Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1 | /* |
| 2 | * Real-Time Scheduling Class (mapped to the SCHED_FIFO and SCHED_RR |
| 3 | * policies) |
| 4 | */ |
| 5 | |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 6 | #ifdef CONFIG_RT_GROUP_SCHED |
| 7 | |
Peter Zijlstra | a1ba4d8 | 2009-04-01 18:40:15 +0200 | [diff] [blame] | 8 | #define rt_entity_is_task(rt_se) (!(rt_se)->my_q) |
| 9 | |
Peter Zijlstra | 8f48894 | 2009-07-24 12:25:30 +0200 | [diff] [blame] | 10 | static 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 Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 18 | static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq) |
| 19 | { |
| 20 | return rt_rq->rq; |
| 21 | } |
| 22 | |
| 23 | static 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 Zijlstra | a1ba4d8 | 2009-04-01 18:40:15 +0200 | [diff] [blame] | 30 | #define rt_entity_is_task(rt_se) (1) |
| 31 | |
Peter Zijlstra | 8f48894 | 2009-07-24 12:25:30 +0200 | [diff] [blame] | 32 | static 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 Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 37 | static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq) |
| 38 | { |
| 39 | return container_of(rt_rq, struct rq, rt); |
| 40 | } |
| 41 | |
| 42 | static 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 Rostedt | 4fd2917 | 2008-01-25 21:08:06 +0100 | [diff] [blame] | 52 | #ifdef CONFIG_SMP |
Ingo Molnar | 84de427 | 2008-01-25 21:08:15 +0100 | [diff] [blame] | 53 | |
Gregory Haskins | 637f508 | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 54 | static inline int rt_overloaded(struct rq *rq) |
Steven Rostedt | 4fd2917 | 2008-01-25 21:08:06 +0100 | [diff] [blame] | 55 | { |
Gregory Haskins | 637f508 | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 56 | return atomic_read(&rq->rd->rto_count); |
Steven Rostedt | 4fd2917 | 2008-01-25 21:08:06 +0100 | [diff] [blame] | 57 | } |
Ingo Molnar | 84de427 | 2008-01-25 21:08:15 +0100 | [diff] [blame] | 58 | |
Steven Rostedt | 4fd2917 | 2008-01-25 21:08:06 +0100 | [diff] [blame] | 59 | static inline void rt_set_overload(struct rq *rq) |
| 60 | { |
Gregory Haskins | 1f11eb6a | 2008-06-04 15:04:05 -0400 | [diff] [blame] | 61 | if (!rq->online) |
| 62 | return; |
| 63 | |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 64 | cpumask_set_cpu(rq->cpu, rq->rd->rto_mask); |
Steven Rostedt | 4fd2917 | 2008-01-25 21:08:06 +0100 | [diff] [blame] | 65 | /* |
| 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 Haskins | 637f508 | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 73 | atomic_inc(&rq->rd->rto_count); |
Steven Rostedt | 4fd2917 | 2008-01-25 21:08:06 +0100 | [diff] [blame] | 74 | } |
Ingo Molnar | 84de427 | 2008-01-25 21:08:15 +0100 | [diff] [blame] | 75 | |
Steven Rostedt | 4fd2917 | 2008-01-25 21:08:06 +0100 | [diff] [blame] | 76 | static inline void rt_clear_overload(struct rq *rq) |
| 77 | { |
Gregory Haskins | 1f11eb6a | 2008-06-04 15:04:05 -0400 | [diff] [blame] | 78 | if (!rq->online) |
| 79 | return; |
| 80 | |
Steven Rostedt | 4fd2917 | 2008-01-25 21:08:06 +0100 | [diff] [blame] | 81 | /* the order here really doesn't matter */ |
Gregory Haskins | 637f508 | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 82 | atomic_dec(&rq->rd->rto_count); |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 83 | cpumask_clear_cpu(rq->cpu, rq->rd->rto_mask); |
Steven Rostedt | 4fd2917 | 2008-01-25 21:08:06 +0100 | [diff] [blame] | 84 | } |
Gregory Haskins | 73fe6aae | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 85 | |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 86 | static void update_rt_migration(struct rt_rq *rt_rq) |
Gregory Haskins | 73fe6aae | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 87 | { |
Peter Zijlstra | a1ba4d8 | 2009-04-01 18:40:15 +0200 | [diff] [blame] | 88 | if (rt_rq->rt_nr_migratory && rt_rq->rt_nr_total > 1) { |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 89 | if (!rt_rq->overloaded) { |
| 90 | rt_set_overload(rq_of_rt_rq(rt_rq)); |
| 91 | rt_rq->overloaded = 1; |
Gregory Haskins | cdc8eb9 | 2008-01-25 21:08:23 +0100 | [diff] [blame] | 92 | } |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 93 | } else if (rt_rq->overloaded) { |
| 94 | rt_clear_overload(rq_of_rt_rq(rt_rq)); |
| 95 | rt_rq->overloaded = 0; |
Gregory Haskins | 637f508 | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 96 | } |
Gregory Haskins | 73fe6aae | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 97 | } |
Steven Rostedt | 4fd2917 | 2008-01-25 21:08:06 +0100 | [diff] [blame] | 98 | |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 99 | static void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
Peter Zijlstra | fa85ae2 | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 100 | { |
Peter Zijlstra | a1ba4d8 | 2009-04-01 18:40:15 +0200 | [diff] [blame] | 101 | 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 Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 107 | if (rt_se->nr_cpus_allowed > 1) |
| 108 | rt_rq->rt_nr_migratory++; |
| 109 | |
| 110 | update_rt_migration(rt_rq); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 111 | } |
| 112 | |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 113 | static void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
| 114 | { |
Peter Zijlstra | a1ba4d8 | 2009-04-01 18:40:15 +0200 | [diff] [blame] | 115 | 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 Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 121 | if (rt_se->nr_cpus_allowed > 1) |
| 122 | rt_rq->rt_nr_migratory--; |
| 123 | |
| 124 | update_rt_migration(rt_rq); |
| 125 | } |
| 126 | |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 127 | static 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 | |
| 134 | static 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 Molnar | bcf08df | 2008-04-19 12:11:10 +0200 | [diff] [blame] | 139 | static inline int has_pushable_tasks(struct rq *rq) |
| 140 | { |
| 141 | return !plist_head_empty(&rq->rt.pushable_tasks); |
| 142 | } |
| 143 | |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 144 | #else |
| 145 | |
Peter Zijlstra | ceacc2c | 2009-01-16 14:46:40 +0100 | [diff] [blame] | 146 | static inline void enqueue_pushable_task(struct rq *rq, struct task_struct *p) |
| 147 | { |
| 148 | } |
| 149 | |
| 150 | static inline void dequeue_pushable_task(struct rq *rq, struct task_struct *p) |
| 151 | { |
| 152 | } |
| 153 | |
Gregory Haskins | b07430a | 2009-01-14 08:55:39 -0500 | [diff] [blame] | 154 | static inline |
Peter Zijlstra | ceacc2c | 2009-01-16 14:46:40 +0100 | [diff] [blame] | 155 | void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
| 156 | { |
| 157 | } |
| 158 | |
Gregory Haskins | b07430a | 2009-01-14 08:55:39 -0500 | [diff] [blame] | 159 | static inline |
Peter Zijlstra | ceacc2c | 2009-01-16 14:46:40 +0100 | [diff] [blame] | 160 | void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
| 161 | { |
| 162 | } |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 163 | |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 164 | #endif /* CONFIG_SMP */ |
| 165 | |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 166 | static inline int on_rt_rq(struct sched_rt_entity *rt_se) |
| 167 | { |
| 168 | return !list_empty(&rt_se->run_list); |
| 169 | } |
| 170 | |
Peter Zijlstra | 052f1dc | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 171 | #ifdef CONFIG_RT_GROUP_SCHED |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 172 | |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 173 | static inline u64 sched_rt_runtime(struct rt_rq *rt_rq) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 174 | { |
| 175 | if (!rt_rq->tg) |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 176 | return RUNTIME_INF; |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 177 | |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 178 | return rt_rq->rt_runtime; |
| 179 | } |
| 180 | |
| 181 | static inline u64 sched_rt_period(struct rt_rq *rt_rq) |
| 182 | { |
| 183 | return ktime_to_ns(rt_rq->tg->rt_bandwidth.rt_period); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 184 | } |
| 185 | |
| 186 | #define for_each_leaf_rt_rq(rt_rq, rq) \ |
Bharata B Rao | 80f40ee | 2008-12-15 11:56:48 +0530 | [diff] [blame] | 187 | list_for_each_entry_rcu(rt_rq, &rq->leaf_rt_rq_list, leaf_rt_rq_list) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 188 | |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 189 | #define for_each_sched_rt_entity(rt_se) \ |
| 190 | for (; rt_se; rt_se = rt_se->parent) |
| 191 | |
| 192 | static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se) |
| 193 | { |
| 194 | return rt_se->my_q; |
| 195 | } |
| 196 | |
Thomas Gleixner | 37dad3f | 2010-01-20 20:59:01 +0000 | [diff] [blame] | 197 | static void enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 198 | static void dequeue_rt_entity(struct sched_rt_entity *rt_se); |
| 199 | |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 200 | static void sched_rt_rq_enqueue(struct rt_rq *rt_rq) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 201 | { |
Yong Zhang | 74b7eb5 | 2010-01-29 14:57:52 +0800 | [diff] [blame] | 202 | int this_cpu = smp_processor_id(); |
Dario Faggioli | f6121f4 | 2008-10-03 17:40:46 +0200 | [diff] [blame] | 203 | struct task_struct *curr = rq_of_rt_rq(rt_rq)->curr; |
Yong Zhang | 74b7eb5 | 2010-01-29 14:57:52 +0800 | [diff] [blame] | 204 | struct sched_rt_entity *rt_se; |
| 205 | |
| 206 | rt_se = rt_rq->tg->rt_se[this_cpu]; |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 207 | |
Dario Faggioli | f6121f4 | 2008-10-03 17:40:46 +0200 | [diff] [blame] | 208 | if (rt_rq->rt_nr_running) { |
| 209 | if (rt_se && !on_rt_rq(rt_se)) |
Thomas Gleixner | 37dad3f | 2010-01-20 20:59:01 +0000 | [diff] [blame] | 210 | enqueue_rt_entity(rt_se, false); |
