~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/kernel/sched/sched.h

Version: ~ [ linux-5.15-rc5 ] ~ [ linux-5.14.11 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.72 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.152 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.210 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.250 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.286 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.288 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /* SPDX-License-Identifier: GPL-2.0 */
  2 /*
  3  * Scheduler internal types and methods:
  4  */
  5 #include <linux/sched.h>
  6 
  7 #include <linux/sched/autogroup.h>
  8 #include <linux/sched/clock.h>
  9 #include <linux/sched/coredump.h>
 10 #include <linux/sched/cpufreq.h>
 11 #include <linux/sched/cputime.h>
 12 #include <linux/sched/deadline.h>
 13 #include <linux/sched/debug.h>
 14 #include <linux/sched/hotplug.h>
 15 #include <linux/sched/idle.h>
 16 #include <linux/sched/init.h>
 17 #include <linux/sched/isolation.h>
 18 #include <linux/sched/jobctl.h>
 19 #include <linux/sched/loadavg.h>
 20 #include <linux/sched/mm.h>
 21 #include <linux/sched/nohz.h>
 22 #include <linux/sched/numa_balancing.h>
 23 #include <linux/sched/prio.h>
 24 #include <linux/sched/rt.h>
 25 #include <linux/sched/signal.h>
 26 #include <linux/sched/smt.h>
 27 #include <linux/sched/stat.h>
 28 #include <linux/sched/sysctl.h>
 29 #include <linux/sched/task.h>
 30 #include <linux/sched/task_stack.h>
 31 #include <linux/sched/topology.h>
 32 #include <linux/sched/user.h>
 33 #include <linux/sched/wake_q.h>
 34 #include <linux/sched/xacct.h>
 35 
 36 #include <uapi/linux/sched/types.h>
 37 
 38 #include <linux/binfmts.h>
 39 #include <linux/bitops.h>
 40 #include <linux/blkdev.h>
 41 #include <linux/compat.h>
 42 #include <linux/context_tracking.h>
 43 #include <linux/cpufreq.h>
 44 #include <linux/cpuidle.h>
 45 #include <linux/cpuset.h>
 46 #include <linux/ctype.h>
 47 #include <linux/debugfs.h>
 48 #include <linux/delayacct.h>
 49 #include <linux/energy_model.h>
 50 #include <linux/init_task.h>
 51 #include <linux/kprobes.h>
 52 #include <linux/kthread.h>
 53 #include <linux/membarrier.h>
 54 #include <linux/migrate.h>
 55 #include <linux/mmu_context.h>
 56 #include <linux/nmi.h>
 57 #include <linux/proc_fs.h>
 58 #include <linux/prefetch.h>
 59 #include <linux/profile.h>
 60 #include <linux/psi.h>
 61 #include <linux/ratelimit.h>
 62 #include <linux/rcupdate_wait.h>
 63 #include <linux/security.h>
 64 #include <linux/stop_machine.h>
 65 #include <linux/suspend.h>
 66 #include <linux/swait.h>
 67 #include <linux/syscalls.h>
 68 #include <linux/task_work.h>
 69 #include <linux/tsacct_kern.h>
 70 
 71 #include <asm/tlb.h>
 72 
 73 #ifdef CONFIG_PARAVIRT
 74 # include <asm/paravirt.h>
 75 #endif
 76 
 77 #include "cpupri.h"
 78 #include "cpudeadline.h"
 79 
 80 #include <trace/events/sched.h>
 81 
 82 #ifdef CONFIG_SCHED_DEBUG
 83 # define SCHED_WARN_ON(x)       WARN_ONCE(x, #x)
 84 #else
 85 # define SCHED_WARN_ON(x)       ({ (void)(x), 0; })
 86 #endif
 87 
 88 struct rq;
 89 struct cpuidle_state;
 90 
 91 /* task_struct::on_rq states: */
 92 #define TASK_ON_RQ_QUEUED       1
 93 #define TASK_ON_RQ_MIGRATING    2
 94 
 95 extern __read_mostly int scheduler_running;
 96 
 97 extern unsigned long calc_load_update;
 98 extern atomic_long_t calc_load_tasks;
 99 
100 extern void calc_global_load_tick(struct rq *this_rq);
101 extern long calc_load_fold_active(struct rq *this_rq, long adjust);
102 
103 extern void call_trace_sched_update_nr_running(struct rq *rq, int count);
104 /*
105  * Helpers for converting nanosecond timing to jiffy resolution
106  */
107 #define NS_TO_JIFFIES(TIME)     ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
108 
109 /*
110  * Increase resolution of nice-level calculations for 64-bit architectures.
111  * The extra resolution improves shares distribution and load balancing of
112  * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
113  * hierarchies, especially on larger systems. This is not a user-visible change
114  * and does not change the user-interface for setting shares/weights.
115  *
116  * We increase resolution only if we have enough bits to allow this increased
117  * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
118  * are pretty high and the returns do not justify the increased costs.
119  *
120  * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
121  * increase coverage and consistency always enable it on 64-bit platforms.
122  */
123 #ifdef CONFIG_64BIT
124 # define NICE_0_LOAD_SHIFT      (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
125 # define scale_load(w)          ((w) << SCHED_FIXEDPOINT_SHIFT)
126 # define scale_load_down(w) \
127 ({ \
128         unsigned long __w = (w); \
129         if (__w) \
130                 __w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \
131         __w; \
132 })
133 #else
134 # define NICE_0_LOAD_SHIFT      (SCHED_FIXEDPOINT_SHIFT)
135 # define scale_load(w)          (w)
136 # define scale_load_down(w)     (w)
137 #endif
138 
139 /*
140  * Task weight (visible to users) and its load (invisible to users) have
141  * independent resolution, but they should be well calibrated. We use
142  * scale_load() and scale_load_down(w) to convert between them. The
143  * following must be true:
144  *
145  *  scale_load(sched_prio_to_weight[NICE_TO_PRIO(0)-MAX_RT_PRIO]) == NICE_0_LOAD
146  *
147  */
148 #define NICE_0_LOAD             (1L << NICE_0_LOAD_SHIFT)
149 
150 /*
151  * Single value that decides SCHED_DEADLINE internal math precision.
152  * 10 -> just above 1us
153  * 9  -> just above 0.5us
154  */
155 #define DL_SCALE                10
156 
157 /*
158  * Single value that denotes runtime == period, ie unlimited time.
159  */
160 #define RUNTIME_INF             ((u64)~0ULL)
161 
162 static inline int idle_policy(int policy)
163 {
164         return policy == SCHED_IDLE;
165 }
166 static inline int fair_policy(int policy)
167 {
168         return policy == SCHED_NORMAL || policy == SCHED_BATCH;
169 }
170 
171 static inline int rt_policy(int policy)
172 {
173         return policy == SCHED_FIFO || policy == SCHED_RR;
174 }
175 
176 static inline int dl_policy(int policy)
177 {
178         return policy == SCHED_DEADLINE;
179 }
180 static inline bool valid_policy(int policy)
181 {
182         return idle_policy(policy) || fair_policy(policy) ||
183                 rt_policy(policy) || dl_policy(policy);
184 }
185 
186 static inline int task_has_idle_policy(struct task_struct *p)
187 {
188         return idle_policy(p->policy);
189 }
190 
191 static inline int task_has_rt_policy(struct task_struct *p)
192 {
193         return rt_policy(p->policy);
194 }
195 
196 static inline int task_has_dl_policy(struct task_struct *p)
197 {
198         return dl_policy(p->policy);
199 }
200 
201 #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
202 
203 static inline void update_avg(u64 *avg, u64 sample)
204 {
205         s64 diff = sample - *avg;
206         *avg += diff / 8;
207 }
208 
209 /*
210  * Shifting a value by an exponent greater *or equal* to the size of said value
211  * is UB; cap at size-1.
212  */
213 #define shr_bound(val, shift)                                                   \
214         (val >> min_t(typeof(shift), shift, BITS_PER_TYPE(typeof(val)) - 1))
215 
216 /*
217  * !! For sched_setattr_nocheck() (kernel) only !!
218  *
219  * This is actually gross. :(
220  *
221  * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
222  * tasks, but still be able to sleep. We need this on platforms that cannot
223  * atomically change clock frequency. Remove once fast switching will be
224  * available on such platforms.
225  *
226  * SUGOV stands for SchedUtil GOVernor.
227  */
228 #define SCHED_FLAG_SUGOV        0x10000000
229 
230 #define SCHED_DL_FLAGS (SCHED_FLAG_RECLAIM | SCHED_FLAG_DL_OVERRUN | SCHED_FLAG_SUGOV)
231 
232 static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se)
233 {
234 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
235         return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
236 #else
237         return false;
238 #endif
239 }
240 
241 /*
242  * Tells if entity @a should preempt entity @b.
243  */
244 static inline bool
245 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
246 {
247         return dl_entity_is_special(a) ||
248                dl_time_before(a->deadline, b->deadline);
249 }
250 
251 /*
252  * This is the priority-queue data structure of the RT scheduling class:
253  */
254 struct rt_prio_array {
255         DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
256         struct list_head queue[MAX_RT_PRIO];
257 };
258 
259 struct rt_bandwidth {
260         /* nests inside the rq lock: */
261         raw_spinlock_t          rt_runtime_lock;
262         ktime_t                 rt_period;
263         u64                     rt_runtime;
264         struct hrtimer          rt_period_timer;
265         unsigned int            rt_period_active;
266 };
267 
268 void __dl_clear_params(struct task_struct *p);
269 
270 struct dl_bandwidth {
271         raw_spinlock_t          dl_runtime_lock;
272         u64                     dl_runtime;
273         u64                     dl_period;
274 };
275 
276 static 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  */
299 struct dl_bw {
300         raw_spinlock_t          lock;
301         u64                     bw;
302         u64                     total_bw;
303 };
304 
305 static inline void __dl_update(struct dl_bw *dl_b, s64 bw);
306 
307 static inline
308 void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
309 {
310         dl_b->total_bw -= tsk_bw;
311         __dl_update(dl_b, (s32)tsk_bw / cpus);
312 }
313 
314 static inline
315 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
316 {
317         dl_b->total_bw += tsk_bw;
318         __dl_update(dl_b, -((s32)tsk_bw / cpus));
319 }
320 
321 static inline bool __dl_overflow(struct dl_bw *dl_b, unsigned long cap,
322                                  u64 old_bw, u64 new_bw)
323 {
324         return dl_b->bw != -1 &&
325                cap_scale(dl_b->bw, cap) < dl_b->total_bw - old_bw + new_bw;
326 }
327 
328 /*
329  * Verify the fitness of task @p to run on @cpu taking into account the
330  * CPU original capacity and the runtime/deadline ratio of the task.
331  *
332  * The function will return true if the CPU original capacity of the
333  * @cpu scaled by SCHED_CAPACITY_SCALE >= runtime/deadline ratio of the
334  * task and false otherwise.
335  */
336 static inline bool dl_task_fits_capacity(struct task_struct *p, int cpu)
337 {
338         unsigned long cap = arch_scale_cpu_capacity(cpu);
339 
340         return cap_scale(p->dl.dl_deadline, cap) >= p->dl.dl_runtime;
341 }
342 
343 extern void init_dl_bw(struct dl_bw *dl_b);
344 extern int  sched_dl_global_validate(void);
345 extern void sched_dl_do_global(void);
346 extern int  sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
347 extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
348 extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
349 extern bool __checkparam_dl(const struct sched_attr *attr);
350 extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
351 extern int  dl_task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed);
352 extern int  dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
353 extern bool dl_cpu_busy(unsigned int cpu);
354 
355 #ifdef CONFIG_CGROUP_SCHED
356 
357 #include <linux/cgroup.h>
358 #include <linux/psi.h>
359 
360 struct cfs_rq;
361 struct rt_rq;
362 
363 extern struct list_head task_groups;
364 
365 struct cfs_bandwidth {
366 #ifdef CONFIG_CFS_BANDWIDTH
367         raw_spinlock_t          lock;
368         ktime_t                 period;
369         u64                     quota;
370         u64                     runtime;
371         u64                     burst;
372         s64                     hierarchical_quota;
373 
374         u8                      idle;
375         u8                      period_active;
376         u8                      slack_started;
377         struct hrtimer          period_timer;
378         struct hrtimer          slack_timer;
379         struct list_head        throttled_cfs_rq;
380 
381         /* Statistics: */
382         int                     nr_periods;
383         int                     nr_throttled;
384         u64                     throttled_time;
385 #endif
386 };
387 
388 /* Task group related information */
389 struct task_group {
390         struct cgroup_subsys_state css;
391 
392 #ifdef CONFIG_FAIR_GROUP_SCHED
393         /* schedulable entities of this group on each CPU */
394         struct sched_entity     **se;
395         /* runqueue "owned" by this group on each CPU */
396         struct cfs_rq           **cfs_rq;
397         unsigned long           shares;
398 
399         /* A positive value indicates that this is a SCHED_IDLE group. */
400         int                     idle;
401 
402 #ifdef  CONFIG_SMP
403         /*
404          * load_avg can be heavily contended at clock tick time, so put
405          * it in its own cacheline separated from the fields above which
406          * will also be accessed at each tick.
