Kernel

Kernel Git Source Tree

Root/kernel/sched_fair.c

1	/*
2	* Completely Fair Scheduling (CFS) Class (SCHED_NORMAL/SCHED_BATCH)
3	*
4	* Copyright (C) 2007 Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
5	*
6	* Interactivity improvements by Mike Galbraith
7	* (C) 2007 Mike Galbraith <efault@gmx.de>
8	*
9	* Various enhancements by Dmitry Adamushko.
10	* (C) 2007 Dmitry Adamushko <dmitry.adamushko@gmail.com>
11	*
12	* Group scheduling enhancements by Srivatsa Vaddagiri
13	* Copyright IBM Corporation, 2007
14	* Author: Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com>
15	*
16	* Scaled math optimizations by Thomas Gleixner
17	* Copyright (C) 2007, Thomas Gleixner <tglx@linutronix.de>
18	*
19	* Adaptive scheduling granularity, math enhancements by Peter Zijlstra
20	* Copyright (C) 2007 Red Hat, Inc., Peter Zijlstra <pzijlstr@redhat.com>
21	*/
22
23	#include <linux/latencytop.h>
24	#include <linux/sched.h>
25
26	/*
27	* Targeted preemption latency for CPU-bound tasks:
28	* (default: 6ms * (1 + ilog(ncpus)), units: nanoseconds)
29	*
30	* NOTE: this latency value is not the same as the concept of
31	* 'timeslice length' - timeslices in CFS are of variable length
32	* and have no persistent notion like in traditional, time-slice
33	* based scheduling concepts.
34	*
35	* (to see the precise effective timeslice length of your workload,
36	* run vmstat and monitor the context-switches (cs) field)
37	*/
38	unsigned int sysctl_sched_latency = 6000000ULL;
39	unsigned int normalized_sysctl_sched_latency = 6000000ULL;
40
41	/*
42	* The initial- and re-scaling of tunables is configurable
43	* (default SCHED_TUNABLESCALING_LOG = *(1+ilog(ncpus))
44	*
45	* Options are:
46	* SCHED_TUNABLESCALING_NONE - unscaled, always *1
47	* SCHED_TUNABLESCALING_LOG - scaled logarithmical, *1+ilog(ncpus)
48	* SCHED_TUNABLESCALING_LINEAR - scaled linear, *ncpus
49	*/
50	enum sched_tunable_scaling sysctl_sched_tunable_scaling
51	= SCHED_TUNABLESCALING_LOG;
52
53	/*
54	* Minimal preemption granularity for CPU-bound tasks:
55	* (default: 0.75 msec * (1 + ilog(ncpus)), units: nanoseconds)
56	*/
57	unsigned int sysctl_sched_min_granularity = 750000ULL;
58	unsigned int normalized_sysctl_sched_min_granularity = 750000ULL;
59
60	/*
61	* is kept at sysctl_sched_latency / sysctl_sched_min_granularity
62	*/
63	static unsigned int sched_nr_latency = 8;
64
65	/*
66	* After fork, child runs first. If set to 0 (default) then
67	* parent will (try to) run first.
68	*/
69	unsigned int sysctl_sched_child_runs_first __read_mostly;
70
71	/*
72	* sys_sched_yield() compat mode
73	*
74	* This option switches the agressive yield implementation of the
75	* old scheduler back on.
76	*/
77	unsigned int __read_mostly sysctl_sched_compat_yield;
78
79	/*
80	* SCHED_OTHER wake-up granularity.
81	* (default: 1 msec * (1 + ilog(ncpus)), units: nanoseconds)
82	*
83	* This option delays the preemption effects of decoupled workloads
84	* and reduces their over-scheduling. Synchronous workloads will still
85	* have immediate wakeup/sleep latencies.
86	*/
87	unsigned int sysctl_sched_wakeup_granularity = 1000000UL;
88	unsigned int normalized_sysctl_sched_wakeup_granularity = 1000000UL;
89
90	const_debug unsigned int sysctl_sched_migration_cost = 500000UL;
91
92	/*
93	* The exponential sliding window over which load is averaged for shares
94	* distribution.
95	* (default: 10msec)
96	*/
97	unsigned int __read_mostly sysctl_sched_shares_window = 10000000UL;
98
99	static const struct sched_class fair_sched_class;
100
101	/**************************************************************
102	* CFS operations on generic schedulable entities:
103	*/
104
105	#ifdef CONFIG_FAIR_GROUP_SCHED
106
107	/* cpu runqueue to which this cfs_rq is attached */
108	static inline struct rq rq_of(struct cfs_rq cfs_rq)
109	{
110	return cfs_rq->rq;
111	}
112
113	/* An entity is a task if it doesn't "own" a runqueue */
114	#define entity_is_task(se) (!se->my_q)
115
116	static inline struct task_struct task_of(struct sched_entity se)
117	{
118	#ifdef CONFIG_SCHED_DEBUG
119	WARN_ON_ONCE(!entity_is_task(se));
120	#endif
121	return container_of(se, struct task_struct, se);
122	}
123
124	/* Walk up scheduling entities hierarchy */
125	#define for_each_sched_entity(se) \
126	for (; se; se = se->parent)
127
128	static inline struct cfs_rq task_cfs_rq(struct task_struct p)
129	{
130	return p->se.cfs_rq;
131	}
132
133	/* runqueue on which this entity is (to be) queued */
134	static inline struct cfs_rq cfs_rq_of(struct sched_entity se)
135	{
136	return se->cfs_rq;
137	}
138
139	/* runqueue "owned" by this group */
140	static inline struct cfs_rq group_cfs_rq(struct sched_entity grp)
141	{
142	return grp->my_q;
143	}
144
145	/* Given a group's cfs_rq on one cpu, return its corresponding cfs_rq on
146	* another cpu ('this_cpu')
147	*/
148	static inline struct cfs_rq cpu_cfs_rq(struct cfs_rq cfs_rq, int this_cpu)
149	{
150	return cfs_rq->tg->cfs_rq[this_cpu];
151	}
152
153	static inline void list_add_leaf_cfs_rq(struct cfs_rq *cfs_rq)
154	{
155	if (!cfs_rq->on_list) {
156	/*
157	* Ensure we either appear before our parent (if already
158	* enqueued) or force our parent to appear after us when it is
159	* enqueued. The fact that we always enqueue bottom-up
160	* reduces this to two cases.
161	*/
162	if (cfs_rq->tg->parent &&
163	cfs_rq->tg->parent->cfs_rq[cpu_of(rq_of(cfs_rq))]->on_list) {
164	list_add_rcu(&cfs_rq->leaf_cfs_rq_list,
165	&rq_of(cfs_rq)->leaf_cfs_rq_list);
166	} else {
167	list_add_tail_rcu(&cfs_rq->leaf_cfs_rq_list,
168	&rq_of(cfs_rq)->leaf_cfs_rq_list);
169	}
170
171	cfs_rq->on_list = 1;
172	}
173	}
174
175	static inline void list_del_leaf_cfs_rq(struct cfs_rq *cfs_rq)
176	{
177	if (cfs_rq->on_list) {
178	list_del_rcu(&cfs_rq->leaf_cfs_rq_list);
179	cfs_rq->on_list = 0;
180	}
181	}
182
183	/* Iterate thr' all leaf cfs_rq's on a runqueue */
184	#define for_each_leaf_cfs_rq(rq, cfs_rq) \
185	list_for_each_entry_rcu(cfs_rq, &rq->leaf_cfs_rq_list, leaf_cfs_rq_list)
186
187	/* Do the two (enqueued) entities belong to the same group ? */
188	static inline int
189	is_same_group(struct sched_entity se, struct sched_entity pse)
190	{
191	if (se->cfs_rq == pse->cfs_rq)
192	return 1;
193
194	return 0;
195	}
196
197	static inline struct sched_entity parent_entity(struct sched_entity se)
198	{
199	return se->parent;
200	}
201
202	/* return depth at which a sched entity is present in the hierarchy */
203	static inline int depth_se(struct sched_entity *se)
204	{
205	int depth = 0;
206
207	for_each_sched_entity(se)
208	depth++;
209
210	return depth;
211	}
212
213	static void
214	find_matching_se(struct sched_entity se, struct sched_entity pse)
215	{
216	int se_depth, pse_depth;
217
218	/*
219	* preemption test can be made between sibling entities who are in the
220	* same cfs_rq i.e who have a common parent. Walk up the hierarchy of
221	* both tasks until we find their ancestors who are siblings of common
222	* parent.
223	*/
224
225	/* First walk up until both entities are at same depth */
226	se_depth = depth_se(*se);
227	pse_depth = depth_se(*pse);
228
229	while (se_depth > pse_depth) {
230	se_depth--;
231	se = parent_entity(se);
232	}
233
234	while (pse_depth > se_depth) {
235	pse_depth--;
236	pse = parent_entity(pse);
237	}
238
239	while (!is_same_group(se, pse)) {
240	se = parent_entity(se);
241	pse = parent_entity(pse);
242	}
243	}
244
245	#else /* !CONFIG_FAIR_GROUP_SCHED */
246
247	static inline struct task_struct task_of(struct sched_entity se)
248	{
249	return container_of(se, struct task_struct, se);
250	}
251
252	static inline struct rq rq_of(struct cfs_rq cfs_rq)
253	{
254	return container_of(cfs_rq, struct rq, cfs);
255	}
256
257	#define entity_is_task(se) 1
258
259	#define for_each_sched_entity(se) \
260	for (; se; se = NULL)
261
262	static inline struct cfs_rq task_cfs_rq(struct task_struct p)
263	{
264	return &task_rq(p)->cfs;
265	}
266
267	static inline struct cfs_rq cfs_rq_of(struct sched_entity se)
268	{
269	struct task_struct *p = task_of(se);
270	struct rq *rq = task_rq(p);
271
272	return &rq->cfs;
273	}
274
275	/* runqueue "owned" by this group */
276	static inline struct cfs_rq group_cfs_rq(struct sched_entity grp)
277	{
278	return NULL;
279	}
280
281	static inline struct cfs_rq cpu_cfs_rq(struct cfs_rq cfs_rq, int this_cpu)
282	{
283	return &cpu_rq(this_cpu)->cfs;
284	}
285
286	static inline void list_add_leaf_cfs_rq(struct cfs_rq *cfs_rq)
287	{
288	}
289
290	static inline void list_del_leaf_cfs_rq(struct cfs_rq *cfs_rq)
291	{
292	}
293
294	#define for_each_leaf_cfs_rq(rq, cfs_rq) \
295	for (cfs_rq = &rq->cfs; cfs_rq; cfs_rq = NULL)
296
297	static inline int
298	is_same_group(struct sched_entity se, struct sched_entity pse)
299	{
300	return 1;
301	}
302
303	static inline struct sched_entity parent_entity(struct sched_entity se)
304	{
305	return NULL;
306	}
307
308	static inline void
309	find_matching_se(struct sched_entity se, struct sched_entity pse)
310	{
311	}
312
313	#endif /* CONFIG_FAIR_GROUP_SCHED */
314
315
316	/**************************************************************
317	* Scheduling class tree data structure manipulation methods:
318	*/
319
320	static inline u64 max_vruntime(u64 min_vruntime, u64 vruntime)
321	{
322	s64 delta = (s64)(vruntime - min_vruntime);
323	if (delta > 0)
324	min_vruntime = vruntime;
325
326	return min_vruntime;
327	}
328
329	static inline u64 min_vruntime(u64 min_vruntime, u64 vruntime)
330	{
331	s64 delta = (s64)(vruntime - min_vruntime);
332	if (delta < 0)
333	min_vruntime = vruntime;
334
335	return min_vruntime;
336	}
337
338	static inline int entity_before(struct sched_entity *a,
339	struct sched_entity *b)
340	{
341	return (s64)(a->vruntime - b->vruntime) < 0;
342	}
343
344	static inline s64 entity_key(struct cfs_rq cfs_rq, struct sched_entity se)
345	{
346	return se->vruntime - cfs_rq->min_vruntime;
347	}
348
349	static void update_min_vruntime(struct cfs_rq *cfs_rq)
350	{
351	u64 vruntime = cfs_rq->min_vruntime;
352
353	if (cfs_rq->curr)
354	vruntime = cfs_rq->curr->vruntime;
355
356	if (cfs_rq->rb_leftmost) {
357	struct sched_entity *se = rb_entry(cfs_rq->rb_leftmost,
358	struct sched_entity,
359	run_node);
360
361	if (!cfs_rq->curr)
362	vruntime = se->vruntime;
363	else
364	vruntime = min_vruntime(vruntime, se->vruntime);
365	}
366
367	cfs_rq->min_vruntime = max_vruntime(cfs_rq->min_vruntime, vruntime);
368	}
369
370	/*
371	* Enqueue an entity into the rb-tree:
372	*/
373	static void __enqueue_entity(struct cfs_rq cfs_rq, struct sched_entity se)
374	{
375	struct rb_node **link = &cfs_rq->tasks_timeline.rb_node;
376	struct rb_node *parent = NULL;
377	struct sched_entity *entry;
378	s64 key = entity_key(cfs_rq, se);
379	int leftmost = 1;
380
381	/*
382	* Find the right place in the rbtree:
383	*/
384	while (*link) {
385	parent = *link;
386	entry = rb_entry(parent, struct sched_entity, run_node);
387	/*
388	* We dont care about collisions. Nodes with
389	* the same key stay together.
390	*/
391	if (key < entity_key(cfs_rq, entry)) {
392	link = &parent->rb_left;
393	} else {
394	link = &parent->rb_right;
395	leftmost = 0;
396	}
397	}
398
399	/*
400	* Maintain a cache of leftmost tree entries (it is frequently
401	* used):
402	*/
403	if (leftmost)
404	cfs_rq->rb_leftmost = &se->run_node;
405
406	rb_link_node(&se->run_node, parent, link);
407	rb_insert_color(&se->run_node, &cfs_rq->tasks_timeline);
408	}
409
410	static void __dequeue_entity(struct cfs_rq cfs_rq, struct sched_entity se)
411	{
412	if (cfs_rq->rb_leftmost == &se->run_node) {
413	struct rb_node *next_node;
414
415	next_node = rb_next(&se->run_node);
416	cfs_rq->rb_leftmost = next_node;
417	}
418
419	rb_erase(&se->run_node, &cfs_rq->tasks_timeline);
420	}
421
422	static struct sched_entity __pick_next_entity(struct cfs_rq cfs_rq)
423	{
424	struct rb_node *left = cfs_rq->rb_leftmost;
425
426	if (!left)
427	return NULL;
428
429	return rb_entry(left, struct sched_entity, run_node);
430	}
431
432	static struct sched_entity __pick_last_entity(struct cfs_rq cfs_rq)
433	{
434	struct rb_node *last = rb_last(&cfs_rq->tasks_timeline);
435
436	if (!last)
437	return NULL;
438
439	return rb_entry(last, struct sched_entity, run_node);
440	}
441
442	/**************************************************************
443	* Scheduling class statistics methods:
444	*/
445
446	#ifdef CONFIG_SCHED_DEBUG
447	int sched_proc_update_handler(struct ctl_table *table, int write,
448	void __user buffer, size_t lenp,
449	loff_t *ppos)
450	{
451	int ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
452	int factor = get_update_sysctl_factor();
453
454	if (ret \|\| !write)
455	return ret;
456
457	sched_nr_latency = DIV_ROUND_UP(sysctl_sched_latency,
458	sysctl_sched_min_granularity);
459
460	#define WRT_SYSCTL(name) \
461	(normalized_sysctl_##name = sysctl_##name / (factor))
462	WRT_SYSCTL(sched_min_granularity);
463	WRT_SYSCTL(sched_latency);
464	WRT_SYSCTL(sched_wakeup_granularity);
465	#undef WRT_SYSCTL
466
467	return 0;
468	}
469	#endif
470
471	/*
472	* delta /= w
473	*/
474	static inline unsigned long
475	calc_delta_fair(unsigned long delta, struct sched_entity *se)
476	{
477	if (unlikely(se->load.weight != NICE_0_LOAD))
478	delta = calc_delta_mine(delta, NICE_0_LOAD, &se->load);
479
480	return delta;
481	}
482
483	/*
484	* The idea is to set a period in which each task runs once.
485	*
486	* When there are too many tasks (sysctl_sched_nr_latency) we have to stretch
487	* this period because otherwise the slices get too small.
488	*
489	* p = (nr <= nl) ? l : l*nr/nl
490	*/
491	static u64 __sched_period(unsigned long nr_running)
492	{
493	u64 period = sysctl_sched_latency;
494	unsigned long nr_latency = sched_nr_latency;
495
496	if (unlikely(nr_running > nr_latency)) {
497	period = sysctl_sched_min_granularity;
498	period *= nr_running;
499	}
500
501	return period;
502	}
503
504	/*
505	* We calculate the wall-time slice from the period by taking a part
506	* proportional to the weight.
507	*
508	* s = p*P[w/rw]
509	*/
510	static u64 sched_slice(struct cfs_rq cfs_rq, struct sched_entity se)
511	{
512	u64 slice = __sched_period(cfs_rq->nr_running + !se->on_rq);
513
514	for_each_sched_entity(se) {
515	struct load_weight *load;
516	struct load_weight lw;
517
518	cfs_rq = cfs_rq_of(se);
519	load = &cfs_rq->load;
520
521	if (unlikely(!se->on_rq)) {
522	lw = cfs_rq->load;
523
524	update_load_add(&lw, se->load.weight);
525	load = &lw;
526	}
527	slice = calc_delta_mine(slice, se->load.weight, load);
528	}
529	return slice;
530	}
531
532	/*
533	* We calculate the vruntime slice of a to be inserted task
534	*
535	* vs = s/w
536	*/
537	static u64 sched_vslice(struct cfs_rq cfs_rq, struct sched_entity se)
538	{
539	return calc_delta_fair(sched_slice(cfs_rq, se), se);
540	}
541
542	static void update_cfs_load(struct cfs_rq *cfs_rq, int global_update);
543	static void update_cfs_shares(struct cfs_rq *cfs_rq, long weight_delta);
544
545	/*
546	* Update the current task's runtime statistics. Skip current tasks that
547	* are not in our scheduling class.
