Linux Audio

Check our new training course

Loading...
v6.2
   1/* SPDX-License-Identifier: GPL-2.0 */
   2/*
   3 * Scheduler internal types and methods:
   4 */
   5#ifndef _KERNEL_SCHED_SCHED_H
   6#define _KERNEL_SCHED_SCHED_H
   7
   8#include <linux/sched/affinity.h>
   9#include <linux/sched/autogroup.h>
  10#include <linux/sched/cpufreq.h>
  11#include <linux/sched/deadline.h>
  12#include <linux/sched.h>
  13#include <linux/sched/loadavg.h>
  14#include <linux/sched/mm.h>
  15#include <linux/sched/rseq_api.h>
  16#include <linux/sched/signal.h>
  17#include <linux/sched/smt.h>
  18#include <linux/sched/stat.h>
  19#include <linux/sched/sysctl.h>
  20#include <linux/sched/task_flags.h>
  21#include <linux/sched/task.h>
  22#include <linux/sched/topology.h>
  23
  24#include <linux/atomic.h>
  25#include <linux/bitmap.h>
  26#include <linux/bug.h>
  27#include <linux/capability.h>
  28#include <linux/cgroup_api.h>
  29#include <linux/cgroup.h>
  30#include <linux/context_tracking.h>
  31#include <linux/cpufreq.h>
  32#include <linux/cpumask_api.h>
  33#include <linux/ctype.h>
  34#include <linux/file.h>
  35#include <linux/fs_api.h>
  36#include <linux/hrtimer_api.h>
  37#include <linux/interrupt.h>
  38#include <linux/irq_work.h>
  39#include <linux/jiffies.h>
  40#include <linux/kref_api.h>
  41#include <linux/kthread.h>
  42#include <linux/ktime_api.h>
  43#include <linux/lockdep_api.h>
  44#include <linux/lockdep.h>
  45#include <linux/minmax.h>
  46#include <linux/mm.h>
  47#include <linux/module.h>
  48#include <linux/mutex_api.h>
  49#include <linux/plist.h>
  50#include <linux/poll.h>
  51#include <linux/proc_fs.h>
  52#include <linux/profile.h>
  53#include <linux/psi.h>
  54#include <linux/rcupdate.h>
  55#include <linux/seq_file.h>
  56#include <linux/seqlock.h>
  57#include <linux/softirq.h>
  58#include <linux/spinlock_api.h>
  59#include <linux/static_key.h>
  60#include <linux/stop_machine.h>
  61#include <linux/syscalls_api.h>
  62#include <linux/syscalls.h>
  63#include <linux/tick.h>
  64#include <linux/topology.h>
  65#include <linux/types.h>
  66#include <linux/u64_stats_sync_api.h>
  67#include <linux/uaccess.h>
  68#include <linux/wait_api.h>
  69#include <linux/wait_bit.h>
  70#include <linux/workqueue_api.h>
  71
  72#include <trace/events/power.h>
  73#include <trace/events/sched.h>
  74
  75#include "../workqueue_internal.h"
  76
  77#ifdef CONFIG_CGROUP_SCHED
  78#include <linux/cgroup.h>
  79#include <linux/psi.h>
  80#endif
  81
  82#ifdef CONFIG_SCHED_DEBUG
  83# include <linux/static_key.h>
  84#endif
  85
  86#ifdef CONFIG_PARAVIRT
  87# include <asm/paravirt.h>
  88# include <asm/paravirt_api_clock.h>
  89#endif
  90
  91#include "cpupri.h"
  92#include "cpudeadline.h"
  93
  94#ifdef CONFIG_SCHED_DEBUG
  95# define SCHED_WARN_ON(x)      WARN_ONCE(x, #x)
  96#else
  97# define SCHED_WARN_ON(x)      ({ (void)(x), 0; })
  98#endif
  99
 100struct rq;
 101struct cpuidle_state;
 102
 103/* task_struct::on_rq states: */
 104#define TASK_ON_RQ_QUEUED	1
 105#define TASK_ON_RQ_MIGRATING	2
 106
 107extern __read_mostly int scheduler_running;
 108
 109extern unsigned long calc_load_update;
 110extern atomic_long_t calc_load_tasks;
 111
 112extern unsigned int sysctl_sched_child_runs_first;
 113
 114extern void calc_global_load_tick(struct rq *this_rq);
 115extern long calc_load_fold_active(struct rq *this_rq, long adjust);
 116
 117extern void call_trace_sched_update_nr_running(struct rq *rq, int count);
 118
 119extern unsigned int sysctl_sched_rt_period;
 120extern int sysctl_sched_rt_runtime;
 121extern int sched_rr_timeslice;
 122
 123/*
 124 * Helpers for converting nanosecond timing to jiffy resolution
 125 */
 126#define NS_TO_JIFFIES(TIME)	((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
 127
 128/*
 129 * Increase resolution of nice-level calculations for 64-bit architectures.
 130 * The extra resolution improves shares distribution and load balancing of
 131 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
 132 * hierarchies, especially on larger systems. This is not a user-visible change
 133 * and does not change the user-interface for setting shares/weights.
 134 *
 135 * We increase resolution only if we have enough bits to allow this increased
 136 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
 137 * are pretty high and the returns do not justify the increased costs.
 138 *
 139 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
 140 * increase coverage and consistency always enable it on 64-bit platforms.
 141 */
 142#ifdef CONFIG_64BIT
 143# define NICE_0_LOAD_SHIFT	(SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
 144# define scale_load(w)		((w) << SCHED_FIXEDPOINT_SHIFT)
 145# define scale_load_down(w) \
 146({ \
 147	unsigned long __w = (w); \
 148	if (__w) \
 149		__w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \
 150	__w; \
 151})
 152#else
 153# define NICE_0_LOAD_SHIFT	(SCHED_FIXEDPOINT_SHIFT)
 154# define scale_load(w)		(w)
 155# define scale_load_down(w)	(w)
 156#endif
 157
 158/*
 159 * Task weight (visible to users) and its load (invisible to users) have
 160 * independent resolution, but they should be well calibrated. We use
 161 * scale_load() and scale_load_down(w) to convert between them. The
 162 * following must be true:
 163 *
 164 *  scale_load(sched_prio_to_weight[NICE_TO_PRIO(0)-MAX_RT_PRIO]) == NICE_0_LOAD
 165 *
 166 */
 167#define NICE_0_LOAD		(1L << NICE_0_LOAD_SHIFT)
 168
 169/*
 170 * Single value that decides SCHED_DEADLINE internal math precision.
 171 * 10 -> just above 1us
 172 * 9  -> just above 0.5us
 173 */
 174#define DL_SCALE		10
 175
 176/*
 177 * Single value that denotes runtime == period, ie unlimited time.
 178 */
 179#define RUNTIME_INF		((u64)~0ULL)
 180
 181static inline int idle_policy(int policy)
 182{
 183	return policy == SCHED_IDLE;
 184}
 185static inline int fair_policy(int policy)
 186{
 187	return policy == SCHED_NORMAL || policy == SCHED_BATCH;
 188}
 189
 190static inline int rt_policy(int policy)
 191{
 192	return policy == SCHED_FIFO || policy == SCHED_RR;
 193}
 194
 195static inline int dl_policy(int policy)
 196{
 197	return policy == SCHED_DEADLINE;
 198}
 199static inline bool valid_policy(int policy)
 200{
 201	return idle_policy(policy) || fair_policy(policy) ||
 202		rt_policy(policy) || dl_policy(policy);
 203}
 204
 205static inline int task_has_idle_policy(struct task_struct *p)
 206{
 207	return idle_policy(p->policy);
 208}
 209
 210static inline int task_has_rt_policy(struct task_struct *p)
 211{
 212	return rt_policy(p->policy);
 213}
 214
 215static inline int task_has_dl_policy(struct task_struct *p)
 216{
 217	return dl_policy(p->policy);
 218}
 219
 220#define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
 221
 222static inline void update_avg(u64 *avg, u64 sample)
 223{
 224	s64 diff = sample - *avg;
 225	*avg += diff / 8;
 226}
 227
 228/*
 229 * Shifting a value by an exponent greater *or equal* to the size of said value
 230 * is UB; cap at size-1.
 231 */
 232#define shr_bound(val, shift)							\
 233	(val >> min_t(typeof(shift), shift, BITS_PER_TYPE(typeof(val)) - 1))
 234
 235/*
 236 * !! For sched_setattr_nocheck() (kernel) only !!
 237 *
 238 * This is actually gross. :(
 239 *
 240 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
 241 * tasks, but still be able to sleep. We need this on platforms that cannot
 242 * atomically change clock frequency. Remove once fast switching will be
 243 * available on such platforms.
 244 *
 245 * SUGOV stands for SchedUtil GOVernor.
 246 */
 247#define SCHED_FLAG_SUGOV	0x10000000
 248
 249#define SCHED_DL_FLAGS (SCHED_FLAG_RECLAIM | SCHED_FLAG_DL_OVERRUN | SCHED_FLAG_SUGOV)
 250
 251static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se)
 252{
 253#ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
 254	return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
 255#else
 256	return false;
 257#endif
 258}
 259
 260/*
 261 * Tells if entity @a should preempt entity @b.
 262 */
 263static inline bool
 264dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
 265{
 266	return dl_entity_is_special(a) ||
 267	       dl_time_before(a->deadline, b->deadline);
 268}
 269
 270/*
 271 * This is the priority-queue data structure of the RT scheduling class:
 272 */
 273struct rt_prio_array {
 274	DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
 275	struct list_head queue[MAX_RT_PRIO];
 276};
 277
 278struct rt_bandwidth {
 279	/* nests inside the rq lock: */
 280	raw_spinlock_t		rt_runtime_lock;
 281	ktime_t			rt_period;
 282	u64			rt_runtime;
 283	struct hrtimer		rt_period_timer;
 284	unsigned int		rt_period_active;
 285};
 286
 287void __dl_clear_params(struct task_struct *p);
 288
 289struct dl_bandwidth {
 290	raw_spinlock_t		dl_runtime_lock;
 291	u64			dl_runtime;
 292	u64			dl_period;
 293};
 294
 295static inline int dl_bandwidth_enabled(void)
 296{
 297	return sysctl_sched_rt_runtime >= 0;
 298}
 299
 300/*
 301 * To keep the bandwidth of -deadline tasks under control
 302 * we need some place where:
 303 *  - store the maximum -deadline bandwidth of each cpu;
 304 *  - cache the fraction of bandwidth that is currently allocated in
 305 *    each root domain;
 306 *
 307 * This is all done in the data structure below. It is similar to the
 308 * one used for RT-throttling (rt_bandwidth), with the main difference
 309 * that, since here we are only interested in admission control, we
 310 * do not decrease any runtime while the group "executes", neither we
 311 * need a timer to replenish it.
 312 *
 313 * With respect to SMP, bandwidth is given on a per root domain basis,
 314 * meaning that:
 315 *  - bw (< 100%) is the deadline bandwidth of each CPU;
 316 *  - total_bw is the currently allocated bandwidth in each root domain;
 317 */
 318struct dl_bw {
 319	raw_spinlock_t		lock;
 320	u64			bw;
 321	u64			total_bw;
 322};
 323
 324extern void init_dl_bw(struct dl_bw *dl_b);
 325extern int  sched_dl_global_validate(void);
 326extern void sched_dl_do_global(void);
 327extern int  sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
 328extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
 329extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
 330extern bool __checkparam_dl(const struct sched_attr *attr);
 331extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
 332extern int  dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
 333extern int  dl_cpu_busy(int cpu, struct task_struct *p);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 334
 335#ifdef CONFIG_CGROUP_SCHED
 336
 337struct cfs_rq;
 338struct rt_rq;
 339
 340extern struct list_head task_groups;
 341
 342struct cfs_bandwidth {
 343#ifdef CONFIG_CFS_BANDWIDTH
 344	raw_spinlock_t		lock;
 345	ktime_t			period;
 346	u64			quota;
 347	u64			runtime;
 348	u64			burst;
 349	u64			runtime_snap;
 350	s64			hierarchical_quota;
 351
 352	u8			idle;
 353	u8			period_active;
 354	u8			slack_started;
 355	struct hrtimer		period_timer;
 356	struct hrtimer		slack_timer;
 357	struct list_head	throttled_cfs_rq;
 358
 359	/* Statistics: */
 360	int			nr_periods;
 361	int			nr_throttled;
 362	int			nr_burst;
 363	u64			throttled_time;
 364	u64			burst_time;
 365#endif
 366};
 367
 368/* Task group related information */
 369struct task_group {
 370	struct cgroup_subsys_state css;
 371
 372#ifdef CONFIG_FAIR_GROUP_SCHED
 373	/* schedulable entities of this group on each CPU */
 374	struct sched_entity	**se;
 375	/* runqueue "owned" by this group on each CPU */
 376	struct cfs_rq		**cfs_rq;
 377	unsigned long		shares;
 378
 379	/* A positive value indicates that this is a SCHED_IDLE group. */
 380	int			idle;
 381
 382#ifdef	CONFIG_SMP
 383	/*
 384	 * load_avg can be heavily contended at clock tick time, so put
 385	 * it in its own cacheline separated from the fields above which
 386	 * will also be accessed at each tick.
 387	 */
 388	atomic_long_t		load_avg ____cacheline_aligned;
 389#endif
 390#endif
 391
 392#ifdef CONFIG_RT_GROUP_SCHED
 393	struct sched_rt_entity	**rt_se;
 394	struct rt_rq		**rt_rq;
 395
 396	struct rt_bandwidth	rt_bandwidth;
 397#endif
 398
 399	struct rcu_head		rcu;
 400	struct list_head	list;
 401
 402	struct task_group	*parent;
 403	struct list_head	siblings;
 404	struct list_head	children;
 405
 406#ifdef CONFIG_SCHED_AUTOGROUP
 407	struct autogroup	*autogroup;
 408#endif
 409
 410	struct cfs_bandwidth	cfs_bandwidth;
 411
 412#ifdef CONFIG_UCLAMP_TASK_GROUP
 413	/* The two decimal precision [%] value requested from user-space */
 414	unsigned int		uclamp_pct[UCLAMP_CNT];
 415	/* Clamp values requested for a task group */
 416	struct uclamp_se	uclamp_req[UCLAMP_CNT];
 417	/* Effective clamp values used for a task group */
 418	struct uclamp_se	uclamp[UCLAMP_CNT];
 419#endif
 420
 421};
 422
 423#ifdef CONFIG_FAIR_GROUP_SCHED
 424#define ROOT_TASK_GROUP_LOAD	NICE_0_LOAD
 425
 426/*
 427 * A weight of 0 or 1 can cause arithmetics problems.
 428 * A weight of a cfs_rq is the sum of weights of which entities
 429 * are queued on this cfs_rq, so a weight of a entity should not be
 430 * too large, so as the shares value of a task group.
 431 * (The default weight is 1024 - so there's no practical
 432 *  limitation from this.)
 433 */
 434#define MIN_SHARES		(1UL <<  1)
 435#define MAX_SHARES		(1UL << 18)
 436#endif
 437
 438typedef int (*tg_visitor)(struct task_group *, void *);
 439
 440extern int walk_tg_tree_from(struct task_group *from,
 441			     tg_visitor down, tg_visitor up, void *data);
 442
 443/*
 444 * Iterate the full tree, calling @down when first entering a node and @up when
 445 * leaving it for the final time.
 446 *
 447 * Caller must hold rcu_lock or sufficient equivalent.
 448 */
 449static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
 450{
 451	return walk_tg_tree_from(&root_task_group, down, up, data);
 452}
 453
 454extern int tg_nop(struct task_group *tg, void *data);
 455
 
 456extern void free_fair_sched_group(struct task_group *tg);
 457extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
 458extern void online_fair_sched_group(struct task_group *tg);
 459extern void unregister_fair_sched_group(struct task_group *tg);
 
 
 
 
 
 
 
 
 
 
 460extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
 461			struct sched_entity *se, int cpu,
 462			struct sched_entity *parent);
 463extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
 464
 465extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
 466extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
 467extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
 
 468
 469extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
 470		struct sched_rt_entity *rt_se, int cpu,
 471		struct sched_rt_entity *parent);
 472extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
 473extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
 474extern long sched_group_rt_runtime(struct task_group *tg);
 475extern long sched_group_rt_period(struct task_group *tg);
 476extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
 477
 478extern struct task_group *sched_create_group(struct task_group *parent);
 479extern void sched_online_group(struct task_group *tg,
 480			       struct task_group *parent);
 481extern void sched_destroy_group(struct task_group *tg);
 482extern void sched_release_group(struct task_group *tg);
 483
 484extern void sched_move_task(struct task_struct *tsk);
 485
 486#ifdef CONFIG_FAIR_GROUP_SCHED
 487extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
 488
 489extern int sched_group_set_idle(struct task_group *tg, long idle);
 490
 491#ifdef CONFIG_SMP
 492extern void set_task_rq_fair(struct sched_entity *se,
 493			     struct cfs_rq *prev, struct cfs_rq *next);
 494#else /* !CONFIG_SMP */
 495static inline void set_task_rq_fair(struct sched_entity *se,
 496			     struct cfs_rq *prev, struct cfs_rq *next) { }
 497#endif /* CONFIG_SMP */
 498#endif /* CONFIG_FAIR_GROUP_SCHED */
 499
 500#else /* CONFIG_CGROUP_SCHED */
 501
 502struct cfs_bandwidth { };
 
 503
 504#endif	/* CONFIG_CGROUP_SCHED */
 505
 506extern void unregister_rt_sched_group(struct task_group *tg);
 507extern void free_rt_sched_group(struct task_group *tg);
 508extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
 509
 510/*
 511 * u64_u32_load/u64_u32_store
 512 *
 513 * Use a copy of a u64 value to protect against data race. This is only
 514 * applicable for 32-bits architectures.
 515 */
 516#ifdef CONFIG_64BIT
 517# define u64_u32_load_copy(var, copy)       var
 518# define u64_u32_store_copy(var, copy, val) (var = val)
 519#else
 520# define u64_u32_load_copy(var, copy)					\
 521({									\
 522	u64 __val, __val_copy;						\
 523	do {								\
 524		__val_copy = copy;					\
 525		/*							\
 526		 * paired with u64_u32_store_copy(), ordering access	\
 527		 * to var and copy.					\
 528		 */							\
 529		smp_rmb();						\
 530		__val = var;						\
 531	} while (__val != __val_copy);					\
 532	__val;								\
 533})
 534# define u64_u32_store_copy(var, copy, val)				\
 535do {									\
 536	typeof(val) __val = (val);					\
 537	var = __val;							\
 538	/*								\
 539	 * paired with u64_u32_load_copy(), ordering access to var and	\
 540	 * copy.							\
 541	 */								\
 542	smp_wmb();							\
 543	copy = __val;							\
 544} while (0)
 545#endif
 546# define u64_u32_load(var)      u64_u32_load_copy(var, var##_copy)
 547# define u64_u32_store(var, val) u64_u32_store_copy(var, var##_copy, val)
 548
 549/* CFS-related fields in a runqueue */
 550struct cfs_rq {
 551	struct load_weight	load;
 552	unsigned int		nr_running;
 553	unsigned int		h_nr_running;      /* SCHED_{NORMAL,BATCH,IDLE} */
 554	unsigned int		idle_nr_running;   /* SCHED_IDLE */
 555	unsigned int		idle_h_nr_running; /* SCHED_IDLE */
 556
 
 
 
 557	u64			exec_clock;
 558	u64			min_vruntime;
 559#ifdef CONFIG_SCHED_CORE
 560	unsigned int		forceidle_seq;
 561	u64			min_vruntime_fi;
 562#endif
 563
 564#ifndef CONFIG_64BIT
 565	u64			min_vruntime_copy;
 566#endif
 567
 568	struct rb_root_cached	tasks_timeline;
 569
 570	/*
 571	 * 'curr' points to currently running entity on this cfs_rq.
 572	 * It is set to NULL otherwise (i.e when none are currently running).
 573	 */
 574	struct sched_entity	*curr;
 575	struct sched_entity	*next;
 576	struct sched_entity	*last;
 577	struct sched_entity	*skip;
 578
 579#ifdef	CONFIG_SCHED_DEBUG
 580	unsigned int		nr_spread_over;
 581#endif
 582
 583#ifdef CONFIG_SMP
 584	/*
 585	 * CFS load tracking
 586	 */
 587	struct sched_avg	avg;
 588#ifndef CONFIG_64BIT
 589	u64			last_update_time_copy;
 590#endif
 591	struct {
 592		raw_spinlock_t	lock ____cacheline_aligned;
 593		int		nr;
 594		unsigned long	load_avg;
 595		unsigned long	util_avg;
 596		unsigned long	runnable_avg;
 597	} removed;
 598
 599#ifdef CONFIG_FAIR_GROUP_SCHED
 
 600	unsigned long		tg_load_avg_contrib;
 601	long			propagate;
 602	long			prop_runnable_sum;
 603
 604	/*
 605	 *   h_load = weight * f(tg)
 606	 *
 607	 * Where f(tg) is the recursive weight fraction assigned to
 608	 * this group.
 609	 */
 610	unsigned long		h_load;
 611	u64			last_h_load_update;
 612	struct sched_entity	*h_load_next;
 613#endif /* CONFIG_FAIR_GROUP_SCHED */
 614#endif /* CONFIG_SMP */
 615
 616#ifdef CONFIG_FAIR_GROUP_SCHED
 617	struct rq		*rq;	/* CPU runqueue to which this cfs_rq is attached */
 618
 619	/*
 620	 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
 621	 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
 622	 * (like users, containers etc.)
 623	 *
 624	 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
 625	 * This list is used during load balance.
 626	 */
 627	int			on_list;
 628	struct list_head	leaf_cfs_rq_list;
 629	struct task_group	*tg;	/* group that "owns" this runqueue */
 630
 631	/* Locally cached copy of our task_group's idle value */
 632	int			idle;
 633
 634#ifdef CONFIG_CFS_BANDWIDTH
 635	int			runtime_enabled;
 636	s64			runtime_remaining;
 637
 638	u64			throttled_pelt_idle;
 639#ifndef CONFIG_64BIT
 640	u64                     throttled_pelt_idle_copy;
 641#endif
 642	u64			throttled_clock;
 643	u64			throttled_clock_pelt;
 644	u64			throttled_clock_pelt_time;
 
 
 645	int			throttled;
 646	int			throttle_count;
 647	struct list_head	throttled_list;
 
 648#endif /* CONFIG_CFS_BANDWIDTH */
 649#endif /* CONFIG_FAIR_GROUP_SCHED */
 650};
 651
 652static inline int rt_bandwidth_enabled(void)
 653{
 654	return sysctl_sched_rt_runtime >= 0;
 655}
 656
 657/* RT IPI pull logic requires IRQ_WORK */
 658#if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
 659# define HAVE_RT_PUSH_IPI
 660#endif
 661
 662/* Real-Time classes' related field in a runqueue: */
 663struct rt_rq {
 664	struct rt_prio_array	active;
 665	unsigned int		rt_nr_running;
 666	unsigned int		rr_nr_running;
 667#if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
 668	struct {
 669		int		curr; /* highest queued rt task prio */
 670#ifdef CONFIG_SMP
 671		int		next; /* next highest */
 672#endif
 673	} highest_prio;
 674#endif
 675#ifdef CONFIG_SMP
 676	unsigned int		rt_nr_migratory;
 677	unsigned int		rt_nr_total;
 678	int			overloaded;
 679	struct plist_head	pushable_tasks;
 680
 681#endif /* CONFIG_SMP */
 682	int			rt_queued;
 683
 684	int			rt_throttled;
 685	u64			rt_time;
 686	u64			rt_runtime;
 687	/* Nests inside the rq lock: */
 688	raw_spinlock_t		rt_runtime_lock;
 689
 690#ifdef CONFIG_RT_GROUP_SCHED
 691	unsigned int		rt_nr_boosted;
 692
 693	struct rq		*rq;
 694	struct task_group	*tg;
 695#endif
 696};
 697
 698static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
 699{
 700	return rt_rq->rt_queued && rt_rq->rt_nr_running;
 701}
 702
 703/* Deadline class' related fields in a runqueue */
 704struct dl_rq {
 705	/* runqueue is an rbtree, ordered by deadline */
 706	struct rb_root_cached	root;
 707
 708	unsigned int		dl_nr_running;
 709
 710#ifdef CONFIG_SMP
 711	/*
 712	 * Deadline values of the currently executing and the
 713	 * earliest ready task on this rq. Caching these facilitates
 714	 * the decision whether or not a ready but not running task
 715	 * should migrate somewhere else.
 716	 */
 717	struct {
 718		u64		curr;
 719		u64		next;
 720	} earliest_dl;
 721
 722	unsigned int		dl_nr_migratory;
 723	int			overloaded;
 724
 725	/*
 726	 * Tasks on this rq that can be pushed away. They are kept in
 727	 * an rb-tree, ordered by tasks' deadlines, with caching
 728	 * of the leftmost (earliest deadline) element.
 729	 */
 730	struct rb_root_cached	pushable_dl_tasks_root;
 731#else
 732	struct dl_bw		dl_bw;
 733#endif
 734	/*
 735	 * "Active utilization" for this runqueue: increased when a
 736	 * task wakes up (becomes TASK_RUNNING) and decreased when a
 737	 * task blocks
 738	 */
 739	u64			running_bw;
 740
 741	/*
 742	 * Utilization of the tasks "assigned" to this runqueue (including
 743	 * the tasks that are in runqueue and the tasks that executed on this
 744	 * CPU and blocked). Increased when a task moves to this runqueue, and
 745	 * decreased when the task moves away (migrates, changes scheduling
 746	 * policy, or terminates).
 747	 * This is needed to compute the "inactive utilization" for the
 748	 * runqueue (inactive utilization = this_bw - running_bw).
 749	 */
 750	u64			this_bw;
 751	u64			extra_bw;
 752
 753	/*
 
 
 
 
 
