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

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

Version: ~ [ linux-5.16-rc1 ] ~ [ linux-5.15.2 ] ~ [ linux-5.14.18 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.79 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.159 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.217 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.255 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.290 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.292 ] ~ [ 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/blkdev.h>
 40 #include <linux/compat.h>
 41 #include <linux/context_tracking.h>
 42 #include <linux/cpufreq.h>
 43 #include <linux/cpuidle.h>
 44 #include <linux/cpuset.h>
 45 #include <linux/ctype.h>
 46 #include <linux/debugfs.h>
 47 #include <linux/delayacct.h>
 48 #include <linux/energy_model.h>
 49 #include <linux/init_task.h>
 50 #include <linux/kprobes.h>
 51 #include <linux/kthread.h>
 52 #include <linux/membarrier.h>
 53 #include <linux/migrate.h>
 54 #include <linux/mmu_context.h>
 55 #include <linux/nmi.h>
 56 #include <linux/proc_fs.h>
 57 #include <linux/prefetch.h>
 58 #include <linux/profile.h>
 59 #include <linux/psi.h>
 60 #include <linux/rcupdate_wait.h>
 61 #include <linux/security.h>
 62 #include <linux/stop_machine.h>
 63 #include <linux/suspend.h>
 64 #include <linux/swait.h>
 65 #include <linux/syscalls.h>
 66 #include <linux/task_work.h>
 67 #include <linux/tsacct_kern.h>
 68 
 69 #include <asm/tlb.h>
 70 
 71 #ifdef CONFIG_PARAVIRT
 72 # include <asm/paravirt.h>
 73 #endif
 74 
 75 #include "cpupri.h"
 76 #include "cpudeadline.h"
 77 
 78 #ifdef CONFIG_SCHED_DEBUG
 79 # define SCHED_WARN_ON(x)       WARN_ONCE(x, #x)
 80 #else
 81 # define SCHED_WARN_ON(x)       ({ (void)(x), 0; })
 82 #endif
 83 
 84 struct rq;
 85 struct cpuidle_state;
 86 
 87 /* task_struct::on_rq states: */
 88 #define TASK_ON_RQ_QUEUED       1
 89 #define TASK_ON_RQ_MIGRATING    2
 90 
 91 extern __read_mostly int scheduler_running;
 92 
 93 extern unsigned long calc_load_update;
 94 extern atomic_long_t calc_load_tasks;
 95 
 96 extern void calc_global_load_tick(struct rq *this_rq);
 97 extern long calc_load_fold_active(struct rq *this_rq, long adjust);
 98 
 99 /*
100  * Helpers for converting nanosecond timing to jiffy resolution
101  */
102 #define NS_TO_JIFFIES(TIME)     ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
103 
104 /*
105  * Increase resolution of nice-level calculations for 64-bit architectures.
106  * The extra resolution improves shares distribution and load balancing of
107  * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
108  * hierarchies, especially on larger systems. This is not a user-visible change
109  * and does not change the user-interface for setting shares/weights.
110  *
111  * We increase resolution only if we have enough bits to allow this increased
112  * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
113  * are pretty high and the returns do not justify the increased costs.
114  *
115  * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
116  * increase coverage and consistency always enable it on 64-bit platforms.
117  */
118 #ifdef CONFIG_64BIT
119 # define NICE_0_LOAD_SHIFT      (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
120 # define scale_load(w)          ((w) << SCHED_FIXEDPOINT_SHIFT)
121 # define scale_load_down(w)     ((w) >> SCHED_FIXEDPOINT_SHIFT)
122 #else
123 # define NICE_0_LOAD_SHIFT      (SCHED_FIXEDPOINT_SHIFT)
124 # define scale_load(w)          (w)
125 # define scale_load_down(w)     (w)
126 #endif
127 
128 /*
129  * Task weight (visible to users) and its load (invisible to users) have
130  * independent resolution, but they should be well calibrated. We use
131  * scale_load() and scale_load_down(w) to convert between them. The
132  * following must be true:
133  *
134  *  scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD
135  *
136  */
137 #define NICE_0_LOAD             (1L << NICE_0_LOAD_SHIFT)
138 
139 /*
140  * Single value that decides SCHED_DEADLINE internal math precision.
141  * 10 -> just above 1us
142  * 9  -> just above 0.5us
143  */
144 #define DL_SCALE                10
145 
146 /*
147  * Single value that denotes runtime == period, ie unlimited time.
148  */
149 #define RUNTIME_INF             ((u64)~0ULL)
150 
151 static inline int idle_policy(int policy)
152 {
153         return policy == SCHED_IDLE;
154 }
155 static inline int fair_policy(int policy)
156 {
157         return policy == SCHED_NORMAL || policy == SCHED_BATCH;
158 }
159 
160 static inline int rt_policy(int policy)
161 {
162         return policy == SCHED_FIFO || policy == SCHED_RR;
163 }
164 
165 static inline int dl_policy(int policy)
166 {
167         return policy == SCHED_DEADLINE;
168 }
169 static inline bool valid_policy(int policy)
170 {
171         return idle_policy(policy) || fair_policy(policy) ||
172                 rt_policy(policy) || dl_policy(policy);
173 }
174 
175 static inline int task_has_idle_policy(struct task_struct *p)
176 {
177         return idle_policy(p->policy);
178 }
179 
180 static inline int task_has_rt_policy(struct task_struct *p)
181 {
182         return rt_policy(p->policy);
183 }
184 
185 static inline int task_has_dl_policy(struct task_struct *p)
186 {
187         return dl_policy(p->policy);
188 }
189 
190 #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
191 
192 /*
193  * !! For sched_setattr_nocheck() (kernel) only !!
194  *
195  * This is actually gross. :(
196  *
197  * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
198  * tasks, but still be able to sleep. We need this on platforms that cannot
199  * atomically change clock frequency. Remove once fast switching will be
200  * available on such platforms.
201  *
202  * SUGOV stands for SchedUtil GOVernor.
203  */
204 #define SCHED_FLAG_SUGOV        0x10000000
205 
206 static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se)
207 {
208 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
209         return unlikely(dl_se->flags & SCHED_FLAG_SUGOV);
210 #else
211         return false;
212 #endif
213 }
214 
215 /*
216  * Tells if entity @a should preempt entity @b.
217  */
218 static inline bool
219 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
220 {
221         return dl_entity_is_special(a) ||
222                dl_time_before(a->deadline, b->deadline);
223 }
224 
225 /*
226  * This is the priority-queue data structure of the RT scheduling class:
227  */
228 struct rt_prio_array {
229         DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
230         struct list_head queue[MAX_RT_PRIO];
231 };
232 
233 struct rt_bandwidth {
234         /* nests inside the rq lock: */
235         raw_spinlock_t          rt_runtime_lock;
236         ktime_t                 rt_period;
237         u64                     rt_runtime;
238         struct hrtimer          rt_period_timer;
239         unsigned int            rt_period_active;
240 };
241 
242 void __dl_clear_params(struct task_struct *p);
243 
244 /*
245  * To keep the bandwidth of -deadline tasks and groups under control
246  * we need some place where:
247  *  - store the maximum -deadline bandwidth of the system (the group);
248  *  - cache the fraction of that bandwidth that is currently allocated.
249  *
250  * This is all done in the data structure below. It is similar to the
251  * one used for RT-throttling (rt_bandwidth), with the main difference
252  * that, since here we are only interested in admission control, we
253  * do not decrease any runtime while the group "executes", neither we
254  * need a timer to replenish it.
255  *
256  * With respect to SMP, the bandwidth is given on a per-CPU basis,
257  * meaning that:
258  *  - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU;
259  *  - dl_total_bw array contains, in the i-eth element, the currently
260  *    allocated bandwidth on the i-eth CPU.
261  * Moreover, groups consume bandwidth on each CPU, while tasks only
262  * consume bandwidth on the CPU they're running on.
263  * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw
264  * that will be shown the next time the proc or cgroup controls will
265  * be red. It on its turn can be changed by writing on its own
266  * control.
267  */
268 struct dl_bandwidth {
269         raw_spinlock_t          dl_runtime_lock;
270         u64                     dl_runtime;
271         u64                     dl_period;
272 };
273 
274 static inline int dl_bandwidth_enabled(void)
275 {
276         return sysctl_sched_rt_runtime >= 0;
277 }
278 
279 struct dl_bw {
280         raw_spinlock_t          lock;
281         u64                     bw;
282         u64                     total_bw;
283 };
284 
285 static inline void __dl_update(struct dl_bw *dl_b, s64 bw);
286 
287 static inline
288 void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
289 {
290         dl_b->total_bw -= tsk_bw;
291         __dl_update(dl_b, (s32)tsk_bw / cpus);
292 }
293 
294 static inline
295 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
296 {
297         dl_b->total_bw += tsk_bw;
298         __dl_update(dl_b, -((s32)tsk_bw / cpus));
299 }
300 
301 static inline
302 bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw)
303 {
304         return dl_b->bw != -1 &&
305                dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw;
306 }
307 
308 extern void dl_change_utilization(struct task_struct *p, u64 new_bw);
309 extern void init_dl_bw(struct dl_bw *dl_b);
310 extern int  sched_dl_global_validate(void);
311 extern void sched_dl_do_global(void);
312 extern int  sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr);
313 extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr);
314 extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr);
315 extern bool __checkparam_dl(const struct sched_attr *attr);
316 extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr);
317 extern int  dl_task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed);
318 extern int  dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
319 extern bool dl_cpu_busy(unsigned int cpu);
320 
321 #ifdef CONFIG_CGROUP_SCHED
322 
323 #include <linux/cgroup.h>
324 #include <linux/psi.h>
325 
326 struct cfs_rq;
327 struct rt_rq;
328 
329 extern struct list_head task_groups;
330 
331 struct cfs_bandwidth {
332 #ifdef CONFIG_CFS_BANDWIDTH
333         raw_spinlock_t          lock;
334         ktime_t                 period;
335         u64                     quota;
336         u64                     runtime;
337         s64                     hierarchical_quota;
338 
339         u8                      idle;
340         u8                      period_active;
341         u8                      distribute_running;
342         u8                      slack_started;
343         struct hrtimer          period_timer;
344         struct hrtimer          slack_timer;
345         struct list_head        throttled_cfs_rq;
346 
347         /* Statistics: */
348         int                     nr_periods;
349         int                     nr_throttled;
350         u64                     throttled_time;
351 #endif
352 };
353 
354 /* Task group related information */
355 struct task_group {
356         struct cgroup_subsys_state css;
357 
358 #ifdef CONFIG_FAIR_GROUP_SCHED
359         /* schedulable entities of this group on each CPU */
360         struct sched_entity     **se;
361         /* runqueue "owned" by this group on each CPU */
362         struct cfs_rq           **cfs_rq;
363         unsigned long           shares;
364 
365 #ifdef  CONFIG_SMP
366         /*
367          * load_avg can be heavily contended at clock tick time, so put
368          * it in its own cacheline separated from the fields above which
369          * will also be accessed at each tick.