407          */
408         atomic_long_t           load_avg ____cacheline_aligned;
409 #endif
410 #endif
411 
412 #ifdef CONFIG_RT_GROUP_SCHED
413         struct sched_rt_entity  **rt_se;
414         struct rt_rq            **rt_rq;
415 
416         struct rt_bandwidth     rt_bandwidth;
417 #endif
418 
419         struct rcu_head         rcu;
420         struct list_head        list;
421 
422         struct task_group       *parent;
423         struct list_head        siblings;
424         struct list_head        children;
425 
426 #ifdef CONFIG_SCHED_AUTOGROUP
427         struct autogroup        *autogroup;
428 #endif
429 
430         struct cfs_bandwidth    cfs_bandwidth;
431 
432 #ifdef CONFIG_UCLAMP_TASK_GROUP
433         /* The two decimal precision [%] value requested from user-space */
434         unsigned int            uclamp_pct[UCLAMP_CNT];
435         /* Clamp values requested for a task group */
436         struct uclamp_se        uclamp_req[UCLAMP_CNT];
437         /* Effective clamp values used for a task group */
438         struct uclamp_se        uclamp[UCLAMP_CNT];
439 #endif
440 
441 };
442 
443 #ifdef CONFIG_FAIR_GROUP_SCHED
444 #define ROOT_TASK_GROUP_LOAD    NICE_0_LOAD
445 
446 /*
447  * A weight of 0 or 1 can cause arithmetics problems.
448  * A weight of a cfs_rq is the sum of weights of which entities
449  * are queued on this cfs_rq, so a weight of a entity should not be
450  * too large, so as the shares value of a task group.
451  * (The default weight is 1024 - so there's no practical
452  *  limitation from this.)
453  */
454 #define MIN_SHARES              (1UL <<  1)
455 #define MAX_SHARES              (1UL << 18)
456 #endif
457 
458 typedef int (*tg_visitor)(struct task_group *, void *);
459 
460 extern int walk_tg_tree_from(struct task_group *from,
461                              tg_visitor down, tg_visitor up, void *data);
462 
463 /*
464  * Iterate the full tree, calling @down when first entering a node and @up when
465  * leaving it for the final time.
466  *
467  * Caller must hold rcu_lock or sufficient equivalent.
468  */
469 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
470 {
471         return walk_tg_tree_from(&root_task_group, down, up, data);
472 }
473 
474 extern int tg_nop(struct task_group *tg, void *data);
475 
476 extern void free_fair_sched_group(struct task_group *tg);
477 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
478 extern void online_fair_sched_group(struct task_group *tg);
479 extern void unregister_fair_sched_group(struct task_group *tg);
480 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
481                         struct sched_entity *se, int cpu,
482                         struct sched_entity *parent);
483 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
484 
485 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
486 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
487 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
488 
489 extern void free_rt_sched_group(struct task_group *tg);
490 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
491 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
492                 struct sched_rt_entity *rt_se, int cpu,
493                 struct sched_rt_entity *parent);
494 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
495 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
496 extern long sched_group_rt_runtime(struct task_group *tg);
497 extern long sched_group_rt_period(struct task_group *tg);
498 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
499 
500 extern struct task_group *sched_create_group(struct task_group *parent);
501 extern void sched_online_group(struct task_group *tg,
502                                struct task_group *parent);
503 extern void sched_destroy_group(struct task_group *tg);
504 extern void sched_offline_group(struct task_group *tg);
505 
506 extern void sched_move_task(struct task_struct *tsk);
507 
508 #ifdef CONFIG_FAIR_GROUP_SCHED
509 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
510 
511 extern int sched_group_set_idle(struct task_group *tg, long idle);
512 
513 #ifdef CONFIG_SMP
514 extern void set_task_rq_fair(struct sched_entity *se,
515                              struct cfs_rq *prev, struct cfs_rq *next);
516 #else /* !CONFIG_SMP */
517 static inline void set_task_rq_fair(struct sched_entity *se,
518                              struct cfs_rq *prev, struct cfs_rq *next) { }
519 #endif /* CONFIG_SMP */
520 #endif /* CONFIG_FAIR_GROUP_SCHED */
521 
522 #else /* CONFIG_CGROUP_SCHED */
523 
524 struct cfs_bandwidth { };
525 
526 #endif  /* CONFIG_CGROUP_SCHED */
527 
528 /* CFS-related fields in a runqueue */
529 struct cfs_rq {
530         struct load_weight      load;
531         unsigned int            nr_running;
532         unsigned int            h_nr_running;      /* SCHED_{NORMAL,BATCH,IDLE} */
533         unsigned int            idle_h_nr_running; /* SCHED_IDLE */
534 
535         u64                     exec_clock;
536         u64                     min_vruntime;
537 #ifdef CONFIG_SCHED_CORE
538         unsigned int            forceidle_seq;
539         u64                     min_vruntime_fi;
540 #endif
541 
542 #ifndef CONFIG_64BIT
543         u64                     min_vruntime_copy;
544 #endif
545 
546         struct rb_root_cached   tasks_timeline;
547 
548         /*
549          * 'curr' points to currently running entity on this cfs_rq.
550          * It is set to NULL otherwise (i.e when none are currently running).
551          */
552         struct sched_entity     *curr;
553         struct sched_entity     *next;
554         struct sched_entity     *last;
555         struct sched_entity     *skip;
556 
557 #ifdef  CONFIG_SCHED_DEBUG
558         unsigned int            nr_spread_over;
559 #endif
560 
561 #ifdef CONFIG_SMP
562         /*
563          * CFS load tracking
564          */
565         struct sched_avg        avg;
566 #ifndef CONFIG_64BIT
567         u64                     load_last_update_time_copy;
568 #endif
569         struct {
570                 raw_spinlock_t  lock ____cacheline_aligned;
571                 int             nr;
572                 unsigned long   load_avg;
573                 unsigned long   util_avg;
574                 unsigned long   runnable_avg;
575         } removed;
576 
577 #ifdef CONFIG_FAIR_GROUP_SCHED
578         unsigned long           tg_load_avg_contrib;
579         long                    propagate;
580         long                    prop_runnable_sum;
581 
582         /*
583          *   h_load = weight * f(tg)
584          *
585          * Where f(tg) is the recursive weight fraction assigned to
586          * this group.
587          */
588         unsigned long           h_load;
589         u64                     last_h_load_update;
590         struct sched_entity     *h_load_next;
591 #endif /* CONFIG_FAIR_GROUP_SCHED */
592 #endif /* CONFIG_SMP */
593 
594 #ifdef CONFIG_FAIR_GROUP_SCHED
595         struct rq               *rq;    /* CPU runqueue to which this cfs_rq is attached */
596 
597         /*
598          * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
599          * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
600          * (like users, containers etc.)
601          *
602          * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
603          * This list is used during load balance.
604          */
605         int                     on_list;
606         struct list_head        leaf_cfs_rq_list;
607         struct task_group       *tg;    /* group that "owns" this runqueue */
608 
609         /* Locally cached copy of our task_group's idle value */
610         int                     idle;
611 
612 #ifdef CONFIG_CFS_BANDWIDTH
613         int                     runtime_enabled;
614         s64                     runtime_remaining;
615 
616         u64                     throttled_clock;
617         u64                     throttled_clock_task;
618         u64                     throttled_clock_task_time;
619         int                     throttled;
620         int                     throttle_count;
621         struct list_head        throttled_list;
622 #endif /* CONFIG_CFS_BANDWIDTH */
623 #endif /* CONFIG_FAIR_GROUP_SCHED */
624 };
625 
626 static inline int rt_bandwidth_enabled(void)
627 {
628         return sysctl_sched_rt_runtime >= 0;
629 }
630 
631 /* RT IPI pull logic requires IRQ_WORK */
632 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
633 # define HAVE_RT_PUSH_IPI
634 #endif
635 
636 /* Real-Time classes' related field in a runqueue: */
637 struct rt_rq {
638         struct rt_prio_array    active;
639         unsigned int            rt_nr_running;
640         unsigned int            rr_nr_running;
641 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
642         struct {
643                 int             curr; /* highest queued rt task prio */
644 #ifdef CONFIG_SMP
645                 int             next; /* next highest */
646 #endif
647         } highest_prio;
648 #endif
649 #ifdef CONFIG_SMP
650         unsigned int            rt_nr_migratory;
651         unsigned int            rt_nr_total;
652         int                     overloaded;
653         struct plist_head       pushable_tasks;
654 
655 #endif /* CONFIG_SMP */
656         int                     rt_queued;
657 
658         int                     rt_throttled;
659         u64                     rt_time;
660         u64                     rt_runtime;
661         /* Nests inside the rq lock: */
662         raw_spinlock_t          rt_runtime_lock;
663 
664 #ifdef CONFIG_RT_GROUP_SCHED
665         unsigned int            rt_nr_boosted;
666 
667         struct rq               *rq;
668         struct task_group       *tg;
669 #endif
670 };
671 
672 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
673 {
674         return rt_rq->rt_queued && rt_rq->rt_nr_running;
675 }
676 
677 /* Deadline class' related fields in a runqueue */
678 struct dl_rq {
679         /* runqueue is an rbtree, ordered by deadline */
680         struct rb_root_cached   root;
681 
682         unsigned int            dl_nr_running;
683 
684 #ifdef CONFIG_SMP
685         /*
686          * Deadline values of the currently executing and the
687          * earliest ready task on this rq. Caching these facilitates
688          * the decision whether or not a ready but not running task
689          * should migrate somewhere else.
690          */
691         struct {
692                 u64             curr;
693                 u64             next;
694         } earliest_dl;
695 
696         unsigned int            dl_nr_migratory;
697         int                     overloaded;
698 
699         /*
700          * Tasks on this rq that can be pushed away. They are kept in
701          * an rb-tree, ordered by tasks' deadlines, with caching
702          * of the leftmost (earliest deadline) element.
703          */
704         struct rb_root_cached   pushable_dl_tasks_root;
705 #else
706         struct dl_bw            dl_bw;
707 #endif
708         /*
709          * "Active utilization" for this runqueue: increased when a
710          * task wakes up (becomes TASK_RUNNING) and decreased when a
711          * task blocks
712          */
713         u64                     running_bw;
714 
715         /*
716          * Utilization of the tasks "assigned" to this runqueue (including
717          * the tasks that are in runqueue and the tasks that executed on this
718          * CPU and blocked). Increased when a task moves to this runqueue, and
719          * decreased when the task moves away (migrates, changes scheduling
720          * policy, or terminates).
721          * This is needed to compute the "inactive utilization" for the
722          * runqueue (inactive utilization = this_bw - running_bw).
723          */
724         u64                     this_bw;
725         u64                     extra_bw;
726 
727         /*
728          * Inverse of the fraction of CPU utilization that can be reclaimed
729          * by the GRUB algorithm.
730          */
731         u64                     bw_ratio;
732 };
733 
734 #ifdef CONFIG_FAIR_GROUP_SCHED
735 /* An entity is a task if it doesn't "own" a runqueue */
736 #define entity_is_task(se)      (!se->my_q)
737 
738 static inline void se_update_runnable(struct sched_entity *se)
739 {
740         if (!entity_is_task(se))
741                 se->runnable_weight = se->my_q->h_nr_running;
742 }
743 
744 static inline long se_runnable(struct sched_entity *se)
745 {
746         if (entity_is_task(se))
747                 return !!se->on_rq;
748         else
749                 return se->runnable_weight;
750 }
751 
752 #else
753 #define entity_is_task(se)      1
754 
755 static inline void se_update_runnable(struct sched_entity *se) {}
756 
757 static inline long se_runnable(struct sched_entity *se)
758 {
759         return !!se->on_rq;
760 }
761 #endif
762 
763 #ifdef CONFIG_SMP
764 /*
765  * XXX we want to get rid of these helpers and use the full load resolution.
