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/smt.h> 27 #include <linux/sched/stat.h> 28 #include <linux/sched/sysctl.h> 29 #include <linux/sched/task.h> 30 #include <linux/sched/task_stack.h> 31 #include <linux/sched/topology.h> 32 #include <linux/sched/user.h> 33 #include <linux/sched/wake_q.h> 34 #include <linux/sched/xacct.h> 35 36 #include <uapi/linux/sched/types.h> 37 38 #include <linux/binfmts.h> 39 #include <linux/blkdev.h> 40 #include <linux/compat.h> 41 #include <linux/context_tracking.h> 42 #include <linux/cpufreq.h> 43 #include <linux/cpuidle.h> 44 #include <linux/cpuset.h> 45 #include <linux/ctype.h> 46 #include <linux/debugfs.h> 47 #include <linux/delayacct.h> 48 #include <linux/energy_model.h> 49 #include <linux/init_task.h> 50 #include <linux/kprobes.h> 51 #include <linux/kthread.h> 52 #include <linux/membarrier.h> 53 #include <linux/migrate.h> 54 #include <linux/mmu_context.h> 55 #include <linux/nmi.h> 56 #include <linux/proc_fs.h> 57 #include <linux/prefetch.h> 58 #include <linux/profile.h> 59 #include <linux/psi.h> 60 #include <linux/rcupdate_wait.h> 61 #include <linux/security.h> 62 #include <linux/stop_machine.h> 63 #include <linux/suspend.h> 64 #include <linux/swait.h> 65 #include <linux/syscalls.h> 66 #include <linux/task_work.h> 67 #include <linux/tsacct_kern.h> 68 69 #include <asm/tlb.h> 70 71 #ifdef CONFIG_PARAVIRT 72 # include <asm/paravirt.h> 73 #endif 74 75 #include "cpupri.h" 76 #include "cpudeadline.h" 77 78 #ifdef CONFIG_SCHED_DEBUG 79 # define SCHED_WARN_ON(x) WARN_ONCE(x, #x) 80 #else 81 # define SCHED_WARN_ON(x) ({ (void)(x), 0; }) 82 #endif 83 84 struct rq; 85 struct cpuidle_state; 86 87 /* task_struct::on_rq states: */ 88 #define TASK_ON_RQ_QUEUED 1 89 #define TASK_ON_RQ_MIGRATING 2 90 91 extern __read_mostly int scheduler_running; 92 93 extern unsigned long calc_load_update; 94 extern atomic_long_t calc_load_tasks; 95 96 extern void calc_global_load_tick(struct rq *this_rq); 97 extern long calc_load_fold_active(struct rq *this_rq, long adjust); 98 99 /* 100 * Helpers for converting nanosecond timing to jiffy resolution 101 */ 102 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) 103 104 /* 105 * Increase resolution of nice-level calculations for 64-bit architectures. 106 * The extra resolution improves shares distribution and load balancing of 107 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup 108 * hierarchies, especially on larger systems. This is not a user-visible change 109 * and does not change the user-interface for setting shares/weights. 110 * 111 * We increase resolution only if we have enough bits to allow this increased 112 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit 113 * are pretty high and the returns do not justify the increased costs. 114 * 115 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to 116 * increase coverage and consistency always enable it on 64-bit platforms. 117 */ 118 #ifdef CONFIG_64BIT 119 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT) 120 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT) 121 # define scale_load_down(w) \ 122 ({ \ 123 unsigned long __w = (w); \ 124 if (__w) \ 125 __w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \ 126 __w; \ 127 }) 128 #else 129 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT) 130 # define scale_load(w) (w) 131 # define scale_load_down(w) (w) 132 #endif 133 134 /* 135 * Task weight (visible to users) and its load (invisible to users) have 136 * independent resolution, but they should be well calibrated. We use 137 * scale_load() and scale_load_down(w) to convert between them. The 138 * following must be true: 139 * 140 * scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD 141 * 142 */ 143 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT) 144 145 /* 146 * Single value that decides SCHED_DEADLINE internal math precision. 147 * 10 -> just above 1us 148 * 9 -> just above 0.5us 149 */ 150 #define DL_SCALE 10 151 152 /* 153 * Single value that denotes runtime == period, ie unlimited time. 154 */ 155 #define RUNTIME_INF ((u64)~0ULL) 156 157 static inline int idle_policy(int policy) 158 { 159 return policy == SCHED_IDLE; 160 } 161 static inline int fair_policy(int policy) 162 { 163 return policy == SCHED_NORMAL || policy == SCHED_BATCH; 164 } 165 166 static inline int rt_policy(int policy) 167 { 168 return policy == SCHED_FIFO || policy == SCHED_RR; 169 } 170 171 static inline int dl_policy(int policy) 172 { 173 return policy == SCHED_DEADLINE; 174 } 175 static inline bool valid_policy(int policy) 176 { 177 return idle_policy(policy) || fair_policy(policy) || 178 rt_policy(policy) || dl_policy(policy); 179 } 180 181 static inline int task_has_idle_policy(struct task_struct *p) 182 { 183 return idle_policy(p->policy); 184 } 185 186 static inline int task_has_rt_policy(struct task_struct *p) 187 { 188 return rt_policy(p->policy); 189 } 190 191 static inline int task_has_dl_policy(struct task_struct *p) 192 { 193 return dl_policy(p->policy); 194 } 195 196 #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT) 197 198 static inline void update_avg(u64 *avg, u64 sample) 199 { 200 s64 diff = sample - *avg; 201 *avg += diff / 8; 202 } 203 204 /* 205 * !! For sched_setattr_nocheck() (kernel) only !! 206 * 207 * This is actually gross. :( 208 * 209 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE 210 * tasks, but still be able to sleep. We need this on platforms that cannot 211 * atomically change clock frequency. Remove once fast switching will be 212 * available on such platforms. 213 * 214 * SUGOV stands for SchedUtil GOVernor. 215 */ 216 #define SCHED_FLAG_SUGOV 0x10000000 217 218 static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se) 219 { 220 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL 221 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV); 222 #else 223 return false; 224 #endif 225 } 226 227 /* 228 * Tells if entity @a should preempt entity @b. 229 */ 230 static inline bool 231 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b) 232 { 233 return dl_entity_is_special(a) || 234 dl_time_before(a->deadline, b->deadline); 235 } 236 237 /* 238 * This is the priority-queue data structure of the RT scheduling class: 239 */ 240 struct rt_prio_array { 241 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ 242 struct list_head queue[MAX_RT_PRIO]; 243 }; 244 245 struct rt_bandwidth { 246 /* nests inside the rq lock: */ 247 raw_spinlock_t rt_runtime_lock; 248 ktime_t rt_period; 249 u64 rt_runtime; 250 struct hrtimer rt_period_timer; 251 unsigned int rt_period_active; 252 }; 253 254 void __dl_clear_params(struct task_struct *p); 255 256 /* 257 * To keep the bandwidth of -deadline tasks and groups under control 258 * we need some place where: 259 * - store the maximum -deadline bandwidth of the system (the group); 260 * - cache the fraction of that bandwidth that is currently allocated. 261 * 262 * This is all done in the data structure below. It is similar to the 263 * one used for RT-throttling (rt_bandwidth), with the main difference 264 * that, since here we are only interested in admission control, we 265 * do not decrease any runtime while the group "executes", neither we 266 * need a timer to replenish it. 267 * 268 * With respect to SMP, the bandwidth is given on a per-CPU basis, 269 * meaning that: 270 * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU; 271 * - dl_total_bw array contains, in the i-eth element, the currently 272 * allocated bandwidth on the i-eth CPU. 273 * Moreover, groups consume bandwidth on each CPU, while tasks only 274 * consume bandwidth on the CPU they're running on. 275 * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw 276 * that will be shown the next time the proc or cgroup controls will 277 * be red. It on its turn can be changed by writing on its own 278 * control. 279 */ 280 struct dl_bandwidth { 281 raw_spinlock_t dl_runtime_lock; 282 u64 dl_runtime; 283 u64 dl_period; 284 }; 285 286 static inline int dl_bandwidth_enabled(void) 287 { 288 return sysctl_sched_rt_runtime >= 0; 289 } 290 291 struct dl_bw { 292 raw_spinlock_t lock; 293 u64 bw; 294 u64 total_bw; 295 }; 296 297 static inline void __dl_update(struct dl_bw *dl_b, s64 bw); 298 299 static inline 300 void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus) 301 { 302 dl_b->total_bw -= tsk_bw; 303 __dl_update(dl_b, (s32)tsk_bw / cpus); 304 } 305 306 static inline 307 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus) 308 { 309 dl_b->total_bw += tsk_bw; 310 __dl_update(dl_b, -((s32)tsk_bw / cpus)); 311 } 312 313 static inline 314 bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw) 315 { 316 return dl_b->bw != -1 && 317 dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw; 318 } 319 320 extern void init_dl_bw(struct dl_bw *dl_b); 321 extern int sched_dl_global_validate(void); 322 extern void sched_dl_do_global(void); 323 extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr); 324 extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr); 325 extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr); 326 extern bool __checkparam_dl(const struct sched_attr *attr); 327 extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr); 328 extern int dl_task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed); 329 extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial); 330 extern bool dl_cpu_busy(unsigned int cpu); 331 332 #ifdef CONFIG_CGROUP_SCHED 333 334 #include <linux/cgroup.h> 335 #include <linux/psi.h> 336 337 struct cfs_rq; 338 struct rt_rq; 339 340 extern struct list_head task_groups; 341 342 struct cfs_bandwidth { 343 #ifdef CONFIG_CFS_BANDWIDTH 344 raw_spinlock_t lock; 345 ktime_t period; 346 u64 quota; 347 u64 runtime; 348 s64 hierarchical_quota; 349 350 u8 idle; 351 u8 period_active; 352 u8 slack_started; 353 struct hrtimer period_timer; 354 struct hrtimer slack_timer; 355 struct list_head throttled_cfs_rq; 356 357 /* Statistics: */ 358 int nr_periods; 359 int nr_throttled; 360 u64 throttled_time; 361 #endif 362 }; 363 364 /* Task group related information */ 365 struct task_group { 366 struct cgroup_subsys_state css; 367 368 #ifdef CONFIG_FAIR_GROUP_SCHED 369 /* schedulable entities of this group on each CPU */ 370 struct sched_entity **se; 371 /* runqueue "owned" by this group on each CPU */ 372 struct cfs_rq **cfs_rq; 373 unsigned long shares; 374 375 #ifdef CONFIG_SMP 376 /* 377 * load_avg can be heavily contended at clock tick time, so put 378 * it in its own cacheline separated from the fields above which 379 * will also be accessed at each tick. 380 */ 381 atomic_long_t load_avg ____cacheline_aligned; 382 #endif 383 #endif 384 385 #ifdef CONFIG_RT_GROUP_SCHED 386 struct sched_rt_entity **rt_se; 387 struct rt_rq **rt_rq; 388 389 struct rt_bandwidth rt_bandwidth; 390 #endif 391 392 struct rcu_head rcu; 393 struct list_head list; 394 395 struct task_group *parent; 396 struct list_head siblings; 397 struct list_head children; 398 399 #ifdef CONFIG_SCHED_AUTOGROUP 400 struct autogroup *autogroup; 401 #endif 402 403 struct cfs_bandwidth cfs_bandwidth; 404 405 #ifdef CONFIG_UCLAMP_TASK_GROUP 406 /* The two decimal precision [%] value requested from user-space */ 407 unsigned int uclamp_pct[UCLAMP_CNT]; 408 /* Clamp values requested for a task group */ 409 struct uclamp_se uclamp_req[UCLAMP_CNT]; 410 /* Effective clamp values used for a task group */ 411 struct uclamp_se uclamp[UCLAMP_CNT]; 412 #endif 413 414 }; 415 416 #ifdef CONFIG_FAIR_GROUP_SCHED 417 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD 418 419 /* 420 * A weight of 0 or 1 can cause arithmetics problems. 421 * A weight of a cfs_rq is the sum of weights of which entities 422 * are queued on this cfs_rq, so a weight of a entity should not be 423 * too large, so as the shares value of a task group. 424 * (The default weight is 1024 - so there's no practical 425 * limitation from this.) 426 */ 427 #define MIN_SHARES (1UL << 1) 428 #define MAX_SHARES (1UL << 18) 429 #endif 430 431 typedef int (*tg_visitor)(struct task_group *, void *); 432 433 extern int walk_tg_tree_from(struct task_group *from, 434 tg_visitor down, tg_visitor up, void *data); 435 436 /* 437 * Iterate the full tree, calling @down when first entering a node and @up when 438 * leaving it for the final time. 439 * 440 * Caller must hold rcu_lock or sufficient equivalent. 441 */ 442 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) 443 { 444 return walk_tg_tree_from(&root_task_group, down, up, data); 445 } 446 447 extern int tg_nop(struct task_group *tg, void *data); 448 449 extern void free_fair_sched_group(struct task_group *tg); 450 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent); 451 extern void online_fair_sched_group(struct task_group *tg); 452 extern void unregister_fair_sched_group(struct task_group *tg); 453 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, 454 struct sched_entity *se, int cpu, 455 struct sched_entity *parent); 456 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b); 457 458 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b); 459 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b); 460 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq); 461 462 extern void free_rt_sched_group(struct task_group *tg); 463 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent); 464 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, 465 struct sched_rt_entity *rt_se, int cpu, 466 struct sched_rt_entity *parent); 467 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us); 468 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us); 469 extern long sched_group_rt_runtime(struct task_group *tg); 470 extern long sched_group_rt_period(struct task_group *tg); 471 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk); 472 473 extern struct task_group *sched_create_group(struct task_group *parent); 474 extern void sched_online_group(struct task_group *tg, 475 struct task_group *parent); 476 extern void sched_destroy_group(struct task_group *tg); 477 extern void sched_offline_group(struct task_group *tg); 478 479 extern void sched_move_task(struct task_struct *tsk); 480 481 #ifdef CONFIG_FAIR_GROUP_SCHED 482 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares); 483 484 #ifdef CONFIG_SMP 485 extern void set_task_rq_fair(struct sched_entity *se, 486 struct cfs_rq *prev, struct cfs_rq *next); 487 #else /* !CONFIG_SMP */ 488 static inline void set_task_rq_fair(struct sched_entity *se, 489 struct cfs_rq *prev, struct cfs_rq *next) { } 490 #endif /* CONFIG_SMP */ 491 #endif /* CONFIG_FAIR_GROUP_SCHED */ 492 493 #else /* CONFIG_CGROUP_SCHED */ 494 495 struct cfs_bandwidth { }; 496 497 #endif /* CONFIG_CGROUP_SCHED */ 498 499 /* CFS-related fields in a runqueue */ 500 struct cfs_rq { 501 struct load_weight load; 502 unsigned int nr_running; 503 unsigned int h_nr_running; /* SCHED_{NORMAL,BATCH,IDLE} */ 504 unsigned int idle_h_nr_running; /* SCHED_IDLE */ 505 506 u64 exec_clock; 507 u64 min_vruntime; 508 #ifndef CONFIG_64BIT 509 u64 min_vruntime_copy; 510 #endif 511 512 struct rb_root_cached tasks_timeline; 513 514 /* 515 * 'curr' points to currently running entity on this cfs_rq. 516 * It is set to NULL otherwise (i.e when none are currently running). 517 */ 518 struct sched_entity *curr; 519 struct sched_entity *next; 520 struct sched_entity *last; 521 struct sched_entity *skip; 522 523 #ifdef CONFIG_SCHED_DEBUG 524 unsigned int nr_spread_over; 525 #endif 526 527 #ifdef CONFIG_SMP 528 /* 529 * CFS load tracking 530 */ 531 struct sched_avg avg; 532 #ifndef CONFIG_64BIT 533 u64 load_last_update_time_copy; 534 #endif 535 struct { 536 raw_spinlock_t lock ____cacheline_aligned; 537 int nr; 538 unsigned long load_avg; 539 unsigned long util_avg; 540 unsigned long runnable_avg; 541 } removed; 542 543 #ifdef CONFIG_FAIR_GROUP_SCHED 544 unsigned long tg_load_avg_contrib; 545 long propagate; 546 long prop_runnable_sum; 547 548 /* 549 * h_load = weight * f(tg) 550 * 551 * Where f(tg) is the recursive weight fraction assigned to 552 * this group. 553 */ 554 unsigned long h_load; 555 u64 last_h_load_update; 556 struct sched_entity *h_load_next; 557 #endif /* CONFIG_FAIR_GROUP_SCHED */ 558 #endif /* CONFIG_SMP */ 559 560 #ifdef CONFIG_FAIR_GROUP_SCHED 561 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */ 562 563 /* 564 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in 565 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities 566 * (like users, containers etc.) 567 * 568 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU. 569 * This list is used during load balance. 570 */ 571 int on_list; 572 struct list_head leaf_cfs_rq_list; 573 struct task_group *tg; /* group that "owns" this runqueue */ 574 575 #ifdef CONFIG_CFS_BANDWIDTH 576 int runtime_enabled; 577 s64 runtime_remaining; 578 579 u64 throttled_clock; 580 u64 throttled_clock_task; 581 u64 throttled_clock_task_time; 582 int throttled; 583 int throttle_count; 584 struct list_head throttled_list; 585 #endif /* CONFIG_CFS_BANDWIDTH */ 586 #endif /* CONFIG_FAIR_GROUP_SCHED */ 587 }; 588 589 static inline int rt_bandwidth_enabled(void) 590 { 591 return sysctl_sched_rt_runtime >= 0; 592 } 593 594 /* RT IPI pull logic requires IRQ_WORK */ 595 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP) 596 # define HAVE_RT_PUSH_IPI 597 #endif 598 599 /* Real-Time classes' related field in a runqueue: */ 600 struct rt_rq { 601 struct rt_prio_array active; 602 unsigned int rt_nr_running; 603 unsigned int rr_nr_running; 604 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED 605 struct { 606 int curr; /* highest queued rt task prio */ 607 #ifdef CONFIG_SMP 608 int next; /* next highest */ 609 #endif 610 } highest_prio; 611 #endif 612 #ifdef CONFIG_SMP 613 unsigned long rt_nr_migratory; 614 unsigned long rt_nr_total; 615 int overloaded; 616 struct plist_head pushable_tasks; 617 618 #endif /* CONFIG_SMP */ 619 int rt_queued; 620 621 int rt_throttled; 622 u64 rt_time; 623 u64 rt_runtime; 624 /* Nests inside the rq lock: */ 625 raw_spinlock_t rt_runtime_lock; 626 627 #ifdef CONFIG_RT_GROUP_SCHED 628 unsigned long rt_nr_boosted; 629 630 struct rq *rq; 631 struct task_group *tg; 632 #endif 633 }; 634 635 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq) 636 { 637 return rt_rq->rt_queued && rt_rq->rt_nr_running; 638 } 639 640 /* Deadline class' related fields in a runqueue */ 641 struct dl_rq { 642 /* runqueue is an rbtree, ordered by deadline */ 643 struct rb_root_cached root; 644 645 unsigned long dl_nr_running; 646 647 #ifdef CONFIG_SMP 648 /* 649 * Deadline values of the currently executing and the 650 * earliest ready task on this rq. Caching these facilitates 651 * the decision whether or not a ready but not running task 652 * should migrate somewhere else. 653 */ 654 struct { 655 u64 curr; 656 u64 next; 657 } earliest_dl; 658 659 unsigned long dl_nr_migratory; 660 int overloaded; 661 662 /* 663 * Tasks on this rq that can be pushed away. They are kept in 664 * an rb-tree, ordered by tasks' deadlines, with caching 665 * of the leftmost (earliest deadline) element. 