 
 754	 * Inverse of the fraction of CPU utilization that can be reclaimed
 755	 * by the GRUB algorithm.
 756	 */
 757	u64			bw_ratio;
 758};
 759
 760#ifdef CONFIG_FAIR_GROUP_SCHED
 761/* An entity is a task if it doesn't "own" a runqueue */
 762#define entity_is_task(se)	(!se->my_q)
 763
 764static inline void se_update_runnable(struct sched_entity *se)
 765{
 766	if (!entity_is_task(se))
 767		se->runnable_weight = se->my_q->h_nr_running;
 768}
 769
 770static inline long se_runnable(struct sched_entity *se)
 771{
 772	if (entity_is_task(se))
 773		return !!se->on_rq;
 774	else
 775		return se->runnable_weight;
 776}
 777
 778#else
 779#define entity_is_task(se)	1
 780
 781static inline void se_update_runnable(struct sched_entity *se) {}
 782
 783static inline long se_runnable(struct sched_entity *se)
 784{
 785	return !!se->on_rq;
 786}
 787#endif
 788
 789#ifdef CONFIG_SMP
 790/*
 791 * XXX we want to get rid of these helpers and use the full load resolution.
 792 */
 793static inline long se_weight(struct sched_entity *se)
 794{
 795	return scale_load_down(se->load.weight);
 796}
 797
 798
 799static inline bool sched_asym_prefer(int a, int b)
 800{
 801	return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
 802}
 803
 804struct perf_domain {
 805	struct em_perf_domain *em_pd;
 806	struct perf_domain *next;
 807	struct rcu_head rcu;
 808};
 809
 810/* Scheduling group status flags */
 811#define SG_OVERLOAD		0x1 /* More than one runnable task on a CPU. */
 812#define SG_OVERUTILIZED		0x2 /* One or more CPUs are over-utilized. */
 813
 814/*
 815 * We add the notion of a root-domain which will be used to define per-domain
 816 * variables. Each exclusive cpuset essentially defines an island domain by
 817 * fully partitioning the member CPUs from any other cpuset. Whenever a new
 818 * exclusive cpuset is created, we also create and attach a new root-domain
 819 * object.
 820 *
 821 */
 822struct root_domain {
 823	atomic_t		refcount;
 824	atomic_t		rto_count;
 825	struct rcu_head		rcu;
 826	cpumask_var_t		span;
 827	cpumask_var_t		online;
 828
 829	/*
 830	 * Indicate pullable load on at least one CPU, e.g:
 831	 * - More than one runnable task
 832	 * - Running task is misfit
 833	 */
 834	int			overload;
 835
 836	/* Indicate one or more cpus over-utilized (tipping point) */
 837	int			overutilized;
 838
 839	/*
 840	 * The bit corresponding to a CPU gets set here if such CPU has more
 841	 * than one runnable -deadline task (as it is below for RT tasks).
 842	 */
 843	cpumask_var_t		dlo_mask;
 844	atomic_t		dlo_count;
 845	struct dl_bw		dl_bw;
 846	struct cpudl		cpudl;
 847
 848	/*
 849	 * Indicate whether a root_domain's dl_bw has been checked or
 850	 * updated. It's monotonously increasing value.
 851	 *
 852	 * Also, some corner cases, like 'wrap around' is dangerous, but given
 853	 * that u64 is 'big enough'. So that shouldn't be a concern.
 854	 */
 855	u64 visit_gen;
 856
 857#ifdef HAVE_RT_PUSH_IPI
 858	/*
 859	 * For IPI pull requests, loop across the rto_mask.
 860	 */
 861	struct irq_work		rto_push_work;
 862	raw_spinlock_t		rto_lock;
 863	/* These are only updated and read within rto_lock */
 864	int			rto_loop;
 865	int			rto_cpu;
 866	/* These atomics are updated outside of a lock */
 867	atomic_t		rto_loop_next;
 868	atomic_t		rto_loop_start;
 869#endif
 870	/*
 871	 * The "RT overload" flag: it gets set if a CPU has more than
 872	 * one runnable RT task.
 873	 */
 874	cpumask_var_t		rto_mask;
 875	struct cpupri		cpupri;
 876
 877	unsigned long		max_cpu_capacity;
 878
 879	/*
 880	 * NULL-terminated list of performance domains intersecting with the
 881	 * CPUs of the rd. Protected by RCU.
 882	 */
 883	struct perf_domain __rcu *pd;
 884};
 885
 886extern void init_defrootdomain(void);
 887extern int sched_init_domains(const struct cpumask *cpu_map);
 888extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
 889extern void sched_get_rd(struct root_domain *rd);
 890extern void sched_put_rd(struct root_domain *rd);
 891
 892#ifdef HAVE_RT_PUSH_IPI
 893extern void rto_push_irq_work_func(struct irq_work *work);
 894#endif
 895#endif /* CONFIG_SMP */
 896
 897#ifdef CONFIG_UCLAMP_TASK
 898/*
 899 * struct uclamp_bucket - Utilization clamp bucket
 900 * @value: utilization clamp value for tasks on this clamp bucket
 901 * @tasks: number of RUNNABLE tasks on this clamp bucket
 902 *
 903 * Keep track of how many tasks are RUNNABLE for a given utilization
 904 * clamp value.
 905 */
 906struct uclamp_bucket {
 907	unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
 908	unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
 909};
 910
 911/*
 912 * struct uclamp_rq - rq's utilization clamp
 913 * @value: currently active clamp values for a rq
 914 * @bucket: utilization clamp buckets affecting a rq
 915 *
 916 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
 917 * A clamp value is affecting a rq when there is at least one task RUNNABLE
 918 * (or actually running) with that value.
 919 *
 920 * There are up to UCLAMP_CNT possible different clamp values, currently there
 921 * are only two: minimum utilization and maximum utilization.
 922 *
 923 * All utilization clamping values are MAX aggregated, since:
 924 * - for util_min: we want to run the CPU at least at the max of the minimum
 925 *   utilization required by its currently RUNNABLE tasks.
 926 * - for util_max: we want to allow the CPU to run up to the max of the
 927 *   maximum utilization allowed by its currently RUNNABLE tasks.
 928 *
 929 * Since on each system we expect only a limited number of different
 930 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
 931 * the metrics required to compute all the per-rq utilization clamp values.
 932 */
 933struct uclamp_rq {
 934	unsigned int value;
 935	struct uclamp_bucket bucket[UCLAMP_BUCKETS];
 936};
 937
 938DECLARE_STATIC_KEY_FALSE(sched_uclamp_used);
 939#endif /* CONFIG_UCLAMP_TASK */
 940
 941struct rq;
 942struct balance_callback {
 943	struct balance_callback *next;
 944	void (*func)(struct rq *rq);
 945};
 946
 947/*
 948 * This is the main, per-CPU runqueue data structure.
 949 *
 950 * Locking rule: those places that want to lock multiple runqueues
 951 * (such as the load balancing or the thread migration code), lock
 952 * acquire operations must be ordered by ascending &runqueue.
 953 */
 954struct rq {
 955	/* runqueue lock: */
 956	raw_spinlock_t		__lock;
 957
 958	/*
 959	 * nr_running and cpu_load should be in the same cacheline because
 960	 * remote CPUs use both these fields when doing load calculation.
 961	 */
 962	unsigned int		nr_running;
 963#ifdef CONFIG_NUMA_BALANCING
 964	unsigned int		nr_numa_running;
 965	unsigned int		nr_preferred_running;
 966	unsigned int		numa_migrate_on;
 967#endif
 968#ifdef CONFIG_NO_HZ_COMMON
 969#ifdef CONFIG_SMP
 970	unsigned long		last_blocked_load_update_tick;
 971	unsigned int		has_blocked_load;
 972	call_single_data_t	nohz_csd;
 973#endif /* CONFIG_SMP */
 974	unsigned int		nohz_tick_stopped;
 975	atomic_t		nohz_flags;
 976#endif /* CONFIG_NO_HZ_COMMON */
 977
 978#ifdef CONFIG_SMP
 979	unsigned int		ttwu_pending;
 980#endif
 981	u64			nr_switches;
 982
 983#ifdef CONFIG_UCLAMP_TASK
 984	/* Utilization clamp values based on CPU's RUNNABLE tasks */
 985	struct uclamp_rq	uclamp[UCLAMP_CNT] ____cacheline_aligned;
 986	unsigned int		uclamp_flags;
 987#define UCLAMP_FLAG_IDLE 0x01
 988#endif
 989
 990	struct cfs_rq		cfs;
 991	struct rt_rq		rt;
 992	struct dl_rq		dl;
 993
 994#ifdef CONFIG_FAIR_GROUP_SCHED
 995	/* list of leaf cfs_rq on this CPU: */
 996	struct list_head	leaf_cfs_rq_list;
 997	struct list_head	*tmp_alone_branch;
 998#endif /* CONFIG_FAIR_GROUP_SCHED */
 999
1000	/*
1001	 * This is part of a global counter where only the total sum
1002	 * over all CPUs matters. A task can increase this counter on
1003	 * one CPU and if it got migrated afterwards it may decrease
1004	 * it on another CPU. Always updated under the runqueue lock:
1005	 */
1006	unsigned int		nr_uninterruptible;
1007
1008	struct task_struct __rcu	*curr;
1009	struct task_struct	*idle;
1010	struct task_struct	*stop;
1011	unsigned long		next_balance;
1012	struct mm_struct	*prev_mm;
1013
1014	unsigned int		clock_update_flags;
1015	u64			clock;
1016	/* Ensure that all clocks are in the same cache line */
1017	u64			clock_task ____cacheline_aligned;
1018	u64			clock_pelt;
1019	unsigned long		lost_idle_time;
1020	u64			clock_pelt_idle;
1021	u64			clock_idle;
1022#ifndef CONFIG_64BIT
1023	u64			clock_pelt_idle_copy;
1024	u64			clock_idle_copy;
1025#endif
1026
1027	atomic_t		nr_iowait;
1028
1029#ifdef CONFIG_SCHED_DEBUG
1030	u64 last_seen_need_resched_ns;
1031	int ticks_without_resched;
1032#endif
1033
1034#ifdef CONFIG_MEMBARRIER
1035	int membarrier_state;
1036#endif
1037
1038#ifdef CONFIG_SMP
1039	struct root_domain		*rd;
1040	struct sched_domain __rcu	*sd;
1041
1042	unsigned long		cpu_capacity;
1043	unsigned long		cpu_capacity_orig;
1044	unsigned long		cpu_capacity_inverted;
1045
1046	struct balance_callback *balance_callback;
1047
1048	unsigned char		nohz_idle_balance;
1049	unsigned char		idle_balance;
1050
1051	unsigned long		misfit_task_load;
1052
1053	/* For active balancing */
1054	int			active_balance;
1055	int			push_cpu;
1056	struct cpu_stop_work	active_balance_work;
1057
1058	/* CPU of this runqueue: */
1059	int			cpu;
1060	int			online;
1061
1062	struct list_head cfs_tasks;
1063
1064	struct sched_avg	avg_rt;
1065	struct sched_avg	avg_dl;
1066#ifdef CONFIG_HAVE_SCHED_AVG_IRQ
1067	struct sched_avg	avg_irq;
1068#endif
1069#ifdef CONFIG_SCHED_THERMAL_PRESSURE
1070	struct sched_avg	avg_thermal;
1071#endif
1072	u64			idle_stamp;
1073	u64			avg_idle;
1074
1075	unsigned long		wake_stamp;
1076	u64			wake_avg_idle;
1077
1078	/* This is used to determine avg_idle's max value */
1079	u64			max_idle_balance_cost;
1080
1081#ifdef CONFIG_HOTPLUG_CPU
1082	struct rcuwait		hotplug_wait;
1083#endif
1084#endif /* CONFIG_SMP */
1085
1086#ifdef CONFIG_IRQ_TIME_ACCOUNTING
1087	u64			prev_irq_time;
1088#endif
1089#ifdef CONFIG_PARAVIRT
1090	u64			prev_steal_time;
1091#endif
1092#ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
1093	u64			prev_steal_time_rq;
1094#endif
1095
1096	/* calc_load related fields */
1097	unsigned long		calc_load_update;
1098	long			calc_load_active;
1099
1100#ifdef CONFIG_SCHED_HRTICK
1101#ifdef CONFIG_SMP
1102	call_single_data_t	hrtick_csd;
1103#endif
1104	struct hrtimer		hrtick_timer;
1105	ktime_t 		hrtick_time;
1106#endif
1107
1108#ifdef CONFIG_SCHEDSTATS
1109	/* latency stats */
1110	struct sched_info	rq_sched_info;
1111	unsigned long long	rq_cpu_time;
1112	/* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
1113
1114	/* sys_sched_yield() stats */
1115	unsigned int		yld_count;
1116
1117	/* schedule() stats */
1118	unsigned int		sched_count;
1119	unsigned int		sched_goidle;
1120
1121	/* try_to_wake_up() stats */
1122	unsigned int		ttwu_count;
1123	unsigned int		ttwu_local;
1124#endif
1125
1126#ifdef CONFIG_CPU_IDLE
1127	/* Must be inspected within a rcu lock section */
1128	struct cpuidle_state	*idle_state;
1129#endif
1130
1131#ifdef CONFIG_SMP
1132	unsigned int		nr_pinned;
1133#endif
1134	unsigned int		push_busy;
1135	struct cpu_stop_work	push_work;
1136
1137#ifdef CONFIG_SCHED_CORE
1138	/* per rq */
1139	struct rq		*core;
1140	struct task_struct	*core_pick;
1141	unsigned int		core_enabled;
1142	unsigned int		core_sched_seq;
1143	struct rb_root		core_tree;
1144
1145	/* shared state -- careful with sched_core_cpu_deactivate() */
1146	unsigned int		core_task_seq;
1147	unsigned int		core_pick_seq;
1148	unsigned long		core_cookie;
1149	unsigned int		core_forceidle_count;
1150	unsigned int		core_forceidle_seq;
1151	unsigned int		core_forceidle_occupation;
1152	u64			core_forceidle_start;
1153#endif
1154
1155	/* Scratch cpumask to be temporarily used under rq_lock */
1156	cpumask_var_t		scratch_mask;
 
 
 
 
 
1157};
1158
1159#ifdef CONFIG_FAIR_GROUP_SCHED
1160
1161/* CPU runqueue to which this cfs_rq is attached */
1162static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1163{
1164	return cfs_rq->rq;
1165}
1166
1167#else
1168
1169static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1170{
1171	return container_of(cfs_rq, struct rq, cfs);
1172}
1173#endif
1174
1175static inline int cpu_of(struct rq *rq)
1176{
1177#ifdef CONFIG_SMP
1178	return rq->cpu;
1179#else
1180	return 0;
1181#endif
1182}
1183
1184#define MDF_PUSH	0x01
1185
1186static inline bool is_migration_disabled(struct task_struct *p)
1187{
1188#ifdef CONFIG_SMP
1189	return p->migration_disabled;
1190#else
1191	return false;
1192#endif
1193}
1194
1195DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1196
1197#define cpu_rq(cpu)		(&per_cpu(runqueues, (cpu)))
1198#define this_rq()		this_cpu_ptr(&runqueues)
1199#define task_rq(p)		cpu_rq(task_cpu(p))
1200#define cpu_curr(cpu)		(cpu_rq(cpu)->curr)
1201#define raw_rq()		raw_cpu_ptr(&runqueues)
1202
1203struct sched_group;
1204#ifdef CONFIG_SCHED_CORE
1205static inline struct cpumask *sched_group_span(struct sched_group *sg);
1206
1207DECLARE_STATIC_KEY_FALSE(__sched_core_enabled);
1208
1209static inline bool sched_core_enabled(struct rq *rq)
1210{
1211	return static_branch_unlikely(&__sched_core_enabled) && rq->core_enabled;
1212}
1213
1214static inline bool sched_core_disabled(void)
1215{
1216	return !static_branch_unlikely(&__sched_core_enabled);
1217}
1218
1219/*
1220 * Be careful with this function; not for general use. The return value isn't
1221 * stable unless you actually hold a relevant rq->__lock.
1222 */
1223static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1224{
1225	if (sched_core_enabled(rq))
1226		return &rq->core->__lock;
1227
1228	return &rq->__lock;
1229}
1230
1231static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1232{
1233	if (rq->core_enabled)
1234		return &rq->core->__lock;
1235
1236	return &rq->__lock;
1237}
1238
1239bool cfs_prio_less(struct task_struct *a, struct task_struct *b, bool fi);
 
 
1240
1241/*
1242 * Helpers to check if the CPU's core cookie matches with the task's cookie
1243 * when core scheduling is enabled.
1244 * A special case is that the task's cookie always matches with CPU's core
1245 * cookie if the CPU is in an idle core.
1246 */
1247static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1248{
1249	/* Ignore cookie match if core scheduler is not enabled on the CPU. */
1250	if (!sched_core_enabled(rq))
1251		return true;
1252
1253	return rq->core->core_cookie == p->core_cookie;
1254}
1255
1256static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1257{
1258	bool idle_core = true;
1259	int cpu;
1260
1261	/* Ignore cookie match if core scheduler is not enabled on the CPU. */
1262	if (!sched_core_enabled(rq))
1263		return true;
1264
1265	for_each_cpu(cpu, cpu_smt_mask(cpu_of(rq))) {
1266		if (!available_idle_cpu(cpu)) {
1267			idle_core = false;
1268			break;
1269		}
1270	}
1271
1272	/*
1273	 * A CPU in an idle core is always the best choice for tasks with
1274	 * cookies.
1275	 */
1276	return idle_core || rq->core->core_cookie == p->core_cookie;
1277}
1278
1279static inline bool sched_group_cookie_match(struct rq *rq,
1280					    struct task_struct *p,
1281					    struct sched_group *group)
1282{
1283	int cpu;
1284
1285	/* Ignore cookie match if core scheduler is not enabled on the CPU. */
1286	if (!sched_core_enabled(rq))
1287		return true;
1288
1289	for_each_cpu_and(cpu, sched_group_span(group), p->cpus_ptr) {
1290		if (sched_core_cookie_match(cpu_rq(cpu), p))
1291			return true;
1292	}
1293	return false;
1294}
1295
1296static inline bool sched_core_enqueued(struct task_struct *p)
1297{
1298	return !RB_EMPTY_NODE(&p->core_node);
1299}
1300
1301extern void sched_core_enqueue(struct rq *rq, struct task_struct *p);
1302extern void sched_core_dequeue(struct rq *rq, struct task_struct *p, int flags);
1303
1304extern void sched_core_get(void);
1305extern void sched_core_put(void);
1306
1307#else /* !CONFIG_SCHED_CORE */
1308
1309static inline bool sched_core_enabled(struct rq *rq)
1310{
1311	return false;
1312}
1313
1314static inline bool sched_core_disabled(void)
1315{
1316	return true;
1317}
1318
1319static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1320{
1321	return &rq->__lock;
1322}
1323
1324static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1325{
1326	return &rq->__lock;
1327}
1328
1329static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1330{
1331	return true;
1332}
1333
1334static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1335{
1336	return true;
1337}
1338
1339static inline bool sched_group_cookie_match(struct rq *rq,
1340					    struct task_struct *p,
1341					    struct sched_group *group)
1342{
1343	return true;
1344}
1345#endif /* CONFIG_SCHED_CORE */
1346
1347static inline void lockdep_assert_rq_held(struct rq *rq)
1348{
1349	lockdep_assert_held(__rq_lockp(rq));
1350}
1351
1352extern void raw_spin_rq_lock_nested(struct rq *rq, int subclass);
1353extern bool raw_spin_rq_trylock(struct rq *rq);
1354extern void raw_spin_rq_unlock(struct rq *rq);
1355
1356static inline void raw_spin_rq_lock(struct rq *rq)
1357{
1358	raw_spin_rq_lock_nested(rq, 0);
1359}
1360
1361static inline void raw_spin_rq_lock_irq(struct rq *rq)
1362{
1363	local_irq_disable();
1364	raw_spin_rq_lock(rq);
1365}
1366
1367static inline void raw_spin_rq_unlock_irq(struct rq *rq)
1368{
1369	raw_spin_rq_unlock(rq);
1370	local_irq_enable();
1371}
1372
1373static inline unsigned long _raw_spin_rq_lock_irqsave(struct rq *rq)
1374{
1375	unsigned long flags;
1376	local_irq_save(flags);
1377	raw_spin_rq_lock(rq);
1378	return flags;
1379}
1380
1381static inline void raw_spin_rq_unlock_irqrestore(struct rq *rq, unsigned long flags)
1382{
1383	raw_spin_rq_unlock(rq);
1384	local_irq_restore(flags);
1385}
1386
1387#define raw_spin_rq_lock_irqsave(rq, flags)	\
1388do {						\
1389	flags = _raw_spin_rq_lock_irqsave(rq);	\
1390} while (0)
1391
1392#ifdef CONFIG_SCHED_SMT
1393extern void __update_idle_core(struct rq *rq);
1394
1395static inline void update_idle_core(struct rq *rq)
1396{
1397	if (static_branch_unlikely(&sched_smt_present))
1398		__update_idle_core(rq);
1399}
1400
1401#else
1402static inline void update_idle_core(struct rq *rq) { }
1403#endif
1404
1405#ifdef CONFIG_FAIR_GROUP_SCHED
1406static inline struct task_struct *task_of(struct sched_entity *se)
1407{
1408	SCHED_WARN_ON(!entity_is_task(se));
1409	return container_of(se, struct task_struct, se);
1410}
1411
1412static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1413{
1414	return p->se.cfs_rq;
1415}
1416
1417/* runqueue on which this entity is (to be) queued */
1418static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
1419{
1420	return se->cfs_rq;
1421}
1422
1423/* runqueue "owned" by this group */
1424static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1425{
1426	return grp->my_q;
1427}
1428
1429#else
1430
1431static inline struct task_struct *task_of(struct sched_entity *se)
1432{
1433	return container_of(se, struct task_struct, se);
1434}
1435
1436static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1437{
1438	return &task_rq(p)->cfs;
1439}
1440
1441static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
1442{
1443	struct task_struct *p = task_of(se);
1444	struct rq *rq = task_rq(p);
1445
1446	return &rq->cfs;
1447}
1448
1449/* runqueue "owned" by this group */
1450static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1451{
1452	return NULL;
1453}
1454#endif
1455
1456extern void update_rq_clock(struct rq *rq);
1457
1458/*
1459 * rq::clock_update_flags bits
1460 *
1461 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1462 *  call to __schedule(). This is an optimisation to avoid
1463 *  neighbouring rq clock updates.
1464 *
1465 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1466 *  in effect and calls to update_rq_clock() are being ignored.
1467 *
1468 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1469 *  made to update_rq_clock() since the last time rq::lock was pinned.
1470 *
1471 * If inside of __schedule(), clock_update_flags will have been
1472 * shifted left (a left shift is a cheap operation for the fast path
1473 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1474 *
1475 *	if (rq-clock_update_flags >= RQCF_UPDATED)
1476 *
1477 * to check if %RQCF_UPDATED is set. It'll never be shifted more than
1478 * one position though, because the next rq_unpin_lock() will shift it
1479 * back.
1480 */
1481#define RQCF_REQ_SKIP		0x01
1482#define RQCF_ACT_SKIP		0x02
1483#define RQCF_UPDATED		0x04
1484
1485static inline void assert_clock_updated(struct rq *rq)
1486{
1487	/*
1488	 * The only reason for not seeing a clock update since the
1489	 * last rq_pin_lock() is if we're currently skipping updates.
1490	 */
1491	SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1492}
1493
1494static inline u64 rq_clock(struct rq *rq)
1495{
1496	lockdep_assert_rq_held(rq);
1497	assert_clock_updated(rq);
1498
1499	return rq->clock;
1500}
1501
1502static inline u64 rq_clock_task(struct rq *rq)
1503{
1504	lockdep_assert_rq_held(rq);
1505	assert_clock_updated(rq);
1506
1507	return rq->clock_task;
1508}
1509
1510/**
1511 * By default the decay is the default pelt decay period.
1512 * The decay shift can change the decay period in
1513 * multiples of 32.
1514 *  Decay shift		Decay period(ms)
1515 *	0			32
1516 *	1			64
1517 *	2			128
1518 *	3			256
1519 *	4			512
1520 */
1521extern int sched_thermal_decay_shift;
1522
1523static inline u64 rq_clock_thermal(struct rq *rq)
1524{
1525	return rq_clock_task(rq) >> sched_thermal_decay_shift;
1526}
1527
1528static inline void rq_clock_skip_update(struct rq *rq)
1529{
1530	lockdep_assert_rq_held(rq);
1531	rq->clock_update_flags |= RQCF_REQ_SKIP;
1532}
1533
1534/*
1535 * See rt task throttling, which is the only time a skip
1536 * request is canceled.
1537 */
1538static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1539{
1540	lockdep_assert_rq_held(rq);
1541	rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1542}
1543
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1544struct rq_flags {
1545	unsigned long flags;
1546	struct pin_cookie cookie;
1547#ifdef CONFIG_SCHED_DEBUG
1548	/*
1549	 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1550	 * current pin context is stashed here in case it needs to be
1551	 * restored in rq_repin_lock().
1552	 */
1553	unsigned int clock_update_flags;
1554#endif
1555};
1556
1557extern struct balance_callback balance_push_callback;
1558
1559/*
1560 * Lockdep annotation that avoids accidental unlocks; it's like a
1561 * sticky/continuous lockdep_assert_held().
1562 *
1563 * This avoids code that has access to 'struct rq *rq' (basically everything in
1564 * the scheduler) from accidentally unlocking the rq if they do not also have a
1565 * copy of the (on-stack) 'struct rq_flags rf'.
1566 *
1567 * Also see Documentation/locking/lockdep-design.rst.
1568 */
1569static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1570{
1571	rf->cookie = lockdep_pin_lock(__rq_lockp(rq));
1572
1573#ifdef CONFIG_SCHED_DEBUG
1574	rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1575	rf->clock_update_flags = 0;
1576#ifdef CONFIG_SMP
1577	SCHED_WARN_ON(rq->balance_callback && rq->balance_callback != &balance_push_callback);
1578#endif
1579#endif
1580}
1581
1582static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1583{
1584#ifdef CONFIG_SCHED_DEBUG
1585	if (rq->clock_update_flags > RQCF_ACT_SKIP)
1586		rf->clock_update_flags = RQCF_UPDATED;
1587#endif
1588
1589	lockdep_unpin_lock(__rq_lockp(rq), rf->cookie);
1590}
1591
1592static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1593{
1594	lockdep_repin_lock(__rq_lockp(rq), rf->cookie);
1595
1596#ifdef CONFIG_SCHED_DEBUG
1597	/*
1598	 * Restore the value we stashed in @rf for this pin context.
1599	 */
1600	rq->clock_update_flags |= rf->clock_update_flags;
1601#endif
1602}
1603
1604struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1605	__acquires(rq->lock);
1606
1607struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1608	__acquires(p->pi_lock)
1609	__acquires(rq->lock);
1610
1611static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1612	__releases(rq->lock)
1613{
1614	rq_unpin_lock(rq, rf);
1615	raw_spin_rq_unlock(rq);
1616}
1617
1618static inline void
1619task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1620	__releases(rq->lock)
1621	__releases(p->pi_lock)
1622{
1623	rq_unpin_lock(rq, rf);
1624	raw_spin_rq_unlock(rq);
1625	raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1626}
1627
 
 
 
 
 
1628static inline void
1629rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1630	__acquires(rq->lock)
1631{
1632	raw_spin_rq_lock_irqsave(rq, rf->flags);
1633	rq_pin_lock(rq, rf);
1634}
1635
1636static inline void
1637rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1638	__acquires(rq->lock)
1639{
1640	raw_spin_rq_lock_irq(rq);
1641	rq_pin_lock(rq, rf);
1642}
1643
1644static inline void
1645rq_lock(struct rq *rq, struct rq_flags *rf)
1646	__acquires(rq->lock)
1647{
1648	raw_spin_rq_lock(rq);
1649	rq_pin_lock(rq, rf);
1650}
1651
1652static inline void
1653rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1654	__releases(rq->lock)
1655{
1656	rq_unpin_lock(rq, rf);
1657	raw_spin_rq_unlock_irqrestore(rq, rf->flags);
1658}
1659
1660static inline void
1661rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1662	__releases(rq->lock)
1663{
1664	rq_unpin_lock(rq, rf);
1665	raw_spin_rq_unlock_irq(rq);
1666}
1667
1668static inline void
1669rq_unlock(struct rq *rq, struct rq_flags *rf)
1670	__releases(rq->lock)
1671{
1672	rq_unpin_lock(rq, rf);
1673	raw_spin_rq_unlock(rq);
1674}
1675
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1676static inline struct rq *
1677this_rq_lock_irq(struct rq_flags *rf)
1678	__acquires(rq->lock)
1679{
1680	struct rq *rq;
1681
1682	local_irq_disable();
1683	rq = this_rq();
1684	rq_lock(rq, rf);
1685	return rq;
1686}
1687
1688#ifdef CONFIG_NUMA
1689enum numa_topology_type {
1690	NUMA_DIRECT,
1691	NUMA_GLUELESS_MESH,
1692	NUMA_BACKPLANE,
1693};
1694extern enum numa_topology_type sched_numa_topology_type;
1695extern int sched_max_numa_distance;
1696extern bool find_numa_distance(int distance);
1697extern void sched_init_numa(int offline_node);
1698extern void sched_update_numa(int cpu, bool online);
1699extern void sched_domains_numa_masks_set(unsigned int cpu);
1700extern void sched_domains_numa_masks_clear(unsigned int cpu);
1701extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1702#else
1703static inline void sched_init_numa(int offline_node) { }
1704static inline void sched_update_numa(int cpu, bool online) { }
1705static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1706static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1707static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1708{
1709	return nr_cpu_ids;
1710}
1711#endif
1712
1713#ifdef CONFIG_NUMA_BALANCING
1714/* The regions in numa_faults array from task_struct */
1715enum numa_faults_stats {
1716	NUMA_MEM = 0,
1717	NUMA_CPU,
1718	NUMA_MEMBUF,
1719	NUMA_CPUBUF
1720};
1721extern void sched_setnuma(struct task_struct *p, int node);
1722extern int migrate_task_to(struct task_struct *p, int cpu);
1723extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1724			int cpu, int scpu);
1725extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1726#else
1727static inline void
1728init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1729{
1730}
1731#endif /* CONFIG_NUMA_BALANCING */
1732
1733#ifdef CONFIG_SMP
1734
1735static inline void
1736queue_balance_callback(struct rq *rq,
1737		       struct balance_callback *head,
1738		       void (*func)(struct rq *rq))
1739{
1740	lockdep_assert_rq_held(rq);
1741
1742	/*
1743	 * Don't (re)queue an already queued item; nor queue anything when
1744	 * balance_push() is active, see the comment with
1745	 * balance_push_callback.
1746	 */
1747	if (unlikely(head->next || rq->balance_callback == &balance_push_callback))
1748		return;
1749
1750	head->func = func;
1751	head->next = rq->balance_callback;
1752	rq->balance_callback = head;
1753}
1754
1755#define rcu_dereference_check_sched_domain(p) \
1756	rcu_dereference_check((p), \
1757			      lockdep_is_held(&sched_domains_mutex))
1758
1759/*
1760 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1761 * See destroy_sched_domains: call_rcu for details.
1762 *
1763 * The domain tree of any CPU may only be accessed from within
1764 * preempt-disabled sections.
1765 */
1766#define for_each_domain(cpu, __sd) \
1767	for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1768			__sd; __sd = __sd->parent)
1769
 
 
 
 
 
 
 
1770/**
1771 * highest_flag_domain - Return highest sched_domain containing flag.
1772 * @cpu:	The CPU whose highest level of sched domain is to
1773 *		be returned.
1774 * @flag:	The flag to check for the highest sched_domain
1775 *		for the given CPU.
1776 *
1777 * Returns the highest sched_domain of a CPU which contains the given flag.
 