766  */
767 static inline long se_weight(struct sched_entity *se)
768 {
769         return scale_load_down(se->load.weight);
770 }
771 
772 
773 static inline bool sched_asym_prefer(int a, int b)
774 {
775         return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
776 }
777 
778 struct perf_domain {
779         struct em_perf_domain *em_pd;
780         struct perf_domain *next;
781         struct rcu_head rcu;
782 };
783 
784 /* Scheduling group status flags */
785 #define SG_OVERLOAD             0x1 /* More than one runnable task on a CPU. */
786 #define SG_OVERUTILIZED         0x2 /* One or more CPUs are over-utilized. */
787 
788 /*
789  * We add the notion of a root-domain which will be used to define per-domain
790  * variables. Each exclusive cpuset essentially defines an island domain by
791  * fully partitioning the member CPUs from any other cpuset. Whenever a new
792  * exclusive cpuset is created, we also create and attach a new root-domain
793  * object.
794  *
795  */
796 struct root_domain {
797         atomic_t                refcount;
798         atomic_t                rto_count;
799         struct rcu_head         rcu;
800         cpumask_var_t           span;
801         cpumask_var_t           online;
802 
803         /*
804          * Indicate pullable load on at least one CPU, e.g:
805          * - More than one runnable task
806          * - Running task is misfit
807          */
808         int                     overload;
809 
810         /* Indicate one or more cpus over-utilized (tipping point) */
811         int                     overutilized;
812 
813         /*
814          * The bit corresponding to a CPU gets set here if such CPU has more
815          * than one runnable -deadline task (as it is below for RT tasks).
816          */
817         cpumask_var_t           dlo_mask;
818         atomic_t                dlo_count;
819         struct dl_bw            dl_bw;
820         struct cpudl            cpudl;
821 
822         /*
823          * Indicate whether a root_domain's dl_bw has been checked or
824          * updated. It's monotonously increasing value.
825          *
826          * Also, some corner cases, like 'wrap around' is dangerous, but given
827          * that u64 is 'big enough'. So that shouldn't be a concern.
828          */
829         u64 visit_gen;
830 
831 #ifdef HAVE_RT_PUSH_IPI
832         /*
833          * For IPI pull requests, loop across the rto_mask.
834          */
835         struct irq_work         rto_push_work;
836         raw_spinlock_t          rto_lock;
837         /* These are only updated and read within rto_lock */
838         int                     rto_loop;
839         int                     rto_cpu;
840         /* These atomics are updated outside of a lock */
841         atomic_t                rto_loop_next;
842         atomic_t                rto_loop_start;
843 #endif
844         /*
845          * The "RT overload" flag: it gets set if a CPU has more than
846          * one runnable RT task.
847          */
848         cpumask_var_t           rto_mask;
849         struct cpupri           cpupri;
850 
851         unsigned long           max_cpu_capacity;
852 
853         /*
854          * NULL-terminated list of performance domains intersecting with the
855          * CPUs of the rd. Protected by RCU.
856          */
857         struct perf_domain __rcu *pd;
858 };
859 
860 extern void init_defrootdomain(void);
861 extern int sched_init_domains(const struct cpumask *cpu_map);
862 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
863 extern void sched_get_rd(struct root_domain *rd);
864 extern void sched_put_rd(struct root_domain *rd);
865 
866 #ifdef HAVE_RT_PUSH_IPI
867 extern void rto_push_irq_work_func(struct irq_work *work);
868 #endif
869 #endif /* CONFIG_SMP */
870 
871 #ifdef CONFIG_UCLAMP_TASK
872 /*
873  * struct uclamp_bucket - Utilization clamp bucket
874  * @value: utilization clamp value for tasks on this clamp bucket
875  * @tasks: number of RUNNABLE tasks on this clamp bucket
876  *
877  * Keep track of how many tasks are RUNNABLE for a given utilization
878  * clamp value.
879  */
880 struct uclamp_bucket {
881         unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
882         unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
883 };
884 
885 /*
886  * struct uclamp_rq - rq's utilization clamp
887  * @value: currently active clamp values for a rq
888  * @bucket: utilization clamp buckets affecting a rq
889  *
890  * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
891  * A clamp value is affecting a rq when there is at least one task RUNNABLE
892  * (or actually running) with that value.
893  *
894  * There are up to UCLAMP_CNT possible different clamp values, currently there
895  * are only two: minimum utilization and maximum utilization.
896  *
897  * All utilization clamping values are MAX aggregated, since:
898  * - for util_min: we want to run the CPU at least at the max of the minimum
899  *   utilization required by its currently RUNNABLE tasks.
900  * - for util_max: we want to allow the CPU to run up to the max of the
901  *   maximum utilization allowed by its currently RUNNABLE tasks.
902  *
903  * Since on each system we expect only a limited number of different
904  * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
905  * the metrics required to compute all the per-rq utilization clamp values.
906  */
907 struct uclamp_rq {
908         unsigned int value;
909         struct uclamp_bucket bucket[UCLAMP_BUCKETS];
910 };
911 
912 DECLARE_STATIC_KEY_FALSE(sched_uclamp_used);
913 #endif /* CONFIG_UCLAMP_TASK */
914 
915 /*
916  * This is the main, per-CPU runqueue data structure.
917  *
918  * Locking rule: those places that want to lock multiple runqueues
919  * (such as the load balancing or the thread migration code), lock
920  * acquire operations must be ordered by ascending &runqueue.
921  */
922 struct rq {
923         /* runqueue lock: */
924         raw_spinlock_t          __lock;
925 
926         /*
927          * nr_running and cpu_load should be in the same cacheline because
928          * remote CPUs use both these fields when doing load calculation.
929          */
930         unsigned int            nr_running;
931 #ifdef CONFIG_NUMA_BALANCING
932         unsigned int            nr_numa_running;
933         unsigned int            nr_preferred_running;
934         unsigned int            numa_migrate_on;
935 #endif
936 #ifdef CONFIG_NO_HZ_COMMON
937 #ifdef CONFIG_SMP
938         unsigned long           last_blocked_load_update_tick;
939         unsigned int            has_blocked_load;
940         call_single_data_t      nohz_csd;
941 #endif /* CONFIG_SMP */
942         unsigned int            nohz_tick_stopped;
943         atomic_t                nohz_flags;
944 #endif /* CONFIG_NO_HZ_COMMON */
945 
946 #ifdef CONFIG_SMP
947         unsigned int            ttwu_pending;
948 #endif
949         u64                     nr_switches;
950 
951 #ifdef CONFIG_UCLAMP_TASK
952         /* Utilization clamp values based on CPU's RUNNABLE tasks */
953         struct uclamp_rq        uclamp[UCLAMP_CNT] ____cacheline_aligned;
954         unsigned int            uclamp_flags;
955 #define UCLAMP_FLAG_IDLE 0x01
956 #endif
957 
958         struct cfs_rq           cfs;
959         struct rt_rq            rt;
960         struct dl_rq            dl;
961 
962 #ifdef CONFIG_FAIR_GROUP_SCHED
963         /* list of leaf cfs_rq on this CPU: */
964         struct list_head        leaf_cfs_rq_list;
965         struct list_head        *tmp_alone_branch;
966 #endif /* CONFIG_FAIR_GROUP_SCHED */
967 
968         /*
969          * This is part of a global counter where only the total sum
970          * over all CPUs matters. A task can increase this counter on
971          * one CPU and if it got migrated afterwards it may decrease
972          * it on another CPU. Always updated under the runqueue lock:
973          */
974         unsigned int            nr_uninterruptible;
975 
976         struct task_struct __rcu        *curr;
977         struct task_struct      *idle;
978         struct task_struct      *stop;
979         unsigned long           next_balance;
980         struct mm_struct        *prev_mm;
981 
982         unsigned int            clock_update_flags;
983         u64                     clock;
984         /* Ensure that all clocks are in the same cache line */
985         u64                     clock_task ____cacheline_aligned;
986         u64                     clock_pelt;
987         unsigned long           lost_idle_time;
988 
989         atomic_t                nr_iowait;
990 
991 #ifdef CONFIG_SCHED_DEBUG
992         u64 last_seen_need_resched_ns;
993         int ticks_without_resched;
994 #endif
995 
996 #ifdef CONFIG_MEMBARRIER
997         int membarrier_state;
998 #endif
999 
1000 #ifdef CONFIG_SMP
1001         struct root_domain              *rd;
1002         struct sched_domain __rcu       *sd;
1003 
1004         unsigned long           cpu_capacity;
1005         unsigned long           cpu_capacity_orig;
1006 
1007         struct callback_head    *balance_callback;
1008 
1009         unsigned char           nohz_idle_balance;
1010         unsigned char           idle_balance;
1011 
1012         unsigned long           misfit_task_load;
1013 
1014         /* For active balancing */
1015         int                     active_balance;
1016         int                     push_cpu;
1017         struct cpu_stop_work    active_balance_work;
1018 
1019         /* CPU of this runqueue: */
1020         int                     cpu;
1021         int                     online;
1022 
1023         struct list_head cfs_tasks;
1024 
1025         struct sched_avg        avg_rt;
1026         struct sched_avg        avg_dl;
1027 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
1028         struct sched_avg        avg_irq;
1029 #endif
1030 #ifdef CONFIG_SCHED_THERMAL_PRESSURE
1031         struct sched_avg        avg_thermal;
1032 #endif
1033         u64                     idle_stamp;
1034         u64                     avg_idle;
1035 
1036         unsigned long           wake_stamp;
1037         u64                     wake_avg_idle;
1038 
1039         /* This is used to determine avg_idle's max value */
1040         u64                     max_idle_balance_cost;
1041 
1042 #ifdef CONFIG_HOTPLUG_CPU
1043         struct rcuwait          hotplug_wait;
1044 #endif
1045 #endif /* CONFIG_SMP */
1046 
1047 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1048         u64                     prev_irq_time;
1049 #endif
1050 #ifdef CONFIG_PARAVIRT
1051         u64                     prev_steal_time;
1052 #endif
1053 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
1054         u64                     prev_steal_time_rq;
1055 #endif
1056 
1057         /* calc_load related fields */
1058         unsigned long           calc_load_update;
1059         long                    calc_load_active;
1060 
1061 #ifdef CONFIG_SCHED_HRTICK
1062 #ifdef CONFIG_SMP
1063         call_single_data_t      hrtick_csd;
1064 #endif
1065         struct hrtimer          hrtick_timer;
1066         ktime_t                 hrtick_time;
1067 #endif
1068 
1069 #ifdef CONFIG_SCHEDSTATS
1070         /* latency stats */
1071         struct sched_info       rq_sched_info;
1072         unsigned long long      rq_cpu_time;
1073         /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
1074 
1075         /* sys_sched_yield() stats */
1076         unsigned int            yld_count;
1077 
1078         /* schedule() stats */
1079         unsigned int            sched_count;
1080         unsigned int            sched_goidle;
1081 
1082         /* try_to_wake_up() stats */
1083         unsigned int            ttwu_count;
1084         unsigned int            ttwu_local;
1085 #endif
1086 
1087 #ifdef CONFIG_CPU_IDLE
1088         /* Must be inspected within a rcu lock section */
1089         struct cpuidle_state    *idle_state;
1090 #endif
1091 
1092 #ifdef CONFIG_SMP
1093         unsigned int            nr_pinned;
1094 #endif
1095         unsigned int            push_busy;
1096         struct cpu_stop_work    push_work;
1097 
1098 #ifdef CONFIG_SCHED_CORE
1099         /* per rq */
1100         struct rq               *core;
1101         struct task_struct      *core_pick;
1102         unsigned int            core_enabled;
1103         unsigned int            core_sched_seq;
1104         struct rb_root          core_tree;
1105 
1106         /* shared state -- careful with sched_core_cpu_deactivate() */
1107         unsigned int            core_task_seq;
1108         unsigned int            core_pick_seq;
1109         unsigned long           core_cookie;
1110         unsigned char           core_forceidle;
1111         unsigned int            core_forceidle_seq;
1112 #endif
1113 };
1114 
1115 #ifdef CONFIG_FAIR_GROUP_SCHED
1116 
1117 /* CPU runqueue to which this cfs_rq is attached */
1118 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1119 {
1120         return cfs_rq->rq;
1121 }
1122 
1123 #else
1124 
1125 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1126 {
1127         return container_of(cfs_rq, struct rq, cfs);
1128 }
1129 #endif
1130 
1131 static inline int cpu_of(struct rq *rq)
1132 {
1133 #ifdef CONFIG_SMP
1134         return rq->cpu;
1135 #else
1136         return 0;
1137 #endif
1138 }
1139 
1140 #define MDF_PUSH        0x01
1141 
1142 static inline bool is_migration_disabled(struct task_struct *p)
1143 {
1144 #ifdef CONFIG_SMP
1145         return p->migration_disabled;
1146 #else
1147         return false;
1148 #endif
1149 }
1150 
1151 struct sched_group;
1152 #ifdef CONFIG_SCHED_CORE
1153 static inline struct cpumask *sched_group_span(struct sched_group *sg);
1154 
1155 DECLARE_STATIC_KEY_FALSE(__sched_core_enabled);
1156 
1157 static inline bool sched_core_enabled(struct rq *rq)
1158 {
1159         return static_branch_unlikely(&__sched_core_enabled) && rq->core_enabled;
1160 }
1161 
1162 static inline bool sched_core_disabled(void)
1163 {
1164         return !static_branch_unlikely(&__sched_core_enabled);
1165 }
1166 
1167 /*
1168  * Be careful with this function; not for general use. The return value isn't
1169  * stable unless you actually hold a relevant rq->__lock.