666 */ 667 struct rb_root_cached pushable_dl_tasks_root; 668 #else 669 struct dl_bw dl_bw; 670 #endif 671 /* 672 * "Active utilization" for this runqueue: increased when a 673 * task wakes up (becomes TASK_RUNNING) and decreased when a 674 * task blocks 675 */ 676 u64 running_bw; 677 678 /* 679 * Utilization of the tasks "assigned" to this runqueue (including 680 * the tasks that are in runqueue and the tasks that executed on this 681 * CPU and blocked). Increased when a task moves to this runqueue, and 682 * decreased when the task moves away (migrates, changes scheduling 683 * policy, or terminates). 684 * This is needed to compute the "inactive utilization" for the 685 * runqueue (inactive utilization = this_bw - running_bw). 686 */ 687 u64 this_bw; 688 u64 extra_bw; 689 690 /* 691 * Inverse of the fraction of CPU utilization that can be reclaimed 692 * by the GRUB algorithm. 693 */ 694 u64 bw_ratio; 695 }; 696 697 #ifdef CONFIG_FAIR_GROUP_SCHED 698 /* An entity is a task if it doesn't "own" a runqueue */ 699 #define entity_is_task(se) (!se->my_q) 700 701 static inline void se_update_runnable(struct sched_entity *se) 702 { 703 if (!entity_is_task(se)) 704 se->runnable_weight = se->my_q->h_nr_running; 705 } 706 707 static inline long se_runnable(struct sched_entity *se) 708 { 709 if (entity_is_task(se)) 710 return !!se->on_rq; 711 else 712 return se->runnable_weight; 713 } 714 715 #else 716 #define entity_is_task(se) 1 717 718 static inline void se_update_runnable(struct sched_entity *se) {} 719 720 static inline long se_runnable(struct sched_entity *se) 721 { 722 return !!se->on_rq; 723 } 724 #endif 725 726 #ifdef CONFIG_SMP 727 /* 728 * XXX we want to get rid of these helpers and use the full load resolution. 729 */ 730 static inline long se_weight(struct sched_entity *se) 731 { 732 return scale_load_down(se->load.weight); 733 } 734 735 736 static inline bool sched_asym_prefer(int a, int b) 737 { 738 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b); 739 } 740 741 struct perf_domain { 742 struct em_perf_domain *em_pd; 743 struct perf_domain *next; 744 struct rcu_head rcu; 745 }; 746 747 /* Scheduling group status flags */ 748 #define SG_OVERLOAD 0x1 /* More than one runnable task on a CPU. */ 749 #define SG_OVERUTILIZED 0x2 /* One or more CPUs are over-utilized. */ 750 751 /* 752 * We add the notion of a root-domain which will be used to define per-domain 753 * variables. Each exclusive cpuset essentially defines an island domain by 754 * fully partitioning the member CPUs from any other cpuset. Whenever a new 755 * exclusive cpuset is created, we also create and attach a new root-domain 756 * object. 757 * 758 */ 759 struct root_domain { 760 atomic_t refcount; 761 atomic_t rto_count; 762 struct rcu_head rcu; 763 cpumask_var_t span; 764 cpumask_var_t online; 765 766 /* 767 * Indicate pullable load on at least one CPU, e.g: 768 * - More than one runnable task 769 * - Running task is misfit 770 */ 771 int overload; 772 773 /* Indicate one or more cpus over-utilized (tipping point) */ 774 int overutilized; 775 776 /* 777 * The bit corresponding to a CPU gets set here if such CPU has more 778 * than one runnable -deadline task (as it is below for RT tasks). 779 */ 780 cpumask_var_t dlo_mask; 781 atomic_t dlo_count; 782 struct dl_bw dl_bw; 783 struct cpudl cpudl; 784 785 #ifdef HAVE_RT_PUSH_IPI 786 /* 787 * For IPI pull requests, loop across the rto_mask. 788 */ 789 struct irq_work rto_push_work; 790 raw_spinlock_t rto_lock; 791 /* These are only updated and read within rto_lock */ 792 int rto_loop; 793 int rto_cpu; 794 /* These atomics are updated outside of a lock */ 795 atomic_t rto_loop_next; 796 atomic_t rto_loop_start; 797 #endif 798 /* 799 * The "RT overload" flag: it gets set if a CPU has more than 800 * one runnable RT task. 801 */ 802 cpumask_var_t rto_mask; 803 struct cpupri cpupri; 804 805 unsigned long max_cpu_capacity; 806 807 /* 808 * NULL-terminated list of performance domains intersecting with the 809 * CPUs of the rd. Protected by RCU. 810 */ 811 struct perf_domain __rcu *pd; 812 }; 813 814 extern void init_defrootdomain(void); 815 extern int sched_init_domains(const struct cpumask *cpu_map); 816 extern void rq_attach_root(struct rq *rq, struct root_domain *rd); 817 extern void sched_get_rd(struct root_domain *rd); 818 extern void sched_put_rd(struct root_domain *rd); 819 820 #ifdef HAVE_RT_PUSH_IPI 821 extern void rto_push_irq_work_func(struct irq_work *work); 822 #endif 823 #endif /* CONFIG_SMP */ 824 825 #ifdef CONFIG_UCLAMP_TASK 826 /* 827 * struct uclamp_bucket - Utilization clamp bucket 828 * @value: utilization clamp value for tasks on this clamp bucket 829 * @tasks: number of RUNNABLE tasks on this clamp bucket 830 * 831 * Keep track of how many tasks are RUNNABLE for a given utilization 832 * clamp value. 833 */ 834 struct uclamp_bucket { 835 unsigned long value : bits_per(SCHED_CAPACITY_SCALE); 836 unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE); 837 }; 838 839 /* 840 * struct uclamp_rq - rq's utilization clamp 841 * @value: currently active clamp values for a rq 842 * @bucket: utilization clamp buckets affecting a rq 843 * 844 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values. 845 * A clamp value is affecting a rq when there is at least one task RUNNABLE 846 * (or actually running) with that value. 847 * 848 * There are up to UCLAMP_CNT possible different clamp values, currently there 849 * are only two: minimum utilization and maximum utilization. 850 * 851 * All utilization clamping values are MAX aggregated, since: 852 * - for util_min: we want to run the CPU at least at the max of the minimum 853 * utilization required by its currently RUNNABLE tasks. 854 * - for util_max: we want to allow the CPU to run up to the max of the 855 * maximum utilization allowed by its currently RUNNABLE tasks. 856 * 857 * Since on each system we expect only a limited number of different 858 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track 859 * the metrics required to compute all the per-rq utilization clamp values. 860 */ 861 struct uclamp_rq { 862 unsigned int value; 863 struct uclamp_bucket bucket[UCLAMP_BUCKETS]; 864 }; 865 #endif /* CONFIG_UCLAMP_TASK */ 866 867 /* 868 * This is the main, per-CPU runqueue data structure. 869 * 870 * Locking rule: those places that want to lock multiple runqueues 871 * (such as the load balancing or the thread migration code), lock 872 * acquire operations must be ordered by ascending &runqueue. 873 */ 874 struct rq { 875 /* runqueue lock: */ 876 raw_spinlock_t lock; 877 878 /* 879 * nr_running and cpu_load should be in the same cacheline because 880 * remote CPUs use both these fields when doing load calculation. 881 */ 882 unsigned int nr_running; 883 #ifdef CONFIG_NUMA_BALANCING 884 unsigned int nr_numa_running; 885 unsigned int nr_preferred_running; 886 unsigned int numa_migrate_on; 887 #endif 888 #ifdef CONFIG_NO_HZ_COMMON 889 #ifdef CONFIG_SMP 890 unsigned long last_blocked_load_update_tick; 891 unsigned int has_blocked_load; 892 call_single_data_t nohz_csd; 893 #endif /* CONFIG_SMP */ 894 unsigned int nohz_tick_stopped; 895 atomic_t nohz_flags; 896 #endif /* CONFIG_NO_HZ_COMMON */ 897 898 #ifdef CONFIG_SMP 899 unsigned int ttwu_pending; 900 #endif 901 u64 nr_switches; 902 903 #ifdef CONFIG_UCLAMP_TASK 904 /* Utilization clamp values based on CPU's RUNNABLE tasks */ 905 struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned; 906 unsigned int uclamp_flags; 907 #define UCLAMP_FLAG_IDLE 0x01 908 #endif 909 910 struct cfs_rq cfs; 911 struct rt_rq rt; 912 struct dl_rq dl; 913 914 #ifdef CONFIG_FAIR_GROUP_SCHED 915 /* list of leaf cfs_rq on this CPU: */ 916 struct list_head leaf_cfs_rq_list; 917 struct list_head *tmp_alone_branch; 918 #endif /* CONFIG_FAIR_GROUP_SCHED */ 919 920 /* 921 * This is part of a global counter where only the total sum 922 * over all CPUs matters. A task can increase this counter on 923 * one CPU and if it got migrated afterwards it may decrease 924 * it on another CPU. Always updated under the runqueue lock: 925 */ 926 unsigned long nr_uninterruptible; 927 928 struct task_struct __rcu *curr; 929 struct task_struct *idle; 930 struct task_struct *stop; 931 unsigned long next_balance; 932 struct mm_struct *prev_mm; 933 934 unsigned int clock_update_flags; 935 u64 clock; 936 /* Ensure that all clocks are in the same cache line */ 937 u64 clock_task ____cacheline_aligned; 938 u64 clock_pelt; 939 unsigned long lost_idle_time; 940 941 atomic_t nr_iowait; 942 943 #ifdef CONFIG_MEMBARRIER 944 int membarrier_state; 945 #endif 946 947 #ifdef CONFIG_SMP 948 struct root_domain *rd; 949 struct sched_domain __rcu *sd; 950 951 unsigned long cpu_capacity; 952 unsigned long cpu_capacity_orig; 953 954 struct callback_head *balance_callback; 955 956 unsigned