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/stop_machine.h> 60 #include <linux/suspend.h> 61 #include <linux/swait.h> 62 #include <linux/syscalls.h> 63 #include <linux/task_work.h> 64 #include <linux/tsacct_kern.h> 65 66 #include <asm/tlb.h> 67 68 #ifdef CONFIG_PARAVIRT 69 # include <asm/paravirt.h> 70 #endif 71 72 #include "cpupri.h" 73 #include "cpudeadline.h" 74 75 #ifdef CONFIG_SCHED_DEBUG 76 # define SCHED_WARN_ON(x) WARN_ONCE(x, #x) 77 #else 78 # define SCHED_WARN_ON(x) ({ (void)(x), 0; }) 79 #endif 80 81 struct rq; 82 struct cpuidle_state; 83 84 /* task_struct::on_rq states: */ 85 #define TASK_ON_RQ_QUEUED 1 86 #define TASK_ON_RQ_MIGRATING 2 87 88 extern __read_mostly int scheduler_running; 89 90 extern unsigned long calc_load_update; 91 extern atomic_long_t calc_load_tasks; 92 93 extern void calc_global_load_tick(struct rq *this_rq); 94 extern long calc_load_fold_active(struct rq *this_rq, long adjust); 95 96 #ifdef CONFIG_SMP 97 extern void cpu_load_update_active(struct rq *this_rq); 98 #else 99 static inline void cpu_load_update_active(struct rq *this_rq) { } 100 #endif 101 102 /* 103 * Helpers for converting nanosecond timing to jiffy resolution 104 */ 105 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) 106 107 /* 108 * Increase resolution of nice-level calculations for 64-bit architectures. 109 * The extra resolution improves shares distribution and load balancing of 110 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup 111 * hierarchies, especially on larger systems. This is not a user-visible change 112 * and does not change the user-interface for setting shares/weights. 113 * 114 * We increase resolution only if we have enough bits to allow this increased 115 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit 116 * are pretty high and the returns do not justify the increased costs. 117 * 118 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to 119 * increase coverage and consistency always enable it on 64-bit platforms. 120 */ 121 #ifdef CONFIG_64BIT 122 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT) 123 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT) 124 # define scale_load_down(w) ((w) >> SCHED_FIXEDPOINT_SHIFT) 125 #else 126 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT) 127 # define scale_load(w) (w) 128 # define scale_load_down(w) (w) 129 #endif 130 131 /* 132 * Task weight (visible to users) and its load (invisible to users) have 133 * independent resolution, but they should be well calibrated. We use 134 * scale_load() and scale_load_down(w) to convert between them. The 135 * following must be true: 136 * 137 * scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD 138 * 139 */ 140 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT) 141 142 /* 143 * Single value that decides SCHED_DEADLINE internal math precision. 144 * 10 -> just above 1us 145 * 9 -> just above 0.5us 146 */ 147 #define DL_SCALE 10 148 149 /* 150 * Single value that denotes runtime == period, ie unlimited time. 151 */ 152 #define RUNTIME_INF ((u64)~0ULL) 153 154 static inline int idle_policy(int policy) 155 { 156 return policy == SCHED_IDLE; 157 } 158 static inline int fair_policy(int policy) 159 { 160 return policy == SCHED_NORMAL || policy == SCHED_BATCH; 161 } 162 163 static inline int rt_policy(int policy) 164 { 165 return policy == SCHED_FIFO || policy == SCHED_RR; 166 } 167 168 static inline int dl_policy(int policy) 169 { 170 return policy == SCHED_DEADLINE; 171 } 172 static inline bool valid_policy(int policy) 173 { 174 return idle_policy(policy) || fair_policy(policy) || 175 rt_policy(policy) || dl_policy(policy); 176 } 177 178 static inline int task_has_rt_policy(struct task_struct *p) 179 { 180 return rt_policy(p->policy); 181 } 182 183 static inline int task_has_dl_policy(struct task_struct *p) 184 { 185 return dl_policy(p->policy); 186 } 187 188 #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT) 189 190 /* 191 * !! For sched_setattr_nocheck() (kernel) only !! 192 * 193 * This is actually gross. :( 194 * 195 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE 196 * tasks, but still be able to sleep. We need this on platforms that cannot 197 * atomically change clock frequency. Remove once fast switching will be 198 * available on such platforms. 199 * 200 * SUGOV stands for SchedUtil GOVernor. 201 */ 202 #define SCHED_FLAG_SUGOV 0x10000000 203 204 static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se) 205 { 206 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL 207 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV); 208 #else 209 return false; 210 #endif 211 } 212 213 /* 214 * Tells if entity @a should preempt entity @b. 215 */ 216 static inline bool 217 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b) 218 { 219 return dl_entity_is_special(a) || 220 dl_time_before(a->deadline, b->deadline); 221 } 222 223 /* 224 * This is the priority-queue data structure of the RT scheduling class: 225 */ 226 struct rt_prio_array { 227 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ 228 struct list_head queue[MAX_RT_PRIO]; 229 }; 230 231 struct rt_bandwidth { 232 /* nests inside the rq lock: */ 233 raw_spinlock_t rt_runtime_lock; 234 ktime_t rt_period; 235 u64 rt_runtime; 236 struct hrtimer rt_period_timer; 237 unsigned int rt_period_active; 238 }; 239 240 void __dl_clear_params(struct task_struct *p); 241 242 /* 243 * To keep the bandwidth of -deadline tasks and groups under control 244 * we need some place where: 245 * - store the maximum -deadline bandwidth of the system (the group); 246 * - cache the fraction of that bandwidth that is currently allocated. 247 * 248 * This is all done in the data structure below. It is similar to the 249 * one used for RT-throttling (rt_bandwidth), with the main difference 250 * that, since here we are only interested in admission control, we 251 * do not decrease any runtime while the group "executes", neither we 252 * need a timer to replenish it. 253 * 254 * With respect to SMP, the bandwidth is given on a per-CPU basis, 255 * meaning that: 256 * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU; 257 * - dl_total_bw array contains, in the i-eth element, the currently 258 * allocated bandwidth on the i-eth CPU. 259 * Moreover, groups consume bandwidth on each CPU, while tasks only 260 * consume bandwidth on the CPU they're running on. 261 * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw 262 * that will be shown the next time the proc or cgroup controls will 263 * be red. It on its turn can be changed by writing on its own 264 * control. 265 */ 266 struct dl_bandwidth { 267 raw_spinlock_t dl_runtime_lock; 268 u64 dl_runtime; 269 u64 dl_period; 270 }; 271 272 static inline int dl_bandwidth_enabled(void) 273 { 274 return sysctl_sched_rt_runtime >= 0; 275 } 276 277 struct dl_bw { 278 raw_spinlock_t lock; 279 u64 bw; 280 u64 total_bw; 281 }; 282 283 static inline void __dl_update(struct dl_bw *dl_b, s64 bw); 284 285 static inline 286 void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus) 287 { 288 dl_b->total_bw -= tsk_bw; 289 __dl_update(dl_b, (s32)tsk_bw / cpus); 290 } 291 292 static inline 293 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus) 294 { 295 dl_b->total_bw += tsk_bw; 296 __dl_update(dl_b, -((s32)tsk_bw / cpus)); 297 } 298 299 static inline 300 bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw) 301 { 302 return dl_b->bw != -1 && 303 dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw; 304 } 305 306 extern void dl_change_utilization(struct task_struct *p, u64 new_bw); 307 extern void init_dl_bw(struct dl_bw *dl_b); 308 extern int sched_dl_global_validate(void); 309 extern void sched_dl_do_global(void); 310 extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr); 311 extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr); 312 extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr); 313 extern bool __checkparam_dl(const struct sched_attr *attr); 314 extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr); 315 