1170  */
1171 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1172 {
1173         if (sched_core_enabled(rq))
1174                 return &rq->core->__lock;
1175 
1176         return &rq->__lock;
1177 }
1178 
1179 static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1180 {
1181         if (rq->core_enabled)
1182                 return &rq->core->__lock;
1183 
1184         return &rq->__lock;
1185 }
1186 
1187 bool cfs_prio_less(struct task_struct *a, struct task_struct *b, bool fi);
1188 
1189 /*
1190  * Helpers to check if the CPU's core cookie matches with the task's cookie
1191  * when core scheduling is enabled.
1192  * A special case is that the task's cookie always matches with CPU's core
1193  * cookie if the CPU is in an idle core.
1194  */
1195 static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1196 {
1197         /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1198         if (!sched_core_enabled(rq))
1199                 return true;
1200 
1201         return rq->core->core_cookie == p->core_cookie;
1202 }
1203 
1204 static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1205 {
1206         bool idle_core = true;
1207         int cpu;
1208 
1209         /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1210         if (!sched_core_enabled(rq))
1211                 return true;
1212 
1213         for_each_cpu(cpu, cpu_smt_mask(cpu_of(rq))) {
1214                 if (!available_idle_cpu(cpu)) {
1215                         idle_core = false;
1216                         break;
1217                 }
1218         }
1219 
1220         /*
1221          * A CPU in an idle core is always the best choice for tasks with
1222          * cookies.
1223          */
1224         return idle_core || rq->core->core_cookie == p->core_cookie;
1225 }
1226 
1227 static inline bool sched_group_cookie_match(struct rq *rq,
1228                                             struct task_struct *p,
1229                                             struct sched_group *group)
1230 {
1231         int cpu;
1232 
1233         /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1234         if (!sched_core_enabled(rq))
1235                 return true;
1236 
1237         for_each_cpu_and(cpu, sched_group_span(group), p->cpus_ptr) {
1238                 if (sched_core_cookie_match(rq, p))
1239                         return true;
1240         }
1241         return false;
1242 }
1243 
1244 extern void queue_core_balance(struct rq *rq);
1245 
1246 static inline bool sched_core_enqueued(struct task_struct *p)
1247 {
1248         return !RB_EMPTY_NODE(&p->core_node);
1249 }
1250 
1251 extern void sched_core_enqueue(struct rq *rq, struct task_struct *p);
1252 extern void sched_core_dequeue(struct rq *rq, struct task_struct *p);
1253 
1254 extern void sched_core_get(void);
1255 extern void sched_core_put(void);
1256 
1257 extern unsigned long sched_core_alloc_cookie(void);
1258 extern void sched_core_put_cookie(unsigned long cookie);
1259 extern unsigned long sched_core_get_cookie(unsigned long cookie);
1260 extern unsigned long sched_core_update_cookie(struct task_struct *p, unsigned long cookie);
1261 
1262 #else /* !CONFIG_SCHED_CORE */
1263 
1264 static inline bool sched_core_enabled(struct rq *rq)
1265 {
1266         return false;
1267 }
1268 
1269 static inline bool sched_core_disabled(void)
1270 {
1271         return true;
1272 }
1273 
1274 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
1275 {
1276         return &rq->__lock;
1277 }
1278 
1279 static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
1280 {
1281         return &rq->__lock;
1282 }
1283 
1284 static inline void queue_core_balance(struct rq *rq)
1285 {
1286 }
1287 
1288 static inline bool sched_cpu_cookie_match(struct rq *rq, struct task_struct *p)
1289 {
1290         return true;
1291 }
1292 
1293 static inline bool sched_core_cookie_match(struct rq *rq, struct task_struct *p)
1294 {
1295         return true;
1296 }
1297 
1298 static inline bool sched_group_cookie_match(struct rq *rq,
1299                                             struct task_struct *p,
1300                                             struct sched_group *group)
1301 {
1302         return true;
1303 }
1304 #endif /* CONFIG_SCHED_CORE */
1305 
1306 static inline void lockdep_assert_rq_held(struct rq *rq)
1307 {
1308         lockdep_assert_held(__rq_lockp(rq));
1309 }
1310 
1311 extern void raw_spin_rq_lock_nested(struct rq *rq, int subclass);
1312 extern bool raw_spin_rq_trylock(struct rq *rq);
1313 extern void raw_spin_rq_unlock(struct rq *rq);
1314 
1315 static inline void raw_spin_rq_lock(struct rq *rq)
1316 {
1317         raw_spin_rq_lock_nested(rq, 0);
1318 }
1319 
1320 static inline void raw_spin_rq_lock_irq(struct rq *rq)
1321 {
1322         local_irq_disable();
1323         raw_spin_rq_lock(rq);
1324 }
1325 
1326 static inline void raw_spin_rq_unlock_irq(struct rq *rq)
1327 {
1328         raw_spin_rq_unlock(rq);
1329         local_irq_enable();
1330 }
1331 
1332 static inline unsigned long _raw_spin_rq_lock_irqsave(struct rq *rq)
1333 {
1334         unsigned long flags;
1335         local_irq_save(flags);
1336         raw_spin_rq_lock(rq);
1337         return flags;
1338 }
1339 
1340 static inline void raw_spin_rq_unlock_irqrestore(struct rq *rq, unsigned long flags)
1341 {
1342         raw_spin_rq_unlock(rq);
1343         local_irq_restore(flags);
1344 }
1345 
1346 #define raw_spin_rq_lock_irqsave(rq, flags)     \
1347 do {                                            \
1348         flags = _raw_spin_rq_lock_irqsave(rq);  \
1349 } while (0)
1350 
1351 #ifdef CONFIG_SCHED_SMT
1352 extern void __update_idle_core(struct rq *rq);
1353 
1354 static inline void update_idle_core(struct rq *rq)
1355 {
1356         if (static_branch_unlikely(&sched_smt_present))
1357                 __update_idle_core(rq);
1358 }
1359 
1360 #else
1361 static inline void update_idle_core(struct rq *rq) { }
1362 #endif
1363 
1364 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1365 
1366 #define cpu_rq(cpu)             (&per_cpu(runqueues, (cpu)))
1367 #define this_rq()               this_cpu_ptr(&runqueues)
1368 #define task_rq(p)              cpu_rq(task_cpu(p))
1369 #define cpu_curr(cpu)           (cpu_rq(cpu)->curr)
1370 #define raw_rq()                raw_cpu_ptr(&runqueues)
1371 
1372 #ifdef CONFIG_FAIR_GROUP_SCHED
1373 static inline struct task_struct *task_of(struct sched_entity *se)
1374 {
1375         SCHED_WARN_ON(!entity_is_task(se));
1376         return container_of(se, struct task_struct, se);
1377 }
1378 
1379 static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1380 {
1381         return p->se.cfs_rq;
1382 }
1383 
1384 /* runqueue on which this entity is (to be) queued */
1385 static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
1386 {
1387         return se->cfs_rq;
1388 }
1389 
1390 /* runqueue "owned" by this group */
1391 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1392 {
1393         return grp->my_q;
1394 }
1395 
1396 #else
1397 
1398 static inline struct task_struct *task_of(struct sched_entity *se)
1399 {
1400         return container_of(se, struct task_struct, se);
1401 }
1402 
1403 static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
1404 {
1405         return &task_rq(p)->cfs;
1406 }
1407 
1408 static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
1409 {
1410         struct task_struct *p = task_of(se);
1411         struct rq *rq = task_rq(p);
1412 
1413         return &rq->cfs;
1414 }
1415 
1416 /* runqueue "owned" by this group */
1417 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
1418 {
1419         return NULL;
1420 }
1421 #endif
1422 
1423 extern void update_rq_clock(struct rq *rq);
1424 
1425 static inline u64 __rq_clock_broken(struct rq *rq)
1426 {
1427         return READ_ONCE(rq->clock);
1428 }
1429 
1430 /*
1431  * rq::clock_update_flags bits
1432  *
1433  * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1434  *  call to __schedule(). This is an optimisation to avoid
1435  *  neighbouring rq clock updates.
1436  *
1437  * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1438  *  in effect and calls to update_rq_clock() are being ignored.
1439  *
1440  * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1441  *  made to update_rq_clock() since the last time rq::lock was pinned.
1442  *
1443  * If inside of __schedule(), clock_update_flags will have been
1444  * shifted left (a left shift is a cheap operation for the fast path
1445  * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1446  *
1447  *      if (rq-clock_update_flags >= RQCF_UPDATED)
1448  *
1449  * to check if %RQCF_UPDATED is set. It'll never be shifted more than
1450  * one position though, because the next rq_unpin_lock() will shift it
1451  * back.
1452  */
1453 #define RQCF_REQ_SKIP           0x01
1454 #define RQCF_ACT_SKIP           0x02
1455 #define RQCF_UPDATED            0x04
1456 
1457 static inline void assert_clock_updated(struct rq *rq)
1458 {
1459         /*
1460          * The only reason for not seeing a clock update since the
1461          * last rq_pin_lock() is if we're currently skipping updates.
1462          */
1463         SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1464 }
1465 
1466 static inline u64 rq_clock(struct rq *rq)
1467 {
1468         lockdep_assert_rq_held(rq);
1469         assert_clock_updated(rq);
1470 
1471         return rq->clock;
1472 }
1473 
1474 static inline u64 rq_clock_task(struct rq *rq)
1475 {
1476         lockdep_assert_rq_held(rq);
1477         assert_clock_updated(rq);
1478 
1479         return rq->clock_task;
1480 }
1481 
1482 /**
1483  * By default the decay is the default pelt decay period.
1484  * The decay shift can change the decay period in
1485  * multiples of 32.
1486  *  Decay shift         Decay period(ms)
1487  *      0                       32
1488  *      1                       64
1489  *      2                       128
1490  *      3                       256
1491  *      4                       512
1492  */
1493 extern int sched_thermal_decay_shift;
1494 
1495 static inline u64 rq_clock_thermal(struct rq *rq)
1496 {
1497         return rq_clock_task(rq) >> sched_thermal_decay_shift;
1498 }
1499 
1500 static inline void rq_clock_skip_update(struct rq *rq)
1501 {
1502         lockdep_assert_rq_held(rq);
1503         rq->clock_update_flags |= RQCF_REQ_SKIP;
1504 }
1505 
1506 /*
1507  * See rt task throttling, which is the only time a skip
1508  * request is canceled.
1509  */
1510 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1511 {
1512         lockdep_assert_rq_held(rq);
1513         rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1514 }
1515 
1516 struct rq_flags {
1517         unsigned long flags;
1518         struct pin_cookie cookie;
1519 #ifdef CONFIG_SCHED_DEBUG
1520         /*
1521          * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1522          * current pin context is stashed here in case it needs to be
1523          * restored in rq_repin_lock().
1524          */
1525         unsigned int clock_update_flags;
1526 #endif
1527 };
1528 
1529 extern struct callback_head balance_push_callback;
1530 
1531 /*
1532  * Lockdep annotation that avoids accidental unlocks; it's like a
1533  * sticky/continuous lockdep_assert_held().
1534  *
1535  * This avoids code that has access to 'struct rq *rq' (basically everything in
1536  * the scheduler) from accidentally unlocking the rq if they do not also have a
1537  * copy of the (on-stack) 'struct rq_flags rf'.
1538  *
1539  * Also see Documentation/locking/lockdep-design.rst.