char nohz_idle_balance; 957 unsigned char idle_balance; 958 959 unsigned long misfit_task_load; 960 961 /* For active balancing */ 962 int active_balance; 963 int push_cpu; 964 struct cpu_stop_work active_balance_work; 965 966 /* CPU of this runqueue: */ 967 int cpu; 968 int online; 969 970 struct list_head cfs_tasks; 971 972 struct sched_avg avg_rt; 973 struct sched_avg avg_dl; 974 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ 975 struct sched_avg avg_irq; 976 #endif 977 #ifdef CONFIG_SCHED_THERMAL_PRESSURE 978 struct sched_avg avg_thermal; 979 #endif 980 u64 idle_stamp; 981 u64 avg_idle; 982 983 /* This is used to determine avg_idle's max value */ 984 u64 max_idle_balance_cost; 985 #endif /* CONFIG_SMP */ 986 987 #ifdef CONFIG_IRQ_TIME_ACCOUNTING 988 u64 prev_irq_time; 989 #endif 990 #ifdef CONFIG_PARAVIRT 991 u64 prev_steal_time; 992 #endif 993 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING 994 u64 prev_steal_time_rq; 995 #endif 996 997 /* calc_load related fields */ 998 unsigned long calc_load_update; 999 long calc_load_active; 1000 1001 #ifdef CONFIG_SCHED_HRTICK 1002 #ifdef CONFIG_SMP 1003 call_single_data_t hrtick_csd; 1004 #endif 1005 struct hrtimer hrtick_timer; 1006 #endif 1007 1008 #ifdef CONFIG_SCHEDSTATS 1009 /* latency stats */ 1010 struct sched_info rq_sched_info; 1011 unsigned long long rq_cpu_time; 1012 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ 1013 1014 /* sys_sched_yield() stats */ 1015 unsigned int yld_count; 1016 1017 /* schedule() stats */ 1018 unsigned int sched_count; 1019 unsigned int sched_goidle; 1020 1021 /* try_to_wake_up() stats */ 1022 unsigned int ttwu_count; 1023 unsigned int ttwu_local; 1024 #endif 1025 1026 #ifdef CONFIG_CPU_IDLE 1027 /* Must be inspected within a rcu lock section */ 1028 struct cpuidle_state *idle_state; 1029 #endif 1030 }; 1031 1032 #ifdef CONFIG_FAIR_GROUP_SCHED 1033 1034 /* CPU runqueue to which this cfs_rq is attached */ 1035 static inline struct rq *rq_of(struct cfs_rq *cfs_rq) 1036 { 1037 return cfs_rq->rq; 1038 } 1039 1040 #else 1041 1042 static inline struct rq *rq_of(struct cfs_rq *cfs_rq) 1043 { 1044 return container_of(cfs_rq, struct rq, cfs); 1045 } 1046 #endif 1047 1048 static inline int cpu_of(struct rq *rq) 1049 { 1050 #ifdef CONFIG_SMP 1051 return rq->cpu; 1052 #else 1053 return 0; 1054 #endif 1055 } 1056 1057 1058 #ifdef CONFIG_SCHED_SMT 1059 extern void __update_idle_core(struct rq *rq); 1060 1061 static inline void update_idle_core(struct rq *rq) 1062 { 1063 if (static_branch_unlikely(&sched_smt_present)) 1064 __update_idle_core(rq); 1065 } 1066 1067 #else 1068 static inline void update_idle_core(struct rq *rq) { } 1069 #endif 1070 1071 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); 1072 1073 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) 1074 #define this_rq() this_cpu_ptr(&runqueues) 1075 #define task_rq(p) cpu_rq(task_cpu(p)) 1076 #define cpu_curr(cpu) (cpu_rq(cpu)->curr) 1077 #define raw_rq() raw_cpu_ptr(&runqueues) 1078 1079 extern void update_rq_clock(struct rq *rq); 1080 1081 static inline u64 __rq_clock_broken(struct rq *rq) 1082 { 1083 return READ_ONCE(rq->clock); 1084 } 1085 1086 /* 1087 * rq::clock_update_flags bits 1088 * 1089 * %RQCF_REQ_SKIP - will request skipping of clock update on the next 1090 * call to __schedule(). This is an optimisation to avoid 1091 * neighbouring rq clock updates. 1092 * 1093 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is 1094 * in effect and calls to update_rq_clock() are being ignored. 1095 * 1096 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been 1097 * made to update_rq_clock() since the last time rq::lock was pinned. 1098 * 1099 * If inside of __schedule(), clock_update_flags will have been 1100 * shifted left (a left shift is a cheap operation for the fast path 1101 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use, 1102 * 1103 * if (rq-clock_update_flags >= RQCF_UPDATED) 1104 * 1105 * to check if %RQCF_UPADTED is set. It'll never be shifted more than 1106 * one position though, because the next rq_unpin_lock() will shift it 1107 * back. 1108 */ 1109 #define RQCF_REQ_SKIP 0x01 1110 #define RQCF_ACT_SKIP 0x02 1111 #define RQCF_UPDATED 0x04 1112 1113 static inline void assert_clock_updated(struct rq *rq) 1114 { 1115 /* 1116 * The only reason for not seeing a clock update since the 1117 * last rq_pin_lock() is if we're currently skipping updates. 1118 */ 1119 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP); 1120 } 1121 1122 static inline u64 rq_clock(struct rq *rq) 1123 { 1124 lockdep_assert_held(&rq->lock); 1125 assert_clock_updated(rq); 1126 1127 return rq->clock; 1128 } 1129 1130 static inline u64 rq_clock_task(struct rq *rq) 1131 { 1132 lockdep_assert_held(&rq->lock); 1133 assert_clock_updated(rq); 1134 1135 return rq->clock_task; 1136 } 1137 1138 /** 1139 * By default the decay is the default pelt decay period. 1140 * The decay shift can change the decay period in 1141 * multiples of 32. 1142 * Decay shift Decay period(ms) 1143 * 0 32 1144 * 1 64 1145 * 2 128 1146 * 3 256 1147 * 4 512 1148 */ 1149 extern int sched_thermal_decay_shift; 1150 1151 static inline u64 rq_clock_thermal(struct rq *rq) 1152 { 1153 return rq_clock_task(rq) >> sched_thermal_decay_shift; 1154 } 1155 1156 static inline void rq_clock_skip_update(struct rq *rq) 1157 { 1158 lockdep_assert_held(&rq->lock); 1159 rq->clock_update_flags |= RQCF_REQ_SKIP; 1160 } 1161 1162 /* 1163 * See rt task throttling, which is the only time a skip 1164 * request is cancelled. 1165 */ 1166 static inline void rq_clock_cancel_skipupdate(struct rq *rq) 1167 { 1168 lockdep_assert_held(&rq->lock); 1169 rq->clock_update_flags &= ~RQCF_REQ_SKIP; 1170 } 1171 1172 struct rq_flags { 1173 unsigned long flags; 1174 struct pin_cookie cookie; 1175 #ifdef CONFIG_SCHED_DEBUG 1176 /* 1177 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the 1178 * current pin context is stashed here in case it needs to be 1179 * restored in rq_repin_lock(). 1180 */ 1181 unsigned int clock_update_flags; 1182 #endif 1183 }; 1184 1185 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf) 1186 { 1187 rf->cookie = lockdep_pin_lock(&rq->lock); 1188 1189 #ifdef CONFIG_SCHED_DEBUG 1190 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP); 1191 rf->clock_update_flags = 0; 1192 #endif 1193 } 1194 1195 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf) 1196 { 1197 #ifdef CONFIG_SCHED_DEBUG 1198 if (rq->clock_update_flags > RQCF_ACT_SKIP) 1199 rf->clock_update_flags = RQCF_UPDATED; 1200 #endif 1201 1202 lockdep_unpin_lock(&rq->lock, rf->cookie); 1203 } 1204 1205 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf) 1206 { 1207 lockdep_repin_lock(&rq->lock, rf->cookie); 1208 1209 #ifdef CONFIG_SCHED_DEBUG 1210 /* 1211 * Restore the value we stashed in @rf for this pin context. 1212 */ 1213 rq->clock_update_flags |= rf->clock_update_flags; 1214 #endif 1215 } 1216 1217 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf) 1218 __acquires(rq->lock); 1219 1220 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf) 1221 __acquires(p->pi_lock) 1222 __acquires(rq->lock); 1223 1224 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf) 1225 __releases(rq->lock) 1226 { 1227 rq_unpin_lock(rq, rf); 1228 raw_spin_unlock(&rq->lock); 1229 } 1230 1231 static inline void 1232 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf) 1233 __releases(rq->lock) 1234 __releases(p->pi_lock) 1235 { 1236 rq_unpin_lock(rq, rf); 1237 raw_spin_unlock(&rq->lock); 1238 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags); 1239 } 1240 1241 static inline void 1242 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf) 1243 __acquires(rq->lock) 1244 { 1245 raw_spin_lock_irqsave(&rq->lock, rf->flags); 1246 rq_pin_lock(rq, rf); 1247 } 1248 1249 static inline void 1250 rq_lock_irq(struct rq *rq, struct rq_flags *rf) 1251 __acquires(rq->lock) 1252 { 1253 raw_spin_lock_irq(&rq->lock); 1254 rq_pin_lock(rq, rf); 1255 } 1256 1257 static inline void 1258 rq_lock(struct rq *rq, struct rq_flags *rf) 1259 __acquires(rq->lock) 1260 { 1261 raw_spin_lock(&rq->lock); 1262 rq_pin_lock(rq, rf); 1263 } 1264 1265 static inline void 1266 rq_relock(struct rq *rq, struct rq_flags *rf) 1267 __acquires(rq->lock) 1268 { 1269 raw_spin_lock(&rq->lock); 1270 rq_repin_lock(rq, rf); 1271 } 1272 1273 static inline void 1274 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf) 1275 __releases(rq->lock) 1276 { 1277 rq_unpin_lock(rq, rf); 1278 raw_spin_unlock_irqrestore(&rq->lock, rf->flags); 1279 } 1280 1281 static inline void 1282 rq_unlock_irq(struct rq *rq, struct rq_flags *rf) 1283 __releases(rq->lock) 1284 { 1285 rq_unpin_lock(rq, rf); 1286 raw_spin_unlock_irq(&rq->lock); 1287 } 1288 1289 static inline void 1290 rq_unlock(struct rq *rq, struct rq_flags *rf) 1291 __releases(rq->lock) 1292 { 1293 rq_unpin_lock(rq, rf); 1294 raw_spin_unlock(&rq->lock); 1295 } 1296 1297 static inline struct rq * 1298 this_rq_lock_irq(struct rq_flags *rf) 1299 __acquires(rq->lock) 1300 { 1301 struct rq *rq; 1302 1303 local_irq_disable(); 1304 rq = this_rq(); 1305 rq_lock(rq, rf); 1306 return rq; 1307 } 1308 1309 #ifdef CONFIG_NUMA 1310 enum numa_topology_type { 1311 NUMA_DIRECT, 1312 NUMA_GLUELESS_MESH, 1313 NUMA_BACKPLANE, 1314 }; 1315 extern enum numa_topology_type sched_numa_topology_type; 