1778 */
1779static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1780{
1781	struct sched_domain *sd, *hsd = NULL;
1782
1783	for_each_domain(cpu, sd) {
1784		if (!(sd->flags & flag))
 
 
 
 
 
 
 
 
 
1785			break;
1786		hsd = sd;
1787	}
1788
1789	return hsd;
1790}
1791
1792static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1793{
1794	struct sched_domain *sd;
1795
1796	for_each_domain(cpu, sd) {
1797		if (sd->flags & flag)
1798			break;
1799	}
1800
1801	return sd;
1802}
1803
1804DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
1805DECLARE_PER_CPU(int, sd_llc_size);
1806DECLARE_PER_CPU(int, sd_llc_id);
 
1807DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
1808DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
1809DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
1810DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
1811extern struct static_key_false sched_asym_cpucapacity;
 
1812
1813static __always_inline bool sched_asym_cpucap_active(void)
1814{
1815	return static_branch_unlikely(&sched_asym_cpucapacity);
1816}
1817
1818struct sched_group_capacity {
1819	atomic_t		ref;
1820	/*
1821	 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1822	 * for a single CPU.
1823	 */
1824	unsigned long		capacity;
1825	unsigned long		min_capacity;		/* Min per-CPU capacity in group */
1826	unsigned long		max_capacity;		/* Max per-CPU capacity in group */
1827	unsigned long		next_update;
1828	int			imbalance;		/* XXX unrelated to capacity but shared group state */
1829
1830#ifdef CONFIG_SCHED_DEBUG
1831	int			id;
1832#endif
1833
1834	unsigned long		cpumask[];		/* Balance mask */
1835};
1836
1837struct sched_group {
1838	struct sched_group	*next;			/* Must be a circular list */
1839	atomic_t		ref;
1840
1841	unsigned int		group_weight;
 
1842	struct sched_group_capacity *sgc;
1843	int			asym_prefer_cpu;	/* CPU of highest priority in group */
1844	int			flags;
1845
1846	/*
1847	 * The CPUs this group covers.
1848	 *
1849	 * NOTE: this field is variable length. (Allocated dynamically
1850	 * by attaching extra space to the end of the structure,
1851	 * depending on how many CPUs the kernel has booted up with)
1852	 */
1853	unsigned long		cpumask[];
1854};
1855
1856static inline struct cpumask *sched_group_span(struct sched_group *sg)
1857{
1858	return to_cpumask(sg->cpumask);
1859}
1860
1861/*
1862 * See build_balance_mask().
1863 */
1864static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1865{
1866	return to_cpumask(sg->sgc->cpumask);
1867}
1868
1869extern int group_balance_cpu(struct sched_group *sg);
1870
1871#ifdef CONFIG_SCHED_DEBUG
1872void update_sched_domain_debugfs(void);
1873void dirty_sched_domain_sysctl(int cpu);
1874#else
1875static inline void update_sched_domain_debugfs(void)
1876{
1877}
1878static inline void dirty_sched_domain_sysctl(int cpu)
1879{
1880}
1881#endif
1882
1883extern int sched_update_scaling(void);
1884
1885static inline const struct cpumask *task_user_cpus(struct task_struct *p)
1886{
1887	if (!p->user_cpus_ptr)
1888		return cpu_possible_mask; /* &init_task.cpus_mask */
1889	return p->user_cpus_ptr;
1890}
1891#endif /* CONFIG_SMP */
1892
1893#include "stats.h"
1894
1895#if defined(CONFIG_SCHED_CORE) && defined(CONFIG_SCHEDSTATS)
1896
1897extern void __sched_core_account_forceidle(struct rq *rq);
1898
1899static inline void sched_core_account_forceidle(struct rq *rq)
1900{
1901	if (schedstat_enabled())
1902		__sched_core_account_forceidle(rq);
1903}
1904
1905extern void __sched_core_tick(struct rq *rq);
1906
1907static inline void sched_core_tick(struct rq *rq)
1908{
1909	if (sched_core_enabled(rq) && schedstat_enabled())
1910		__sched_core_tick(rq);
1911}
1912
1913#else
1914
1915static inline void sched_core_account_forceidle(struct rq *rq) {}
1916
1917static inline void sched_core_tick(struct rq *rq) {}
1918
1919#endif /* CONFIG_SCHED_CORE && CONFIG_SCHEDSTATS */
1920
1921#ifdef CONFIG_CGROUP_SCHED
1922
1923/*
1924 * Return the group to which this tasks belongs.
1925 *
1926 * We cannot use task_css() and friends because the cgroup subsystem
1927 * changes that value before the cgroup_subsys::attach() method is called,
1928 * therefore we cannot pin it and might observe the wrong value.
1929 *
1930 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1931 * core changes this before calling sched_move_task().
1932 *
1933 * Instead we use a 'copy' which is updated from sched_move_task() while
1934 * holding both task_struct::pi_lock and rq::lock.
1935 */
1936static inline struct task_group *task_group(struct task_struct *p)
1937{
1938	return p->sched_task_group;
1939}
1940
1941/* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1942static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1943{
1944#if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1945	struct task_group *tg = task_group(p);
1946#endif
1947
1948#ifdef CONFIG_FAIR_GROUP_SCHED
1949	set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1950	p->se.cfs_rq = tg->cfs_rq[cpu];
1951	p->se.parent = tg->se[cpu];
1952	p->se.depth = tg->se[cpu] ? tg->se[cpu]->depth + 1 : 0;
1953#endif
1954
1955#ifdef CONFIG_RT_GROUP_SCHED
1956	p->rt.rt_rq  = tg->rt_rq[cpu];
1957	p->rt.parent = tg->rt_se[cpu];
1958#endif
1959}
1960
1961#else /* CONFIG_CGROUP_SCHED */
1962
1963static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1964static inline struct task_group *task_group(struct task_struct *p)
1965{
1966	return NULL;
1967}
1968
1969#endif /* CONFIG_CGROUP_SCHED */
1970
1971static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1972{
1973	set_task_rq(p, cpu);
1974#ifdef CONFIG_SMP
1975	/*
1976	 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1977	 * successfully executed on another CPU. We must ensure that updates of
1978	 * per-task data have been completed by this moment.
1979	 */
1980	smp_wmb();
1981	WRITE_ONCE(task_thread_info(p)->cpu, cpu);
1982	p->wake_cpu = cpu;
1983#endif
1984}
1985
1986/*
1987 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1988 */
1989#ifdef CONFIG_SCHED_DEBUG
1990# define const_debug __read_mostly
1991#else
1992# define const_debug const
1993#endif
1994
1995#define SCHED_FEAT(name, enabled)	\
1996	__SCHED_FEAT_##name ,
1997
1998enum {
1999#include "features.h"
2000	__SCHED_FEAT_NR,
2001};
2002
2003#undef SCHED_FEAT
2004
2005#ifdef CONFIG_SCHED_DEBUG
2006
2007/*
2008 * To support run-time toggling of sched features, all the translation units
2009 * (but core.c) reference the sysctl_sched_features defined in core.c.
2010 */
2011extern const_debug unsigned int sysctl_sched_features;
2012
2013#ifdef CONFIG_JUMP_LABEL
2014#define SCHED_FEAT(name, enabled)					\
2015static __always_inline bool static_branch_##name(struct static_key *key) \
2016{									\
2017	return static_key_##enabled(key);				\
2018}
2019
2020#include "features.h"
2021#undef SCHED_FEAT
2022
2023extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
2024#define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
2025
2026#else /* !CONFIG_JUMP_LABEL */
2027
2028#define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
2029
2030#endif /* CONFIG_JUMP_LABEL */
2031
2032#else /* !SCHED_DEBUG */
2033
2034/*
2035 * Each translation unit has its own copy of sysctl_sched_features to allow
2036 * constants propagation at compile time and compiler optimization based on
2037 * features default.
2038 */
2039#define SCHED_FEAT(name, enabled)	\
2040	(1UL << __SCHED_FEAT_##name) * enabled |
2041static const_debug __maybe_unused unsigned int sysctl_sched_features =
2042#include "features.h"
2043	0;
2044#undef SCHED_FEAT
2045
2046#define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
2047
2048#endif /* SCHED_DEBUG */
2049
2050extern struct static_key_false sched_numa_balancing;
2051extern struct static_key_false sched_schedstats;
2052
2053static inline u64 global_rt_period(void)
2054{
2055	return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
2056}
2057
2058static inline u64 global_rt_runtime(void)
2059{
2060	if (sysctl_sched_rt_runtime < 0)
2061		return RUNTIME_INF;
2062
2063	return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
2064}
2065
2066static inline int task_current(struct rq *rq, struct task_struct *p)
2067{
2068	return rq->curr == p;
2069}
2070
2071static inline int task_on_cpu(struct rq *rq, struct task_struct *p)
2072{
2073#ifdef CONFIG_SMP
2074	return p->on_cpu;
2075#else
2076	return task_current(rq, p);
2077#endif
2078}
2079
2080static inline int task_on_rq_queued(struct task_struct *p)
2081{
2082	return p->on_rq == TASK_ON_RQ_QUEUED;
2083}
2084
2085static inline int task_on_rq_migrating(struct task_struct *p)
2086{
2087	return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
2088}
2089
2090/* Wake flags. The first three directly map to some SD flag value */
2091#define WF_EXEC     0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */
2092#define WF_FORK     0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */
2093#define WF_TTWU     0x08 /* Wakeup;            maps to SD_BALANCE_WAKE */
2094
2095#define WF_SYNC     0x10 /* Waker goes to sleep after wakeup */
2096#define WF_MIGRATED 0x20 /* Internal use, task got migrated */
 
2097
2098#ifdef CONFIG_SMP
2099static_assert(WF_EXEC == SD_BALANCE_EXEC);
2100static_assert(WF_FORK == SD_BALANCE_FORK);
2101static_assert(WF_TTWU == SD_BALANCE_WAKE);
2102#endif
2103
2104/*
2105 * To aid in avoiding the subversion of "niceness" due to uneven distribution
2106 * of tasks with abnormal "nice" values across CPUs the contribution that
2107 * each task makes to its run queue's load is weighted according to its
2108 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
2109 * scaled version of the new time slice allocation that they receive on time
2110 * slice expiry etc.
2111 */
2112
2113#define WEIGHT_IDLEPRIO		3
2114#define WMULT_IDLEPRIO		1431655765
2115
2116extern const int		sched_prio_to_weight[40];
2117extern const u32		sched_prio_to_wmult[40];
2118
2119/*
2120 * {de,en}queue flags:
2121 *
2122 * DEQUEUE_SLEEP  - task is no longer runnable
2123 * ENQUEUE_WAKEUP - task just became runnable
2124 *
2125 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
2126 *                are in a known state which allows modification. Such pairs
2127 *                should preserve as much state as possible.
2128 *
2129 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
2130 *        in the runqueue.
2131 *
 
 
 
 
2132 * ENQUEUE_HEAD      - place at front of runqueue (tail if not specified)
2133 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
2134 * ENQUEUE_MIGRATED  - the task was migrated during wakeup
2135 *
2136 */
2137
2138#define DEQUEUE_SLEEP		0x01
2139#define DEQUEUE_SAVE		0x02 /* Matches ENQUEUE_RESTORE */
2140#define DEQUEUE_MOVE		0x04 /* Matches ENQUEUE_MOVE */
2141#define DEQUEUE_NOCLOCK		0x08 /* Matches ENQUEUE_NOCLOCK */
 
2142
2143#define ENQUEUE_WAKEUP		0x01
2144#define ENQUEUE_RESTORE		0x02
2145#define ENQUEUE_MOVE		0x04
2146#define ENQUEUE_NOCLOCK		0x08
2147
2148#define ENQUEUE_HEAD		0x10
2149#define ENQUEUE_REPLENISH	0x20
2150#ifdef CONFIG_SMP
2151#define ENQUEUE_MIGRATED	0x40
2152#else
2153#define ENQUEUE_MIGRATED	0x00
2154#endif
 
 
2155
2156#define RETRY_TASK		((void *)-1UL)
2157
2158struct affinity_context {
2159	const struct cpumask *new_mask;
2160	struct cpumask *user_mask;
2161	unsigned int flags;
2162};
2163
 
 
2164struct sched_class {
2165
2166#ifdef CONFIG_UCLAMP_TASK
2167	int uclamp_enabled;
2168#endif
2169
2170	void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
2171	void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
2172	void (*yield_task)   (struct rq *rq);
2173	bool (*yield_to_task)(struct rq *rq, struct task_struct *p);
2174
2175	void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
2176
2177	struct task_struct *(*pick_next_task)(struct rq *rq);
2178
2179	void (*put_prev_task)(struct rq *rq, struct task_struct *p);
2180	void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
2181
2182#ifdef CONFIG_SMP
2183	int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2184	int  (*select_task_rq)(struct task_struct *p, int task_cpu, int flags);
2185
2186	struct task_struct * (*pick_task)(struct rq *rq);
2187
2188	void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
2189
2190	void (*task_woken)(struct rq *this_rq, struct task_struct *task);
2191
2192	void (*set_cpus_allowed)(struct task_struct *p, struct affinity_context *ctx);
2193
2194	void (*rq_online)(struct rq *rq);
2195	void (*rq_offline)(struct rq *rq);
2196
2197	struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq);
2198#endif
2199
2200	void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
2201	void (*task_fork)(struct task_struct *p);
2202	void (*task_dead)(struct task_struct *p);
2203
2204	/*
2205	 * The switched_from() call is allowed to drop rq->lock, therefore we
2206	 * cannot assume the switched_from/switched_to pair is serialized by
2207	 * rq->lock. They are however serialized by p->pi_lock.
2208	 */
2209	void (*switched_from)(struct rq *this_rq, struct task_struct *task);
2210	void (*switched_to)  (struct rq *this_rq, struct task_struct *task);
2211	void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
2212			      int oldprio);
2213
2214	unsigned int (*get_rr_interval)(struct rq *rq,
2215					struct task_struct *task);
2216
2217	void (*update_curr)(struct rq *rq);
2218
2219#ifdef CONFIG_FAIR_GROUP_SCHED
2220	void (*task_change_group)(struct task_struct *p);
2221#endif
 
 
 
 
2222};
2223
2224static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
2225{
2226	WARN_ON_ONCE(rq->curr != prev);
2227	prev->sched_class->put_prev_task(rq, prev);
2228}
2229
2230static inline void set_next_task(struct rq *rq, struct task_struct *next)
2231{
2232	next->sched_class->set_next_task(rq, next, false);
2233}
2234
2235
2236/*
2237 * Helper to define a sched_class instance; each one is placed in a separate
2238 * section which is ordered by the linker script:
2239 *
2240 *   include/asm-generic/vmlinux.lds.h
2241 *
2242 * *CAREFUL* they are laid out in *REVERSE* order!!!
2243 *
2244 * Also enforce alignment on the instance, not the type, to guarantee layout.
2245 */
2246#define DEFINE_SCHED_CLASS(name) \
2247const struct sched_class name##_sched_class \
2248	__aligned(__alignof__(struct sched_class)) \
2249	__section("__" #name "_sched_class")
2250
2251/* Defined in include/asm-generic/vmlinux.lds.h */
2252extern struct sched_class __sched_class_highest[];
2253extern struct sched_class __sched_class_lowest[];
2254
2255#define for_class_range(class, _from, _to) \
2256	for (class = (_from); class < (_to); class++)
2257
2258#define for_each_class(class) \
2259	for_class_range(class, __sched_class_highest, __sched_class_lowest)
2260
2261#define sched_class_above(_a, _b)	((_a) < (_b))
2262
2263extern const struct sched_class stop_sched_class;
2264extern const struct sched_class dl_sched_class;
2265extern const struct sched_class rt_sched_class;
2266extern const struct sched_class fair_sched_class;
2267extern const struct sched_class idle_sched_class;
2268
2269static inline bool sched_stop_runnable(struct rq *rq)
2270{
2271	return rq->stop && task_on_rq_queued(rq->stop);
2272}
2273
2274static inline bool sched_dl_runnable(struct rq *rq)
2275{
2276	return rq->dl.dl_nr_running > 0;
2277}
2278
2279static inline bool sched_rt_runnable(struct rq *rq)
2280{
2281	return rq->rt.rt_queued > 0;
2282}
2283
2284static inline bool sched_fair_runnable(struct rq *rq)
2285{
2286	return rq->cfs.nr_running > 0;
2287}
2288
2289extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2290extern struct task_struct *pick_next_task_idle(struct rq *rq);
2291
2292#define SCA_CHECK		0x01
2293#define SCA_MIGRATE_DISABLE	0x02
2294#define SCA_MIGRATE_ENABLE	0x04
2295#define SCA_USER		0x08
2296
2297#ifdef CONFIG_SMP
2298
2299extern void update_group_capacity(struct sched_domain *sd, int cpu);
2300
2301extern void trigger_load_balance(struct rq *rq);
2302
2303extern void set_cpus_allowed_common(struct task_struct *p, struct affinity_context *ctx);
2304
2305static inline struct task_struct *get_push_task(struct rq *rq)
2306{
2307	struct task_struct *p = rq->curr;
2308
2309	lockdep_assert_rq_held(rq);
2310
2311	if (rq->push_busy)
2312		return NULL;
2313
2314	if (p->nr_cpus_allowed == 1)
2315		return NULL;
2316
2317	if (p->migration_disabled)
2318		return NULL;
2319
2320	rq->push_busy = true;
2321	return get_task_struct(p);
2322}
2323
2324extern int push_cpu_stop(void *arg);
2325
2326#endif
2327
2328#ifdef CONFIG_CPU_IDLE
2329static inline void idle_set_state(struct rq *rq,
2330				  struct cpuidle_state *idle_state)
2331{
2332	rq->idle_state = idle_state;
2333}
2334
2335static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2336{
2337	SCHED_WARN_ON(!rcu_read_lock_held());
2338
2339	return rq->idle_state;
2340}
2341#else
2342static inline void idle_set_state(struct rq *rq,
2343				  struct cpuidle_state *idle_state)
2344{
2345}
2346
2347static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2348{
2349	return NULL;
2350}
2351#endif
2352
2353extern void schedule_idle(void);
 
2354
2355extern void sysrq_sched_debug_show(void);
2356extern void sched_init_granularity(void);
2357extern void update_max_interval(void);
2358
2359extern void init_sched_dl_class(void);
2360extern void init_sched_rt_class(void);
2361extern void init_sched_fair_class(void);
2362
2363extern void reweight_task(struct task_struct *p, int prio);
2364
2365extern void resched_curr(struct rq *rq);
2366extern void resched_cpu(int cpu);
2367
2368extern struct rt_bandwidth def_rt_bandwidth;
2369extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
2370extern bool sched_rt_bandwidth_account(struct rt_rq *rt_rq);
2371
2372extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
2373extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
2374extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
2375
2376#define BW_SHIFT		20
2377#define BW_UNIT			(1 << BW_SHIFT)
2378#define RATIO_SHIFT		8
2379#define MAX_BW_BITS		(64 - BW_SHIFT)
2380#define MAX_BW			((1ULL << MAX_BW_BITS) - 1)
2381unsigned long to_ratio(u64 period, u64 runtime);
2382
2383extern void init_entity_runnable_average(struct sched_entity *se);
2384extern void post_init_entity_util_avg(struct task_struct *p);
2385
2386#ifdef CONFIG_NO_HZ_FULL
2387extern bool sched_can_stop_tick(struct rq *rq);
2388extern int __init sched_tick_offload_init(void);
2389
2390/*
2391 * Tick may be needed by tasks in the runqueue depending on their policy and
2392 * requirements. If tick is needed, lets send the target an IPI to kick it out of
2393 * nohz mode if necessary.
2394 */
2395static inline void sched_update_tick_dependency(struct rq *rq)
2396{
2397	int cpu = cpu_of(rq);
2398
2399	if (!tick_nohz_full_cpu(cpu))
2400		return;
2401
2402	if (sched_can_stop_tick(rq))
2403		tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
2404	else
2405		tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
2406}
2407#else
2408static inline int sched_tick_offload_init(void) { return 0; }
2409static inline void sched_update_tick_dependency(struct rq *rq) { }
2410#endif
2411
2412static inline void add_nr_running(struct rq *rq, unsigned count)
2413{
2414	unsigned prev_nr = rq->nr_running;
2415
2416	rq->nr_running = prev_nr + count;
2417	if (trace_sched_update_nr_running_tp_enabled()) {
2418		call_trace_sched_update_nr_running(rq, count);
2419	}
2420
2421#ifdef CONFIG_SMP
2422	if (prev_nr < 2 && rq->nr_running >= 2) {
2423		if (!READ_ONCE(rq->rd->overload))
2424			WRITE_ONCE(rq->rd->overload, 1);
2425	}
2426#endif
2427
2428	sched_update_tick_dependency(rq);
2429}
2430
2431static inline void sub_nr_running(struct rq *rq, unsigned count)
2432{
2433	rq->nr_running -= count;
2434	if (trace_sched_update_nr_running_tp_enabled()) {
2435		call_trace_sched_update_nr_running(rq, -count);
2436	}
2437
2438	/* Check if we still need preemption */
2439	sched_update_tick_dependency(rq);
2440}
2441
2442extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
2443extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
2444
2445extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
2446
2447#ifdef CONFIG_PREEMPT_RT
2448#define SCHED_NR_MIGRATE_BREAK 8
2449#else
2450#define SCHED_NR_MIGRATE_BREAK 32
2451#endif
2452
2453extern const_debug unsigned int sysctl_sched_nr_migrate;
2454extern const_debug unsigned int sysctl_sched_migration_cost;
2455
 
 
2456#ifdef CONFIG_SCHED_DEBUG
2457extern unsigned int sysctl_sched_latency;
2458extern unsigned int sysctl_sched_min_granularity;
2459extern unsigned int sysctl_sched_idle_min_granularity;
2460extern unsigned int sysctl_sched_wakeup_granularity;
2461extern int sysctl_resched_latency_warn_ms;
2462extern int sysctl_resched_latency_warn_once;
2463
2464extern unsigned int sysctl_sched_tunable_scaling;
2465
2466extern unsigned int sysctl_numa_balancing_scan_delay;
2467extern unsigned int sysctl_numa_balancing_scan_period_min;
2468extern unsigned int sysctl_numa_balancing_scan_period_max;
2469extern unsigned int sysctl_numa_balancing_scan_size;
2470extern unsigned int sysctl_numa_balancing_hot_threshold;
2471#endif
2472
2473#ifdef CONFIG_SCHED_HRTICK
2474
2475/*
2476 * Use hrtick when:
2477 *  - enabled by features
2478 *  - hrtimer is actually high res
2479 */
2480static inline int hrtick_enabled(struct rq *rq)
2481{
2482	if (!cpu_active(cpu_of(rq)))
2483		return 0;
2484	return hrtimer_is_hres_active(&rq->hrtick_timer);
2485}
2486
2487static inline int hrtick_enabled_fair(struct rq *rq)
2488{
2489	if (!sched_feat(HRTICK))
2490		return 0;
2491	return hrtick_enabled(rq);
2492}
2493
2494static inline int hrtick_enabled_dl(struct rq *rq)
2495{
2496	if (!sched_feat(HRTICK_DL))
2497		return 0;
2498	return hrtick_enabled(rq);
2499}
2500
2501void hrtick_start(struct rq *rq, u64 delay);
2502
2503#else
2504
2505static inline int hrtick_enabled_fair(struct rq *rq)
2506{
2507	return 0;
2508}
2509
2510static inline int hrtick_enabled_dl(struct rq *rq)
2511{
2512	return 0;
2513}
2514
2515static inline int hrtick_enabled(struct rq *rq)
2516{
2517	return 0;
2518}
2519
2520#endif /* CONFIG_SCHED_HRTICK */
2521
2522#ifndef arch_scale_freq_tick
2523static __always_inline
2524void arch_scale_freq_tick(void)
2525{
2526}
2527#endif
2528
2529#ifndef arch_scale_freq_capacity
2530/**
2531 * arch_scale_freq_capacity - get the frequency scale factor of a given CPU.
2532 * @cpu: the CPU in question.
2533 *
2534 * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
2535 *
2536 *     f_curr
2537 *     ------ * SCHED_CAPACITY_SCALE
2538 *     f_max
2539 */
2540static __always_inline
2541unsigned long arch_scale_freq_capacity(int cpu)
2542{
2543	return SCHED_CAPACITY_SCALE;
2544}
2545#endif
2546
2547#ifdef CONFIG_SCHED_DEBUG
2548/*
2549 * In double_lock_balance()/double_rq_lock(), we use raw_spin_rq_lock() to
2550 * acquire rq lock instead of rq_lock(). So at the end of these two functions
2551 * we need to call double_rq_clock_clear_update() to clear RQCF_UPDATED of
2552 * rq->clock_update_flags to avoid the WARN_DOUBLE_CLOCK warning.
2553 */
2554static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2)
2555{
2556	rq1->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
2557	/* rq1 == rq2 for !CONFIG_SMP, so just clear RQCF_UPDATED once. */
2558#ifdef CONFIG_SMP
2559	rq2->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
2560#endif
2561}
2562#else
2563static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2) {}
2564#endif
2565
 
 
 
 
 