extern int dl_task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed); 316 extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial); 317 extern bool dl_cpu_busy(unsigned int cpu); 318 319 #ifdef CONFIG_CGROUP_SCHED 320 321 #include <linux/cgroup.h> 322 323 struct cfs_rq; 324 struct rt_rq; 325 326 extern struct list_head task_groups; 327 328 struct cfs_bandwidth { 329 #ifdef CONFIG_CFS_BANDWIDTH 330 raw_spinlock_t lock; 331 ktime_t period; 332 u64 quota; 333 u64 runtime; 334 s64 hierarchical_quota; 335 u64 runtime_expires; 336 int expires_seq; 337 338 short idle; 339 short period_active; 340 struct hrtimer period_timer; 341 struct hrtimer slack_timer; 342 struct list_head throttled_cfs_rq; 343 344 /* Statistics: */ 345 int nr_periods; 346 int nr_throttled; 347 u64 throttled_time; 348 349 bool distribute_running; 350 #endif 351 }; 352 353 /* Task group related information */ 354 struct task_group { 355 struct cgroup_subsys_state css; 356 357 #ifdef CONFIG_FAIR_GROUP_SCHED 358 /* schedulable entities of this group on each CPU */ 359 struct sched_entity **se; 360 /* runqueue "owned" by this group on each CPU */ 361 struct cfs_rq **cfs_rq; 362 unsigned long shares; 363 364 #ifdef CONFIG_SMP 365 /* 366 * load_avg can be heavily contended at clock tick time, so put 367 * it in its own cacheline separated from the fields above which 368 * will also be accessed at each tick. 369 */ 370 atomic_long_t load_avg ____cacheline_aligned; 371 #endif 372 #endif 373 374 #ifdef CONFIG_RT_GROUP_SCHED 375 struct sched_rt_entity **rt_se; 376 struct rt_rq **rt_rq; 377 378 struct rt_bandwidth rt_bandwidth; 379 #endif 380 381 struct rcu_head rcu; 382 struct list_head list; 383 384 struct task_group *parent; 385 struct list_head siblings; 386 struct list_head children; 387 388 #ifdef CONFIG_SCHED_AUTOGROUP 389 struct autogroup *autogroup; 390 #endif 391 392 struct cfs_bandwidth cfs_bandwidth; 393 }; 394 395 #ifdef CONFIG_FAIR_GROUP_SCHED 396 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD 397 398 /* 399 * A weight of 0 or 1 can cause arithmetics problems. 400 * A weight of a cfs_rq is the sum of weights of which entities 401 * are queued on this cfs_rq, so a weight of a entity should not be 402 * too large, so as the shares value of a task group. 403 * (The default weight is 1024 - so there's no practical 404 * limitation from this.) 405 */ 406 #define MIN_SHARES (1UL << 1) 407 #define MAX_SHARES (1UL << 18) 408 #endif 409 410 typedef int (*tg_visitor)(struct task_group *, void *); 411 412 extern int walk_tg_tree_from(struct task_group *from, 413 tg_visitor down, tg_visitor up, void *data); 414 415 /* 416 * Iterate the full tree, calling @down when first entering a node and @up when 417 * leaving it for the final time. 418 * 419 * Caller must hold rcu_lock or sufficient equivalent. 420 */ 421 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) 422 { 423 return walk_tg_tree_from(&root_task_group, down, up, data); 424 } 425 426 extern int tg_nop(struct task_group *tg, void *data); 427 428 extern void free_fair_sched_group(struct task_group *tg); 429 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent); 430 extern void online_fair_sched_group(struct task_group *tg); 431 extern void unregister_fair_sched_group(struct task_group *tg); 432 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, 433 struct sched_entity *se, int cpu, 434 struct sched_entity *parent); 435 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b); 436 437 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b); 438 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b); 439 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq); 440 441 extern void free_rt_sched_group(struct task_group *tg); 442 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent); 443 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, 444 struct sched_rt_entity *rt_se, int cpu, 445 struct sched_rt_entity *parent); 446 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us); 447 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us); 448 extern long sched_group_rt_runtime(struct task_group *tg); 449 extern long sched_group_rt_period(struct task_group *tg); 450 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk); 451 452 extern struct task_group *sched_create_group(struct task_group *parent); 453 extern void sched_online_group(struct task_group *tg, 454 struct task_group *parent); 455 extern void sched_destroy_group(struct task_group *tg); 456 extern void sched_offline_group(struct task_group *tg); 457 458 extern void sched_move_task(struct task_struct *tsk); 459 460 #ifdef CONFIG_FAIR_GROUP_SCHED 461 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares); 462 463 #ifdef CONFIG_SMP 464 extern void set_task_rq_fair(struct sched_entity *se, 465 struct cfs_rq *prev, struct cfs_rq *next); 466 #else /* !CONFIG_SMP */ 467 static inline void set_task_rq_fair(struct sched_entity *se, 468 struct cfs_rq *prev, struct cfs_rq *next) { } 469 #endif /* CONFIG_SMP */ 470 #endif /* CONFIG_FAIR_GROUP_SCHED */ 471 472 #else /* CONFIG_CGROUP_SCHED */ 473 474 struct cfs_bandwidth { }; 475 476 #endif /* CONFIG_CGROUP_SCHED */ 477 478 /* CFS-related fields in a runqueue */ 479 struct cfs_rq { 480 struct load_weight load; 481 unsigned long runnable_weight; 482 unsigned int nr_running; 483 unsigned int h_nr_running; 484 485 u64 exec_clock; 486 u64 min_vruntime; 487 #ifndef CONFIG_64BIT 488 u64 min_vruntime_copy; 489 #endif 490 491 struct rb_root_cached tasks_timeline; 492 493 /* 494 * 'curr' points to currently running entity on this cfs_rq. 495 * It is set to NULL otherwise (i.e when none are currently running). 496 */ 497 struct sched_entity *curr; 498 struct sched_entity *next; 499 struct sched_entity *last; 500 struct sched_entity *skip; 501 502 #ifdef CONFIG_SCHED_DEBUG 503 unsigned int nr_spread_over; 504 #endif 505 506 #ifdef CONFIG_SMP 507 /* 508 * CFS load tracking 509 */ 510 struct sched_avg avg; 511 #ifndef CONFIG_64BIT 512 u64 load_last_update_time_copy; 513 #endif 514 struct { 515 raw_spinlock_t lock ____cacheline_aligned; 516 int nr; 517 unsigned long load_avg; 518 unsigned long util_avg; 519 unsigned long runnable_sum; 520 } removed; 521 522 #ifdef CONFIG_FAIR_GROUP_SCHED 523 unsigned long tg_load_avg_contrib; 524 long propagate; 525 long prop_runnable_sum; 526 527 /* 528 * h_load = weight * f(tg) 529 * 530 * Where f(tg) is the recursive weight fraction assigned to 531 * this group. 532 */ 533 unsigned long h_load; 534 u64 last_h_load_update; 535 struct sched_entity *h_load_next; 536 #endif /* CONFIG_FAIR_GROUP_SCHED */ 537 #endif /* CONFIG_SMP */ 538 539 #ifdef CONFIG_FAIR_GROUP_SCHED 540 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */ 541 542 /* 543 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in 544 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities 545 * (like users, containers etc.) 546 * 547 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU. 548 * This list is used during load balance. 