blob: bf3e38fdbe6dc6ca0160d7f408881525928da21a [file] [log] [blame]
Ingo Molnarbb44e5d2007-07-09 18:51:58 +02001/*
2 * Real-Time Scheduling Class (mapped to the SCHED_FIFO and SCHED_RR
3 * policies)
4 */
5
Gregory Haskins398a1532009-01-14 09:10:04 -05006#ifdef CONFIG_RT_GROUP_SCHED
7
Peter Zijlstraa1ba4d82009-04-01 18:40:15 +02008#define rt_entity_is_task(rt_se) (!(rt_se)->my_q)
9
Peter Zijlstra8f488942009-07-24 12:25:30 +020010static inline struct task_struct *rt_task_of(struct sched_rt_entity *rt_se)
11{
12#ifdef CONFIG_SCHED_DEBUG
13 WARN_ON_ONCE(!rt_entity_is_task(rt_se));
14#endif
15 return container_of(rt_se, struct task_struct, rt);
16}
17
Gregory Haskins398a1532009-01-14 09:10:04 -050018static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq)
19{
20 return rt_rq->rq;
21}
22
23static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se)
24{
25 return rt_se->rt_rq;
26}
27
28#else /* CONFIG_RT_GROUP_SCHED */
29
Peter Zijlstraa1ba4d82009-04-01 18:40:15 +020030#define rt_entity_is_task(rt_se) (1)
31
Peter Zijlstra8f488942009-07-24 12:25:30 +020032static inline struct task_struct *rt_task_of(struct sched_rt_entity *rt_se)
33{
34 return container_of(rt_se, struct task_struct, rt);
35}
36
Gregory Haskins398a1532009-01-14 09:10:04 -050037static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq)
38{
39 return container_of(rt_rq, struct rq, rt);
40}
41
42static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se)
43{
44 struct task_struct *p = rt_task_of(rt_se);
45 struct rq *rq = task_rq(p);
46
47 return &rq->rt;
48}
49
50#endif /* CONFIG_RT_GROUP_SCHED */
51
Steven Rostedt4fd29172008-01-25 21:08:06 +010052#ifdef CONFIG_SMP
Ingo Molnar84de4272008-01-25 21:08:15 +010053
Gregory Haskins637f5082008-01-25 21:08:18 +010054static inline int rt_overloaded(struct rq *rq)
Steven Rostedt4fd29172008-01-25 21:08:06 +010055{
Gregory Haskins637f5082008-01-25 21:08:18 +010056 return atomic_read(&rq->rd->rto_count);
Steven Rostedt4fd29172008-01-25 21:08:06 +010057}
Ingo Molnar84de4272008-01-25 21:08:15 +010058
Steven Rostedt4fd29172008-01-25 21:08:06 +010059static inline void rt_set_overload(struct rq *rq)
60{
Gregory Haskins1f11eb6a2008-06-04 15:04:05 -040061 if (!rq->online)
62 return;
63
Rusty Russellc6c49272008-11-25 02:35:05 +103064 cpumask_set_cpu(rq->cpu, rq->rd->rto_mask);
Steven Rostedt4fd29172008-01-25 21:08:06 +010065 /*
66 * Make sure the mask is visible before we set
67 * the overload count. That is checked to determine
68 * if we should look at the mask. It would be a shame
69 * if we looked at the mask, but the mask was not
70 * updated yet.
71 */
72 wmb();
Gregory Haskins637f5082008-01-25 21:08:18 +010073 atomic_inc(&rq->rd->rto_count);
Steven Rostedt4fd29172008-01-25 21:08:06 +010074}
Ingo Molnar84de4272008-01-25 21:08:15 +010075
Steven Rostedt4fd29172008-01-25 21:08:06 +010076static inline void rt_clear_overload(struct rq *rq)
77{
Gregory Haskins1f11eb6a2008-06-04 15:04:05 -040078 if (!rq->online)
79 return;
80
Steven Rostedt4fd29172008-01-25 21:08:06 +010081 /* the order here really doesn't matter */
Gregory Haskins637f5082008-01-25 21:08:18 +010082 atomic_dec(&rq->rd->rto_count);
Rusty Russellc6c49272008-11-25 02:35:05 +103083 cpumask_clear_cpu(rq->cpu, rq->rd->rto_mask);
Steven Rostedt4fd29172008-01-25 21:08:06 +010084}
Gregory Haskins73fe6aae2008-01-25 21:08:07 +010085
Gregory Haskins398a1532009-01-14 09:10:04 -050086static void update_rt_migration(struct rt_rq *rt_rq)
Gregory Haskins73fe6aae2008-01-25 21:08:07 +010087{
Peter Zijlstraa1ba4d82009-04-01 18:40:15 +020088 if (rt_rq->rt_nr_migratory && rt_rq->rt_nr_total > 1) {
Gregory Haskins398a1532009-01-14 09:10:04 -050089 if (!rt_rq->overloaded) {
90 rt_set_overload(rq_of_rt_rq(rt_rq));
91 rt_rq->overloaded = 1;
Gregory Haskinscdc8eb92008-01-25 21:08:23 +010092 }
Gregory Haskins398a1532009-01-14 09:10:04 -050093 } else if (rt_rq->overloaded) {
94 rt_clear_overload(rq_of_rt_rq(rt_rq));
95 rt_rq->overloaded = 0;
Gregory Haskins637f5082008-01-25 21:08:18 +010096 }
Gregory Haskins73fe6aae2008-01-25 21:08:07 +010097}
Steven Rostedt4fd29172008-01-25 21:08:06 +010098
Gregory Haskins398a1532009-01-14 09:10:04 -050099static void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100100{
Peter Zijlstraa1ba4d82009-04-01 18:40:15 +0200101 if (!rt_entity_is_task(rt_se))
102 return;
103
104 rt_rq = &rq_of_rt_rq(rt_rq)->rt;
105
106 rt_rq->rt_nr_total++;
Gregory Haskins398a1532009-01-14 09:10:04 -0500107 if (rt_se->nr_cpus_allowed > 1)
108 rt_rq->rt_nr_migratory++;
109
110 update_rt_migration(rt_rq);
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100111}
112
Gregory Haskins398a1532009-01-14 09:10:04 -0500113static void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
114{
Peter Zijlstraa1ba4d82009-04-01 18:40:15 +0200115 if (!rt_entity_is_task(rt_se))
116 return;
117
118 rt_rq = &rq_of_rt_rq(rt_rq)->rt;
119
120 rt_rq->rt_nr_total--;
Gregory Haskins398a1532009-01-14 09:10:04 -0500121 if (rt_se->nr_cpus_allowed > 1)
122 rt_rq->rt_nr_migratory--;
123
124 update_rt_migration(rt_rq);
125}
126
Gregory Haskins917b6272008-12-29 09:39:53 -0500127static void enqueue_pushable_task(struct rq *rq, struct task_struct *p)
128{
129 plist_del(&p->pushable_tasks, &rq->rt.pushable_tasks);
130 plist_node_init(&p->pushable_tasks, p->prio);
131 plist_add(&p->pushable_tasks, &rq->rt.pushable_tasks);
132}
133
134static void dequeue_pushable_task(struct rq *rq, struct task_struct *p)
135{
136 plist_del(&p->pushable_tasks, &rq->rt.pushable_tasks);
137}
138
Ingo Molnarbcf08df2008-04-19 12:11:10 +0200139static inline int has_pushable_tasks(struct rq *rq)
140{
141 return !plist_head_empty(&rq->rt.pushable_tasks);
142}
143
Gregory Haskins917b6272008-12-29 09:39:53 -0500144#else
145
Peter Zijlstraceacc2c2009-01-16 14:46:40 +0100146static inline void enqueue_pushable_task(struct rq *rq, struct task_struct *p)
147{
148}
149
150static inline void dequeue_pushable_task(struct rq *rq, struct task_struct *p)
151{
152}
153
Gregory Haskinsb07430a2009-01-14 08:55:39 -0500154static inline
Peter Zijlstraceacc2c2009-01-16 14:46:40 +0100155void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
156{
157}
158
Gregory Haskinsb07430a2009-01-14 08:55:39 -0500159static inline
Peter Zijlstraceacc2c2009-01-16 14:46:40 +0100160void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
161{
162}
Gregory Haskins917b6272008-12-29 09:39:53 -0500163
Ingo Molnarbb44e5d2007-07-09 18:51:58 +0200164#endif /* CONFIG_SMP */
165
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100166static inline int on_rt_rq(struct sched_rt_entity *rt_se)
167{
168 return !list_empty(&rt_se->run_list);
169}
170
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100171#ifdef CONFIG_RT_GROUP_SCHED
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100172
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +0100173static inline u64 sched_rt_runtime(struct rt_rq *rt_rq)
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100174{
175 if (!rt_rq->tg)
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +0100176 return RUNTIME_INF;
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100177
Peter Zijlstraac086bc2008-04-19 19:44:58 +0200178 return rt_rq->rt_runtime;
179}
180
181static inline u64 sched_rt_period(struct rt_rq *rt_rq)
182{
183 return ktime_to_ns(rt_rq->tg->rt_bandwidth.rt_period);
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100184}
185
186#define for_each_leaf_rt_rq(rt_rq, rq) \
Bharata B Rao80f40ee2008-12-15 11:56:48 +0530187 list_for_each_entry_rcu(rt_rq, &rq->leaf_rt_rq_list, leaf_rt_rq_list)
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100188
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100189#define for_each_sched_rt_entity(rt_se) \
190 for (; rt_se; rt_se = rt_se->parent)
191
192static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se)
193{
194 return rt_se->my_q;
195}
196
Thomas Gleixner37dad3f2010-01-20 20:59:01 +0000197static void enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head);
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100198static void dequeue_rt_entity(struct sched_rt_entity *rt_se);
199
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +0100200static void sched_rt_rq_enqueue(struct rt_rq *rt_rq)
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100201{
Yong Zhang74b7eb52010-01-29 14:57:52 +0800202 int this_cpu = smp_processor_id();
Dario Faggiolif6121f42008-10-03 17:40:46 +0200203 struct task_struct *curr = rq_of_rt_rq(rt_rq)->curr;
Yong Zhang74b7eb52010-01-29 14:57:52 +0800204 struct sched_rt_entity *rt_se;
205
206 rt_se = rt_rq->tg->rt_se[this_cpu];
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100207
Dario Faggiolif6121f42008-10-03 17:40:46 +0200208 if (rt_rq->rt_nr_running) {
209 if (rt_se && !on_rt_rq(rt_se))
Thomas Gleixner37dad3f2010-01-20 20:59:01 +0000210 enqueue_rt_entity(rt_se, false);
Gregory Haskinse864c492008-12-29 09:39:49 -0500211 if (rt_rq->highest_prio.curr < curr->prio)
Peter Zijlstra10203872008-01-25 21:08:32 +0100212 resched_task(curr);
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100213 }
214}
215
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +0100216static void sched_rt_rq_dequeue(struct rt_rq *rt_rq)
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100217{
Yong Zhang74b7eb52010-01-29 14:57:52 +0800218 int this_cpu = smp_processor_id();
219 struct sched_rt_entity *rt_se;
220
221 rt_se = rt_rq->tg->rt_se[this_cpu];
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100222
223 if (rt_se && on_rt_rq(rt_se))
224 dequeue_rt_entity(rt_se);
225}
226
Peter Zijlstra23b0fdf2008-02-13 15:45:39 +0100227static inline int rt_rq_throttled(struct rt_rq *rt_rq)
228{
229 return rt_rq->rt_throttled && !rt_rq->rt_nr_boosted;
230}
231
232static int rt_se_boosted(struct sched_rt_entity *rt_se)
233{
234 struct rt_rq *rt_rq = group_rt_rq(rt_se);
235 struct task_struct *p;
236
237 if (rt_rq)
238 return !!rt_rq->rt_nr_boosted;
239
240 p = rt_task_of(rt_se);
241 return p->prio != p->normal_prio;
242}
243
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200244#ifdef CONFIG_SMP
Rusty Russellc6c49272008-11-25 02:35:05 +1030245static inline const struct cpumask *sched_rt_period_mask(void)
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200246{
247 return cpu_rq(smp_processor_id())->rd->span;
248}
249#else
Rusty Russellc6c49272008-11-25 02:35:05 +1030250static inline const struct cpumask *sched_rt_period_mask(void)
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200251{
Rusty Russellc6c49272008-11-25 02:35:05 +1030252 return cpu_online_mask;
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200253}
254#endif
255
256static inline
257struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu)
258{
259 return container_of(rt_b, struct task_group, rt_bandwidth)->rt_rq[cpu];
260}
261
Peter Zijlstraac086bc2008-04-19 19:44:58 +0200262static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq)
263{
264 return &rt_rq->tg->rt_bandwidth;
265}
266
Dhaval Giani55e12e52008-06-24 23:39:43 +0530267#else /* !CONFIG_RT_GROUP_SCHED */
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100268
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +0100269static inline u64 sched_rt_runtime(struct rt_rq *rt_rq)
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100270{
Peter Zijlstraac086bc2008-04-19 19:44:58 +0200271 return rt_rq->rt_runtime;
272}
273
274static inline u64 sched_rt_period(struct rt_rq *rt_rq)
275{
276 return ktime_to_ns(def_rt_bandwidth.rt_period);
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100277}
278
279#define for_each_leaf_rt_rq(rt_rq, rq) \
280 for (rt_rq = &rq->rt; rt_rq; rt_rq = NULL)
281
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100282#define for_each_sched_rt_entity(rt_se) \
283 for (; rt_se; rt_se = NULL)
284
285static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se)
286{
287 return NULL;
288}
289
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +0100290static inline void sched_rt_rq_enqueue(struct rt_rq *rt_rq)
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100291{
John Blackwoodf3ade832008-08-26 15:09:43 -0400292 if (rt_rq->rt_nr_running)
293 resched_task(rq_of_rt_rq(rt_rq)->curr);
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100294}
295
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +0100296static inline void sched_rt_rq_dequeue(struct rt_rq *rt_rq)
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100297{
298}
299
Peter Zijlstra23b0fdf2008-02-13 15:45:39 +0100300static inline int rt_rq_throttled(struct rt_rq *rt_rq)
301{
302 return rt_rq->rt_throttled;
303}
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200304
Rusty Russellc6c49272008-11-25 02:35:05 +1030305static inline const struct cpumask *sched_rt_period_mask(void)
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200306{
Rusty Russellc6c49272008-11-25 02:35:05 +1030307 return cpu_online_mask;
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200308}
309
310static inline
311struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu)
312{
313 return &cpu_rq(cpu)->rt;
314}
315
Peter Zijlstraac086bc2008-04-19 19:44:58 +0200316static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq)
317{
318 return &def_rt_bandwidth;
319}
320
Dhaval Giani55e12e52008-06-24 23:39:43 +0530321#endif /* CONFIG_RT_GROUP_SCHED */
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100322
Peter Zijlstrab79f3832008-06-19 14:22:25 +0200323#ifdef CONFIG_SMP
Peter Zijlstra78333cd2008-09-23 15:33:43 +0200324/*
325 * We ran out of runtime, see if we can borrow some from our neighbours.
