1 /* SPDX-License-Identifier: GPL-2.0 */ 2 /* 3 * Scheduler internal types and methods: 4 */ 5 #include <linux/sched.h> 6 7 #include <linux/sched/autogroup.h> 8 #include <linux/sched/clock.h> 9 #include <linux/sched/coredump.h> 10 #include <linux/sched/cpufreq.h> 11 #include <linux/sched/cputime.h> 12 #include <linux/sched/deadline.h> 13 #include <linux/sched/debug.h> 14 #include <linux/sched/hotplug.h> 15 #include <linux/sched/idle.h> 16 #include <linux/sched/init.h> 17 #include <linux/sched/isolation.h> 18 #include <linux/sched/jobctl.h> 19 #include <linux/sched/loadavg.h> 20 #include <linux/sched/mm.h> 21 #include <linux/sched/nohz.h> 22 #include <linux/sched/numa_balancing.h> 23 #include <linux/sched/prio.h> 24 #include <linux/sched/rt.h> 25 #include <linux/sched/signal.h> 26 #include <linux/sched/smt.h> 27 #include <linux/sched/stat.h> 28 #include <linux/sched/sysctl.h> 29 #include <linux/sched/task.h> 30 #include <linux/sched/task_stack.h> 31 #include <linux/sched/topology.h> 32 #include <linux/sched/user.h> 33 #include <linux/sched/wake_q.h> 34 #include <linux/sched/xacct.h> 35 36 #include <uapi/linux/sched/types.h> 37 38 #include <linux/binfmts.h> 39 #include <linux/blkdev.h> 40 #include <linux/compat.h> 41 #include <linux/context_tracking.h> 42 #include <linux/cpufreq.h> 43 #include <linux/cpuidle.h> 44 #include <linux/cpuset.h> 45 #include <linux/ctype.h> 46 #include <linux/debugfs.h> 47 #include <linux/delayacct.h> 48 #include <linux/energy_model.h> 49 #include <linux/init_task.h> 50 #include <linux/kprobes.h> 51 #include <linux/kthread.h> 52 #include <linux/membarrier.h> 53 #include <linux/migrate.h> 54 #include <linux/mmu_context.h> 55 #include <linux/nmi.h> 56 #include <linux/proc_fs.h> 57 #include <linux/prefetch.h> 58 #include <linux/profile.h> 59 #include <linux/psi.h> 60 #include <linux/rcupdate_wait.h> 61 #include <linux/security.h> 62 #include <linux/stop_machine.h> 63 #include <linux/suspend.h> 64 #include <linux/swait.h> 65 #include <linux/syscalls.h> 66 #include <linux/task_work.h> 67 #include <linux/tsacct_kern.h> 68 69 #include <asm/tlb.h> 70 71 #ifdef CONFIG_PARAVIRT 72 # include <asm/paravirt.h> 73 #endif 74 75 #include "cpupri.h" 76 #include "cpudeadline.h" 77 78 #ifdef CONFIG_SCHED_DEBUG 79 # define SCHED_WARN_ON(x) WARN_ONCE(x, #x) 80 #else 81 # define SCHED_WARN_ON(x) ({ (void)(x), 0; }) 82 #endif 83 84 struct rq; 85 struct cpuidle_state; 86 87 /* task_struct::on_rq states: */ 88 #define TASK_ON_RQ_QUEUED 1 89 #define TASK_ON_RQ_MIGRATING 2 90 91 extern __read_mostly int scheduler_running; 92 93 extern unsigned long calc_load_update; 94 extern atomic_long_t calc_load_tasks; 95 96 extern void calc_global_load_tick(struct rq *this_rq); 97 extern long calc_load_fold_active(struct rq *this_rq, long adjust); 98 99 /* 100 * Helpers for converting nanosecond timing to jiffy resolution 101 */ 102 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) 103 104 /* 105 * Increase resolution of nice-level calculations for 64-bit architectures. 106 * The extra resolution improves shares distribution and load balancing of 107 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup 108 * hierarchies, especially on larger systems. This is not a user-visible change 109 * and does not change the user-interface for setting shares/weights. 110 * 111 * We increase resolution only if we have enough bits to allow this increased 112 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit 113 * are pretty high and the returns do not justify the increased costs. 114 * 115 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to 116 * increase coverage and consistency always enable it on 64-bit platforms. 117 */ 118 #ifdef CONFIG_64BIT 119 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT) 120 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT) 121 # define scale_load_down(w) \ 122 ({ \ 123 unsigned long __w = (w); \ 124 if (__w) \ 125 __w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \ 126 __w; \ 127 }) 128 #else 129 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT) 130 # define scale_load(w) (w) 131 # define scale_load_down(w) (w) 132 #endif 133 134 /* 135 * Task weight (visible to users) and its load (invisible to users) have 136 * independent resolution, but they should be well calibrated. We use 137 * scale_load() and scale_load_down(w) to convert between them. The 138 * following must be true: 139 * 140 * scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD 141 * 142 */ 143 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT) 144 145 /* 146 * Single value that decides SCHED_DEADLINE internal math precision. 147 * 10 -> just above 1us 148 * 9 -> just above 0.5us 149 */ 150 #define DL_SCALE 10 151 152 /* 153 * Single value that denotes runtime == period, ie unlimited time. 154 */ 155 #define RUNTIME_INF ((u64)~0ULL) 156 157 static inline int idle_policy(int policy) 158 { 159 return policy == SCHED_IDLE; 160 } 161 static inline int fair_policy(int policy) 162 { 163 return policy == SCHED_NORMAL || policy == SCHED_BATCH; 164 } 165 166 static inline int rt_policy(int policy) 167 { 168 return policy == SCHED_FIFO || policy == SCHED_RR; 169 } 170 171 static inline int dl_policy(int policy) 172 { 173 return policy == SCHED_DEADLINE; 174 } 175 static inline bool valid_policy(int policy) 176 { 177 return idle_policy(policy) || fair_policy(policy) || 178 rt_policy(policy) || dl_policy(policy); 179 } 180 181 static inline int task_has_idle_policy(struct task_struct *p) 182 { 183 return idle_policy(p->policy); 184 } 185 186 static inline int task_has_rt_policy(struct task_struct *p) 187 { 188 return rt_policy(p->policy); 189 } 190 191 static inline int task_has_dl_policy(struct task_struct *p) 192 { 193 return dl_policy(p->policy); 194 } 195 196 #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT) 197 198 static inline void update_avg(u64 *avg, u64 sample) 199 { 200 s64 diff = sample - *avg; 201 *avg += diff / 8; 202 } 203 204 /* 205 * !! For sched_setattr_nocheck() (kernel) only !! 206 * 207 * This is actually gross. :( 208 * 209 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE 210 * tasks, but still be able to sleep. We need this on platforms that cannot 211 * atomically change clock frequency. Remove once fast switching will be 212 * available on such platforms. 213 * 214 * SUGOV stands for SchedUtil GOVernor. 215 */ 216 #define SCHED_FLAG_SUGOV 0x10000000 217 218 static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se) 219 { 220 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL 221 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV); 222 #else 223 return false; 224 #endif 225 } 226 227 /* 228 * Tells if entity @a should preempt entity @b. 229 */ 230 static inline bool 231 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b) 232 { 233 return dl_entity_is_special(a) || 234 dl_time_before(a->deadline, b->deadline); 235 } 236 237 /* 238 * This is the priority-queue data structure of the RT scheduling class: 239 */ 240 struct rt_prio_array { 241 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ 242 struct list_head queue[MAX_RT_PRIO]; 243 }; 244 245 struct rt_bandwidth { 246 /* nests inside the rq lock: */ 247 raw_spinlock_t rt_runtime_lock; 248 ktime_t rt_period; 249 u64 rt_runtime; 250 struct hrtimer rt_period_timer; 251 unsigned int rt_period_active; 252 }; 253 254 void __dl_clear_params(struct task_struct *p); 255 256 /* 257 * To keep the bandwidth of -deadline tasks and groups under control 258 * we need some place where: 259 * - store the maximum -deadline bandwidth of the system (the group); 260 * - cache the fraction of that bandwidth that is currently allocated. 261 * 262 * This is all done in the data structure below. It is similar to the 263 * one used for RT-throttling (rt_bandwidth), with the main difference 264 * that, since here we are only interested in admission control, we 265 * do not decrease any runtime while the group "executes", neither we 266 * need a timer to replenish it. 267 * 268 * With respect to SMP, the bandwidth is given on a per-CPU basis, 269 * meaning that: 270 * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU; 271 * - dl_total_bw array contains, in the i-eth element, the currently 272 * allocated bandwidth on the i-eth CPU. 273 * Moreover, groups consume bandwidth on each CPU, while tasks only 274 * consume bandwidth on the CPU they're running on. 275 * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw 276 * that will be shown the next time the proc or cgroup controls will 277 * be red. It on its turn can be changed by writing on its own 278 * control. 