549 */ 550 int on_list; 551 struct list_head leaf_cfs_rq_list; 552 struct task_group *tg; /* group that "owns" this runqueue */ 553 554 #ifdef CONFIG_CFS_BANDWIDTH 555 int runtime_enabled; 556 int expires_seq; 557 u64 runtime_expires; 558 s64 runtime_remaining; 559 560 u64 throttled_clock; 561 u64 throttled_clock_task; 562 u64 throttled_clock_task_time; 563 int throttled; 564 int throttle_count; 565 struct list_head throttled_list; 566 #endif /* CONFIG_CFS_BANDWIDTH */ 567 #endif /* CONFIG_FAIR_GROUP_SCHED */ 568 }; 569 570 static inline int rt_bandwidth_enabled(void) 571 { 572 return sysctl_sched_rt_runtime >= 0; 573 } 574 575 /* RT IPI pull logic requires IRQ_WORK */ 576 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP) 577 # define HAVE_RT_PUSH_IPI 578 #endif 579 580 /* Real-Time classes' related field in a runqueue: */ 581 struct rt_rq { 582 struct rt_prio_array active; 583 unsigned int rt_nr_running; 584 unsigned int rr_nr_running; 585 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED 586 struct { 587 int curr; /* highest queued rt task prio */ 588 #ifdef CONFIG_SMP 589 int next; /* next highest */ 590 #endif 591 } highest_prio; 592 #endif 593 #ifdef CONFIG_SMP 594 unsigned long rt_nr_migratory; 595 unsigned long rt_nr_total; 596 int overloaded; 597 struct plist_head pushable_tasks; 598 599 #endif /* CONFIG_SMP */ 600 int rt_queued; 601 602 int rt_throttled; 603 u64 rt_time; 604 u64 rt_runtime; 605 /* Nests inside the rq lock: */ 606 raw_spinlock_t rt_runtime_lock; 607 608 #ifdef CONFIG_RT_GROUP_SCHED 609 unsigned long rt_nr_boosted; 610 611 struct rq *rq; 612 struct task_group *tg; 613 #endif 614 }; 615 616 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq) 617 { 618 return rt_rq->rt_queued && rt_rq->rt_nr_running; 619 } 620 621 /* Deadline class' related fields in a runqueue */ 622 struct dl_rq { 623 /* runqueue is an rbtree, ordered by deadline */ 624 struct rb_root_cached root; 625 626 unsigned long dl_nr_running; 627 628 #ifdef CONFIG_SMP 629 /* 630 * Deadline values of the currently executing and the 631 * earliest ready task on this rq. Caching these facilitates 632 * the decision wether or not a ready but not running task 633 * should migrate somewhere else. 634 */ 635 struct { 636 u64 curr; 637 u64 next; 638 } earliest_dl; 639 640 unsigned long dl_nr_migratory; 641 int overloaded; 642 643 /* 644 * Tasks on this rq that can be pushed away. They are kept in 645 * an rb-tree, ordered by tasks' deadlines, with caching 646 * of the leftmost (earliest deadline) element. 647 */ 648 struct rb_root_cached pushable_dl_tasks_root; 649 #else 650 struct dl_bw dl_bw; 651 #endif 652 /* 653 * "Active utilization" for this runqueue: increased when a 654 * task wakes up (becomes TASK_RUNNING) and decreased when a 655 * task blocks 656 */ 657 u64 running_bw; 658 659 /* 660 * Utilization of the tasks "assigned" to this runqueue (including 661 * the tasks that are in runqueue and the tasks that executed on this 662 * CPU and blocked). Increased when a task moves to this runqueue, and 663 * decreased when the task moves away (migrates, changes scheduling 664 * policy, or terminates). 665 * This is needed to compute the "inactive utilization" for the 666 * runqueue (inactive utilization = this_bw - running_bw). 667 */ 668 u64 this_bw; 669 u64 extra_bw; 670 671 /* 672 * Inverse of the fraction of CPU utilization that can be reclaimed 673 * by the GRUB algorithm. 674 */ 675 u64 bw_ratio; 676 }; 677 678 #ifdef CONFIG_FAIR_GROUP_SCHED 679 /* An entity is a task if it doesn't "own" a runqueue */ 680 #define entity_is_task(se) (!se->my_q) 681 #else 682 #define entity_is_task(se) 1 683 #endif 684 685 #ifdef CONFIG_SMP 686 /* 687 * XXX we want to get rid of these helpers and use the full load resolution. 688 */ 689 static inline long se_weight(struct sched_entity *se) 690 { 691 return scale_load_down(se->load.weight); 692 } 693 694 static inline long se_runnable(struct sched_entity *se) 695 { 696 return scale_load_down(se->runnable_weight); 697 } 698 699 static inline bool sched_asym_prefer(int a, int b) 700 { 701 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b); 702 } 703 704 /* 705 * We add the notion of a root-domain which will be used to define per-domain 706 * variables. Each exclusive cpuset essentially defines an island domain by 707 * fully partitioning the member CPUs from any other cpuset. Whenever a new 708 * exclusive cpuset is created, we also create and attach a new root-domain 709 * object. 710 * 711 */ 712 struct root_domain { 713 atomic_t refcount; 714 atomic_t rto_count; 715 struct rcu_head rcu; 716 cpumask_var_t span; 717 cpumask_var_t online; 718 719 /* 720 * Indicate pullable load on at least one CPU, e.g: 721 * - More than one runnable task 722 * - Running task is misfit 723 */ 724 int overload; 725 726 /* 727 * The bit corresponding to a CPU gets set here if such CPU has more 728 * than one runnable -deadline task (as it is below for RT tasks). 729 */ 730 cpumask_var_t dlo_mask; 731 atomic_t dlo_count; 732 struct dl_bw dl_bw; 733 struct cpudl cpudl; 734 735 #ifdef HAVE_RT_PUSH_IPI 736 /* 737 * For IPI pull requests, loop across the rto_mask. 738 */ 739 struct irq_work rto_push_work; 740 raw_spinlock_t rto_lock; 741 /* These are only updated and read within rto_lock */ 742 int rto_loop; 743 int rto_cpu; 744 /* These atomics are updated outside of a lock */ 745 atomic_t rto_loop_next; 746 atomic_t rto_loop_start; 747 #endif 748 /* 749 * The "RT overload" flag: it gets set if a CPU has more than 750 * one runnable RT task. 751 */ 752 cpumask_var_t rto_mask; 753 struct cpupri cpupri; 754 755 unsigned long max_cpu_capacity; 756 }; 757 758 extern struct root_domain def_root_domain; 759 extern struct mutex sched_domains_mutex; 760 761 extern void init_defrootdomain(void); 762 extern int sched_init_domains(const struct cpumask *cpu_map); 763 extern void rq_attach_root(struct rq *rq, struct root_domain *rd); 764 extern void sched_get_rd(struct root_domain *rd); 765 extern void sched_put_rd(struct root_domain *rd); 766 767 #ifdef HAVE_RT_PUSH_IPI 768 extern void rto_push_irq_work_func(struct irq_work *work); 769 #endif 770 #endif /* CONFIG_SMP */ 771 772 /* 773 * This is the main, per-CPU runqueue data structure. 774 * 775 * Locking rule: those places that want to lock multiple runqueues 776 * (such as the load balancing or the thread migration code), lock 777 * acquire operations must be ordered by ascending &runqueue. 778 */ 779 struct rq { 780 /* runqueue lock: */ 781 raw_spinlock_t lock; 782 783 /* 784 * nr_running and cpu_load should be in the same cacheline because 785 * remote CPUs use both these fields when doing load calculation. 786 */ 787 unsigned int nr_running; 788 #ifdef CONFIG_NUMA_BALANCING 789 unsigned int nr_numa_running; 790 unsigned int nr_preferred_running; 791 unsigned int numa_migrate_on; 792 #endif 793 #define CPU_LOAD_IDX_MAX 5 794 unsigned long cpu_load[CPU_LOAD_IDX_MAX]; 795 #ifdef CONFIG_NO_HZ_COMMON 796 #ifdef CONFIG_SMP 797 unsigned long last_load_update_tick; 798 unsigned long last_blocked_load_update_tick; 799 unsigned int has_blocked_load; 800 #endif /* CONFIG_SMP */ 801 unsigned int nohz_tick_stopped; 802 atomic_t nohz_flags; 803 #endif /* CONFIG_NO_HZ_COMMON */ 804 805 /* capture load from *all* tasks on this CPU: */ 806 struct load_weight load; 807 unsigned long nr_load_updates; 808 u64 nr_switches; 809 810 struct cfs_rq cfs; 811 struct rt_rq rt; 812 struct dl_rq dl; 813 814 #ifdef CONFIG_FAIR_GROUP_SCHED 815 /* list of leaf cfs_rq on this CPU: */ 816 struct list_head leaf_cfs_rq_list; 817 struct list_head *tmp_alone_branch; 818 #endif /* CONFIG_FAIR_GROUP_SCHED */ 819 820 /* 821 * This is part of a global counter where only the total sum 822 * over all CPUs matters. A task can increase this counter on 823 * one CPU and if it got migrated afterwards it may decrease 824 * it on another CPU. Always updated under the runqueue lock: 825 */ 826 unsigned long nr_uninterruptible; 827 828 struct task_struct *curr; 829 struct task_struct *idle; 830 struct task_struct *stop; 831 unsigned long next_balance; 832 struct mm_struct *prev_mm; 833 834 unsigned int clock_update_flags; 835 u64 clock; 836 u64 clock_task; 837 838 atomic_t nr_iowait; 839 840 #ifdef CONFIG_SMP 841 struct root_domain *rd; 842 struct sched_domain *sd; 843 844 unsigned long cpu_capacity; 845 unsigned long cpu_capacity_orig; 846 847 struct callback_head *balance_callback; 848 849 unsigned char idle_balance; 850 851 unsigned long misfit_task_load; 852 853 /* For active balancing */ 854 int active_balance; 855 int push_cpu; 856 struct cpu_stop_work active_balance_work; 857 858 /* CPU of this runqueue: */ 859 int cpu; 860 int online; 861 862 struct list_head cfs_tasks; 863 864 struct sched_avg avg_rt; 865 struct sched_avg avg_dl; 866 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ 867 struct sched_avg avg_irq; 868 #endif 869 u64 idle_stamp; 870 u64 avg_idle; 871 872 /* This is used to determine avg_idle's max value */ 873 u64 max_idle_balance_cost; 874 #endif 875 876 #ifdef