548	*/
549	static inline void
550	__update_curr(struct cfs_rq cfs_rq, struct sched_entity curr,
551	unsigned long delta_exec)
552	{
553	unsigned long delta_exec_weighted;
554
555	schedstat_set(curr->statistics.exec_max,
556	max((u64)delta_exec, curr->statistics.exec_max));
557
558	curr->sum_exec_runtime += delta_exec;
559	schedstat_add(cfs_rq, exec_clock, delta_exec);
560	delta_exec_weighted = calc_delta_fair(delta_exec, curr);
561
562	curr->vruntime += delta_exec_weighted;
563	update_min_vruntime(cfs_rq);
564
565	#if defined CONFIG_SMP && defined CONFIG_FAIR_GROUP_SCHED
566	cfs_rq->load_unacc_exec_time += delta_exec;
567	#endif
568	}
569
570	static void update_curr(struct cfs_rq *cfs_rq)
571	{
572	struct sched_entity *curr = cfs_rq->curr;
573	u64 now = rq_of(cfs_rq)->clock_task;
574	unsigned long delta_exec;
575
576	if (unlikely(!curr))
577	return;
578
579	/*
580	* Get the amount of time the current task was running
581	* since the last time we changed load (this cannot
582	* overflow on 32 bits):
583	*/
584	delta_exec = (unsigned long)(now - curr->exec_start);
585	if (!delta_exec)
586	return;
587
588	__update_curr(cfs_rq, curr, delta_exec);
589	curr->exec_start = now;
590
591	if (entity_is_task(curr)) {
592	struct task_struct *curtask = task_of(curr);
593
594	trace_sched_stat_runtime(curtask, delta_exec, curr->vruntime);
595	cpuacct_charge(curtask, delta_exec);
596	account_group_exec_runtime(curtask, delta_exec);
597	}
598	}
599
600	static inline void
601	update_stats_wait_start(struct cfs_rq cfs_rq, struct sched_entity se)
602	{
603	schedstat_set(se->statistics.wait_start, rq_of(cfs_rq)->clock);
604	}
605
606	/*
607	* Task is being enqueued - update stats:
608	*/
609	static void update_stats_enqueue(struct cfs_rq cfs_rq, struct sched_entity se)
610	{
611	/*
612	* Are we enqueueing a waiting task? (for current tasks
613	* a dequeue/enqueue event is a NOP)
614	*/
615	if (se != cfs_rq->curr)
616	update_stats_wait_start(cfs_rq, se);
617	}
618
619	static void
620	update_stats_wait_end(struct cfs_rq cfs_rq, struct sched_entity se)
621	{
622	schedstat_set(se->statistics.wait_max, max(se->statistics.wait_max,
623	rq_of(cfs_rq)->clock - se->statistics.wait_start));
624	schedstat_set(se->statistics.wait_count, se->statistics.wait_count + 1);
625	schedstat_set(se->statistics.wait_sum, se->statistics.wait_sum +
626	rq_of(cfs_rq)->clock - se->statistics.wait_start);
627	#ifdef CONFIG_SCHEDSTATS
628	if (entity_is_task(se)) {
629	trace_sched_stat_wait(task_of(se),
630	rq_of(cfs_rq)->clock - se->statistics.wait_start);
631	}
632	#endif
633	schedstat_set(se->statistics.wait_start, 0);
634	}
635
636	static inline void
637	update_stats_dequeue(struct cfs_rq cfs_rq, struct sched_entity se)
638	{
639	/*
640	* Mark the end of the wait period if dequeueing a
641	* waiting task:
642	*/
643	if (se != cfs_rq->curr)
644	update_stats_wait_end(cfs_rq, se);
645	}
646
647	/*
648	* We are picking a new current task - update its stats:
649	*/
650	static inline void
651	update_stats_curr_start(struct cfs_rq cfs_rq, struct sched_entity se)
652	{
653	/*
654	* We are starting a new run period:
655	*/
656	se->exec_start = rq_of(cfs_rq)->clock_task;
657	}
658
659	/**************************************************
660	* Scheduling class queueing methods:
661	*/
662
663	#if defined CONFIG_SMP && defined CONFIG_FAIR_GROUP_SCHED
664	static void
665	add_cfs_task_weight(struct cfs_rq *cfs_rq, unsigned long weight)
666	{
667	cfs_rq->task_weight += weight;
668	}
669	#else
670	static inline void
671	add_cfs_task_weight(struct cfs_rq *cfs_rq, unsigned long weight)
672	{
673	}
674	#endif
675
676	static void
677	account_entity_enqueue(struct cfs_rq cfs_rq, struct sched_entity se)
678	{
679	update_load_add(&cfs_rq->load, se->load.weight);
680	if (!parent_entity(se))
681	inc_cpu_load(rq_of(cfs_rq), se->load.weight);
682	if (entity_is_task(se)) {
683	add_cfs_task_weight(cfs_rq, se->load.weight);
684	list_add(&se->group_node, &cfs_rq->tasks);
685	}
686	cfs_rq->nr_running++;
687	}
688
689	static void
690	account_entity_dequeue(struct cfs_rq cfs_rq, struct sched_entity se)
691	{
692	update_load_sub(&cfs_rq->load, se->load.weight);
693	if (!parent_entity(se))
694	dec_cpu_load(rq_of(cfs_rq), se->load.weight);
695	if (entity_is_task(se)) {
696	add_cfs_task_weight(cfs_rq, -se->load.weight);
697	list_del_init(&se->group_node);
698	}
699	cfs_rq->nr_running--;
700	}
701
702	#ifdef CONFIG_FAIR_GROUP_SCHED
703	# ifdef CONFIG_SMP
704	static void update_cfs_rq_load_contribution(struct cfs_rq *cfs_rq,
705	int global_update)
706	{
707	struct task_group *tg = cfs_rq->tg;
708	long load_avg;
709
710	load_avg = div64_u64(cfs_rq->load_avg, cfs_rq->load_period+1);
711	load_avg -= cfs_rq->load_contribution;
712
713	if (global_update \|\| abs(load_avg) > cfs_rq->load_contribution / 8) {
714	atomic_add(load_avg, &tg->load_weight);
715	cfs_rq->load_contribution += load_avg;
716	}
717	}
718
719	static void update_cfs_load(struct cfs_rq *cfs_rq, int global_update)
720	{
721	u64 period = sysctl_sched_shares_window;
722	u64 now, delta;
723	unsigned long load = cfs_rq->load.weight;
724
725	if (cfs_rq->tg == &root_task_group)
726	return;
727
728	now = rq_of(cfs_rq)->clock_task;
729	delta = now - cfs_rq->load_stamp;
730
731	/* truncate load history at 4 idle periods */
732	if (cfs_rq->load_stamp > cfs_rq->load_last &&
733	now - cfs_rq->load_last > 4 * period) {
734	cfs_rq->load_period = 0;
735	cfs_rq->load_avg = 0;
736	}
737
738	cfs_rq->load_stamp = now;
739	cfs_rq->load_unacc_exec_time = 0;
740	cfs_rq->load_period += delta;
741	if (load) {
742	cfs_rq->load_last = now;
743	cfs_rq->load_avg += delta * load;
744	}
745
746	/* consider updating load contribution on each fold or truncate */
747	if (global_update \|\| cfs_rq->load_period > period
748	\|\| !cfs_rq->load_period)
749	update_cfs_rq_load_contribution(cfs_rq, global_update);
750
751	while (cfs_rq->load_period > period) {
752	/*
753	* Inline assembly required to prevent the compiler
754	* optimising this loop into a divmod call.
755	* See __iter_div_u64_rem() for another example of this.
756	*/
757	asm("" : "+rm" (cfs_rq->load_period));
758	cfs_rq->load_period /= 2;
759	cfs_rq->load_avg /= 2;
760	}
761
762	if (!cfs_rq->curr && !cfs_rq->nr_running && !cfs_rq->load_avg)
763	list_del_leaf_cfs_rq(cfs_rq);
764	}
765
766	static long calc_cfs_shares(struct cfs_rq cfs_rq, struct task_group tg,
767	long weight_delta)
768	{
769	long load_weight, load, shares;
770
771	load = cfs_rq->load.weight + weight_delta;
772
773	load_weight = atomic_read(&tg->load_weight);
774	load_weight -= cfs_rq->load_contribution;
775	load_weight += load;
776
777	shares = (tg->shares * load);
778	if (load_weight)
779	shares /= load_weight;
780
781	if (shares < MIN_SHARES)
782	shares = MIN_SHARES;
783	if (shares > tg->shares)
784	shares = tg->shares;
785
786	return shares;
787	}
788
789	static void update_entity_shares_tick(struct cfs_rq *cfs_rq)
790	{
791	if (cfs_rq->load_unacc_exec_time > sysctl_sched_shares_window) {
792	update_cfs_load(cfs_rq, 0);
793	update_cfs_shares(cfs_rq, 0);
794	}
795	}
796	# else /* CONFIG_SMP */
797	static void update_cfs_load(struct cfs_rq *cfs_rq, int global_update)
798	{
799	}
800
801	static inline long calc_cfs_shares(struct cfs_rq cfs_rq, struct task_group tg,
802	long weight_delta)
803	{
804	return tg->shares;
805	}
806
807	static inline void update_entity_shares_tick(struct cfs_rq *cfs_rq)
808	{
809	}
810	# endif /* CONFIG_SMP */
811	static void reweight_entity(struct cfs_rq cfs_rq, struct sched_entity se,
812	unsigned long weight)
813	{
814	if (se->on_rq) {
815	/* commit outstanding execution time */
816	if (cfs_rq->curr == se)
817	update_curr(cfs_rq);
818	account_entity_dequeue(cfs_rq, se);
819	}
820
821	update_load_set(&se->load, weight);
822
823	if (se->on_rq)
824	account_entity_enqueue(cfs_rq, se);
825	}
826
827	static void update_cfs_shares(struct cfs_rq *cfs_rq, long weight_delta)
828	{
829	struct task_group *tg;
830	struct sched_entity *se;
831	long shares;
832
833	tg = cfs_rq->tg;
834	se = tg->se[cpu_of(rq_of(cfs_rq))];
835	if (!se)
836	return;
837	#ifndef CONFIG_SMP
838	if (likely(se->load.weight == tg->shares))
839	return;
840	#endif
841	shares = calc_cfs_shares(cfs_rq, tg, weight_delta);
842
843	reweight_entity(cfs_rq_of(se), se, shares);
844	}
845	#else /* CONFIG_FAIR_GROUP_SCHED */
846	static void update_cfs_load(struct cfs_rq *cfs_rq, int global_update)
847	{
848	}
849
850	static inline void update_cfs_shares(struct cfs_rq *cfs_rq, long weight_delta)
851	{
852	}
853
854	static inline void update_entity_shares_tick(struct cfs_rq *cfs_rq)
855	{
856	}
857	#endif /* CONFIG_FAIR_GROUP_SCHED */
858
859	static void enqueue_sleeper(struct cfs_rq cfs_rq, struct sched_entity se)
860	{
861	#ifdef CONFIG_SCHEDSTATS
862	struct task_struct *tsk = NULL;
863
864	if (entity_is_task(se))
865	tsk = task_of(se);
866
867	if (se->statistics.sleep_start) {
868	u64 delta = rq_of(cfs_rq)->clock - se->statistics.sleep_start;
869
870	if ((s64)delta < 0)
871	delta = 0;
872
873	if (unlikely(delta > se->statistics.sleep_max))
874	se->statistics.sleep_max = delta;
875
876	se->statistics.sleep_start = 0;
877	se->statistics.sum_sleep_runtime += delta;
878
879	if (tsk) {
880	account_scheduler_latency(tsk, delta >> 10, 1);
881	trace_sched_stat_sleep(tsk, delta);
882	}
883	}
884	if (se->statistics.block_start) {
885	u64 delta = rq_of(cfs_rq)->clock - se->statistics.block_start;
886
887	if ((s64)delta < 0)
888	delta = 0;
889
890	if (unlikely(delta > se->statistics.block_max))
891	se->statistics.block_max = delta;
892
893	se->statistics.block_start = 0;
894	se->statistics.sum_sleep_runtime += delta;
895
896	if (tsk) {
897	if (tsk->in_iowait) {
898	se->statistics.iowait_sum += delta;
899	se->statistics.iowait_count++;
900	trace_sched_stat_iowait(tsk, delta);
901	}
902
903	/*
904	* Blocking time is in units of nanosecs, so shift by
905	* 20 to get a milliseconds-range estimation of the
906	* amount of time that the task spent sleeping:
907	*/
908	if (unlikely(prof_on == SLEEP_PROFILING)) {
909	profile_hits(SLEEP_PROFILING,
910	(void *)get_wchan(tsk),
911	delta >> 20);
912	}
913	account_scheduler_latency(tsk, delta >> 10, 0);
914	}
915	}
916	#endif
917	}
918
919	static void check_spread(struct cfs_rq cfs_rq, struct sched_entity se)
920	{
921	#ifdef CONFIG_SCHED_DEBUG
922	s64 d = se->vruntime - cfs_rq->min_vruntime;
923
924	if (d < 0)
925	d = -d;
926
927	if (d > 3*sysctl_sched_latency)
928	schedstat_inc(cfs_rq, nr_spread_over);
929	#endif
930	}
931
932	static void
933	place_entity(struct cfs_rq cfs_rq, struct sched_entity se, int initial)
934	{
935	u64 vruntime = cfs_rq->min_vruntime;
936
937	/*
938	* The 'current' period is already promised to the current tasks,
939	* however the extra weight of the new task will slow them down a
940	* little, place the new task so that it fits in the slot that
941	* stays open at the end.
942	*/
943	if (initial && sched_feat(START_DEBIT))
944	vruntime += sched_vslice(cfs_rq, se);
945
946	/* sleeps up to a single latency don't count. */
947	if (!initial) {
948	unsigned long thresh = sysctl_sched_latency;
949
950	/*
951	* Halve their sleep time's effect, to allow
952	* for a gentler effect of sleepers:
953	*/
954	if (sched_feat(GENTLE_FAIR_SLEEPERS))
955	thresh >>= 1;
956
957	vruntime -= thresh;
958	}
959
960	/* ensure we never gain time by being placed backwards. */
961	vruntime = max_vruntime(se->vruntime, vruntime);
962
963	se->vruntime = vruntime;
964	}
965
966	static void
967	enqueue_entity(struct cfs_rq cfs_rq, struct sched_entity se, int flags)
968	{
969	/*
970	* Update the normalized vruntime before updating min_vruntime
971	* through callig update_curr().
972	*/
973	if (!(flags & ENQUEUE_WAKEUP) \|\| (flags & ENQUEUE_WAKING))
974	se->vruntime += cfs_rq->min_vruntime;
975
976	/*
977	* Update run-time statistics of the 'current'.
978	*/
979	update_curr(cfs_rq);
980	update_cfs_load(cfs_rq, 0);
981	update_cfs_shares(cfs_rq, se->load.weight);
982	account_entity_enqueue(cfs_rq, se);
983
984	if (flags & ENQUEUE_WAKEUP) {
985	place_entity(cfs_rq, se, 0);
986	enqueue_sleeper(cfs_rq, se);
987	}
988
989	update_stats_enqueue(cfs_rq, se);
990	check_spread(cfs_rq, se);
991	if (se != cfs_rq->curr)
992	__enqueue_entity(cfs_rq, se);
993	se->on_rq = 1;
994
995	if (cfs_rq->nr_running == 1)
996	list_add_leaf_cfs_rq(cfs_rq);
997	}
998
999	static void __clear_buddies(struct cfs_rq cfs_rq, struct sched_entity se)
1000	{
1001	if (!se \|\| cfs_rq->last == se)
1002	cfs_rq->last = NULL;
1003
1004	if (!se \|\| cfs_rq->next == se)
1005	cfs_rq->next = NULL;
1006	}
1007
1008	static void clear_buddies(struct cfs_rq cfs_rq, struct sched_entity se)
1009	{
1010	for_each_sched_entity(se)
1011	__clear_buddies(cfs_rq_of(se), se);
1012	}
1013
1014	static void
1015	dequeue_entity(struct cfs_rq cfs_rq, struct sched_entity se, int flags)
1016	{
1017	/*
1018	* Update run-time statistics of the 'current'.
1019	*/
1020	update_curr(cfs_rq);
1021
1022	update_stats_dequeue(cfs_rq, se);
1023	if (flags & DEQUEUE_SLEEP) {
1024	#ifdef CONFIG_SCHEDSTATS
1025	if (entity_is_task(se)) {
1026	struct task_struct *tsk = task_of(se);
1027
1028	if (tsk->state & TASK_INTERRUPTIBLE)
1029	se->statistics.sleep_start = rq_of(cfs_rq)->clock;
1030	if (tsk->state & TASK_UNINTERRUPTIBLE)
1031	se->statistics.block_start = rq_of(cfs_rq)->clock;
1032	}
1033	#endif
1034	}
1035
1036	clear_buddies(cfs_rq, se);
1037
1038	if (se != cfs_rq->curr)
1039	__dequeue_entity(cfs_rq, se);
1040	se->on_rq = 0;
1041	update_cfs_load(cfs_rq, 0);
1042	account_entity_dequeue(cfs_rq, se);
1043	update_min_vruntime(cfs_rq);
1044	update_cfs_shares(cfs_rq, 0);
1045
1046	/*
1047	* Normalize the entity after updating the min_vruntime because the
1048	* update can refer to the ->curr item and we need to reflect this
1049	* movement in our normalized position.
1050	*/
1051	if (!(flags & DEQUEUE_SLEEP))
1052	se->vruntime -= cfs_rq->min_vruntime;
1053	}
1054
1055	/*
1056	* Preempt the current task with a newly woken task if needed:
1057	*/
1058	static void
1059	check_preempt_tick(struct cfs_rq cfs_rq, struct sched_entity curr)
1060	{
1061	unsigned long ideal_runtime, delta_exec;
1062
1063	ideal_runtime = sched_slice(cfs_rq, curr);
1064	delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime;
1065	if (delta_exec > ideal_runtime) {
1066	resched_task(rq_of(cfs_rq)->curr);
1067	/*
1068	* The current task ran long enough, ensure it doesn't get
1069	* re-elected due to buddy favours.
1070	*/
1071	clear_buddies(cfs_rq, curr);
1072	return;
1073	}
1074
1075	/*
1076	* Ensure that a task that missed wakeup preemption by a
1077	* narrow margin doesn't have to wait for a full slice.
1078	* This also mitigates buddy induced latencies under load.
1079	*/
1080	if (!sched_feat(WAKEUP_PREEMPT))
1081	return;
1082
1083	if (delta_exec < sysctl_sched_min_granularity)
1084	return;
1085
1086	if (cfs_rq->nr_running > 1) {
1087	struct sched_entity *se = __pick_next_entity(cfs_rq);
1088	s64 delta = curr->vruntime - se->vruntime;
1089
1090	if (delta < 0)
1091	return;
1092
1093	if (delta > ideal_runtime)
1094	resched_task(rq_of(cfs_rq)->curr);
1095	}
1096	}
1097
1098	static void
1099	set_next_entity(struct cfs_rq cfs_rq, struct sched_entity se)
1100	{
1101	/* 'current' is not kept within the tree. */
1102	if (se->on_rq) {
1103	/*
1104	* Any task has to be enqueued before it get to execute on
1105	* a CPU. So account for the time it spent waiting on the
1106	* runqueue.
