1 /* SPDX-License-Identifier: GPL-2.0 */ 2 /* 3 * Scheduler internal types and methods: 4 */ 5 #ifndef _KERNEL_SCHED_SCHED_H 6 #define _KERNEL_SCHED_SCHED_H 7 8 #include <linux/sched/affinity.h> 9 #include <linux/sched/autogroup.h> 10 #include <linux/sched/cpufreq.h> 11 #include <linux/sched/deadline.h> 12 #include <linux/sched.h> 13 #include <linux/sched/loadavg.h> 14 #include <linux/sched/mm.h> 15 #include <linux/sched/rseq_api.h> 16 #include <linux/sched/signal.h> 17 #include <linux/sched/smt.h> 18 #include <linux/sched/stat.h> 19 #include <linux/sched/sysctl.h> 20 #include <linux/sched/task_flags.h> 21 #include <linux/sched/task.h> 22 #include <linux/sched/topology.h> 23 24 #include <linux/atomic.h> 25 #include <linux/bitmap.h> 26 #include <linux/bug.h> 27 #include <linux/capability.h> 28 #include <linux/cgroup_api.h> 29 #include <linux/cgroup.h> 30 #include <linux/context_tracking.h> 31 #include <linux/cpufreq.h> 32 #include <linux/cpumask_api.h> 33 #include <linux/ctype.h> 34 #include <linux/file.h> 35 #include <linux/fs_api.h> 36 #include <linux/hrtimer_api.h> 37 #include <linux/interrupt.h> 38 #include <linux/irq_work.h> 39 #include <linux/jiffies.h> 40 #include <linux/kref_api.h> 41 #include <linux/kthread.h> 42 #include <linux/ktime_api.h> 43 #include <linux/lockdep_api.h> 44 #include <linux/lockdep.h> 45 #include <linux/minmax.h> 46 #include <linux/mm.h> 47 #include <linux/module.h> 48 #include <linux/mutex_api.h> 49 #include <linux/plist.h> 50 #include <linux/poll.h> 51 #include <linux/proc_fs.h> 52 #include <linux/profile.h> 53 #include <linux/psi.h> 54 #include <linux/rcupdate.h> 55 #include <linux/seq_file.h> 56 #include <linux/seqlock.h> 57 #include <linux/softirq.h> 58 #include <linux/spinlock_api.h> 59 #include <linux/static_key.h> 60 #include <linux/stop_machine.h> 61 #include <linux/syscalls_api.h> 62 #include <linux/syscalls.h> 63 #include <linux/tick.h> 64 #include <linux/topology.h> 65 #include <linux/types.h> 66 #include <linux/u64_stats_sync_api.h> 67 #include <linux/uaccess.h> 68 #include <linux/wait_api.h> 69 #include <linux/wait_bit.h> 70 #include <linux/workqueue_api.h> 71 72 #include <trace/events/power.h> 73 #include <trace/events/sched.h> 74 75 #include "../workqueue_internal.h" 76 77 #ifdef CONFIG_CGROUP_SCHED 78 #include <linux/cgroup.h> 79 #include <linux/psi.h> 80 #endif 81 82 #ifdef CONFIG_SCHED_DEBUG 83 # include <linux/static_key.h> 84 #endif 85 86 #ifdef CONFIG_PARAVIRT 87 # include <asm/paravirt.h> 88 # include <asm/paravirt_api_clock.h> 89 #endif 90 91 #include "cpupri.h" 92 #include "cpudeadline.h" 93 94 #ifdef CONFIG_SCHED_DEBUG 95 # define SCHED_WARN_ON(x) WARN_ONCE(x, #x) 96 #else 97 # define SCHED_WARN_ON(x) ({ (void)(x), 0; }) 98 #endif 99 100 struct rq; 101 struct cpuidle_state; 102 103 /* task_struct::on_rq states: */ 104 #define TASK_ON_RQ_QUEUED 1 105 #define TASK_ON_RQ_MIGRATING 2 106 107 extern __read_mostly int scheduler_running; 108 109 extern unsigned long calc_load_update; 110 extern atomic_long_t calc_load_tasks; 111 112 extern unsigned int sysctl_sched_child_runs_first; 113 114 extern void calc_global_load_tick(struct rq *this_rq); 115 extern long calc_load_fold_active(struct rq *this_rq, long adjust); 116 117 extern void call_trace_sched_update_nr_running(struct rq *rq, int count); 118 119 extern unsigned int sysctl_sched_rt_period; 120 extern int sysctl_sched_rt_runtime; 121 extern int sched_rr_timeslice; 122 123 /* 124 * Helpers for converting nanosecond timing to jiffy resolution 125 */ 126 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) 127 128 /* 129 * Increase resolution of nice-level calculations for 64-bit architectures. 130 * The extra resolution improves shares distribution and load balancing of 131 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup 132 * hierarchies, especially on larger systems. This is not a user-visible change 133 * and does not change the user-interface for setting shares/weights. 134 * 135 * We increase resolution only if we have enough bits to allow this increased 136 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit 137 * are pretty high and the returns do not justify the increased costs. 138 * 139 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to 140 * increase coverage and consistency always enable it on 64-bit platforms. 141 */ 142 #ifdef CONFIG_64BIT 143 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT) 144 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT) 145 # define scale_load_down(w) \ 146 ({ \ 147 unsigned long __w = (w); \ 148 if (__w) \ 149 __w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \ 150 __w; \ 151 }) 152 #else 153 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT) 154 # define scale_load(w) (w) 155 # define scale_load_down(w) (w) 156 #endif 157 158 /* 159 * Task weight (visible to users) and its load (invisible to users) have 160 * independent resolution, but they should be well calibrated. We use 161 * scale_load() and scale_load_down(w) to convert between them. The 162 * following must be true: 163 * 164 * scale_load(sched_prio_to_weight[NICE_TO_PRIO(0)-MAX_RT_PRIO]) == NICE_0_LOAD 165 * 166 */ 167 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT) 168 169 /* 170 * Single value that decides SCHED_DEADLINE internal math precision. 171 * 10 -> just above 1us 172 * 9 -> just above 0.5us 173 */ 174 #define DL_SCALE 10 175 176 /* 177 * Single value that denotes runtime == period, ie unlimited time. 178 */ 179 #define RUNTIME_INF ((u64)~0ULL) 180 181 static inline int idle_policy(int policy) 182 { 183 return policy == SCHED_IDLE; 184 } 185 static inline int fair_policy(int policy) 186 { 187 return policy == SCHED_NORMAL || policy == SCHED_BATCH; 188 } 189 190 static inline int rt_policy(int policy) 191 { 192 return policy == SCHED_FIFO || policy == SCHED_RR; 193 } 194 195 static inline int dl_policy(int policy) 196 { 197 return policy == SCHED_DEADLINE; 198 } 199 static inline bool valid_policy(int policy) 200 { 201 return idle_policy(policy) || fair_policy(policy) || 202 rt_policy(policy) || dl_policy(policy); 203 } 204 205 static inline int task_has_idle_policy(struct task_struct *p) 206 { 207 return idle_policy(p->policy); 208 } 209 210 static inline int task_has_rt_policy(struct task_struct *p) 211 { 212 return rt_policy(p->policy); 213 } 214 215 static inline int task_has_dl_policy(struct task_struct *p) 216 { 217 return dl_policy(p->policy); 218 } 219 220 #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT) 221 222 static inline void update_avg(u64 *avg, u64 sample) 223 { 224 s64 diff = sample - *avg; 225 *avg += diff / 8; 226 } 227 228 /* 229 * Shifting a value by an exponent greater *or equal* to the size of said value 230 * is UB; cap at size-1. 231 */ 232 #define shr_bound(val, shift) \ 233 (val >> min_t(typeof(shift), shift, BITS_PER_TYPE(typeof(val)) - 1)) 234 235 /* 236 * !! For sched_setattr_nocheck() (kernel) only !! 237 * 238 * This is actually gross. :( 239 * 240 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE 241 * tasks, but still be able to sleep. We need this on platforms that cannot 242 * atomically change clock frequency. Remove once fast switching will be 243 * available on such platforms. 244 * 245 * SUGOV stands for SchedUtil GOVernor. 246 */ 247 #define SCHED_FLAG_SUGOV 0x10000000 248 249 #define SCHED_DL_FLAGS (SCHED_FLAG_RECLAIM | SCHED_FLAG_DL_OVERRUN | SCHED_FLAG_SUGOV) 250 251 static inline bool dl_entity_is_special(const struct sched_dl_entity *dl_se) 252 { 253 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL 254 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV); 255 #else 256 return false; 257 #endif 258 } 259 260 /* 261 * Tells if entity @a should preempt entity @b. 262 */ 263 static inline bool dl_entity_preempt(const struct sched_dl_entity *a, 264 const struct sched_dl_entity *b) 265 { 266 return dl_entity_is_special(a) || 267 dl_time_before(a->deadline, b->deadline); 268 } 269 270 /* 271 * This is the priority-queue data structure of the RT scheduling class: 272 */ 273 struct rt_prio_array { 274 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ 275 struct list_head queue[MAX_RT_PRIO]; 276 }; 277 278 struct rt_bandwidth { 279 /* nests inside the rq lock: */ 280 raw_spinlock_t rt_runtime_lock; 281 ktime_t rt_period; 282 u64 rt_runtime; 283 struct hrtimer rt_period_timer; 284 unsigned int rt_period_active; 285 }; 286 287 void __dl_clear_params(struct task_struct *p); 288 289 struct dl_bandwidth { 290 raw_spinlock_t dl_runtime_lock; 291 u64 dl_runtime; 292 u64 dl_period; 293 }; 294 295 static inline int dl_bandwidth_enabled(void) 296 { 297 return sysctl_sched_rt_runtime >= 0; 298 } 299 300 /* 301 * To keep the bandwidth of -deadline tasks under control 302 * we need some place where: 303 * - store the maximum -deadline bandwidth of each cpu; 304 * - cache the fraction of bandwidth that is currently allocated in 305 * each root domain; 306 * 307 * This is all done in the data structure below. It is similar to the 308 * one used for RT-throttling (rt_bandwidth), with the main difference 309 * that, since here we are only interested in admission control, we 310 * do not decrease any runtime while the group "executes", neither we 311 * need a timer to replenish it. 312 * 313 * With respect to SMP, bandwidth is given on a per root domain basis, 314 * meaning that: 315 * - bw (< 100%) is the deadline bandwidth of each CPU; 316 * - total_bw is the currently allocated bandwidth in each root domain; 317 */ 318 struct dl_bw { 319 raw_spinlock_t lock; 320 u64 bw; 321 u64 total_bw; 322 }; 323 324 extern void init_dl_bw(struct dl_bw *dl_b); 325 extern int sched_dl_global_validate(void); 326 extern void sched_dl_do_global(void); 327 extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr); 328 extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr); 329 extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr); 330 extern bool __checkparam_dl(const struct sched_attr *attr); 331 extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr); 332 extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial); 333 extern int dl_cpu_busy(int cpu, struct task_struct *p); 334 335 #ifdef CONFIG_CGROUP_SCHED 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 u64 burst; 349 u64 runtime_snap; 350 s64 hierarchical_quota; 351 352 u8 idle; 353 u8 period_active; 354 u8 slack_started; 355 struct hrtimer period_timer; 356 struct hrtimer slack_timer; 357 struct list_head throttled_cfs_rq; 358 359 /* Statistics: */ 360 int nr_periods; 361 int nr_throttled; 362 int nr_burst; 363 u64 throttled_time; 364 u64 burst_time; 365 #endif 366 }; 367 368 /* Task group related information */ 369 struct task_group { 370 struct cgroup_subsys_state css; 371 372 #ifdef CONFIG_FAIR_GROUP_SCHED 373 /* schedulable entities of this group on each CPU */ 374 struct sched_entity **se; 375 /* runqueue "owned" by this group on each CPU */ 376 struct cfs_rq **cfs_rq; 377 unsigned long shares; 378 379 /* A positive value indicates that this is a SCHED_IDLE group. */ 380 int idle; 381 382 #ifdef CONFIG_SMP 383 /* 384 * load_avg can be heavily contended at clock tick time, so put 385 * it in its own cacheline separated from the fields above which 386 * will also be accessed at each tick. 387 */ 388 atomic_long_t load_avg ____cacheline_aligned; 389 #endif 390 #endif 391 392 #ifdef CONFIG_RT_GROUP_SCHED 393 struct sched_rt_entity **rt_se; 394 struct rt_rq **rt_rq; 395 396 struct rt_bandwidth rt_bandwidth; 397 #endif 398 399 struct rcu_head rcu; 400 struct list_head list; 401 402 struct task_group *parent; 403 struct list_head siblings; 404 struct list_head children; 405 406 #ifdef CONFIG_SCHED_AUTOGROUP 407 struct autogroup *autogroup; 408 #endif 409 410 struct cfs_bandwidth cfs_bandwidth; 411 412 #ifdef CONFIG_UCLAMP_TASK_GROUP 413 /* The two decimal precision [%] value requested from user-space */ 414 unsigned int uclamp_pct[UCLAMP_CNT]; 415 /* Clamp values requested for a task group */ 416 struct uclamp_se uclamp_req[UCLAMP_CNT]; 417 /* Effective clamp values used for a task group */ 418 struct uclamp_se uclamp[UCLAMP_CNT]; 419 #endif 420 421 }; 422 423 #ifdef CONFIG_FAIR_GROUP_SCHED 424 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD 425 426 /* 427 * A weight of 0 or 1 can cause arithmetics problems. 