CONFIG_IRQ_TIME_ACCOUNTING 877 u64 prev_irq_time; 878 #endif 879 #ifdef CONFIG_PARAVIRT 880 u64 prev_steal_time; 881 #endif 882 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING 883 u64 prev_steal_time_rq; 884 #endif 885 886 /* calc_load related fields */ 887 unsigned long calc_load_update; 888 long calc_load_active; 889 890 #ifdef CONFIG_SCHED_HRTICK 891 #ifdef CONFIG_SMP 892 int hrtick_csd_pending; 893 call_single_data_t hrtick_csd; 894 #endif 895 struct hrtimer hrtick_timer; 896 #endif 897 898 #ifdef CONFIG_SCHEDSTATS 899 /* latency stats */ 900 struct sched_info rq_sched_info; 901 unsigned long long rq_cpu_time; 902 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ 903 904 /* sys_sched_yield() stats */ 905 unsigned int yld_count; 906 907 /* schedule() stats */ 908 unsigned int sched_count; 909 unsigned int sched_goidle; 910 911 /* try_to_wake_up() stats */ 912 unsigned int ttwu_count; 913 unsigned int ttwu_local; 914 #endif 915 916 #ifdef CONFIG_SMP 917 struct llist_head wake_list; 918 #endif 919 920 #ifdef CONFIG_CPU_IDLE 921 /* Must be inspected within a rcu lock section */ 922 struct cpuidle_state *idle_state; 923 #endif 924 }; 925 926 static inline int cpu_of(struct rq *rq) 927 { 928 #ifdef CONFIG_SMP 929 return rq->cpu; 930 #else 931 return 0; 932 #endif 933 } 934 935 936 #ifdef CONFIG_SCHED_SMT 937 938 extern struct static_key_false sched_smt_present; 939 940 extern void __update_idle_core(struct rq *rq); 941 942 static inline void update_idle_core(struct rq *rq) 943 { 944 if (static_branch_unlikely(&sched_smt_present)) 945 __update_idle_core(rq); 946 } 947 948 #else 949 static inline void update_idle_core(struct rq *rq) { } 950 #endif 951 952 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); 953 954 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) 955 #define this_rq() this_cpu_ptr(&runqueues) 956 #define task_rq(p) cpu_rq(task_cpu(p)) 957 #define cpu_curr(cpu) (cpu_rq(cpu)->curr) 958 #define raw_rq() raw_cpu_ptr(&runqueues) 959 960 static inline u64 __rq_clock_broken(struct rq *rq) 961 { 962 return READ_ONCE(rq->clock); 963 } 964 965 /* 966 * rq::clock_update_flags bits 967 * 968 * %RQCF_REQ_SKIP - will request skipping of clock update on the next 969 * call to __schedule(). This is an optimisation to avoid 970 * neighbouring rq clock updates. 971 * 972 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is 973 * in effect and calls to update_rq_clock() are being ignored. 974 * 975 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been 976 * made to update_rq_clock() since the last time rq::lock was pinned. 977 * 978 * If inside of __schedule(), clock_update_flags will have been 979 * shifted left (a left shift is a cheap operation for the fast path 980 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use, 981 * 982 * if (rq-clock_update_flags >= RQCF_UPDATED) 983 * 984 * to check if %RQCF_UPADTED is set. It'll never be shifted more than 985 * one position though, because the next rq_unpin_lock() will shift it 986 * back. 987 */ 988 #define RQCF_REQ_SKIP 0x01 989 #define RQCF_ACT_SKIP 0x02 990 #define RQCF_UPDATED 0x04 991 992 static inline void assert_clock_updated(struct rq *rq) 993 { 994 /* 995 * The only reason for not seeing a clock update since the 996 * last rq_pin_lock() is if we're currently skipping updates. 997 */ 998 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP); 999 } 1000 1001 static inline u64 rq_clock(struct rq *rq) 1002 { 1003 lockdep_assert_held(&rq->lock); 1004 assert_clock_updated(rq); 1005 1006 return rq->clock; 1007 } 1008 1009 static inline u64 rq_clock_task(struct rq *rq) 1010 { 1011 lockdep_assert_held(&rq->lock); 1012 assert_clock_updated(rq); 1013 1014 return rq->clock_task; 1015 } 1016 1017 static inline void rq_clock_skip_update(struct rq *rq) 1018 { 1019 lockdep_assert_held(&rq->lock); 1020 rq->clock_update_flags |= RQCF_REQ_SKIP; 1021 } 1022 1023 /* 1024 * See rt task throttling, which is the only time a skip 1025 * request is cancelled. 1026 */ 1027 static inline void rq_clock_cancel_skipupdate(struct rq *rq) 1028 { 1029 lockdep_assert_held(&rq->lock); 1030 rq->clock_update_flags &= ~RQCF_REQ_SKIP; 1031 } 1032 1033 struct rq_flags { 1034 unsigned long flags; 1035 struct pin_cookie cookie; 1036 #ifdef CONFIG_SCHED_DEBUG 1037 /* 1038 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the 1039 * current pin context is stashed here in case it needs to be 1040 * restored in rq_repin_lock(). 1041 */ 1042 unsigned int clock_update_flags; 1043 #endif 1044 }; 1045 1046 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf) 1047 { 1048 rf->cookie = lockdep_pin_lock(&rq->lock); 1049 1050 #ifdef CONFIG_SCHED_DEBUG 1051 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP); 1052 rf->clock_update_flags = 0; 1053 #endif 1054 } 1055 1056 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf) 1057 { 1058 #ifdef CONFIG_SCHED_DEBUG 1059 if (rq->clock_update_flags > RQCF_ACT_SKIP) 1060 rf->clock_update_flags = RQCF_UPDATED; 1061 #endif 1062 1063 lockdep_unpin_lock(&rq->lock, rf->cookie); 1064 } 1065 1066 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf) 1067 { 1068 lockdep_repin_lock(&rq->lock, rf->cookie); 1069 1070 #ifdef CONFIG_SCHED_DEBUG 1071 /* 1072 * Restore the value we stashed in @rf for this pin context. 1073 */ 1074 rq->clock_update_flags |= rf->clock_update_flags; 1075 #endif 1076 } 1077 1078 #ifdef CONFIG_NUMA 1079 enum numa_topology_type { 1080 NUMA_DIRECT, 1081 NUMA_GLUELESS_MESH, 1082 NUMA_BACKPLANE, 1083 }; 1084 extern enum numa_topology_type sched_numa_topology_type; 1085 extern int sched_max_numa_distance; 1086 extern bool find_numa_distance(int distance); 1087 #endif 1088 1089 #ifdef CONFIG_NUMA 1090 extern void sched_init_numa(void); 1091 extern void sched_domains_numa_masks_set(unsigned int cpu); 1092 extern void sched_domains_numa_masks_clear(unsigned int cpu); 1093 #else 1094 static inline void sched_init_numa(void) { } 1095 static inline void sched_domains_numa_masks_set(unsigned int cpu) { } 1096 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { } 1097 #endif 1098 1099 #ifdef CONFIG_NUMA_BALANCING 1100 /* The regions in numa_faults array from task_struct */ 1101 enum numa_faults_stats { 1102 NUMA_MEM = 0, 1103 NUMA_CPU, 1104 NUMA_MEMBUF, 1105 NUMA_CPUBUF 1106 }; 1107 extern void sched_setnuma(struct task_struct *p, int node); 1108 extern int migrate_task_to(struct task_struct *p, int cpu); 1109 extern int migrate_swap(struct task_struct *p, struct task_struct *t, 1110 int cpu, int scpu); 1111 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p); 1112 #else 1113 static inline void 1114 init_numa_balancing(unsigned long clone_flags, struct task_struct *p) 1115 { 1116 } 1117 #endif /* CONFIG_NUMA_BALANCING */ 1118 1119 #ifdef CONFIG_SMP 1120 1121 static inline void 1122 queue_balance_callback(struct rq *rq, 1123 struct callback_head *head, 1124 void (*func)(struct rq *rq)) 1125 { 1126 lockdep_assert_held(&rq->lock); 1127 1128 if (unlikely(head->next)) 1129 return; 1130 1131 head->func = (void (*)(struct callback_head *))func; 1132 head->next = rq->balance_callback; 1133 rq->balance_callback = head; 1134 } 1135 1136 extern void sched_ttwu_pending(void); 1137 1138 #define rcu_dereference_check_sched_domain(p) \ 1139 rcu_dereference_check((p), \ 1140 lockdep_is_held(&sched_domains_mutex)) 1141 1142 /* 1143 * The domain tree (rq->sd) is protected by RCU's quiescent state transition. 1144 * See detach_destroy_domains: synchronize_sched for details. 1145 * 1146 * The domain tree of any CPU may only be accessed from within 1147 * preempt-disabled sections. 1148 */ 1149 #define for_each_domain(cpu, __sd) \ 1150 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \ 1151 __sd; __sd = __sd->parent) 1152 1153 #define for_each_lower_domain(sd) for (; sd; sd = sd->child) 1154 1155 /** 1156 * highest_flag_domain - Return highest sched_domain containing flag. 1157 * @cpu: The CPU whose highest level of sched domain is to 1158 * be returned. 