Gregory Haskins | e864c49 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 211 | if (rt_rq->highest_prio.curr < curr->prio) |
Peter Zijlstra | 1020387 | 2008-01-25 21:08:32 +0100 | [diff] [blame] | 212 | resched_task(curr); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 213 | } |
| 214 | } |
| 215 | |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 216 | static void sched_rt_rq_dequeue(struct rt_rq *rt_rq) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 217 | { |
Yong Zhang | 74b7eb5 | 2010-01-29 14:57:52 +0800 | [diff] [blame] | 218 | 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 Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 222 | |
| 223 | if (rt_se && on_rt_rq(rt_se)) |
| 224 | dequeue_rt_entity(rt_se); |
| 225 | } |
| 226 | |
Peter Zijlstra | 23b0fdf | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 227 | static inline int rt_rq_throttled(struct rt_rq *rt_rq) |
| 228 | { |
| 229 | return rt_rq->rt_throttled && !rt_rq->rt_nr_boosted; |
| 230 | } |
| 231 | |
| 232 | static 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 Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 244 | #ifdef CONFIG_SMP |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 245 | static inline const struct cpumask *sched_rt_period_mask(void) |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 246 | { |
| 247 | return cpu_rq(smp_processor_id())->rd->span; |
| 248 | } |
| 249 | #else |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 250 | static inline const struct cpumask *sched_rt_period_mask(void) |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 251 | { |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 252 | return cpu_online_mask; |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 253 | } |
| 254 | #endif |
| 255 | |
| 256 | static inline |
| 257 | struct 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 Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 262 | static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq) |
| 263 | { |
| 264 | return &rt_rq->tg->rt_bandwidth; |
| 265 | } |
| 266 | |
Dhaval Giani | 55e12e5 | 2008-06-24 23:39:43 +0530 | [diff] [blame] | 267 | #else /* !CONFIG_RT_GROUP_SCHED */ |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 268 | |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 269 | static inline u64 sched_rt_runtime(struct rt_rq *rt_rq) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 270 | { |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 271 | return rt_rq->rt_runtime; |
| 272 | } |
| 273 | |
| 274 | static inline u64 sched_rt_period(struct rt_rq *rt_rq) |
| 275 | { |
| 276 | return ktime_to_ns(def_rt_bandwidth.rt_period); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 277 | } |
| 278 | |
| 279 | #define for_each_leaf_rt_rq(rt_rq, rq) \ |
| 280 | for (rt_rq = &rq->rt; rt_rq; rt_rq = NULL) |
| 281 | |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 282 | #define for_each_sched_rt_entity(rt_se) \ |
| 283 | for (; rt_se; rt_se = NULL) |
| 284 | |
| 285 | static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se) |
| 286 | { |
| 287 | return NULL; |
| 288 | } |
| 289 | |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 290 | static inline void sched_rt_rq_enqueue(struct rt_rq *rt_rq) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 291 | { |
John Blackwood | f3ade83 | 2008-08-26 15:09:43 -0400 | [diff] [blame] | 292 | if (rt_rq->rt_nr_running) |
| 293 | resched_task(rq_of_rt_rq(rt_rq)->curr); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 294 | } |
| 295 | |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 296 | static inline void sched_rt_rq_dequeue(struct rt_rq *rt_rq) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 297 | { |
| 298 | } |
| 299 | |
Peter Zijlstra | 23b0fdf | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 300 | static inline int rt_rq_throttled(struct rt_rq *rt_rq) |
| 301 | { |
| 302 | return rt_rq->rt_throttled; |
| 303 | } |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 304 | |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 305 | static inline const struct cpumask *sched_rt_period_mask(void) |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 306 | { |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 307 | return cpu_online_mask; |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 308 | } |
| 309 | |
| 310 | static inline |
| 311 | struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu) |
| 312 | { |
| 313 | return &cpu_rq(cpu)->rt; |
| 314 | } |
| 315 | |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 316 | static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq) |
| 317 | { |
| 318 | return &def_rt_bandwidth; |
| 319 | } |
| 320 | |
Dhaval Giani | 55e12e5 | 2008-06-24 23:39:43 +0530 | [diff] [blame] | 321 | #endif /* CONFIG_RT_GROUP_SCHED */ |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 322 | |
Peter Zijlstra | b79f383 | 2008-06-19 14:22:25 +0200 | [diff] [blame] | 323 | #ifdef CONFIG_SMP |
Peter Zijlstra | 78333cd | 2008-09-23 15:33:43 +0200 | [diff] [blame] | 324 | /* |
| 325 | * We ran out of runtime, see if we can borrow some from our neighbours. |
| 326 | */ |
Peter Zijlstra | b79f383 | 2008-06-19 14:22:25 +0200 | [diff] [blame] | 327 | static int do_balance_runtime(struct rt_rq *rt_rq) |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 328 | { |
| 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 Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 334 | weight = cpumask_weight(rd->span); |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 335 | |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 336 | raw_spin_lock(&rt_b->rt_runtime_lock); |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 337 | rt_period = ktime_to_ns(rt_b->rt_period); |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 338 | for_each_cpu(i, rd->span) { |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 339 | 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 Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 345 | raw_spin_lock(&iter->rt_runtime_lock); |
Peter Zijlstra | 78333cd | 2008-09-23 15:33:43 +0200 | [diff] [blame] | 346 | /* |
| 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 Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 351 | if (iter->rt_runtime == RUNTIME_INF) |
| 352 | goto next; |
| 353 | |
Peter Zijlstra | 78333cd | 2008-09-23 15:33:43 +0200 | [diff] [blame] | 354 | /* |
| 355 | * From runqueues with spare time, take 1/n part of their |
| 356 | * spare time, but no more than our period. |
| 357 | */ |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 358 | diff = iter->rt_runtime - iter->rt_time; |
| 359 | if (diff > 0) { |
Peter Zijlstra | 58838cf | 2008-07-24 12:43:13 +0200 | [diff] [blame] | 360 | diff = div_u64((u64)diff, weight); |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 361 | 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 Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 367 | raw_spin_unlock(&iter->rt_runtime_lock); |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 368 | break; |
| 369 | } |
| 370 | } |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 371 | next: |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 372 | raw_spin_unlock(&iter->rt_runtime_lock); |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 373 | } |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 374 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 375 | |
| 376 | return more; |
| 377 | } |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 378 | |
Peter Zijlstra | 78333cd | 2008-09-23 15:33:43 +0200 | [diff] [blame] | 379 | /* |
| 380 | * Ensure this RQ takes back all the runtime it lend to its neighbours. |
| 381 | */ |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 382 | static 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 Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 395 | raw_spin_lock(&rt_b->rt_runtime_lock); |
| 396 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
Peter Zijlstra | 78333cd | 2008-09-23 15:33:43 +0200 | [diff] [blame] | 397 | /* |
| 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 Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 402 | if (rt_rq->rt_runtime == RUNTIME_INF || |
| 403 | rt_rq->rt_runtime == rt_b->rt_runtime) |
| 404 | goto balanced; |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 405 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 406 | |
Peter Zijlstra | 78333cd | 2008-09-23 15:33:43 +0200 | [diff] [blame] | 407 | /* |
| 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 Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 412 | want = rt_b->rt_runtime - rt_rq->rt_runtime; |
| 413 | |
Peter Zijlstra | 78333cd | 2008-09-23 15:33:43 +0200 | [diff] [blame] | 414 | /* |
| 415 | * Greedy reclaim, take back as much as we can. |
| 416 | */ |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 417 | for_each_cpu(i, rd->span) { |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 418 | struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i); |
| 419 | s64 diff; |
| 420 | |
Peter Zijlstra | 78333cd | 2008-09-23 15:33:43 +0200 | [diff] [blame] | 421 | /* |
| 422 | * Can't reclaim from ourselves or disabled runqueues. |
| 423 | */ |
Peter Zijlstra | f1679d0 | 2008-08-14 15:49:00 +0200 | [diff] [blame] | 424 | if (iter == rt_rq || iter->rt_runtime == RUNTIME_INF) |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 425 | continue; |
| 426 | |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 427 | raw_spin_lock(&iter->rt_runtime_lock); |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 428 | 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 Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 436 | raw_spin_unlock(&iter->rt_runtime_lock); |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 437 | |
| 438 | if (!want) |
| 439 | break; |
| 440 | } |
