FILE
sched.h
File preview with syntax highlighting.
linux-6.19.14/kernel/sched/sched.h
1/* SPDX-License-Identifier: GPL-2.0 */ 2/* 3 * Scheduler internal types and methods: 4 */ 5#ifndef _KERNEL_SCHED_SCHED_H 6#define _KERNEL_SCHED_SCHED_H 7 8#include <linux/prandom.h> 9#include <linux/sched/affinity.h> 10#include <linux/sched/autogroup.h> 11#include <linux/sched/cpufreq.h> 12#include <linux/sched/deadline.h> 13#include <linux/sched.h> 14#include <linux/sched/loadavg.h> 15#include <linux/sched/mm.h> 16#include <linux/sched/rseq_api.h> 17#include <linux/sched/signal.h> 18#include <linux/sched/smt.h> 19#include <linux/sched/stat.h> 20#include <linux/sched/sysctl.h> 21#include <linux/sched/task_flags.h> 22#include <linux/sched/task.h> 23#include <linux/sched/topology.h> 24#include <linux/atomic.h> 25#include <linux/bitmap.h> 26#include <linux/bug.h> 27#include <linux/capability.h> 28#include <linux/cgroup_api.h> 29#include <linux/cgroup.h> 30#include <linux/context_tracking.h> 31#include <linux/cpufreq.h> 32#include <linux/cpumask_api.h> 33#include <linux/ctype.h> 34#include <linux/file.h> 35#include <linux/fs_api.h> 36#include <linux/hrtimer_api.h> 37#include <linux/interrupt.h> 38#include <linux/irq_work.h> 39#include <linux/jiffies.h> 40#include <linux/kref_api.h> 41#include <linux/kthread.h> 42#include <linux/ktime_api.h> 43#include <linux/lockdep_api.h> 44#include <linux/lockdep.h> 45#include <linux/minmax.h> 46#include <linux/mm.h> 47#include <linux/module.h> 48#include <linux/mutex_api.h> 49#include <linux/plist.h> 50#include <linux/poll.h> 51#include <linux/proc_fs.h> 52#include <linux/profile.h> 53#include <linux/psi.h> 54#include <linux/rcupdate.h> 55#include <linux/seq_file.h> 56#include <linux/seqlock.h> 57#include <linux/softirq.h> 58#include <linux/spinlock_api.h> 59#include <linux/static_key.h> 60#include <linux/stop_machine.h> 61#include <linux/syscalls_api.h> 62#include <linux/syscalls.h> 63#include <linux/tick.h> 64#include <linux/topology.h> 65#include <linux/types.h> 66#include <linux/u64_stats_sync_api.h> 67#include <linux/uaccess.h> 68#include <linux/wait_api.h> 69#include <linux/wait_bit.h> 70#include <linux/workqueue_api.h> 71#include <linux/delayacct.h> 72#include <linux/mmu_context.h> 73 74#include <trace/events/power.h> 75#include <trace/events/sched.h> 76 77#include "../workqueue_internal.h" 78 79struct rq; 80struct cfs_rq; 81struct rt_rq; 82struct sched_group; 83struct cpuidle_state; 84 85#ifdef CONFIG_PARAVIRT 86# include <asm/paravirt.h> 87# include <asm/paravirt_api_clock.h> 88#endif 89 90#include <asm/barrier.h> 91 92#include "cpupri.h" 93#include "cpudeadline.h" 94 95/* task_struct::on_rq states: */ 96#define TASK_ON_RQ_QUEUED 1 97#define TASK_ON_RQ_MIGRATING 2 98 99extern __read_mostly int scheduler_running; 100 101extern unsigned long calc_load_update; 102extern atomic_long_t calc_load_tasks; 103 104extern void calc_global_load_tick(struct rq *this_rq); 105extern long calc_load_fold_active(struct rq *this_rq, long adjust); 106 107extern void call_trace_sched_update_nr_running(struct rq *rq, int count); 108 109extern int sysctl_sched_rt_period; 110extern int sysctl_sched_rt_runtime; 111extern int sched_rr_timeslice; 112 113/* 114 * Asymmetric CPU capacity bits 115 */ 116struct asym_cap_data { 117 struct list_head link; 118 struct rcu_head rcu; 119 unsigned long capacity; 120 unsigned long cpus[]; 121}; 122 123extern struct list_head asym_cap_list; 124 125#define cpu_capacity_span(asym_data) to_cpumask((asym_data)->cpus) 126 127/* 128 * Helpers for converting nanosecond timing to jiffy resolution 129 */ 130#define NS_TO_JIFFIES(time) ((unsigned long)(time) / (NSEC_PER_SEC/HZ)) 131 132/* 133 * Increase resolution of nice-level calculations for 64-bit architectures. 134 * The extra resolution improves shares distribution and load balancing of 135 * low-weight task groups (eg. nice +19 on an autogroup), deeper task-group 136 * hierarchies, especially on larger systems. This is not a user-visible change 137 * and does not change the user-interface for setting shares/weights. 138 * 139 * We increase resolution only if we have enough bits to allow this increased 140 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit 141 * are pretty high and the returns do not justify the increased costs. 142 * 143 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to 144 * increase coverage and consistency always enable it on 64-bit platforms. 145 */ 146#ifdef CONFIG_64BIT 147# define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT) 148# define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT) 149# define scale_load_down(w) \ 150({ \ 151 unsigned long __w = (w); \ 152 \ 153 if (__w) \ 154 __w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \ 155 __w; \ 156}) 157#else 158# define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT) 159# define scale_load(w) (w) 160# define scale_load_down(w) (w) 161#endif 162 163/* 164 * Task weight (visible to users) and its load (invisible to users) have 165 * independent resolution, but they should be well calibrated. We use 166 * scale_load() and scale_load_down(w) to convert between them. The 167 * following must be true: 168 * 169 * scale_load(sched_prio_to_weight[NICE_TO_PRIO(0)-MAX_RT_PRIO]) == NICE_0_LOAD 170 * 171 */ 172#define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT) 173 174/* 175 * Single value that decides SCHED_DEADLINE internal math precision. 176 * 10 -> just above 1us 177 * 9 -> just above 0.5us 178 */ 179#define DL_SCALE 10 180 181/* 182 * Single value that denotes runtime == period, ie unlimited time. 183 */ 184#define RUNTIME_INF ((u64)~0ULL) 185 186static inline int idle_policy(int policy) 187{ 188 return policy == SCHED_IDLE; 189} 190 191static inline int normal_policy(int policy) 192{ 193#ifdef CONFIG_SCHED_CLASS_EXT 194 if (policy == SCHED_EXT) 195 return true; 196#endif 197 return policy == SCHED_NORMAL; 198} 199 200static inline int fair_policy(int policy) 201{ 202 return normal_policy(policy) || policy == SCHED_BATCH; 203} 204 205static inline int rt_policy(int policy) 206{ 207 return policy == SCHED_FIFO || policy == SCHED_RR; 208} 209 210static inline int dl_policy(int policy) 211{ 212 return policy == SCHED_DEADLINE; 213} 214 215static inline bool valid_policy(int policy) 216{ 217 return idle_policy(policy) || fair_policy(policy) || 218 rt_policy(policy) || dl_policy(policy); 219} 220 221static inline int task_has_idle_policy(struct task_struct *p) 222{ 223 return idle_policy(p->policy); 224} 225 226static inline int task_has_rt_policy(struct task_struct *p) 227{ 228 return rt_policy(p->policy); 229} 230 231static inline int task_has_dl_policy(struct task_struct *p) 232{ 233 return dl_policy(p->policy); 234} 235 236#define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT) 237 238static inline void update_avg(u64 *avg, u64 sample) 239{ 240 s64 diff = sample - *avg; 241 242 *avg += diff / 8; 243} 244 245/* 246 * Shifting a value by an exponent greater *or equal* to the size of said value 247 * is UB; cap at size-1. 248 */ 249#define shr_bound(val, shift) \ 250 (val >> min_t(typeof(shift), shift, BITS_PER_TYPE(typeof(val)) - 1)) 251 252/* 253 * cgroup weight knobs should use the common MIN, DFL and MAX values which are 254 * 1, 100 and 10000 respectively. While it loses a bit of range on both ends, it 255 * maps pretty well onto the shares value used by scheduler and the round-trip 256 * conversions preserve the original value over the entire range. 257 */ 258static inline unsigned long sched_weight_from_cgroup(unsigned long cgrp_weight) 259{ 260 return DIV_ROUND_CLOSEST_ULL(cgrp_weight * 1024, CGROUP_WEIGHT_DFL); 261} 262 263static inline unsigned long sched_weight_to_cgroup(unsigned long weight) 264{ 265 return clamp_t(unsigned long, 266 DIV_ROUND_CLOSEST_ULL(weight * CGROUP_WEIGHT_DFL, 1024), 267 CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); 268} 269 270/* 271 * !! For sched_setattr_nocheck() (kernel) only !! 272 * 273 * This is actually gross. :( 274 * 275 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE 276 * tasks, but still be able to sleep. We need this on platforms that cannot 277 * atomically change clock frequency. Remove once fast switching will be 278 * available on such platforms. 279 * 280 * SUGOV stands for SchedUtil GOVernor. 281 */ 282#define SCHED_FLAG_SUGOV 0x10000000 283 284#define SCHED_DL_FLAGS (SCHED_FLAG_RECLAIM | SCHED_FLAG_DL_OVERRUN | SCHED_FLAG_SUGOV) 285 286static inline bool dl_entity_is_special(const struct sched_dl_entity *dl_se) 287{ 288#ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL 289 return unlikely(dl_se->flags & SCHED_FLAG_SUGOV); 290#else 291 return false; 292#endif 293} 294 295/* 296 * Tells if entity @a should preempt entity @b. 297 */ 298static inline bool dl_entity_preempt(const struct sched_dl_entity *a, 299 const struct sched_dl_entity *b) 300{ 301 return dl_entity_is_special(a) || 302 dl_time_before(a->deadline, b->deadline); 303} 304 305/* 306 * This is the priority-queue data structure of the RT scheduling class: 307 */ 308struct rt_prio_array { 309 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ 310 struct list_head queue[MAX_RT_PRIO]; 311}; 312 313struct rt_bandwidth { 314 /* nests inside the rq lock: */ 315 raw_spinlock_t rt_runtime_lock; 316 ktime_t rt_period; 317 u64 rt_runtime; 318 struct hrtimer rt_period_timer; 319 unsigned int rt_period_active; 320}; 321 322static inline int dl_bandwidth_enabled(void) 323{ 324 return sysctl_sched_rt_runtime >= 0; 325} 326 327/* 328 * To keep the bandwidth of -deadline tasks under control 329 * we need some place where: 330 * - store the maximum -deadline bandwidth of each cpu; 331 * - cache the fraction of bandwidth that is currently allocated in 332 * each root domain; 333 * 334 * This is all done in the data structure below. It is similar to the 335 * one used for RT-throttling (rt_bandwidth), with the main difference 336 * that, since here we are only interested in admission control, we 337 * do not decrease any runtime while the group "executes", neither we 338 * need a timer to replenish it. 339 * 340 * With respect to SMP, bandwidth is given on a per root domain basis, 341 * meaning that: 342 * - bw (< 100%) is the deadline bandwidth of each CPU; 343 * - total_bw is the currently allocated bandwidth in each root domain; 344 */ 345struct dl_bw { 346 raw_spinlock_t lock; 347 u64 bw; 348 u64 total_bw; 349}; 350 351extern void init_dl_bw(struct dl_bw *dl_b); 352extern int sched_dl_global_validate(void); 353extern void sched_dl_do_global(void); 354extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr); 355extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr); 356extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr); 357extern bool __checkparam_dl(const struct sched_attr *attr); 358extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr); 359extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial); 360extern int dl_bw_deactivate(int cpu); 361extern s64 dl_scaled_delta_exec(struct rq *rq, struct sched_dl_entity *dl_se, s64 delta_exec); 362/* 363 * SCHED_DEADLINE supports servers (nested scheduling) with the following 364 * interface: 365 * 366 * dl_se::rq -- runqueue we belong to. 367 * 368 * dl_se::server_pick() -- nested pick_next_task(); we yield the period if this 369 * returns NULL. 370 * 371 * dl_server_update() -- called from update_curr_common(), propagates runtime 372 * to the server. 373 * 374 * dl_server_start() -- start the server when it has tasks; it will stop 375 * automatically when there are no more tasks, per 376 * dl_se::server_pick() returning NULL. 377 * 378 * dl_server_stop() -- (force) stop the server; use when updating 379 * parameters. 380 * 381 * dl_server_init() -- initializes the server. 382 * 383 * When started the dl_server will (per dl_defer) schedule a timer for its 384 * zero-laxity point -- that is, unlike regular EDF tasks which run ASAP, a 385 * server will run at the very end of its period. 386 * 387 * This is done such that any runtime from the target class can be accounted 388 * against the server -- through dl_server_update() above -- such that when it 389 * becomes time to run, it might already be out of runtime and get deferred 390 * until the next period. In this case dl_server_timer() will alternate 391 * between defer and replenish but never actually enqueue the server. 392 * 393 * Only when the target class does not manage to exhaust the server's runtime 394 * (there's actualy starvation in the given period), will the dl_server get on 395 * the runqueue. Once queued it will pick tasks from the target class and run 396 * them until either its runtime is exhaused, at which point its back to 397 * dl_server_timer, or until there are no more tasks to run, at which point 398 * the dl_server stops itself. 399 * 400 * By stopping at this point the dl_server retains bandwidth, which, if a new 401 * task wakes up imminently (starting the server again), can be used -- 402 * subject to CBS wakeup rules -- without having to wait for the next period. 403 * 404 * Additionally, because of the dl_defer behaviour the start/stop behaviour is 405 * naturally thottled to once per period, avoiding high context switch 406 * workloads from spamming the hrtimer program/cancel paths. 407 */ 408extern void dl_server_update_idle(struct sched_dl_entity *dl_se, s64 delta_exec); 409extern void dl_server_update(struct sched_dl_entity *dl_se, s64 delta_exec); 410extern void dl_server_start(struct sched_dl_entity *dl_se); 411extern void dl_server_stop(struct sched_dl_entity *dl_se); 412extern void dl_server_init(struct sched_dl_entity *dl_se, struct rq *rq, 413 dl_server_pick_f pick_task); 414extern void sched_init_dl_servers(void); 415 416extern void fair_server_init(struct rq *rq); 417extern void __dl_server_attach_root(struct sched_dl_entity *dl_se, struct rq *rq); 418extern int dl_server_apply_params(struct sched_dl_entity *dl_se, 419 u64 runtime, u64 period, bool init); 420 421static inline bool dl_server_active(struct sched_dl_entity *dl_se) 422{ 423 return dl_se->dl_server_active; 424} 425 426#ifdef CONFIG_CGROUP_SCHED 427 428extern struct list_head task_groups; 429 430#ifdef CONFIG_GROUP_SCHED_BANDWIDTH 431extern const u64 max_bw_quota_period_us; 432 433/* 434 * default period for group bandwidth. 