326 */
Peter Zijlstrab79f3832008-06-19 14:22:25 +0200327static int do_balance_runtime(struct rt_rq *rt_rq)
Peter Zijlstraac086bc2008-04-19 19:44:58 +0200328{
329 struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
330 struct root_domain *rd = cpu_rq(smp_processor_id())->rd;
331 int i, weight, more = 0;
332 u64 rt_period;
333
Rusty Russellc6c49272008-11-25 02:35:05 +1030334 weight = cpumask_weight(rd->span);
Peter Zijlstraac086bc2008-04-19 19:44:58 +0200335
Thomas Gleixner0986b112009-11-17 15:32:06 +0100336 raw_spin_lock(&rt_b->rt_runtime_lock);
Peter Zijlstraac086bc2008-04-19 19:44:58 +0200337 rt_period = ktime_to_ns(rt_b->rt_period);
Rusty Russellc6c49272008-11-25 02:35:05 +1030338 for_each_cpu(i, rd->span) {
Peter Zijlstraac086bc2008-04-19 19:44:58 +0200339 struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i);
340 s64 diff;
341
342 if (iter == rt_rq)
343 continue;
344
Thomas Gleixner0986b112009-11-17 15:32:06 +0100345 raw_spin_lock(&iter->rt_runtime_lock);
Peter Zijlstra78333cd2008-09-23 15:33:43 +0200346 /*
347 * Either all rqs have inf runtime and there's nothing to steal
348 * or __disable_runtime() below sets a specific rq to inf to
349 * indicate its been disabled and disalow stealing.
350 */
Peter Zijlstra7def2be2008-06-05 14:49:58 +0200351 if (iter->rt_runtime == RUNTIME_INF)
352 goto next;
353
Peter Zijlstra78333cd2008-09-23 15:33:43 +0200354 /*
355 * From runqueues with spare time, take 1/n part of their
356 * spare time, but no more than our period.
357 */
Peter Zijlstraac086bc2008-04-19 19:44:58 +0200358 diff = iter->rt_runtime - iter->rt_time;
359 if (diff > 0) {
Peter Zijlstra58838cf2008-07-24 12:43:13 +0200360 diff = div_u64((u64)diff, weight);
Peter Zijlstraac086bc2008-04-19 19:44:58 +0200361 if (rt_rq->rt_runtime + diff > rt_period)
362 diff = rt_period - rt_rq->rt_runtime;
363 iter->rt_runtime -= diff;
364 rt_rq->rt_runtime += diff;
365 more = 1;
366 if (rt_rq->rt_runtime == rt_period) {
Thomas Gleixner0986b112009-11-17 15:32:06 +0100367 raw_spin_unlock(&iter->rt_runtime_lock);
Peter Zijlstraac086bc2008-04-19 19:44:58 +0200368 break;
369 }
370 }
Peter Zijlstra7def2be2008-06-05 14:49:58 +0200371next:
Thomas Gleixner0986b112009-11-17 15:32:06 +0100372 raw_spin_unlock(&iter->rt_runtime_lock);
Peter Zijlstraac086bc2008-04-19 19:44:58 +0200373 }
Thomas Gleixner0986b112009-11-17 15:32:06 +0100374 raw_spin_unlock(&rt_b->rt_runtime_lock);
Peter Zijlstraac086bc2008-04-19 19:44:58 +0200375
376 return more;
377}
Peter Zijlstra7def2be2008-06-05 14:49:58 +0200378
Peter Zijlstra78333cd2008-09-23 15:33:43 +0200379/*
380 * Ensure this RQ takes back all the runtime it lend to its neighbours.
381 */
Peter Zijlstra7def2be2008-06-05 14:49:58 +0200382static void __disable_runtime(struct rq *rq)
383{
384 struct root_domain *rd = rq->rd;
385 struct rt_rq *rt_rq;
386
387 if (unlikely(!scheduler_running))
388 return;
389
390 for_each_leaf_rt_rq(rt_rq, rq) {
391 struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
392 s64 want;
393 int i;
394
Thomas Gleixner0986b112009-11-17 15:32:06 +0100395 raw_spin_lock(&rt_b->rt_runtime_lock);
396 raw_spin_lock(&rt_rq->rt_runtime_lock);
Peter Zijlstra78333cd2008-09-23 15:33:43 +0200397 /*
398 * Either we're all inf and nobody needs to borrow, or we're
399 * already disabled and thus have nothing to do, or we have
400 * exactly the right amount of runtime to take out.
401 */
Peter Zijlstra7def2be2008-06-05 14:49:58 +0200402 if (rt_rq->rt_runtime == RUNTIME_INF ||
403 rt_rq->rt_runtime == rt_b->rt_runtime)
404 goto balanced;
Thomas Gleixner0986b112009-11-17 15:32:06 +0100405 raw_spin_unlock(&rt_rq->rt_runtime_lock);
Peter Zijlstra7def2be2008-06-05 14:49:58 +0200406
Peter Zijlstra78333cd2008-09-23 15:33:43 +0200407 /*
408 * Calculate the difference between what we started out with
409 * and what we current have, that's the amount of runtime
410 * we lend and now have to reclaim.
411 */
Peter Zijlstra7def2be2008-06-05 14:49:58 +0200412 want = rt_b->rt_runtime - rt_rq->rt_runtime;
413
Peter Zijlstra78333cd2008-09-23 15:33:43 +0200414 /*
415 * Greedy reclaim, take back as much as we can.
416 */
Rusty Russellc6c49272008-11-25 02:35:05 +1030417 for_each_cpu(i, rd->span) {
Peter Zijlstra7def2be2008-06-05 14:49:58 +0200418 struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i);
419 s64 diff;
420
Peter Zijlstra78333cd2008-09-23 15:33:43 +0200421 /*
422 * Can't reclaim from ourselves or disabled runqueues.
423 */
Peter Zijlstraf1679d02008-08-14 15:49:00 +0200424 if (iter == rt_rq || iter->rt_runtime == RUNTIME_INF)
Peter Zijlstra7def2be2008-06-05 14:49:58 +0200425 continue;
426
Thomas Gleixner0986b112009-11-17 15:32:06 +0100427 raw_spin_lock(&iter->rt_runtime_lock);
Peter Zijlstra7def2be2008-06-05 14:49:58 +0200428 if (want > 0) {
429 diff = min_t(s64, iter->rt_runtime, want);
430 iter->rt_runtime -= diff;
431 want -= diff;
432 } else {
433 iter->rt_runtime -= want;
434 want -= want;
435 }
Thomas Gleixner0986b112009-11-17 15:32:06 +0100436 raw_spin_unlock(&iter->rt_runtime_lock);
Peter Zijlstra7def2be2008-06-05 14:49:58 +0200437
438 if (!want)
439 break;
440 }
441
Thomas Gleixner0986b112009-11-17 15:32:06 +0100442 raw_spin_lock(&rt_rq->rt_runtime_lock);
Peter Zijlstra78333cd2008-09-23 15:33:43 +0200443 /*
444 * We cannot be left wanting - that would mean some runtime
445 * leaked out of the system.
446 */
Peter Zijlstra7def2be2008-06-05 14:49:58 +0200447 BUG_ON(want);
448balanced:
Peter Zijlstra78333cd2008-09-23 15:33:43 +0200449 /*
450 * Disable all the borrow logic by pretending we have inf
451 * runtime - in which case borrowing doesn't make sense.