| 441 | |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 442 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
Peter Zijlstra | 78333cd | 2008-09-23 15:33:43 +0200 | [diff] [blame] | 443 | /* |
| 444 | * We cannot be left wanting - that would mean some runtime |
| 445 | * leaked out of the system. |
| 446 | */ |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 447 | BUG_ON(want); |
| 448 | balanced: |
Peter Zijlstra | 78333cd | 2008-09-23 15:33:43 +0200 | [diff] [blame] | 449 | /* |
| 450 | * Disable all the borrow logic by pretending we have inf |
| 451 | * runtime - in which case borrowing doesn't make sense. |
| 452 | */ |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 453 | rt_rq->rt_runtime = RUNTIME_INF; |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 454 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
| 455 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 456 | } |
| 457 | } |
| 458 | |
| 459 | static void disable_runtime(struct rq *rq) |
| 460 | { |
| 461 | unsigned long flags; |
| 462 | |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 463 | raw_spin_lock_irqsave(&rq->lock, flags); |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 464 | __disable_runtime(rq); |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 465 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 466 | } |
| 467 | |
| 468 | static void __enable_runtime(struct rq *rq) |
| 469 | { |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 470 | struct rt_rq *rt_rq; |
| 471 | |
| 472 | if (unlikely(!scheduler_running)) |
| 473 | return; |
| 474 | |
Peter Zijlstra | 78333cd | 2008-09-23 15:33:43 +0200 | [diff] [blame] | 475 | /* |
| 476 | * Reset each runqueue's bandwidth settings |
| 477 | */ |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 478 | for_each_leaf_rt_rq(rt_rq, rq) { |
| 479 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); |
| 480 | |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 481 | raw_spin_lock(&rt_b->rt_runtime_lock); |
| 482 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 483 | rt_rq->rt_runtime = rt_b->rt_runtime; |
| 484 | rt_rq->rt_time = 0; |
Zhang, Yanmin | baf2573 | 2008-09-09 11:26:33 +0800 | [diff] [blame] | 485 | rt_rq->rt_throttled = 0; |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 486 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
| 487 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 488 | } |
| 489 | } |
| 490 | |
| 491 | static void enable_runtime(struct rq *rq) |
| 492 | { |
| 493 | unsigned long flags; |
| 494 | |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 495 | raw_spin_lock_irqsave(&rq->lock, flags); |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 496 | __enable_runtime(rq); |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 497 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 498 | } |
| 499 | |
Peter Zijlstra | eff6549 | 2008-06-19 14:22:26 +0200 | [diff] [blame] | 500 | static 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 Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 505 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
Peter Zijlstra | eff6549 | 2008-06-19 14:22:26 +0200 | [diff] [blame] | 506 | more = do_balance_runtime(rt_rq); |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 507 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
Peter Zijlstra | eff6549 | 2008-06-19 14:22:26 +0200 | [diff] [blame] | 508 | } |
| 509 | |
| 510 | return more; |
| 511 | } |
Dhaval Giani | 55e12e5 | 2008-06-24 23:39:43 +0530 | [diff] [blame] | 512 | #else /* !CONFIG_SMP */ |
Peter Zijlstra | eff6549 | 2008-06-19 14:22:26 +0200 | [diff] [blame] | 513 | static inline int balance_runtime(struct rt_rq *rt_rq) |
| 514 | { |
| 515 | return 0; |
| 516 | } |
Dhaval Giani | 55e12e5 | 2008-06-24 23:39:43 +0530 | [diff] [blame] | 517 | #endif /* CONFIG_SMP */ |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 518 | |
| 519 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun) |
| 520 | { |
| 521 | int i, idle = 1; |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 522 | const struct cpumask *span; |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 523 | |
Peter Zijlstra | 0b148fa | 2008-08-19 12:33:04 +0200 | [diff] [blame] | 524 | if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF) |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 525 | return 1; |
| 526 | |
| 527 | span = sched_rt_period_mask(); |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 528 | for_each_cpu(i, span) { |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 529 | 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 Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 533 | raw_spin_lock(&rq->lock); |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 534 | if (rt_rq->rt_time) { |
| 535 | u64 runtime; |
| 536 | |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 537 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
Peter Zijlstra | eff6549 | 2008-06-19 14:22:26 +0200 | [diff] [blame] | 538 | if (rt_rq->rt_throttled) |
| 539 | balance_runtime(rt_rq); |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 540 | 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 Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 548 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
Peter Zijlstra | 8a8cde1 | 2008-06-19 14:22:28 +0200 | [diff] [blame] | 549 | } else if (rt_rq->rt_nr_running) |
| 550 | idle = 0; |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 551 | |
| 552 | if (enqueue) |
| 553 | sched_rt_rq_enqueue(rt_rq); |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 554 | raw_spin_unlock(&rq->lock); |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 555 | } |
| 556 | |
| 557 | return idle; |
| 558 | } |
| 559 | |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 560 | static inline int rt_se_prio(struct sched_rt_entity *rt_se) |
| 561 | { |
Peter Zijlstra | 052f1dc | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 562 | #ifdef CONFIG_RT_GROUP_SCHED |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 563 | struct rt_rq *rt_rq = group_rt_rq(rt_se); |
| 564 | |
| 565 | if (rt_rq) |
Gregory Haskins | e864c49 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 566 | return rt_rq->highest_prio.curr; |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 567 | #endif |
| 568 | |
| 569 | return rt_task_of(rt_se)->prio; |
| 570 | } |
| 571 | |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 572 | static int sched_rt_runtime_exceeded(struct rt_rq *rt_rq) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 573 | { |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 574 | u64 runtime = sched_rt_runtime(rt_rq); |
Peter Zijlstra | fa85ae2 | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 575 | |
Peter Zijlstra | fa85ae2 | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 576 | if (rt_rq->rt_throttled) |
Peter Zijlstra | 23b0fdf | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 577 | return rt_rq_throttled(rt_rq); |
Peter Zijlstra | fa85ae2 | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 578 | |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 579 | if (sched_rt_runtime(rt_rq) >= sched_rt_period(rt_rq)) |
| 580 | return 0; |
| 581 | |
Peter Zijlstra | b79f383 | 2008-06-19 14:22:25 +0200 | [diff] [blame] | 582 | balance_runtime(rt_rq); |
| 583 | runtime = sched_rt_runtime(rt_rq); |
| 584 | if (runtime == RUNTIME_INF) |
| 585 | return 0; |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 586 | |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 587 | if (rt_rq->rt_time > runtime) { |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 588 | rt_rq->rt_throttled = 1; |
Peter Zijlstra | 23b0fdf | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 589 | if (rt_rq_throttled(rt_rq)) { |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 590 | sched_rt_rq_dequeue(rt_rq); |
Peter Zijlstra | 23b0fdf | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 591 | return 1; |
| 592 | } |
Peter Zijlstra | fa85ae2 | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 593 | } |
| 594 | |
| 595 | return 0; |
| 596 | } |
| 597 | |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 598 | /* |
| 599 | * Update the current task's runtime statistics. Skip current tasks that |
| 600 | * are not in our scheduling class. |
| 601 | */ |
Alexey Dobriyan | a995744 | 2007-10-15 17:00:13 +0200 | [diff] [blame] | 602 | static void update_curr_rt(struct rq *rq) |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 603 | { |
| 604 | struct task_struct *curr = rq->curr; |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 605 | struct sched_rt_entity *rt_se = &curr->rt; |
| 606 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 607 | u64 delta_exec; |
| 608 | |
| 609 | if (!task_has_rt_policy(curr)) |
| 610 | return; |
| 611 | |
Ingo Molnar | d281918 | 2007-08-09 11:16:47 +0200 | [diff] [blame] | 612 | delta_exec = rq->clock - curr->se.exec_start; |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 613 | if (unlikely((s64)delta_exec < 0)) |
| 614 | delta_exec = 0; |
Ingo Molnar | 6cfb0d5 | 2007-08-02 17:41:40 +0200 | [diff] [blame] | 615 | |
| 616 | schedstat_set(curr->se.exec_max, max(curr->se.exec_max, delta_exec)); |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 617 | |
| 618 | curr->se.sum_exec_runtime += delta_exec; |
Frank Mayhar | f06febc | 2008-09-12 09:54:39 -0700 | [diff] [blame] | 619 | account_group_exec_runtime(curr, delta_exec); |
| 620 | |
Ingo Molnar | d281918 | 2007-08-09 11:16:47 +0200 | [diff] [blame] | 621 | curr->se.exec_start = rq->clock; |
Srivatsa Vaddagiri | d842de8 | 2007-12-02 20:04:49 +0100 | [diff] [blame] | 622 | cpuacct_charge(curr, delta_exec); |
Peter Zijlstra | fa85ae2 | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 623 | |
Peter Zijlstra | e9e9250 | 2009-09-01 10:34:37 +0200 | [diff] [blame] | 624 | sched_rt_avg_update(rq, delta_exec); |