370          */
371         atomic_long_t           load_avg ____cacheline_aligned;
372 #endif
373 #endif
374 
375 #ifdef CONFIG_RT_GROUP_SCHED
376         struct sched_rt_entity  **rt_se;
377         struct rt_rq            **rt_rq;
378 
379         struct rt_bandwidth     rt_bandwidth;
380 #endif
381 
382         struct rcu_head         rcu;
383         struct list_head        list;
384 
385         struct task_group       *parent;
386         struct list_head        siblings;
387         struct list_head        children;
388 
389 #ifdef CONFIG_SCHED_AUTOGROUP
390         struct autogroup        *autogroup;
391 #endif
392 
393         struct cfs_bandwidth    cfs_bandwidth;
394 };
395 
396 #ifdef CONFIG_FAIR_GROUP_SCHED
397 #define ROOT_TASK_GROUP_LOAD    NICE_0_LOAD
398 
399 /*
400  * A weight of 0 or 1 can cause arithmetics problems.
401  * A weight of a cfs_rq is the sum of weights of which entities
402  * are queued on this cfs_rq, so a weight of a entity should not be
403  * too large, so as the shares value of a task group.
404  * (The default weight is 1024 - so there's no practical
405  *  limitation from this.)
406  */
407 #define MIN_SHARES              (1UL <<  1)
408 #define MAX_SHARES              (1UL << 18)
409 #endif
410 
411 typedef int (*tg_visitor)(struct task_group *, void *);
412 
413 extern int walk_tg_tree_from(struct task_group *from,
414                              tg_visitor down, tg_visitor up, void *data);
415 
416 /*
417  * Iterate the full tree, calling @down when first entering a node and @up when
418  * leaving it for the final time.
419  *
420  * Caller must hold rcu_lock or sufficient equivalent.
421  */
422 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
423 {
424         return walk_tg_tree_from(&root_task_group, down, up, data);
425 }
426 
427 extern int tg_nop(struct task_group *tg, void *data);
428 
429 extern void free_fair_sched_group(struct task_group *tg);
430 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
431 extern void online_fair_sched_group(struct task_group *tg);
432 extern void unregister_fair_sched_group(struct task_group *tg);
433 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
434                         struct sched_entity *se, int cpu,
435                         struct sched_entity *parent);
436 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
437 
438 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
439 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
440 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
441 
442 extern void free_rt_sched_group(struct task_group *tg);
443 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
444 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
445                 struct sched_rt_entity *rt_se, int cpu,
446                 struct sched_rt_entity *parent);
447 extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us);
448 extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us);
449 extern long sched_group_rt_runtime(struct task_group *tg);
450 extern long sched_group_rt_period(struct task_group *tg);
451 extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk);
452 
453 extern struct task_group *sched_create_group(struct task_group *parent);
454 extern void sched_online_group(struct task_group *tg,
455                                struct task_group *parent);
456 extern void sched_destroy_group(struct task_group *tg);
457 extern void sched_offline_group(struct task_group *tg);
458 
459 extern void sched_move_task(struct task_struct *tsk);
460 
461 #ifdef CONFIG_FAIR_GROUP_SCHED
462 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
463 
464 #ifdef CONFIG_SMP
465 extern void set_task_rq_fair(struct sched_entity *se,
466                              struct cfs_rq *prev, struct cfs_rq *next);
467 #else /* !CONFIG_SMP */
468 static inline void set_task_rq_fair(struct sched_entity *se,
469                              struct cfs_rq *prev, struct cfs_rq *next) { }
470 #endif /* CONFIG_SMP */
471 #endif /* CONFIG_FAIR_GROUP_SCHED */
472 
473 #else /* CONFIG_CGROUP_SCHED */
474 
475 struct cfs_bandwidth { };
476 
477 #endif  /* CONFIG_CGROUP_SCHED */
478 
479 /* CFS-related fields in a runqueue */
480 struct cfs_rq {
481         struct load_weight      load;
482         unsigned long           runnable_weight;
483         unsigned int            nr_running;
484         unsigned int            h_nr_running;
485 
486         u64                     exec_clock;
487         u64                     min_vruntime;
488 #ifndef CONFIG_64BIT
489         u64                     min_vruntime_copy;
490 #endif
491 
492         struct rb_root_cached   tasks_timeline;
493 
494         /*
495          * 'curr' points to currently running entity on this cfs_rq.
496          * It is set to NULL otherwise (i.e when none are currently running).
497          */
498         struct sched_entity     *curr;
499         struct sched_entity     *next;
500         struct sched_entity     *last;
501         struct sched_entity     *skip;
502 
503 #ifdef  CONFIG_SCHED_DEBUG
504         unsigned int            nr_spread_over;
505 #endif
506 
507 #ifdef CONFIG_SMP
508         /*
509          * CFS load tracking
510          */
511         struct sched_avg        avg;
512 #ifndef CONFIG_64BIT
513         u64                     load_last_update_time_copy;
514 #endif
515         struct {
516                 raw_spinlock_t  lock ____cacheline_aligned;
517                 int             nr;
518                 unsigned long   load_avg;
519                 unsigned long   util_avg;
520                 unsigned long   runnable_sum;
521         } removed;
522 
523 #ifdef CONFIG_FAIR_GROUP_SCHED
524         unsigned long           tg_load_avg_contrib;
525         long                    propagate;
526         long                    prop_runnable_sum;
527 
528         /*
529          *   h_load = weight * f(tg)
530          *
531          * Where f(tg) is the recursive weight fraction assigned to
532          * this group.
533          */
534         unsigned long           h_load;
535         u64                     last_h_load_update;
536         struct sched_entity     *h_load_next;
537 #endif /* CONFIG_FAIR_GROUP_SCHED */
538 #endif /* CONFIG_SMP */
539 
540 #ifdef CONFIG_FAIR_GROUP_SCHED
541         struct rq               *rq;    /* CPU runqueue to which this cfs_rq is attached */
542 
543         /*
544          * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
545          * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
546          * (like users, containers etc.)
547          *
548          * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
549          * This list is used during load balance.
550          */
551         int                     on_list;
552         struct list_head        leaf_cfs_rq_list;
553         struct task_group       *tg;    /* group that "owns" this runqueue */
554 
555 #ifdef CONFIG_CFS_BANDWIDTH
556         int                     runtime_enabled;
557         s64                     runtime_remaining;
558 
559         u64                     throttled_clock;
560         u64                     throttled_clock_task;
561         u64                     throttled_clock_task_time;
562         int                     throttled;
563         int                     throttle_count;
564         struct list_head        throttled_list;
565 #endif /* CONFIG_CFS_BANDWIDTH */
566 #endif /* CONFIG_FAIR_GROUP_SCHED */
567 };
568 
569 static inline int rt_bandwidth_enabled(void)
570 {
571         return sysctl_sched_rt_runtime >= 0;
572 }
573 
574 /* RT IPI pull logic requires IRQ_WORK */
575 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
576 # define HAVE_RT_PUSH_IPI
577 #endif
578 
579 /* Real-Time classes' related field in a runqueue: */
580 struct rt_rq {
581         struct rt_prio_array    active;
582         unsigned int            rt_nr_running;
583         unsigned int            rr_nr_running;
584 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
585         struct {
586                 int             curr; /* highest queued rt task prio */
587 #ifdef CONFIG_SMP
588                 int             next; /* next highest */
589 #endif
590         } highest_prio;
591 #endif
592 #ifdef CONFIG_SMP
593         unsigned long           rt_nr_migratory;
594         unsigned long           rt_nr_total;
595         int                     overloaded;
596         struct plist_head       pushable_tasks;
597 
598 #endif /* CONFIG_SMP */
599         int                     rt_queued;
600 
601         int                     rt_throttled;
602         u64                     rt_time;
603         u64                     rt_runtime;
604         /* Nests inside the rq lock: */
605         raw_spinlock_t          rt_runtime_lock;
606 
607 #ifdef CONFIG_RT_GROUP_SCHED
608         unsigned long           rt_nr_boosted;
609 
610         struct rq               *rq;
611         struct task_group       *tg;
612 #endif
613 };
614 
615 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
616 {
617         return rt_rq->rt_queued && rt_rq->rt_nr_running;
618 }
619 
620 /* Deadline class' related fields in a runqueue */
621 struct dl_rq {
622         /* runqueue is an rbtree, ordered by deadline */
623         struct rb_root_cached   root;
624 
625         unsigned long           dl_nr_running;
626 
627 #ifdef CONFIG_SMP
628         /*
629          * Deadline values of the currently executing and the
630          * earliest ready task on this rq. Caching these facilitates
631          * the decision whether or not a ready but not running task
632          * should migrate somewhere else.