452 */
Peter Zijlstra7def2be2008-06-05 14:49:58 +0200453 rt_rq->rt_runtime = RUNTIME_INF;
Thomas Gleixner0986b112009-11-17 15:32:06 +0100454 raw_spin_unlock(&rt_rq->rt_runtime_lock);
455 raw_spin_unlock(&rt_b->rt_runtime_lock);
Peter Zijlstra7def2be2008-06-05 14:49:58 +0200456 }
457}
458
459static void disable_runtime(struct rq *rq)
460{
461 unsigned long flags;
462
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100463 raw_spin_lock_irqsave(&rq->lock, flags);
Peter Zijlstra7def2be2008-06-05 14:49:58 +0200464 __disable_runtime(rq);
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100465 raw_spin_unlock_irqrestore(&rq->lock, flags);
Peter Zijlstra7def2be2008-06-05 14:49:58 +0200466}
467
468static void __enable_runtime(struct rq *rq)
469{
Peter Zijlstra7def2be2008-06-05 14:49:58 +0200470 struct rt_rq *rt_rq;
471
472 if (unlikely(!scheduler_running))
473 return;
474
Peter Zijlstra78333cd2008-09-23 15:33:43 +0200475 /*
476 * Reset each runqueue's bandwidth settings
477 */
Peter Zijlstra7def2be2008-06-05 14:49:58 +0200478 for_each_leaf_rt_rq(rt_rq, rq) {
479 struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
480
Thomas Gleixner0986b112009-11-17 15:32:06 +0100481 raw_spin_lock(&rt_b->rt_runtime_lock);
482 raw_spin_lock(&rt_rq->rt_runtime_lock);
Peter Zijlstra7def2be2008-06-05 14:49:58 +0200483 rt_rq->rt_runtime = rt_b->rt_runtime;
484 rt_rq->rt_time = 0;
Zhang, Yanminbaf25732008-09-09 11:26:33 +0800485 rt_rq->rt_throttled = 0;
Thomas Gleixner0986b112009-11-17 15:32:06 +0100486 raw_spin_unlock(&rt_rq->rt_runtime_lock);
487 raw_spin_unlock(&rt_b->rt_runtime_lock);
Peter Zijlstra7def2be2008-06-05 14:49:58 +0200488 }
489}
490
491static void enable_runtime(struct rq *rq)
492{
493 unsigned long flags;
494
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100495 raw_spin_lock_irqsave(&rq->lock, flags);
Peter Zijlstra7def2be2008-06-05 14:49:58 +0200496 __enable_runtime(rq);
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100497 raw_spin_unlock_irqrestore(&rq->lock, flags);
Peter Zijlstra7def2be2008-06-05 14:49:58 +0200498}
499
Peter Zijlstraeff65492008-06-19 14:22:26 +0200500static int balance_runtime(struct rt_rq *rt_rq)
501{
502 int more = 0;
503
504 if (rt_rq->rt_time > rt_rq->rt_runtime) {
Thomas Gleixner0986b112009-11-17 15:32:06 +0100505 raw_spin_unlock(&rt_rq->rt_runtime_lock);
Peter Zijlstraeff65492008-06-19 14:22:26 +0200506 more = do_balance_runtime(rt_rq);
Thomas Gleixner0986b112009-11-17 15:32:06 +0100507 raw_spin_lock(&rt_rq->rt_runtime_lock);
Peter Zijlstraeff65492008-06-19 14:22:26 +0200508 }
509
510 return more;
511}
Dhaval Giani55e12e52008-06-24 23:39:43 +0530512#else /* !CONFIG_SMP */
Peter Zijlstraeff65492008-06-19 14:22:26 +0200513static inline int balance_runtime(struct rt_rq *rt_rq)
514{
515 return 0;
516}
Dhaval Giani55e12e52008-06-24 23:39:43 +0530517#endif /* CONFIG_SMP */
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100518
519static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun)
520{
521 int i, idle = 1;
Rusty Russellc6c49272008-11-25 02:35:05 +1030522 const struct cpumask *span;
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200523
Peter Zijlstra0b148fa2008-08-19 12:33:04 +0200524 if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF)
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200525 return 1;
526
527 span = sched_rt_period_mask();
Rusty Russellc6c49272008-11-25 02:35:05 +1030528 for_each_cpu(i, span) {
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200529 int enqueue = 0;
530 struct rt_rq *rt_rq = sched_rt_period_rt_rq(rt_b, i);
531 struct rq *rq = rq_of_rt_rq(rt_rq);
532
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100533 raw_spin_lock(&rq->lock);
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200534 if (rt_rq->rt_time) {
535 u64 runtime;
536
Thomas Gleixner0986b112009-11-17 15:32:06 +0100537 raw_spin_lock(&rt_rq->rt_runtime_lock);
Peter Zijlstraeff65492008-06-19 14:22:26 +0200538 if (rt_rq->rt_throttled)
539 balance_runtime(rt_rq);
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200540 runtime = rt_rq->rt_runtime;
541 rt_rq->rt_time -= min(rt_rq->rt_time, overrun*runtime);
542 if (rt_rq->rt_throttled && rt_rq->rt_time < runtime) {
543 rt_rq->rt_throttled = 0;
544 enqueue = 1;
545 }
546 if (rt_rq->rt_time || rt_rq->rt_nr_running)
547 idle = 0;
Thomas Gleixner0986b112009-11-17 15:32:06 +0100548 raw_spin_unlock(&rt_rq->rt_runtime_lock);
Peter Zijlstra8a8cde12008-06-19 14:22:28 +0200549 } else if (rt_rq->rt_nr_running)
550 idle = 0;
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200551
552 if (enqueue)
553 sched_rt_rq_enqueue(rt_rq);
Thomas Gleixner05fa7852009-11-17 14:28:38 +0100554 raw_spin_unlock(&rq->lock);
Peter Zijlstrad0b27fa2008-04-19 19:44:57 +0200555 }
556
557 return idle;
558}
559
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100560static inline int rt_se_prio(struct sched_rt_entity *rt_se)
561{
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100562#ifdef CONFIG_RT_GROUP_SCHED
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100563 struct rt_rq *rt_rq = group_rt_rq(rt_se);
564
565 if (rt_rq)
Gregory Haskinse864c492008-12-29 09:39:49 -0500566 return rt_rq->highest_prio.curr;
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100567#endif
568
569 return rt_task_of(rt_se)->prio;
570}
571
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +0100572static int sched_rt_runtime_exceeded(struct rt_rq *rt_rq)
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100573{
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +0100574 u64 runtime = sched_rt_runtime(rt_rq);
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100575
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100576 if (rt_rq->rt_throttled)
Peter Zijlstra23b0fdf2008-02-13 15:45:39 +0100577 return rt_rq_throttled(rt_rq);
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100578
Peter Zijlstraac086bc2008-04-19 19:44:58 +0200579 if (sched_rt_runtime(rt_rq) >= sched_rt_period(rt_rq))
580 return 0;
581
Peter Zijlstrab79f3832008-06-19 14:22:25 +0200582 balance_runtime(rt_rq);
583 runtime = sched_rt_runtime(rt_rq);
584 if (runtime == RUNTIME_INF)
585 return 0;
Peter Zijlstraac086bc2008-04-19 19:44:58 +0200586
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +0100587 if (rt_rq->rt_time > runtime) {
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100588 rt_rq->rt_throttled = 1;
Peter Zijlstra23b0fdf2008-02-13 15:45:39 +0100589 if (rt_rq_throttled(rt_rq)) {
Peter Zijlstra9f0c1e52008-02-13 15:45:39 +0100590 sched_rt_rq_dequeue(rt_rq);
Peter Zijlstra23b0fdf2008-02-13 15:45:39 +0100591 return 1;
592 }
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100593 }
594
595 return 0;
596}
597
Ingo Molnarbb44e5d2007-07-09 18:51:58 +0200598/*
599 * Update the current task's runtime statistics. Skip current tasks that
600 * are not in our scheduling class.
601 */
Alexey Dobriyana9957442007-10-15 17:00:13 +0200602static void update_curr_rt(struct rq *rq)
Ingo Molnarbb44e5d2007-07-09 18:51:58 +0200603{
604 struct task_struct *curr = rq->curr;
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100605 struct sched_rt_entity *rt_se = &curr->rt;
606 struct rt_rq *rt_rq = rt_rq_of_se(rt_se);
Ingo Molnarbb44e5d2007-07-09 18:51:58 +0200607 u64 delta_exec;
608
609 if (!task_has_rt_policy(curr))
610 return;
611
Ingo Molnard2819182007-08-09 11:16:47 +0200612 delta_exec = rq->clock - curr->se.exec_start;
Ingo Molnarbb44e5d2007-07-09 18:51:58 +0200613 if (unlikely((s64)delta_exec < 0))
614 delta_exec = 0;
Ingo Molnar6cfb0d52007-08-02 17:41:40 +0200615
616 schedstat_set(curr->se.exec_max, max(curr->se.exec_max, delta_exec));
Ingo Molnarbb44e5d2007-07-09 18:51:58 +0200617
618 curr->se.sum_exec_runtime += delta_exec;
Frank Mayharf06febc2008-09-12 09:54:39 -0700619 account_group_exec_runtime(curr, delta_exec);
620
Ingo Molnard2819182007-08-09 11:16:47 +0200621 curr->se.exec_start = rq->clock;
Srivatsa Vaddagirid842de82007-12-02 20:04:49 +0100622 cpuacct_charge(curr, delta_exec);
Peter Zijlstrafa85ae22008-01-25 21:08:29 +0100623
Peter Zijlstrae9e92502009-09-01 10:34:37 +0200624 sched_rt_avg_update(rq, delta_exec);
625
Peter Zijlstra0b148fa2008-08-19 12:33:04 +0200626 if (!rt_bandwidth_enabled())
627 return;
628
Dhaval Giani354d60c2008-04-19 19:44:59 +0200629 for_each_sched_rt_entity(rt_se) {
630 rt_rq = rt_rq_of_se(rt_se);
631
Peter Zijlstracc2991c2008-08-19 12:33:03 +0200632 if (sched_rt_runtime(rt_rq) != RUNTIME_INF) {
Thomas Gleixner0986b112009-11-17 15:32:06 +0100633 raw_spin_lock(&rt_rq->rt_runtime_lock);
Peter Zijlstracc2991c2008-08-19 12:33:03 +0200634 rt_rq->rt_time += delta_exec;
635 if (sched_rt_runtime_exceeded(rt_rq))
636 resched_task(curr);
Thomas Gleixner0986b112009-11-17 15:32:06 +0100637 raw_spin_unlock(&rt_rq->rt_runtime_lock);
Peter Zijlstracc2991c2008-08-19 12:33:03 +0200638 }
Dhaval Giani354d60c2008-04-19 19:44:59 +0200639 }
Ingo Molnarbb44e5d2007-07-09 18:51:58 +0200640}
641
Gregory Haskins398a1532009-01-14 09:10:04 -0500642#if defined CONFIG_SMP
Gregory Haskinse864c492008-12-29 09:39:49 -0500643
644static struct task_struct *pick_next_highest_task_rt(struct rq *rq, int cpu);
645
646static inline int next_prio(struct rq *rq)
Steven Rostedt63489e42008-01-25 21:08:03 +0100647{
Gregory Haskinse864c492008-12-29 09:39:49 -0500648 struct task_struct *next = pick_next_highest_task_rt(rq, rq->cpu);
Gregory Haskins1f11eb6a2008-06-04 15:04:05 -0400649
Gregory Haskinse864c492008-12-29 09:39:49 -0500650 if (next && rt_prio(next->prio))
651 return next->prio;
652 else
653 return MAX_RT_PRIO;
654}
Gregory Haskinse864c492008-12-29 09:39:49 -0500655
Gregory Haskins398a1532009-01-14 09:10:04 -0500656static void
657inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio)
Ingo Molnarbb44e5d2007-07-09 18:51:58 +0200658{
Gregory Haskins4d984272008-12-29 09:39:49 -0500659 struct rq *rq = rq_of_rt_rq(rt_rq);
Gregory Haskins4d984272008-12-29 09:39:49 -0500660
Gregory Haskins398a1532009-01-14 09:10:04 -0500661 if (prio < prev_prio) {
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100662
Gregory Haskinse864c492008-12-29 09:39:49 -0500663 /*
664 * If the new task is higher in priority than anything on the
Gregory Haskins398a1532009-01-14 09:10:04 -0500665 * run-queue, we know that the previous high becomes our
666 * next-highest.