435 * default: 0.1s, units: microseconds 436 */ 437static inline u64 default_bw_period_us(void) 438{ 439 return 100000ULL; 440} 441#endif /* CONFIG_GROUP_SCHED_BANDWIDTH */ 442 443struct cfs_bandwidth { 444#ifdef CONFIG_CFS_BANDWIDTH 445 raw_spinlock_t lock; 446 ktime_t period; 447 u64 quota; 448 u64 runtime; 449 u64 burst; 450 u64 runtime_snap; 451 s64 hierarchical_quota; 452 453 u8 idle; 454 u8 period_active; 455 u8 slack_started; 456 struct hrtimer period_timer; 457 struct hrtimer slack_timer; 458 struct list_head throttled_cfs_rq; 459 460 /* Statistics: */ 461 int nr_periods; 462 int nr_throttled; 463 int nr_burst; 464 u64 throttled_time; 465 u64 burst_time; 466#endif /* CONFIG_CFS_BANDWIDTH */ 467}; 468 469/* Task group related information */ 470struct task_group { 471 struct cgroup_subsys_state css; 472 473#ifdef CONFIG_GROUP_SCHED_WEIGHT 474 /* A positive value indicates that this is a SCHED_IDLE group. */ 475 int idle; 476#endif 477 478#ifdef CONFIG_FAIR_GROUP_SCHED 479 /* schedulable entities of this group on each CPU */ 480 struct sched_entity **se; 481 /* runqueue "owned" by this group on each CPU */ 482 struct cfs_rq **cfs_rq; 483 unsigned long shares; 484 /* 485 * load_avg can be heavily contended at clock tick time, so put 486 * it in its own cache-line separated from the fields above which 487 * will also be accessed at each tick. 488 */ 489 atomic_long_t load_avg ____cacheline_aligned; 490#endif /* CONFIG_FAIR_GROUP_SCHED */ 491 492#ifdef CONFIG_RT_GROUP_SCHED 493 struct sched_rt_entity **rt_se; 494 struct rt_rq **rt_rq; 495 496 struct rt_bandwidth rt_bandwidth; 497#endif 498 499 struct scx_task_group scx; 500 501 struct rcu_head rcu; 502 struct list_head list; 503 504 struct task_group *parent; 505 struct list_head siblings; 506 struct list_head children; 507 508#ifdef CONFIG_SCHED_AUTOGROUP 509 struct autogroup *autogroup; 510#endif 511 512 struct cfs_bandwidth cfs_bandwidth; 513 514#ifdef CONFIG_UCLAMP_TASK_GROUP 515 /* The two decimal precision [%] value requested from user-space */ 516 unsigned int uclamp_pct[UCLAMP_CNT]; 517 /* Clamp values requested for a task group */ 518 struct uclamp_se uclamp_req[UCLAMP_CNT]; 519 /* Effective clamp values used for a task group */ 520 struct uclamp_se uclamp[UCLAMP_CNT]; 521#endif 522 523}; 524 525#ifdef CONFIG_GROUP_SCHED_WEIGHT 526#define ROOT_TASK_GROUP_LOAD NICE_0_LOAD 527 528/* 529 * A weight of 0 or 1 can cause arithmetics problems. 530 * A weight of a cfs_rq is the sum of weights of which entities 531 * are queued on this cfs_rq, so a weight of a entity should not be 532 * too large, so as the shares value of a task group. 533 * (The default weight is 1024 - so there's no practical 534 * limitation from this.) 535 */ 536#define MIN_SHARES (1UL << 1) 537#define MAX_SHARES (1UL << 18) 538#endif 539 540typedef int (*tg_visitor)(struct task_group *, void *); 541 542extern int walk_tg_tree_from(struct task_group *from, 543 tg_visitor down, tg_visitor up, void *data); 544 545/* 546 * Iterate the full tree, calling @down when first entering a node and @up when 547 * leaving it for the final time. 548 * 549 * Caller must hold rcu_lock or sufficient equivalent. 550 */ 551static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) 552{ 553 return walk_tg_tree_from(&root_task_group, down, up, data); 554} 555 556static inline struct task_group *css_tg(struct cgroup_subsys_state *css) 557{ 558 return css ? container_of(css, struct task_group, css) : NULL; 559} 560 561extern int tg_nop(struct task_group *tg, void *data); 562 563#ifdef CONFIG_FAIR_GROUP_SCHED 564extern void free_fair_sched_group(struct task_group *tg); 565extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent); 566extern void online_fair_sched_group(struct task_group *tg); 567extern void unregister_fair_sched_group(struct task_group *tg); 568#else /* !CONFIG_FAIR_GROUP_SCHED: */ 569static inline void free_fair_sched_group(struct task_group *tg) { } 570static inline int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) 571{ 572 return 1; 573} 574static inline void online_fair_sched_group(struct task_group *tg) { } 575static inline void unregister_fair_sched_group(struct task_group *tg) { } 576#endif /* !CONFIG_FAIR_GROUP_SCHED */ 577 578extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, 579 struct sched_entity *se, int cpu, 580 struct sched_entity *parent); 581extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b, struct cfs_bandwidth *parent); 582 583extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b); 584extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b); 585extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq); 586extern bool cfs_task_bw_constrained(struct task_struct *p); 587 588extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, 589 struct sched_rt_entity *rt_se, int cpu, 590 struct sched_rt_entity *parent); 591extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us); 592extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us); 593extern long sched_group_rt_runtime(struct task_group *tg); 594extern long sched_group_rt_period(struct task_group *tg); 595extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk); 596 597extern struct task_group *sched_create_group(struct task_group *parent); 598extern void sched_online_group(struct task_group *tg, 599 struct task_group *parent); 600extern void sched_destroy_group(struct task_group *tg); 601extern void sched_release_group(struct task_group *tg); 602 603extern void sched_move_task(struct task_struct *tsk, bool for_autogroup); 604 605#ifdef CONFIG_FAIR_GROUP_SCHED 606extern int sched_group_set_shares(struct task_group *tg, unsigned long shares); 607 608extern int sched_group_set_idle(struct task_group *tg, long idle); 609 610extern void set_task_rq_fair(struct sched_entity *se, 611 struct cfs_rq *prev, struct cfs_rq *next); 612#else /* !CONFIG_FAIR_GROUP_SCHED: */ 613static inline int sched_group_set_shares(struct task_group *tg, unsigned long shares) { return 0; } 614static inline int sched_group_set_idle(struct task_group *tg, long idle) { return 0; } 615#endif /* !CONFIG_FAIR_GROUP_SCHED */ 616 617#else /* !CONFIG_CGROUP_SCHED: */ 618 619struct cfs_bandwidth { }; 620 621static inline bool cfs_task_bw_constrained(struct task_struct *p) { return false; } 622 623#endif /* !CONFIG_CGROUP_SCHED */ 624 625extern void unregister_rt_sched_group(struct task_group *tg); 626extern void free_rt_sched_group(struct task_group *tg); 627extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent); 628 629/* 630 * u64_u32_load/u64_u32_store 631 * 632 * Use a copy of a u64 value to protect against data race. This is only 633 * applicable for 32-bits architectures. 634 */ 635#ifdef CONFIG_64BIT 636# define u64_u32_load_copy(var, copy) var 637# define u64_u32_store_copy(var, copy, val) (var = val) 638#else 639# define u64_u32_load_copy(var, copy) \ 640({ \ 641 u64 __val, __val_copy; \ 642 do { \ 643 __val_copy = copy; \ 644 /* \ 645 * paired with u64_u32_store_copy(), ordering access \ 646 * to var and copy. \ 647 */ \ 648 smp_rmb(); \ 649 __val = var; \ 650 } while (__val != __val_copy); \ 651 __val; \ 652}) 653# define u64_u32_store_copy(var, copy, val) \ 654do { \ 655 typeof(val) __val = (val); \ 656 var = __val; \ 657 /* \ 658 * paired with u64_u32_load_copy(), ordering access to var and \ 659 * copy. \ 660 */ \ 661 smp_wmb(); \ 662 copy = __val; \ 663} while (0) 664#endif 665# define u64_u32_load(var) u64_u32_load_copy(var, var##_copy) 666# define u64_u32_store(var, val) u64_u32_store_copy(var, var##_copy, val) 667 668struct balance_callback { 669 struct balance_callback *next; 670 void (*func)(struct rq *rq); 671}; 672 673/* CFS-related fields in a runqueue */ 674struct cfs_rq { 675 struct load_weight load; 676 unsigned int nr_queued; 677 unsigned int h_nr_queued; /* SCHED_{NORMAL,BATCH,IDLE} */ 678 unsigned int h_nr_runnable; /* SCHED_{NORMAL,BATCH,IDLE} */ 679 unsigned int h_nr_idle; /* SCHED_IDLE */ 680 681 s64 sum_w_vruntime; 682 u64 sum_weight; 683 684 u64 zero_vruntime; 685#ifdef CONFIG_SCHED_CORE 686 unsigned int forceidle_seq; 687 u64 zero_vruntime_fi; 688#endif 689 690 struct rb_root_cached tasks_timeline; 691 692 /* 693 * 'curr' points to currently running entity on this cfs_rq. 694 * It is set to NULL otherwise (i.e when none are currently running). 695 */ 696 struct sched_entity *curr; 697 struct sched_entity *next; 698 699 /* 700 * CFS load tracking 701 */ 702 struct sched_avg avg; 703#ifndef CONFIG_64BIT 704 u64 last_update_time_copy; 705#endif 706 struct { 707 raw_spinlock_t lock ____cacheline_aligned; 708 int nr; 709 unsigned long load_avg; 710 unsigned long util_avg; 711 unsigned long runnable_avg; 712 } removed; 713 714#ifdef CONFIG_FAIR_GROUP_SCHED 715 u64 last_update_tg_load_avg; 716 unsigned long tg_load_avg_contrib; 717 long propagate; 718 long prop_runnable_sum; 719 720 /* 721 * h_load = weight * f(tg) 722 * 723 * Where f(tg) is the recursive weight fraction assigned to 724 * this group. 725 */ 726 unsigned long h_load; 727 u64 last_h_load_update; 728 struct sched_entity *h_load_next; 729#endif /* CONFIG_FAIR_GROUP_SCHED */ 730 731#ifdef CONFIG_FAIR_GROUP_SCHED 732 struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */ 733 734 /* 735 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in 736 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities 737 * (like users, containers etc.) 738 * 739 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU. 740 * This list is used during load balance. 741 */ 742 int on_list; 743 struct list_head leaf_cfs_rq_list; 744 struct task_group *tg; /* group that "owns" this runqueue */ 745 746 /* Locally cached copy of our task_group's idle value */ 747 int idle; 748 749#ifdef CONFIG_CFS_BANDWIDTH 750 int runtime_enabled; 751 s64 runtime_remaining; 752 753 u64 throttled_pelt_idle; 754#ifndef CONFIG_64BIT 755 u64 throttled_pelt_idle_copy; 756#endif 757 u64 throttled_clock; 758 u64 throttled_clock_pelt; 759 u64 throttled_clock_pelt_time; 760 u64 throttled_clock_self; 761 u64 throttled_clock_self_time; 762 bool throttled:1; 763 bool pelt_clock_throttled:1; 764 int throttle_count; 765 struct list_head throttled_list; 766 struct list_head throttled_csd_list; 767 struct list_head throttled_limbo_list; 768#endif /* CONFIG_CFS_BANDWIDTH */ 769#endif /* CONFIG_FAIR_GROUP_SCHED */ 770}; 771 772#ifdef CONFIG_SCHED_CLASS_EXT 773/* scx_rq->flags, protected by the rq lock */ 774enum scx_rq_flags { 775 /* 776 * A hotplugged CPU starts scheduling before rq_online_scx(). Track 777 * ops.cpu_on/offline() state so that ops.enqueue/dispatch() are called 778 * only while the BPF scheduler considers the CPU to be online. 779 */ 780 SCX_RQ_ONLINE = 1 << 0, 781 SCX_RQ_CAN_STOP_TICK = 1 << 1, 782 SCX_RQ_BAL_KEEP = 1 << 3, /* balance decided to keep current */ 783 SCX_RQ_BYPASSING = 1 << 4, 784 SCX_RQ_CLK_VALID = 1 << 5, /* RQ clock is fresh and valid */ 785 SCX_RQ_BAL_CB_PENDING = 1 << 6, /* must queue a cb after dispatching */ 786 787 SCX_RQ_IN_WAKEUP = 1 << 16, 788 SCX_RQ_IN_BALANCE = 1 << 17, 789}; 790 791struct scx_rq { 792 struct scx_dispatch_q local_dsq; 793 struct list_head runnable_list; /* runnable tasks on this rq */ 794 struct list_head ddsp_deferred_locals; /* deferred ddsps from enq */ 795 unsigned long ops_qseq; 796 u64 extra_enq_flags; /* see move_task_to_local_dsq() */ 797 u32 nr_running; 798 u32 cpuperf_target; /* [0, SCHED_CAPACITY_SCALE] */ 799 bool cpu_released; 800 u32 flags; 801 u64 clock; /* current per-rq clock -- see scx_bpf_now() */ 802 cpumask_var_t cpus_to_kick; 803 cpumask_var_t cpus_to_kick_if_idle; 804 cpumask_var_t cpus_to_preempt; 805 cpumask_var_t cpus_to_wait; 806 cpumask_var_t cpus_to_sync; 807 bool kick_sync_pending; 808 unsigned long kick_sync; 809 local_t reenq_local_deferred; 810 struct balance_callback deferred_bal_cb; 811 struct balance_callback kick_sync_bal_cb; 812 struct irq_work deferred_irq_work; 813 struct irq_work kick_cpus_irq_work; 814 struct scx_dispatch_q bypass_dsq; 815}; 816#endif /* CONFIG_SCHED_CLASS_EXT */ 817 818static inline int rt_bandwidth_enabled(void) 819{ 820 return sysctl_sched_rt_runtime >= 0; 821} 822 823/* RT IPI pull logic requires IRQ_WORK */ 824#if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP) 825# define HAVE_RT_PUSH_IPI 826#endif 827 828/* Real-Time classes' related field in a runqueue: */ 829struct rt_rq { 830 struct rt_prio_array active; 831 unsigned int rt_nr_running; 832 unsigned int rr_nr_running; 833 struct { 834 int curr; /* highest queued rt task prio */ 835 int next; /* next highest */ 836 } highest_prio; 837 bool overloaded; 838 struct plist_head pushable_tasks; 839 840 int rt_queued; 841 842#ifdef CONFIG_RT_GROUP_SCHED 843 int rt_throttled; 844 u64 rt_time; /* consumed RT time, goes up in update_curr_rt */ 845 u64 rt_runtime; /* allotted RT time, "slice" from rt_bandwidth, RT sharing/balancing */ 846 /* Nests inside the rq lock: */ 847 raw_spinlock_t rt_runtime_lock; 848 849 unsigned int rt_nr_boosted; 850 851 struct rq *rq; /* this is always top-level rq, cache? */ 852#endif 853#ifdef CONFIG_CGROUP_SCHED 854 struct task_group *tg; /* this tg has "this" rt_rq on given CPU for runnable entities */ 855#endif 856}; 857 858static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq) 859{ 860 return rt_rq->rt_queued && rt_rq->rt_nr_running; 861} 862 863/* Deadline class' related fields in a runqueue */ 864struct dl_rq { 865 /* runqueue is an rbtree, ordered by deadline */ 866 struct rb_root_cached root; 867 868 unsigned int dl_nr_running; 869 870 /* 871 * Deadline values of the currently executing and the 872 * earliest ready task on this rq. Caching these facilitates 873 * the decision whether or not a ready but not running task 874 * should migrate somewhere else. 875 */ 876 struct { 877 u64 curr; 878 u64 next; 879 } earliest_dl; 880 881 bool overloaded; 882 883 /* 884 * Tasks on this rq that can be pushed away. They are kept in 885 * an rb-tree, ordered by tasks' deadlines, with caching 886 * of the leftmost (earliest deadline) element. 887 */ 888 struct rb_root_cached pushable_dl_tasks_root; 889 890 /* 891 * "Active utilization" for this runqueue: increased when a 892 * task wakes up (becomes TASK_RUNNING) and decreased when a 893 * task blocks 894 */ 895 u64 running_bw; 896 897 /* 898 * Utilization of the tasks "assigned" to this runqueue (including 899 * the tasks that are in runqueue and the tasks that executed on this 900 * CPU and blocked). Increased when a task moves to this runqueue, and 901 * decreased when the task moves away (migrates, changes scheduling 902 * policy, or terminates). 903 * This is needed to compute the "inactive utilization" for the 904 * runqueue (inactive utilization = this_bw - running_bw). 905 */ 906 u64 this_bw; 907 u64 extra_bw; 908 909 /* 910 * Maximum available bandwidth for reclaiming by SCHED_FLAG_RECLAIM 911 * tasks of this rq. Used in calculation of reclaimable bandwidth(GRUB). 912 */ 913 u64 max_bw; 914 915 /* 916 * Inverse of the fraction of CPU utilization that can be reclaimed 917 * by the GRUB algorithm. 918 */ 919 u64 bw_ratio; 920}; 921 922#ifdef CONFIG_FAIR_GROUP_SCHED 923 924/* An entity is a task if it doesn't "own" a runqueue */ 925#define entity_is_task(se) (!se->my_q) 926 927static inline void se_update_runnable(struct sched_entity *se) 928{ 929 if (!entity_is_task(se)) 930 se->runnable_weight = se->my_q->h_nr_runnable; 931} 932 933static inline long se_runnable(struct sched_entity *se) 934{ 935 if (se->sched_delayed) 936 return false; 937 938 if (entity_is_task(se)) 939 return !!se->on_rq; 940 else 941 return se->runnable_weight; 942} 943 944#else /* !CONFIG_FAIR_GROUP_SCHED: */ 945 946#define entity_is_task(se) 1 947 948static inline void se_update_runnable(struct sched_entity *se) { } 949 950static inline long se_runnable(struct sched_entity *se) 951{ 952 if (se->sched_delayed) 953 return false; 954 955 return !!se->on_rq; 956} 957 958#endif /* !CONFIG_FAIR_GROUP_SCHED */ 959 960/* 961 * XXX we want to get rid of these helpers and use the full load resolution. 