428 * A weight of a cfs_rq is the sum of weights of which entities 429 * are queued on this cfs_rq, so a weight of a entity should not be 430 * too large, so as the shares value of a task group. 431 * (The default weight is 1024 - so there's no practical 432 * limitation from this.) 433 */ 434 #define MIN_SHARES (1UL << 1) 435 #define MAX_SHARES (1UL << 18) 436 #endif 437 438 typedef int (*tg_visitor)(struct task_group *, void *); 439 440 extern int walk_tg_tree_from(struct task_group *from, 441 tg_visitor down, tg_visitor up, void *data); 442 443 /* 444 * Iterate the full tree, calling @down when first entering a node and @up when 445 * leaving it for the final time. 446 * 447 * Caller must hold rcu_lock or sufficient equivalent. 448 */ 449 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) 450 { 451 return walk_tg_tree_from(&root_task_group, down, up, data); 452 } 453 454 extern int tg_nop(struct task_group *tg, void *data); 455 456 extern void free_fair_sched_group(struct task_group *tg); 457 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent); 458 extern void online_fair_sched_group(struct task_group *tg); 459 extern void unregister_fair_sched_group(struct task_group *tg); 460 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, 461 struct sched_entity *se, int cpu, 462 struct sched_entity *parent); 463 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b); 464 465 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b); 466 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b); 467 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq); 468 469 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, 470 struct sched_rt_entity *rt_se, int cpu, 471 struct sched_rt_entity *parent); 472 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us); 473 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us); 474 extern long sched_group_rt_runtime(struct task_group *tg); 475 extern long sched_group_rt_period(struct task_group *tg); 476 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk); 477 478 extern struct task_group *sched_create_group(struct task_group *parent); 479 extern void sched_online_group(struct task_group *tg, 480 struct task_group *parent); 481 extern void sched_destroy_group(struct task_group *tg); 482 extern void sched_release_group(struct task_group *tg); 483 484 extern void sched_move_task(struct task_struct *tsk); 485 486 #ifdef CONFIG_FAIR_GROUP_SCHED 487 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares); 488 489 extern int sched_group_set_idle(struct task_group *tg, long idle); 490 491 #ifdef CONFIG_SMP 492 extern void set_task_rq_fair(struct sched_entity *se, 493 struct cfs_rq *prev, struct cfs_rq *next); 494 #else /* !CONFIG_SMP */ 495 static inline void set_task_rq_fair(struct sched_entity *se, 496 struct cfs_rq *prev, struct cfs_rq *next) { } 497 #endif /* CONFIG_SMP */ 498 #endif /* CONFIG_FAIR_GROUP_SCHED */ 499 500 #else /* CONFIG_CGROUP_SCHED */ 501 502 struct cfs_bandwidth { }; 503 504 #endif /* CONFIG_CGROUP_SCHED */ 505 506 extern void unregister_rt_sched_group(struct task_group *tg); 507 extern void free_rt_sched_group(struct task_group *tg); 508 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent); 509 510 /* 511 * u64_u32_load/u64_u32_store 512 * 513 * Use a copy of a u64 value to protect against data race. This is only 514 * applicable for 32-bits architectures. 515 */ 516 #ifdef CONFIG_64BIT 517 # define u64_u32_load_copy(var, copy) var 518 # define u64_u32_store_copy(var, copy, val) (var = val) 519 #else 520 # define u64_u32_load_copy(var, copy) \ 521 ({ \ 522 u64 __val, __val_copy; \ 523 do { \ 524 __val_copy = copy; \ 525 /* \ 526 * paired with u64_u32_store_copy(), ordering access \ 527 * to var and copy. \ 528 */ \ 529 smp_rmb(); \ 530 __val = var; \ 531 } while (__val != __val_copy); \ 532 __val; \ 533 }) 534 # define u64_u32_store_copy(var, copy, val) \ 535 do { \ 536 typeof(val) __val = (val); \ 537 var = __val; \ 538 /* \ 539 * paired with u64_u32_load_copy(), ordering access to var and \ 540 * copy. \ 541 */ \ 542 smp_wmb(); \ 543 copy = __val; \ 544 } while (0) 545 #endif 546 # define u64_u32_load(var) u64_u32_load_copy(var, var##_copy) 547 # define u64_u32_store(var, val) u64_u32_store_copy(var, var##_copy, val) 548 549 /* CFS-related fields in a runqueue */ 550 struct cfs_rq { 551 struct load_weight load; 552 unsigned int nr_running; 553 unsigned int h_nr_running; /* SCHED_{NORMAL,BATCH,IDLE} */ 554 unsigned int idle_nr_running; /* SCHED_IDLE */ 555 unsigned int idle_h_nr_running; /* SCHED_IDLE */ 556 557 u64 exec_clock; 558 u64 min_vruntime; 559 #ifdef CONFIG_SCHED_CORE 560 unsigned int forceidle_seq; 561 u64 min_vruntime_fi; 562 #endif 563 564 #ifndef CONFIG_64BIT 565 u64 min_vruntime_copy; 566 #endif 567 568 struct rb_root_cached tasks_timeline; 569 570 /* 571 * 'curr' points to currently running entity on this cfs_rq. 572 * It is set to NULL otherwise (i.e when none are currently running). 573 */ 574 struct sched_entity *curr; 575 struct sched_entity *next; 576 struct sched_entity *last; 577 struct sched_entity *skip; 578 579 #ifdef CONFIG_SCHED_DEBUG 580 unsigned int nr_spread_over; 581 #endif 582 583 #ifdef CONFIG_SMP 584 /* 585 * CFS load tracking 586 */ 587 struct sched_avg avg; 588 #ifndef CONFIG_64BIT 589 u64 last_update_time_copy; 590 #endif 591 struct { 592 raw_spinlock_t lock ____cacheline_aligned; 593 int nr; 594 unsigned long load_avg; 595 unsigned long util_avg; 596 unsigned long runnable_avg; 597 } removed; 598 599 #ifdef CONFIG_FAIR_GROUP_SCHED 600 unsigned long tg_load_avg_contrib; 601 long propagate; 602 long prop_runnable_sum; 603 604 /* 605 * h_load = weight * f(tg) 606 * 607 * Where f(tg) is the recursive weight fraction assigned to 608 * this group. 609 */ 610 unsigned long h_load; 611 u64 last_h_load_update; 612 struct sched_entity *h_load_next; 613 #endif /* CONFIG_FAIR_GROUP_SCHED */ 614 #endif /* CONFIG_SMP */ 615 616 #ifdef CONFIG_FAIR_GROUP_SCHED 617 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */ 618 619 /* 620 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in 621 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities 622 * (like users, containers etc.) 623 * 624 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU. 625 * This list is used during load balance. 626 */ 627 int on_list; 628 struct list_head leaf_cfs_rq_list; 629 struct task_group *tg; /* group that "owns" this runqueue */ 630 631 /* Locally cached copy of our task_group's idle value */ 632 int idle; 633 634 #ifdef CONFIG_CFS_BANDWIDTH 635 int runtime_enabled; 636 s64 runtime_remaining; 637 638 u64 throttled_pelt_idle; 639 #ifndef CONFIG_64BIT 640 u64 throttled_pelt_idle_copy; 641 #endif 642 u64 throttled_clock; 643 u64 throttled_clock_pelt; 644 u64 throttled_clock_pelt_time; 645 int throttled; 646 int throttle_count; 647 struct list_head throttled_list; 648 #ifdef CONFIG_SMP 649 struct list_head throttled_csd_list; 650 #endif 651 #endif /* CONFIG_CFS_BANDWIDTH */ 652 #endif /* CONFIG_FAIR_GROUP_SCHED */ 653 }; 654 655 static inline int rt_bandwidth_enabled(void) 656 { 657 return sysctl_sched_rt_runtime >= 0; 658 } 659 660 /* RT IPI pull logic requires IRQ_WORK */ 661 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP) 662 # define HAVE_RT_PUSH_IPI 663 #endif 664 665 /* Real-Time classes' related field in a runqueue: */ 666 struct rt_rq { 667 struct rt_prio_array active; 668 unsigned int rt_nr_running; 669 unsigned int rr_nr_running; 670 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED 671 struct { 672 int curr; /* highest queued rt task prio */ 673 #ifdef CONFIG_SMP 674 int next; /* next highest */ 675 #endif 676 } highest_prio; 677 #endif 678 #ifdef CONFIG_SMP 679 unsigned int rt_nr_migratory; 680 unsigned int rt_nr_total; 681 int overloaded; 682 struct plist_head pushable_tasks; 683 684 #endif /* CONFIG_SMP */ 685 int rt_queued; 686 687 int rt_throttled; 688 u64 rt_time; 689 u64 rt_runtime; 690 /* Nests inside the rq lock: */ 691 raw_spinlock_t rt_runtime_lock; 692 693 #ifdef CONFIG_RT_GROUP_SCHED 694 unsigned int rt_nr_boosted; 695 696 struct rq *rq; 697 struct task_group *tg; 698 #endif 699 }; 700 701 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq) 702 { 703 return rt_rq->rt_queued && rt_rq->rt_nr_running; 704 } 705 706 /* Deadline class' related fields in a runqueue */ 707 struct dl_rq { 708 /* runqueue is an rbtree, ordered by deadline */ 709 struct rb_root_cached root; 710 711 unsigned int dl_nr_running; 712 713 #ifdef CONFIG_SMP 714 /* 715 * Deadline values of the currently executing and the 716 * earliest ready task on this rq. Caching these facilitates 717 * the decision whether or not a ready but not running task 718 * should migrate somewhere else. 719 */ 720 struct { 721 u64 curr; 722 u64 next; 723 } earliest_dl; 724 725 unsigned int dl_nr_migratory; 726 int overloaded; 727 728 /* 729 * Tasks on this rq that can be pushed away. They are kept in 730 * an rb-tree, ordered by tasks' deadlines, with caching 731 * of the leftmost (earliest deadline) element. 732 */ 733 struct rb_root_cached pushable_dl_tasks_root; 734 #else 735 struct dl_bw dl_bw; 736 #endif 737 /* 738 * "Active utilization" for this runqueue: increased when a 739 * task wakes up (becomes TASK_RUNNING) and decreased when a 740 * task blocks 741 */ 742 u64 running_bw; 743 744 /* 745 * Utilization of the tasks "assigned" to this runqueue (including 746 * the tasks that are in runqueue and the tasks that executed on this 747 * CPU and blocked). Increased when a task moves to this runqueue, and 748 * decreased when the task moves away (migrates, changes scheduling 749 * policy, or terminates). 750 * This is needed to compute the "inactive utilization" for the 751 * runqueue (inactive utilization = this_bw - running_bw). 752 */ 753 u64 this_bw; 754 u64 extra_bw; 755 756 /* 757 * Inverse of the fraction of CPU utilization that can be reclaimed 758 * by the GRUB algorithm. 759 */ 760 u64 bw_ratio; 761 }; 762 763 #ifdef CONFIG_FAIR_GROUP_SCHED 764 /* An entity is a task if it doesn't "own" a runqueue */ 765 #define entity_is_task(se) (!se->my_q) 766 767 static inline void se_update_runnable(struct sched_entity *se) 768 { 769 if (!entity_is_task(se)) 770 se->runnable_weight = se->my_q->h_nr_running; 771 } 772 773 static inline long se_runnable(struct sched_entity *se) 774 { 775 if (entity_is_task(se)) 776 return !!se->on_rq; 777 else 778 return se->runnable_weight; 779 } 780 781 #else 782 #define entity_is_task(se) 1 783 784 static inline void se_update_runnable(struct sched_entity *se) {} 785 786 static inline long se_runnable(struct sched_entity *se) 787 { 788 return !!se->on_rq; 789 } 790 #endif 791 792 #ifdef CONFIG_SMP 793 /* 794 * XXX we want to get rid of these helpers and use the full load resolution. 795 */ 796 static inline long se_weight(struct sched_entity *se) 797 { 798 return scale_load_down(se->load.weight); 799 } 800 801 802 static inline bool sched_asym_prefer(int a, int b) 803 { 804 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b); 805 } 806 807 struct perf_domain { 808 struct em_perf_domain *em_pd; 809 struct perf_domain *next; 810 struct rcu_head rcu; 811 }; 812 813 /* Scheduling group status flags */ 814 #define SG_OVERLOAD 0x1 /* More than one runnable task on a CPU. */ 815 #define SG_OVERUTILIZED 0x2 /* One or more CPUs are over-utilized. */ 816 817 /* 818 * We add the notion of a root-domain which will be used to define per-domain 819 * variables. Each exclusive cpuset essentially defines an island domain by 820 * fully partitioning the member CPUs from any other cpuset. Whenever a new 821 * exclusive cpuset is created, we also create and attach a new root-domain 822 * object. 823 * 824 */ 825 struct root_domain { 826 atomic_t refcount; 827 atomic_t rto_count; 828 struct rcu_head rcu; 829 cpumask_var_t span; 830 cpumask_var_t online; 831 832 /* 833 * Indicate pullable load on at least one CPU, e.g: 834 * - More than one runnable task 835 * - Running task is misfit 836 */ 837 int overload; 838 839 /* Indicate one or more cpus over-utilized (tipping point) */ 840 int overutilized; 841 842 /* 843 * The bit corresponding to a CPU gets set here if such CPU has more 844 * than one runnable -deadline task (as it is below for RT tasks). 845 */ 846 cpumask_var_t dlo_mask; 847 atomic_t dlo_count; 848 struct dl_bw dl_bw; 849 struct cpudl cpudl; 850 851 /* 852 * Indicate whether a root_domain's dl_bw has been checked or 853 * updated. It's monotonously increasing value. 