, struct task_struct *task); 1550 void (*prio_changed) (struct rq *this_rq, struct task_struct *task, 1551 int oldprio); 1552 1553 unsigned int (*get_rr_interval)(struct rq *rq, 1554 struct task_struct *task); 1555 1556 void (*update_curr)(struct rq *rq); 1557 1558 #define TASK_SET_GROUP 0 1559 #define TASK_MOVE_GROUP 1 1560 1561 #ifdef CONFIG_FAIR_GROUP_SCHED 1562 void (*task_change_group)(struct task_struct *p, int type); 1563 #endif 1564 }; 1565 1566 static inline void put_prev_task(struct rq *rq, struct task_struct *prev) 1567 { 1568 prev->sched_class->put_prev_task(rq, prev); 1569 } 1570 1571 static inline void set_curr_task(struct rq *rq, struct task_struct *curr) 1572 { 1573 curr->sched_class->set_curr_task(rq); 1574 } 1575 1576 #ifdef CONFIG_SMP 1577 #define sched_class_highest (&stop_sched_class) 1578 #else 1579 #define sched_class_highest (&dl_sched_class) 1580 #endif 1581 #define for_each_class(class) \ 1582 for (class = sched_class_highest; class; class = class->next) 1583 1584 extern const struct sched_class stop_sched_class; 1585 extern const struct sched_class dl_sched_class; 1586 extern const struct sched_class rt_sched_class; 1587 extern const struct sched_class fair_sched_class; 1588 extern const struct sched_class idle_sched_class; 1589 1590 1591 #ifdef CONFIG_SMP 1592 1593 extern void update_group_capacity(struct sched_domain *sd, int cpu); 1594 1595 extern void trigger_load_balance(struct rq *rq); 1596 1597 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask); 1598 1599 #endif 1600 1601 #ifdef CONFIG_CPU_IDLE 1602 static inline void idle_set_state(struct rq *rq, 1603 struct cpuidle_state *idle_state) 1604 { 1605 rq->idle_state = idle_state; 1606 } 1607 1608 static inline struct cpuidle_state *idle_get_state(struct rq *rq) 1609 { 1610 SCHED_WARN_ON(!rcu_read_lock_held()); 1611 1612 return rq->idle_state; 1613 } 1614 #else 1615 static inline void idle_set_state(struct rq *rq, 1616 struct cpuidle_state *idle_state) 1617 { 1618 } 1619 1620 static inline struct cpuidle_state *idle_get_state(struct rq *rq) 1621 { 1622 return NULL; 1623 } 1624 #endif 1625 1626 extern void schedule_idle(void); 1627 1628 extern void sysrq_sched_debug_show(void); 1629 extern void sched_init_granularity(void); 1630 extern void update_max_interval(void); 1631 1632 extern void init_sched_dl_class(void); 1633 extern void init_sched_rt_class(void); 1634 extern void init_sched_fair_class(void); 1635 1636 extern void reweight_task(struct task_struct *p, int prio); 1637 1638 extern void resched_curr(struct rq *rq); 1639 extern void resched_cpu(int cpu); 1640 1641 extern struct rt_bandwidth def_rt_bandwidth; 1642 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime); 1643 1644 extern struct dl_bandwidth def_dl_bandwidth; 1645 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime); 1646 extern void init_dl_task_timer(struct sched_dl_entity *dl_se); 1647 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se); 1648 extern void init_dl_rq_bw_ratio(struct dl_rq *dl_rq); 1649 1650 #define BW_SHIFT 20 1651 #define BW_UNIT (1 << BW_SHIFT) 1652 #define RATIO_SHIFT 8 1653 unsigned long to_ratio(u64 period, u64 runtime); 1654 1655 extern void init_entity_runnable_average(struct sched_entity *se); 1656 extern void post_init_entity_util_avg(struct sched_entity *se); 1657 1658 #ifdef CONFIG_NO_HZ_FULL 1659 extern bool sched_can_stop_tick(struct rq *rq); 1660 extern int __init sched_tick_offload_init(void); 1661 1662 /* 1663 * Tick may be needed by tasks in the runqueue depending on their policy and 1664 * requirements. If tick is needed, lets send the target an IPI to kick it out of 1665 * nohz mode if necessary. 