| 625 | |
Peter Zijlstra | 0b148fa | 2008-08-19 12:33:04 +0200 | [diff] [blame] | 626 | if (!rt_bandwidth_enabled()) |
| 627 | return; |
| 628 | |
Dhaval Giani | 354d60c | 2008-04-19 19:44:59 +0200 | [diff] [blame] | 629 | for_each_sched_rt_entity(rt_se) { |
| 630 | rt_rq = rt_rq_of_se(rt_se); |
| 631 | |
Peter Zijlstra | cc2991c | 2008-08-19 12:33:03 +0200 | [diff] [blame] | 632 | if (sched_rt_runtime(rt_rq) != RUNTIME_INF) { |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 633 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
Peter Zijlstra | cc2991c | 2008-08-19 12:33:03 +0200 | [diff] [blame] | 634 | rt_rq->rt_time += delta_exec; |
| 635 | if (sched_rt_runtime_exceeded(rt_rq)) |
| 636 | resched_task(curr); |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 637 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
Peter Zijlstra | cc2991c | 2008-08-19 12:33:03 +0200 | [diff] [blame] | 638 | } |
Dhaval Giani | 354d60c | 2008-04-19 19:44:59 +0200 | [diff] [blame] | 639 | } |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 640 | } |
| 641 | |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 642 | #if defined CONFIG_SMP |
Gregory Haskins | e864c49 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 643 | |
| 644 | static struct task_struct *pick_next_highest_task_rt(struct rq *rq, int cpu); |
| 645 | |
| 646 | static inline int next_prio(struct rq *rq) |
Steven Rostedt | 63489e4 | 2008-01-25 21:08:03 +0100 | [diff] [blame] | 647 | { |
Gregory Haskins | e864c49 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 648 | struct task_struct *next = pick_next_highest_task_rt(rq, rq->cpu); |
Gregory Haskins | 1f11eb6a | 2008-06-04 15:04:05 -0400 | [diff] [blame] | 649 | |
Gregory Haskins | e864c49 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 650 | if (next && rt_prio(next->prio)) |
| 651 | return next->prio; |
| 652 | else |
| 653 | return MAX_RT_PRIO; |
| 654 | } |
Gregory Haskins | e864c49 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 655 | |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 656 | static void |
| 657 | inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 658 | { |
Gregory Haskins | 4d98427 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 659 | struct rq *rq = rq_of_rt_rq(rt_rq); |
Gregory Haskins | 4d98427 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 660 | |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 661 | if (prio < prev_prio) { |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 662 | |
Gregory Haskins | e864c49 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 663 | /* |
| 664 | * If the new task is higher in priority than anything on the |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 665 | * run-queue, we know that the previous high becomes our |
| 666 | * next-highest. |
Gregory Haskins | e864c49 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 667 | */ |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 668 | rt_rq->highest_prio.next = prev_prio; |
| 669 | |
Gregory Haskins | 1f11eb6a | 2008-06-04 15:04:05 -0400 | [diff] [blame] | 670 | if (rq->online) |
Gregory Haskins | 4d98427 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 671 | cpupri_set(&rq->rd->cpupri, rq->cpu, prio); |
Ingo Molnar | 1100ac9 | 2008-06-05 12:25:37 +0200 | [diff] [blame] | 672 | |
Gregory Haskins | e864c49 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 673 | } 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 Rostedt | 63489e4 | 2008-01-25 21:08:03 +0100 | [diff] [blame] | 685 | } |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 686 | |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 687 | static void |
| 688 | dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) |
Steven Rostedt | 63489e4 | 2008-01-25 21:08:03 +0100 | [diff] [blame] | 689 | { |
Gregory Haskins | 4d98427 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 690 | struct rq *rq = rq_of_rt_rq(rt_rq); |
Gregory Haskins | 6e0534f | 2008-05-12 21:21:01 +0200 | [diff] [blame] | 691 | |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 692 | 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 | |
| 701 | static inline |
| 702 | void inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {} |
| 703 | static inline |
| 704 | void dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {} |
| 705 | |
| 706 | #endif /* CONFIG_SMP */ |
| 707 | |
Steven Rostedt | 63489e4 | 2008-01-25 21:08:03 +0100 | [diff] [blame] | 708 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 709 | static void |
| 710 | inc_rt_prio(struct rt_rq *rt_rq, int prio) |
| 711 | { |
| 712 | int prev_prio = rt_rq->highest_prio.curr; |
Steven Rostedt | 63489e4 | 2008-01-25 21:08:03 +0100 | [diff] [blame] | 713 | |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 714 | if (prio < prev_prio) |
| 715 | rt_rq->highest_prio.curr = prio; |
| 716 | |
| 717 | inc_rt_prio_smp(rt_rq, prio, prev_prio); |
| 718 | } |
| 719 | |
| 720 | static void |
| 721 | dec_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 Haskins | e864c49 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 728 | |
| 729 | /* |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 730 | * This may have been our highest task, and therefore |
| 731 | * we may have some recomputation to do |
Gregory Haskins | e864c49 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 732 | */ |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 733 | if (prio == prev_prio) { |
Gregory Haskins | e864c49 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 734 | struct rt_prio_array *array = &rt_rq->active; |
| 735 | |
| 736 | rt_rq->highest_prio.curr = |
Steven Rostedt | 764a9d6 | 2008-01-25 21:08:04 +0100 | [diff] [blame] | 737 | sched_find_first_bit(array->bitmap); |
Gregory Haskins | e864c49 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 738 | } |
| 739 | |
Steven Rostedt | 764a9d6 | 2008-01-25 21:08:04 +0100 | [diff] [blame] | 740 | } else |
Gregory Haskins | e864c49 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 741 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
Gregory Haskins | 73fe6aae | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 742 | |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 743 | dec_rt_prio_smp(rt_rq, prio, prev_prio); |
| 744 | } |
Gregory Haskins | 1f11eb6a | 2008-06-04 15:04:05 -0400 | [diff] [blame] | 745 | |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 746 | #else |
| 747 | |
| 748 | static inline void inc_rt_prio(struct rt_rq *rt_rq, int prio) {} |
| 749 | static inline void dec_rt_prio(struct rt_rq *rt_rq, int prio) {} |
| 750 | |
| 751 | #endif /* CONFIG_SMP || CONFIG_RT_GROUP_SCHED */ |
| 752 | |
Gregory Haskins | 73fe6aae | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 753 | #ifdef CONFIG_RT_GROUP_SCHED |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 754 | |
| 755 | static void |
| 756 | inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
| 757 | { |
Gregory Haskins | 73fe6aae | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 758 | if (rt_se_boosted(rt_se)) |
Steven Rostedt | 764a9d6 | 2008-01-25 21:08:04 +0100 | [diff] [blame] | 759 | rt_rq->rt_nr_boosted++; |
Peter Zijlstra | 052f1dc | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 760 | |
Peter Zijlstra | 23b0fdf | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 761 | if (rt_rq->tg) |
| 762 | start_rt_bandwidth(&rt_rq->tg->rt_bandwidth); |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 763 | } |
| 764 | |
| 765 | static void |
| 766 | dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
| 767 | { |
Peter Zijlstra | 23b0fdf | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 768 | 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 Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 772 | } |
| 773 | |
| 774 | #else /* CONFIG_RT_GROUP_SCHED */ |
| 775 | |
| 776 | static void |
| 777 | inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
| 778 | { |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 779 | start_rt_bandwidth(&def_rt_bandwidth); |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 780 | } |
| 781 | |
| 782 | static inline |
| 783 | void dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) {} |
| 784 | |
| 785 | #endif /* CONFIG_RT_GROUP_SCHED */ |
| 786 | |
| 787 | static inline |
| 788 | void 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 Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 798 | } |
| 799 | |
Peter Zijlstra | 23b0fdf | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 800 | static inline |
| 801 | void dec_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
| 802 | { |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 803 | WARN_ON(!rt_prio(rt_se_prio(rt_se))); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 804 | WARN_ON(!rt_rq->rt_nr_running); |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 805 | rt_rq->rt_nr_running--; |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 806 | |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 807 | 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 Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 810 | } |
| 811 | |
Thomas Gleixner | 37dad3f | 2010-01-20 20:59:01 +0000 | [diff] [blame] | 812 | static void __enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head) |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 813 | { |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 814 | 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 Adamushko | 20b6331 | 2008-06-11 00:58:30 +0200 | [diff] [blame] | 817 | struct list_head *queue = array->queue + rt_se_prio(rt_se); |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 818 | |
Peter Zijlstra | ad2a3f1 | 2008-06-19 09:06:57 +0200 | [diff] [blame] | 819 | /* |