Gregory Haskinse864c492008-12-29 09:39:49 -0500667 */
Gregory Haskins398a1532009-01-14 09:10:04 -0500668 rt_rq->highest_prio.next = prev_prio;
669
Gregory Haskins1f11eb6a2008-06-04 15:04:05 -0400670 if (rq->online)
Gregory Haskins4d984272008-12-29 09:39:49 -0500671 cpupri_set(&rq->rd->cpupri, rq->cpu, prio);
Ingo Molnar1100ac92008-06-05 12:25:37 +0200672
Gregory Haskinse864c492008-12-29 09:39:49 -0500673 } else if (prio == rt_rq->highest_prio.curr)
674 /*
675 * If the next task is equal in priority to the highest on
676 * the run-queue, then we implicitly know that the next highest
677 * task cannot be any lower than current
678 */
679 rt_rq->highest_prio.next = prio;
680 else if (prio < rt_rq->highest_prio.next)
681 /*
682 * Otherwise, we need to recompute next-highest
683 */
684 rt_rq->highest_prio.next = next_prio(rq);
Steven Rostedt63489e42008-01-25 21:08:03 +0100685}
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100686
Gregory Haskins398a1532009-01-14 09:10:04 -0500687static void
688dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio)
Steven Rostedt63489e42008-01-25 21:08:03 +0100689{
Gregory Haskins4d984272008-12-29 09:39:49 -0500690 struct rq *rq = rq_of_rt_rq(rt_rq);
Gregory Haskins6e0534f2008-05-12 21:21:01 +0200691
Gregory Haskins398a1532009-01-14 09:10:04 -0500692 if (rt_rq->rt_nr_running && (prio <= rt_rq->highest_prio.next))
693 rt_rq->highest_prio.next = next_prio(rq);
694
695 if (rq->online && rt_rq->highest_prio.curr != prev_prio)
696 cpupri_set(&rq->rd->cpupri, rq->cpu, rt_rq->highest_prio.curr);
697}
698
699#else /* CONFIG_SMP */
700
701static inline
702void inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {}
703static inline
704void dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {}
705
706#endif /* CONFIG_SMP */
707
Steven Rostedt63489e42008-01-25 21:08:03 +0100708#if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
Gregory Haskins398a1532009-01-14 09:10:04 -0500709static void
710inc_rt_prio(struct rt_rq *rt_rq, int prio)
711{
712 int prev_prio = rt_rq->highest_prio.curr;
Steven Rostedt63489e42008-01-25 21:08:03 +0100713
Gregory Haskins398a1532009-01-14 09:10:04 -0500714 if (prio < prev_prio)
715 rt_rq->highest_prio.curr = prio;
716
717 inc_rt_prio_smp(rt_rq, prio, prev_prio);
718}
719
720static void
721dec_rt_prio(struct rt_rq *rt_rq, int prio)
722{
723 int prev_prio = rt_rq->highest_prio.curr;
724
725 if (rt_rq->rt_nr_running) {
726
727 WARN_ON(prio < prev_prio);
Gregory Haskinse864c492008-12-29 09:39:49 -0500728
729 /*
Gregory Haskins398a1532009-01-14 09:10:04 -0500730 * This may have been our highest task, and therefore
731 * we may have some recomputation to do
Gregory Haskinse864c492008-12-29 09:39:49 -0500732 */
Gregory Haskins398a1532009-01-14 09:10:04 -0500733 if (prio == prev_prio) {
Gregory Haskinse864c492008-12-29 09:39:49 -0500734 struct rt_prio_array *array = &rt_rq->active;
735
736 rt_rq->highest_prio.curr =
Steven Rostedt764a9d62008-01-25 21:08:04 +0100737 sched_find_first_bit(array->bitmap);
Gregory Haskinse864c492008-12-29 09:39:49 -0500738 }
739
Steven Rostedt764a9d62008-01-25 21:08:04 +0100740 } else
Gregory Haskinse864c492008-12-29 09:39:49 -0500741 rt_rq->highest_prio.curr = MAX_RT_PRIO;
Gregory Haskins73fe6aae2008-01-25 21:08:07 +0100742
Gregory Haskins398a1532009-01-14 09:10:04 -0500743 dec_rt_prio_smp(rt_rq, prio, prev_prio);
744}
Gregory Haskins1f11eb6a2008-06-04 15:04:05 -0400745
Gregory Haskins398a1532009-01-14 09:10:04 -0500746#else
747
748static inline void inc_rt_prio(struct rt_rq *rt_rq, int prio) {}
749static inline void dec_rt_prio(struct rt_rq *rt_rq, int prio) {}
750
751#endif /* CONFIG_SMP || CONFIG_RT_GROUP_SCHED */
752
Gregory Haskins73fe6aae2008-01-25 21:08:07 +0100753#ifdef CONFIG_RT_GROUP_SCHED
Gregory Haskins398a1532009-01-14 09:10:04 -0500754
755static void
756inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
757{
Gregory Haskins73fe6aae2008-01-25 21:08:07 +0100758 if (rt_se_boosted(rt_se))
Steven Rostedt764a9d62008-01-25 21:08:04 +0100759 rt_rq->rt_nr_boosted++;
Peter Zijlstra052f1dc2008-02-13 15:45:40 +0100760
Peter Zijlstra23b0fdf2008-02-13 15:45:39 +0100761 if (rt_rq->tg)
762 start_rt_bandwidth(&rt_rq->tg->rt_bandwidth);
Gregory Haskins398a1532009-01-14 09:10:04 -0500763}
764
765static void
766dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
767{
Peter Zijlstra23b0fdf2008-02-13 15:45:39 +0100768 if (rt_se_boosted(rt_se))
769 rt_rq->rt_nr_boosted--;
770
771 WARN_ON(!rt_rq->rt_nr_running && rt_rq->rt_nr_boosted);
Gregory Haskins398a1532009-01-14 09:10:04 -0500772}
773
774#else /* CONFIG_RT_GROUP_SCHED */
775
776static void
777inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
778{
Ingo Molnarbb44e5d2007-07-09 18:51:58 +0200779 start_rt_bandwidth(&def_rt_bandwidth);
Gregory Haskins398a1532009-01-14 09:10:04 -0500780}
781
782static inline
783void dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) {}
784
785#endif /* CONFIG_RT_GROUP_SCHED */
786
787static inline
788void inc_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
789{
790 int prio = rt_se_prio(rt_se);
791
792 WARN_ON(!rt_prio(prio));
793 rt_rq->rt_nr_running++;
794
795 inc_rt_prio(rt_rq, prio);
796 inc_rt_migration(rt_se, rt_rq);
797 inc_rt_group(rt_se, rt_rq);
Ingo Molnarbb44e5d2007-07-09 18:51:58 +0200798}
799
Peter Zijlstra23b0fdf2008-02-13 15:45:39 +0100800static inline
801void dec_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
802{
Ingo Molnarbb44e5d2007-07-09 18:51:58 +0200803 WARN_ON(!rt_prio(rt_se_prio(rt_se)));
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100804 WARN_ON(!rt_rq->rt_nr_running);
Ingo Molnarbb44e5d2007-07-09 18:51:58 +0200805 rt_rq->rt_nr_running--;
Ingo Molnarbb44e5d2007-07-09 18:51:58 +0200806
Gregory Haskins398a1532009-01-14 09:10:04 -0500807 dec_rt_prio(rt_rq, rt_se_prio(rt_se));
808 dec_rt_migration(rt_se, rt_rq);
809 dec_rt_group(rt_se, rt_rq);
Ingo Molnarbb44e5d2007-07-09 18:51:58 +0200810}
811
Thomas Gleixner37dad3f2010-01-20 20:59:01 +0000812static void __enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head)
Ingo Molnarbb44e5d2007-07-09 18:51:58 +0200813{
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100814 struct rt_rq *rt_rq = rt_rq_of_se(rt_se);
815 struct rt_prio_array *array = &rt_rq->active;
816 struct rt_rq *group_rq = group_rt_rq(rt_se);
Dmitry Adamushko20b63312008-06-11 00:58:30 +0200817 struct list_head *queue = array->queue + rt_se_prio(rt_se);
Ingo Molnarbb44e5d2007-07-09 18:51:58 +0200818
Peter Zijlstraad2a3f12008-06-19 09:06:57 +0200819 /*
820 * Don't enqueue the group if its throttled, or when empty.
821 * The latter is a consequence of the former when a child group
822 * get throttled and the current group doesn't have any other
823 * active members.
824 */
825 if (group_rq && (rt_rq_throttled(group_rq) || !group_rq->rt_nr_running))
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100826 return;
Ingo Molnarbb44e5d2007-07-09 18:51:58 +0200827
Thomas Gleixner37dad3f2010-01-20 20:59:01 +0000828 if (head)
829 list_add(&rt_se->run_list, queue);
830 else
831 list_add_tail(&rt_se->run_list, queue);
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100832 __set_bit(rt_se_prio(rt_se), array->bitmap);
Peter Zijlstra78f2c7d2008-01-25 21:08:27 +0100833
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100834 inc_rt_tasks(rt_se, rt_rq);
835}
836
Peter Zijlstraad2a3f12008-06-19 09:06:57 +0200837static void __dequeue_rt_entity(struct sched_rt_entity *rt_se)
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100838{
839 struct rt_rq *rt_rq = rt_rq_of_se(rt_se);
840 struct rt_prio_array *array = &rt_rq->active;
841
842 list_del_init(&rt_se->run_list);
843 if (list_empty(array->queue + rt_se_prio(rt_se)))
844 __clear_bit(rt_se_prio(rt_se), array->bitmap);
845
846 dec_rt_tasks(rt_se, rt_rq);
847}
848
849/*
850 * Because the prio of an upper entry depends on the lower
851 * entries, we must remove entries top - down.
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100852 */
Peter Zijlstraad2a3f12008-06-19 09:06:57 +0200853static void dequeue_rt_stack(struct sched_rt_entity *rt_se)
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100854{
Peter Zijlstraad2a3f12008-06-19 09:06:57 +0200855 struct sched_rt_entity *back = NULL;
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100856
Peter Zijlstra58d6c2d2008-04-19 19:45:00 +0200857 for_each_sched_rt_entity(rt_se) {
858 rt_se->back = back;
859 back = rt_se;
860 }
861
862 for (rt_se = back; rt_se; rt_se = rt_se->back) {
863 if (on_rt_rq(rt_se))
Peter Zijlstraad2a3f12008-06-19 09:06:57 +0200864 __dequeue_rt_entity(rt_se);
865 }
866}
867
Thomas Gleixner37dad3f2010-01-20 20:59:01 +0000868static void enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head)
Peter Zijlstraad2a3f12008-06-19 09:06:57 +0200869{
870 dequeue_rt_stack(rt_se);
871 for_each_sched_rt_entity(rt_se)
Thomas Gleixner37dad3f2010-01-20 20:59:01 +0000872 __enqueue_rt_entity(rt_se, head);
Peter Zijlstraad2a3f12008-06-19 09:06:57 +0200873}
874
875static void dequeue_rt_entity(struct sched_rt_entity *rt_se)
876{
877 dequeue_rt_stack(rt_se);
878
879 for_each_sched_rt_entity(rt_se) {
880 struct rt_rq *rt_rq = group_rt_rq(rt_se);
881
882 if (rt_rq && rt_rq->rt_nr_running)
Thomas Gleixner37dad3f2010-01-20 20:59:01 +0000883 __enqueue_rt_entity(rt_se, false);
Peter Zijlstra58d6c2d2008-04-19 19:45:00 +0200884 }
Ingo Molnarbb44e5d2007-07-09 18:51:58 +0200885}
886
887/*
888 * Adding/removing a task to/from a priority array:
889 */
Thomas Gleixnerea87bb72010-01-20 20:58:57 +0000890static void
891enqueue_task_rt(struct rq *rq, struct task_struct *p, int wakeup, bool head)
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100892{
893 struct sched_rt_entity *rt_se = &p->rt;
894
895 if (wakeup)
896 rt_se->timeout = 0;
897
Thomas Gleixner37dad3f2010-01-20 20:59:01 +0000898 enqueue_rt_entity(rt_se, head);
Peter Zijlstrac09595f2008-06-27 13:41:14 +0200899
Gregory Haskins917b6272008-12-29 09:39:53 -0500900 if (!task_current(rq, p) && p->rt.nr_cpus_allowed > 1)
901 enqueue_pushable_task(rq, p);
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100902}
903
Ingo Molnarbb44e5d2007-07-09 18:51:58 +0200904static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int sleep)
905{
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100906 struct sched_rt_entity *rt_se = &p->rt;
Ingo Molnarbb44e5d2007-07-09 18:51:58 +0200907
908 update_curr_rt(rq);
Peter Zijlstraad2a3f12008-06-19 09:06:57 +0200909 dequeue_rt_entity(rt_se);
Peter Zijlstrac09595f2008-06-27 13:41:14 +0200910
Gregory Haskins917b6272008-12-29 09:39:53 -0500911 dequeue_pushable_task(rq, p);
Ingo Molnarbb44e5d2007-07-09 18:51:58 +0200912}
913
914/*
915 * Put task to the end of the run list without the overhead of dequeue
916 * followed by enqueue.