1666 */ 1667 static inline void sched_update_tick_dependency(struct rq *rq) 1668 { 1669 int cpu; 1670 1671 if (!tick_nohz_full_enabled()) 1672 return; 1673 1674 cpu = cpu_of(rq); 1675 1676 if (!tick_nohz_full_cpu(cpu)) 1677 return; 1678 1679 if (sched_can_stop_tick(rq)) 1680 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED); 1681 else 1682 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED); 1683 } 1684 #else 1685 static inline int sched_tick_offload_init(void) { return 0; } 1686 static inline void sched_update_tick_dependency(struct rq *rq) { } 1687 #endif 1688 1689 static inline void add_nr_running(struct rq *rq, unsigned count) 1690 { 1691 unsigned prev_nr = rq->nr_running; 1692 1693 rq->nr_running = prev_nr + count; 1694 1695 if (prev_nr < 2 && rq->nr_running >= 2) { 1696 #ifdef CONFIG_SMP 1697 if (!rq->rd->overload) 1698 rq->rd->overload = true; 1699 #endif 1700 } 1701 1702 sched_update_tick_dependency(rq); 1703 } 1704 1705 static inline void sub_nr_running(struct rq *rq, unsigned count) 1706 { 1707 rq->nr_running -= count; 1708 /* Check if we still need preemption */ 1709 sched_update_tick_dependency(rq); 1710 } 1711 1712 extern void update_rq_clock(struct rq *rq); 1713 1714 extern void activate_task(struct rq *rq, struct task_struct *p, int flags); 1715 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags); 1716 1717 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags); 1718 1719 extern const_debug unsigned int sysctl_sched_nr_migrate; 1720 extern const_debug unsigned int sysctl_sched_migration_cost; 1721 1722 #ifdef CONFIG_SCHED_HRTICK 1723 1724 /* 1725 * Use hrtick when: 1726 * - enabled by features 1727 * - hrtimer is actually high res 1728 */ 1729 static inline int hrtick_enabled(struct rq *rq) 1730 { 1731 if (!sched_feat(HRTICK)) 1732 return 0; 1733 if (!cpu_active(cpu_of(rq))) 1734 return 0; 1735 return hrtimer_is_hres_active(&rq->hrtick_timer); 1736 } 1737 1738 void hrtick_start(struct rq *rq, u64 delay); 1739 1740 #else 1741 1742 static inline int hrtick_enabled(struct rq *rq) 1743 { 1744 return 0; 1745 } 1746 1747 #endif /* CONFIG_SCHED_HRTICK */ 1748 1749 #ifndef arch_scale_freq_capacity 1750 static __always_inline 1751 unsigned long arch_scale_freq_capacity(int cpu) 1752 { 1753 return SCHED_CAPACITY_SCALE; 1754 } 1755 #endif 1756 1757 #ifdef CONFIG_SMP 1758 #ifndef arch_scale_cpu_capacity 1759 static __always_inline 1760 unsigned long arch_scale_cpu_capacity(struct sched_domain *sd, int cpu) 1761 { 1762 if (sd && (sd->flags & SD_SHARE_CPUCAPACITY) && (sd->span_weight > 1)) 1763 return sd->smt_gain / sd->span_weight; 1764 1765 return SCHED_CAPACITY_SCALE; 1766 } 1767 #endif 1768 #else 1769 #ifndef arch_scale_cpu_capacity 1770 static __always_inline 1771 unsigned long arch_scale_cpu_capacity(void __always_unused *sd, int cpu) 1772 { 1773 return SCHED_CAPACITY_SCALE; 1774 } 1775 #endif 1776 #endif 1777 1778 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf) 1779 __acquires(rq->lock); 1780 1781 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf) 1782 __acquires(p->pi_lock) 1783 __acquires(rq->lock); 1784 1785 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf) 1786 __releases(rq->lock) 1787 { 1788 rq_unpin_lock(rq, rf); 1789 raw_spin_unlock(&rq->lock); 1790 } 1791 1792 static inline void 1793 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf) 1794 __releases(rq->lock) 1795 __releases(p->pi_lock) 1796 { 1797 rq_unpin_lock(rq, rf); 1798 raw_spin_unlock(&rq->lock); 1799 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags); 1800 } 1801 1802 static inline void 1803 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf) 1804 __acquires(rq->lock) 1805 { 1806 raw_spin_lock_irqsave(&rq->lock, rf->flags); 1807 rq_pin_lock(rq, rf); 1808 } 1809 1810 static inline void 1811 rq_lock_irq(struct rq *rq, struct rq_flags *rf) 1812 __acquires(rq->lock) 1813 { 1814 raw_spin_lock_irq(&rq->lock); 1815 rq_pin_lock(rq, rf); 1816 } 1817 1818 static inline void 1819 rq_lock(struct rq *rq, struct rq_flags *rf) 1820 __acquires(rq->lock) 1821 { 1822 raw_spin_lock(&rq->lock); 1823 rq_pin_lock(rq, rf); 1824 } 1825 1826 static inline void 1827 rq_relock(struct rq *rq, struct rq_flags *rf) 1828 __acquires(rq->lock) 1829 { 1830 raw_spin_lock(&rq->lock); 1831 rq_repin_lock(rq, rf); 1832 } 1833 1834 static inline void 1835 