 
2566#ifdef CONFIG_SMP
2567
2568static inline bool rq_order_less(struct rq *rq1, struct rq *rq2)
2569{
2570#ifdef CONFIG_SCHED_CORE
2571	/*
2572	 * In order to not have {0,2},{1,3} turn into into an AB-BA,
2573	 * order by core-id first and cpu-id second.
2574	 *
2575	 * Notably:
2576	 *
2577	 *	double_rq_lock(0,3); will take core-0, core-1 lock
2578	 *	double_rq_lock(1,2); will take core-1, core-0 lock
2579	 *
2580	 * when only cpu-id is considered.
2581	 */
2582	if (rq1->core->cpu < rq2->core->cpu)
2583		return true;
2584	if (rq1->core->cpu > rq2->core->cpu)
2585		return false;
2586
2587	/*
2588	 * __sched_core_flip() relies on SMT having cpu-id lock order.
2589	 */
2590#endif
2591	return rq1->cpu < rq2->cpu;
2592}
2593
2594extern void double_rq_lock(struct rq *rq1, struct rq *rq2);
2595
2596#ifdef CONFIG_PREEMPTION
2597
2598/*
2599 * fair double_lock_balance: Safely acquires both rq->locks in a fair
2600 * way at the expense of forcing extra atomic operations in all
2601 * invocations.  This assures that the double_lock is acquired using the
2602 * same underlying policy as the spinlock_t on this architecture, which
2603 * reduces latency compared to the unfair variant below.  However, it
2604 * also adds more overhead and therefore may reduce throughput.
2605 */
2606static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2607	__releases(this_rq->lock)
2608	__acquires(busiest->lock)
2609	__acquires(this_rq->lock)
2610{
2611	raw_spin_rq_unlock(this_rq);
2612	double_rq_lock(this_rq, busiest);
2613
2614	return 1;
2615}
2616
2617#else
2618/*
2619 * Unfair double_lock_balance: Optimizes throughput at the expense of
2620 * latency by eliminating extra atomic operations when the locks are
2621 * already in proper order on entry.  This favors lower CPU-ids and will
2622 * grant the double lock to lower CPUs over higher ids under contention,
2623 * regardless of entry order into the function.
2624 */
2625static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2626	__releases(this_rq->lock)
2627	__acquires(busiest->lock)
2628	__acquires(this_rq->lock)
2629{
2630	if (__rq_lockp(this_rq) == __rq_lockp(busiest) ||
2631	    likely(raw_spin_rq_trylock(busiest))) {
2632		double_rq_clock_clear_update(this_rq, busiest);
2633		return 0;
2634	}
2635
2636	if (rq_order_less(this_rq, busiest)) {
2637		raw_spin_rq_lock_nested(busiest, SINGLE_DEPTH_NESTING);
2638		double_rq_clock_clear_update(this_rq, busiest);
2639		return 0;
2640	}
2641
2642	raw_spin_rq_unlock(this_rq);
2643	double_rq_lock(this_rq, busiest);
2644
2645	return 1;
2646}
2647
2648#endif /* CONFIG_PREEMPTION */
2649
2650/*
2651 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2652 */
2653static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2654{
2655	lockdep_assert_irqs_disabled();
2656
2657	return _double_lock_balance(this_rq, busiest);
2658}
2659
2660static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2661	__releases(busiest->lock)
2662{
2663	if (__rq_lockp(this_rq) != __rq_lockp(busiest))
2664		raw_spin_rq_unlock(busiest);
2665	lock_set_subclass(&__rq_lockp(this_rq)->dep_map, 0, _RET_IP_);
2666}
2667
2668static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
2669{
2670	if (l1 > l2)
2671		swap(l1, l2);
2672
2673	spin_lock(l1);
2674	spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2675}
2676
2677static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
2678{
2679	if (l1 > l2)
2680		swap(l1, l2);
2681
2682	spin_lock_irq(l1);
2683	spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2684}
2685
2686static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2687{
2688	if (l1 > l2)
2689		swap(l1, l2);
2690
2691	raw_spin_lock(l1);
2692	raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2693}
2694
 
 
 
 
 
 
 
 
 
 
2695/*
2696 * double_rq_unlock - safely unlock two runqueues
2697 *
2698 * Note this does not restore interrupts like task_rq_unlock,
2699 * you need to do so manually after calling.
2700 */
2701static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2702	__releases(rq1->lock)
2703	__releases(rq2->lock)
2704{
2705	if (__rq_lockp(rq1) != __rq_lockp(rq2))
2706		raw_spin_rq_unlock(rq2);
2707	else
2708		__release(rq2->lock);
2709	raw_spin_rq_unlock(rq1);
2710}
2711
2712extern void set_rq_online (struct rq *rq);
2713extern void set_rq_offline(struct rq *rq);
2714extern bool sched_smp_initialized;
2715
2716#else /* CONFIG_SMP */
2717
2718/*
2719 * double_rq_lock - safely lock two runqueues
2720 *
2721 * Note this does not disable interrupts like task_rq_lock,
2722 * you need to do so manually before calling.
2723 */
2724static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2725	__acquires(rq1->lock)
2726	__acquires(rq2->lock)
2727{
2728	WARN_ON_ONCE(!irqs_disabled());
2729	WARN_ON_ONCE(rq1 != rq2);
2730	raw_spin_rq_lock(rq1);
2731	__acquire(rq2->lock);	/* Fake it out ;) */
2732	double_rq_clock_clear_update(rq1, rq2);
2733}
2734
2735/*
2736 * double_rq_unlock - safely unlock two runqueues
2737 *
2738 * Note this does not restore interrupts like task_rq_unlock,
2739 * you need to do so manually after calling.
2740 */
2741static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2742	__releases(rq1->lock)
2743	__releases(rq2->lock)
2744{
2745	WARN_ON_ONCE(rq1 != rq2);
2746	raw_spin_rq_unlock(rq1);
2747	__release(rq2->lock);
2748}
2749
2750#endif
2751
 
 
 
 
 
2752extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2753extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2754
2755#ifdef	CONFIG_SCHED_DEBUG
2756extern bool sched_debug_verbose;
2757
2758extern void print_cfs_stats(struct seq_file *m, int cpu);
2759extern void print_rt_stats(struct seq_file *m, int cpu);
2760extern void print_dl_stats(struct seq_file *m, int cpu);
2761extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2762extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2763extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2764
2765extern void resched_latency_warn(int cpu, u64 latency);
2766#ifdef CONFIG_NUMA_BALANCING
2767extern void
2768show_numa_stats(struct task_struct *p, struct seq_file *m);
2769extern void
2770print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2771	unsigned long tpf, unsigned long gsf, unsigned long gpf);
2772#endif /* CONFIG_NUMA_BALANCING */
2773#else
2774static inline void resched_latency_warn(int cpu, u64 latency) {}
2775#endif /* CONFIG_SCHED_DEBUG */
2776
2777extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2778extern void init_rt_rq(struct rt_rq *rt_rq);
2779extern void init_dl_rq(struct dl_rq *dl_rq);
2780
2781extern void cfs_bandwidth_usage_inc(void);
2782extern void cfs_bandwidth_usage_dec(void);
2783
2784#ifdef CONFIG_NO_HZ_COMMON
2785#define NOHZ_BALANCE_KICK_BIT	0
2786#define NOHZ_STATS_KICK_BIT	1
2787#define NOHZ_NEWILB_KICK_BIT	2
2788#define NOHZ_NEXT_KICK_BIT	3
2789
2790/* Run rebalance_domains() */
2791#define NOHZ_BALANCE_KICK	BIT(NOHZ_BALANCE_KICK_BIT)
2792/* Update blocked load */
2793#define NOHZ_STATS_KICK		BIT(NOHZ_STATS_KICK_BIT)
2794/* Update blocked load when entering idle */
2795#define NOHZ_NEWILB_KICK	BIT(NOHZ_NEWILB_KICK_BIT)
2796/* Update nohz.next_balance */
2797#define NOHZ_NEXT_KICK		BIT(NOHZ_NEXT_KICK_BIT)
2798
2799#define NOHZ_KICK_MASK	(NOHZ_BALANCE_KICK | NOHZ_STATS_KICK | NOHZ_NEXT_KICK)
2800
2801#define nohz_flags(cpu)	(&cpu_rq(cpu)->nohz_flags)
2802
2803extern void nohz_balance_exit_idle(struct rq *rq);
2804#else
2805static inline void nohz_balance_exit_idle(struct rq *rq) { }
2806#endif
2807
2808#if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
2809extern void nohz_run_idle_balance(int cpu);
2810#else
2811static inline void nohz_run_idle_balance(int cpu) { }
2812#endif
2813
2814#ifdef CONFIG_IRQ_TIME_ACCOUNTING
2815struct irqtime {
2816	u64			total;
2817	u64			tick_delta;
2818	u64			irq_start_time;
2819	struct u64_stats_sync	sync;
2820};
2821
2822DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2823
2824/*
2825 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2826 * Otherwise ksoftirqd's sum_exec_runtime is subtracted its own runtime
2827 * and never move forward.
2828 */
2829static inline u64 irq_time_read(int cpu)
2830{
2831	struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2832	unsigned int seq;
2833	u64 total;
2834
2835	do {
2836		seq = __u64_stats_fetch_begin(&irqtime->sync);
2837		total = irqtime->total;
2838	} while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2839
2840	return total;
2841}
2842#endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2843
2844#ifdef CONFIG_CPU_FREQ
2845DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
2846
2847/**
2848 * cpufreq_update_util - Take a note about CPU utilization changes.
2849 * @rq: Runqueue to carry out the update for.
2850 * @flags: Update reason flags.
2851 *
2852 * This function is called by the scheduler on the CPU whose utilization is
2853 * being updated.
2854 *
2855 * It can only be called from RCU-sched read-side critical sections.
2856 *
2857 * The way cpufreq is currently arranged requires it to evaluate the CPU
2858 * performance state (frequency/voltage) on a regular basis to prevent it from
2859 * being stuck in a completely inadequate performance level for too long.
2860 * That is not guaranteed to happen if the updates are only triggered from CFS
2861 * and DL, though, because they may not be coming in if only RT tasks are
2862 * active all the time (or there are RT tasks only).
2863 *
2864 * As a workaround for that issue, this function is called periodically by the
2865 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2866 * but that really is a band-aid.  Going forward it should be replaced with
2867 * solutions targeted more specifically at RT tasks.
2868 */
2869static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2870{
2871	struct update_util_data *data;
2872
2873	data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2874						  cpu_of(rq)));
2875	if (data)
2876		data->func(data, rq_clock(rq), flags);
2877}
2878#else
2879static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2880#endif /* CONFIG_CPU_FREQ */
2881
2882#ifdef arch_scale_freq_capacity
2883# ifndef arch_scale_freq_invariant
2884#  define arch_scale_freq_invariant()	true
2885# endif
2886#else
2887# define arch_scale_freq_invariant()	false
2888#endif
2889
2890#ifdef CONFIG_SMP
2891static inline unsigned long capacity_orig_of(int cpu)
2892{
2893	return cpu_rq(cpu)->cpu_capacity_orig;
2894}
2895
2896/*
2897 * Returns inverted capacity if the CPU is in capacity inversion state.
2898 * 0 otherwise.
2899 *
2900 * Capacity inversion detection only considers thermal impact where actual
2901 * performance points (OPPs) gets dropped.
2902 *
2903 * Capacity inversion state happens when another performance domain that has
2904 * equal or lower capacity_orig_of() becomes effectively larger than the perf
2905 * domain this CPU belongs to due to thermal pressure throttling it hard.
2906 *
2907 * See comment in update_cpu_capacity().
2908 */
2909static inline unsigned long cpu_in_capacity_inversion(int cpu)
2910{
2911	return cpu_rq(cpu)->cpu_capacity_inverted;
2912}
2913
2914/**
2915 * enum cpu_util_type - CPU utilization type
2916 * @FREQUENCY_UTIL:	Utilization used to select frequency
2917 * @ENERGY_UTIL:	Utilization used during energy calculation
2918 *
2919 * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
2920 * need to be aggregated differently depending on the usage made of them. This
2921 * enum is used within effective_cpu_util() to differentiate the types of
2922 * utilization expected by the callers, and adjust the aggregation accordingly.
2923 */
2924enum cpu_util_type {
2925	FREQUENCY_UTIL,
2926	ENERGY_UTIL,
2927};
2928
2929unsigned long effective_cpu_util(int cpu, unsigned long util_cfs,
2930				 enum cpu_util_type type,
2931				 struct task_struct *p);
2932
2933/*
2934 * Verify the fitness of task @p to run on @cpu taking into account the
2935 * CPU original capacity and the runtime/deadline ratio of the task.
2936 *
2937 * The function will return true if the original capacity of @cpu is
2938 * greater than or equal to task's deadline density right shifted by
2939 * (BW_SHIFT - SCHED_CAPACITY_SHIFT) and false otherwise.
2940 */
2941static inline bool dl_task_fits_capacity(struct task_struct *p, int cpu)
2942{
2943	unsigned long cap = arch_scale_cpu_capacity(cpu);
2944
2945	return cap >= p->dl.dl_density >> (BW_SHIFT - SCHED_CAPACITY_SHIFT);
2946}
2947
2948static inline unsigned long cpu_bw_dl(struct rq *rq)
2949{
2950	return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2951}
2952
2953static inline unsigned long cpu_util_dl(struct rq *rq)
2954{
2955	return READ_ONCE(rq->avg_dl.util_avg);
2956}
2957
2958/**
2959 * cpu_util_cfs() - Estimates the amount of CPU capacity used by CFS tasks.
2960 * @cpu: the CPU to get the utilization for.
2961 *
2962 * The unit of the return value must be the same as the one of CPU capacity
2963 * so that CPU utilization can be compared with CPU capacity.
2964 *
2965 * CPU utilization is the sum of running time of runnable tasks plus the
2966 * recent utilization of currently non-runnable tasks on that CPU.
2967 * It represents the amount of CPU capacity currently used by CFS tasks in
2968 * the range [0..max CPU capacity] with max CPU capacity being the CPU
2969 * capacity at f_max.
2970 *
2971 * The estimated CPU utilization is defined as the maximum between CPU
2972 * utilization and sum of the estimated utilization of the currently
2973 * runnable tasks on that CPU. It preserves a utilization "snapshot" of
2974 * previously-executed tasks, which helps better deduce how busy a CPU will
2975 * be when a long-sleeping task wakes up. The contribution to CPU utilization
2976 * of such a task would be significantly decayed at this point of time.
2977 *
2978 * CPU utilization can be higher than the current CPU capacity
2979 * (f_curr/f_max * max CPU capacity) or even the max CPU capacity because
2980 * of rounding errors as well as task migrations or wakeups of new tasks.
2981 * CPU utilization has to be capped to fit into the [0..max CPU capacity]
2982 * range. Otherwise a group of CPUs (CPU0 util = 121% + CPU1 util = 80%)
2983 * could be seen as over-utilized even though CPU1 has 20% of spare CPU
2984 * capacity. CPU utilization is allowed to overshoot current CPU capacity
2985 * though since this is useful for predicting the CPU capacity required
2986 * after task migrations (scheduler-driven DVFS).
2987 *
2988 * Return: (Estimated) utilization for the specified CPU.
2989 */
2990static inline unsigned long cpu_util_cfs(int cpu)
2991{
2992	struct cfs_rq *cfs_rq;
2993	unsigned long util;
2994
2995	cfs_rq = &cpu_rq(cpu)->cfs;
2996	util = READ_ONCE(cfs_rq->avg.util_avg);
2997
2998	if (sched_feat(UTIL_EST)) {
2999		util = max_t(unsigned long, util,
3000			     READ_ONCE(cfs_rq->avg.util_est.enqueued));
3001	}
3002
3003	return min(util, capacity_orig_of(cpu));
3004}
3005
3006static inline unsigned long cpu_util_rt(struct rq *rq)
3007{
3008	return READ_ONCE(rq->avg_rt.util_avg);
3009}
3010#endif
3011
3012#ifdef CONFIG_UCLAMP_TASK
3013unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
3014
3015static inline unsigned long uclamp_rq_get(struct rq *rq,
3016					  enum uclamp_id clamp_id)
3017{
3018	return READ_ONCE(rq->uclamp[clamp_id].value);
3019}
3020
3021static inline void uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id,
3022				 unsigned int value)
3023{
3024	WRITE_ONCE(rq->uclamp[clamp_id].value, value);
3025}
3026
3027static inline bool uclamp_rq_is_idle(struct rq *rq)
3028{
3029	return rq->uclamp_flags & UCLAMP_FLAG_IDLE;
3030}
3031
3032/**
3033 * uclamp_rq_util_with - clamp @util with @rq and @p effective uclamp values.
3034 * @rq:		The rq to clamp against. Must not be NULL.
3035 * @util:	The util value to clamp.
3036 * @p:		The task to clamp against. Can be NULL if you want to clamp
3037 *		against @rq only.
3038 *
3039 * Clamps the passed @util to the max(@rq, @p) effective uclamp values.
3040 *
3041 * If sched_uclamp_used static key is disabled, then just return the util
3042 * without any clamping since uclamp aggregation at the rq level in the fast
3043 * path is disabled, rendering this operation a NOP.
3044 *
3045 * Use uclamp_eff_value() if you don't care about uclamp values at rq level. It
3046 * will return the correct effective uclamp value of the task even if the
3047 * static key is disabled.
3048 */
3049static __always_inline
3050unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
3051				  struct task_struct *p)
3052{
3053	unsigned long min_util = 0;
3054	unsigned long max_util = 0;
3055
3056	if (!static_branch_likely(&sched_uclamp_used))
3057		return util;
3058
3059	if (p) {
3060		min_util = uclamp_eff_value(p, UCLAMP_MIN);
3061		max_util = uclamp_eff_value(p, UCLAMP_MAX);
3062
3063		/*
3064		 * Ignore last runnable task's max clamp, as this task will
3065		 * reset it. Similarly, no need to read the rq's min clamp.
3066		 */
3067		if (uclamp_rq_is_idle(rq))
3068			goto out;
3069	}
3070
3071	min_util = max_t(unsigned long, min_util, uclamp_rq_get(rq, UCLAMP_MIN));
3072	max_util = max_t(unsigned long, max_util, uclamp_rq_get(rq, UCLAMP_MAX));
3073out:
3074	/*
3075	 * Since CPU's {min,max}_util clamps are MAX aggregated considering
3076	 * RUNNABLE tasks with _different_ clamps, we can end up with an
3077	 * inversion. Fix it now when the clamps are applied.
3078	 */
3079	if (unlikely(min_util >= max_util))
3080		return min_util;
3081
3082	return clamp(util, min_util, max_util);
3083}
3084
3085/* Is the rq being capped/throttled by uclamp_max? */
3086static inline bool uclamp_rq_is_capped(struct rq *rq)
3087{
3088	unsigned long rq_util;
3089	unsigned long max_util;
3090
3091	if (!static_branch_likely(&sched_uclamp_used))
3092		return false;
3093
3094	rq_util = cpu_util_cfs(cpu_of(rq)) + cpu_util_rt(rq);
3095	max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value);
3096
3097	return max_util != SCHED_CAPACITY_SCALE && rq_util >= max_util;
3098}
3099
3100/*
3101 * When uclamp is compiled in, the aggregation at rq level is 'turned off'
3102 * by default in the fast path and only gets turned on once userspace performs
3103 * an operation that requires it.
3104 *
3105 * Returns true if userspace opted-in to use uclamp and aggregation at rq level
3106 * hence is active.
3107 */
3108static inline bool uclamp_is_used(void)
3109{
3110	return static_branch_likely(&sched_uclamp_used);
3111}
3112#else /* CONFIG_UCLAMP_TASK */
3113static inline unsigned long uclamp_eff_value(struct task_struct *p,
3114					     enum uclamp_id clamp_id)
3115{
3116	if (clamp_id == UCLAMP_MIN)
3117		return 0;
3118
3119	return SCHED_CAPACITY_SCALE;
3120}
3121
3122static inline
3123unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
3124				  struct task_struct *p)
3125{
3126	return util;
3127}
3128
3129static inline bool uclamp_rq_is_capped(struct rq *rq) { return false; }
3130
3131static inline bool uclamp_is_used(void)
3132{
3133	return false;
3134}
3135
3136static inline unsigned long uclamp_rq_get(struct rq *rq,
3137					  enum uclamp_id clamp_id)
3138{
3139	if (clamp_id == UCLAMP_MIN)
3140		return 0;
3141
3142	return SCHED_CAPACITY_SCALE;
3143}
3144
3145static inline void uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id,
3146				 unsigned int value)
3147{
3148}
3149
3150static inline bool uclamp_rq_is_idle(struct rq *rq)
3151{
3152	return false;
3153}
3154#endif /* CONFIG_UCLAMP_TASK */
3155
3156#ifdef CONFIG_HAVE_SCHED_AVG_IRQ
3157static inline unsigned long cpu_util_irq(struct rq *rq)
3158{
3159	return rq->avg_irq.util_avg;
3160}
3161
3162static inline
3163unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3164{
3165	util *= (max - irq);
3166	util /= max;
3167
3168	return util;
3169
3170}
3171#else
3172static inline unsigned long cpu_util_irq(struct rq *rq)
3173{
3174	return 0;
3175}
3176
3177static inline
3178unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3179{
3180	return util;
3181}
3182#endif
3183
3184#if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
3185
3186#define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
3187
3188DECLARE_STATIC_KEY_FALSE(sched_energy_present);
3189
3190static inline bool sched_energy_enabled(void)
3191{
3192	return static_branch_unlikely(&sched_energy_present);
3193}
3194
 
 
3195#else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
3196
3197#define perf_domain_span(pd) NULL
3198static inline bool sched_energy_enabled(void) { return false; }
3199
3200#endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
3201
3202#ifdef CONFIG_MEMBARRIER
3203/*
3204 * The scheduler provides memory barriers required by membarrier between:
3205 * - prior user-space memory accesses and store to rq->membarrier_state,
3206 * - store to rq->membarrier_state and following user-space memory accesses.
3207 * In the same way it provides those guarantees around store to rq->curr.
3208 */
3209static inline void membarrier_switch_mm(struct rq *rq,
3210					struct mm_struct *prev_mm,
3211					struct mm_struct *next_mm)
3212{
3213	int membarrier_state;
3214
3215	if (prev_mm == next_mm)
3216		return;
3217
3218	membarrier_state = atomic_read(&next_mm->membarrier_state);
3219	if (READ_ONCE(rq->membarrier_state) == membarrier_state)
3220		return;
3221
3222	WRITE_ONCE(rq->membarrier_state, membarrier_state);
3223}
3224#else
3225static inline void membarrier_switch_mm(struct rq *rq,
3226					struct mm_struct *prev_mm,
3227					struct mm_struct *next_mm)
3228{
3229}
3230#endif
3231
3232#ifdef CONFIG_SMP
3233static inline bool is_per_cpu_kthread(struct task_struct *p)
3234{
3235	if (!(p->flags & PF_KTHREAD))
3236		return false;
3237
3238	if (p->nr_cpus_allowed != 1)
3239		return false;
3240
3241	return true;
3242}
3243#endif
3244
3245extern void swake_up_all_locked(struct swait_queue_head *q);
3246extern void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait);
3247
 
 
3248#ifdef CONFIG_PREEMPT_DYNAMIC
3249extern int preempt_dynamic_mode;
3250extern int sched_dynamic_mode(const char *str);
3251extern void sched_dynamic_update(int mode);
3252#endif
3253
3254static inline void update_current_exec_runtime(struct task_struct *curr,
3255						u64 now, u64 delta_exec)
 
 
 
 
 
 
 
 
 
 
 
 
3256{
3257	curr->se.sum_exec_runtime += delta_exec;
3258	account_group_exec_runtime(curr, delta_exec);
 
 
3259
3260	curr->se.exec_start = now;
3261	cgroup_account_cputime(curr, delta_exec);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3262}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3263
3264#endif /* _KERNEL_SCHED_SCHED_H */
v6.8
   1/* SPDX-License-Identifier: GPL-2.0 */
   2/*
   3 * Scheduler internal types and methods:
   4 */
   5#ifndef _KERNEL_SCHED_SCHED_H
   6#define _KERNEL_SCHED_SCHED_H
   7
   8#include <linux/sched/affinity.h>
   9#include <linux/sched/autogroup.h>
  10#include <linux/sched/cpufreq.h>
  11#include <linux/sched/deadline.h>
  12#include <linux/sched.h>
  13#include <linux/sched/loadavg.h>
  14#include <linux/sched/mm.h>
  15#include <linux/sched/rseq_api.h>
  16#include <linux/sched/signal.h>
  17#include <linux/sched/smt.h>
  18#include <linux/sched/stat.h>
  19#include <linux/sched/sysctl.h>
  20#include <linux/sched/task_flags.h>
  21#include <linux/sched/task.h>
  22#include <linux/sched/topology.h>
  23
  24#include <linux/atomic.h>
  25#include <linux/bitmap.h>
  26#include <linux/bug.h>
  27#include <linux/capability.h>
  28#include <linux/cgroup_api.h>
  29#include <linux/cgroup.h>
  30#include <linux/context_tracking.h>
  31#include <linux/cpufreq.h>
  32#include <linux/cpumask_api.h>
  33#include <linux/ctype.h>
  34#include <linux/file.h>
  35#include <linux/fs_api.h>
  36#include <linux/hrtimer_api.h>
  37#include <linux/interrupt.h>
  38#include <linux/irq_work.h>
  39#include <linux/jiffies.h>
  40#include <linux/kref_api.h>
  41#include <linux/kthread.h>
  42#include <linux/ktime_api.h>
  43#include <linux/lockdep_api.h>
  44#include <linux/lockdep.h>
  45#include <linux/minmax.h>
  46#include <linux/mm.h>
  47#include <linux/module.h>
  48#include <linux/mutex_api.h>
  49#include <linux/plist.h>
  50#include <linux/poll.h>
  51#include <linux/proc_fs.h>
  52#include <linux/profile.h>
  53#include <linux/psi.h>
  54#include <linux/rcupdate.h>
  55#include <linux/seq_file.h>
  56#include <linux/seqlock.h>
  57#include <linux/softirq.h>
  58#include <linux/spinlock_api.h>
  59#include <linux/static_key.h>
  60#include <linux/stop_machine.h>
  61#include <linux/syscalls_api.h>
  62#include <linux/syscalls.h>
  63#include <linux/tick.h>
  64#include <linux/topology.h>
  65#include <linux/types.h>
  66#include <linux/u64_stats_sync_api.h>
  67#include <linux/uaccess.h>
  68#include <linux/wait_api.h>
  69#include <linux/wait_bit.h>
  70#include <linux/workqueue_api.h>
  71
  72#include <trace/events/power.h>
  73#include <trace/events/sched.h>
  74
  75#include "../workqueue_internal.h"
  76
 
 
 
 
 
 
 
 
 
  77#ifdef CONFIG_PARAVIRT
  78# include <asm/paravirt.h>
  79# include <asm/paravirt_api_clock.h>
  80#endif
  81
  82#include "cpupri.h"
  83#include "cpudeadline.h"
  84
  85#ifdef CONFIG_SCHED_DEBUG
  86# define SCHED_WARN_ON(x)      WARN_ONCE(x, #x)
  87#else
  88# define SCHED_WARN_ON(x)      ({ (void)(x), 0; })
  89#endif
  90
  91struct rq;
  92struct cpuidle_state;
  93
  94/* task_struct::on_rq states: */
  95#define TASK_ON_RQ_QUEUED	1
  96#define TASK_ON_RQ_MIGRATING	2
  97
  98extern __read_mostly int scheduler_running;
  99
 100extern unsigned long calc_load_update;
 101extern atomic_long_t calc_load_tasks;
 102
 