279 */ 280 struct dl_bandwidth { 281 raw_spinlock_t dl_runtime_lock; 282 u64 dl_runtime; 283 u64 dl_period; 284 }; 285 286 static inline int dl_bandwidth_enabled(void) 287 { 288 return sysctl_sched_rt_runtime >= 0; 289 } 290 291 struct dl_bw { 292 raw_spinlock_t lock; 293 u64 bw; 294 u64 total_bw; 295 }; 296 297 static inline void __dl_update(struct dl_bw *dl_b, s64 bw); 298 299 static inline 300 void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus) 301 { 302 dl_b->total_bw -= tsk_bw; 303 __dl_update(dl_b, (s32)tsk_bw / cpus); 304 } 305 306 static inline 307 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus) 308 { 309 dl_b->total_bw += tsk_bw; 310 __dl_update(dl_b, -((s32)tsk_bw / cpus)); 311 } 312 313 static inline 314 bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw) 315 { 316 return dl_b->bw != -1 && 317 dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw; 318 } 319 320 extern void init_dl_bw(struct dl_bw *dl_b); 321 extern int sched_dl_global_validate(void); 322 extern void sched_dl_do_global(void); 323 extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr); 324 extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr); 325 extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr); 326 extern bool __checkparam_dl(const struct sched_attr *attr); 327 extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr); 328 extern int dl_task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed); 329 extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial); 330 extern bool dl_cpu_busy(unsigned int cpu); 331 332 #ifdef CONFIG_CGROUP_SCHED 333 334 #include <linux/cgroup.h> 335 #include <linux/psi.h> 336 337 struct cfs_rq; 338 struct rt_rq; 339 340 extern struct list_head task_groups; 341 342 struct cfs_bandwidth { 343 #ifdef CONFIG_CFS_BANDWIDTH 344 raw_spinlock_t lock; 345 ktime_t period; 346 u64 quota; 347 u64 runtime; 348 s64 hierarchical_quota; 349 350 u8 idle; 351 u8 period_active; 352 u8 distribute_running; 353 u8 slack_started; 354 struct hrtimer period_timer; 355 struct hrtimer slack_timer; 356 struct list_head throttled_cfs_rq; 357 358 /* Statistics: */ 359 int nr_periods; 360 int nr_throttled; 361 u64 throttled_time; 362 #endif 363 }; 364 365 /* Task group related information */ 366 struct task_group { 367 struct cgroup_subsys_state css; 368 369 #ifdef CONFIG_FAIR_GROUP_SCHED 370 /* schedulable entities of this group on each CPU */ 371 struct sched_entity **se; 372 /* runqueue "owned" by this group on each CPU */ 373 struct cfs_rq **cfs_rq; 374 unsigned long shares; 375 376 #ifdef CONFIG_SMP 377 /* 378 * load_avg can be heavily contended at clock tick time, so put 379 * it in its own cacheline separated from the fields above which 380 * will also be accessed at each tick. 381 */ 382 atomic_long_t load_avg ____cacheline_aligned; 383 #endif 384 #endif 385 386 #ifdef CONFIG_RT_GROUP_SCHED 387 struct sched_rt_entity **rt_se; 388 struct rt_rq **rt_rq; 389 390 struct rt_bandwidth rt_bandwidth; 391 #endif 392 393 struct rcu_head rcu; 394 struct list_head list; 395 396 struct task_group *parent; 397 struct list_head siblings; 398 struct list_head children; 399 400 #ifdef CONFIG_SCHED_AUTOGROUP 401 struct autogroup *autogroup; 402 #endif 403 404 struct cfs_bandwidth cfs_bandwidth; 405 406 #ifdef CONFIG_UCLAMP_TASK_GROUP 407 /* The two decimal precision [%] value requested from user-space */ 408 unsigned int uclamp_pct[UCLAMP_CNT]; 409 /* Clamp values requested for a task group */ 410 struct uclamp_se uclamp_req[UCLAMP_CNT]; 411 /* Effective clamp values used for a task group */ 412 struct uclamp_se uclamp[UCLAMP_CNT]; 413 #endif 414 415 }; 416 417 #ifdef CONFIG_FAIR_GROUP_SCHED 418 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD 419 420 /* 421 * A weight of 0 or 1 can cause arithmetics problems. 422 * A weight of a cfs_rq is the sum of weights of which entities 423 * are queued on this cfs_rq, so a weight of a entity should not be 424 * too large, so as the shares value of a task group. 425 * (The default weight is 1024 - so there's no practical 426 * limitation from this.) 427 */ 428 #define MIN_SHARES (1UL << 1) 429 #define MAX_SHARES (1UL << 18) 430 #endif 431 432 typedef int (*tg_visitor)(struct task_group *, void *); 433 434 extern int walk_tg_tree_from(struct task_group *from, 435 tg_visitor down, tg_visitor up, void *data); 436 437 /* 438 * Iterate the full tree, calling @down when first entering a node and @up when 439 * leaving it for the final time. 440 * 441 * Caller must hold rcu_lock or sufficient equivalent. 442 */ 443 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) 444 { 445 return walk_tg_tree_from(&root_task_group, down, up, data); 446 } 447 448 extern int tg_nop(struct task_group *tg, void *data); 449 450 extern void free_fair_sched_group(struct task_group *tg); 451 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent); 452 extern void online_fair_sched_group(struct task_group *tg); 453 extern void unregister_fair_sched_group(struct task_group *tg); 454 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, 455 struct sched_entity *se, int cpu, 456 struct sched_entity *parent); 457 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b); 458 459 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b); 460 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b); 461 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq); 462 463 extern void free_rt_sched_group(struct task_group *tg); 464 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent); 465 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, 466 struct sched_rt_entity *rt_se, int cpu, 467 struct sched_rt_entity *parent); 468 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us); 469 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us); 470 extern long sched_group_rt_runtime(struct task_group *tg); 471 extern long sched_group_rt_period(struct task_group *tg); 472 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk); 473 474 extern struct task_group *sched_create_group(struct task_group *parent); 475 extern void sched_online_group(struct task_group *tg, 476 struct task_group *parent); 477 extern void sched_destroy_group(struct task_group *tg); 478 extern void sched_offline_group(struct task_group *tg); 479 480 extern void sched_move_task(struct task_struct *tsk); 481 482 #ifdef CONFIG_FAIR_GROUP_SCHED 483 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares); 484 485 #ifdef CONFIG_SMP 486 extern void set_task_rq_fair(struct sched_entity *se, 487 struct cfs_rq *prev, struct cfs_rq *next); 488 #else /* !CONFIG_SMP */ 489 static inline void set_task_rq_fair(struct sched_entity *se, 490 struct cfs_rq *prev, struct cfs_rq *next) { } 491 #endif /* CONFIG_SMP */ 492 #endif /* CONFIG_FAIR_GROUP_SCHED */ 493 494 #else /* CONFIG_CGROUP_SCHED */ 495 496 struct cfs_bandwidth { }; 497 498 #endif /* CONFIG_CGROUP_SCHED */ 499 500 /* CFS-related fields in a runqueue */ 501 struct cfs_rq { 502 struct load_weight load; 503 unsigned int nr_running; 504 unsigned int h_nr_running; /* SCHED_{NORMAL,BATCH,IDLE} */ 505 unsigned int idle_h_nr_running; /* SCHED_IDLE */ 506 507 u64 exec_clock; 508 u64 min_vruntime; 509 #ifndef CONFIG_64BIT 510 u64 min_vruntime_copy; 511 #endif 512 513 struct rb_root_cached tasks_timeline; 514 515 /* 516 * 'curr' points to currently running entity on this cfs_rq. 517 * It is set to NULL otherwise (i.e when none are currently running). 518 */ 519 struct sched_entity *curr; 520 struct sched_entity *next; 521 struct sched_entity *last; 522 struct sched_entity *skip; 523 524 #ifdef CONFIG_SCHED_DEBUG 525 unsigned int nr_spread_over; 526 #endif 527 528 #ifdef CONFIG_SMP 529 /* 530 * CFS load tracking 531 */ 532 struct sched_avg avg; 533 #ifndef CONFIG_64BIT 534 u64 load_last_update_time_copy; 535 #endif 536 struct { 537 raw_spinlock_t lock ____cacheline_aligned; 538 int nr; 539 unsigned long load_avg; 540 unsigned long util_avg; 541 unsigned long runnable_avg; 542 } removed; 543 544 #ifdef CONFIG_FAIR_GROUP_SCHED 545 unsigned long tg_load_avg_contrib; 546 long propagate; 547 long prop_runnable_sum; 548 549 /* 550 * h_load = weight * f(tg) 551 * 552 * Where f(tg) is the recursive weight fraction assigned to 553 * this group. 554 */ 555 unsigned long h_load; 556 u64 last_h_load_update; 557 struct sched_entity *h_load_next; 558 #endif /* CONFIG_FAIR_GROUP_SCHED */ 559 #endif /* CONFIG_SMP */ 560 561 #ifdef CONFIG_FAIR_GROUP_SCHED 562 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */ 563 564 /* 565 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in 566 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities 567 * (like users, containers etc.) 568 * 569 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU. 570 * This list is used during load balance. 571 */ 572 int on_list; 573 struct list_head leaf_cfs_rq_list; 574 struct task_group *tg; /* group that "owns" this runqueue */ 575 576 #ifdef CONFIG_CFS_BANDWIDTH 577 int runtime_enabled; 578 s64 runtime_remaining; 579 580 u64 throttled_clock; 581 u64 throttled_clock_task; 582 u64 throttled_clock_task_time; 583 int throttled; 584 int throttle_count; 585 struct list_head throttled_list; 586 #endif /* CONFIG_CFS_BANDWIDTH */ 587 #endif /* CONFIG_FAIR_GROUP_SCHED */ 588 }; 589 590 static inline int rt_bandwidth_enabled(void) 591 { 592 return sysctl_sched_rt_runtime >= 0; 593 } 594 595 /* RT IPI pull logic requires IRQ_WORK */ 596 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP) 597 # define HAVE_RT_PUSH_IPI 598 #endif 599 600 /* Real-Time classes' related field in a runqueue: */ 601 struct rt_rq { 602 struct rt_prio_array active; 603 unsigned int rt_nr_running; 604 unsigned int rr_nr_running; 605 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED 606 struct { 607 int curr; /* highest queued rt task prio */ 608 #ifdef CONFIG_SMP 609 int next; /* next highest */ 610 #endif 611 } highest_prio; 612 #endif 613 #ifdef CONFIG_SMP 614 unsigned long rt_nr_migratory; 615 unsigned long rt_nr_total; 616 int overloaded; 617 struct plist_head pushable_tasks; 618 619 #endif /* CONFIG_SMP */ 620 int rt_queued; 621 622 int rt_throttled; 623 u64 rt_time; 624 u64 rt_runtime; 625 /* Nests inside the rq lock: */ 626 raw_spinlock_t rt_runtime_lock; 627 628 #ifdef CONFIG_RT_GROUP_SCHED 629 unsigned long rt_nr_boosted; 630 631 struct rq *rq; 632 struct task_group *tg; 633 #endif 634 }; 635 636 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq) 637 { 638 return rt_rq->rt_queued && rt_rq->rt_nr_running; 639 } 640 641 /* Deadline class' related fields in a runqueue */ 642 struct dl_rq { 643 /* runqueue is an rbtree, ordered by deadline */ 644 struct rb_root_cached root; 645 646 unsigned long dl_nr_running; 647 648 #ifdef CONFIG_SMP 649 /* 650 * Deadline values of the currently executing and the 651 * earliest ready task on this rq. Caching these facilitates 652 * the decision whether or not a ready but not running task 653 * should migrate somewhere else. 654 */ 655 struct { 656 u64 curr; 657 u64 next; 658 } earliest_dl; 659 660 unsigned long dl_nr_migratory; 661 int overloaded; 662 663 /* 664 * Tasks on this rq that can be pushed away. They are kept in 665 * an rb-tree, ordered by tasks' deadlines, with caching 666 * of the leftmost (earliest deadline) element. 