917 */
Dmitry Adamushko7ebefa82008-07-01 23:32:15 +0200918static void
919requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se, int head)
Ingo Molnarbb44e5d2007-07-09 18:51:58 +0200920{
Ingo Molnar1cdad712008-06-19 09:09:15 +0200921 if (on_rt_rq(rt_se)) {
Dmitry Adamushko7ebefa82008-07-01 23:32:15 +0200922 struct rt_prio_array *array = &rt_rq->active;
923 struct list_head *queue = array->queue + rt_se_prio(rt_se);
924
925 if (head)
926 list_move(&rt_se->run_list, queue);
927 else
928 list_move_tail(&rt_se->run_list, queue);
Ingo Molnar1cdad712008-06-19 09:09:15 +0200929 }
Ingo Molnarbb44e5d2007-07-09 18:51:58 +0200930}
931
Dmitry Adamushko7ebefa82008-07-01 23:32:15 +0200932static void requeue_task_rt(struct rq *rq, struct task_struct *p, int head)
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100933{
934 struct sched_rt_entity *rt_se = &p->rt;
935 struct rt_rq *rt_rq;
936
937 for_each_sched_rt_entity(rt_se) {
938 rt_rq = rt_rq_of_se(rt_se);
Dmitry Adamushko7ebefa82008-07-01 23:32:15 +0200939 requeue_rt_entity(rt_rq, rt_se, head);
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100940 }
941}
942
943static void yield_task_rt(struct rq *rq)
Ingo Molnarbb44e5d2007-07-09 18:51:58 +0200944{
Dmitry Adamushko7ebefa82008-07-01 23:32:15 +0200945 requeue_task_rt(rq, rq->curr, 0);
Ingo Molnarbb44e5d2007-07-09 18:51:58 +0200946}
947
Gregory Haskinse7693a32008-01-25 21:08:09 +0100948#ifdef CONFIG_SMP
Gregory Haskins318e0892008-01-25 21:08:10 +0100949static int find_lowest_rq(struct task_struct *task);
950
Peter Zijlstra7d478722009-09-14 19:55:44 +0200951static int select_task_rq_rt(struct task_struct *p, int sd_flag, int flags)
Gregory Haskinse7693a32008-01-25 21:08:09 +0100952{
Gregory Haskins318e0892008-01-25 21:08:10 +0100953 struct rq *rq = task_rq(p);
954
Peter Zijlstra0763a662009-09-14 19:37:39 +0200955 if (sd_flag != SD_BALANCE_WAKE)
Peter Zijlstra5f3edc12009-09-10 13:42:00 +0200956 return smp_processor_id();
957
Gregory Haskins318e0892008-01-25 21:08:10 +0100958 /*
Steven Rostedte1f47d82008-01-25 21:08:12 +0100959 * If the current task is an RT task, then
960 * try to see if we can wake this RT task up on another
961 * runqueue. Otherwise simply start this RT task
962 * on its current runqueue.
963 *
964 * We want to avoid overloading runqueues. Even if
965 * the RT task is of higher priority than the current RT task.
966 * RT tasks behave differently than other tasks. If
967 * one gets preempted, we try to push it off to another queue.
968 * So trying to keep a preempting RT task on the same
969 * cache hot CPU will force the running RT task to
970 * a cold CPU. So we waste all the cache for the lower
971 * RT task in hopes of saving some of a RT task
972 * that is just being woken and probably will have
973 * cold cache anyway.
Gregory Haskins318e0892008-01-25 21:08:10 +0100974 */
Gregory Haskins17b32792008-01-25 21:08:13 +0100975 if (unlikely(rt_task(rq->curr)) &&
Peter Zijlstra6f505b12008-01-25 21:08:30 +0100976 (p->rt.nr_cpus_allowed > 1)) {
Gregory Haskins318e0892008-01-25 21:08:10 +0100977 int cpu = find_lowest_rq(p);
978
979 return (cpu == -1) ? task_cpu(p) : cpu;
980 }
981
982 /*
983 * Otherwise, just let it ride on the affined RQ and the
984 * post-schedule router will push the preempted task away
985 */
Gregory Haskinse7693a32008-01-25 21:08:09 +0100986 return task_cpu(p);
987}
Dmitry Adamushko7ebefa82008-07-01 23:32:15 +0200988
989static void check_preempt_equal_prio(struct rq *rq, struct task_struct *p)
990{
Dmitry Adamushko7ebefa82008-07-01 23:32:15 +0200991 if (rq->curr->rt.nr_cpus_allowed == 1)
992 return;
993
Rusty Russell13b8bd02009-03-25 15:01:22 +1030994 if (p->rt.nr_cpus_allowed != 1
995 && cpupri_find(&rq->rd->cpupri, p, NULL))
Dmitry Adamushko7ebefa82008-07-01 23:32:15 +0200996 return;
997
Rusty Russell13b8bd02009-03-25 15:01:22 +1030998 if (!cpupri_find(&rq->rd->cpupri, rq->curr, NULL))
999 return;
Dmitry Adamushko7ebefa82008-07-01 23:32:15 +02001000
1001 /*
1002 * There appears to be other cpus that can accept
1003 * current and none to run 'p', so lets reschedule
1004 * to try and push current away:
1005 */
1006 requeue_task_rt(rq, p, 1);
1007 resched_task(rq->curr);
1008}
1009
Gregory Haskinse7693a32008-01-25 21:08:09 +01001010#endif /* CONFIG_SMP */
1011
Ingo Molnarbb44e5d2007-07-09 18:51:58 +02001012/*
1013 * Preempt the current task with a newly woken task if needed:
1014 */
Peter Zijlstra7d478722009-09-14 19:55:44 +02001015static void check_preempt_curr_rt(struct rq *rq, struct task_struct *p, int flags)
Ingo Molnarbb44e5d2007-07-09 18:51:58 +02001016{
Gregory Haskins45c01e82008-05-12 21:20:41 +02001017 if (p->prio < rq->curr->prio) {
Ingo Molnarbb44e5d2007-07-09 18:51:58 +02001018 resched_task(rq->curr);
Gregory Haskins45c01e82008-05-12 21:20:41 +02001019 return;
1020 }
1021
1022#ifdef CONFIG_SMP
1023 /*
1024 * If:
1025 *
1026 * - the newly woken task is of equal priority to the current task
1027 * - the newly woken task is non-migratable while current is migratable
1028 * - current will be preempted on the next reschedule
1029 *
1030 * we should check to see if current can readily move to a different
1031 * cpu. If so, we will reschedule to allow the push logic to try
1032 * to move current somewhere else, making room for our non-migratable
1033 * task.
1034 */
Dmitry Adamushko7ebefa82008-07-01 23:32:15 +02001035 if (p->prio == rq->curr->prio && !need_resched())
1036 check_preempt_equal_prio(rq, p);
Gregory Haskins45c01e82008-05-12 21:20:41 +02001037#endif
Ingo Molnarbb44e5d2007-07-09 18:51:58 +02001038}
1039
Peter Zijlstra6f505b12008-01-25 21:08:30 +01001040static struct sched_rt_entity *pick_next_rt_entity(struct rq *rq,
1041 struct rt_rq *rt_rq)
Ingo Molnarbb44e5d2007-07-09 18:51:58 +02001042{
Peter Zijlstra6f505b12008-01-25 21:08:30 +01001043 struct rt_prio_array *array = &rt_rq->active;
1044 struct sched_rt_entity *next = NULL;
Ingo Molnarbb44e5d2007-07-09 18:51:58 +02001045 struct list_head *queue;
1046 int idx;
1047
1048 idx = sched_find_first_bit(array->bitmap);
Peter Zijlstra6f505b12008-01-25 21:08:30 +01001049 BUG_ON(idx >= MAX_RT_PRIO);
Ingo Molnarbb44e5d2007-07-09 18:51:58 +02001050
1051 queue = array->queue + idx;
Peter Zijlstra6f505b12008-01-25 21:08:30 +01001052 next = list_entry(queue->next, struct sched_rt_entity, run_list);
Dmitry Adamushko326587b2008-01-25 21:08:34 +01001053
Ingo Molnarbb44e5d2007-07-09 18:51:58 +02001054 return next;
1055}
1056
Gregory Haskins917b6272008-12-29 09:39:53 -05001057static struct task_struct *_pick_next_task_rt(struct rq *rq)
Peter Zijlstra6f505b12008-01-25 21:08:30 +01001058{
1059 struct sched_rt_entity *rt_se;
1060 struct task_struct *p;
1061 struct rt_rq *rt_rq;
1062
Peter Zijlstra6f505b12008-01-25 21:08:30 +01001063 rt_rq = &rq->rt;
1064
1065 if (unlikely(!rt_rq->rt_nr_running))
1066 return NULL;
1067
Peter Zijlstra23b0fdf2008-02-13 15:45:39 +01001068 if (rt_rq_throttled(rt_rq))
Peter Zijlstra6f505b12008-01-25 21:08:30 +01001069 return NULL;
1070
1071 do {
1072 rt_se = pick_next_rt_entity(rq, rt_rq);
Dmitry Adamushko326587b2008-01-25 21:08:34 +01001073 BUG_ON(!rt_se);
Peter Zijlstra6f505b12008-01-25 21:08:30 +01001074 rt_rq = group_rt_rq(rt_se);
1075 } while (rt_rq);
1076
1077 p = rt_task_of(rt_se);
1078 p->se.exec_start = rq->clock;
Gregory Haskins917b6272008-12-29 09:39:53 -05001079
1080 return p;
1081}
1082
1083static struct task_struct *pick_next_task_rt(struct rq *rq)
1084{
1085 struct task_struct *p = _pick_next_task_rt(rq);
1086
1087 /* The running task is never eligible for pushing */
1088 if (p)
1089 dequeue_pushable_task(rq, p);
1090
Ingo Molnarbcf08df2008-04-19 12:11:10 +02001091#ifdef CONFIG_SMP
Gregory Haskins3f029d32009-07-29 11:08:47 -04001092 /*
1093 * We detect this state here so that we can avoid taking the RQ
1094 * lock again later if there is no need to push
1095 */
1096 rq->post_schedule = has_pushable_tasks(rq);
Ingo Molnarbcf08df2008-04-19 12:11:10 +02001097#endif
Gregory Haskins3f029d32009-07-29 11:08:47 -04001098
Peter Zijlstra6f505b12008-01-25 21:08:30 +01001099 return p;
1100}
1101
Ingo Molnar31ee5292007-08-09 11:16:49 +02001102static void put_prev_task_rt(struct rq *rq, struct task_struct *p)
Ingo Molnarbb44e5d2007-07-09 18:51:58 +02001103{
Ingo Molnarf1e14ef2007-08-09 11:16:48 +02001104 update_curr_rt(rq);
Ingo Molnarbb44e5d2007-07-09 18:51:58 +02001105 p->se.exec_start = 0;
Gregory Haskins917b6272008-12-29 09:39:53 -05001106
1107 /*
1108 * The previous task needs to be made eligible for pushing
1109 * if it is still active
1110 */
1111 if (p->se.on_rq && p->rt.nr_cpus_allowed > 1)
1112 enqueue_pushable_task(rq, p);
Ingo Molnarbb44e5d2007-07-09 18:51:58 +02001113}
1114
Peter Williams681f3e62007-10-24 18:23:51 +02001115#ifdef CONFIG_SMP
Peter Zijlstra6f505b12008-01-25 21:08:30 +01001116
Steven Rostedte8fa1362008-01-25 21:08:05 +01001117/* Only try algorithms three times */
1118#define RT_MAX_TRIES 3
1119
Steven Rostedte8fa1362008-01-25 21:08:05 +01001120static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep);
1121
Steven Rostedtf65eda42008-01-25 21:08:07 +01001122static int pick_rt_task(struct rq *rq, struct task_struct *p, int cpu)
1123{
1124 if (!task_running(rq, p) &&
Rusty Russell96f874e22008-11-25 02:35:14 +10301125 (cpu < 0 || cpumask_test_cpu(cpu, &p->cpus_allowed)) &&
Peter Zijlstra6f505b12008-01-25 21:08:30 +01001126 (p->rt.nr_cpus_allowed > 1))
Steven Rostedtf65eda42008-01-25 21:08:07 +01001127 return 1;
1128 return 0;
1129}
1130
Steven Rostedte8fa1362008-01-25 21:08:05 +01001131/* Return the second highest RT task, NULL otherwise */
Ingo Molnar79064fb2008-01-25 21:08:14 +01001132static struct task_struct *pick_next_highest_task_rt(struct rq *rq, int cpu)
Steven Rostedte8fa1362008-01-25 21:08:05 +01001133{
Peter Zijlstra6f505b12008-01-25 21:08:30 +01001134 struct task_struct *next = NULL;
1135 struct sched_rt_entity *rt_se;
1136 struct rt_prio_array *array;
1137 struct rt_rq *rt_rq;
Steven Rostedte8fa1362008-01-25 21:08:05 +01001138 int idx;
1139
Peter Zijlstra6f505b12008-01-25 21:08:30 +01001140 for_each_leaf_rt_rq(rt_rq, rq) {
1141 array = &rt_rq->active;
1142 idx = sched_find_first_bit(array->bitmap);
1143 next_idx:
1144 if (idx >= MAX_RT_PRIO)
1145 continue;
1146 if (next && next->prio < idx)
1147 continue;
1148 list_for_each_entry(rt_se, array->queue + idx, run_list) {
1149 struct task_struct *p = rt_task_of(rt_se);
1150 if (pick_rt_task(rq, p, cpu)) {
1151 next = p;
1152 break;
1153 }
1154 }
1155 if (!next) {
1156 idx = find_next_bit(array->bitmap, MAX_RT_PRIO, idx+1);
1157 goto next_idx;
1158 }
Steven Rostedte8fa1362008-01-25 21:08:05 +01001159 }
1160
Steven Rostedte8fa1362008-01-25 21:08:05 +01001161 return next;
1162}
1163
Rusty Russell0e3900e2008-11-25 02:35:13 +10301164static DEFINE_PER_CPU(cpumask_var_t, local_cpu_mask);
Steven Rostedte8fa1362008-01-25 21:08:05 +01001165
Gregory Haskins6e1254d2008-01-25 21:08:11 +01001166static int find_lowest_rq(struct task_struct *task)
1167{
1168 struct sched_domain *sd;
Rusty Russell96f874e22008-11-25 02:35:14 +10301169 struct cpumask *lowest_mask = __get_cpu_var(local_cpu_mask);
Gregory Haskins6e1254d2008-01-25 21:08:11 +01001170 int this_cpu = smp_processor_id();
1171 int cpu = task_cpu(task);
1172
Gregory Haskins6e0534f2008-05-12 21:21:01 +02001173 if (task->rt.nr_cpus_allowed == 1)
1174 return -1; /* No other targets possible */
1175
1176 if (!cpupri_find(&task_rq(task)->rd->cpupri, task, lowest_mask))
Gregory Haskins06f90db2008-01-25 21:08:13 +01001177 return -1; /* No targets found */
1178
1179 /*
Gregory Haskins6e1254d2008-01-25 21:08:11 +01001180 * At this point we have built a mask of cpus representing the
1181 * lowest priority tasks in the system. Now we want to elect
1182 * the best one based on our affinity and topology.