| 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 Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 826 | return; |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 827 | |
Thomas Gleixner | 37dad3f | 2010-01-20 20:59:01 +0000 | [diff] [blame] | 828 | if (head) |
| 829 | list_add(&rt_se->run_list, queue); |
| 830 | else |
| 831 | list_add_tail(&rt_se->run_list, queue); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 832 | __set_bit(rt_se_prio(rt_se), array->bitmap); |
Peter Zijlstra | 78f2c7d | 2008-01-25 21:08:27 +0100 | [diff] [blame] | 833 | |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 834 | inc_rt_tasks(rt_se, rt_rq); |
| 835 | } |
| 836 | |
Peter Zijlstra | ad2a3f1 | 2008-06-19 09:06:57 +0200 | [diff] [blame] | 837 | static void __dequeue_rt_entity(struct sched_rt_entity *rt_se) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 838 | { |
| 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 Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 852 | */ |
Peter Zijlstra | ad2a3f1 | 2008-06-19 09:06:57 +0200 | [diff] [blame] | 853 | static void dequeue_rt_stack(struct sched_rt_entity *rt_se) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 854 | { |
Peter Zijlstra | ad2a3f1 | 2008-06-19 09:06:57 +0200 | [diff] [blame] | 855 | struct sched_rt_entity *back = NULL; |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 856 | |
Peter Zijlstra | 58d6c2d | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 857 | 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 Zijlstra | ad2a3f1 | 2008-06-19 09:06:57 +0200 | [diff] [blame] | 864 | __dequeue_rt_entity(rt_se); |
| 865 | } |
| 866 | } |
| 867 | |
Thomas Gleixner | 37dad3f | 2010-01-20 20:59:01 +0000 | [diff] [blame] | 868 | static void enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head) |
Peter Zijlstra | ad2a3f1 | 2008-06-19 09:06:57 +0200 | [diff] [blame] | 869 | { |
| 870 | dequeue_rt_stack(rt_se); |
| 871 | for_each_sched_rt_entity(rt_se) |
Thomas Gleixner | 37dad3f | 2010-01-20 20:59:01 +0000 | [diff] [blame] | 872 | __enqueue_rt_entity(rt_se, head); |
Peter Zijlstra | ad2a3f1 | 2008-06-19 09:06:57 +0200 | [diff] [blame] | 873 | } |
| 874 | |
| 875 | static 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 Gleixner | 37dad3f | 2010-01-20 20:59:01 +0000 | [diff] [blame] | 883 | __enqueue_rt_entity(rt_se, false); |
Peter Zijlstra | 58d6c2d | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 884 | } |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 885 | } |
| 886 | |
| 887 | /* |
| 888 | * Adding/removing a task to/from a priority array: |
| 889 | */ |
Thomas Gleixner | ea87bb7 | 2010-01-20 20:58:57 +0000 | [diff] [blame] | 890 | static void |
| 891 | enqueue_task_rt(struct rq *rq, struct task_struct *p, int wakeup, bool head) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 892 | { |
| 893 | struct sched_rt_entity *rt_se = &p->rt; |
| 894 | |
| 895 | if (wakeup) |
| 896 | rt_se->timeout = 0; |
| 897 | |
Thomas Gleixner | 37dad3f | 2010-01-20 20:59:01 +0000 | [diff] [blame] | 898 | enqueue_rt_entity(rt_se, head); |
Peter Zijlstra | c09595f | 2008-06-27 13:41:14 +0200 | [diff] [blame] | 899 | |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 900 | if (!task_current(rq, p) && p->rt.nr_cpus_allowed > 1) |
| 901 | enqueue_pushable_task(rq, p); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 902 | } |
| 903 | |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 904 | static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int sleep) |
| 905 | { |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 906 | struct sched_rt_entity *rt_se = &p->rt; |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 907 | |
| 908 | update_curr_rt(rq); |
Peter Zijlstra | ad2a3f1 | 2008-06-19 09:06:57 +0200 | [diff] [blame] | 909 | dequeue_rt_entity(rt_se); |
Peter Zijlstra | c09595f | 2008-06-27 13:41:14 +0200 | [diff] [blame] | 910 | |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 911 | dequeue_pushable_task(rq, p); |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 912 | } |
| 913 | |
| 914 | /* |
| 915 | * Put task to the end of the run list without the overhead of dequeue |
| 916 | * followed by enqueue. |
| 917 | */ |
Dmitry Adamushko | 7ebefa8 | 2008-07-01 23:32:15 +0200 | [diff] [blame] | 918 | static void |
| 919 | requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se, int head) |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 920 | { |
Ingo Molnar | 1cdad71 | 2008-06-19 09:09:15 +0200 | [diff] [blame] | 921 | if (on_rt_rq(rt_se)) { |
Dmitry Adamushko | 7ebefa8 | 2008-07-01 23:32:15 +0200 | [diff] [blame] | 922 | 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 Molnar | 1cdad71 | 2008-06-19 09:09:15 +0200 | [diff] [blame] | 929 | } |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 930 | } |
| 931 | |
Dmitry Adamushko | 7ebefa8 | 2008-07-01 23:32:15 +0200 | [diff] [blame] | 932 | static void requeue_task_rt(struct rq *rq, struct task_struct *p, int head) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 933 | { |
| 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 Adamushko | 7ebefa8 | 2008-07-01 23:32:15 +0200 | [diff] [blame] | 939 | requeue_rt_entity(rt_rq, rt_se, head); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 940 | } |
| 941 | } |
| 942 | |
| 943 | static void yield_task_rt(struct rq *rq) |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 944 | { |
Dmitry Adamushko | 7ebefa8 | 2008-07-01 23:32:15 +0200 | [diff] [blame] | 945 | requeue_task_rt(rq, rq->curr, 0); |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 946 | } |
| 947 | |
Gregory Haskins | e7693a3 | 2008-01-25 21:08:09 +0100 | [diff] [blame] | 948 | #ifdef CONFIG_SMP |
Gregory Haskins | 318e089 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 949 | static int find_lowest_rq(struct task_struct *task); |
| 950 | |
Peter Zijlstra | 7d47872 | 2009-09-14 19:55:44 +0200 | [diff] [blame] | 951 | static int select_task_rq_rt(struct task_struct *p, int sd_flag, int flags) |
Gregory Haskins | e7693a3 | 2008-01-25 21:08:09 +0100 | [diff] [blame] | 952 | { |
Gregory Haskins | 318e089 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 953 | struct rq *rq = task_rq(p); |
| 954 | |
Peter Zijlstra | 0763a66 | 2009-09-14 19:37:39 +0200 | [diff] [blame] | 955 | if (sd_flag != SD_BALANCE_WAKE) |
Peter Zijlstra | 5f3edc1 | 2009-09-10 13:42:00 +0200 | [diff] [blame] | 956 | return smp_processor_id(); |
| 957 | |
Gregory Haskins | 318e089 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 958 | /* |
Steven Rostedt | e1f47d8 | 2008-01-25 21:08:12 +0100 | [diff] [blame] | 959 | * 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 Haskins | 318e089 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 974 | */ |
Gregory Haskins | 17b3279 | 2008-01-25 21:08:13 +0100 | [diff] [blame] | 975 | if (unlikely(rt_task(rq->curr)) && |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 976 | (p->rt.nr_cpus_allowed > 1)) { |
Gregory Haskins | 318e089 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 977 | 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 Haskins | e7693a3 | 2008-01-25 21:08:09 +0100 | [diff] [blame] | 986 | return task_cpu(p); |
| 987 | } |
Dmitry Adamushko | 7ebefa8 | 2008-07-01 23:32:15 +0200 | [diff] [blame] | 988 | |
| 989 | static void check_preempt_equal_prio(struct rq *rq, struct task_struct *p) |
| 990 | { |
Dmitry Adamushko | 7ebefa8 | 2008-07-01 23:32:15 +0200 | [diff] [blame] | 991 | if (rq->curr->rt.nr_cpus_allowed == 1) |
| 992 | return; |
| 993 | |
Rusty Russell | 13b8bd0 | 2009-03-25 15:01:22 +1030 | [diff] [blame] | 994 | if (p->rt.nr_cpus_allowed != 1 |
| 995 | && cpupri_find(&rq->rd->cpupri, p, NULL)) |
Dmitry Adamushko | 7ebefa8 | 2008-07-01 23:32:15 +0200 | [diff] [blame] | 996 | return; |
| 997 | |
Rusty Russell | 13b8bd0 | 2009-03-25 15:01:22 +1030 | [diff] [blame] | 998 | if (!cpupri_find(&rq->rd->cpupri, rq->curr, NULL)) |
| 999 | return; |
Dmitry Adamushko | 7ebefa8 | 2008-07-01 23:32:15 +0200 | [diff] [blame] | 1000 | |
| 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 Haskins | e7693a3 | 2008-01-25 21:08:09 +0100 | [diff] [blame] | 1010 | #endif /* CONFIG_SMP */ |
| 1011 | |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1012 | /* |
| 1013 | * Preempt the current task with a newly woken task if needed: |
| 1014 | */ |
Peter Zijlstra | 7d47872 | 2009-09-14 19:55:44 +0200 | [diff] [blame] | 1015 | static void check_preempt_curr_rt(struct rq *rq, struct task_struct *p, int flags) |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1016 | { |
Gregory Haskins | 45c01e8 | 2008-05-12 21:20:41 +0200 | [diff] [blame] | 1017 | if (p->prio < rq->curr->prio) { |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1018 | resched_task(rq->curr); |
Gregory Haskins | 45c01e8 | 2008-05-12 21:20:41 +0200 | [diff] [blame] | 1019 | 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 Adamushko | 7ebefa8 | 2008-07-01 23:32:15 +0200 | [diff] [blame] | 1035 | if (p->prio == rq->curr->prio && !need_resched()) |
| 1036 | check_preempt_equal_prio(rq, p); |
Gregory Haskins | 45c01e8 | 2008-05-12 21:20:41 +0200 | [diff] [blame] | 1037 | #endif |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1038 | } |
| 1039 | |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1040 | static struct sched_rt_entity *pick_next_rt_entity(struct rq *rq, |
| 1041 | struct rt_rq *rt_rq) |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1042 | { |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1043 | struct rt_prio_array *array = &rt_rq->active; |
| 1044 | struct sched_rt_entity *next = NULL; |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1045 | struct list_head *queue; |
| 1046 | int idx; |
| 1047 | |
| 1048 | idx = sched_find_first_bit(array->bitmap); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1049 | BUG_ON(idx >= MAX_RT_PRIO); |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1050 | |