667 */ 668 struct rb_root_cached pushable_dl_tasks_root; 669 #else 670 struct dl_bw dl_bw; 671 #endif 672 /* 673 * "Active utilization" for this runqueue: increased when a 674 * task wakes up (becomes TASK_RUNNING) and decreased when a 675 * task blocks 676 */ 677 u64 running_bw; 678 679 /* 680 * Utilization of the tasks "assigned" to this runqueue (including 681 * the tasks that are in runqueue and the tasks that executed on this 682 * CPU and blocked). Increased when a task moves to this runqueue, and 683 * decreased when the task moves away (migrates, changes scheduling 684 * policy, or terminates). 685 * This is needed to compute the "inactive utilization" for the 686 * runqueue (inactive utilization = this_bw - running_bw). 687 */ 688 u64 this_bw; 689 u64 extra_bw; 690 691 /* 692 * Inverse of the fraction of CPU utilization that can be reclaimed 693 * by the GRUB algorithm. 694 */ 695 u64 bw_ratio; 696 }; 697 698 #ifdef CONFIG_FAIR_GROUP_SCHED 699 /* An entity is a task if it doesn't "own" a runqueue */ 700 #define entity_is_task(se) (!se->my_q) 701 702 static inline void se_update_runnable(struct sched_entity *se) 703 { 704 if (!entity_is_task(se)) 705 se->runnable_weight = se->my_q->h_nr_running; 706 } 707 708 static inline long se_runnable(struct sched_entity *se) 709 { 710 if (entity_is_task(se)) 711 return !!se->on_rq; 712 else 713 return se->runnable_weight; 714 } 715 716 #else 717 #define entity_is_task(se) 1 718 719 static inline void se_update_runnable(struct sched_entity *se) {} 720 721 static inline long se_runnable(struct sched_entity *se) 722 { 723 return !!se->on_rq; 724 } 725 #endif 726 727 #ifdef CONFIG_SMP 728 /* 729 * XXX we want to get rid of these helpers and use the full load resolution. 730 */ 731 static inline long se_weight(struct sched_entity *se) 732 { 733 return scale_load_down(se->load.weight); 734 } 735 736 737 static inline bool sched_asym_prefer(int a, int b) 738 { 739 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b); 740 } 741 742 struct perf_domain { 743 struct em_perf_domain *em_pd; 744 struct perf_domain *next; 745 struct rcu_head rcu; 746 }; 747 748 /* Scheduling group status flags */ 749 #define SG_OVERLOAD 0x1 /* More than one runnable task on a CPU. */ 750 #define SG_OVERUTILIZED 0x2 /* One or more CPUs are over-utilized. */ 751 752 /* 753 * We add the notion of a root-domain which will be used to define per-domain 754 * variables. Each exclusive cpuset essentially defines an island domain by 755 * fully partitioning the member CPUs from any other cpuset. Whenever a new 756 * exclusive cpuset is created, we also create and attach a new root-domain 757 * object. 758 * 759 */ 760 struct root_domain { 761 atomic_t refcount; 762 atomic_t rto_count; 763 struct rcu_head rcu; 764 cpumask_var_t span; 765 cpumask_var_t online; 766 767 /* 768 * Indicate pullable load on at least one CPU, e.g: 769 * - More than one runnable task 770 * - Running task is misfit 771 */ 772 int overload; 773 774 /* Indicate one or more cpus over-utilized (tipping point) */ 775 int overutilized; 776 777 /* 778 * The bit corresponding to a CPU gets set here if such CPU has more 779 * than one runnable -deadline task (as it is below for RT tasks). 780 */ 781 cpumask_var_t dlo_mask; 782 atomic_t dlo_count; 783 struct dl_bw dl_bw; 784 struct cpudl cpudl; 785 786 #ifdef HAVE_RT_PUSH_IPI 787 /* 788 * For IPI pull requests, loop across the rto_mask. 789 */ 790 struct irq_work rto_push_work; 791 raw_spinlock_t rto_lock; 792 /* These are only updated and read within rto_lock */ 793 int rto_loop; 794 int rto_cpu; 795 /* These atomics are updated outside of a lock */ 796 atomic_t rto_loop_next; 797 atomic_t rto_loop_start; 798 #endif 799 /* 800 * The "RT overload" flag: it gets set if a CPU has more than 801 * one runnable RT task. 802 */ 803 cpumask_var_t rto_mask; 804 struct cpupri cpupri; 805 806 unsigned long max_cpu_capacity; 807 808 /* 809 * NULL-terminated list of performance domains intersecting with the 810 * CPUs of the rd. Protected by RCU. 811 */ 812 struct perf_domain __rcu *pd; 813 }; 814 815 extern void init_defrootdomain(void); 816 extern int sched_init_domains(const struct cpumask *cpu_map); 817 extern void rq_attach_root(struct rq *rq, struct root_domain *rd); 818 extern void sched_get_rd(struct root_domain *rd); 819 extern void sched_put_rd(struct root_domain *rd); 820 821 #ifdef HAVE_RT_PUSH_IPI 822 extern void rto_push_irq_work_func(struct irq_work *work); 823 #endif 824 #endif /* CONFIG_SMP */ 825 826 #ifdef CONFIG_UCLAMP_TASK 827 /* 828 * struct uclamp_bucket - Utilization clamp bucket 829 * @value: utilization clamp value for tasks on this clamp bucket 830 * @tasks: number of RUNNABLE tasks on this clamp bucket 831 * 832 * Keep track of how many tasks are RUNNABLE for a given utilization 833 * clamp value. 834 */ 835 struct uclamp_bucket { 836 unsigned long value : bits_per(SCHED_CAPACITY_SCALE); 837 unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE); 838 }; 839 840 /* 841 * struct uclamp_rq - rq's utilization clamp 842 * @value: currently active clamp values for a rq 843 * @bucket: utilization clamp buckets affecting a rq 844 * 845 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values. 846 * A clamp value is affecting a rq when there is at least one task RUNNABLE 847 * (or actually running) with that value. 848 * 849 * There are up to UCLAMP_CNT possible different clamp values, currently there 850 * are only two: minimum utilization and maximum utilization. 851 * 852 * All utilization clamping values are MAX aggregated, since: 853 * - for util_min: we want to run the CPU at least at the max of the minimum 854 * utilization required by its currently RUNNABLE tasks. 855 * - for util_max: we want to allow the CPU to run up to the max of the 856 * maximum utilization allowed by its currently RUNNABLE tasks. 857 * 858 * Since on each system we expect only a limited number of different 859 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track 860 * the metrics required to compute all the per-rq utilization clamp values. 861 */ 862 struct uclamp_rq { 863 unsigned int value; 864 struct uclamp_bucket bucket[UCLAMP_BUCKETS]; 865 }; 866 #endif /* CONFIG_UCLAMP_TASK */ 867 868 /* 869 * This is the main, per-CPU runqueue data structure. 870 * 871 * Locking rule: those places that want to lock multiple runqueues 872 * (such as the load balancing or the thread migration code), lock 873 * acquire operations must be ordered by ascending &runqueue. 874 */ 875 struct rq { 876 /* runqueue lock: */ 877 raw_spinlock_t lock; 878 879 /* 880 * nr_running and cpu_load should be in the same cacheline because 881 * remote CPUs use both these fields when doing load calculation. 882 */ 883 unsigned int nr_running; 884 #ifdef CONFIG_NUMA_BALANCING 885 unsigned int nr_numa_running; 886 unsigned int nr_preferred_running; 887 unsigned int numa_migrate_on; 888 #endif 889 #ifdef CONFIG_NO_HZ_COMMON 890 #ifdef CONFIG_SMP 891 unsigned long last_blocked_load_update_tick; 892 unsigned int has_blocked_load; 893 #endif /* CONFIG_SMP */ 894 unsigned int nohz_tick_stopped; 895 atomic_t nohz_flags; 896 #endif /* CONFIG_NO_HZ_COMMON */ 897 898 unsigned long nr_load_updates; 899 u64 nr_switches; 900 901 #ifdef CONFIG_UCLAMP_TASK 902 /* Utilization clamp values based on CPU's RUNNABLE tasks */ 903 struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned; 904 unsigned int uclamp_flags; 905 #define UCLAMP_FLAG_IDLE 0x01 906 #endif 907 908 struct cfs_rq cfs; 909 struct rt_rq rt; 910 struct dl_rq dl; 911 912 #ifdef CONFIG_FAIR_GROUP_SCHED 913 /* list of leaf cfs_rq on this CPU: */ 914 struct list_head leaf_cfs_rq_list; 915 struct list_head *tmp_alone_branch; 916 #endif /* CONFIG_FAIR_GROUP_SCHED */ 917 918 /* 919 * This is part of a global counter where only the total sum 920 * over all CPUs matters. A task can increase this counter on 921 * one CPU and if it got migrated afterwards it may decrease 922 * it on another CPU. Always updated under the runqueue lock: 923 */ 924 unsigned long nr_uninterruptible; 925 926 struct task_struct __rcu *curr; 927 struct task_struct *idle; 928 struct task_struct *stop; 929 unsigned long next_balance; 930 struct mm_struct *prev_mm; 931 932 unsigned int clock_update_flags; 933 u64 clock; 934 /* Ensure that all clocks are in the