1316 extern int sched_max_numa_distance; 1317 extern bool find_numa_distance(int distance); 1318 extern void sched_init_numa(void); 1319 extern void sched_domains_numa_masks_set(unsigned int cpu); 1320 extern void sched_domains_numa_masks_clear(unsigned int cpu); 1321 extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu); 1322 #else 1323 static inline void sched_init_numa(void) { } 1324 static inline void sched_domains_numa_masks_set(unsigned int cpu) { } 1325 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { } 1326 static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu) 1327 { 1328 return nr_cpu_ids; 1329 } 1330 #endif 1331 1332 #ifdef CONFIG_NUMA_BALANCING 1333 /* The regions in numa_faults array from task_struct */ 1334 enum numa_faults_stats { 1335 NUMA_MEM = 0, 1336 NUMA_CPU, 1337 NUMA_MEMBUF, 1338 NUMA_CPUBUF 1339 }; 1340 extern void sched_setnuma(struct task_struct *p, int node); 1341 extern int migrate_task_to(struct task_struct *p, int cpu); 1342 extern int migrate_swap(struct task_struct *p, struct task_struct *t, 1343 int cpu, int scpu); 1344 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p); 1345 #else 1346 static inline void 1347 init_numa_balancing(unsigned long clone_flags, struct task_struct *p) 1348 { 1349 } 1350 #endif /* CONFIG_NUMA_BALANCING */ 1351 1352 #ifdef CONFIG_SMP 1353 1354 static inline void 1355 queue_balance_callback(struct rq *rq, 1356 struct callback_head *head, 1357 void (*func)(struct rq *rq)) 1358 { 1359 lockdep_assert_held(&rq->lock); 1360 1361 if (unlikely(head->next)) 1362 return; 1363 1364 head->func = (void (*)(struct callback_head *))func; 1365 head->next = rq->balance_callback; 1366 rq->balance_callback = head; 1367 } 1368 1369 #define rcu_dereference_check_sched_domain(p) \ 1370 rcu_dereference_check((p), \ 1371 lockdep_is_held(&sched_domains_mutex)) 1372 1373 /* 1374 * The domain tree (rq->sd) is protected by RCU's quiescent state transition. 1375 * See destroy_sched_domains: call_rcu for details. 1376 * 1377 * The domain tree of any CPU may only be accessed from within 1378 * preempt-disabled sections. 1379 */ 1380 #define for_each_domain(cpu, __sd) \ 1381 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \ 1382 __sd; __sd = __sd->parent) 1383 1384 /** 1385 * highest_flag_domain - Return highest sched_domain containing flag. 1386 * @cpu: The CPU whose highest level of sched domain is to 1387 * be returned. 1388 * @flag: The flag to check for the highest sched_domain 1389 * for the given CPU. 1390 * 1391 * Returns the highest sched_domain of a CPU which contains the given flag. 1392 */ 1393 static inline struct sched_domain *highest_flag_domain(int cpu, int flag) 1394 { 1395 struct sched_domain *sd, *hsd = NULL; 1396 1397 for_each_domain(cpu, sd) { 1398 if (!(sd->flags & flag)) 1399 break; 1400 hsd = sd; 1401 } 1402 1403 return hsd; 1404 } 1405 1406 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag) 1407 { 1408 struct sched_domain *sd; 1409 1410 for_each_domain(cpu, sd) { 1411 if (sd->flags & flag) 1412 break; 1413 } 1414 1415 return sd; 1416 } 1417 1418 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc); 1419 DECLARE_PER_CPU(int, sd_llc_size); 1420 DECLARE_PER_CPU(int, sd_llc_id); 1421 DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared); 1422 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa); 1423 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing); 1424 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity); 1425 extern struct static_key_false sched_asym_cpucapacity; 1426 1427 struct sched_group_capacity { 1428 atomic_t ref; 1429 /* 1430 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity 1431 * for a single CPU. 1432 */ 1433 unsigned long capacity; 1434 unsigned long min_capacity; /* Min per-CPU capacity in group */ 1435 unsigned long max_capacity; /* Max per-CPU capacity in group */ 1436 unsigned long next_update; 1437 int imbalance; /* XXX unrelated to capacity but shared group state */ 1438 1439 #ifdef CONFIG_SCHED_DEBUG 1440 int id; 1441 #endif 1442 1443 unsigned long cpumask[0]; /* Balance mask */ 1444 }; 1445 1446 struct sched_group { 1447 struct sched_group *next; /* Must be a circular list */ 1448 atomic_t ref; 1449 1450 unsigned int group_weight; 1451 struct sched_group_capacity *sgc; 1452 int asym_prefer_cpu; /* CPU of highest priority in group */ 1453 1454 /* 1455 * The CPUs this group covers. 1456 * 1457 * NOTE: this field is variable length. (Allocated dynamically 1458 * by attaching extra space to the end of the structure, 1459 * depending on how many CPUs the kernel has booted up with) 1460 */ 1461 unsigned long cpumask[]; 1462 }; 1463 1464 static inline struct cpumask *sched_group_span(struct sched_group *sg) 1465 { 1466 return to_cpumask(sg->cpumask); 1467 } 1468 1469 /* 1470 * See build_balance_mask(). 1471 */ 1472 static inline struct cpumask *group_balance_mask(struct sched_group *sg) 1473 { 1474 return to_cpumask(sg->sgc->cpumask); 1475 } 1476 1477 /** 1478 * group_first_cpu - Returns the first CPU in the cpumask of a sched_group. 1479 * @group: The group whose first CPU is to be returned. 1480 */ 1481 static inline unsigned int group_first_cpu(struct sched_group *group) 1482 { 1483 return cpumask_first(sched_group_span(group)); 1484 } 1485 1486 extern int group_balance_cpu(struct sched_group *sg); 1487 1488 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) 1489 void register_sched_domain_sysctl(void); 1490 void dirty_sched_domain_sysctl(int cpu); 1491 void unregister_sched_domain_sysctl(void); 1492 #else 1493 static inline void register_sched_domain_sysctl(void) 1494 { 1495 } 1496 static inline void dirty_sched_domain_sysctl(int cpu) 1497 { 1498 } 1499 static inline void unregister_sched_domain_sysctl(void) 1500 { 1501 } 1502 #endif 1503 1504 extern void flush_smp_call_function_from_idle(void); 1505 1506 #else /* !CONFIG_SMP: */ 1507 static inline void flush_smp_call_function_from_idle(void) { } 1508 #endif 1509 1510 #include "stats.h" 1511 #include "autogroup.h" 1512 1513 #ifdef CONFIG_CGROUP_SCHED 1514 1515 /* 1516 * Return the group to which this tasks belongs. 1517 * 1518 * We cannot use task_css() and friends because the cgroup subsystem 1519 * changes that value before the cgroup_subsys::attach() method is called, 1520 * therefore we cannot pin it and might observe the wrong value. 1521 * 1522 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup 1523 * core changes this before calling sched_move_task(). 1524 * 1525 * Instead we use a 'copy' which is updated from sched_move_task() while 1526 * holding both task_struct::pi_lock and rq::lock. 1527 */ 1528 static inline struct task_group *task_group(struct task_struct *p) 1529 { 1530 return p->sched_task_group; 1531 } 1532 1533 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ 1534 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) 1535 { 1536 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED) 1537 struct task_group *tg = task_group(p); 1538 #endif 1539 1540 #ifdef CONFIG_FAIR_GROUP_SCHED 1541 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]); 1542 p->se.cfs_rq = tg->cfs_rq[cpu]; 1543 p->se.parent = tg->se[cpu]; 1544 #endif 1545 1546 #ifdef CONFIG_RT_GROUP_SCHED 1547 p->rt.rt_rq = tg->rt_rq[cpu]; 1548 p->rt.parent = tg->rt_se[cpu]; 1549 #endif 1550 } 1551 1552 #else /* CONFIG_CGROUP_SCHED */ 1553 1554 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } 1555 static inline struct task_group *task_group(struct task_struct *p) 1556 { 1557 return NULL; 1558 } 1559 1560 #endif /* CONFIG_CGROUP_SCHED */ 1561 1562 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) 1563 { 1564 set_task_rq(p, cpu); 1565 #ifdef CONFIG_SMP 1566 /* 1567 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be 1568 * successfully executed on another CPU. We must ensure that updates of 1569 * per-task data have been completed by this moment. 1570 */ 1571 smp_wmb(); 1572 #ifdef CONFIG_THREAD_INFO_IN_TASK 1573 WRITE_ONCE(p->cpu, cpu); 1574 #else 1575 WRITE_ONCE(task_thread_info(p)->cpu, cpu); 1576 #endif 1577 p->wake_cpu = cpu; 1578 #endif 1579 } 1580 1581 /* 1582 * Tunables that become constants when CONFIG_SCHED_DEBUG is off: 1583 */ 1584 #ifdef CONFIG_SCHED_DEBUG 1585 # include <linux/static_key.h> 1586 # define const_debug __read_mostly 1587 #else 1588 # define const_debug const 1589 #endif 1590 1591 #define SCHED_FEAT(name, enabled) \ 1592 __SCHED_FEAT_##name , 1593 1594 enum { 1595 #include "features.h" 1596 __SCHED_FEAT_NR, 1597 }; 1598 1599 #undef SCHED_FEAT 1600 1601 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_JUMP_LABEL) 1602 1603 /* 1604 * To support run-time toggling of sched features, all the translation units 1605 * (but core.c) reference the sysctl_sched_features defined in core.c. 1606 */ 1607 extern const_debug unsigned int sysctl_sched_features; 1608 1609 #define SCHED_FEAT(name, enabled) \ 1610 static __always_inline bool static_branch_##name(struct static_key *key) \ 1611 { \ 1612 return static_key_##enabled(key); \ 1613 } 1614 1615 #include "features.h" 1616 #undef SCHED_FEAT 1617 1618 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR]; 1619 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x])) 1620 1621 #else /* !