 
 103extern void calc_global_load_tick(struct rq *this_rq);
 104extern long calc_load_fold_active(struct rq *this_rq, long adjust);
 105
 106extern void call_trace_sched_update_nr_running(struct rq *rq, int count);
 107
 108extern int sysctl_sched_rt_period;
 109extern int sysctl_sched_rt_runtime;
 110extern int sched_rr_timeslice;
 111
 112/*
 113 * Helpers for converting nanosecond timing to jiffy resolution
 114 */
 115#define NS_TO_JIFFIES(TIME)	((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
 116
 117/*
 118 * Increase resolution of nice-level calculations for 64-bit architectures.
 119 * The extra resolution improves shares distribution and load balancing of
 120 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
 121 * hierarchies, especially on larger systems. This is not a user-visible change
 122 * and does not change the user-interface for setting shares/weights.
 123 *
 124 * We increase resolution only if we have enough bits to allow this increased
 125 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
 126 * are pretty high and the returns do not justify the increased costs.
 127 *
 128 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
 129 * increase coverage and consistency always enable it on 64-bit platforms.
 130 */
 131#ifdef CONFIG_64BIT
 132# define NICE_0_LOAD_SHIFT	(SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
 133# define scale_load(w)		((w) << SCHED_FIXEDPOINT_SHIFT)
 134# define scale_load_down(w) \
 135({ \
 136	unsigned long __w = (w); \
 137	if (__w) \
 138		__w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \
 139	__w; \
 140})
 141#else
 142# define NICE_0_LOAD_SHIFT	(SCHED_FIXEDPOINT_SHIFT)
 143# define scale_load(w)		(w)
 144# define scale_load_down(w)	(w)
 145#endif
 146
 147/*
 148 * Task weight (visible to users) and its load (invisible to users) have
 149 * independent resolution, but they should be well calibrated. We use
 150 * scale_load() and scale_load_down(w) to convert between them. The
 151 * following must be true:
 152 *
 153 *  scale_load(sched_prio_to_weight[NICE_TO_PRIO(0)-MAX_RT_PRIO]) == NICE_0_LOAD
 154 *
 155 */
 156#define NICE_0_LOAD		(1L << NICE_0_LOAD_SHIFT)
 157
 158/*
 159 * Single value that decides SCHED_DEADLINE internal math precision.
 160 * 10 -> just above 1us
 161 * 9  -> just above 0.5us
 162 */
 163#define DL_SCALE		10
 164
 165/*
 166 * Single value that denotes runtime == period, ie unlimited time.
 167 */
 168#define RUNTIME_INF		((u64)~0ULL)
 169
 170static inline int idle_policy(int policy)
 171{
 172	return policy == SCHED_IDLE;
 173}
 174static inline int fair_policy(int policy)
 175{
 176	return policy == SCHED_NORMAL || policy == SCHED_BATCH;
 177}
 178
 179static inline int rt_policy(int policy)
 180{
 181	return policy == SCHED_FIFO || policy == SCHED_RR;
 182}
 183
 184static inline int dl_policy(int policy)
 185{
 186	return policy == SCHED_DEADLINE;
 187}
 188static inline bool valid_policy(int policy)
 189{
 190	return idle_policy(policy) || fair_policy(policy) ||
 191		rt_policy(policy) || dl_policy(policy);
 192}
 193
 194static inline int task_has_idle_policy(struct task_struct *p)
 195{
 196	return idle_policy(p->policy);
 197}
 198
 199static inline int task_has_rt_policy(struct task_struct *p)
 200{
 201	return rt_policy(p->policy);
 202}
 203
 204static inline int task_has_dl_policy(struct task_struct *p)
 205{
 206	return dl_policy(p->policy);
 207}
 208
 209#define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
 210
 211static inline void update_avg(u64 *avg, u64 sample)
 212{
 213	s64 diff = sample - *avg;
 214	*avg += diff / 8;
 215}
 216
 217/*
 218 * Shifting a value by an exponent greater *or equal* to the size of said value
 219 * is UB; cap at size-1.
 220 */
 221#define shr_bound(val, shift)							\
 222	(val >> min_t(typeof(shift), shift, BITS_PER_TYPE(typeof(val)) - 1))
 223
 224/*
 225 * !! For sched_setattr_nocheck() (kernel) only !!
 226 *
 227 * This is actually gross. :(
 228 *
 229 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
 230 * tasks, but still be able to sleep. We need this on platforms that cannot
 231 * atomically change clock frequency. Remove once fast switching will be
 232 * available on such platforms.
 233 *
 234 * SUGOV stands for SchedUtil GOVernor.
 235 */
 236#define SCHED_FLAG_SUGOV	0x10000000
 237
 238#define SCHED_DL_FLAGS (SCHED_FLAG_RECLAIM | SCHED_FLAG_DL_OVERRUN | SCHED_FLAG_SUGOV)
 239
 240static inline bool dl_entity_is_special(const struct sched_dl_entity *dl_se)
 241{
 242#ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
 243	return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
 244#else
 245	return false;
 246#endif
 247}
 248
 249/*
 250 * Tells if entity @a should preempt entity @b.
 251 */
 252static inline bool dl_entity_preempt(const struct sched_dl_entity *a,
 253				     const struct sched_dl_entity *b)
 254{
 255	return dl_entity_is_special(a) ||
 256	       dl_time_before(a->deadline, b->deadline);
 257}
 258
 259/*
 260 * This is the priority-queue data structure of the RT scheduling class:
 261 */
 262struct rt_prio_array {
 263	DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
 264	struct list_head queue[MAX_RT_PRIO];
 265};
 266
 267struct rt_bandwidth {
 268	/* nests inside the rq lock: */
 269	raw_spinlock_t		rt_runtime_lock;
 270	ktime_t			rt_period;
 271	u64			rt_runtime;
 272	struct hrtimer		rt_period_timer;
 273	unsigned int		rt_period_active;
 274};
 275
 
 
 
 
 
 
 
 
 276static inline int dl_bandwidth_enabled(void)
 277{
 278	return sysctl_sched_rt_runtime >= 0;
 279}
 280
 281/*
 282 * To keep the bandwidth of -deadline tasks under control
 283 * we need some place where:
 284 *  - store the maximum -deadline bandwidth of each cpu;
 285 *  - cache the fraction of bandwidth that is currently allocated in
 286 *    each root domain;
 287 *
 288 * This is all done in the data structure below. It is similar to the
 289 * one used for RT-throttling (rt_bandwidth), with the main difference
 290 * that, since here we are only interested in admission control, we
 291 * do not decrease any runtime while the group "executes", neither we
 292 * need a timer to replenish it.
 293 *
 294 * With respect to SMP, bandwidth is given on a per root domain basis,
 295 * meaning that:
 296 *  - bw (< 100%) is the deadline bandwidth of each CPU;
 297 *  - total_bw is the currently allocated bandwidth in each root domain;
 298 */
 299struct dl_bw {
 300	raw_spinlock_t		lock;
 301	u64			bw;
 302	u64			total_bw;
 303};
 304
 305extern void init_dl_bw(struct dl_bw *dl_b);
 306extern int  sched_dl_global_validate(void);
 307extern void sched_dl_do_global(void);
 308extern int  sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
 309extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
 310extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
 311extern bool __checkparam_dl(const struct sched_attr *attr);
 312extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
 313extern int  dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
 314extern int  dl_bw_check_overflow(int cpu);
 315
 316/*
 317 * SCHED_DEADLINE supports servers (nested scheduling) with the following
 318 * interface:
 319 *
 320 *   dl_se::rq -- runqueue we belong to.
 321 *
 322 *   dl_se::server_has_tasks() -- used on bandwidth enforcement; we 'stop' the
 323 *                                server when it runs out of tasks to run.
 324 *
 325 *   dl_se::server_pick() -- nested pick_next_task(); we yield the period if this
 326 *                           returns NULL.
 327 *
 328 *   dl_server_update() -- called from update_curr_common(), propagates runtime
 329 *                         to the server.
 330 *
 331 *   dl_server_start()
 332 *   dl_server_stop()  -- start/stop the server when it has (no) tasks.
 333 *
 334 *   dl_server_init() -- initializes the server.
 335 */
 336extern void dl_server_update(struct sched_dl_entity *dl_se, s64 delta_exec);
 337extern void dl_server_start(struct sched_dl_entity *dl_se);
 338extern void dl_server_stop(struct sched_dl_entity *dl_se);
 339extern void dl_server_init(struct sched_dl_entity *dl_se, struct rq *rq,
 340		    dl_server_has_tasks_f has_tasks,
 341		    dl_server_pick_f pick);
 342
 343#ifdef CONFIG_CGROUP_SCHED
 344
 345struct cfs_rq;
 346struct rt_rq;
 347
 348extern struct list_head task_groups;
 349
 350struct cfs_bandwidth {
 351#ifdef CONFIG_CFS_BANDWIDTH
 352	raw_spinlock_t		lock;
 353	ktime_t			period;
 354	u64			quota;
 355	u64			runtime;
 356	u64			burst;
 357	u64			runtime_snap;
 358	s64			hierarchical_quota;
 359
 360	u8			idle;
 361	u8			period_active;
 362	u8			slack_started;
 363	struct hrtimer		period_timer;
 364	struct hrtimer		slack_timer;
 365	struct list_head	throttled_cfs_rq;
 366
 367	/* Statistics: */
 368	int			nr_periods;
 369	int			nr_throttled;
 370	int			nr_burst;
 371	u64			throttled_time;
 372	u64			burst_time;
 373#endif
 374};
 375
 376/* Task group related information */
 377struct task_group {
 378	struct cgroup_subsys_state css;
 379
 380#ifdef CONFIG_FAIR_GROUP_SCHED
 381	/* schedulable entities of this group on each CPU */
 382	struct sched_entity	**se;
 383	/* runqueue "owned" by this group on each CPU */
 384	struct cfs_rq		**cfs_rq;
 385	unsigned long		shares;
 386
 387	/* A positive value indicates that this is a SCHED_IDLE group. */
 388	int			idle;
 389
 390#ifdef	CONFIG_SMP
 391	/*
 392	 * load_avg can be heavily contended at clock tick time, so put
 393	 * it in its own cacheline separated from the fields above which
 394	 * will also be accessed at each tick.
 395	 */
 396	atomic_long_t		load_avg ____cacheline_aligned;
 397#endif
 398#endif
 399
 400#ifdef CONFIG_RT_GROUP_SCHED
 401	struct sched_rt_entity	**rt_se;
 402	struct rt_rq		**rt_rq;
 403
 404	struct rt_bandwidth	rt_bandwidth;
 405#endif
 406
 407	struct rcu_head		rcu;
 408	struct list_head	list;
 409
 410	struct task_group	*parent;
 411	struct list_head	siblings;
 412	struct list_head	children;
 413
 414#ifdef CONFIG_SCHED_AUTOGROUP
 415	struct autogroup	*autogroup;
 416#endif
 417
 418	struct cfs_bandwidth	cfs_bandwidth;
 419
 420#ifdef CONFIG_UCLAMP_TASK_GROUP
 421	/* The two decimal precision [%] value requested from user-space */
 422	unsigned int		uclamp_pct[UCLAMP_CNT];
 423	/* Clamp values requested for a task group */
 424	struct uclamp_se	uclamp_req[UCLAMP_CNT];
 425	/* Effective clamp values used for a task group */
 426	struct uclamp_se	uclamp[UCLAMP_CNT];
 427#endif
 428
 429};
 430
 431#ifdef CONFIG_FAIR_GROUP_SCHED
 432#define ROOT_TASK_GROUP_LOAD	NICE_0_LOAD
 433
 434/*
 435 * A weight of 0 or 1 can cause arithmetics problems.
 436 * A weight of a cfs_rq is the sum of weights of which entities
 437 * are queued on this cfs_rq, so a weight of a entity should not be
 438 * too large, so as the shares value of a task group.
 439 * (The default weight is 1024 - so there's no practical
 440 *  limitation from this.)
 441 */
 442#define MIN_SHARES		(1UL <<  1)
 443#define MAX_SHARES		(1UL << 18)
 444#endif
 445
 446typedef int (*tg_visitor)(struct task_group *, void *);
 447
 448extern int walk_tg_tree_from(struct task_group *from,
 449			     tg_visitor down, tg_visitor up, void *data);
 450
 451/*
 452 * Iterate the full tree, calling @down when first entering a node and @up when
 453 * leaving it for the final time.
 454 *
 455 * Caller must hold rcu_lock or sufficient equivalent.
 456 */
 457static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
 458{
 459	return walk_tg_tree_from(&root_task_group, down, up, data);
 460}
 461
 462extern int tg_nop(struct task_group *tg, void *data);
 463
 464#ifdef CONFIG_FAIR_GROUP_SCHED
 465extern void free_fair_sched_group(struct task_group *tg);
 466extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
 467extern void online_fair_sched_group(struct task_group *tg);
 468extern void unregister_fair_sched_group(struct task_group *tg);
 469#else
 470static inline void free_fair_sched_group(struct task_group *tg) { }
 471static inline int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent)
 472{
 473       return 1;
 474}
 475static inline void online_fair_sched_group(struct task_group *tg) { }
 476static inline void unregister_fair_sched_group(struct task_group *tg) { }
 477#endif
 478
 479extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
 480			struct sched_entity *se, int cpu,
 481			struct sched_entity *parent);
 482extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b, struct cfs_bandwidth *parent);
 483
 484extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
 485extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
 486extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
 487extern bool cfs_task_bw_constrained(struct task_struct *p);
 488
 489extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
 490		struct sched_rt_entity *rt_se, int cpu,
 491		struct sched_rt_entity *parent);
 492extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
 493extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
 494extern long sched_group_rt_runtime(struct task_group *tg);
 495extern long sched_group_rt_period(struct task_group *tg);
 496extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
 497
 498extern struct task_group *sched_create_group(struct task_group *parent);
 499extern void sched_online_group(struct task_group *tg,
 500			       struct task_group *parent);
 501extern void sched_destroy_group(struct task_group *tg);
 502extern void sched_release_group(struct task_group *tg);
 503
 504extern void sched_move_task(struct task_struct *tsk);
 505
 506#ifdef CONFIG_FAIR_GROUP_SCHED
 507extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
 508
 509extern int sched_group_set_idle(struct task_group *tg, long idle);
 510
 511#ifdef CONFIG_SMP
 512extern void set_task_rq_fair(struct sched_entity *se,
 513			     struct cfs_rq *prev, struct cfs_rq *next);
 514#else /* !CONFIG_SMP */
 515static inline void set_task_rq_fair(struct sched_entity *se,
 516			     struct cfs_rq *prev, struct cfs_rq *next) { }
 517#endif /* CONFIG_SMP */
 518#endif /* CONFIG_FAIR_GROUP_SCHED */
 519
 520#else /* CONFIG_CGROUP_SCHED */
 521
 522struct cfs_bandwidth { };
 523static inline bool cfs_task_bw_constrained(struct task_struct *p) { return false; }
 524
 525#endif	/* CONFIG_CGROUP_SCHED */
 526
 527extern void unregister_rt_sched_group(struct task_group *tg);
 528extern void free_rt_sched_group(struct task_group *tg);
 529extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
 530
 531/*
 532 * u64_u32_load/u64_u32_store
 533 *
 534 * Use a copy of a u64 value to protect against data race. This is only
 535 * applicable for 32-bits architectures.
 536 */
 537#ifdef CONFIG_64BIT
 538# define u64_u32_load_copy(var, copy)       var
 539# define u64_u32_store_copy(var, copy, val) (var = val)
 540#else
 541# define u64_u32_load_copy(var, copy)					\
 542({									\
 543	u64 __val, __val_copy;						\
 544	do {								\
 545		__val_copy = copy;					\
 546		/*							\
 547		 * paired with u64_u32_store_copy(), ordering access	\
 548		 * to var and copy.					\
 549		 */							\
 550		smp_rmb();						\
 551		__val = var;						\
 552	} while (__val != __val_copy);					\
 553	__val;								\
 554})
 555# define u64_u32_store_copy(var, copy, val)				\
 556do {									\
 557	typeof(val) __val = (val);					\
 558	var = __val;							\
 559	/*								\
 560	 * paired with u64_u32_load_copy(), ordering access to var and	\
 561	 * copy.							\
 562	 */								\
 563	smp_wmb();							\
 564	copy = __val;							\
 565} while (0)
 566#endif
 567# define u64_u32_load(var)      u64_u32_load_copy(var, var##_copy)
 568# define u64_u32_store(var, val) u64_u32_store_copy(var, var##_copy, val)
 569
 570/* CFS-related fields in a runqueue */
 571struct cfs_rq {
 572	struct load_weight	load;
 573	unsigned int		nr_running;
 574	unsigned int		h_nr_running;      /* SCHED_{NORMAL,BATCH,IDLE} */
 575	unsigned int		idle_nr_running;   /* SCHED_IDLE */
 576	unsigned int		idle_h_nr_running; /* SCHED_IDLE */
 577
 578	s64			avg_vruntime;
 579	u64			avg_load;
 580
 581	u64			exec_clock;
 582	u64			min_vruntime;
 583#ifdef CONFIG_SCHED_CORE
 584	unsigned int		forceidle_seq;
 585	u64			min_vruntime_fi;
 586#endif
 587
 588#ifndef CONFIG_64BIT
 589	u64			min_vruntime_copy;
 590#endif
 591
 592	struct rb_root_cached	tasks_timeline;
 593
 594	/*
 595	 * 'curr' points to currently running entity on this cfs_rq.
 596	 * It is set to NULL otherwise (i.e when none are currently running).
 597	 */
 598	struct sched_entity	*curr;
 599	struct sched_entity	*next;
 
 
 600
 601#ifdef	CONFIG_SCHED_DEBUG
 602	unsigned int		nr_spread_over;
 603#endif
 604
 605#ifdef CONFIG_SMP
 606	/*
 607	 * CFS load tracking
 608	 */
 609	struct sched_avg	avg;
 610#ifndef CONFIG_64BIT
 611	u64			last_update_time_copy;
 612#endif
 613	struct {
 614		raw_spinlock_t	lock ____cacheline_aligned;
 615		int		nr;
 616		unsigned long	load_avg;
 617		unsigned long	util_avg;
 618		unsigned long	runnable_avg;
 619	} removed;
 620
 621#ifdef CONFIG_FAIR_GROUP_SCHED
 622	u64			last_update_tg_load_avg;
 623	unsigned long		tg_load_avg_contrib;
 624	long			propagate;
 625	long			prop_runnable_sum;
 626
 627	/*
 628	 *   h_load = weight * f(tg)
 629	 *
 630	 * Where f(tg) is the recursive weight fraction assigned to
 631	 * this group.
 632	 */
 633	unsigned long		h_load;
 634	u64			last_h_load_update;
 635	struct sched_entity	*h_load_next;
 636#endif /* CONFIG_FAIR_GROUP_SCHED */
 637#endif /* CONFIG_SMP */
 638
 639#ifdef CONFIG_FAIR_GROUP_SCHED
 640	struct rq		*rq;	/* CPU runqueue to which this cfs_rq is attached */
 641
 642	/*
 643	 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
 644	 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
 645	 * (like users, containers etc.)
 646	 *
 647	 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
 648	 * This list is used during load balance.
 649	 */
 650	int			on_list;
 651	struct list_head	leaf_cfs_rq_list;
 652	struct task_group	*tg;	/* group that "owns" this runqueue */
 653
 654	/* Locally cached copy of our task_group's idle value */
 655	int			idle;
 656
 657#ifdef CONFIG_CFS_BANDWIDTH
 658	int			runtime_enabled;
 659	s64			runtime_remaining;
 660
 661	u64			throttled_pelt_idle;
 662#ifndef CONFIG_64BIT
 663	u64                     throttled_pelt_idle_copy;
 664#endif
 665	u64			throttled_clock;
 666	u64			throttled_clock_pelt;
 667	u64			throttled_clock_pelt_time;
 668	u64			throttled_clock_self;
 669	u64			throttled_clock_self_time;
 670	int			throttled;
 671	int			throttle_count;
 672	struct list_head	throttled_list;
 673	struct list_head	throttled_csd_list;
 674#endif /* CONFIG_CFS_BANDWIDTH */
 675#endif /* CONFIG_FAIR_GROUP_SCHED */
 676};
 677
 678static inline int rt_bandwidth_enabled(void)
 679{
 680	return sysctl_sched_rt_runtime >= 0;
 681}
 682
 683/* RT IPI pull logic requires IRQ_WORK */
 684#if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
 685# define HAVE_RT_PUSH_IPI
 686#endif
 687
 688/* Real-Time classes' related field in a runqueue: */
 689struct rt_rq {
 690	struct rt_prio_array	active;
 691	unsigned int		rt_nr_running;
 692	unsigned int		rr_nr_running;
 693#if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
 694	struct {
 695		int		curr; /* highest queued rt task prio */
 696#ifdef CONFIG_SMP
 697		int		next; /* next highest */
 698#endif
 699	} highest_prio;
 700#endif
 701#ifdef CONFIG_SMP
 
 
 702	int			overloaded;
 703	struct plist_head	pushable_tasks;
 704
 705#endif /* CONFIG_SMP */
 706	int			rt_queued;
 707
 708	int			rt_throttled;
 709	u64			rt_time;
 710	u64			rt_runtime;
 711	/* Nests inside the rq lock: */
 712	raw_spinlock_t		rt_runtime_lock;
 713
 714#ifdef CONFIG_RT_GROUP_SCHED
 715	unsigned int		rt_nr_boosted;
 716
 717	struct rq		*rq;
 718	struct task_group	*tg;
 719#endif
 720};
 721
 722static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
 723{
 724	return rt_rq->rt_queued && rt_rq->rt_nr_running;
 725}
 726
 727/* Deadline class' related fields in a runqueue */
 728struct dl_rq {
 729	/* runqueue is an rbtree, ordered by deadline */
 730	struct rb_root_cached	root;
 731
 732	unsigned int		dl_nr_running;
 733
 734#ifdef CONFIG_SMP
 735	/*
 736	 * Deadline values of the currently executing and the
 737	 * earliest ready task on this rq. Caching these facilitates
 738	 * the decision whether or not a ready but not running task
 739	 * should migrate somewhere else.
 740	 */
 741	struct {
 742		u64		curr;
 743		u64		next;
 744	} earliest_dl;
 745
 
 746	int			overloaded;
 747
 748	/*
 749	 * Tasks on this rq that can be pushed away. They are kept in
 750	 * an rb-tree, ordered by tasks' deadlines, with caching
 751	 * of the leftmost (earliest deadline) element.
 752	 */
 753	struct rb_root_cached	pushable_dl_tasks_root;
 754#else
 755	struct dl_bw		dl_bw;
 756#endif
 757	/*
 758	 * "Active utilization" for this runqueue: increased when a
 759	 * task wakes up (becomes TASK_RUNNING) and decreased when a
 760	 * task blocks
 761	 */
 762	u64			running_bw;
 763
 764	/*
 765	 * Utilization of the tasks "assigned" to this runqueue (including
 766	 * the tasks that are in runqueue and the tasks that executed on this
 767	 * CPU and blocked). Increased when a task moves to this runqueue, and
 768	 * decreased when the task moves away (migrates, changes scheduling
 769	 * policy, or terminates).
 770	 * This is needed to compute the "inactive utilization" for the
 771	 * runqueue (inactive utilization = this_bw - running_bw).
 772	 */
 773	u64			this_bw;
 774	u64			extra_bw;
 775
 776	/*
 777	 * Maximum available bandwidth for reclaiming by SCHED_FLAG_RECLAIM
 778	 * tasks of this rq. Used in calculation of reclaimable bandwidth(GRUB).
 779	 */
 780	u64			max_bw;
 781
 782	/*
 783	 * Inverse of the fraction of CPU utilization that can be reclaimed
 784	 * by the GRUB algorithm.
 785	 */
 786	u64			bw_ratio;
 787};
 788
 789#ifdef CONFIG_FAIR_GROUP_SCHED
 790/* An entity is a task if it doesn't "own" a runqueue */
 791#define entity_is_task(se)	(!se->my_q)
 792
 793static inline void se_update_runnable(struct sched_entity *se)
 794{
 795	if (!entity_is_task(se))
 796		se->runnable_weight = se->my_q->h_nr_running;
 797}
 798
 799static inline long se_runnable(struct sched_entity *se)
 800{
 801	if (entity_is_task(se))
 802		return !!se->on_rq;
 803	else
 804		return se->runnable_weight;
 805}
 806
 807#else
 808#define entity_is_task(se)	1
 809
 810static inline void se_update_runnable(struct sched_entity *se) {}
 811
 812static inline long se_runnable(struct sched_entity *se)
 813{
 814	return !!se->on_rq;
 815}
 816#endif
 817
 818#ifdef CONFIG_SMP
 819/*
 820 * XXX we want to get rid of these helpers and use the full load resolution.
 821 */
 822static inline long se_weight(struct sched_entity *se)
 823{
 824	return scale_load_down(se->load.weight);
 825}
 826
 827
 828static inline bool sched_asym_prefer(int a, int b)
 829{
 830	return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
 831}
 832
 833struct perf_domain {
 834	struct em_perf_domain *em_pd;
 835	struct perf_domain *next;
 836	struct rcu_head rcu;
 837};
 838
 839/* Scheduling group status flags */
 840#define SG_OVERLOAD		0x1 /* More than one runnable task on a CPU. */
 841#define SG_OVERUTILIZED		0x2 /* One or more CPUs are over-utilized. */
 842
 843/*
 844 * We add the notion of a root-domain which will be used to define per-domain
 845 * variables. Each exclusive cpuset essentially defines an island domain by
 846 * fully partitioning the member CPUs from any other cpuset. Whenever a new
 847 * exclusive cpuset is created, we also create and attach a new root-domain
 848 * object.
 849 *
 850 */
 851struct root_domain {
 852	atomic_t		refcount;
 853	atomic_t		rto_count;
 854	struct rcu_head		rcu;
 855	cpumask_var_t		span;
 856	cpumask_var_t		online;
 857
 858	/*
 859	 * Indicate pullable load on at least one CPU, e.g:
 860	 * - More than one runnable task
 861	 * - Running task is misfit
 862	 */
 863	int			overload;
 864
 865	/* Indicate one or more cpus over-utilized (tipping point) */
 866	int			overutilized;
 867
 868	/*
 869	 * The bit corresponding to a CPU gets set here if such CPU has more
 870	 * than one runnable -deadline task (as it is below for RT tasks).
 871	 */
 872	cpumask_var_t		dlo_mask;
 873	atomic_t		dlo_count;
 874	struct dl_bw		dl_bw;
 875	struct cpudl		cpudl;
 876
 877	/*
 878	 * Indicate whether a root_domain's dl_bw has been checked or
 879	 * updated. It's monotonously increasing value.
 880	 *
 881	 * Also, some corner cases, like 'wrap around' is dangerous, but given
 882	 * that u64 is 'big enough'. So that shouldn't be a concern.
 883	 */
 884	u64 visit_gen;
 885
 886#ifdef HAVE_RT_PUSH_IPI
 887	/*
 888	 * For IPI pull requests, loop across the rto_mask.
 889	 */
 890	struct irq_work		rto_push_work;
 891	raw_spinlock_t		rto_lock;
 892	/* These are only updated and read within rto_lock */
 893	int			rto_loop;
 894	int			rto_cpu;
 895	/* These atomics are updated outside of a lock */
 896	atomic_t		rto_loop_next;
 897	atomic_t		rto_loop_start;
 898#endif
 899	/*
 900	 * The "RT overload" flag: it gets set if a CPU has more than
 901	 * one runnable RT task.
 902	 */
 903	cpumask_var_t		rto_mask;
 904	struct cpupri		cpupri;
 905
 906	unsigned long		max_cpu_capacity;
 907
 908	/*
 909	 * NULL-terminated list of performance domains intersecting with the
 910	 * CPUs of the rd. Protected by RCU.
 911	 */
 912	struct perf_domain __rcu *pd;
 913};
 914
 915extern void init_defrootdomain(void);
 916extern int sched_init_domains(const struct cpumask *cpu_map);
 917extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
 918extern void sched_get_rd(struct root_domain *rd);
 919extern void sched_put_rd(struct root_domain *rd);
 920
 921#ifdef HAVE_RT_PUSH_IPI
 922extern void rto_push_irq_work_func(struct irq_work *work);
 923#endif
 924#endif /* CONFIG_SMP */
 925
 926#ifdef CONFIG_UCLAMP_TASK
 927/*
 928 * struct uclamp_bucket - Utilization clamp bucket
 929 * @value: utilization clamp value for tasks on this clamp bucket
 930 * @tasks: number of RUNNABLE tasks on this clamp bucket
 931 *
 932 * Keep track of how many tasks are RUNNABLE for a given utilization
 933 * clamp value.
 934 */
 935struct uclamp_bucket {
 936	unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
 937	unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
 938};
 939
 940/*
 941 * struct uclamp_rq - rq's utilization clamp
 942 * @value: currently active clamp values for a rq
 943 * @bucket: utilization clamp buckets affecting a rq
 944 *
 945 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
 946 * A clamp value is affecting a rq when there is at least one task RUNNABLE
 947 * (or actually running) with that value.
 948 *
 949 * There are up to UCLAMP_CNT possible different clamp values, currently there
 950 * are only two: minimum utilization and maximum utilization.
 951 *
 952 * All utilization clamping values are MAX aggregated, since:
 953 * - for util_min: we want to run the CPU at least at the max of the minimum
 954 *   utilization required by its currently RUNNABLE tasks.
 955 * - for util_max: we want to allow the CPU to run up to the max of the
 956 *   maximum utilization allowed by its currently RUNNABLE tasks.
 957 *
 958 * Since on each system we expect only a limited number of different
 959 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
 960 * the metrics required to compute all the per-rq utilization clamp values.
 961 */
 962struct uclamp_rq {
 963	unsigned int value;
 964	struct uclamp_bucket bucket[UCLAMP_BUCKETS];
 965};
 966
 967DECLARE_STATIC_KEY_FALSE(sched_uclamp_used);
 968#endif /* CONFIG_UCLAMP_TASK */
 969
 970struct rq;
 971struct balance_callback {
 972	struct balance_callback *next;
 973	void (*func)(struct rq *rq);
 974};
 975
 976/*
 977 * This is the main, per-CPU runqueue data structure.
 978 *
 979 * Locking rule: those places that want to lock multiple runqueues
 980 * (such as the load balancing or the thread migration code), lock
 981 * acquire operations must be ordered by ascending &runqueue.
 982 */
 983struct rq {
 984	/* runqueue lock: */
 985	raw_spinlock_t		__lock;
 986
 