same cache line */ 935 u64 clock_task ____cacheline_aligned; 936 u64 clock_pelt; 937 unsigned long lost_idle_time; 938 939 atomic_t nr_iowait; 940 941 #ifdef CONFIG_MEMBARRIER 942 int membarrier_state; 943 #endif 944 945 #ifdef CONFIG_SMP 946 struct root_domain *rd; 947 struct sched_domain __rcu *sd; 948 949 unsigned long cpu_capacity; 950 unsigned long cpu_capacity_orig; 951 952 struct callback_head *balance_callback; 953 954 unsigned char idle_balance; 955 956 unsigned long misfit_task_load; 957 958 /* For active balancing */ 959 int active_balance; 960 int push_cpu; 961 struct cpu_stop_work active_balance_work; 962 963 /* CPU of this runqueue: */ 964 int cpu; 965 int online; 966 967 struct list_head cfs_tasks; 968 969 struct sched_avg avg_rt; 970 struct sched_avg avg_dl; 971 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ 972 struct sched_avg avg_irq; 973 #endif 974 #ifdef CONFIG_SCHED_THERMAL_PRESSURE 975 struct sched_avg avg_thermal; 976 #endif 977 u64 idle_stamp; 978 u64 avg_idle; 979 980 /* This is used to determine avg_idle's max value */ 981 u64 max_idle_balance_cost; 982 #endif 983 984 #ifdef CONFIG_IRQ_TIME_ACCOUNTING 985 u64 prev_irq_time; 986 #endif 987 #ifdef CONFIG_PARAVIRT 988 u64 prev_steal_time; 989 #endif 990 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING 991 u64 prev_steal_time_rq; 992 #endif 993 994 /* calc_load related fields */ 995 unsigned long calc_load_update; 996 long calc_load_active; 997 998 #ifdef CONFIG_SCHED_HRTICK 999 #ifdef CONFIG_SMP 1000 call_single_data_t hrtick_csd; 1001 #endif 1002 struct hrtimer hrtick_timer; 1003 #endif 1004 1005 #ifdef CONFIG_SCHEDSTATS 1006 /* latency stats */ 1007 struct sched_info rq_sched_info; 1008 unsigned long long rq_cpu_time; 1009 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ 1010 1011 /* sys_sched_yield() stats */ 1012 unsigned int yld_count; 1013 1014 /* schedule() stats */ 1015 unsigned int sched_count; 1016 unsigned int sched_goidle; 1017 1018 /* try_to_wake_up() stats */ 1019 unsigned int ttwu_count; 1020 unsigned int ttwu_local; 1021 #endif 1022 1023 #ifdef CONFIG_SMP 1024 struct llist_head wake_list; 1025 #endif 1026 1027 #ifdef CONFIG_CPU_IDLE 1028 /* Must be inspected within a rcu lock section */ 1029 struct cpuidle_state *idle_state; 1030 #endif 1031 }; 1032 1033 #ifdef CONFIG_FAIR_GROUP_SCHED 1034 1035 /* CPU runqueue to which this cfs_rq is attached */ 1036 static inline struct rq *rq_of(struct cfs_rq *cfs_rq) 1037 { 1038 return cfs_rq->rq; 1039 } 1040 1041 #else 1042 1043 static inline struct rq *rq_of(struct cfs_rq *cfs_rq) 1044 { 1045 return container_of(cfs_rq, struct rq, cfs); 1046 } 1047 #endif 1048 1049 static inline int cpu_of(struct rq *rq) 1050 { 1051 #ifdef CONFIG_SMP 1052 return rq->cpu; 1053 #else 1054 return 0; 1055 #endif 1056 } 1057 1058 1059 #ifdef CONFIG_SCHED_SMT 1060 extern void __update_idle_core(struct rq *rq); 1061 1062 static inline void update_idle_core(struct rq *rq) 1063 { 1064 if (static_branch_unlikely(&sched_smt_present)) 1065 __update_idle_core(rq); 1066 } 1067 1068 #else 1069 static inline void update_idle_core(struct rq *rq) { } 1070 #endif 1071 1072 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); 1073 1074 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) 1075 #define this_rq() this_cpu_ptr(&runqueues) 1076 #define task_rq(p) cpu_rq(task_cpu(p)) 1077 #define cpu_curr(cpu) (cpu_rq(cpu)->curr) 1078 #define raw_rq() raw_cpu_ptr(&runqueues) 1079 1080 extern void update_rq_clock(struct rq *rq); 1081 1082 static inline u64 __rq_clock_broken(struct rq *rq) 1083 { 1084 return READ_ONCE(rq->clock); 1085 } 1086 1087 /* 1088 * rq::clock_update_flags bits 1089 * 1090 * %RQCF_REQ_SKIP - will request skipping of clock update on the next 1091 * call to __schedule(). This is an optimisation to avoid 1092 * neighbouring rq clock updates. 1093 * 1094 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is 1095 * in effect and calls to update_rq_clock() are being ignored. 1096 * 1097 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been 1098 * made to update_rq_clock() since the last time rq::lock was pinned. 1099 * 1100 * If inside of __schedule(), clock_update_flags will have been 1101 * shifted left (a left shift is a cheap operation for the fast path 1102 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use, 1103 * 1104 * if (rq-clock_update_flags >= RQCF_UPDATED) 1105 * 1106 * to check if %RQCF_UPADTED is set. It'll never be shifted more than 1107 * one position though, because the next rq_unpin_lock() will shift it 1108 * back. 1109 */ 1110 #define RQCF_REQ_SKIP 0x01 1111 #define RQCF_ACT_SKIP 0x02 1112 #define RQCF_UPDATED 0x04 1113 1114 static inline void assert_clock_updated(struct rq *rq) 1115 { 1116 /* 1117 * The only reason for not seeing a clock update since the 1118 * last rq_pin_lock() is if we're currently skipping updates. 1119 */ 1120 SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP); 1121 } 1122 1123 static inline u64 rq_clock(struct rq *rq) 1124 { 1125 lockdep_assert_held(&rq->lock); 1126 assert_clock_updated(rq); 1127 1128 return rq->clock; 1129 } 1130 1131 static inline u64 rq_clock_task(struct rq *rq) 1132 { 1133 lockdep_assert_held(&rq->lock); 1134 assert_clock_updated(rq); 1135 1136 return rq->clock_task; 1137 } 1138 1139 /** 1140 * By default the decay is the default pelt decay period. 1141 * The decay shift can change the decay period in 1142 * multiples of 32. 1143 * Decay shift Decay period(ms) 1144 * 0 32 1145 * 1 64 1146 * 2 128 1147 * 3 256 1148 * 4 512 1149 */ 1150 extern int sched_thermal_decay_shift; 1151 1152 static inline u64 rq_clock_thermal(struct rq *rq) 1153 { 1154 return rq_clock_task(rq) >> sched_thermal_decay_shift; 1155 } 1156 1157 static inline void rq_clock_skip_update(struct rq *rq) 1158 { 1159 lockdep_assert_held(&rq->lock); 1160 rq->clock_update_flags |= RQCF_REQ_SKIP; 1161 } 1162 1163 /* 1164 * See rt task throttling, which is the only time a skip 1165 * request is cancelled. 1166 */ 1167 static inline void rq_clock_cancel_skipupdate(struct rq *rq) 1168 { 1169 lockdep_assert_held(&rq->lock); 1170 rq->clock_update_flags &= ~RQCF_REQ_SKIP; 1171 } 1172 1173 struct rq_flags { 1174 unsigned long flags; 1175 struct pin_cookie cookie; 1176 #ifdef CONFIG_SCHED_DEBUG 1177 /* 1178 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the 1179 * current pin context is stashed here in case it needs to be 1180 * restored in rq_repin_lock(). 1181 */ 1182 unsigned int clock_update_flags; 1183 #endif 1184 }; 1185 1186 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf) 1187 { 1188 rf->cookie = lockdep_pin_lock(&rq->lock); 1189 1190 #ifdef CONFIG_SCHED_DEBUG 1191 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP); 1192 rf->clock_update_flags = 0; 1193 #endif 1194 } 1195 1196 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf) 1197 { 1198 #ifdef CONFIG_SCHED_DEBUG 1199 if (rq->clock_update_flags > RQCF_ACT_SKIP) 1200 rf->clock_update_flags = RQCF_UPDATED; 1201 #endif 1202 1203 lockdep_unpin_lock(&rq->lock, rf->cookie); 1204 } 1205 1206 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf) 1207 { 1208 lockdep_repin_lock(&rq->lock, rf->cookie); 1209 1210 #ifdef CONFIG_SCHED_DEBUG 1211 /* 1212 * Restore the value we stashed in @rf for this pin context. 1213 */ 1214 rq->clock_update_flags |= rf->clock_update_flags; 1215 #endif 1216 } 1217 1218 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf) 1219 __acquires(rq->lock); 1220 1221 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf) 1222 __acquires(p->pi_lock) 1223 __acquires(rq->lock); 1224 1225 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf) 1226 __releases(rq->lock) 1227 { 1228 rq_unpin_lock(rq, rf); 1229 raw_spin_unlock(&rq->lock); 1230 } 1231 1232 static inline void 1233 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf) 1234 __releases(rq->lock) 1235 __releases(p->pi_lock) 1236 { 1237 rq_unpin_lock(rq, rf); 1238 raw_spin_unlock(&rq->lock); 1239 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags); 1240 } 1241 1242 static inline void 1243 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf) 1244 __acquires(rq->lock) 1245 { 1246 raw_spin_lock_irqsave(&rq->lock, rf->flags); 1247 rq_pin_lock(rq, rf); 1248 } 1249 1250 static inline void 1251 rq_lock_irq(struct rq *rq, struct rq_flags *rf) 1252 __acquires(rq->lock) 1253 { 1254 raw_spin_lock_irq(&rq->lock); 1255 rq_pin_lock(rq, rf); 1256 } 1257 1258 static inline void 1259 rq_lock(struct rq *rq, struct rq_flags *rf) 1260 __acquires(rq->lock) 1261 { 1262 raw_spin_lock(&rq->lock); 1263 rq_pin_lock(rq, rf); 1264 } 1265 1266 static inline void 1267 rq_relock(struct rq *rq, struct rq_flags *rf) 1268 __acquires(rq->lock) 1269 { 1270 raw_spin_lock(&rq->lock); 1271 rq_repin_lock(rq, rf); 1272 } 1273 1274 static inline void 1275 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf) 1276 __releases(rq->lock) 1277 { 1278 rq_unpin_lock(rq, rf); 1279 raw_spin_unlock_irqrestore(&rq->lock, rf->flags); 1280 } 1281 1282 static inline void 1283 rq_unlock_irq(struct rq *rq, struct rq_flags *rf) 1284 __releases(rq->lock) 1285 { 1286 rq_unpin_lock(rq, rf); 1287 raw_spin_unlock_irq(&rq->lock); 1288 } 1289 1290 static inline void 1291 rq_unlock(struct rq *rq, struct rq_flags *rf) 1292 __releases(rq->lock) 1293 { 1294 rq_unpin_lock(rq, rf); 1295 raw_spin_unlock(&rq->lock); 1296 } 1297 1298 static inline struct rq * 1299 