633          */
634         struct {
635                 u64             curr;
636                 u64             next;
637         } earliest_dl;
638 
639         unsigned long           dl_nr_migratory;
640         int                     overloaded;
641 
642         /*
643          * Tasks on this rq that can be pushed away. They are kept in
644          * an rb-tree, ordered by tasks' deadlines, with caching
645          * of the leftmost (earliest deadline) element.
646          */
647         struct rb_root_cached   pushable_dl_tasks_root;
648 #else
649         struct dl_bw            dl_bw;
650 #endif
651         /*
652          * "Active utilization" for this runqueue: increased when a
653          * task wakes up (becomes TASK_RUNNING) and decreased when a
654          * task blocks
655          */
656         u64                     running_bw;
657 
658         /*
659          * Utilization of the tasks "assigned" to this runqueue (including
660          * the tasks that are in runqueue and the tasks that executed on this
661          * CPU and blocked). Increased when a task moves to this runqueue, and
662          * decreased when the task moves away (migrates, changes scheduling
663          * policy, or terminates).
664          * This is needed to compute the "inactive utilization" for the
665          * runqueue (inactive utilization = this_bw - running_bw).
666          */
667         u64                     this_bw;
668         u64                     extra_bw;
669 
670         /*
671          * Inverse of the fraction of CPU utilization that can be reclaimed
672          * by the GRUB algorithm.
673          */
674         u64                     bw_ratio;
675 };
676 
677 #ifdef CONFIG_FAIR_GROUP_SCHED
678 /* An entity is a task if it doesn't "own" a runqueue */
679 #define entity_is_task(se)      (!se->my_q)
680 #else
681 #define entity_is_task(se)      1
682 #endif
683 
684 #ifdef CONFIG_SMP
685 /*
686  * XXX we want to get rid of these helpers and use the full load resolution.
687  */
688 static inline long se_weight(struct sched_entity *se)
689 {
690         return scale_load_down(se->load.weight);
691 }
692 
693 static inline long se_runnable(struct sched_entity *se)
694 {
695         return scale_load_down(se->runnable_weight);
696 }
697 
698 static inline bool sched_asym_prefer(int a, int b)
699 {
700         return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b);
701 }
702 
703 struct perf_domain {
704         struct em_perf_domain *em_pd;
705         struct perf_domain *next;
706         struct rcu_head rcu;
707 };
708 
709 /* Scheduling group status flags */
710 #define SG_OVERLOAD             0x1 /* More than one runnable task on a CPU. */
711 #define SG_OVERUTILIZED         0x2 /* One or more CPUs are over-utilized. */
712 
713 /*
714  * We add the notion of a root-domain which will be used to define per-domain
715  * variables. Each exclusive cpuset essentially defines an island domain by
716  * fully partitioning the member CPUs from any other cpuset. Whenever a new
717  * exclusive cpuset is created, we also create and attach a new root-domain
718  * object.
719  *
720  */
721 struct root_domain {
722         atomic_t                refcount;
723         atomic_t                rto_count;
724         struct rcu_head         rcu;
725         cpumask_var_t           span;
726         cpumask_var_t           online;
727 
728         /*
729          * Indicate pullable load on at least one CPU, e.g:
730          * - More than one runnable task
731          * - Running task is misfit
732          */
733         int                     overload;
734 
735         /* Indicate one or more cpus over-utilized (tipping point) */
736         int                     overutilized;
737 
738         /*
739          * The bit corresponding to a CPU gets set here if such CPU has more
740          * than one runnable -deadline task (as it is below for RT tasks).
741          */
742         cpumask_var_t           dlo_mask;
743         atomic_t                dlo_count;
744         struct dl_bw            dl_bw;
745         struct cpudl            cpudl;
746 
747 #ifdef HAVE_RT_PUSH_IPI
748         /*
749          * For IPI pull requests, loop across the rto_mask.
750          */
751         struct irq_work         rto_push_work;
752         raw_spinlock_t          rto_lock;
753         /* These are only updated and read within rto_lock */
754         int                     rto_loop;
755         int                     rto_cpu;
756         /* These atomics are updated outside of a lock */
757         atomic_t                rto_loop_next;
758         atomic_t                rto_loop_start;
759 #endif
760         /*
761          * The "RT overload" flag: it gets set if a CPU has more than
762          * one runnable RT task.
763          */
764         cpumask_var_t           rto_mask;
765         struct cpupri           cpupri;
766 
767         unsigned long           max_cpu_capacity;
768 
769         /*
770          * NULL-terminated list of performance domains intersecting with the
771          * CPUs of the rd. Protected by RCU.
772          */
773         struct perf_domain __rcu *pd;
774 };
775 
776 extern struct root_domain def_root_domain;
777 extern struct mutex sched_domains_mutex;
778 
779 extern void init_defrootdomain(void);
780 extern int sched_init_domains(const struct cpumask *cpu_map);
781 extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
782 extern void sched_get_rd(struct root_domain *rd);
783 extern void sched_put_rd(struct root_domain *rd);
784 
785 #ifdef HAVE_RT_PUSH_IPI
786 extern void rto_push_irq_work_func(struct irq_work *work);
787 #endif
788 #endif /* CONFIG_SMP */
789 
790 #ifdef CONFIG_UCLAMP_TASK
791 /*
792  * struct uclamp_bucket - Utilization clamp bucket
793  * @value: utilization clamp value for tasks on this clamp bucket
794  * @tasks: number of RUNNABLE tasks on this clamp bucket
795  *
796  * Keep track of how many tasks are RUNNABLE for a given utilization
797  * clamp value.
798  */
799 struct uclamp_bucket {
800         unsigned long value : bits_per(SCHED_CAPACITY_SCALE);
801         unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE);
802 };
803 
804 /*
805  * struct uclamp_rq - rq's utilization clamp
806  * @value: currently active clamp values for a rq
807  * @bucket: utilization clamp buckets affecting a rq
808  *
809  * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
810  * A clamp value is affecting a rq when there is at least one task RUNNABLE
811  * (or actually running) with that value.
812  *
813  * There are up to UCLAMP_CNT possible different clamp values, currently there
814  * are only two: minimum utilization and maximum utilization.
815  *
816  * All utilization clamping values are MAX aggregated, since:
817  * - for util_min: we want to run the CPU at least at the max of the minimum
818  *   utilization required by its currently RUNNABLE tasks.
819  * - for util_max: we want to allow the CPU to run up to the max of the
820  *   maximum utilization allowed by its currently RUNNABLE tasks.
821  *
822  * Since on each system we expect only a limited number of different
823  * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
824  * the metrics required to compute all the per-rq utilization clamp values.
825  */
826 struct uclamp_rq {
827         unsigned int value;
828         struct uclamp_bucket bucket[UCLAMP_BUCKETS];
829 };
830 #endif /* CONFIG_UCLAMP_TASK */
831 
832 /*
833  * This is the main, per-CPU runqueue data structure.
834  *
835  * Locking rule: those places that want to lock multiple runqueues
836  * (such as the load balancing or the thread migration code), lock
837  * acquire operations must be ordered by ascending &runqueue.
838  */
839 struct rq {
840         /* runqueue lock: */
841         raw_spinlock_t          lock;
842 
843         /*
844          * nr_running and cpu_load should be in the same cacheline because
845          * remote CPUs use both these fields when doing load calculation.
