Greg Kroah-Hartman | b244131 | 2017-11-01 15:07:57 +0100 | [diff] [blame] | 1 | // SPDX-License-Identifier: GPL-2.0 |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 2 | /* |
| 3 | * Real-Time Scheduling Class (mapped to the SCHED_FIFO and SCHED_RR |
| 4 | * policies) |
| 5 | */ |
Vincent Guittot | 371bf42 | 2018-06-28 17:45:05 +0200 | [diff] [blame] | 6 | |
Clark Williams | ce0dbbb | 2013-02-07 09:47:04 -0600 | [diff] [blame] | 7 | int sched_rr_timeslice = RR_TIMESLICE; |
Huaixin Chang | d505b8a | 2020-04-25 18:52:48 +0800 | [diff] [blame] | 8 | /* More than 4 hours if BW_SHIFT equals 20. */ |
| 9 | static const u64 max_rt_runtime = MAX_BW; |
Clark Williams | ce0dbbb | 2013-02-07 09:47:04 -0600 | [diff] [blame] | 10 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 11 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); |
| 12 | |
| 13 | struct rt_bandwidth def_rt_bandwidth; |
| 14 | |
Zhen Ni | d9ab0e6 | 2022-02-15 19:45:59 +0800 | [diff] [blame] | 15 | /* |
| 16 | * period over which we measure -rt task CPU usage in us. |
| 17 | * default: 1s |
| 18 | */ |
| 19 | unsigned int sysctl_sched_rt_period = 1000000; |
| 20 | |
| 21 | /* |
| 22 | * part of the period that we allow rt tasks to run in us. |
| 23 | * default: 0.95s |
| 24 | */ |
| 25 | int sysctl_sched_rt_runtime = 950000; |
| 26 | |
Baisong Zhong | 28f152c | 2022-03-18 10:54:17 +0800 | [diff] [blame] | 27 | #ifdef CONFIG_SYSCTL |
Cyril Hrubis | c7fcb99 | 2023-08-02 17:19:05 +0200 | [diff] [blame] | 28 | static int sysctl_sched_rr_timeslice = (MSEC_PER_SEC * RR_TIMESLICE) / HZ; |
Zhen Ni | d9ab0e6 | 2022-02-15 19:45:59 +0800 | [diff] [blame] | 29 | static int sched_rt_handler(struct ctl_table *table, int write, void *buffer, |
| 30 | size_t *lenp, loff_t *ppos); |
Zhen Ni | dafd7a9 | 2022-02-15 19:46:01 +0800 | [diff] [blame] | 31 | static int sched_rr_handler(struct ctl_table *table, int write, void *buffer, |
| 32 | size_t *lenp, loff_t *ppos); |
Zhen Ni | d9ab0e6 | 2022-02-15 19:45:59 +0800 | [diff] [blame] | 33 | static struct ctl_table sched_rt_sysctls[] = { |
| 34 | { |
| 35 | .procname = "sched_rt_period_us", |
| 36 | .data = &sysctl_sched_rt_period, |
| 37 | .maxlen = sizeof(unsigned int), |
| 38 | .mode = 0644, |
| 39 | .proc_handler = sched_rt_handler, |
| 40 | }, |
| 41 | { |
| 42 | .procname = "sched_rt_runtime_us", |
| 43 | .data = &sysctl_sched_rt_runtime, |
| 44 | .maxlen = sizeof(int), |
| 45 | .mode = 0644, |
| 46 | .proc_handler = sched_rt_handler, |
| 47 | }, |
Zhen Ni | dafd7a9 | 2022-02-15 19:46:01 +0800 | [diff] [blame] | 48 | { |
| 49 | .procname = "sched_rr_timeslice_ms", |
| 50 | .data = &sysctl_sched_rr_timeslice, |
| 51 | .maxlen = sizeof(int), |
| 52 | .mode = 0644, |
| 53 | .proc_handler = sched_rr_handler, |
| 54 | }, |
Zhen Ni | d9ab0e6 | 2022-02-15 19:45:59 +0800 | [diff] [blame] | 55 | {} |
| 56 | }; |
| 57 | |
| 58 | static int __init sched_rt_sysctl_init(void) |
| 59 | { |
| 60 | register_sysctl_init("kernel", sched_rt_sysctls); |
| 61 | return 0; |
| 62 | } |
| 63 | late_initcall(sched_rt_sysctl_init); |
| 64 | #endif |
| 65 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 66 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) |
| 67 | { |
| 68 | struct rt_bandwidth *rt_b = |
| 69 | container_of(timer, struct rt_bandwidth, rt_period_timer); |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 70 | int idle = 0; |
Peter Zijlstra | 77a4d1a | 2015-04-15 11:41:57 +0200 | [diff] [blame] | 71 | int overrun; |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 72 | |
Peter Zijlstra | 77a4d1a | 2015-04-15 11:41:57 +0200 | [diff] [blame] | 73 | raw_spin_lock(&rt_b->rt_runtime_lock); |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 74 | for (;;) { |
Peter Zijlstra | 77a4d1a | 2015-04-15 11:41:57 +0200 | [diff] [blame] | 75 | overrun = hrtimer_forward_now(timer, rt_b->rt_period); |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 76 | if (!overrun) |
| 77 | break; |
| 78 | |
Peter Zijlstra | 77a4d1a | 2015-04-15 11:41:57 +0200 | [diff] [blame] | 79 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 80 | idle = do_sched_rt_period_timer(rt_b, overrun); |
Peter Zijlstra | 77a4d1a | 2015-04-15 11:41:57 +0200 | [diff] [blame] | 81 | raw_spin_lock(&rt_b->rt_runtime_lock); |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 82 | } |
Peter Zijlstra | 4cfafd3 | 2015-05-14 12:23:11 +0200 | [diff] [blame] | 83 | if (idle) |
| 84 | rt_b->rt_period_active = 0; |
Peter Zijlstra | 77a4d1a | 2015-04-15 11:41:57 +0200 | [diff] [blame] | 85 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 86 | |
| 87 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; |
| 88 | } |
| 89 | |
| 90 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) |
| 91 | { |
| 92 | rt_b->rt_period = ns_to_ktime(period); |
| 93 | rt_b->rt_runtime = runtime; |
| 94 | |
| 95 | raw_spin_lock_init(&rt_b->rt_runtime_lock); |
| 96 | |
Sebastian Andrzej Siewior | d5096aa | 2019-07-26 20:30:52 +0200 | [diff] [blame] | 97 | hrtimer_init(&rt_b->rt_period_timer, CLOCK_MONOTONIC, |
| 98 | HRTIMER_MODE_REL_HARD); |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 99 | rt_b->rt_period_timer.function = sched_rt_period_timer; |