(SCHED_DEBUG && CONFIG_JUMP_LABEL) */ 1622 1623 /* 1624 * Each translation unit has its own copy of sysctl_sched_features to allow 1625 * constants propagation at compile time and compiler optimization based on 1626 * features default. 1627 */ 1628 #define SCHED_FEAT(name, enabled) \ 1629 (1UL << __SCHED_FEAT_##name) * enabled | 1630 static const_debug __maybe_unused unsigned int sysctl_sched_features = 1631 #include "features.h" 1632 0; 1633 #undef SCHED_FEAT 1634 1635 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) 1636 1637 #endif /* SCHED_DEBUG && CONFIG_JUMP_LABEL */ 1638 1639 extern struct static_key_false sched_numa_balancing; 1640 extern struct static_key_false sched_schedstats; 1641 1642 static inline u64 global_rt_period(void) 1643 { 1644 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; 1645 } 1646 1647 static inline u64 global_rt_runtime(void) 1648 { 1649 if (sysctl_sched_rt_runtime < 0) 1650 return RUNTIME_INF; 1651 1652 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; 1653 } 1654 1655 static inline int task_current(struct rq *rq, struct task_struct *p) 1656 { 1657 return rq->curr == p; 1658 } 1659 1660 static inline int task_running(struct rq *rq, struct task_struct *p) 1661 { 1662 #ifdef CONFIG_SMP 1663 return p->on_cpu; 1664 #else 1665 return task_current(rq, p); 1666 #endif 1667 } 1668 1669 static inline int task_on_rq_queued(struct task_struct *p) 1670 { 1671 return p->on_rq == TASK_ON_RQ_QUEUED; 1672 } 1673 1674 static inline int task_on_rq_migrating(struct task_struct *p) 1675 { 1676 return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING; 1677 } 1678 1679 /* 1680 * wake flags 1681 */ 1682 #define WF_SYNC 0x01 /* Waker goes to sleep after wakeup */ 1683 #define WF_FORK 0x02 /* Child wakeup after fork */ 1684 #define WF_MIGRATED 0x04 /* Internal use, task got migrated */ 1685 #define WF_ON_CPU 0x08 /* Wakee is on_cpu */ 1686 1687 /* 1688 * To aid in avoiding the subversion of "niceness" due to uneven distribution 1689 * of tasks with abnormal "nice" values across CPUs the contribution that 1690 * each task makes to its run queue's load is weighted according to its 1691 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a 1692 * scaled version of the new time slice allocation that they receive on time 1693 * slice expiry etc. 1694 */ 1695 1696 #define WEIGHT_IDLEPRIO 3 1697 #define WMULT_IDLEPRIO 1431655765 1698 1699 extern const int sched_prio_to_weight[40]; 1700 extern const u32 sched_prio_to_wmult[40]; 1701 1702 /* 1703 * {de,en}queue flags: 1704 * 1705 * DEQUEUE_SLEEP - task is no longer runnable 1706 * ENQUEUE_WAKEUP - task just became runnable 1707 * 1708 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks 1709 * are in a known state which allows modification. Such pairs 1710 * should preserve as much state as possible. 1711 * 1712 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location 1713 * in the runqueue. 1714 * 1715 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified) 1716 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline) 1717 * ENQUEUE_MIGRATED - the task was migrated during wakeup 1718 * 1719 */ 1720 1721 #define DEQUEUE_SLEEP 0x01 1722 #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */ 1723 #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */ 1724 #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */ 1725 1726 #define ENQUEUE_WAKEUP 0x01 1727 #define ENQUEUE_RESTORE 0x02 1728 #define ENQUEUE_MOVE 0x04 1729 #define ENQUEUE_NOCLOCK 0x08 1730 1731 #define ENQUEUE_HEAD 0x10 1732 #define ENQUEUE_REPLENISH 0x20 1733 #ifdef CONFIG_SMP 1734 #define ENQUEUE_MIGRATED 0x40 1735 #else 1736 #define ENQUEUE_MIGRATED 0x00 1737 #endif 1738 1739 #define RETRY_TASK ((void *)-1UL) 1740 1741 struct sched_class { 1742 const struct sched_class *next; 1743 1744 #ifdef CONFIG_UCLAMP_TASK 1745 int uclamp_enabled; 1746 #endif 1747 1748 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags); 1749 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags); 1750 void (*yield_task) (struct rq *rq); 1751 bool (*yield_to_task)(struct rq *rq, struct task_struct *p, bool preempt); 1752 1753 void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags); 1754 1755 struct task_struct *(*pick_next_task)(struct rq *rq); 1756 1757 void (*put_prev_task)(struct rq *rq, struct task_struct *p); 1758 void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first); 1759 1760 #ifdef CONFIG_SMP 1761 int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf); 1762 int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags); 1763 void (*migrate_task_rq)(struct task_struct *p, int new_cpu); 1764 1765 void (*task_woken)(struct rq *this_rq, struct task_struct *task); 1766 1767 void (*set_cpus_allowed)(struct task_struct *p, 1768 const struct cpumask *newmask); 1769 1770 void (*rq_online)(struct rq *rq); 1771 void (*rq_offline)(struct rq *rq); 1772 #endif 1773 1774 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued); 1775 void (*task_fork)(struct task_struct *p); 1776 void (*task_dead)(struct task_struct *p); 1777 1778 /* 1779 * The switched_from() call is allowed to drop rq->lock, therefore we 1780 * cannot assume the switched_from/switched_to pair is serliazed by 1781 * rq->lock. They are however serialized by p->pi_lock. 1782 */ 1783 void (*switched_from)(struct rq *this_rq, struct task_struct *task); 1784 void (*switched_to) (struct rq *this_rq, struct task_struct *task); 1785 void (*prio_changed) (struct rq *this_rq, struct task_struct *task, 1786 int oldprio); 1787 1788 unsigned int (*get_rr_interval)(struct rq *rq, 1789 struct task_struct *task); 1790 1791 void (*update_curr)(struct rq *rq); 1792 1793 #define TASK_SET_GROUP 0 1794 #define TASK_MOVE_GROUP 1 1795 1796 #ifdef CONFIG_FAIR_GROUP_SCHED 1797 void (*task_change_group)(struct task_struct *p, int type); 1798 #endif 1799 }; 1800 1801 static inline void put_prev_task(struct rq *rq, struct task_struct *prev) 1802 { 1803 WARN_ON_ONCE(rq->curr != prev); 1804 prev->sched_class->put_prev_task(rq, prev); 1805 } 1806 1807 static inline void set_next_task(struct rq *rq, struct task_struct *next) 1808 { 1809 WARN_ON_ONCE(rq->curr != next); 1810 next->sched_class->set_next_task(rq, next, false); 1811 } 1812 1813 #ifdef CONFIG_SMP 1814 #define sched_class_highest (&stop_sched_class) 1815 #else 1816 #define sched_class_highest (&dl_sched_class) 1817 #endif 1818 1819 #define for_class_range(class, _from, _to) \ 1820 for (class = (_from); class != (_to); class = class->next) 1821 1822 #define for_each_class(class) \ 1823 for_class_range(class, sched_class_highest, NULL) 1824 1825 extern const struct sched_class stop_sched_class; 1826 extern const struct sched_class dl_sched_class; 1827 extern const struct sched_class rt_sched_class; 1828 extern const struct sched_class fair_sched_class; 1829 extern const struct sched_class idle_sched_class; 1830 1831 static inline bool sched_stop_runnable(struct rq *rq) 1832 { 1833 return rq->stop && task_on_rq_queued(rq->stop); 1834 } 1835 1836 static inline bool sched_dl_runnable(struct rq *rq) 1837 { 1838 return rq->dl.dl_nr_running > 0; 1839 } 1840 1841 static inline bool sched_rt_runnable(struct rq *rq) 1842 { 1843 return rq->rt.rt_queued > 0; 1844 } 1845 1846 static inline bool sched_fair_runnable(struct rq *rq) 1847 { 1848 return rq->cfs.nr_running > 0; 1849 } 1850 1851 extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf); 1852 extern struct task_struct *pick_next_task_idle(struct rq *rq); 1853 1854 #ifdef CONFIG_SMP 1855 1856 extern void update_group_capacity(struct sched_domain *sd, int cpu); 1857 1858 extern void trigger_load_balance(struct rq *rq); 1859 1860 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask); 1861 1862 #endif 1863 1864 #ifdef CONFIG_CPU_IDLE 1865 static inline void idle_set_state(struct rq *rq, 1866 struct cpuidle_state *idle_state) 1867 { 1868 rq->idle_state = idle_state; 1869 } 1870 1871 static inline struct cpuidle_state *idle_get_state(struct rq *rq) 1872 { 1873 SCHED_WARN_ON(!rcu_read_lock_held()); 1874 1875 return rq->idle_state; 1876 } 1877 #else 1878 static inline void idle_set_state(struct rq *rq, 1879 struct cpuidle_state *idle_state) 1880 { 1881 } 1882 1883 static inline struct cpuidle_state *idle_get_state(struct rq *rq) 1884 { 1885 return NULL; 1886 } 1887 #endif 1888 1889 extern void schedule_idle(void); 1890 1891 extern void sysrq_sched_debug_show(void); 1892 extern void sched_init_granularity(void); 1893 extern void update_max_interval(void); 1894 1895 extern void init_sched_dl_class(void); 1896 extern void init_sched_rt_class(void); 1897 extern void init_sched_fair_class(void); 1898 1899 extern void reweight_task(struct task_struct *p, int prio); 1900 1901 extern void resched_curr(struct rq *rq); 1902 extern void resched_cpu(int cpu); 1903 1904 extern struct rt_bandwidth def_rt_bandwidth; 1905 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime); 1906 1907 extern struct dl_bandwidth def_dl_bandwidth; 1908 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime); 1909 extern void init_dl_task_timer(struct sched_dl_entity *dl_se); 1910 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se); 1911 1912 #define BW_SHIFT 20 1913 #define BW_UNIT (1 << BW_SHIFT) 1914 #define RATIO_SHIFT 8 1915 #define MAX_BW_BITS (64 - BW_SHIFT) 1916 #define MAX_BW ((1ULL << MAX_BW_BITS) - 1) 1917 unsigned long to_ratio(u64 period, u64 runtime); 1918 1919 extern void init_entity_runnable_average(struct sched_entity *se); 1920 extern void post_init_entity_util_avg(struct task_struct *p); 1921 1922 #ifdef CONFIG_NO_HZ_FULL 1923 extern bool sched_can_stop_tick(struct rq *rq); 1924 extern int __init sched_tick_offload_init(void); 1925 1926 /* 1927 * Tick may be needed by tasks in the runqueue depending on their policy and 1928 * requirements. If tick is needed, lets send the target an IPI to kick it out of 1929 * nohz mode if necessary. 