846          */
847         unsigned int            nr_running;
848 #ifdef CONFIG_NUMA_BALANCING
849         unsigned int            nr_numa_running;
850         unsigned int            nr_preferred_running;
851         unsigned int            numa_migrate_on;
852 #endif
853 #ifdef CONFIG_NO_HZ_COMMON
854 #ifdef CONFIG_SMP
855         unsigned long           last_load_update_tick;
856         unsigned long           last_blocked_load_update_tick;
857         unsigned int            has_blocked_load;
858 #endif /* CONFIG_SMP */
859         unsigned int            nohz_tick_stopped;
860         atomic_t nohz_flags;
861 #endif /* CONFIG_NO_HZ_COMMON */
862 
863         unsigned long           nr_load_updates;
864         u64                     nr_switches;
865 
866 #ifdef CONFIG_UCLAMP_TASK
867         /* Utilization clamp values based on CPU's RUNNABLE tasks */
868         struct uclamp_rq        uclamp[UCLAMP_CNT] ____cacheline_aligned;
869         unsigned int            uclamp_flags;
870 #define UCLAMP_FLAG_IDLE 0x01
871 #endif
872 
873         struct cfs_rq           cfs;
874         struct rt_rq            rt;
875         struct dl_rq            dl;
876 
877 #ifdef CONFIG_FAIR_GROUP_SCHED
878         /* list of leaf cfs_rq on this CPU: */
879         struct list_head        leaf_cfs_rq_list;
880         struct list_head        *tmp_alone_branch;
881 #endif /* CONFIG_FAIR_GROUP_SCHED */
882 
883         /*
884          * This is part of a global counter where only the total sum
885          * over all CPUs matters. A task can increase this counter on
886          * one CPU and if it got migrated afterwards it may decrease
887          * it on another CPU. Always updated under the runqueue lock:
888          */
889         unsigned long           nr_uninterruptible;
890 
891         struct task_struct      *curr;
892         struct task_struct      *idle;
893         struct task_struct      *stop;
894         unsigned long           next_balance;
895         struct mm_struct        *prev_mm;
896 
897         unsigned int            clock_update_flags;
898         u64                     clock;
899         /* Ensure that all clocks are in the same cache line */
900         u64                     clock_task ____cacheline_aligned;
901         u64                     clock_pelt;
902         unsigned long           lost_idle_time;
903 
904         atomic_t                nr_iowait;
905 
906 #ifdef CONFIG_SMP
907         struct root_domain              *rd;
908         struct sched_domain __rcu       *sd;
909 
910         unsigned long           cpu_capacity;
911         unsigned long           cpu_capacity_orig;
912 
913         struct callback_head    *balance_callback;
914 
915         unsigned char           idle_balance;
916 
917         unsigned long           misfit_task_load;
918 
919         /* For active balancing */
920         int                     active_balance;
921         int                     push_cpu;
922         struct cpu_stop_work    active_balance_work;
923 
924         /* CPU of this runqueue: */
925         int                     cpu;
926         int                     online;
927 
928         struct list_head cfs_tasks;
929 
930         struct sched_avg        avg_rt;
931         struct sched_avg        avg_dl;
932 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
933         struct sched_avg        avg_irq;
934 #endif
935         u64                     idle_stamp;
936         u64                     avg_idle;
937 
938         /* This is used to determine avg_idle's max value */
939         u64                     max_idle_balance_cost;
940 #endif
941 
942 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
943         u64                     prev_irq_time;
944 #endif
945 #ifdef CONFIG_PARAVIRT
946         u64                     prev_steal_time;
947 #endif
948 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
949         u64                     prev_steal_time_rq;
950 #endif
951 
952         /* calc_load related fields */
953         unsigned long           calc_load_update;
954         long                    calc_load_active;
955 
956 #ifdef CONFIG_SCHED_HRTICK
957 #ifdef CONFIG_SMP
958         int                     hrtick_csd_pending;
959         call_single_data_t      hrtick_csd;
960 #endif
961         struct hrtimer          hrtick_timer;
962 #endif
963 
964 #ifdef CONFIG_SCHEDSTATS
965         /* latency stats */
966         struct sched_info       rq_sched_info;
967         unsigned long long      rq_cpu_time;
968         /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
969 
970         /* sys_sched_yield() stats */
971         unsigned int            yld_count;
972 
973         /* schedule() stats */
974         unsigned int            sched_count;
975         unsigned int            sched_goidle;
976 
977         /* try_to_wake_up() stats */
978         unsigned int            ttwu_count;
979         unsigned int            ttwu_local;
980 #endif
981 
982 #ifdef CONFIG_SMP
983         struct llist_head       wake_list;
984 #endif
985 
986 #ifdef CONFIG_CPU_IDLE
987         /* Must be inspected within a rcu lock section */
988         struct cpuidle_state    *idle_state;
989 #endif
990 };
991 
992 #ifdef CONFIG_FAIR_GROUP_SCHED
993 
994 /* CPU runqueue to which this cfs_rq is attached */
995 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
996 {
997         return cfs_rq->rq;
998 }
999 
1000 #else
1001 
1002 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
1003 {
1004         return container_of(cfs_rq, struct rq, cfs);
1005 }
1006 #endif
1007 
1008 static inline int cpu_of(struct rq *rq)
1009 {
1010 #ifdef CONFIG_SMP
1011         return rq->cpu;
1012 #else
1013         return 0;
1014 #endif
1015 }
1016 
1017 
1018 #ifdef CONFIG_SCHED_SMT
1019 extern void __update_idle_core(struct rq *rq);
1020 
1021 static inline void update_idle_core(struct rq *rq)
1022 {
1023         if (static_branch_unlikely(&sched_smt_present))
1024                 __update_idle_core(rq);
1025 }
1026 
1027 #else
1028 static inline void update_idle_core(struct rq *rq) { }
1029 #endif
1030 
1031 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1032 
1033 #define cpu_rq(cpu)             (&per_cpu(runqueues, (cpu)))
1034 #define this_rq()               this_cpu_ptr(&runqueues)
1035 #define task_rq(p)              cpu_rq(task_cpu(p))
1036 #define cpu_curr(cpu)           (cpu_rq(cpu)->curr)
1037 #define raw_rq()                raw_cpu_ptr(&runqueues)
1038 
1039 extern void update_rq_clock(struct rq *rq);
1040 
1041 static inline u64 __rq_clock_broken(struct rq *rq)
1042 {
1043         return READ_ONCE(rq->clock);
1044 }
1045 
1046 /*
1047  * rq::clock_update_flags bits
1048  *
1049  * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1050  *  call to __schedule(). This is an optimisation to avoid
1051  *  neighbouring rq clock updates.
1052  *
1053  * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1054  *  in effect and calls to update_rq_clock() are being ignored.
1055  *
1056  * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1057  *  made to update_rq_clock() since the last time rq::lock was pinned.
1058  *
1059  * If inside of __schedule(), clock_update_flags will have been
1060  * shifted left (a left shift is a cheap operation for the fast path
1061  * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1062  *
1063  *      if (rq-clock_update_flags >= RQCF_UPDATED)
1064  *
1065  * to check if %RQCF_UPADTED is set. It'll never be shifted more than
1066  * one position though, because the next rq_unpin_lock() will shift it
1067  * back.
1068  */
1069 #define RQCF_REQ_SKIP           0x01
1070 #define RQCF_ACT_SKIP           0x02
1071 #define RQCF_UPDATED            0x04
1072 
1073 static inline void assert_clock_updated(struct rq *rq)
1074 {
1075         /*
1076          * The only reason for not seeing a clock update since the
1077          * last rq_pin_lock() is if we're currently skipping updates.
1078          */
1079         SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP);
1080 }
1081 
1082 static inline u64 rq_clock(struct rq *rq)
1083 {
1084         lockdep_assert_held(&rq->lock);
1085         assert_clock_updated(rq);
1086 
1087         return rq->clock;
1088 }
1089 
1090 static inline u64 rq_clock_task(struct rq *rq)
1091 {
1092         lockdep_assert_held(&rq->lock);
1093         assert_clock_updated(rq);
1094 
1095         return rq->clock_task;
1096 }
1097 
1098 static inline void rq_clock_skip_update(struct rq *rq)
1099 {
1100         lockdep_assert_held(&rq->lock);
1101         rq->clock_update_flags |= RQCF_REQ_SKIP;
1102 }
1103 
1104 /*
1105  * See rt task throttling, which is the only time a skip
1106  * request is cancelled.
1107  */
1108 static inline void rq_clock_cancel_skipupdate(struct rq *rq)
1109 {
1110         lockdep_assert_held(&rq->lock);
1111         rq->clock_update_flags &= ~RQCF_REQ_SKIP;
1112 }
1113 
1114 struct rq_flags {
1115         unsigned long flags;
1116         struct pin_cookie cookie;
1117 #ifdef CONFIG_SCHED_DEBUG
1118         /*
1119          * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1120          * current pin context is stashed here in case it needs to be
1121          * restored in rq_repin_lock().
1122          */
1123         unsigned int clock_update_flags;
1124 #endif
1125 };
1126 
1127 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
1128 {
1129         rf->cookie = lockdep_pin_lock(&rq->lock);
1130 
1131 #ifdef CONFIG_SCHED_DEBUG
1132         rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
1133         rf->clock_update_flags = 0;
1134 #endif
1135 }
1136 
1137 static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
1138 {
1139 #ifdef CONFIG_SCHED_DEBUG
1140         if (rq->clock_update_flags > RQCF_ACT_SKIP)
1141                 rf->clock_update_flags = RQCF_UPDATED;
1142 #endif
1143 
1144         lockdep_unpin_lock(&rq->lock, rf->cookie);
1145 }
1146 
1147 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
1148 {
1149         lockdep_repin_lock(&rq->lock, rf->cookie);
1150 
1151 #ifdef CONFIG_SCHED_DEBUG
1152         /*
1153          * Restore the value we stashed in @rf for this pin context.