962 */ 963static inline long se_weight(struct sched_entity *se) 964{ 965 return scale_load_down(se->load.weight); 966} 967 968 969static inline bool sched_asym_prefer(int a, int b) 970{ 971 return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b); 972} 973 974struct perf_domain { 975 struct em_perf_domain *em_pd; 976 struct perf_domain *next; 977 struct rcu_head rcu; 978}; 979 980/* 981 * We add the notion of a root-domain which will be used to define per-domain 982 * variables. Each exclusive cpuset essentially defines an island domain by 983 * fully partitioning the member CPUs from any other cpuset. Whenever a new 984 * exclusive cpuset is created, we also create and attach a new root-domain 985 * object. 986 * 987 */ 988struct root_domain { 989 atomic_t refcount; 990 atomic_t rto_count; 991 struct rcu_head rcu; 992 cpumask_var_t span; 993 cpumask_var_t online; 994 995 /* 996 * Indicate pullable load on at least one CPU, e.g: 997 * - More than one runnable task 998 * - Running task is misfit 999 */ 1000 bool overloaded; 1001 1002 /* Indicate one or more CPUs over-utilized (tipping point) */ 1003 bool overutilized; 1004 1005 /* 1006 * The bit corresponding to a CPU gets set here if such CPU has more 1007 * than one runnable -deadline task (as it is below for RT tasks). 1008 */ 1009 cpumask_var_t dlo_mask; 1010 atomic_t dlo_count; 1011 struct dl_bw dl_bw; 1012 struct cpudl cpudl; 1013 1014 /* 1015 * Indicate whether a root_domain's dl_bw has been checked or 1016 * updated. It's monotonously increasing value. 1017 * 1018 * Also, some corner cases, like 'wrap around' is dangerous, but given 1019 * that u64 is 'big enough'. So that shouldn't be a concern. 1020 */ 1021 u64 visit_cookie; 1022 1023#ifdef HAVE_RT_PUSH_IPI 1024 /* 1025 * For IPI pull requests, loop across the rto_mask. 1026 */ 1027 struct irq_work rto_push_work; 1028 raw_spinlock_t rto_lock; 1029 /* These are only updated and read within rto_lock */ 1030 int rto_loop; 1031 int rto_cpu; 1032 /* These atomics are updated outside of a lock */ 1033 atomic_t rto_loop_next; 1034 atomic_t rto_loop_start; 1035#endif /* HAVE_RT_PUSH_IPI */ 1036 /* 1037 * The "RT overload" flag: it gets set if a CPU has more than 1038 * one runnable RT task. 1039 */ 1040 cpumask_var_t rto_mask; 1041 struct cpupri cpupri; 1042 1043 /* 1044 * NULL-terminated list of performance domains intersecting with the 1045 * CPUs of the rd. Protected by RCU. 1046 */ 1047 struct perf_domain __rcu *pd; 1048}; 1049 1050extern void init_defrootdomain(void); 1051extern int sched_init_domains(const struct cpumask *cpu_map); 1052extern void rq_attach_root(struct rq *rq, struct root_domain *rd); 1053extern void sched_get_rd(struct root_domain *rd); 1054extern void sched_put_rd(struct root_domain *rd); 1055 1056static inline int get_rd_overloaded(struct root_domain *rd) 1057{ 1058 return READ_ONCE(rd->overloaded); 1059} 1060 1061static inline void set_rd_overloaded(struct root_domain *rd, int status) 1062{ 1063 if (get_rd_overloaded(rd) != status) 1064 WRITE_ONCE(rd->overloaded, status); 1065} 1066 1067#ifdef HAVE_RT_PUSH_IPI 1068extern void rto_push_irq_work_func(struct irq_work *work); 1069#endif 1070 1071#ifdef CONFIG_UCLAMP_TASK 1072/* 1073 * struct uclamp_bucket - Utilization clamp bucket 1074 * @value: utilization clamp value for tasks on this clamp bucket 1075 * @tasks: number of RUNNABLE tasks on this clamp bucket 1076 * 1077 * Keep track of how many tasks are RUNNABLE for a given utilization 1078 * clamp value. 1079 */ 1080struct uclamp_bucket { 1081 unsigned long value : bits_per(SCHED_CAPACITY_SCALE); 1082 unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE); 1083}; 1084 1085/* 1086 * struct uclamp_rq - rq's utilization clamp 1087 * @value: currently active clamp values for a rq 1088 * @bucket: utilization clamp buckets affecting a rq 1089 * 1090 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values. 1091 * A clamp value is affecting a rq when there is at least one task RUNNABLE 1092 * (or actually running) with that value. 1093 * 1094 * There are up to UCLAMP_CNT possible different clamp values, currently there 1095 * are only two: minimum utilization and maximum utilization. 1096 * 1097 * All utilization clamping values are MAX aggregated, since: 1098 * - for util_min: we want to run the CPU at least at the max of the minimum 1099 * utilization required by its currently RUNNABLE tasks. 1100 * - for util_max: we want to allow the CPU to run up to the max of the 1101 * maximum utilization allowed by its currently RUNNABLE tasks. 1102 * 1103 * Since on each system we expect only a limited number of different 1104 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track 1105 * the metrics required to compute all the per-rq utilization clamp values. 1106 */ 1107struct uclamp_rq { 1108 unsigned int value; 1109 struct uclamp_bucket bucket[UCLAMP_BUCKETS]; 1110}; 1111 1112DECLARE_STATIC_KEY_FALSE(sched_uclamp_used); 1113#endif /* CONFIG_UCLAMP_TASK */ 1114 1115/* 1116 * This is the main, per-CPU runqueue data structure. 1117 * 1118 * Locking rule: those places that want to lock multiple runqueues 1119 * (such as the load balancing or the thread migration code), lock 1120 * acquire operations must be ordered by ascending &runqueue. 1121 */ 1122struct rq { 1123 /* runqueue lock: */ 1124 raw_spinlock_t __lock; 1125 1126 /* Per class runqueue modification mask; bits in class order. */ 1127 unsigned int queue_mask; 1128 unsigned int nr_running; 1129#ifdef CONFIG_NUMA_BALANCING 1130 unsigned int nr_numa_running; 1131 unsigned int nr_preferred_running; 1132 unsigned int numa_migrate_on; 1133#endif 1134#ifdef CONFIG_NO_HZ_COMMON 1135 unsigned long last_blocked_load_update_tick; 1136 unsigned int has_blocked_load; 1137 call_single_data_t nohz_csd; 1138 unsigned int nohz_tick_stopped; 1139 atomic_t nohz_flags; 1140#endif /* CONFIG_NO_HZ_COMMON */ 1141 1142 unsigned int ttwu_pending; 1143 u64 nr_switches; 1144 1145#ifdef CONFIG_UCLAMP_TASK 1146 /* Utilization clamp values based on CPU's RUNNABLE tasks */ 1147 struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned; 1148 unsigned int uclamp_flags; 1149#define UCLAMP_FLAG_IDLE 0x01 1150#endif 1151 1152 struct cfs_rq cfs; 1153 struct rt_rq rt; 1154 struct dl_rq dl; 1155#ifdef CONFIG_SCHED_CLASS_EXT 1156 struct scx_rq scx; 1157#endif 1158 1159 struct sched_dl_entity fair_server; 1160 1161#ifdef CONFIG_FAIR_GROUP_SCHED 1162 /* list of leaf cfs_rq on this CPU: */ 1163 struct list_head leaf_cfs_rq_list; 1164 struct list_head *tmp_alone_branch; 1165#endif /* CONFIG_FAIR_GROUP_SCHED */ 1166 1167 /* 1168 * This is part of a global counter where only the total sum 1169 * over all CPUs matters. A task can increase this counter on 1170 * one CPU and if it got migrated afterwards it may decrease 1171 * it on another CPU. Always updated under the runqueue lock: 1172 */ 1173 unsigned long nr_uninterruptible; 1174 1175#ifdef CONFIG_SCHED_PROXY_EXEC 1176 struct task_struct __rcu *donor; /* Scheduling context */ 1177 struct task_struct __rcu *curr; /* Execution context */ 1178#else 1179 union { 1180 struct task_struct __rcu *donor; /* Scheduler context */ 1181 struct task_struct __rcu *curr; /* Execution context */ 1182 }; 1183#endif 1184 struct sched_dl_entity *dl_server; 1185 struct task_struct *idle; 1186 struct task_struct *stop; 1187 unsigned long next_balance; 1188 struct mm_struct *prev_mm; 1189 1190 unsigned int clock_update_flags; 1191 u64 clock; 1192 /* Ensure that all clocks are in the same cache line */ 1193 u64 clock_task ____cacheline_aligned; 1194 u64 clock_pelt; 1195 unsigned long lost_idle_time; 1196 u64 clock_pelt_idle; 1197 u64 clock_idle; 1198#ifndef CONFIG_64BIT 1199 u64 clock_pelt_idle_copy; 1200 u64 clock_idle_copy; 1201#endif 1202 1203 atomic_t nr_iowait; 1204 1205 u64 last_seen_need_resched_ns; 1206 int ticks_without_resched; 1207 1208#ifdef CONFIG_MEMBARRIER 1209 int membarrier_state; 1210#endif 1211 1212 struct root_domain *rd; 1213 struct sched_domain __rcu *sd; 1214 1215 unsigned long cpu_capacity; 1216 1217 struct balance_callback *balance_callback; 1218 1219 unsigned char nohz_idle_balance; 1220 unsigned char idle_balance; 1221 1222 unsigned long misfit_task_load; 1223 1224 /* For active balancing */ 1225 int active_balance; 1226 int push_cpu; 1227 struct cpu_stop_work active_balance_work; 1228 1229 /* CPU of this runqueue: */ 1230 int cpu; 1231 int online; 1232 1233 struct list_head cfs_tasks; 1234 1235 struct sched_avg avg_rt; 1236 struct sched_avg avg_dl; 1237#ifdef CONFIG_HAVE_SCHED_AVG_IRQ 1238 struct sched_avg avg_irq; 1239#endif 1240#ifdef CONFIG_SCHED_HW_PRESSURE 1241 struct sched_avg avg_hw; 1242#endif 1243 u64 idle_stamp; 1244 u64 avg_idle; 1245 1246 /* This is used to determine avg_idle's max value */ 1247 u64 max_idle_balance_cost; 1248 1249#ifdef CONFIG_HOTPLUG_CPU 1250 struct rcuwait hotplug_wait; 1251#endif 1252 1253#ifdef CONFIG_IRQ_TIME_ACCOUNTING 1254 u64 prev_irq_time; 1255 u64 psi_irq_time; 1256#endif 1257#ifdef CONFIG_PARAVIRT 1258 u64 prev_steal_time; 1259#endif 1260#ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING 1261 u64 prev_steal_time_rq; 1262#endif 1263 1264 /* calc_load related fields */ 1265 unsigned long calc_load_update; 1266 long calc_load_active; 1267 1268#ifdef CONFIG_SCHED_HRTICK 1269 call_single_data_t hrtick_csd; 1270 struct hrtimer hrtick_timer; 1271 ktime_t hrtick_time; 1272#endif 1273 1274#ifdef CONFIG_SCHEDSTATS 1275 /* latency stats */ 1276 struct sched_info rq_sched_info; 1277 unsigned long long rq_cpu_time; 1278 1279 /* sys_sched_yield() stats */ 1280 unsigned int yld_count; 1281 1282 /* schedule() stats */ 1283 unsigned int sched_count; 1284 unsigned int sched_goidle; 1285 1286 /* try_to_wake_up() stats */ 1287 unsigned int ttwu_count; 1288 unsigned int ttwu_local; 1289#endif 1290 1291#ifdef CONFIG_CPU_IDLE 1292 /* Must be inspected within a RCU lock section */ 1293 struct cpuidle_state *idle_state; 1294#endif 1295 1296 unsigned int nr_pinned; 1297 unsigned int push_busy; 1298 struct cpu_stop_work push_work; 1299 1300#ifdef CONFIG_SCHED_CORE 1301 /* per rq */ 1302 struct rq *core; 1303 struct task_struct *core_pick; 1304 struct sched_dl_entity *core_dl_server; 1305 unsigned int core_enabled; 1306 unsigned int core_sched_seq; 1307 struct rb_root core_tree; 1308 1309 /* shared state -- careful with sched_core_cpu_deactivate() */ 1310 unsigned int core_task_seq; 1311 unsigned int core_pick_seq; 1312 unsigned long core_cookie; 1313 unsigned int core_forceidle_count; 1314 unsigned int core_forceidle_seq; 1315 unsigned int core_forceidle_occupation; 1316 u64 core_forceidle_start; 1317#endif /* CONFIG_SCHED_CORE */ 1318 1319 /* Scratch cpumask to be temporarily used under rq_lock */ 1320 cpumask_var_t scratch_mask; 1321 1322#ifdef CONFIG_CFS_BANDWIDTH 1323 call_single_data_t cfsb_csd; 1324 struct list_head cfsb_csd_list; 1325#endif 1326}; 1327 1328#ifdef CONFIG_FAIR_GROUP_SCHED 1329 1330/* CPU runqueue to which this cfs_rq is attached */ 1331static inline struct rq *rq_of(struct cfs_rq *cfs_rq) 1332{ 1333 return cfs_rq->rq; 1334} 1335 1336#else /* !CONFIG_FAIR_GROUP_SCHED: */ 1337 1338static inline struct rq *rq_of(struct cfs_rq *cfs_rq) 1339{ 1340 return container_of(cfs_rq, struct rq, cfs); 1341} 1342#endif /* !CONFIG_FAIR_GROUP_SCHED */ 1343 1344static inline int cpu_of(struct rq *rq) 1345{ 1346 return rq->cpu; 1347} 1348 1349#define MDF_PUSH 0x01 1350 1351static inline bool is_migration_disabled(struct task_struct *p) 1352{ 1353 return p->migration_disabled; 1354} 1355 1356DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); 1357DECLARE_PER_CPU(struct rnd_state, sched_rnd_state); 1358 1359static inline u32 sched_rng(void) 1360{ 1361 return prandom_u32_state(this_cpu_ptr(&sched_rnd_state)); 1362} 1363 1364#define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) 1365#define this_rq() this_cpu_ptr(&runqueues) 1366#define task_rq(p) cpu_rq(task_cpu(p)) 1367#define cpu_curr(cpu) (cpu_rq(cpu)->curr) 1368#define raw_rq() raw_cpu_ptr(&runqueues) 1369 1370static inline bool idle_rq(struct rq *rq) 1371{ 1372 return rq->curr == rq->idle && !rq->nr_running && !rq->ttwu_pending; 1373} 1374 1375/** 1376 * available_idle_cpu - is a given CPU idle for enqueuing work. 1377 * @cpu: the CPU in question. 1378 * 1379 * Return: 1 if the CPU is currently idle. 0 otherwise. 1380 */ 1381static inline bool available_idle_cpu(int cpu) 1382{ 1383 if (!idle_rq(cpu_rq(cpu))) 1384 return 0; 1385 1386 if (vcpu_is_preempted(cpu)) 1387 return 0; 1388 1389 return 1; 1390} 1391 1392#ifdef CONFIG_SCHED_PROXY_EXEC 1393static inline void rq_set_donor(struct rq *rq, struct task_struct *t) 1394{ 1395 rcu_assign_pointer(rq->donor, t); 1396} 1397#else 1398static inline void rq_set_donor(struct rq *rq, struct task_struct *t) 1399{ 1400 /* Do nothing */ 1401} 1402#endif 1403 1404#ifdef CONFIG_SCHED_CORE 1405static inline struct cpumask *sched_group_span(struct sched_group *sg); 1406 1407DECLARE_STATIC_KEY_FALSE(__sched_core_enabled); 1408 1409static inline bool sched_core_enabled(struct rq *rq) 1410{ 1411 return static_branch_unlikely(&__sched_core_enabled) && rq->core_enabled; 1412} 1413 1414static inline bool sched_core_disabled(void) 1415{ 1416 return !static_branch_unlikely(&__sched_core_enabled); 1417} 1418 1419/* 1420 * Be careful with this function; not for general use. The return value isn't 1421 * stable unless you actually hold a relevant rq->__lock. 1422 */ 1423static inline raw_spinlock_t *rq_lockp(struct rq *rq) 1424{ 1425 if (sched_core_enabled(rq)) 1426 return &rq->core->__lock; 1427 1428 return &rq->__lock; 1429} 1430 1431static inline raw_spinlock_t *__rq_lockp(struct rq *rq) 1432{ 1433 if (rq->core_enabled) 1434 return &rq->core->__lock; 1435 1436 return &rq->__lock; 1437} 1438 1439extern bool 1440cfs_prio_less(const struct task_struct *a, const struct task_struct *b, bool fi); 1441 1442extern void task_vruntime_update(struct rq *rq, struct task_struct *p, bool in_fi); 1443 1444/* 1445 * Helpers to check if the CPU's core cookie matches with the task's cookie 1446 * when core scheduling is enabled. 1447 * A special case is that the task's cookie always matches with CPU's core 1448 * cookie if the CPU is in an idle core. 1449 */ 1450static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p) 1451{ 1452 /* Ignore cookie match if core scheduler is not enabled on the CPU. */ 1453 if (!sched_core_enabled(rq)) 1454 return true; 1455 1456 return rq->core->core_cookie == p->core_cookie; 1457} 1458 1459static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p) 1460{ 1461 bool idle_core = true; 1462 int cpu; 1463 1464 /* Ignore cookie match if core scheduler is not enabled on the CPU. */ 1465 if (!sched_core_enabled(rq)) 1466 return true; 1467 1468 if (rq->core->core_cookie == p->core_cookie) 1469 return true; 1470 1471 for_each_cpu(cpu, cpu_smt_mask(cpu_of(rq))) { 1472 if (!available_idle_cpu(cpu)) { 1473 idle_core = false; 1474 break; 1475 } 1476 } 1477 1478 /* 1479 * A CPU in an idle core is always the best choice for tasks with 1480 * cookies. 