1540  */
1541 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1542 {
1543         rf->cookie = lockdep_pin_lock(__rq_lockp(rq));
1544 
1545 #ifdef CONFIG_SCHED_DEBUG
1546         rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1547         rf->clock_update_flags = 0;
1548 #ifdef CONFIG_SMP
1549         SCHED_WARN_ON(rq->balance_callback && rq->balance_callback != &balance_push_callback);
1550 #endif
1551 #endif
1552 }
1553 
1554 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1555 {
1556 #ifdef CONFIG_SCHED_DEBUG
1557         if (rq->clock_update_flags > RQCF_ACT_SKIP)
1558                 rf->clock_update_flags = RQCF_UPDATED;
1559 #endif
1560 
1561         lockdep_unpin_lock(__rq_lockp(rq), rf->cookie);
1562 }
1563 
1564 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1565 {
1566         lockdep_repin_lock(__rq_lockp(rq), rf->cookie);
1567 
1568 #ifdef CONFIG_SCHED_DEBUG
1569         /*
1570          * Restore the value we stashed in @rf for this pin context.
1571          */
1572         rq->clock_update_flags |= rf->clock_update_flags;
1573 #endif
1574 }
1575 
1576 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1577         __acquires(rq->lock);
1578 
1579 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1580         __acquires(p->pi_lock)
1581         __acquires(rq->lock);
1582 
1583 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1584         __releases(rq->lock)
1585 {
1586         rq_unpin_lock(rq, rf);
1587         raw_spin_rq_unlock(rq);
1588 }
1589 
1590 static inline void
1591 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1592         __releases(rq->lock)
1593         __releases(p->pi_lock)
1594 {
1595         rq_unpin_lock(rq, rf);
1596         raw_spin_rq_unlock(rq);
1597         raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1598 }
1599 
1600 static inline void
1601 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1602         __acquires(rq->lock)
1603 {
1604         raw_spin_rq_lock_irqsave(rq, rf->flags);
1605         rq_pin_lock(rq, rf);
1606 }
1607 
1608 static inline void
1609 rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1610         __acquires(rq->lock)
1611 {
1612         raw_spin_rq_lock_irq(rq);
1613         rq_pin_lock(rq, rf);
1614 }
1615 
1616 static inline void
1617 rq_lock(struct rq *rq, struct rq_flags *rf)
1618         __acquires(rq->lock)
1619 {
1620         raw_spin_rq_lock(rq);
1621         rq_pin_lock(rq, rf);
1622 }
1623 
1624 static inline void
1625 rq_relock(struct rq *rq, struct rq_flags *rf)
1626         __acquires(rq->lock)
1627 {
1628         raw_spin_rq_lock(rq);
1629         rq_repin_lock(rq, rf);
1630 }
1631 
1632 static inline void
1633 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1634         __releases(rq->lock)
1635 {
1636         rq_unpin_lock(rq, rf);
1637         raw_spin_rq_unlock_irqrestore(rq, rf->flags);
1638 }
1639 
1640 static inline void
1641 rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1642         __releases(rq->lock)
1643 {
1644         rq_unpin_lock(rq, rf);
1645         raw_spin_rq_unlock_irq(rq);
1646 }
1647 
1648 static inline void
1649 rq_unlock(struct rq *rq, struct rq_flags *rf)
1650         __releases(rq->lock)
1651 {
1652         rq_unpin_lock(rq, rf);
1653         raw_spin_rq_unlock(rq);
1654 }
1655 
1656 static inline struct rq *
1657 this_rq_lock_irq(struct rq_flags *rf)
1658         __acquires(rq->lock)
1659 {
1660         struct rq *rq;
1661 
1662         local_irq_disable();
1663         rq = this_rq();
1664         rq_lock(rq, rf);
1665         return rq;
1666 }
1667 
1668 #ifdef CONFIG_NUMA
1669 enum numa_topology_type {
1670         NUMA_DIRECT,
1671         NUMA_GLUELESS_MESH,
1672         NUMA_BACKPLANE,
1673 };
1674 extern enum numa_topology_type sched_numa_topology_type;
1675 extern int sched_max_numa_distance;
1676 extern bool find_numa_distance(int distance);
1677 extern void sched_init_numa(void);
1678 extern void sched_domains_numa_masks_set(unsigned int cpu);
1679 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1680 extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
1681 #else
1682 static inline void sched_init_numa(void) { }
1683 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1684 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1685 static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
1686 {
1687         return nr_cpu_ids;
1688 }
1689 #endif
1690 
1691 #ifdef CONFIG_NUMA_BALANCING
1692 /* The regions in numa_faults array from task_struct */
1693 enum numa_faults_stats {
1694         NUMA_MEM = 0,
1695         NUMA_CPU,
1696         NUMA_MEMBUF,
1697         NUMA_CPUBUF
1698 };
1699 extern void sched_setnuma(struct task_struct *p, int node);
1700 extern int migrate_task_to(struct task_struct *p, int cpu);
1701 extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1702                         int cpu, int scpu);
1703 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1704 #else
1705 static inline void
1706 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1707 {
1708 }
1709 #endif /* CONFIG_NUMA_BALANCING */
1710 
1711 #ifdef CONFIG_SMP
1712 
1713 static inline void
1714 queue_balance_callback(struct rq *rq,
1715                        struct callback_head *head,
1716                        void (*func)(struct rq *rq))
1717 {
1718         lockdep_assert_rq_held(rq);
1719 
1720         if (unlikely(head->next || rq->balance_callback == &balance_push_callback))
1721                 return;
1722 
1723         head->func = (void (*)(struct callback_head *))func;
1724         head->next = rq->balance_callback;
1725         rq->balance_callback = head;
1726 }
1727 
1728 #define rcu_dereference_check_sched_domain(p) \
1729         rcu_dereference_check((p), \
1730                               lockdep_is_held(&sched_domains_mutex))
1731 
1732 /*
1733  * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1734  * See destroy_sched_domains: call_rcu for details.
1735  *
1736  * The domain tree of any CPU may only be accessed from within
1737  * preempt-disabled sections.
1738  */
1739 #define for_each_domain(cpu, __sd) \
1740         for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1741                         __sd; __sd = __sd->parent)
1742 
1743 /**
1744  * highest_flag_domain - Return highest sched_domain containing flag.
1745  * @cpu:        The CPU whose highest level of sched domain is to
1746  *              be returned.
1747  * @flag:       The flag to check for the highest sched_domain
1748  *              for the given CPU.
1749  *
1750  * Returns the highest sched_domain of a CPU which contains the given flag.
1751  */
1752 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1753 {
1754         struct sched_domain *sd, *hsd = NULL;
1755 
1756         for_each_domain(cpu, sd) {
1757                 if (!(sd->flags & flag))
1758                         break;
1759                 hsd = sd;
1760         }
1761 
1762         return hsd;
1763 }
1764 
1765 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1766 {
1767         struct sched_domain *sd;
1768 
1769         for_each_domain(cpu, sd) {
1770                 if (sd->flags & flag)
1771                         break;
1772         }
1773 
1774         return sd;
1775 }
1776 
1777 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
1778 DECLARE_PER_CPU(int, sd_llc_size);
1779 DECLARE_PER_CPU(int, sd_llc_id);
1780 DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
1781 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
1782 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
1783 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
1784 extern struct static_key_false sched_asym_cpucapacity;
1785 
1786 struct sched_group_capacity {
1787         atomic_t                ref;
1788         /*
1789          * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1790          * for a single CPU.
1791          */
1792         unsigned long           capacity;
1793         unsigned long           min_capacity;           /* Min per-CPU capacity in group */
1794         unsigned long           max_capacity;           /* Max per-CPU capacity in group */
1795         unsigned long           next_update;
1796         int                     imbalance;              /* XXX unrelated to capacity but shared group state */
1797 
1798 #ifdef CONFIG_SCHED_DEBUG
1799         int                     id;
1800 #endif
1801 
1802         unsigned long           cpumask[];              /* Balance mask */
1803 };
1804 
1805 struct sched_group {
1806         struct sched_group      *next;                  /* Must be a circular list */
1807         atomic_t                ref;
1808 
1809         unsigned int            group_weight;
1810         struct sched_group_capacity *sgc;
1811         int                     asym_prefer_cpu;        /* CPU of highest priority in group */
1812 
1813         /*
1814          * The CPUs this group covers.
1815          *
1816          * NOTE: this field is variable length. (Allocated dynamically
1817          * by attaching extra space to the end of the structure,
1818          * depending on how many CPUs the kernel has booted up with)
1819          */
1820         unsigned long           cpumask[];
1821 };
1822 
1823 static inline struct cpumask *sched_group_span(struct sched_group *sg)
1824 {
1825         return to_cpumask(sg->cpumask);
1826 }
1827 
1828 /*
1829  * See build_balance_mask().
1830  */
1831 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1832 {
1833         return to_cpumask(sg->sgc->cpumask);
1834 }
1835 
1836 /**
1837  * group_first_cpu - Returns the first CPU in the cpumask of a sched_group.
1838  * @group: The group whose first CPU is to be returned.
1839  */
1840 static inline unsigned int group_first_cpu(struct sched_group *group)
1841 {
1842         return cpumask_first(sched_group_span(group));
1843 }
1844 
1845 extern int group_balance_cpu(struct sched_group *sg);
1846 
1847 #ifdef CONFIG_SCHED_DEBUG
1848 void update_sched_domain_debugfs(void);
1849 void dirty_sched_domain_sysctl(int cpu);
1850 #else
1851 static inline void update_sched_domain_debugfs(void)
1852 {
1853 }
1854 static inline void dirty_sched_domain_sysctl(int cpu)
1855 {
1856 }
1857 #endif
1858 
1859 extern int sched_update_scaling(void);
1860 
1861 extern void flush_smp_call_function_from_idle(void);
1862 
1863 #else /* !CONFIG_SMP: */
1864 static inline void flush_smp_call_function_from_idle(void) { }
1865 #endif
1866 
1867 #include "stats.h"
1868 #include "autogroup.h"
1869 
1870 #ifdef CONFIG_CGROUP_SCHED
1871 
1872 /*
1873  * Return the group to which this tasks belongs.
1874  *
1875  * We cannot use task_css() and friends because the cgroup subsystem
1876  * changes that value before the cgroup_subsys::attach() method is called,
1877  * therefore we cannot pin it and might observe the wrong value.
1878  *
1879  * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1880  * core changes this before calling sched_move_task().
1881  *
1882  * Instead we use a 'copy' which is updated from sched_move_task() while
1883  * holding both task_struct::pi_lock and rq::lock.
1884  */
1885 static inline struct task_group *task_group(struct task_struct *p)
1886 {
1887         return p->sched_task_group;
1888 }
1889 
1890 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1891 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1892 {
1893 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1894         struct task_group *tg = task_group(p);
1895 #endif
1896 
1897 #ifdef CONFIG_FAIR_GROUP_SCHED
1898         set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1899         p->se.cfs_rq = tg->cfs_rq[cpu];
1900         p->se.parent = tg->se[cpu];
1901 #endif
1902 
1903 #ifdef CONFIG_RT_GROUP_SCHED
1904         p->rt.rt_rq  = tg->rt_rq[cpu];
1905         p->rt.parent = tg->rt_se[cpu];
1906 #endif
1907 }
1908 
1909 #else /* CONFIG_CGROUP_SCHED */
1910 
1911 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1912 static inline struct task_group *task_group(struct task_struct *p)
1913 {
1914         return NULL;
1915 }
1916 
1917 #endif /* CONFIG_CGROUP_SCHED */
1918 
1919 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1920 {
1921         set_task_rq(p, cpu);
1922 #ifdef CONFIG_SMP
1923         /*
1924          * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1925          * successfully executed on another CPU. We must ensure that updates of
1926          * per-task data have been completed by this moment.
1927          */
1928         smp_wmb();
1929 #ifdef CONFIG_THREAD_INFO_IN_TASK
1930         WRITE_ONCE(p->cpu, cpu);
1931 #else
1932         WRITE_ONCE(task_thread_info(p)->cpu, cpu);
1933 #endif
1934         p->wake_cpu = cpu;
1935 #endif
1936 }
1937 
1938 /*
1939  * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1940  */
1941 #ifdef CONFIG_SCHED_DEBUG
1942 # include <linux/static_key.h>
1943 # define const_debug __read_mostly
1944 #else
1945 # define const_debug const
1946 #endif
1947 
1948 #define SCHED_FEAT(name, enabled)       \
1949         __SCHED_FEAT_##name ,
1950 
1951 enum {
1952 #include "features.h"
1953         __SCHED_FEAT_NR,
1954 };
1955 
1956 #undef SCHED_FEAT
1957 
1958 #ifdef CONFIG_SCHED_DEBUG
1959 
1960 /*
1961  * To support run-time toggling of sched features, all the translation units
1962  * (but core.c) reference the sysctl_sched_features defined in core.c.