1183 *
1184 * We prioritize the last cpu that the task executed on since
1185 * it is most likely cache-hot in that location.
1186 */
Rusty Russell96f874e22008-11-25 02:35:14 +10301187 if (cpumask_test_cpu(cpu, lowest_mask))
Gregory Haskins6e1254d2008-01-25 21:08:11 +01001188 return cpu;
1189
1190 /*
1191 * Otherwise, we consult the sched_domains span maps to figure
1192 * out which cpu is logically closest to our hot cache data.
1193 */
Rusty Russelle2c88062009-11-03 14:53:15 +10301194 if (!cpumask_test_cpu(this_cpu, lowest_mask))
1195 this_cpu = -1; /* Skip this_cpu opt if not among lowest */
Gregory Haskins6e1254d2008-01-25 21:08:11 +01001196
Rusty Russelle2c88062009-11-03 14:53:15 +10301197 for_each_domain(cpu, sd) {
1198 if (sd->flags & SD_WAKE_AFFINE) {
1199 int best_cpu;
Gregory Haskins6e1254d2008-01-25 21:08:11 +01001200
Rusty Russelle2c88062009-11-03 14:53:15 +10301201 /*
1202 * "this_cpu" is cheaper to preempt than a
1203 * remote processor.
1204 */
1205 if (this_cpu != -1 &&
1206 cpumask_test_cpu(this_cpu, sched_domain_span(sd)))
1207 return this_cpu;
Gregory Haskins6e1254d2008-01-25 21:08:11 +01001208
Rusty Russelle2c88062009-11-03 14:53:15 +10301209 best_cpu = cpumask_first_and(lowest_mask,
1210 sched_domain_span(sd));
1211 if (best_cpu < nr_cpu_ids)
1212 return best_cpu;
Gregory Haskins6e1254d2008-01-25 21:08:11 +01001213 }
1214 }
1215
1216 /*
1217 * And finally, if there were no matches within the domains
1218 * just give the caller *something* to work with from the compatible
1219 * locations.
1220 */
Rusty Russelle2c88062009-11-03 14:53:15 +10301221 if (this_cpu != -1)
1222 return this_cpu;
1223
1224 cpu = cpumask_any(lowest_mask);
1225 if (cpu < nr_cpu_ids)
1226 return cpu;
1227 return -1;
Gregory Haskins07b40322008-01-25 21:08:10 +01001228}
1229
Steven Rostedte8fa1362008-01-25 21:08:05 +01001230/* Will lock the rq it finds */
Ingo Molnar4df64c02008-01-25 21:08:15 +01001231static struct rq *find_lock_lowest_rq(struct task_struct *task, struct rq *rq)
Steven Rostedte8fa1362008-01-25 21:08:05 +01001232{
1233 struct rq *lowest_rq = NULL;
Steven Rostedte8fa1362008-01-25 21:08:05 +01001234 int tries;
Ingo Molnar4df64c02008-01-25 21:08:15 +01001235 int cpu;
Steven Rostedte8fa1362008-01-25 21:08:05 +01001236
1237 for (tries = 0; tries < RT_MAX_TRIES; tries++) {
Gregory Haskins07b40322008-01-25 21:08:10 +01001238 cpu = find_lowest_rq(task);
Steven Rostedte8fa1362008-01-25 21:08:05 +01001239
Gregory Haskins2de0b462008-01-25 21:08:10 +01001240 if ((cpu == -1) || (cpu == rq->cpu))
Steven Rostedte8fa1362008-01-25 21:08:05 +01001241 break;
1242
Gregory Haskins07b40322008-01-25 21:08:10 +01001243 lowest_rq = cpu_rq(cpu);
1244
Steven Rostedte8fa1362008-01-25 21:08:05 +01001245 /* if the prio of this runqueue changed, try again */
Gregory Haskins07b40322008-01-25 21:08:10 +01001246 if (double_lock_balance(rq, lowest_rq)) {
Steven Rostedte8fa1362008-01-25 21:08:05 +01001247 /*
1248 * We had to unlock the run queue. In
1249 * the mean time, task could have
1250 * migrated already or had its affinity changed.
1251 * Also make sure that it wasn't scheduled on its rq.
1252 */
Gregory Haskins07b40322008-01-25 21:08:10 +01001253 if (unlikely(task_rq(task) != rq ||
Rusty Russell96f874e22008-11-25 02:35:14 +10301254 !cpumask_test_cpu(lowest_rq->cpu,
1255 &task->cpus_allowed) ||
Gregory Haskins07b40322008-01-25 21:08:10 +01001256 task_running(rq, task) ||
Steven Rostedte8fa1362008-01-25 21:08:05 +01001257 !task->se.on_rq)) {
Ingo Molnar4df64c02008-01-25 21:08:15 +01001258
Thomas Gleixner05fa7852009-11-17 14:28:38 +01001259 raw_spin_unlock(&lowest_rq->lock);
Steven Rostedte8fa1362008-01-25 21:08:05 +01001260 lowest_rq = NULL;
1261 break;
1262 }
1263 }
1264
1265 /* If this rq is still suitable use it. */
Gregory Haskinse864c492008-12-29 09:39:49 -05001266 if (lowest_rq->rt.highest_prio.curr > task->prio)
Steven Rostedte8fa1362008-01-25 21:08:05 +01001267 break;
1268
1269 /* try again */
Peter Zijlstra1b12bbc2008-08-11 09:30:22 +02001270 double_unlock_balance(rq, lowest_rq);
Steven Rostedte8fa1362008-01-25 21:08:05 +01001271 lowest_rq = NULL;
1272 }
1273
1274 return lowest_rq;
1275}
1276
Gregory Haskins917b6272008-12-29 09:39:53 -05001277static struct task_struct *pick_next_pushable_task(struct rq *rq)
1278{
1279 struct task_struct *p;
1280
1281 if (!has_pushable_tasks(rq))
1282 return NULL;
1283
1284 p = plist_first_entry(&rq->rt.pushable_tasks,
1285 struct task_struct, pushable_tasks);
1286
1287 BUG_ON(rq->cpu != task_cpu(p));
1288 BUG_ON(task_current(rq, p));
1289 BUG_ON(p->rt.nr_cpus_allowed <= 1);
1290
1291 BUG_ON(!p->se.on_rq);
1292 BUG_ON(!rt_task(p));
1293
1294 return p;
1295}
1296
Steven Rostedte8fa1362008-01-25 21:08:05 +01001297/*
1298 * If the current CPU has more than one RT task, see if the non
1299 * running task can migrate over to a CPU that is running a task
1300 * of lesser priority.
1301 */
Gregory Haskins697f0a42008-01-25 21:08:09 +01001302static int push_rt_task(struct rq *rq)
Steven Rostedte8fa1362008-01-25 21:08:05 +01001303{
1304 struct task_struct *next_task;
1305 struct rq *lowest_rq;
Steven Rostedte8fa1362008-01-25 21:08:05 +01001306
Gregory Haskinsa22d7fc2008-01-25 21:08:12 +01001307 if (!rq->rt.overloaded)
1308 return 0;
1309
Gregory Haskins917b6272008-12-29 09:39:53 -05001310 next_task = pick_next_pushable_task(rq);
Steven Rostedte8fa1362008-01-25 21:08:05 +01001311 if (!next_task)
1312 return 0;
1313
1314 retry:
Gregory Haskins697f0a42008-01-25 21:08:09 +01001315 if (unlikely(next_task == rq->curr)) {
Steven Rostedtf65eda42008-01-25 21:08:07 +01001316 WARN_ON(1);
Steven Rostedte8fa1362008-01-25 21:08:05 +01001317 return 0;
Steven Rostedtf65eda42008-01-25 21:08:07 +01001318 }
Steven Rostedte8fa1362008-01-25 21:08:05 +01001319
1320 /*
1321 * It's possible that the next_task slipped in of
1322 * higher priority than current. If that's the case
1323 * just reschedule current.
1324 */
Gregory Haskins697f0a42008-01-25 21:08:09 +01001325 if (unlikely(next_task->prio < rq->curr->prio)) {
1326 resched_task(rq->curr);
Steven Rostedte8fa1362008-01-25 21:08:05 +01001327 return 0;
1328 }
1329
Gregory Haskins697f0a42008-01-25 21:08:09 +01001330 /* We might release rq lock */
Steven Rostedte8fa1362008-01-25 21:08:05 +01001331 get_task_struct(next_task);
1332
1333 /* find_lock_lowest_rq locks the rq if found */
Gregory Haskins697f0a42008-01-25 21:08:09 +01001334 lowest_rq = find_lock_lowest_rq(next_task, rq);
Steven Rostedte8fa1362008-01-25 21:08:05 +01001335 if (!lowest_rq) {
1336 struct task_struct *task;
1337 /*
Gregory Haskins697f0a42008-01-25 21:08:09 +01001338 * find lock_lowest_rq releases rq->lock
Gregory Haskins15635132008-12-29 09:39:53 -05001339 * so it is possible that next_task has migrated.
1340 *
1341 * We need to make sure that the task is still on the same
1342 * run-queue and is also still the next task eligible for
1343 * pushing.