1481 */ 1482 return idle_core; 1483} 1484 1485static inline bool sched_group_cookie_match(struct rq *rq, 1486 struct task_struct *p, 1487 struct sched_group *group) 1488{ 1489 int cpu; 1490 1491 /* Ignore cookie match if core scheduler is not enabled on the CPU. */ 1492 if (!sched_core_enabled(rq)) 1493 return true; 1494 1495 for_each_cpu_and(cpu, sched_group_span(group), p->cpus_ptr) { 1496 if (sched_core_cookie_match(cpu_rq(cpu), p)) 1497 return true; 1498 } 1499 return false; 1500} 1501 1502static inline bool sched_core_enqueued(struct task_struct *p) 1503{ 1504 return !RB_EMPTY_NODE(&p->core_node); 1505} 1506 1507extern void sched_core_enqueue(struct rq *rq, struct task_struct *p); 1508extern void sched_core_dequeue(struct rq *rq, struct task_struct *p, int flags); 1509 1510extern void sched_core_get(void); 1511extern void sched_core_put(void); 1512 1513#else /* !CONFIG_SCHED_CORE: */ 1514 1515static inline bool sched_core_enabled(struct rq *rq) 1516{ 1517 return false; 1518} 1519 1520static inline bool sched_core_disabled(void) 1521{ 1522 return true; 1523} 1524 1525static inline raw_spinlock_t *rq_lockp(struct rq *rq) 1526{ 1527 return &rq->__lock; 1528} 1529 1530static inline raw_spinlock_t *__rq_lockp(struct rq *rq) 1531{ 1532 return &rq->__lock; 1533} 1534 1535static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p) 1536{ 1537 return true; 1538} 1539 1540static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p) 1541{ 1542 return true; 1543} 1544 1545static inline bool sched_group_cookie_match(struct rq *rq, 1546 struct task_struct *p, 1547 struct sched_group *group) 1548{ 1549 return true; 1550} 1551 1552#endif /* !CONFIG_SCHED_CORE */ 1553 1554#ifdef CONFIG_RT_GROUP_SCHED 1555# ifdef CONFIG_RT_GROUP_SCHED_DEFAULT_DISABLED 1556DECLARE_STATIC_KEY_FALSE(rt_group_sched); 1557static inline bool rt_group_sched_enabled(void) 1558{ 1559 return static_branch_unlikely(&rt_group_sched); 1560} 1561# else /* !CONFIG_RT_GROUP_SCHED_DEFAULT_DISABLED: */ 1562DECLARE_STATIC_KEY_TRUE(rt_group_sched); 1563static inline bool rt_group_sched_enabled(void) 1564{ 1565 return static_branch_likely(&rt_group_sched); 1566} 1567# endif /* !CONFIG_RT_GROUP_SCHED_DEFAULT_DISABLED */ 1568#else /* !CONFIG_RT_GROUP_SCHED: */ 1569# define rt_group_sched_enabled() false 1570#endif /* !CONFIG_RT_GROUP_SCHED */ 1571 1572static inline void lockdep_assert_rq_held(struct rq *rq) 1573{ 1574 lockdep_assert_held(__rq_lockp(rq)); 1575} 1576 1577extern void raw_spin_rq_lock_nested(struct rq *rq, int subclass); 1578extern bool raw_spin_rq_trylock(struct rq *rq); 1579extern void raw_spin_rq_unlock(struct rq *rq); 1580 1581static inline void raw_spin_rq_lock(struct rq *rq) 1582{ 1583 raw_spin_rq_lock_nested(rq, 0); 1584} 1585 1586static inline void raw_spin_rq_lock_irq(struct rq *rq) 1587{ 1588 local_irq_disable(); 1589 raw_spin_rq_lock(rq); 1590} 1591 1592static inline void raw_spin_rq_unlock_irq(struct rq *rq) 1593{ 1594 raw_spin_rq_unlock(rq); 1595 local_irq_enable(); 1596} 1597 1598static inline unsigned long _raw_spin_rq_lock_irqsave(struct rq *rq) 1599{ 1600 unsigned long flags; 1601 1602 local_irq_save(flags); 1603 raw_spin_rq_lock(rq); 1604 1605 return flags; 1606} 1607 1608static inline void raw_spin_rq_unlock_irqrestore(struct rq *rq, unsigned long flags) 1609{ 1610 raw_spin_rq_unlock(rq); 1611 local_irq_restore(flags); 1612} 1613 1614#define raw_spin_rq_lock_irqsave(rq, flags) \ 1615do { \ 1616 flags = _raw_spin_rq_lock_irqsave(rq); \ 1617} while (0) 1618 1619#ifdef CONFIG_SCHED_SMT 1620extern void __update_idle_core(struct rq *rq); 1621 1622static inline void update_idle_core(struct rq *rq) 1623{ 1624 if (static_branch_unlikely(&sched_smt_present)) 1625 __update_idle_core(rq); 1626} 1627 1628#else /* !CONFIG_SCHED_SMT: */ 1629static inline void update_idle_core(struct rq *rq) { } 1630#endif /* !CONFIG_SCHED_SMT */ 1631 1632#ifdef CONFIG_FAIR_GROUP_SCHED 1633 1634static inline struct task_struct *task_of(struct sched_entity *se) 1635{ 1636 WARN_ON_ONCE(!entity_is_task(se)); 1637 return container_of(se, struct task_struct, se); 1638} 1639 1640static inline struct cfs_rq *task_cfs_rq(struct task_struct *p) 1641{ 1642 return p->se.cfs_rq; 1643} 1644 1645/* runqueue on which this entity is (to be) queued */ 1646static inline struct cfs_rq *cfs_rq_of(const struct sched_entity *se) 1647{ 1648 return se->cfs_rq; 1649} 1650 1651/* runqueue "owned" by this group */ 1652static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp) 1653{ 1654 return grp->my_q; 1655} 1656 1657#else /* !CONFIG_FAIR_GROUP_SCHED: */ 1658 1659#define task_of(_se) container_of(_se, struct task_struct, se) 1660 1661static inline struct cfs_rq *task_cfs_rq(const struct task_struct *p) 1662{ 1663 return &task_rq(p)->cfs; 1664} 1665 1666static inline struct cfs_rq *cfs_rq_of(const struct sched_entity *se) 1667{ 1668 const struct task_struct *p = task_of(se); 1669 struct rq *rq = task_rq(p); 1670 1671 return &rq->cfs; 1672} 1673 1674/* runqueue "owned" by this group */ 1675static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp) 1676{ 1677 return NULL; 1678} 1679 1680#endif /* !CONFIG_FAIR_GROUP_SCHED */ 1681 1682extern void update_rq_clock(struct rq *rq); 1683 1684/* 1685 * rq::clock_update_flags bits 1686 * 1687 * %RQCF_REQ_SKIP - will request skipping of clock update on the next 1688 * call to __schedule(). This is an optimisation to avoid 1689 * neighbouring rq clock updates. 1690 * 1691 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is 1692 * in effect and calls to update_rq_clock() are being ignored. 1693 * 1694 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been 1695 * made to update_rq_clock() since the last time rq::lock was pinned. 1696 * 1697 * If inside of __schedule(), clock_update_flags will have been 1698 * shifted left (a left shift is a cheap operation for the fast path 1699 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use, 1700 * 1701 * if (rq-clock_update_flags >= RQCF_UPDATED) 1702 * 1703 * to check if %RQCF_UPDATED is set. It'll never be shifted more than 1704 * one position though, because the next rq_unpin_lock() will shift it 1705 * back. 1706 */ 1707#define RQCF_REQ_SKIP 0x01 1708#define RQCF_ACT_SKIP 0x02 1709#define RQCF_UPDATED 0x04 1710 1711static inline void assert_clock_updated(struct rq *rq) 1712{ 1713 /* 1714 * The only reason for not seeing a clock update since the 1715 * last rq_pin_lock() is if we're currently skipping updates. 1716 */ 1717 WARN_ON_ONCE(rq->clock_update_flags < RQCF_ACT_SKIP); 1718} 1719 1720static inline u64 rq_clock(struct rq *rq) 1721{ 1722 lockdep_assert_rq_held(rq); 1723 assert_clock_updated(rq); 1724 1725 return rq->clock; 1726} 1727 1728static inline u64 rq_clock_task(struct rq *rq) 1729{ 1730 lockdep_assert_rq_held(rq); 1731 assert_clock_updated(rq); 1732 1733 return rq->clock_task; 1734} 1735 1736static inline void rq_clock_skip_update(struct rq *rq) 1737{ 1738 lockdep_assert_rq_held(rq); 1739 rq->clock_update_flags |= RQCF_REQ_SKIP; 1740} 1741 1742/* 1743 * See rt task throttling, which is the only time a skip 1744 * request is canceled. 1745 */ 1746static inline void rq_clock_cancel_skipupdate(struct rq *rq) 1747{ 1748 lockdep_assert_rq_held(rq); 1749 rq->clock_update_flags &= ~RQCF_REQ_SKIP; 1750} 1751 1752/* 1753 * During cpu offlining and rq wide unthrottling, we can trigger 1754 * an update_rq_clock() for several cfs and rt runqueues (Typically 1755 * when using list_for_each_entry_*) 1756 * rq_clock_start_loop_update() can be called after updating the clock 1757 * once and before iterating over the list to prevent multiple update. 1758 * After the iterative traversal, we need to call rq_clock_stop_loop_update() 1759 * to clear RQCF_ACT_SKIP of rq->clock_update_flags. 1760 */ 1761static inline void rq_clock_start_loop_update(struct rq *rq) 1762{ 1763 lockdep_assert_rq_held(rq); 1764 WARN_ON_ONCE(rq->clock_update_flags & RQCF_ACT_SKIP); 1765 rq->clock_update_flags |= RQCF_ACT_SKIP; 1766} 1767 1768static inline void rq_clock_stop_loop_update(struct rq *rq) 1769{ 1770 lockdep_assert_rq_held(rq); 1771 rq->clock_update_flags &= ~RQCF_ACT_SKIP; 1772} 1773 1774struct rq_flags { 1775 unsigned long flags; 1776 struct pin_cookie cookie; 1777 /* 1778 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the 1779 * current pin context is stashed here in case it needs to be 1780 * restored in rq_repin_lock(). 1781 */ 1782 unsigned int clock_update_flags; 1783}; 1784 1785extern struct balance_callback balance_push_callback; 1786 1787#ifdef CONFIG_SCHED_CLASS_EXT 1788extern const struct sched_class ext_sched_class; 1789 1790DECLARE_STATIC_KEY_FALSE(__scx_enabled); /* SCX BPF scheduler loaded */ 1791DECLARE_STATIC_KEY_FALSE(__scx_switched_all); /* all fair class tasks on SCX */ 1792 1793#define scx_enabled() static_branch_unlikely(&__scx_enabled) 1794#define scx_switched_all() static_branch_unlikely(&__scx_switched_all) 1795 1796static inline void scx_rq_clock_update(struct rq *rq, u64 clock) 1797{ 1798 if (!scx_enabled()) 1799 return; 1800 WRITE_ONCE(rq->scx.clock, clock); 1801 smp_store_release(&rq->scx.flags, rq->scx.flags | SCX_RQ_CLK_VALID); 1802} 1803 1804static inline void scx_rq_clock_invalidate(struct rq *rq) 1805{ 1806 if (!scx_enabled()) 1807 return; 1808 WRITE_ONCE(rq->scx.flags, rq->scx.flags & ~SCX_RQ_CLK_VALID); 1809} 1810 1811#else /* !CONFIG_SCHED_CLASS_EXT: */ 1812#define scx_enabled() false 1813#define scx_switched_all() false 1814 1815static inline void scx_rq_clock_update(struct rq *rq, u64 clock) {} 1816static inline void scx_rq_clock_invalidate(struct rq *rq) {} 1817#endif /* !CONFIG_SCHED_CLASS_EXT */ 1818 1819/* 1820 * Lockdep annotation that avoids accidental unlocks; it's like a 1821 * sticky/continuous lockdep_assert_held(). 1822 * 1823 * This avoids code that has access to 'struct rq *rq' (basically everything in 1824 * the scheduler) from accidentally unlocking the rq if they do not also have a 1825 * copy of the (on-stack) 'struct rq_flags rf'. 1826 * 1827 * Also see Documentation/locking/lockdep-design.rst. 1828 */ 1829static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf) 1830{ 1831 rf->cookie = lockdep_pin_lock(__rq_lockp(rq)); 1832 1833 rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP); 1834 rf->clock_update_flags = 0; 1835 WARN_ON_ONCE(rq->balance_callback && rq->balance_callback != &balance_push_callback); 1836} 1837 1838static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf) 1839{ 1840 if (rq->clock_update_flags > RQCF_ACT_SKIP) 1841 rf->clock_update_flags = RQCF_UPDATED; 1842 1843 scx_rq_clock_invalidate(rq); 1844 lockdep_unpin_lock(__rq_lockp(rq), rf->cookie); 1845} 1846 1847static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf) 1848{ 1849 lockdep_repin_lock(__rq_lockp(rq), rf->cookie); 1850 1851 /* 1852 * Restore the value we stashed in @rf for this pin context. 1853 */ 1854 rq->clock_update_flags |= rf->clock_update_flags; 1855} 1856 1857extern 1858struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf) 1859 __acquires(rq->lock); 1860 1861extern 1862struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf) 1863 __acquires(p->pi_lock) 1864 __acquires(rq->lock); 1865 1866static inline void 1867__task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf) 1868 __releases(rq->lock) 1869{ 1870 rq_unpin_lock(rq, rf); 1871 raw_spin_rq_unlock(rq); 1872} 1873 1874static inline void 1875task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf) 1876 __releases(rq->lock) 1877 __releases(p->pi_lock) 1878{ 1879 __task_rq_unlock(rq, p, rf); 1880 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags); 1881} 1882 1883DEFINE_LOCK_GUARD_1(task_rq_lock, struct task_struct, 1884 _T->rq = task_rq_lock(_T->lock, &_T->rf), 1885 task_rq_unlock(_T->rq, _T->lock, &_T->rf), 1886 struct rq *rq; struct rq_flags rf) 1887 1888DEFINE_LOCK_GUARD_1(__task_rq_lock, struct task_struct, 1889 _T->rq = __task_rq_lock(_T->lock, &_T->rf), 1890 __task_rq_unlock(_T->rq, _T->lock, &_T->rf), 1891 struct rq *rq; struct rq_flags rf) 1892 1893static inline void rq_lock_irqsave(struct rq *rq, struct rq_flags *rf) 1894 __acquires(rq->lock) 1895{ 1896 raw_spin_rq_lock_irqsave(rq, rf->flags); 1897 rq_pin_lock(rq, rf); 1898} 1899 1900static inline void rq_lock_irq(struct rq *rq, struct rq_flags *rf) 1901 __acquires(rq->lock) 1902{ 1903 raw_spin_rq_lock_irq(rq); 1904 rq_pin_lock(rq, rf); 1905} 1906 1907static inline void rq_lock(struct rq *rq, struct rq_flags *rf) 1908 __acquires(rq->lock) 1909{ 1910 raw_spin_rq_lock(rq); 1911 rq_pin_lock(rq, rf); 1912} 1913 1914static inline void rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf) 1915 __releases(rq->lock) 1916{ 1917 rq_unpin_lock(rq, rf); 1918 raw_spin_rq_unlock_irqrestore(rq, rf->flags); 1919} 1920 1921static inline void rq_unlock_irq(struct rq *rq, struct rq_flags *rf) 1922 __releases(rq->lock) 1923{ 1924 rq_unpin_lock(rq, rf); 1925 raw_spin_rq_unlock_irq(rq); 1926} 1927 1928static inline void rq_unlock(struct rq *rq, struct rq_flags *rf) 1929 __releases(rq->lock) 1930{ 1931 rq_unpin_lock(rq, rf); 1932 raw_spin_rq_unlock(rq); 1933} 1934 1935DEFINE_LOCK_GUARD_1(rq_lock, struct rq, 1936 rq_lock(_T->lock, &_T->rf), 1937 rq_unlock(_T->lock, &_T->rf), 1938 struct rq_flags rf) 1939 1940DEFINE_LOCK_GUARD_1(rq_lock_irq, struct rq, 1941 rq_lock_irq(_T->lock, &_T->rf), 1942 rq_unlock_irq(_T->lock, &_T->rf), 1943 struct rq_flags rf) 1944 1945DEFINE_LOCK_GUARD_1(rq_lock_irqsave, struct rq, 1946 rq_lock_irqsave(_T->lock, &_T->rf), 1947 rq_unlock_irqrestore(_T->lock, &_T->rf), 1948 struct rq_flags rf) 1949 1950static inline struct rq *this_rq_lock_irq(struct rq_flags *rf) 1951 __acquires(rq->lock) 1952{ 1953 struct rq *rq; 1954 1955 local_irq_disable(); 1956 rq = this_rq(); 1957 rq_lock(rq, rf); 1958 1959 return rq; 1960} 1961 1962#ifdef CONFIG_NUMA 1963 1964enum numa_topology_type { 1965 NUMA_DIRECT, 1966 NUMA_GLUELESS_MESH, 1967 NUMA_BACKPLANE, 1968}; 1969 1970extern enum numa_topology_type sched_numa_topology_type; 1971extern int sched_max_numa_distance; 1972extern bool find_numa_distance(int distance); 1973extern void sched_init_numa(int offline_node); 1974extern void sched_update_numa(int cpu, bool online); 1975extern void sched_domains_numa_masks_set(unsigned int cpu); 1976extern void sched_domains_numa_masks_clear(unsigned int cpu); 1977extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu); 1978 1979#else /* !CONFIG_NUMA: */ 1980 1981static inline void sched_init_numa(int offline_node) { } 1982static inline void sched_update_numa(int cpu, bool online) { } 1983static inline void sched_domains_numa_masks_set(unsigned int cpu) { } 1984static inline void sched_domains_numa_masks_clear(unsigned int cpu) { } 1985 1986static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu) 1987{ 1988 return nr_cpu_ids; 1989} 1990 1991#endif /* !CONFIG_NUMA */ 1992 1993#ifdef CONFIG_NUMA_BALANCING 1994 1995/* The regions in numa_faults array from task_struct */ 1996enum numa_faults_stats { 1997 NUMA_MEM = 0, 1998 NUMA_CPU, 1999 NUMA_MEMBUF, 2000 NUMA_CPUBUF 2001}; 2002 2003extern void sched_setnuma(struct task_struct *p, int node); 2004extern int migrate_task_to(struct task_struct *p, int cpu); 2005extern int migrate_swap(struct task_struct *p, struct task_struct *t, 2006 int cpu, int scpu); 2007extern void init_numa_balancing(u64 clone_flags, struct task_struct *p); 2008 2009#else /* !CONFIG_NUMA_BALANCING: */ 2010 2011static inline void 2012init_numa_balancing(u64 clone_flags, struct task_struct *p) 2013{ 2014} 2015 2016#endif /* !CONFIG_NUMA_BALANCING */ 2017 2018static inline void 2019queue_balance_callback(struct rq *rq, 2020 struct balance_callback *head, 2021 void (*func)(struct rq *rq)) 2022{ 2023 lockdep_assert_rq_held(rq); 2024 2025 /* 2026 * Don't (re)queue an already queued item; nor queue anything when 2027 * balance_push() is active, see the comment with 2028 * balance_push_callback. 