1963  */
1964 extern const_debug unsigned int sysctl_sched_features;
1965 
1966 #ifdef CONFIG_JUMP_LABEL
1967 #define SCHED_FEAT(name, enabled)                                       \
1968 static __always_inline bool static_branch_##name(struct static_key *key) \
1969 {                                                                       \
1970         return static_key_##enabled(key);                               \
1971 }
1972 
1973 #include "features.h"
1974 #undef SCHED_FEAT
1975 
1976 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1977 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1978 
1979 #else /* !CONFIG_JUMP_LABEL */
1980 
1981 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1982 
1983 #endif /* CONFIG_JUMP_LABEL */
1984 
1985 #else /* !SCHED_DEBUG */
1986 
1987 /*
1988  * Each translation unit has its own copy of sysctl_sched_features to allow
1989  * constants propagation at compile time and compiler optimization based on
1990  * features default.
1991  */
1992 #define SCHED_FEAT(name, enabled)       \
1993         (1UL << __SCHED_FEAT_##name) * enabled |
1994 static const_debug __maybe_unused unsigned int sysctl_sched_features =
1995 #include "features.h"
1996         0;
1997 #undef SCHED_FEAT
1998 
1999 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
2000 
2001 #endif /* SCHED_DEBUG */
2002 
2003 extern struct static_key_false sched_numa_balancing;
2004 extern struct static_key_false sched_schedstats;
2005 
2006 static inline u64 global_rt_period(void)
2007 {
2008         return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
2009 }
2010 
2011 static inline u64 global_rt_runtime(void)
2012 {
2013         if (sysctl_sched_rt_runtime < 0)
2014                 return RUNTIME_INF;
2015 
2016         return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
2017 }
2018 
2019 static inline int task_current(struct rq *rq, struct task_struct *p)
2020 {
2021         return rq->curr == p;
2022 }
2023 
2024 static inline int task_running(struct rq *rq, struct task_struct *p)
2025 {
2026 #ifdef CONFIG_SMP
2027         return p->on_cpu;
2028 #else
2029         return task_current(rq, p);
2030 #endif
2031 }
2032 
2033 static inline int task_on_rq_queued(struct task_struct *p)
2034 {
2035         return p->on_rq == TASK_ON_RQ_QUEUED;
2036 }
2037 
2038 static inline int task_on_rq_migrating(struct task_struct *p)
2039 {
2040         return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
2041 }
2042 
2043 /* Wake flags. The first three directly map to some SD flag value */
2044 #define WF_EXEC     0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */
2045 #define WF_FORK     0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */
2046 #define WF_TTWU     0x08 /* Wakeup;            maps to SD_BALANCE_WAKE */
2047 
2048 #define WF_SYNC     0x10 /* Waker goes to sleep after wakeup */
2049 #define WF_MIGRATED 0x20 /* Internal use, task got migrated */
2050 #define WF_ON_CPU   0x40 /* Wakee is on_cpu */
2051 
2052 #ifdef CONFIG_SMP
2053 static_assert(WF_EXEC == SD_BALANCE_EXEC);
2054 static_assert(WF_FORK == SD_BALANCE_FORK);
2055 static_assert(WF_TTWU == SD_BALANCE_WAKE);
2056 #endif
2057 
2058 /*
2059  * To aid in avoiding the subversion of "niceness" due to uneven distribution
2060  * of tasks with abnormal "nice" values across CPUs the contribution that
2061  * each task makes to its run queue's load is weighted according to its
2062  * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
2063  * scaled version of the new time slice allocation that they receive on time
2064  * slice expiry etc.
2065  */
2066 
2067 #define WEIGHT_IDLEPRIO         3
2068 #define WMULT_IDLEPRIO          1431655765
2069 
2070 extern const int                sched_prio_to_weight[40];
2071 extern const u32                sched_prio_to_wmult[40];
2072 
2073 /*
2074  * {de,en}queue flags:
2075  *
2076  * DEQUEUE_SLEEP  - task is no longer runnable
2077  * ENQUEUE_WAKEUP - task just became runnable
2078  *
2079  * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
2080  *                are in a known state which allows modification. Such pairs
2081  *                should preserve as much state as possible.
2082  *
2083  * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
2084  *        in the runqueue.
2085  *
2086  * ENQUEUE_HEAD      - place at front of runqueue (tail if not specified)
2087  * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
2088  * ENQUEUE_MIGRATED  - the task was migrated during wakeup
2089  *
2090  */
2091 
2092 #define DEQUEUE_SLEEP           0x01
2093 #define DEQUEUE_SAVE            0x02 /* Matches ENQUEUE_RESTORE */
2094 #define DEQUEUE_MOVE            0x04 /* Matches ENQUEUE_MOVE */
2095 #define DEQUEUE_NOCLOCK         0x08 /* Matches ENQUEUE_NOCLOCK */
2096 
2097 #define ENQUEUE_WAKEUP          0x01
2098 #define ENQUEUE_RESTORE         0x02
2099 #define ENQUEUE_MOVE            0x04
2100 #define ENQUEUE_NOCLOCK         0x08
2101 
2102 #define ENQUEUE_HEAD            0x10
2103 #define ENQUEUE_REPLENISH       0x20
2104 #ifdef CONFIG_SMP
2105 #define ENQUEUE_MIGRATED        0x40
2106 #else
2107 #define ENQUEUE_MIGRATED        0x00
2108 #endif
2109 
2110 #define RETRY_TASK              ((void *)-1UL)
2111 
2112 struct sched_class {
2113 
2114 #ifdef CONFIG_UCLAMP_TASK
2115         int uclamp_enabled;
2116 #endif
2117 
2118         void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
2119         void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
2120         void (*yield_task)   (struct rq *rq);
2121         bool (*yield_to_task)(struct rq *rq, struct task_struct *p);
2122 
2123         void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
2124 
2125         struct task_struct *(*pick_next_task)(struct rq *rq);
2126 
2127         void (*put_prev_task)(struct rq *rq, struct task_struct *p);
2128         void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
2129 
2130 #ifdef CONFIG_SMP
2131         int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2132         int  (*select_task_rq)(struct task_struct *p, int task_cpu, int flags);
2133 
2134         struct task_struct * (*pick_task)(struct rq *rq);
2135 
2136         void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
2137 
2138         void (*task_woken)(struct rq *this_rq, struct task_struct *task);
2139 
2140         void (*set_cpus_allowed)(struct task_struct *p,
2141                                  const struct cpumask *newmask,
2142                                  u32 flags);
2143 
2144         void (*rq_online)(struct rq *rq);
2145         void (*rq_offline)(struct rq *rq);
2146 
2147         struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq);
2148 #endif
2149 
2150         void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
2151         void (*task_fork)(struct task_struct *p);
2152         void (*task_dead)(struct task_struct *p);
2153 
2154         /*
2155          * The switched_from() call is allowed to drop rq->lock, therefore we
2156          * cannot assume the switched_from/switched_to pair is serialized by
2157          * rq->lock. They are however serialized by p->pi_lock.
2158          */
2159         void (*switched_from)(struct rq *this_rq, struct task_struct *task);
2160         void (*switched_to)  (struct rq *this_rq, struct task_struct *task);
2161         void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
2162                               int oldprio);
2163 
2164         unsigned int (*get_rr_interval)(struct rq *rq,
2165                                         struct task_struct *task);
2166 
2167         void (*update_curr)(struct rq *rq);
2168 
2169 #define TASK_SET_GROUP          0
2170 #define TASK_MOVE_GROUP         1
2171 
2172 #ifdef CONFIG_FAIR_GROUP_SCHED
2173         void (*task_change_group)(struct task_struct *p, int type);
2174 #endif
2175 };
2176 
2177 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
2178 {
2179         WARN_ON_ONCE(rq->curr != prev);
2180         prev->sched_class->put_prev_task(rq, prev);
2181 }
2182 
2183 static inline void set_next_task(struct rq *rq, struct task_struct *next)
2184 {
2185         next->sched_class->set_next_task(rq, next, false);
2186 }
2187 
2188 
2189 /*
2190  * Helper to define a sched_class instance; each one is placed in a separate
2191  * section which is ordered by the linker script:
2192  *
2193  *   include/asm-generic/vmlinux.lds.h
2194  *
2195  * Also enforce alignment on the instance, not the type, to guarantee layout.
2196  */
2197 #define DEFINE_SCHED_CLASS(name) \
2198 const struct sched_class name##_sched_class \
2199         __aligned(__alignof__(struct sched_class)) \
2200         __section("__" #name "_sched_class")
2201 
2202 /* Defined in include/asm-generic/vmlinux.lds.h */
2203 extern struct sched_class __begin_sched_classes[];
2204 extern struct sched_class __end_sched_classes[];
2205 
2206 #define sched_class_highest (__end_sched_classes - 1)
2207 #define sched_class_lowest  (__begin_sched_classes - 1)
2208 
2209 #define for_class_range(class, _from, _to) \
2210         for (class = (_from); class != (_to); class--)
2211 
2212 #define for_each_class(class) \
2213         for_class_range(class, sched_class_highest, sched_class_lowest)
2214 
2215 extern const struct sched_class stop_sched_class;
2216 extern const struct sched_class dl_sched_class;
2217 extern const struct sched_class rt_sched_class;
2218 extern const struct sched_class fair_sched_class;
2219 extern const struct sched_class idle_sched_class;
2220 
2221 static inline bool sched_stop_runnable(struct rq *rq)
2222 {
2223         return rq->stop && task_on_rq_queued(rq->stop);
2224 }
2225 
2226 static inline bool sched_dl_runnable(struct rq *rq)
2227 {
2228         return rq->dl.dl_nr_running > 0;
2229 }
2230 
2231 static inline bool sched_rt_runnable(struct rq *rq)
2232 {
2233         return rq->rt.rt_queued > 0;
2234 }
2235 
2236 static inline bool sched_fair_runnable(struct rq *rq)
2237 {
2238         return rq->cfs.nr_running > 0;
2239 }
2240 
2241 extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
2242 extern struct task_struct *pick_next_task_idle(struct rq *rq);
2243 
2244 #define SCA_CHECK               0x01
2245 #define SCA_MIGRATE_DISABLE     0x02
2246 #define SCA_MIGRATE_ENABLE      0x04
2247 #define SCA_USER                0x08
2248 
2249 #ifdef CONFIG_SMP
2250 
2251 extern void update_group_capacity(struct sched_domain *sd, int cpu);
2252 
2253 extern void trigger_load_balance(struct rq *rq);
2254 
2255 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask, u32 flags);
2256 
2257 static inline struct task_struct *get_push_task(struct rq *rq)
2258 {
2259         struct task_struct *p = rq->curr;
2260 
2261         lockdep_assert_rq_held(rq);
2262 
2263         if (rq->push_busy)
2264                 return NULL;
2265 
2266         if (p->nr_cpus_allowed == 1)
2267                 return NULL;
2268 
2269         if (p->migration_disabled)
2270                 return NULL;
2271 
2272         rq->push_busy = true;
2273         return get_task_struct(p);
2274 }
2275 
2276 extern int push_cpu_stop(void *arg);
2277 
2278 #endif
2279 
2280 #ifdef CONFIG_CPU_IDLE
2281 static inline void idle_set_state(struct rq *rq,
2282                                   struct cpuidle_state *idle_state)
2283 {
2284         rq->idle_state = idle_state;
2285 }
2286 
2287 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2288 {
2289         SCHED_WARN_ON(!rcu_read_lock_held());
2290 
2291         return rq->idle_state;
2292 }
2293 #else
2294 static inline void idle_set_state(struct rq *rq,
2295                                   struct cpuidle_state *idle_state)
2296 {
2297 }
2298 
2299 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
2300 {
2301         return NULL;
2302 }
2303 #endif
2304 
2305 extern void schedule_idle(void);
2306 
2307 extern void sysrq_sched_debug_show(void);
2308 extern void sched_init_granularity(void);
2309 extern void update_max_interval(void);
2310 
2311 extern void init_sched_dl_class(void);
2312 extern void init_sched_rt_class(void);
2313 extern void init_sched_fair_class(void);
2314 
2315 extern void reweight_task(struct task_struct *p, int prio);
2316 
2317 extern void resched_curr(struct rq *rq);
2318 extern void resched_cpu(int cpu);
2319 
2320 extern struct rt_bandwidth def_rt_bandwidth;
2321 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
2322 
2323 extern struct dl_bandwidth def_dl_bandwidth;
2324 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
2325 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
2326 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
2327 
2328 #define BW_SHIFT                20
2329 #define BW_UNIT                 (1 << BW_SHIFT)
2330 #define RATIO_SHIFT             8
2331 #define MAX_BW_BITS             (64 - BW_SHIFT)
2332 #define MAX_BW                  ((1ULL << MAX_BW_BITS) - 1)
2333 unsigned long to_ratio(u64 period, u64 runtime);
2334 
2335 extern void init_entity_runnable_average(struct sched_entity *se);
2336 extern void post_init_entity_util_avg(struct task_struct *p);
2337 
2338 #ifdef CONFIG_NO_HZ_FULL
2339 extern bool sched_can_stop_tick(struct rq *rq);
2340 extern int __init sched_tick_offload_init(void);
2341 
2342 /*
2343  * Tick may be needed by tasks in the runqueue depending on their policy and
2344  * requirements. If tick is needed, lets send the target an IPI to kick it out of
2345  * nohz mode if necessary.