1159 * @flag: The flag to check for the highest sched_domain 1160 * for the given CPU. 1161 * 1162 * Returns the highest sched_domain of a CPU which contains the given flag. 1163 */ 1164 static inline struct sched_domain *highest_flag_domain(int cpu, int flag) 1165 { 1166 struct sched_domain *sd, *hsd = NULL; 1167 1168 for_each_domain(cpu, sd) { 1169 if (!(sd->flags & flag)) 1170 break; 1171 hsd = sd; 1172 } 1173 1174 return hsd; 1175 } 1176 1177 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag) 1178 { 1179 struct sched_domain *sd; 1180 1181 for_each_domain(cpu, sd) { 1182 if (sd->flags & flag) 1183 break; 1184 } 1185 1186 return sd; 1187 } 1188 1189 DECLARE_PER_CPU(struct sched_domain *, sd_llc); 1190 DECLARE_PER_CPU(int, sd_llc_size); 1191 DECLARE_PER_CPU(int, sd_llc_id); 1192 DECLARE_PER_CPU(struct sched_domain_shared *, sd_llc_shared); 1193 DECLARE_PER_CPU(struct sched_domain *, sd_numa); 1194 DECLARE_PER_CPU(struct sched_domain *, sd_asym); 1195 extern struct static_key_false sched_asym_cpucapacity; 1196 1197 struct sched_group_capacity { 1198 atomic_t ref; 1199 /* 1200 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity 1201 * for a single CPU. 1202 */ 1203 unsigned long capacity; 1204 unsigned long min_capacity; /* Min per-CPU capacity in group */ 1205 unsigned long max_capacity; /* Max per-CPU capacity in group */ 1206 unsigned long next_update; 1207 int imbalance; /* XXX unrelated to capacity but shared group state */ 1208 1209 #ifdef CONFIG_SCHED_DEBUG 1210 int id; 1211 #endif 1212 1213 unsigned long cpumask[0]; /* Balance mask */ 1214 }; 1215 1216 struct sched_group { 1217 struct sched_group *next; /* Must be a circular list */ 1218 atomic_t ref; 1219 1220 unsigned int group_weight; 1221 struct sched_group_capacity *sgc; 1222 int asym_prefer_cpu; /* CPU of highest priority in group */ 1223 1224 /* 1225 * The CPUs this group covers. 1226 * 1227 * NOTE: this field is variable length. (Allocated dynamically 1228 * by attaching extra space to the end of the structure, 1229 * depending on how many CPUs the kernel has booted up with) 1230 */ 1231 unsigned long cpumask[0]; 1232 }; 1233 1234 static inline struct cpumask *sched_group_span(struct sched_group *sg) 1235 { 1236 return to_cpumask(sg->cpumask); 1237 } 1238 1239 /* 1240 * See build_balance_mask(). 1241 */ 1242 static inline struct cpumask *group_balance_mask(struct sched_group *sg) 1243 { 1244 return to_cpumask(sg->sgc->cpumask); 1245 } 1246 1247 /** 1248 * group_first_cpu - Returns the first CPU in the cpumask of a sched_group. 1249 * @group: The group whose first CPU is to be returned. 1250 */ 1251 static inline unsigned int group_first_cpu(struct sched_group *group) 1252 { 1253 return cpumask_first(sched_group_span(group)); 1254 } 1255 1256 extern int group_balance_cpu(struct sched_group *sg); 1257 1258 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) 1259 void register_sched_domain_sysctl(void); 1260 void dirty_sched_domain_sysctl(int cpu); 1261 void unregister_sched_domain_sysctl(void); 1262 #else 1263 static inline void register_sched_domain_sysctl(void) 1264 { 1265 } 1266 static inline void dirty_sched_domain_sysctl(int cpu) 1267 { 1268 } 1269 static inline void unregister_sched_domain_sysctl(void) 1270 { 1271 } 1272 #endif 1273 1274 #else 1275 1276 static inline void sched_ttwu_pending(void) { } 1277 1278 #endif /* CONFIG_SMP */ 1279 1280 #include "stats.h" 1281 #include "autogroup.h" 1282 1283 #ifdef CONFIG_CGROUP_SCHED 1284 1285 /* 1286 * Return the group to which this tasks belongs. 1287 * 1288 * We cannot use task_css() and friends because the cgroup subsystem 1289 * changes that value before the cgroup_subsys::attach() method is called, 1290 * therefore we cannot pin it and might observe the wrong value. 1291 * 1292 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup 1293 * core changes this before calling sched_move_task(). 1294 * 1295 * Instead we use a 'copy' which is updated from sched_move_task() while 1296 * holding both task_struct::pi_lock and rq::lock. 1297 */ 1298 static inline struct task_group *task_group(struct task_struct *p) 1299 { 1300 return p->sched_task_group; 1301 } 1302 1303 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ 1304 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) 1305 { 1306 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED) 1307 struct task_group *tg = task_group(p); 1308 #endif 1309 1310 #ifdef CONFIG_FAIR_GROUP_SCHED 1311 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]); 1312 p->se.cfs_rq = tg->cfs_rq[cpu]; 1313 p->se.parent = tg->se[cpu]; 1314 #endif 1315 1316 #ifdef CONFIG_RT_GROUP_SCHED 1317 p->rt.rt_rq = tg->rt_rq[cpu]; 1318 p->rt.parent = tg->rt_se[cpu]; 1319 #endif 1320 } 1321 1322 #else /* CONFIG_CGROUP_SCHED */ 1323 1324 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } 1325 static inline struct task_group *task_group(struct task_struct *p) 1326 { 1327 return NULL; 1328 } 1329 1330 #endif /* CONFIG_CGROUP_SCHED */ 1331 1332 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) 1333 { 1334 set_task_rq(p, cpu); 1335 #ifdef CONFIG_SMP 1336 /* 1337 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be 1338 * successfuly executed on another CPU. We must ensure that updates of 1339 * per-task data have been completed by this moment. 1340 */ 1341 smp_wmb(); 1342 #ifdef CONFIG_THREAD_INFO_IN_TASK 1343 p->cpu = cpu; 1344 #else 1345 task_thread_info(p)->cpu = cpu; 1346 #endif 1347 p->wake_cpu = cpu; 1348 #endif 1349 } 1350 1351 /* 1352 * Tunables that become constants when CONFIG_SCHED_DEBUG is off: 1353 */ 1354 #ifdef CONFIG_SCHED_DEBUG 1355 # include <linux/static_key.h> 1356 # define const_debug __read_mostly 1357 #else 1358 # define const_debug const 1359 #endif 1360 1361 #define SCHED_FEAT(name, enabled) \ 1362 __SCHED_FEAT_##name , 1363 1364 enum { 1365 #include "features.h" 1366 __SCHED_FEAT_NR, 1367 }; 1368 1369 #undef SCHED_FEAT 1370 1371 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL) 1372 1373 /* 1374 * To support run-time toggling of sched features, all the translation units 1375 * (but core.c) reference the sysctl_sched_features defined in core.c. 1376 */ 1377 extern const_debug unsigned int sysctl_sched_features; 1378 1379 #define SCHED_FEAT(name, enabled) \ 1380 static __always_inline bool static_branch_##name(struct static_key *key) \ 1381 { \ 1382 return static_key_##enabled(key); \ 1383 } 1384 1385 #include "features.h" 1386 #undef SCHED_FEAT 1387 1388 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR]; 1389 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x])) 1390 1391 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */ 1392 1393 /* 1394 * Each translation unit has its own copy of sysctl_sched_features to allow 1395 * constants propagation at compile time and compiler optimization based on 1396 * features default. 1397 */ 1398 #define SCHED_FEAT(name, enabled) \ 1399 (1UL << __SCHED_FEAT_##name) * enabled | 1400 static const_debug __maybe_unused unsigned int sysctl_sched_features = 1401 #include "features.h" 1402 0; 1403 #undef SCHED_FEAT 1404 1405 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) 1406 1407 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */ 1408 1409 extern struct static_key_false sched_numa_balancing; 1410 extern struct static_key_false sched_schedstats; 1411 1412 static inline u64 global_rt_period(void) 1413 { 1414 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; 1415 } 1416 1417 static inline u64 global_rt_runtime(void) 1418 { 1419 if (sysctl_sched_rt_runtime < 0) 1420 return RUNTIME_INF; 1421 1422 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; 1423 } 1424 1425 static inline int task_current(struct rq *rq, struct task_struct *p) 1426 { 1427 return rq->curr == p; 1428 } 1429 1430 static inline int task_running(struct rq *rq, struct task_struct *p) 1431 { 1432 #ifdef CONFIG_SMP 1433 return p->on_cpu; 1434 #else 1435 return task_current(rq, p); 1436 #endif 1437 } 1438 1439 static inline int task_on_rq_queued(struct task_struct *p) 1440 { 1441 return p->on_rq == TASK_ON_RQ_QUEUED; 1442 } 1443 1444 static