| 100 | } |
| 101 | |
Li Hua | 9b58e97 | 2021-12-03 03:36:18 +0000 | [diff] [blame] | 102 | static inline void do_start_rt_bandwidth(struct rt_bandwidth *rt_b) |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 103 | { |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 104 | raw_spin_lock(&rt_b->rt_runtime_lock); |
Peter Zijlstra | 4cfafd3 | 2015-05-14 12:23:11 +0200 | [diff] [blame] | 105 | if (!rt_b->rt_period_active) { |
| 106 | rt_b->rt_period_active = 1; |
Steven Rostedt | c3a990d | 2016-02-16 18:37:46 -0500 | [diff] [blame] | 107 | /* |
| 108 | * SCHED_DEADLINE updates the bandwidth, as a run away |
| 109 | * RT task with a DL task could hog a CPU. But DL does |
| 110 | * not reset the period. If a deadline task was running |
| 111 | * without an RT task running, it can cause RT tasks to |
| 112 | * throttle when they start up. Kick the timer right away |
| 113 | * to update the period. |
| 114 | */ |
| 115 | hrtimer_forward_now(&rt_b->rt_period_timer, ns_to_ktime(0)); |
Sebastian Andrzej Siewior | d5096aa | 2019-07-26 20:30:52 +0200 | [diff] [blame] | 116 | hrtimer_start_expires(&rt_b->rt_period_timer, |
| 117 | HRTIMER_MODE_ABS_PINNED_HARD); |
Peter Zijlstra | 4cfafd3 | 2015-05-14 12:23:11 +0200 | [diff] [blame] | 118 | } |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 119 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
| 120 | } |
| 121 | |
Li Hua | 9b58e97 | 2021-12-03 03:36:18 +0000 | [diff] [blame] | 122 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) |
| 123 | { |
| 124 | if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF) |
| 125 | return; |
| 126 | |
| 127 | do_start_rt_bandwidth(rt_b); |
| 128 | } |
| 129 | |
Abel Vesa | 07c54f7 | 2015-03-03 13:50:27 +0200 | [diff] [blame] | 130 | void init_rt_rq(struct rt_rq *rt_rq) |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 131 | { |
| 132 | struct rt_prio_array *array; |
| 133 | int i; |
| 134 | |
| 135 | array = &rt_rq->active; |
| 136 | for (i = 0; i < MAX_RT_PRIO; i++) { |
| 137 | INIT_LIST_HEAD(array->queue + i); |
| 138 | __clear_bit(i, array->bitmap); |
| 139 | } |
| 140 | /* delimiter for bitsearch: */ |
| 141 | __set_bit(MAX_RT_PRIO, array->bitmap); |
| 142 | |
| 143 | #if defined CONFIG_SMP |
Peter Zijlstra | 934fc33 | 2020-10-14 21:06:49 +0200 | [diff] [blame] | 144 | rt_rq->highest_prio.curr = MAX_RT_PRIO-1; |
| 145 | rt_rq->highest_prio.next = MAX_RT_PRIO-1; |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 146 | rt_rq->rt_nr_migratory = 0; |
| 147 | rt_rq->overloaded = 0; |
| 148 | plist_head_init(&rt_rq->pushable_tasks); |
Steven Rostedt | b6366f0 | 2015-03-18 14:49:46 -0400 | [diff] [blame] | 149 | #endif /* CONFIG_SMP */ |
Kirill Tkhai | f4ebcbc | 2014-03-15 02:15:00 +0400 | [diff] [blame] | 150 | /* We start is dequeued state, because no RT tasks are queued */ |
| 151 | rt_rq->rt_queued = 0; |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 152 | |
| 153 | rt_rq->rt_time = 0; |
| 154 | rt_rq->rt_throttled = 0; |
| 155 | rt_rq->rt_runtime = 0; |
| 156 | raw_spin_lock_init(&rt_rq->rt_runtime_lock); |
| 157 | } |
| 158 | |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 159 | #ifdef CONFIG_RT_GROUP_SCHED |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 160 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) |
| 161 | { |
| 162 | hrtimer_cancel(&rt_b->rt_period_timer); |
| 163 | } |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 164 | |
Peter Zijlstra | a1ba4d8 | 2009-04-01 18:40:15 +0200 | [diff] [blame] | 165 | #define rt_entity_is_task(rt_se) (!(rt_se)->my_q) |
| 166 | |
Peter Zijlstra | 8f48894 | 2009-07-24 12:25:30 +0200 | [diff] [blame] | 167 | static inline struct task_struct *rt_task_of(struct sched_rt_entity *rt_se) |
| 168 | { |
| 169 | #ifdef CONFIG_SCHED_DEBUG |
| 170 | WARN_ON_ONCE(!rt_entity_is_task(rt_se)); |
| 171 | #endif |
| 172 | return container_of(rt_se, struct task_struct, rt); |
| 173 | } |
| 174 | |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 175 | static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq) |
| 176 | { |
| 177 | return rt_rq->rq; |
| 178 | } |
| 179 | |
| 180 | static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se) |
| 181 | { |
| 182 | return rt_se->rt_rq; |
| 183 | } |
| 184 | |
Kirill Tkhai | 653d07a | 2014-03-15 02:14:55 +0400 | [diff] [blame] | 185 | static inline struct rq *rq_of_rt_se(struct sched_rt_entity *rt_se) |
| 186 | { |
| 187 | struct rt_rq *rt_rq = rt_se->rt_rq; |
| 188 | |
| 189 | return rt_rq->rq; |
| 190 | } |
| 191 | |
Mathias Krause | b027789 | 2021-11-03 20:06:13 +0100 | [diff] [blame] | 192 | void unregister_rt_sched_group(struct task_group *tg) |
| 193 | { |
| 194 | if (tg->rt_se) |
| 195 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
| 196 | |
| 197 | } |
| 198 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 199 | void free_rt_sched_group(struct task_group *tg) |
| 200 | { |
| 201 | int i; |
| 202 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 203 | for_each_possible_cpu(i) { |
| 204 | if (tg->rt_rq) |
| 205 | kfree(tg->rt_rq[i]); |
| 206 | if (tg->rt_se) |
| 207 | kfree(tg->rt_se[i]); |
| 208 | } |
| 209 | |
| 210 | kfree(tg->rt_rq); |
| 211 | kfree(tg->rt_se); |
| 212 | } |
| 213 | |
| 214 | void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
| 215 | struct sched_rt_entity *rt_se, int cpu, |
| 216 | struct sched_rt_entity *parent) |
| 217 | { |
| 218 | struct rq *rq = cpu_rq(cpu); |
| 219 | |
Peter Zijlstra | 934fc33 | 2020-10-14 21:06:49 +0200 | [diff] [blame] | 220 | rt_rq->highest_prio.curr = MAX_RT_PRIO-1; |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 221 | rt_rq->rt_nr_boosted = 0; |
| 222 | rt_rq->rq = rq; |
| 223 | rt_rq->tg = tg; |
| 224 | |
| 225 | tg->rt_rq[cpu] = rt_rq; |
| 226 | tg->rt_se[cpu] = rt_se; |
| 227 | |
| 228 | if (!rt_se) |
| 229 | return; |
| 230 | |
| 231 | if (!parent) |
| 232 | rt_se->rt_rq = &rq->rt; |
| 233 | else |
| 234 | rt_se->rt_rq = parent->my_q; |
| 235 | |
| 236 | rt_se->my_q = rt_rq; |
| 237 | rt_se->parent = parent; |
| 238 | INIT_LIST_HEAD(&rt_se->run_list); |
| 239 | } |
| 240 | |
| 241 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) |
| 242 | { |
| 243 | struct rt_rq *rt_rq; |
| 244 | struct sched_rt_entity *rt_se; |
| 245 | int i; |
| 246 | |
Kees Cook | 6396bb2 | 2018-06-12 14:03:40 -0700 | [diff] [blame] | 247 | tg->rt_rq = kcalloc(nr_cpu_ids, sizeof(rt_rq), GFP_KERNEL); |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 248 | if (!tg->rt_rq) |
| 249 | goto err; |
Kees Cook | 6396bb2 | 2018-06-12 14:03:40 -0700 | [diff] [blame] | 250 | tg->rt_se = kcalloc(nr_cpu_ids, sizeof(rt_se), GFP_KERNEL); |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 251 | if (!tg->rt_se) |
| 252 | goto err; |
| 253 | |
| 254 | init_rt_bandwidth(&tg->rt_bandwidth, |
| 255 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); |
| 256 | |
| 257 | for_each_possible_cpu(i) { |
| 258 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
| 259 | GFP_KERNEL, cpu_to_node(i)); |
| 260 | if (!rt_rq) |
| 261 | goto err; |
| 262 | |
| 263 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
| 264 | GFP_KERNEL, cpu_to_node(i)); |
| 265 | if (!rt_se) |
| 266 | goto err_free_rq; |
| 267 | |
Abel Vesa | 07c54f7 | 2015-03-03 13:50:27 +0200 | [diff] [blame] | 268 | init_rt_rq(rt_rq); |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 269 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
| 270 | init_tg_rt_entry(tg, rt_rq, rt_se, i, parent->rt_se[i]); |
| 271 | } |
| 272 | |
| 273 | return 1; |
| 274 | |
| 275 | err_free_rq: |
| 276 | kfree(rt_rq); |
| 277 | err: |
| 278 | return 0; |
| 279 | } |
| 280 | |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 281 | #else /* CONFIG_RT_GROUP_SCHED */ |
| 282 | |
Peter Zijlstra | a1ba4d8 | 2009-04-01 18:40:15 +0200 | [diff] [blame] | 283 | #define rt_entity_is_task(rt_se) (1) |
| 284 | |
Peter Zijlstra | 8f48894 | 2009-07-24 12:25:30 +0200 | [diff] [blame] | 285 | static inline struct task_struct *rt_task_of(struct sched_rt_entity *rt_se) |
| 286 | { |
| 287 | return container_of(rt_se, struct task_struct, rt); |
| 288 | } |
| 289 | |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 290 | static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq) |
| 291 | { |
| 292 | return container_of(rt_rq, struct rq, rt); |
| 293 | } |
| 294 | |
Kirill Tkhai | 653d07a | 2014-03-15 02:14:55 +0400 | [diff] [blame] | 295 | static inline struct rq *rq_of_rt_se(struct sched_rt_entity *rt_se) |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 296 | { |
| 297 | struct task_struct *p = rt_task_of(rt_se); |
Kirill Tkhai | 653d07a | 2014-03-15 02:14:55 +0400 | [diff] [blame] | 298 | |
| 299 | return task_rq(p); |
| 300 | } |
| 301 | |
| 302 | static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se) |
| 303 | { |
| 304 | struct rq *rq = rq_of_rt_se(rt_se); |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 305 | |
| 306 | return &rq->rt; |
| 307 | } |
| 308 | |
Mathias Krause | b027789 | 2021-11-03 20:06:13 +0100 | [diff] [blame] | 309 | void unregister_rt_sched_group(struct task_group *tg) { } |
| 310 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 311 | void free_rt_sched_group(struct task_group *tg) { } |
| 312 | |
| 313 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) |
| 314 | { |
| 315 | return 1; |
| 316 | } |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 317 | #endif /* CONFIG_RT_GROUP_SCHED */ |
| 318 | |
Steven Rostedt | 4fd2917 | 2008-01-25 21:08:06 +0100 | [diff] [blame] | 319 | #ifdef CONFIG_SMP |
Ingo Molnar | 84de427 | 2008-01-25 21:08:15 +0100 | [diff] [blame] | 320 | |
Peter Zijlstra | dc87734 | 2014-02-12 15:47:29 +0100 | [diff] [blame] | 321 | static inline bool need_pull_rt_task(struct rq *rq, struct task_struct *prev) |
| 322 | { |
| 323 | /* Try to pull RT tasks here if we lower this rq's prio */ |
Peter Zijlstra | 120455c5 | 2020-09-25 16:42:31 +0200 | [diff] [blame] | 324 | return rq->online && rq->rt.highest_prio.curr > prev->prio; |
Peter Zijlstra | dc87734 | 2014-02-12 15:47:29 +0100 | [diff] [blame] | 325 | } |
| 326 | |
Gregory Haskins | 637f508 | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 327 | static inline int rt_overloaded(struct rq *rq) |
Steven Rostedt | 4fd2917 | 2008-01-25 21:08:06 +0100 | [diff] [blame] | 328 | { |
Gregory Haskins | 637f508 | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 329 | return atomic_read(&rq->rd->rto_count); |
Steven Rostedt | 4fd2917 | 2008-01-25 21:08:06 +0100 | [diff] [blame] | 330 | } |
Ingo Molnar | 84de427 | 2008-01-25 21:08:15 +0100 | [diff] [blame] | 331 | |
Steven Rostedt | 4fd2917 | 2008-01-25 21:08:06 +0100 | [diff] [blame] | 332 | static inline void rt_set_overload(struct rq *rq) |
| 333 | { |
Gregory Haskins | 1f11eb6a | 2008-06-04 15:04:05 -0400 | [diff] [blame] | 334 | if (!rq->online) |
| 335 | return; |
| 336 | |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 337 | cpumask_set_cpu(rq->cpu, rq->rd->rto_mask); |
Steven Rostedt | 4fd2917 | 2008-01-25 21:08:06 +0100 | [diff] [blame] | 338 | /* |
| 339 | * Make sure the mask is visible before we set |
| 340 | * the overload count. That is checked to determine |
| 341 | * if we should look at the mask. It would be a shame |
| 342 | * if we looked at the mask, but the mask was not |
| 343 | * updated yet. |
Peter Zijlstra | 7c3f2ab | 2013-10-15 12:35:07 +0200 | [diff] [blame] | 344 | * |
| 345 | * Matched by the barrier in pull_rt_task(). |
Steven Rostedt | 4fd2917 | 2008-01-25 21:08:06 +0100 | [diff] [blame] | 346 | */ |
Peter Zijlstra | 7c3f2ab | 2013-10-15 12:35:07 +0200 | [diff] [blame] | 347 | smp_wmb(); |
Gregory Haskins | 637f508 | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 348 | atomic_inc(&rq->rd->rto_count); |
Steven Rostedt | 4fd2917 | 2008-01-25 21:08:06 +0100 | [diff] [blame] | 349 | } |
Ingo Molnar | 84de427 | 2008-01-25 21:08:15 +0100 | [diff] [blame] | 350 | |
Steven Rostedt | 4fd2917 | 2008-01-25 21:08:06 +0100 | [diff] [blame] | 351 | static inline void rt_clear_overload(struct rq *rq) |
| 352 | { |
Gregory Haskins | 1f11eb6a | 2008-06-04 15:04:05 -0400 | [diff] [blame] | 353 | if (!rq->online) |
| 354 | return; |
| 355 | |
Steven Rostedt | 4fd2917 | 2008-01-25 21:08:06 +0100 | [diff] [blame] | 356 | /* the order here really doesn't matter */ |
Gregory Haskins | 637f508 | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 357 | atomic_dec(&rq->rd->rto_count); |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 358 | cpumask_clear_cpu(rq->cpu, rq->rd->rto_mask); |
Steven Rostedt | 4fd2917 | 2008-01-25 21:08:06 +0100 | [diff] [blame] | 359 | } |
Gregory Haskins | 73fe6aae | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 360 | |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 361 | static void update_rt_migration(struct rt_rq *rt_rq) |
Gregory Haskins | 73fe6aae | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 362 | { |
Peter Zijlstra | a1ba4d8 | 2009-04-01 18:40:15 +0200 | [diff] [blame] | 363 | if (rt_rq->rt_nr_migratory && rt_rq->rt_nr_total > 1) { |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 364 | if (!rt_rq->overloaded) { |
| 365 | rt_set_overload(rq_of_rt_rq(rt_rq)); |
| 366 | rt_rq->overloaded = 1; |
Gregory Haskins | cdc8eb9 | 2008-01-25 21:08:23 +0100 | [diff] [blame] | 367 | } |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 368 | } else if (rt_rq->overloaded) { |
| 369 | rt_clear_overload(rq_of_rt_rq(rt_rq)); |
| 370 | rt_rq->overloaded = 0; |
Gregory Haskins | 637f508 | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 371 | } |
Gregory Haskins | 73fe6aae | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 372 | } |
Steven Rostedt | 4fd2917 | 2008-01-25 21:08:06 +0100 | [diff] [blame] | 373 | |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 374 | static void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
Peter Zijlstra | fa85ae2 | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 375 | { |
Peter Zijlstra | 29baa74 | 2012-04-23 12:11:21 +0200 | [diff] [blame] | 376 | struct task_struct *p; |
| 377 | |
Peter Zijlstra | a1ba4d8 | 2009-04-01 18:40:15 +0200 | [diff] [blame] | 378 | if (!rt_entity_is_task(rt_se)) |
| 379 | return; |
| 380 | |
Peter Zijlstra | 29baa74 | 2012-04-23 12:11:21 +0200 | [diff] [blame] | 381 | p = rt_task_of(rt_se); |
Peter Zijlstra | a1ba4d8 | 2009-04-01 18:40:15 +0200 | [diff] [blame] | 382 | rt_rq = &rq_of_rt_rq(rt_rq)->rt; |
| 383 | |
| 384 | rt_rq->rt_nr_total++; |
Ingo Molnar | 4b53a34 | 2017-02-05 15:41:03 +0100 | [diff] [blame] | 385 | if (p->nr_cpus_allowed > 1) |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 386 | rt_rq->rt_nr_migratory++; |
| 387 | |
| 388 | update_rt_migration(rt_rq); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 389 | } |
| 390 | |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 391 | static void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
| 392 | { |
Peter Zijlstra | 29baa74 | 2012-04-23 12:11:21 +0200 | [diff] [blame] | 393 | struct task_struct *p; |
| 394 | |
Peter Zijlstra | a1ba4d8 | 2009-04-01 18:40:15 +0200 | [diff] [blame] | 395 | if (!rt_entity_is_task(rt_se)) |
| 396 | return; |
| 397 | |
Peter Zijlstra | 29baa74 | 2012-04-23 12:11:21 +0200 | [diff] [blame] | 398 | p = rt_task_of(rt_se); |
Peter Zijlstra | a1ba4d8 | 2009-04-01 18:40:15 +0200 | [diff] [blame] | 399 | rt_rq = &rq_of_rt_rq(rt_rq)->rt; |
| 400 | |
| 401 | rt_rq->rt_nr_total--; |
Ingo Molnar | 4b53a34 | 2017-02-05 15:41:03 +0100 | [diff] [blame] | 402 | if (p->nr_cpus_allowed > 1) |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 403 | rt_rq->rt_nr_migratory--; |
| 404 | |
| 405 | update_rt_migration(rt_rq); |
| 406 | } |
| 407 | |
Steven Rostedt | 5181f4a4 | 2011-06-16 21:55:23 -0400 | [diff] [blame] | 408 | static inline int has_pushable_tasks(struct rq *rq) |
| 409 | { |
| 410 | return !plist_head_empty(&rq->rt.pushable_tasks); |
| 411 | } |
| 412 | |
Kees Cook | 8e5bad7 | 2022-10-07 17:07:58 -0700 | [diff] [blame] | 413 | static DEFINE_PER_CPU(struct balance_callback, rt_push_head); |
| 414 | static DEFINE_PER_CPU(struct balance_callback, rt_pull_head); |
Peter Zijlstra | e3fca9e | 2015-06-11 14:46:37 +0200 | [diff] [blame] | 415 | |
| 416 | static void push_rt_tasks(struct rq *); |
Peter Zijlstra | fd7a4be | 2015-06-11 14:46:41 +0200 | [diff] [blame] | 417 | static void pull_rt_task(struct rq *); |
Peter Zijlstra | e3fca9e | 2015-06-11 14:46:37 +0200 | [diff] [blame] | 418 | |
Ingo Molnar | 02d8ec9 | 2018-03-03 16:27:54 +0100 | [diff] [blame] | 419 | static inline void rt_queue_push_tasks(struct rq *rq) |
Peter Zijlstra | dc87734 | 2014-02-12 15:47:29 +0100 | [diff] [blame] | 420 | { |
Peter Zijlstra | e3fca9e | 2015-06-11 14:46:37 +0200 | [diff] [blame] | 421 | if (!has_pushable_tasks(rq)) |
| 422 | return; |
| 423 | |
Peter Zijlstra | fd7a4be | 2015-06-11 14:46:41 +0200 | [diff] [blame] | 424 | queue_balance_callback(rq, &per_cpu(rt_push_head, rq->cpu), push_rt_tasks); |
| 425 | } |
| 426 | |
Ingo Molnar | 02d8ec9 | 2018-03-03 16:27:54 +0100 | [diff] [blame] | 427 | static inline void rt_queue_pull_task(struct rq *rq) |
Peter Zijlstra | fd7a4be | 2015-06-11 14:46:41 +0200 | [diff] [blame] | 428 | { |
| 429 | queue_balance_callback(rq, &per_cpu(rt_pull_head, rq->cpu), pull_rt_task); |
Peter Zijlstra | dc87734 | 2014-02-12 15:47:29 +0100 | [diff] [blame] | 430 | } |
| 431 | |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 432 | static void enqueue_pushable_task(struct rq *rq, struct task_struct *p) |
| 433 | { |
| 434 | plist_del(&p->pushable_tasks, &rq->rt.pushable_tasks); |
| 435 | plist_node_init(&p->pushable_tasks, p->prio); |
| 436 | plist_add(&p->pushable_tasks, &rq->rt.pushable_tasks); |
Steven Rostedt | 5181f4a4 | 2011-06-16 21:55:23 -0400 | [diff] [blame] | 437 | |
| 438 | /* Update the highest prio pushable task */ |
| 439 | if (p->prio < rq->rt.highest_prio.next) |
| 440 | rq->rt.highest_prio.next = p->prio; |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 441 | } |
| 442 | |
| 443 | static void dequeue_pushable_task(struct rq *rq, struct task_struct *p) |
| 444 | { |
| 445 | plist_del(&p->pushable_tasks, &rq->rt.pushable_tasks); |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 446 | |
Steven Rostedt | 5181f4a4 | 2011-06-16 21:55:23 -0400 | [diff] [blame] | 447 | /* Update the new highest prio pushable task */ |
| 448 | if (has_pushable_tasks(rq)) { |
| 449 | p = plist_first_entry(&rq->rt.pushable_tasks, |
| 450 | struct task_struct, pushable_tasks); |
| 451 | rq->rt.highest_prio.next = p->prio; |
Peter Zijlstra | 934fc33 | 2020-10-14 21:06:49 +0200 | [diff] [blame] | 452 | } else { |
| 453 | rq->rt.highest_prio.next = MAX_RT_PRIO-1; |
| 454 | } |
Ingo Molnar | bcf08df | 2008-04-19 12:11:10 +0200 | [diff] [blame] | 455 | } |
| 456 | |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 457 | #else |
| 458 | |
Peter Zijlstra | ceacc2c | 2009-01-16 14:46:40 +0100 | [diff] [blame] | 459 | static inline void enqueue_pushable_task(struct rq *rq, struct task_struct *p) |
| 460 | { |
| 461 | } |
| 462 | |
| 463 | static inline void dequeue_pushable_task(struct rq *rq, struct task_struct *p) |
| 464 | { |
| 465 | } |
| 466 | |
Gregory Haskins | b07430a | 2009-01-14 08:55:39 -0500 | [diff] [blame] | 467 | static inline |
Peter Zijlstra | ceacc2c | 2009-01-16 14:46:40 +0100 | [diff] [blame] | 468 | void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
| 469 | { |
| 470 | } |
| 471 | |
Gregory Haskins | b07430a | 2009-01-14 08:55:39 -0500 | [diff] [blame] | 472 | static inline |
Peter Zijlstra | ceacc2c | 2009-01-16 14:46:40 +0100 | [diff] [blame] | 473 | void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
| 474 | { |
| 475 | } |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 476 | |
Ingo Molnar | 02d8ec9 | 2018-03-03 16:27:54 +0100 | [diff] [blame] | 477 | static inline void rt_queue_push_tasks(struct rq *rq) |
Peter Zijlstra | dc87734 | 2014-02-12 15:47:29 +0100 | [diff] [blame] | 478 | { |
| 479 | } |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 480 | #endif /* CONFIG_SMP */ |
| 481 | |
Kirill Tkhai | f4ebcbc | 2014-03-15 02:15:00 +0400 | [diff] [blame] | 482 | static void enqueue_top_rt_rq(struct rt_rq *rt_rq); |
Nicolas Saenz Julienne | 5c66d1b | 2022-06-28 11:22:59 +0200 | [diff] [blame] | 483 | static void dequeue_top_rt_rq(struct rt_rq *rt_rq, unsigned int count); |
Kirill Tkhai | f4ebcbc | 2014-03-15 02:15:00 +0400 | [diff] [blame] | 484 | |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 485 | static inline int on_rt_rq(struct sched_rt_entity *rt_se) |
| 486 | { |
Peter Zijlstra | ff77e46 | 2016-01-18 15:27:07 +0100 | [diff] [blame] | 487 | return rt_se->on_rq; |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 488 | } |
| 489 | |
Qais Yousef | 804d402 | 2019-10-09 11:46:11 +0100 | [diff] [blame] | 490 | #ifdef CONFIG_UCLAMP_TASK |
| 491 | /* |
| 492 | * Verify the fitness of task @p to run on @cpu taking into account the uclamp |
| 493 | * settings. |
| 494 | * |
| 495 | * This check is only important for heterogeneous systems where uclamp_min value |
| 496 | * is higher than the capacity of a @cpu. For non-heterogeneous system this |
| 497 | * function will always return true. |
| 498 | * |
| 499 | * The function will return true if the capacity of the @cpu is >= the |
| 500 | * uclamp_min and false otherwise. |
| 501 | * |
| 502 | * Note that uclamp_min will be clamped to uclamp_max if uclamp_min |
| 503 | * > uclamp_max. |
| 504 | */ |
| 505 | static inline bool rt_task_fits_capacity(struct task_struct *p, int cpu) |
| 506 | { |
| 507 | unsigned int min_cap; |
| 508 | unsigned int max_cap; |
| 509 | unsigned int cpu_cap; |
| 510 | |
| 511 | /* Only heterogeneous systems can benefit from this check */ |
Dietmar Eggemann | 740cf8a | 2022-07-29 13:13:03 +0200 | [diff] [blame] | 512 | if (!sched_asym_cpucap_active()) |
Qais Yousef | 804d402 | 2019-10-09 11:46:11 +0100 | [diff] [blame] | 513 | return true; |
| 514 | |
| 515 | min_cap = uclamp_eff_value(p, UCLAMP_MIN); |
| 516 | max_cap = uclamp_eff_value(p, UCLAMP_MAX); |
| 517 | |
| 518 | cpu_cap = capacity_orig_of(cpu); |
| 519 | |
| 520 | return cpu_cap >= min(min_cap, max_cap); |
| 521 | } |
| 522 | #else |
| 523 | static inline bool rt_task_fits_capacity(struct task_struct *p, int cpu) |
| 524 | { |
| 525 | return true; |
| 526 | } |
| 527 | #endif |
| 528 | |
Peter Zijlstra | 052f1dc | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 529 | #ifdef CONFIG_RT_GROUP_SCHED |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 530 | |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 531 | static inline u64 sched_rt_runtime(struct rt_rq *rt_rq) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 532 | { |
| 533 | if (!rt_rq->tg) |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 534 | return RUNTIME_INF; |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 535 | |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 536 | return rt_rq->rt_runtime; |
| 537 | } |
| 538 | |
| 539 | static inline u64 sched_rt_period(struct rt_rq *rt_rq) |
| 540 | { |
| 541 | return ktime_to_ns(rt_rq->tg->rt_bandwidth.rt_period); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 542 | } |
| 543 | |
Cheng Xu | ec514c4 | 2011-05-14 14:20:02 +0800 | [diff] [blame] | 544 | typedef struct task_group *rt_rq_iter_t; |
| 545 | |
Yong Zhang | 1c09ab0 | 2011-06-28 10:51:31 +0800 | [diff] [blame] | 546 | static inline struct task_group *next_task_group(struct task_group *tg) |
| 547 | { |
| 548 | do { |
| 549 | tg = list_entry_rcu(tg->list.next, |
| 550 | typeof(struct task_group), list); |
| 551 | } while (&tg->list != &task_groups && task_group_is_autogroup(tg)); |
| 552 | |
| 553 | if (&tg->list == &task_groups) |
| 554 | tg = NULL; |
| 555 | |
| 556 | return tg; |
| 557 | } |
| 558 | |
| 559 | #define for_each_rt_rq(rt_rq, iter, rq) \ |
| 560 | for (iter = container_of(&task_groups, typeof(*iter), list); \ |
| 561 | (iter = next_task_group(iter)) && \ |
| 562 | (rt_rq = iter->rt_rq[cpu_of(rq)]);) |
Cheng Xu | ec514c4 | 2011-05-14 14:20:02 +0800 | [diff] [blame] | 563 | |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 564 | #define for_each_sched_rt_entity(rt_se) \ |
| 565 | for (; rt_se; rt_se = rt_se->parent) |
| 566 | |
| 567 | static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se) |
| 568 | { |
| 569 | return rt_se->my_q; |
| 570 | } |
| 571 | |
Peter Zijlstra | ff77e46 | 2016-01-18 15:27:07 +0100 | [diff] [blame] | 572 | static void enqueue_rt_entity(struct sched_rt_entity *rt_se, unsigned int flags); |
| 573 | static void dequeue_rt_entity(struct sched_rt_entity *rt_se, unsigned int flags); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 574 | |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 575 | static void sched_rt_rq_enqueue(struct rt_rq *rt_rq) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 576 | { |
Dario Faggioli | f6121f4 | 2008-10-03 17:40:46 +0200 | [diff] [blame] | 577 | struct task_struct *curr = rq_of_rt_rq(rt_rq)->curr; |
Kirill Tkhai | 8875125 | 2014-06-29 00:03:57 +0400 | [diff] [blame] | 578 | struct rq *rq = rq_of_rt_rq(rt_rq); |
Yong Zhang | 74b7eb5 | 2010-01-29 14:57:52 +0800 | [diff] [blame] | 579 | struct sched_rt_entity *rt_se; |
| 580 | |
Kirill Tkhai | 8875125 | 2014-06-29 00:03:57 +0400 | [diff] [blame] | 581 | int cpu = cpu_of(rq); |
Balbir Singh | 0c3b916 | 2011-03-03 17:04:35 +0530 | [diff] [blame] | 582 | |
| 583 | rt_se = rt_rq->tg->rt_se[cpu]; |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 584 | |
Dario Faggioli | f6121f4 | 2008-10-03 17:40:46 +0200 | [diff] [blame] | 585 | if (rt_rq->rt_nr_running) { |
Kirill Tkhai | f4ebcbc | 2014-03-15 02:15:00 +0400 | [diff] [blame] | 586 | if (!rt_se) |
| 587 | enqueue_top_rt_rq(rt_rq); |
| 588 | else if (!on_rt_rq(rt_se)) |
Peter Zijlstra | ff77e46 | 2016-01-18 15:27:07 +0100 | [diff] [blame] | 589 | enqueue_rt_entity(rt_se, 0); |
Kirill Tkhai | f4ebcbc | 2014-03-15 02:15:00 +0400 | [diff] [blame] | 590 | |
Gregory Haskins | e864c49 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 591 | if (rt_rq->highest_prio.curr < curr->prio) |
Kirill Tkhai | 8875125 | 2014-06-29 00:03:57 +0400 | [diff] [blame] | 592 | resched_curr(rq); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 593 | } |
| 594 | } |
| 595 | |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 596 | static void sched_rt_rq_dequeue(struct rt_rq *rt_rq) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 597 | { |
Yong Zhang | 74b7eb5 | 2010-01-29 14:57:52 +0800 | [diff] [blame] | 598 | struct sched_rt_entity *rt_se; |
Balbir Singh | 0c3b916 | 2011-03-03 17:04:35 +0530 | [diff] [blame] | 599 | int cpu = cpu_of(rq_of_rt_rq(rt_rq)); |
Yong Zhang | 74b7eb5 | 2010-01-29 14:57:52 +0800 | [diff] [blame] | 600 | |
Balbir Singh | 0c3b916 | 2011-03-03 17:04:35 +0530 | [diff] [blame] | 601 | rt_se = rt_rq->tg->rt_se[cpu]; |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 602 | |
Vincent Guittot | 296b2ff | 2018-06-26 15:53:22 +0200 | [diff] [blame] | 603 | if (!rt_se) { |
Nicolas Saenz Julienne | 5c66d1b | 2022-06-28 11:22:59 +0200 | [diff] [blame] | 604 | dequeue_top_rt_rq(rt_rq, rt_rq->rt_nr_running); |
Vincent Guittot | 296b2ff | 2018-06-26 15:53:22 +0200 | [diff] [blame] | 605 | /* Kick cpufreq (see the comment in kernel/sched/sched.h). */ |
| 606 | cpufreq_update_util(rq_of_rt_rq(rt_rq), 0); |
| 607 | } |
Kirill Tkhai | f4ebcbc | 2014-03-15 02:15:00 +0400 | [diff] [blame] | 608 | else if (on_rt_rq(rt_se)) |
Peter Zijlstra | ff77e46 | 2016-01-18 15:27:07 +0100 | [diff] [blame] | 609 | dequeue_rt_entity(rt_se, 0); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 610 | } |
| 611 | |
Kirill Tkhai | 4638364 | 2014-03-15 02:15:07 +0400 | [diff] [blame] | 612 | static inline int rt_rq_throttled(struct rt_rq *rt_rq) |
| 613 | { |
| 614 | return rt_rq->rt_throttled && !rt_rq->rt_nr_boosted; |
| 615 | } |
| 616 | |
Peter Zijlstra | 23b0fdf | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 617 | static int rt_se_boosted(struct sched_rt_entity *rt_se) |
| 618 | { |
| 619 | struct rt_rq *rt_rq = group_rt_rq(rt_se); |
| 620 | struct task_struct *p; |
| 621 | |
| 622 | if (rt_rq) |
| 623 | return !!rt_rq->rt_nr_boosted; |
| 624 | |
| 625 | p = rt_task_of(rt_se); |
| 626 | return p->prio != p->normal_prio; |
| 627 | } |
| 628 | |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 629 | #ifdef CONFIG_SMP |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 630 | static inline const struct cpumask *sched_rt_period_mask(void) |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 631 | { |
Nathan Zimmer | 424c93f | 2013-05-09 11:24:03 -0500 | [diff] [blame] | 632 | return this_rq()->rd->span; |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 633 | } |
| 634 | #else |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 635 | static inline const struct cpumask *sched_rt_period_mask(void) |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 636 | { |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 637 | return cpu_online_mask; |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 638 | } |
| 639 | #endif |
| 640 | |
| 641 | static inline |
| 642 | struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu) |
| 643 | { |
| 644 | return container_of(rt_b, struct task_group, rt_bandwidth)->rt_rq[cpu]; |
| 645 | } |
| 646 | |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 647 | static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq) |
| 648 | { |
| 649 | return &rt_rq->tg->rt_bandwidth; |
| 650 | } |
| 651 | |
Dhaval Giani | 55e12e5 | 2008-06-24 23:39:43 +0530 | [diff] [blame] | 652 | #else /* !CONFIG_RT_GROUP_SCHED */ |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 653 | |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 654 | static inline u64 sched_rt_runtime(struct rt_rq *rt_rq) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 655 | { |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 656 | return rt_rq->rt_runtime; |
| 657 | } |
| 658 | |
| 659 | static inline u64 sched_rt_period(struct rt_rq *rt_rq) |
| 660 | { |
| 661 | return ktime_to_ns(def_rt_bandwidth.rt_period); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 662 | } |
| 663 | |
Cheng Xu | ec514c4 | 2011-05-14 14:20:02 +0800 | [diff] [blame] | 664 | typedef struct rt_rq *rt_rq_iter_t; |
| 665 | |
| 666 | #define for_each_rt_rq(rt_rq, iter, rq) \ |
| 667 | for ((void) iter, rt_rq = &rq->rt; rt_rq; rt_rq = NULL) |
| 668 | |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 669 | #define for_each_sched_rt_entity(rt_se) \ |
| 670 | for (; rt_se; rt_se = NULL) |
| 671 | |
| 672 | static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se) |
| 673 | { |
| 674 | return NULL; |
| 675 | } |
| 676 | |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 677 | static inline void sched_rt_rq_enqueue(struct rt_rq *rt_rq) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 678 | { |