1107	*/
1108	update_stats_wait_end(cfs_rq, se);
1109	__dequeue_entity(cfs_rq, se);
1110	}
1111
1112	update_stats_curr_start(cfs_rq, se);
1113	cfs_rq->curr = se;
1114	#ifdef CONFIG_SCHEDSTATS
1115	/*
1116	* Track our maximum slice length, if the CPU's load is at
1117	* least twice that of our own weight (i.e. dont track it
1118	* when there are only lesser-weight tasks around):
1119	*/
1120	if (rq_of(cfs_rq)->load.weight >= 2*se->load.weight) {
1121	se->statistics.slice_max = max(se->statistics.slice_max,
1122	se->sum_exec_runtime - se->prev_sum_exec_runtime);
1123	}
1124	#endif
1125	se->prev_sum_exec_runtime = se->sum_exec_runtime;
1126	}
1127
1128	static int
1129	wakeup_preempt_entity(struct sched_entity curr, struct sched_entity se);
1130
1131	static struct sched_entity pick_next_entity(struct cfs_rq cfs_rq)
1132	{
1133	struct sched_entity *se = __pick_next_entity(cfs_rq);
1134	struct sched_entity *left = se;
1135
1136	if (cfs_rq->next && wakeup_preempt_entity(cfs_rq->next, left) < 1)
1137	se = cfs_rq->next;
1138
1139	/*
1140	* Prefer last buddy, try to return the CPU to a preempted task.
1141	*/
1142	if (cfs_rq->last && wakeup_preempt_entity(cfs_rq->last, left) < 1)
1143	se = cfs_rq->last;
1144
1145	clear_buddies(cfs_rq, se);
1146
1147	return se;
1148	}
1149
1150	static void put_prev_entity(struct cfs_rq cfs_rq, struct sched_entity prev)
1151	{
1152	/*
1153	* If still on the runqueue then deactivate_task()
1154	* was not called and update_curr() has to be done:
1155	*/
1156	if (prev->on_rq)
1157	update_curr(cfs_rq);
1158
1159	check_spread(cfs_rq, prev);
1160	if (prev->on_rq) {
1161	update_stats_wait_start(cfs_rq, prev);
1162	/* Put 'current' back into the tree. */
1163	__enqueue_entity(cfs_rq, prev);
1164	}
1165	cfs_rq->curr = NULL;
1166	}
1167
1168	static void
1169	entity_tick(struct cfs_rq cfs_rq, struct sched_entity curr, int queued)
1170	{
1171	/*
1172	* Update run-time statistics of the 'current'.
1173	*/
1174	update_curr(cfs_rq);
1175
1176	/*
1177	* Update share accounting for long-running entities.
1178	*/
1179	update_entity_shares_tick(cfs_rq);
1180
1181	#ifdef CONFIG_SCHED_HRTICK
1182	/*
1183	* queued ticks are scheduled to match the slice, so don't bother
1184	* validating it and just reschedule.
1185	*/
1186	if (queued) {
1187	resched_task(rq_of(cfs_rq)->curr);
1188	return;
1189	}
1190	/*
1191	* don't let the period tick interfere with the hrtick preemption
1192	*/
1193	if (!sched_feat(DOUBLE_TICK) &&
1194	hrtimer_active(&rq_of(cfs_rq)->hrtick_timer))
1195	return;
1196	#endif
1197
1198	if (cfs_rq->nr_running > 1 \|\| !sched_feat(WAKEUP_PREEMPT))
1199	check_preempt_tick(cfs_rq, curr);
1200	}
1201
1202	/**************************************************
1203	* CFS operations on tasks:
1204	*/
1205
1206	#ifdef CONFIG_SCHED_HRTICK
1207	static void hrtick_start_fair(struct rq rq, struct task_struct p)
1208	{
1209	struct sched_entity *se = &p->se;
1210	struct cfs_rq *cfs_rq = cfs_rq_of(se);
1211
1212	WARN_ON(task_rq(p) != rq);
1213
1214	if (hrtick_enabled(rq) && cfs_rq->nr_running > 1) {
1215	u64 slice = sched_slice(cfs_rq, se);
1216	u64 ran = se->sum_exec_runtime - se->prev_sum_exec_runtime;
1217	s64 delta = slice - ran;
1218
1219	if (delta < 0) {
1220	if (rq->curr == p)
1221	resched_task(p);
1222	return;
1223	}
1224
1225	/*
1226	* Don't schedule slices shorter than 10000ns, that just
1227	* doesn't make sense. Rely on vruntime for fairness.
1228	*/
1229	if (rq->curr != p)
1230	delta = max_t(s64, 10000LL, delta);
1231
1232	hrtick_start(rq, delta);
1233	}
1234	}
1235
1236	/*
1237	* called from enqueue/dequeue and updates the hrtick when the
1238	* current task is from our class and nr_running is low enough
1239	* to matter.
1240	*/
1241	static void hrtick_update(struct rq *rq)
1242	{
1243	struct task_struct *curr = rq->curr;
1244
1245	if (curr->sched_class != &fair_sched_class)
1246	return;
1247
1248	if (cfs_rq_of(&curr->se)->nr_running < sched_nr_latency)
1249	hrtick_start_fair(rq, curr);
1250	}
1251	#else /* !CONFIG_SCHED_HRTICK */
1252	static inline void
1253	hrtick_start_fair(struct rq rq, struct task_struct p)
1254	{
1255	}
1256
1257	static inline void hrtick_update(struct rq *rq)
1258	{
1259	}
1260	#endif
1261
1262	/*
1263	* The enqueue_task method is called before nr_running is
1264	* increased. Here we update the fair scheduling stats and
1265	* then put the task into the rbtree:
1266	*/
1267	static void
1268	enqueue_task_fair(struct rq rq, struct task_struct p, int flags)
1269	{
1270	struct cfs_rq *cfs_rq;
1271	struct sched_entity *se = &p->se;
1272
1273	for_each_sched_entity(se) {
1274	if (se->on_rq)
1275	break;
1276	cfs_rq = cfs_rq_of(se);
1277	enqueue_entity(cfs_rq, se, flags);
1278	flags = ENQUEUE_WAKEUP;
1279	}
1280
1281	for_each_sched_entity(se) {
1282	struct cfs_rq *cfs_rq = cfs_rq_of(se);
1283
1284	update_cfs_load(cfs_rq, 0);
1285	update_cfs_shares(cfs_rq, 0);
1286	}
1287
1288	hrtick_update(rq);
1289	}
1290
1291	/*
1292	* The dequeue_task method is called before nr_running is
1293	* decreased. We remove the task from the rbtree and
1294	* update the fair scheduling stats:
1295	*/
1296	static void dequeue_task_fair(struct rq rq, struct task_struct p, int flags)
1297	{
1298	struct cfs_rq *cfs_rq;
1299	struct sched_entity *se = &p->se;
1300
1301	for_each_sched_entity(se) {
1302	cfs_rq = cfs_rq_of(se);
1303	dequeue_entity(cfs_rq, se, flags);
1304
1305	/* Don't dequeue parent if it has other entities besides us */
1306	if (cfs_rq->load.weight)
1307	break;
1308	flags \|= DEQUEUE_SLEEP;
1309	}
1310
1311	for_each_sched_entity(se) {
1312	struct cfs_rq *cfs_rq = cfs_rq_of(se);
1313
1314	update_cfs_load(cfs_rq, 0);
1315	update_cfs_shares(cfs_rq, 0);
1316	}
1317
1318	hrtick_update(rq);
1319	}
1320
1321	/*
1322	* sched_yield() support is very simple - we dequeue and enqueue.
1323	*
1324	* If compat_yield is turned on then we requeue to the end of the tree.
1325	*/
1326	static void yield_task_fair(struct rq *rq)
1327	{
1328	struct task_struct *curr = rq->curr;
1329	struct cfs_rq *cfs_rq = task_cfs_rq(curr);
1330	struct sched_entity rightmost, se = &curr->se;
1331
1332	/*
1333	* Are we the only task in the tree?
1334	*/
1335	if (unlikely(cfs_rq->nr_running == 1))
1336	return;
1337
1338	clear_buddies(cfs_rq, se);
1339
1340	if (likely(!sysctl_sched_compat_yield) && curr->policy != SCHED_BATCH) {
1341	update_rq_clock(rq);
1342	/*
1343	* Update run-time statistics of the 'current'.
1344	*/
1345	update_curr(cfs_rq);
1346
1347	return;
1348	}
1349	/*
1350	* Find the rightmost entry in the rbtree:
1351	*/
1352	rightmost = __pick_last_entity(cfs_rq);
1353	/*
1354	* Already in the rightmost position?
1355	*/
1356	if (unlikely(!rightmost \|\| entity_before(rightmost, se)))
1357	return;
1358
1359	/*
1360	* Minimally necessary key value to be last in the tree:
1361	* Upon rescheduling, sched_class::put_prev_task() will place
1362	* 'current' within the tree based on its new key value.
1363	*/
1364	se->vruntime = rightmost->vruntime + 1;
1365	}
1366
1367	#ifdef CONFIG_SMP
1368
1369	static void task_waking_fair(struct rq rq, struct task_struct p)
1370	{
1371	struct sched_entity *se = &p->se;
1372	struct cfs_rq *cfs_rq = cfs_rq_of(se);
1373
1374	se->vruntime -= cfs_rq->min_vruntime;
1375	}
1376
1377	#ifdef CONFIG_FAIR_GROUP_SCHED
1378	/*
1379	* effective_load() calculates the load change as seen from the root_task_group
1380	*
1381	* Adding load to a group doesn't make a group heavier, but can cause movement
1382	* of group shares between cpus. Assuming the shares were perfectly aligned one
1383	* can calculate the shift in shares.
1384	*/
1385	static long effective_load(struct task_group *tg, int cpu, long wl, long wg)
1386	{
1387	struct sched_entity *se = tg->se[cpu];
1388
1389	if (!tg->parent)
1390	return wl;
1391
1392	for_each_sched_entity(se) {
1393	long lw, w;
1394
1395	tg = se->my_q->tg;
1396	w = se->my_q->load.weight;
1397
1398	/* use this cpu's instantaneous contribution */
1399	lw = atomic_read(&tg->load_weight);
1400	lw -= se->my_q->load_contribution;
1401	lw += w + wg;
1402
1403	wl += w;
1404
1405	if (lw > 0 && wl < lw)
1406	wl = (wl * tg->shares) / lw;
1407	else
1408	wl = tg->shares;
1409
1410	/* zero point is MIN_SHARES */
1411	if (wl < MIN_SHARES)
1412	wl = MIN_SHARES;
1413	wl -= se->load.weight;
1414	wg = 0;
1415	}
1416
1417	return wl;
1418	}
1419
1420	#else
1421
1422	static inline unsigned long effective_load(struct task_group *tg, int cpu,
1423	unsigned long wl, unsigned long wg)
1424	{
1425	return wl;
1426	}
1427
1428	#endif
1429
1430	static int wake_affine(struct sched_domain sd, struct task_struct p, int sync)
1431	{
1432	s64 this_load, load;
1433	int idx, this_cpu, prev_cpu;
1434	unsigned long tl_per_task;
1435	struct task_group *tg;
1436	unsigned long weight;
1437	int balanced;
1438
1439	idx = sd->wake_idx;
1440	this_cpu = smp_processor_id();
1441	prev_cpu = task_cpu(p);
1442	load = source_load(prev_cpu, idx);
1443	this_load = target_load(this_cpu, idx);
1444
1445	/*
1446	* If sync wakeup then subtract the (maximum possible)
1447	* effect of the currently running task from the load
1448	* of the current CPU:
1449	*/
1450	rcu_read_lock();
1451	if (sync) {
1452	tg = task_group(current);
1453	weight = current->se.load.weight;
1454
1455	this_load += effective_load(tg, this_cpu, -weight, -weight);
1456	load += effective_load(tg, prev_cpu, 0, -weight);
1457	}
1458
1459	tg = task_group(p);
1460	weight = p->se.load.weight;
1461
1462	/*
1463	* In low-load situations, where prev_cpu is idle and this_cpu is idle
1464	* due to the sync cause above having dropped this_load to 0, we'll
1465	* always have an imbalance, but there's really nothing you can do
1466	* about that, so that's good too.
1467	*
1468	* Otherwise check if either cpus are near enough in load to allow this
1469	* task to be woken on this_cpu.
1470	*/
1471	if (this_load > 0) {
1472	s64 this_eff_load, prev_eff_load;
1473
1474	this_eff_load = 100;
1475	this_eff_load *= power_of(prev_cpu);
1476	this_eff_load *= this_load +
1477	effective_load(tg, this_cpu, weight, weight);
1478
1479	prev_eff_load = 100 + (sd->imbalance_pct - 100) / 2;
1480	prev_eff_load *= power_of(this_cpu);
1481	prev_eff_load *= load + effective_load(tg, prev_cpu, 0, weight);
1482
1483	balanced = this_eff_load <= prev_eff_load;
1484	} else
1485	balanced = true;
1486	rcu_read_unlock();
1487
1488	/*
1489	* If the currently running task will sleep within
1490	* a reasonable amount of time then attract this newly
1491	* woken task:
1492	*/
1493	if (sync && balanced)
1494	return 1;
1495
1496	schedstat_inc(p, se.statistics.nr_wakeups_affine_attempts);
1497	tl_per_task = cpu_avg_load_per_task(this_cpu);
1498
1499	if (balanced \|\|
1500	(this_load <= load &&
1501	this_load + target_load(prev_cpu, idx) <= tl_per_task)) {
1502	/*
1503	* This domain has SD_WAKE_AFFINE and
1504	* p is cache cold in this domain, and
1505	* there is no bad imbalance.
1506	*/
1507	schedstat_inc(sd, ttwu_move_affine);
1508	schedstat_inc(p, se.statistics.nr_wakeups_affine);
1509
1510	return 1;
1511	}
1512	return 0;
1513	}
1514
1515	/*
1516	* find_idlest_group finds and returns the least busy CPU group within the
1517	* domain.
1518	*/
1519	static struct sched_group *
1520	find_idlest_group(struct sched_domain sd, struct task_struct p,
1521	int this_cpu, int load_idx)
1522	{
1523	struct sched_group idlest = NULL, group = sd->groups;
1524	unsigned long min_load = ULONG_MAX, this_load = 0;
1525	int imbalance = 100 + (sd->imbalance_pct-100)/2;
1526
1527	do {
1528	unsigned long load, avg_load;
1529	int local_group;
1530	int i;
1531
1532	/* Skip over this group if it has no CPUs allowed */
1533	if (!cpumask_intersects(sched_group_cpus(group),
1534	&p->cpus_allowed))
1535	continue;
1536
1537	local_group = cpumask_test_cpu(this_cpu,
1538	sched_group_cpus(group));
1539
1540	/* Tally up the load of all CPUs in the group */
1541	avg_load = 0;
1542
1543	for_each_cpu(i, sched_group_cpus(group)) {
1544	/* Bias balancing toward cpus of our domain */
1545	if (local_group)
1546	load = source_load(i, load_idx);
1547	else
1548	load = target_load(i, load_idx);
1549
1550	avg_load += load;
1551	}
1552
1553	/* Adjust by relative CPU power of the group */
1554	avg_load = (avg_load * SCHED_LOAD_SCALE) / group->cpu_power;
1555
1556	if (local_group) {
1557	this_load = avg_load;
1558	} else if (avg_load < min_load) {
1559	min_load = avg_load;
1560	idlest = group;
1561	}
1562	} while (group = group->next, group != sd->groups);
1563
1564	if (!idlest \|\| 100this_load < imbalancemin_load)
1565	return NULL;
1566	return idlest;
1567	}
1568
1569	/*
1570	* find_idlest_cpu - find the idlest cpu among the cpus in group.
1571	*/
1572	static int
1573	find_idlest_cpu(struct sched_group group, struct task_struct p, int this_cpu)
1574	{
1575	unsigned long load, min_load = ULONG_MAX;
1576	int idlest = -1;
1577	int i;
1578
1579	/* Traverse only the allowed CPUs */
1580	for_each_cpu_and(i, sched_group_cpus(group), &p->cpus_allowed) {
1581	load = weighted_cpuload(i);
1582
1583	if (load < min_load \|\| (load == min_load && i == this_cpu)) {
1584	min_load = load;
1585	idlest = i;
1586	}
1587	}
1588
1589	return idlest;
1590	}
1591
1592	/*
1593	* Try and locate an idle CPU in the sched_domain.
1594	*/
1595	static int select_idle_sibling(struct task_struct *p, int target)
1596	{
1597	int cpu = smp_processor_id();
1598	int prev_cpu = task_cpu(p);
1599	struct sched_domain *sd;
1600	int i;
1601
1602	/*
1603	* If the task is going to be woken-up on this cpu and if it is
1604	* already idle, then it is the right target.
1605	*/
1606	if (target == cpu && idle_cpu(cpu))
1607	return cpu;
1608
1609	/*
1610	* If the task is going to be woken-up on the cpu where it previously
1611	* ran and if it is currently idle, then it the right target.
1612	*/
1613	if (target == prev_cpu && idle_cpu(prev_cpu))
1614	return prev_cpu;
1615
1616	/*
1617	* Otherwise, iterate the domains and find an elegible idle cpu.
1618	*/
1619	for_each_domain(target, sd) {
1620	if (!(sd->flags & SD_SHARE_PKG_RESOURCES))
1621	break;
1622
1623	for_each_cpu_and(i, sched_domain_span(sd), &p->cpus_allowed) {
1624	if (idle_cpu(i)) {
1625	target = i;
1626	break;
1627	}
1628	}
1629
1630	/*
1631	* Lets stop looking for an idle sibling when we reached
1632	* the domain that spans the current cpu and prev_cpu.
1633	*/
1634	if (cpumask_test_cpu(cpu, sched_domain_span(sd)) &&
1635	cpumask_test_cpu(prev_cpu, sched_domain_span(sd)))
1636	break;
1637	}
1638
1639	return target;
1640	}
1641
1642	/*
1643	* sched_balance_self: balance the current task (running on cpu) in domains
1644	* that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and
1645	* SD_BALANCE_EXEC.
1646	*
1647	* Balance, ie. select the least loaded group.
1648	*
1649	* Returns the target CPU number, or the same CPU if no balancing is needed.
1650	*
1651	* preempt must be disabled.
1652	*/
1653	static int
1654	select_task_rq_fair(struct rq rq, struct task_struct p, int sd_flag, int wake_flags)
1655	{
1656	struct sched_domain tmp, affine_sd = NULL, *sd = NULL;
1657	int cpu = smp_processor_id();
1658	int prev_cpu = task_cpu(p);
1659	int new_cpu = cpu;
1660	int want_affine = 0;
1661	int want_sd = 1;
1662	int sync = wake_flags & WF_SYNC;
1663
1664	if (sd_flag & SD_BALANCE_WAKE) {
1665	if (cpumask_test_cpu(cpu, &p->cpus_allowed))
1666	want_affine = 1;
1667	new_cpu = prev_cpu;
1668	}
1669
1670	for_each_domain(cpu, tmp) {
1671	if (!(tmp->flags & SD_LOAD_BALANCE))
1672	continue;
1673
1674	/*
1675	* If power savings logic is enabled for a domain, see if we
1676	* are not overloaded, if so, don't balance wider.
