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 | |
Steven Rostedt | 4fd2917 | 2008-01-25 21:08:06 +0100 | [diff] [blame] | 6 | #ifdef CONFIG_SMP |
Ingo Molnar | 84de427 | 2008-01-25 21:08:15 +0100 | [diff] [blame] | 7 | |
Gregory Haskins | 637f508 | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 8 | static inline int rt_overloaded(struct rq *rq) |
Steven Rostedt | 4fd2917 | 2008-01-25 21:08:06 +0100 | [diff] [blame] | 9 | { |
Gregory Haskins | 637f508 | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 10 | return atomic_read(&rq->rd->rto_count); |
Steven Rostedt | 4fd2917 | 2008-01-25 21:08:06 +0100 | [diff] [blame] | 11 | } |
Ingo Molnar | 84de427 | 2008-01-25 21:08:15 +0100 | [diff] [blame] | 12 | |
Steven Rostedt | 4fd2917 | 2008-01-25 21:08:06 +0100 | [diff] [blame] | 13 | static inline void rt_set_overload(struct rq *rq) |
| 14 | { |
Gregory Haskins | 637f508 | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 15 | cpu_set(rq->cpu, rq->rd->rto_mask); |
Steven Rostedt | 4fd2917 | 2008-01-25 21:08:06 +0100 | [diff] [blame] | 16 | /* |
| 17 | * Make sure the mask is visible before we set |
| 18 | * the overload count. That is checked to determine |
| 19 | * if we should look at the mask. It would be a shame |
| 20 | * if we looked at the mask, but the mask was not |
| 21 | * updated yet. |
| 22 | */ |
| 23 | wmb(); |
Gregory Haskins | 637f508 | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 24 | atomic_inc(&rq->rd->rto_count); |
Steven Rostedt | 4fd2917 | 2008-01-25 21:08:06 +0100 | [diff] [blame] | 25 | } |
Ingo Molnar | 84de427 | 2008-01-25 21:08:15 +0100 | [diff] [blame] | 26 | |
Steven Rostedt | 4fd2917 | 2008-01-25 21:08:06 +0100 | [diff] [blame] | 27 | static inline void rt_clear_overload(struct rq *rq) |
| 28 | { |
| 29 | /* the order here really doesn't matter */ |
Gregory Haskins | 637f508 | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 30 | atomic_dec(&rq->rd->rto_count); |
| 31 | cpu_clear(rq->cpu, rq->rd->rto_mask); |
Steven Rostedt | 4fd2917 | 2008-01-25 21:08:06 +0100 | [diff] [blame] | 32 | } |
Gregory Haskins | 73fe6aa | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 33 | |
| 34 | static void update_rt_migration(struct rq *rq) |
| 35 | { |
Gregory Haskins | 637f508 | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 36 | if (rq->rt.rt_nr_migratory && (rq->rt.rt_nr_running > 1)) { |
Gregory Haskins | cdc8eb9 | 2008-01-25 21:08:23 +0100 | [diff] [blame] | 37 | if (!rq->rt.overloaded) { |
| 38 | rt_set_overload(rq); |
| 39 | rq->rt.overloaded = 1; |
| 40 | } |
| 41 | } else if (rq->rt.overloaded) { |
Gregory Haskins | 73fe6aa | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 42 | rt_clear_overload(rq); |
Gregory Haskins | 637f508 | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 43 | rq->rt.overloaded = 0; |
| 44 | } |
Gregory Haskins | 73fe6aa | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 45 | } |
Steven Rostedt | 4fd2917 | 2008-01-25 21:08:06 +0100 | [diff] [blame] | 46 | #endif /* CONFIG_SMP */ |
| 47 | |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 48 | static inline struct task_struct *rt_task_of(struct sched_rt_entity *rt_se) |
Peter Zijlstra | fa85ae2 | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 49 | { |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 50 | return container_of(rt_se, struct task_struct, rt); |
| 51 | } |
| 52 | |
| 53 | static inline int on_rt_rq(struct sched_rt_entity *rt_se) |
| 54 | { |
| 55 | return !list_empty(&rt_se->run_list); |
| 56 | } |
| 57 | |
Peter Zijlstra | 052f1dc | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 58 | #ifdef CONFIG_RT_GROUP_SCHED |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 59 | |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 60 | static inline u64 sched_rt_runtime(struct rt_rq *rt_rq) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 61 | { |
| 62 | if (!rt_rq->tg) |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 63 | return RUNTIME_INF; |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 64 | |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 65 | return rt_rq->tg->rt_runtime; |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 66 | } |
| 67 | |
| 68 | #define for_each_leaf_rt_rq(rt_rq, rq) \ |
| 69 | list_for_each_entry(rt_rq, &rq->leaf_rt_rq_list, leaf_rt_rq_list) |
| 70 | |
| 71 | static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq) |
| 72 | { |
| 73 | return rt_rq->rq; |
| 74 | } |
| 75 | |
| 76 | static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se) |
| 77 | { |
| 78 | return rt_se->rt_rq; |
| 79 | } |
| 80 | |
| 81 | #define for_each_sched_rt_entity(rt_se) \ |
| 82 | for (; rt_se; rt_se = rt_se->parent) |
| 83 | |
| 84 | static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se) |
| 85 | { |
| 86 | return rt_se->my_q; |
| 87 | } |
| 88 | |
| 89 | static void enqueue_rt_entity(struct sched_rt_entity *rt_se); |
| 90 | static void dequeue_rt_entity(struct sched_rt_entity *rt_se); |
| 91 | |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 92 | static void sched_rt_rq_enqueue(struct rt_rq *rt_rq) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 93 | { |
| 94 | struct sched_rt_entity *rt_se = rt_rq->rt_se; |
| 95 | |
| 96 | if (rt_se && !on_rt_rq(rt_se) && rt_rq->rt_nr_running) { |
Peter Zijlstra | 1020387 | 2008-01-25 21:08:32 +0100 | [diff] [blame] | 97 | struct task_struct *curr = rq_of_rt_rq(rt_rq)->curr; |
| 98 | |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 99 | enqueue_rt_entity(rt_se); |
Peter Zijlstra | 1020387 | 2008-01-25 21:08:32 +0100 | [diff] [blame] | 100 | if (rt_rq->highest_prio < curr->prio) |
| 101 | resched_task(curr); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 102 | } |
| 103 | } |
| 104 | |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 105 | static void sched_rt_rq_dequeue(struct rt_rq *rt_rq) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 106 | { |
| 107 | struct sched_rt_entity *rt_se = rt_rq->rt_se; |
| 108 | |
| 109 | if (rt_se && on_rt_rq(rt_se)) |
| 110 | dequeue_rt_entity(rt_se); |
| 111 | } |
| 112 | |
Peter Zijlstra | 23b0fdf | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 113 | static inline int rt_rq_throttled(struct rt_rq *rt_rq) |
| 114 | { |
| 115 | return rt_rq->rt_throttled && !rt_rq->rt_nr_boosted; |
| 116 | } |
| 117 | |
| 118 | static int rt_se_boosted(struct sched_rt_entity *rt_se) |
| 119 | { |
| 120 | struct rt_rq *rt_rq = group_rt_rq(rt_se); |
| 121 | struct task_struct *p; |
| 122 | |
| 123 | if (rt_rq) |
| 124 | return !!rt_rq->rt_nr_boosted; |
| 125 | |
| 126 | p = rt_task_of(rt_se); |
| 127 | return p->prio != p->normal_prio; |
| 128 | } |
| 129 | |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 130 | #else |
| 131 | |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 132 | static inline u64 sched_rt_runtime(struct rt_rq *rt_rq) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 133 | { |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 134 | if (sysctl_sched_rt_runtime == -1) |
| 135 | return RUNTIME_INF; |
| 136 | |
| 137 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 138 | } |
| 139 | |
| 140 | #define for_each_leaf_rt_rq(rt_rq, rq) \ |
| 141 | for (rt_rq = &rq->rt; rt_rq; rt_rq = NULL) |
| 142 | |
| 143 | static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq) |
| 144 | { |
| 145 | return container_of(rt_rq, struct rq, rt); |
| 146 | } |
| 147 | |
| 148 | static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se) |
| 149 | { |
| 150 | struct task_struct *p = rt_task_of(rt_se); |
| 151 | struct rq *rq = task_rq(p); |
| 152 | |
| 153 | return &rq->rt; |
| 154 | } |
| 155 | |
| 156 | #define for_each_sched_rt_entity(rt_se) \ |
| 157 | for (; rt_se; rt_se = NULL) |
| 158 | |
| 159 | static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se) |
| 160 | { |
| 161 | return NULL; |
| 162 | } |
| 163 | |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 164 | static inline void sched_rt_rq_enqueue(struct rt_rq *rt_rq) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 165 | { |
| 166 | } |
| 167 | |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 168 | static inline void sched_rt_rq_dequeue(struct rt_rq *rt_rq) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 169 | { |
| 170 | } |
| 171 | |
Peter Zijlstra | 23b0fdf | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 172 | static inline int rt_rq_throttled(struct rt_rq *rt_rq) |
| 173 | { |
| 174 | return rt_rq->rt_throttled; |
| 175 | } |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 176 | #endif |
| 177 | |
| 178 | static inline int rt_se_prio(struct sched_rt_entity *rt_se) |
| 179 | { |
Peter Zijlstra | 052f1dc | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 180 | #ifdef CONFIG_RT_GROUP_SCHED |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 181 | struct rt_rq *rt_rq = group_rt_rq(rt_se); |
| 182 | |
| 183 | if (rt_rq) |
| 184 | return rt_rq->highest_prio; |
| 185 | #endif |
| 186 | |
| 187 | return rt_task_of(rt_se)->prio; |
| 188 | } |
| 189 | |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 190 | static int sched_rt_runtime_exceeded(struct rt_rq *rt_rq) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 191 | { |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 192 | u64 runtime = sched_rt_runtime(rt_rq); |
Peter Zijlstra | fa85ae2 | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 193 | |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 194 | if (runtime == RUNTIME_INF) |
Peter Zijlstra | fa85ae2 | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 195 | return 0; |
| 196 | |
| 197 | if (rt_rq->rt_throttled) |
Peter Zijlstra | 23b0fdf | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 198 | return rt_rq_throttled(rt_rq); |
Peter Zijlstra | fa85ae2 | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 199 | |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 200 | if (rt_rq->rt_time > runtime) { |
Peter Zijlstra | 48d5e25 | 2008-01-25 21:08:31 +0100 | [diff] [blame] | 201 | struct rq *rq = rq_of_rt_rq(rt_rq); |
| 202 | |
| 203 | rq->rt_throttled = 1; |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 204 | rt_rq->rt_throttled = 1; |
Peter Zijlstra | 48d5e25 | 2008-01-25 21:08:31 +0100 | [diff] [blame] | 205 | |
Peter Zijlstra | 23b0fdf | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 206 | if (rt_rq_throttled(rt_rq)) { |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 207 | sched_rt_rq_dequeue(rt_rq); |
Peter Zijlstra | 23b0fdf | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 208 | return 1; |
| 209 | } |
Peter Zijlstra | fa85ae2 | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 210 | } |
| 211 | |
| 212 | return 0; |
| 213 | } |
| 214 | |
| 215 | static void update_sched_rt_period(struct rq *rq) |
| 216 | { |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 217 | struct rt_rq *rt_rq; |
| 218 | u64 period; |
| 219 | |
Peter Zijlstra | fa85ae2 | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 220 | while (rq->clock > rq->rt_period_expire) { |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 221 | period = (u64)sysctl_sched_rt_period * NSEC_PER_USEC; |
Peter Zijlstra | fa85ae2 | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 222 | rq->rt_period_expire += period; |
Peter Zijlstra | fa85ae2 | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 223 | |
Peter Zijlstra | 48d5e25 | 2008-01-25 21:08:31 +0100 | [diff] [blame] | 224 | for_each_leaf_rt_rq(rt_rq, rq) { |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 225 | u64 runtime = sched_rt_runtime(rt_rq); |
Peter Zijlstra | 48d5e25 | 2008-01-25 21:08:31 +0100 | [diff] [blame] | 226 | |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 227 | rt_rq->rt_time -= min(rt_rq->rt_time, runtime); |
| 228 | if (rt_rq->rt_throttled && rt_rq->rt_time < runtime) { |
Peter Zijlstra | 48d5e25 | 2008-01-25 21:08:31 +0100 | [diff] [blame] | 229 | rt_rq->rt_throttled = 0; |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 230 | sched_rt_rq_enqueue(rt_rq); |
Peter Zijlstra | 48d5e25 | 2008-01-25 21:08:31 +0100 | [diff] [blame] | 231 | } |
| 232 | } |
| 233 | |
| 234 | rq->rt_throttled = 0; |
Peter Zijlstra | fa85ae2 | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 235 | } |
| 236 | } |
| 237 | |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 238 | /* |
| 239 | * Update the current task's runtime statistics. Skip current tasks that |
| 240 | * are not in our scheduling class. |
| 241 | */ |
Alexey Dobriyan | a995744 | 2007-10-15 17:00:13 +0200 | [diff] [blame] | 242 | static void update_curr_rt(struct rq *rq) |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 243 | { |
| 244 | struct task_struct *curr = rq->curr; |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 245 | struct sched_rt_entity *rt_se = &curr->rt; |
| 246 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 247 | u64 delta_exec; |
| 248 | |
| 249 | if (!task_has_rt_policy(curr)) |
| 250 | return; |
| 251 | |
Ingo Molnar | d281918 | 2007-08-09 11:16:47 +0200 | [diff] [blame] | 252 | delta_exec = rq->clock - curr->se.exec_start; |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 253 | if (unlikely((s64)delta_exec < 0)) |
| 254 | delta_exec = 0; |
Ingo Molnar | 6cfb0d5 | 2007-08-02 17:41:40 +0200 | [diff] [blame] | 255 | |
| 256 | 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] | 257 | |
| 258 | curr->se.sum_exec_runtime += delta_exec; |
Ingo Molnar | d281918 | 2007-08-09 11:16:47 +0200 | [diff] [blame] | 259 | curr->se.exec_start = rq->clock; |
Srivatsa Vaddagiri | d842de8 | 2007-12-02 20:04:49 +0100 | [diff] [blame] | 260 | cpuacct_charge(curr, delta_exec); |
Peter Zijlstra | fa85ae2 | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 261 | |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 262 | rt_rq->rt_time += delta_exec; |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 263 | if (sched_rt_runtime_exceeded(rt_rq)) |
Peter Zijlstra | fa85ae2 | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 264 | resched_task(curr); |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 265 | } |
| 266 | |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 267 | static inline |
| 268 | void inc_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
Steven Rostedt | 63489e4 | 2008-01-25 21:08:03 +0100 | [diff] [blame] | 269 | { |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 270 | WARN_ON(!rt_prio(rt_se_prio(rt_se))); |
| 271 | rt_rq->rt_nr_running++; |
Peter Zijlstra | 052f1dc | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 272 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 273 | if (rt_se_prio(rt_se) < rt_rq->highest_prio) |
| 274 | rt_rq->highest_prio = rt_se_prio(rt_se); |
| 275 | #endif |
Steven Rostedt | 764a9d6 | 2008-01-25 21:08:04 +0100 | [diff] [blame] | 276 | #ifdef CONFIG_SMP |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 277 | if (rt_se->nr_cpus_allowed > 1) { |
| 278 | struct rq *rq = rq_of_rt_rq(rt_rq); |
Gregory Haskins | 73fe6aa | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 279 | rq->rt.rt_nr_migratory++; |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 280 | } |
Gregory Haskins | 73fe6aa | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 281 | |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 282 | update_rt_migration(rq_of_rt_rq(rt_rq)); |
| 283 | #endif |
Peter Zijlstra | 052f1dc | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 284 | #ifdef CONFIG_RT_GROUP_SCHED |
Peter Zijlstra | 23b0fdf | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 285 | if (rt_se_boosted(rt_se)) |
| 286 | rt_rq->rt_nr_boosted++; |
| 287 | #endif |
Steven Rostedt | 63489e4 | 2008-01-25 21:08:03 +0100 | [diff] [blame] | 288 | } |
| 289 | |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 290 | static inline |
| 291 | void dec_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
Steven Rostedt | 63489e4 | 2008-01-25 21:08:03 +0100 | [diff] [blame] | 292 | { |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 293 | WARN_ON(!rt_prio(rt_se_prio(rt_se))); |
| 294 | WARN_ON(!rt_rq->rt_nr_running); |
| 295 | rt_rq->rt_nr_running--; |
Peter Zijlstra | 052f1dc | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 296 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 297 | if (rt_rq->rt_nr_running) { |
Steven Rostedt | 764a9d6 | 2008-01-25 21:08:04 +0100 | [diff] [blame] | 298 | struct rt_prio_array *array; |
| 299 | |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 300 | WARN_ON(rt_se_prio(rt_se) < rt_rq->highest_prio); |
| 301 | if (rt_se_prio(rt_se) == rt_rq->highest_prio) { |
Steven Rostedt | 764a9d6 | 2008-01-25 21:08:04 +0100 | [diff] [blame] | 302 | /* recalculate */ |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 303 | array = &rt_rq->active; |
| 304 | rt_rq->highest_prio = |
Steven Rostedt | 764a9d6 | 2008-01-25 21:08:04 +0100 | [diff] [blame] | 305 | sched_find_first_bit(array->bitmap); |
| 306 | } /* otherwise leave rq->highest prio alone */ |
| 307 | } else |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 308 | rt_rq->highest_prio = MAX_RT_PRIO; |
| 309 | #endif |
| 310 | #ifdef CONFIG_SMP |
| 311 | if (rt_se->nr_cpus_allowed > 1) { |
| 312 | struct rq *rq = rq_of_rt_rq(rt_rq); |
Gregory Haskins | 73fe6aa | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 313 | rq->rt.rt_nr_migratory--; |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 314 | } |
Gregory Haskins | 73fe6aa | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 315 | |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 316 | update_rt_migration(rq_of_rt_rq(rt_rq)); |
Steven Rostedt | 764a9d6 | 2008-01-25 21:08:04 +0100 | [diff] [blame] | 317 | #endif /* CONFIG_SMP */ |
Peter Zijlstra | 052f1dc | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 318 | #ifdef CONFIG_RT_GROUP_SCHED |
Peter Zijlstra | 23b0fdf | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 319 | if (rt_se_boosted(rt_se)) |
| 320 | rt_rq->rt_nr_boosted--; |
| 321 | |
| 322 | WARN_ON(!rt_rq->rt_nr_running && rt_rq->rt_nr_boosted); |
| 323 | #endif |
Steven Rostedt | 63489e4 | 2008-01-25 21:08:03 +0100 | [diff] [blame] | 324 | } |
| 325 | |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 326 | static void enqueue_rt_entity(struct sched_rt_entity *rt_se) |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 327 | { |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 328 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); |
| 329 | struct rt_prio_array *array = &rt_rq->active; |
| 330 | struct rt_rq *group_rq = group_rt_rq(rt_se); |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 331 | |
Peter Zijlstra | 23b0fdf | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 332 | if (group_rq && rt_rq_throttled(group_rq)) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 333 | return; |
Steven Rostedt | 63489e4 | 2008-01-25 21:08:03 +0100 | [diff] [blame] | 334 | |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 335 | list_add_tail(&rt_se->run_list, array->queue + rt_se_prio(rt_se)); |
| 336 | __set_bit(rt_se_prio(rt_se), array->bitmap); |
Peter Zijlstra | 78f2c7d | 2008-01-25 21:08:27 +0100 | [diff] [blame] | 337 | |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 338 | inc_rt_tasks(rt_se, rt_rq); |
| 339 | } |
| 340 | |
| 341 | static void dequeue_rt_entity(struct sched_rt_entity *rt_se) |
| 342 | { |
| 343 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); |
| 344 | struct rt_prio_array *array = &rt_rq->active; |
| 345 | |
| 346 | list_del_init(&rt_se->run_list); |
| 347 | if (list_empty(array->queue + rt_se_prio(rt_se))) |
| 348 | __clear_bit(rt_se_prio(rt_se), array->bitmap); |
| 349 | |
| 350 | dec_rt_tasks(rt_se, rt_rq); |
| 351 | } |
| 352 | |
| 353 | /* |
| 354 | * Because the prio of an upper entry depends on the lower |
| 355 | * entries, we must remove entries top - down. |
| 356 | * |
| 357 | * XXX: O(1/2 h^2) because we can only walk up, not down the chain. |
| 358 | * doesn't matter much for now, as h=2 for GROUP_SCHED. |
| 359 | */ |
| 360 | static void dequeue_rt_stack(struct task_struct *p) |
| 361 | { |
| 362 | struct sched_rt_entity *rt_se, *top_se; |
| 363 | |
| 364 | /* |
| 365 | * dequeue all, top - down. |
| 366 | */ |
| 367 | do { |
| 368 | rt_se = &p->rt; |
| 369 | top_se = NULL; |
| 370 | for_each_sched_rt_entity(rt_se) { |
| 371 | if (on_rt_rq(rt_se)) |
| 372 | top_se = rt_se; |
| 373 | } |
| 374 | if (top_se) |
| 375 | dequeue_rt_entity(top_se); |
| 376 | } while (top_se); |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 377 | } |
| 378 | |
| 379 | /* |
| 380 | * Adding/removing a task to/from a priority array: |
| 381 | */ |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 382 | static void enqueue_task_rt(struct rq *rq, struct task_struct *p, int wakeup) |
| 383 | { |
| 384 | struct sched_rt_entity *rt_se = &p->rt; |
| 385 | |
| 386 | if (wakeup) |
| 387 | rt_se->timeout = 0; |
| 388 | |
| 389 | dequeue_rt_stack(p); |
| 390 | |
| 391 | /* |
| 392 | * enqueue everybody, bottom - up. |
| 393 | */ |
| 394 | for_each_sched_rt_entity(rt_se) |
| 395 | enqueue_rt_entity(rt_se); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 396 | } |
| 397 | |
Ingo Molnar | f02231e | 2007-08-09 11:16:48 +0200 | [diff] [blame] | 398 | static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int sleep) |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 399 | { |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 400 | struct sched_rt_entity *rt_se = &p->rt; |
| 401 | struct rt_rq *rt_rq; |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 402 | |
Ingo Molnar | f1e14ef | 2007-08-09 11:16:48 +0200 | [diff] [blame] | 403 | update_curr_rt(rq); |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 404 | |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 405 | dequeue_rt_stack(p); |
Steven Rostedt | 63489e4 | 2008-01-25 21:08:03 +0100 | [diff] [blame] | 406 | |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 407 | /* |
| 408 | * re-enqueue all non-empty rt_rq entities. |
| 409 | */ |
| 410 | for_each_sched_rt_entity(rt_se) { |
| 411 | rt_rq = group_rt_rq(rt_se); |
| 412 | if (rt_rq && rt_rq->rt_nr_running) |
| 413 | enqueue_rt_entity(rt_se); |
| 414 | } |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 415 | } |
| 416 | |
| 417 | /* |
| 418 | * Put task to the end of the run list without the overhead of dequeue |
| 419 | * followed by enqueue. |
| 420 | */ |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 421 | static |
| 422 | void requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se) |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 423 | { |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 424 | struct rt_prio_array *array = &rt_rq->active; |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 425 | |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 426 | list_move_tail(&rt_se->run_list, array->queue + rt_se_prio(rt_se)); |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 427 | } |
| 428 | |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 429 | static void requeue_task_rt(struct rq *rq, struct task_struct *p) |
| 430 | { |
| 431 | struct sched_rt_entity *rt_se = &p->rt; |
| 432 | struct rt_rq *rt_rq; |
| 433 | |
| 434 | for_each_sched_rt_entity(rt_se) { |
| 435 | rt_rq = rt_rq_of_se(rt_se); |
| 436 | requeue_rt_entity(rt_rq, rt_se); |
| 437 | } |
| 438 | } |
| 439 | |
| 440 | static void yield_task_rt(struct rq *rq) |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 441 | { |
Dmitry Adamushko | 4530d7a | 2007-10-15 17:00:08 +0200 | [diff] [blame] | 442 | requeue_task_rt(rq, rq->curr); |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 443 | } |
| 444 | |
Gregory Haskins | e7693a3 | 2008-01-25 21:08:09 +0100 | [diff] [blame] | 445 | #ifdef CONFIG_SMP |
Gregory Haskins | 318e089 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 446 | static int find_lowest_rq(struct task_struct *task); |
| 447 | |
Gregory Haskins | e7693a3 | 2008-01-25 21:08:09 +0100 | [diff] [blame] | 448 | static int select_task_rq_rt(struct task_struct *p, int sync) |
| 449 | { |
Gregory Haskins | 318e089 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 450 | struct rq *rq = task_rq(p); |
| 451 | |
| 452 | /* |
Steven Rostedt | e1f47d8 | 2008-01-25 21:08:12 +0100 | [diff] [blame] | 453 | * If the current task is an RT task, then |
| 454 | * try to see if we can wake this RT task up on another |
| 455 | * runqueue. Otherwise simply start this RT task |
| 456 | * on its current runqueue. |
| 457 | * |
| 458 | * We want to avoid overloading runqueues. Even if |
| 459 | * the RT task is of higher priority than the current RT task. |
| 460 | * RT tasks behave differently than other tasks. If |
| 461 | * one gets preempted, we try to push it off to another queue. |
| 462 | * So trying to keep a preempting RT task on the same |
| 463 | * cache hot CPU will force the running RT task to |
| 464 | * a cold CPU. So we waste all the cache for the lower |
| 465 | * RT task in hopes of saving some of a RT task |
| 466 | * that is just being woken and probably will have |
| 467 | * cold cache anyway. |
Gregory Haskins | 318e089 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 468 | */ |
Gregory Haskins | 17b3279 | 2008-01-25 21:08:13 +0100 | [diff] [blame] | 469 | if (unlikely(rt_task(rq->curr)) && |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 470 | (p->rt.nr_cpus_allowed > 1)) { |
Gregory Haskins | 318e089 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 471 | int cpu = find_lowest_rq(p); |
| 472 | |
| 473 | return (cpu == -1) ? task_cpu(p) : cpu; |
| 474 | } |
| 475 | |
| 476 | /* |
| 477 | * Otherwise, just let it ride on the affined RQ and the |
| 478 | * post-schedule router will push the preempted task away |
| 479 | */ |
Gregory Haskins | e7693a3 | 2008-01-25 21:08:09 +0100 | [diff] [blame] | 480 | return task_cpu(p); |
| 481 | } |
| 482 | #endif /* CONFIG_SMP */ |
| 483 | |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 484 | /* |
| 485 | * Preempt the current task with a newly woken task if needed: |
| 486 | */ |
| 487 | static void check_preempt_curr_rt(struct rq *rq, struct task_struct *p) |
| 488 | { |
| 489 | if (p->prio < rq->curr->prio) |
| 490 | resched_task(rq->curr); |
| 491 | } |
| 492 | |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 493 | static struct sched_rt_entity *pick_next_rt_entity(struct rq *rq, |
| 494 | struct rt_rq *rt_rq) |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 495 | { |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 496 | struct rt_prio_array *array = &rt_rq->active; |
| 497 | struct sched_rt_entity *next = NULL; |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 498 | struct list_head *queue; |
| 499 | int idx; |
| 500 | |
| 501 | idx = sched_find_first_bit(array->bitmap); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 502 | BUG_ON(idx >= MAX_RT_PRIO); |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 503 | |
| 504 | queue = array->queue + idx; |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 505 | next = list_entry(queue->next, struct sched_rt_entity, run_list); |
Dmitry Adamushko | 326587b | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 506 | |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 507 | return next; |
| 508 | } |
| 509 | |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 510 | static struct task_struct *pick_next_task_rt(struct rq *rq) |
| 511 | { |
| 512 | struct sched_rt_entity *rt_se; |
| 513 | struct task_struct *p; |
| 514 | struct rt_rq *rt_rq; |
| 515 | |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 516 | rt_rq = &rq->rt; |
| 517 | |
| 518 | if (unlikely(!rt_rq->rt_nr_running)) |
| 519 | return NULL; |
| 520 | |
Peter Zijlstra | 23b0fdf | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 521 | if (rt_rq_throttled(rt_rq)) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 522 | return NULL; |
| 523 | |
| 524 | do { |
| 525 | rt_se = pick_next_rt_entity(rq, rt_rq); |
Dmitry Adamushko | 326587b | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 526 | BUG_ON(!rt_se); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 527 | rt_rq = group_rt_rq(rt_se); |
| 528 | } while (rt_rq); |
| 529 | |
| 530 | p = rt_task_of(rt_se); |
| 531 | p->se.exec_start = rq->clock; |
| 532 | return p; |
| 533 | } |
| 534 | |
Ingo Molnar | 31ee529 | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 535 | 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] | 536 | { |
Ingo Molnar | f1e14ef | 2007-08-09 11:16:48 +0200 | [diff] [blame] | 537 | update_curr_rt(rq); |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 538 | p->se.exec_start = 0; |
| 539 | } |
| 540 | |
Peter Williams | 681f3e6 | 2007-10-24 18:23:51 +0200 | [diff] [blame] | 541 | #ifdef CONFIG_SMP |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 542 | |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 543 | /* Only try algorithms three times */ |
| 544 | #define RT_MAX_TRIES 3 |
| 545 | |
| 546 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest); |
| 547 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep); |
| 548 | |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 549 | static int pick_rt_task(struct rq *rq, struct task_struct *p, int cpu) |
| 550 | { |
| 551 | if (!task_running(rq, p) && |
Gregory Haskins | 73fe6aa | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 552 | (cpu < 0 || cpu_isset(cpu, p->cpus_allowed)) && |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 553 | (p->rt.nr_cpus_allowed > 1)) |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 554 | return 1; |
| 555 | return 0; |
| 556 | } |
| 557 | |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 558 | /* Return the second highest RT task, NULL otherwise */ |
Ingo Molnar | 79064fb | 2008-01-25 21:08:14 +0100 | [diff] [blame] | 559 | 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] | 560 | { |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 561 | struct task_struct *next = NULL; |
| 562 | struct sched_rt_entity *rt_se; |
| 563 | struct rt_prio_array *array; |
| 564 | struct rt_rq *rt_rq; |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 565 | int idx; |
| 566 | |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 567 | for_each_leaf_rt_rq(rt_rq, rq) { |
| 568 | array = &rt_rq->active; |
| 569 | idx = sched_find_first_bit(array->bitmap); |
| 570 | next_idx: |
| 571 | if (idx >= MAX_RT_PRIO) |
| 572 | continue; |
| 573 | if (next && next->prio < idx) |
| 574 | continue; |
| 575 | list_for_each_entry(rt_se, array->queue + idx, run_list) { |
| 576 | struct task_struct *p = rt_task_of(rt_se); |
| 577 | if (pick_rt_task(rq, p, cpu)) { |
| 578 | next = p; |
| 579 | break; |
| 580 | } |
| 581 | } |
| 582 | if (!next) { |
| 583 | idx = find_next_bit(array->bitmap, MAX_RT_PRIO, idx+1); |
| 584 | goto next_idx; |
| 585 | } |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 586 | } |
| 587 | |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 588 | return next; |
| 589 | } |
| 590 | |
| 591 | static DEFINE_PER_CPU(cpumask_t, local_cpu_mask); |
| 592 | |
Gregory Haskins | 6e1254d | 2008-01-25 21:08:11 +0100 | [diff] [blame] | 593 | static int find_lowest_cpus(struct task_struct *task, cpumask_t *lowest_mask) |
Gregory Haskins | 07b4032 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 594 | { |
Gregory Haskins | 6e1254d | 2008-01-25 21:08:11 +0100 | [diff] [blame] | 595 | int lowest_prio = -1; |
Steven Rostedt | 610bf05 | 2008-01-25 21:08:13 +0100 | [diff] [blame] | 596 | int lowest_cpu = -1; |
Gregory Haskins | 06f90db | 2008-01-25 21:08:13 +0100 | [diff] [blame] | 597 | int count = 0; |
Steven Rostedt | 610bf05 | 2008-01-25 21:08:13 +0100 | [diff] [blame] | 598 | int cpu; |
Gregory Haskins | 07b4032 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 599 | |
Gregory Haskins | 637f508 | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 600 | cpus_and(*lowest_mask, task_rq(task)->rd->online, task->cpus_allowed); |
Gregory Haskins | 07b4032 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 601 | |
| 602 | /* |
| 603 | * Scan each rq for the lowest prio. |
| 604 | */ |
Steven Rostedt | 610bf05 | 2008-01-25 21:08:13 +0100 | [diff] [blame] | 605 | for_each_cpu_mask(cpu, *lowest_mask) { |
Gregory Haskins | 07b4032 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 606 | struct rq *rq = cpu_rq(cpu); |
| 607 | |
Gregory Haskins | 07b4032 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 608 | /* We look for lowest RT prio or non-rt CPU */ |
| 609 | if (rq->rt.highest_prio >= MAX_RT_PRIO) { |
Steven Rostedt | 610bf05 | 2008-01-25 21:08:13 +0100 | [diff] [blame] | 610 | /* |
| 611 | * if we already found a low RT queue |
| 612 | * and now we found this non-rt queue |
| 613 | * clear the mask and set our bit. |
| 614 | * Otherwise just return the queue as is |
| 615 | * and the count==1 will cause the algorithm |
| 616 | * to use the first bit found. |
| 617 | */ |
| 618 | if (lowest_cpu != -1) { |
Gregory Haskins | 6e1254d | 2008-01-25 21:08:11 +0100 | [diff] [blame] | 619 | cpus_clear(*lowest_mask); |
Steven Rostedt | 610bf05 | 2008-01-25 21:08:13 +0100 | [diff] [blame] | 620 | cpu_set(rq->cpu, *lowest_mask); |
| 621 | } |
Gregory Haskins | 6e1254d | 2008-01-25 21:08:11 +0100 | [diff] [blame] | 622 | return 1; |
Gregory Haskins | 07b4032 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 623 | } |
| 624 | |
| 625 | /* no locking for now */ |
Gregory Haskins | 6e1254d | 2008-01-25 21:08:11 +0100 | [diff] [blame] | 626 | if ((rq->rt.highest_prio > task->prio) |
| 627 | && (rq->rt.highest_prio >= lowest_prio)) { |
| 628 | if (rq->rt.highest_prio > lowest_prio) { |
| 629 | /* new low - clear old data */ |
| 630 | lowest_prio = rq->rt.highest_prio; |
Steven Rostedt | 610bf05 | 2008-01-25 21:08:13 +0100 | [diff] [blame] | 631 | lowest_cpu = cpu; |
| 632 | count = 0; |
Gregory Haskins | 6e1254d | 2008-01-25 21:08:11 +0100 | [diff] [blame] | 633 | } |
Gregory Haskins | 06f90db | 2008-01-25 21:08:13 +0100 | [diff] [blame] | 634 | count++; |
Steven Rostedt | 610bf05 | 2008-01-25 21:08:13 +0100 | [diff] [blame] | 635 | } else |
| 636 | cpu_clear(cpu, *lowest_mask); |
| 637 | } |
| 638 | |
| 639 | /* |
| 640 | * Clear out all the set bits that represent |
| 641 | * runqueues that were of higher prio than |
| 642 | * the lowest_prio. |
| 643 | */ |
| 644 | if (lowest_cpu > 0) { |
| 645 | /* |
| 646 | * Perhaps we could add another cpumask op to |
| 647 | * zero out bits. Like cpu_zero_bits(cpumask, nrbits); |
| 648 | * Then that could be optimized to use memset and such. |
| 649 | */ |
| 650 | for_each_cpu_mask(cpu, *lowest_mask) { |
| 651 | if (cpu >= lowest_cpu) |
| 652 | break; |
| 653 | cpu_clear(cpu, *lowest_mask); |
Gregory Haskins | 07b4032 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 654 | } |
| 655 | } |
| 656 | |
Gregory Haskins | 06f90db | 2008-01-25 21:08:13 +0100 | [diff] [blame] | 657 | return count; |
Gregory Haskins | 6e1254d | 2008-01-25 21:08:11 +0100 | [diff] [blame] | 658 | } |
| 659 | |
| 660 | static inline int pick_optimal_cpu(int this_cpu, cpumask_t *mask) |
| 661 | { |
| 662 | int first; |
| 663 | |
| 664 | /* "this_cpu" is cheaper to preempt than a remote processor */ |
| 665 | if ((this_cpu != -1) && cpu_isset(this_cpu, *mask)) |
| 666 | return this_cpu; |
| 667 | |
| 668 | first = first_cpu(*mask); |
| 669 | if (first != NR_CPUS) |
| 670 | return first; |
| 671 | |
| 672 | return -1; |
| 673 | } |
| 674 | |
| 675 | static int find_lowest_rq(struct task_struct *task) |
| 676 | { |
| 677 | struct sched_domain *sd; |
| 678 | cpumask_t *lowest_mask = &__get_cpu_var(local_cpu_mask); |
| 679 | int this_cpu = smp_processor_id(); |
| 680 | int cpu = task_cpu(task); |
Gregory Haskins | 06f90db | 2008-01-25 21:08:13 +0100 | [diff] [blame] | 681 | int count = find_lowest_cpus(task, lowest_mask); |
Gregory Haskins | 6e1254d | 2008-01-25 21:08:11 +0100 | [diff] [blame] | 682 | |
Gregory Haskins | 06f90db | 2008-01-25 21:08:13 +0100 | [diff] [blame] | 683 | if (!count) |
| 684 | return -1; /* No targets found */ |
| 685 | |
| 686 | /* |
| 687 | * There is no sense in performing an optimal search if only one |
| 688 | * target is found. |
| 689 | */ |
| 690 | if (count == 1) |
| 691 | return first_cpu(*lowest_mask); |
Gregory Haskins | 6e1254d | 2008-01-25 21:08:11 +0100 | [diff] [blame] | 692 | |
| 693 | /* |
| 694 | * At this point we have built a mask of cpus representing the |
| 695 | * lowest priority tasks in the system. Now we want to elect |
| 696 | * the best one based on our affinity and topology. |
| 697 | * |
| 698 | * We prioritize the last cpu that the task executed on since |
| 699 | * it is most likely cache-hot in that location. |
| 700 | */ |
| 701 | if (cpu_isset(cpu, *lowest_mask)) |
| 702 | return cpu; |
| 703 | |
| 704 | /* |
| 705 | * Otherwise, we consult the sched_domains span maps to figure |
| 706 | * out which cpu is logically closest to our hot cache data. |
| 707 | */ |
| 708 | if (this_cpu == cpu) |
| 709 | this_cpu = -1; /* Skip this_cpu opt if the same */ |
| 710 | |
| 711 | for_each_domain(cpu, sd) { |
| 712 | if (sd->flags & SD_WAKE_AFFINE) { |
| 713 | cpumask_t domain_mask; |
| 714 | int best_cpu; |
| 715 | |
| 716 | cpus_and(domain_mask, sd->span, *lowest_mask); |
| 717 | |
| 718 | best_cpu = pick_optimal_cpu(this_cpu, |
| 719 | &domain_mask); |
| 720 | if (best_cpu != -1) |
| 721 | return best_cpu; |
| 722 | } |
| 723 | } |
| 724 | |
| 725 | /* |
| 726 | * And finally, if there were no matches within the domains |
| 727 | * just give the caller *something* to work with from the compatible |
| 728 | * locations. |
| 729 | */ |
| 730 | return pick_optimal_cpu(this_cpu, lowest_mask); |
Gregory Haskins | 07b4032 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 731 | } |
| 732 | |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 733 | /* Will lock the rq it finds */ |
Ingo Molnar | 4df64c0 | 2008-01-25 21:08:15 +0100 | [diff] [blame] | 734 | 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] | 735 | { |
| 736 | struct rq *lowest_rq = NULL; |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 737 | int tries; |
Ingo Molnar | 4df64c0 | 2008-01-25 21:08:15 +0100 | [diff] [blame] | 738 | int cpu; |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 739 | |
| 740 | for (tries = 0; tries < RT_MAX_TRIES; tries++) { |
Gregory Haskins | 07b4032 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 741 | cpu = find_lowest_rq(task); |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 742 | |
Gregory Haskins | 2de0b46 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 743 | if ((cpu == -1) || (cpu == rq->cpu)) |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 744 | break; |
| 745 | |
Gregory Haskins | 07b4032 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 746 | lowest_rq = cpu_rq(cpu); |
| 747 | |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 748 | /* if the prio of this runqueue changed, try again */ |
Gregory Haskins | 07b4032 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 749 | if (double_lock_balance(rq, lowest_rq)) { |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 750 | /* |
| 751 | * We had to unlock the run queue. In |
| 752 | * the mean time, task could have |
| 753 | * migrated already or had its affinity changed. |
| 754 | * Also make sure that it wasn't scheduled on its rq. |
| 755 | */ |
Gregory Haskins | 07b4032 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 756 | if (unlikely(task_rq(task) != rq || |
Ingo Molnar | 4df64c0 | 2008-01-25 21:08:15 +0100 | [diff] [blame] | 757 | !cpu_isset(lowest_rq->cpu, |
| 758 | task->cpus_allowed) || |
Gregory Haskins | 07b4032 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 759 | task_running(rq, task) || |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 760 | !task->se.on_rq)) { |
Ingo Molnar | 4df64c0 | 2008-01-25 21:08:15 +0100 | [diff] [blame] | 761 | |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 762 | spin_unlock(&lowest_rq->lock); |
| 763 | lowest_rq = NULL; |
| 764 | break; |
| 765 | } |
| 766 | } |
| 767 | |
| 768 | /* If this rq is still suitable use it. */ |
| 769 | if (lowest_rq->rt.highest_prio > task->prio) |
| 770 | break; |
| 771 | |
| 772 | /* try again */ |
| 773 | spin_unlock(&lowest_rq->lock); |
| 774 | lowest_rq = NULL; |
| 775 | } |
| 776 | |
| 777 | return lowest_rq; |
| 778 | } |
| 779 | |
| 780 | /* |
| 781 | * If the current CPU has more than one RT task, see if the non |
| 782 | * running task can migrate over to a CPU that is running a task |
| 783 | * of lesser priority. |
| 784 | */ |
Gregory Haskins | 697f0a4 | 2008-01-25 21:08:09 +0100 | [diff] [blame] | 785 | static int push_rt_task(struct rq *rq) |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 786 | { |
| 787 | struct task_struct *next_task; |
| 788 | struct rq *lowest_rq; |
| 789 | int ret = 0; |
| 790 | int paranoid = RT_MAX_TRIES; |
| 791 | |
Gregory Haskins | a22d7fc | 2008-01-25 21:08:12 +0100 | [diff] [blame] | 792 | if (!rq->rt.overloaded) |
| 793 | return 0; |
| 794 | |
Gregory Haskins | 697f0a4 | 2008-01-25 21:08:09 +0100 | [diff] [blame] | 795 | next_task = pick_next_highest_task_rt(rq, -1); |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 796 | if (!next_task) |
| 797 | return 0; |
| 798 | |
| 799 | retry: |
Gregory Haskins | 697f0a4 | 2008-01-25 21:08:09 +0100 | [diff] [blame] | 800 | if (unlikely(next_task == rq->curr)) { |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 801 | WARN_ON(1); |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 802 | return 0; |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 803 | } |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 804 | |
| 805 | /* |
| 806 | * It's possible that the next_task slipped in of |
| 807 | * higher priority than current. If that's the case |
| 808 | * just reschedule current. |
| 809 | */ |
Gregory Haskins | 697f0a4 | 2008-01-25 21:08:09 +0100 | [diff] [blame] | 810 | if (unlikely(next_task->prio < rq->curr->prio)) { |
| 811 | resched_task(rq->curr); |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 812 | return 0; |
| 813 | } |
| 814 | |
Gregory Haskins | 697f0a4 | 2008-01-25 21:08:09 +0100 | [diff] [blame] | 815 | /* We might release rq lock */ |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 816 | get_task_struct(next_task); |
| 817 | |
| 818 | /* find_lock_lowest_rq locks the rq if found */ |
Gregory Haskins | 697f0a4 | 2008-01-25 21:08:09 +0100 | [diff] [blame] | 819 | lowest_rq = find_lock_lowest_rq(next_task, rq); |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 820 | if (!lowest_rq) { |
| 821 | struct task_struct *task; |
| 822 | /* |
Gregory Haskins | 697f0a4 | 2008-01-25 21:08:09 +0100 | [diff] [blame] | 823 | * find lock_lowest_rq releases rq->lock |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 824 | * so it is possible that next_task has changed. |
| 825 | * If it has, then try again. |
| 826 | */ |
Gregory Haskins | 697f0a4 | 2008-01-25 21:08:09 +0100 | [diff] [blame] | 827 | task = pick_next_highest_task_rt(rq, -1); |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 828 | if (unlikely(task != next_task) && task && paranoid--) { |
| 829 | put_task_struct(next_task); |
| 830 | next_task = task; |
| 831 | goto retry; |
| 832 | } |
| 833 | goto out; |
| 834 | } |
| 835 | |
Gregory Haskins | 697f0a4 | 2008-01-25 21:08:09 +0100 | [diff] [blame] | 836 | deactivate_task(rq, next_task, 0); |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 837 | set_task_cpu(next_task, lowest_rq->cpu); |
| 838 | activate_task(lowest_rq, next_task, 0); |
| 839 | |
| 840 | resched_task(lowest_rq->curr); |
| 841 | |
| 842 | spin_unlock(&lowest_rq->lock); |
| 843 | |
| 844 | ret = 1; |
| 845 | out: |
| 846 | put_task_struct(next_task); |
| 847 | |
| 848 | return ret; |
| 849 | } |
| 850 | |
| 851 | /* |
| 852 | * TODO: Currently we just use the second highest prio task on |
| 853 | * the queue, and stop when it can't migrate (or there's |
| 854 | * no more RT tasks). There may be a case where a lower |
| 855 | * priority RT task has a different affinity than the |
| 856 | * higher RT task. In this case the lower RT task could |
| 857 | * possibly be able to migrate where as the higher priority |
| 858 | * RT task could not. We currently ignore this issue. |
| 859 | * Enhancements are welcome! |
| 860 | */ |
| 861 | static void push_rt_tasks(struct rq *rq) |
| 862 | { |
| 863 | /* push_rt_task will return true if it moved an RT */ |
| 864 | while (push_rt_task(rq)) |
| 865 | ; |
| 866 | } |
| 867 | |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 868 | static int pull_rt_task(struct rq *this_rq) |
| 869 | { |
Ingo Molnar | 80bf317 | 2008-01-25 21:08:17 +0100 | [diff] [blame] | 870 | int this_cpu = this_rq->cpu, ret = 0, cpu; |
| 871 | struct task_struct *p, *next; |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 872 | struct rq *src_rq; |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 873 | |
Gregory Haskins | 637f508 | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 874 | if (likely(!rt_overloaded(this_rq))) |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 875 | return 0; |
| 876 | |
| 877 | next = pick_next_task_rt(this_rq); |
| 878 | |
Gregory Haskins | 637f508 | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 879 | for_each_cpu_mask(cpu, this_rq->rd->rto_mask) { |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 880 | if (this_cpu == cpu) |
| 881 | continue; |
| 882 | |
| 883 | src_rq = cpu_rq(cpu); |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 884 | /* |
| 885 | * We can potentially drop this_rq's lock in |
| 886 | * double_lock_balance, and another CPU could |
| 887 | * steal our next task - hence we must cause |
| 888 | * the caller to recalculate the next task |
| 889 | * in that case: |
| 890 | */ |
| 891 | if (double_lock_balance(this_rq, src_rq)) { |
| 892 | struct task_struct *old_next = next; |
Ingo Molnar | 80bf317 | 2008-01-25 21:08:17 +0100 | [diff] [blame] | 893 | |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 894 | next = pick_next_task_rt(this_rq); |
| 895 | if (next != old_next) |
| 896 | ret = 1; |
| 897 | } |
| 898 | |
| 899 | /* |
| 900 | * Are there still pullable RT tasks? |
| 901 | */ |
Mike Galbraith | 614ee1f | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 902 | if (src_rq->rt.rt_nr_running <= 1) |
| 903 | goto skip; |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 904 | |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 905 | p = pick_next_highest_task_rt(src_rq, this_cpu); |
| 906 | |
| 907 | /* |
| 908 | * Do we have an RT task that preempts |
| 909 | * the to-be-scheduled task? |
| 910 | */ |
| 911 | if (p && (!next || (p->prio < next->prio))) { |
| 912 | WARN_ON(p == src_rq->curr); |
| 913 | WARN_ON(!p->se.on_rq); |
| 914 | |
| 915 | /* |
| 916 | * There's a chance that p is higher in priority |
| 917 | * than what's currently running on its cpu. |
| 918 | * This is just that p is wakeing up and hasn't |
| 919 | * had a chance to schedule. We only pull |
| 920 | * p if it is lower in priority than the |
| 921 | * current task on the run queue or |
| 922 | * this_rq next task is lower in prio than |
| 923 | * the current task on that rq. |
| 924 | */ |
| 925 | if (p->prio < src_rq->curr->prio || |
| 926 | (next && next->prio < src_rq->curr->prio)) |
Mike Galbraith | 614ee1f | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 927 | goto skip; |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 928 | |
| 929 | ret = 1; |
| 930 | |
| 931 | deactivate_task(src_rq, p, 0); |
| 932 | set_task_cpu(p, this_cpu); |
| 933 | activate_task(this_rq, p, 0); |
| 934 | /* |
| 935 | * We continue with the search, just in |
| 936 | * case there's an even higher prio task |
| 937 | * in another runqueue. (low likelyhood |
| 938 | * but possible) |
Ingo Molnar | 80bf317 | 2008-01-25 21:08:17 +0100 | [diff] [blame] | 939 | * |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 940 | * Update next so that we won't pick a task |
| 941 | * on another cpu with a priority lower (or equal) |
| 942 | * than the one we just picked. |
| 943 | */ |
| 944 | next = p; |
| 945 | |
| 946 | } |
Mike Galbraith | 614ee1f | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 947 | skip: |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 948 | spin_unlock(&src_rq->lock); |
| 949 | } |
| 950 | |
| 951 | return ret; |
| 952 | } |
| 953 | |
Steven Rostedt | 9a897c5 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 954 | static void pre_schedule_rt(struct rq *rq, struct task_struct *prev) |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 955 | { |
| 956 | /* Try to pull RT tasks here if we lower this rq's prio */ |
Ingo Molnar | 7f51f29 | 2008-01-25 21:08:17 +0100 | [diff] [blame] | 957 | if (unlikely(rt_task(prev)) && rq->rt.highest_prio > prev->prio) |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 958 | pull_rt_task(rq); |
| 959 | } |
| 960 | |
Steven Rostedt | 9a897c5 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 961 | static void post_schedule_rt(struct rq *rq) |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 962 | { |
| 963 | /* |
| 964 | * If we have more than one rt_task queued, then |
| 965 | * see if we can push the other rt_tasks off to other CPUS. |
| 966 | * Note we may release the rq lock, and since |
| 967 | * the lock was owned by prev, we need to release it |
| 968 | * first via finish_lock_switch and then reaquire it here. |
| 969 | */ |
Gregory Haskins | a22d7fc | 2008-01-25 21:08:12 +0100 | [diff] [blame] | 970 | if (unlikely(rq->rt.overloaded)) { |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 971 | spin_lock_irq(&rq->lock); |
| 972 | push_rt_tasks(rq); |
| 973 | spin_unlock_irq(&rq->lock); |
| 974 | } |
| 975 | } |
| 976 | |
Steven Rostedt | 4642daf | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 977 | |
Steven Rostedt | 9a897c5 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 978 | static void task_wake_up_rt(struct rq *rq, struct task_struct *p) |
Steven Rostedt | 4642daf | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 979 | { |
Steven Rostedt | 9a897c5 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 980 | if (!task_running(rq, p) && |
Gregory Haskins | a22d7fc | 2008-01-25 21:08:12 +0100 | [diff] [blame] | 981 | (p->prio >= rq->rt.highest_prio) && |
| 982 | rq->rt.overloaded) |
Steven Rostedt | 4642daf | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 983 | push_rt_tasks(rq); |
| 984 | } |
| 985 | |
Peter Williams | 4301065 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 986 | static unsigned long |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 987 | load_balance_rt(struct rq *this_rq, int this_cpu, struct rq *busiest, |
Peter Williams | e1d1484 | 2007-10-24 18:23:51 +0200 | [diff] [blame] | 988 | unsigned long max_load_move, |
| 989 | struct sched_domain *sd, enum cpu_idle_type idle, |
| 990 | int *all_pinned, int *this_best_prio) |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 991 | { |
Steven Rostedt | c7a1e46 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 992 | /* don't touch RT tasks */ |
| 993 | return 0; |
Peter Williams | e1d1484 | 2007-10-24 18:23:51 +0200 | [diff] [blame] | 994 | } |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 995 | |
Peter Williams | e1d1484 | 2007-10-24 18:23:51 +0200 | [diff] [blame] | 996 | static int |
| 997 | move_one_task_rt(struct rq *this_rq, int this_cpu, struct rq *busiest, |
| 998 | struct sched_domain *sd, enum cpu_idle_type idle) |
| 999 | { |
Steven Rostedt | c7a1e46 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1000 | /* don't touch RT tasks */ |
| 1001 | return 0; |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1002 | } |
Ingo Molnar | deeeccd | 2008-01-25 21:08:15 +0100 | [diff] [blame] | 1003 | |
Gregory Haskins | 73fe6aa | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1004 | static void set_cpus_allowed_rt(struct task_struct *p, cpumask_t *new_mask) |
| 1005 | { |
| 1006 | int weight = cpus_weight(*new_mask); |
| 1007 | |
| 1008 | BUG_ON(!rt_task(p)); |
| 1009 | |
| 1010 | /* |
| 1011 | * Update the migration status of the RQ if we have an RT task |
| 1012 | * which is running AND changing its weight value. |
| 1013 | */ |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1014 | if (p->se.on_rq && (weight != p->rt.nr_cpus_allowed)) { |
Gregory Haskins | 73fe6aa | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1015 | struct rq *rq = task_rq(p); |
| 1016 | |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1017 | if ((p->rt.nr_cpus_allowed <= 1) && (weight > 1)) { |
Gregory Haskins | 73fe6aa | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1018 | rq->rt.rt_nr_migratory++; |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1019 | } else if ((p->rt.nr_cpus_allowed > 1) && (weight <= 1)) { |
Gregory Haskins | 73fe6aa | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1020 | BUG_ON(!rq->rt.rt_nr_migratory); |
| 1021 | rq->rt.rt_nr_migratory--; |
| 1022 | } |
| 1023 | |
| 1024 | update_rt_migration(rq); |
| 1025 | } |
| 1026 | |
| 1027 | p->cpus_allowed = *new_mask; |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1028 | p->rt.nr_cpus_allowed = weight; |
Gregory Haskins | 73fe6aa | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1029 | } |
Ingo Molnar | deeeccd | 2008-01-25 21:08:15 +0100 | [diff] [blame] | 1030 | |
Ingo Molnar | bdd7c81 | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 1031 | /* Assumes rq->lock is held */ |
| 1032 | static void join_domain_rt(struct rq *rq) |
| 1033 | { |
| 1034 | if (rq->rt.overloaded) |
| 1035 | rt_set_overload(rq); |
| 1036 | } |
| 1037 | |
| 1038 | /* Assumes rq->lock is held */ |
| 1039 | static void leave_domain_rt(struct rq *rq) |
| 1040 | { |
| 1041 | if (rq->rt.overloaded) |
| 1042 | rt_clear_overload(rq); |
| 1043 | } |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 1044 | |
| 1045 | /* |
| 1046 | * When switch from the rt queue, we bring ourselves to a position |
| 1047 | * that we might want to pull RT tasks from other runqueues. |
| 1048 | */ |
| 1049 | static void switched_from_rt(struct rq *rq, struct task_struct *p, |
| 1050 | int running) |
| 1051 | { |
| 1052 | /* |
| 1053 | * If there are other RT tasks then we will reschedule |
| 1054 | * and the scheduling of the other RT tasks will handle |
| 1055 | * the balancing. But if we are the last RT task |
| 1056 | * we may need to handle the pulling of RT tasks |
| 1057 | * now. |
| 1058 | */ |
| 1059 | if (!rq->rt.rt_nr_running) |
| 1060 | pull_rt_task(rq); |
| 1061 | } |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1062 | #endif /* CONFIG_SMP */ |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1063 | |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 1064 | /* |
| 1065 | * When switching a task to RT, we may overload the runqueue |
| 1066 | * with RT tasks. In this case we try to push them off to |
| 1067 | * other runqueues. |
| 1068 | */ |
| 1069 | static void switched_to_rt(struct rq *rq, struct task_struct *p, |
| 1070 | int running) |
| 1071 | { |
| 1072 | int check_resched = 1; |
| 1073 | |
| 1074 | /* |
| 1075 | * If we are already running, then there's nothing |
| 1076 | * that needs to be done. But if we are not running |
| 1077 | * we may need to preempt the current running task. |
| 1078 | * If that current running task is also an RT task |
| 1079 | * then see if we can move to another run queue. |
| 1080 | */ |
| 1081 | if (!running) { |
| 1082 | #ifdef CONFIG_SMP |
| 1083 | if (rq->rt.overloaded && push_rt_task(rq) && |
| 1084 | /* Don't resched if we changed runqueues */ |
| 1085 | rq != task_rq(p)) |
| 1086 | check_resched = 0; |
| 1087 | #endif /* CONFIG_SMP */ |
| 1088 | if (check_resched && p->prio < rq->curr->prio) |
| 1089 | resched_task(rq->curr); |
| 1090 | } |
| 1091 | } |
| 1092 | |
| 1093 | /* |
| 1094 | * Priority of the task has changed. This may cause |
| 1095 | * us to initiate a push or pull. |
| 1096 | */ |
| 1097 | static void prio_changed_rt(struct rq *rq, struct task_struct *p, |
| 1098 | int oldprio, int running) |
| 1099 | { |
| 1100 | if (running) { |
| 1101 | #ifdef CONFIG_SMP |
| 1102 | /* |
| 1103 | * If our priority decreases while running, we |
| 1104 | * may need to pull tasks to this runqueue. |
| 1105 | */ |
| 1106 | if (oldprio < p->prio) |
| 1107 | pull_rt_task(rq); |
| 1108 | /* |
| 1109 | * If there's a higher priority task waiting to run |
Steven Rostedt | 6fa46fa | 2008-03-05 10:00:12 -0500 | [diff] [blame] | 1110 | * then reschedule. Note, the above pull_rt_task |
| 1111 | * can release the rq lock and p could migrate. |
| 1112 | * Only reschedule if p is still on the same runqueue. |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 1113 | */ |
Steven Rostedt | 6fa46fa | 2008-03-05 10:00:12 -0500 | [diff] [blame] | 1114 | if (p->prio > rq->rt.highest_prio && rq->curr == p) |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 1115 | resched_task(p); |
| 1116 | #else |
| 1117 | /* For UP simply resched on drop of prio */ |
| 1118 | if (oldprio < p->prio) |
| 1119 | resched_task(p); |
| 1120 | #endif /* CONFIG_SMP */ |
| 1121 | } else { |
| 1122 | /* |
| 1123 | * This task is not running, but if it is |
| 1124 | * greater than the current running task |
| 1125 | * then reschedule. |
| 1126 | */ |
| 1127 | if (p->prio < rq->curr->prio) |
| 1128 | resched_task(rq->curr); |
| 1129 | } |
| 1130 | } |
| 1131 | |
Peter Zijlstra | 78f2c7d | 2008-01-25 21:08:27 +0100 | [diff] [blame] | 1132 | static void watchdog(struct rq *rq, struct task_struct *p) |
| 1133 | { |
| 1134 | unsigned long soft, hard; |
| 1135 | |
| 1136 | if (!p->signal) |
| 1137 | return; |
| 1138 | |
| 1139 | soft = p->signal->rlim[RLIMIT_RTTIME].rlim_cur; |
| 1140 | hard = p->signal->rlim[RLIMIT_RTTIME].rlim_max; |
| 1141 | |
| 1142 | if (soft != RLIM_INFINITY) { |
| 1143 | unsigned long next; |
| 1144 | |
| 1145 | p->rt.timeout++; |
| 1146 | next = DIV_ROUND_UP(min(soft, hard), USEC_PER_SEC/HZ); |
Peter Zijlstra | 5a52dd5 | 2008-01-25 21:08:32 +0100 | [diff] [blame] | 1147 | if (p->rt.timeout > next) |
Peter Zijlstra | 78f2c7d | 2008-01-25 21:08:27 +0100 | [diff] [blame] | 1148 | p->it_sched_expires = p->se.sum_exec_runtime; |
| 1149 | } |
| 1150 | } |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 1151 | |
Peter Zijlstra | 8f4d37e | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 1152 | 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] | 1153 | { |
Peter Zijlstra | 67e2be0 | 2007-12-20 15:01:17 +0100 | [diff] [blame] | 1154 | update_curr_rt(rq); |
| 1155 | |
Peter Zijlstra | 78f2c7d | 2008-01-25 21:08:27 +0100 | [diff] [blame] | 1156 | watchdog(rq, p); |
| 1157 | |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1158 | /* |
| 1159 | * RR tasks need a special form of timeslice management. |
| 1160 | * FIFO tasks have no timeslices. |
| 1161 | */ |
| 1162 | if (p->policy != SCHED_RR) |
| 1163 | return; |
| 1164 | |
Peter Zijlstra | fa71706 | 2008-01-25 21:08:27 +0100 | [diff] [blame] | 1165 | if (--p->rt.time_slice) |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1166 | return; |
| 1167 | |
Peter Zijlstra | fa71706 | 2008-01-25 21:08:27 +0100 | [diff] [blame] | 1168 | p->rt.time_slice = DEF_TIMESLICE; |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1169 | |
Dmitry Adamushko | 98fbc79 | 2007-08-24 20:39:10 +0200 | [diff] [blame] | 1170 | /* |
| 1171 | * Requeue to the end of queue if we are not the only element |
| 1172 | * on the queue: |
| 1173 | */ |
Peter Zijlstra | fa71706 | 2008-01-25 21:08:27 +0100 | [diff] [blame] | 1174 | if (p->rt.run_list.prev != p->rt.run_list.next) { |
Dmitry Adamushko | 98fbc79 | 2007-08-24 20:39:10 +0200 | [diff] [blame] | 1175 | requeue_task_rt(rq, p); |
| 1176 | set_tsk_need_resched(p); |
| 1177 | } |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1178 | } |
| 1179 | |
Srivatsa Vaddagiri | 83b699e | 2007-10-15 17:00:08 +0200 | [diff] [blame] | 1180 | static void set_curr_task_rt(struct rq *rq) |
| 1181 | { |
| 1182 | struct task_struct *p = rq->curr; |
| 1183 | |
| 1184 | p->se.exec_start = rq->clock; |
| 1185 | } |
| 1186 | |
Ingo Molnar | 5522d5d | 2007-10-15 17:00:12 +0200 | [diff] [blame] | 1187 | const struct sched_class rt_sched_class = { |
| 1188 | .next = &fair_sched_class, |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1189 | .enqueue_task = enqueue_task_rt, |
| 1190 | .dequeue_task = dequeue_task_rt, |
| 1191 | .yield_task = yield_task_rt, |
Gregory Haskins | e7693a3 | 2008-01-25 21:08:09 +0100 | [diff] [blame] | 1192 | #ifdef CONFIG_SMP |
| 1193 | .select_task_rq = select_task_rq_rt, |
| 1194 | #endif /* CONFIG_SMP */ |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1195 | |
| 1196 | .check_preempt_curr = check_preempt_curr_rt, |
| 1197 | |
| 1198 | .pick_next_task = pick_next_task_rt, |
| 1199 | .put_prev_task = put_prev_task_rt, |
| 1200 | |
Peter Williams | 681f3e6 | 2007-10-24 18:23:51 +0200 | [diff] [blame] | 1201 | #ifdef CONFIG_SMP |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1202 | .load_balance = load_balance_rt, |
Peter Williams | e1d1484 | 2007-10-24 18:23:51 +0200 | [diff] [blame] | 1203 | .move_one_task = move_one_task_rt, |
Gregory Haskins | 73fe6aa | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1204 | .set_cpus_allowed = set_cpus_allowed_rt, |
Ingo Molnar | bdd7c81 | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 1205 | .join_domain = join_domain_rt, |
| 1206 | .leave_domain = leave_domain_rt, |
Steven Rostedt | 9a897c5 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 1207 | .pre_schedule = pre_schedule_rt, |
| 1208 | .post_schedule = post_schedule_rt, |
| 1209 | .task_wake_up = task_wake_up_rt, |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 1210 | .switched_from = switched_from_rt, |
Peter Williams | 681f3e6 | 2007-10-24 18:23:51 +0200 | [diff] [blame] | 1211 | #endif |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1212 | |
Srivatsa Vaddagiri | 83b699e | 2007-10-15 17:00:08 +0200 | [diff] [blame] | 1213 | .set_curr_task = set_curr_task_rt, |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1214 | .task_tick = task_tick_rt, |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 1215 | |
| 1216 | .prio_changed = prio_changed_rt, |
| 1217 | .switched_to = switched_to_rt, |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1218 | }; |