inline int task_on_rq_migrating(struct task_struct *p) 1445 { 1446 return p->on_rq == TASK_ON_RQ_MIGRATING; 1447 } 1448 1449 /* 1450 * wake flags 1451 */ 1452 #define WF_SYNC 0x01 /* Waker goes to sleep after wakeup */ 1453 #define WF_FORK 0x02 /* Child wakeup after fork */ 1454 #define WF_MIGRATED 0x4 /* Internal use, task got migrated */ 1455 1456 /* 1457 * To aid in avoiding the subversion of "niceness" due to uneven distribution 1458 * of tasks with abnormal "nice" values across CPUs the contribution that 1459 * each task makes to its run queue's load is weighted according to its 1460 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a 1461 * scaled version of the new time slice allocation that they receive on time 1462 * slice expiry etc. 1463 */ 1464 1465 #define WEIGHT_IDLEPRIO 3 1466 #define WMULT_IDLEPRIO 1431655765 1467 1468 extern const int sched_prio_to_weight[40]; 1469 extern const u32 sched_prio_to_wmult[40]; 1470 1471 /* 1472 * {de,en}queue flags: 1473 * 1474 * DEQUEUE_SLEEP - task is no longer runnable 1475 * ENQUEUE_WAKEUP - task just became runnable 1476 * 1477 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks 1478 * are in a known state which allows modification. Such pairs 1479 * should preserve as much state as possible. 1480 * 1481 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location 1482 * in the runqueue. 1483 * 1484 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified) 1485 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline) 1486 * ENQUEUE_MIGRATED - the task was migrated during wakeup 1487 * 1488 */ 1489 1490 #define DEQUEUE_SLEEP 0x01 1491 #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */ 1492 #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */ 1493 #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */ 1494 1495 #define ENQUEUE_WAKEUP 0x01 1496 #define ENQUEUE_RESTORE 0x02 1497 #define ENQUEUE_MOVE 0x04 1498 #define ENQUEUE_NOCLOCK 0x08 1499 1500 #define ENQUEUE_HEAD 0x10 1501 #define ENQUEUE_REPLENISH 0x20 1502 #ifdef CONFIG_SMP 1503 #define ENQUEUE_MIGRATED 0x40 1504 #else 1505 #define ENQUEUE_MIGRATED 0x00 1506 #endif 1507 1508 #define RETRY_TASK ((void *)-1UL) 1509 1510 struct sched_class { 1511 const struct sched_class *next; 1512 1513 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags); 1514 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags); 1515 void (*yield_task) (struct rq *rq); 1516 bool (*yield_to_task)(struct rq *rq, struct task_struct *p, bool preempt); 1517 1518 void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags); 1519 1520 /* 1521 * It is the responsibility of the pick_next_task() method that will 1522 * return the next task to call put_prev_task() on the @prev task or 1523 * something equivalent. 1524 * 1525 * May return RETRY_TASK when it finds a higher prio class has runnable 1526 * tasks. 1527 */ 1528 struct task_struct * (*pick_next_task)(struct rq *rq, 1529 struct task_struct *prev, 1530 struct rq_flags *rf); 1531 void (*put_prev_task)(struct rq *rq, struct task_struct *p); 1532 1533 #ifdef CONFIG_SMP 1534 int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags); 1535 void (*migrate_task_rq)(struct task_struct *p, int new_cpu); 1536 1537 void (*task_woken)(struct rq *this_rq, struct task_struct *task); 1538 1539 void (*set_cpus_allowed)(struct task_struct *p, 1540 const struct cpumask *newmask); 1541 1542 void (*rq_online)(struct rq *rq); 1543 void (*rq_offline)(struct rq *rq); 1544 #endif 1545 1546 void (*set_curr_task)(struct rq *rq); 1547 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued); 1548 void (*task_fork)(struct task_struct *p); 1549 void (*task_dead)(struct task_struct *p); 1550 1551 /* 1552 * The switched_from() call is allowed to drop rq->lock, therefore we 1553 * cannot assume the switched_from/switched_to pair is serliazed by 1554 * rq->lock. They are however serialized by p->pi_lock. 1555 */ 1556 void (*switched_from)(struct rq *this_rq, struct task_struct *task); 1557 void (*switched_to) (struct rq *this_rq, struct task_struct *task); 1558 void (*prio_changed) (struct rq *this_rq, struct task_struct *task, 1559 int oldprio); 1560 1561 unsigned int (*get_rr_interval)(struct rq *rq, 1562 struct task_struct *task); 1563 1564 void (*update_curr)(struct rq *rq); 1565 1566 #define TASK_SET_GROUP 0 1567 #define TASK_MOVE_GROUP 1 1568 1569 #ifdef CONFIG_FAIR_GROUP_SCHED 1570 void (*task_change_group)(struct task_struct *p, int type); 1571 #endif 1572 }; 1573 1574 static inline void put_prev_task(struct rq *rq, struct task_struct *prev) 1575 { 1576 prev->sched_class->put_prev_task(rq, prev); 1577 } 1578 1579 static inline void set_curr_task(struct rq *rq, struct task_struct *curr) 1580 { 1581 curr->sched_class->set_curr_task(rq); 1582 } 1583 1584 #ifdef CONFIG_SMP 1585 #define sched_class_highest (&stop_sched_class) 1586 #else 1587 #define sched_class_highest (&dl_sched_class) 1588 #endif 1589 #define for_each_class(class) \ 1590 for (class = sched_class_highest; class; class = class->next) 1591 1592 extern const struct sched_class stop_sched_class; 1593 extern const struct sched_class dl_sched_class; 1594 extern const struct sched_class rt_sched_class; 1595 extern const struct sched_class fair_sched_class; 1596 extern const struct sched_class idle_sched_class; 1597 1598 1599 #ifdef CONFIG_SMP 1600 1601 extern void update_group_capacity(struct sched_domain *sd, int cpu); 1602 1603 extern void trigger_load_balance(struct rq *rq); 1604 1605 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask); 1606 1607 #endif 1608 1609 #ifdef CONFIG_CPU_IDLE 1610 static inline void idle_set_state(struct rq *rq, 1611 struct cpuidle_state *idle_state) 1612 { 1613 rq->idle_state = idle_state; 1614 } 1615 1616 static inline struct cpuidle_state *idle_get_state(struct rq *rq) 1617 { 1618 SCHED_WARN_ON(!rcu_read_lock_held()); 1619 1620 return rq->idle_state; 1621 } 1622 #else 1623 static inline void idle_set_state(struct rq *rq, 1624 struct cpuidle_state *idle_state) 1625 { 1626 } 1627 1628 static inline struct cpuidle_state *idle_get_state(struct rq *rq) 1629 { 1630 return NULL; 1631 } 1632 #endif 1633 1634 extern void schedule_idle(void); 1635 1636 extern void sysrq_sched_debug_show(void); 1637 extern void sched_init_granularity(void); 1638 extern void update_max_interval(void); 1639 1640 extern void init_sched_dl_class(void); 1641 extern void init_sched_rt_class(void); 1642 extern void init_sched_fair_class(void); 1643 1644 extern void reweight_task(struct task_struct *p, int prio); 1645 1646 extern void resched_curr(struct rq *rq); 1647 extern void resched_cpu(int cpu); 1648 1649 extern struct rt_bandwidth def_rt_bandwidth; 1650 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime); 1651 1652 extern struct dl_bandwidth def_dl_bandwidth; 1653 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime); 1654 extern void init_dl_task_timer(struct sched_dl_entity *dl_se); 1655 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se); 1656 extern void init_dl_rq_bw_ratio(struct dl_rq *dl_rq); 1657 1658 #define BW_SHIFT 20 1659 #define BW_UNIT (1 << BW_SHIFT) 1660 #define RATIO_SHIFT 8 1661 unsigned long to_ratio(u64 period, u64 runtime); 1662 1663 extern void init_entity_runnable_average(struct sched_entity *se); 1664 extern void post_init_entity_util_avg(struct sched_entity *se); 1665 1666 #ifdef CONFIG_NO_HZ_FULL 1667 extern bool sched_can_stop_tick(struct rq *rq); 1668 extern int __init sched_tick_offload_init(void); 1669 1670 /* 1671 * Tick may be needed by tasks in the runqueue depending on their policy and 1672 * requirements. If tick is needed, lets send the target an IPI to kick it out of 1673 * nohz mode if necessary. 