1677	*/
1678	if (tmp->flags & (SD_POWERSAVINGS_BALANCE\|SD_PREFER_LOCAL)) {
1679	unsigned long power = 0;
1680	unsigned long nr_running = 0;
1681	unsigned long capacity;
1682	int i;
1683
1684	for_each_cpu(i, sched_domain_span(tmp)) {
1685	power += power_of(i);
1686	nr_running += cpu_rq(i)->cfs.nr_running;
1687	}
1688
1689	capacity = DIV_ROUND_CLOSEST(power, SCHED_LOAD_SCALE);
1690
1691	if (tmp->flags & SD_POWERSAVINGS_BALANCE)
1692	nr_running /= 2;
1693
1694	if (nr_running < capacity)
1695	want_sd = 0;
1696	}
1697
1698	/*
1699	* If both cpu and prev_cpu are part of this domain,
1700	* cpu is a valid SD_WAKE_AFFINE target.
1701	*/
1702	if (want_affine && (tmp->flags & SD_WAKE_AFFINE) &&
1703	cpumask_test_cpu(prev_cpu, sched_domain_span(tmp))) {
1704	affine_sd = tmp;
1705	want_affine = 0;
1706	}
1707
1708	if (!want_sd && !want_affine)
1709	break;
1710
1711	if (!(tmp->flags & sd_flag))
1712	continue;
1713
1714	if (want_sd)
1715	sd = tmp;
1716	}
1717
1718	if (affine_sd) {
1719	if (cpu == prev_cpu \|\| wake_affine(affine_sd, p, sync))
1720	return select_idle_sibling(p, cpu);
1721	else
1722	return select_idle_sibling(p, prev_cpu);
1723	}
1724
1725	while (sd) {
1726	int load_idx = sd->forkexec_idx;
1727	struct sched_group *group;
1728	int weight;
1729
1730	if (!(sd->flags & sd_flag)) {
1731	sd = sd->child;
1732	continue;
1733	}
1734
1735	if (sd_flag & SD_BALANCE_WAKE)
1736	load_idx = sd->wake_idx;
1737
1738	group = find_idlest_group(sd, p, cpu, load_idx);
1739	if (!group) {
1740	sd = sd->child;
1741	continue;
1742	}
1743
1744	new_cpu = find_idlest_cpu(group, p, cpu);
1745	if (new_cpu == -1 \|\| new_cpu == cpu) {
1746	/* Now try balancing at a lower domain level of cpu */
1747	sd = sd->child;
1748	continue;
1749	}
1750
1751	/* Now try balancing at a lower domain level of new_cpu */
1752	cpu = new_cpu;
1753	weight = sd->span_weight;
1754	sd = NULL;
1755	for_each_domain(cpu, tmp) {
1756	if (weight <= tmp->span_weight)
1757	break;
1758	if (tmp->flags & sd_flag)
1759	sd = tmp;
1760	}
1761	/* while loop will break here if sd == NULL */
1762	}
1763
1764	return new_cpu;
1765	}
1766	#endif /* CONFIG_SMP */
1767
1768	static unsigned long
1769	wakeup_gran(struct sched_entity curr, struct sched_entity se)
1770	{
1771	unsigned long gran = sysctl_sched_wakeup_granularity;
1772
1773	/*
1774	* Since its curr running now, convert the gran from real-time
1775	* to virtual-time in his units.
1776	*
1777	* By using 'se' instead of 'curr' we penalize light tasks, so
1778	* they get preempted easier. That is, if 'se' < 'curr' then
1779	* the resulting gran will be larger, therefore penalizing the
1780	* lighter, if otoh 'se' > 'curr' then the resulting gran will
1781	* be smaller, again penalizing the lighter task.
1782	*
1783	* This is especially important for buddies when the leftmost
1784	* task is higher priority than the buddy.
1785	*/
1786	if (unlikely(se->load.weight != NICE_0_LOAD))
1787	gran = calc_delta_fair(gran, se);
1788
1789	return gran;
1790	}
1791
1792	/*
1793	* Should 'se' preempt 'curr'.
1794	*
1795	* \|s1
1796	* \|s2
1797	* \|s3
1798	* g
1799	* \|<--->\|c
1800	*
1801	* w(c, s1) = -1
1802	* w(c, s2) = 0
1803	* w(c, s3) = 1
1804	*
1805	*/
1806	static int
1807	wakeup_preempt_entity(struct sched_entity curr, struct sched_entity se)
1808	{
1809	s64 gran, vdiff = curr->vruntime - se->vruntime;
1810
1811	if (vdiff <= 0)
1812	return -1;
1813
1814	gran = wakeup_gran(curr, se);
1815	if (vdiff > gran)
1816	return 1;
1817
1818	return 0;
1819	}
1820
1821	static void set_last_buddy(struct sched_entity *se)
1822	{
1823	if (likely(task_of(se)->policy != SCHED_IDLE)) {
1824	for_each_sched_entity(se)
1825	cfs_rq_of(se)->last = se;
1826	}
1827	}
1828
1829	static void set_next_buddy(struct sched_entity *se)
1830	{
1831	if (likely(task_of(se)->policy != SCHED_IDLE)) {
1832	for_each_sched_entity(se)
1833	cfs_rq_of(se)->next = se;
1834	}
1835	}
1836
1837	/*
1838	* Preempt the current task with a newly woken task if needed:
1839	*/
1840	static void check_preempt_wakeup(struct rq rq, struct task_struct p, int wake_flags)
1841	{
1842	struct task_struct *curr = rq->curr;
1843	struct sched_entity se = &curr->se, pse = &p->se;
1844	struct cfs_rq *cfs_rq = task_cfs_rq(curr);
1845	int scale = cfs_rq->nr_running >= sched_nr_latency;
1846
1847	if (unlikely(se == pse))
1848	return;
1849
1850	if (sched_feat(NEXT_BUDDY) && scale && !(wake_flags & WF_FORK))
1851	set_next_buddy(pse);
1852
1853	/*
1854	* We can come here with TIF_NEED_RESCHED already set from new task
1855	* wake up path.
1856	*/
1857	if (test_tsk_need_resched(curr))
1858	return;
1859
1860	/*
1861	* Batch and idle tasks do not preempt (their preemption is driven by
1862	* the tick):
1863	*/
1864	if (unlikely(p->policy != SCHED_NORMAL))
1865	return;
1866
1867	/* Idle tasks are by definition preempted by everybody. */
1868	if (unlikely(curr->policy == SCHED_IDLE))
1869	goto preempt;
1870
1871	if (!sched_feat(WAKEUP_PREEMPT))
1872	return;
1873
1874	update_curr(cfs_rq);
1875	find_matching_se(&se, &pse);
1876	BUG_ON(!pse);
1877	if (wakeup_preempt_entity(se, pse) == 1)
1878	goto preempt;
1879
1880	return;
1881
1882	preempt:
1883	resched_task(curr);
1884	/*
1885	* Only set the backward buddy when the current task is still
1886	* on the rq. This can happen when a wakeup gets interleaved
1887	* with schedule on the ->pre_schedule() or idle_balance()
1888	* point, either of which can * drop the rq lock.
1889	*
1890	* Also, during early boot the idle thread is in the fair class,
1891	* for obvious reasons its a bad idea to schedule back to it.
1892	*/
1893	if (unlikely(!se->on_rq \|\| curr == rq->idle))
1894	return;
1895
1896	if (sched_feat(LAST_BUDDY) && scale && entity_is_task(se))
1897	set_last_buddy(se);
1898	}
1899
1900	static struct task_struct pick_next_task_fair(struct rq rq)
1901	{
1902	struct task_struct *p;
1903	struct cfs_rq *cfs_rq = &rq->cfs;
1904	struct sched_entity *se;
1905
1906	if (!cfs_rq->nr_running)
1907	return NULL;
1908
1909	do {
1910	se = pick_next_entity(cfs_rq);
1911	set_next_entity(cfs_rq, se);
1912	cfs_rq = group_cfs_rq(se);
1913	} while (cfs_rq);
1914
1915	p = task_of(se);
1916	hrtick_start_fair(rq, p);
1917
1918	return p;
1919	}
1920
1921	/*
1922	* Account for a descheduled task:
1923	*/
1924	static void put_prev_task_fair(struct rq rq, struct task_struct prev)
1925	{
1926	struct sched_entity *se = &prev->se;
1927	struct cfs_rq *cfs_rq;
1928
1929	for_each_sched_entity(se) {
1930	cfs_rq = cfs_rq_of(se);
1931	put_prev_entity(cfs_rq, se);
1932	}
1933	}
1934
1935	#ifdef CONFIG_SMP
1936	/**************************************************
1937	* Fair scheduling class load-balancing methods:
1938	*/
1939
1940	/*
1941	* pull_task - move a task from a remote runqueue to the local runqueue.
1942	* Both runqueues must be locked.
1943	*/
1944	static void pull_task(struct rq src_rq, struct task_struct p,
1945	struct rq *this_rq, int this_cpu)
1946	{
1947	deactivate_task(src_rq, p, 0);
1948	set_task_cpu(p, this_cpu);
1949	activate_task(this_rq, p, 0);
1950	check_preempt_curr(this_rq, p, 0);
1951	}
1952
1953	/*
1954	* can_migrate_task - may task p from runqueue rq be migrated to this_cpu?
1955	*/
1956	static
1957	int can_migrate_task(struct task_struct p, struct rq rq, int this_cpu,
1958	struct sched_domain *sd, enum cpu_idle_type idle,
1959	int *all_pinned)
1960	{
1961	int tsk_cache_hot = 0;
1962	/*
1963	* We do not migrate tasks that are:
1964	* 1) running (obviously), or
1965	* 2) cannot be migrated to this CPU due to cpus_allowed, or
1966	* 3) are cache-hot on their current CPU.
1967	*/
1968	if (!cpumask_test_cpu(this_cpu, &p->cpus_allowed)) {
1969	schedstat_inc(p, se.statistics.nr_failed_migrations_affine);
1970	return 0;
1971	}
1972	*all_pinned = 0;
1973
1974	if (task_running(rq, p)) {
1975	schedstat_inc(p, se.statistics.nr_failed_migrations_running);
1976	return 0;
1977	}
1978
1979	/*
1980	* Aggressive migration if:
1981	* 1) task is cache cold, or
1982	* 2) too many balance attempts have failed.
1983	*/
1984
1985	tsk_cache_hot = task_hot(p, rq->clock_task, sd);
1986	if (!tsk_cache_hot \|\|
1987	sd->nr_balance_failed > sd->cache_nice_tries) {
1988	#ifdef CONFIG_SCHEDSTATS
1989	if (tsk_cache_hot) {
1990	schedstat_inc(sd, lb_hot_gained[idle]);
1991	schedstat_inc(p, se.statistics.nr_forced_migrations);
1992	}
1993	#endif
1994	return 1;
1995	}
1996
1997	if (tsk_cache_hot) {
1998	schedstat_inc(p, se.statistics.nr_failed_migrations_hot);
1999	return 0;
2000	}
2001	return 1;
2002	}
2003
2004	/*
2005	* move_one_task tries to move exactly one task from busiest to this_rq, as
2006	* part of active balancing operations within "domain".
2007	* Returns 1 if successful and 0 otherwise.
2008	*
2009	* Called with both runqueues locked.
2010	*/
2011	static int
2012	move_one_task(struct rq this_rq, int this_cpu, struct rq busiest,
2013	struct sched_domain *sd, enum cpu_idle_type idle)
2014	{
2015	struct task_struct p, n;
2016	struct cfs_rq *cfs_rq;
2017	int pinned = 0;
2018
2019	for_each_leaf_cfs_rq(busiest, cfs_rq) {
2020	list_for_each_entry_safe(p, n, &cfs_rq->tasks, se.group_node) {
2021
2022	if (!can_migrate_task(p, busiest, this_cpu,
2023	sd, idle, &pinned))
2024	continue;
2025
2026	pull_task(busiest, p, this_rq, this_cpu);
2027	/*
2028	* Right now, this is only the second place pull_task()
2029	* is called, so we can safely collect pull_task()
2030	* stats here rather than inside pull_task().
2031	*/
2032	schedstat_inc(sd, lb_gained[idle]);
2033	return 1;
2034	}
2035	}
2036
2037	return 0;
2038	}
2039
2040	static unsigned long
2041	balance_tasks(struct rq this_rq, int this_cpu, struct rq busiest,
2042	unsigned long max_load_move, struct sched_domain *sd,
2043	enum cpu_idle_type idle, int *all_pinned,
2044	int this_best_prio, struct cfs_rq busiest_cfs_rq)
2045	{
2046	int loops = 0, pulled = 0, pinned = 0;
2047	long rem_load_move = max_load_move;
2048	struct task_struct p, n;
2049
2050	if (max_load_move == 0)
2051	goto out;
2052
2053	pinned = 1;
2054
2055	list_for_each_entry_safe(p, n, &busiest_cfs_rq->tasks, se.group_node) {
2056	if (loops++ > sysctl_sched_nr_migrate)
2057	break;
2058
2059	if ((p->se.load.weight >> 1) > rem_load_move \|\|
2060	!can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned))
2061	continue;
2062
2063	pull_task(busiest, p, this_rq, this_cpu);
2064	pulled++;
2065	rem_load_move -= p->se.load.weight;
2066
2067	#ifdef CONFIG_PREEMPT
2068	/*
2069	* NEWIDLE balancing is a source of latency, so preemptible
2070	* kernels will stop after the first task is pulled to minimize
2071	* the critical section.
2072	*/
2073	if (idle == CPU_NEWLY_IDLE)
2074	break;
2075	#endif
2076
2077	/*
2078	* We only want to steal up to the prescribed amount of
2079	* weighted load.
2080	*/
2081	if (rem_load_move <= 0)
2082	break;
2083
2084	if (p->prio < *this_best_prio)
2085	*this_best_prio = p->prio;
2086	}
2087	out:
2088	/*
2089	* Right now, this is one of only two places pull_task() is called,
2090	* so we can safely collect pull_task() stats here rather than
2091	* inside pull_task().
2092	*/
2093	schedstat_add(sd, lb_gained[idle], pulled);
2094
2095	if (all_pinned)
2096	*all_pinned = pinned;
2097
2098	return max_load_move - rem_load_move;
2099	}
2100
2101	#ifdef CONFIG_FAIR_GROUP_SCHED
2102	/*
2103	* update tg->load_weight by folding this cpu's load_avg
2104	*/
2105	static int update_shares_cpu(struct task_group *tg, int cpu)
2106	{
2107	struct cfs_rq *cfs_rq;
2108	unsigned long flags;
2109	struct rq *rq;
2110
2111	if (!tg->se[cpu])
2112	return 0;
2113
2114	rq = cpu_rq(cpu);
2115	cfs_rq = tg->cfs_rq[cpu];
2116
2117	raw_spin_lock_irqsave(&rq->lock, flags);
2118
2119	update_rq_clock(rq);
2120	update_cfs_load(cfs_rq, 1);
2121
2122	/*
2123	* We need to update shares after updating tg->load_weight in
2124	* order to adjust the weight of groups with long running tasks.
2125	*/
2126	update_cfs_shares(cfs_rq, 0);
2127
2128	raw_spin_unlock_irqrestore(&rq->lock, flags);
2129
2130	return 0;
2131	}
2132
2133	static void update_shares(int cpu)
2134	{
2135	struct cfs_rq *cfs_rq;
2136	struct rq *rq = cpu_rq(cpu);
2137
2138	rcu_read_lock();
2139	for_each_leaf_cfs_rq(rq, cfs_rq)
2140	update_shares_cpu(cfs_rq->tg, cpu);
2141	rcu_read_unlock();
2142	}
2143
2144	static unsigned long
2145	load_balance_fair(struct rq this_rq, int this_cpu, struct rq busiest,
2146	unsigned long max_load_move,
2147	struct sched_domain *sd, enum cpu_idle_type idle,
2148	int all_pinned, int this_best_prio)
2149	{
2150	long rem_load_move = max_load_move;
2151	int busiest_cpu = cpu_of(busiest);
2152	struct task_group *tg;
2153
2154	rcu_read_lock();
2155	update_h_load(busiest_cpu);
2156
2157	list_for_each_entry_rcu(tg, &task_groups, list) {
2158	struct cfs_rq *busiest_cfs_rq = tg->cfs_rq[busiest_cpu];
2159	unsigned long busiest_h_load = busiest_cfs_rq->h_load;
2160	unsigned long busiest_weight = busiest_cfs_rq->load.weight;
2161	u64 rem_load, moved_load;
2162
2163	/*
2164	* empty group
2165	*/
2166	if (!busiest_cfs_rq->task_weight)
2167	continue;
2168
2169	rem_load = (u64)rem_load_move * busiest_weight;
2170	rem_load = div_u64(rem_load, busiest_h_load + 1);
2171
2172	moved_load = balance_tasks(this_rq, this_cpu, busiest,
2173	rem_load, sd, idle, all_pinned, this_best_prio,
2174	busiest_cfs_rq);
2175
2176	if (!moved_load)
2177	continue;
2178
2179	moved_load *= busiest_h_load;
2180	moved_load = div_u64(moved_load, busiest_weight + 1);
2181
2182	rem_load_move -= moved_load;
2183	if (rem_load_move < 0)
2184	break;
2185	}
2186	rcu_read_unlock();
2187
2188	return max_load_move - rem_load_move;
2189	}
2190	#else
2191	static inline void update_shares(int cpu)
2192	{
2193	}
2194
2195	static unsigned long
2196	load_balance_fair(struct rq this_rq, int this_cpu, struct rq busiest,
2197	unsigned long max_load_move,
2198	struct sched_domain *sd, enum cpu_idle_type idle,
2199	int all_pinned, int this_best_prio)
2200	{
2201	return balance_tasks(this_rq, this_cpu, busiest,
2202	max_load_move, sd, idle, all_pinned,
2203	this_best_prio, &busiest->cfs);
2204	}
2205	#endif
2206
2207	/*
2208	* move_tasks tries to move up to max_load_move weighted load from busiest to
2209	* this_rq, as part of a balancing operation within domain "sd".
2210	* Returns 1 if successful and 0 otherwise.
2211	*
2212	* Called with both runqueues locked.