2029 */ 2030 if (unlikely(head->next || rq->balance_callback == &balance_push_callback)) 2031 return; 2032 2033 head->func = func; 2034 head->next = rq->balance_callback; 2035 rq->balance_callback = head; 2036} 2037 2038#define rcu_dereference_check_sched_domain(p) \ 2039 rcu_dereference_check((p), lockdep_is_held(&sched_domains_mutex)) 2040 2041/* 2042 * The domain tree (rq->sd) is protected by RCU's quiescent state transition. 2043 * See destroy_sched_domains: call_rcu for details. 2044 * 2045 * The domain tree of any CPU may only be accessed from within 2046 * preempt-disabled sections. 2047 */ 2048#define for_each_domain(cpu, __sd) \ 2049 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \ 2050 __sd; __sd = __sd->parent) 2051 2052/* A mask of all the SD flags that have the SDF_SHARED_CHILD metaflag */ 2053#define SD_FLAG(name, mflags) (name * !!((mflags) & SDF_SHARED_CHILD)) | 2054static const unsigned int SD_SHARED_CHILD_MASK = 2055#include <linux/sched/sd_flags.h> 20560; 2057#undef SD_FLAG 2058 2059/** 2060 * highest_flag_domain - Return highest sched_domain containing flag. 2061 * @cpu: The CPU whose highest level of sched domain is to 2062 * be returned. 2063 * @flag: The flag to check for the highest sched_domain 2064 * for the given CPU. 2065 * 2066 * Returns the highest sched_domain of a CPU which contains @flag. If @flag has 2067 * the SDF_SHARED_CHILD metaflag, all the children domains also have @flag. 2068 */ 2069static inline struct sched_domain *highest_flag_domain(int cpu, int flag) 2070{ 2071 struct sched_domain *sd, *hsd = NULL; 2072 2073 for_each_domain(cpu, sd) { 2074 if (sd->flags & flag) { 2075 hsd = sd; 2076 continue; 2077 } 2078 2079 /* 2080 * Stop the search if @flag is known to be shared at lower 2081 * levels. It will not be found further up. 2082 */ 2083 if (flag & SD_SHARED_CHILD_MASK) 2084 break; 2085 } 2086 2087 return hsd; 2088} 2089 2090static inline struct sched_domain *lowest_flag_domain(int cpu, int flag) 2091{ 2092 struct sched_domain *sd; 2093 2094 for_each_domain(cpu, sd) { 2095 if (sd->flags & flag) 2096 break; 2097 } 2098 2099 return sd; 2100} 2101 2102DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc); 2103DECLARE_PER_CPU(int, sd_llc_size); 2104DECLARE_PER_CPU(int, sd_llc_id); 2105DECLARE_PER_CPU(int, sd_share_id); 2106DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared); 2107DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa); 2108DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing); 2109DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity); 2110 2111extern struct static_key_false sched_asym_cpucapacity; 2112extern struct static_key_false sched_cluster_active; 2113 2114static __always_inline bool sched_asym_cpucap_active(void) 2115{ 2116 return static_branch_unlikely(&sched_asym_cpucapacity); 2117} 2118 2119struct sched_group_capacity { 2120 atomic_t ref; 2121 /* 2122 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity 2123 * for a single CPU. 2124 */ 2125 unsigned long capacity; 2126 unsigned long min_capacity; /* Min per-CPU capacity in group */ 2127 unsigned long max_capacity; /* Max per-CPU capacity in group */ 2128 unsigned long next_update; 2129 int imbalance; /* XXX unrelated to capacity but shared group state */ 2130 2131 int id; 2132 2133 unsigned long cpumask[]; /* Balance mask */ 2134}; 2135 2136struct sched_group { 2137 struct sched_group *next; /* Must be a circular list */ 2138 atomic_t ref; 2139 2140 unsigned int group_weight; 2141 unsigned int cores; 2142 struct sched_group_capacity *sgc; 2143 int asym_prefer_cpu; /* CPU of highest priority in group */ 2144 int flags; 2145 2146 /* 2147 * The CPUs this group covers. 2148 * 2149 * NOTE: this field is variable length. (Allocated dynamically 2150 * by attaching extra space to the end of the structure, 2151 * depending on how many CPUs the kernel has booted up with) 2152 */ 2153 unsigned long cpumask[]; 2154}; 2155 2156static inline struct cpumask *sched_group_span(struct sched_group *sg) 2157{ 2158 return to_cpumask(sg->cpumask); 2159} 2160 2161/* 2162 * See build_balance_mask(). 2163 */ 2164static inline struct cpumask *group_balance_mask(struct sched_group *sg) 2165{ 2166 return to_cpumask(sg->sgc->cpumask); 2167} 2168 2169extern int group_balance_cpu(struct sched_group *sg); 2170 2171extern void update_sched_domain_debugfs(void); 2172extern void dirty_sched_domain_sysctl(int cpu); 2173 2174extern int sched_update_scaling(void); 2175 2176static inline const struct cpumask *task_user_cpus(struct task_struct *p) 2177{ 2178 if (!p->user_cpus_ptr) 2179 return cpu_possible_mask; /* &init_task.cpus_mask */ 2180 return p->user_cpus_ptr; 2181} 2182 2183#ifdef CONFIG_CGROUP_SCHED 2184 2185/* 2186 * Return the group to which this tasks belongs. 2187 * 2188 * We cannot use task_css() and friends because the cgroup subsystem 2189 * changes that value before the cgroup_subsys::attach() method is called, 2190 * therefore we cannot pin it and might observe the wrong value. 2191 * 2192 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup 2193 * core changes this before calling sched_move_task(). 2194 * 2195 * Instead we use a 'copy' which is updated from sched_move_task() while 2196 * holding both task_struct::pi_lock and rq::lock. 2197 */ 2198static inline struct task_group *task_group(struct task_struct *p) 2199{ 2200 return p->sched_task_group; 2201} 2202 2203/* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ 2204static inline void set_task_rq(struct task_struct *p, unsigned int cpu) 2205{ 2206#if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED) 2207 struct task_group *tg = task_group(p); 2208#endif 2209 2210#ifdef CONFIG_FAIR_GROUP_SCHED 2211 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]); 2212 p->se.cfs_rq = tg->cfs_rq[cpu]; 2213 p->se.parent = tg->se[cpu]; 2214 p->se.depth = tg->se[cpu] ? tg->se[cpu]->depth + 1 : 0; 2215#endif 2216 2217#ifdef CONFIG_RT_GROUP_SCHED 2218 /* 2219 * p->rt.rt_rq is NULL initially and it is easier to assign 2220 * root_task_group's rt_rq than switching in rt_rq_of_se() 2221 * Clobbers tg(!) 2222 */ 2223 if (!rt_group_sched_enabled()) 2224 tg = &root_task_group; 2225 p->rt.rt_rq = tg->rt_rq[cpu]; 2226 p->rt.parent = tg->rt_se[cpu]; 2227#endif /* CONFIG_RT_GROUP_SCHED */ 2228} 2229 2230#else /* !CONFIG_CGROUP_SCHED: */ 2231 2232static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } 2233 2234static inline struct task_group *task_group(struct task_struct *p) 2235{ 2236 return NULL; 2237} 2238 2239#endif /* !CONFIG_CGROUP_SCHED */ 2240 2241static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) 2242{ 2243 set_task_rq(p, cpu); 2244#ifdef CONFIG_SMP 2245 /* 2246 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be 2247 * successfully executed on another CPU. We must ensure that updates of 2248 * per-task data have been completed by this moment. 2249 */ 2250 smp_wmb(); 2251 WRITE_ONCE(task_thread_info(p)->cpu, cpu); 2252 p->wake_cpu = cpu; 2253 rseq_sched_set_ids_changed(p); 2254#endif /* CONFIG_SMP */ 2255} 2256 2257/* 2258 * Tunables: 2259 */ 2260 2261#define SCHED_FEAT(name, enabled) \ 2262 __SCHED_FEAT_##name , 2263 2264enum { 2265#include "features.h" 2266 __SCHED_FEAT_NR, 2267}; 2268 2269#undef SCHED_FEAT 2270 2271/* 2272 * To support run-time toggling of sched features, all the translation units 2273 * (but core.c) reference the sysctl_sched_features defined in core.c. 2274 */ 2275extern __read_mostly unsigned int sysctl_sched_features; 2276 2277#ifdef CONFIG_JUMP_LABEL 2278 2279#define SCHED_FEAT(name, enabled) \ 2280static __always_inline bool static_branch_##name(struct static_key *key) \ 2281{ \ 2282 return static_key_##enabled(key); \ 2283} 2284 2285#include "features.h" 2286#undef SCHED_FEAT 2287 2288extern struct static_key sched_feat_keys[__SCHED_FEAT_NR]; 2289#define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x])) 2290 2291#else /* !CONFIG_JUMP_LABEL: */ 2292 2293#define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) 2294 2295#endif /* !CONFIG_JUMP_LABEL */ 2296 2297extern struct static_key_false sched_numa_balancing; 2298extern struct static_key_false sched_schedstats; 2299 2300static inline u64 global_rt_period(void) 2301{ 2302 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; 2303} 2304 2305static inline u64 global_rt_runtime(void) 2306{ 2307 if (sysctl_sched_rt_runtime < 0) 2308 return RUNTIME_INF; 2309 2310 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; 2311} 2312 2313/* 2314 * Is p the current execution context? 2315 */ 2316static inline int task_current(struct rq *rq, struct task_struct *p) 2317{ 2318 return rq->curr == p; 2319} 2320 2321/* 2322 * Is p the current scheduling context? 2323 * 2324 * Note that it might be the current execution context at the same time if 2325 * rq->curr == rq->donor == p. 2326 */ 2327static inline int task_current_donor(struct rq *rq, struct task_struct *p) 2328{ 2329 return rq->donor == p; 2330} 2331 2332static inline bool task_is_blocked(struct task_struct *p) 2333{ 2334 if (!sched_proxy_exec()) 2335 return false; 2336 2337 return !!p->blocked_on; 2338} 2339 2340static inline int task_on_cpu(struct rq *rq, struct task_struct *p) 2341{ 2342 return p->on_cpu; 2343} 2344 2345static inline int task_on_rq_queued(struct task_struct *p) 2346{ 2347 return READ_ONCE(p->on_rq) == TASK_ON_RQ_QUEUED; 2348} 2349 2350static inline int task_on_rq_migrating(struct task_struct *p) 2351{ 2352 return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING; 2353} 2354 2355/* Wake flags. The first three directly map to some SD flag value */ 2356#define WF_EXEC 0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */ 2357#define WF_FORK 0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */ 2358#define WF_TTWU 0x08 /* Wakeup; maps to SD_BALANCE_WAKE */ 2359 2360#define WF_SYNC 0x10 /* Waker goes to sleep after wakeup */ 2361#define WF_MIGRATED 0x20 /* Internal use, task got migrated */ 2362#define WF_CURRENT_CPU 0x40 /* Prefer to move the wakee to the current CPU. */ 2363#define WF_RQ_SELECTED 0x80 /* ->select_task_rq() was called */ 2364 2365static_assert(WF_EXEC == SD_BALANCE_EXEC); 2366static_assert(WF_FORK == SD_BALANCE_FORK); 2367static_assert(WF_TTWU == SD_BALANCE_WAKE); 2368 2369/* 2370 * To aid in avoiding the subversion of "niceness" due to uneven distribution 2371 * of tasks with abnormal "nice" values across CPUs the contribution that 2372 * each task makes to its run queue's load is weighted according to its 2373 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a 2374 * scaled version of the new time slice allocation that they receive on time 2375 * slice expiry etc. 2376 */ 2377 2378#define WEIGHT_IDLEPRIO 3 2379#define WMULT_IDLEPRIO 1431655765 2380 2381extern const int sched_prio_to_weight[40]; 2382extern const u32 sched_prio_to_wmult[40]; 2383 2384/* 2385 * {de,en}queue flags: 2386 * 2387 * SLEEP/WAKEUP - task is no-longer/just-became runnable 2388 * 2389 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks 2390 * are in a known state which allows modification. Such pairs 2391 * should preserve as much state as possible. 2392 * 2393 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location 2394 * in the runqueue. IOW the priority is allowed to change. Callers 2395 * must expect to deal with balance callbacks. 2396 * 2397 * NOCLOCK - skip the update_rq_clock() (avoids double updates) 2398 * 2399 * MIGRATION - p->on_rq == TASK_ON_RQ_MIGRATING (used for DEADLINE) 2400 * 2401 * DELAYED - de/re-queue a sched_delayed task 2402 * 2403 * CLASS - going to update p->sched_class; makes sched_change call the 2404 * various switch methods. 2405 * 2406 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified) 2407 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline) 2408 * ENQUEUE_MIGRATED - the task was migrated during wakeup 2409 * ENQUEUE_RQ_SELECTED - ->select_task_rq() was called 2410 * 2411 * XXX SAVE/RESTORE in combination with CLASS doesn't really make sense, but 2412 * SCHED_DEADLINE seems to rely on this for now. 2413 */ 2414 2415#define DEQUEUE_SLEEP 0x0001 /* Matches ENQUEUE_WAKEUP */ 2416#define DEQUEUE_SAVE 0x0002 /* Matches ENQUEUE_RESTORE */ 2417#define DEQUEUE_MOVE 0x0004 /* Matches ENQUEUE_MOVE */ 2418#define DEQUEUE_NOCLOCK 0x0008 /* Matches ENQUEUE_NOCLOCK */ 2419 2420#define DEQUEUE_MIGRATING 0x0010 /* Matches ENQUEUE_MIGRATING */ 2421#define DEQUEUE_DELAYED 0x0020 /* Matches ENQUEUE_DELAYED */ 2422#define DEQUEUE_CLASS 0x0040 /* Matches ENQUEUE_CLASS */ 2423 2424#define DEQUEUE_SPECIAL 0x00010000 2425#define DEQUEUE_THROTTLE 0x00020000 2426 2427#define ENQUEUE_WAKEUP 0x0001 2428#define ENQUEUE_RESTORE 0x0002 2429#define ENQUEUE_MOVE 0x0004 2430#define ENQUEUE_NOCLOCK 0x0008 2431 2432#define ENQUEUE_MIGRATING 0x0010 2433#define ENQUEUE_DELAYED 0x0020 2434#define ENQUEUE_CLASS 0x0040 2435 2436#define ENQUEUE_HEAD 0x00010000 2437#define ENQUEUE_REPLENISH 0x00020000 2438#define ENQUEUE_MIGRATED 0x00040000 2439#define ENQUEUE_INITIAL 0x00080000 2440#define ENQUEUE_RQ_SELECTED 0x00100000 2441 2442#define RETRY_TASK ((void *)-1UL) 2443 2444struct affinity_context { 2445 const struct cpumask *new_mask; 2446 struct cpumask *user_mask; 2447 unsigned int flags; 2448}; 2449 2450extern s64 update_curr_common(struct rq *rq); 2451 2452struct sched_class { 2453 2454#ifdef CONFIG_UCLAMP_TASK 2455 int uclamp_enabled; 2456#endif 2457 /* 2458 * idle: 0 2459 * ext: 1 2460 * fair: 2 2461 * rt: 4 2462 * dl: 8 2463 * stop: 16 2464 */ 2465 unsigned int queue_mask; 2466 2467 /* 2468 * move_queued_task/activate_task/enqueue_task: rq->lock 2469 * ttwu_do_activate/activate_task/enqueue_task: rq->lock 2470 * wake_up_new_task/activate_task/enqueue_task: task_rq_lock 2471 * ttwu_runnable/enqueue_task: task_rq_lock 2472 * proxy_task_current: rq->lock 2473 * sched_change_end 2474 */ 2475 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags); 2476 /* 2477 * move_queued_task/deactivate_task/dequeue_task: rq->lock 2478 * __schedule/block_task/dequeue_task: rq->lock 2479 * proxy_task_current: rq->lock 2480 * wait_task_inactive: task_rq_lock 2481 * sched_change_begin 2482 */ 2483 bool (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags); 2484 2485 /* 2486 * do_sched_yield: rq->lock 2487 */ 2488 void (*yield_task) (struct rq *rq); 2489 /* 2490 * yield_to: rq->lock (double) 2491 */ 2492 bool (*yield_to_task)(struct rq *rq, struct task_struct *p); 2493 2494 /* 2495 * move_queued_task: rq->lock 2496 * __migrate_swap_task: rq->lock 2497 * ttwu_do_activate: rq->lock 2498 * ttwu_runnable: task_rq_lock 2499 * wake_up_new_task: task_rq_lock 2500 */ 2501 void (*wakeup_preempt)(struct rq *rq, struct task_struct *p, int flags); 2502 2503 /* 2504 * schedule/pick_next_task/prev_balance: rq->lock 2505 */ 2506 int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf); 2507 2508 /* 2509 * schedule/pick_next_task: rq->lock 2510 */ 2511 struct task_struct *(*pick_task)(struct rq *rq, struct rq_flags *rf); 2512 /* 2513 * Optional! When implemented pick_next_task() should be equivalent to: 2514 * 2515 * next = pick_task(); 2516 * if (next) { 2517 * put_prev_task(prev); 2518 * set_next_task_first(next); 2519 * } 2520 */ 2521 struct task_struct *(*pick_next_task)(struct rq *rq, struct task_struct *prev, 2522 struct rq_flags *rf); 2523 2524 /* 2525 * sched_change: 2526 * __schedule: rq->lock 2527 */ 2528 void (*put_prev_task)(struct rq *rq, struct task_struct *p, struct task_struct *next); 2529 void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first); 2530 2531 /* 2532 * select_task_rq: p->pi_lock 2533 * sched_exec: p->pi_lock 2534 */ 2535 int (*select_task_rq)(struct task_struct *p, int task_cpu, int flags); 2536 2537 /* 2538 * set_task_cpu: p->pi_lock || rq->lock (ttwu like) 2539 */ 2540 void (*migrate_task_rq)(struct task_struct *p, int new_cpu); 2541 2542 /* 2543 * ttwu_do_activate: rq->lock 2544 * wake_up_new_task: task_rq_lock 2545 */ 2546 void (*task_woken)(struct rq *this_rq, struct task_struct *task); 2547 2548 /* 2549 * do_set_cpus_allowed: task_rq_lock + sched_change 2550 */ 2551 void (*set_cpus_allowed)(struct task_struct *p, struct affinity_context *ctx); 2552 2553 /* 2554 * sched_set_rq_{on,off}line: rq->lock 2555 */ 2556 void (*rq_online)(struct rq *rq); 2557 void (*rq_offline)(struct rq *rq); 2558 2559 /* 2560 * push_cpu_stop: p->pi_lock && rq->lock 2561 */ 2562 struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq); 2563 2564 /* 2565 * hrtick: rq->lock 2566 * sched_tick: rq->lock 2567 * sched_tick_remote: rq->lock 2568 */ 2569 void (*task_tick)(struct rq *rq, struct task_struct *p, int queued); 2570 /* 2571 * sched_cgroup_fork: p->pi_lock 2572 */ 2573 void (*task_fork)(struct task_struct *p); 2574 /* 2575 * finish_task_switch: no locks 2576 */ 2577 void (*task_dead)(struct task_struct *p); 2578 2579 /* 2580 * sched_change 2581 */ 2582 void (*switching_from)(struct rq *this_rq, struct task_struct *task); 2583 void (*switched_from) (struct rq *this_rq, struct task_struct *task); 2584 void (*switching_to) (struct rq *this_rq, struct task_struct *task); 2585 void (*switched_to) (struct rq *this_rq, struct task_struct *task); 2586 u64 (*get_prio) (struct rq *this_rq, struct task_struct *task); 2587 void (*prio_changed) (struct rq *this_rq, struct task_struct *task, 2588 u64 oldprio); 2589 2590 /* 2591 * set_load_weight: task_rq_lock + sched_change 2592 * __setscheduler_parms: task_rq_lock + sched_change 2593 */ 2594 void (*reweight_task)(struct rq *this_rq, struct task_struct *task, 2595 const struct load_weight *lw); 2596 2597 /* 2598 * sched_rr_get_interval: task_rq_lock 2599 */ 2600 unsigned int (*get_rr_interval)(struct rq *rq, 2601 struct task_struct *task); 2602 2603 /* 2604 * task_sched_runtime: task_rq_lock 2605 */ 2606 void (*update_curr)(struct rq *rq); 2607 2608#ifdef CONFIG_FAIR_GROUP_SCHED 2609 /* 2610 * sched_change_group: task_rq_lock + sched_change 2611 */ 2612 void (*task_change_group)(struct task_struct *p); 2613#endif 2614 2615#ifdef CONFIG_SCHED_CORE 2616 /* 2617 * pick_next_task: rq->lock 2618 * try_steal_cookie: rq->lock (double) 2619 */ 2620 int (*task_is_throttled)(struct task_struct *p, int cpu); 2621#endif 2622}; 2623 2624/* 2625 * Does not nest; only used around sched_class::pick_task() rq-lock-breaks. 2626 */ 2627static inline void rq_modified_clear(struct rq *rq) 2628{ 2629 rq->queue_mask = 0; 2630} 2631 2632static inline bool rq_modified_above(struct rq *rq, const struct sched_class * class) 2633{ 2634 unsigned int mask = class->queue_mask; 2635 return rq->queue_mask & ~((mask << 1) - 1); 2636} 2637 2638static inline void put_prev_task(struct rq *rq, struct task_struct *prev) 2639{ 2640 WARN_ON_ONCE(rq->donor != prev); 2641 prev->sched_class->put_prev_task(rq, prev, NULL); 2642} 2643 2644static inline void set_next_task(struct rq *rq, struct task_struct *next) 2645{ 2646 next->sched_class->set_next_task(rq, next, false); 2647} 2648 2649static inline void 2650__put_prev_set_next_dl_server(struct rq *rq, 2651 struct task_struct *prev, 2652 struct task_struct *next) 2653{ 2654 prev->dl_server = NULL; 2655 next->dl_server = rq->dl_server; 2656 rq->dl_server = NULL; 2657} 2658 2659static inline void put_prev_set_next_task(struct rq *rq, 2660 struct task_struct *prev, 2661 struct task_struct *next) 2662{ 2663 WARN_ON_ONCE(rq->donor != prev); 2664 2665 __put_prev_set_next_dl_server(rq, prev, next); 2666 2667 if (next == prev) 2668 return; 2669 2670 prev->sched_class->put_prev_task(rq, prev, next); 2671 next->sched_class->set_next_task(rq, next, true); 2672} 2673 2674/* 2675 * Helper to define a sched_class instance; each one is placed in a separate 2676 * section which is ordered by the linker script: 2677 * 2678 * include/asm-generic/vmlinux.lds.h 2679 * 2680 * *CAREFUL* they are laid out in *REVERSE* order!!! 2681 * 2682 * Also enforce alignment on the instance, not the type, to guarantee layout. 2683 */ 2684#define DEFINE_SCHED_CLASS(name) \ 2685const struct sched_class name##_sched_class \ 2686 __aligned(__alignof__(struct sched_class)) \ 2687 __section("__" #name "_sched_class") 2688 2689/* Defined in include/asm-generic/vmlinux.lds.h */ 2690extern struct sched_class __sched_class_highest[]; 2691extern struct sched_class __sched_class_lowest[]; 2692 2693extern const struct sched_class stop_sched_class; 2694extern const struct sched_class dl_sched_class; 2695extern const struct sched_class rt_sched_class; 2696extern const struct sched_class fair_sched_class; 2697extern const struct sched_class idle_sched_class; 2698 2699/* 2700 * Iterate only active classes. SCX can take over all fair tasks or be 2701 * completely disabled. If the former, skip fair. If the latter, skip SCX. 2702 */ 2703static inline const struct sched_class *next_active_class(const struct sched_class *class) 2704{ 2705 class++; 2706#ifdef CONFIG_SCHED_CLASS_EXT 2707 if (scx_switched_all() && class == &fair_sched_class) 2708 class++; 2709 if (!scx_enabled() && class == &ext_sched_class) 2710 class++; 2711#endif 2712 return class; 2713} 2714 2715#define for_class_range(class, _from, _to) \ 2716 for (class = (_from); class < (_to); class++) 2717 2718#define for_each_class(class) \ 2719 for_class_range(class, __sched_class_highest, __sched_class_lowest) 2720 2721#define for_active_class_range(class, _from, _to) \ 2722 for (class = (_from); class != (_to); class = next_active_class(class)) 2723 2724#define for_each_active_class(class) \ 2725 for_active_class_range(class, __sched_class_highest, __sched_class_lowest) 2726 2727#define sched_class_above(_a, _b) ((_a) < (_b)) 2728 2729static inline bool sched_stop_runnable(struct rq *rq) 2730{ 2731 return rq->stop && task_on_rq_queued(rq->stop); 2732} 2733 2734static inline bool sched_dl_runnable(struct rq *rq) 2735{ 2736 return rq->dl.dl_nr_running > 0; 2737} 2738 2739static inline bool sched_rt_runnable(struct rq *rq) 2740{ 2741 return rq->rt.rt_queued > 0; 2742} 2743 2744static inline bool sched_fair_runnable(struct rq *rq) 2745{ 2746 return rq->cfs.nr_queued > 0; 2747} 2748 2749extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, 2750 struct rq_flags *rf); 2751extern struct task_struct *pick_task_idle(struct rq *rq, struct rq_flags *rf); 2752 2753#define SCA_CHECK 0x01 2754#define SCA_MIGRATE_DISABLE 0x02 2755#define SCA_MIGRATE_ENABLE 0x04 2756#define SCA_USER 0x08 2757 2758extern void update_group_capacity(struct sched_domain *sd, int cpu); 2759 2760extern void sched_balance_trigger(struct rq *rq); 2761 2762extern int __set_cpus_allowed_ptr(struct task_struct *p, struct affinity_context *ctx); 2763extern void set_cpus_allowed_common(struct task_struct *p, struct affinity_context *ctx); 2764 2765static inline bool task_allowed_on_cpu(struct task_struct *p, int cpu) 2766{ 2767 /* When not in the task's cpumask, no point in looking further. */ 2768 if (!cpumask_test_cpu(cpu, p->cpus_ptr)) 2769 return false; 2770 2771 /* Can @cpu run a user thread? */ 2772 if (!(p->flags & PF_KTHREAD) && !task_cpu_possible(cpu, p)) 2773 return false; 2774 2775 return true; 2776} 2777 2778static inline cpumask_t *alloc_user_cpus_ptr(int node) 2779{ 2780 /* 2781 * See set_cpus_allowed_force() above for the rcu_head usage. 2782 */ 2783 int size = max_t(int, cpumask_size(), sizeof(struct rcu_head)); 2784 2785 return kmalloc_node(size, GFP_KERNEL, node); 2786} 2787 2788static inline struct task_struct *get_push_task(struct rq *rq) 2789{ 2790 struct task_struct *p = rq->donor; 2791 2792 lockdep_assert_rq_held(rq); 2793 2794 if (rq->push_busy) 2795 return NULL; 2796 2797 if (p->nr_cpus_allowed == 1) 2798 return NULL; 2799 2800 if (p->migration_disabled) 2801 return NULL; 2802 2803 rq->push_busy = true; 2804 return get_task_struct(p); 2805} 2806 2807extern int push_cpu_stop(void *arg); 2808 2809#ifdef CONFIG_CPU_IDLE 2810 2811static inline void idle_set_state(struct rq *rq, 2812 struct cpuidle_state *idle_state) 2813{ 2814 rq->idle_state = idle_state; 2815} 2816 2817static inline struct cpuidle_state *idle_get_state(struct rq *rq) 2818{ 2819 WARN_ON_ONCE(!rcu_read_lock_held()); 2820 2821 return rq->idle_state; 2822} 2823 2824#else /* !CONFIG_CPU_IDLE: */ 2825 2826static inline void idle_set_state(struct rq *rq, 2827 struct cpuidle_state *idle_state) 2828{ 2829} 2830 2831static inline struct cpuidle_state *idle_get_state(struct rq *rq) 2832{ 2833 return NULL; 2834} 2835 2836#endif /* !CONFIG_CPU_IDLE */ 2837 2838extern void schedule_idle(void); 2839asmlinkage void schedule_user(void); 2840 2841extern void sysrq_sched_debug_show(void); 2842extern void sched_init_granularity(void); 2843extern void update_max_interval(void); 2844 2845extern void init_sched_dl_class(void); 2846extern void init_sched_rt_class(void); 2847extern void init_sched_fair_class(void); 2848 2849extern void resched_curr(struct rq *rq); 2850extern void resched_curr_lazy(struct rq *rq); 2851extern void resched_cpu(int cpu); 2852 2853extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime); 2854extern bool sched_rt_bandwidth_account(struct rt_rq *rt_rq); 2855 2856extern void init_dl_entity(struct sched_dl_entity *dl_se); 2857 2858extern void init_cfs_throttle_work(struct task_struct *p); 2859 2860#define BW_SHIFT 20 2861#define BW_UNIT (1 << BW_SHIFT) 2862#define RATIO_SHIFT 8 2863#define MAX_BW_BITS (64 - BW_SHIFT) 2864#define MAX_BW ((1ULL << MAX_BW_BITS) - 1) 2865 2866extern unsigned long to_ratio(u64 period, u64 runtime); 2867 2868extern void init_entity_runnable_average(struct sched_entity *se); 2869extern void post_init_entity_util_avg(struct task_struct *p); 2870 2871#ifdef CONFIG_NO_HZ_FULL 2872extern bool sched_can_stop_tick(struct rq *rq); 2873extern int __init sched_tick_offload_init(void); 2874 2875/* 2876 * Tick may be needed by tasks in the runqueue depending on their policy and 2877 * requirements. If tick is needed, lets send the target an IPI to kick it out of 2878 * nohz mode if necessary. 2879 */ 2880static inline void sched_update_tick_dependency(struct rq *rq) 2881{ 2882 int cpu = cpu_of(rq); 2883 2884 if (!tick_nohz_full_cpu(cpu)) 2885 return; 2886 2887 if (sched_can_stop_tick(rq)) 2888 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED); 2889 else 2890 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED); 2891} 2892#else /* !CONFIG_NO_HZ_FULL: */ 2893static inline int sched_tick_offload_init(void) { return 0; } 2894static inline void sched_update_tick_dependency(struct rq *rq) { } 2895#endif /* !CONFIG_NO_HZ_FULL */ 2896 2897static inline void add_nr_running(struct rq *rq, unsigned count) 2898{ 2899 unsigned prev_nr = rq->nr_running; 2900 2901 rq->nr_running = prev_nr + count; 2902 if (trace_sched_update_nr_running_tp_enabled()) { 2903 call_trace_sched_update_nr_running(rq, count); 2904 } 2905 2906 if (prev_nr < 2 && rq->nr_running >= 2) 2907 set_rd_overloaded(rq->rd, 1); 2908 2909 sched_update_tick_dependency(rq); 2910} 2911 2912static inline void sub_nr_running(struct rq *rq, unsigned count) 2913{ 2914 rq->nr_running -= count; 2915 if (trace_sched_update_nr_running_tp_enabled()) { 2916 call_trace_sched_update_nr_running(rq, -count); 2917 } 2918 2919 /* Check if we still need preemption */ 2920 sched_update_tick_dependency(rq); 2921} 2922 2923static inline void __block_task(struct rq *rq, struct task_struct *p) 2924{ 2925 if (p->sched_contributes_to_load) 2926 rq->nr_uninterruptible++; 2927 2928 if (p->in_iowait) { 2929 atomic_inc(&rq->nr_iowait); 2930 delayacct_blkio_start(); 2931 } 2932 2933 ASSERT_EXCLUSIVE_WRITER(p->on_rq); 2934 2935 /* 2936 * The moment this write goes through, ttwu() can swoop in and migrate 2937 * this task, rendering our rq->__lock ineffective. 2938 * 2939 * __schedule() try_to_wake_up() 2940 * LOCK rq->__lock LOCK p->pi_lock 2941 * pick_next_task() 2942 * pick_next_task_fair() 2943 * pick_next_entity() 2944 * dequeue_entities() 2945 * __block_task() 2946 * RELEASE p->on_rq = 0 if (p->on_rq && ...) 2947 * break; 2948 * 2949 * ACQUIRE (after ctrl-dep) 2950 * 2951 * cpu = select_task_rq(); 2952 * set_task_cpu(p, cpu); 2953 * ttwu_queue() 2954 * ttwu_do_activate() 2955 * LOCK rq->__lock 2956 * activate_task() 2957 * STORE p->on_rq = 1 2958 * UNLOCK rq->__lock 2959 * 2960 * Callers must ensure to not reference @p after this -- we no longer 2961 * own it. 2962 */ 2963 smp_store_release(&p->on_rq, 0); 2964} 2965 2966extern void activate_task(struct rq *rq, struct task_struct *p, int flags); 2967extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags); 2968 2969extern void wakeup_preempt(struct rq *rq, struct task_struct *p, int flags); 2970 2971#ifdef CONFIG_PREEMPT_RT 2972# define SCHED_NR_MIGRATE_BREAK 8 2973#else 2974# define SCHED_NR_MIGRATE_BREAK 32 2975#endif 2976 2977extern __read_mostly unsigned int sysctl_sched_nr_migrate; 2978extern __read_mostly unsigned int sysctl_sched_migration_cost; 2979 2980extern unsigned int sysctl_sched_base_slice; 2981 2982extern int sysctl_resched_latency_warn_ms; 2983extern int sysctl_resched_latency_warn_once; 2984 2985extern unsigned int sysctl_sched_tunable_scaling; 2986 2987extern unsigned int sysctl_numa_balancing_scan_delay; 2988extern unsigned int sysctl_numa_balancing_scan_period_min; 2989extern unsigned int sysctl_numa_balancing_scan_period_max; 2990extern unsigned int sysctl_numa_balancing_scan_size; 2991extern unsigned int sysctl_numa_balancing_hot_threshold; 2992 2993#ifdef CONFIG_SCHED_HRTICK 2994 2995/* 2996 * Use hrtick when: 2997 * - enabled by features 2998 * - hrtimer is actually high res 2999 */ 3000static inline int hrtick_enabled(struct rq *rq) 3001{ 3002 if (!cpu_active(cpu_of(rq))) 3003 return 0; 3004 return hrtimer_is_hres_active(&rq->hrtick_timer); 3005} 3006 3007static inline int hrtick_enabled_fair(struct rq *rq) 3008{ 3009 if (!sched_feat(HRTICK)) 3010 return 0; 3011 return hrtick_enabled(rq); 3012} 3013 3014static inline int hrtick_enabled_dl(struct rq *rq) 3015{ 3016 if (!sched_feat(HRTICK_DL)) 3017 return 0; 3018 return hrtick_enabled(rq); 3019} 3020 3021extern void hrtick_start(struct rq *rq, u64 delay); 3022 3023#else /* !CONFIG_SCHED_HRTICK: */ 3024 3025static inline int hrtick_enabled_fair(struct rq *rq) 3026{ 3027 return 0; 3028} 3029 3030static inline int hrtick_enabled_dl(struct rq *rq) 3031{ 3032 return 0; 3033} 3034 3035static inline int hrtick_enabled(struct rq *rq) 3036{ 3037 return 0; 3038} 3039 3040#endif /* !CONFIG_SCHED_HRTICK */ 3041 3042#ifndef arch_scale_freq_tick 3043static __always_inline void arch_scale_freq_tick(void) { } 3044#endif 3045 3046#ifndef arch_scale_freq_capacity 3047/** 3048 * arch_scale_freq_capacity - get the frequency scale factor of a given CPU. 3049 * @cpu: the CPU in question. 