1154          */
1155         rq->clock_update_flags |= rf->clock_update_flags;
1156 #endif
1157 }
1158 
1159 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1160         __acquires(rq->lock);
1161 
1162 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1163         __acquires(p->pi_lock)
1164         __acquires(rq->lock);
1165 
1166 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1167         __releases(rq->lock)
1168 {
1169         rq_unpin_lock(rq, rf);
1170         raw_spin_unlock(&rq->lock);
1171 }
1172 
1173 static inline void
1174 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1175         __releases(rq->lock)
1176         __releases(p->pi_lock)
1177 {
1178         rq_unpin_lock(rq, rf);
1179         raw_spin_unlock(&rq->lock);
1180         raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1181 }
1182 
1183 static inline void
1184 rq_lock_irqsave(struct rq *rq, struct rq_flags *rf)
1185         __acquires(rq->lock)
1186 {
1187         raw_spin_lock_irqsave(&rq->lock, rf->flags);
1188         rq_pin_lock(rq, rf);
1189 }
1190 
1191 static inline void
1192 rq_lock_irq(struct rq *rq, struct rq_flags *rf)
1193         __acquires(rq->lock)
1194 {
1195         raw_spin_lock_irq(&rq->lock);
1196         rq_pin_lock(rq, rf);
1197 }
1198 
1199 static inline void
1200 rq_lock(struct rq *rq, struct rq_flags *rf)
1201         __acquires(rq->lock)
1202 {
1203         raw_spin_lock(&rq->lock);
1204         rq_pin_lock(rq, rf);
1205 }
1206 
1207 static inline void
1208 rq_relock(struct rq *rq, struct rq_flags *rf)
1209         __acquires(rq->lock)
1210 {
1211         raw_spin_lock(&rq->lock);
1212         rq_repin_lock(rq, rf);
1213 }
1214 
1215 static inline void
1216 rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf)
1217         __releases(rq->lock)
1218 {
1219         rq_unpin_lock(rq, rf);
1220         raw_spin_unlock_irqrestore(&rq->lock, rf->flags);
1221 }
1222 
1223 static inline void
1224 rq_unlock_irq(struct rq *rq, struct rq_flags *rf)
1225         __releases(rq->lock)
1226 {
1227         rq_unpin_lock(rq, rf);
1228         raw_spin_unlock_irq(&rq->lock);
1229 }
1230 
1231 static inline void
1232 rq_unlock(struct rq *rq, struct rq_flags *rf)
1233         __releases(rq->lock)
1234 {
1235         rq_unpin_lock(rq, rf);
1236         raw_spin_unlock(&rq->lock);
1237 }
1238 
1239 static inline struct rq *
1240 this_rq_lock_irq(struct rq_flags *rf)
1241         __acquires(rq->lock)
1242 {
1243         struct rq *rq;
1244 
1245         local_irq_disable();
1246         rq = this_rq();
1247         rq_lock(rq, rf);
1248         return rq;
1249 }
1250 
1251 #ifdef CONFIG_NUMA
1252 enum numa_topology_type {
1253         NUMA_DIRECT,
1254         NUMA_GLUELESS_MESH,
1255         NUMA_BACKPLANE,
1256 };
1257 extern enum numa_topology_type sched_numa_topology_type;
1258 extern int sched_max_numa_distance;
1259 extern bool find_numa_distance(int distance);
1260 #endif
1261 
1262 #ifdef CONFIG_NUMA
1263 extern void sched_init_numa(void);
1264 extern void sched_domains_numa_masks_set(unsigned int cpu);
1265 extern void sched_domains_numa_masks_clear(unsigned int cpu);
1266 #else
1267 static inline void sched_init_numa(void) { }
1268 static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
1269 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
1270 #endif
1271 
1272 #ifdef CONFIG_NUMA_BALANCING
1273 /* The regions in numa_faults array from task_struct */
1274 enum numa_faults_stats {
1275         NUMA_MEM = 0,
1276         NUMA_CPU,
1277         NUMA_MEMBUF,
1278         NUMA_CPUBUF
1279 };
1280 extern void sched_setnuma(struct task_struct *p, int node);
1281 extern int migrate_task_to(struct task_struct *p, int cpu);
1282 extern int migrate_swap(struct task_struct *p, struct task_struct *t,
1283                         int cpu, int scpu);
1284 extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p);
1285 #else
1286 static inline void
1287 init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
1288 {
1289 }
1290 #endif /* CONFIG_NUMA_BALANCING */
1291 
1292 #ifdef CONFIG_SMP
1293 
1294 static inline void
1295 queue_balance_callback(struct rq *rq,
1296                        struct callback_head *head,
1297                        void (*func)(struct rq *rq))
1298 {
1299         lockdep_assert_held(&rq->lock);
1300 
1301         if (unlikely(head->next))
1302                 return;
1303 
1304         head->func = (void (*)(struct callback_head *))func;
1305         head->next = rq->balance_callback;
1306         rq->balance_callback = head;
1307 }
1308 
1309 extern void sched_ttwu_pending(void);
1310 
1311 #define rcu_dereference_check_sched_domain(p) \
1312         rcu_dereference_check((p), \
1313                               lockdep_is_held(&sched_domains_mutex))
1314 
1315 /*
1316  * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1317  * See destroy_sched_domains: call_rcu for details.
1318  *
1319  * The domain tree of any CPU may only be accessed from within
1320  * preempt-disabled sections.
1321  */
1322 #define for_each_domain(cpu, __sd) \
1323         for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1324                         __sd; __sd = __sd->parent)
1325 
1326 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
1327 
1328 /**
1329  * highest_flag_domain - Return highest sched_domain containing flag.
1330  * @cpu:        The CPU whose highest level of sched domain is to
1331  *              be returned.
1332  * @flag:       The flag to check for the highest sched_domain
1333  *              for the given CPU.
1334  *
1335  * Returns the highest sched_domain of a CPU which contains the given flag.
1336  */
1337 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
1338 {
1339         struct sched_domain *sd, *hsd = NULL;
1340 
1341         for_each_domain(cpu, sd) {
1342                 if (!(sd->flags & flag))
1343                         break;
1344                 hsd = sd;
1345         }
1346 
1347         return hsd;
1348 }
1349 
1350 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
1351 {
1352         struct sched_domain *sd;
1353 
1354         for_each_domain(cpu, sd) {
1355                 if (sd->flags & flag)
1356                         break;
1357         }
1358 
1359         return sd;
1360 }
1361 
1362 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc);
1363 DECLARE_PER_CPU(int, sd_llc_size);
1364 DECLARE_PER_CPU(int, sd_llc_id);
1365 DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared);
1366 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa);
1367 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing);
1368 DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity);
1369 extern struct static_key_false sched_asym_cpucapacity;
1370 
1371 struct sched_group_capacity {
1372         atomic_t                ref;
1373         /*
1374          * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1375          * for a single CPU.
1376          */
1377         unsigned long           capacity;
1378         unsigned long           min_capacity;           /* Min per-CPU capacity in group */
1379         unsigned long           max_capacity;           /* Max per-CPU capacity in group */
1380         unsigned long           next_update;
1381         int                     imbalance;              /* XXX unrelated to capacity but shared group state */
1382 
1383 #ifdef CONFIG_SCHED_DEBUG
1384         int                     id;
1385 #endif
1386 
1387         unsigned long           cpumask[0];             /* Balance mask */
1388 };
1389 
1390 struct sched_group {
1391         struct sched_group      *next;                  /* Must be a circular list */
1392         atomic_t                ref;
1393 
1394         unsigned int            group_weight;
1395         struct sched_group_capacity *sgc;
1396         int                     asym_prefer_cpu;        /* CPU of highest priority in group */
1397 
1398         /*
1399          * The CPUs this group covers.
1400          *
1401          * NOTE: this field is variable length. (Allocated dynamically
1402          * by attaching extra space to the end of the structure,
1403          * depending on how many CPUs the kernel has booted up with)
1404          */
1405         unsigned long           cpumask[0];
1406 };
1407 
1408 static inline struct cpumask *sched_group_span(struct sched_group *sg)
1409 {
1410         return to_cpumask(sg->cpumask);
1411 }
1412 
1413 /*
1414  * See build_balance_mask().
1415  */
1416 static inline struct cpumask *group_balance_mask(struct sched_group *sg)
1417 {
1418         return to_cpumask(sg->sgc->cpumask);
1419 }
1420 
1421 /**
1422  * group_first_cpu - Returns the first CPU in the cpumask of a sched_group.
1423  * @group: The group whose first CPU is to be returned.
1424  */
1425 static inline unsigned int group_first_cpu(struct sched_group *group)
1426 {
1427         return cpumask_first(sched_group_span(group));
1428 }
1429 
1430 extern int group_balance_cpu(struct sched_group *sg);
1431 
1432 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
1433 void register_sched_domain_sysctl(void);
1434 void dirty_sched_domain_sysctl(int cpu);
1435 void unregister_sched_domain_sysctl(void);
1436 #else
1437 static inline void register_sched_domain_sysctl(void)
1438 {
1439 }
1440 static inline void dirty_sched_domain_sysctl(int cpu)
1441 {
1442 }
1443 static inline void unregister_sched_domain_sysctl(void)
1444 {
1445 }
1446 #endif
1447 
1448 #else
1449 
1450 static inline void sched_ttwu_pending(void) { }
1451 
1452 #endif /* CONFIG_SMP */
1453 
1454 #include "stats.h"
1455 #include "autogroup.h"
1456 
1457 #ifdef CONFIG_CGROUP_SCHED
1458 
1459 /*
1460  * Return the group to which this tasks belongs.
1461  *
1462  * We cannot use task_css() and friends because the cgroup subsystem
1463  * changes that value before the cgroup_subsys::attach() method is called,
1464  * therefore we cannot pin it and might observe the wrong value.
1465  *
1466  * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1467  * core changes this before calling sched_move_task().
1468  *
1469  * Instead we use a 'copy' which is updated from sched_move_task() while
1470  * holding both task_struct::pi_lock and rq::lock.
1471  */
1472 static inline struct task_group *task_group(struct task_struct *p)
1473 {
1474         return p->sched_task_group;
1475 }
1476 
1477 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1478 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
1479 {
1480 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1481         struct task_group *tg = task_group(p);
1482 #endif
1483 
1484 #ifdef CONFIG_FAIR_GROUP_SCHED
1485         set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
1486         p->se.cfs_rq = tg->cfs_rq[cpu];
1487         p->se.parent = tg->se[cpu];
1488 #endif
1489 
1490 #ifdef CONFIG_RT_GROUP_SCHED
1491         p->rt.rt_rq  = tg->rt_rq[cpu];
1492         p->rt.parent = tg->rt_se[cpu];
1493 #endif
1494 }
1495 
1496 #else /* CONFIG_CGROUP_SCHED */
1497 
1498 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1499 static inline struct task_group *task_group(struct task_struct *p)
1500 {
1501         return NULL;
1502 }
1503 
1504 #endif /* CONFIG_CGROUP_SCHED */
1505 
1506 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1507 {
1508         set_task_rq(p, cpu);
1509 #ifdef CONFIG_SMP
1510         /*
1511          * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1512          * successfully executed on another CPU. We must ensure that updates of
1513          * per-task data have been completed by this moment.
1514          */
1515         smp_wmb();
1516 #ifdef CONFIG_THREAD_INFO_IN_TASK
1517         WRITE_ONCE(p->cpu, cpu);
1518 #else
1519         WRITE_ONCE(task_thread_info(p)->cpu, cpu);
1520 #endif
1521         p->wake_cpu = cpu;
1522 #endif
1523 }
1524 
1525 /*
1526  * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1527  */
1528 #ifdef CONFIG_SCHED_DEBUG
1529 # include <linux/static_key.h>
1530 # define const_debug __read_mostly
1531 #else
1532 # define const_debug const
1533 #endif
1534 
1535 #define SCHED_FEAT(name, enabled)       \
1536         __SCHED_FEAT_##name ,
1537 
1538 enum {
1539 #include "features.h"
1540         __SCHED_FEAT_NR,
1541 };
1542 
1543 #undef SCHED_FEAT
1544 
1545 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_JUMP_LABEL)
1546 
1547 /*
1548  * To support run-time toggling of sched features, all the translation units
1549  * (but core.c) reference the sysctl_sched_features defined in core.c.