854 * 855 * Also, some corner cases, like 'wrap around' is dangerous, but given 856 * that u64 is 'big enough'. So that shouldn't be a concern. 857 */ 858 u64 visit_gen; 859 860 #ifdef HAVE_RT_PUSH_IPI 861 /* 862 * For IPI pull requests, loop across the rto_mask. 863 */ 864 struct irq_work rto_push_work; 865 raw_spinlock_t rto_lock; 866 /* These are only updated and read within rto_lock */ 867 int rto_loop; 868 int rto_cpu; 869 /* These atomics are updated outside of a lock */ 870 atomic_t rto_loop_next; 871 atomic_t rto_loop_start; 872 #endif 873 /* 874 * The "RT overload" flag: it gets set if a CPU has more than 875 * one runnable RT task. 876 */ 877 cpumask_var_t rto_mask; 878 struct cpupri cpupri; 879 880 unsigned long max_cpu_capacity; 881 882 /* 883 * NULL-terminated list of performance domains intersecting with the 884 * CPUs of the rd. Protected by RCU. 885 */ 886 struct perf_domain __rcu *pd; 887 }; 888 889 extern void init_defrootdomain(void); 890 extern int sched_init_domains(const struct cpumask *cpu_map); 891 extern void rq_attach_root(struct rq *rq, struct root_domain *rd); 892 extern void sched_get_rd(struct root_domain *rd); 893 extern void sched_put_rd(struct root_domain *rd); 894 895 #ifdef HAVE_RT_PUSH_IPI 896 extern void rto_push_irq_work_func(struct irq_work *work); 897 #endif 898 #endif /* CONFIG_SMP */ 899 900 #ifdef CONFIG_UCLAMP_TASK 901 /* 902 * struct uclamp_bucket - Utilization clamp bucket 903 * @value: utilization clamp value for tasks on this clamp bucket 904 * @tasks: number of RUNNABLE tasks on this clamp bucket 905 * 906 * Keep track of how many tasks are RUNNABLE for a given utilization 907 * clamp value. 908 */ 909 struct uclamp_bucket { 910 unsigned long value : bits_per(SCHED_CAPACITY_SCALE); 911 unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE); 912 }; 913 914 /* 915 * struct uclamp_rq - rq's utilization clamp 916 * @value: currently active clamp values for a rq 917 * @bucket: utilization clamp buckets affecting a rq 918 * 919 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values. 920 * A clamp value is affecting a rq when there is at least one task RUNNABLE 921 * (or actually running) with that value. 922 * 923 * There are up to UCLAMP_CNT possible different clamp values, currently there 924 * are only two: minimum utilization and maximum utilization. 925 * 926 * All utilization clamping values are MAX aggregated, since: 927 * - for util_min: we want to run the CPU at least at the max of the minimum 928 * utilization required by its currently RUNNABLE tasks. 929 * - for util_max: we want to allow the CPU to run up to the max of the 930 * maximum utilization allowed by its currently RUNNABLE tasks. 931 * 932 * Since on each system we expect only a limited number of different 933 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track 934 * the metrics required to compute all the per-rq utilization clamp values. 935 */ 936 struct uclamp_rq { 937 unsigned int value; 938 struct uclamp_bucket bucket[UCLAMP_BUCKETS]; 939 }; 940 941 DECLARE_STATIC_KEY_FALSE(sched_uclamp_used); 942 #endif /* CONFIG_UCLAMP_TASK */ 943 944 struct rq; 945 struct balance_callback { 946 struct balance_callback *next; 947 void (*func)(struct rq *rq); 948 }; 949 950 /* 951 * This is the main, per-CPU runqueue data structure. 952 * 953 * Locking rule: those places that want to lock multiple runqueues 954 * (such as the load balancing or the thread migration code), lock 955 * acquire operations must be ordered by ascending &runqueue. 956 */ 957 struct rq { 958 /* runqueue lock: */ 959 raw_spinlock_t __lock; 960 961 /* 962 * nr_running and cpu_load should be in the same cacheline because 963 * remote CPUs use both these fields when doing load calculation. 964 */ 965 unsigned int nr_running; 966 #ifdef CONFIG_NUMA_BALANCING 967 unsigned int nr_numa_running; 968 unsigned int nr_preferred_running; 969 unsigned int numa_migrate_on; 970 #endif 971 #ifdef CONFIG_NO_HZ_COMMON 972 #ifdef CONFIG_SMP 973 unsigned long last_blocked_load_update_tick; 974 unsigned int has_blocked_load; 975 call_single_data_t nohz_csd; 976 #endif /* CONFIG_SMP */ 977 unsigned int nohz_tick_stopped; 978 atomic_t nohz_flags; 979 #endif /* CONFIG_NO_HZ_COMMON */ 980 981 #ifdef CONFIG_SMP 982 unsigned int ttwu_pending; 983 #endif 984 u64 nr_switches; 985 986 #ifdef CONFIG_UCLAMP_TASK 987 /* Utilization clamp values based on CPU's RUNNABLE tasks */ 988 struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned; 989 unsigned int uclamp_flags; 990 #define UCLAMP_FLAG_IDLE 0x01 991 #endif 992 993 struct cfs_rq cfs; 994 struct rt_rq rt; 995 struct dl_rq dl; 996 997 #ifdef CONFIG_FAIR_GROUP_SCHED 998 /* list of leaf cfs_rq on this CPU: */ 999 struct list_head leaf_cfs_rq_list; 1000 struct list_head *tmp_alone_branch; 1001 #endif /* CONFIG_FAIR_GROUP_SCHED */ 1002 1003 /* 1004 * This is part of a global counter where only the total sum 1005 * over all CPUs matters. A task can increase this counter on 1006 * one CPU and if it got migrated afterwards it may decrease 1007 * it on another CPU. Always updated under the runqueue lock: 1008 */ 1009 unsigned int nr_uninterruptible; 1010 1011 struct task_struct __rcu *curr; 1012 struct task_struct *idle; 1013 struct task_struct *stop; 1014 unsigned long next_balance; 1015 struct mm_struct *prev_mm; 1016 1017 unsigned int clock_update_flags; 1018 u64 clock; 1019 /* Ensure that all clocks are in the same cache line */ 1020 u64 clock_task ____cacheline_aligned; 1021 u64 clock_pelt; 1022 unsigned long lost_idle_time; 1023 u64 clock_pelt_idle; 1024 u64 clock_idle; 1025 #ifndef CONFIG_64BIT 1026 u64 clock_pelt_idle_copy; 1027 u64 clock_idle_copy; 1028 #endif 1029 1030 atomic_t nr_iowait; 1031 1032 #ifdef CONFIG_SCHED_DEBUG 1033 u64 last_seen_need_resched_ns; 1034 int ticks_without_resched; 1035 #endif 1036 1037 #ifdef CONFIG_MEMBARRIER 1038 int membarrier_state; 1039 #endif 1040 1041 #ifdef CONFIG_SMP 1042 struct root_domain *rd; 1043 struct sched_domain __rcu *sd; 1044 1045 unsigned long cpu_capacity; 1046 unsigned long cpu_capacity_orig; 1047 unsigned long cpu_capacity_inverted; 1048 1049 struct balance_callback *balance_callback; 1050 1051 unsigned char nohz_idle_balance; 1052 unsigned char idle_balance; 1053 1054 unsigned long misfit_task_load; 1055 1056 /* For active balancing */ 1057 int active_balance; 1058 int push_cpu; 1059 struct cpu_stop_work active_balance_work; 1060 1061 /* CPU of this runqueue: */ 1062 int cpu; 1063 int online; 1064 1065 struct list_head cfs_tasks; 1066 1067 struct sched_avg avg_rt; 1068 struct sched_avg avg_dl; 1069 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ 1070 struct sched_avg avg_irq; 1071 #endif 1072 #ifdef CONFIG_SCHED_THERMAL_PRESSURE 1073 struct sched_avg avg_thermal; 1074 #endif 1075 u64 idle_stamp; 1076 u64 avg_idle; 1077 1078 unsigned long wake_stamp; 1079 u64 wake_avg_idle; 1080 1081 /* This is used to determine avg_idle's max value */ 1082 u64 max_idle_balance_cost; 1083 1084 #ifdef CONFIG_HOTPLUG_CPU 1085 struct rcuwait hotplug_wait; 1086 #endif 1087 #endif /* CONFIG_SMP */ 1088 1089 #ifdef CONFIG_IRQ_TIME_ACCOUNTING 1090 u64 prev_irq_time; 1091 #endif 1092 #ifdef CONFIG_PARAVIRT 1093 u64 prev_steal_time; 1094 #endif 1095 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING 1096 u64 prev_steal_time_rq; 1097 #endif 1098 1099 /* calc_load related fields */ 1100 unsigned long calc_load_update; 1101 long calc_load_active; 1102 1103 #ifdef CONFIG_SCHED_HRTICK 1104 #ifdef CONFIG_SMP 1105 call_single_data_t hrtick_csd; 1106 #endif 1107 struct hrtimer hrtick_timer; 1108 ktime_t hrtick_time; 1109 #endif 1110 1111 #ifdef CONFIG_SCHEDSTATS 1112 /* latency stats */ 1113 struct sched_info rq_sched_info; 1114 unsigned long long rq_cpu_time; 1115 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ 1116 1117 /* sys_sched_yield() stats */ 1118 unsigned int yld_count; 1119 1120 /* schedule() stats */ 1121 unsigned int sched_count; 1122 unsigned int sched_goidle; 1123 1124 /* try_to_wake_up() stats */ 1125 unsigned int ttwu_count; 1126 unsigned int ttwu_local; 1127 #endif 1128 1129 #ifdef CONFIG_CPU_IDLE 1130 /* Must be inspected within a rcu lock section */ 1131 struct cpuidle_state *idle_state; 1132 #endif 1133 1134 #ifdef CONFIG_SMP 1135 unsigned int nr_pinned; 1136 #endif 1137 unsigned int push_busy; 1138 struct cpu_stop_work push_work; 1139 1140 #ifdef CONFIG_SCHED_CORE 1141 /* per rq */ 1142 struct rq *core; 1143 struct task_struct *core_pick; 1144 unsigned int core_enabled; 1145 unsigned int core_sched_seq; 1146 struct rb_root core_tree; 1147 1148 /* shared state -- careful with sched_core_cpu_deactivate() */ 1149 unsigned int core_task_seq; 1150 unsigned int core_pick_seq; 1151 unsigned long core_cookie; 1152 unsigned int core_forceidle_count; 1153 unsigned int core_forceidle_seq; 1154 unsigned int core_forceidle_occupation; 1155 u64 core_forceidle_start; 1156 #endif 1157 1158 /* Scratch cpumask to be temporarily used under rq_lock */ 1159 cpumask_var_t scratch_mask; 1160 1161 #if defined(CONFIG_CFS_BANDWIDTH) && defined(CONFIG_SMP) 1162 call_single_data_t cfsb_csd; 1163 struct list_head cfsb_csd_list; 1164 #endif 1165 }; 1166 1167 #ifdef CONFIG_FAIR_GROUP_SCHED 1168 1169 /* CPU runqueue to which this cfs_rq is attached */ 1170 static inline struct rq *rq_of(struct cfs_rq *cfs_rq) 1171 { 1172 return cfs_rq->rq; 1173 } 1174 1175 #else 1176 1177 static inline struct rq *rq_of(struct cfs_rq *cfs_rq) 1178 { 1179 return container_of(cfs_rq, struct rq, cfs); 1180 } 1181 #endif 1182 1183 static inline int cpu_of(struct rq *rq) 1184 { 1185 #ifdef CONFIG_SMP 1186 return rq->cpu; 1187 #else 1188 return 0; 1189 #endif 1190 } 1191 1192 #define MDF_PUSH 0x01 1193 1194 static inline bool is_migration_disabled(struct task_struct *p) 1195 { 1196 #ifdef CONFIG_SMP 1197 return p->migration_disabled; 1198 #else 1199 return false; 1200 #endif 1201 } 1202 1203 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); 1204 1205 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) 1206 #define this_rq() this_cpu_ptr(&runqueues) 1207 #define task_rq(p) cpu_rq(task_cpu(p)) 1208 #define cpu_curr(cpu) (cpu_rq(cpu)->curr) 1209 #define raw_rq() raw_cpu_ptr(&runqueues) 1210 1211 struct sched_group; 1212 #ifdef CONFIG_SCHED_CORE 1213 static inline struct cpumask *sched_group_span(struct sched_group *sg); 1214 1215 DECLARE_STATIC_KEY_FALSE(__sched_core_enabled); 1216 1217 static inline bool sched_core_enabled(struct rq *rq) 1218 { 1219 return static_branch_unlikely(&__sched_core_enabled) && rq->core_enabled; 1220 } 1221 1222 static inline bool sched_core_disabled(void) 1223 { 1224 return !static_branch_unlikely(&__sched_core_enabled); 1225 } 1226 1227 /* 1228 * Be careful with this function; not for general use. The return value isn't 1229 * stable unless you actually hold a relevant rq->__lock. 1230 */ 1231 static inline raw_spinlock_t *rq_lockp(struct rq *rq) 1232 { 1233 if (sched_core_enabled(rq)) 1234 return &rq->core->__lock; 1235 1236 return &rq->__lock; 1237 } 1238 1239 static inline raw_spinlock_t *__rq_lockp(struct rq *rq) 1240 { 1241 if (rq->core_enabled) 1242 return &rq->core->__lock; 1243 1244 return &rq->__lock; 1245 } 1246 1247 bool cfs_prio_less(const struct task_struct *a, const struct task_struct *b, 1248 bool fi); 1249 1250 /* 1251 * Helpers to check if the CPU's core cookie matches with the task's cookie 1252 * when core scheduling is enabled. 1253 * A special case is that the task's cookie always matches with CPU's core 1254 * cookie if the CPU is in an idle core. 1255 */ 1256 static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p) 1257 { 1258 /* Ignore cookie match if core scheduler is not enabled on the CPU. */ 1259 if (!sched_core_enabled(rq)) 1260 return true; 1261 1262 return rq->core->core_cookie == p->core_cookie; 1263 } 1264 1265 static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p) 1266 { 1267 bool idle_core = true; 1268 int cpu; 1269 1270 /* Ignore cookie match if core scheduler is not enabled on the CPU. */ 1271 if (!sched_core_enabled(rq)) 1272 return true; 1273 1274 for_each_cpu(cpu, cpu_smt_mask(cpu_of(rq))) { 1275 if (!available_idle_cpu(cpu)) { 1276 idle_core = false; 1277 break; 1278 } 1279 } 1280 1281 /* 1282 * A CPU in an idle core is always the best choice for tasks with 1283 * cookies. 