3050 * 3051 * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e. 3052 * 3053 * f_curr 3054 * ------ * SCHED_CAPACITY_SCALE 3055 * f_max 3056 */ 3057static __always_inline 3058unsigned long arch_scale_freq_capacity(int cpu) 3059{ 3060 return SCHED_CAPACITY_SCALE; 3061} 3062#endif 3063 3064/* 3065 * In double_lock_balance()/double_rq_lock(), we use raw_spin_rq_lock() to 3066 * acquire rq lock instead of rq_lock(). So at the end of these two functions 3067 * we need to call double_rq_clock_clear_update() to clear RQCF_UPDATED of 3068 * rq->clock_update_flags to avoid the WARN_DOUBLE_CLOCK warning. 3069 */ 3070static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2) 3071{ 3072 rq1->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP); 3073 rq2->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP); 3074} 3075 3076#define DEFINE_LOCK_GUARD_2(name, type, _lock, _unlock, ...) \ 3077__DEFINE_UNLOCK_GUARD(name, type, _unlock, type *lock2; __VA_ARGS__) \ 3078static inline class_##name##_t class_##name##_constructor(type *lock, type *lock2) \ 3079{ class_##name##_t _t = { .lock = lock, .lock2 = lock2 }, *_T = &_t; \ 3080 _lock; return _t; } 3081 3082static inline bool rq_order_less(struct rq *rq1, struct rq *rq2) 3083{ 3084#ifdef CONFIG_SCHED_CORE 3085 /* 3086 * In order to not have {0,2},{1,3} turn into into an AB-BA, 3087 * order by core-id first and cpu-id second. 3088 * 3089 * Notably: 3090 * 3091 * double_rq_lock(0,3); will take core-0, core-1 lock 3092 * double_rq_lock(1,2); will take core-1, core-0 lock 3093 * 3094 * when only cpu-id is considered. 3095 */ 3096 if (rq1->core->cpu < rq2->core->cpu) 3097 return true; 3098 if (rq1->core->cpu > rq2->core->cpu) 3099 return false; 3100 3101 /* 3102 * __sched_core_flip() relies on SMT having cpu-id lock order. 3103 */ 3104#endif /* CONFIG_SCHED_CORE */ 3105 return rq1->cpu < rq2->cpu; 3106} 3107 3108extern void double_rq_lock(struct rq *rq1, struct rq *rq2); 3109 3110#ifdef CONFIG_PREEMPTION 3111 3112/* 3113 * fair double_lock_balance: Safely acquires both rq->locks in a fair 3114 * way at the expense of forcing extra atomic operations in all 3115 * invocations. This assures that the double_lock is acquired using the 3116 * same underlying policy as the spinlock_t on this architecture, which 3117 * reduces latency compared to the unfair variant below. However, it 3118 * also adds more overhead and therefore may reduce throughput. 3119 */ 3120static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) 3121 __releases(this_rq->lock) 3122 __acquires(busiest->lock) 3123 __acquires(this_rq->lock) 3124{ 3125 raw_spin_rq_unlock(this_rq); 3126 double_rq_lock(this_rq, busiest); 3127 3128 return 1; 3129} 3130 3131#else /* !CONFIG_PREEMPTION: */ 3132/* 3133 * Unfair double_lock_balance: Optimizes throughput at the expense of 3134 * latency by eliminating extra atomic operations when the locks are 3135 * already in proper order on entry. This favors lower CPU-ids and will 3136 * grant the double lock to lower CPUs over higher ids under contention, 3137 * regardless of entry order into the function. 3138 */ 3139static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) 3140 __releases(this_rq->lock) 3141 __acquires(busiest->lock) 3142 __acquires(this_rq->lock) 3143{ 3144 if (__rq_lockp(this_rq) == __rq_lockp(busiest) || 3145 likely(raw_spin_rq_trylock(busiest))) { 3146 double_rq_clock_clear_update(this_rq, busiest); 3147 return 0; 3148 } 3149 3150 if (rq_order_less(this_rq, busiest)) { 3151 raw_spin_rq_lock_nested(busiest, SINGLE_DEPTH_NESTING); 3152 double_rq_clock_clear_update(this_rq, busiest); 3153 return 0; 3154 } 3155 3156 raw_spin_rq_unlock(this_rq); 3157 double_rq_lock(this_rq, busiest); 3158 3159 return 1; 3160} 3161 3162#endif /* !CONFIG_PREEMPTION */ 3163 3164/* 3165 * double_lock_balance - lock the busiest runqueue, this_rq is locked already. 3166 */ 3167static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest) 3168{ 3169 lockdep_assert_irqs_disabled(); 3170 3171 return _double_lock_balance(this_rq, busiest); 3172} 3173 3174static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) 3175 __releases(busiest->lock) 3176{ 3177 if (__rq_lockp(this_rq) != __rq_lockp(busiest)) 3178 raw_spin_rq_unlock(busiest); 3179 lock_set_subclass(&__rq_lockp(this_rq)->dep_map, 0, _RET_IP_); 3180} 3181 3182static inline void double_lock(spinlock_t *l1, spinlock_t *l2) 3183{ 3184 if (l1 > l2) 3185 swap(l1, l2); 3186 3187 spin_lock(l1); 3188 spin_lock_nested(l2, SINGLE_DEPTH_NESTING); 3189} 3190 3191static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2) 3192{ 3193 if (l1 > l2) 3194 swap(l1, l2); 3195 3196 spin_lock_irq(l1); 3197 spin_lock_nested(l2, SINGLE_DEPTH_NESTING); 3198} 3199 3200static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2) 3201{ 3202 if (l1 > l2) 3203 swap(l1, l2); 3204 3205 raw_spin_lock(l1); 3206 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING); 3207} 3208 3209static inline void double_raw_unlock(raw_spinlock_t *l1, raw_spinlock_t *l2) 3210{ 3211 raw_spin_unlock(l1); 3212 raw_spin_unlock(l2); 3213} 3214 3215DEFINE_LOCK_GUARD_2(double_raw_spinlock, raw_spinlock_t, 3216 double_raw_lock(_T->lock, _T->lock2), 3217 double_raw_unlock(_T->lock, _T->lock2)) 3218 3219/* 3220 * double_rq_unlock - safely unlock two runqueues 3221 * 3222 * Note this does not restore interrupts like task_rq_unlock, 3223 * you need to do so manually after calling. 3224 */ 3225static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) 3226 __releases(rq1->lock) 3227 __releases(rq2->lock) 3228{ 3229 if (__rq_lockp(rq1) != __rq_lockp(rq2)) 3230 raw_spin_rq_unlock(rq2); 3231 else 3232 __release(rq2->lock); 3233 raw_spin_rq_unlock(rq1); 3234} 3235 3236extern void set_rq_online (struct rq *rq); 3237extern void set_rq_offline(struct rq *rq); 3238 3239extern bool sched_smp_initialized; 3240 3241DEFINE_LOCK_GUARD_2(double_rq_lock, struct rq, 3242 double_rq_lock(_T->lock, _T->lock2), 3243 double_rq_unlock(_T->lock, _T->lock2)) 3244 3245extern struct sched_entity *__pick_root_entity(struct cfs_rq *cfs_rq); 3246extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq); 3247extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq); 3248 3249extern bool sched_debug_verbose; 3250 3251extern void print_cfs_stats(struct seq_file *m, int cpu); 3252extern void print_rt_stats(struct seq_file *m, int cpu); 3253extern void print_dl_stats(struct seq_file *m, int cpu); 3254extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq); 3255extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq); 3256extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq); 3257 3258extern void resched_latency_warn(int cpu, u64 latency); 3259 3260#ifdef CONFIG_NUMA_BALANCING 3261extern void show_numa_stats(struct task_struct *p, struct seq_file *m); 3262extern void 3263print_numa_stats(struct seq_file *m, int node, unsigned long tsf, 3264 unsigned long tpf, unsigned long gsf, unsigned long gpf); 3265#endif /* CONFIG_NUMA_BALANCING */ 3266 3267extern void init_cfs_rq(struct cfs_rq *cfs_rq); 3268extern void init_rt_rq(struct rt_rq *rt_rq); 3269extern void init_dl_rq(struct dl_rq *dl_rq); 3270 3271extern void cfs_bandwidth_usage_inc(void); 3272extern void cfs_bandwidth_usage_dec(void); 3273 3274#ifdef CONFIG_NO_HZ_COMMON 3275 3276#define NOHZ_BALANCE_KICK_BIT 0 3277#define NOHZ_STATS_KICK_BIT 1 3278#define NOHZ_NEWILB_KICK_BIT 2 3279#define NOHZ_NEXT_KICK_BIT 3 3280 3281/* Run sched_balance_domains() */ 3282#define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT) 3283/* Update blocked load */ 3284#define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT) 3285/* Update blocked load when entering idle */ 3286#define NOHZ_NEWILB_KICK BIT(NOHZ_NEWILB_KICK_BIT) 3287/* Update nohz.next_balance */ 3288#define NOHZ_NEXT_KICK BIT(NOHZ_NEXT_KICK_BIT) 3289 3290#define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK | NOHZ_NEXT_KICK) 3291 3292#define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags) 3293 3294extern void nohz_balance_exit_idle(struct rq *rq); 3295#else /* !CONFIG_NO_HZ_COMMON: */ 3296static inline void nohz_balance_exit_idle(struct rq *rq) { } 3297#endif /* !CONFIG_NO_HZ_COMMON */ 3298 3299#ifdef CONFIG_NO_HZ_COMMON 3300extern void nohz_run_idle_balance(int cpu); 3301#else 3302static inline void nohz_run_idle_balance(int cpu) { } 3303#endif 3304 3305#include "stats.h" 3306 3307#if defined(CONFIG_SCHED_CORE) && defined(CONFIG_SCHEDSTATS) 3308 3309extern void __sched_core_account_forceidle(struct rq *rq); 3310 3311static inline void sched_core_account_forceidle(struct rq *rq) 3312{ 3313 if (schedstat_enabled()) 3314 __sched_core_account_forceidle(rq); 3315} 3316 3317extern void __sched_core_tick(struct rq *rq); 3318 3319static inline void sched_core_tick(struct rq *rq) 3320{ 3321 if (sched_core_enabled(rq) && schedstat_enabled()) 3322 __sched_core_tick(rq); 3323} 3324 3325#else /* !(CONFIG_SCHED_CORE && CONFIG_SCHEDSTATS): */ 3326 3327static inline void sched_core_account_forceidle(struct rq *rq) { } 3328 3329static inline void sched_core_tick(struct rq *rq) { } 3330 3331#endif /* !(CONFIG_SCHED_CORE && CONFIG_SCHEDSTATS) */ 3332 3333#ifdef CONFIG_IRQ_TIME_ACCOUNTING 3334 3335struct irqtime { 3336 u64 total; 3337 u64 tick_delta; 3338 u64 irq_start_time; 3339 struct u64_stats_sync sync; 3340}; 3341 3342DECLARE_PER_CPU(struct irqtime, cpu_irqtime); 3343extern int sched_clock_irqtime; 3344 3345static inline int irqtime_enabled(void) 3346{ 3347 return sched_clock_irqtime; 3348} 3349 3350/* 3351 * Returns the irqtime minus the softirq time computed by ksoftirqd. 3352 * Otherwise ksoftirqd's sum_exec_runtime is subtracted its own runtime 3353 * and never move forward. 3354 */ 3355static inline u64 irq_time_read(int cpu) 3356{ 3357 struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu); 3358 unsigned int seq; 3359 u64 total; 3360 3361 do { 3362 seq = __u64_stats_fetch_begin(&irqtime->sync); 3363 total = irqtime->total; 3364 } while (__u64_stats_fetch_retry(&irqtime->sync, seq)); 3365 3366 return total; 3367} 3368 3369#else /* !CONFIG_IRQ_TIME_ACCOUNTING: */ 3370 3371static inline int irqtime_enabled(void) 3372{ 3373 return 0; 3374} 3375 3376#endif /* !CONFIG_IRQ_TIME_ACCOUNTING */ 3377 3378#ifdef CONFIG_CPU_FREQ 3379 3380DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data); 3381 3382/** 3383 * cpufreq_update_util - Take a note about CPU utilization changes. 3384 * @rq: Runqueue to carry out the update for. 3385 * @flags: Update reason flags. 3386 * 3387 * This function is called by the scheduler on the CPU whose utilization is 3388 * being updated. 3389 * 3390 * It can only be called from RCU-sched read-side critical sections. 3391 * 3392 * The way cpufreq is currently arranged requires it to evaluate the CPU 3393 * performance state (frequency/voltage) on a regular basis to prevent it from 3394 * being stuck in a completely inadequate performance level for too long. 3395 * That is not guaranteed to happen if the updates are only triggered from CFS 3396 * and DL, though, because they may not be coming in if only RT tasks are 3397 * active all the time (or there are RT tasks only). 3398 * 3399 * As a workaround for that issue, this function is called periodically by the 3400 * RT sched class to trigger extra cpufreq updates to prevent it from stalling, 3401 * but that really is a band-aid. Going forward it should be replaced with 3402 * solutions targeted more specifically at RT tasks. 3403 */ 3404static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) 3405{ 3406 struct update_util_data *data; 3407 3408 data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data, 3409 cpu_of(rq))); 3410 if (data) 3411 data->func(data, rq_clock(rq), flags); 3412} 3413#else /* !CONFIG_CPU_FREQ: */ 3414static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) { } 3415#endif /* !CONFIG_CPU_FREQ */ 3416 3417#ifdef arch_scale_freq_capacity 3418# ifndef arch_scale_freq_invariant 3419# define arch_scale_freq_invariant() true 3420# endif 3421#else 3422# define arch_scale_freq_invariant() false 3423#endif 3424 3425unsigned long effective_cpu_util(int cpu, unsigned long util_cfs, 3426 unsigned long *min, 3427 unsigned long *max); 3428 3429unsigned long sugov_effective_cpu_perf(int cpu, unsigned long actual, 3430 unsigned long min, 3431 unsigned long max); 3432 3433 3434/* 3435 * Verify the fitness of task @p to run on @cpu taking into account the 3436 * CPU original capacity and the runtime/deadline ratio of the task. 3437 * 3438 * The function will return true if the original capacity of @cpu is 3439 * greater than or equal to task's deadline density right shifted by 3440 * (BW_SHIFT - SCHED_CAPACITY_SHIFT) and false otherwise. 3441 */ 3442static inline bool dl_task_fits_capacity(struct task_struct *p, int cpu) 3443{ 3444 unsigned long cap = arch_scale_cpu_capacity(cpu); 3445 3446 return cap >= p->dl.dl_density >> (BW_SHIFT - SCHED_CAPACITY_SHIFT); 3447} 3448 3449static inline unsigned long cpu_bw_dl(struct rq *rq) 3450{ 3451 return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT; 3452} 3453 3454static inline unsigned long cpu_util_dl(struct rq *rq) 3455{ 3456 return READ_ONCE(rq->avg_dl.util_avg); 3457} 3458 3459 3460extern unsigned long cpu_util_cfs(int cpu); 3461extern unsigned long cpu_util_cfs_boost(int cpu); 3462 3463static inline unsigned long cpu_util_rt(struct rq *rq) 3464{ 3465 return READ_ONCE(rq->avg_rt.util_avg); 3466} 3467 3468#ifdef CONFIG_UCLAMP_TASK 3469 3470unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id); 3471 3472/* 3473 * When uclamp is compiled in, the aggregation at rq level is 'turned off' 3474 * by default in the fast path and only gets turned on once userspace performs 3475 * an operation that requires it. 3476 * 3477 * Returns true if userspace opted-in to use uclamp and aggregation at rq level 3478 * hence is active. 