1550  */
1551 extern const_debug unsigned int sysctl_sched_features;
1552 
1553 #define SCHED_FEAT(name, enabled)                                       \
1554 static __always_inline bool static_branch_##name(struct static_key *key) \
1555 {                                                                       \
1556         return static_key_##enabled(key);                               \
1557 }
1558 
1559 #include "features.h"
1560 #undef SCHED_FEAT
1561 
1562 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1563 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1564 
1565 #else /* !(SCHED_DEBUG && CONFIG_JUMP_LABEL) */
1566 
1567 /*
1568  * Each translation unit has its own copy of sysctl_sched_features to allow
1569  * constants propagation at compile time and compiler optimization based on
1570  * features default.
1571  */
1572 #define SCHED_FEAT(name, enabled)       \
1573         (1UL << __SCHED_FEAT_##name) * enabled |
1574 static const_debug __maybe_unused unsigned int sysctl_sched_features =
1575 #include "features.h"
1576         0;
1577 #undef SCHED_FEAT
1578 
1579 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1580 
1581 #endif /* SCHED_DEBUG && CONFIG_JUMP_LABEL */
1582 
1583 extern struct static_key_false sched_numa_balancing;
1584 extern struct static_key_false sched_schedstats;
1585 
1586 static inline u64 global_rt_period(void)
1587 {
1588         return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1589 }
1590 
1591 static inline u64 global_rt_runtime(void)
1592 {
1593         if (sysctl_sched_rt_runtime < 0)
1594                 return RUNTIME_INF;
1595 
1596         return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1597 }
1598 
1599 static inline int task_current(struct rq *rq, struct task_struct *p)
1600 {
1601         return rq->curr == p;
1602 }
1603 
1604 static inline int task_running(struct rq *rq, struct task_struct *p)
1605 {
1606 #ifdef CONFIG_SMP
1607         return p->on_cpu;
1608 #else
1609         return task_current(rq, p);
1610 #endif
1611 }
1612 
1613 static inline int task_on_rq_queued(struct task_struct *p)
1614 {
1615         return p->on_rq == TASK_ON_RQ_QUEUED;
1616 }
1617 
1618 static inline int task_on_rq_migrating(struct task_struct *p)
1619 {
1620         return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING;
1621 }
1622 
1623 /*
1624  * wake flags
1625  */
1626 #define WF_SYNC                 0x01            /* Waker goes to sleep after wakeup */
1627 #define WF_FORK                 0x02            /* Child wakeup after fork */
1628 #define WF_MIGRATED             0x4             /* Internal use, task got migrated */
1629 
1630 /*
1631  * To aid in avoiding the subversion of "niceness" due to uneven distribution
1632  * of tasks with abnormal "nice" values across CPUs the contribution that
1633  * each task makes to its run queue's load is weighted according to its
1634  * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1635  * scaled version of the new time slice allocation that they receive on time
1636  * slice expiry etc.
1637  */
1638 
1639 #define WEIGHT_IDLEPRIO         3
1640 #define WMULT_IDLEPRIO          1431655765
1641 
1642 extern const int                sched_prio_to_weight[40];
1643 extern const u32                sched_prio_to_wmult[40];
1644 
1645 /*
1646  * {de,en}queue flags:
1647  *
1648  * DEQUEUE_SLEEP  - task is no longer runnable
1649  * ENQUEUE_WAKEUP - task just became runnable
1650  *
1651  * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1652  *                are in a known state which allows modification. Such pairs
1653  *                should preserve as much state as possible.
1654  *
1655  * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1656  *        in the runqueue.
1657  *
1658  * ENQUEUE_HEAD      - place at front of runqueue (tail if not specified)
1659  * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1660  * ENQUEUE_MIGRATED  - the task was migrated during wakeup
1661  *
1662  */
1663 
1664 #define DEQUEUE_SLEEP           0x01
1665 #define DEQUEUE_SAVE            0x02 /* Matches ENQUEUE_RESTORE */
1666 #define DEQUEUE_MOVE            0x04 /* Matches ENQUEUE_MOVE */
1667 #define DEQUEUE_NOCLOCK         0x08 /* Matches ENQUEUE_NOCLOCK */
1668 
1669 #define ENQUEUE_WAKEUP          0x01
1670 #define ENQUEUE_RESTORE         0x02
1671 #define ENQUEUE_MOVE            0x04
1672 #define ENQUEUE_NOCLOCK         0x08
1673 
1674 #define ENQUEUE_HEAD            0x10
1675 #define ENQUEUE_REPLENISH       0x20
1676 #ifdef CONFIG_SMP
1677 #define ENQUEUE_MIGRATED        0x40
1678 #else
1679 #define ENQUEUE_MIGRATED        0x00
1680 #endif
1681 
1682 #define RETRY_TASK              ((void *)-1UL)
1683 
1684 struct sched_class {
1685         const struct sched_class *next;
1686 
1687 #ifdef CONFIG_UCLAMP_TASK
1688         int uclamp_enabled;
1689 #endif
1690 
1691         void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1692         void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1693         void (*yield_task)   (struct rq *rq);
1694         bool (*yield_to_task)(struct rq *rq, struct task_struct *p, bool preempt);
1695 
1696         void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags);
1697 
1698         /*
1699          * It is the responsibility of the pick_next_task() method that will
1700          * return the next task to call put_prev_task() on the @prev task or
1701          * something equivalent.
1702          *
1703          * May return RETRY_TASK when it finds a higher prio class has runnable
1704          * tasks.
1705          */
1706         struct task_struct * (*pick_next_task)(struct rq *rq,
1707                                                struct task_struct *prev,
1708                                                struct rq_flags *rf);
1709         void (*put_prev_task)(struct rq *rq, struct task_struct *p);
1710 
1711 #ifdef CONFIG_SMP
1712         int  (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags);
1713         void (*migrate_task_rq)(struct task_struct *p, int new_cpu);
1714 
1715         void (*task_woken)(struct rq *this_rq, struct task_struct *task);
1716 
1717         void (*set_cpus_allowed)(struct task_struct *p,
1718                                  const struct cpumask *newmask);
1719 
1720         void (*rq_online)(struct rq *rq);
1721         void (*rq_offline)(struct rq *rq);
1722 #endif
1723 
1724         void (*set_curr_task)(struct rq *rq);
1725         void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
1726         void (*task_fork)(struct task_struct *p);
1727         void (*task_dead)(struct task_struct *p);
1728 
1729         /*
1730          * The switched_from() call is allowed to drop rq->lock, therefore we
1731          * cannot assume the switched_from/switched_to pair is serliazed by
1732          * rq->lock. They are however serialized by p->pi_lock.
1733          */
1734         void (*switched_from)(struct rq *this_rq, struct task_struct *task);
1735         void (*switched_to)  (struct rq *this_rq, struct task_struct *task);
1736         void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1737                               int oldprio);
1738 
1739         unsigned int (*get_rr_interval)(struct rq *rq,
1740                                         struct task_struct *task);
1741 
1742         void (*update_curr)(struct rq *rq);
1743 
1744 #define TASK_SET_GROUP          0
1745 #define TASK_MOVE_GROUP         1
1746 
1747 #ifdef CONFIG_FAIR_GROUP_SCHED
1748         void (*task_change_group)(struct task_struct *p, int type);
1749 #endif
1750 };
1751 
1752 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1753 {
1754         prev->sched_class->put_prev_task(rq, prev);
1755 }
1756 
1757 static inline void set_curr_task(struct rq *rq, struct task_struct *curr)
1758 {
1759         curr->sched_class->set_curr_task(rq);
1760 }
1761 
1762 #ifdef CONFIG_SMP
1763 #define sched_class_highest (&stop_sched_class)
1764 #else
1765 #define sched_class_highest (&dl_sched_class)
1766 #endif
1767 #define for_each_class(class) \
1768    for (class = sched_class_highest; class; class = class->next)
1769 
1770 extern const struct sched_class stop_sched_class;
1771 extern const struct sched_class dl_sched_class;
1772 extern const struct sched_class rt_sched_class;
1773 extern const struct sched_class fair_sched_class;
1774 extern const struct sched_class idle_sched_class;
1775 
1776 
1777 #ifdef CONFIG_SMP
1778 
1779 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1780 
1781 extern void trigger_load_balance(struct rq *rq);
1782 
1783 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask);
1784 
1785 #endif
1786 
1787 #ifdef CONFIG_CPU_IDLE
1788 static inline void idle_set_state(struct rq *rq,
1789                                   struct cpuidle_state *idle_state)
1790 {
1791         rq->idle_state = idle_state;
1792 }
1793 
1794 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1795 {
1796         SCHED_WARN_ON(!rcu_read_lock_held());
1797 
1798         return rq->idle_state;
1799 }
1800 #else
1801 static inline void idle_set_state(struct rq *rq,
1802                                   struct cpuidle_state *idle_state)
1803 {
1804 }
1805 
1806 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1807 {
1808         return NULL;
1809 }
1810 #endif
1811 
1812 extern void schedule_idle(void);
1813 
1814 extern void sysrq_sched_debug_show(void);
1815 extern void sched_init_granularity(void);
1816 extern void update_max_interval(void);
1817 
1818 extern void init_sched_dl_class(void);
1819 extern void init_sched_rt_class(void);
1820 extern void init_sched_fair_class(void);
1821 
1822 extern void reweight_task(struct task_struct *p, int prio);
1823 
1824 extern void resched_curr(struct rq *rq);
1825 extern void resched_cpu(int cpu);
1826 
1827 extern struct rt_bandwidth def_rt_bandwidth;
1828 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1829 
1830 extern struct dl_bandwidth def_dl_bandwidth;
1831 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
1832 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
1833 extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se);
1834 extern void init_dl_rq_bw_ratio(struct dl_rq *dl_rq);
1835 
1836 #define BW_SHIFT                20
1837 #define BW_UNIT                 (1 << BW_SHIFT)
1838 #define RATIO_SHIFT             8
1839 unsigned long to_ratio(u64 period, u64 runtime);
1840 
1841 extern void init_entity_runnable_average(struct sched_entity *se);
1842 extern void post_init_entity_util_avg(struct task_struct *p);
1843 
1844 #ifdef CONFIG_NO_HZ_FULL
1845 extern bool sched_can_stop_tick(struct rq *rq);
1846 extern int __init sched_tick_offload_init(void);
1847 
1848 /*
1849  * Tick may be needed by tasks in the runqueue depending on their policy and
1850  * requirements. If tick is needed, lets send the target an IPI to kick it out of
1851  * nohz mode if necessary.