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf) 1836 __releases(rq->lock) 1837 { 1838 rq_unpin_lock(rq, rf); 1839 raw_spin_unlock_irqrestore(&rq->lock, rf->flags); 1840 } 1841 1842 static inline void 1843 rq_unlock_irq(struct rq *rq, struct rq_flags *rf) 1844 __releases(rq->lock) 1845 { 1846 rq_unpin_lock(rq, rf); 1847 raw_spin_unlock_irq(&rq->lock); 1848 } 1849 1850 static inline void 1851 rq_unlock(struct rq *rq, struct rq_flags *rf) 1852 __releases(rq->lock) 1853 { 1854 rq_unpin_lock(rq, rf); 1855 raw_spin_unlock(&rq->lock); 1856 } 1857 1858 #ifdef CONFIG_SMP 1859 #ifdef CONFIG_PREEMPT 1860 1861 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2); 1862 1863 /* 1864 * fair double_lock_balance: Safely acquires both rq->locks in a fair 1865 * way at the expense of forcing extra atomic operations in all 1866 * invocations. This assures that the double_lock is acquired using the 1867 * same underlying policy as the spinlock_t on this architecture, which 1868 * reduces latency compared to the unfair variant below. However, it 1869 * also adds more overhead and therefore may reduce throughput. 1870 */ 1871 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) 1872 __releases(this_rq->lock) 1873 __acquires(busiest->lock) 1874 __acquires(this_rq->lock) 1875 { 1876 raw_spin_unlock(&this_rq->lock); 1877 double_rq_lock(this_rq, busiest); 1878 1879 return 1; 1880 } 1881 1882 #else 1883 /* 1884 * Unfair double_lock_balance: Optimizes throughput at the expense of 1885 * latency by eliminating extra atomic operations when the locks are 1886 * already in proper order on entry. This favors lower CPU-ids and will 1887 * grant the double lock to lower CPUs over higher ids under contention, 1888 * regardless of entry order into the function. 1889 */ 1890 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) 1891 __releases(this_rq->lock) 1892 __acquires(busiest->lock) 1893 __acquires(this_rq->lock) 1894 { 1895 int ret = 0; 1896 1897 if (unlikely(!raw_spin_trylock(&busiest->lock))) { 1898 if (busiest < this_rq) { 1899 raw_spin_unlock(&this_rq->lock); 1900 raw_spin_lock(&busiest->lock); 1901 raw_spin_lock_nested(&this_rq->lock, 1902 SINGLE_DEPTH_NESTING); 1903 ret = 1; 1904 } else 1905 raw_spin_lock_nested(&busiest->lock, 1906 SINGLE_DEPTH_NESTING); 1907 } 1908 return ret; 1909 } 1910 1911 #endif /* CONFIG_PREEMPT */ 1912 1913 /* 1914 * double_lock_balance - lock the busiest runqueue, this_rq is locked already. 1915 */ 1916 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest) 1917 { 1918 if (unlikely(!irqs_disabled())) { 1919 /* printk() doesn't work well under rq->lock */ 1920 raw_spin_unlock(&this_rq->lock); 1921 BUG_ON(1); 1922 } 1923 1924 return _double_lock_balance(this_rq, busiest); 1925 } 1926 1927 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) 1928 __releases(busiest->lock) 1929 { 1930 raw_spin_unlock(&busiest->lock); 1931 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); 1932 } 1933 1934 static inline void double_lock(spinlock_t *l1, spinlock_t *l2) 1935 { 1936 if (l1 > l2) 1937 swap(l1, l2); 1938 1939 spin_lock(l1); 1940 spin_lock_nested(l2, SINGLE_DEPTH_NESTING); 1941 } 1942 1943 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2) 1944 { 1945 if (l1 > l2) 1946 swap(l1, l2); 1947 1948 spin_lock_irq(l1); 1949 spin_lock_nested(l2, SINGLE_DEPTH_NESTING); 1950 } 1951 1952 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2) 1953 { 1954 if (l1 > l2) 1955 swap(l1, l2); 1956 1957 raw_spin_lock(l1); 1958 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING); 1959 } 1960 1961 /* 1962 * double_rq_lock - safely lock two runqueues 1963 * 1964 * Note this does not disable interrupts like task_rq_lock, 1965 * you need to do so manually before calling. 