2213	*/
2214	static int move_tasks(struct rq this_rq, int this_cpu, struct rq busiest,
2215	unsigned long max_load_move,
2216	struct sched_domain *sd, enum cpu_idle_type idle,
2217	int *all_pinned)
2218	{
2219	unsigned long total_load_moved = 0, load_moved;
2220	int this_best_prio = this_rq->curr->prio;
2221
2222	do {
2223	load_moved = load_balance_fair(this_rq, this_cpu, busiest,
2224	max_load_move - total_load_moved,
2225	sd, idle, all_pinned, &this_best_prio);
2226
2227	total_load_moved += load_moved;
2228
2229	#ifdef CONFIG_PREEMPT
2230	/*
2231	* NEWIDLE balancing is a source of latency, so preemptible
2232	* kernels will stop after the first task is pulled to minimize
2233	* the critical section.
2234	*/
2235	if (idle == CPU_NEWLY_IDLE && this_rq->nr_running)
2236	break;
2237
2238	if (raw_spin_is_contended(&this_rq->lock) \|\|
2239	raw_spin_is_contended(&busiest->lock))
2240	break;
2241	#endif
2242	} while (load_moved && max_load_move > total_load_moved);
2243
2244	return total_load_moved > 0;
2245	}
2246
2247	/******** Helpers for find_busiest_group **********************/
2248	/*
2249	* sd_lb_stats - Structure to store the statistics of a sched_domain
2250	* during load balancing.
2251	*/
2252	struct sd_lb_stats {
2253	struct sched_group busiest; / Busiest group in this sd */
2254	struct sched_group this; / Local group in this sd */
2255	unsigned long total_load; /* Total load of all groups in sd */
2256	unsigned long total_pwr; /* Total power of all groups in sd */
2257	unsigned long avg_load; /* Average load across all groups in sd */
2258
2259	/** Statistics of this group */
2260	unsigned long this_load;
2261	unsigned long this_load_per_task;
2262	unsigned long this_nr_running;
2263	unsigned long this_has_capacity;
2264	unsigned int this_idle_cpus;
2265
2266	/* Statistics of the busiest group */
2267	unsigned int busiest_idle_cpus;
2268	unsigned long max_load;
2269	unsigned long busiest_load_per_task;
2270	unsigned long busiest_nr_running;
2271	unsigned long busiest_group_capacity;
2272	unsigned long busiest_has_capacity;
2273	unsigned int busiest_group_weight;
2274
2275	int group_imb; /* Is there imbalance in this sd */
2276	#if defined(CONFIG_SCHED_MC) \|\| defined(CONFIG_SCHED_SMT)
2277	int power_savings_balance; /* Is powersave balance needed for this sd */
2278	struct sched_group group_min; / Least loaded group in sd */
2279	struct sched_group group_leader; / Group which relieves group_min */
2280	unsigned long min_load_per_task; /* load_per_task in group_min */
2281	unsigned long leader_nr_running; /* Nr running of group_leader */
2282	unsigned long min_nr_running; /* Nr running of group_min */
2283	#endif
2284	};
2285
2286	/*
2287	* sg_lb_stats - stats of a sched_group required for load_balancing
2288	*/
2289	struct sg_lb_stats {
2290	unsigned long avg_load; /Avg load across the CPUs of the group /
2291	unsigned long group_load; /* Total load over the CPUs of the group */
2292	unsigned long sum_nr_running; /* Nr tasks running in the group */
2293	unsigned long sum_weighted_load; /* Weighted load of group's tasks */
2294	unsigned long group_capacity;
2295	unsigned long idle_cpus;
2296	unsigned long group_weight;
2297	int group_imb; /* Is there an imbalance in the group ? */
2298	int group_has_capacity; /* Is there extra capacity in the group? */
2299	};
2300
2301	/**
2302	* group_first_cpu - Returns the first cpu in the cpumask of a sched_group.
2303	* @group: The group whose first cpu is to be returned.
2304	*/
2305	static inline unsigned int group_first_cpu(struct sched_group *group)
2306	{
2307	return cpumask_first(sched_group_cpus(group));
2308	}
2309
2310	/**
2311	* get_sd_load_idx - Obtain the load index for a given sched domain.
2312	* @sd: The sched_domain whose load_idx is to be obtained.
2313	* @idle: The Idle status of the CPU for whose sd load_icx is obtained.
2314	*/
2315	static inline int get_sd_load_idx(struct sched_domain *sd,
2316	enum cpu_idle_type idle)
2317	{
2318	int load_idx;
2319
2320	switch (idle) {
2321	case CPU_NOT_IDLE:
2322	load_idx = sd->busy_idx;
2323	break;
2324
2325	case CPU_NEWLY_IDLE:
2326	load_idx = sd->newidle_idx;
2327	break;
2328	default:
2329	load_idx = sd->idle_idx;
2330	break;
2331	}
2332
2333	return load_idx;
2334	}
2335
2336
2337	#if defined(CONFIG_SCHED_MC) \|\| defined(CONFIG_SCHED_SMT)
2338	/**
2339	* init_sd_power_savings_stats - Initialize power savings statistics for
2340	* the given sched_domain, during load balancing.
2341	*
2342	* @sd: Sched domain whose power-savings statistics are to be initialized.
2343	* @sds: Variable containing the statistics for sd.
2344	* @idle: Idle status of the CPU at which we're performing load-balancing.
2345	*/
2346	static inline void init_sd_power_savings_stats(struct sched_domain *sd,
2347	struct sd_lb_stats *sds, enum cpu_idle_type idle)
2348	{
2349	/*
2350	* Busy processors will not participate in power savings
2351	* balance.
2352	*/
2353	if (idle == CPU_NOT_IDLE \|\| !(sd->flags & SD_POWERSAVINGS_BALANCE))
2354	sds->power_savings_balance = 0;
2355	else {
2356	sds->power_savings_balance = 1;
2357	sds->min_nr_running = ULONG_MAX;
2358	sds->leader_nr_running = 0;
2359	}
2360	}
2361
2362	/**
2363	* update_sd_power_savings_stats - Update the power saving stats for a
2364	* sched_domain while performing load balancing.
2365	*
2366	* @group: sched_group belonging to the sched_domain under consideration.
2367	* @sds: Variable containing the statistics of the sched_domain
2368	* @local_group: Does group contain the CPU for which we're performing
2369	* load balancing ?
2370	* @sgs: Variable containing the statistics of the group.
2371	*/
2372	static inline void update_sd_power_savings_stats(struct sched_group *group,
2373	struct sd_lb_stats sds, int local_group, struct sg_lb_stats sgs)
2374	{
2375
2376	if (!sds->power_savings_balance)
2377	return;
2378
2379	/*
2380	* If the local group is idle or completely loaded
2381	* no need to do power savings balance at this domain
2382	*/
2383	if (local_group && (sds->this_nr_running >= sgs->group_capacity \|\|
2384	!sds->this_nr_running))
2385	sds->power_savings_balance = 0;
2386
2387	/*
2388	* If a group is already running at full capacity or idle,
2389	* don't include that group in power savings calculations
2390	*/
2391	if (!sds->power_savings_balance \|\|
2392	sgs->sum_nr_running >= sgs->group_capacity \|\|
2393	!sgs->sum_nr_running)
2394	return;
2395
2396	/*
2397	* Calculate the group which has the least non-idle load.
2398	* This is the group from where we need to pick up the load
2399	* for saving power
2400	*/
2401	if ((sgs->sum_nr_running < sds->min_nr_running) \|\|
2402	(sgs->sum_nr_running == sds->min_nr_running &&
2403	group_first_cpu(group) > group_first_cpu(sds->group_min))) {
2404	sds->group_min = group;
2405	sds->min_nr_running = sgs->sum_nr_running;
2406	sds->min_load_per_task = sgs->sum_weighted_load /
2407	sgs->sum_nr_running;
2408	}
2409
2410	/*
2411	* Calculate the group which is almost near its
2412	* capacity but still has some space to pick up some load
2413	* from other group and save more power
2414	*/
2415	if (sgs->sum_nr_running + 1 > sgs->group_capacity)
2416	return;
2417
2418	if (sgs->sum_nr_running > sds->leader_nr_running \|\|
2419	(sgs->sum_nr_running == sds->leader_nr_running &&
2420	group_first_cpu(group) < group_first_cpu(sds->group_leader))) {
2421	sds->group_leader = group;
2422	sds->leader_nr_running = sgs->sum_nr_running;
2423	}
2424	}
2425
2426	/**
2427	* check_power_save_busiest_group - see if there is potential for some power-savings balance
2428	* @sds: Variable containing the statistics of the sched_domain
2429	* under consideration.
2430	* @this_cpu: Cpu at which we're currently performing load-balancing.
2431	* @imbalance: Variable to store the imbalance.
2432	*
2433	* Description:
2434	* Check if we have potential to perform some power-savings balance.
2435	* If yes, set the busiest group to be the least loaded group in the
2436	* sched_domain, so that it's CPUs can be put to idle.
2437	*
2438	* Returns 1 if there is potential to perform power-savings balance.
2439	* Else returns 0.
2440	*/
2441	static inline int check_power_save_busiest_group(struct sd_lb_stats *sds,
2442	int this_cpu, unsigned long *imbalance)
2443	{
2444	if (!sds->power_savings_balance)
2445	return 0;
2446
2447	if (sds->this != sds->group_leader \|\|
2448	sds->group_leader == sds->group_min)
2449	return 0;
2450
2451	*imbalance = sds->min_load_per_task;
2452	sds->busiest = sds->group_min;
2453
2454	return 1;
2455
2456	}
2457	#else /* CONFIG_SCHED_MC \|\| CONFIG_SCHED_SMT */
2458	static inline void init_sd_power_savings_stats(struct sched_domain *sd,
2459	struct sd_lb_stats *sds, enum cpu_idle_type idle)
2460	{
2461	return;
2462	}
2463
2464	static inline void update_sd_power_savings_stats(struct sched_group *group,
2465	struct sd_lb_stats sds, int local_group, struct sg_lb_stats sgs)
2466	{
2467	return;
2468	}
2469
2470	static inline int check_power_save_busiest_group(struct sd_lb_stats *sds,
2471	int this_cpu, unsigned long *imbalance)
2472	{
2473	return 0;
2474	}
2475	#endif /* CONFIG_SCHED_MC \|\| CONFIG_SCHED_SMT */
2476
2477
2478	unsigned long default_scale_freq_power(struct sched_domain *sd, int cpu)
2479	{
2480	return SCHED_LOAD_SCALE;
2481	}
2482
2483	unsigned long __weak arch_scale_freq_power(struct sched_domain *sd, int cpu)
2484	{
2485	return default_scale_freq_power(sd, cpu);
2486	}
2487
2488	unsigned long default_scale_smt_power(struct sched_domain *sd, int cpu)
2489	{
2490	unsigned long weight = sd->span_weight;
2491	unsigned long smt_gain = sd->smt_gain;
2492
2493	smt_gain /= weight;
2494
2495	return smt_gain;
2496	}
2497
2498	unsigned long __weak arch_scale_smt_power(struct sched_domain *sd, int cpu)
2499	{
2500	return default_scale_smt_power(sd, cpu);
2501	}
2502
2503	unsigned long scale_rt_power(int cpu)
2504	{
2505	struct rq *rq = cpu_rq(cpu);
2506	u64 total, available;
2507
2508	total = sched_avg_period() + (rq->clock - rq->age_stamp);
2509
2510	if (unlikely(total < rq->rt_avg)) {
2511	/* Ensures that power won't end up being negative */
2512	available = 0;
2513	} else {
2514	available = total - rq->rt_avg;
2515	}
2516
2517	if (unlikely((s64)total < SCHED_LOAD_SCALE))
2518	total = SCHED_LOAD_SCALE;
2519
2520	total >>= SCHED_LOAD_SHIFT;
2521
2522	return div_u64(available, total);
2523	}
2524
2525	static void update_cpu_power(struct sched_domain *sd, int cpu)
2526	{
2527	unsigned long weight = sd->span_weight;
2528	unsigned long power = SCHED_LOAD_SCALE;
2529	struct sched_group *sdg = sd->groups;
2530
2531	if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) {
2532	if (sched_feat(ARCH_POWER))
2533	power *= arch_scale_smt_power(sd, cpu);
2534	else
2535	power *= default_scale_smt_power(sd, cpu);
2536
2537	power >>= SCHED_LOAD_SHIFT;
2538	}
2539
2540	sdg->cpu_power_orig = power;
2541
2542	if (sched_feat(ARCH_POWER))
2543	power *= arch_scale_freq_power(sd, cpu);
2544	else
2545	power *= default_scale_freq_power(sd, cpu);
2546
2547	power >>= SCHED_LOAD_SHIFT;
2548
2549	power *= scale_rt_power(cpu);
2550	power >>= SCHED_LOAD_SHIFT;
2551
2552	if (!power)
2553	power = 1;
2554
2555	cpu_rq(cpu)->cpu_power = power;
2556	sdg->cpu_power = power;
2557	}
2558
2559	static void update_group_power(struct sched_domain *sd, int cpu)
2560	{
2561	struct sched_domain *child = sd->child;
2562	struct sched_group group, sdg = sd->groups;
2563	unsigned long power;
2564
2565	if (!child) {
2566	update_cpu_power(sd, cpu);
2567	return;
2568	}
2569
2570	power = 0;
2571
2572	group = child->groups;
2573	do {
2574	power += group->cpu_power;
2575	group = group->next;
2576	} while (group != child->groups);
2577
2578	sdg->cpu_power = power;
2579	}
2580
2581	/*
2582	* Try and fix up capacity for tiny siblings, this is needed when
2583	* things like SD_ASYM_PACKING need f_b_g to select another sibling
2584	* which on its own isn't powerful enough.
2585	*
2586	* See update_sd_pick_busiest() and check_asym_packing().
2587	*/
2588	static inline int
2589	fix_small_capacity(struct sched_domain sd, struct sched_group group)
2590	{
2591	/*
2592	* Only siblings can have significantly less than SCHED_LOAD_SCALE
2593	*/
2594	if (sd->level != SD_LV_SIBLING)
2595	return 0;
2596
2597	/*
2598	* If ~90% of the cpu_power is still there, we're good.
2599	*/
2600	if (group->cpu_power * 32 > group->cpu_power_orig * 29)
2601	return 1;
2602
2603	return 0;
2604	}
2605
2606	/**
2607	* update_sg_lb_stats - Update sched_group's statistics for load balancing.
2608	* @sd: The sched_domain whose statistics are to be updated.
2609	* @group: sched_group whose statistics are to be updated.
2610	* @this_cpu: Cpu for which load balance is currently performed.
2611	* @idle: Idle status of this_cpu
2612	* @load_idx: Load index of sched_domain of this_cpu for load calc.
2613	* @sd_idle: Idle status of the sched_domain containing group.
2614	* @local_group: Does group contain this_cpu.
2615	* @cpus: Set of cpus considered for load balancing.
2616	* @balance: Should we balance.
2617	* @sgs: variable to hold the statistics for this group.
2618	*/
2619	static inline void update_sg_lb_stats(struct sched_domain *sd,
2620	struct sched_group *group, int this_cpu,
2621	enum cpu_idle_type idle, int load_idx, int *sd_idle,
2622	int local_group, const struct cpumask *cpus,
2623	int balance, struct sg_lb_stats sgs)
2624	{
2625	unsigned long load, max_cpu_load, min_cpu_load, max_nr_running;
2626	int i;
2627	unsigned int balance_cpu = -1, first_idle_cpu = 0;
2628	unsigned long avg_load_per_task = 0;
2629
2630	if (local_group)
2631	balance_cpu = group_first_cpu(group);
2632
2633	/* Tally up the load of all CPUs in the group */
2634	max_cpu_load = 0;
2635	min_cpu_load = ~0UL;
2636	max_nr_running = 0;
2637
2638	for_each_cpu_and(i, sched_group_cpus(group), cpus) {
2639	struct rq *rq = cpu_rq(i);
2640
2641	if (*sd_idle && rq->nr_running)
2642	*sd_idle = 0;
2643
2644	/* Bias balancing toward cpus of our domain */
2645	if (local_group) {
2646	if (idle_cpu(i) && !first_idle_cpu) {
2647	first_idle_cpu = 1;
2648	balance_cpu = i;
2649	}
2650
2651	load = target_load(i, load_idx);
2652	} else {
2653	load = source_load(i, load_idx);
2654	if (load > max_cpu_load) {
2655	max_cpu_load = load;
2656	max_nr_running = rq->nr_running;
2657	}
2658	if (min_cpu_load > load)
2659	min_cpu_load = load;
2660	}
2661
2662	sgs->group_load += load;
2663	sgs->sum_nr_running += rq->nr_running;
2664	sgs->sum_weighted_load += weighted_cpuload(i);
2665	if (idle_cpu(i))
2666	sgs->idle_cpus++;
2667	}
2668
2669	/*
2670	* First idle cpu or the first cpu(busiest) in this sched group
2671	* is eligible for doing load balancing at this and above
2672	* domains. In the newly idle case, we will allow all the cpu's
2673	* to do the newly idle load balance.
2674	*/
2675	if (idle != CPU_NEWLY_IDLE && local_group) {
2676	if (balance_cpu != this_cpu) {
2677	*balance = 0;
2678	return;
2679	}
2680	update_group_power(sd, this_cpu);
2681	}
2682
2683	/* Adjust by relative CPU power of the group */
2684	sgs->avg_load = (sgs->group_load * SCHED_LOAD_SCALE) / group->cpu_power;
2685
2686	/*
2687	* Consider the group unbalanced when the imbalance is larger
2688	* than the average weight of two tasks.
2689	*
2690	* APZ: with cgroup the avg task weight can vary wildly and
2691	* might not be a suitable number - should we keep a
2692	* normalized nr_running number somewhere that negates
2693	* the hierarchy?
2694	*/
2695	if (sgs->sum_nr_running)
2696	avg_load_per_task = sgs->sum_weighted_load / sgs->sum_nr_running;
2697
2698	if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task && max_nr_running > 1)
2699	sgs->group_imb = 1;
2700
2701	sgs->group_capacity = DIV_ROUND_CLOSEST(group->cpu_power, SCHED_LOAD_SCALE);
2702	if (!sgs->group_capacity)
2703	sgs->group_capacity = fix_small_capacity(sd, group);
2704	sgs->group_weight = group->group_weight;
2705
2706	if (sgs->group_capacity > sgs->sum_nr_running)
2707	sgs->group_has_capacity = 1;
2708	}
2709
2710	/**
2711	* update_sd_pick_busiest - return 1 on busiest group
2712	* @sd: sched_domain whose statistics are to be checked
2713	* @sds: sched_domain statistics
2714	* @sg: sched_group candidate to be checked for being the busiest
2715	* @sgs: sched_group statistics
2716	* @this_cpu: the current cpu
2717	*
2718	* Determine if @sg is a busier group than the previously selected
2719	* busiest group.