1852  */
1853 static inline void sched_update_tick_dependency(struct rq *rq)
1854 {
1855         int cpu;
1856 
1857         if (!tick_nohz_full_enabled())
1858                 return;
1859 
1860         cpu = cpu_of(rq);
1861 
1862         if (!tick_nohz_full_cpu(cpu))
1863                 return;
1864 
1865         if (sched_can_stop_tick(rq))
1866                 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
1867         else
1868                 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
1869 }
1870 #else
1871 static inline int sched_tick_offload_init(void) { return 0; }
1872 static inline void sched_update_tick_dependency(struct rq *rq) { }
1873 #endif
1874 
1875 static inline void add_nr_running(struct rq *rq, unsigned count)
1876 {
1877         unsigned prev_nr = rq->nr_running;
1878 
1879         rq->nr_running = prev_nr + count;
1880 
1881 #ifdef CONFIG_SMP
1882         if (prev_nr < 2 && rq->nr_running >= 2) {
1883                 if (!READ_ONCE(rq->rd->overload))
1884                         WRITE_ONCE(rq->rd->overload, 1);
1885         }
1886 #endif
1887 
1888         sched_update_tick_dependency(rq);
1889 }
1890 
1891 static inline void sub_nr_running(struct rq *rq, unsigned count)
1892 {
1893         rq->nr_running -= count;
1894         /* Check if we still need preemption */
1895         sched_update_tick_dependency(rq);
1896 }
1897 
1898 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1899 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1900 
1901 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1902 
1903 extern const_debug unsigned int sysctl_sched_nr_migrate;
1904 extern const_debug unsigned int sysctl_sched_migration_cost;
1905 
1906 #ifdef CONFIG_SCHED_HRTICK
1907 
1908 /*
1909  * Use hrtick when:
1910  *  - enabled by features
1911  *  - hrtimer is actually high res
1912  */
1913 static inline int hrtick_enabled(struct rq *rq)
1914 {
1915         if (!sched_feat(HRTICK))
1916                 return 0;
1917         if (!cpu_active(cpu_of(rq)))
1918                 return 0;
1919         return hrtimer_is_hres_active(&rq->hrtick_timer);
1920 }
1921 
1922 void hrtick_start(struct rq *rq, u64 delay);
1923 
1924 #else
1925 
1926 static inline int hrtick_enabled(struct rq *rq)
1927 {
1928         return 0;
1929 }
1930 
1931 #endif /* CONFIG_SCHED_HRTICK */
1932 
1933 #ifndef arch_scale_freq_capacity
1934 static __always_inline
1935 unsigned long arch_scale_freq_capacity(int cpu)
1936 {
1937         return SCHED_CAPACITY_SCALE;
1938 }
1939 #endif
1940 
1941 #ifdef CONFIG_SMP
1942 #ifdef CONFIG_PREEMPT
1943 
1944 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1945 
1946 /*
1947  * fair double_lock_balance: Safely acquires both rq->locks in a fair
1948  * way at the expense of forcing extra atomic operations in all
1949  * invocations.  This assures that the double_lock is acquired using the
1950  * same underlying policy as the spinlock_t on this architecture, which
1951  * reduces latency compared to the unfair variant below.  However, it
1952  * also adds more overhead and therefore may reduce throughput.
1953  */
1954 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1955         __releases(this_rq->lock)
1956         __acquires(busiest->lock)
1957         __acquires(this_rq->lock)
1958 {
1959         raw_spin_unlock(&this_rq->lock);
1960         double_rq_lock(this_rq, busiest);
1961 
1962         return 1;
1963 }
1964 
1965 #else
1966 /*
1967  * Unfair double_lock_balance: Optimizes throughput at the expense of
1968  * latency by eliminating extra atomic operations when the locks are
1969  * already in proper order on entry.  This favors lower CPU-ids and will
1970  * grant the double lock to lower CPUs over higher ids under contention,
1971  * regardless of entry order into the function.
1972  */
1973 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1974         __releases(this_rq->lock)
1975         __acquires(busiest->lock)
1976         __acquires(this_rq->lock)
1977 {
1978         int ret = 0;
1979 
1980         if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1981                 if (busiest < this_rq) {
1982                         raw_spin_unlock(&this_rq->lock);
1983                         raw_spin_lock(&busiest->lock);
1984                         raw_spin_lock_nested(&this_rq->lock,
1985                                               SINGLE_DEPTH_NESTING);
1986                         ret = 1;
1987                 } else
1988                         raw_spin_lock_nested(&busiest->lock,
1989                                               SINGLE_DEPTH_NESTING);
1990         }
1991         return ret;
1992 }
1993 
1994 #endif /* CONFIG_PREEMPT */
1995 
1996 /*
1997  * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1998  */
1999 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2000 {
2001         if (unlikely(!irqs_disabled())) {
2002                 /* printk() doesn't work well under rq->lock */
2003                 raw_spin_unlock(&this_rq->lock);
2004                 BUG_ON(1);
2005         }
2006 
2007         return _double_lock_balance(this_rq, busiest);
2008 }
2009 
2010 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2011         __releases(busiest->lock)
2012 {
2013         raw_spin_unlock(&busiest->lock);
2014         lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
2015 }
2016 
2017 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
2018 {
2019         if (l1 > l2)
2020                 swap(l1, l2);
2021 
2022         spin_lock(l1);
2023         spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2024 }
2025 
2026 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
2027 {
2028         if (l1 > l2)
2029                 swap(l1, l2);
2030 
2031         spin_lock_irq(l1);
2032         spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2033 }
2034 
2035 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
2036 {
2037         if (l1 > l2)
2038                 swap(l1, l2);
2039 
2040         raw_spin_lock(l1);
2041         raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
2042 }
2043 
2044 /*
2045  * double_rq_lock - safely lock two runqueues
2046  *
2047  * Note this does not disable interrupts like task_rq_lock,
2048  * you need to do so manually before calling.
2049  */
2050 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2051         __acquires(rq1->lock)
2052         __acquires(rq2->lock)
2053 {
2054         BUG_ON(!irqs_disabled());
2055         if (rq1 == rq2) {
2056                 raw_spin_lock(&rq1->lock);
2057                 __acquire(rq2->lock);   /* Fake it out ;) */
2058         } else {
2059                 if (rq1 < rq2) {
2060                         raw_spin_lock(&rq1->lock);
2061                         raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
2062                 } else {
2063                         raw_spin_lock(&rq2->lock);
2064                         raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
2065                 }
2066         }
2067 }
2068 
2069 /*
2070  * double_rq_unlock - safely unlock two runqueues
2071  *
2072  * Note this does not restore interrupts like task_rq_unlock,
2073  * you need to do so manually after calling.
2074  */
2075 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2076         __releases(rq1->lock)
2077         __releases(rq2->lock)
2078 {
2079         raw_spin_unlock(&rq1->lock);
2080         if (rq1 != rq2)
2081                 raw_spin_unlock(&rq2->lock);
2082         else
2083                 __release(rq2->lock);
2084 }
2085 
2086 extern void set_rq_online (struct rq *rq);
2087 extern void set_rq_offline(struct rq *rq);
2088 extern bool sched_smp_initialized;
2089 
2090 #else /* CONFIG_SMP */
2091 
2092 /*
2093  * double_rq_lock - safely lock two runqueues
2094  *
2095  * Note this does not disable interrupts like task_rq_lock,
2096  * you need to do so manually before calling.
2097  */
2098 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
2099         __acquires(rq1->lock)
2100         __acquires(rq2->lock)
2101 {
2102         BUG_ON(!irqs_disabled());
2103         BUG_ON(rq1 != rq2);
2104         raw_spin_lock(&rq1->lock);
2105         __acquire(rq2->lock);   /* Fake it out ;) */
2106 }
2107 
2108 /*
2109  * double_rq_unlock - safely unlock two runqueues
2110  *
2111  * Note this does not restore interrupts like task_rq_unlock,
2112  * you need to do so manually after calling.