1674 */ 1675 static inline void sched_update_tick_dependency(struct rq *rq) 1676 { 1677 int cpu; 1678 1679 if (!tick_nohz_full_enabled()) 1680 return; 1681 1682 cpu = cpu_of(rq); 1683 1684 if (!tick_nohz_full_cpu(cpu)) 1685 return; 1686 1687 if (sched_can_stop_tick(rq)) 1688 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED); 1689 else 1690 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED); 1691 } 1692 #else 1693 static inline int sched_tick_offload_init(void) { return 0; } 1694 static inline void sched_update_tick_dependency(struct rq *rq) { } 1695 #endif 1696 1697 static inline void add_nr_running(struct rq *rq, unsigned count) 1698 { 1699 unsigned prev_nr = rq->nr_running; 1700 1701 rq->nr_running = prev_nr + count; 1702 1703 if (prev_nr < 2 && rq->nr_running >= 2) { 1704 #ifdef CONFIG_SMP 1705 if (!READ_ONCE(rq->rd->overload)) 1706 WRITE_ONCE(rq->rd->overload, 1); 1707 #endif 1708 } 1709 1710 sched_update_tick_dependency(rq); 1711 } 1712 1713 static inline void sub_nr_running(struct rq *rq, unsigned count) 1714 { 1715 rq->nr_running -= count; 1716 /* Check if we still need preemption */ 1717 sched_update_tick_dependency(rq); 1718 } 1719 1720 extern void update_rq_clock(struct rq *rq); 1721 1722 extern void activate_task(struct rq *rq, struct task_struct *p, int flags); 1723 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags); 1724 1725 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags); 1726 1727 extern const_debug unsigned int sysctl_sched_nr_migrate; 1728 extern const_debug unsigned int sysctl_sched_migration_cost; 1729 1730 #ifdef CONFIG_SCHED_HRTICK 1731 1732 /* 1733 * Use hrtick when: 1734 * - enabled by features 1735 * - hrtimer is actually high res 1736 */ 1737 static inline int hrtick_enabled(struct rq *rq) 1738 { 1739 if (!sched_feat(HRTICK)) 1740 return 0; 1741 if (!cpu_active(cpu_of(rq))) 1742 return 0; 1743 return hrtimer_is_hres_active(&rq->hrtick_timer); 1744 } 1745 1746 void hrtick_start(struct rq *rq, u64 delay); 1747 1748 #else 1749 1750 static inline int hrtick_enabled(struct rq *rq) 1751 { 1752 return 0; 1753 } 1754 1755 #endif /* CONFIG_SCHED_HRTICK */ 1756 1757 #ifndef arch_scale_freq_capacity 1758 static __always_inline 1759 unsigned long arch_scale_freq_capacity(int cpu) 1760 { 1761 return SCHED_CAPACITY_SCALE; 1762 } 1763 #endif 1764 1765 #ifdef CONFIG_SMP 1766 #ifndef arch_scale_cpu_capacity 1767 static __always_inline 1768 unsigned long arch_scale_cpu_capacity(struct sched_domain *sd, int cpu) 1769 { 1770 if (sd && (sd->flags & SD_SHARE_CPUCAPACITY) && (sd->span_weight > 1)) 1771 return sd->smt_gain / sd->span_weight; 1772 1773 return SCHED_CAPACITY_SCALE; 1774 } 1775 #endif 1776 #else 1777 #ifndef arch_scale_cpu_capacity 1778 static __always_inline 1779 unsigned long arch_scale_cpu_capacity(void __always_unused *sd, int cpu) 1780 { 1781 return SCHED_CAPACITY_SCALE; 1782 } 1783 #endif 1784 #endif 1785 1786 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf) 1787 __acquires(rq->lock); 1788 1789 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf) 1790 __acquires(p->pi_lock) 1791 __acquires(rq->lock); 1792 1793 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf) 1794 __releases(rq->lock) 1795 { 1796 rq_unpin_lock(rq, rf); 1797 raw_spin_unlock(&rq->lock); 1798 } 1799 1800 static inline void 1801 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf) 1802 __releases(rq->lock) 1803 __releases(p->pi_lock) 1804 { 1805 rq_unpin_lock(rq, rf); 1806 raw_spin_unlock(&rq->lock); 1807 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags); 1808 } 1809 1810 static inline void 1811 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf) 1812 __acquires(rq->lock) 1813 { 1814 raw_spin_lock_irqsave(&rq->lock, rf->flags); 1815 rq_pin_lock(rq, rf); 1816 } 1817 1818 static inline void 1819 rq_lock_irq(struct rq *rq, struct rq_flags *rf) 1820 __acquires(rq->lock) 1821 { 1822 raw_spin_lock_irq(&rq->lock); 1823 rq_pin_lock(rq, rf); 1824 } 1825 1826 static inline void 1827 rq_lock(struct rq *rq, struct rq_flags *rf) 1828 __acquires(rq->lock) 1829 { 1830 raw_spin_lock(&rq->lock); 1831 rq_pin_lock(rq, rf); 1832 } 1833 1834 static inline void 1835 rq_relock(struct rq *rq, struct rq_flags *rf) 1836 __acquires(rq->lock) 1837 { 1838 raw_spin_lock(&rq->lock); 1839 rq_repin_lock(rq, rf); 1840 } 1841 1842 static inline void 1843 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf) 1844 __releases(rq->lock) 1845 { 1846 rq_unpin_lock(rq, rf); 1847 raw_spin_unlock_irqrestore(&rq->lock, rf->flags); 1848 } 1849 1850 static inline void 1851 rq_unlock_irq(struct rq *rq, struct rq_flags *rf) 1852 __releases(rq->lock) 1853 { 1854 rq_unpin_lock(rq, rf); 1855 raw_spin_unlock_irq(&rq->lock); 1856 } 1857 1858 static inline void 1859 rq_unlock(struct rq *rq, struct rq_flags *rf) 1860 __releases(rq->lock) 1861 { 1862 rq_unpin_lock(rq, rf); 1863 raw_spin_unlock(&rq->lock); 1864 } 1865 1866 #ifdef CONFIG_SMP 1867 #ifdef CONFIG_PREEMPT 1868 1869 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2); 1870 1871 /* 1872 * fair double_lock_balance: Safely acquires both rq->locks in a fair 1873 * way at the expense of forcing extra atomic operations in all 1874 * invocations. This assures that the double_lock is acquired using the 1875 * same underlying policy as the spinlock_t on this architecture, which 1876 * reduces latency compared to the unfair variant below. However, it 1877 * also adds more overhead and therefore may reduce throughput. 1878 */ 1879 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) 1880 __releases(this_rq->lock) 1881 __acquires(busiest->lock) 1882 __acquires(this_rq->lock) 1883 { 1884 raw_spin_unlock(&this_rq->lock); 1885 double_rq_lock(this_rq, busiest); 1886 1887 return 1; 1888 } 1889 1890 #else 1891 /* 1892 * Unfair double_lock_balance: Optimizes throughput at the expense of 1893 * latency by eliminating extra atomic operations when the locks are 1894 * already in proper order on entry. This favors lower CPU-ids and will 1895 * grant the double lock to lower CPUs over higher ids under contention, 1896 * regardless of entry order into the function. 1897 */ 1898 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) 1899 __releases(this_rq->lock) 1900 __acquires(busiest->lock) 1901 __acquires(this_rq->lock) 1902 { 1903 int ret = 0; 1904 1905 if (unlikely(!raw_spin_trylock(&busiest->lock))) { 1906 if (busiest < this_rq) { 1907 raw_spin_unlock(&this_rq->lock); 1908 raw_spin_lock(&busiest->lock); 1909 raw_spin_lock_nested(&this_rq->lock, 1910 SINGLE_DEPTH_NESTING); 1911 ret = 1; 1912 } else 1913 raw_spin_lock_nested(&busiest->lock, 1914 SINGLE_DEPTH_NESTING); 1915 } 1916 return ret; 1917 } 1918 1919 #endif /* CONFIG_PREEMPT */ 1920 1921 /* 1922 * double_lock_balance - lock the busiest runqueue, this_rq is locked already. 1923 */ 1924 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest) 1925 { 1926 if (unlikely(!irqs_disabled())) { 1927 /* printk() doesn't work well under rq->lock */ 1928 raw_spin_unlock(&this_rq->lock); 1929 BUG_ON(1); 1930 } 1931 1932 return _double_lock_balance(this_rq, busiest); 1933 } 1934 1935 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) 1936 __releases(busiest->lock) 1937 { 1938 raw_spin_unlock(&busiest->lock); 1939 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); 1940 } 1941 1942 static inline void double_lock(spinlock_t *l1, spinlock_t *l2) 1943 { 1944 if (l1 > l2) 1945 swap(l1, l2); 1946 1947 spin_lock(l1); 1948 spin_lock_nested(l2, SINGLE_DEPTH_NESTING); 1949 } 1950 1951 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2) 1952 { 1953 if (l1 > l2) 1954 swap(l1, l2); 1955 1956 spin_lock_irq(l1); 1957 spin_lock_nested(l2, SINGLE_DEPTH_NESTING); 1958 } 1959 1960 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2) 1961 { 1962 if (l1 > l2) 1963 swap(l1, l2); 1964 1965 raw_spin_lock(l1); 1966 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING); 1967 } 1968 1969 /* 1970 * double_rq_lock - safely lock two runqueues 1971 * 1972 * Note this does not disable interrupts like task_rq_lock, 1973 * you need to do so manually before calling. 