| 1051 | queue = array->queue + idx; |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1052 | next = list_entry(queue->next, struct sched_rt_entity, run_list); |
Dmitry Adamushko | 326587b | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 1053 | |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1054 | return next; |
| 1055 | } |
| 1056 | |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 1057 | static struct task_struct *_pick_next_task_rt(struct rq *rq) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1058 | { |
| 1059 | struct sched_rt_entity *rt_se; |
| 1060 | struct task_struct *p; |
| 1061 | struct rt_rq *rt_rq; |
| 1062 | |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1063 | rt_rq = &rq->rt; |
| 1064 | |
| 1065 | if (unlikely(!rt_rq->rt_nr_running)) |
| 1066 | return NULL; |
| 1067 | |
Peter Zijlstra | 23b0fdf | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 1068 | if (rt_rq_throttled(rt_rq)) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1069 | return NULL; |
| 1070 | |
| 1071 | do { |
| 1072 | rt_se = pick_next_rt_entity(rq, rt_rq); |
Dmitry Adamushko | 326587b | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 1073 | BUG_ON(!rt_se); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1074 | 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 Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 1079 | |
| 1080 | return p; |
| 1081 | } |
| 1082 | |
| 1083 | static 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 Molnar | bcf08df | 2008-04-19 12:11:10 +0200 | [diff] [blame] | 1091 | #ifdef CONFIG_SMP |
Gregory Haskins | 3f029d3 | 2009-07-29 11:08:47 -0400 | [diff] [blame] | 1092 | /* |
| 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 Molnar | bcf08df | 2008-04-19 12:11:10 +0200 | [diff] [blame] | 1097 | #endif |
Gregory Haskins | 3f029d3 | 2009-07-29 11:08:47 -0400 | [diff] [blame] | 1098 | |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1099 | return p; |
| 1100 | } |
| 1101 | |
Ingo Molnar | 31ee529 | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 1102 | static void put_prev_task_rt(struct rq *rq, struct task_struct *p) |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1103 | { |
Ingo Molnar | f1e14ef | 2007-08-09 11:16:48 +0200 | [diff] [blame] | 1104 | update_curr_rt(rq); |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1105 | p->se.exec_start = 0; |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 1106 | |
| 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 Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1113 | } |
| 1114 | |
Peter Williams | 681f3e6 | 2007-10-24 18:23:51 +0200 | [diff] [blame] | 1115 | #ifdef CONFIG_SMP |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1116 | |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1117 | /* Only try algorithms three times */ |
| 1118 | #define RT_MAX_TRIES 3 |
| 1119 | |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1120 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep); |
| 1121 | |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1122 | static int pick_rt_task(struct rq *rq, struct task_struct *p, int cpu) |
| 1123 | { |
| 1124 | if (!task_running(rq, p) && |
Rusty Russell | 96f874e2 | 2008-11-25 02:35:14 +1030 | [diff] [blame] | 1125 | (cpu < 0 || cpumask_test_cpu(cpu, &p->cpus_allowed)) && |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1126 | (p->rt.nr_cpus_allowed > 1)) |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1127 | return 1; |
| 1128 | return 0; |
| 1129 | } |
| 1130 | |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1131 | /* Return the second highest RT task, NULL otherwise */ |
Ingo Molnar | 79064fb | 2008-01-25 21:08:14 +0100 | [diff] [blame] | 1132 | static struct task_struct *pick_next_highest_task_rt(struct rq *rq, int cpu) |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1133 | { |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1134 | 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 Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1138 | int idx; |
| 1139 | |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1140 | 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 Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1159 | } |
| 1160 | |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1161 | return next; |
| 1162 | } |
| 1163 | |
Rusty Russell | 0e3900e | 2008-11-25 02:35:13 +1030 | [diff] [blame] | 1164 | static DEFINE_PER_CPU(cpumask_var_t, local_cpu_mask); |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1165 | |
Gregory Haskins | 6e1254d | 2008-01-25 21:08:11 +0100 | [diff] [blame] | 1166 | static int find_lowest_rq(struct task_struct *task) |
| 1167 | { |
| 1168 | struct sched_domain *sd; |
Rusty Russell | 96f874e2 | 2008-11-25 02:35:14 +1030 | [diff] [blame] | 1169 | struct cpumask *lowest_mask = __get_cpu_var(local_cpu_mask); |
Gregory Haskins | 6e1254d | 2008-01-25 21:08:11 +0100 | [diff] [blame] | 1170 | int this_cpu = smp_processor_id(); |
| 1171 | int cpu = task_cpu(task); |
| 1172 | |
Gregory Haskins | 6e0534f | 2008-05-12 21:21:01 +0200 | [diff] [blame] | 1173 | 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 Haskins | 06f90db | 2008-01-25 21:08:13 +0100 | [diff] [blame] | 1177 | return -1; /* No targets found */ |
| 1178 | |
| 1179 | /* |
Gregory Haskins | 6e1254d | 2008-01-25 21:08:11 +0100 | [diff] [blame] | 1180 | * 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 Russell | 96f874e2 | 2008-11-25 02:35:14 +1030 | [diff] [blame] | 1187 | if (cpumask_test_cpu(cpu, lowest_mask)) |
Gregory Haskins | 6e1254d | 2008-01-25 21:08:11 +0100 | [diff] [blame] | 1188 | 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 Russell | e2c8806 | 2009-11-03 14:53:15 +1030 | [diff] [blame] | 1194 | if (!cpumask_test_cpu(this_cpu, lowest_mask)) |
| 1195 | this_cpu = -1; /* Skip this_cpu opt if not among lowest */ |
Gregory Haskins | 6e1254d | 2008-01-25 21:08:11 +0100 | [diff] [blame] | 1196 | |
Rusty Russell | e2c8806 | 2009-11-03 14:53:15 +1030 | [diff] [blame] | 1197 | for_each_domain(cpu, sd) { |
| 1198 | if (sd->flags & SD_WAKE_AFFINE) { |
| 1199 | int best_cpu; |
Gregory Haskins | 6e1254d | 2008-01-25 21:08:11 +0100 | [diff] [blame] | 1200 | |
Rusty Russell | e2c8806 | 2009-11-03 14:53:15 +1030 | [diff] [blame] | 1201 | /* |
| 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 Haskins | 6e1254d | 2008-01-25 21:08:11 +0100 | [diff] [blame] | 1208 | |
Rusty Russell | e2c8806 | 2009-11-03 14:53:15 +1030 | [diff] [blame] | 1209 | best_cpu = cpumask_first_and(lowest_mask, |
| 1210 | sched_domain_span(sd)); |
| 1211 | if (best_cpu < nr_cpu_ids) |
| 1212 | return best_cpu; |
Gregory Haskins | 6e1254d | 2008-01-25 21:08:11 +0100 | [diff] [blame] | 1213 | } |
| 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 Russell | e2c8806 | 2009-11-03 14:53:15 +1030 | [diff] [blame] | 1221 | 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 Haskins | 07b4032 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 1228 | } |
| 1229 | |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1230 | /* Will lock the rq it finds */ |
Ingo Molnar | 4df64c0 | 2008-01-25 21:08:15 +0100 | [diff] [blame] | 1231 | static struct rq *find_lock_lowest_rq(struct task_struct *task, struct rq *rq) |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1232 | { |
| 1233 | struct rq *lowest_rq = NULL; |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1234 | int tries; |
Ingo Molnar | 4df64c0 | 2008-01-25 21:08:15 +0100 | [diff] [blame] | 1235 | int cpu; |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1236 | |
| 1237 | for (tries = 0; tries < RT_MAX_TRIES; tries++) { |
Gregory Haskins | 07b4032 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 1238 | cpu = find_lowest_rq(task); |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1239 | |
Gregory Haskins | 2de0b46 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 1240 | if ((cpu == -1) || (cpu == rq->cpu)) |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1241 | break; |
| 1242 | |
Gregory Haskins | 07b4032 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 1243 | lowest_rq = cpu_rq(cpu); |
| 1244 | |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1245 | /* if the prio of this runqueue changed, try again */ |
Gregory Haskins | 07b4032 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 1246 | if (double_lock_balance(rq, lowest_rq)) { |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1247 | /* |
| 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 Haskins | 07b4032 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 1253 | if (unlikely(task_rq(task) != rq || |
Rusty Russell | 96f874e2 | 2008-11-25 02:35:14 +1030 | [diff] [blame] | 1254 | !cpumask_test_cpu(lowest_rq->cpu, |
| 1255 | &task->cpus_allowed) || |
Gregory Haskins | 07b4032 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 1256 | task_running(rq, task) || |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1257 | !task->se.on_rq)) { |
Ingo Molnar | 4df64c0 | 2008-01-25 21:08:15 +0100 | [diff] [blame] | 1258 | |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 1259 | raw_spin_unlock(&lowest_rq->lock); |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1260 | lowest_rq = NULL; |
| 1261 | break; |
| 1262 | } |
| 1263 | } |
| 1264 | |
| 1265 | /* If this rq is still suitable use it. */ |
Gregory Haskins | e864c49 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 1266 | if (lowest_rq->rt.highest_prio.curr > task->prio) |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1267 | break; |
| 1268 | |
| 1269 | /* try again */ |
Peter Zijlstra | 1b12bbc | 2008-08-11 09:30:22 +0200 | [diff] [blame] | 1270 | double_unlock_balance(rq, lowest_rq); |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1271 | lowest_rq = NULL; |
| 1272 | } |
| 1273 | |
| 1274 | return lowest_rq; |
| 1275 | } |
| 1276 | |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 1277 | static 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 Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1297 | /* |
| 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 Haskins | 697f0a4 | 2008-01-25 21:08:09 +0100 | [diff] [blame] | 1302 | static int push_rt_task(struct rq *rq) |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1303 | { |
| 1304 | struct task_struct *next_task; |
| 1305 | struct rq *lowest_rq; |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1306 | |
Gregory Haskins | a22d7fc | 2008-01-25 21:08:12 +0100 | [diff] [blame] | 1307 | if (!rq->rt.overloaded) |
| 1308 | return 0; |
| 1309 | |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 1310 | next_task = pick_next_pushable_task(rq); |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1311 | if (!next_task) |
| 1312 | return 0; |
| 1313 | |
| 1314 | retry: |
Gregory Haskins | 697f0a4 | 2008-01-25 21:08:09 +0100 | [diff] [blame] | 1315 | if (unlikely(next_task == rq->curr)) { |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1316 | WARN_ON(1); |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1317 | return 0; |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1318 | } |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1319 | |
| 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 Haskins | 697f0a4 | 2008-01-25 21:08:09 +0100 | [diff] [blame] | 1325 | if (unlikely(next_task->prio < rq->curr->prio)) { |
| 1326 | resched_task(rq->curr); |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1327 | return 0; |
| 1328 | } |
| 1329 | |
Gregory Haskins | 697f0a4 | 2008-01-25 21:08:09 +0100 | [diff] [blame] | 1330 | /* We might release rq lock */ |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1331 | get_task_struct(next_task); |
| 1332 | |
| 1333 | /* find_lock_lowest_rq locks the rq if found */ |
Gregory Haskins | 697f0a4 | 2008-01-25 21:08:09 +0100 | [diff] [blame] | 1334 | lowest_rq = find_lock_lowest_rq(next_task, rq); |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1335 | if (!lowest_rq) { |
| 1336 | struct task_struct *task; |
| 1337 | /* |
Gregory Haskins | 697f0a4 | 2008-01-25 21:08:09 +0100 | [diff] [blame] | 1338 | * find lock_lowest_rq releases rq->lock |
Gregory Haskins | 1563513 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 1339 | * 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 Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1344 | */ |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 1345 | task = pick_next_pushable_task(rq); |
Gregory Haskins | 1563513 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 1346 | 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 Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1355 | } |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 1356 | |
Gregory Haskins | 1563513 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 1357 | if (!task) |
| 1358 | /* No more tasks, just exit */ |
| 1359 | goto out; |
| 1360 | |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 1361 | /* |
Gregory Haskins | 1563513 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 1362 | * Something has shifted, try again. |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 1363 | */ |
Gregory Haskins | 1563513 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 1364 | put_task_struct(next_task); |
| 1365 | next_task = task; |
| 1366 | goto retry; |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1367 | } |
| 1368 | |
Gregory Haskins | 697f0a4 | 2008-01-25 21:08:09 +0100 | [diff] [blame] | 1369 | deactivate_task(rq, next_task, 0); |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1370 | 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 Zijlstra | 1b12bbc | 2008-08-11 09:30:22 +0200 | [diff] [blame] | 1375 | double_unlock_balance(rq, lowest_rq); |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1376 | |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1377 | out: |
| 1378 | put_task_struct(next_task); |
| 1379 | |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 1380 | return 1; |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1381 | } |
| 1382 | |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1383 | static 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 Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1390 | static int pull_rt_task(struct rq *this_rq) |
| 1391 | { |
Ingo Molnar | 80bf317 | 2008-01-25 21:08:17 +0100 | [diff] [blame] | 1392 | int this_cpu = this_rq->cpu, ret = 0, cpu; |
Gregory Haskins | a872894 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 1393 | struct task_struct *p; |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1394 | struct rq *src_rq; |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1395 | |
Gregory Haskins | 637f508 | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 1396 | if (likely(!rt_overloaded(this_rq))) |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1397 | return 0; |
| 1398 | |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 1399 | for_each_cpu(cpu, this_rq->rd->rto_mask) { |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1400 | if (this_cpu == cpu) |
| 1401 | continue; |
| 1402 | |
| 1403 | src_rq = cpu_rq(cpu); |
Gregory Haskins | 74ab8e4 | 2008-12-29 09:39:50 -0500 | [diff] [blame] | 1404 | |
| 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 Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1416 | /* |
| 1417 | * We can potentially drop this_rq's lock in |
| 1418 | * double_lock_balance, and another CPU could |
Gregory Haskins | a872894 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 1419 | * alter this_rq |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1420 | */ |
Gregory Haskins | a872894 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 1421 | double_lock_balance(this_rq, src_rq); |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1422 | |
| 1423 | /* |
| 1424 | * Are there still pullable RT tasks? |
| 1425 | */ |
Mike Galbraith | 614ee1f | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1426 | if (src_rq->rt.rt_nr_running <= 1) |
| 1427 | goto skip; |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1428 | |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1429 | 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 Haskins | a872894 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 1435 | if (p && (p->prio < this_rq->rt.highest_prio.curr)) { |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1436 | 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 Haskins | a872894 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 1445 | * current task on the run queue |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1446 | */ |
Gregory Haskins | a872894 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 1447 | if (p->prio < src_rq->curr->prio) |
Mike Galbraith | 614ee1f | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1448 | goto skip; |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1449 | |
| 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 Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1460 | */ |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1461 | } |
Mike Galbraith | 614ee1f | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1462 | skip: |
Peter Zijlstra | 1b12bbc | 2008-08-11 09:30:22 +0200 | [diff] [blame] | 1463 | double_unlock_balance(this_rq, src_rq); |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1464 | } |
| 1465 | |
| 1466 | return ret; |
| 1467 | } |
| 1468 | |
Steven Rostedt | 9a897c5 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 1469 | static void pre_schedule_rt(struct rq *rq, struct task_struct *prev) |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1470 | { |
| 1471 | /* Try to pull RT tasks here if we lower this rq's prio */ |
Gregory Haskins | e864c49 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 1472 | if (unlikely(rt_task(prev)) && rq->rt.highest_prio.curr > prev->prio) |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1473 | pull_rt_task(rq); |
| 1474 | } |
| 1475 | |
Steven Rostedt | 9a897c5 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 1476 | static void post_schedule_rt(struct rq *rq) |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1477 | { |
Gregory Haskins | 967fc04 | 2008-12-29 09:39:52 -0500 | [diff] [blame] | 1478 | push_rt_tasks(rq); |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1479 | } |
| 1480 | |
Gregory Haskins | 8ae121a | 2008-04-23 07:13:29 -0400 | [diff] [blame] | 1481 | /* |
| 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 Zijlstra | efbbd05 | 2009-12-16 18:04:40 +0100 | [diff] [blame] | 1485 | static void task_woken_rt(struct rq *rq, struct task_struct *p) |
Steven Rostedt | 4642daf | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1486 | { |
Steven Rostedt | 9a897c5 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 1487 | if (!task_running(rq, p) && |
Gregory Haskins | 8ae121a | 2008-04-23 07:13:29 -0400 | [diff] [blame] | 1488 | !test_tsk_need_resched(rq->curr) && |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 1489 | has_pushable_tasks(rq) && |
Gregory Haskins | 777c2f3 | 2008-12-29 09:39:50 -0500 | [diff] [blame] | 1490 | p->rt.nr_cpus_allowed > 1) |
Steven Rostedt | 4642daf | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1491 | push_rt_tasks(rq); |
| 1492 | } |
| 1493 | |
Mike Travis | cd8ba7c | 2008-03-26 14:23:49 -0700 | [diff] [blame] | 1494 | static void set_cpus_allowed_rt(struct task_struct *p, |
Rusty Russell | 96f874e2 | 2008-11-25 02:35:14 +1030 | [diff] [blame] | 1495 | const struct cpumask *new_mask) |
Gregory Haskins | 73fe6aae | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1496 | { |
Rusty Russell | 96f874e2 | 2008-11-25 02:35:14 +1030 | [diff] [blame] | 1497 | int weight = cpumask_weight(new_mask); |
Gregory Haskins | 73fe6aae | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1498 | |
| 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 Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1505 | if (p->se.on_rq && (weight != p->rt.nr_cpus_allowed)) { |
Gregory Haskins | 73fe6aae | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1506 | struct rq *rq = task_rq(p); |
| 1507 | |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 1508 | 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 Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1526 | if ((p->rt.nr_cpus_allowed <= 1) && (weight > 1)) { |
Gregory Haskins | 73fe6aae | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1527 | rq->rt.rt_nr_migratory++; |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1528 | } else if ((p->rt.nr_cpus_allowed > 1) && (weight <= 1)) { |
Gregory Haskins | 73fe6aae | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1529 | BUG_ON(!rq->rt.rt_nr_migratory); |
| 1530 | rq->rt.rt_nr_migratory--; |
| 1531 | } |
| 1532 | |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 1533 | update_rt_migration(&rq->rt); |
Gregory Haskins | 73fe6aae | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1534 | } |
| 1535 | |
Rusty Russell | 96f874e2 | 2008-11-25 02:35:14 +1030 | [diff] [blame] | 1536 | cpumask_copy(&p->cpus_allowed, new_mask); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1537 | p->rt.nr_cpus_allowed = weight; |
Gregory Haskins | 73fe6aae | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1538 | } |
Ingo Molnar | deeeccd | 2008-01-25 21:08:15 +0100 | [diff] [blame] | 1539 | |
Ingo Molnar | bdd7c81 | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 1540 | /* Assumes rq->lock is held */ |
Gregory Haskins | 1f11eb6a | 2008-06-04 15:04:05 -0400 | [diff] [blame] | 1541 | static void rq_online_rt(struct rq *rq) |
Ingo Molnar | bdd7c81 | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 1542 | { |
| 1543 | if (rq->rt.overloaded) |
| 1544 | rt_set_overload(rq); |
Gregory Haskins | 6e0534f | 2008-05-12 21:21:01 +0200 | [diff] [blame] | 1545 | |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 1546 | __enable_runtime(rq); |
| 1547 | |
Gregory Haskins | e864c49 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 1548 | cpupri_set(&rq->rd->cpupri, rq->cpu, rq->rt.highest_prio.curr); |
Ingo Molnar | bdd7c81 | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 1549 | } |
| 1550 | |
| 1551 | /* Assumes rq->lock is held */ |
Gregory Haskins | 1f11eb6a | 2008-06-04 15:04:05 -0400 | [diff] [blame] | 1552 | static void rq_offline_rt(struct rq *rq) |
Ingo Molnar | bdd7c81 | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 1553 | { |
| 1554 | if (rq->rt.overloaded) |
| 1555 | rt_clear_overload(rq); |
Gregory Haskins | 6e0534f | 2008-05-12 21:21:01 +0200 | [diff] [blame] | 1556 | |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 1557 | __disable_runtime(rq); |
| 1558 | |
Gregory Haskins | 6e0534f | 2008-05-12 21:21:01 +0200 | [diff] [blame] | 1559 | cpupri_set(&rq->rd->cpupri, rq->cpu, CPUPRI_INVALID); |
Ingo Molnar | bdd7c81 | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 1560 | } |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 1561 | |
| 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 | */ |
| 1566 | static 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 Russell | 3d8cbdf | 2008-11-25 09:58:41 +1030 | [diff] [blame] | 1579 | |
| 1580 | static inline void init_sched_rt_class(void) |
| 1581 | { |
| 1582 | unsigned int i; |
| 1583 | |
| 1584 | for_each_possible_cpu(i) |
Yinghai Lu | eaa9584 | 2009-06-06 14:51:36 -0700 | [diff] [blame] | 1585 | zalloc_cpumask_var_node(&per_cpu(local_cpu_mask, i), |
Mike Travis | 6ca09df | 2008-12-31 18:08:45 -0800 | [diff] [blame] | 1586 | GFP_KERNEL, cpu_to_node(i)); |
Rusty Russell | 3d8cbdf | 2008-11-25 09:58:41 +1030 | [diff] [blame] | 1587 | } |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1588 | #endif /* CONFIG_SMP */ |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1589 | |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 1590 | /* |
| 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 | */ |
| 1595 | static 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 | */ |
| 1623 | static 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 Rostedt | 6fa46fa | 2008-03-05 10:00:12 -0500 | [diff] [blame] | 1636 | * 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 Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 1639 | */ |
Gregory Haskins | e864c49 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 1640 | if (p->prio > rq->rt.highest_prio.curr && rq->curr == p) |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 1641 | 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 Zijlstra | 78f2c7d | 2008-01-25 21:08:27 +0100 | [diff] [blame] | 1658 | static 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 Zijlstra | 5a52dd5 | 2008-01-25 21:08:32 +0100 | [diff] [blame] | 1673 | if (p->rt.timeout > next) |
Frank Mayhar | f06febc | 2008-09-12 09:54:39 -0700 | [diff] [blame] | 1674 | p->cputime_expires.sched_exp = p->se.sum_exec_runtime; |
Peter Zijlstra | 78f2c7d | 2008-01-25 21:08:27 +0100 | [diff] [blame] | 1675 | } |
| 1676 | } |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 1677 | |
Peter Zijlstra | 8f4d37e | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 1678 | static void task_tick_rt(struct rq *rq, struct task_struct *p, int queued) |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1679 | { |
Peter Zijlstra | 67e2be0 | 2007-12-20 15:01:17 +0100 | [diff] [blame] | 1680 | update_curr_rt(rq); |
| 1681 | |
Peter Zijlstra | 78f2c7d | 2008-01-25 21:08:27 +0100 | [diff] [blame] | 1682 | watchdog(rq, p); |
| 1683 | |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1684 | /* |
| 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 Zijlstra | fa71706 | 2008-01-25 21:08:27 +0100 | [diff] [blame] | 1691 | if (--p->rt.time_slice) |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1692 | return; |
| 1693 | |
Peter Zijlstra | fa71706 | 2008-01-25 21:08:27 +0100 | [diff] [blame] | 1694 | p->rt.time_slice = DEF_TIMESLICE; |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1695 | |
Dmitry Adamushko | 98fbc79 | 2007-08-24 20:39:10 +0200 | [diff] [blame] | 1696 | /* |
| 1697 | * Requeue to the end of queue if we are not the only element |
| 1698 | * on the queue: |
| 1699 | */ |
Peter Zijlstra | fa71706 | 2008-01-25 21:08:27 +0100 | [diff] [blame] | 1700 | if (p->rt.run_list.prev != p->rt.run_list.next) { |
Dmitry Adamushko | 7ebefa8 | 2008-07-01 23:32:15 +0200 | [diff] [blame] | 1701 | requeue_task_rt(rq, p, 0); |
Dmitry Adamushko | 98fbc79 | 2007-08-24 20:39:10 +0200 | [diff] [blame] | 1702 | set_tsk_need_resched(p); |
| 1703 | } |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1704 | } |
| 1705 | |
Srivatsa Vaddagiri | 83b699e | 2007-10-15 17:00:08 +0200 | [diff] [blame] | 1706 | static 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 Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 1711 | |
| 1712 | /* The running task is never eligible for pushing */ |
| 1713 | dequeue_pushable_task(rq, p); |
Srivatsa Vaddagiri | 83b699e | 2007-10-15 17:00:08 +0200 | [diff] [blame] | 1714 | } |
| 1715 | |
H Hartley Sweeten | 6d686f4 | 2010-01-13 20:21:52 -0700 | [diff] [blame] | 1716 | static unsigned int get_rr_interval_rt(struct rq *rq, struct task_struct *task) |
Peter Williams | 0d721ce | 2009-09-21 01:31:53 +0000 | [diff] [blame] | 1717 | { |
| 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 Harrison | 2abdad0 | 2008-04-25 10:53:13 -0700 | [diff] [blame] | 1727 | static const struct sched_class rt_sched_class = { |
Ingo Molnar | 5522d5d | 2007-10-15 17:00:12 +0200 | [diff] [blame] | 1728 | .next = &fair_sched_class, |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1729 | .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 Williams | 681f3e6 | 2007-10-24 18:23:51 +0200 | [diff] [blame] | 1738 | #ifdef CONFIG_SMP |
Li Zefan | 4ce72a2 | 2008-10-22 15:25:26 +0800 | [diff] [blame] | 1739 | .select_task_rq = select_task_rq_rt, |
| 1740 | |
Gregory Haskins | 73fe6aae | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1741 | .set_cpus_allowed = set_cpus_allowed_rt, |
Gregory Haskins | 1f11eb6a | 2008-06-04 15:04:05 -0400 | [diff] [blame] | 1742 | .rq_online = rq_online_rt, |
| 1743 | .rq_offline = rq_offline_rt, |
Steven Rostedt | 9a897c5 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 1744 | .pre_schedule = pre_schedule_rt, |
| 1745 | .post_schedule = post_schedule_rt, |
Peter Zijlstra | efbbd05 | 2009-12-16 18:04:40 +0100 | [diff] [blame] | 1746 | .task_woken = task_woken_rt, |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 1747 | .switched_from = switched_from_rt, |
Peter Williams | 681f3e6 | 2007-10-24 18:23:51 +0200 | [diff] [blame] | 1748 | #endif |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1749 | |
Srivatsa Vaddagiri | 83b699e | 2007-10-15 17:00:08 +0200 | [diff] [blame] | 1750 | .set_curr_task = set_curr_task_rt, |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1751 | .task_tick = task_tick_rt, |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 1752 | |
Peter Williams | 0d721ce | 2009-09-21 01:31:53 +0000 | [diff] [blame] | 1753 | .get_rr_interval = get_rr_interval_rt, |
| 1754 | |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 1755 | .prio_changed = prio_changed_rt, |
| 1756 | .switched_to = switched_to_rt, |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1757 | }; |
Peter Zijlstra | ada18de | 2008-06-19 14:22:24 +0200 | [diff] [blame] | 1758 | |
| 1759 | #ifdef CONFIG_SCHED_DEBUG |
| 1760 | extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq); |
| 1761 | |
| 1762 | static 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 Giani | 55e12e5 | 2008-06-24 23:39:43 +0530 | [diff] [blame] | 1771 | #endif /* CONFIG_SCHED_DEBUG */ |
Rusty Russell | 0e3900e | 2008-11-25 02:35:13 +1030 | [diff] [blame] | 1772 | |