1966 */ 1967 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) 1968 __acquires(rq1->lock) 1969 __acquires(rq2->lock) 1970 { 1971 BUG_ON(!irqs_disabled()); 1972 if (rq1 == rq2) { 1973 raw_spin_lock(&rq1->lock); 1974 __acquire(rq2->lock); /* Fake it out ;) */ 1975 } else { 1976 if (rq1 < rq2) { 1977 raw_spin_lock(&rq1->lock); 1978 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); 1979 } else { 1980 raw_spin_lock(&rq2->lock); 1981 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); 1982 } 1983 } 1984 } 1985 1986 /* 1987 * double_rq_unlock - safely unlock two runqueues 1988 * 1989 * Note this does not restore interrupts like task_rq_unlock, 1990 * you need to do so manually after calling. 1991 */ 1992 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) 1993 __releases(rq1->lock) 1994 __releases(rq2->lock) 1995 { 1996 raw_spin_unlock(&rq1->lock); 1997 if (rq1 != rq2) 1998 raw_spin_unlock(&rq2->lock); 1999 else 2000 __release(rq2->lock); 2001 } 2002 2003 extern void set_rq_online (struct rq *rq); 2004 extern void set_rq_offline(struct rq *rq); 2005 extern bool sched_smp_initialized; 2006 2007 #else /* CONFIG_SMP */ 2008 2009 /* 2010 * double_rq_lock - safely lock two runqueues 2011 * 2012 * Note this does not disable interrupts like task_rq_lock, 2013 * you need to do so manually before calling. 2014 */ 2015 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) 2016 __acquires(rq1->lock) 2017 __acquires(rq2->lock) 2018 { 2019 BUG_ON(!irqs_disabled()); 2020 BUG_ON(rq1 != rq2); 2021 raw_spin_lock(&rq1->lock); 2022 __acquire(rq2->lock); /* Fake it out ;) */ 2023 } 2024 2025 /* 2026 * double_rq_unlock - safely unlock two runqueues 2027 * 2028 * Note this does not restore interrupts like task_rq_unlock, 2029 * you need to do so manually after calling. 2030 */ 2031 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) 2032 __releases(rq1->lock) 2033 __releases(rq2->lock) 2034 { 2035 BUG_ON(rq1 != rq2); 2036 raw_spin_unlock(&rq1->lock); 2037 __release(rq2->lock); 2038 } 2039 2040 #endif 2041 2042 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq); 2043 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq); 2044 2045 #ifdef CONFIG_SCHED_DEBUG 2046 extern bool sched_debug_enabled; 2047 2048 extern void print_cfs_stats(struct seq_file *m, int cpu); 2049 extern void print_rt_stats(struct seq_file *m, int cpu); 2050 extern void print_dl_stats(struct seq_file *m, int cpu); 2051 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq); 2052 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq); 2053 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq); 2054 #ifdef CONFIG_NUMA_BALANCING 2055 extern void 2056 show_numa_stats(struct task_struct *p, struct seq_file *m); 2057 extern void 2058 print_numa_stats(struct seq_file *m, int node, unsigned long tsf, 2059 unsigned long tpf, unsigned long gsf, unsigned long gpf); 2060 #endif /* CONFIG_NUMA_BALANCING */ 2061 #endif /* CONFIG_SCHED_DEBUG */ 2062 2063 extern void init_cfs_rq(struct cfs_rq *cfs_rq); 2064 extern void init_rt_rq(struct rt_rq *rt_rq); 2065 extern void init_dl_rq(struct dl_rq *dl_rq); 2066 2067 extern void cfs_bandwidth_usage_inc(void); 2068 extern void cfs_bandwidth_usage_dec(void); 2069 2070 #ifdef CONFIG_NO_HZ_COMMON 2071 #define NOHZ_BALANCE_KICK_BIT 0 2072 #define NOHZ_STATS_KICK_BIT 1 2073 2074 #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT) 2075 #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT) 2076 2077 #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK) 2078 2079 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags) 2080 2081 extern void nohz_balance_exit_idle(struct rq *rq); 2082 #else 2083 static inline void nohz_balance_exit_idle(struct rq *rq) { } 2084 #endif 2085 2086 2087 #ifdef CONFIG_SMP 2088 static inline 2089 void __dl_update(struct dl_bw *dl_b, s64 bw) 2090 { 2091 struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw); 2092 int i; 2093 2094 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(), 2095 "sched RCU must be held"); 2096 for_each_cpu_and(i, rd->span, cpu_active_mask) { 2097 struct rq *rq = cpu_rq(i); 2098 2099 rq->dl.extra_bw += bw; 2100 } 2101 } 2102 #else 2103 static inline 2104 void __dl_update(struct dl_bw *dl_b, s64 bw) 2105 { 2106 struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw); 2107 2108 dl->extra_bw += bw; 2109 } 2110 #endif 2111 2112 2113 #ifdef CONFIG_IRQ_TIME_ACCOUNTING 2114 struct irqtime { 2115 u64 total; 2116 u64 tick_delta; 2117 u64 irq_start_time; 2118 struct u64_stats_sync sync; 2119 }; 2120 2121 DECLARE_PER_CPU(struct irqtime, cpu_irqtime); 2122 2123 /* 2124 * Returns the irqtime minus the softirq time computed by ksoftirqd. 