2720	*/
2721	static bool update_sd_pick_busiest(struct sched_domain *sd,
2722	struct sd_lb_stats *sds,
2723	struct sched_group *sg,
2724	struct sg_lb_stats *sgs,
2725	int this_cpu)
2726	{
2727	if (sgs->avg_load <= sds->max_load)
2728	return false;
2729
2730	if (sgs->sum_nr_running > sgs->group_capacity)
2731	return true;
2732
2733	if (sgs->group_imb)
2734	return true;
2735
2736	/*
2737	* ASYM_PACKING needs to move all the work to the lowest
2738	* numbered CPUs in the group, therefore mark all groups
2739	* higher than ourself as busy.
2740	*/
2741	if ((sd->flags & SD_ASYM_PACKING) && sgs->sum_nr_running &&
2742	this_cpu < group_first_cpu(sg)) {
2743	if (!sds->busiest)
2744	return true;
2745
2746	if (group_first_cpu(sds->busiest) > group_first_cpu(sg))
2747	return true;
2748	}
2749
2750	return false;
2751	}
2752
2753	/**
2754	* update_sd_lb_stats - Update sched_group's statistics for load balancing.
2755	* @sd: sched_domain whose statistics are to be updated.
2756	* @this_cpu: Cpu for which load balance is currently performed.
2757	* @idle: Idle status of this_cpu
2758	* @sd_idle: Idle status of the sched_domain containing sg.
2759	* @cpus: Set of cpus considered for load balancing.
2760	* @balance: Should we balance.
2761	* @sds: variable to hold the statistics for this sched_domain.
2762	*/
2763	static inline void update_sd_lb_stats(struct sched_domain *sd, int this_cpu,
2764	enum cpu_idle_type idle, int *sd_idle,
2765	const struct cpumask cpus, int balance,
2766	struct sd_lb_stats *sds)
2767	{
2768	struct sched_domain *child = sd->child;
2769	struct sched_group *sg = sd->groups;
2770	struct sg_lb_stats sgs;
2771	int load_idx, prefer_sibling = 0;
2772
2773	if (child && child->flags & SD_PREFER_SIBLING)
2774	prefer_sibling = 1;
2775
2776	init_sd_power_savings_stats(sd, sds, idle);
2777	load_idx = get_sd_load_idx(sd, idle);
2778
2779	do {
2780	int local_group;
2781
2782	local_group = cpumask_test_cpu(this_cpu, sched_group_cpus(sg));
2783	memset(&sgs, 0, sizeof(sgs));
2784	update_sg_lb_stats(sd, sg, this_cpu, idle, load_idx, sd_idle,
2785	local_group, cpus, balance, &sgs);
2786
2787	if (local_group && !(*balance))
2788	return;
2789
2790	sds->total_load += sgs.group_load;
2791	sds->total_pwr += sg->cpu_power;
2792
2793	/*
2794	* In case the child domain prefers tasks go to siblings
2795	* first, lower the sg capacity to one so that we'll try
2796	* and move all the excess tasks away. We lower the capacity
2797	* of a group only if the local group has the capacity to fit
2798	* these excess tasks, i.e. nr_running < group_capacity. The
2799	* extra check prevents the case where you always pull from the
2800	* heaviest group when it is already under-utilized (possible
2801	* with a large weight task outweighs the tasks on the system).
2802	*/
2803	if (prefer_sibling && !local_group && sds->this_has_capacity)
2804	sgs.group_capacity = min(sgs.group_capacity, 1UL);
2805
2806	if (local_group) {
2807	sds->this_load = sgs.avg_load;
2808	sds->this = sg;
2809	sds->this_nr_running = sgs.sum_nr_running;
2810	sds->this_load_per_task = sgs.sum_weighted_load;
2811	sds->this_has_capacity = sgs.group_has_capacity;
2812	sds->this_idle_cpus = sgs.idle_cpus;
2813	} else if (update_sd_pick_busiest(sd, sds, sg, &sgs, this_cpu)) {
2814	sds->max_load = sgs.avg_load;
2815	sds->busiest = sg;
2816	sds->busiest_nr_running = sgs.sum_nr_running;
2817	sds->busiest_idle_cpus = sgs.idle_cpus;
2818	sds->busiest_group_capacity = sgs.group_capacity;
2819	sds->busiest_load_per_task = sgs.sum_weighted_load;
2820	sds->busiest_has_capacity = sgs.group_has_capacity;
2821	sds->busiest_group_weight = sgs.group_weight;
2822	sds->group_imb = sgs.group_imb;
2823	}
2824
2825	update_sd_power_savings_stats(sg, sds, local_group, &sgs);
2826	sg = sg->next;
2827	} while (sg != sd->groups);
2828	}
2829
2830	int __weak arch_sd_sibling_asym_packing(void)
2831	{
2832	return 0*SD_ASYM_PACKING;
2833	}
2834
2835	/**
2836	* check_asym_packing - Check to see if the group is packed into the
2837	* sched doman.
2838	*
2839	* This is primarily intended to used at the sibling level. Some
2840	* cores like POWER7 prefer to use lower numbered SMT threads. In the
2841	* case of POWER7, it can move to lower SMT modes only when higher
2842	* threads are idle. When in lower SMT modes, the threads will
2843	* perform better since they share less core resources. Hence when we
2844	* have idle threads, we want them to be the higher ones.
2845	*
2846	* This packing function is run on idle threads. It checks to see if
2847	* the busiest CPU in this domain (core in the P7 case) has a higher
2848	* CPU number than the packing function is being run on. Here we are
2849	* assuming lower CPU number will be equivalent to lower a SMT thread
2850	* number.
2851	*
2852	* Returns 1 when packing is required and a task should be moved to
2853	* this CPU. The amount of the imbalance is returned in *imbalance.
2854	*
2855	* @sd: The sched_domain whose packing is to be checked.
2856	* @sds: Statistics of the sched_domain which is to be packed
2857	* @this_cpu: The cpu at whose sched_domain we're performing load-balance.
2858	* @imbalance: returns amount of imbalanced due to packing.
2859	*/
2860	static int check_asym_packing(struct sched_domain *sd,
2861	struct sd_lb_stats *sds,
2862	int this_cpu, unsigned long *imbalance)
2863	{
2864	int busiest_cpu;
2865
2866	if (!(sd->flags & SD_ASYM_PACKING))
2867	return 0;
2868
2869	if (!sds->busiest)
2870	return 0;
2871
2872	busiest_cpu = group_first_cpu(sds->busiest);
2873	if (this_cpu > busiest_cpu)
2874	return 0;
2875
2876	imbalance = DIV_ROUND_CLOSEST(sds->max_load sds->busiest->cpu_power,
2877	SCHED_LOAD_SCALE);
2878	return 1;
2879	}
2880
2881	/**
2882	* fix_small_imbalance - Calculate the minor imbalance that exists
2883	* amongst the groups of a sched_domain, during
2884	* load balancing.
2885	* @sds: Statistics of the sched_domain whose imbalance is to be calculated.
2886	* @this_cpu: The cpu at whose sched_domain we're performing load-balance.
2887	* @imbalance: Variable to store the imbalance.
2888	*/
2889	static inline void fix_small_imbalance(struct sd_lb_stats *sds,
2890	int this_cpu, unsigned long *imbalance)
2891	{
2892	unsigned long tmp, pwr_now = 0, pwr_move = 0;
2893	unsigned int imbn = 2;
2894	unsigned long scaled_busy_load_per_task;
2895
2896	if (sds->this_nr_running) {
2897	sds->this_load_per_task /= sds->this_nr_running;
2898	if (sds->busiest_load_per_task >
2899	sds->this_load_per_task)
2900	imbn = 1;
2901	} else
2902	sds->this_load_per_task =
2903	cpu_avg_load_per_task(this_cpu);
2904
2905	scaled_busy_load_per_task = sds->busiest_load_per_task
2906	* SCHED_LOAD_SCALE;
2907	scaled_busy_load_per_task /= sds->busiest->cpu_power;
2908
2909	if (sds->max_load - sds->this_load + scaled_busy_load_per_task >=
2910	(scaled_busy_load_per_task * imbn)) {
2911	*imbalance = sds->busiest_load_per_task;
2912	return;
2913	}
2914
2915	/*
2916	* OK, we don't have enough imbalance to justify moving tasks,
2917	* however we may be able to increase total CPU power used by
2918	* moving them.
2919	*/
2920
2921	pwr_now += sds->busiest->cpu_power *
2922	min(sds->busiest_load_per_task, sds->max_load);
2923	pwr_now += sds->this->cpu_power *
2924	min(sds->this_load_per_task, sds->this_load);
2925	pwr_now /= SCHED_LOAD_SCALE;
2926
2927	/* Amount of load we'd subtract */
2928	tmp = (sds->busiest_load_per_task * SCHED_LOAD_SCALE) /
2929	sds->busiest->cpu_power;
2930	if (sds->max_load > tmp)
2931	pwr_move += sds->busiest->cpu_power *
2932	min(sds->busiest_load_per_task, sds->max_load - tmp);
2933
2934	/* Amount of load we'd add */
2935	if (sds->max_load * sds->busiest->cpu_power <
2936	sds->busiest_load_per_task * SCHED_LOAD_SCALE)
2937	tmp = (sds->max_load * sds->busiest->cpu_power) /
2938	sds->this->cpu_power;
2939	else
2940	tmp = (sds->busiest_load_per_task * SCHED_LOAD_SCALE) /
2941	sds->this->cpu_power;
2942	pwr_move += sds->this->cpu_power *
2943	min(sds->this_load_per_task, sds->this_load + tmp);
2944	pwr_move /= SCHED_LOAD_SCALE;
2945
2946	/* Move if we gain throughput */
2947	if (pwr_move > pwr_now)
2948	*imbalance = sds->busiest_load_per_task;
2949	}
2950
2951	/**
2952	* calculate_imbalance - Calculate the amount of imbalance present within the
2953	* groups of a given sched_domain during load balance.
2954	* @sds: statistics of the sched_domain whose imbalance is to be calculated.
2955	* @this_cpu: Cpu for which currently load balance is being performed.
2956	* @imbalance: The variable to store the imbalance.
2957	*/
2958	static inline void calculate_imbalance(struct sd_lb_stats *sds, int this_cpu,
2959	unsigned long *imbalance)
2960	{
2961	unsigned long max_pull, load_above_capacity = ~0UL;
2962
2963	sds->busiest_load_per_task /= sds->busiest_nr_running;
2964	if (sds->group_imb) {
2965	sds->busiest_load_per_task =
2966	min(sds->busiest_load_per_task, sds->avg_load);
2967	}
2968
2969	/*
2970	* In the presence of smp nice balancing, certain scenarios can have
2971	* max load less than avg load(as we skip the groups at or below
2972	* its cpu_power, while calculating max_load..)
2973	*/
2974	if (sds->max_load < sds->avg_load) {
2975	*imbalance = 0;
2976	return fix_small_imbalance(sds, this_cpu, imbalance);
2977	}
2978
2979	if (!sds->group_imb) {
2980	/*
2981	* Don't want to pull so many tasks that a group would go idle.
2982	*/
2983	load_above_capacity = (sds->busiest_nr_running -
2984	sds->busiest_group_capacity);
2985
2986	load_above_capacity = (SCHED_LOAD_SCALE SCHED_LOAD_SCALE);
2987
2988	load_above_capacity /= sds->busiest->cpu_power;
2989	}
2990
2991	/*
2992	* We're trying to get all the cpus to the average_load, so we don't
2993	* want to push ourselves above the average load, nor do we wish to
2994	* reduce the max loaded cpu below the average load. At the same time,
2995	* we also don't want to reduce the group load below the group capacity
2996	* (so that we can implement power-savings policies etc). Thus we look
2997	* for the minimum possible imbalance.
2998	* Be careful of negative numbers as they'll appear as very large values
2999	* with unsigned longs.
3000	*/
3001	max_pull = min(sds->max_load - sds->avg_load, load_above_capacity);
3002
3003	/* How much load to actually move to equalise the imbalance */
3004	imbalance = min(max_pull sds->busiest->cpu_power,
3005	(sds->avg_load - sds->this_load) * sds->this->cpu_power)
3006	/ SCHED_LOAD_SCALE;
3007
3008	/*
3009	* if *imbalance is less than the average load per runnable task
3010	* there is no gaurantee that any tasks will be moved so we'll have
3011	* a think about bumping its value to force at least one task to be
3012	* moved
3013	*/
3014	if (*imbalance < sds->busiest_load_per_task)
3015	return fix_small_imbalance(sds, this_cpu, imbalance);
3016
3017	}
3018
3019	/***** find_busiest_group() helpers end here *******************/
3020
3021	/**
3022	* find_busiest_group - Returns the busiest group within the sched_domain
3023	* if there is an imbalance. If there isn't an imbalance, and
3024	* the user has opted for power-savings, it returns a group whose
3025	* CPUs can be put to idle by rebalancing those tasks elsewhere, if
3026	* such a group exists.
3027	*
3028	* Also calculates the amount of weighted load which should be moved
3029	* to restore balance.
3030	*
3031	* @sd: The sched_domain whose busiest group is to be returned.
3032	* @this_cpu: The cpu for which load balancing is currently being performed.
3033	* @imbalance: Variable which stores amount of weighted load which should
3034	* be moved to restore balance/put a group to idle.
3035	* @idle: The idle status of this_cpu.
3036	* @sd_idle: The idleness of sd
3037	* @cpus: The set of CPUs under consideration for load-balancing.
3038	* @balance: Pointer to a variable indicating if this_cpu
3039	* is the appropriate cpu to perform load balancing at this_level.
3040	*
3041	* Returns: - the busiest group if imbalance exists.
3042	* - If no imbalance and user has opted for power-savings balance,
3043	* return the least loaded group whose CPUs can be
3044	* put to idle by rebalancing its tasks onto our group.
3045	*/
3046	static struct sched_group *
3047	find_busiest_group(struct sched_domain *sd, int this_cpu,
3048	unsigned long *imbalance, enum cpu_idle_type idle,
3049	int sd_idle, const struct cpumask cpus, int *balance)
3050	{
3051	struct sd_lb_stats sds;
3052
3053	memset(&sds, 0, sizeof(sds));
3054
3055	/*
3056	* Compute the various statistics relavent for load balancing at
3057	* this level.
3058	*/
3059	update_sd_lb_stats(sd, this_cpu, idle, sd_idle, cpus,
3060	balance, &sds);
3061
3062	/* Cases where imbalance does not exist from POV of this_cpu */
3063	/* 1) this_cpu is not the appropriate cpu to perform load balancing
3064	* at this level.
3065	* 2) There is no busy sibling group to pull from.
3066	* 3) This group is the busiest group.
3067	* 4) This group is more busy than the avg busieness at this
3068	* sched_domain.
3069	* 5) The imbalance is within the specified limit.
3070	*
3071	* Note: when doing newidle balance, if the local group has excess
3072	* capacity (i.e. nr_running < group_capacity) and the busiest group
3073	* does not have any capacity, we force a load balance to pull tasks
3074	* to the local group. In this case, we skip past checks 3, 4 and 5.
3075	*/
3076	if (!(*balance))
3077	goto ret;
3078
3079	if ((idle == CPU_IDLE \|\| idle == CPU_NEWLY_IDLE) &&
3080	check_asym_packing(sd, &sds, this_cpu, imbalance))
3081	return sds.busiest;
3082
3083	if (!sds.busiest \|\| sds.busiest_nr_running == 0)
3084	goto out_balanced;
3085
3086	/* SD_BALANCE_NEWIDLE trumps SMP nice when underutilized */
3087	if (idle == CPU_NEWLY_IDLE && sds.this_has_capacity &&
3088	!sds.busiest_has_capacity)
3089	goto force_balance;
3090
3091	if (sds.this_load >= sds.max_load)
3092	goto out_balanced;
3093
3094	sds.avg_load = (SCHED_LOAD_SCALE * sds.total_load) / sds.total_pwr;
3095
3096	if (sds.this_load >= sds.avg_load)
3097	goto out_balanced;
3098
3099	/*
3100	* In the CPU_NEWLY_IDLE, use imbalance_pct to be conservative.
3101	* And to check for busy balance use !idle_cpu instead of
3102	* CPU_NOT_IDLE. This is because HT siblings will use CPU_NOT_IDLE
3103	* even when they are idle.
3104	*/
3105	if (idle == CPU_NEWLY_IDLE \|\| !idle_cpu(this_cpu)) {
3106	if (100 * sds.max_load <= sd->imbalance_pct * sds.this_load)
3107	goto out_balanced;
3108	} else {
3109	/*
3110	* This cpu is idle. If the busiest group load doesn't
3111	* have more tasks than the number of available cpu's and
3112	* there is no imbalance between this and busiest group
3113	* wrt to idle cpu's, it is balanced.
3114	*/
3115	if ((sds.this_idle_cpus <= sds.busiest_idle_cpus + 1) &&
3116	sds.busiest_nr_running <= sds.busiest_group_weight)
3117	goto out_balanced;
3118	}
3119
3120	force_balance:
3121	/* Looks like there is an imbalance. Compute it */
3122	calculate_imbalance(&sds, this_cpu, imbalance);
3123	return sds.busiest;
3124
3125	out_balanced:
3126	/*
3127	* There is no obvious imbalance. But check if we can do some balancing
3128	* to save power.
3129	*/
3130	if (check_power_save_busiest_group(&sds, this_cpu, imbalance))
3131	return sds.busiest;
3132	ret:
3133	*imbalance = 0;
3134	return NULL;
3135	}
3136
3137	/*
3138	* find_busiest_queue - find the busiest runqueue among the cpus in group.
3139	*/
3140	static struct rq *
3141	find_busiest_queue(struct sched_domain sd, struct sched_group group,
3142	enum cpu_idle_type idle, unsigned long imbalance,
3143	const struct cpumask *cpus)
3144	{
3145	struct rq busiest = NULL, rq;
3146	unsigned long max_load = 0;
3147	int i;
3148
3149	for_each_cpu(i, sched_group_cpus(group)) {
3150	unsigned long power = power_of(i);
3151	unsigned long capacity = DIV_ROUND_CLOSEST(power, SCHED_LOAD_SCALE);
3152	unsigned long wl;
3153
3154	if (!capacity)
3155	capacity = fix_small_capacity(sd, group);
3156
3157	if (!cpumask_test_cpu(i, cpus))
3158	continue;
3159
3160	rq = cpu_rq(i);
3161	wl = weighted_cpuload(i);
3162
3163	/*
3164	* When comparing with imbalance, use weighted_cpuload()
3165	* which is not scaled with the cpu power.
3166	*/
3167	if (capacity && rq->nr_running == 1 && wl > imbalance)
3168	continue;
3169
3170	/*
3171	* For the load comparisons with the other cpu's, consider
3172	* the weighted_cpuload() scaled with the cpu power, so that
3173	* the load can be moved away from the cpu that is potentially
3174	* running at a lower capacity.
3175	*/
3176	wl = (wl * SCHED_LOAD_SCALE) / power;
3177
3178	if (wl > max_load) {
3179	max_load = wl;
3180	busiest = rq;
3181	}
3182	}
3183
3184	return busiest;
3185	}
3186
3187	/*
3188	* Max backoff if we encounter pinned tasks. Pretty arbitrary value, but
3189	* so long as it is large enough.