 
 
 
 987	unsigned int		nr_running;
 988#ifdef CONFIG_NUMA_BALANCING
 989	unsigned int		nr_numa_running;
 990	unsigned int		nr_preferred_running;
 991	unsigned int		numa_migrate_on;
 992#endif
 993#ifdef CONFIG_NO_HZ_COMMON
 994#ifdef CONFIG_SMP
 995	unsigned long		last_blocked_load_update_tick;
 996	unsigned int		has_blocked_load;
 997	call_single_data_t	nohz_csd;
 998#endif /* CONFIG_SMP */
 999	unsigned int		nohz_tick_stopped;
1000	atomic_t		nohz_flags;
1001#endif /* CONFIG_NO_HZ_COMMON */
1002
1003#ifdef CONFIG_SMP
1004	unsigned int		ttwu_pending;
1005#endif
1006	u64			nr_switches;
1007
1008#ifdef CONFIG_UCLAMP_TASK
1009	/* Utilization clamp values based on CPU's RUNNABLE tasks */
1010	struct uclamp_rq	uclamp[UCLAMP_CNT] ____cacheline_aligned;
1011	unsigned int		uclamp_flags;
1012#define UCLAMP_FLAG_IDLE 0x01
1013#endif
1014
1015	struct cfs_rq		cfs;
1016	struct rt_rq		rt;
1017	struct dl_rq		dl;
1018
1019#ifdef CONFIG_FAIR_GROUP_SCHED
1020	/* list of leaf cfs_rq on this CPU: */
1021	struct list_head	leaf_cfs_rq_list;
1022	struct list_head	*tmp_alone_branch;
1023#endif /* CONFIG_FAIR_GROUP_SCHED */
1024
1025	/*
1026	 * This is part of a global counter where only the total sum
1027	 * over all CPUs matters. A task can increase this counter on
1028	 * one CPU and if it got migrated afterwards it may decrease
1029	 * it on another CPU. Always updated under the runqueue lock:
1030	 */
1031	unsigned int		nr_uninterruptible;
1032
1033	struct task_struct __rcu	*curr;
1034	struct task_struct	*idle;
1035	struct task_struct	*stop;
1036	unsigned long		next_balance;
1037	struct mm_struct	*prev_mm;
1038
1039	unsigned int		clock_update_flags;
1040	u64			clock;
1041	/* Ensure that all clocks are in the same cache line */
1042	u64			clock_task ____cacheline_aligned;
1043	u64			clock_pelt;
1044	unsigned long		lost_idle_time;
1045	u64			clock_pelt_idle;
1046	u64			clock_idle;
1047#ifndef CONFIG_64BIT
1048	u64			clock_pelt_idle_copy;
1049	u64			clock_idle_copy;
1050#endif
1051
1052	atomic_t		nr_iowait;
1053
1054#ifdef CONFIG_SCHED_DEBUG
1055	u64 last_seen_need_resched_ns;
1056	int ticks_without_resched;
1057#endif
1058
1059#ifdef CONFIG_MEMBARRIER
1060	int membarrier_state;
1061#endif
1062
1063#ifdef CONFIG_SMP
1064	struct root_domain		*rd;
1065	struct sched_domain __rcu	*sd;
1066
1067	unsigned long		cpu_capacity;
 
 
1068
1069	struct balance_callback *balance_callback;
1070
1071	unsigned char		nohz_idle_balance;
1072	unsigned char		idle_balance;
1073
1074	unsigned long		misfit_task_load;
1075
1076	/* For active balancing */
1077	int			active_balance;
1078	int			push_cpu;
1079	struct cpu_stop_work	active_balance_work;
1080
1081	/* CPU of this runqueue: */
1082	int			cpu;
1083	int			online;
1084
1085	struct list_head cfs_tasks;
1086
1087	struct sched_avg	avg_rt;
1088	struct sched_avg	avg_dl;
1089#ifdef CONFIG_HAVE_SCHED_AVG_IRQ
1090	struct sched_avg	avg_irq;
1091#endif
1092#ifdef CONFIG_SCHED_THERMAL_PRESSURE
1093	struct sched_avg	avg_thermal;
1094#endif
1095	u64			idle_stamp;
1096	u64			avg_idle;
1097
 
 
 
1098	/* This is used to determine avg_idle's max value */
1099	u64			max_idle_balance_cost;
1100
1101#ifdef CONFIG_HOTPLUG_CPU
1102	struct rcuwait		hotplug_wait;
1103#endif
1104#endif /* CONFIG_SMP */
1105
1106#ifdef CONFIG_IRQ_TIME_ACCOUNTING
1107	u64			prev_irq_time;
1108#endif
1109#ifdef CONFIG_PARAVIRT
1110	u64			prev_steal_time;
1111#endif
1112#ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
1113	u64			prev_steal_time_rq;
1114#endif
1115
1116	/* calc_load related fields */
1117	unsigned long		calc_load_update;
1118	long			calc_load_active;
1119
1120#ifdef CONFIG_SCHED_HRTICK
1121#ifdef CONFIG_SMP
1122	call_single_data_t	hrtick_csd;
1123#endif
1124	struct hrtimer		hrtick_timer;
1125	ktime_t 		hrtick_time;
1126#endif
1127
1128#ifdef CONFIG_SCHEDSTATS
1129	/* latency stats */
1130	struct sched_info	rq_sched_info;
1131	unsigned long long	rq_cpu_time;
1132	/* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
1133
1134	/* sys_sched_yield() stats */
1135	unsigned int		yld_count;
1136
1137	/* schedule() stats */
1138	unsigned int		sched_count;
1139	unsigned int		sched_goidle;
1140
1141	/* try_to_wake_up() stats */
1142	unsigned int		ttwu_count;
1143	unsigned int		ttwu_local;
1144#endif
1145
1146#ifdef CONFIG_CPU_IDLE
1147	/* Must be inspected within a rcu lock section */
1148	struct cpuidle_state	*idle_state;
1149#endif
1150
1151#ifdef CONFIG_SMP
1152	unsigned int		nr_pinned;
1153#endif
1154	unsigned int		push_busy;
1155	struct cpu_stop_work	push_work;
1156
1157#ifdef CONFIG_SCHED_CORE
1158	/* per rq */
1159	struct rq		*core;
1160	struct task_struct	*core_pick;
1161	unsigned int		core_enabled;
1162	unsigned int		core_sched_seq;
1163	struct rb_root		core_tree;
1164
1165	/* shared state -- careful with sched_core_cpu_deactivate() */
1166	unsigned int		core_task_seq;
1167	unsigned int		core_pick_seq;
1168	unsigned long		core_cookie;
1169	unsigned int		core_forceidle_count;
1170	unsigned int		core_forceidle_seq;
1171	unsigned int		core_forceidle_occupation;
1172	u64			core_forceidle_start;
1173#endif
1174
1175	/* Scratch cpumask to be temporarily used under rq_lock */
1176	cpumask_var_t		scratch_mask;
1177
1178#if defined(CONFIG_CFS_BANDWIDTH) && defined(CONFIG_SMP)
1179	call_single_data_t	cfsb_csd;
1180	struct list_head	cfsb_csd_list;
1181#endif
1182};
1183
1184#ifdef CONFIG_FAIR_GROUP_SCHED
1185
1186/* CPU runqueue to which this cfs_rq is attached */
1187static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1188{
1189	return cfs_rq->rq;
1190}
1191
1192#else
1193
1194static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1195{
1196	return container_of(cfs_rq, struct rq, cfs);
1197}
1198#endif
1199
1200static inline int cpu_of(struct rq *rq)
1201{
1202#ifdef CONFIG_SMP
1203	return rq->cpu;
1204#else
1205	return 0;
1206#endif
1207}
1208
1209#define MDF_PUSH	0x01
1210
1211static inline bool is_migration_disabled(struct task_struct *p)
1212{
1213#ifdef CONFIG_SMP
1214	return p->migration_disabled;
1215#else
1216	return false;
1217#endif
1218}
1219
1220DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1221
1222#define cpu_rq(cpu)		(&per_cpu(runqueues, (cpu)))
1223#define this_rq()		this_cpu_ptr(&runqueues)
1224#define task_rq(p)		cpu_rq(task_cpu(p))
1225#define cpu_curr(cpu)		(cpu_rq(cpu)->curr)
1226#define raw_rq()		raw_cpu_ptr(&runqueues)
1227
1228struct sched_group;
1229#ifdef CONFIG_SCHED_CORE
1230static inline struct cpumask *sched_group_span(struct sched_group *sg);
1231
1232DECLARE_STATIC_KEY_FALSE(__sched_core_enabled);
1233
1234static inline bool sched_core_enabled(struct rq *rq)
1235{
1236	return static_branch_unlikely(&__sched_core_enabled) && rq->core_enabled;
1237}
1238
1239static inline bool sched_core_disabled(void)
1240{
1241	return !static_branch_unlikely(&__sched_core_enabled);
1242}
1243
1244/*
1245 * Be careful with this function; not for general use. The return value isn't
1246 * stable unless you actually hold a relevant rq->__lock.
1247 */
1248static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1249{
1250	if (sched_core_enabled(rq))
1251		return &rq->core->__lock;
1252
1253	return &rq->__lock;
1254}
1255
1256static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1257{
1258	if (rq->core_enabled)
1259		return &rq->core->__lock;
1260
1261	return &rq->__lock;
1262}
1263
1264bool cfs_prio_less(const struct task_struct *a, const struct task_struct *b,
1265			bool fi);
1266void task_vruntime_update(struct rq *rq, struct task_struct *p, bool in_fi);
1267
1268/*
1269 * Helpers to check if the CPU's core cookie matches with the task's cookie
1270 * when core scheduling is enabled.
1271 * A special case is that the task's cookie always matches with CPU's core
1272 * cookie if the CPU is in an idle core.
1273 */
1274static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1275{
1276	/* Ignore cookie match if core scheduler is not enabled on the CPU. */
1277	if (!sched_core_enabled(rq))
1278		return true;
1279
1280	return rq->core->core_cookie == p->core_cookie;
1281}
1282
1283static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1284{
1285	bool idle_core = true;
1286	int cpu;
1287
1288	/* Ignore cookie match if core scheduler is not enabled on the CPU. */
1289	if (!sched_core_enabled(rq))
1290		return true;
1291
1292	for_each_cpu(cpu, cpu_smt_mask(cpu_of(rq))) {
1293		if (!available_idle_cpu(cpu)) {
1294			idle_core = false;
1295			break;
1296		}
1297	}
1298
1299	/*
1300	 * A CPU in an idle core is always the best choice for tasks with
1301	 * cookies.
1302	 */
1303	return idle_core || rq->core->core_cookie == p->core_cookie;
1304}
1305
1306static inline bool sched_group_cookie_match(struct rq *rq,
1307					    struct task_struct *p,
1308					    struct sched_group *group)
1309{
1310	int cpu;
1311
1312	/* Ignore cookie match if core scheduler is not enabled on the CPU. */
1313	if (!sched_core_enabled(rq))
1314		return true;
1315
1316	for_each_cpu_and(cpu, sched_group_span(group), p->cpus_ptr) {
1317		if (sched_core_cookie_match(cpu_rq(cpu), p))
1318			return true;
1319	}
1320	return false;
1321}
1322
1323static inline bool sched_core_enqueued(struct task_struct *p)
1324{
1325	return !RB_EMPTY_NODE(&p->core_node);
1326}
1327
1328extern void sched_core_enqueue(struct rq *rq, struct task_struct *p);
1329extern void sched_core_dequeue(struct rq *rq, struct task_struct *p, int flags);
1330
1331extern void sched_core_get(void);
1332extern void sched_core_put(void);
1333
1334#else /* !CONFIG_SCHED_CORE */
1335
1336static inline bool sched_core_enabled(struct rq *rq)
1337{
1338	return false;
1339}
1340
1341static inline bool sched_core_disabled(void)
1342{
1343	return true;
1344}
1345
1346static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1347{
1348	return &rq->__lock;
1349}
1350
1351static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1352{
1353	return &rq->__lock;
1354}
1355
1356static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1357{
1358	return true;
1359}
1360
1361static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1362{
1363	return true;
1364}
1365
1366static inline bool sched_group_cookie_match(struct rq *rq,
1367					    struct task_struct *p,
1368					    struct sched_group *group)
1369{
1370	return true;
1371}
1372#endif /* CONFIG_SCHED_CORE */
1373
1374static inline void lockdep_assert_rq_held(struct rq *rq)
1375{
1376	lockdep_assert_held(__rq_lockp(rq));
1377}
1378
1379extern void raw_spin_rq_lock_nested(struct rq *rq, int subclass);
1380extern bool raw_spin_rq_trylock(struct rq *rq);
1381extern void raw_spin_rq_unlock(struct rq *rq);
1382
1383static inline void raw_spin_rq_lock(struct rq *rq)
1384{
1385	raw_spin_rq_lock_nested(rq, 0);
1386}
1387
1388static inline void raw_spin_rq_lock_irq(struct rq *rq)
1389{
1390	local_irq_disable();
1391	raw_spin_rq_lock(rq);
1392}
1393
1394static inline void raw_spin_rq_unlock_irq(struct rq *rq)
1395{
1396	raw_spin_rq_unlock(rq);
1397	local_irq_enable();
1398}
1399
1400static inline unsigned long _raw_spin_rq_lock_irqsave(struct rq *rq)
1401{
1402	unsigned long flags;
1403	local_irq_save(flags);
1404	raw_spin_rq_lock(rq);
1405	return flags;
1406}
1407
1408static inline void raw_spin_rq_unlock_irqrestore(struct rq *rq, unsigned long flags)
1409{
1410	raw_spin_rq_unlock(rq);
1411	local_irq_restore(flags);
1412}
1413
1414#define raw_spin_rq_lock_irqsave(rq, flags)	\
1415do {						\
1416	flags = _raw_spin_rq_lock_irqsave(rq);	\
1417} while (0)
1418
1419#ifdef CONFIG_SCHED_SMT
1420extern void __update_idle_core(struct rq *rq);
1421
1422static inline void update_idle_core(struct rq *rq)
1423{
1424	if (static_branch_unlikely(&sched_smt_present))
1425		__update_idle_core(rq);
1426}
1427
1428#else
1429static inline void update_idle_core(struct rq *rq) { }
1430#endif
1431
1432#ifdef CONFIG_FAIR_GROUP_SCHED
1433static inline struct task_struct *task_of(struct sched_entity *se)
1434{
1435	SCHED_WARN_ON(!entity_is_task(se));
1436	return container_of(se, struct task_struct, se);
1437}
1438
1439static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1440{
1441	return p->se.cfs_rq;
1442}
1443
1444/* runqueue on which this entity is (to be) queued */
1445static inline struct cfs_rq *cfs_rq_of(const struct sched_entity *se)
1446{
1447	return se->cfs_rq;
1448}
1449
1450/* runqueue "owned" by this group */
1451static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1452{
1453	return grp->my_q;
1454}
1455
1456#else
1457
1458#define task_of(_se)	container_of(_se, struct task_struct, se)
 
 
 
1459
1460static inline struct cfs_rq *task_cfs_rq(const struct task_struct *p)
1461{
1462	return &task_rq(p)->cfs;
1463}
1464
1465static inline struct cfs_rq *cfs_rq_of(const struct sched_entity *se)
1466{
1467	const struct task_struct *p = task_of(se);
1468	struct rq *rq = task_rq(p);
1469
1470	return &rq->cfs;
1471}
1472
1473/* runqueue "owned" by this group */
1474static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1475{
1476	return NULL;
1477}
1478#endif
1479
1480extern void update_rq_clock(struct rq *rq);
1481
1482/*
1483 * rq::clock_update_flags bits
1484 *
1485 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1486 *  call to __schedule(). This is an optimisation to avoid
1487 *  neighbouring rq clock updates.
1488 *
1489 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1490 *  in effect and calls to update_rq_clock() are being ignored.
1491 *
1492 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1493 *  made to update_rq_clock() since the last time rq::lock was pinned.
1494 *
1495 * If inside of __schedule(), clock_update_flags will have been
1496 * shifted left (a left shift is a cheap operation for the fast path
1497 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1498 *
1499 *	if (rq-clock_update_flags >= RQCF_UPDATED)
1500 *
1501 * to check if %RQCF_UPDATED is set. It'll never be shifted more than
1502 * one position though, because the next rq_unpin_lock() will shift it
1503 * back.
1504 */
1505#define RQCF_REQ_SKIP		0x01
1506#define RQCF_ACT_SKIP		0x02
1507#define RQCF_UPDATED		0x04
1508
1509static inline void assert_clock_updated(struct rq *rq)
1510{
1511	/*
1512	 * The only reason for not seeing a clock update since the
1513	 * last rq_pin_lock() is if we're currently skipping updates.
1514	 */
1515	SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1516}
1517
1518static inline u64 rq_clock(struct rq *rq)
1519{
1520	lockdep_assert_rq_held(rq);
1521	assert_clock_updated(rq);
1522
1523	return rq->clock;
1524}
1525
1526static inline u64 rq_clock_task(struct rq *rq)
1527{
1528	lockdep_assert_rq_held(rq);
1529	assert_clock_updated(rq);
1530
1531	return rq->clock_task;
1532}
1533
1534/**
1535 * By default the decay is the default pelt decay period.
1536 * The decay shift can change the decay period in
1537 * multiples of 32.
1538 *  Decay shift		Decay period(ms)
1539 *	0			32
1540 *	1			64
1541 *	2			128
1542 *	3			256
1543 *	4			512
1544 */
1545extern int sched_thermal_decay_shift;
1546
1547static inline u64 rq_clock_thermal(struct rq *rq)
1548{
1549	return rq_clock_task(rq) >> sched_thermal_decay_shift;
1550}
1551
1552static inline void rq_clock_skip_update(struct rq *rq)
1553{
1554	lockdep_assert_rq_held(rq);
1555	rq->clock_update_flags |= RQCF_REQ_SKIP;
1556}
1557
1558/*
1559 * See rt task throttling, which is the only time a skip
1560 * request is canceled.
1561 */
1562static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1563{
1564	lockdep_assert_rq_held(rq);
1565	rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1566}
1567
1568/*
1569 * During cpu offlining and rq wide unthrottling, we can trigger
1570 * an update_rq_clock() for several cfs and rt runqueues (Typically
1571 * when using list_for_each_entry_*)
1572 * rq_clock_start_loop_update() can be called after updating the clock
1573 * once and before iterating over the list to prevent multiple update.
1574 * After the iterative traversal, we need to call rq_clock_stop_loop_update()
1575 * to clear RQCF_ACT_SKIP of rq->clock_update_flags.
1576 */
1577static inline void rq_clock_start_loop_update(struct rq *rq)
1578{
1579	lockdep_assert_rq_held(rq);
1580	SCHED_WARN_ON(rq->clock_update_flags & RQCF_ACT_SKIP);
1581	rq->clock_update_flags |= RQCF_ACT_SKIP;
1582}
1583
1584static inline void rq_clock_stop_loop_update(struct rq *rq)
1585{
1586	lockdep_assert_rq_held(rq);
1587	rq->clock_update_flags &= ~RQCF_ACT_SKIP;
1588}
1589
1590struct rq_flags {
1591	unsigned long flags;
1592	struct pin_cookie cookie;
1593#ifdef CONFIG_SCHED_DEBUG
1594	/*
1595	 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1596	 * current pin context is stashed here in case it needs to be
1597	 * restored in rq_repin_lock().
1598	 */
1599	unsigned int clock_update_flags;
1600#endif
1601};
1602
1603extern struct balance_callback balance_push_callback;
1604
1605/*
1606 * Lockdep annotation that avoids accidental unlocks; it's like a
1607 * sticky/continuous lockdep_assert_held().
1608 *
1609 * This avoids code that has access to 'struct rq *rq' (basically everything in
1610 * the scheduler) from accidentally unlocking the rq if they do not also have a
1611 * copy of the (on-stack) 'struct rq_flags rf'.
1612 *
1613 * Also see Documentation/locking/lockdep-design.rst.
1614 */
1615static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1616{
1617	rf->cookie = lockdep_pin_lock(__rq_lockp(rq));
1618
1619#ifdef CONFIG_SCHED_DEBUG
1620	rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1621	rf->clock_update_flags = 0;
1622#ifdef CONFIG_SMP
1623	SCHED_WARN_ON(rq->balance_callback && rq->balance_callback != &balance_push_callback);
1624#endif
1625#endif
1626}
1627
1628static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1629{
1630#ifdef CONFIG_SCHED_DEBUG
1631	if (rq->clock_update_flags > RQCF_ACT_SKIP)
1632		rf->clock_update_flags = RQCF_UPDATED;
1633#endif
1634
1635	lockdep_unpin_lock(__rq_lockp(rq), rf->cookie);
1636}
1637
1638static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1639{
1640	lockdep_repin_lock(__rq_lockp(rq), rf->cookie);
1641
1642#ifdef CONFIG_SCHED_DEBUG
1643	/*
1644	 * Restore the value we stashed in @rf for this pin context.
1645	 */
1646	rq->clock_update_flags |= rf->clock_update_flags;
1647#endif
1648}
1649
1650struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1651	__acquires(rq->lock);
1652
1653struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1654	__acquires(p->pi_lock)
1655	__acquires(rq->lock);
1656
1657static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1658	__releases(rq->lock)
1659{
1660	rq_unpin_lock(rq, rf);
1661	raw_spin_rq_unlock(rq);
1662}
1663
1664static inline void
1665task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1666	__releases(rq->lock)
1667	__releases(p->pi_lock)
1668{
1669	rq_unpin_lock(rq, rf);
1670	raw_spin_rq_unlock(rq);
1671	raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1672}
1673
1674DEFINE_LOCK_GUARD_1(task_rq_lock, struct task_struct,
1675		    _T->rq = task_rq_lock(_T->lock, &_T->rf),
1676		    task_rq_unlock(_T->rq, _T->lock, &_T->rf),
1677		    struct rq *rq; struct rq_flags rf)
1678
1679static inline void
1680rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1681	__acquires(rq->lock)
1682{
1683	raw_spin_rq_lock_irqsave(rq, rf->flags);
1684	rq_pin_lock(rq, rf);
1685}
1686
1687static inline void
1688rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1689	__acquires(rq->lock)
1690{
1691	raw_spin_rq_lock_irq(rq);
1692	rq_pin_lock(rq, rf);
1693}
1694
1695static inline void
1696rq_lock(struct rq *rq, struct rq_flags *rf)
1697	__acquires(rq->lock)
1698{
1699	raw_spin_rq_lock(rq);
1700	rq_pin_lock(rq, rf);
1701}
1702
1703static inline void
1704rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1705	__releases(rq->lock)
1706{
1707	rq_unpin_lock(rq, rf);
1708	raw_spin_rq_unlock_irqrestore(rq, rf->flags);
1709}
1710
1711static inline void
1712rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1713	__releases(rq->lock)
1714{
1715	rq_unpin_lock(rq, rf);
1716	raw_spin_rq_unlock_irq(rq);
1717}
1718
1719static inline void
1720rq_unlock(struct rq *rq, struct rq_flags *rf)
1721	__releases(rq->lock)
1722{
1723	rq_unpin_lock(rq, rf);
1724	raw_spin_rq_unlock(rq);
1725}
1726
1727DEFINE_LOCK_GUARD_1(rq_lock, struct rq,
1728		    rq_lock(_T->lock, &_T->rf),
1729		    rq_unlock(_T->lock, &_T->rf),
1730		    struct rq_flags rf)
1731
1732DEFINE_LOCK_GUARD_1(rq_lock_irq, struct rq,
1733		    rq_lock_irq(_T->lock, &_T->rf),
1734		    rq_unlock_irq(_T->lock, &_T->rf),
1735		    struct rq_flags rf)
1736
1737DEFINE_LOCK_GUARD_1(rq_lock_irqsave, struct rq,
1738		    rq_lock_irqsave(_T->lock, &_T->rf),
1739		    rq_unlock_irqrestore(_T->lock, &_T->rf),
1740		    struct rq_flags rf)
1741
1742static inline struct rq *
1743this_rq_lock_irq(struct rq_flags *rf)
1744	__acquires(rq->lock)
1745{
1746	struct rq *rq;
1747
1748	local_irq_disable();
1749	rq = this_rq();
1750	rq_lock(rq, rf);
1751	return rq;
1752}
1753
1754#ifdef CONFIG_NUMA
1755enum numa_topology_type {
1756	NUMA_DIRECT,
1757	NUMA_GLUELESS_MESH,
1758	NUMA_BACKPLANE,
1759};
1760extern enum numa_topology_type sched_numa_topology_type;
1761extern int sched_max_numa_distance;
1762extern bool find_numa_distance(int distance);
1763extern void sched_init_numa(int offline_node);
1764extern void sched_update_numa(int cpu, bool online);
1765extern void sched_domains_numa_masks_set(unsigned int cpu);
1766extern void sched_domains_numa_masks_clear(unsigned int cpu);
1767extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1768#else
1769static inline void sched_init_numa(int offline_node) { }
1770static inline void sched_update_numa(int cpu, bool online) { }
1771static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1772static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1773static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1774{
1775	return nr_cpu_ids;
1776}
1777#endif
1778
1779#ifdef CONFIG_NUMA_BALANCING
1780/* The regions in numa_faults array from task_struct */
1781enum numa_faults_stats {
1782	NUMA_MEM = 0,
1783	NUMA_CPU,
1784	NUMA_MEMBUF,
1785	NUMA_CPUBUF
1786};
1787extern void sched_setnuma(struct task_struct *p, int node);
1788extern int migrate_task_to(struct task_struct *p, int cpu);
1789extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1790			int cpu, int scpu);
1791extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1792#else
1793static inline void
1794init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1795{
1796}
1797#endif /* CONFIG_NUMA_BALANCING */
1798
1799#ifdef CONFIG_SMP
1800
1801static inline void
1802queue_balance_callback(struct rq *rq,
1803		       struct balance_callback *head,
1804		       void (*func)(struct rq *rq))
1805{
1806	lockdep_assert_rq_held(rq);
1807
1808	/*
1809	 * Don't (re)queue an already queued item; nor queue anything when
1810	 * balance_push() is active, see the comment with
1811	 * balance_push_callback.
1812	 */
1813	if (unlikely(head->next || rq->balance_callback == &balance_push_callback))
1814		return;
1815
1816	head->func = func;
1817	head->next = rq->balance_callback;
1818	rq->balance_callback = head;
1819}
1820
1821#define rcu_dereference_check_sched_domain(p) \
1822	rcu_dereference_check((p), \
1823			      lockdep_is_held(&sched_domains_mutex))
1824
1825/*
1826 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1827 * See destroy_sched_domains: call_rcu for details.
1828 *
1829 * The domain tree of any CPU may only be accessed from within
1830 * preempt-disabled sections.
1831 */
1832#define for_each_domain(cpu, __sd) \
1833	for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1834			__sd; __sd = __sd->parent)
1835
1836/* A mask of all the SD flags that have the SDF_SHARED_CHILD metaflag */
1837#define SD_FLAG(name, mflags) (name * !!((mflags) & SDF_SHARED_CHILD)) |
1838static const unsigned int SD_SHARED_CHILD_MASK =
1839#include <linux/sched/sd_flags.h>
18400;
1841#undef SD_FLAG
1842
1843/**
1844 * highest_flag_domain - Return highest sched_domain containing flag.
1845 * @cpu:	The CPU whose highest level of sched domain is to
1846 *		be returned.
1847 * @flag:	The flag to check for the highest sched_domain
1848 *		for the given CPU.
1849 *
1850 * Returns the highest sched_domain of a CPU which contains @flag. If @flag has
1851 * the SDF_SHARED_CHILD metaflag, all the children domains also have @flag.
1852 */
1853static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1854{
1855	struct sched_domain *sd, *hsd = NULL;
1856
1857	for_each_domain(cpu, sd) {
1858		if (sd->flags & flag) {
1859			hsd = sd;
1860			continue;
1861		}
1862
1863		/*
1864		 * Stop the search if @flag is known to be shared at lower
1865		 * levels. It will not be found further up.
1866		 */
1867		if (flag & SD_SHARED_CHILD_MASK)
1868			break;
 