Steven Rostedte8fa1362008-01-25 21:08:05 +01001344 */
Gregory Haskins917b6272008-12-29 09:39:53 -05001345 task = pick_next_pushable_task(rq);
Gregory Haskins15635132008-12-29 09:39:53 -05001346 if (task_cpu(next_task) == rq->cpu && task == next_task) {
1347 /*
1348 * If we get here, the task hasnt moved at all, but
1349 * it has failed to push. We will not try again,
1350 * since the other cpus will pull from us when they
1351 * are ready.
1352 */
1353 dequeue_pushable_task(rq, next_task);
1354 goto out;
Steven Rostedte8fa1362008-01-25 21:08:05 +01001355 }
Gregory Haskins917b6272008-12-29 09:39:53 -05001356
Gregory Haskins15635132008-12-29 09:39:53 -05001357 if (!task)
1358 /* No more tasks, just exit */
1359 goto out;
1360
Gregory Haskins917b6272008-12-29 09:39:53 -05001361 /*
Gregory Haskins15635132008-12-29 09:39:53 -05001362 * Something has shifted, try again.
Gregory Haskins917b6272008-12-29 09:39:53 -05001363 */
Gregory Haskins15635132008-12-29 09:39:53 -05001364 put_task_struct(next_task);
1365 next_task = task;
1366 goto retry;
Steven Rostedte8fa1362008-01-25 21:08:05 +01001367 }
1368
Gregory Haskins697f0a42008-01-25 21:08:09 +01001369 deactivate_task(rq, next_task, 0);
Steven Rostedte8fa1362008-01-25 21:08:05 +01001370 set_task_cpu(next_task, lowest_rq->cpu);
1371 activate_task(lowest_rq, next_task, 0);
1372
1373 resched_task(lowest_rq->curr);
1374
Peter Zijlstra1b12bbc2008-08-11 09:30:22 +02001375 double_unlock_balance(rq, lowest_rq);
Steven Rostedte8fa1362008-01-25 21:08:05 +01001376
Steven Rostedte8fa1362008-01-25 21:08:05 +01001377out:
1378 put_task_struct(next_task);
1379
Gregory Haskins917b6272008-12-29 09:39:53 -05001380 return 1;
Steven Rostedte8fa1362008-01-25 21:08:05 +01001381}
1382
Steven Rostedte8fa1362008-01-25 21:08:05 +01001383static void push_rt_tasks(struct rq *rq)
1384{
1385 /* push_rt_task will return true if it moved an RT */
1386 while (push_rt_task(rq))
1387 ;
1388}
1389
Steven Rostedtf65eda42008-01-25 21:08:07 +01001390static int pull_rt_task(struct rq *this_rq)
1391{
Ingo Molnar80bf3172008-01-25 21:08:17 +01001392 int this_cpu = this_rq->cpu, ret = 0, cpu;
Gregory Haskinsa8728942008-12-29 09:39:49 -05001393 struct task_struct *p;
Steven Rostedtf65eda42008-01-25 21:08:07 +01001394 struct rq *src_rq;
Steven Rostedtf65eda42008-01-25 21:08:07 +01001395
Gregory Haskins637f5082008-01-25 21:08:18 +01001396 if (likely(!rt_overloaded(this_rq)))
Steven Rostedtf65eda42008-01-25 21:08:07 +01001397 return 0;
1398
Rusty Russellc6c49272008-11-25 02:35:05 +10301399 for_each_cpu(cpu, this_rq->rd->rto_mask) {
Steven Rostedtf65eda42008-01-25 21:08:07 +01001400 if (this_cpu == cpu)
1401 continue;
1402
1403 src_rq = cpu_rq(cpu);
Gregory Haskins74ab8e42008-12-29 09:39:50 -05001404
1405 /*
1406 * Don't bother taking the src_rq->lock if the next highest
1407 * task is known to be lower-priority than our current task.
1408 * This may look racy, but if this value is about to go
1409 * logically higher, the src_rq will push this task away.
1410 * And if its going logically lower, we do not care
1411 */
1412 if (src_rq->rt.highest_prio.next >=
1413 this_rq->rt.highest_prio.curr)
1414 continue;
1415
Steven Rostedtf65eda42008-01-25 21:08:07 +01001416 /*
1417 * We can potentially drop this_rq's lock in
1418 * double_lock_balance, and another CPU could
Gregory Haskinsa8728942008-12-29 09:39:49 -05001419 * alter this_rq
Steven Rostedtf65eda42008-01-25 21:08:07 +01001420 */
Gregory Haskinsa8728942008-12-29 09:39:49 -05001421 double_lock_balance(this_rq, src_rq);
Steven Rostedtf65eda42008-01-25 21:08:07 +01001422
1423 /*
1424 * Are there still pullable RT tasks?
1425 */
Mike Galbraith614ee1f2008-01-25 21:08:30 +01001426 if (src_rq->rt.rt_nr_running <= 1)
1427 goto skip;
Steven Rostedtf65eda42008-01-25 21:08:07 +01001428
Steven Rostedtf65eda42008-01-25 21:08:07 +01001429 p = pick_next_highest_task_rt(src_rq, this_cpu);
1430
1431 /*
1432 * Do we have an RT task that preempts
1433 * the to-be-scheduled task?
1434 */
Gregory Haskinsa8728942008-12-29 09:39:49 -05001435 if (p && (p->prio < this_rq->rt.highest_prio.curr)) {
Steven Rostedtf65eda42008-01-25 21:08:07 +01001436 WARN_ON(p == src_rq->curr);
1437 WARN_ON(!p->se.on_rq);
1438
1439 /*
1440 * There's a chance that p is higher in priority
1441 * than what's currently running on its cpu.
1442 * This is just that p is wakeing up and hasn't
1443 * had a chance to schedule. We only pull
1444 * p if it is lower in priority than the
Gregory Haskinsa8728942008-12-29 09:39:49 -05001445 * current task on the run queue
Steven Rostedtf65eda42008-01-25 21:08:07 +01001446 */
Gregory Haskinsa8728942008-12-29 09:39:49 -05001447 if (p->prio < src_rq->curr->prio)
Mike Galbraith614ee1f2008-01-25 21:08:30 +01001448 goto skip;
Steven Rostedtf65eda42008-01-25 21:08:07 +01001449
1450 ret = 1;
1451
1452 deactivate_task(src_rq, p, 0);
1453 set_task_cpu(p, this_cpu);
1454 activate_task(this_rq, p, 0);
1455 /*
1456 * We continue with the search, just in
1457 * case there's an even higher prio task
1458 * in another runqueue. (low likelyhood
1459 * but possible)
Steven Rostedtf65eda42008-01-25 21:08:07 +01001460 */
Steven Rostedtf65eda42008-01-25 21:08:07 +01001461 }
Mike Galbraith614ee1f2008-01-25 21:08:30 +01001462 skip:
Peter Zijlstra1b12bbc2008-08-11 09:30:22 +02001463 double_unlock_balance(this_rq, src_rq);
Steven Rostedtf65eda42008-01-25 21:08:07 +01001464 }
1465
1466 return ret;
1467}
1468
Steven Rostedt9a897c52008-01-25 21:08:22 +01001469static void pre_schedule_rt(struct rq *rq, struct task_struct *prev)
Steven Rostedtf65eda42008-01-25 21:08:07 +01001470{
1471 /* Try to pull RT tasks here if we lower this rq's prio */
Gregory Haskinse864c492008-12-29 09:39:49 -05001472 if (unlikely(rt_task(prev)) && rq->rt.highest_prio.curr > prev->prio)
Steven Rostedtf65eda42008-01-25 21:08:07 +01001473 pull_rt_task(rq);
1474}
1475
Steven Rostedt9a897c52008-01-25 21:08:22 +01001476static void post_schedule_rt(struct rq *rq)
Steven Rostedte8fa1362008-01-25 21:08:05 +01001477{
Gregory Haskins967fc042008-12-29 09:39:52 -05001478 push_rt_tasks(rq);
Steven Rostedte8fa1362008-01-25 21:08:05 +01001479}
1480
Gregory Haskins8ae121a2008-04-23 07:13:29 -04001481/*
1482 * If we are not running and we are not going to reschedule soon, we should
1483 * try to push tasks away now
1484 */
Peter Zijlstraefbbd052009-12-16 18:04:40 +01001485static void task_woken_rt(struct rq *rq, struct task_struct *p)
Steven Rostedt4642daf2008-01-25 21:08:07 +01001486{
Steven Rostedt9a897c52008-01-25 21:08:22 +01001487 if (!task_running(rq, p) &&
Gregory Haskins8ae121a2008-04-23 07:13:29 -04001488 !test_tsk_need_resched(rq->curr) &&
Gregory Haskins917b6272008-12-29 09:39:53 -05001489 has_pushable_tasks(rq) &&
Gregory Haskins777c2f32008-12-29 09:39:50 -05001490 p->rt.nr_cpus_allowed > 1)
Steven Rostedt4642daf2008-01-25 21:08:07 +01001491 push_rt_tasks(rq);
1492}
1493
Mike Traviscd8ba7c2008-03-26 14:23:49 -07001494static void set_cpus_allowed_rt(struct task_struct *p,
Rusty Russell96f874e22008-11-25 02:35:14 +10301495 const struct cpumask *new_mask)
Gregory Haskins73fe6aae2008-01-25 21:08:07 +01001496{
Rusty Russell96f874e22008-11-25 02:35:14 +10301497 int weight = cpumask_weight(new_mask);
Gregory Haskins73fe6aae2008-01-25 21:08:07 +01001498
1499 BUG_ON(!rt_task(p));
1500
1501 /*
1502 * Update the migration status of the RQ if we have an RT task
1503 * which is running AND changing its weight value.
1504 */
Peter Zijlstra6f505b12008-01-25 21:08:30 +01001505 if (p->se.on_rq && (weight != p->rt.nr_cpus_allowed)) {
Gregory Haskins73fe6aae2008-01-25 21:08:07 +01001506 struct rq *rq = task_rq(p);
1507
Gregory Haskins917b6272008-12-29 09:39:53 -05001508 if (!task_current(rq, p)) {
1509 /*
1510 * Make sure we dequeue this task from the pushable list
1511 * before going further. It will either remain off of
1512 * the list because we are no longer pushable, or it
1513 * will be requeued.
1514 */
1515 if (p->rt.nr_cpus_allowed > 1)
1516 dequeue_pushable_task(rq, p);
1517
1518 /*
1519 * Requeue if our weight is changing and still > 1
1520 */
1521 if (weight > 1)
1522 enqueue_pushable_task(rq, p);
1523
1524 }
1525
Peter Zijlstra6f505b12008-01-25 21:08:30 +01001526 if ((p->rt.nr_cpus_allowed <= 1) && (weight > 1)) {
Gregory Haskins73fe6aae2008-01-25 21:08:07 +01001527 rq->rt.rt_nr_migratory++;
Peter Zijlstra6f505b12008-01-25 21:08:30 +01001528 } else if ((p->rt.nr_cpus_allowed > 1) && (weight <= 1)) {
Gregory Haskins73fe6aae2008-01-25 21:08:07 +01001529 BUG_ON(!rq->rt.rt_nr_migratory);
1530 rq->rt.rt_nr_migratory--;
1531 }
1532
Gregory Haskins398a1532009-01-14 09:10:04 -05001533 update_rt_migration(&rq->rt);
Gregory Haskins73fe6aae2008-01-25 21:08:07 +01001534 }
1535
Rusty Russell96f874e22008-11-25 02:35:14 +10301536 cpumask_copy(&p->cpus_allowed, new_mask);
Peter Zijlstra6f505b12008-01-25 21:08:30 +01001537 p->rt.nr_cpus_allowed = weight;
Gregory Haskins73fe6aae2008-01-25 21:08:07 +01001538}
Ingo Molnardeeeccd2008-01-25 21:08:15 +01001539
Ingo Molnarbdd7c812008-01-25 21:08:18 +01001540/* Assumes rq->lock is held */
Gregory Haskins1f11eb6a2008-06-04 15:04:05 -04001541static void rq_online_rt(struct rq *rq)
Ingo Molnarbdd7c812008-01-25 21:08:18 +01001542{
1543 if (rq->rt.overloaded)
1544 rt_set_overload(rq);
Gregory Haskins6e0534f2008-05-12 21:21:01 +02001545
Peter Zijlstra7def2be2008-06-05 14:49:58 +02001546 __enable_runtime(rq);
1547
Gregory Haskinse864c492008-12-29 09:39:49 -05001548 cpupri_set(&rq->rd->cpupri, rq->cpu, rq->rt.highest_prio.curr);
Ingo Molnarbdd7c812008-01-25 21:08:18 +01001549}
1550
1551/* Assumes rq->lock is held */
Gregory Haskins1f11eb6a2008-06-04 15:04:05 -04001552static void rq_offline_rt(struct rq *rq)
Ingo Molnarbdd7c812008-01-25 21:08:18 +01001553{
1554 if (rq->rt.overloaded)
1555 rt_clear_overload(rq);
Gregory Haskins6e0534f2008-05-12 21:21:01 +02001556
Peter Zijlstra7def2be2008-06-05 14:49:58 +02001557 __disable_runtime(rq);
1558
Gregory Haskins6e0534f2008-05-12 21:21:01 +02001559 cpupri_set(&rq->rd->cpupri, rq->cpu, CPUPRI_INVALID);
Ingo Molnarbdd7c812008-01-25 21:08:18 +01001560}
Steven Rostedtcb469842008-01-25 21:08:22 +01001561
1562/*
1563 * When switch from the rt queue, we bring ourselves to a position
1564 * that we might want to pull RT tasks from other runqueues.
