root/kernel/sched/sched.h

INCLUDED FROM

DEFINITIONS

1. idle_policy
2. fair_policy
3. rt_policy
4. dl_policy
5. valid_policy
6. task_has_idle_policy
7. task_has_rt_policy
8. task_has_dl_policy
9. dl_entity_is_special
10. dl_entity_preempt
11. dl_bandwidth_enabled
12. __dl_sub
13. __dl_add
14. __dl_overflow
15. walk_tg_tree
16. set_task_rq_fair
17. rt_bandwidth_enabled
18. rt_rq_is_runnable
19. se_weight
20. se_runnable
21. sched_asym_prefer
22. rq_of
23. rq_of
24. cpu_of
25. update_idle_core
26. update_idle_core
27. __rq_clock_broken
28. assert_clock_updated
29. rq_clock
30. rq_clock_task
31. rq_clock_skip_update
32. rq_clock_cancel_skipupdate
33. rq_pin_lock
34. rq_unpin_lock
35. rq_repin_lock
36. __task_rq_lock
37. task_rq_unlock
38. rq_lock_irqsave
39. rq_lock_irq
40. rq_lock
41. rq_relock
42. rq_unlock_irqrestore
43. rq_unlock_irq
44. rq_unlock
45. this_rq_lock_irq
46. sched_init_numa
47. sched_domains_numa_masks_set
48. sched_domains_numa_masks_clear
49. sched_numa_find_closest
50. init_numa_balancing
51. queue_balance_callback
52. highest_flag_domain
53. lowest_flag_domain
54. sched_group_span
55. group_balance_mask
56. group_first_cpu
57. register_sched_domain_sysctl
58. dirty_sched_domain_sysctl
59. unregister_sched_domain_sysctl
60. sched_ttwu_pending
61. newidle_balance
62. set_task_rq
63. set_task_rq
64. __set_task_cpu
65. global_rt_period
66. global_rt_runtime
67. task_current
68. task_running
69. task_on_rq_queued
70. task_on_rq_migrating
71. put_prev_task
72. set_next_task
73. sched_stop_runnable
74. sched_dl_runnable
75. sched_rt_runnable
76. sched_fair_runnable
77. idle_set_state
78. idle_get_state
79. idle_set_state
80. idle_get_state
81. sched_update_tick_dependency
82. sched_tick_offload_init
83. sched_update_tick_dependency
84. add_nr_running
85. sub_nr_running
86. hrtick_enabled
87. hrtick_enabled
88. arch_scale_freq_capacity
89. _double_lock_balance
90. _double_lock_balance
91. double_lock_balance
92. double_unlock_balance
93. double_lock
94. double_lock_irq
95. double_raw_lock
96. double_rq_lock
97. double_rq_unlock
98. double_rq_lock
99. double_rq_unlock
100. nohz_balance_exit_idle
101. __dl_update
102. __dl_update
103. irq_time_read
104. cpufreq_update_util
105. cpufreq_update_util
106. uclamp_util_with
107. uclamp_util
108. uclamp_util_with
109. uclamp_util
110. capacity_orig_of
111. cpu_bw_dl
112. cpu_util_dl
113. cpu_util_cfs
114. cpu_util_rt
115. schedutil_cpu_util
116. cpu_util_irq
117. scale_irq_capacity
118. cpu_util_irq
119. scale_irq_capacity
120. sched_energy_enabled
121. sched_energy_enabled
122. membarrier_switch_mm
123. membarrier_switch_mm

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