1869	}
1870
1871	return hsd;
1872}
1873
1874static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1875{
1876	struct sched_domain *sd;
1877
1878	for_each_domain(cpu, sd) {
1879		if (sd->flags & flag)
1880			break;
1881	}
1882
1883	return sd;
1884}
1885
1886DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
1887DECLARE_PER_CPU(int, sd_llc_size);
1888DECLARE_PER_CPU(int, sd_llc_id);
1889DECLARE_PER_CPU(int, sd_share_id);
1890DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
1891DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
1892DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
1893DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
1894extern struct static_key_false sched_asym_cpucapacity;
1895extern struct static_key_false sched_cluster_active;
1896
1897static __always_inline bool sched_asym_cpucap_active(void)
1898{
1899	return static_branch_unlikely(&sched_asym_cpucapacity);
1900}
1901
1902struct sched_group_capacity {
1903	atomic_t		ref;
1904	/*
1905	 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1906	 * for a single CPU.
1907	 */
1908	unsigned long		capacity;
1909	unsigned long		min_capacity;		/* Min per-CPU capacity in group */
1910	unsigned long		max_capacity;		/* Max per-CPU capacity in group */
1911	unsigned long		next_update;
1912	int			imbalance;		/* XXX unrelated to capacity but shared group state */
1913
1914#ifdef CONFIG_SCHED_DEBUG
1915	int			id;
1916#endif
1917
1918	unsigned long		cpumask[];		/* Balance mask */
1919};
1920
1921struct sched_group {
1922	struct sched_group	*next;			/* Must be a circular list */
1923	atomic_t		ref;
1924
1925	unsigned int		group_weight;
1926	unsigned int		cores;
1927	struct sched_group_capacity *sgc;
1928	int			asym_prefer_cpu;	/* CPU of highest priority in group */
1929	int			flags;
1930
1931	/*
1932	 * The CPUs this group covers.
1933	 *
1934	 * NOTE: this field is variable length. (Allocated dynamically
1935	 * by attaching extra space to the end of the structure,
1936	 * depending on how many CPUs the kernel has booted up with)
1937	 */
1938	unsigned long		cpumask[];
1939};
1940
1941static inline struct cpumask *sched_group_span(struct sched_group *sg)
1942{
1943	return to_cpumask(sg->cpumask);
1944}
1945
1946/*
1947 * See build_balance_mask().
1948 */
1949static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1950{
1951	return to_cpumask(sg->sgc->cpumask);
1952}
1953
1954extern int group_balance_cpu(struct sched_group *sg);
1955
1956#ifdef CONFIG_SCHED_DEBUG
1957void update_sched_domain_debugfs(void);
1958void dirty_sched_domain_sysctl(int cpu);
1959#else
1960static inline void update_sched_domain_debugfs(void)
1961{
1962}
1963static inline void dirty_sched_domain_sysctl(int cpu)
1964{
1965}
1966#endif
1967
1968extern int sched_update_scaling(void);
1969
1970static inline const struct cpumask *task_user_cpus(struct task_struct *p)
1971{
1972	if (!p->user_cpus_ptr)
1973		return cpu_possible_mask; /* &init_task.cpus_mask */
1974	return p->user_cpus_ptr;
1975}
1976#endif /* CONFIG_SMP */
1977
1978#include "stats.h"
1979
1980#if defined(CONFIG_SCHED_CORE) && defined(CONFIG_SCHEDSTATS)
1981
1982extern void __sched_core_account_forceidle(struct rq *rq);
1983
1984static inline void sched_core_account_forceidle(struct rq *rq)
1985{
1986	if (schedstat_enabled())
1987		__sched_core_account_forceidle(rq);
1988}
1989
1990extern void __sched_core_tick(struct rq *rq);
1991
1992static inline void sched_core_tick(struct rq *rq)
1993{
1994	if (sched_core_enabled(rq) && schedstat_enabled())
1995		__sched_core_tick(rq);
1996}
1997
1998#else
1999
2000static inline void sched_core_account_forceidle(struct rq *rq) {}
2001
2002static inline void sched_core_tick(struct rq *rq) {}
2003
2004#endif /* CONFIG_SCHED_CORE && CONFIG_SCHEDSTATS */
2005
2006#ifdef CONFIG_CGROUP_SCHED
2007
2008/*
2009 * Return the group to which this tasks belongs.
2010 *
2011 * We cannot use task_css() and friends because the cgroup subsystem
2012 * changes that value before the cgroup_subsys::attach() method is called,
2013 * therefore we cannot pin it and might observe the wrong value.
2014 *
2015 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
2016 * core changes this before calling sched_move_task().
2017 *
2018 * Instead we use a 'copy' which is updated from sched_move_task() while
2019 * holding both task_struct::pi_lock and rq::lock.
2020 */
2021static inline struct task_group *task_group(struct task_struct *p)
2022{
2023	return p->sched_task_group;
2024}
2025
2026/* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
2027static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
2028{
2029#if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
2030	struct task_group *tg = task_group(p);
2031#endif
2032
2033#ifdef CONFIG_FAIR_GROUP_SCHED
2034	set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
2035	p->se.cfs_rq = tg->cfs_rq[cpu];
2036	p->se.parent = tg->se[cpu];
2037	p->se.depth = tg->se[cpu] ? tg->se[cpu]->depth + 1 : 0;
2038#endif
2039
2040#ifdef CONFIG_RT_GROUP_SCHED
2041	p->rt.rt_rq  = tg->rt_rq[cpu];
2042	p->rt.parent = tg->rt_se[cpu];
2043#endif
2044}
2045
2046#else /* CONFIG_CGROUP_SCHED */
2047
2048static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
2049static inline struct task_group *task_group(struct task_struct *p)
2050{
2051	return NULL;
2052}
2053
2054#endif /* CONFIG_CGROUP_SCHED */
2055
2056static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
2057{
2058	set_task_rq(p, cpu);
2059#ifdef CONFIG_SMP
2060	/*
2061	 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
2062	 * successfully executed on another CPU. We must ensure that updates of
2063	 * per-task data have been completed by this moment.
2064	 */
2065	smp_wmb();
2066	WRITE_ONCE(task_thread_info(p)->cpu, cpu);
2067	p->wake_cpu = cpu;
2068#endif
2069}
2070
2071/*
2072 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
2073 */
2074#ifdef CONFIG_SCHED_DEBUG
2075# define const_debug __read_mostly
2076#else
2077# define const_debug const
2078#endif
2079
2080#define SCHED_FEAT(name, enabled)	\
2081	__SCHED_FEAT_##name ,
2082
2083enum {
2084#include "features.h"
2085	__SCHED_FEAT_NR,
2086};
2087
2088#undef SCHED_FEAT
2089
2090#ifdef CONFIG_SCHED_DEBUG
2091
2092/*
2093 * To support run-time toggling of sched features, all the translation units
2094 * (but core.c) reference the sysctl_sched_features defined in core.c.
2095 */
2096extern const_debug unsigned int sysctl_sched_features;
2097
2098#ifdef CONFIG_JUMP_LABEL
2099#define SCHED_FEAT(name, enabled)					\
2100static __always_inline bool static_branch_##name(struct static_key *key) \
2101{									\
2102	return static_key_##enabled(key);				\
2103}
2104
2105#include "features.h"
2106#undef SCHED_FEAT
2107
2108extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
2109#define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
2110
2111#else /* !CONFIG_JUMP_LABEL */
2112
2113#define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
2114
2115#endif /* CONFIG_JUMP_LABEL */
2116
2117#else /* !SCHED_DEBUG */
2118
2119/*
2120 * Each translation unit has its own copy of sysctl_sched_features to allow
2121 * constants propagation at compile time and compiler optimization based on
2122 * features default.
2123 */
2124#define SCHED_FEAT(name, enabled)	\
2125	(1UL << __SCHED_FEAT_##name) * enabled |
2126static const_debug __maybe_unused unsigned int sysctl_sched_features =
2127#include "features.h"
2128	0;
2129#undef SCHED_FEAT
2130
2131#define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
2132
2133#endif /* SCHED_DEBUG */
2134
2135extern struct static_key_false sched_numa_balancing;
2136extern struct static_key_false sched_schedstats;
2137
2138static inline u64 global_rt_period(void)
2139{
2140	return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
2141}
2142
2143static inline u64 global_rt_runtime(void)
2144{
2145	if (sysctl_sched_rt_runtime < 0)
2146		return RUNTIME_INF;
2147
2148	return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
2149}
2150
2151static inline int task_current(struct rq *rq, struct task_struct *p)
2152{
2153	return rq->curr == p;
2154}
2155
2156static inline int task_on_cpu(struct rq *rq, struct task_struct *p)
2157{
2158#ifdef CONFIG_SMP
2159	return p->on_cpu;
2160#else
2161	return task_current(rq, p);
2162#endif
2163}
2164
2165static inline int task_on_rq_queued(struct task_struct *p)
2166{
2167	return p->on_rq == TASK_ON_RQ_QUEUED;
2168}
2169
2170static inline int task_on_rq_migrating(struct task_struct *p)
2171{
2172	return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
2173}
2174
2175/* Wake flags. The first three directly map to some SD flag value */
2176#define WF_EXEC         0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */
2177#define WF_FORK         0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */
2178#define WF_TTWU         0x08 /* Wakeup;            maps to SD_BALANCE_WAKE */
2179
2180#define WF_SYNC         0x10 /* Waker goes to sleep after wakeup */
2181#define WF_MIGRATED     0x20 /* Internal use, task got migrated */
2182#define WF_CURRENT_CPU  0x40 /* Prefer to move the wakee to the current CPU. */
2183
2184#ifdef CONFIG_SMP
2185static_assert(WF_EXEC == SD_BALANCE_EXEC);
2186static_assert(WF_FORK == SD_BALANCE_FORK);
2187static_assert(WF_TTWU == SD_BALANCE_WAKE);
2188#endif
2189
2190/*
2191 * To aid in avoiding the subversion of "niceness" due to uneven distribution
2192 * of tasks with abnormal "nice" values across CPUs the contribution that
2193 * each task makes to its run queue's load is weighted according to its
2194 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
2195 * scaled version of the new time slice allocation that they receive on time
2196 * slice expiry etc.
2197 */
2198
2199#define WEIGHT_IDLEPRIO		3
2200#define WMULT_IDLEPRIO		1431655765
2201
2202extern const int		sched_prio_to_weight[40];
2203extern const u32		sched_prio_to_wmult[40];
2204
2205/*
2206 * {de,en}queue flags:
2207 *
2208 * DEQUEUE_SLEEP  - task is no longer runnable
2209 * ENQUEUE_WAKEUP - task just became runnable
2210 *
2211 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
2212 *                are in a known state which allows modification. Such pairs
2213 *                should preserve as much state as possible.
2214 *
2215 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
2216 *        in the runqueue.
2217 *
2218 * NOCLOCK - skip the update_rq_clock() (avoids double updates)
2219 *
2220 * MIGRATION - p->on_rq == TASK_ON_RQ_MIGRATING (used for DEADLINE)
2221 *
2222 * ENQUEUE_HEAD      - place at front of runqueue (tail if not specified)
2223 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
2224 * ENQUEUE_MIGRATED  - the task was migrated during wakeup
2225 *
2226 */
2227
2228#define DEQUEUE_SLEEP		0x01
2229#define DEQUEUE_SAVE		0x02 /* Matches ENQUEUE_RESTORE */
2230#define DEQUEUE_MOVE		0x04 /* Matches ENQUEUE_MOVE */
2231#define DEQUEUE_NOCLOCK		0x08 /* Matches ENQUEUE_NOCLOCK */
2232#define DEQUEUE_MIGRATING	0x100 /* Matches ENQUEUE_MIGRATING */
2233
2234#define ENQUEUE_WAKEUP		0x01
2235#define ENQUEUE_RESTORE		0x02
2236#define ENQUEUE_MOVE		0x04
2237#define ENQUEUE_NOCLOCK		0x08
2238
2239#define ENQUEUE_HEAD		0x10
2240#define ENQUEUE_REPLENISH	0x20
2241#ifdef CONFIG_SMP
2242#define ENQUEUE_MIGRATED	0x40
2243#else
2244#define ENQUEUE_MIGRATED	0x00
2245#endif
2246#define ENQUEUE_INITIAL		0x80
2247#define ENQUEUE_MIGRATING	0x100
2248
2249#define RETRY_TASK		((void *)-1UL)
2250
2251struct affinity_context {
2252	const struct cpumask *new_mask;
2253	struct cpumask *user_mask;
2254	unsigned int flags;
2255};
2256
2257extern s64 update_curr_common(struct rq *rq);
2258
2259struct sched_class {
2260
2261#ifdef CONFIG_UCLAMP_TASK
2262	int uclamp_enabled;
2263#endif
2264
2265	void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
2266	void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
2267	void (*yield_task)   (struct rq *rq);
2268	bool (*yield_to_task)(struct rq *rq, struct task_struct *p);
2269
2270	void (*wakeup_preempt)(struct rq *rq, struct task_struct *p, int flags);
2271
2272	struct task_struct *(*pick_next_task)(struct rq *rq);
2273
2274	void (*put_prev_task)(struct rq *rq, struct task_struct *p);
2275	void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
2276
2277#ifdef CONFIG_SMP
2278	int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2279	int  (*select_task_rq)(struct task_struct *p, int task_cpu, int flags);
2280
2281	struct task_struct * (*pick_task)(struct rq *rq);
2282
2283	void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
2284
2285	void (*task_woken)(struct rq *this_rq, struct task_struct *task);
2286
2287	void (*set_cpus_allowed)(struct task_struct *p, struct affinity_context *ctx);
2288
2289	void (*rq_online)(struct rq *rq);
2290	void (*rq_offline)(struct rq *rq);
2291
2292	struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq);
2293#endif
2294
2295	void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
2296	void (*task_fork)(struct task_struct *p);
2297	void (*task_dead)(struct task_struct *p);
2298
2299	/*
2300	 * The switched_from() call is allowed to drop rq->lock, therefore we
2301	 * cannot assume the switched_from/switched_to pair is serialized by
2302	 * rq->lock. They are however serialized by p->pi_lock.
2303	 */
2304	void (*switched_from)(struct rq *this_rq, struct task_struct *task);
2305	void (*switched_to)  (struct rq *this_rq, struct task_struct *task);
2306	void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
2307			      int oldprio);
2308
2309	unsigned int (*get_rr_interval)(struct rq *rq,
2310					struct task_struct *task);
2311
2312	void (*update_curr)(struct rq *rq);
2313
2314#ifdef CONFIG_FAIR_GROUP_SCHED
2315	void (*task_change_group)(struct task_struct *p);
2316#endif
2317
2318#ifdef CONFIG_SCHED_CORE
2319	int (*task_is_throttled)(struct task_struct *p, int cpu);
2320#endif
2321};
2322
2323static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
2324{
2325	WARN_ON_ONCE(rq->curr != prev);
2326	prev->sched_class->put_prev_task(rq, prev);
2327}
2328
2329static inline void set_next_task(struct rq *rq, struct task_struct *next)
2330{
2331	next->sched_class->set_next_task(rq, next, false);
2332}
2333
2334
2335/*
2336 * Helper to define a sched_class instance; each one is placed in a separate
2337 * section which is ordered by the linker script:
2338 *
2339 *   include/asm-generic/vmlinux.lds.h
2340 *
2341 * *CAREFUL* they are laid out in *REVERSE* order!!!
2342 *
2343 * Also enforce alignment on the instance, not the type, to guarantee layout.
2344 */
2345#define DEFINE_SCHED_CLASS(name) \
2346const struct sched_class name##_sched_class \
2347	__aligned(__alignof__(struct sched_class)) \
2348	__section("__" #name "_sched_class")
2349
2350/* Defined in include/asm-generic/vmlinux.lds.h */
2351extern struct sched_class __sched_class_highest[];
2352extern struct sched_class __sched_class_lowest[];
2353
2354#define for_class_range(class, _from, _to) \
2355	for (class = (_from); class < (_to); class++)
2356
2357#define for_each_class(class) \
2358	for_class_range(class, __sched_class_highest, __sched_class_lowest)
2359
2360#define sched_class_above(_a, _b)	((_a) < (_b))
2361
2362extern const struct sched_class stop_sched_class;
2363extern const struct sched_class dl_sched_class;
2364extern const struct sched_class rt_sched_class;
2365extern const struct sched_class fair_sched_class;
2366extern const struct sched_class idle_sched_class;
2367
2368static inline bool sched_stop_runnable(struct rq *rq)
2369{
2370	return rq->stop && task_on_rq_queued(rq->stop);
2371}
2372
2373static inline bool sched_dl_runnable(struct rq *rq)
2374{
2375	return rq->dl.dl_nr_running > 0;
2376}
2377
2378static inline bool sched_rt_runnable(struct rq *rq)
2379{
2380	return rq->rt.rt_queued > 0;
2381}
2382
2383static inline bool sched_fair_runnable(struct rq *rq)
2384{
2385	return rq->cfs.nr_running > 0;
2386}
2387
2388extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2389extern struct task_struct *pick_next_task_idle(struct rq *rq);
2390
2391#define SCA_CHECK		0x01
2392#define SCA_MIGRATE_DISABLE	0x02
2393#define SCA_MIGRATE_ENABLE	0x04
2394#define SCA_USER		0x08
2395
2396#ifdef CONFIG_SMP
2397
2398extern void update_group_capacity(struct sched_domain *sd, int cpu);
2399
2400extern void trigger_load_balance(struct rq *rq);
2401
2402extern void set_cpus_allowed_common(struct task_struct *p, struct affinity_context *ctx);
2403
2404static inline struct task_struct *get_push_task(struct rq *rq)
2405{
2406	struct task_struct *p = rq->curr;
2407
2408	lockdep_assert_rq_held(rq);
2409
2410	if (rq->push_busy)
2411		return NULL;
2412
2413	if (p->nr_cpus_allowed == 1)
2414		return NULL;
2415
2416	if (p->migration_disabled)
2417		return NULL;
2418
2419	rq->push_busy = true;
2420	return get_task_struct(p);
2421}
2422
2423extern int push_cpu_stop(void *arg);
2424
2425#endif
2426
2427#ifdef CONFIG_CPU_IDLE
2428static inline void idle_set_state(struct rq *rq,
2429				  struct cpuidle_state *idle_state)
2430{
2431	rq->idle_state = idle_state;
2432}
2433
2434static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2435{
2436	SCHED_WARN_ON(!rcu_read_lock_held());
2437
2438	return rq->idle_state;
2439}
2440#else
2441static inline void idle_set_state(struct rq *rq,
2442				  struct cpuidle_state *idle_state)
2443{
2444}
2445
2446static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2447{
2448	return NULL;
2449}
2450#endif
2451
2452extern void schedule_idle(void);
2453asmlinkage void schedule_user(void);
2454
2455extern void sysrq_sched_debug_show(void);
2456extern void sched_init_granularity(void);
2457extern void update_max_interval(void);
2458
2459extern void init_sched_dl_class(void);
2460extern void init_sched_rt_class(void);
2461extern void init_sched_fair_class(void);
2462
2463extern void reweight_task(struct task_struct *p, int prio);
2464
2465extern void resched_curr(struct rq *rq);
2466extern void resched_cpu(int cpu);
2467
2468extern struct rt_bandwidth def_rt_bandwidth;
2469extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
2470extern bool sched_rt_bandwidth_account(struct rt_rq *rt_rq);
2471
2472extern void init_dl_entity(struct sched_dl_entity *dl_se);
 
 
2473
2474#define BW_SHIFT		20
2475#define BW_UNIT			(1 << BW_SHIFT)
2476#define RATIO_SHIFT		8
2477#define MAX_BW_BITS		(64 - BW_SHIFT)
2478#define MAX_BW			((1ULL << MAX_BW_BITS) - 1)
2479unsigned long to_ratio(u64 period, u64 runtime);
2480
2481extern void init_entity_runnable_average(struct sched_entity *se);
2482extern void post_init_entity_util_avg(struct task_struct *p);
2483
2484#ifdef CONFIG_NO_HZ_FULL
2485extern bool sched_can_stop_tick(struct rq *rq);
2486extern int __init sched_tick_offload_init(void);
2487
2488/*
2489 * Tick may be needed by tasks in the runqueue depending on their policy and
2490 * requirements. If tick is needed, lets send the target an IPI to kick it out of
2491 * nohz mode if necessary.
2492 */
2493static inline void sched_update_tick_dependency(struct rq *rq)
2494{
2495	int cpu = cpu_of(rq);
2496
2497	if (!tick_nohz_full_cpu(cpu))
2498		return;
2499
2500	if (sched_can_stop_tick(rq))
2501		tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
2502	else
2503		tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
2504}
2505#else
2506static inline int sched_tick_offload_init(void) { return 0; }
2507static inline void sched_update_tick_dependency(struct rq *rq) { }
2508#endif
2509
2510static inline void add_nr_running(struct rq *rq, unsigned count)
2511{
2512	unsigned prev_nr = rq->nr_running;
2513
2514	rq->nr_running = prev_nr + count;
2515	if (trace_sched_update_nr_running_tp_enabled()) {
2516		call_trace_sched_update_nr_running(rq, count);
2517	}
2518
2519#ifdef CONFIG_SMP
2520	if (prev_nr < 2 && rq->nr_running >= 2) {
2521		if (!READ_ONCE(rq->rd->overload))
2522			WRITE_ONCE(rq->rd->overload, 1);
2523	}
2524#endif
2525
2526	sched_update_tick_dependency(rq);
2527}
2528
2529static inline void sub_nr_running(struct rq *rq, unsigned count)
2530{
2531	rq->nr_running -= count;
2532	if (trace_sched_update_nr_running_tp_enabled()) {
2533		call_trace_sched_update_nr_running(rq, -count);
2534	}
2535
2536	/* Check if we still need preemption */
2537	sched_update_tick_dependency(rq);
2538}
2539
2540extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
2541extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
2542
2543extern void wakeup_preempt(struct rq *rq, struct task_struct *p, int flags);
2544
2545#ifdef CONFIG_PREEMPT_RT
2546#define SCHED_NR_MIGRATE_BREAK 8
2547#else
2548#define SCHED_NR_MIGRATE_BREAK 32
2549#endif
2550
2551extern const_debug unsigned int sysctl_sched_nr_migrate;
2552extern const_debug unsigned int sysctl_sched_migration_cost;
2553
2554extern unsigned int sysctl_sched_base_slice;
2555
2556#ifdef CONFIG_SCHED_DEBUG
 
 
 