2113  */
2114 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2115         __releases(rq1->lock)
2116         __releases(rq2->lock)
2117 {
2118         BUG_ON(rq1 != rq2);
2119         raw_spin_unlock(&rq1->lock);
2120         __release(rq2->lock);
2121 }
2122 
2123 #endif
2124 
2125 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
2126 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
2127 
2128 #ifdef  CONFIG_SCHED_DEBUG
2129 extern bool sched_debug_enabled;
2130 
2131 extern void print_cfs_stats(struct seq_file *m, int cpu);
2132 extern void print_rt_stats(struct seq_file *m, int cpu);
2133 extern void print_dl_stats(struct seq_file *m, int cpu);
2134 extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
2135 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2136 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
2137 #ifdef CONFIG_NUMA_BALANCING
2138 extern void
2139 show_numa_stats(struct task_struct *p, struct seq_file *m);
2140 extern void
2141 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
2142         unsigned long tpf, unsigned long gsf, unsigned long gpf);
2143 #endif /* CONFIG_NUMA_BALANCING */
2144 #endif /* CONFIG_SCHED_DEBUG */
2145 
2146 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
2147 extern void init_rt_rq(struct rt_rq *rt_rq);
2148 extern void init_dl_rq(struct dl_rq *dl_rq);
2149 
2150 extern void cfs_bandwidth_usage_inc(void);
2151 extern void cfs_bandwidth_usage_dec(void);
2152 
2153 #ifdef CONFIG_NO_HZ_COMMON
2154 #define NOHZ_BALANCE_KICK_BIT   0
2155 #define NOHZ_STATS_KICK_BIT     1
2156 
2157 #define NOHZ_BALANCE_KICK       BIT(NOHZ_BALANCE_KICK_BIT)
2158 #define NOHZ_STATS_KICK         BIT(NOHZ_STATS_KICK_BIT)
2159 
2160 #define NOHZ_KICK_MASK  (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK)
2161 
2162 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2163 
2164 extern void nohz_balance_exit_idle(struct rq *rq);
2165 #else
2166 static inline void nohz_balance_exit_idle(struct rq *rq) { }
2167 #endif
2168 
2169 
2170 #ifdef CONFIG_SMP
2171 static inline
2172 void __dl_update(struct dl_bw *dl_b, s64 bw)
2173 {
2174         struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw);
2175         int i;
2176 
2177         RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
2178                          "sched RCU must be held");
2179         for_each_cpu_and(i, rd->span, cpu_active_mask) {
2180                 struct rq *rq = cpu_rq(i);
2181 
2182                 rq->dl.extra_bw += bw;
2183         }
2184 }
2185 #else
2186 static inline
2187 void __dl_update(struct dl_bw *dl_b, s64 bw)
2188 {
2189         struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw);
2190 
2191         dl->extra_bw += bw;
2192 }
2193 #endif
2194 
2195 
2196 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2197 struct irqtime {
2198         u64                     total;
2199         u64                     tick_delta;
2200         u64                     irq_start_time;
2201         struct u64_stats_sync   sync;
2202 };
2203 
2204 DECLARE_PER_CPU(struct irqtime, cpu_irqtime);
2205 
2206 /*
2207  * Returns the irqtime minus the softirq time computed by ksoftirqd.
2208  * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime
2209  * and never move forward.
2210  */
2211 static inline u64 irq_time_read(int cpu)
2212 {
2213         struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu);
2214         unsigned int seq;
2215         u64 total;
2216 
2217         do {
2218                 seq = __u64_stats_fetch_begin(&irqtime->sync);
2219                 total = irqtime->total;
2220         } while (__u64_stats_fetch_retry(&irqtime->sync, seq));
2221 
2222         return total;
2223 }
2224 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2225 
2226 #ifdef CONFIG_CPU_FREQ
2227 DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data);
2228 
2229 /**
2230  * cpufreq_update_util - Take a note about CPU utilization changes.
2231  * @rq: Runqueue to carry out the update for.
2232  * @flags: Update reason flags.
2233  *
2234  * This function is called by the scheduler on the CPU whose utilization is
2235  * being updated.
2236  *
2237  * It can only be called from RCU-sched read-side critical sections.
2238  *
2239  * The way cpufreq is currently arranged requires it to evaluate the CPU
2240  * performance state (frequency/voltage) on a regular basis to prevent it from
2241  * being stuck in a completely inadequate performance level for too long.
2242  * That is not guaranteed to happen if the updates are only triggered from CFS
2243  * and DL, though, because they may not be coming in if only RT tasks are
2244  * active all the time (or there are RT tasks only).
2245  *
2246  * As a workaround for that issue, this function is called periodically by the
2247  * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2248  * but that really is a band-aid.  Going forward it should be replaced with
2249  * solutions targeted more specifically at RT tasks.
2250  */
2251 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2252 {
2253         struct update_util_data *data;
2254 
2255         data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data,
2256                                                   cpu_of(rq)));
2257         if (data)
2258                 data->func(data, rq_clock(rq), flags);
2259 }
2260 #else
2261 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2262 #endif /* CONFIG_CPU_FREQ */
2263 
2264 #ifdef CONFIG_UCLAMP_TASK
2265 unsigned int uclamp_eff_value(struct task_struct *p, unsigned int clamp_id);
2266 
2267 static __always_inline
2268 unsigned int uclamp_util_with(struct rq *rq, unsigned int util,
2269                               struct task_struct *p)
2270 {
2271         unsigned int min_util = READ_ONCE(rq->uclamp[UCLAMP_MIN].value);
2272         unsigned int max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value);
2273 
2274         if (p) {
2275                 min_util = max(min_util, uclamp_eff_value(p, UCLAMP_MIN));
2276                 max_util = max(max_util, uclamp_eff_value(p, UCLAMP_MAX));
2277         }
2278 
2279         /*
2280          * Since CPU's {min,max}_util clamps are MAX aggregated considering
2281          * RUNNABLE tasks with _different_ clamps, we can end up with an
2282          * inversion. Fix it now when the clamps are applied.
2283          */
2284         if (unlikely(min_util >= max_util))
2285                 return min_util;
2286 
2287         return clamp(util, min_util, max_util);
2288 }
2289 
2290 static inline unsigned int uclamp_util(struct rq *rq, unsigned int util)
2291 {
2292         return uclamp_util_with(rq, util, NULL);
2293 }
2294 #else /* CONFIG_UCLAMP_TASK */
2295 static inline unsigned int uclamp_util_with(struct rq *rq, unsigned int util,
2296                                             struct task_struct *p)
2297 {
2298         return util;
2299 }
2300 static inline unsigned int uclamp_util(struct rq *rq, unsigned int util)
2301 {
2302         return util;
2303 }
2304 #endif /* CONFIG_UCLAMP_TASK */
2305 
2306 #ifdef arch_scale_freq_capacity
2307 # ifndef arch_scale_freq_invariant
2308 #  define arch_scale_freq_invariant()   true
2309 # endif
2310 #else
2311 # define arch_scale_freq_invariant()    false
2312 #endif
2313 
2314 #ifdef CONFIG_SMP
2315 static inline unsigned long capacity_orig_of(int cpu)
2316 {
2317         return cpu_rq(cpu)->cpu_capacity_orig;
2318 }
2319 #endif
2320 
2321 /**
2322  * enum schedutil_type - CPU utilization type
2323  * @FREQUENCY_UTIL:     Utilization used to select frequency
2324  * @ENERGY_UTIL:        Utilization used during energy calculation
2325  *
2326  * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
2327  * need to be aggregated differently depending on the usage made of them. This
2328  * enum is used within schedutil_freq_util() to differentiate the types of
2329  * utilization expected by the callers, and adjust the aggregation accordingly.
2330  */
2331 enum schedutil_type {
2332         FREQUENCY_UTIL,
2333         ENERGY_UTIL,
2334 };
2335 
2336 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
2337 
2338 unsigned long schedutil_cpu_util(int cpu, unsigned long util_cfs,
2339                                  unsigned long max, enum schedutil_type type,
2340                                  struct task_struct *p);
2341 
2342 static inline unsigned long cpu_bw_dl(struct rq *rq)
2343 {
2344         return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT;
2345 }
2346 
2347 static inline unsigned long cpu_util_dl(struct rq *rq)
2348 {
2349         return READ_ONCE(rq->avg_dl.util_avg);
2350 }
2351 
2352 static inline unsigned long cpu_util_cfs(struct rq *rq)
2353 {
2354         unsigned long util = READ_ONCE(rq->cfs.avg.util_avg);
2355 
2356         if (sched_feat(UTIL_EST)) {
2357                 util = max_t(unsigned long, util,
2358                              READ_ONCE(rq->cfs.avg.util_est.enqueued));
2359         }
2360 
2361         return util;
2362 }
2363 
2364 static inline unsigned long cpu_util_rt(struct rq *rq)
2365 {
2366         return READ_ONCE(rq->avg_rt.util_avg);
2367 }
2368 #else /* CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
2369 static inline unsigned long schedutil_cpu_util(int cpu, unsigned long util_cfs,
2370                                  unsigned long max, enum schedutil_type type,
2371                                  struct task_struct *p)
2372 {
2373         return 0;
2374 }
2375 #endif /* CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
2376 
2377 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
2378 static inline unsigned long cpu_util_irq(struct rq *rq)
2379 {
2380         return rq->avg_irq.util_avg;
2381 }
2382 
2383 static inline
2384 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2385 {
2386         util *= (max - irq);
2387         util /= max;
2388 
2389         return util;
2390 
2391 }
2392 #else
2393 static inline unsigned long cpu_util_irq(struct rq *rq)
2394 {
2395         return 0;
2396 }
2397 
2398 static inline
2399 unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max)
2400 {
2401         return util;
2402 }
2403 #endif
2404 
2405 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
2406 
2407 #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
2408 
2409 DECLARE_STATIC_KEY_FALSE(sched_energy_present);
2410 
2411 static inline bool sched_energy_enabled(void)
2412 {
2413         return static_branch_unlikely(&sched_energy_present);
2414 }
2415 
2416 #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
2417 
2418 #define perf_domain_span(pd) NULL
2419 static inline bool sched_energy_enabled(void) { return false; }
2420 
2421 #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
2422 

~ [ 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