1930 */ 1931 static inline void sched_update_tick_dependency(struct rq *rq) 1932 { 1933 int cpu; 1934 1935 if (!tick_nohz_full_enabled()) 1936 return; 1937 1938 cpu = cpu_of(rq); 1939 1940 if (!tick_nohz_full_cpu(cpu)) 1941 return; 1942 1943 if (sched_can_stop_tick(rq)) 1944 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED); 1945 else 1946 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED); 1947 } 1948 #else 1949 static inline int sched_tick_offload_init(void) { return 0; } 1950 static inline void sched_update_tick_dependency(struct rq *rq) { } 1951 #endif 1952 1953 static inline void add_nr_running(struct rq *rq, unsigned count) 1954 { 1955 unsigned prev_nr = rq->nr_running; 1956 1957 rq->nr_running = prev_nr + count; 1958 1959 #ifdef CONFIG_SMP 1960 if (prev_nr < 2 && rq->nr_running >= 2) { 1961 if (!READ_ONCE(rq->rd->overload)) 1962 WRITE_ONCE(rq->rd->overload, 1); 1963 } 1964 #endif 1965 1966 sched_update_tick_dependency(rq); 1967 } 1968 1969 static inline void sub_nr_running(struct rq *rq, unsigned count) 1970 { 1971 rq->nr_running -= count; 1972 /* Check if we still need preemption */ 1973 sched_update_tick_dependency(rq); 1974 } 1975 1976 extern void activate_task(struct rq *rq, struct task_struct *p, int flags); 1977 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags); 1978 1979 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags); 1980 1981 extern const_debug unsigned int sysctl_sched_nr_migrate; 1982 extern const_debug unsigned int sysctl_sched_migration_cost; 1983 1984 #ifdef CONFIG_SCHED_HRTICK 1985 1986 /* 1987 * Use hrtick when: 1988 * - enabled by features 1989 * - hrtimer is actually high res 1990 */ 1991 static inline int hrtick_enabled(struct rq *rq) 1992 { 1993 if (!sched_feat(HRTICK)) 1994 return 0; 1995 if (!cpu_active(cpu_of(rq))) 1996 return 0; 1997 return hrtimer_is_hres_active(&rq->hrtick_timer); 1998 } 1999 2000 void hrtick_start(struct rq *rq, u64 delay); 2001 2002 #else 2003 2004 static inline int hrtick_enabled(struct rq *rq) 2005 { 2006 return 0; 2007 } 2008 2009 #endif /* CONFIG_SCHED_HRTICK */ 2010 2011 #ifndef arch_scale_freq_tick 2012 static __always_inline 2013 void arch_scale_freq_tick(void) 2014 { 2015 } 2016 #endif 2017 2018 #ifndef arch_scale_freq_capacity 2019 static __always_inline 2020 unsigned long arch_scale_freq_capacity(int cpu) 2021 { 2022 return SCHED_CAPACITY_SCALE; 2023 } 2024 #endif 2025 2026 #ifdef CONFIG_SMP 2027 #ifdef CONFIG_PREEMPTION 2028 2029 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2); 2030 2031 /* 2032 * fair double_lock_balance: Safely acquires both rq->locks in a fair 2033 * way at the expense of forcing extra atomic operations in all 2034 * invocations. This assures that the double_lock is acquired using the 2035 * same underlying policy as the spinlock_t on this architecture, which 2036 * reduces latency compared to the unfair variant below. However, it 2037 * also adds more overhead and therefore may reduce throughput. 2038 */ 2039 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) 2040 __releases(this_rq->lock) 2041 __acquires(busiest->lock) 2042 __acquires(this_rq->lock) 2043 { 2044 raw_spin_unlock(&this_rq->lock); 2045 double_rq_lock(this_rq, busiest); 2046 2047 return 1; 2048 } 2049 2050 #else 2051 /* 2052 * Unfair double_lock_balance: Optimizes throughput at the expense of 2053 * latency by eliminating extra atomic operations when the locks are 2054 * already in proper order on entry. This favors lower CPU-ids and will 2055 * grant the double lock to lower CPUs over higher ids under contention, 2056 * regardless of entry order into the function. 2057 */ 2058 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) 2059 __releases(this_rq->lock) 2060 __acquires(busiest->lock) 2061 __acquires(this_rq->lock) 2062 { 2063 int ret = 0; 2064 2065 if (unlikely(!raw_spin_trylock(&busiest->lock))) { 2066 if (busiest < this_rq) { 2067 raw_spin_unlock(&this_rq->lock); 2068 raw_spin_lock(&busiest->lock); 2069 raw_spin_lock_nested(&this_rq->lock, 2070 SINGLE_DEPTH_NESTING); 2071 ret = 1; 2072 } else 2073 raw_spin_lock_nested(&busiest->lock, 2074 SINGLE_DEPTH_NESTING); 2075 } 2076 return ret; 2077 } 2078 2079 #endif /* CONFIG_PREEMPTION */ 2080 2081 /* 2082 * double_lock_balance - lock the busiest runqueue, this_rq is locked already. 2083 */ 2084 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest) 2085 { 2086 if (unlikely(!irqs_disabled())) { 2087 /* printk() doesn't work well under rq->lock */ 2088 raw_spin_unlock(&this_rq->lock); 2089 BUG_ON(1); 2090 } 2091 2092 return _double_lock_balance(this_rq, busiest); 2093 } 2094 2095 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) 2096 __releases(busiest->lock) 2097 { 2098 raw_spin_unlock(&busiest->lock); 2099 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); 2100 } 2101 2102 static inline void double_lock(spinlock_t *l1, spinlock_t *l2) 2103 { 2104 if (l1 > l2) 2105 swap(l1, l2); 2106 2107 spin_lock(l1); 2108 spin_lock_nested(l2, SINGLE_DEPTH_NESTING); 2109 } 2110 2111 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2) 2112 { 2113 if (l1 > l2) 2114 swap(l1, l2); 2115 2116 spin_lock_irq(l1); 2117 spin_lock_nested(l2, SINGLE_DEPTH_NESTING); 2118 } 2119 2120 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2) 2121 { 2122 if (l1 > l2) 2123 swap(l1, l2); 2124 2125 raw_spin_lock(l1); 2126 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING); 2127 } 2128 2129 /* 2130 * double_rq_lock - safely lock two runqueues 2131 * 2132 * Note this does not disable interrupts like task_rq_lock, 2133 * you need to do so manually before calling. 2134 */ 2135 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) 2136 __acquires(rq1->lock) 2137 __acquires(rq2->lock) 2138 { 2139 BUG_ON(!irqs_disabled()); 2140 if (rq1 == rq2) { 2141 raw_spin_lock(&rq1->lock); 2142 __acquire(rq2->lock); /* Fake it out ;) */ 2143 } else { 2144 if (rq1 < rq2) { 2145 raw_spin_lock(&rq1->lock); 2146 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); 2147 } else { 2148 raw_spin_lock(&rq2->lock); 2149 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); 2150 } 2151 } 2152 } 2153 2154 /* 2155 * double_rq_unlock - safely unlock two runqueues 2156 * 2157 * Note this does not restore interrupts like task_rq_unlock, 2158 * you need to do so manually after calling. 2159 */ 2160 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) 2161 __releases(rq1->lock) 2162 __releases(rq2->lock) 2163 { 2164 raw_spin_unlock(&rq1->lock); 2165 if (rq1 != rq2) 2166 raw_spin_unlock(&rq2->lock); 2167 else 2168 __release(rq2->lock); 2169 } 2170 2171 extern void set_rq_online (struct rq *rq); 2172 extern void set_rq_offline(struct rq *rq); 2173 extern bool sched_smp_initialized; 2174 2175 #else /* CONFIG_SMP */ 2176 2177 /* 2178 * double_rq_lock - safely lock two runqueues 2179 * 2180 * Note this does not disable interrupts like task_rq_lock, 2181 * you need to do so manually before calling. 2182 */ 2183 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) 2184 __acquires(rq1->lock) 2185 __acquires(rq2->lock) 2186 { 2187 BUG_ON(!irqs_disabled()); 2188 BUG_ON(rq1 != rq2); 2189 raw_spin_lock(&rq1->lock); 2190 __acquire(rq2->lock); /* Fake it out ;) */ 2191 } 2192 2193 /* 2194 * double_rq_unlock - safely unlock two runqueues 2195 * 2196 * Note this does not restore interrupts like task_rq_unlock, 2197 * you need to do so manually after calling. 2198 */ 2199 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) 2200 __releases(rq1->lock) 2201 __releases(rq2->lock) 2202 { 2203 BUG_ON(rq1 != rq2); 2204 raw_spin_unlock(&rq1->lock); 2205 __release(rq2->lock); 2206 } 2207 2208 #endif 2209 2210 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq); 2211 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq); 2212 2213 #ifdef CONFIG_SCHED_DEBUG 2214 extern bool sched_debug_enabled; 2215 2216 extern void print_cfs_stats(struct seq_file *m, int cpu); 2217 extern void print_rt_stats(struct seq_file *m, int cpu); 2218 extern void print_dl_stats(struct seq_file *m, int cpu); 2219 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq); 2220 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq); 2221 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq); 2222 #ifdef CONFIG_NUMA_BALANCING 2223 extern void 2224 show_numa_stats(struct task_struct *p, struct seq_file *m); 2225 extern void 2226 print_numa_stats(struct seq_file *m, int node, unsigned long tsf, 2227 unsigned long tpf, unsigned long gsf, unsigned long gpf); 2228 #endif /* CONFIG_NUMA_BALANCING */ 2229 #endif /* CONFIG_SCHED_DEBUG */ 2230 2231 extern void init_cfs_rq(struct cfs_rq *cfs_rq); 2232 extern void init_rt_rq(struct rt_rq *rt_rq); 2233 extern void init_dl_rq(struct dl_rq *dl_rq); 2234 2235 extern void cfs_bandwidth_usage_inc(void); 2236 extern void cfs_bandwidth_usage_dec(void); 2237 2238 #ifdef CONFIG_NO_HZ_COMMON 2239 #define NOHZ_BALANCE_KICK_BIT 0 2240 #define NOHZ_STATS_KICK_BIT 1 2241 2242 #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT) 2243 #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT) 2244 2245 #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK) 2246 2247 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags) 2248 2249 extern void nohz_balance_exit_idle(struct rq *rq); 2250 #else 2251 static inline void nohz_balance_exit_idle(struct rq *rq) { } 2252 #endif 2253 2254 2255 #ifdef CONFIG_SMP 2256 static inline 2257 void __dl_update(struct dl_bw *dl_b, s64 bw) 2258 { 2259 struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw); 2260 int i; 2261 2262 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(), 2263 "sched RCU must be held"); 2264 for_each_cpu_and(i, rd->span, cpu_active_mask) { 2265 struct rq *rq = cpu_rq(i); 2266 2267 rq->dl.extra_bw += bw; 2268 } 2269 } 2270 #else 2271 static inline 2272 void __dl_update(struct dl_bw *dl_b, s64 bw) 2273 { 2274 struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw); 2275 2276 dl->extra_bw += bw; 2277 } 2278 #endif 2279 2280 2281 #ifdef CONFIG_IRQ_TIME_ACCOUNTING 2282 struct irqtime { 2283 u64 total; 2284 u64 tick_delta; 2285 u64 irq_start_time; 2286 struct u64_stats_sync sync; 2287 }; 2288 2289 DECLARE_PER_CPU(struct irqtime, cpu_irqtime); 2290 2291 /* 2292 * Returns the irqtime minus the softirq time computed by ksoftirqd. 2293 * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime 2294 * and never move forward. 2295 */ 2296 static inline u64 irq_time_read(int cpu) 2297 { 2298 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu); 2299 unsigned int seq; 2300 u64 total; 2301 2302 do { 2303 seq = __u64_stats_fetch_begin(&irqtime->sync); 2304 total = irqtime->total; 2305 } while (__u64_stats_fetch_retry(&irqtime->sync, seq)); 2306 2307 return total; 2308 } 2309 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ 2310 2311 #ifdef CONFIG_CPU_FREQ 2312 DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data); 2313 2314 /** 2315 * cpufreq_update_util - Take a note about CPU utilization changes. 2316 * @rq: Runqueue to carry out the update for. 2317 * @flags: Update reason flags. 2318 * 2319 * This function is called by the scheduler on the CPU whose utilization is 2320 * being updated. 2321 * 2322 * It can only be called from RCU-sched read-side critical sections. 2323 * 2324 * The way cpufreq is currently arranged requires it to evaluate the CPU 2325 * performance state (frequency/voltage) on a regular basis to prevent it from 2326 * being stuck in a completely inadequate performance level for too long. 2327 * That is not guaranteed to happen if the updates are only triggered from CFS 2328 * and DL, though, because they may not be coming in if only RT tasks are 2329 * active all the time (or there are RT tasks only). 2330 * 2331 * As a workaround for that issue, this function is called periodically by the 2332 * RT sched class to trigger extra cpufreq updates to prevent it from stalling, 2333 * but that really is a band-aid. Going forward it should be replaced with 2334 * solutions targeted more specifically at RT tasks. 2335 */ 2336 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) 2337 { 2338 struct update_util_data *data; 2339 2340 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data, 2341 cpu_of(rq))); 2342 if (data) 2343 data->func(data, rq_clock(rq), flags); 2344 } 2345 #else 2346 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {} 2347 #endif /* CONFIG_CPU_FREQ */ 2348 2349 #ifdef CONFIG_UCLAMP_TASK 2350 unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id); 2351 2352 static __always_inline 2353 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util, 2354 struct task_struct *p) 2355 { 2356 unsigned long min_util = READ_ONCE(rq->uclamp[UCLAMP_MIN].value); 2357 unsigned long max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value); 2358 2359 if (p) { 2360 min_util = max(min_util, uclamp_eff_value(p, UCLAMP_MIN)); 2361 max_util = max(max_util, uclamp_eff_value(p, UCLAMP_MAX)); 2362 } 2363 2364 /* 2365 * Since CPU's {min,max}_util clamps are MAX aggregated considering 2366 * RUNNABLE tasks with _different_ clamps, we can end up with an 2367 * inversion. Fix it now when the clamps are applied. 2368 */ 2369 if (unlikely(min_util >= max_util)) 2370 return min_util; 2371 2372 return clamp(util, min_util, max_util); 2373 } 2374 #else /* CONFIG_UCLAMP_TASK */ 2375 static inline 2376 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util, 2377 struct task_struct *p) 2378 { 2379 return util; 2380 } 2381 #endif /* CONFIG_UCLAMP_TASK */ 2382 2383 #ifdef arch_scale_freq_capacity 2384 # ifndef arch_scale_freq_invariant 2385 # define arch_scale_freq_invariant() true 2386 # endif 2387 #else 2388 # define arch_scale_freq_invariant() false 2389 #endif 2390 2391 #ifdef CONFIG_SMP 2392 static inline unsigned long capacity_orig_of(int cpu) 2393 { 2394 return cpu_rq(cpu)->cpu_capacity_orig; 2395 } 2396 #endif 2397 2398 /** 2399 * enum schedutil_type - CPU utilization type 2400 * @FREQUENCY_UTIL: Utilization used to select frequency 2401 * @ENERGY_UTIL: Utilization used during energy calculation 2402 * 2403 * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time 2404 * need to be aggregated differently depending on the usage made of them. This 2405 * enum is used within schedutil_freq_util() to differentiate the types of 2406 * utilization expected by the callers, and adjust the aggregation accordingly. 2407 */ 2408 enum schedutil_type { 2409 FREQUENCY_UTIL, 2410 ENERGY_UTIL, 2411 }; 2412 2413 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL 2414 2415 unsigned long schedutil_cpu_util(int cpu, unsigned long util_cfs, 2416 unsigned long max, enum schedutil_type type, 2417 struct task_struct *p); 2418 2419 static inline unsigned long cpu_bw_dl(struct rq *rq) 2420 { 2421 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT; 2422 } 2423 2424 static inline unsigned long cpu_util_dl(struct rq *rq) 2425 { 2426 return READ_ONCE(rq->avg_dl.util_avg); 2427 } 2428 2429 static inline unsigned long cpu_util_cfs(struct rq *rq) 2430 { 2431 unsigned long util = READ_ONCE(rq->cfs.avg.util_avg); 2432 2433 if (sched_feat(UTIL_EST)) { 2434 util = max_t(unsigned long, util, 2435 READ_ONCE(rq->cfs.avg.util_est.enqueued)); 2436 } 2437 2438 return util; 2439 } 2440 2441 static inline unsigned long cpu_util_rt(struct rq *rq) 2442 { 2443 return READ_ONCE(rq->avg_rt.util_avg); 2444 } 2445 #else /* CONFIG_CPU_FREQ_GOV_SCHEDUTIL */ 2446 static inline unsigned long schedutil_cpu_util(int cpu, unsigned long util_cfs, 2447 unsigned long max, enum schedutil_type type, 2448 struct task_struct *p) 2449 { 2450 return 0; 2451 } 2452 #endif /* CONFIG_CPU_FREQ_GOV_SCHEDUTIL */ 2453 2454 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ 2455 static inline unsigned long cpu_util_irq(struct rq *rq) 2456 { 2457 return rq->avg_irq.util_avg; 2458 } 2459 2460 static inline 2461 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max) 2462 { 2463 util *= (max - irq); 2464 util /= max; 2465 2466 return util; 2467 2468 } 2469 #else 2470 static inline unsigned long cpu_util_irq(struct rq *rq) 2471 { 2472 return 0; 2473 } 2474 2475 static inline 2476 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max) 2477 { 2478 return util; 2479 } 2480 #endif 2481 2482 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL) 2483 2484 #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus))) 2485 2486 DECLARE_STATIC_KEY_FALSE(sched_energy_present); 2487 2488 static inline bool sched_energy_enabled(void) 2489 { 2490 return static_branch_unlikely(&sched_energy_present); 2491 } 2492 2493 #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */ 2494 2495 #define perf_domain_span(pd) NULL 2496 static inline bool sched_energy_enabled(void) { return false; } 2497 2498 #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */ 2499 2500 #ifdef CONFIG_MEMBARRIER 2501 /* 2502 * The scheduler provides memory barriers required by membarrier between: 2503 * - prior user-space memory accesses and store to rq->membarrier_state, 2504 * - store to rq->membarrier_state and following user-space memory accesses. 2505 * In the same way it provides those guarantees around store to rq->curr. 2506 */ 2507 static inline void membarrier_switch_mm(struct rq *rq, 2508 struct mm_struct *prev_mm, 2509 struct mm_struct *next_mm) 2510 { 2511 int membarrier_state; 2512 2513 if (prev_mm == next_mm) 2514 return; 2515 2516 membarrier_state = atomic_read(&next_mm->membarrier_state); 2517 if (READ_ONCE(rq->membarrier_state) == membarrier_state) 2518 return; 2519 2520 WRITE_ONCE(rq->membarrier_state, membarrier_state); 2521 } 2522 #else 2523 static inline void membarrier_switch_mm(struct rq *rq, 2524 struct mm_struct *prev_mm, 2525 struct mm_struct *next_mm) 2526 { 2527 } 2528 #endif 2529 2530 #ifdef CONFIG_SMP 2531 static inline bool is_per_cpu_kthread(struct task_struct *p) 2532 { 2533 if (!(p->flags & PF_KTHREAD)) 2534 return false; 2535 2536 if (p->nr_cpus_allowed != 1) 2537 return false; 2538 2539 return true; 2540 } 2541 #endif 2542 2543 void swake_up_all_locked(struct swait_queue_head *q); 2544 void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait); 2545