3190	*/
3191	#define MAX_PINNED_INTERVAL 512
3192
3193	/* Working cpumask for load_balance and load_balance_newidle. */
3194	static DEFINE_PER_CPU(cpumask_var_t, load_balance_tmpmask);
3195
3196	static int need_active_balance(struct sched_domain *sd, int sd_idle, int idle,
3197	int busiest_cpu, int this_cpu)
3198	{
3199	if (idle == CPU_NEWLY_IDLE) {
3200
3201	/*
3202	* ASYM_PACKING needs to force migrate tasks from busy but
3203	* higher numbered CPUs in order to pack all tasks in the
3204	* lowest numbered CPUs.
3205	*/
3206	if ((sd->flags & SD_ASYM_PACKING) && busiest_cpu > this_cpu)
3207	return 1;
3208
3209	/*
3210	* The only task running in a non-idle cpu can be moved to this
3211	* cpu in an attempt to completely freeup the other CPU
3212	* package.
3213	*
3214	* The package power saving logic comes from
3215	* find_busiest_group(). If there are no imbalance, then
3216	* f_b_g() will return NULL. However when sched_mc={1,2} then
3217	* f_b_g() will select a group from which a running task may be
3218	* pulled to this cpu in order to make the other package idle.
3219	* If there is no opportunity to make a package idle and if
3220	* there are no imbalance, then f_b_g() will return NULL and no
3221	* action will be taken in load_balance_newidle().
3222	*
3223	* Under normal task pull operation due to imbalance, there
3224	* will be more than one task in the source run queue and
3225	* move_tasks() will succeed. ld_moved will be true and this
3226	* active balance code will not be triggered.
3227	*/
3228	if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER &&
3229	!test_sd_parent(sd, SD_POWERSAVINGS_BALANCE))
3230	return 0;
3231
3232	if (sched_mc_power_savings < POWERSAVINGS_BALANCE_WAKEUP)
3233	return 0;
3234	}
3235
3236	return unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2);
3237	}
3238
3239	static int active_load_balance_cpu_stop(void *data);
3240
3241	/*
3242	* Check this_cpu to ensure it is balanced within domain. Attempt to move
3243	* tasks if there is an imbalance.
3244	*/
3245	static int load_balance(int this_cpu, struct rq *this_rq,
3246	struct sched_domain *sd, enum cpu_idle_type idle,
3247	int *balance)
3248	{
3249	int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0;
3250	struct sched_group *group;
3251	unsigned long imbalance;
3252	struct rq *busiest;
3253	unsigned long flags;
3254	struct cpumask *cpus = __get_cpu_var(load_balance_tmpmask);
3255
3256	cpumask_copy(cpus, cpu_active_mask);
3257
3258	/*
3259	* When power savings policy is enabled for the parent domain, idle
3260	* sibling can pick up load irrespective of busy siblings. In this case,
3261	* let the state of idle sibling percolate up as CPU_IDLE, instead of
3262	* portraying it as CPU_NOT_IDLE.
3263	*/
3264	if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER &&
3265	!test_sd_parent(sd, SD_POWERSAVINGS_BALANCE))
3266	sd_idle = 1;
3267
3268	schedstat_inc(sd, lb_count[idle]);
3269
3270	redo:
3271	group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle,
3272	cpus, balance);
3273
3274	if (*balance == 0)
3275	goto out_balanced;
3276
3277	if (!group) {
3278	schedstat_inc(sd, lb_nobusyg[idle]);
3279	goto out_balanced;
3280	}
3281
3282	busiest = find_busiest_queue(sd, group, idle, imbalance, cpus);
3283	if (!busiest) {
3284	schedstat_inc(sd, lb_nobusyq[idle]);
3285	goto out_balanced;
3286	}
3287
3288	BUG_ON(busiest == this_rq);
3289
3290	schedstat_add(sd, lb_imbalance[idle], imbalance);
3291
3292	ld_moved = 0;
3293	if (busiest->nr_running > 1) {
3294	/*
3295	* Attempt to move tasks. If find_busiest_group has found
3296	* an imbalance but busiest->nr_running <= 1, the group is
3297	* still unbalanced. ld_moved simply stays zero, so it is
3298	* correctly treated as an imbalance.
3299	*/
3300	local_irq_save(flags);
3301	double_rq_lock(this_rq, busiest);
3302	ld_moved = move_tasks(this_rq, this_cpu, busiest,
3303	imbalance, sd, idle, &all_pinned);
3304	double_rq_unlock(this_rq, busiest);
3305	local_irq_restore(flags);
3306
3307	/*
3308	* some other cpu did the load balance for us.
3309	*/
3310	if (ld_moved && this_cpu != smp_processor_id())
3311	resched_cpu(this_cpu);
3312
3313	/* All tasks on this runqueue were pinned by CPU affinity */
3314	if (unlikely(all_pinned)) {
3315	cpumask_clear_cpu(cpu_of(busiest), cpus);
3316	if (!cpumask_empty(cpus))
3317	goto redo;
3318	goto out_balanced;
3319	}
3320	}
3321
3322	if (!ld_moved) {
3323	schedstat_inc(sd, lb_failed[idle]);
3324	/*
3325	* Increment the failure counter only on periodic balance.
3326	* We do not want newidle balance, which can be very
3327	* frequent, pollute the failure counter causing
3328	* excessive cache_hot migrations and active balances.
3329	*/
3330	if (idle != CPU_NEWLY_IDLE)
3331	sd->nr_balance_failed++;
3332
3333	if (need_active_balance(sd, sd_idle, idle, cpu_of(busiest),
3334	this_cpu)) {
3335	raw_spin_lock_irqsave(&busiest->lock, flags);
3336
3337	/* don't kick the active_load_balance_cpu_stop,
3338	* if the curr task on busiest cpu can't be
3339	* moved to this_cpu
3340	*/
3341	if (!cpumask_test_cpu(this_cpu,
3342	&busiest->curr->cpus_allowed)) {
3343	raw_spin_unlock_irqrestore(&busiest->lock,
3344	flags);
3345	all_pinned = 1;
3346	goto out_one_pinned;
3347	}
3348
3349	/*
3350	* ->active_balance synchronizes accesses to
3351	* ->active_balance_work. Once set, it's cleared
3352	* only after active load balance is finished.
3353	*/
3354	if (!busiest->active_balance) {
3355	busiest->active_balance = 1;
3356	busiest->push_cpu = this_cpu;
3357	active_balance = 1;
3358	}
3359	raw_spin_unlock_irqrestore(&busiest->lock, flags);
3360
3361	if (active_balance)
3362	stop_one_cpu_nowait(cpu_of(busiest),
3363	active_load_balance_cpu_stop, busiest,
3364	&busiest->active_balance_work);
3365
3366	/*
3367	* We've kicked active balancing, reset the failure
3368	* counter.
3369	*/
3370	sd->nr_balance_failed = sd->cache_nice_tries+1;
3371	}
3372	} else
3373	sd->nr_balance_failed = 0;
3374
3375	if (likely(!active_balance)) {
3376	/* We were unbalanced, so reset the balancing interval */
3377	sd->balance_interval = sd->min_interval;
3378	} else {
3379	/*
3380	* If we've begun active balancing, start to back off. This
3381	* case may not be covered by the all_pinned logic if there
3382	* is only 1 task on the busy runqueue (because we don't call
3383	* move_tasks).
3384	*/
3385	if (sd->balance_interval < sd->max_interval)
3386	sd->balance_interval *= 2;
3387	}
3388
3389	if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER &&
3390	!test_sd_parent(sd, SD_POWERSAVINGS_BALANCE))
3391	ld_moved = -1;
3392
3393	goto out;
3394
3395	out_balanced:
3396	schedstat_inc(sd, lb_balanced[idle]);
3397
3398	sd->nr_balance_failed = 0;
3399
3400	out_one_pinned:
3401	/* tune up the balancing interval */
3402	if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) \|\|
3403	(sd->balance_interval < sd->max_interval))
3404	sd->balance_interval *= 2;
3405
3406	if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER &&
3407	!test_sd_parent(sd, SD_POWERSAVINGS_BALANCE))
3408	ld_moved = -1;
3409	else
3410	ld_moved = 0;
3411	out:
3412	return ld_moved;
3413	}
3414
3415	/*
3416	* idle_balance is called by schedule() if this_cpu is about to become
3417	* idle. Attempts to pull tasks from other CPUs.
3418	*/
3419	static void idle_balance(int this_cpu, struct rq *this_rq)
3420	{
3421	struct sched_domain *sd;
3422	int pulled_task = 0;
3423	unsigned long next_balance = jiffies + HZ;
3424
3425	this_rq->idle_stamp = this_rq->clock;
3426
3427	if (this_rq->avg_idle < sysctl_sched_migration_cost)
3428	return;
3429
3430	/*
3431	* Drop the rq->lock, but keep IRQ/preempt disabled.
3432	*/
3433	raw_spin_unlock(&this_rq->lock);
3434
3435	update_shares(this_cpu);
3436	for_each_domain(this_cpu, sd) {
3437	unsigned long interval;
3438	int balance = 1;
3439
3440	if (!(sd->flags & SD_LOAD_BALANCE))
3441	continue;
3442
3443	if (sd->flags & SD_BALANCE_NEWIDLE) {
3444	/* If we've pulled tasks over stop searching: */
3445	pulled_task = load_balance(this_cpu, this_rq,
3446	sd, CPU_NEWLY_IDLE, &balance);
3447	}
3448
3449	interval = msecs_to_jiffies(sd->balance_interval);
3450	if (time_after(next_balance, sd->last_balance + interval))
3451	next_balance = sd->last_balance + interval;
3452	if (pulled_task) {
3453	this_rq->idle_stamp = 0;
3454	break;
3455	}
3456	}
3457
3458	raw_spin_lock(&this_rq->lock);
3459
3460	if (pulled_task \|\| time_after(jiffies, this_rq->next_balance)) {
3461	/*
3462	* We are going idle. next_balance may be set based on
3463	* a busy processor. So reset next_balance.
3464	*/
3465	this_rq->next_balance = next_balance;
3466	}
3467	}
3468
3469	/*
3470	* active_load_balance_cpu_stop is run by cpu stopper. It pushes
3471	* running tasks off the busiest CPU onto idle CPUs. It requires at
3472	* least 1 task to be running on each physical CPU where possible, and
3473	* avoids physical / logical imbalances.
3474	*/
3475	static int active_load_balance_cpu_stop(void *data)
3476	{
3477	struct rq *busiest_rq = data;
3478	int busiest_cpu = cpu_of(busiest_rq);
3479	int target_cpu = busiest_rq->push_cpu;
3480	struct rq *target_rq = cpu_rq(target_cpu);
3481	struct sched_domain *sd;
3482
3483	raw_spin_lock_irq(&busiest_rq->lock);
3484
3485	/* make sure the requested cpu hasn't gone down in the meantime */
3486	if (unlikely(busiest_cpu != smp_processor_id() \|\|
3487	!busiest_rq->active_balance))
3488	goto out_unlock;
3489
3490	/* Is there any task to move? */
3491	if (busiest_rq->nr_running <= 1)
3492	goto out_unlock;
3493
3494	/*
3495	* This condition is "impossible", if it occurs
3496	* we need to fix it. Originally reported by
3497	* Bjorn Helgaas on a 128-cpu setup.
3498	*/
3499	BUG_ON(busiest_rq == target_rq);
3500
3501	/* move a task from busiest_rq to target_rq */
3502	double_lock_balance(busiest_rq, target_rq);
3503
3504	/* Search for an sd spanning us and the target CPU. */
3505	for_each_domain(target_cpu, sd) {
3506	if ((sd->flags & SD_LOAD_BALANCE) &&
3507	cpumask_test_cpu(busiest_cpu, sched_domain_span(sd)))
3508	break;
3509	}
3510
3511	if (likely(sd)) {
3512	schedstat_inc(sd, alb_count);
3513
3514	if (move_one_task(target_rq, target_cpu, busiest_rq,
3515	sd, CPU_IDLE))
3516	schedstat_inc(sd, alb_pushed);
3517	else
3518	schedstat_inc(sd, alb_failed);
3519	}
3520	double_unlock_balance(busiest_rq, target_rq);
3521	out_unlock:
3522	busiest_rq->active_balance = 0;
3523	raw_spin_unlock_irq(&busiest_rq->lock);
3524	return 0;
3525	}
3526
3527	#ifdef CONFIG_NO_HZ
3528
3529	static DEFINE_PER_CPU(struct call_single_data, remote_sched_softirq_cb);
3530
3531	static void trigger_sched_softirq(void *data)
3532	{
3533	raise_softirq_irqoff(SCHED_SOFTIRQ);
3534	}
3535
3536	static inline void init_sched_softirq_csd(struct call_single_data *csd)
3537	{
3538	csd->func = trigger_sched_softirq;
3539	csd->info = NULL;
3540	csd->flags = 0;
3541	csd->priv = 0;
3542	}
3543
3544	/*
3545	* idle load balancing details
3546	* - One of the idle CPUs nominates itself as idle load_balancer, while
3547	* entering idle.
3548	* - This idle load balancer CPU will also go into tickless mode when
3549	* it is idle, just like all other idle CPUs
3550	* - When one of the busy CPUs notice that there may be an idle rebalancing
3551	* needed, they will kick the idle load balancer, which then does idle
3552	* load balancing for all the idle CPUs.
3553	*/
3554	static struct {
3555	atomic_t load_balancer;
3556	atomic_t first_pick_cpu;
3557	atomic_t second_pick_cpu;
3558	cpumask_var_t idle_cpus_mask;
3559	cpumask_var_t grp_idle_mask;
3560	unsigned long next_balance; /* in jiffy units */
3561	} nohz ____cacheline_aligned;
3562
3563	int get_nohz_load_balancer(void)
3564	{
3565	return atomic_read(&nohz.load_balancer);
3566	}
3567
3568	#if defined(CONFIG_SCHED_MC) \|\| defined(CONFIG_SCHED_SMT)
3569	/**
3570	* lowest_flag_domain - Return lowest sched_domain containing flag.
3571	* @cpu: The cpu whose lowest level of sched domain is to
3572	* be returned.
3573	* @flag: The flag to check for the lowest sched_domain
3574	* for the given cpu.
3575	*
3576	* Returns the lowest sched_domain of a cpu which contains the given flag.
3577	*/
3578	static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
3579	{
3580	struct sched_domain *sd;
3581
3582	for_each_domain(cpu, sd)
3583	if (sd && (sd->flags & flag))
3584	break;
3585
3586	return sd;
3587	}
3588
3589	/**
3590	* for_each_flag_domain - Iterates over sched_domains containing the flag.
3591	* @cpu: The cpu whose domains we're iterating over.
3592	* @sd: variable holding the value of the power_savings_sd
3593	* for cpu.
3594	* @flag: The flag to filter the sched_domains to be iterated.
3595	*
3596	* Iterates over all the scheduler domains for a given cpu that has the 'flag'
3597	* set, starting from the lowest sched_domain to the highest.
3598	*/
3599	#define for_each_flag_domain(cpu, sd, flag) \
3600	for (sd = lowest_flag_domain(cpu, flag); \
3601	(sd && (sd->flags & flag)); sd = sd->parent)
3602
3603	/**
3604	* is_semi_idle_group - Checks if the given sched_group is semi-idle.
3605	* @ilb_group: group to be checked for semi-idleness
3606	*
3607	* Returns: 1 if the group is semi-idle. 0 otherwise.
3608	*
3609	* We define a sched_group to be semi idle if it has atleast one idle-CPU
3610	* and atleast one non-idle CPU. This helper function checks if the given
3611	* sched_group is semi-idle or not.
3612	*/
3613	static inline int is_semi_idle_group(struct sched_group *ilb_group)
3614	{
3615	cpumask_and(nohz.grp_idle_mask, nohz.idle_cpus_mask,
3616	sched_group_cpus(ilb_group));
3617
3618	/*
3619	* A sched_group is semi-idle when it has atleast one busy cpu
3620	* and atleast one idle cpu.
3621	*/
3622	if (cpumask_empty(nohz.grp_idle_mask))
3623	return 0;
3624
3625	if (cpumask_equal(nohz.grp_idle_mask, sched_group_cpus(ilb_group)))
3626	return 0;
3627
3628	return 1;
3629	}
3630	/**
3631	* find_new_ilb - Finds the optimum idle load balancer for nomination.
3632	* @cpu: The cpu which is nominating a new idle_load_balancer.
3633	*
3634	* Returns: Returns the id of the idle load balancer if it exists,
3635	* Else, returns >= nr_cpu_ids.
3636	*
3637	* This algorithm picks the idle load balancer such that it belongs to a
3638	* semi-idle powersavings sched_domain. The idea is to try and avoid
3639	* completely idle packages/cores just for the purpose of idle load balancing
3640	* when there are other idle cpu's which are better suited for that job.
3641	*/
3642	static int find_new_ilb(int cpu)
3643	{
3644	struct sched_domain *sd;
3645	struct sched_group *ilb_group;
3646
3647	/*
3648	* Have idle load balancer selection from semi-idle packages only
3649	* when power-aware load balancing is enabled
3650	*/
3651	if (!(sched_smt_power_savings \|\| sched_mc_power_savings))
3652	goto out_done;
3653
3654	/*
3655	* Optimize for the case when we have no idle CPUs or only one
3656	* idle CPU. Don't walk the sched_domain hierarchy in such cases
3657	*/
3658	if (cpumask_weight(nohz.idle_cpus_mask) < 2)
3659	goto out_done;
3660
3661	for_each_flag_domain(cpu, sd, SD_POWERSAVINGS_BALANCE) {
3662	ilb_group = sd->groups;
3663
3664	do {
3665	if (is_semi_idle_group(ilb_group))
3666	return cpumask_first(nohz.grp_idle_mask);
3667
3668	ilb_group = ilb_group->next;
3669
3670	} while (ilb_group != sd->groups);
3671	}
3672
3673	out_done:
3674	return nr_cpu_ids;
3675	}
3676	#else /* (CONFIG_SCHED_MC \|\| CONFIG_SCHED_SMT) */
3677	static inline int find_new_ilb(int call_cpu)
3678	{
3679	return nr_cpu_ids;
3680	}
3681	#endif
3682
3683	/*
3684	* Kick a CPU to do the nohz balancing, if it is time for it. We pick the
3685	* nohz_load_balancer CPU (if there is one) otherwise fallback to any idle
3686	* CPU (if there is one).