this_rq_lock_irq(struct rq_flags *rf) 1300 __acquires(rq->lock) 1301 { 1302 struct rq *rq; 1303 1304 local_irq_disable(); 1305 rq = this_rq(); 1306 rq_lock(rq, rf); 1307 return rq; 1308 } 1309 1310 #ifdef CONFIG_NUMA 1311 enum numa_topology_type { 1312 NUMA_DIRECT, 1313 NUMA_GLUELESS_MESH, 1314 NUMA_BACKPLANE, 1315 }; 1316 extern enum numa_topology_type sched_numa_topology_type; 1317 extern int sched_max_numa_distance; 1318 extern bool find_numa_distance(int distance); 1319 extern void sched_init_numa(void); 1320 extern void sched_domains_numa_masks_set(unsigned int cpu); 1321 extern void sched_domains_numa_masks_clear(unsigned int cpu); 1322 extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu); 1323 #else 1324 static inline void sched_init_numa(void) { } 1325 static inline void sched_domains_numa_masks_set(unsigned int cpu) { } 1326 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { } 1327 static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu) 1328 { 1329 return nr_cpu_ids; 1330 } 1331 #endif 1332 1333 #ifdef CONFIG_NUMA_BALANCING 1334 /* The regions in numa_faults array from task_struct */ 1335 enum numa_faults_stats { 1336 NUMA_MEM = 0, 1337 NUMA_CPU, 1338 NUMA_MEMBUF, 1339 NUMA_CPUBUF 1340 }; 1341 extern void sched_setnuma(struct task_struct *p, int node); 1342 extern int migrate_task_to(struct task_struct *p, int cpu); 1343 extern int migrate_swap(struct task_struct *p, struct task_struct *t, 1344 int cpu, int scpu); 1345 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p); 1346 #else 1347 static inline void 1348 init_numa_balancing(unsigned long clone_flags, struct task_struct *p) 1349 { 1350 } 1351 #endif /* CONFIG_NUMA_BALANCING */ 1352 1353 #ifdef CONFIG_SMP 1354 1355 static inline void 1356 queue_balance_callback(struct rq *rq, 1357 struct callback_head *head, 1358 void (*func)(struct rq *rq)) 1359 { 1360 lockdep_assert_held(&rq->lock); 1361 1362 if (unlikely(head->next)) 1363 return; 1364 1365 head->func = (void (*)(struct callback_head *))func; 1366 head->next = rq->balance_callback; 1367 rq->balance_callback = head; 1368 } 1369 1370 extern void sched_ttwu_pending(void); 1371 1372 #define rcu_dereference_check_sched_domain(p) \ 1373 rcu_dereference_check((p), \ 1374 lockdep_is_held(&sched_domains_mutex)) 1375 1376 /* 1377 * The domain tree (rq->sd) is protected by RCU's quiescent state transition. 1378 * See destroy_sched_domains: call_rcu for details. 1379 * 1380 * The domain tree of any CPU may only be accessed from within 1381 * preempt-disabled sections. 1382 */ 1383 #define for_each_domain(cpu, __sd) \ 1384 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \ 1385 __sd; __sd = __sd->parent) 1386 1387 /** 1388 * highest_flag_domain - Return highest sched_domain containing flag. 1389 * @cpu: The CPU whose highest level of sched domain is to 1390 * be returned. 1391 * @flag: The flag to check for the highest sched_domain 1392 * for the given CPU. 1393 * 1394 * Returns the highest sched_domain of a CPU which contains the given flag. 1395 */ 1396 static inline struct sched_domain *highest_flag_domain(int cpu, int flag) 1397 { 1398 struct sched_domain *sd, *hsd = NULL; 1399 1400 for_each_domain(cpu, sd) { 1401 if (!(sd->flags & flag)) 1402 break; 1403 hsd = sd; 1404 } 1405 1406 return hsd; 1407 } 1408 1409 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag) 1410 { 1411 struct sched_domain *sd; 1412 1413 for_each_domain(cpu, sd) { 1414 if (sd->flags & flag) 1415 break; 1416 } 1417 1418 return sd; 1419 } 1420 1421 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc); 1422 DECLARE_PER_CPU(int, sd_llc_size); 1423 DECLARE_PER_CPU(int, sd_llc_id); 1424 DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared); 1425 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa); 1426 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing); 1427 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity); 1428 extern struct static_key_false sched_asym_cpucapacity; 1429 1430 struct sched_group_capacity { 1431 atomic_t ref; 1432 /* 1433 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity 1434 * for a single CPU. 1435 */ 1436 unsigned long capacity; 1437 unsigned long min_capacity; /* Min per-CPU capacity in group */ 1438 unsigned long max_capacity; /* Max per-CPU capacity in group */ 1439 unsigned long next_update; 1440 int imbalance; /* XXX unrelated to capacity but shared group state */ 1441 1442 #ifdef CONFIG_SCHED_DEBUG 1443 int id; 1444 #endif 1445 1446 unsigned long cpumask[0]; /* Balance mask */ 1447 }; 1448 1449 struct sched_group { 1450 struct sched_group *next; /* Must be a circular list */ 1451 atomic_t ref; 1452 1453 unsigned int group_weight; 1454 struct sched_group_capacity *sgc; 1455 int asym_prefer_cpu; /* CPU of highest priority in group */ 1456 1457 /* 1458 * The CPUs this group covers. 1459 * 1460 * NOTE: this field is variable length. (Allocated dynamically 1461 * by attaching extra space to the end of the structure, 1462 * depending on how many CPUs the kernel has booted up with) 1463 */ 1464 unsigned long cpumask[0]; 1465 }; 1466 1467 static inline struct cpumask *sched_group_span(struct sched_group *sg) 1468 { 1469 return to_cpumask(sg->cpumask); 1470 } 1471 1472 /* 1473 * See build_balance_mask(). 1474 */ 1475 static inline struct cpumask *group_balance_mask(struct sched_group *sg) 1476 { 1477 return to_cpumask(sg->sgc->cpumask); 1478 } 1479 1480 /** 1481 * group_first_cpu - Returns the first CPU in the cpumask of a sched_group. 1482 * @group: The group whose first CPU is to be returned. 1483 */ 1484 static inline unsigned int group_first_cpu(struct sched_group *group) 1485 { 1486 return cpumask_first(sched_group_span(group)); 1487 } 1488 1489 extern int group_balance_cpu(struct sched_group *sg); 1490 1491 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) 1492 void register_sched_domain_sysctl(void); 1493 void dirty_sched_domain_sysctl(int cpu); 1494 void unregister_sched_domain_sysctl(void); 1495 #else 1496 static inline void register_sched_domain_sysctl(void) 1497 { 1498 } 1499 static inline void dirty_sched_domain_sysctl(int cpu) 1500 { 1501 } 1502 static inline void unregister_sched_domain_sysctl(void) 1503 { 1504 } 1505 #endif 1506 1507 extern int newidle_balance(struct rq *this_rq, struct rq_flags *rf); 1508 1509 #else 1510 1511 static inline void sched_ttwu_pending(void) { } 1512 1513 static inline int newidle_balance(struct rq *this_rq, struct rq_flags *rf) { return 0; } 1514 1515 #endif /* CONFIG_SMP */ 1516 1517 #include "stats.h" 1518 #include "autogroup.h" 1519 1520 #ifdef CONFIG_CGROUP_SCHED 1521 1522 /* 1523 * Return the group to which this tasks belongs. 1524 * 1525 * We cannot use task_css() and friends because the cgroup subsystem 1526 * changes that value before the cgroup_subsys::attach() method is called, 1527 * therefore we cannot pin it and might observe the wrong value. 1528 * 1529 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup 1530 * core changes this before calling sched_move_task(). 1531 * 1532 * Instead we use a 'copy' which is updated from sched_move_task() while 1533 * holding both task_struct::pi_lock and rq::lock. 1534 */ 1535 static inline struct task_group *task_group(struct task_struct *p) 1536 { 1537 return p->sched_task_group; 1538 } 1539 1540 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ 1541 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) 1542 { 1543 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED) 1544 struct task_group *tg = task_group(p); 1545 #endif 1546 1547 #ifdef CONFIG_FAIR_GROUP_SCHED 1548 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]); 1549 p->se.cfs_rq = tg->cfs_rq[cpu]; 1550 p->se.parent = tg->se[cpu]; 1551 #endif 1552 1553 #ifdef CONFIG_RT_GROUP_SCHED 1554 p->rt.rt_rq = tg->rt_rq[cpu]; 1555 p->rt.parent = tg->rt_se[cpu]; 1556 #endif 1557 } 1558 1559 #else /* CONFIG_CGROUP_SCHED */ 1560 1561 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } 1562 static inline struct task_group *task_group(struct task_struct *p) 1563 { 1564 return NULL; 1565 } 1566 1567 #endif /* CONFIG_CGROUP_SCHED */ 1568 1569 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) 1570 { 1571 set_task_rq(p, cpu); 1572 #ifdef CONFIG_SMP 1573 /* 1574 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be 1575 * successfully executed on another CPU. We must ensure that updates of 1576 * per-task data have been completed by this moment. 1577 */ 1578 smp_wmb(); 1579 #ifdef CONFIG_THREAD_INFO_IN_TASK 1580 WRITE_ONCE(p->cpu, cpu); 1581 #else 1582 WRITE_ONCE(task_thread_info(p)->cpu, cpu); 1583 #endif 1584 p->wake_cpu = cpu; 1585 #endif 1586 } 1587 1588 /* 1589 * Tunables that become constants when CONFIG_SCHED_DEBUG is off: 1590 */ 1591 #ifdef CONFIG_SCHED_DEBUG 1592 # include <linux/static_key.h> 1593 # define const_debug __read_mostly 1594 #else 1595 # define const_debug const 1596 #endif 1597 1598 #define SCHED_FEAT(name, enabled) \ 1599 __SCHED_FEAT_##name , 1600 1601 enum { 1602 #include "features.h" 1603 __SCHED_FEAT_NR, 1604 }; 1605 1606 #undef SCHED_FEAT 1607 1608 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_JUMP_LABEL) 1609 1610 /* 1611 * To support run-time toggling of sched features, all the translation units 1612 * (but core.c) reference the sysctl_sched_features defined in core.c. 1613 */ 1614 extern const_debug unsigned int sysctl_sched_features; 1615 1616 #define SCHED_FEAT(name, enabled) \ 1617 static __always_inline bool static_branch_##name(struct static_key *key) \ 1618 { \ 1619 return static_key_##enabled(key); \ 1620 } 1621 1622 #include "features.h" 1623 #undef SCHED_FEAT 1624 1625 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR]; 1626 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x])) 1627 1628 #else /* !