2125 * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime 2126 * and never move forward. 2127 */ 2128 static inline u64 irq_time_read(int cpu) 2129 { 2130 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu); 2131 unsigned int seq; 2132 u64 total; 2133 2134 do { 2135 seq = __u64_stats_fetch_begin(&irqtime->sync); 2136 total = irqtime->total; 2137 } while (__u64_stats_fetch_retry(&irqtime->sync, seq)); 2138 2139 return total; 2140 } 2141 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ 2142 2143 #ifdef CONFIG_CPU_FREQ 2144 DECLARE_PER_CPU(struct update_util_data *, cpufreq_update_util_data); 2145 2146 /** 2147 * cpufreq_update_util - Take a note about CPU utilization changes. 2148 * @rq: Runqueue to carry out the update for. 2149 * @flags: Update reason flags. 2150 * 2151 * This function is called by the scheduler on the CPU whose utilization is 2152 * being updated. 2153 * 2154 * It can only be called from RCU-sched read-side critical sections. 2155 * 2156 * The way cpufreq is currently arranged requires it to evaluate the CPU 2157 * performance state (frequency/voltage) on a regular basis to prevent it from 2158 * being stuck in a completely inadequate performance level for too long. 2159 * That is not guaranteed to happen if the updates are only triggered from CFS 2160 * and DL, though, because they may not be coming in if only RT tasks are 2161 * active all the time (or there are RT tasks only). 2162 * 2163 * As a workaround for that issue, this function is called periodically by the 2164 * RT sched class to trigger extra cpufreq updates to prevent it from stalling, 2165 * but that really is a band-aid. Going forward it should be replaced with 2166 * solutions targeted more specifically at RT tasks. 2167 */ 2168 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) 2169 { 2170 struct update_util_data *data; 2171 2172 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data, 2173 cpu_of(rq))); 2174 if (data) 2175 data->func(data, rq_clock(rq), flags); 2176 } 2177 #else 2178 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {} 2179 #endif /* CONFIG_CPU_FREQ */ 2180 2181 #ifdef arch_scale_freq_capacity 2182 # ifndef arch_scale_freq_invariant 2183 # define arch_scale_freq_invariant() true 2184 # endif 2185 #else 2186 # define arch_scale_freq_invariant() false 2187 #endif 2188 2189 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL 2190 static inline unsigned long cpu_bw_dl(struct rq *rq) 2191 { 2192 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT; 2193 } 2194 2195 static inline unsigned long cpu_util_dl(struct rq *rq) 2196 { 2197 return READ_ONCE(rq->avg_dl.util_avg); 2198 } 2199 2200 static inline unsigned long cpu_util_cfs(struct rq *rq) 2201 { 2202 unsigned long util = READ_ONCE(rq->cfs.avg.util_avg); 2203 2204 if (sched_feat(UTIL_EST)) { 2205 util = max_t(unsigned long, util, 2206 READ_ONCE(rq->cfs.avg.util_est.enqueued)); 2207 } 2208 2209 return util; 2210 } 2211 2212 static inline unsigned long cpu_util_rt(struct rq *rq) 2213 { 2214 return READ_ONCE(rq->avg_rt.util_avg); 2215 } 2216 #endif 2217 2218 #ifdef HAVE_SCHED_AVG_IRQ 2219 static inline unsigned long cpu_util_irq(struct rq *rq) 2220 { 2221 return rq->avg_irq.util_avg; 2222 } 2223 2224 static inline 2225 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max) 2226 { 2227 util *= (max - irq); 2228 util /= max; 2229 2230 return util; 2231 2232 } 2233 #else 2234 static inline unsigned long cpu_util_irq(struct rq *rq) 2235 { 2236 return 0; 2237 } 2238 2239 static inline 2240 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max) 2241 { 2242 return util; 2243 } 2244 #endif 2245