3687	*/
3688	static void nohz_balancer_kick(int cpu)
3689	{
3690	int ilb_cpu;
3691
3692	nohz.next_balance++;
3693
3694	ilb_cpu = get_nohz_load_balancer();
3695
3696	if (ilb_cpu >= nr_cpu_ids) {
3697	ilb_cpu = cpumask_first(nohz.idle_cpus_mask);
3698	if (ilb_cpu >= nr_cpu_ids)
3699	return;
3700	}
3701
3702	if (!cpu_rq(ilb_cpu)->nohz_balance_kick) {
3703	struct call_single_data *cp;
3704
3705	cpu_rq(ilb_cpu)->nohz_balance_kick = 1;
3706	cp = &per_cpu(remote_sched_softirq_cb, cpu);
3707	__smp_call_function_single(ilb_cpu, cp, 0);
3708	}
3709	return;
3710	}
3711
3712	/*
3713	* This routine will try to nominate the ilb (idle load balancing)
3714	* owner among the cpus whose ticks are stopped. ilb owner will do the idle
3715	* load balancing on behalf of all those cpus.
3716	*
3717	* When the ilb owner becomes busy, we will not have new ilb owner until some
3718	* idle CPU wakes up and goes back to idle or some busy CPU tries to kick
3719	* idle load balancing by kicking one of the idle CPUs.
3720	*
3721	* Ticks are stopped for the ilb owner as well, with busy CPU kicking this
3722	* ilb owner CPU in future (when there is a need for idle load balancing on
3723	* behalf of all idle CPUs).
3724	*/
3725	void select_nohz_load_balancer(int stop_tick)
3726	{
3727	int cpu = smp_processor_id();
3728
3729	if (stop_tick) {
3730	if (!cpu_active(cpu)) {
3731	if (atomic_read(&nohz.load_balancer) != cpu)
3732	return;
3733
3734	/*
3735	* If we are going offline and still the leader,
3736	* give up!
3737	*/
3738	if (atomic_cmpxchg(&nohz.load_balancer, cpu,
3739	nr_cpu_ids) != cpu)
3740	BUG();
3741
3742	return;
3743	}
3744
3745	cpumask_set_cpu(cpu, nohz.idle_cpus_mask);
3746
3747	if (atomic_read(&nohz.first_pick_cpu) == cpu)
3748	atomic_cmpxchg(&nohz.first_pick_cpu, cpu, nr_cpu_ids);
3749	if (atomic_read(&nohz.second_pick_cpu) == cpu)
3750	atomic_cmpxchg(&nohz.second_pick_cpu, cpu, nr_cpu_ids);
3751
3752	if (atomic_read(&nohz.load_balancer) >= nr_cpu_ids) {
3753	int new_ilb;
3754
3755	/* make me the ilb owner */
3756	if (atomic_cmpxchg(&nohz.load_balancer, nr_cpu_ids,
3757	cpu) != nr_cpu_ids)
3758	return;
3759
3760	/*
3761	* Check to see if there is a more power-efficient
3762	* ilb.
3763	*/
3764	new_ilb = find_new_ilb(cpu);
3765	if (new_ilb < nr_cpu_ids && new_ilb != cpu) {
3766	atomic_set(&nohz.load_balancer, nr_cpu_ids);
3767	resched_cpu(new_ilb);
3768	return;
3769	}
3770	return;
3771	}
3772	} else {
3773	if (!cpumask_test_cpu(cpu, nohz.idle_cpus_mask))
3774	return;
3775
3776	cpumask_clear_cpu(cpu, nohz.idle_cpus_mask);
3777
3778	if (atomic_read(&nohz.load_balancer) == cpu)
3779	if (atomic_cmpxchg(&nohz.load_balancer, cpu,
3780	nr_cpu_ids) != cpu)
3781	BUG();
3782	}
3783	return;
3784	}
3785	#endif
3786
3787	static DEFINE_SPINLOCK(balancing);
3788
3789	/*
3790	* It checks each scheduling domain to see if it is due to be balanced,
3791	* and initiates a balancing operation if so.
3792	*
3793	* Balancing parameters are set up in arch_init_sched_domains.
3794	*/
3795	static void rebalance_domains(int cpu, enum cpu_idle_type idle)
3796	{
3797	int balance = 1;
3798	struct rq *rq = cpu_rq(cpu);
3799	unsigned long interval;
3800	struct sched_domain *sd;
3801	/* Earliest time when we have to do rebalance again */
3802	unsigned long next_balance = jiffies + 60*HZ;
3803	int update_next_balance = 0;
3804	int need_serialize;
3805
3806	update_shares(cpu);
3807
3808	for_each_domain(cpu, sd) {
3809	if (!(sd->flags & SD_LOAD_BALANCE))
3810	continue;
3811
3812	interval = sd->balance_interval;
3813	if (idle != CPU_IDLE)
3814	interval *= sd->busy_factor;
3815
3816	/* scale ms to jiffies */
3817	interval = msecs_to_jiffies(interval);
3818	if (unlikely(!interval))
3819	interval = 1;
3820	if (interval > HZ*NR_CPUS/10)
3821	interval = HZ*NR_CPUS/10;
3822
3823	need_serialize = sd->flags & SD_SERIALIZE;
3824
3825	if (need_serialize) {
3826	if (!spin_trylock(&balancing))
3827	goto out;
3828	}
3829
3830	if (time_after_eq(jiffies, sd->last_balance + interval)) {
3831	if (load_balance(cpu, rq, sd, idle, &balance)) {
3832	/*
3833	* We've pulled tasks over so either we're no
3834	* longer idle, or one of our SMT siblings is
3835	* not idle.
3836	*/
3837	idle = CPU_NOT_IDLE;
3838	}
3839	sd->last_balance = jiffies;
3840	}
3841	if (need_serialize)
3842	spin_unlock(&balancing);
3843	out:
3844	if (time_after(next_balance, sd->last_balance + interval)) {
3845	next_balance = sd->last_balance + interval;
3846	update_next_balance = 1;
3847	}
3848
3849	/*
3850	* Stop the load balance at this level. There is another
3851	* CPU in our sched group which is doing load balancing more
3852	* actively.
3853	*/
3854	if (!balance)
3855	break;
3856	}
3857
3858	/*
3859	* next_balance will be updated only when there is a need.
3860	* When the cpu is attached to null domain for ex, it will not be
3861	* updated.
3862	*/
3863	if (likely(update_next_balance))
3864	rq->next_balance = next_balance;
3865	}
3866
3867	#ifdef CONFIG_NO_HZ
3868	/*
3869	* In CONFIG_NO_HZ case, the idle balance kickee will do the
3870	* rebalancing for all the cpus for whom scheduler ticks are stopped.
3871	*/
3872	static void nohz_idle_balance(int this_cpu, enum cpu_idle_type idle)
3873	{
3874	struct rq *this_rq = cpu_rq(this_cpu);
3875	struct rq *rq;
3876	int balance_cpu;
3877
3878	if (idle != CPU_IDLE \|\| !this_rq->nohz_balance_kick)
3879	return;
3880
3881	for_each_cpu(balance_cpu, nohz.idle_cpus_mask) {
3882	if (balance_cpu == this_cpu)
3883	continue;
3884
3885	/*
3886	* If this cpu gets work to do, stop the load balancing
3887	* work being done for other cpus. Next load
3888	* balancing owner will pick it up.
3889	*/
3890	if (need_resched()) {
3891	this_rq->nohz_balance_kick = 0;
3892	break;
3893	}
3894
3895	raw_spin_lock_irq(&this_rq->lock);
3896	update_rq_clock(this_rq);
3897	update_cpu_load(this_rq);
3898	raw_spin_unlock_irq(&this_rq->lock);
3899
3900	rebalance_domains(balance_cpu, CPU_IDLE);
3901
3902	rq = cpu_rq(balance_cpu);
3903	if (time_after(this_rq->next_balance, rq->next_balance))
3904	this_rq->next_balance = rq->next_balance;
3905	}
3906	nohz.next_balance = this_rq->next_balance;
3907	this_rq->nohz_balance_kick = 0;
3908	}
3909
3910	/*
3911	* Current heuristic for kicking the idle load balancer
3912	* - first_pick_cpu is the one of the busy CPUs. It will kick
3913	* idle load balancer when it has more than one process active. This
3914	* eliminates the need for idle load balancing altogether when we have
3915	* only one running process in the system (common case).
3916	* - If there are more than one busy CPU, idle load balancer may have
3917	* to run for active_load_balance to happen (i.e., two busy CPUs are
3918	* SMT or core siblings and can run better if they move to different
3919	* physical CPUs). So, second_pick_cpu is the second of the busy CPUs
3920	* which will kick idle load balancer as soon as it has any load.
3921	*/
3922	static inline int nohz_kick_needed(struct rq *rq, int cpu)
3923	{
3924	unsigned long now = jiffies;
3925	int ret;
3926	int first_pick_cpu, second_pick_cpu;
3927
3928	if (time_before(now, nohz.next_balance))
3929	return 0;
3930
3931	if (rq->idle_at_tick)
3932	return 0;
3933
3934	first_pick_cpu = atomic_read(&nohz.first_pick_cpu);
3935	second_pick_cpu = atomic_read(&nohz.second_pick_cpu);
3936
3937	if (first_pick_cpu < nr_cpu_ids && first_pick_cpu != cpu &&
3938	second_pick_cpu < nr_cpu_ids && second_pick_cpu != cpu)
3939	return 0;
3940
3941	ret = atomic_cmpxchg(&nohz.first_pick_cpu, nr_cpu_ids, cpu);
3942	if (ret == nr_cpu_ids \|\| ret == cpu) {
3943	atomic_cmpxchg(&nohz.second_pick_cpu, cpu, nr_cpu_ids);
3944	if (rq->nr_running > 1)
3945	return 1;
3946	} else {
3947	ret = atomic_cmpxchg(&nohz.second_pick_cpu, nr_cpu_ids, cpu);
3948	if (ret == nr_cpu_ids \|\| ret == cpu) {
3949	if (rq->nr_running)
3950	return 1;
3951	}
3952	}
3953	return 0;
3954	}
3955	#else
3956	static void nohz_idle_balance(int this_cpu, enum cpu_idle_type idle) { }
3957	#endif
3958
3959	/*
3960	* run_rebalance_domains is triggered when needed from the scheduler tick.
3961	* Also triggered for nohz idle balancing (with nohz_balancing_kick set).
3962	*/
3963	static void run_rebalance_domains(struct softirq_action *h)
3964	{
3965	int this_cpu = smp_processor_id();
3966	struct rq *this_rq = cpu_rq(this_cpu);
3967	enum cpu_idle_type idle = this_rq->idle_at_tick ?
3968	CPU_IDLE : CPU_NOT_IDLE;
3969
3970	rebalance_domains(this_cpu, idle);
3971
3972	/*
3973	* If this cpu has a pending nohz_balance_kick, then do the
3974	* balancing on behalf of the other idle cpus whose ticks are
3975	* stopped.
3976	*/
3977	nohz_idle_balance(this_cpu, idle);
3978	}
3979
3980	static inline int on_null_domain(int cpu)
3981	{
3982	return !rcu_dereference_sched(cpu_rq(cpu)->sd);
3983	}
3984
3985	/*
3986	* Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing.
3987	*/
3988	static inline void trigger_load_balance(struct rq *rq, int cpu)
3989	{
3990	/* Don't need to rebalance while attached to NULL domain */
3991	if (time_after_eq(jiffies, rq->next_balance) &&
3992	likely(!on_null_domain(cpu)))
3993	raise_softirq(SCHED_SOFTIRQ);
3994	#ifdef CONFIG_NO_HZ
3995	else if (nohz_kick_needed(rq, cpu) && likely(!on_null_domain(cpu)))
3996	nohz_balancer_kick(cpu);
3997	#endif
3998	}
3999
4000	static void rq_online_fair(struct rq *rq)
4001	{
4002	update_sysctl();
4003	}
4004
4005	static void rq_offline_fair(struct rq *rq)
4006	{
4007	update_sysctl();
4008	}
4009
4010	#else /* CONFIG_SMP */
4011
4012	/*
4013	* on UP we do not need to balance between CPUs:
4014	*/
4015	static inline void idle_balance(int cpu, struct rq *rq)
4016	{
4017	}
4018
4019	#endif /* CONFIG_SMP */
4020
4021	/*
4022	* scheduler tick hitting a task of our scheduling class:
4023	*/
4024	static void task_tick_fair(struct rq rq, struct task_struct curr, int queued)
4025	{
4026	struct cfs_rq *cfs_rq;
4027	struct sched_entity *se = &curr->se;
4028
4029	for_each_sched_entity(se) {
4030	cfs_rq = cfs_rq_of(se);
4031	entity_tick(cfs_rq, se, queued);
4032	}
4033	}
4034
4035	/*
4036	* called on fork with the child task as argument from the parent's context
4037	* - child not yet on the tasklist
4038	* - preemption disabled
4039	*/
4040	static void task_fork_fair(struct task_struct *p)
4041	{
4042	struct cfs_rq *cfs_rq = task_cfs_rq(current);
4043	struct sched_entity se = &p->se, curr = cfs_rq->curr;
4044	int this_cpu = smp_processor_id();
4045	struct rq *rq = this_rq();
4046	unsigned long flags;
4047
4048	raw_spin_lock_irqsave(&rq->lock, flags);
4049
4050	update_rq_clock(rq);
4051
4052	if (unlikely(task_cpu(p) != this_cpu)) {
4053	rcu_read_lock();
4054	__set_task_cpu(p, this_cpu);
4055	rcu_read_unlock();
4056	}
4057
4058	update_curr(cfs_rq);
4059
4060	if (curr)
4061	se->vruntime = curr->vruntime;
4062	place_entity(cfs_rq, se, 1);
4063
4064	if (sysctl_sched_child_runs_first && curr && entity_before(curr, se)) {
4065	/*
4066	* Upon rescheduling, sched_class::put_prev_task() will place
4067	* 'current' within the tree based on its new key value.
4068	*/
4069	swap(curr->vruntime, se->vruntime);
4070	resched_task(rq->curr);
4071	}
4072
4073	se->vruntime -= cfs_rq->min_vruntime;
4074
4075	raw_spin_unlock_irqrestore(&rq->lock, flags);
4076	}
4077
4078	/*
4079	* Priority of the task has changed. Check to see if we preempt
4080	* the current task.
4081	*/
4082	static void prio_changed_fair(struct rq rq, struct task_struct p,
4083	int oldprio, int running)
4084	{
4085	/*
4086	* Reschedule if we are currently running on this runqueue and
4087	* our priority decreased, or if we are not currently running on
4088	* this runqueue and our priority is higher than the current's
4089	*/
4090	if (running) {
4091	if (p->prio > oldprio)
4092	resched_task(rq->curr);
4093	} else
4094	check_preempt_curr(rq, p, 0);
4095	}
4096
4097	/*
4098	* We switched to the sched_fair class.
4099	*/
4100	static void switched_to_fair(struct rq rq, struct task_struct p,
4101	int running)
4102	{
4103	/*
4104	* We were most likely switched from sched_rt, so
4105	* kick off the schedule if running, otherwise just see
4106	* if we can still preempt the current task.
4107	*/
4108	if (running)
4109	resched_task(rq->curr);
4110	else
4111	check_preempt_curr(rq, p, 0);
4112	}
4113
4114	/* Account for a task changing its policy or group.
4115	*
4116	* This routine is mostly called to set cfs_rq->curr field when a task
4117	* migrates between groups/classes.
4118	*/
4119	static void set_curr_task_fair(struct rq *rq)
4120	{
4121	struct sched_entity *se = &rq->curr->se;
4122
4123	for_each_sched_entity(se)
4124	set_next_entity(cfs_rq_of(se), se);
4125	}
4126
4127	#ifdef CONFIG_FAIR_GROUP_SCHED
4128	static void task_move_group_fair(struct task_struct *p, int on_rq)
4129	{
4130	/*
4131	* If the task was not on the rq at the time of this cgroup movement
4132	* it must have been asleep, sleeping tasks keep their ->vruntime
4133	* absolute on their old rq until wakeup (needed for the fair sleeper
4134	* bonus in place_entity()).
4135	*
4136	* If it was on the rq, we've just 'preempted' it, which does convert
4137	* ->vruntime to a relative base.
4138	*
4139	* Make sure both cases convert their relative position when migrating
4140	* to another cgroup's rq. This does somewhat interfere with the
4141	* fair sleeper stuff for the first placement, but who cares.
4142	*/
4143	if (!on_rq)
4144	p->se.vruntime -= cfs_rq_of(&p->se)->min_vruntime;
4145	set_task_rq(p, task_cpu(p));
4146	if (!on_rq)
4147	p->se.vruntime += cfs_rq_of(&p->se)->min_vruntime;
4148	}
4149	#endif
4150
4151	static unsigned int get_rr_interval_fair(struct rq rq, struct task_struct task)
4152	{
4153	struct sched_entity *se = &task->se;
4154	unsigned int rr_interval = 0;
4155
4156	/*
4157	* Time slice is 0 for SCHED_OTHER tasks that are on an otherwise
4158	* idle runqueue:
4159	*/
4160	if (rq->cfs.load.weight)
4161	rr_interval = NS_TO_JIFFIES(sched_slice(&rq->cfs, se));
4162
4163	return rr_interval;
4164	}
4165
4166	/*
4167	* All the scheduling class methods:
4168	*/
4169	static const struct sched_class fair_sched_class = {
4170	.next = &idle_sched_class,
4171	.enqueue_task = enqueue_task_fair,
4172	.dequeue_task = dequeue_task_fair,
4173	.yield_task = yield_task_fair,
4174
4175	.check_preempt_curr = check_preempt_wakeup,
4176
4177	.pick_next_task = pick_next_task_fair,
4178	.put_prev_task = put_prev_task_fair,
4179
4180	#ifdef CONFIG_SMP
4181	.select_task_rq = select_task_rq_fair,
4182
4183	.rq_online = rq_online_fair,
4184	.rq_offline = rq_offline_fair,
4185
4186	.task_waking = task_waking_fair,
4187	#endif
4188
4189	.set_curr_task = set_curr_task_fair,
4190	.task_tick = task_tick_fair,
4191	.task_fork = task_fork_fair,
4192
4193	.prio_changed = prio_changed_fair,
4194	.switched_to = switched_to_fair,
4195
4196	.get_rr_interval = get_rr_interval_fair,
4197
4198	#ifdef CONFIG_FAIR_GROUP_SCHED
4199	.task_move_group = task_move_group_fair,
4200	#endif
4201	};
4202
4203	#ifdef CONFIG_SCHED_DEBUG
4204	static void print_cfs_stats(struct seq_file *m, int cpu)
4205	{
4206	struct cfs_rq *cfs_rq;
4207
4208	rcu_read_lock();
4209	for_each_leaf_cfs_rq(cpu_rq(cpu), cfs_rq)
4210	print_cfs_rq(m, cpu, cfs_rq);
4211	rcu_read_unlock();
4212	}
4213	#endif
4214

Download this file

Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master

Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9

Kernel

Kernel Git Source Tree

Root/kernel/sched_fair.c

interactive