1565 */
1566static void switched_from_rt(struct rq *rq, struct task_struct *p,
1567 int running)
1568{
1569 /*
1570 * If there are other RT tasks then we will reschedule
1571 * and the scheduling of the other RT tasks will handle
1572 * the balancing. But if we are the last RT task
1573 * we may need to handle the pulling of RT tasks
1574 * now.
1575 */
1576 if (!rq->rt.rt_nr_running)
1577 pull_rt_task(rq);
1578}
Rusty Russell3d8cbdf2008-11-25 09:58:41 +10301579
1580static inline void init_sched_rt_class(void)
1581{
1582 unsigned int i;
1583
1584 for_each_possible_cpu(i)
Yinghai Lueaa95842009-06-06 14:51:36 -07001585 zalloc_cpumask_var_node(&per_cpu(local_cpu_mask, i),
Mike Travis6ca09df2008-12-31 18:08:45 -08001586 GFP_KERNEL, cpu_to_node(i));
Rusty Russell3d8cbdf2008-11-25 09:58:41 +10301587}
Steven Rostedte8fa1362008-01-25 21:08:05 +01001588#endif /* CONFIG_SMP */
Ingo Molnarbb44e5d2007-07-09 18:51:58 +02001589
Steven Rostedtcb469842008-01-25 21:08:22 +01001590/*
1591 * When switching a task to RT, we may overload the runqueue
1592 * with RT tasks. In this case we try to push them off to
1593 * other runqueues.
1594 */
1595static void switched_to_rt(struct rq *rq, struct task_struct *p,
1596 int running)
1597{
1598 int check_resched = 1;
1599
1600 /*
1601 * If we are already running, then there's nothing
1602 * that needs to be done. But if we are not running
1603 * we may need to preempt the current running task.
1604 * If that current running task is also an RT task
1605 * then see if we can move to another run queue.
1606 */
1607 if (!running) {
1608#ifdef CONFIG_SMP
1609 if (rq->rt.overloaded && push_rt_task(rq) &&
1610 /* Don't resched if we changed runqueues */
1611 rq != task_rq(p))
1612 check_resched = 0;
1613#endif /* CONFIG_SMP */
1614 if (check_resched && p->prio < rq->curr->prio)
1615 resched_task(rq->curr);
1616 }
1617}
1618
1619/*
1620 * Priority of the task has changed. This may cause
1621 * us to initiate a push or pull.
1622 */
1623static void prio_changed_rt(struct rq *rq, struct task_struct *p,
1624 int oldprio, int running)
1625{
1626 if (running) {
1627#ifdef CONFIG_SMP
1628 /*
1629 * If our priority decreases while running, we
1630 * may need to pull tasks to this runqueue.
1631 */
1632 if (oldprio < p->prio)
1633 pull_rt_task(rq);
1634 /*
1635 * If there's a higher priority task waiting to run
Steven Rostedt6fa46fa2008-03-05 10:00:12 -05001636 * then reschedule. Note, the above pull_rt_task
1637 * can release the rq lock and p could migrate.
1638 * Only reschedule if p is still on the same runqueue.
Steven Rostedtcb469842008-01-25 21:08:22 +01001639 */
Gregory Haskinse864c492008-12-29 09:39:49 -05001640 if (p->prio > rq->rt.highest_prio.curr && rq->curr == p)
Steven Rostedtcb469842008-01-25 21:08:22 +01001641 resched_task(p);
1642#else
1643 /* For UP simply resched on drop of prio */
1644 if (oldprio < p->prio)
1645 resched_task(p);
1646#endif /* CONFIG_SMP */
1647 } else {
1648 /*
1649 * This task is not running, but if it is
1650 * greater than the current running task
1651 * then reschedule.
1652 */
1653 if (p->prio < rq->curr->prio)
1654 resched_task(rq->curr);
1655 }
1656}
1657
Peter Zijlstra78f2c7d2008-01-25 21:08:27 +01001658static void watchdog(struct rq *rq, struct task_struct *p)
1659{
1660 unsigned long soft, hard;
1661
1662 if (!p->signal)
1663 return;
1664
1665 soft = p->signal->rlim[RLIMIT_RTTIME].rlim_cur;
1666 hard = p->signal->rlim[RLIMIT_RTTIME].rlim_max;
1667
1668 if (soft != RLIM_INFINITY) {
1669 unsigned long next;
1670
1671 p->rt.timeout++;
1672 next = DIV_ROUND_UP(min(soft, hard), USEC_PER_SEC/HZ);
Peter Zijlstra5a52dd52008-01-25 21:08:32 +01001673 if (p->rt.timeout > next)
Frank Mayharf06febc2008-09-12 09:54:39 -07001674 p->cputime_expires.sched_exp = p->se.sum_exec_runtime;
Peter Zijlstra78f2c7d2008-01-25 21:08:27 +01001675 }
1676}
Steven Rostedtcb469842008-01-25 21:08:22 +01001677
Peter Zijlstra8f4d37e2008-01-25 21:08:29 +01001678static void task_tick_rt(struct rq *rq, struct task_struct *p, int queued)
Ingo Molnarbb44e5d2007-07-09 18:51:58 +02001679{
Peter Zijlstra67e2be02007-12-20 15:01:17 +01001680 update_curr_rt(rq);
1681
Peter Zijlstra78f2c7d2008-01-25 21:08:27 +01001682 watchdog(rq, p);
1683
Ingo Molnarbb44e5d2007-07-09 18:51:58 +02001684 /*
1685 * RR tasks need a special form of timeslice management.
1686 * FIFO tasks have no timeslices.
1687 */
1688 if (p->policy != SCHED_RR)
1689 return;
1690
Peter Zijlstrafa717062008-01-25 21:08:27 +01001691 if (--p->rt.time_slice)
Ingo Molnarbb44e5d2007-07-09 18:51:58 +02001692 return;
1693
Peter Zijlstrafa717062008-01-25 21:08:27 +01001694 p->rt.time_slice = DEF_TIMESLICE;
Ingo Molnarbb44e5d2007-07-09 18:51:58 +02001695
Dmitry Adamushko98fbc792007-08-24 20:39:10 +02001696 /*
1697 * Requeue to the end of queue if we are not the only element
1698 * on the queue:
1699 */
Peter Zijlstrafa717062008-01-25 21:08:27 +01001700 if (p->rt.run_list.prev != p->rt.run_list.next) {
Dmitry Adamushko7ebefa82008-07-01 23:32:15 +02001701 requeue_task_rt(rq, p, 0);
Dmitry Adamushko98fbc792007-08-24 20:39:10 +02001702 set_tsk_need_resched(p);
1703 }
Ingo Molnarbb44e5d2007-07-09 18:51:58 +02001704}
1705
Srivatsa Vaddagiri83b699e2007-10-15 17:00:08 +02001706static void set_curr_task_rt(struct rq *rq)
1707{
1708 struct task_struct *p = rq->curr;
1709
1710 p->se.exec_start = rq->clock;
Gregory Haskins917b6272008-12-29 09:39:53 -05001711
1712 /* The running task is never eligible for pushing */
1713 dequeue_pushable_task(rq, p);
Srivatsa Vaddagiri83b699e2007-10-15 17:00:08 +02001714}
1715
H Hartley Sweeten6d686f42010-01-13 20:21:52 -07001716static unsigned int get_rr_interval_rt(struct rq *rq, struct task_struct *task)
Peter Williams0d721ce2009-09-21 01:31:53 +00001717{
1718 /*
1719 * Time slice is 0 for SCHED_FIFO tasks
1720 */
1721 if (task->policy == SCHED_RR)
1722 return DEF_TIMESLICE;
1723 else
1724 return 0;
1725}
1726
Harvey Harrison2abdad02008-04-25 10:53:13 -07001727static const struct sched_class rt_sched_class = {
Ingo Molnar5522d5d2007-10-15 17:00:12 +02001728 .next = &fair_sched_class,
Ingo Molnarbb44e5d2007-07-09 18:51:58 +02001729 .enqueue_task = enqueue_task_rt,
1730 .dequeue_task = dequeue_task_rt,
1731 .yield_task = yield_task_rt,
1732
1733 .check_preempt_curr = check_preempt_curr_rt,
1734
1735 .pick_next_task = pick_next_task_rt,
1736 .put_prev_task = put_prev_task_rt,
1737
Peter Williams681f3e62007-10-24 18:23:51 +02001738#ifdef CONFIG_SMP
Li Zefan4ce72a22008-10-22 15:25:26 +08001739 .select_task_rq = select_task_rq_rt,
1740
Gregory Haskins73fe6aae2008-01-25 21:08:07 +01001741 .set_cpus_allowed = set_cpus_allowed_rt,
Gregory Haskins1f11eb6a2008-06-04 15:04:05 -04001742 .rq_online = rq_online_rt,
1743 .rq_offline = rq_offline_rt,
Steven Rostedt9a897c52008-01-25 21:08:22 +01001744 .pre_schedule = pre_schedule_rt,
1745 .post_schedule = post_schedule_rt,
Peter Zijlstraefbbd052009-12-16 18:04:40 +01001746 .task_woken = task_woken_rt,
Steven Rostedtcb469842008-01-25 21:08:22 +01001747 .switched_from = switched_from_rt,
Peter Williams681f3e62007-10-24 18:23:51 +02001748#endif
Ingo Molnarbb44e5d2007-07-09 18:51:58 +02001749
Srivatsa Vaddagiri83b699e2007-10-15 17:00:08 +02001750 .set_curr_task = set_curr_task_rt,
Ingo Molnarbb44e5d2007-07-09 18:51:58 +02001751 .task_tick = task_tick_rt,
Steven Rostedtcb469842008-01-25 21:08:22 +01001752
Peter Williams0d721ce2009-09-21 01:31:53 +00001753 .get_rr_interval = get_rr_interval_rt,
1754
Steven Rostedtcb469842008-01-25 21:08:22 +01001755 .prio_changed = prio_changed_rt,
1756 .switched_to = switched_to_rt,
Ingo Molnarbb44e5d2007-07-09 18:51:58 +02001757};
Peter Zijlstraada18de2008-06-19 14:22:24 +02001758
1759#ifdef CONFIG_SCHED_DEBUG
1760extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
1761
1762static void print_rt_stats(struct seq_file *m, int cpu)
1763{
1764 struct rt_rq *rt_rq;
1765
1766 rcu_read_lock();
1767 for_each_leaf_rt_rq(rt_rq, cpu_rq(cpu))
1768 print_rt_rq(m, cpu, rt_rq);
1769 rcu_read_unlock();
1770}
Dhaval Giani55e12e52008-06-24 23:39:43 +05301771#endif /* CONFIG_SCHED_DEBUG */
Rusty Russell0e3900e2008-11-25 02:35:13 +10301772