2346  */
2347 static inline void sched_update_tick_dependency(struct rq *rq)
2348 {
2349         int cpu = cpu_of(rq);
2350 
2351         if (!tick_nohz_full_cpu(cpu))
2352                 return;
2353 
2354         if (sched_can_stop_tick(rq))
2355                 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
2356         else
2357                 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
2358 }
2359 #else
2360 static inline int sched_tick_offload_init(void) { return 0; }
2361 static inline void sched_update_tick_dependency(struct rq *rq) { }
2362 #endif
2363 
2364 static inline void add_nr_running(struct rq *rq, unsigned count)
2365 {
2366         unsigned prev_nr = rq->nr_running;
2367 
2368         rq->nr_running = prev_nr + count;
2369         if (trace_sched_update_nr_running_tp_enabled()) {
2370                 call_trace_sched_update_nr_running(rq, count);
2371         }
2372 
2373 #ifdef CONFIG_SMP
2374         if (prev_nr < 2 && rq->nr_running >= 2) {
2375                 if (!READ_ONCE(rq->rd->overload))
2376                         WRITE_ONCE(rq->rd->overload, 1);
2377         }
2378 #endif
2379 
2380         sched_update_tick_dependency(rq);
2381 }
2382 
2383 static inline void sub_nr_running(struct rq *rq, unsigned count)
2384 {
2385         rq->nr_running -= count;
2386         if (trace_sched_update_nr_running_tp_enabled()) {
2387                 call_trace_sched_update_nr_running(rq, -count);
2388         }
2389 
2390         /* Check if we still need preemption */
2391         sched_update_tick_dependency(rq);
2392 }
2393 
2394 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
2395 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
2396 
2397 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
2398 
2399 extern const_debug unsigned int sysctl_sched_nr_migrate;
2400 extern const_debug unsigned int sysctl_sched_migration_cost;
2401 
2402 #ifdef CONFIG_SCHED_DEBUG
2403 extern unsigned int sysctl_sched_latency;
2404 extern unsigned int sysctl_sched_min_granularity;
2405 extern unsigned int sysctl_sched_wakeup_granularity;
2406 extern int sysctl_resched_latency_warn_ms;
2407 extern int sysctl_resched_latency_warn_once;
2408 
2409 extern unsigned int sysctl_sched_tunable_scaling;
2410 
2411 extern unsigned int sysctl_numa_balancing_scan_delay;
2412 extern unsigned int sysctl_numa_balancing_scan_period_min;
2413 extern unsigned int sysctl_numa_balancing_scan_period_max;
2414 extern unsigned int sysctl_numa_balancing_scan_size;
2415 #endif
2416 
2417 #ifdef CONFIG_SCHED_HRTICK
2418 
2419 /*
2420  * Use hrtick when:
2421  *  - enabled by features
2422  *  - hrtimer is actually high res
2423  */
2424 static inline int hrtick_enabled(struct rq *rq)
2425 {
2426         if (!cpu_active(cpu_of(rq)))
2427                 return 0;
2428         return hrtimer_is_hres_active(&rq->hrtick_timer);
2429 }
2430 
2431 static inline int hrtick_enabled_fair(struct rq *rq)
2432 {
2433         if (!sched_feat(HRTICK))
2434                 return 0;
2435         return hrtick_enabled(rq);
2436 }
2437 
2438 static inline int hrtick_enabled_dl(struct rq *rq)
2439 {
2440         if (!sched_feat(HRTICK_DL))
2441                 return 0;
2442         return hrtick_enabled(rq);
2443 }
2444 
2445 void hrtick_start(struct rq *rq, u64 delay);
2446 
2447 #else
2448 
2449 static inline int hrtick_enabled_fair(struct rq *rq)
2450 {
2451         return 0;
2452 }
2453 
2454 static inline int hrtick_enabled_dl(struct rq *rq)
2455 {
2456         return 0;
2457 }
2458 
2459 static inline int hrtick_enabled(struct rq *rq)
2460 {
2461         return 0;
2462 }
2463 
2464 #endif /* CONFIG_SCHED_HRTICK */
2465 
2466 #ifndef arch_scale_freq_tick
2467 static __always_inline
2468 void arch_scale_freq_tick(void)
2469 {
2470 }
2471 #endif
2472 
2473 #ifndef arch_scale_freq_capacity
2474 /**
2475  * arch_scale_freq_capacity - get the frequency scale factor of a given CPU.
2476  * @cpu: the CPU in question.
2477  *
2478  * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
2479  *
2480  *     f_curr
2481  *     ------ * SCHED_CAPACITY_SCALE
2482  *     f_max
2483  */
2484 static __always_inline
2485 unsigned long arch_scale_freq_capacity(int cpu)
2486 {
2487         return SCHED_CAPACITY_SCALE;
2488 }
2489 #endif
2490 
2491 
2492 #ifdef CONFIG_SMP
2493 
2494 static inline bool rq_order_less(struct rq *rq1, struct rq *rq2)
2495 {
2496 #ifdef CONFIG_SCHED_CORE
2497         /*
2498          * In order to not have {0,2},{1,3} turn into into an AB-BA,
2499          * order by core-id first and cpu-id second.
2500          *
2501          * Notably:
2502          *
2503          *      double_rq_lock(0,3); will take core-0, core-1 lock
2504          *      double_rq_lock(1,2); will take core-1, core-0 lock
2505          *
2506          * when only cpu-id is considered.
2507          */
2508         if (rq1->core->cpu < rq2->core->cpu)
2509                 return true;
2510         if (rq1->core->cpu > rq2->core->cpu)
2511                 return false;
2512 
2513         /*
2514          * __sched_core_flip() relies on SMT having cpu-id lock order.
2515          */
2516 #endif
2517         return rq1->cpu < rq2->cpu;
2518 }
2519 
2520 extern void double_rq_lock(struct rq *rq1, struct rq *rq2);
2521 
2522 #ifdef CONFIG_PREEMPTION
2523 
2524 /*
2525  * fair double_lock_balance: Safely acquires both rq->locks in a fair
2526  * way at the expense of forcing extra atomic operations in all
2527  * invocations.  This assures that the double_lock is acquired using the
2528  * same underlying policy as the spinlock_t on this architecture, which
2529  * reduces latency compared to the unfair variant below.  However, it
2530  * also adds more overhead and therefore may reduce throughput.
2531  */
2532 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2533         __releases(this_rq->lock)
2534         __acquires(busiest->lock)
2535         __acquires(this_rq->lock)
2536 {
2537         raw_spin_rq_unlock(this_rq);
2538         double_rq_lock(this_rq, busiest);
2539 
2540         return 1;
2541 }
2542 
2543 #else
2544 /*
2545  * Unfair double_lock_balance: Optimizes throughput at the expense of
2546  * latency by eliminating extra atomic operations when the locks are
2547  * already in proper order on entry.  This favors lower CPU-ids and will
2548  * grant the double lock to lower CPUs over higher ids under contention,
2549  * regardless of entry order into the function.
2550  */
2551 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
2552         __releases(this_rq->lock)
2553         __acquires(busiest->lock)
2554         __acquires(this_rq->lock)
2555 {
2556         if (__rq_lockp(this_rq) == __rq_lockp(busiest))
2557                 return 0;
2558 
2559         if (likely(raw_spin_rq_trylock(busiest)))
2560                 return 0;
2561 
2562         if (rq_order_less(this_rq, busiest)) {
2563                 raw_spin_rq_lock_nested(busiest, SINGLE_DEPTH_NESTING);
2564                 return 0;
2565         }
2566 
2567         raw_spin_rq_unlock(this_rq);
2568         double_rq_lock(this_rq, busiest);
2569 
2570         return 1;
2571 }
2572 
2573 #endif /* CONFIG_PREEMPTION */
2574 
2575 /*
2576  * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2577  */
2578 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2579 {
2580         lockdep_assert_irqs_disabled();
2581 
2582         return _double_lock_balance(this_rq, busiest);
2583 }
2584 
2585 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2586         __releases(busiest->lock)
2587 {
2588         if (__rq_lockp(this_rq) != __rq_lockp(busiest))
2589                 raw_spin_rq_unlock(busiest);
2590         lock_set_subclass(&__rq_lockp(this_rq)->dep_map, 0, _RET_IP_);
2591 }
2592 
2593 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
2594 {
2595         if (l1 > l2)
2596                 swap(l1, l2);
2597 
2598         spin_lock(l1);
2599         spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2600 }
2601 
2602 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
2603 {
2604         if (l1 > l2)
2605                 swap(l1, l2);
2606 
2607         spin_lock_irq(l1);
2608         spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2609 }
2610 
2611 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2612 {
2613         if (l1 > l2)
2614                 swap(l1, l2);
2615 
2616         raw_spin_lock(l1);
2617         raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2618 }
2619 
2620 /*
2621  * double_rq_unlock - safely unlock two runqueues
2622  *
2623  * Note this does not restore interrupts like task_rq_unlock,
2624  * you need to do so manually after calling.
2625  */
2626 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2627         __releases(rq1->lock)
2628         __releases(rq2->lock)
2629 {
2630         if (__rq_lockp(rq1) != __rq_lockp(rq2))
2631                 raw_spin_rq_unlock(rq2);
2632         else
2633                 __release(rq2->lock);
2634         raw_spin_rq_unlock(rq1);
2635 }
2636 
2637 extern void set_rq_online (struct rq *rq);
2638 extern void set_rq_offline(struct rq *rq);
2639 extern bool sched_smp_initialized;
2640 
2641 #else /* CONFIG_SMP */
2642 
2643 /*
2644  * double_rq_lock - safely lock two runqueues
2645  *
2646  * Note this does not disable interrupts like task_rq_lock,
2647  * you need to do so manually before calling.
2648  */
2649 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2650         __acquires(rq1->lock)
2651         __acquires(rq2->lock)
2652 {
2653         BUG_ON(!irqs_disabled());
2654         BUG_ON(rq1 != rq2);
2655         raw_spin_rq_lock(rq1);
2656         __acquire(rq2->lock);   /* Fake it out ;) */
2657 }
2658 
2659 /*
2660  * double_rq_unlock - safely unlock two runqueues
2661  *
2662  * Note this does not restore interrupts like task_rq_unlock,
2663  * you need to do so manually after calling.