1284 */ 1285 return idle_core || rq->core->core_cookie == p->core_cookie; 1286 } 1287 1288 static inline bool sched_group_cookie_match(struct rq *rq, 1289 struct task_struct *p, 1290 struct sched_group *group) 1291 { 1292 int cpu; 1293 1294 /* Ignore cookie match if core scheduler is not enabled on the CPU. */ 1295 if (!sched_core_enabled(rq)) 1296 return true; 1297 1298 for_each_cpu_and(cpu, sched_group_span(group), p->cpus_ptr) { 1299 if (sched_core_cookie_match(cpu_rq(cpu), p)) 1300 return true; 1301 } 1302 return false; 1303 } 1304 1305 static inline bool sched_core_enqueued(struct task_struct *p) 1306 { 1307 return !RB_EMPTY_NODE(&p->core_node); 1308 } 1309 1310 extern void sched_core_enqueue(struct rq *rq, struct task_struct *p); 1311 extern void sched_core_dequeue(struct rq *rq, struct task_struct *p, int flags); 1312 1313 extern void sched_core_get(void); 1314 extern void sched_core_put(void); 1315 1316 #else /* !CONFIG_SCHED_CORE */ 1317 1318 static inline bool sched_core_enabled(struct rq *rq) 1319 { 1320 return false; 1321 } 1322 1323 static inline bool sched_core_disabled(void) 1324 { 1325 return true; 1326 } 1327 1328 static inline raw_spinlock_t *rq_lockp(struct rq *rq) 1329 { 1330 return &rq->__lock; 1331 } 1332 1333 static inline raw_spinlock_t *__rq_lockp(struct rq *rq) 1334 { 1335 return &rq->__lock; 1336 } 1337 1338 static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p) 1339 { 1340 return true; 1341 } 1342 1343 static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p) 1344 { 1345 return true; 1346 } 1347 1348 static inline bool sched_group_cookie_match(struct rq *rq, 1349 struct task_struct *p, 1350 struct sched_group *group) 1351 { 1352 return true; 1353 } 1354 #endif /* CONFIG_SCHED_CORE */ 1355 1356 static inline void lockdep_assert_rq_held(struct rq *rq) 1357 { 1358 lockdep_assert_held(__rq_lockp(rq)); 1359 } 1360 1361 extern void raw_spin_rq_lock_nested(struct rq *rq, int subclass); 1362 extern bool raw_spin_rq_trylock(struct rq *rq); 1363 extern void raw_spin_rq_unlock(struct rq *rq); 1364 1365 static inline void raw_spin_rq_lock(struct rq *rq) 1366 { 1367 raw_spin_rq_lock_nested(rq, 0); 1368 } 1369 1370 static inline void raw_spin_rq_lock_irq(struct rq *rq) 1371 { 1372 local_irq_disable(); 1373 raw_spin_rq_lock(rq); 1374 } 1375 1376 static inline void raw_spin_rq_unlock_irq(struct rq *rq) 1377 { 1378 raw_spin_rq_unlock(rq); 1379 local_irq_enable(); 1380 } 1381 1382 static inline unsigned long _raw_spin_rq_lock_irqsave(struct rq *rq) 1383 { 1384 unsigned long flags; 1385 local_irq_save(flags); 1386 raw_spin_rq_lock(rq); 1387 return flags; 1388 } 1389 1390 static inline void raw_spin_rq_unlock_irqrestore(struct rq *rq, unsigned long flags) 1391 { 1392 raw_spin_rq_unlock(rq); 1393 local_irq_restore(flags); 1394 } 1395 1396 #define raw_spin_rq_lock_irqsave(rq, flags) \ 1397 do { \ 1398 flags = _raw_spin_rq_lock_irqsave(rq); \ 1399 } while (0) 1400 1401 #ifdef CONFIG_SCHED_SMT 1402 extern void __update_idle_core(struct rq *rq); 1403 1404 static inline void update_idle_core(struct rq *rq) 1405 { 1406 if (static_branch_unlikely(&sched_smt_present)) 1407 __update_idle_core(rq); 1408 } 1409 1410 #else 1411 static inline void update_idle_core(struct rq *rq) { } 1412 #endif 1413 1414 #ifdef CONFIG_FAIR_GROUP_SCHED 1415 static inline struct task_struct *task_of(struct sched_entity *se) 1416 { 1417 SCHED_WARN_ON(!entity_is_task(se)); 1418 return container_of(se, struct task_struct, se); 1419 } 1420 1421 static inline struct cfs_rq *task_cfs_rq(struct task_struct *p) 1422 { 1423 return p->se.cfs_rq; 1424 } 1425 1426 /* runqueue on which this entity is (to be) queued */ 1427 static inline struct cfs_rq *cfs_rq_of(const struct sched_entity *se) 1428 { 1429 return se->cfs_rq; 1430 } 1431 1432 /* runqueue "owned" by this group */ 1433 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp) 1434 { 1435 return grp->my_q; 1436 } 1437 1438 #else 1439 1440 #define task_of(_se) container_of(_se, struct task_struct, se) 1441 1442 static inline struct cfs_rq *task_cfs_rq(const struct task_struct *p) 1443 { 1444 return &task_rq(p)->cfs; 1445 } 1446 1447 static inline struct cfs_rq *cfs_rq_of(const struct sched_entity *se) 1448 { 1449 const struct task_struct *p = task_of(se); 1450 struct rq *rq = task_rq(p); 1451 1452 return &rq->cfs; 1453 } 1454 1455 /* runqueue "owned" by this group */ 1456 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp) 1457 { 1458 return NULL; 1459 } 1460 #endif 1461 1462 extern void update_rq_clock(struct rq *rq); 1463 1464 /* 1465 * rq::clock_update_flags bits 1466 * 1467 * %RQCF_REQ_SKIP - will request skipping of clock update on the next 1468 * call to __schedule(). This is an optimisation to avoid 1469 * neighbouring rq clock updates. 1470 * 1471 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is 1472 * in effect and calls to update_rq_clock() are being ignored. 1473 * 1474 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been 1475 * made to update_rq_clock() since the last time rq::lock was pinned. 1476 * 1477 * If inside of __schedule(), clock_update_flags will have been 1478 * shifted left (a left shift is a cheap operation for the fast path 1479 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use, 1480 * 1481 * if (rq-clock_update_flags >= RQCF_UPDATED) 1482 * 1483 * to check if %RQCF_UPDATED is set. It'll never be shifted more than 1484 * one position though, because the next rq_unpin_lock() will shift it 1485 * back. 1486 */ 1487 #define RQCF_REQ_SKIP 0x01 1488 #define RQCF_ACT_SKIP 0x02 1489 #define RQCF_UPDATED 0x04 1490 1491 static inline void assert_clock_updated(struct rq *rq) 1492 { 1493 /* 1494 * The only reason for not seeing a clock update since the 1495 * last rq_pin_lock() is if we're currently skipping updates. 1496 */ 1497 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP); 1498 } 1499 1500 static inline u64 rq_clock(struct rq *rq) 1501 { 1502 lockdep_assert_rq_held(rq); 1503 assert_clock_updated(rq); 1504 1505 return rq->clock; 1506 } 1507 1508 static inline u64 rq_clock_task(struct rq *rq) 1509 { 1510 lockdep_assert_rq_held(rq); 1511 assert_clock_updated(rq); 1512 1513 return rq->clock_task; 1514 } 1515 1516 /** 1517 * By default the decay is the default pelt decay period. 1518 * The decay shift can change the decay period in 1519 * multiples of 32. 1520 * Decay shift Decay period(ms) 1521 * 0 32 1522 * 1 64 1523 * 2 128 1524 * 3 256 1525 * 4 512 1526 */ 1527 extern int sched_thermal_decay_shift; 1528 1529 static inline u64 rq_clock_thermal(struct rq *rq) 1530 { 1531 return rq_clock_task(rq) >> sched_thermal_decay_shift; 1532 } 1533 1534 static inline void rq_clock_skip_update(struct rq *rq) 1535 { 1536 lockdep_assert_rq_held(rq); 1537 rq->clock_update_flags |= RQCF_REQ_SKIP; 1538 } 1539 1540 /* 1541 * See rt task throttling, which is the only time a skip 1542 * request is canceled. 1543 */ 1544 static inline void rq_clock_cancel_skipupdate(struct rq *rq) 1545 { 1546 lockdep_assert_rq_held(rq); 1547 rq->clock_update_flags &= ~RQCF_REQ_SKIP; 1548 } 1549 1550 struct rq_flags { 1551 unsigned long flags; 1552 struct pin_cookie cookie; 1553 #ifdef CONFIG_SCHED_DEBUG 1554 /* 1555 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the 1556 * current pin context is stashed here in case it needs to be 1557 * restored in rq_repin_lock(). 1558 */ 1559 unsigned int clock_update_flags; 1560 #endif 1561 }; 1562 1563 extern struct balance_callback balance_push_callback; 1564 1565 /* 1566 * Lockdep annotation that avoids accidental unlocks; it's like a 1567 * sticky/continuous lockdep_assert_held(). 1568 * 1569 * This avoids code that has access to 'struct rq *rq' (basically everything in 1570 * the scheduler) from accidentally unlocking the rq if they do not also have a 1571 * copy of the (on-stack) 'struct rq_flags rf'. 1572 * 1573 * Also see Documentation/locking/lockdep-design.rst. 1574 */ 1575 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf) 1576 { 1577 rf->cookie = lockdep_pin_lock(__rq_lockp(rq)); 1578 1579 #ifdef CONFIG_SCHED_DEBUG 1580 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP); 1581 rf->clock_update_flags = 0; 1582 #ifdef CONFIG_SMP 1583 SCHED_WARN_ON(rq->balance_callback && rq->balance_callback != &balance_push_callback); 1584 #endif 1585 #endif 1586 } 1587 1588 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf) 1589 { 1590 #ifdef CONFIG_SCHED_DEBUG 1591 if (rq->clock_update_flags > RQCF_ACT_SKIP) 1592 rf->clock_update_flags = RQCF_UPDATED; 1593 #endif 1594 1595 lockdep_unpin_lock(__rq_lockp(rq), rf->cookie); 1596 } 1597 1598 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf) 1599 { 1600 lockdep_repin_lock(__rq_lockp(rq), rf->cookie); 1601 1602 #ifdef CONFIG_SCHED_DEBUG 1603 /* 1604 * Restore the value we stashed in @rf for this pin context. 1605 */ 1606 rq->clock_update_flags |= rf->clock_update_flags; 1607 #endif 1608 } 1609 1610 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf) 1611 __acquires(rq->lock); 1612 1613 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf) 1614 __acquires(p->pi_lock) 1615 __acquires(rq->lock); 1616 1617 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf) 1618 __releases(rq->lock) 1619 { 1620 rq_unpin_lock(rq, rf); 1621 raw_spin_rq_unlock(rq); 1622 } 1623 1624 static inline void 1625 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf) 1626 __releases(rq->lock) 1627 __releases(p->pi_lock) 1628 { 1629 rq_unpin_lock(rq, rf); 1630 raw_spin_rq_unlock(rq); 1631 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags); 1632 } 1633 1634 static inline void 1635 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf) 1636 __acquires(rq->lock) 1637 { 1638 raw_spin_rq_lock_irqsave(rq, rf->flags); 1639 rq_pin_lock(rq, rf); 1640 } 1641 1642 static inline void 1643 rq_lock_irq(struct rq *rq, struct rq_flags *rf) 1644 __acquires(rq->lock) 1645 { 1646 raw_spin_rq_lock_irq(rq); 1647 rq_pin_lock(rq, rf); 1648 } 1649 1650 static inline void 1651 rq_lock(struct rq *rq, struct rq_flags *rf) 1652 __acquires(rq->lock) 1653 { 1654 raw_spin_rq_lock(rq); 1655 rq_pin_lock(rq, rf); 1656 } 1657 1658 static inline void 1659 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf) 1660 __releases(rq->lock) 1661 { 1662 rq_unpin_lock(rq, rf); 1663 raw_spin_rq_unlock_irqrestore(rq, rf->flags); 1664 } 1665 1666 static inline void 1667 rq_unlock_irq(struct rq *rq, struct rq_flags *rf) 1668 __releases(rq->lock) 1669 { 1670 rq_unpin_lock(rq, rf); 1671 raw_spin_rq_unlock_irq(rq); 1672 } 1673 1674 static inline void 1675 rq_unlock(struct rq *rq, struct rq_flags *rf) 1676 __releases(rq->lock) 1677 { 1678 rq_unpin_lock(rq, rf); 1679 raw_spin_rq_unlock(rq); 1680 } 1681 1682 static inline struct rq * 1683 this_rq_lock_irq(struct rq_flags *rf) 1684 __acquires(rq->lock) 1685 { 1686 struct rq *rq; 1687 1688 local_irq_disable(); 1689 rq = this_rq(); 1690 rq_lock(rq, rf); 1691 return rq; 1692 } 1693 1694 #ifdef CONFIG_NUMA 1695 enum numa_topology_type { 1696 NUMA_DIRECT, 1697 NUMA_GLUELESS_MESH, 1698 NUMA_BACKPLANE, 1699 }; 1700 extern enum numa_topology_type sched_numa_topology_type; 1701 extern int sched_max_numa_distance; 1702 extern bool find_numa_distance(int distance); 1703 extern void sched_init_numa(int offline_node); 1704 extern void sched_update_numa(int cpu, bool online); 1705 extern void sched_domains_numa_masks_set(unsigned int cpu); 1706 extern void sched_domains_numa_masks_clear(unsigned int cpu); 1707 extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu); 1708 #else 1709 static inline void sched_init_numa(int offline_node) { } 1710 static inline void sched_update_numa(int cpu, bool online) { } 1711 static inline void sched_domains_numa_masks_set(unsigned int cpu) { } 1712 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { } 1713 static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu) 1714 { 1715 return nr_cpu_ids; 1716 } 1717 #endif 1718 1719 #ifdef CONFIG_NUMA_BALANCING 1720 /* The regions in numa_faults array from task_struct */ 1721 enum numa_faults_stats { 1722 NUMA_MEM = 0, 1723 NUMA_CPU, 1724 NUMA_MEMBUF, 1725 NUMA_CPUBUF 1726 }; 1727 extern void sched_setnuma(struct task_struct *p, int node); 1728 extern int migrate_task_to(struct task_struct *p, int cpu); 1729 extern int migrate_swap(struct task_struct *p, struct task_struct *t, 1730 int cpu, int scpu); 1731 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p); 1732 #else 1733 static inline void 1734 init_numa_balancing(unsigned long clone_flags, struct task_struct *p) 1735 { 1736 } 1737 #endif /* CONFIG_NUMA_BALANCING */ 1738 1739 #ifdef CONFIG_SMP 1740 1741 static inline void 1742 queue_balance_callback(struct rq *rq, 1743 struct balance_callback *head, 1744 void (*func)(struct rq *rq)) 1745 { 1746 lockdep_assert_rq_held(rq); 1747 1748 /* 1749 * Don't (re)queue an already queued item; nor queue anything when 1750 * balance_push() is active, see the comment with 1751 * balance_push_callback. 1752 */ 1753 if (unlikely(head->next || rq->balance_callback == &balance_push_callback)) 1754 return; 1755 1756 head->func = func; 1757 head->next = rq->balance_callback; 1758 rq->balance_callback = head; 1759 } 1760 1761 #define rcu_dereference_check_sched_domain(p) \ 1762 rcu_dereference_check((p), \ 1763 lockdep_is_held(&sched_domains_mutex)) 1764 1765 /* 1766 * The domain tree (rq->sd) is protected by RCU's quiescent state transition. 1767 * See destroy_sched_domains: call_rcu for details. 1768 * 1769 * The domain tree of any CPU may only be accessed from within 1770 * preempt-disabled sections. 1771 */ 1772 #define for_each_domain(cpu, __sd) \ 1773 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \ 1774 __sd; __sd = __sd->parent) 1775 1776 /** 1777 * highest_flag_domain - Return highest sched_domain containing flag. 1778 * @cpu: The CPU whose highest level of sched domain is to 1779 * be returned. 1780 * @flag: The flag to check for the highest sched_domain 1781 * for the given CPU. 1782 * 1783 * Returns the highest sched_domain of a CPU which contains the given flag. 1784 */ 1785 static inline struct sched_domain *highest_flag_domain(int cpu, int flag) 1786 { 1787 struct sched_domain *sd, *hsd = NULL; 1788 1789 for_each_domain(cpu, sd) { 1790 if (!(sd->flags & flag)) 1791 break; 1792 hsd = sd; 1793 } 1794 1795 return hsd; 1796 } 1797 1798 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag) 1799 { 1800 struct sched_domain *sd; 1801 1802 for_each_domain(cpu, sd) { 1803 if (sd->flags & flag) 1804 break; 1805 } 1806 1807 return sd; 1808 } 1809 1810 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc); 1811 DECLARE_PER_CPU(int, sd_llc_size); 1812 DECLARE_PER_CPU(int, sd_llc_id); 1813 DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared); 1814 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa); 1815 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing); 1816 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity); 1817 extern struct static_key_false sched_asym_cpucapacity; 1818 1819 static __always_inline bool sched_asym_cpucap_active(void) 1820 { 1821 return static_branch_unlikely(&sched_asym_cpucapacity); 1822 } 1823 1824 struct sched_group_capacity { 1825 atomic_t ref; 1826 /* 1827 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity 1828 * for a single CPU. 1829 */ 1830 unsigned long capacity; 1831 unsigned long min_capacity; /* Min per-CPU capacity in group */ 1832 unsigned long max_capacity; /* Max per-CPU capacity in group */ 1833 unsigned long next_update; 1834 int imbalance; /* XXX unrelated to capacity but shared group state */ 1835 1836 #ifdef CONFIG_SCHED_DEBUG 1837 int id; 1838 #endif 1839 1840 unsigned long cpumask[]; /* Balance mask */ 1841 }; 1842 1843 struct sched_group { 1844 struct sched_group *next; /* Must be a circular list */ 1845 atomic_t ref; 1846 1847 unsigned int group_weight; 1848 struct sched_group_capacity *sgc; 1849 int asym_prefer_cpu; /* CPU of highest priority in group */ 1850 int flags; 1851 1852 /* 1853 * The CPUs this group covers. 1854 * 1855 * NOTE: this field is variable length. (Allocated dynamically 1856 * by attaching extra space to the end of the structure, 1857 * depending on how many CPUs the kernel has booted up with) 1858 */ 1859 unsigned long cpumask[]; 1860 }; 1861 1862 static inline struct cpumask *sched_group_span(struct sched_group *sg) 1863 { 1864 return to_cpumask(sg->cpumask); 1865 } 1866 1867 /* 1868 * See build_balance_mask(). 1869 */ 1870 static inline struct cpumask *group_balance_mask(struct sched_group *sg) 1871 { 1872 return to_cpumask(sg->sgc->cpumask); 1873 } 1874 1875 extern int group_balance_cpu(struct sched_group *sg); 1876 1877 #ifdef CONFIG_SCHED_DEBUG 1878 void update_sched_domain_debugfs(void); 1879 void dirty_sched_domain_sysctl(int cpu); 1880 #else 1881 static inline void update_sched_domain_debugfs(void) 1882 { 1883 } 1884 static inline void dirty_sched_domain_sysctl(int cpu) 1885 { 1886 } 1887 #endif 1888 1889 extern int sched_update_scaling(void); 1890 1891 static inline const struct cpumask *task_user_cpus(struct task_struct *p) 1892 { 1893 if (!p->user_cpus_ptr) 1894 return cpu_possible_mask; /* &init_task.cpus_mask */ 1895 return p->user_cpus_ptr; 1896 } 1897 #endif /* CONFIG_SMP */ 1898 1899 #include "stats.h" 1900 1901 #if defined(CONFIG_SCHED_CORE) && defined(CONFIG_SCHEDSTATS) 1902 1903 extern void __sched_core_account_forceidle(struct rq *rq); 1904 1905 static inline void sched_core_account_forceidle(struct rq *rq) 1906 { 1907 if (schedstat_enabled()) 1908 __sched_core_account_forceidle(rq); 1909 } 1910 1911 extern void __sched_core_tick(struct rq *rq); 1912 1913 static inline void sched_core_tick(struct rq *rq) 1914 { 1915 if (sched_core_enabled(rq) && schedstat_enabled()) 1916 __sched_core_tick(rq); 1917 } 1918 1919 #else 1920 1921 static inline void sched_core_account_forceidle(struct rq *rq) {} 1922 1923 static inline void sched_core_tick(struct rq *rq) {} 1924 1925 #endif /* CONFIG_SCHED_CORE && CONFIG_SCHEDSTATS */ 1926 1927 #ifdef CONFIG_CGROUP_SCHED 1928 1929 /* 1930 * Return the group to which this tasks belongs. 1931 * 1932 * We cannot use task_css() and friends because the cgroup subsystem 1933 * changes that value before the cgroup_subsys::attach() method is called, 1934 * therefore we cannot pin it and might observe the wrong value. 1935 * 1936 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup 1937 * core changes this before calling sched_move_task(). 1938 * 1939 * Instead we use a 'copy' which is updated from sched_move_task() while 1940 * holding both task_struct::pi_lock and rq::lock. 1941 */ 1942 static inline struct task_group *task_group(struct task_struct *p) 1943 { 1944 return p->sched_task_group; 1945 } 1946 1947 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ 1948 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) 1949 { 1950 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED) 1951 struct task_group *tg = task_group(p); 1952 #endif 1953 1954 #ifdef CONFIG_FAIR_GROUP_SCHED 1955 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]); 1956 p->se.cfs_rq = tg->cfs_rq[cpu]; 1957 p->se.parent = tg->se[cpu]; 1958 p->se.depth = tg->se[cpu] ? tg->se[cpu]->depth + 1 : 0; 1959 #endif 1960 1961 #ifdef CONFIG_RT_GROUP_SCHED 1962 p->rt.rt_rq = tg->rt_rq[cpu]; 1963 p->rt.parent = tg->rt_se[cpu]; 1964 #endif 1965 } 1966 1967 #else /* CONFIG_CGROUP_SCHED */ 1968 1969 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } 1970 static inline struct task_group *task_group(struct task_struct *p) 1971 { 1972 return NULL; 1973 } 1974 1975 #endif /* CONFIG_CGROUP_SCHED */ 1976 1977 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) 1978 { 1979 set_task_rq(p, cpu); 1980 #ifdef CONFIG_SMP 1981 /* 1982 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be 1983 * successfully executed on another CPU. We must ensure that updates of 1984 * per-task data have been completed by this moment. 1985 */ 1986 smp_wmb(); 1987 WRITE_ONCE(task_thread_info(p)->cpu, cpu); 1988 p->wake_cpu = cpu; 1989 #endif 1990 } 1991 1992 /* 1993 * Tunables that become constants when CONFIG_SCHED_DEBUG is off: 1994 */ 1995 #ifdef CONFIG_SCHED_DEBUG 1996 # define const_debug __read_mostly 1997 #else 1998 # define const_debug const 1999 #endif 2000 2001 #define SCHED_FEAT(name, enabled) \ 2002 __SCHED_FEAT_##name , 2003 2004 enum { 2005 #include "features.h" 2006 __SCHED_FEAT_NR, 2007 }; 2008 2009 #undef SCHED_FEAT 2010 2011 #ifdef CONFIG_SCHED_DEBUG 2012 2013 /* 2014 * To support run-time toggling of sched features, all the translation units 2015 * (but core.c) reference the sysctl_sched_features defined in core.c. 2016 */ 2017 extern const_debug unsigned int sysctl_sched_features; 2018 2019 #ifdef CONFIG_JUMP_LABEL 2020 #define SCHED_FEAT(name, enabled) \ 2021 static __always_inline bool static_branch_##name(struct static_key *key) \ 2022 { \ 2023 return static_key_##enabled(key); \ 2024 } 2025 2026 #include "features.h" 2027 #undef SCHED_FEAT 2028 2029 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR]; 2030 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x])) 2031 2032 #else /* !CONFIG_JUMP_LABEL */ 2033 2034 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) 2035 2036 #endif /* CONFIG_JUMP_LABEL */ 2037 2038 #else /* !SCHED_DEBUG */ 2039 2040 /* 2041 * Each translation unit has its own copy of sysctl_sched_features to allow 2042 * constants propagation at compile time and compiler optimization based on 2043 * features default. 2044 */ 2045 #define SCHED_FEAT(name, enabled) \ 2046 (1UL << __SCHED_FEAT_##name) * enabled | 2047 static const_debug __maybe_unused unsigned int sysctl_sched_features = 2048 #include "features.h" 2049 0; 2050 #undef SCHED_FEAT 2051 2052 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) 2053 2054 #endif /* SCHED_DEBUG */ 2055 2056 extern struct static_key_false sched_numa_balancing; 2057 extern struct static_key_false sched_schedstats; 2058 2059 static inline u64 global_rt_period(void) 2060 { 2061 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; 2062 } 2063 2064 static inline u64 global_rt_runtime(void) 2065 { 2066 if (sysctl_sched_rt_runtime < 0) 2067 return RUNTIME_INF; 2068 2069 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; 2070 } 2071 2072 static inline int task_current(struct rq *rq, struct task_struct *p) 2073 { 2074 return rq->curr == p; 2075 } 2076 2077 static inline int task_on_cpu(struct rq *rq, struct task_struct *p) 2078 { 2079 #ifdef CONFIG_SMP 2080 return p->on_cpu; 2081 #else 2082 return task_current(rq, p); 2083 #endif 2084 } 2085 2086 static inline int task_on_rq_queued(struct task_struct *p) 2087 { 2088 return p->on_rq == TASK_ON_RQ_QUEUED; 2089 } 2090 2091 static inline int task_on_rq_migrating(struct task_struct *p) 2092 { 2093 return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING; 2094 } 2095 2096 /* Wake flags. The first three directly map to some SD flag value */ 2097 #define WF_EXEC 0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */ 2098 #define WF_FORK 0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */ 2099 #define WF_TTWU 0x08 /* Wakeup; maps to SD_BALANCE_WAKE */ 2100 2101 #define WF_SYNC 0x10 /* Waker goes to sleep after wakeup */ 2102 #define WF_MIGRATED 0x20 /* Internal use, task got migrated */ 2103 2104 #ifdef CONFIG_SMP 2105 static_assert(WF_EXEC == SD_BALANCE_EXEC); 2106 static_assert(WF_FORK == SD_BALANCE_FORK); 2107 static_assert(WF_TTWU == SD_BALANCE_WAKE); 2108 #endif 2109 2110 /* 2111 * To aid in avoiding the subversion of "niceness" due to uneven distribution 2112 * of tasks with abnormal "nice" values across CPUs the contribution that 2113 * each task makes to its run queue's load is weighted according to its 2114 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a 2115 * scaled version of the new time slice allocation that they receive on time 2116 * slice expiry etc. 2117 */ 2118 2119 #define WEIGHT_IDLEPRIO 3 2120 #define WMULT_IDLEPRIO 1431655765 2121 2122 extern const int sched_prio_to_weight[40]; 2123 extern const u32 sched_prio_to_wmult[40]; 2124 2125 /* 2126 * {de,en}queue flags: 2127 * 2128 * DEQUEUE_SLEEP - task is no longer runnable 2129 * ENQUEUE_WAKEUP - task just became runnable 2130 * 2131 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks 2132 * are in a known state which allows modification. Such pairs 2133 * should preserve as much state as possible. 2134 * 2135 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location 2136 * in the runqueue. 