(SCHED_DEBUG && CONFIG_JUMP_LABEL) */ 1629 1630 /* 1631 * Each translation unit has its own copy of sysctl_sched_features to allow 1632 * constants propagation at compile time and compiler optimization based on 1633 * features default. 1634 */ 1635 #define SCHED_FEAT(name, enabled) \ 1636 (1UL << __SCHED_FEAT_##name) * enabled | 1637 static const_debug __maybe_unused unsigned int sysctl_sched_features = 1638 #include "features.h" 1639 0; 1640 #undef SCHED_FEAT 1641 1642 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) 1643 1644 #endif /* SCHED_DEBUG && CONFIG_JUMP_LABEL */ 1645 1646 extern struct static_key_false sched_numa_balancing; 1647 extern struct static_key_false sched_schedstats; 1648 1649 static inline u64 global_rt_period(void) 1650 { 1651 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; 1652 } 1653 1654 static inline u64 global_rt_runtime(void) 1655 { 1656 if (sysctl_sched_rt_runtime < 0) 1657 return RUNTIME_INF; 1658 1659 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; 1660 } 1661 1662 static inline int task_current(struct rq *rq, struct task_struct *p) 1663 { 1664 return rq->curr == p; 1665 } 1666 1667 static inline int task_running(struct rq *rq, struct task_struct *p) 1668 { 1669 #ifdef CONFIG_SMP 1670 return p->on_cpu; 1671 #else 1672 return task_current(rq, p); 1673 #endif 1674 } 1675 1676 static inline int task_on_rq_queued(struct task_struct *p) 1677 { 1678 return p->on_rq == TASK_ON_RQ_QUEUED; 1679 } 1680 1681 static inline int task_on_rq_migrating(struct task_struct *p) 1682 { 1683 return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING; 1684 } 1685 1686 /* 1687 * wake flags 1688 */ 1689 #define WF_SYNC 0x01 /* Waker goes to sleep after wakeup */ 1690 #define WF_FORK 0x02 /* Child wakeup after fork */ 1691 #define WF_MIGRATED 0x4 /* Internal use, task got migrated */ 1692 1693 /* 1694 * To aid in avoiding the subversion of "niceness" due to uneven distribution 1695 * of tasks with abnormal "nice" values across CPUs the contribution that 1696 * each task makes to its run queue's load is weighted according to its 1697 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a 1698 * scaled version of the new time slice allocation that they receive on time 1699 * slice expiry etc. 1700 */ 1701 1702 #define WEIGHT_IDLEPRIO 3 1703 #define WMULT_IDLEPRIO 1431655765 1704 1705 extern const int sched_prio_to_weight[40]; 1706 extern const u32 sched_prio_to_wmult[40]; 1707 1708 /* 1709 * {de,en}queue flags: 1710 * 1711 * DEQUEUE_SLEEP - task is no longer runnable 1712 * ENQUEUE_WAKEUP - task just became runnable 1713 * 1714 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks 1715 * are in a known state which allows modification. Such pairs 1716 * should preserve as much state as possible. 1717 * 1718 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location 1719 * in the runqueue. 1720 * 1721 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified) 1722 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline) 1723 * ENQUEUE_MIGRATED - the task was migrated during wakeup 1724 * 1725 */ 1726 1727 #define DEQUEUE_SLEEP 0x01 1728 #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */ 1729 #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */ 1730 #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */ 1731 1732 #define ENQUEUE_WAKEUP 0x01 1733 #define ENQUEUE_RESTORE 0x02 1734 #define ENQUEUE_MOVE 0x04 1735 #define ENQUEUE_NOCLOCK 0x08 1736 1737 #define ENQUEUE_HEAD 0x10 1738 #define ENQUEUE_REPLENISH 0x20 1739 #ifdef CONFIG_SMP 1740 #define ENQUEUE_MIGRATED 0x40 1741 #else 1742 #define ENQUEUE_MIGRATED 0x00 1743 #endif 1744 1745 #define RETRY_TASK ((void *)-1UL) 1746 1747 struct sched_class { 1748 const struct sched_class *next; 1749 1750 #ifdef CONFIG_UCLAMP_TASK 1751 int uclamp_enabled; 1752 #endif 1753 1754 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags); 1755 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags); 1756 void (*yield_task) (struct rq *rq); 1757 bool (*yield_to_task)(struct rq *rq, struct task_struct *p, bool preempt); 1758 1759 void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags); 1760 1761 struct task_struct *(*pick_next_task)(struct rq *rq); 1762 1763 void (*put_prev_task)(struct rq *rq, struct task_struct *p); 1764 void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first); 1765 1766 #ifdef CONFIG_SMP 1767 int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf); 1768 int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags); 1769 void (*migrate_task_rq)(struct task_struct *p, int new_cpu); 1770 1771 void (*task_woken)(struct rq *this_rq, struct task_struct *task); 1772 1773 void (*set_cpus_allowed)(struct task_struct *p, 1774 const struct cpumask *newmask); 1775 1776 void (*rq_online)(struct rq *rq); 1777 void (*rq_offline)(struct rq *rq); 1778 #endif 1779 1780 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued); 1781 void (*task_fork)(struct task_struct *p); 1782 void (*task_dead)(struct task_struct *p); 1783 1784 /* 1785 * The switched_from() call is allowed to drop rq->lock, therefore we 1786 * cannot assume the switched_from/switched_to pair is serliazed by 1787 * rq->lock. They are however serialized by p->pi_lock. 1788 */ 1789 void (*switched_from)(struct rq *this_rq, struct task_struct *task); 1790 void (*switched_to) (struct rq *this_rq, struct task_struct *task); 1791 void (*prio_changed) (struct rq *this_rq, struct task_struct *task, 1792 int oldprio); 1793 1794 unsigned int (*get_rr_interval)(struct rq *rq, 1795 struct task_struct *task); 1796 1797 void (*update_curr)(struct rq *rq); 1798 1799 #define TASK_SET_GROUP 0 1800 #define TASK_MOVE_GROUP 1 1801 1802 #ifdef CONFIG_FAIR_GROUP_SCHED 1803 void (*task_change_group)(struct task_struct *p, int type); 1804 #endif 1805 }; 1806 1807 static inline void put_prev_task(struct rq *rq, struct task_struct *prev) 1808 { 1809 WARN_ON_ONCE(rq->curr != prev); 1810 prev->sched_class->put_prev_task(rq, prev); 1811 } 1812 1813 static inline void set_next_task(struct rq *rq, struct task_struct *next) 1814 { 1815 WARN_ON_ONCE(rq->curr != next); 1816 next->sched_class->set_next_task(rq, next, false); 1817 } 1818 1819 #ifdef CONFIG_SMP 1820 #define sched_class_highest (&stop_sched_class) 1821 #else 1822 #define sched_class_highest (&dl_sched_class) 1823 #endif 1824 1825 #define for_class_range(class, _from, _to) \ 1826 for (class = (_from); class != (_to); class = class->next) 1827 1828 #define for_each_class(class) \ 1829 for_class_range(class, sched_class_highest, NULL) 1830 1831 extern const struct sched_class stop_sched_class; 1832 extern const struct sched_class dl_sched_class; 1833 extern const struct sched_class rt_sched_class; 1834 extern const struct sched_class fair_sched_class; 1835 extern const struct sched_class idle_sched_class; 1836 1837 static inline bool sched_stop_runnable(struct rq *rq) 1838 { 1839 return rq->stop && task_on_rq_queued(rq->stop); 1840 } 1841 1842 static inline bool sched_dl_runnable(struct rq *rq) 1843 { 1844 return rq->dl.dl_nr_running > 0; 1845 } 1846 1847 static inline bool sched_rt_runnable(struct rq *rq) 1848 { 1849 return rq->rt.rt_queued > 0; 1850 } 1851 1852 static inline bool sched_fair_runnable(struct rq *rq) 1853 { 1854 return rq->cfs.nr_running > 0; 1855 } 1856 1857 extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf); 1858 extern struct task_struct *pick_next_task_idle(struct rq *rq); 1859 1860 #ifdef CONFIG_SMP 1861 1862 extern void update_group_capacity(struct sched_domain *sd, int cpu); 1863 1864 extern void trigger_load_balance(struct rq *rq); 1865 1866 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask); 1867 1868 #endif 1869 1870 #ifdef CONFIG_CPU_IDLE 1871 static inline void idle_set_state(struct rq *rq, 1872 struct cpuidle_state *idle_state) 1873 { 1874 rq->idle_state = idle_state; 1875 } 1876 1877 static inline struct cpuidle_state *idle_get_state(struct rq *rq) 1878 { 1879 SCHED_WARN_ON(!rcu_read_lock_held()); 1880 1881 return rq->idle_state; 1882 } 1883 #else 1884 static inline void idle_set_state(struct rq *rq, 1885 struct cpuidle_state *idle_state) 1886 { 1887 } 1888 1889 static inline struct cpuidle_state *idle_get_state(struct rq *rq) 1890 { 1891 return NULL; 1892 } 1893 #endif 1894 1895 extern void schedule_idle(void); 1896 1897 extern void sysrq_sched_debug_show(void); 1898 extern void sched_init_granularity(void); 1899 extern void update_max_interval(void); 1900 1901 extern void init_sched_dl_class(void); 1902 extern void init_sched_rt_class(void); 1903 extern void init_sched_fair_class(void); 1904 1905 extern void reweight_task(struct task_struct *p, int prio); 1906 1907 extern void resched_curr(struct rq *rq); 1908 extern void resched_cpu(int cpu); 1909 1910 extern struct rt_bandwidth def_rt_bandwidth; 1911 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime); 1912 1913 extern struct dl_bandwidth def_dl_bandwidth; 1914 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime); 1915 extern void init_dl_task_timer(struct sched_dl_entity *dl_se); 1916 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se); 1917 1918 #define BW_SHIFT 20 1919 #define BW_UNIT (1 << BW_SHIFT) 1920 #define RATIO_SHIFT 8 1921 unsigned long to_ratio(u64 period, u64 runtime); 1922 1923 extern void init_entity_runnable_average(struct sched_entity *se); 1924 extern void post_init_entity_util_avg(struct task_struct *p); 1925 1926 #ifdef CONFIG_NO_HZ_FULL 1927 extern bool sched_can_stop_tick(struct rq *rq); 1928 extern int __init sched_tick_offload_init(void); 1929 1930 /* 1931 * Tick may be needed by tasks in the runqueue depending on their policy and 1932 * requirements. If tick is needed, lets send the target an IPI to kick it out of 1933 * nohz mode if necessary. 