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