2137 * 2138 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified) 2139 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline) 2140 * ENQUEUE_MIGRATED - the task was migrated during wakeup 2141 * 2142 */ 2143 2144 #define DEQUEUE_SLEEP 0x01 2145 #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */ 2146 #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */ 2147 #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */ 2148 2149 #define ENQUEUE_WAKEUP 0x01 2150 #define ENQUEUE_RESTORE 0x02 2151 #define ENQUEUE_MOVE 0x04 2152 #define ENQUEUE_NOCLOCK 0x08 2153 2154 #define ENQUEUE_HEAD 0x10 2155 #define ENQUEUE_REPLENISH 0x20 2156 #ifdef CONFIG_SMP 2157 #define ENQUEUE_MIGRATED 0x40 2158 #else 2159 #define ENQUEUE_MIGRATED 0x00 2160 #endif 2161 2162 #define RETRY_TASK ((void *)-1UL) 2163 2164 struct affinity_context { 2165 const struct cpumask *new_mask; 2166 struct cpumask *user_mask; 2167 unsigned int flags; 2168 }; 2169 2170 struct sched_class { 2171 2172 #ifdef CONFIG_UCLAMP_TASK 2173 int uclamp_enabled; 2174 #endif 2175 2176 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags); 2177 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags); 2178 void (*yield_task) (struct rq *rq); 2179 bool (*yield_to_task)(struct rq *rq, struct task_struct *p); 2180 2181 void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags); 2182 2183 struct task_struct *(*pick_next_task)(struct rq *rq); 2184 2185 void (*put_prev_task)(struct rq *rq, struct task_struct *p); 2186 void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first); 2187 2188 #ifdef CONFIG_SMP 2189 int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf); 2190 int (*select_task_rq)(struct task_struct *p, int task_cpu, int flags); 2191 2192 struct task_struct * (*pick_task)(struct rq *rq); 2193 2194 void (*migrate_task_rq)(struct task_struct *p, int new_cpu); 2195 2196 void (*task_woken)(struct rq *this_rq, struct task_struct *task); 2197 2198 void (*set_cpus_allowed)(struct task_struct *p, struct affinity_context *ctx); 2199 2200 void (*rq_online)(struct rq *rq); 2201 void (*rq_offline)(struct rq *rq); 2202 2203 struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq); 2204 #endif 2205 2206 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued); 2207 void (*task_fork)(struct task_struct *p); 2208 void (*task_dead)(struct task_struct *p); 2209 2210 /* 2211 * The switched_from() call is allowed to drop rq->lock, therefore we 2212 * cannot assume the switched_from/switched_to pair is serialized by 2213 * rq->lock. They are however serialized by p->pi_lock. 2214 */ 2215 void (*switched_from)(struct rq *this_rq, struct task_struct *task); 2216 void (*switched_to) (struct rq *this_rq, struct task_struct *task); 2217 void (*prio_changed) (struct rq *this_rq, struct task_struct *task, 2218 int oldprio); 2219 2220 unsigned int (*get_rr_interval)(struct rq *rq, 2221 struct task_struct *task); 2222 2223 void (*update_curr)(struct rq *rq); 2224 2225 #ifdef CONFIG_FAIR_GROUP_SCHED 2226 void (*task_change_group)(struct task_struct *p); 2227 #endif 2228 }; 2229 2230 static inline void put_prev_task(struct rq *rq, struct task_struct *prev) 2231 { 2232 WARN_ON_ONCE(rq->curr != prev); 2233 prev->sched_class->put_prev_task(rq, prev); 2234 } 2235 2236 static inline void set_next_task(struct rq *rq, struct task_struct *next) 2237 { 2238 next->sched_class->set_next_task(rq, next, false); 2239 } 2240 2241 2242 /* 2243 * Helper to define a sched_class instance; each one is placed in a separate 2244 * section which is ordered by the linker script: 2245 * 2246 * include/asm-generic/vmlinux.lds.h 2247 * 2248 * *CAREFUL* they are laid out in *REVERSE* order!!! 2249 * 2250 * Also enforce alignment on the instance, not the type, to guarantee layout. 2251 */ 2252 #define DEFINE_SCHED_CLASS(name) \ 2253 const struct sched_class name##_sched_class \ 2254 __aligned(__alignof__(struct sched_class)) \ 2255 __section("__" #name "_sched_class") 2256 2257 /* Defined in include/asm-generic/vmlinux.lds.h */ 2258 extern struct sched_class __sched_class_highest[]; 2259 extern struct sched_class __sched_class_lowest[]; 2260 2261 #define for_class_range(class, _from, _to) \ 2262 for (class = (_from); class < (_to); class++) 2263 2264 #define for_each_class(class) \ 2265 for_class_range(class, __sched_class_highest, __sched_class_lowest) 2266 2267 #define sched_class_above(_a, _b) ((_a) < (_b)) 2268 2269 extern const struct sched_class stop_sched_class; 2270 extern const struct sched_class dl_sched_class; 2271 extern const struct sched_class rt_sched_class; 2272 extern const struct sched_class fair_sched_class; 2273 extern const struct sched_class idle_sched_class; 2274 2275 static inline bool sched_stop_runnable(struct rq *rq) 2276 { 2277 return rq->stop && task_on_rq_queued(rq->stop); 2278 } 2279 2280 static inline bool sched_dl_runnable(struct rq *rq) 2281 { 2282 return rq->dl.dl_nr_running > 0; 2283 } 2284 2285 static inline bool sched_rt_runnable(struct rq *rq) 2286 { 2287 return rq->rt.rt_queued > 0; 2288 } 2289 2290 static inline bool sched_fair_runnable(struct rq *rq) 2291 { 2292 return rq->cfs.nr_running > 0; 2293 } 2294 2295 extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf); 2296 extern struct task_struct *pick_next_task_idle(struct rq *rq); 2297 2298 #define SCA_CHECK 0x01 2299 #define SCA_MIGRATE_DISABLE 0x02 2300 #define SCA_MIGRATE_ENABLE 0x04 2301 #define SCA_USER 0x08 2302 2303 #ifdef CONFIG_SMP 2304 2305 extern void update_group_capacity(struct sched_domain *sd, int cpu); 2306 2307 extern void trigger_load_balance(struct rq *rq); 2308 2309 extern void set_cpus_allowed_common(struct task_struct *p, struct affinity_context *ctx); 2310 2311 static inline struct task_struct *get_push_task(struct rq *rq) 2312 { 2313 struct task_struct *p = rq->curr; 2314 2315 lockdep_assert_rq_held(rq); 2316 2317 if (rq->push_busy) 2318 return NULL; 2319 2320 if (p->nr_cpus_allowed == 1) 2321 return NULL; 2322 2323 if (p->migration_disabled) 2324 return NULL; 2325 2326 rq->push_busy = true; 2327 return get_task_struct(p); 2328 } 2329 2330 extern int push_cpu_stop(void *arg); 2331 2332 #endif 2333 2334 #ifdef CONFIG_CPU_IDLE 2335 static inline void idle_set_state(struct rq *rq, 2336 struct cpuidle_state *idle_state) 2337 { 2338 rq->idle_state = idle_state; 2339 } 2340 2341 static inline struct cpuidle_state *idle_get_state(struct rq *rq) 2342 { 2343 SCHED_WARN_ON(!rcu_read_lock_held()); 2344 2345 return rq->idle_state; 2346 } 2347 #else 2348 static inline void idle_set_state(struct rq *rq, 2349 struct cpuidle_state *idle_state) 2350 { 2351 } 2352 2353 static inline struct cpuidle_state *idle_get_state(struct rq *rq) 2354 { 2355 return NULL; 2356 } 2357 #endif 2358 2359 extern void schedule_idle(void); 2360 2361 extern void sysrq_sched_debug_show(void); 2362 extern void sched_init_granularity(void); 2363 extern void update_max_interval(void); 2364 2365 extern void init_sched_dl_class(void); 2366 extern void init_sched_rt_class(void); 2367 extern void init_sched_fair_class(void); 2368 2369 extern void reweight_task(struct task_struct *p, int prio); 2370 2371 extern void resched_curr(struct rq *rq); 2372 extern void resched_cpu(int cpu); 2373 2374 extern struct rt_bandwidth def_rt_bandwidth; 2375 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime); 2376 extern bool sched_rt_bandwidth_account(struct rt_rq *rt_rq); 2377 2378 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime); 2379 extern void init_dl_task_timer(struct sched_dl_entity *dl_se); 2380 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se); 2381 2382 #define BW_SHIFT 20 2383 #define BW_UNIT (1 << BW_SHIFT) 2384 #define RATIO_SHIFT 8 2385 #define MAX_BW_BITS (64 - BW_SHIFT) 2386 #define MAX_BW ((1ULL << MAX_BW_BITS) - 1) 2387 unsigned long to_ratio(u64 period, u64 runtime); 2388 2389 extern void init_entity_runnable_average(struct sched_entity *se); 2390 extern void post_init_entity_util_avg(struct task_struct *p); 2391 2392 #ifdef CONFIG_NO_HZ_FULL 2393 extern bool sched_can_stop_tick(struct rq *rq); 2394 extern int __init sched_tick_offload_init(void); 2395 2396 /* 2397 * Tick may be needed by tasks in the runqueue depending on their policy and 2398 * requirements. If tick is needed, lets send the target an IPI to kick it out of 2399 * nohz mode if necessary. 2400 */ 2401 static inline void sched_update_tick_dependency(struct rq *rq) 2402 { 2403 int cpu = cpu_of(rq); 2404 2405 if (!tick_nohz_full_cpu(cpu)) 2406 return; 2407 2408 if (sched_can_stop_tick(rq)) 2409 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED); 2410 else 2411 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED); 2412 } 2413 #else 2414 static inline int sched_tick_offload_init(void) { return 0; } 2415 static inline void sched_update_tick_dependency(struct rq *rq) { } 2416 #endif 2417 2418 static inline void add_nr_running(struct rq *rq, unsigned count) 2419 { 2420 unsigned prev_nr = rq->nr_running; 2421 2422 rq->nr_running = prev_nr + count; 2423 if (trace_sched_update_nr_running_tp_enabled()) { 2424 call_trace_sched_update_nr_running(rq, count); 2425 } 2426 2427 #ifdef CONFIG_SMP 2428 if (prev_nr < 2 && rq->nr_running >= 2) { 2429 if (!READ_ONCE(rq->rd->overload)) 2430 WRITE_ONCE(rq->rd->overload, 1); 2431 } 2432 #endif 2433 2434 sched_update_tick_dependency(rq); 2435 } 2436 2437 static inline void sub_nr_running(struct rq *rq, unsigned count) 2438 { 2439 rq->nr_running -= count; 2440 if (trace_sched_update_nr_running_tp_enabled()) { 2441 call_trace_sched_update_nr_running(rq, -count); 2442 } 2443 2444 /* Check if we still need preemption */ 2445 sched_update_tick_dependency(rq); 2446 } 2447 2448 extern void activate_task(struct rq *rq, struct task_struct *p, int flags); 2449 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags); 2450 2451 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags); 2452 2453 #ifdef CONFIG_PREEMPT_RT 2454 #define SCHED_NR_MIGRATE_BREAK 8 2455 #else 2456 #define SCHED_NR_MIGRATE_BREAK 32 2457 #endif 2458 2459 extern const_debug unsigned int sysctl_sched_nr_migrate; 2460 extern const_debug unsigned int sysctl_sched_migration_cost; 2461 2462 #ifdef CONFIG_SCHED_DEBUG 2463 extern unsigned int sysctl_sched_latency; 2464 extern unsigned int sysctl_sched_min_granularity; 2465 extern unsigned int sysctl_sched_idle_min_granularity; 2466 extern unsigned int sysctl_sched_wakeup_granularity; 2467 extern int sysctl_resched_latency_warn_ms; 2468 extern int sysctl_resched_latency_warn_once; 2469 2470 extern unsigned int sysctl_sched_tunable_scaling; 2471 2472 extern unsigned int sysctl_numa_balancing_scan_delay; 2473 extern unsigned int sysctl_numa_balancing_scan_period_min; 2474 extern unsigned int sysctl_numa_balancing_scan_period_max; 2475 extern unsigned int sysctl_numa_balancing_scan_size; 2476 extern unsigned int sysctl_numa_balancing_hot_threshold; 2477 #endif 2478 2479 #ifdef CONFIG_SCHED_HRTICK 2480 2481 /* 2482 * Use hrtick when: 2483 * - enabled by features 2484 * - hrtimer is actually high res 2485 */ 2486 static inline int hrtick_enabled(struct rq *rq) 2487 { 2488 if (!cpu_active(cpu_of(rq))) 2489 return 0; 2490 return hrtimer_is_hres_active(&rq->hrtick_timer); 2491 } 2492 2493 static inline int hrtick_enabled_fair(struct rq *rq) 2494 { 2495 if (!sched_feat(HRTICK)) 2496 return 0; 2497 return hrtick_enabled(rq); 2498 } 2499 2500 static inline int hrtick_enabled_dl(struct rq *rq) 2501 { 2502 if (!sched_feat(HRTICK_DL)) 2503 return 0; 2504 return hrtick_enabled(rq); 2505 } 2506 2507 void hrtick_start(struct rq *rq, u64 delay); 2508 2509 #else 2510 2511 static inline int hrtick_enabled_fair(struct rq *rq) 2512 { 2513 return 0; 2514 } 2515 2516 static inline int hrtick_enabled_dl(struct rq *rq) 2517 { 2518 return 0; 2519 } 2520 2521 static inline int hrtick_enabled(struct rq *rq) 2522 { 2523 return 0; 2524 } 2525 2526 #endif /* CONFIG_SCHED_HRTICK */ 2527 2528 #ifndef arch_scale_freq_tick 2529 static __always_inline 2530 void arch_scale_freq_tick(void) 2531 { 2532 } 2533 #endif 2534 2535 #ifndef arch_scale_freq_capacity 2536 /** 2537 * arch_scale_freq_capacity - get the frequency scale factor of a given CPU. 2538 * @cpu: the CPU in question. 2539 * 2540 * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e. 2541 * 2542 * f_curr 2543 * ------ * SCHED_CAPACITY_SCALE 2544 * f_max 2545 */ 2546 static __always_inline 2547 unsigned long arch_scale_freq_capacity(int cpu) 2548 { 2549 return SCHED_CAPACITY_SCALE; 2550 } 2551 #endif 2552 2553 #ifdef CONFIG_SCHED_DEBUG 2554 /* 2555 * In double_lock_balance()/double_rq_lock(), we use raw_spin_rq_lock() to 2556 * acquire rq lock instead of rq_lock(). So at the end of these two functions 2557 * we need to call double_rq_clock_clear_update() to clear RQCF_UPDATED of 2558 * rq->clock_update_flags to avoid the WARN_DOUBLE_CLOCK warning. 