1934 */ 1935 static inline void sched_update_tick_dependency(struct rq *rq) 1936 { 1937 int cpu; 1938 1939 if (!tick_nohz_full_enabled()) 1940 return; 1941 1942 cpu = cpu_of(rq); 1943 1944 if (!tick_nohz_full_cpu(cpu)) 1945 return; 1946 1947 if (sched_can_stop_tick(rq)) 1948 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED); 1949 else 1950 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED); 1951 } 1952 #else 1953 static inline int sched_tick_offload_init(void) { return 0; } 1954 static inline void sched_update_tick_dependency(struct rq *rq) { } 1955 #endif 1956 1957 static inline void add_nr_running(struct rq *rq, unsigned count) 1958 { 1959 unsigned prev_nr = rq->nr_running; 1960 1961 rq->nr_running = prev_nr + count; 1962 1963 #ifdef CONFIG_SMP 1964 if (prev_nr < 2 && rq->nr_running >= 2) { 1965 if (!READ_ONCE(rq->rd->overload)) 1966 WRITE_ONCE(rq->rd->overload, 1); 1967 } 1968 #endif 1969 1970 sched_update_tick_dependency(rq); 1971 } 1972 1973 static inline void sub_nr_running(struct rq *rq, unsigned count) 1974 { 1975 rq->nr_running -= count; 1976 /* Check if we still need preemption */ 1977 sched_update_tick_dependency(rq); 1978 } 1979 1980 extern void activate_task(struct rq *rq, struct task_struct *p, int flags); 1981 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags); 1982 1983 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags); 1984 1985 extern const_debug unsigned int sysctl_sched_nr_migrate; 1986 extern const_debug unsigned int sysctl_sched_migration_cost; 1987 1988 #ifdef CONFIG_SCHED_HRTICK 1989 1990 /* 1991 * Use hrtick when: 1992 * - enabled by features 1993 * - hrtimer is actually high res 1994 */ 1995 static inline int hrtick_enabled(struct rq *rq) 1996 { 1997 if (!sched_feat(HRTICK)) 1998 return 0; 1999 if (!cpu_active(cpu_of(rq))) 2000 return 0; 2001 return hrtimer_is_hres_active(&rq->hrtick_timer); 2002 } 2003 2004 void hrtick_start(struct rq *rq, u64 delay); 2005 2006 #else 2007 2008 static inline int hrtick_enabled(struct rq *rq) 2009 { 2010 return 0; 2011 } 2012 2013 #endif /* CONFIG_SCHED_HRTICK */ 2014 2015 #ifndef arch_scale_freq_tick 2016 static __always_inline 2017 void arch_scale_freq_tick(void) 2018 { 2019 } 2020 #endif 2021 2022 #ifndef arch_scale_freq_capacity 2023 static __always_inline 2024 unsigned long arch_scale_freq_capacity(int cpu) 2025 { 2026 return SCHED_CAPACITY_SCALE; 2027 } 2028 #endif 2029 2030 #ifdef CONFIG_SMP 2031 #ifdef CONFIG_PREEMPTION 2032 2033 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2); 2034 2035 /* 2036 * fair double_lock_balance: Safely acquires both rq->locks in a fair 2037 * way at the expense of forcing extra atomic operations in all 2038 * invocations. This assures that the double_lock is acquired using the 2039 * same underlying policy as the spinlock_t on this architecture, which 2040 * reduces latency compared to the unfair variant below. However, it 2041 * also adds more overhead and therefore may reduce throughput. 2042 */ 2043 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) 2044 __releases(this_rq->lock) 2045 __acquires(busiest->lock) 2046 __acquires(this_rq->lock) 2047 { 2048 raw_spin_unlock(&this_rq->lock); 2049 double_rq_lock(this_rq, busiest); 2050 2051 return 1; 2052 } 2053 2054 #else 2055 /* 2056 * Unfair double_lock_balance: Optimizes throughput at the expense of 2057 * latency by eliminating extra atomic operations when the locks are 2058 * already in proper order on entry. This favors lower CPU-ids and will 2059 * grant the double lock to lower CPUs over higher ids under contention, 2060 * regardless of entry order into the function. 2061 */ 2062 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) 2063 __releases(this_rq->lock) 2064 __acquires(busiest->lock) 2065 __acquires(this_rq->lock) 2066 { 2067 int ret = 0; 2068 2069 if (unlikely(!raw_spin_trylock(&busiest->lock))) { 2070 if (busiest < this_rq) { 2071 raw_spin_unlock(&this_rq->lock); 2072 raw_spin_lock(&busiest->lock); 2073 raw_spin_lock_nested(&this_rq->lock, 2074 SINGLE_DEPTH_NESTING); 2075 ret = 1; 2076 } else 2077 raw_spin_lock_nested(&busiest->lock, 2078 SINGLE_DEPTH_NESTING); 2079 } 2080 return ret; 2081 } 2082 2083 #endif /* CONFIG_PREEMPTION */ 2084 2085 /* 2086 * double_lock_balance - lock the busiest runqueue, this_rq is locked already. 2087 */ 2088 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest) 2089 { 2090 if (unlikely(!irqs_disabled())) { 2091 /* printk() doesn't work well under rq->lock */ 2092 raw_spin_unlock(&this_rq->lock); 2093 BUG_ON(1); 2094 } 2095 2096 return _double_lock_balance(this_rq, busiest); 2097 } 2098 2099 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) 2100 __releases(busiest->lock) 2101 { 2102 raw_spin_unlock(&busiest->lock); 2103 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); 2104 } 2105 2106 static inline void double_lock(spinlock_t *l1, spinlock_t *l2) 2107 { 2108 if (l1 > l2) 2109 swap(l1, l2); 2110 2111 spin_lock(l1); 2112 spin_lock_nested(l2, SINGLE_DEPTH_NESTING); 2113 } 2114 2115 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2) 2116 { 2117 if (l1 > l2) 2118 swap(l1, l2); 2119 2120 spin_lock_irq(l1); 2121 spin_lock_nested(l2, SINGLE_DEPTH_NESTING); 2122 } 2123 2124 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2) 2125 { 2126 if (l1 > l2) 2127 swap(l1, l2); 2128 2129 raw_spin_lock(l1); 2130 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING); 2131 } 2132 2133 /* 2134 * double_rq_lock - safely lock two runqueues 2135 * 2136 * Note this does not disable interrupts like task_rq_lock, 2137 * you need to do so manually before calling. 2138 */ 2139 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) 2140 __acquires(rq1->lock) 2141 __acquires(rq2->lock) 2142 { 2143 BUG_ON(!irqs_disabled()); 2144 if (rq1 == rq2) { 2145 raw_spin_lock(&rq1->lock); 2146 __acquire(rq2->lock); /* Fake it out ;) */ 2147 } else { 2148 if (rq1 < rq2) { 2149 raw_spin_lock(&rq1->lock); 2150 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); 2151 } else { 2152 raw_spin_lock(&rq2->lock); 2153 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); 2154 } 2155 } 2156 } 2157 2158 /* 2159 * double_rq_unlock - safely unlock two runqueues 2160 * 2161 * Note this does not restore interrupts like task_rq_unlock, 2162 * you need to do so manually after calling. 2163 */ 2164 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) 2165 __releases(rq1->lock) 2166 __releases(rq2->lock) 2167 { 2168 raw_spin_unlock(&rq1->lock); 2169 if (rq1 != rq2) 2170 raw_spin_unlock(&rq2->lock); 2171 else 2172 __release(rq2->lock); 2173 } 2174 2175 extern void set_rq_online (struct rq *rq); 2176 extern void set_rq_offline(struct rq *rq); 2177 extern bool sched_smp_initialized; 2178 2179 #else /* CONFIG_SMP */ 2180 2181 /* 2182 * double_rq_lock - safely lock two runqueues 2183 * 2184 * Note this does not disable interrupts like task_rq_lock, 2185 * you need to do so manually before calling. 2186 */ 2187 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) 2188 __acquires(rq1->lock) 2189 __acquires(rq2->lock) 2190 { 2191 BUG_ON(!irqs_disabled()); 2192 BUG_ON(rq1 != rq2); 2193 raw_spin_lock(&rq1->lock); 2194 __acquire(rq2->lock); /* Fake it out ;) */ 2195 } 2196 2197 /* 2198 * double_rq_unlock - safely unlock two runqueues 2199 * 2200 * Note this does not restore interrupts like task_rq_unlock, 2201 * you need to do so manually after calling. 