1974 */ 1975 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) 1976 __acquires(rq1->lock) 1977 __acquires(rq2->lock) 1978 { 1979 BUG_ON(!irqs_disabled()); 1980 if (rq1 == rq2) { 1981 raw_spin_lock(&rq1->lock); 1982 __acquire(rq2->lock); /* Fake it out ;) */ 1983 } else { 1984 if (rq1 < rq2) { 1985 raw_spin_lock(&rq1->lock); 1986 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); 1987 } else { 1988 raw_spin_lock(&rq2->lock); 1989 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); 1990 } 1991 } 1992 } 1993 1994 /* 1995 * double_rq_unlock - safely unlock two runqueues 1996 * 1997 * Note this does not restore interrupts like task_rq_unlock, 1998 * you need to do so manually after calling. 1999 */ 2000 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) 2001 __releases(rq1->lock) 2002 __releases(rq2->lock) 2003 { 2004 raw_spin_unlock(&rq1->lock); 2005 if (rq1 != rq2) 2006 raw_spin_unlock(&rq2->lock); 2007 else 2008 __release(rq2->lock); 2009 } 2010 2011 extern void set_rq_online (struct rq *rq); 2012 extern void set_rq_offline(struct rq *rq); 2013 extern bool sched_smp_initialized; 2014 2015 #else /* CONFIG_SMP */ 2016 2017 /* 2018 * double_rq_lock - safely lock two runqueues 2019 * 2020 * Note this does not disable interrupts like task_rq_lock, 2021 * you need to do so manually before calling. 2022 */ 2023 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) 2024 __acquires(rq1->lock) 2025 __acquires(rq2->lock) 2026 { 2027 BUG_ON(!irqs_disabled()); 2028 BUG_ON(rq1 != rq2); 2029 raw_spin_lock(&rq1->lock); 2030 __acquire(rq2->lock); /* Fake it out ;) */ 2031 } 2032 2033 /* 2034 * double_rq_unlock - safely unlock two runqueues 2035 * 2036 * Note this does not restore interrupts like task_rq_unlock, 2037 * you need to do so manually after calling. 2038 */ 2039 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) 2040 __releases(rq1->lock) 2041 __releases(rq2->lock) 2042 { 2043 BUG_ON(rq1 != rq2); 2044 raw_spin_unlock(&rq1->lock); 2045 __release(rq2->lock); 2046 } 2047 2048 #endif 2049 2050 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq); 2051 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq); 2052 2053 #ifdef CONFIG_SCHED_DEBUG 2054 extern bool sched_debug_enabled; 2055 2056 extern void print_cfs_stats(struct seq_file *m, int cpu); 2057 extern void print_rt_stats(struct seq_file *m, int cpu); 2058 extern void print_dl_stats(struct seq_file *m, int cpu); 2059 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq); 2060 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq); 2061 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq); 2062 #ifdef CONFIG_NUMA_BALANCING 2063 extern void 2064 show_numa_stats(struct task_struct *p, struct seq_file *m); 2065 extern void 2066 print_numa_stats(struct seq_file *m, int node, unsigned long tsf, 2067 unsigned long tpf, unsigned long gsf, unsigned long gpf); 2068 #endif /* CONFIG_NUMA_BALANCING */ 2069 #endif /* CONFIG_SCHED_DEBUG */ 2070 2071 extern void init_cfs_rq(struct cfs_rq *cfs_rq); 2072 extern void init_rt_rq(struct rt_rq *rt_rq); 2073 extern void init_dl_rq(struct dl_rq *dl_rq); 2074 2075 extern void cfs_bandwidth_usage_inc(void); 2076 extern void cfs_bandwidth_usage_dec(void); 2077 2078 #ifdef CONFIG_NO_HZ_COMMON 2079 #define NOHZ_BALANCE_KICK_BIT 0 2080 #define NOHZ_STATS_KICK_BIT 1 2081 2082 #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT) 2083 #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT) 2084 2085 #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK) 2086 2087 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags) 2088 2089 extern void nohz_balance_exit_idle(struct rq *rq); 2090 #else 2091 static inline void nohz_balance_exit_idle(struct rq *rq) { } 2092 #endif 2093 2094 2095 #ifdef CONFIG_SMP 2096 static inline 2097 void __dl_update(struct dl_bw *dl_b, s64 bw) 2098 { 2099 struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw); 2100 int i; 2101 2102 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(), 2103 "sched RCU must be held"); 2104 for_each_cpu_and(i, rd->span, cpu_active_mask) { 2105 struct rq *rq = cpu_rq(i); 2106 2107 rq->dl.extra_bw += bw; 2108 } 2109 } 2110 #else 2111 static inline 2112 void __dl_update(struct dl_bw *dl_b, s64 bw) 2113 { 2114 struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw); 2115 2116 dl->extra_bw += bw; 2117 } 2118 #endif 2119 2120 2121 #ifdef CONFIG_IRQ_TIME_ACCOUNTING 2122 struct irqtime { 2123 u64 total; 2124 u64 tick_delta; 2125 u64 irq_start_time; 2126 struct u64_stats_sync sync; 2127 }; 2128 2129 DECLARE_PER_CPU(struct irqtime, cpu_irqtime); 2130 2131 /* 2132 * Returns the irqtime minus the softirq time computed by ksoftirqd. 2133 * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime 2134 * and never move forward. 2135 */ 2136 static inline u64 irq_time_read(int cpu) 2137 { 2138 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu); 2139 unsigned int seq; 2140 u64 total; 2141 2142 do { 2143 seq = __u64_stats_fetch_begin(&irqtime->sync); 2144 total = irqtime->total; 2145 } while (__u64_stats_fetch_retry(&irqtime->sync, seq)); 2146 2147 return total; 2148 } 2149 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ 2150 2151 #ifdef CONFIG_CPU_FREQ 2152 DECLARE_PER_CPU(struct update_util_data *, cpufreq_update_util_data); 2153 2154 /** 2155 * cpufreq_update_util - Take a note about CPU utilization changes. 2156 * @rq: Runqueue to carry out the update for. 2157 * @flags: Update reason flags. 2158 * 2159 * This function is called by the scheduler on the CPU whose utilization is 2160 * being updated. 2161 * 2162 * It can only be called from RCU-sched read-side critical sections. 2163 * 2164 * The way cpufreq is currently arranged requires it to evaluate the CPU 2165 * performance state (frequency/voltage) on a regular basis to prevent it from 2166 * being stuck in a completely inadequate performance level for too long. 2167 * That is not guaranteed to happen if the updates are only triggered from CFS 2168 * and DL, though, because they may not be coming in if only RT tasks are 2169 * active all the time (or there are RT tasks only). 2170 * 2171 * As a workaround for that issue, this function is called periodically by the 2172 * RT sched class to trigger extra cpufreq updates to prevent it from stalling, 2173 * but that really is a band-aid. Going forward it should be replaced with 2174 * solutions targeted more specifically at RT tasks. 2175 */ 2176 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) 2177 { 2178 struct update_util_data *data; 2179 2180 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data, 2181 cpu_of(rq))); 2182 if (data) 2183 data->func(data, rq_clock(rq), flags); 2184 } 2185 #else 2186 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {} 2187 #endif /* CONFIG_CPU_FREQ */ 2188 2189 #ifdef arch_scale_freq_capacity 2190 # ifndef arch_scale_freq_invariant 2191 # define arch_scale_freq_invariant() true 2192 # endif 2193 #else 2194 # define arch_scale_freq_invariant() false 2195 #endif 2196 2197 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL 2198 static inline unsigned long cpu_bw_dl(struct rq *rq) 2199 { 2200 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT; 2201 } 2202 2203 static inline unsigned long cpu_util_dl(struct rq *rq) 2204 { 2205 return READ_ONCE(rq->avg_dl.util_avg); 2206 } 2207 2208 static inline unsigned long cpu_util_cfs(struct rq *rq) 2209 { 2210 unsigned long util = READ_ONCE(rq->cfs.avg.util_avg); 2211 2212 if (sched_feat(UTIL_EST)) { 2213 util = max_t(unsigned long, util, 2214 READ_ONCE(rq->cfs.avg.util_est.enqueued)); 2215 } 2216 2217 return util; 2218 } 2219 2220 static inline unsigned long cpu_util_rt(struct rq *rq) 2221 { 2222 return READ_ONCE(rq->avg_rt.util_avg); 2223 } 2224 #endif 2225 2226 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ 2227 static inline unsigned long cpu_util_irq(struct rq *rq) 2228 { 2229 return rq->avg_irq.util_avg; 2230 } 2231 2232 static inline 2233 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max) 2234 { 2235 util *= (max - irq); 2236 util /= max; 2237 2238 return util; 2239 2240 } 2241 #else 2242 static inline unsigned long cpu_util_irq(struct rq *rq) 2243 { 2244 return 0; 2245 } 2246 2247 static inline 2248 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max) 2249 { 2250 return util; 2251 } 2252 #endif 2253