 
2557extern int sysctl_resched_latency_warn_ms;
2558extern int sysctl_resched_latency_warn_once;
2559
2560extern unsigned int sysctl_sched_tunable_scaling;
2561
2562extern unsigned int sysctl_numa_balancing_scan_delay;
2563extern unsigned int sysctl_numa_balancing_scan_period_min;
2564extern unsigned int sysctl_numa_balancing_scan_period_max;
2565extern unsigned int sysctl_numa_balancing_scan_size;
2566extern unsigned int sysctl_numa_balancing_hot_threshold;
2567#endif
2568
2569#ifdef CONFIG_SCHED_HRTICK
2570
2571/*
2572 * Use hrtick when:
2573 *  - enabled by features
2574 *  - hrtimer is actually high res
2575 */
2576static inline int hrtick_enabled(struct rq *rq)
2577{
2578	if (!cpu_active(cpu_of(rq)))
2579		return 0;
2580	return hrtimer_is_hres_active(&rq->hrtick_timer);
2581}
2582
2583static inline int hrtick_enabled_fair(struct rq *rq)
2584{
2585	if (!sched_feat(HRTICK))
2586		return 0;
2587	return hrtick_enabled(rq);
2588}
2589
2590static inline int hrtick_enabled_dl(struct rq *rq)
2591{
2592	if (!sched_feat(HRTICK_DL))
2593		return 0;
2594	return hrtick_enabled(rq);
2595}
2596
2597void hrtick_start(struct rq *rq, u64 delay);
2598
2599#else
2600
2601static inline int hrtick_enabled_fair(struct rq *rq)
2602{
2603	return 0;
2604}
2605
2606static inline int hrtick_enabled_dl(struct rq *rq)
2607{
2608	return 0;
2609}
2610
2611static inline int hrtick_enabled(struct rq *rq)
2612{
2613	return 0;
2614}
2615
2616#endif /* CONFIG_SCHED_HRTICK */
2617
2618#ifndef arch_scale_freq_tick
2619static __always_inline
2620void arch_scale_freq_tick(void)
2621{
2622}
2623#endif
2624
2625#ifndef arch_scale_freq_capacity
2626/**
2627 * arch_scale_freq_capacity - get the frequency scale factor of a given CPU.
2628 * @cpu: the CPU in question.
2629 *
2630 * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
2631 *
2632 *     f_curr
2633 *     ------ * SCHED_CAPACITY_SCALE
2634 *     f_max
2635 */
2636static __always_inline
2637unsigned long arch_scale_freq_capacity(int cpu)
2638{
2639	return SCHED_CAPACITY_SCALE;
2640}
2641#endif
2642
2643#ifdef CONFIG_SCHED_DEBUG
2644/*
2645 * In double_lock_balance()/double_rq_lock(), we use raw_spin_rq_lock() to
2646 * acquire rq lock instead of rq_lock(). So at the end of these two functions
2647 * we need to call double_rq_clock_clear_update() to clear RQCF_UPDATED of
2648 * rq->clock_update_flags to avoid the WARN_DOUBLE_CLOCK warning.
2649 */
2650static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2)
2651{
2652	rq1->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
2653	/* rq1 == rq2 for !CONFIG_SMP, so just clear RQCF_UPDATED once. */
2654#ifdef CONFIG_SMP
2655	rq2->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
2656#endif
2657}
2658#else
2659static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2) {}
2660#endif
2661
2662#define DEFINE_LOCK_GUARD_2(name, type, _lock, _unlock, ...)		\
2663__DEFINE_UNLOCK_GUARD(name, type, _unlock, type *lock2; __VA_ARGS__) \
2664static inline class_##name##_t class_##name##_constructor(type *lock, type *lock2) \
2665{ class_##name##_t _t = { .lock = lock, .lock2 = lock2 }, *_T = &_t;	\
2666  _lock; return _t; }
2667
2668#ifdef CONFIG_SMP
2669
2670static inline bool rq_order_less(struct rq *rq1, struct rq *rq2)
2671{
2672#ifdef CONFIG_SCHED_CORE
2673	/*
2674	 * In order to not have {0,2},{1,3} turn into into an AB-BA,
2675	 * order by core-id first and cpu-id second.
2676	 *
2677	 * Notably:
2678	 *
2679	 *	double_rq_lock(0,3); will take core-0, core-1 lock
2680	 *	double_rq_lock(1,2); will take core-1, core-0 lock
2681	 *
2682	 * when only cpu-id is considered.
2683	 */
2684	if (rq1->core->cpu < rq2->core->cpu)
2685		return true;
2686	if (rq1->core->cpu > rq2->core->cpu)
2687		return false;
2688
2689	/*
2690	 * __sched_core_flip() relies on SMT having cpu-id lock order.
2691	 */
2692#endif
2693	return rq1->cpu < rq2->cpu;
2694}
2695
2696extern void double_rq_lock(struct rq *rq1, struct rq *rq2);
2697
2698#ifdef CONFIG_PREEMPTION
2699
2700/*
2701 * fair double_lock_balance: Safely acquires both rq->locks in a fair
2702 * way at the expense of forcing extra atomic operations in all
2703 * invocations.  This assures that the double_lock is acquired using the
2704 * same underlying policy as the spinlock_t on this architecture, which
2705 * reduces latency compared to the unfair variant below.  However, it
2706 * also adds more overhead and therefore may reduce throughput.
2707 */
2708static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2709	__releases(this_rq->lock)
2710	__acquires(busiest->lock)
2711	__acquires(this_rq->lock)
2712{
2713	raw_spin_rq_unlock(this_rq);
2714	double_rq_lock(this_rq, busiest);
2715
2716	return 1;
2717}
2718
2719#else
2720/*
2721 * Unfair double_lock_balance: Optimizes throughput at the expense of
2722 * latency by eliminating extra atomic operations when the locks are
2723 * already in proper order on entry.  This favors lower CPU-ids and will
2724 * grant the double lock to lower CPUs over higher ids under contention,
2725 * regardless of entry order into the function.
2726 */
2727static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2728	__releases(this_rq->lock)
2729	__acquires(busiest->lock)
2730	__acquires(this_rq->lock)
2731{
2732	if (__rq_lockp(this_rq) == __rq_lockp(busiest) ||
2733	    likely(raw_spin_rq_trylock(busiest))) {
2734		double_rq_clock_clear_update(this_rq, busiest);
2735		return 0;
2736	}
2737
2738	if (rq_order_less(this_rq, busiest)) {
2739		raw_spin_rq_lock_nested(busiest, SINGLE_DEPTH_NESTING);
2740		double_rq_clock_clear_update(this_rq, busiest);
2741		return 0;
2742	}
2743
2744	raw_spin_rq_unlock(this_rq);
2745	double_rq_lock(this_rq, busiest);
2746
2747	return 1;
2748}
2749
2750#endif /* CONFIG_PREEMPTION */
2751
2752/*
2753 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2754 */
2755static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2756{
2757	lockdep_assert_irqs_disabled();
2758
2759	return _double_lock_balance(this_rq, busiest);
2760}
2761
2762static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2763	__releases(busiest->lock)
2764{
2765	if (__rq_lockp(this_rq) != __rq_lockp(busiest))
2766		raw_spin_rq_unlock(busiest);
2767	lock_set_subclass(&__rq_lockp(this_rq)->dep_map, 0, _RET_IP_);
2768}
2769
2770static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
2771{
2772	if (l1 > l2)
2773		swap(l1, l2);
2774
2775	spin_lock(l1);
2776	spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2777}
2778
2779static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
2780{
2781	if (l1 > l2)
2782		swap(l1, l2);
2783
2784	spin_lock_irq(l1);
2785	spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2786}
2787
2788static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2789{
2790	if (l1 > l2)
2791		swap(l1, l2);
2792
2793	raw_spin_lock(l1);
2794	raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2795}
2796
2797static inline void double_raw_unlock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2798{
2799	raw_spin_unlock(l1);
2800	raw_spin_unlock(l2);
2801}
2802
2803DEFINE_LOCK_GUARD_2(double_raw_spinlock, raw_spinlock_t,
2804		    double_raw_lock(_T->lock, _T->lock2),
2805		    double_raw_unlock(_T->lock, _T->lock2))
2806
2807/*
2808 * double_rq_unlock - safely unlock two runqueues
2809 *
2810 * Note this does not restore interrupts like task_rq_unlock,
2811 * you need to do so manually after calling.
2812 */
2813static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2814	__releases(rq1->lock)
2815	__releases(rq2->lock)
2816{
2817	if (__rq_lockp(rq1) != __rq_lockp(rq2))
2818		raw_spin_rq_unlock(rq2);
2819	else
2820		__release(rq2->lock);
2821	raw_spin_rq_unlock(rq1);
2822}
2823
2824extern void set_rq_online (struct rq *rq);
2825extern void set_rq_offline(struct rq *rq);
2826extern bool sched_smp_initialized;
2827
2828#else /* CONFIG_SMP */
2829
2830/*
2831 * double_rq_lock - safely lock two runqueues
2832 *
2833 * Note this does not disable interrupts like task_rq_lock,
2834 * you need to do so manually before calling.
2835 */
2836static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2837	__acquires(rq1->lock)
2838	__acquires(rq2->lock)
2839{
2840	WARN_ON_ONCE(!irqs_disabled());
2841	WARN_ON_ONCE(rq1 != rq2);
2842	raw_spin_rq_lock(rq1);
2843	__acquire(rq2->lock);	/* Fake it out ;) */
2844	double_rq_clock_clear_update(rq1, rq2);
2845}
2846
2847/*
2848 * double_rq_unlock - safely unlock two runqueues
2849 *
2850 * Note this does not restore interrupts like task_rq_unlock,
2851 * you need to do so manually after calling.
2852 */
2853static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2854	__releases(rq1->lock)
2855	__releases(rq2->lock)
2856{
2857	WARN_ON_ONCE(rq1 != rq2);
2858	raw_spin_rq_unlock(rq1);
2859	__release(rq2->lock);
2860}
2861
2862#endif
2863
2864DEFINE_LOCK_GUARD_2(double_rq_lock, struct rq,
2865		    double_rq_lock(_T->lock, _T->lock2),
2866		    double_rq_unlock(_T->lock, _T->lock2))
2867
2868extern struct sched_entity *__pick_root_entity(struct cfs_rq *cfs_rq);
2869extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2870extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2871
2872#ifdef	CONFIG_SCHED_DEBUG
2873extern bool sched_debug_verbose;
2874
2875extern void print_cfs_stats(struct seq_file *m, int cpu);
2876extern void print_rt_stats(struct seq_file *m, int cpu);
2877extern void print_dl_stats(struct seq_file *m, int cpu);
2878extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2879extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2880extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2881
2882extern void resched_latency_warn(int cpu, u64 latency);
2883#ifdef CONFIG_NUMA_BALANCING
2884extern void
2885show_numa_stats(struct task_struct *p, struct seq_file *m);
2886extern void
2887print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2888	unsigned long tpf, unsigned long gsf, unsigned long gpf);
2889#endif /* CONFIG_NUMA_BALANCING */
2890#else
2891static inline void resched_latency_warn(int cpu, u64 latency) {}
2892#endif /* CONFIG_SCHED_DEBUG */
2893
2894extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2895extern void init_rt_rq(struct rt_rq *rt_rq);
2896extern void init_dl_rq(struct dl_rq *dl_rq);
2897
2898extern void cfs_bandwidth_usage_inc(void);
2899extern void cfs_bandwidth_usage_dec(void);
2900
2901#ifdef CONFIG_NO_HZ_COMMON
2902#define NOHZ_BALANCE_KICK_BIT	0
2903#define NOHZ_STATS_KICK_BIT	1
2904#define NOHZ_NEWILB_KICK_BIT	2
2905#define NOHZ_NEXT_KICK_BIT	3
2906
2907/* Run rebalance_domains() */
2908#define NOHZ_BALANCE_KICK	BIT(NOHZ_BALANCE_KICK_BIT)
2909/* Update blocked load */
2910#define NOHZ_STATS_KICK		BIT(NOHZ_STATS_KICK_BIT)
2911/* Update blocked load when entering idle */
2912#define NOHZ_NEWILB_KICK	BIT(NOHZ_NEWILB_KICK_BIT)
2913/* Update nohz.next_balance */
2914#define NOHZ_NEXT_KICK		BIT(NOHZ_NEXT_KICK_BIT)
2915
2916#define NOHZ_KICK_MASK	(NOHZ_BALANCE_KICK | NOHZ_STATS_KICK | NOHZ_NEXT_KICK)
2917
2918#define nohz_flags(cpu)	(&cpu_rq(cpu)->nohz_flags)
2919
2920extern void nohz_balance_exit_idle(struct rq *rq);
2921#else
2922static inline void nohz_balance_exit_idle(struct rq *rq) { }
2923#endif
2924
2925#if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
2926extern void nohz_run_idle_balance(int cpu);
2927#else
2928static inline void nohz_run_idle_balance(int cpu) { }
2929#endif
2930
2931#ifdef CONFIG_IRQ_TIME_ACCOUNTING
2932struct irqtime {
2933	u64			total;
2934	u64			tick_delta;
2935	u64			irq_start_time;
2936	struct u64_stats_sync	sync;
2937};
2938
2939DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2940
2941/*
2942 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2943 * Otherwise ksoftirqd's sum_exec_runtime is subtracted its own runtime
2944 * and never move forward.
2945 */
2946static inline u64 irq_time_read(int cpu)
2947{
2948	struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2949	unsigned int seq;
2950	u64 total;
2951
2952	do {
2953		seq = __u64_stats_fetch_begin(&irqtime->sync);
2954		total = irqtime->total;
2955	} while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2956
2957	return total;
2958}
2959#endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2960
2961#ifdef CONFIG_CPU_FREQ
2962DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
2963
2964/**
2965 * cpufreq_update_util - Take a note about CPU utilization changes.
2966 * @rq: Runqueue to carry out the update for.
2967 * @flags: Update reason flags.
2968 *
2969 * This function is called by the scheduler on the CPU whose utilization is
2970 * being updated.
2971 *
2972 * It can only be called from RCU-sched read-side critical sections.
2973 *
2974 * The way cpufreq is currently arranged requires it to evaluate the CPU
2975 * performance state (frequency/voltage) on a regular basis to prevent it from
2976 * being stuck in a completely inadequate performance level for too long.
2977 * That is not guaranteed to happen if the updates are only triggered from CFS
2978 * and DL, though, because they may not be coming in if only RT tasks are
2979 * active all the time (or there are RT tasks only).
2980 *
2981 * As a workaround for that issue, this function is called periodically by the
2982 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2983 * but that really is a band-aid.  Going forward it should be replaced with
2984 * solutions targeted more specifically at RT tasks.
2985 */
2986static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2987{
2988	struct update_util_data *data;
2989
2990	data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2991						  cpu_of(rq)));
2992	if (data)
2993		data->func(data, rq_clock(rq), flags);
2994}
2995#else
2996static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2997#endif /* CONFIG_CPU_FREQ */
2998
2999#ifdef arch_scale_freq_capacity
3000# ifndef arch_scale_freq_invariant
3001#  define arch_scale_freq_invariant()	true
3002# endif
3003#else
3004# define arch_scale_freq_invariant()	false
3005#endif
3006
3007#ifdef CONFIG_SMP
3008unsigned long effective_cpu_util(int cpu, unsigned long util_cfs,
3009				 unsigned long *min,
3010				 unsigned long *max);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3011
3012unsigned long sugov_effective_cpu_perf(int cpu, unsigned long actual,
3013				 unsigned long min,
3014				 unsigned long max);
 
 
 
 
 
 
 
 
 
 
 
3015
 
 
 
3016
3017/*
3018 * Verify the fitness of task @p to run on @cpu taking into account the
3019 * CPU original capacity and the runtime/deadline ratio of the task.
3020 *
3021 * The function will return true if the original capacity of @cpu is
3022 * greater than or equal to task's deadline density right shifted by
3023 * (BW_SHIFT - SCHED_CAPACITY_SHIFT) and false otherwise.
3024 */
3025static inline bool dl_task_fits_capacity(struct task_struct *p, int cpu)
3026{
3027	unsigned long cap = arch_scale_cpu_capacity(cpu);
3028
3029	return cap >= p->dl.dl_density >> (BW_SHIFT - SCHED_CAPACITY_SHIFT);
3030}
3031
3032static inline unsigned long cpu_bw_dl(struct rq *rq)
3033{
3034	return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
3035}
3036
3037static inline unsigned long cpu_util_dl(struct rq *rq)
3038{
3039	return READ_ONCE(rq->avg_dl.util_avg);
3040}
3041
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3042
3043extern unsigned long cpu_util_cfs(int cpu);
3044extern unsigned long cpu_util_cfs_boost(int cpu);
3045
3046static inline unsigned long cpu_util_rt(struct rq *rq)
3047{
3048	return READ_ONCE(rq->avg_rt.util_avg);
3049}
3050#endif
3051
3052#ifdef CONFIG_UCLAMP_TASK
3053unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
3054
3055static inline unsigned long uclamp_rq_get(struct rq *rq,
3056					  enum uclamp_id clamp_id)
3057{
3058	return READ_ONCE(rq->uclamp[clamp_id].value);
3059}
3060
3061static inline void uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id,
3062				 unsigned int value)
3063{
3064	WRITE_ONCE(rq->uclamp[clamp_id].value, value);
3065}
3066
3067static inline bool uclamp_rq_is_idle(struct rq *rq)
3068{
3069	return rq->uclamp_flags & UCLAMP_FLAG_IDLE;
3070}
3071
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3072/* Is the rq being capped/throttled by uclamp_max? */
3073static inline bool uclamp_rq_is_capped(struct rq *rq)
3074{
3075	unsigned long rq_util;
3076	unsigned long max_util;
3077
3078	if (!static_branch_likely(&sched_uclamp_used))
3079		return false;
3080
3081	rq_util = cpu_util_cfs(cpu_of(rq)) + cpu_util_rt(rq);
3082	max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value);
3083
3084	return max_util != SCHED_CAPACITY_SCALE && rq_util >= max_util;
3085}
3086
3087/*
3088 * When uclamp is compiled in, the aggregation at rq level is 'turned off'
3089 * by default in the fast path and only gets turned on once userspace performs
3090 * an operation that requires it.
3091 *
3092 * Returns true if userspace opted-in to use uclamp and aggregation at rq level
3093 * hence is active.
3094 */
3095static inline bool uclamp_is_used(void)
3096{
3097	return static_branch_likely(&sched_uclamp_used);
3098}
3099#else /* CONFIG_UCLAMP_TASK */
3100static inline unsigned long uclamp_eff_value(struct task_struct *p,
3101					     enum uclamp_id clamp_id)
3102{
3103	if (clamp_id == UCLAMP_MIN)
3104		return 0;
3105
3106	return SCHED_CAPACITY_SCALE;
3107}
3108
 
 
 
 
 
 
 
3109static inline bool uclamp_rq_is_capped(struct rq *rq) { return false; }
3110
3111static inline bool uclamp_is_used(void)
3112{
3113	return false;
3114}
3115
3116static inline unsigned long uclamp_rq_get(struct rq *rq,
3117					  enum uclamp_id clamp_id)
3118{
3119	if (clamp_id == UCLAMP_MIN)
3120		return 0;
3121
3122	return SCHED_CAPACITY_SCALE;
3123}
3124
3125static inline void uclamp_rq_set(struct rq *rq, enum uclamp_id clamp_id,
3126				 unsigned int value)
3127{
3128}
3129
3130static inline bool uclamp_rq_is_idle(struct rq *rq)
3131{
3132	return false;
3133}
3134#endif /* CONFIG_UCLAMP_TASK */
3135
3136#ifdef CONFIG_HAVE_SCHED_AVG_IRQ
3137static inline unsigned long cpu_util_irq(struct rq *rq)
3138{
3139	return rq->avg_irq.util_avg;
3140}
3141
3142static inline
3143unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3144{
3145	util *= (max - irq);
3146	util /= max;
3147
3148	return util;
3149
3150}
3151#else
3152static inline unsigned long cpu_util_irq(struct rq *rq)
3153{
3154	return 0;
3155}
3156
3157static inline
3158unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
3159{
3160	return util;
3161}
3162#endif
3163
3164#if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
3165
3166#define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
3167
3168DECLARE_STATIC_KEY_FALSE(sched_energy_present);
3169
3170static inline bool sched_energy_enabled(void)
3171{
3172	return static_branch_unlikely(&sched_energy_present);
3173}
3174
3175extern struct cpufreq_governor schedutil_gov;
3176
3177#else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
3178
3179#define perf_domain_span(pd) NULL
3180static inline bool sched_energy_enabled(void) { return false; }
3181
3182#endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
3183
3184#ifdef CONFIG_MEMBARRIER
3185/*
3186 * The scheduler provides memory barriers required by membarrier between:
3187 * - prior user-space memory accesses and store to rq->membarrier_state,
3188 * - store to rq->membarrier_state and following user-space memory accesses.
3189 * In the same way it provides those guarantees around store to rq->curr.
3190 */
3191static inline void membarrier_switch_mm(struct rq *rq,
3192					struct mm_struct *prev_mm,
3193					struct mm_struct *next_mm)
3194{
3195	int membarrier_state;
3196
3197	if (prev_mm == next_mm)
3198		return;
3199
3200	membarrier_state = atomic_read(&next_mm->membarrier_state);
3201	if (READ_ONCE(rq->membarrier_state) == membarrier_state)
3202		return;
3203
3204	WRITE_ONCE(rq->membarrier_state, membarrier_state);
3205}
3206#else
3207static inline void membarrier_switch_mm(struct rq *rq,
3208					struct mm_struct *prev_mm,
3209					struct mm_struct *next_mm)
3210{
3211}
3212#endif
3213
3214#ifdef CONFIG_SMP
3215static inline bool is_per_cpu_kthread(struct task_struct *p)
3216{
3217	if (!(p->flags & PF_KTHREAD))
3218		return false;
3219
3220	if (p->nr_cpus_allowed != 1)
3221		return false;
3222
3223	return true;
3224}
3225#endif
3226
3227extern void swake_up_all_locked(struct swait_queue_head *q);
3228extern void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait);
3229
3230extern int try_to_wake_up(struct task_struct *tsk, unsigned int state, int wake_flags);
3231
3232#ifdef CONFIG_PREEMPT_DYNAMIC
3233extern int preempt_dynamic_mode;
3234extern int sched_dynamic_mode(const char *str);
3235extern void sched_dynamic_update(int mode);
3236#endif
3237
3238#ifdef CONFIG_SCHED_MM_CID
3239
3240#define SCHED_MM_CID_PERIOD_NS	(100ULL * 1000000)	/* 100ms */
3241#define MM_CID_SCAN_DELAY	100			/* 100ms */
3242
3243extern raw_spinlock_t cid_lock;
3244extern int use_cid_lock;
3245
3246extern void sched_mm_cid_migrate_from(struct task_struct *t);
3247extern void sched_mm_cid_migrate_to(struct rq *dst_rq, struct task_struct *t);
3248extern void task_tick_mm_cid(struct rq *rq, struct task_struct *curr);
3249extern void init_sched_mm_cid(struct task_struct *t);
3250
3251static inline void __mm_cid_put(struct mm_struct *mm, int cid)
3252{
3253	if (cid < 0)
3254		return;
3255	cpumask_clear_cpu(cid, mm_cidmask(mm));
3256}
3257
3258/*
3259 * The per-mm/cpu cid can have the MM_CID_LAZY_PUT flag set or transition to
3260 * the MM_CID_UNSET state without holding the rq lock, but the rq lock needs to
3261 * be held to transition to other states.
3262 *
3263 * State transitions synchronized with cmpxchg or try_cmpxchg need to be
3264 * consistent across cpus, which prevents use of this_cpu_cmpxchg.
3265 */
3266static inline void mm_cid_put_lazy(struct task_struct *t)
3267{
3268	struct mm_struct *mm = t->mm;
3269	struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3270	int cid;
3271
3272	lockdep_assert_irqs_disabled();
3273	cid = __this_cpu_read(pcpu_cid->cid);
3274	if (!mm_cid_is_lazy_put(cid) ||
3275	    !try_cmpxchg(&this_cpu_ptr(pcpu_cid)->cid, &cid, MM_CID_UNSET))
3276		return;
3277	__mm_cid_put(mm, mm_cid_clear_lazy_put(cid));
3278}
3279
3280static inline int mm_cid_pcpu_unset(struct mm_struct *mm)
3281{
3282	struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3283	int cid, res;
3284
3285	lockdep_assert_irqs_disabled();
3286	cid = __this_cpu_read(pcpu_cid->cid);
3287	for (;;) {
3288		if (mm_cid_is_unset(cid))
3289			return MM_CID_UNSET;
3290		/*
3291		 * Attempt transition from valid or lazy-put to unset.
3292		 */
3293		res = cmpxchg(&this_cpu_ptr(pcpu_cid)->cid, cid, MM_CID_UNSET);
3294		if (res == cid)
3295			break;
3296		cid = res;
3297	}
3298	return cid;
3299}
3300
3301static inline void mm_cid_put(struct mm_struct *mm)
3302{
3303	int cid;
3304
3305	lockdep_assert_irqs_disabled();
3306	cid = mm_cid_pcpu_unset(mm);
3307	if (cid == MM_CID_UNSET)
3308		return;
3309	__mm_cid_put(mm, mm_cid_clear_lazy_put(cid));
3310}
3311
3312static inline int __mm_cid_try_get(struct mm_struct *mm)
3313{
3314	struct cpumask *cpumask;
3315	int cid;
3316
3317	cpumask = mm_cidmask(mm);
3318	/*
3319	 * Retry finding first zero bit if the mask is temporarily
3320	 * filled. This only happens during concurrent remote-clear
3321	 * which owns a cid without holding a rq lock.
3322	 */
3323	for (;;) {
3324		cid = cpumask_first_zero(cpumask);
3325		if (cid < nr_cpu_ids)
3326			break;
3327		cpu_relax();
3328	}
3329	if (cpumask_test_and_set_cpu(cid, cpumask))
3330		return -1;
3331	return cid;
3332}
3333
3334/*
3335 * Save a snapshot of the current runqueue time of this cpu
3336 * with the per-cpu cid value, allowing to estimate how recently it was used.
3337 */
3338static inline void mm_cid_snapshot_time(struct rq *rq, struct mm_struct *mm)
3339{
3340	struct mm_cid *pcpu_cid = per_cpu_ptr(mm->pcpu_cid, cpu_of(rq));
3341
3342	lockdep_assert_rq_held(rq);
3343	WRITE_ONCE(pcpu_cid->time, rq->clock);
3344}
3345
3346static inline int __mm_cid_get(struct rq *rq, struct mm_struct *mm)
3347{
3348	int cid;
3349
3350	/*
3351	 * All allocations (even those using the cid_lock) are lock-free. If
3352	 * use_cid_lock is set, hold the cid_lock to perform cid allocation to
3353	 * guarantee forward progress.
3354	 */
3355	if (!READ_ONCE(use_cid_lock)) {
3356		cid = __mm_cid_try_get(mm);
3357		if (cid >= 0)
3358			goto end;
3359		raw_spin_lock(&cid_lock);
3360	} else {
3361		raw_spin_lock(&cid_lock);
3362		cid = __mm_cid_try_get(mm);
3363		if (cid >= 0)
3364			goto unlock;
3365	}
3366
3367	/*
3368	 * cid concurrently allocated. Retry while forcing following
3369	 * allocations to use the cid_lock to ensure forward progress.
3370	 */
3371	WRITE_ONCE(use_cid_lock, 1);
3372	/*
3373	 * Set use_cid_lock before allocation. Only care about program order
3374	 * because this is only required for forward progress.
3375	 */
3376	barrier();
3377	/*
3378	 * Retry until it succeeds. It is guaranteed to eventually succeed once
3379	 * all newcoming allocations observe the use_cid_lock flag set.
3380	 */
3381	do {
3382		cid = __mm_cid_try_get(mm);
3383		cpu_relax();
3384	} while (cid < 0);
3385	/*
3386	 * Allocate before clearing use_cid_lock. Only care about
3387	 * program order because this is for forward progress.
3388	 */
3389	barrier();
3390	WRITE_ONCE(use_cid_lock, 0);
3391unlock:
3392	raw_spin_unlock(&cid_lock);
3393end:
3394	mm_cid_snapshot_time(rq, mm);
3395	return cid;
3396}
3397
3398static inline int mm_cid_get(struct rq *rq, struct mm_struct *mm)
3399{
3400	struct mm_cid __percpu *pcpu_cid = mm->pcpu_cid;
3401	struct cpumask *cpumask;
3402	int cid;
3403
3404	lockdep_assert_rq_held(rq);
3405	cpumask = mm_cidmask(mm);
3406	cid = __this_cpu_read(pcpu_cid->cid);
3407	if (mm_cid_is_valid(cid)) {
3408		mm_cid_snapshot_time(rq, mm);
3409		return cid;
3410	}
3411	if (mm_cid_is_lazy_put(cid)) {
3412		if (try_cmpxchg(&this_cpu_ptr(pcpu_cid)->cid, &cid, MM_CID_UNSET))
3413			__mm_cid_put(mm, mm_cid_clear_lazy_put(cid));
3414	}
3415	cid = __mm_cid_get(rq, mm);
3416	__this_cpu_write(pcpu_cid->cid, cid);
3417	return cid;
3418}
3419
3420static inline void switch_mm_cid(struct rq *rq,
3421				 struct task_struct *prev,
3422				 struct task_struct *next)
3423{
3424	/*
3425	 * Provide a memory barrier between rq->curr store and load of
3426	 * {prev,next}->mm->pcpu_cid[cpu] on rq->curr->mm transition.
3427	 *
3428	 * Should be adapted if context_switch() is modified.
3429	 */
3430	if (!next->mm) {                                // to kernel
3431		/*
3432		 * user -> kernel transition does not guarantee a barrier, but
3433		 * we can use the fact that it performs an atomic operation in
3434		 * mmgrab().
3435		 */
3436		if (prev->mm)                           // from user
3437			smp_mb__after_mmgrab();
3438		/*
3439		 * kernel -> kernel transition does not change rq->curr->mm
3440		 * state. It stays NULL.
3441		 */
3442	} else {                                        // to user
3443		/*
3444		 * kernel -> user transition does not provide a barrier
3445		 * between rq->curr store and load of {prev,next}->mm->pcpu_cid[cpu].
3446		 * Provide it here.
3447		 */
3448		if (!prev->mm)                          // from kernel
3449			smp_mb();
3450		/*
3451		 * user -> user transition guarantees a memory barrier through
3452		 * switch_mm() when current->mm changes. If current->mm is
3453		 * unchanged, no barrier is needed.
3454		 */
3455	}
3456	if (prev->mm_cid_active) {
3457		mm_cid_snapshot_time(rq, prev->mm);
3458		mm_cid_put_lazy(prev);
3459		prev->mm_cid = -1;
3460	}
3461	if (next->mm_cid_active)
3462		next->last_mm_cid = next->mm_cid = mm_cid_get(rq, next->mm);
3463}
3464
3465#else
3466static inline void switch_mm_cid(struct rq *rq, struct task_struct *prev, struct task_struct *next) { }
3467static inline void sched_mm_cid_migrate_from(struct task_struct *t) { }
3468static inline void sched_mm_cid_migrate_to(struct rq *dst_rq, struct task_struct *t) { }
3469static inline void task_tick_mm_cid(struct rq *rq, struct task_struct *curr) { }
3470static inline void init_sched_mm_cid(struct task_struct *t) { }
3471#endif
3472
3473extern u64 avg_vruntime(struct cfs_rq *cfs_rq);
3474extern int entity_eligible(struct cfs_rq *cfs_rq, struct sched_entity *se);
3475
3476#endif /* _KERNEL_SCHED_SCHED_H */