2202 */ 2203 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) 2204 __releases(rq1->lock) 2205 __releases(rq2->lock) 2206 { 2207 BUG_ON(rq1 != rq2); 2208 raw_spin_unlock(&rq1->lock); 2209 __release(rq2->lock); 2210 } 2211 2212 #endif 2213 2214 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq); 2215 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq); 2216 2217 #ifdef CONFIG_SCHED_DEBUG 2218 extern bool sched_debug_enabled; 2219 2220 extern void print_cfs_stats(struct seq_file *m, int cpu); 2221 extern void print_rt_stats(struct seq_file *m, int cpu); 2222 extern void print_dl_stats(struct seq_file *m, int cpu); 2223 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq); 2224 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq); 2225 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq); 2226 #ifdef CONFIG_NUMA_BALANCING 2227 extern void 2228 show_numa_stats(struct task_struct *p, struct seq_file *m); 2229 extern void 2230 print_numa_stats(struct seq_file *m, int node, unsigned long tsf, 2231 unsigned long tpf, unsigned long gsf, unsigned long gpf); 2232 #endif /* CONFIG_NUMA_BALANCING */ 2233 #endif /* CONFIG_SCHED_DEBUG */ 2234 2235 extern void init_cfs_rq(struct cfs_rq *cfs_rq); 2236 extern void init_rt_rq(struct rt_rq *rt_rq); 2237 extern void init_dl_rq(struct dl_rq *dl_rq); 2238 2239 extern void cfs_bandwidth_usage_inc(void); 2240 extern void cfs_bandwidth_usage_dec(void); 2241 2242 #ifdef CONFIG_NO_HZ_COMMON 2243 #define NOHZ_BALANCE_KICK_BIT 0 2244 #define NOHZ_STATS_KICK_BIT 1 2245 2246 #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT) 2247 #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT) 2248 2249 #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK) 2250 2251 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags) 2252 2253 extern void nohz_balance_exit_idle(struct rq *rq); 2254 #else 2255 static inline void nohz_balance_exit_idle(struct rq *rq) { } 2256 #endif 2257 2258 2259 #ifdef CONFIG_SMP 2260 static inline 2261 void __dl_update(struct dl_bw *dl_b, s64 bw) 2262 { 2263 struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw); 2264 int i; 2265 2266 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(), 2267 "sched RCU must be held"); 2268 for_each_cpu_and(i, rd->span, cpu_active_mask) { 2269 struct rq *rq = cpu_rq(i); 2270 2271 rq->dl.extra_bw += bw; 2272 } 2273 } 2274 #else 2275 static inline 2276 void __dl_update(struct dl_bw *dl_b, s64 bw) 2277 { 2278 struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw); 2279 2280 dl->extra_bw += bw; 2281 } 2282 #endif 2283 2284 2285 #ifdef CONFIG_IRQ_TIME_ACCOUNTING 2286 struct irqtime { 2287 u64 total; 2288 u64 tick_delta; 2289 u64 irq_start_time; 2290 struct u64_stats_sync sync; 2291 }; 2292 2293 DECLARE_PER_CPU(struct irqtime, cpu_irqtime); 2294 2295 /* 2296 * Returns the irqtime minus the softirq time computed by ksoftirqd. 2297 * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime 2298 * and never move forward. 2299 */ 2300 static inline u64 irq_time_read(int cpu) 2301 { 2302 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu); 2303 unsigned int seq; 2304 u64 total; 2305 2306 do { 2307 seq = __u64_stats_fetch_begin(&irqtime->sync); 2308 total = irqtime->total; 2309 } while (__u64_stats_fetch_retry(&irqtime->sync, seq)); 2310 2311 return total; 2312 } 2313 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ 2314 2315 #ifdef CONFIG_CPU_FREQ 2316 DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data); 2317 2318 /** 2319 * cpufreq_update_util - Take a note about CPU utilization changes. 2320 * @rq: Runqueue to carry out the update for. 2321 * @flags: Update reason flags. 2322 * 2323 * This function is called by the scheduler on the CPU whose utilization is 2324 * being updated. 2325 * 2326 * It can only be called from RCU-sched read-side critical sections. 2327 * 2328 * The way cpufreq is currently arranged requires it to evaluate the CPU 2329 * performance state (frequency/voltage) on a regular basis to prevent it from 2330 * being stuck in a completely inadequate performance level for too long. 2331 * That is not guaranteed to happen if the updates are only triggered from CFS 2332 * and DL, though, because they may not be coming in if only RT tasks are 2333 * active all the time (or there are RT tasks only). 2334 * 2335 * As a workaround for that issue, this function is called periodically by the 2336 * RT sched class to trigger extra cpufreq updates to prevent it from stalling, 2337 * but that really is a band-aid. Going forward it should be replaced with 2338 * solutions targeted more specifically at RT tasks. 2339 */ 2340 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) 2341 { 2342 struct update_util_data *data; 2343 2344 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data, 2345 cpu_of(rq))); 2346 if (data) 2347 data->func(data, rq_clock(rq), flags); 2348 } 2349 #else 2350 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {} 2351 #endif /* CONFIG_CPU_FREQ */ 2352 2353 #ifdef CONFIG_UCLAMP_TASK 2354 unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id); 2355 2356 static __always_inline 2357 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util, 2358 struct task_struct *p) 2359 { 2360 unsigned long min_util = READ_ONCE(rq->uclamp[UCLAMP_MIN].value); 2361 unsigned long max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value); 2362 2363 if (p) { 2364 min_util = max(min_util, uclamp_eff_value(p, UCLAMP_MIN)); 2365 max_util = max(max_util, uclamp_eff_value(p, UCLAMP_MAX)); 2366 } 2367 2368 /* 2369 * Since CPU's {min,max}_util clamps are MAX aggregated considering 2370 * RUNNABLE tasks with _different_ clamps, we can end up with an 2371 * inversion. Fix it now when the clamps are applied. 2372 */ 2373 if (unlikely(min_util >= max_util)) 2374 return min_util; 2375 2376 return clamp(util, min_util, max_util); 2377 } 2378 #else /* CONFIG_UCLAMP_TASK */ 2379 static inline 2380 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util, 2381 struct task_struct *p) 2382 { 2383 return util; 2384 } 2385 #endif /* CONFIG_UCLAMP_TASK */ 2386 2387 #ifdef arch_scale_freq_capacity 2388 # ifndef arch_scale_freq_invariant 2389 # define arch_scale_freq_invariant() true 2390 # endif 2391 #else 2392 # define arch_scale_freq_invariant() false 2393 #endif 2394 2395 #ifdef CONFIG_SMP 2396 static inline unsigned long capacity_orig_of(int cpu) 2397 { 2398 return cpu_rq(cpu)->cpu_capacity_orig; 2399 } 2400 #endif 2401 2402 /** 2403 * enum schedutil_type - CPU utilization type 2404 * @FREQUENCY_UTIL: Utilization used to select frequency 2405 * @ENERGY_UTIL: Utilization used during energy calculation 2406 * 2407 * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time 2408 * need to be aggregated differently depending on the usage made of them. This 2409 * enum is used within schedutil_freq_util() to differentiate the types of 2410 * utilization expected by the callers, and adjust the aggregation accordingly. 2411 */ 2412 enum schedutil_type { 2413 FREQUENCY_UTIL, 2414 ENERGY_UTIL, 2415 }; 2416 2417 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL 2418 2419 unsigned long schedutil_cpu_util(int cpu, unsigned long util_cfs, 2420 unsigned long max, enum schedutil_type type, 2421 struct task_struct *p); 2422 2423 static inline unsigned long cpu_bw_dl(struct rq *rq) 2424 { 2425 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT; 2426 } 2427 2428 static inline unsigned long cpu_util_dl(struct rq *rq) 2429 { 2430 return READ_ONCE(rq->avg_dl.util_avg); 2431 } 2432 2433 static inline unsigned long cpu_util_cfs(struct rq *rq) 2434 { 2435 unsigned long util = READ_ONCE(rq->cfs.avg.util_avg); 2436 2437 if (sched_feat(UTIL_EST)) { 2438 util = max_t(unsigned long, util, 2439 READ_ONCE(rq->cfs.avg.util_est.enqueued)); 2440 } 2441 2442 return util; 2443 } 2444 2445 static inline unsigned long cpu_util_rt(struct rq *rq) 2446 { 2447 return READ_ONCE(rq->avg_rt.util_avg); 2448 } 2449 #else /* CONFIG_CPU_FREQ_GOV_SCHEDUTIL */ 2450 static inline unsigned long schedutil_cpu_util(int cpu, unsigned long util_cfs, 2451 unsigned long max, enum schedutil_type type, 2452 struct task_struct *p) 2453 { 2454 return 0; 2455 } 2456 #endif /* CONFIG_CPU_FREQ_GOV_SCHEDUTIL */ 2457 2458 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ 2459 static inline unsigned long cpu_util_irq(struct rq *rq) 2460 { 2461 return rq->avg_irq.util_avg; 2462 } 2463 2464 static inline 2465 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max) 2466 { 2467 util *= (max - irq); 2468 util /= max; 2469 2470 return util; 2471 2472 } 2473 #else 2474 static inline unsigned long cpu_util_irq(struct rq *rq) 2475 { 2476 return 0; 2477 } 2478 2479 static inline 2480 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max) 2481 { 2482 return util; 2483 } 2484 #endif 2485 2486 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL) 2487 2488 #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus))) 2489 2490 DECLARE_STATIC_KEY_FALSE(sched_energy_present); 2491 2492 static inline bool sched_energy_enabled(void) 2493 { 2494 return static_branch_unlikely(&sched_energy_present); 2495 } 2496 2497 #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */ 2498 2499 #define perf_domain_span(pd) NULL 2500 static inline bool sched_energy_enabled(void) { return false; } 2501 2502 #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */ 2503 2504 #ifdef CONFIG_MEMBARRIER 2505 /* 2506 * The scheduler provides memory barriers required by membarrier between: 2507 * - prior user-space memory accesses and store to rq->membarrier_state, 2508 * - store to rq->membarrier_state and following user-space memory accesses. 2509 * In the same way it provides those guarantees around store to rq->curr. 2510 */ 2511 static inline void membarrier_switch_mm(struct rq *rq, 2512 struct mm_struct *prev_mm, 2513 struct mm_struct *next_mm) 2514 { 2515 int membarrier_state; 2516 2517 if (prev_mm == next_mm) 2518 return; 2519 2520 membarrier_state = atomic_read(&next_mm->membarrier_state); 2521 if (READ_ONCE(rq->membarrier_state) == membarrier_state) 2522 return; 2523 2524 WRITE_ONCE(rq->membarrier_state, membarrier_state); 2525 } 2526 #else 2527 static inline void membarrier_switch_mm(struct rq *rq, 2528 struct mm_struct *prev_mm, 2529 struct mm_struct *next_mm) 2530 { 2531 } 2532 #endif 2533 2534 #ifdef CONFIG_SMP 2535 static inline bool is_per_cpu_kthread(struct task_struct *p) 2536 { 2537 if (!(p->flags & PF_KTHREAD)) 2538 return false; 2539 2540 if (p->nr_cpus_allowed != 1) 2541 return false; 2542 2543 return true; 2544 } 2545 #endif 2546 2547 void swake_up_all_locked(struct swait_queue_head *q); 2548 void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait); 2549