2559 */ 2560 static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2) 2561 { 2562 rq1->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP); 2563 /* rq1 == rq2 for !CONFIG_SMP, so just clear RQCF_UPDATED once. */ 2564 #ifdef CONFIG_SMP 2565 rq2->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP); 2566 #endif 2567 } 2568 #else 2569 static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2) {} 2570 #endif 2571 2572 #ifdef CONFIG_SMP 2573 2574 static inline bool rq_order_less(struct rq *rq1, struct rq *rq2) 2575 { 2576 #ifdef CONFIG_SCHED_CORE 2577 /* 2578 * In order to not have {0,2},{1,3} turn into into an AB-BA, 2579 * order by core-id first and cpu-id second. 2580 * 2581 * Notably: 2582 * 2583 * double_rq_lock(0,3); will take core-0, core-1 lock 2584 * double_rq_lock(1,2); will take core-1, core-0 lock 2585 * 2586 * when only cpu-id is considered. 2587 */ 2588 if (rq1->core->cpu < rq2->core->cpu) 2589 return true; 2590 if (rq1->core->cpu > rq2->core->cpu) 2591 return false; 2592 2593 /* 2594 * __sched_core_flip() relies on SMT having cpu-id lock order. 2595 */ 2596 #endif 2597 return rq1->cpu < rq2->cpu; 2598 } 2599 2600 extern void double_rq_lock(struct rq *rq1, struct rq *rq2); 2601 2602 #ifdef CONFIG_PREEMPTION 2603 2604 /* 2605 * fair double_lock_balance: Safely acquires both rq->locks in a fair 2606 * way at the expense of forcing extra atomic operations in all 2607 * invocations. This assures that the double_lock is acquired using the 2608 * same underlying policy as the spinlock_t on this architecture, which 2609 * reduces latency compared to the unfair variant below. However, it 2610 * also adds more overhead and therefore may reduce throughput. 2611 */ 2612 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) 2613 __releases(this_rq->lock) 2614 __acquires(busiest->lock) 2615 __acquires(this_rq->lock) 2616 { 2617 raw_spin_rq_unlock(this_rq); 2618 double_rq_lock(this_rq, busiest); 2619 2620 return 1; 2621 } 2622 2623 #else 2624 /* 2625 * Unfair double_lock_balance: Optimizes throughput at the expense of 2626 * latency by eliminating extra atomic operations when the locks are 2627 * already in proper order on entry. This favors lower CPU-ids and will 2628 * grant the double lock to lower CPUs over higher ids under contention, 2629 * regardless of entry order into the function. 2630 */ 2631 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) 2632 __releases(this_rq->lock) 2633 __acquires(busiest->lock) 2634 __acquires(this_rq->lock) 2635 { 2636 if (__rq_lockp(this_rq) == __rq_lockp(busiest) || 2637 likely(raw_spin_rq_trylock(busiest))) { 2638 double_rq_clock_clear_update(this_rq, busiest); 2639 return 0; 2640 } 2641 2642 if (rq_order_less(this_rq, busiest)) { 2643 raw_spin_rq_lock_nested(busiest, SINGLE_DEPTH_NESTING); 2644 double_rq_clock_clear_update(this_rq, busiest); 2645 return 0; 2646 } 2647 2648 raw_spin_rq_unlock(this_rq); 2649 double_rq_lock(this_rq, busiest); 2650 2651 return 1; 2652 } 2653 2654 #endif /* CONFIG_PREEMPTION */ 2655 2656 /* 2657 * double_lock_balance - lock the busiest runqueue, this_rq is locked already. 2658 */ 2659 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest) 2660 { 2661 lockdep_assert_irqs_disabled(); 2662 2663 return _double_lock_balance(this_rq, busiest); 2664 } 2665 2666 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) 2667 __releases(busiest->lock) 2668 { 2669 if (__rq_lockp(this_rq) != __rq_lockp(busiest)) 2670 raw_spin_rq_unlock(busiest); 2671 lock_set_subclass(&__rq_lockp(this_rq)->dep_map, 0, _RET_IP_); 2672 } 2673 2674 static inline void double_lock(spinlock_t *l1, spinlock_t *l2) 2675 { 2676 if (l1 > l2) 2677 swap(l1, l2); 2678 2679 spin_lock(l1); 2680 spin_lock_nested(l2, SINGLE_DEPTH_NESTING); 2681 } 2682 2683 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2) 2684 { 2685 if (l1 > l2) 2686 swap(l1, l2); 2687 2688 spin_lock_irq(l1); 2689 spin_lock_nested(l2, SINGLE_DEPTH_NESTING); 2690 } 2691 2692 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2) 2693 { 2694 if (l1 > l2) 2695 swap(l1, l2); 2696 2697 raw_spin_lock(l1); 2698 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING); 2699 } 2700 2701 /* 2702 * double_rq_unlock - safely unlock two runqueues 2703 * 2704 * Note this does not restore interrupts like task_rq_unlock, 2705 * you need to do so manually after calling. 2706 */ 2707 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) 2708 __releases(rq1->lock) 2709 __releases(rq2->lock) 2710 { 2711 if (__rq_lockp(rq1) != __rq_lockp(rq2)) 2712 raw_spin_rq_unlock(rq2); 2713 else 2714 __release(rq2->lock); 2715 raw_spin_rq_unlock(rq1); 2716 } 2717 2718 extern void set_rq_online (struct rq *rq); 2719 extern void set_rq_offline(struct rq *rq); 2720 extern bool sched_smp_initialized; 2721 2722 #else /* CONFIG_SMP */ 2723 2724 /* 2725 * double_rq_lock - safely lock two runqueues 2726 * 2727 * Note this does not disable interrupts like task_rq_lock, 2728 * you need to do so manually before calling. 2729 */ 2730 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) 2731 __acquires(rq1->lock) 2732 __acquires(rq2->lock) 2733 { 2734 WARN_ON_ONCE(!irqs_disabled()); 2735 WARN_ON_ONCE(rq1 != rq2); 2736 raw_spin_rq_lock(rq1); 2737 __acquire(rq2->lock); /* Fake it out ;) */ 2738 double_rq_clock_clear_update(rq1, rq2); 2739 } 2740 2741 /* 2742 * double_rq_unlock - safely unlock two runqueues 2743 * 2744 * Note this does not restore interrupts like task_rq_unlock, 2745 * you need to do so manually after calling. 2746 */ 2747 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) 2748 __releases(rq1->lock) 2749 __releases(rq2->lock) 2750 { 2751 WARN_ON_ONCE(rq1 != rq2); 2752 raw_spin_rq_unlock(rq1); 2753 __release(rq2->lock); 2754 } 2755 2756 #endif 2757 2758 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq); 2759 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq); 2760 2761 #ifdef CONFIG_SCHED_DEBUG 2762 extern bool sched_debug_verbose; 2763 2764 extern void print_cfs_stats(struct seq_file *m, int cpu); 2765 extern void print_rt_stats(struct seq_file *m, int cpu); 2766 extern void print_dl_stats(struct seq_file *m, int cpu); 2767 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq); 2768 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq); 2769 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq); 2770 2771 extern void resched_latency_warn(int cpu, u64 latency); 2772 #ifdef CONFIG_NUMA_BALANCING 2773 extern void 2774 show_numa_stats(struct task_struct *p, struct seq_file *m); 2775 extern void 2776 print_numa_stats(struct seq_file *m, int node, unsigned long tsf, 2777 unsigned long tpf, unsigned long gsf, unsigned long gpf); 2778 #endif /* CONFIG_NUMA_BALANCING */ 2779 #else 2780 static inline void resched_latency_warn(int cpu, u64 latency) {} 2781 #endif /* CONFIG_SCHED_DEBUG */ 2782 2783 extern void init_cfs_rq(struct cfs_rq *cfs_rq); 2784 extern void init_rt_rq(struct rt_rq *rt_rq); 2785 extern void init_dl_rq(struct dl_rq *dl_rq); 2786 2787 extern void cfs_bandwidth_usage_inc(void); 2788 extern void cfs_bandwidth_usage_dec(void); 2789 2790 #ifdef CONFIG_NO_HZ_COMMON 2791 #define NOHZ_BALANCE_KICK_BIT 0 2792 #define NOHZ_STATS_KICK_BIT 1 2793 #define NOHZ_NEWILB_KICK_BIT 2 2794 #define NOHZ_NEXT_KICK_BIT 3 2795 2796 /* Run rebalance_domains() */ 2797 #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT) 2798 /* Update blocked load */ 2799 #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT) 2800 /* Update blocked load when entering idle */ 2801 #define NOHZ_NEWILB_KICK BIT(NOHZ_NEWILB_KICK_BIT) 2802 /* Update nohz.next_balance */ 2803 #define NOHZ_NEXT_KICK BIT(NOHZ_NEXT_KICK_BIT) 2804 2805 #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK | NOHZ_NEXT_KICK) 2806 2807 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags) 2808 2809 extern void nohz_balance_exit_idle(struct rq *rq); 2810 #else 2811 static inline void nohz_balance_exit_idle(struct rq *rq) { } 2812 #endif 2813 2814 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON) 2815 extern void nohz_run_idle_balance(int cpu); 2816 #else 2817 static inline void nohz_run_idle_balance(int cpu) { } 2818 #endif 2819 2820 #ifdef CONFIG_IRQ_TIME_ACCOUNTING 2821 struct irqtime { 2822 u64 total; 2823 u64 tick_delta; 2824 u64 irq_start_time; 2825 struct u64_stats_sync sync; 2826 }; 2827 2828 DECLARE_PER_CPU(struct irqtime, cpu_irqtime); 2829 2830 /* 2831 * Returns the irqtime minus the softirq time computed by ksoftirqd. 2832 * Otherwise ksoftirqd's sum_exec_runtime is subtracted its own runtime 2833 * and never move forward. 2834 */ 2835 static inline u64 irq_time_read(int cpu) 2836 { 2837 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu); 2838 unsigned int seq; 2839 u64 total; 2840 2841 do { 2842 seq = __u64_stats_fetch_begin(&irqtime->sync); 2843 total = irqtime->total; 2844 } while (__u64_stats_fetch_retry(&irqtime->sync, seq)); 2845 2846 return total; 2847 } 2848 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ 2849 2850 #ifdef CONFIG_CPU_FREQ 2851 DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data); 2852 2853 /** 2854 * cpufreq_update_util - Take a note about CPU utilization changes. 2855 * @rq: Runqueue to carry out the update for. 2856 * @flags: Update reason flags. 2857 * 2858 * This function is called by the scheduler on the CPU whose utilization is 2859 * being updated. 2860 * 2861 * It can only be called from RCU-sched read-side critical sections. 2862 * 2863 * The way cpufreq is currently arranged requires it to evaluate the CPU 2864 * performance state (frequency/voltage) on a regular basis to prevent it from 2865 * being stuck in a completely inadequate performance level for too long. 2866 * That is not guaranteed to happen if the updates are only triggered from CFS 2867 * and DL, though, because they may not be coming in if only RT tasks are 2868 * active all the time (or there are RT tasks only). 2869 * 2870 * As a workaround for that issue, this function is called periodically by the 2871 * RT sched class to trigger extra cpufreq updates to prevent it from stalling, 2872 * but that really is a band-aid. Going forward it should be replaced with 2873 * solutions targeted more specifically at RT tasks. 2874 */ 2875 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) 2876 { 2877 struct update_util_data *data; 2878 2879 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data, 2880 cpu_of(rq))); 2881 if (data) 2882 data->func(data, rq_clock(rq), flags); 2883 } 2884 #else 2885 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {} 2886 #endif /* CONFIG_CPU_FREQ */ 2887 2888 #ifdef arch_scale_freq_capacity 2889 # ifndef arch_scale_freq_invariant 2890 # define arch_scale_freq_invariant() true 2891 # endif 2892 #else 2893 # define arch_scale_freq_invariant() false 2894 #endif 2895 2896 #ifdef CONFIG_SMP 2897 static inline unsigned long capacity_orig_of(int cpu) 2898 { 2899 return cpu_rq(cpu)->cpu_capacity_orig; 2900 } 2901 2902 /* 2903 * Returns inverted capacity if the CPU is in capacity inversion state. 2904 * 0 otherwise. 2905 * 2906 * Capacity inversion detection only considers thermal impact where actual 2907 * performance points (OPPs) gets dropped. 2908 * 2909 * Capacity inversion state happens when another performance domain that has 2910 * equal or lower capacity_orig_of() becomes effectively larger than the perf 2911 * domain this CPU belongs to due to thermal pressure throttling it hard. 2912 * 2913 * See comment in update_cpu_capacity(). 2914 */ 2915 static inline unsigned long cpu_in_capacity_inversion(int cpu) 2916 { 2917 return cpu_rq(cpu)->cpu_capacity_inverted; 2918 } 2919 2920 /** 2921 * enum cpu_util_type - CPU utilization type 2922 * @FREQUENCY_UTIL: Utilization used to select frequency 2923 * @ENERGY_UTIL: Utilization used during energy calculation 2924 * 2925 * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time 2926 * need to be aggregated differently depending on the usage made of them. This 2927 * enum is used within effective_cpu_util() to differentiate the types of 2928 * utilization expected by the callers, and adjust the aggregation accordingly. 2929 */ 2930 enum cpu_util_type { 2931 FREQUENCY_UTIL, 2932 ENERGY_UTIL, 2933 }; 2934 2935 unsigned long effective_cpu_util(int cpu, unsigned long util_cfs, 2936 enum cpu_util_type type, 2937 struct task_struct *p); 2938 2939 /* 2940 * Verify the fitness of task @p to run on @cpu taking into account the 2941 * CPU original capacity and the runtime/deadline ratio of the task. 2942 * 2943 * The function will return true if the original capacity of @cpu is 2944 * greater than or equal to task's deadline density right shifted by 2945 * (BW_SHIFT - SCHED_CAPACITY_SHIFT) and false otherwise. 2946 */ 2947 static inline bool dl_task_fits_capacity(struct task_struct *p, int cpu) 2948 { 2949 unsigned long cap = arch_scale_cpu_capacity(cpu); 2950 2951 return cap >= p->dl.dl_density >> (BW_SHIFT - SCHED_CAPACITY_SHIFT); 2952 } 2953 2954 static inline unsigned long cpu_bw_dl(struct rq *rq) 2955 { 2956 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT; 2957 } 2958 2959 static inline unsigned long cpu_util_dl(struct rq *rq) 2960 { 2961 return READ_ONCE(rq->avg_dl.util_avg); 2962 } 2963 2964 /** 2965 * cpu_util_cfs() - Estimates the amount of CPU capacity used by CFS tasks. 2966 * @cpu: the CPU to get the utilization for. 2967 * 2968 * The unit of the return value must be the same as the one of CPU capacity 2969 * so that CPU utilization can be compared with CPU capacity. 2970 * 2971 * CPU utilization is the sum of running time of runnable tasks plus the 2972 * recent utilization of currently non-runnable tasks on that CPU. 2973 * It represents the amount of CPU capacity currently used by CFS tasks in 2974 * the range [0..max CPU capacity] with max CPU capacity being the CPU 2975 * capacity at f_max. 2976 * 2977 * The estimated CPU utilization is defined as the maximum between CPU 2978 * utilization and sum of the estimated utilization of the currently 2979 * runnable tasks on that CPU. It preserves a utilization "snapshot" of 2980 * previously-executed tasks, which helps better deduce how busy a CPU will 2981 * be when a long-sleeping task wakes up. The contribution to CPU utilization 2982 * of such a task would be significantly decayed at this point of time. 2983 * 2984 * CPU utilization can be higher than the current CPU capacity 2985 * (f_curr/f_max * max CPU capacity) or even the max CPU capacity because 2986 * of rounding errors as well as task migrations or wakeups of new tasks. 2987 * CPU utilization has to be capped to fit into the [0..max CPU capacity] 2988 * range. Otherwise a group of CPUs (CPU0 util = 121% + CPU1 util = 80%) 2989 * could be seen as over-utilized even though CPU1 has 20% of spare CPU 2990 * capacity. CPU utilization is allowed to overshoot current CPU capacity 2991 * though since this is useful for predicting the CPU capacity required 2992 * after task migrations (scheduler-driven DVFS). 2993 * 2994 * Return: (Estimated) utilization for the specified CPU. 2995 */ 2996 static inline unsigned long cpu_util_cfs(int cpu) 2997 { 2998 struct cfs_rq *cfs_rq; 2999 unsigned long util; 3000 3001 cfs_rq = &cpu_rq(cpu)->cfs; 3002 util = READ_ONCE(cfs_rq->avg.util_avg); 3003 3004 if (sched_feat(UTIL_EST)) { 3005 util = max_t(unsigned long, util, 3006 READ_ONCE(cfs_rq->avg.util_est.enqueued)); 3007 } 3008 3009 return min(util, capacity_orig_of(cpu)); 3010 } 3011 3012 static inline unsigned long cpu_util_rt(struct rq *rq) 3013 { 3014 return READ_ONCE(rq->avg_rt.util_avg); 3015 } 3016 #endif 3017 3018 #ifdef CONFIG_UCLAMP_TASK 3019 unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id); 3020 3021 static inline unsigned long uclamp_rq_get(struct rq *rq, 3022 enum uclamp_id clamp_id) 3023 { 3024 return READ_ONCE(rq->uclamp[clamp_id].value); 3025 } 3026 3027 static inline void uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id, 3028 unsigned int value) 3029 { 3030 WRITE_ONCE(rq->uclamp[clamp_id].value, value); 3031 } 3032 3033 static inline bool uclamp_rq_is_idle(struct rq *rq) 3034 { 3035 return rq->uclamp_flags & UCLAMP_FLAG_IDLE; 3036 } 3037 3038 /** 3039 * uclamp_rq_util_with - clamp @util with @rq and @p effective uclamp values. 3040 * @rq: The rq to clamp against. Must not be NULL. 3041 * @util: The util value to clamp. 3042 * @p: The task to clamp against. Can be NULL if you want to clamp 3043 * against @rq only. 3044 * 3045 * Clamps the passed @util to the max(@rq, @p) effective uclamp values. 3046 * 3047 * If sched_uclamp_used static key is disabled, then just return the util 3048 * without any clamping since uclamp aggregation at the rq level in the fast 3049 * path is disabled, rendering this operation a NOP. 3050 * 3051 * Use uclamp_eff_value() if you don't care about uclamp values at rq level. It 3052 * will return the correct effective uclamp value of the task even if the 3053 * static key is disabled. 3054 */ 3055 static __always_inline 3056 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util, 3057 struct task_struct *p) 3058 { 3059 unsigned long min_util = 0; 3060 unsigned long max_util = 0; 3061 3062 if (!static_branch_likely(&sched_uclamp_used)) 3063 return util; 3064 3065 if (p) { 3066 min_util = uclamp_eff_value(p, UCLAMP_MIN); 3067 max_util = uclamp_eff_value(p, UCLAMP_MAX); 3068 3069 /* 3070 * Ignore last runnable task's max clamp, as this task will 3071 * reset it. Similarly, no need to read the rq's min clamp. 3072 */ 3073 if (uclamp_rq_is_idle(rq)) 3074 goto out; 3075 } 3076 3077 min_util = max_t(unsigned long, min_util, uclamp_rq_get(rq, UCLAMP_MIN)); 3078 max_util = max_t(unsigned long, max_util, uclamp_rq_get(rq, UCLAMP_MAX)); 3079 out: 3080 /* 3081 * Since CPU's {min,max}_util clamps are MAX aggregated considering 3082 * RUNNABLE tasks with _different_ clamps, we can end up with an 3083 * inversion. Fix it now when the clamps are applied. 3084 */ 3085 if (unlikely(min_util >= max_util)) 3086 return min_util; 3087 3088 return clamp(util, min_util, max_util); 3089 } 3090 3091 /* Is the rq being capped/throttled by uclamp_max? */ 3092 static inline bool uclamp_rq_is_capped(struct rq *rq) 3093 { 3094 unsigned long rq_util; 3095 unsigned long max_util; 3096 3097 if (!static_branch_likely(&sched_uclamp_used)) 3098 return false; 3099 3100 rq_util = cpu_util_cfs(cpu_of(rq)) + cpu_util_rt(rq); 3101 max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value); 3102 3103 return max_util != SCHED_CAPACITY_SCALE && rq_util >= max_util; 3104 } 3105 3106 /* 3107 * When uclamp is compiled in, the aggregation at rq level is 'turned off' 3108 * by default in the fast path and only gets turned on once userspace performs 3109 * an operation that requires it. 3110 * 3111 * Returns true if userspace opted-in to use uclamp and aggregation at rq level 3112 * hence is active. 3113 */ 3114 static inline bool uclamp_is_used(void) 3115 { 3116 return static_branch_likely(&sched_uclamp_used); 3117 } 3118 #else /* CONFIG_UCLAMP_TASK */ 3119 static inline unsigned long uclamp_eff_value(struct task_struct *p, 3120 enum uclamp_id clamp_id) 3121 { 3122 if (clamp_id == UCLAMP_MIN) 3123 return 0; 3124 3125 return SCHED_CAPACITY_SCALE; 3126 } 3127 3128 static inline 3129 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util, 3130 struct task_struct *p) 3131 { 3132 return util; 3133 } 3134 3135 static inline bool uclamp_rq_is_capped(struct rq *rq) { return false; } 3136 3137 static inline bool uclamp_is_used(void) 3138 { 3139 return false; 3140 } 3141 3142 static inline unsigned long uclamp_rq_get(struct rq *rq, 3143 enum uclamp_id clamp_id) 3144 { 3145 if (clamp_id == UCLAMP_MIN) 3146 return 0; 3147 3148 return SCHED_CAPACITY_SCALE; 3149 } 3150 3151 static inline void uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id, 3152 unsigned int value) 3153 { 3154 } 3155 3156 static inline bool uclamp_rq_is_idle(struct rq *rq) 3157 { 3158 return false; 3159 } 3160 #endif /* CONFIG_UCLAMP_TASK */ 3161 3162 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ 3163 static inline unsigned long cpu_util_irq(struct rq *rq) 3164 { 3165 return rq->avg_irq.util_avg; 3166 } 3167 3168 static inline 3169 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max) 3170 { 3171 util *= (max - irq); 3172 util /= max; 3173 3174 return util; 3175 3176 } 3177 #else 3178 static inline unsigned long cpu_util_irq(struct rq *rq) 3179 { 3180 return 0; 3181 } 3182 3183 static inline 3184 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max) 3185 { 3186 return util; 3187 } 3188 #endif 3189 3190 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL) 3191 3192 #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus))) 3193 3194 DECLARE_STATIC_KEY_FALSE(sched_energy_present); 3195 3196 static inline bool sched_energy_enabled(void) 3197 { 3198 return static_branch_unlikely(&sched_energy_present); 3199 } 3200 3201 #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */ 3202 3203 #define perf_domain_span(pd) NULL 3204 static inline bool sched_energy_enabled(void) { return false; } 3205 3206 #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */ 3207 3208 #ifdef CONFIG_MEMBARRIER 3209 /* 3210 * The scheduler provides memory barriers required by membarrier between: 3211 * - prior user-space memory accesses and store to rq->membarrier_state, 3212 * - store to rq->membarrier_state and following user-space memory accesses. 3213 * In the same way it provides those guarantees around store to rq->curr. 3214 */ 3215 static inline void membarrier_switch_mm(struct rq *rq, 3216 struct mm_struct *prev_mm, 3217 struct mm_struct *next_mm) 3218 { 3219 int membarrier_state; 3220 3221 if (prev_mm == next_mm) 3222 return; 3223 3224 membarrier_state = atomic_read(&next_mm->membarrier_state); 3225 if (READ_ONCE(rq->membarrier_state) == membarrier_state) 3226 return; 3227 3228 WRITE_ONCE(rq->membarrier_state, membarrier_state); 3229 } 3230 #else 3231 static inline void membarrier_switch_mm(struct rq *rq, 3232 struct mm_struct *prev_mm, 3233 struct mm_struct *next_mm) 3234 { 3235 } 3236 #endif 3237 3238 #ifdef CONFIG_SMP 3239 static inline bool is_per_cpu_kthread(struct task_struct *p) 3240 { 3241 if (!(p->flags & PF_KTHREAD)) 3242 return false; 3243 3244 if (p->nr_cpus_allowed != 1) 3245 return false; 3246 3247 return true; 3248 } 3249 #endif 3250 3251 extern void swake_up_all_locked(struct swait_queue_head *q); 3252 extern void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait); 3253 3254 #ifdef CONFIG_PREEMPT_DYNAMIC 3255 extern int preempt_dynamic_mode; 3256 extern int sched_dynamic_mode(const char *str); 3257 extern void sched_dynamic_update(int mode); 3258 #endif 3259 3260 static inline void update_current_exec_runtime(struct task_struct *curr, 3261 u64 now, u64 delta_exec) 3262 { 3263 curr->se.sum_exec_runtime += delta_exec; 3264 account_group_exec_runtime(curr, delta_exec); 3265 3266 curr->se.exec_start = now; 3267 cgroup_account_cputime(curr, delta_exec); 3268 } 3269 3270 #ifdef CONFIG_SCHED_MM_CID 3271 static inline int __mm_cid_get(struct mm_struct *mm) 3272 { 3273 struct cpumask *cpumask; 3274 int cid; 3275 3276 cpumask = mm_cidmask(mm); 3277 cid = cpumask_first_zero(cpumask); 3278 if (cid >= nr_cpu_ids) 3279 return -1; 3280 __cpumask_set_cpu(cid, cpumask); 3281 return cid; 3282 } 3283 3284 static inline void mm_cid_put(struct mm_struct *mm, int cid) 3285 { 3286 lockdep_assert_irqs_disabled(); 3287 if (cid < 0) 3288 return; 3289 raw_spin_lock(&mm->cid_lock); 3290 __cpumask_clear_cpu(cid, mm_cidmask(mm)); 3291 raw_spin_unlock(&mm->cid_lock); 3292 } 3293 3294 static inline int mm_cid_get(struct mm_struct *mm) 3295 { 3296 int ret; 3297 3298 lockdep_assert_irqs_disabled(); 3299 raw_spin_lock(&mm->cid_lock); 3300 ret = __mm_cid_get(mm); 3301 raw_spin_unlock(&mm->cid_lock); 3302 return ret; 3303 } 3304 3305 static inline void switch_mm_cid(struct task_struct *prev, struct task_struct *next) 3306 { 3307 if (prev->mm_cid_active) { 3308 if (next->mm_cid_active && next->mm == prev->mm) { 3309 /* 3310 * Context switch between threads in same mm, hand over 3311 * the mm_cid from prev to next. 3312 */ 3313 next->mm_cid = prev->mm_cid; 3314 prev->mm_cid = -1; 3315 return; 3316 } 3317 mm_cid_put(prev->mm, prev->mm_cid); 3318 prev->mm_cid = -1; 3319 } 3320 if (next->mm_cid_active) 3321 next->mm_cid = mm_cid_get(next->mm); 3322 } 3323 3324 #else 3325 static inline void switch_mm_cid(struct task_struct *prev, struct task_struct *next) { } 3326 #endif 3327 3328 #endif /* _KERNEL_SCHED_SCHED_H */ 3329