2664  */
2665 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2666         __releases(rq1->lock)
2667         __releases(rq2->lock)
2668 {
2669         BUG_ON(rq1 != rq2);
2670         raw_spin_rq_unlock(rq1);
2671         __release(rq2->lock);
2672 }
2673 
2674 #endif
2675 
2676 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2677 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2678 
2679 #ifdef  CONFIG_SCHED_DEBUG
2680 extern bool sched_debug_verbose;
2681 
2682 extern void print_cfs_stats(struct seq_file *m, int cpu);
2683 extern void print_rt_stats(struct seq_file *m, int cpu);
2684 extern void print_dl_stats(struct seq_file *m, int cpu);
2685 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2686 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2687 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2688 
2689 extern void resched_latency_warn(int cpu, u64 latency);
2690 #ifdef CONFIG_NUMA_BALANCING
2691 extern void
2692 show_numa_stats(struct task_struct *p, struct seq_file *m);
2693 extern void
2694 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2695         unsigned long tpf, unsigned long gsf, unsigned long gpf);
2696 #endif /* CONFIG_NUMA_BALANCING */
2697 #else
2698 static inline void resched_latency_warn(int cpu, u64 latency) {}
2699 #endif /* CONFIG_SCHED_DEBUG */
2700 
2701 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2702 extern void init_rt_rq(struct rt_rq *rt_rq);
2703 extern void init_dl_rq(struct dl_rq *dl_rq);
2704 
2705 extern void cfs_bandwidth_usage_inc(void);
2706 extern void cfs_bandwidth_usage_dec(void);
2707 
2708 #ifdef CONFIG_NO_HZ_COMMON
2709 #define NOHZ_BALANCE_KICK_BIT   0
2710 #define NOHZ_STATS_KICK_BIT     1
2711 #define NOHZ_NEWILB_KICK_BIT    2
2712 
2713 #define NOHZ_BALANCE_KICK       BIT(NOHZ_BALANCE_KICK_BIT)
2714 #define NOHZ_STATS_KICK         BIT(NOHZ_STATS_KICK_BIT)
2715 #define NOHZ_NEWILB_KICK        BIT(NOHZ_NEWILB_KICK_BIT)
2716 
2717 #define NOHZ_KICK_MASK  (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK)
2718 
2719 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2720 
2721 extern void nohz_balance_exit_idle(struct rq *rq);
2722 #else
2723 static inline void nohz_balance_exit_idle(struct rq *rq) { }
2724 #endif
2725 
2726 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
2727 extern void nohz_run_idle_balance(int cpu);
2728 #else
2729 static inline void nohz_run_idle_balance(int cpu) { }
2730 #endif
2731 
2732 #ifdef CONFIG_SMP
2733 static inline
2734 void __dl_update(struct dl_bw *dl_b, s64 bw)
2735 {
2736         struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw);
2737         int i;
2738 
2739         RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
2740                          "sched RCU must be held");
2741         for_each_cpu_and(i, rd->span, cpu_active_mask) {
2742                 struct rq *rq = cpu_rq(i);
2743 
2744                 rq->dl.extra_bw += bw;
2745         }
2746 }
2747 #else
2748 static inline
2749 void __dl_update(struct dl_bw *dl_b, s64 bw)
2750 {
2751         struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw);
2752 
2753         dl->extra_bw += bw;
2754 }
2755 #endif
2756 
2757 
2758 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2759 struct irqtime {
2760         u64                     total;
2761         u64                     tick_delta;
2762         u64                     irq_start_time;
2763         struct u64_stats_sync   sync;
2764 };
2765 
2766 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2767 
2768 /*
2769  * Returns the irqtime minus the softirq time computed by ksoftirqd.
2770  * Otherwise ksoftirqd's sum_exec_runtime is subtracted its own runtime
2771  * and never move forward.
2772  */
2773 static inline u64 irq_time_read(int cpu)
2774 {
2775         struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2776         unsigned int seq;
2777         u64 total;
2778 
2779         do {
2780                 seq = __u64_stats_fetch_begin(&irqtime->sync);
2781                 total = irqtime->total;
2782         } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2783 
2784         return total;
2785 }
2786 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2787 
2788 #ifdef CONFIG_CPU_FREQ
2789 DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
2790 
2791 /**
2792  * cpufreq_update_util - Take a note about CPU utilization changes.
2793  * @rq: Runqueue to carry out the update for.
2794  * @flags: Update reason flags.
2795  *
2796  * This function is called by the scheduler on the CPU whose utilization is
2797  * being updated.
2798  *
2799  * It can only be called from RCU-sched read-side critical sections.
2800  *
2801  * The way cpufreq is currently arranged requires it to evaluate the CPU
2802  * performance state (frequency/voltage) on a regular basis to prevent it from
2803  * being stuck in a completely inadequate performance level for too long.
2804  * That is not guaranteed to happen if the updates are only triggered from CFS
2805  * and DL, though, because they may not be coming in if only RT tasks are
2806  * active all the time (or there are RT tasks only).
2807  *
2808  * As a workaround for that issue, this function is called periodically by the
2809  * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2810  * but that really is a band-aid.  Going forward it should be replaced with
2811  * solutions targeted more specifically at RT tasks.
2812  */
2813 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2814 {
2815         struct update_util_data *data;
2816 
2817         data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2818                                                   cpu_of(rq)));
2819         if (data)
2820                 data->func(data, rq_clock(rq), flags);
2821 }
2822 #else
2823 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2824 #endif /* CONFIG_CPU_FREQ */
2825 
2826 #ifdef CONFIG_UCLAMP_TASK
2827 unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id);
2828 
2829 /**
2830  * uclamp_rq_util_with - clamp @util with @rq and @p effective uclamp values.
2831  * @rq:         The rq to clamp against. Must not be NULL.
2832  * @util:       The util value to clamp.
2833  * @p:          The task to clamp against. Can be NULL if you want to clamp
2834  *              against @rq only.
2835  *
2836  * Clamps the passed @util to the max(@rq, @p) effective uclamp values.
2837  *
2838  * If sched_uclamp_used static key is disabled, then just return the util
2839  * without any clamping since uclamp aggregation at the rq level in the fast
2840  * path is disabled, rendering this operation a NOP.
2841  *
2842  * Use uclamp_eff_value() if you don't care about uclamp values at rq level. It
2843  * will return the correct effective uclamp value of the task even if the
2844  * static key is disabled.
2845  */
2846 static __always_inline
2847 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2848                                   struct task_struct *p)
2849 {
2850         unsigned long min_util = 0;
2851         unsigned long max_util = 0;
2852 
2853         if (!static_branch_likely(&sched_uclamp_used))
2854                 return util;
2855 
2856         if (p) {
2857                 min_util = uclamp_eff_value(p, UCLAMP_MIN);
2858                 max_util = uclamp_eff_value(p, UCLAMP_MAX);
2859 
2860                 /*
2861                  * Ignore last runnable task's max clamp, as this task will
2862                  * reset it. Similarly, no need to read the rq's min clamp.
2863                  */
2864                 if (rq->uclamp_flags & UCLAMP_FLAG_IDLE)
2865                         goto out;
2866         }
2867 
2868         min_util = max_t(unsigned long, min_util, READ_ONCE(rq->uclamp[UCLAMP_MIN].value));
2869         max_util = max_t(unsigned long, max_util, READ_ONCE(rq->uclamp[UCLAMP_MAX].value));
2870 out:
2871         /*
2872          * Since CPU's {min,max}_util clamps are MAX aggregated considering
2873          * RUNNABLE tasks with _different_ clamps, we can end up with an
2874          * inversion. Fix it now when the clamps are applied.
2875          */
2876         if (unlikely(min_util >= max_util))
2877                 return min_util;
2878 
2879         return clamp(util, min_util, max_util);
2880 }
2881 
2882 /*
2883  * When uclamp is compiled in, the aggregation at rq level is 'turned off'
2884  * by default in the fast path and only gets turned on once userspace performs
2885  * an operation that requires it.
2886  *
2887  * Returns true if userspace opted-in to use uclamp and aggregation at rq level
2888  * hence is active.
2889  */
2890 static inline bool uclamp_is_used(void)
2891 {
2892         return static_branch_likely(&sched_uclamp_used);
2893 }
2894 #else /* CONFIG_UCLAMP_TASK */
2895 static inline
2896 unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util,
2897                                   struct task_struct *p)
2898 {
2899         return util;
2900 }
2901 
2902 static inline bool uclamp_is_used(void)
2903 {
2904         return false;
2905 }
2906 #endif /* CONFIG_UCLAMP_TASK */
2907 
2908 #ifdef arch_scale_freq_capacity
2909 # ifndef arch_scale_freq_invariant
2910 #  define arch_scale_freq_invariant()   true
2911 # endif
2912 #else
2913 # define arch_scale_freq_invariant()    false
2914 #endif
2915 
2916 #ifdef CONFIG_SMP
2917 static inline unsigned long capacity_orig_of(int cpu)
2918 {
2919         return cpu_rq(cpu)->cpu_capacity_orig;
2920 }
2921 
2922 /**
2923  * enum cpu_util_type - CPU utilization type
2924  * @FREQUENCY_UTIL:     Utilization used to select frequency
2925  * @ENERGY_UTIL:        Utilization used during energy calculation
2926  *
2927  * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
2928  * need to be aggregated differently depending on the usage made of them. This
2929  * enum is used within effective_cpu_util() to differentiate the types of
2930  * utilization expected by the callers, and adjust the aggregation accordingly.
2931  */
2932 enum cpu_util_type {
2933         FREQUENCY_UTIL,
2934         ENERGY_UTIL,
2935 };
2936 
2937 unsigned long effective_cpu_util(int cpu, unsigned long util_cfs,
2938                                  unsigned long max, enum cpu_util_type type,
2939                                  struct task_struct *p);
2940 
2941 static inline unsigned long cpu_bw_dl(struct rq *rq)
2942 {
2943         return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2944 }
2945 
2946 static inline unsigned long cpu_util_dl(struct rq *rq)
2947 {
2948         return READ_ONCE(rq->avg_dl.util_avg);
2949 }
2950 
2951 static inline unsigned long cpu_util_cfs(struct rq *rq)
2952 {
2953         unsigned long util = READ_ONCE(rq->cfs.avg.util_avg);
2954 
2955         if (sched_feat(UTIL_EST)) {
2956                 util = max_t(unsigned long, util,
2957                              READ_ONCE(rq->cfs.avg.util_est.enqueued));
2958         }
2959 
2960         return util;
2961 }
2962 
2963 static inline unsigned long cpu_util_rt(struct rq *rq)
2964 {
2965         return READ_ONCE(rq->avg_rt.util_avg);
2966 }
2967 #endif
2968 
2969 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
2970 static inline unsigned long cpu_util_irq(struct rq *rq)
2971 {
2972         return rq->avg_irq.util_avg;
2973 }
2974 
2975 static inline
2976 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2977 {
2978         util *= (max - irq);
2979         util /= max;
2980 
2981         return util;
2982 
2983 }
2984 #else
2985 static inline unsigned long cpu_util_irq(struct rq *rq)
2986 {
2987         return 0;
2988 }
2989 
2990 static inline
2991 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2992 {
2993         return util;
2994 }
2995 #endif
2996 
2997 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
2998 
2999 #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
3000 
3001 DECLARE_STATIC_KEY_FALSE(sched_energy_present);
3002 
3003 static inline bool sched_energy_enabled(void)
3004 {
3005         return static_branch_unlikely(&sched_energy_present);
3006 }
3007 
3008 #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
3009 
3010 #define perf_domain_span(pd) NULL
3011 static inline bool sched_energy_enabled(void) { return false; }
3012 
3013 #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
3014 
3015 #ifdef CONFIG_MEMBARRIER
3016 /*
3017  * The scheduler provides memory barriers required by membarrier between:
3018  * - prior user-space memory accesses and store to rq->membarrier_state,
3019  * - store to rq->membarrier_state and following user-space memory accesses.
3020  * In the same way it provides those guarantees around store to rq->curr.
3021  */
3022 static inline void membarrier_switch_mm(struct rq *rq,
3023                                         struct mm_struct *prev_mm,
3024                                         struct mm_struct *next_mm)
3025 {
3026         int membarrier_state;
3027 
3028         if (prev_mm == next_mm)
3029                 return;
3030 
3031         membarrier_state = atomic_read(&next_mm->membarrier_state);
3032         if (READ_ONCE(rq->membarrier_state) == membarrier_state)
3033                 return;
3034 
3035         WRITE_ONCE(rq->membarrier_state, membarrier_state);
3036 }
3037 #else
3038 static inline void membarrier_switch_mm(struct rq *rq,
3039                                         struct mm_struct *prev_mm,
3040                                         struct mm_struct *next_mm)
3041 {
3042 }
3043 #endif
3044 
3045 #ifdef CONFIG_SMP
3046 static inline bool is_per_cpu_kthread(struct task_struct *p)
3047 {
3048         if (!(p->flags & PF_KTHREAD))
3049                 return false;
3050 
3051         if (p->nr_cpus_allowed != 1)
3052                 return false;
3053 
3054         return true;
3055 }
3056 #endif
3057 
3058 extern void swake_up_all_locked(struct swait_queue_head *q);
3059 extern void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait);
3060 
3061 #ifdef CONFIG_PREEMPT_DYNAMIC
3062 extern int preempt_dynamic_mode;
3063 extern int sched_dynamic_mode(const char *str);
3064 extern void sched_dynamic_update(int mode);
3065 #endif
3066 
3067 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp