Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1 | /* |
| 2 | * Completely Fair Scheduling (CFS) Class (SCHED_NORMAL/SCHED_BATCH) |
| 3 | * |
| 4 | * Copyright (C) 2007 Red Hat, Inc., Ingo Molnar <mingo@redhat.com> |
| 5 | * |
| 6 | * Interactivity improvements by Mike Galbraith |
| 7 | * (C) 2007 Mike Galbraith <efault@gmx.de> |
| 8 | * |
| 9 | * Various enhancements by Dmitry Adamushko. |
| 10 | * (C) 2007 Dmitry Adamushko <dmitry.adamushko@gmail.com> |
| 11 | * |
| 12 | * Group scheduling enhancements by Srivatsa Vaddagiri |
| 13 | * Copyright IBM Corporation, 2007 |
| 14 | * Author: Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com> |
| 15 | * |
| 16 | * Scaled math optimizations by Thomas Gleixner |
| 17 | * Copyright (C) 2007, Thomas Gleixner <tglx@linutronix.de> |
Peter Zijlstra | 2180508 | 2007-08-25 18:41:53 +0200 | [diff] [blame] | 18 | * |
| 19 | * Adaptive scheduling granularity, math enhancements by Peter Zijlstra |
| 20 | * Copyright (C) 2007 Red Hat, Inc., Peter Zijlstra <pzijlstr@redhat.com> |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 21 | */ |
| 22 | |
| 23 | /* |
Peter Zijlstra | 2180508 | 2007-08-25 18:41:53 +0200 | [diff] [blame] | 24 | * Targeted preemption latency for CPU-bound tasks: |
| 25 | * (default: 20ms, units: nanoseconds) |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 26 | * |
Peter Zijlstra | 2180508 | 2007-08-25 18:41:53 +0200 | [diff] [blame] | 27 | * NOTE: this latency value is not the same as the concept of |
| 28 | * 'timeslice length' - timeslices in CFS are of variable length. |
| 29 | * (to see the precise effective timeslice length of your workload, |
| 30 | * run vmstat and monitor the context-switches field) |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 31 | * |
| 32 | * On SMP systems the value of this is multiplied by the log2 of the |
| 33 | * number of CPUs. (i.e. factor 2x on 2-way systems, 3x on 4-way |
| 34 | * systems, 4x on 8-way systems, 5x on 16-way systems, etc.) |
Peter Zijlstra | 2180508 | 2007-08-25 18:41:53 +0200 | [diff] [blame] | 35 | * Targeted preemption latency for CPU-bound tasks: |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 36 | */ |
Ingo Molnar | 2bd8e6d | 2007-10-15 17:00:02 +0200 | [diff] [blame] | 37 | const_debug unsigned int sysctl_sched_latency = 20000000ULL; |
| 38 | |
| 39 | /* |
| 40 | * After fork, child runs first. (default) If set to 0 then |
| 41 | * parent will (try to) run first. |
| 42 | */ |
| 43 | const_debug unsigned int sysctl_sched_child_runs_first = 1; |
Peter Zijlstra | 2180508 | 2007-08-25 18:41:53 +0200 | [diff] [blame] | 44 | |
| 45 | /* |
| 46 | * Minimal preemption granularity for CPU-bound tasks: |
| 47 | * (default: 2 msec, units: nanoseconds) |
| 48 | */ |
Ingo Molnar | 172ac3d | 2007-08-25 18:41:53 +0200 | [diff] [blame] | 49 | unsigned int sysctl_sched_min_granularity __read_mostly = 2000000ULL; |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 50 | |
| 51 | /* |
Ingo Molnar | 1799e35 | 2007-09-19 23:34:46 +0200 | [diff] [blame] | 52 | * sys_sched_yield() compat mode |
| 53 | * |
| 54 | * This option switches the agressive yield implementation of the |
| 55 | * old scheduler back on. |
| 56 | */ |
| 57 | unsigned int __read_mostly sysctl_sched_compat_yield; |
| 58 | |
| 59 | /* |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 60 | * SCHED_BATCH wake-up granularity. |
Ingo Molnar | 71fd371 | 2007-08-24 20:39:10 +0200 | [diff] [blame] | 61 | * (default: 25 msec, units: nanoseconds) |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 62 | * |
| 63 | * This option delays the preemption effects of decoupled workloads |
| 64 | * and reduces their over-scheduling. Synchronous workloads will still |
| 65 | * have immediate wakeup/sleep latencies. |
| 66 | */ |
Ingo Molnar | 2bd8e6d | 2007-10-15 17:00:02 +0200 | [diff] [blame] | 67 | const_debug unsigned int sysctl_sched_batch_wakeup_granularity = 25000000UL; |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 68 | |
| 69 | /* |
| 70 | * SCHED_OTHER wake-up granularity. |
| 71 | * (default: 1 msec, units: nanoseconds) |
| 72 | * |
| 73 | * This option delays the preemption effects of decoupled workloads |
| 74 | * and reduces their over-scheduling. Synchronous workloads will still |
| 75 | * have immediate wakeup/sleep latencies. |
| 76 | */ |
Ingo Molnar | 2e09bf5 | 2007-10-15 17:00:05 +0200 | [diff] [blame] | 77 | const_debug unsigned int sysctl_sched_wakeup_granularity = 2000000UL; |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 78 | |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 79 | extern struct sched_class fair_sched_class; |
| 80 | |
| 81 | /************************************************************** |
| 82 | * CFS operations on generic schedulable entities: |
| 83 | */ |
| 84 | |
| 85 | #ifdef CONFIG_FAIR_GROUP_SCHED |
| 86 | |
| 87 | /* cpu runqueue to which this cfs_rq is attached */ |
| 88 | static inline struct rq *rq_of(struct cfs_rq *cfs_rq) |
| 89 | { |
| 90 | return cfs_rq->rq; |
| 91 | } |
| 92 | |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 93 | /* An entity is a task if it doesn't "own" a runqueue */ |
| 94 | #define entity_is_task(se) (!se->my_q) |
| 95 | |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 96 | #else /* CONFIG_FAIR_GROUP_SCHED */ |
| 97 | |
| 98 | static inline struct rq *rq_of(struct cfs_rq *cfs_rq) |
| 99 | { |
| 100 | return container_of(cfs_rq, struct rq, cfs); |
| 101 | } |
| 102 | |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 103 | #define entity_is_task(se) 1 |
| 104 | |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 105 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
| 106 | |
| 107 | static inline struct task_struct *task_of(struct sched_entity *se) |
| 108 | { |
| 109 | return container_of(se, struct task_struct, se); |
| 110 | } |
| 111 | |
| 112 | |
| 113 | /************************************************************** |
| 114 | * Scheduling class tree data structure manipulation methods: |
| 115 | */ |
| 116 | |
Peter Zijlstra | 02e0431 | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 117 | static inline u64 |
| 118 | max_vruntime(u64 min_vruntime, u64 vruntime) |
| 119 | { |
| 120 | if ((vruntime > min_vruntime) || |
| 121 | (min_vruntime > (1ULL << 61) && vruntime < (1ULL << 50))) |
| 122 | min_vruntime = vruntime; |
| 123 | |
| 124 | return min_vruntime; |
| 125 | } |
| 126 | |
Ingo Molnar | e9acbff | 2007-10-15 17:00:04 +0200 | [diff] [blame] | 127 | static inline void |
| 128 | set_leftmost(struct cfs_rq *cfs_rq, struct rb_node *leftmost) |
| 129 | { |
| 130 | struct sched_entity *se; |
| 131 | |
| 132 | cfs_rq->rb_leftmost = leftmost; |
Peter Zijlstra | 02e0431 | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 133 | if (leftmost) |
Ingo Molnar | e9acbff | 2007-10-15 17:00:04 +0200 | [diff] [blame] | 134 | se = rb_entry(leftmost, struct sched_entity, run_node); |
Ingo Molnar | e9acbff | 2007-10-15 17:00:04 +0200 | [diff] [blame] | 135 | } |
| 136 | |
Peter Zijlstra | 02e0431 | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 137 | static inline s64 |
| 138 | entity_key(struct cfs_rq *cfs_rq, struct sched_entity *se) |
Peter Zijlstra | 9014623 | 2007-10-15 17:00:05 +0200 | [diff] [blame] | 139 | { |
Dmitry Adamushko | 30cfdcf | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 140 | return se->vruntime - cfs_rq->min_vruntime; |
Peter Zijlstra | 9014623 | 2007-10-15 17:00:05 +0200 | [diff] [blame] | 141 | } |
| 142 | |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 143 | /* |
| 144 | * Enqueue an entity into the rb-tree: |
| 145 | */ |
Ingo Molnar | 19ccd97 | 2007-10-15 17:00:04 +0200 | [diff] [blame] | 146 | static void |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 147 | __enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) |
| 148 | { |
| 149 | struct rb_node **link = &cfs_rq->tasks_timeline.rb_node; |
| 150 | struct rb_node *parent = NULL; |
| 151 | struct sched_entity *entry; |
Peter Zijlstra | 9014623 | 2007-10-15 17:00:05 +0200 | [diff] [blame] | 152 | s64 key = entity_key(cfs_rq, se); |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 153 | int leftmost = 1; |
| 154 | |
| 155 | /* |
| 156 | * Find the right place in the rbtree: |
| 157 | */ |
| 158 | while (*link) { |
| 159 | parent = *link; |
| 160 | entry = rb_entry(parent, struct sched_entity, run_node); |
| 161 | /* |
| 162 | * We dont care about collisions. Nodes with |
| 163 | * the same key stay together. |
| 164 | */ |
Peter Zijlstra | 9014623 | 2007-10-15 17:00:05 +0200 | [diff] [blame] | 165 | if (key < entity_key(cfs_rq, entry)) { |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 166 | link = &parent->rb_left; |
| 167 | } else { |
| 168 | link = &parent->rb_right; |
| 169 | leftmost = 0; |
| 170 | } |
| 171 | } |
| 172 | |
| 173 | /* |
| 174 | * Maintain a cache of leftmost tree entries (it is frequently |
| 175 | * used): |
| 176 | */ |
| 177 | if (leftmost) |
Ingo Molnar | e9acbff | 2007-10-15 17:00:04 +0200 | [diff] [blame] | 178 | set_leftmost(cfs_rq, &se->run_node); |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 179 | |
| 180 | rb_link_node(&se->run_node, parent, link); |
| 181 | rb_insert_color(&se->run_node, &cfs_rq->tasks_timeline); |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 182 | } |
| 183 | |
Ingo Molnar | 19ccd97 | 2007-10-15 17:00:04 +0200 | [diff] [blame] | 184 | static void |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 185 | __dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) |
| 186 | { |
| 187 | if (cfs_rq->rb_leftmost == &se->run_node) |
Ingo Molnar | e9acbff | 2007-10-15 17:00:04 +0200 | [diff] [blame] | 188 | set_leftmost(cfs_rq, rb_next(&se->run_node)); |
| 189 | |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 190 | rb_erase(&se->run_node, &cfs_rq->tasks_timeline); |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 191 | } |
| 192 | |
| 193 | static inline struct rb_node *first_fair(struct cfs_rq *cfs_rq) |
| 194 | { |
| 195 | return cfs_rq->rb_leftmost; |
| 196 | } |
| 197 | |
| 198 | static struct sched_entity *__pick_next_entity(struct cfs_rq *cfs_rq) |
| 199 | { |
| 200 | return rb_entry(first_fair(cfs_rq), struct sched_entity, run_node); |
| 201 | } |
| 202 | |
Peter Zijlstra | aeb73b0 | 2007-10-15 17:00:05 +0200 | [diff] [blame] | 203 | static inline struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq) |
| 204 | { |
| 205 | struct rb_node **link = &cfs_rq->tasks_timeline.rb_node; |
| 206 | struct sched_entity *se = NULL; |
| 207 | struct rb_node *parent; |
| 208 | |
| 209 | while (*link) { |
| 210 | parent = *link; |
| 211 | se = rb_entry(parent, struct sched_entity, run_node); |
| 212 | link = &parent->rb_right; |
| 213 | } |
| 214 | |
| 215 | return se; |
| 216 | } |
| 217 | |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 218 | /************************************************************** |
| 219 | * Scheduling class statistics methods: |
| 220 | */ |
| 221 | |
Peter Zijlstra | 4d78e7b | 2007-10-15 17:00:04 +0200 | [diff] [blame] | 222 | static u64 __sched_period(unsigned long nr_running) |
| 223 | { |
| 224 | u64 period = sysctl_sched_latency; |
| 225 | unsigned long nr_latency = |
| 226 | sysctl_sched_latency / sysctl_sched_min_granularity; |
| 227 | |
| 228 | if (unlikely(nr_running > nr_latency)) { |
| 229 | period *= nr_running; |
| 230 | do_div(period, nr_latency); |
| 231 | } |
| 232 | |
| 233 | return period; |
| 234 | } |
| 235 | |
Peter Zijlstra | 6d0f0ebd | 2007-10-15 17:00:05 +0200 | [diff] [blame] | 236 | static u64 sched_slice(struct cfs_rq *cfs_rq, struct sched_entity *se) |
Peter Zijlstra | 2180508 | 2007-08-25 18:41:53 +0200 | [diff] [blame] | 237 | { |
Peter Zijlstra | 6d0f0ebd | 2007-10-15 17:00:05 +0200 | [diff] [blame] | 238 | u64 period = __sched_period(cfs_rq->nr_running); |
Peter Zijlstra | 2180508 | 2007-08-25 18:41:53 +0200 | [diff] [blame] | 239 | |
Peter Zijlstra | 6d0f0ebd | 2007-10-15 17:00:05 +0200 | [diff] [blame] | 240 | period *= se->load.weight; |
| 241 | do_div(period, cfs_rq->load.weight); |
Peter Zijlstra | 2180508 | 2007-08-25 18:41:53 +0200 | [diff] [blame] | 242 | |
Peter Zijlstra | 6d0f0ebd | 2007-10-15 17:00:05 +0200 | [diff] [blame] | 243 | return period; |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 244 | } |
| 245 | |
Peter Zijlstra | 67e9fb2 | 2007-10-15 17:00:10 +0200 | [diff] [blame] | 246 | static u64 __sched_vslice(unsigned long nr_running) |
| 247 | { |
| 248 | u64 period = __sched_period(nr_running); |
| 249 | |
| 250 | do_div(period, nr_running); |
| 251 | |
| 252 | return period; |
| 253 | } |
| 254 | |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 255 | /* |
| 256 | * Update the current task's runtime statistics. Skip current tasks that |
| 257 | * are not in our scheduling class. |
| 258 | */ |
| 259 | static inline void |
Ingo Molnar | 8ebc91d | 2007-10-15 17:00:03 +0200 | [diff] [blame] | 260 | __update_curr(struct cfs_rq *cfs_rq, struct sched_entity *curr, |
| 261 | unsigned long delta_exec) |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 262 | { |
Ingo Molnar | bbdba7c | 2007-10-15 17:00:06 +0200 | [diff] [blame] | 263 | unsigned long delta_exec_weighted; |
Peter Zijlstra | 02e0431 | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 264 | u64 next_vruntime, min_vruntime; |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 265 | |
Ingo Molnar | 8179ca23 | 2007-08-02 17:41:40 +0200 | [diff] [blame] | 266 | schedstat_set(curr->exec_max, max((u64)delta_exec, curr->exec_max)); |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 267 | |
| 268 | curr->sum_exec_runtime += delta_exec; |
Ingo Molnar | 7a62eab | 2007-10-15 17:00:06 +0200 | [diff] [blame] | 269 | schedstat_add(cfs_rq, exec_clock, delta_exec); |
Ingo Molnar | e9acbff | 2007-10-15 17:00:04 +0200 | [diff] [blame] | 270 | delta_exec_weighted = delta_exec; |
| 271 | if (unlikely(curr->load.weight != NICE_0_LOAD)) { |
| 272 | delta_exec_weighted = calc_delta_fair(delta_exec_weighted, |
| 273 | &curr->load); |
| 274 | } |
| 275 | curr->vruntime += delta_exec_weighted; |
Peter Zijlstra | 02e0431 | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 276 | |
| 277 | /* |
| 278 | * maintain cfs_rq->min_vruntime to be a monotonic increasing |
| 279 | * value tracking the leftmost vruntime in the tree. |
| 280 | */ |
| 281 | if (first_fair(cfs_rq)) { |
| 282 | next_vruntime = __pick_next_entity(cfs_rq)->vruntime; |
| 283 | |
| 284 | /* min_vruntime() := !max_vruntime() */ |
| 285 | min_vruntime = max_vruntime(curr->vruntime, next_vruntime); |
| 286 | if (min_vruntime == next_vruntime) |
| 287 | min_vruntime = curr->vruntime; |
| 288 | else |
| 289 | min_vruntime = next_vruntime; |
| 290 | } else |
| 291 | min_vruntime = curr->vruntime; |
| 292 | |
| 293 | cfs_rq->min_vruntime = |
| 294 | max_vruntime(cfs_rq->min_vruntime, min_vruntime); |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 295 | } |
| 296 | |
Ingo Molnar | b7cc089 | 2007-08-09 11:16:47 +0200 | [diff] [blame] | 297 | static void update_curr(struct cfs_rq *cfs_rq) |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 298 | { |
Ingo Molnar | 429d43b | 2007-10-15 17:00:03 +0200 | [diff] [blame] | 299 | struct sched_entity *curr = cfs_rq->curr; |
Ingo Molnar | 8ebc91d | 2007-10-15 17:00:03 +0200 | [diff] [blame] | 300 | u64 now = rq_of(cfs_rq)->clock; |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 301 | unsigned long delta_exec; |
| 302 | |
| 303 | if (unlikely(!curr)) |
| 304 | return; |
| 305 | |
| 306 | /* |
| 307 | * Get the amount of time the current task was running |
| 308 | * since the last time we changed load (this cannot |
| 309 | * overflow on 32 bits): |
| 310 | */ |
Ingo Molnar | 8ebc91d | 2007-10-15 17:00:03 +0200 | [diff] [blame] | 311 | delta_exec = (unsigned long)(now - curr->exec_start); |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 312 | |
Ingo Molnar | 8ebc91d | 2007-10-15 17:00:03 +0200 | [diff] [blame] | 313 | __update_curr(cfs_rq, curr, delta_exec); |
| 314 | curr->exec_start = now; |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 315 | } |
| 316 | |
| 317 | static inline void |
Ingo Molnar | 5870db5 | 2007-08-09 11:16:47 +0200 | [diff] [blame] | 318 | update_stats_wait_start(struct cfs_rq *cfs_rq, struct sched_entity *se) |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 319 | { |
Ingo Molnar | d281918 | 2007-08-09 11:16:47 +0200 | [diff] [blame] | 320 | schedstat_set(se->wait_start, rq_of(cfs_rq)->clock); |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 321 | } |
| 322 | |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 323 | static inline unsigned long |
Ingo Molnar | 08e2388 | 2007-10-15 17:00:04 +0200 | [diff] [blame] | 324 | calc_weighted(unsigned long delta, struct sched_entity *se) |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 325 | { |
Ingo Molnar | 08e2388 | 2007-10-15 17:00:04 +0200 | [diff] [blame] | 326 | unsigned long weight = se->load.weight; |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 327 | |
Ingo Molnar | 08e2388 | 2007-10-15 17:00:04 +0200 | [diff] [blame] | 328 | if (unlikely(weight != NICE_0_LOAD)) |
| 329 | return (u64)delta * se->load.weight >> NICE_0_SHIFT; |
| 330 | else |
| 331 | return delta; |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 332 | } |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 333 | |
| 334 | /* |
| 335 | * Task is being enqueued - update stats: |
| 336 | */ |
Ingo Molnar | d2417e5 | 2007-08-09 11:16:47 +0200 | [diff] [blame] | 337 | static void update_stats_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se) |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 338 | { |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 339 | /* |
| 340 | * Are we enqueueing a waiting task? (for current tasks |
| 341 | * a dequeue/enqueue event is a NOP) |
| 342 | */ |
Ingo Molnar | 429d43b | 2007-10-15 17:00:03 +0200 | [diff] [blame] | 343 | if (se != cfs_rq->curr) |
Ingo Molnar | 5870db5 | 2007-08-09 11:16:47 +0200 | [diff] [blame] | 344 | update_stats_wait_start(cfs_rq, se); |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 345 | } |
| 346 | |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 347 | static void |
Ingo Molnar | 9ef0a96 | 2007-08-09 11:16:47 +0200 | [diff] [blame] | 348 | update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se) |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 349 | { |
Ingo Molnar | bbdba7c | 2007-10-15 17:00:06 +0200 | [diff] [blame] | 350 | schedstat_set(se->wait_max, max(se->wait_max, |
| 351 | rq_of(cfs_rq)->clock - se->wait_start)); |
Ingo Molnar | 6cfb0d5 | 2007-08-02 17:41:40 +0200 | [diff] [blame] | 352 | schedstat_set(se->wait_start, 0); |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 353 | } |
| 354 | |
| 355 | static inline void |
Ingo Molnar | 19b6a2e | 2007-08-09 11:16:48 +0200 | [diff] [blame] | 356 | update_stats_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se) |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 357 | { |
Ingo Molnar | b7cc089 | 2007-08-09 11:16:47 +0200 | [diff] [blame] | 358 | update_curr(cfs_rq); |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 359 | /* |
| 360 | * Mark the end of the wait period if dequeueing a |
| 361 | * waiting task: |
| 362 | */ |
Ingo Molnar | 429d43b | 2007-10-15 17:00:03 +0200 | [diff] [blame] | 363 | if (se != cfs_rq->curr) |
Ingo Molnar | 9ef0a96 | 2007-08-09 11:16:47 +0200 | [diff] [blame] | 364 | update_stats_wait_end(cfs_rq, se); |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 365 | } |
| 366 | |
| 367 | /* |
| 368 | * We are picking a new current task - update its stats: |
| 369 | */ |
| 370 | static inline void |
Ingo Molnar | 79303e9 | 2007-08-09 11:16:47 +0200 | [diff] [blame] | 371 | update_stats_curr_start(struct cfs_rq *cfs_rq, struct sched_entity *se) |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 372 | { |
| 373 | /* |
| 374 | * We are starting a new run period: |
| 375 | */ |
Ingo Molnar | d281918 | 2007-08-09 11:16:47 +0200 | [diff] [blame] | 376 | se->exec_start = rq_of(cfs_rq)->clock; |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 377 | } |
| 378 | |
| 379 | /* |
| 380 | * We are descheduling a task - update its stats: |
| 381 | */ |
| 382 | static inline void |
Ingo Molnar | c7e9b5b | 2007-08-09 11:16:48 +0200 | [diff] [blame] | 383 | update_stats_curr_end(struct cfs_rq *cfs_rq, struct sched_entity *se) |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 384 | { |
| 385 | se->exec_start = 0; |
| 386 | } |
| 387 | |
| 388 | /************************************************** |
| 389 | * Scheduling class queueing methods: |
| 390 | */ |
| 391 | |
Dmitry Adamushko | 30cfdcf | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 392 | static void |
| 393 | account_entity_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se) |
| 394 | { |
| 395 | update_load_add(&cfs_rq->load, se->load.weight); |
| 396 | cfs_rq->nr_running++; |
| 397 | se->on_rq = 1; |
| 398 | } |
| 399 | |
| 400 | static void |
| 401 | account_entity_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se) |
| 402 | { |
| 403 | update_load_sub(&cfs_rq->load, se->load.weight); |
| 404 | cfs_rq->nr_running--; |
| 405 | se->on_rq = 0; |
| 406 | } |
| 407 | |
Ingo Molnar | 2396af6 | 2007-08-09 11:16:48 +0200 | [diff] [blame] | 408 | static void enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se) |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 409 | { |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 410 | #ifdef CONFIG_SCHEDSTATS |
| 411 | if (se->sleep_start) { |
Ingo Molnar | d281918 | 2007-08-09 11:16:47 +0200 | [diff] [blame] | 412 | u64 delta = rq_of(cfs_rq)->clock - se->sleep_start; |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 413 | |
| 414 | if ((s64)delta < 0) |
| 415 | delta = 0; |
| 416 | |
| 417 | if (unlikely(delta > se->sleep_max)) |
| 418 | se->sleep_max = delta; |
| 419 | |
| 420 | se->sleep_start = 0; |
| 421 | se->sum_sleep_runtime += delta; |
| 422 | } |
| 423 | if (se->block_start) { |
Ingo Molnar | d281918 | 2007-08-09 11:16:47 +0200 | [diff] [blame] | 424 | u64 delta = rq_of(cfs_rq)->clock - se->block_start; |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 425 | |
| 426 | if ((s64)delta < 0) |
| 427 | delta = 0; |
| 428 | |
| 429 | if (unlikely(delta > se->block_max)) |
| 430 | se->block_max = delta; |
| 431 | |
| 432 | se->block_start = 0; |
| 433 | se->sum_sleep_runtime += delta; |
Ingo Molnar | 30084fb | 2007-10-02 14:13:08 +0200 | [diff] [blame] | 434 | |
| 435 | /* |
| 436 | * Blocking time is in units of nanosecs, so shift by 20 to |
| 437 | * get a milliseconds-range estimation of the amount of |
| 438 | * time that the task spent sleeping: |
| 439 | */ |
| 440 | if (unlikely(prof_on == SLEEP_PROFILING)) { |
Ingo Molnar | e22f5bb | 2007-10-15 17:00:06 +0200 | [diff] [blame] | 441 | struct task_struct *tsk = task_of(se); |
| 442 | |
Ingo Molnar | 30084fb | 2007-10-02 14:13:08 +0200 | [diff] [blame] | 443 | profile_hits(SLEEP_PROFILING, (void *)get_wchan(tsk), |
| 444 | delta >> 20); |
| 445 | } |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 446 | } |
| 447 | #endif |
| 448 | } |
| 449 | |
Peter Zijlstra | ddc9729 | 2007-10-15 17:00:10 +0200 | [diff] [blame^] | 450 | static void check_spread(struct cfs_rq *cfs_rq, struct sched_entity *se) |
| 451 | { |
| 452 | #ifdef CONFIG_SCHED_DEBUG |
| 453 | s64 d = se->vruntime - cfs_rq->min_vruntime; |
| 454 | |
| 455 | if (d < 0) |
| 456 | d = -d; |
| 457 | |
| 458 | if (d > 3*sysctl_sched_latency) |
| 459 | schedstat_inc(cfs_rq, nr_spread_over); |
| 460 | #endif |
| 461 | } |
| 462 | |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 463 | static void |
Peter Zijlstra | aeb73b0 | 2007-10-15 17:00:05 +0200 | [diff] [blame] | 464 | place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial) |
| 465 | { |
Peter Zijlstra | 67e9fb2 | 2007-10-15 17:00:10 +0200 | [diff] [blame] | 466 | u64 vruntime; |
Peter Zijlstra | aeb73b0 | 2007-10-15 17:00:05 +0200 | [diff] [blame] | 467 | |
Peter Zijlstra | 67e9fb2 | 2007-10-15 17:00:10 +0200 | [diff] [blame] | 468 | vruntime = cfs_rq->min_vruntime; |
Peter Zijlstra | 94dfb5e | 2007-10-15 17:00:05 +0200 | [diff] [blame] | 469 | |
| 470 | if (sched_feat(USE_TREE_AVG)) { |
| 471 | struct sched_entity *last = __pick_last_entity(cfs_rq); |
| 472 | if (last) { |
Peter Zijlstra | 67e9fb2 | 2007-10-15 17:00:10 +0200 | [diff] [blame] | 473 | vruntime += last->vruntime; |
| 474 | vruntime >>= 1; |
Peter Zijlstra | 94dfb5e | 2007-10-15 17:00:05 +0200 | [diff] [blame] | 475 | } |
Peter Zijlstra | 67e9fb2 | 2007-10-15 17:00:10 +0200 | [diff] [blame] | 476 | } else if (sched_feat(APPROX_AVG) && cfs_rq->nr_running) |
| 477 | vruntime += __sched_vslice(cfs_rq->nr_running)/2; |
Peter Zijlstra | 94dfb5e | 2007-10-15 17:00:05 +0200 | [diff] [blame] | 478 | |
| 479 | if (initial && sched_feat(START_DEBIT)) |
Peter Zijlstra | 67e9fb2 | 2007-10-15 17:00:10 +0200 | [diff] [blame] | 480 | vruntime += __sched_vslice(cfs_rq->nr_running + 1); |
Peter Zijlstra | aeb73b0 | 2007-10-15 17:00:05 +0200 | [diff] [blame] | 481 | |
| 482 | if (!initial && sched_feat(NEW_FAIR_SLEEPERS)) { |
Peter Zijlstra | 67e9fb2 | 2007-10-15 17:00:10 +0200 | [diff] [blame] | 483 | s64 latency = cfs_rq->min_vruntime - se->last_min_vruntime; |
| 484 | if (latency < 0 || !cfs_rq->nr_running) |
| 485 | latency = 0; |
Peter Zijlstra | aeb73b0 | 2007-10-15 17:00:05 +0200 | [diff] [blame] | 486 | else |
Peter Zijlstra | 67e9fb2 | 2007-10-15 17:00:10 +0200 | [diff] [blame] | 487 | latency = min_t(s64, latency, sysctl_sched_latency); |
| 488 | vruntime -= latency; |
Peter Zijlstra | aeb73b0 | 2007-10-15 17:00:05 +0200 | [diff] [blame] | 489 | } |
| 490 | |
Peter Zijlstra | 67e9fb2 | 2007-10-15 17:00:10 +0200 | [diff] [blame] | 491 | se->vruntime = vruntime; |
| 492 | |
Peter Zijlstra | aeb73b0 | 2007-10-15 17:00:05 +0200 | [diff] [blame] | 493 | } |
| 494 | |
| 495 | static void |
Srivatsa Vaddagiri | 83b699e | 2007-10-15 17:00:08 +0200 | [diff] [blame] | 496 | enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int wakeup) |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 497 | { |
| 498 | /* |
| 499 | * Update the fair clock. |
| 500 | */ |
Ingo Molnar | b7cc089 | 2007-08-09 11:16:47 +0200 | [diff] [blame] | 501 | update_curr(cfs_rq); |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 502 | |
Ingo Molnar | e9acbff | 2007-10-15 17:00:04 +0200 | [diff] [blame] | 503 | if (wakeup) { |
Peter Zijlstra | 67e9fb2 | 2007-10-15 17:00:10 +0200 | [diff] [blame] | 504 | /* se->vruntime += cfs_rq->min_vruntime; */ |
Peter Zijlstra | aeb73b0 | 2007-10-15 17:00:05 +0200 | [diff] [blame] | 505 | place_entity(cfs_rq, se, 0); |
Ingo Molnar | 2396af6 | 2007-08-09 11:16:48 +0200 | [diff] [blame] | 506 | enqueue_sleeper(cfs_rq, se); |
Ingo Molnar | e9acbff | 2007-10-15 17:00:04 +0200 | [diff] [blame] | 507 | } |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 508 | |
Ingo Molnar | d2417e5 | 2007-08-09 11:16:47 +0200 | [diff] [blame] | 509 | update_stats_enqueue(cfs_rq, se); |
Peter Zijlstra | ddc9729 | 2007-10-15 17:00:10 +0200 | [diff] [blame^] | 510 | check_spread(cfs_rq, se); |
Srivatsa Vaddagiri | 83b699e | 2007-10-15 17:00:08 +0200 | [diff] [blame] | 511 | if (se != cfs_rq->curr) |
| 512 | __enqueue_entity(cfs_rq, se); |
Dmitry Adamushko | 30cfdcf | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 513 | account_entity_enqueue(cfs_rq, se); |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 514 | } |
| 515 | |
| 516 | static void |
Ingo Molnar | 525c271 | 2007-08-09 11:16:48 +0200 | [diff] [blame] | 517 | dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int sleep) |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 518 | { |
Ingo Molnar | 19b6a2e | 2007-08-09 11:16:48 +0200 | [diff] [blame] | 519 | update_stats_dequeue(cfs_rq, se); |
Dmitry Adamushko | db36cc7 | 2007-10-15 17:00:06 +0200 | [diff] [blame] | 520 | if (sleep) { |
Peter Zijlstra | 67e9fb2 | 2007-10-15 17:00:10 +0200 | [diff] [blame] | 521 | #ifdef CONFIG_SCHEDSTATS |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 522 | if (entity_is_task(se)) { |
| 523 | struct task_struct *tsk = task_of(se); |
| 524 | |
| 525 | if (tsk->state & TASK_INTERRUPTIBLE) |
Ingo Molnar | d281918 | 2007-08-09 11:16:47 +0200 | [diff] [blame] | 526 | se->sleep_start = rq_of(cfs_rq)->clock; |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 527 | if (tsk->state & TASK_UNINTERRUPTIBLE) |
Ingo Molnar | d281918 | 2007-08-09 11:16:47 +0200 | [diff] [blame] | 528 | se->block_start = rq_of(cfs_rq)->clock; |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 529 | } |
Dmitry Adamushko | db36cc7 | 2007-10-15 17:00:06 +0200 | [diff] [blame] | 530 | #endif |
Peter Zijlstra | 67e9fb2 | 2007-10-15 17:00:10 +0200 | [diff] [blame] | 531 | /* se->vruntime = entity_key(cfs_rq, se); */ |
| 532 | se->last_min_vruntime = cfs_rq->min_vruntime; |
| 533 | } |
| 534 | |
Srivatsa Vaddagiri | 83b699e | 2007-10-15 17:00:08 +0200 | [diff] [blame] | 535 | if (se != cfs_rq->curr) |
Dmitry Adamushko | 30cfdcf | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 536 | __dequeue_entity(cfs_rq, se); |
| 537 | account_entity_dequeue(cfs_rq, se); |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 538 | } |
| 539 | |
| 540 | /* |
| 541 | * Preempt the current task with a newly woken task if needed: |
| 542 | */ |
Peter Zijlstra | 7c92e54 | 2007-09-05 14:32:49 +0200 | [diff] [blame] | 543 | static void |
Ingo Molnar | 2e09bf5 | 2007-10-15 17:00:05 +0200 | [diff] [blame] | 544 | check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr) |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 545 | { |
Peter Zijlstra | 1169783 | 2007-09-05 14:32:49 +0200 | [diff] [blame] | 546 | unsigned long ideal_runtime, delta_exec; |
| 547 | |
Peter Zijlstra | 6d0f0ebd | 2007-10-15 17:00:05 +0200 | [diff] [blame] | 548 | ideal_runtime = sched_slice(cfs_rq, curr); |
Peter Zijlstra | 1169783 | 2007-09-05 14:32:49 +0200 | [diff] [blame] | 549 | delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime; |
| 550 | if (delta_exec > ideal_runtime) |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 551 | resched_task(rq_of(cfs_rq)->curr); |
| 552 | } |
| 553 | |
Srivatsa Vaddagiri | 83b699e | 2007-10-15 17:00:08 +0200 | [diff] [blame] | 554 | static void |
Ingo Molnar | 8494f41 | 2007-08-09 11:16:48 +0200 | [diff] [blame] | 555 | set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 556 | { |
Srivatsa Vaddagiri | 83b699e | 2007-10-15 17:00:08 +0200 | [diff] [blame] | 557 | /* 'current' is not kept within the tree. */ |
| 558 | if (se->on_rq) { |
| 559 | /* |
| 560 | * Any task has to be enqueued before it get to execute on |
| 561 | * a CPU. So account for the time it spent waiting on the |
| 562 | * runqueue. |
| 563 | */ |
| 564 | update_stats_wait_end(cfs_rq, se); |
| 565 | __dequeue_entity(cfs_rq, se); |
| 566 | } |
| 567 | |
Ingo Molnar | 79303e9 | 2007-08-09 11:16:47 +0200 | [diff] [blame] | 568 | update_stats_curr_start(cfs_rq, se); |
Ingo Molnar | 429d43b | 2007-10-15 17:00:03 +0200 | [diff] [blame] | 569 | cfs_rq->curr = se; |
Ingo Molnar | eba1ed4 | 2007-10-15 17:00:02 +0200 | [diff] [blame] | 570 | #ifdef CONFIG_SCHEDSTATS |
| 571 | /* |
| 572 | * Track our maximum slice length, if the CPU's load is at |
| 573 | * least twice that of our own weight (i.e. dont track it |
| 574 | * when there are only lesser-weight tasks around): |
| 575 | */ |
Dmitry Adamushko | 495eca4 | 2007-10-15 17:00:06 +0200 | [diff] [blame] | 576 | if (rq_of(cfs_rq)->load.weight >= 2*se->load.weight) { |
Ingo Molnar | eba1ed4 | 2007-10-15 17:00:02 +0200 | [diff] [blame] | 577 | se->slice_max = max(se->slice_max, |
| 578 | se->sum_exec_runtime - se->prev_sum_exec_runtime); |
| 579 | } |
| 580 | #endif |
Peter Zijlstra | 4a55b45 | 2007-09-05 14:32:49 +0200 | [diff] [blame] | 581 | se->prev_sum_exec_runtime = se->sum_exec_runtime; |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 582 | } |
| 583 | |
Ingo Molnar | 9948f4b | 2007-08-09 11:16:48 +0200 | [diff] [blame] | 584 | static struct sched_entity *pick_next_entity(struct cfs_rq *cfs_rq) |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 585 | { |
| 586 | struct sched_entity *se = __pick_next_entity(cfs_rq); |
| 587 | |
Ingo Molnar | 8494f41 | 2007-08-09 11:16:48 +0200 | [diff] [blame] | 588 | set_next_entity(cfs_rq, se); |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 589 | |
| 590 | return se; |
| 591 | } |
| 592 | |
Ingo Molnar | ab6cde2 | 2007-08-09 11:16:48 +0200 | [diff] [blame] | 593 | static void put_prev_entity(struct cfs_rq *cfs_rq, struct sched_entity *prev) |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 594 | { |
| 595 | /* |
| 596 | * If still on the runqueue then deactivate_task() |
| 597 | * was not called and update_curr() has to be done: |
| 598 | */ |
| 599 | if (prev->on_rq) |
Ingo Molnar | b7cc089 | 2007-08-09 11:16:47 +0200 | [diff] [blame] | 600 | update_curr(cfs_rq); |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 601 | |
Ingo Molnar | c7e9b5b | 2007-08-09 11:16:48 +0200 | [diff] [blame] | 602 | update_stats_curr_end(cfs_rq, prev); |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 603 | |
Peter Zijlstra | ddc9729 | 2007-10-15 17:00:10 +0200 | [diff] [blame^] | 604 | check_spread(cfs_rq, prev); |
Dmitry Adamushko | 30cfdcf | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 605 | if (prev->on_rq) { |
Ingo Molnar | 5870db5 | 2007-08-09 11:16:47 +0200 | [diff] [blame] | 606 | update_stats_wait_start(cfs_rq, prev); |
Dmitry Adamushko | 30cfdcf | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 607 | /* Put 'current' back into the tree. */ |
| 608 | __enqueue_entity(cfs_rq, prev); |
| 609 | } |
Ingo Molnar | 429d43b | 2007-10-15 17:00:03 +0200 | [diff] [blame] | 610 | cfs_rq->curr = NULL; |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 611 | } |
| 612 | |
| 613 | static void entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr) |
| 614 | { |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 615 | /* |
Dmitry Adamushko | 30cfdcf | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 616 | * Update run-time statistics of the 'current'. |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 617 | */ |
Dmitry Adamushko | 30cfdcf | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 618 | update_curr(cfs_rq); |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 619 | |
Ingo Molnar | 2e09bf5 | 2007-10-15 17:00:05 +0200 | [diff] [blame] | 620 | if (cfs_rq->nr_running > 1) |
| 621 | check_preempt_tick(cfs_rq, curr); |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 622 | } |
| 623 | |
| 624 | /************************************************** |
| 625 | * CFS operations on tasks: |
| 626 | */ |
| 627 | |
| 628 | #ifdef CONFIG_FAIR_GROUP_SCHED |
| 629 | |
| 630 | /* Walk up scheduling entities hierarchy */ |
| 631 | #define for_each_sched_entity(se) \ |
| 632 | for (; se; se = se->parent) |
| 633 | |
| 634 | static inline struct cfs_rq *task_cfs_rq(struct task_struct *p) |
| 635 | { |
| 636 | return p->se.cfs_rq; |
| 637 | } |
| 638 | |
| 639 | /* runqueue on which this entity is (to be) queued */ |
| 640 | static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se) |
| 641 | { |
| 642 | return se->cfs_rq; |
| 643 | } |
| 644 | |
| 645 | /* runqueue "owned" by this group */ |
| 646 | static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp) |
| 647 | { |
| 648 | return grp->my_q; |
| 649 | } |
| 650 | |
| 651 | /* Given a group's cfs_rq on one cpu, return its corresponding cfs_rq on |
| 652 | * another cpu ('this_cpu') |
| 653 | */ |
| 654 | static inline struct cfs_rq *cpu_cfs_rq(struct cfs_rq *cfs_rq, int this_cpu) |
| 655 | { |
Srivatsa Vaddagiri | 29f59db | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 656 | return cfs_rq->tg->cfs_rq[this_cpu]; |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 657 | } |
| 658 | |
| 659 | /* Iterate thr' all leaf cfs_rq's on a runqueue */ |
| 660 | #define for_each_leaf_cfs_rq(rq, cfs_rq) \ |
| 661 | list_for_each_entry(cfs_rq, &rq->leaf_cfs_rq_list, leaf_cfs_rq_list) |
| 662 | |
| 663 | /* Do the two (enqueued) tasks belong to the same group ? */ |
| 664 | static inline int is_same_group(struct task_struct *curr, struct task_struct *p) |
| 665 | { |
| 666 | if (curr->se.cfs_rq == p->se.cfs_rq) |
| 667 | return 1; |
| 668 | |
| 669 | return 0; |
| 670 | } |
| 671 | |
| 672 | #else /* CONFIG_FAIR_GROUP_SCHED */ |
| 673 | |
| 674 | #define for_each_sched_entity(se) \ |
| 675 | for (; se; se = NULL) |
| 676 | |
| 677 | static inline struct cfs_rq *task_cfs_rq(struct task_struct *p) |
| 678 | { |
| 679 | return &task_rq(p)->cfs; |
| 680 | } |
| 681 | |
| 682 | static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se) |
| 683 | { |
| 684 | struct task_struct *p = task_of(se); |
| 685 | struct rq *rq = task_rq(p); |
| 686 | |
| 687 | return &rq->cfs; |
| 688 | } |
| 689 | |
| 690 | /* runqueue "owned" by this group */ |
| 691 | static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp) |
| 692 | { |
| 693 | return NULL; |
| 694 | } |
| 695 | |
| 696 | static inline struct cfs_rq *cpu_cfs_rq(struct cfs_rq *cfs_rq, int this_cpu) |
| 697 | { |
| 698 | return &cpu_rq(this_cpu)->cfs; |
| 699 | } |
| 700 | |
| 701 | #define for_each_leaf_cfs_rq(rq, cfs_rq) \ |
| 702 | for (cfs_rq = &rq->cfs; cfs_rq; cfs_rq = NULL) |
| 703 | |
| 704 | static inline int is_same_group(struct task_struct *curr, struct task_struct *p) |
| 705 | { |
| 706 | return 1; |
| 707 | } |
| 708 | |
| 709 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
| 710 | |
| 711 | /* |
| 712 | * The enqueue_task method is called before nr_running is |
| 713 | * increased. Here we update the fair scheduling stats and |
| 714 | * then put the task into the rbtree: |
| 715 | */ |
Ingo Molnar | fd390f6 | 2007-08-09 11:16:48 +0200 | [diff] [blame] | 716 | static void enqueue_task_fair(struct rq *rq, struct task_struct *p, int wakeup) |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 717 | { |
| 718 | struct cfs_rq *cfs_rq; |
| 719 | struct sched_entity *se = &p->se; |
| 720 | |
| 721 | for_each_sched_entity(se) { |
| 722 | if (se->on_rq) |
| 723 | break; |
| 724 | cfs_rq = cfs_rq_of(se); |
Srivatsa Vaddagiri | 83b699e | 2007-10-15 17:00:08 +0200 | [diff] [blame] | 725 | enqueue_entity(cfs_rq, se, wakeup); |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 726 | } |
| 727 | } |
| 728 | |
| 729 | /* |
| 730 | * The dequeue_task method is called before nr_running is |
| 731 | * decreased. We remove the task from the rbtree and |
| 732 | * update the fair scheduling stats: |
| 733 | */ |
Ingo Molnar | f02231e | 2007-08-09 11:16:48 +0200 | [diff] [blame] | 734 | static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int sleep) |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 735 | { |
| 736 | struct cfs_rq *cfs_rq; |
| 737 | struct sched_entity *se = &p->se; |
| 738 | |
| 739 | for_each_sched_entity(se) { |
| 740 | cfs_rq = cfs_rq_of(se); |
Ingo Molnar | 525c271 | 2007-08-09 11:16:48 +0200 | [diff] [blame] | 741 | dequeue_entity(cfs_rq, se, sleep); |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 742 | /* Don't dequeue parent if it has other entities besides us */ |
| 743 | if (cfs_rq->load.weight) |
| 744 | break; |
| 745 | } |
| 746 | } |
| 747 | |
| 748 | /* |
Ingo Molnar | 1799e35 | 2007-09-19 23:34:46 +0200 | [diff] [blame] | 749 | * sched_yield() support is very simple - we dequeue and enqueue. |
| 750 | * |
| 751 | * If compat_yield is turned on then we requeue to the end of the tree. |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 752 | */ |
Dmitry Adamushko | 4530d7a | 2007-10-15 17:00:08 +0200 | [diff] [blame] | 753 | static void yield_task_fair(struct rq *rq) |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 754 | { |
Srivatsa Vaddagiri | 72ea22f | 2007-10-15 17:00:08 +0200 | [diff] [blame] | 755 | struct cfs_rq *cfs_rq = task_cfs_rq(rq->curr); |
Ingo Molnar | 1799e35 | 2007-09-19 23:34:46 +0200 | [diff] [blame] | 756 | struct rb_node **link = &cfs_rq->tasks_timeline.rb_node; |
Dmitry Adamushko | 4530d7a | 2007-10-15 17:00:08 +0200 | [diff] [blame] | 757 | struct sched_entity *rightmost, *se = &rq->curr->se; |
Ingo Molnar | 1799e35 | 2007-09-19 23:34:46 +0200 | [diff] [blame] | 758 | struct rb_node *parent; |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 759 | |
| 760 | /* |
Ingo Molnar | 1799e35 | 2007-09-19 23:34:46 +0200 | [diff] [blame] | 761 | * Are we the only task in the tree? |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 762 | */ |
Ingo Molnar | 1799e35 | 2007-09-19 23:34:46 +0200 | [diff] [blame] | 763 | if (unlikely(cfs_rq->nr_running == 1)) |
| 764 | return; |
| 765 | |
| 766 | if (likely(!sysctl_sched_compat_yield)) { |
| 767 | __update_rq_clock(rq); |
| 768 | /* |
| 769 | * Dequeue and enqueue the task to update its |
| 770 | * position within the tree: |
| 771 | */ |
Dmitry Adamushko | 4530d7a | 2007-10-15 17:00:08 +0200 | [diff] [blame] | 772 | dequeue_entity(cfs_rq, se, 0); |
Srivatsa Vaddagiri | 83b699e | 2007-10-15 17:00:08 +0200 | [diff] [blame] | 773 | enqueue_entity(cfs_rq, se, 0); |
Ingo Molnar | 1799e35 | 2007-09-19 23:34:46 +0200 | [diff] [blame] | 774 | |
| 775 | return; |
| 776 | } |
| 777 | /* |
| 778 | * Find the rightmost entry in the rbtree: |
| 779 | */ |
| 780 | do { |
| 781 | parent = *link; |
| 782 | link = &parent->rb_right; |
| 783 | } while (*link); |
| 784 | |
| 785 | rightmost = rb_entry(parent, struct sched_entity, run_node); |
| 786 | /* |
| 787 | * Already in the rightmost position? |
| 788 | */ |
| 789 | if (unlikely(rightmost == se)) |
| 790 | return; |
| 791 | |
| 792 | /* |
| 793 | * Minimally necessary key value to be last in the tree: |
| 794 | */ |
Dmitry Adamushko | 30cfdcf | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 795 | se->vruntime = rightmost->vruntime + 1; |
Ingo Molnar | 1799e35 | 2007-09-19 23:34:46 +0200 | [diff] [blame] | 796 | |
| 797 | if (cfs_rq->rb_leftmost == &se->run_node) |
| 798 | cfs_rq->rb_leftmost = rb_next(&se->run_node); |
| 799 | /* |
| 800 | * Relink the task to the rightmost position: |
| 801 | */ |
| 802 | rb_erase(&se->run_node, &cfs_rq->tasks_timeline); |
| 803 | rb_link_node(&se->run_node, parent, link); |
| 804 | rb_insert_color(&se->run_node, &cfs_rq->tasks_timeline); |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 805 | } |
| 806 | |
| 807 | /* |
| 808 | * Preempt the current task with a newly woken task if needed: |
| 809 | */ |
Ingo Molnar | 2e09bf5 | 2007-10-15 17:00:05 +0200 | [diff] [blame] | 810 | static void check_preempt_wakeup(struct rq *rq, struct task_struct *p) |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 811 | { |
| 812 | struct task_struct *curr = rq->curr; |
| 813 | struct cfs_rq *cfs_rq = task_cfs_rq(curr); |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 814 | |
| 815 | if (unlikely(rt_prio(p->prio))) { |
Ingo Molnar | a8e504d | 2007-08-09 11:16:47 +0200 | [diff] [blame] | 816 | update_rq_clock(rq); |
Ingo Molnar | b7cc089 | 2007-08-09 11:16:47 +0200 | [diff] [blame] | 817 | update_curr(cfs_rq); |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 818 | resched_task(curr); |
| 819 | return; |
| 820 | } |
Ingo Molnar | 2e09bf5 | 2007-10-15 17:00:05 +0200 | [diff] [blame] | 821 | if (is_same_group(curr, p)) { |
| 822 | s64 delta = curr->se.vruntime - p->se.vruntime; |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 823 | |
Ingo Molnar | 2e09bf5 | 2007-10-15 17:00:05 +0200 | [diff] [blame] | 824 | if (delta > (s64)sysctl_sched_wakeup_granularity) |
| 825 | resched_task(curr); |
| 826 | } |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 827 | } |
| 828 | |
Ingo Molnar | fb8d472 | 2007-08-09 11:16:48 +0200 | [diff] [blame] | 829 | static struct task_struct *pick_next_task_fair(struct rq *rq) |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 830 | { |
| 831 | struct cfs_rq *cfs_rq = &rq->cfs; |
| 832 | struct sched_entity *se; |
| 833 | |
| 834 | if (unlikely(!cfs_rq->nr_running)) |
| 835 | return NULL; |
| 836 | |
| 837 | do { |
Ingo Molnar | 9948f4b | 2007-08-09 11:16:48 +0200 | [diff] [blame] | 838 | se = pick_next_entity(cfs_rq); |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 839 | cfs_rq = group_cfs_rq(se); |
| 840 | } while (cfs_rq); |
| 841 | |
| 842 | return task_of(se); |
| 843 | } |
| 844 | |
| 845 | /* |
| 846 | * Account for a descheduled task: |
| 847 | */ |
Ingo Molnar | 31ee529 | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 848 | static void put_prev_task_fair(struct rq *rq, struct task_struct *prev) |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 849 | { |
| 850 | struct sched_entity *se = &prev->se; |
| 851 | struct cfs_rq *cfs_rq; |
| 852 | |
| 853 | for_each_sched_entity(se) { |
| 854 | cfs_rq = cfs_rq_of(se); |
Ingo Molnar | ab6cde2 | 2007-08-09 11:16:48 +0200 | [diff] [blame] | 855 | put_prev_entity(cfs_rq, se); |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 856 | } |
| 857 | } |
| 858 | |
| 859 | /************************************************** |
| 860 | * Fair scheduling class load-balancing methods: |
| 861 | */ |
| 862 | |
| 863 | /* |
| 864 | * Load-balancing iterator. Note: while the runqueue stays locked |
| 865 | * during the whole iteration, the current task might be |
| 866 | * dequeued so the iterator has to be dequeue-safe. Here we |
| 867 | * achieve that by always pre-iterating before returning |
| 868 | * the current task: |
| 869 | */ |
| 870 | static inline struct task_struct * |
| 871 | __load_balance_iterator(struct cfs_rq *cfs_rq, struct rb_node *curr) |
| 872 | { |
| 873 | struct task_struct *p; |
| 874 | |
| 875 | if (!curr) |
| 876 | return NULL; |
| 877 | |
| 878 | p = rb_entry(curr, struct task_struct, se.run_node); |
| 879 | cfs_rq->rb_load_balance_curr = rb_next(curr); |
| 880 | |
| 881 | return p; |
| 882 | } |
| 883 | |
| 884 | static struct task_struct *load_balance_start_fair(void *arg) |
| 885 | { |
| 886 | struct cfs_rq *cfs_rq = arg; |
| 887 | |
| 888 | return __load_balance_iterator(cfs_rq, first_fair(cfs_rq)); |
| 889 | } |
| 890 | |
| 891 | static struct task_struct *load_balance_next_fair(void *arg) |
| 892 | { |
| 893 | struct cfs_rq *cfs_rq = arg; |
| 894 | |
| 895 | return __load_balance_iterator(cfs_rq, cfs_rq->rb_load_balance_curr); |
| 896 | } |
| 897 | |
Peter Williams | a4ac01c | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 898 | #ifdef CONFIG_FAIR_GROUP_SCHED |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 899 | static int cfs_rq_best_prio(struct cfs_rq *cfs_rq) |
| 900 | { |
| 901 | struct sched_entity *curr; |
| 902 | struct task_struct *p; |
| 903 | |
| 904 | if (!cfs_rq->nr_running) |
| 905 | return MAX_PRIO; |
| 906 | |
Srivatsa Vaddagiri | 9b5b775 | 2007-10-15 17:00:09 +0200 | [diff] [blame] | 907 | curr = cfs_rq->curr; |
| 908 | if (!curr) |
| 909 | curr = __pick_next_entity(cfs_rq); |
| 910 | |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 911 | p = task_of(curr); |
| 912 | |
| 913 | return p->prio; |
| 914 | } |
Peter Williams | a4ac01c | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 915 | #endif |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 916 | |
Peter Williams | 4301065 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 917 | static unsigned long |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 918 | load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest, |
Peter Williams | a4ac01c | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 919 | unsigned long max_nr_move, unsigned long max_load_move, |
| 920 | struct sched_domain *sd, enum cpu_idle_type idle, |
| 921 | int *all_pinned, int *this_best_prio) |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 922 | { |
| 923 | struct cfs_rq *busy_cfs_rq; |
| 924 | unsigned long load_moved, total_nr_moved = 0, nr_moved; |
| 925 | long rem_load_move = max_load_move; |
| 926 | struct rq_iterator cfs_rq_iterator; |
| 927 | |
| 928 | cfs_rq_iterator.start = load_balance_start_fair; |
| 929 | cfs_rq_iterator.next = load_balance_next_fair; |
| 930 | |
| 931 | for_each_leaf_cfs_rq(busiest, busy_cfs_rq) { |
Peter Williams | a4ac01c | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 932 | #ifdef CONFIG_FAIR_GROUP_SCHED |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 933 | struct cfs_rq *this_cfs_rq; |
Ingo Molnar | e56f31a | 2007-08-10 23:05:11 +0200 | [diff] [blame] | 934 | long imbalance; |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 935 | unsigned long maxload; |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 936 | |
| 937 | this_cfs_rq = cpu_cfs_rq(busy_cfs_rq, this_cpu); |
| 938 | |
Ingo Molnar | e56f31a | 2007-08-10 23:05:11 +0200 | [diff] [blame] | 939 | imbalance = busy_cfs_rq->load.weight - this_cfs_rq->load.weight; |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 940 | /* Don't pull if this_cfs_rq has more load than busy_cfs_rq */ |
| 941 | if (imbalance <= 0) |
| 942 | continue; |
| 943 | |
| 944 | /* Don't pull more than imbalance/2 */ |
| 945 | imbalance /= 2; |
| 946 | maxload = min(rem_load_move, imbalance); |
| 947 | |
Peter Williams | a4ac01c | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 948 | *this_best_prio = cfs_rq_best_prio(this_cfs_rq); |
| 949 | #else |
Ingo Molnar | e56f31a | 2007-08-10 23:05:11 +0200 | [diff] [blame] | 950 | # define maxload rem_load_move |
Peter Williams | a4ac01c | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 951 | #endif |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 952 | /* pass busy_cfs_rq argument into |
| 953 | * load_balance_[start|next]_fair iterators |
| 954 | */ |
| 955 | cfs_rq_iterator.arg = busy_cfs_rq; |
| 956 | nr_moved = balance_tasks(this_rq, this_cpu, busiest, |
| 957 | max_nr_move, maxload, sd, idle, all_pinned, |
Peter Williams | a4ac01c | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 958 | &load_moved, this_best_prio, &cfs_rq_iterator); |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 959 | |
| 960 | total_nr_moved += nr_moved; |
| 961 | max_nr_move -= nr_moved; |
| 962 | rem_load_move -= load_moved; |
| 963 | |
| 964 | if (max_nr_move <= 0 || rem_load_move <= 0) |
| 965 | break; |
| 966 | } |
| 967 | |
Peter Williams | 4301065 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 968 | return max_load_move - rem_load_move; |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 969 | } |
| 970 | |
| 971 | /* |
| 972 | * scheduler tick hitting a task of our scheduling class: |
| 973 | */ |
| 974 | static void task_tick_fair(struct rq *rq, struct task_struct *curr) |
| 975 | { |
| 976 | struct cfs_rq *cfs_rq; |
| 977 | struct sched_entity *se = &curr->se; |
| 978 | |
| 979 | for_each_sched_entity(se) { |
| 980 | cfs_rq = cfs_rq_of(se); |
| 981 | entity_tick(cfs_rq, se); |
| 982 | } |
| 983 | } |
| 984 | |
Peter Zijlstra | 4d78e7b | 2007-10-15 17:00:04 +0200 | [diff] [blame] | 985 | #define swap(a,b) do { typeof(a) tmp = (a); (a) = (b); (b) = tmp; } while (0) |
| 986 | |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 987 | /* |
| 988 | * Share the fairness runtime between parent and child, thus the |
| 989 | * total amount of pressure for CPU stays equal - new tasks |
| 990 | * get a chance to run but frequent forkers are not allowed to |
| 991 | * monopolize the CPU. Note: the parent runqueue is locked, |
| 992 | * the child is not running yet. |
| 993 | */ |
Ingo Molnar | ee0827d | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 994 | static void task_new_fair(struct rq *rq, struct task_struct *p) |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 995 | { |
| 996 | struct cfs_rq *cfs_rq = task_cfs_rq(p); |
Ingo Molnar | 429d43b | 2007-10-15 17:00:03 +0200 | [diff] [blame] | 997 | struct sched_entity *se = &p->se, *curr = cfs_rq->curr; |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 998 | |
| 999 | sched_info_queued(p); |
| 1000 | |
Ting Yang | 7109c442 | 2007-08-28 12:53:24 +0200 | [diff] [blame] | 1001 | update_curr(cfs_rq); |
Peter Zijlstra | aeb73b0 | 2007-10-15 17:00:05 +0200 | [diff] [blame] | 1002 | place_entity(cfs_rq, se, 1); |
Peter Zijlstra | 4d78e7b | 2007-10-15 17:00:04 +0200 | [diff] [blame] | 1003 | |
Peter Zijlstra | 4d78e7b | 2007-10-15 17:00:04 +0200 | [diff] [blame] | 1004 | if (sysctl_sched_child_runs_first && |
| 1005 | curr->vruntime < se->vruntime) { |
Dmitry Adamushko | 87fefa3 | 2007-10-15 17:00:08 +0200 | [diff] [blame] | 1006 | /* |
Ingo Molnar | edcb60a | 2007-10-15 17:00:08 +0200 | [diff] [blame] | 1007 | * Upon rescheduling, sched_class::put_prev_task() will place |
| 1008 | * 'current' within the tree based on its new key value. |
| 1009 | */ |
Peter Zijlstra | 4d78e7b | 2007-10-15 17:00:04 +0200 | [diff] [blame] | 1010 | swap(curr->vruntime, se->vruntime); |
Peter Zijlstra | 4d78e7b | 2007-10-15 17:00:04 +0200 | [diff] [blame] | 1011 | } |
| 1012 | |
Ingo Molnar | e9acbff | 2007-10-15 17:00:04 +0200 | [diff] [blame] | 1013 | update_stats_enqueue(cfs_rq, se); |
Peter Zijlstra | ddc9729 | 2007-10-15 17:00:10 +0200 | [diff] [blame^] | 1014 | check_spread(cfs_rq, se); |
| 1015 | check_spread(cfs_rq, curr); |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1016 | __enqueue_entity(cfs_rq, se); |
Dmitry Adamushko | 30cfdcf | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 1017 | account_entity_enqueue(cfs_rq, se); |
Ingo Molnar | bb61c21 | 2007-10-15 17:00:02 +0200 | [diff] [blame] | 1018 | resched_task(rq->curr); |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1019 | } |
| 1020 | |
Srivatsa Vaddagiri | 83b699e | 2007-10-15 17:00:08 +0200 | [diff] [blame] | 1021 | /* Account for a task changing its policy or group. |
| 1022 | * |
| 1023 | * This routine is mostly called to set cfs_rq->curr field when a task |
| 1024 | * migrates between groups/classes. |
| 1025 | */ |
| 1026 | static void set_curr_task_fair(struct rq *rq) |
| 1027 | { |
| 1028 | struct sched_entity *se = &rq->curr->se; |
| 1029 | |
| 1030 | for_each_sched_entity(se) |
| 1031 | set_next_entity(cfs_rq_of(se), se); |
| 1032 | } |
| 1033 | |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1034 | /* |
| 1035 | * All the scheduling class methods: |
| 1036 | */ |
| 1037 | struct sched_class fair_sched_class __read_mostly = { |
| 1038 | .enqueue_task = enqueue_task_fair, |
| 1039 | .dequeue_task = dequeue_task_fair, |
| 1040 | .yield_task = yield_task_fair, |
| 1041 | |
Ingo Molnar | 2e09bf5 | 2007-10-15 17:00:05 +0200 | [diff] [blame] | 1042 | .check_preempt_curr = check_preempt_wakeup, |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1043 | |
| 1044 | .pick_next_task = pick_next_task_fair, |
| 1045 | .put_prev_task = put_prev_task_fair, |
| 1046 | |
| 1047 | .load_balance = load_balance_fair, |
| 1048 | |
Srivatsa Vaddagiri | 83b699e | 2007-10-15 17:00:08 +0200 | [diff] [blame] | 1049 | .set_curr_task = set_curr_task_fair, |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1050 | .task_tick = task_tick_fair, |
| 1051 | .task_new = task_new_fair, |
| 1052 | }; |
| 1053 | |
| 1054 | #ifdef CONFIG_SCHED_DEBUG |
Ingo Molnar | 5cef9ec | 2007-08-09 11:16:47 +0200 | [diff] [blame] | 1055 | static void print_cfs_stats(struct seq_file *m, int cpu) |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1056 | { |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1057 | struct cfs_rq *cfs_rq; |
| 1058 | |
Srivatsa Vaddagiri | 75c28ac | 2007-10-15 17:00:09 +0200 | [diff] [blame] | 1059 | #ifdef CONFIG_FAIR_GROUP_SCHED |
| 1060 | print_cfs_rq(m, cpu, &cpu_rq(cpu)->cfs); |
| 1061 | #endif |
Ingo Molnar | c3b64f1 | 2007-08-09 11:16:51 +0200 | [diff] [blame] | 1062 | for_each_leaf_cfs_rq(cpu_rq(cpu), cfs_rq) |
Ingo Molnar | 5cef9ec | 2007-08-09 11:16:47 +0200 | [diff] [blame] | 1063 | print_cfs_rq(m, cpu, cfs_rq); |
Ingo Molnar | bf0f6f2 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1064 | } |
| 1065 | #endif |