3479 */ 3480static inline bool uclamp_is_used(void) 3481{ 3482 return static_branch_likely(&sched_uclamp_used); 3483} 3484 3485/* 3486 * Enabling static branches would get the cpus_read_lock(), 3487 * check whether uclamp_is_used before enable it to avoid always 3488 * calling cpus_read_lock(). Because we never disable this 3489 * static key once enable it. 3490 */ 3491static inline void sched_uclamp_enable(void) 3492{ 3493 if (!uclamp_is_used()) 3494 static_branch_enable(&sched_uclamp_used); 3495} 3496 3497static inline unsigned long uclamp_rq_get(struct rq *rq, 3498 enum uclamp_id clamp_id) 3499{ 3500 return READ_ONCE(rq->uclamp[clamp_id].value); 3501} 3502 3503static inline void uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id, 3504 unsigned int value) 3505{ 3506 WRITE_ONCE(rq->uclamp[clamp_id].value, value); 3507} 3508 3509static inline bool uclamp_rq_is_idle(struct rq *rq) 3510{ 3511 return rq->uclamp_flags & UCLAMP_FLAG_IDLE; 3512} 3513 3514/* Is the rq being capped/throttled by uclamp_max? */ 3515static inline bool uclamp_rq_is_capped(struct rq *rq) 3516{ 3517 unsigned long rq_util; 3518 unsigned long max_util; 3519 3520 if (!uclamp_is_used()) 3521 return false; 3522 3523 rq_util = cpu_util_cfs(cpu_of(rq)) + cpu_util_rt(rq); 3524 max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value); 3525 3526 return max_util != SCHED_CAPACITY_SCALE && rq_util >= max_util; 3527} 3528 3529#define for_each_clamp_id(clamp_id) \ 3530 for ((clamp_id) = 0; (clamp_id) < UCLAMP_CNT; (clamp_id)++) 3531 3532extern unsigned int sysctl_sched_uclamp_util_min_rt_default; 3533 3534 3535static inline unsigned int uclamp_none(enum uclamp_id clamp_id) 3536{ 3537 if (clamp_id == UCLAMP_MIN) 3538 return 0; 3539 return SCHED_CAPACITY_SCALE; 3540} 3541 3542/* Integer rounded range for each bucket */ 3543#define UCLAMP_BUCKET_DELTA DIV_ROUND_CLOSEST(SCHED_CAPACITY_SCALE, UCLAMP_BUCKETS) 3544 3545static inline unsigned int uclamp_bucket_id(unsigned int clamp_value) 3546{ 3547 return min_t(unsigned int, clamp_value / UCLAMP_BUCKET_DELTA, UCLAMP_BUCKETS - 1); 3548} 3549 3550static inline void 3551uclamp_se_set(struct uclamp_se *uc_se, unsigned int value, bool user_defined) 3552{ 3553 uc_se->value = value; 3554 uc_se->bucket_id = uclamp_bucket_id(value); 3555 uc_se->user_defined = user_defined; 3556} 3557 3558#else /* !CONFIG_UCLAMP_TASK: */ 3559 3560static inline unsigned long 3561uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id) 3562{ 3563 if (clamp_id == UCLAMP_MIN) 3564 return 0; 3565 3566 return SCHED_CAPACITY_SCALE; 3567} 3568 3569static inline bool uclamp_rq_is_capped(struct rq *rq) { return false; } 3570 3571static inline bool uclamp_is_used(void) 3572{ 3573 return false; 3574} 3575 3576static inline void sched_uclamp_enable(void) {} 3577 3578static inline unsigned long 3579uclamp_rq_get(struct rq *rq, enum uclamp_id clamp_id) 3580{ 3581 if (clamp_id == UCLAMP_MIN) 3582 return 0; 3583 3584 return SCHED_CAPACITY_SCALE; 3585} 3586 3587static inline void 3588uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id, unsigned int value) 3589{ 3590} 3591 3592static inline bool uclamp_rq_is_idle(struct rq *rq) 3593{ 3594 return false; 3595} 3596 3597#endif /* !CONFIG_UCLAMP_TASK */ 3598 3599#ifdef CONFIG_HAVE_SCHED_AVG_IRQ 3600 3601static inline unsigned long cpu_util_irq(struct rq *rq) 3602{ 3603 return READ_ONCE(rq->avg_irq.util_avg); 3604} 3605 3606static inline 3607unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max) 3608{ 3609 util *= (max - irq); 3610 util /= max; 3611 3612 return util; 3613 3614} 3615 3616#else /* !CONFIG_HAVE_SCHED_AVG_IRQ: */ 3617 3618static inline unsigned long cpu_util_irq(struct rq *rq) 3619{ 3620 return 0; 3621} 3622 3623static inline 3624unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max) 3625{ 3626 return util; 3627} 3628 3629#endif /* !CONFIG_HAVE_SCHED_AVG_IRQ */ 3630 3631extern void __setparam_fair(struct task_struct *p, const struct sched_attr *attr); 3632 3633#if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL) 3634 3635#define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus))) 3636 3637DECLARE_STATIC_KEY_FALSE(sched_energy_present); 3638 3639static inline bool sched_energy_enabled(void) 3640{ 3641 return static_branch_unlikely(&sched_energy_present); 3642} 3643 3644#else /* !(CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL): */ 3645 3646#define perf_domain_span(pd) NULL 3647 3648static inline bool sched_energy_enabled(void) { return false; } 3649 3650#endif /* !(CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */ 3651 3652#ifdef CONFIG_MEMBARRIER 3653 3654/* 3655 * The scheduler provides memory barriers required by membarrier between: 3656 * - prior user-space memory accesses and store to rq->membarrier_state, 3657 * - store to rq->membarrier_state and following user-space memory accesses. 3658 * In the same way it provides those guarantees around store to rq->curr. 3659 */ 3660static inline void membarrier_switch_mm(struct rq *rq, 3661 struct mm_struct *prev_mm, 3662 struct mm_struct *next_mm) 3663{ 3664 int membarrier_state; 3665 3666 if (prev_mm == next_mm) 3667 return; 3668 3669 membarrier_state = atomic_read(&next_mm->membarrier_state); 3670 if (READ_ONCE(rq->membarrier_state) == membarrier_state) 3671 return; 3672 3673 WRITE_ONCE(rq->membarrier_state, membarrier_state); 3674} 3675 3676#else /* !CONFIG_MEMBARRIER: */ 3677 3678static inline void membarrier_switch_mm(struct rq *rq, 3679 struct mm_struct *prev_mm, 3680 struct mm_struct *next_mm) 3681{ 3682} 3683 3684#endif /* !CONFIG_MEMBARRIER */ 3685 3686static inline bool is_per_cpu_kthread(struct task_struct *p) 3687{ 3688 if (!(p->flags & PF_KTHREAD)) 3689 return false; 3690 3691 if (p->nr_cpus_allowed != 1) 3692 return false; 3693 3694 return true; 3695} 3696 3697extern void swake_up_all_locked(struct swait_queue_head *q); 3698extern void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait); 3699 3700extern int try_to_wake_up(struct task_struct *tsk, unsigned int state, int wake_flags); 3701 3702#ifdef CONFIG_PREEMPT_DYNAMIC 3703extern int preempt_dynamic_mode; 3704extern int sched_dynamic_mode(const char *str); 3705extern void sched_dynamic_update(int mode); 3706#endif 3707extern const char *preempt_modes[]; 3708 3709#ifdef CONFIG_SCHED_MM_CID 3710 3711static __always_inline bool cid_on_cpu(unsigned int cid) 3712{ 3713 return cid & MM_CID_ONCPU; 3714} 3715 3716static __always_inline bool cid_in_transit(unsigned int cid) 3717{ 3718 return cid & MM_CID_TRANSIT; 3719} 3720 3721static __always_inline unsigned int cpu_cid_to_cid(unsigned int cid) 3722{ 3723 return cid & ~MM_CID_ONCPU; 3724} 3725 3726static __always_inline unsigned int cid_to_cpu_cid(unsigned int cid) 3727{ 3728 return cid | MM_CID_ONCPU; 3729} 3730 3731static __always_inline unsigned int cid_to_transit_cid(unsigned int cid) 3732{ 3733 return cid | MM_CID_TRANSIT; 3734} 3735 3736static __always_inline unsigned int cid_from_transit_cid(unsigned int cid) 3737{ 3738 return cid & ~MM_CID_TRANSIT; 3739} 3740 3741static __always_inline bool cid_on_task(unsigned int cid) 3742{ 3743 /* True if none of the MM_CID_ONCPU, MM_CID_TRANSIT, MM_CID_UNSET bits is set */ 3744 return cid < MM_CID_TRANSIT; 3745} 3746 3747static __always_inline void mm_drop_cid(struct mm_struct *mm, unsigned int cid) 3748{ 3749 clear_bit(cid, mm_cidmask(mm)); 3750} 3751 3752static __always_inline void mm_unset_cid_on_task(struct task_struct *t) 3753{ 3754 unsigned int cid = t->mm_cid.cid; 3755 3756 t->mm_cid.cid = MM_CID_UNSET; 3757 if (cid_on_task(cid)) 3758 mm_drop_cid(t->mm, cid); 3759} 3760 3761static __always_inline void mm_drop_cid_on_cpu(struct mm_struct *mm, struct mm_cid_pcpu *pcp) 3762{ 3763 /* Clear the ONCPU bit, but do not set UNSET in the per CPU storage */ 3764 if (cid_on_cpu(pcp->cid)) { 3765 pcp->cid = cpu_cid_to_cid(pcp->cid); 3766 mm_drop_cid(mm, pcp->cid); 3767 } 3768} 3769 3770static inline unsigned int __mm_get_cid(struct mm_struct *mm, unsigned int max_cids) 3771{ 3772 unsigned int cid = find_first_zero_bit(mm_cidmask(mm), max_cids); 3773 3774 if (cid >= max_cids) 3775 return MM_CID_UNSET; 3776 if (test_and_set_bit(cid, mm_cidmask(mm))) 3777 return MM_CID_UNSET; 3778 return cid; 3779} 3780 3781static inline unsigned int mm_get_cid(struct mm_struct *mm) 3782{ 3783 unsigned int cid = __mm_get_cid(mm, READ_ONCE(mm->mm_cid.max_cids)); 3784 3785 while (cid == MM_CID_UNSET) { 3786 cpu_relax(); 3787 cid = __mm_get_cid(mm, num_possible_cpus()); 3788 } 3789 return cid; 3790} 3791 3792static inline unsigned int mm_cid_converge(struct mm_struct *mm, unsigned int orig_cid, 3793 unsigned int max_cids) 3794{ 3795 unsigned int new_cid, cid = cpu_cid_to_cid(orig_cid); 3796 3797 /* Is it in the optimal CID space? */ 3798 if (likely(cid < max_cids)) 3799 return orig_cid; 3800 3801 /* Try to find one in the optimal space. Otherwise keep the provided. */ 3802 new_cid = __mm_get_cid(mm, max_cids); 3803 if (new_cid != MM_CID_UNSET) { 3804 mm_drop_cid(mm, cid); 3805 /* Preserve the ONCPU mode of the original CID */ 3806 return new_cid | (orig_cid & MM_CID_ONCPU); 3807 } 3808 return orig_cid; 3809} 3810 3811static __always_inline void mm_cid_update_task_cid(struct task_struct *t, unsigned int cid) 3812{ 3813 if (t->mm_cid.cid != cid) { 3814 t->mm_cid.cid = cid; 3815 rseq_sched_set_ids_changed(t); 3816 } 3817} 3818 3819static __always_inline void mm_cid_update_pcpu_cid(struct mm_struct *mm, unsigned int cid) 3820{ 3821 __this_cpu_write(mm->mm_cid.pcpu->cid, cid); 3822} 3823 3824static __always_inline void mm_cid_from_cpu(struct task_struct *t, unsigned int cpu_cid, 3825 unsigned int mode) 3826{ 3827 unsigned int max_cids, tcid = t->mm_cid.cid; 3828 struct mm_struct *mm = t->mm; 3829 3830 max_cids = READ_ONCE(mm->mm_cid.max_cids); 3831 /* Optimize for the common case where both have the ONCPU bit set */ 3832 if (likely(cid_on_cpu(cpu_cid & tcid))) { 3833 if (likely(cpu_cid_to_cid(cpu_cid) < max_cids)) { 3834 mm_cid_update_task_cid(t, cpu_cid); 3835 return; 3836 } 3837 /* Try to converge into the optimal CID space */ 3838 cpu_cid = mm_cid_converge(mm, cpu_cid, max_cids); 3839 } else { 3840 /* Hand over or drop the task owned CID */ 3841 if (cid_on_task(tcid)) { 3842 if (cid_on_cpu(cpu_cid)) 3843 mm_unset_cid_on_task(t); 3844 else 3845 cpu_cid = cid_to_cpu_cid(tcid); 3846 } 3847 /* Still nothing, allocate a new one */ 3848 if (!cid_on_cpu(cpu_cid)) 3849 cpu_cid = cid_to_cpu_cid(mm_get_cid(mm)); 3850 3851 /* Handle the transition mode flag if required */ 3852 if (mode & MM_CID_TRANSIT) 3853 cpu_cid = cpu_cid_to_cid(cpu_cid) | MM_CID_TRANSIT; 3854 } 3855 mm_cid_update_pcpu_cid(mm, cpu_cid); 3856 mm_cid_update_task_cid(t, cpu_cid); 3857} 3858 3859static __always_inline void mm_cid_from_task(struct task_struct *t, unsigned int cpu_cid, 3860 unsigned int mode) 3861{ 3862 unsigned int max_cids, tcid = t->mm_cid.cid; 3863 struct mm_struct *mm = t->mm; 3864 3865 max_cids = READ_ONCE(mm->mm_cid.max_cids); 3866 /* Optimize for the common case, where both have the ONCPU bit clear */ 3867 if (likely(cid_on_task(tcid | cpu_cid))) { 3868 if (likely(tcid < max_cids)) { 3869 mm_cid_update_pcpu_cid(mm, tcid); 3870 return; 3871 } 3872 /* Try to converge into the optimal CID space */ 3873 tcid = mm_cid_converge(mm, tcid, max_cids); 3874 } else { 3875 /* Hand over or drop the CPU owned CID */ 3876 if (cid_on_cpu(cpu_cid)) { 3877 if (cid_on_task(tcid)) 3878 mm_drop_cid_on_cpu(mm, this_cpu_ptr(mm->mm_cid.pcpu)); 3879 else 3880 tcid = cpu_cid_to_cid(cpu_cid); 3881 } 3882 /* Still nothing, allocate a new one */ 3883 if (!cid_on_task(tcid)) 3884 tcid = mm_get_cid(mm); 3885 /* Set the transition mode flag if required */ 3886 tcid |= mode & MM_CID_TRANSIT; 3887 } 3888 mm_cid_update_pcpu_cid(mm, tcid); 3889 mm_cid_update_task_cid(t, tcid); 3890} 3891 3892static __always_inline void mm_cid_schedin(struct task_struct *next) 3893{ 3894 struct mm_struct *mm = next->mm; 3895 unsigned int cpu_cid, mode; 3896 3897 if (!next->mm_cid.active) 3898 return; 3899 3900 cpu_cid = __this_cpu_read(mm->mm_cid.pcpu->cid); 3901 mode = READ_ONCE(mm->mm_cid.mode); 3902 if (likely(!cid_on_cpu(mode))) 3903 mm_cid_from_task(next, cpu_cid, mode); 3904 else 3905 mm_cid_from_cpu(next, cpu_cid, mode); 3906} 3907 3908static __always_inline void mm_cid_schedout(struct task_struct *prev) 3909{ 3910 struct mm_struct *mm = prev->mm; 3911 unsigned int mode, cid; 3912 3913 /* During mode transitions CIDs are temporary and need to be dropped */ 3914 if (likely(!cid_in_transit(prev->mm_cid.cid))) 3915 return; 3916 3917 mode = READ_ONCE(mm->mm_cid.mode); 3918 cid = cid_from_transit_cid(prev->mm_cid.cid); 3919 3920 /* 3921 * If transition mode is done, transfer ownership when the CID is 3922 * within the convergence range to optimize the next schedule in. 3923 */ 3924 if (!cid_in_transit(mode) && cid < READ_ONCE(mm->mm_cid.max_cids)) { 3925 if (cid_on_cpu(mode)) 3926 cid = cid_to_cpu_cid(cid); 3927 3928 /* Update both so that the next schedule in goes into the fast path */ 3929 mm_cid_update_pcpu_cid(mm, cid); 3930 prev->mm_cid.cid = cid; 3931 } else { 3932 mm_drop_cid(mm, cid); 3933 prev->mm_cid.cid = MM_CID_UNSET; 3934 } 3935} 3936 3937static inline void mm_cid_switch_to(struct task_struct *prev, struct task_struct *next) 3938{ 3939 mm_cid_schedout(prev); 3940 mm_cid_schedin(next); 3941} 3942 3943#else /* !CONFIG_SCHED_MM_CID: */ 3944static inline void mm_cid_switch_to(struct task_struct *prev, struct task_struct *next) { } 3945#endif /* !CONFIG_SCHED_MM_CID */ 3946 3947extern u64 avg_vruntime(struct cfs_rq *cfs_rq); 3948extern int entity_eligible(struct cfs_rq *cfs_rq, struct sched_entity *se); 3949static inline 3950void move_queued_task_locked(struct rq *src_rq, struct rq *dst_rq, struct task_struct *task) 3951{ 3952 lockdep_assert_rq_held(src_rq); 3953 lockdep_assert_rq_held(dst_rq); 3954 3955 deactivate_task(src_rq, task, 0); 3956 set_task_cpu(task, dst_rq->cpu); 3957 activate_task(dst_rq, task, 0); 3958} 3959 3960static inline 3961bool task_is_pushable(struct rq *rq, struct task_struct *p, int cpu) 3962{ 3963 if (!task_on_cpu(rq, p) && 3964 cpumask_test_cpu(cpu, &p->cpus_mask)) 3965 return true; 3966 3967 return false; 3968} 3969 3970#ifdef CONFIG_RT_MUTEXES 3971 3972static inline int __rt_effective_prio(struct task_struct *pi_task, int prio) 3973{ 3974 if (pi_task) 3975 prio = min(prio, pi_task->prio); 3976 3977 return prio; 3978} 3979 3980static inline int rt_effective_prio(struct task_struct *p, int prio) 3981{ 3982 struct task_struct *pi_task = rt_mutex_get_top_task(p); 3983 3984 return __rt_effective_prio(pi_task, prio); 3985} 3986 3987#else /* !CONFIG_RT_MUTEXES: */ 3988 3989static inline int rt_effective_prio(struct task_struct *p, int prio) 3990{ 3991 return prio; 3992} 3993 3994#endif /* !CONFIG_RT_MUTEXES */ 3995 3996extern int __sched_setscheduler(struct task_struct *p, const struct sched_attr *attr, bool user, bool pi); 3997extern int __sched_setaffinity(struct task_struct *p, struct affinity_context *ctx); 3998extern const struct sched_class *__setscheduler_class(int policy, int prio); 3999extern void set_load_weight(struct task_struct *p, bool update_load); 4000extern void enqueue_task(struct rq *rq, struct task_struct *p, int flags); 4001extern bool dequeue_task(struct rq *rq, struct task_struct *p, int flags); 4002 4003extern struct balance_callback *splice_balance_callbacks(struct rq *rq); 4004 4005extern void __balance_callbacks(struct rq *rq, struct rq_flags *rf); 4006extern void balance_callbacks(struct rq *rq, struct balance_callback *head); 4007 4008/* 4009 * The 'sched_change' pattern is the safe, easy and slow way of changing a 4010 * task's scheduling properties. It dequeues a task, such that the scheduler 4011 * is fully unaware of it; at which point its properties can be modified; 4012 * after which it is enqueued again. 4013 * 4014 * Typically this must be called while holding task_rq_lock, since most/all 4015 * properties are serialized under those locks. There is currently one 4016 * exception to this rule in sched/ext which only holds rq->lock. 4017 */ 4018 4019/* 4020 * This structure is a temporary, used to preserve/convey the queueing state 4021 * of the task between sched_change_begin() and sched_change_end(). Ensuring 4022 * the task's queueing state is idempotent across the operation. 4023 */ 4024struct sched_change_ctx { 4025 u64 prio; 4026 struct task_struct *p; 4027 int flags; 4028 bool queued; 4029 bool running; 4030}; 4031 4032struct sched_change_ctx *sched_change_begin(struct task_struct *p, unsigned int flags); 4033void sched_change_end(struct sched_change_ctx *ctx); 4034 4035DEFINE_CLASS(sched_change, struct sched_change_ctx *, 4036 sched_change_end(_T), 4037 sched_change_begin(p, flags), 4038 struct task_struct *p, unsigned int flags) 4039 4040DEFINE_CLASS_IS_UNCONDITIONAL(sched_change) 4041 4042#include "ext.h" 4043 4044#endif /* _KERNEL_SCHED_SCHED_H */
~/
Read-only file preview.