Kirill Tkhai | f4ebcbc | 2014-03-15 02:15:00 +0400 | [diff] [blame] | 679 | struct rq *rq = rq_of_rt_rq(rt_rq); |
| 680 | |
| 681 | if (!rt_rq->rt_nr_running) |
| 682 | return; |
| 683 | |
| 684 | enqueue_top_rt_rq(rt_rq); |
Kirill Tkhai | 8875125 | 2014-06-29 00:03:57 +0400 | [diff] [blame] | 685 | resched_curr(rq); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 686 | } |
| 687 | |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 688 | static inline void sched_rt_rq_dequeue(struct rt_rq *rt_rq) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 689 | { |
Nicolas Saenz Julienne | 5c66d1b | 2022-06-28 11:22:59 +0200 | [diff] [blame] | 690 | dequeue_top_rt_rq(rt_rq, rt_rq->rt_nr_running); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 691 | } |
| 692 | |
Kirill Tkhai | 4638364 | 2014-03-15 02:15:07 +0400 | [diff] [blame] | 693 | static inline int rt_rq_throttled(struct rt_rq *rt_rq) |
| 694 | { |
| 695 | return rt_rq->rt_throttled; |
| 696 | } |
| 697 | |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 698 | static inline const struct cpumask *sched_rt_period_mask(void) |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 699 | { |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 700 | return cpu_online_mask; |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 701 | } |
| 702 | |
| 703 | static inline |
| 704 | struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu) |
| 705 | { |
| 706 | return &cpu_rq(cpu)->rt; |
| 707 | } |
| 708 | |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 709 | static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq) |
| 710 | { |
| 711 | return &def_rt_bandwidth; |
| 712 | } |
| 713 | |
Dhaval Giani | 55e12e5 | 2008-06-24 23:39:43 +0530 | [diff] [blame] | 714 | #endif /* CONFIG_RT_GROUP_SCHED */ |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 715 | |
Juri Lelli | faa5993 | 2014-02-21 11:37:15 +0100 | [diff] [blame] | 716 | bool sched_rt_bandwidth_account(struct rt_rq *rt_rq) |
| 717 | { |
| 718 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); |
| 719 | |
| 720 | return (hrtimer_active(&rt_b->rt_period_timer) || |
| 721 | rt_rq->rt_time < rt_b->rt_runtime); |
| 722 | } |
| 723 | |
Peter Zijlstra | b79f383 | 2008-06-19 14:22:25 +0200 | [diff] [blame] | 724 | #ifdef CONFIG_SMP |
Peter Zijlstra | 78333cd | 2008-09-23 15:33:43 +0200 | [diff] [blame] | 725 | /* |
| 726 | * We ran out of runtime, see if we can borrow some from our neighbours. |
| 727 | */ |
Juri Lelli | 269b26a | 2015-09-02 11:01:36 +0100 | [diff] [blame] | 728 | static void do_balance_runtime(struct rt_rq *rt_rq) |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 729 | { |
| 730 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); |
Shawn Bohrer | aa7f673 | 2013-01-14 11:55:31 -0600 | [diff] [blame] | 731 | struct root_domain *rd = rq_of_rt_rq(rt_rq)->rd; |
Juri Lelli | 269b26a | 2015-09-02 11:01:36 +0100 | [diff] [blame] | 732 | int i, weight; |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 733 | u64 rt_period; |
| 734 | |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 735 | weight = cpumask_weight(rd->span); |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 736 | |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 737 | raw_spin_lock(&rt_b->rt_runtime_lock); |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 738 | rt_period = ktime_to_ns(rt_b->rt_period); |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 739 | for_each_cpu(i, rd->span) { |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 740 | struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i); |
| 741 | s64 diff; |
| 742 | |
| 743 | if (iter == rt_rq) |
| 744 | continue; |
| 745 | |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 746 | raw_spin_lock(&iter->rt_runtime_lock); |
Peter Zijlstra | 78333cd | 2008-09-23 15:33:43 +0200 | [diff] [blame] | 747 | /* |
| 748 | * Either all rqs have inf runtime and there's nothing to steal |
| 749 | * or __disable_runtime() below sets a specific rq to inf to |
Ingo Molnar | 3b03706 | 2021-03-18 13:38:50 +0100 | [diff] [blame] | 750 | * indicate its been disabled and disallow stealing. |
Peter Zijlstra | 78333cd | 2008-09-23 15:33:43 +0200 | [diff] [blame] | 751 | */ |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 752 | if (iter->rt_runtime == RUNTIME_INF) |
| 753 | goto next; |
| 754 | |
Peter Zijlstra | 78333cd | 2008-09-23 15:33:43 +0200 | [diff] [blame] | 755 | /* |
| 756 | * From runqueues with spare time, take 1/n part of their |
| 757 | * spare time, but no more than our period. |
| 758 | */ |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 759 | diff = iter->rt_runtime - iter->rt_time; |
| 760 | if (diff > 0) { |
Peter Zijlstra | 58838cf | 2008-07-24 12:43:13 +0200 | [diff] [blame] | 761 | diff = div_u64((u64)diff, weight); |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 762 | if (rt_rq->rt_runtime + diff > rt_period) |
| 763 | diff = rt_period - rt_rq->rt_runtime; |
| 764 | iter->rt_runtime -= diff; |
| 765 | rt_rq->rt_runtime += diff; |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 766 | if (rt_rq->rt_runtime == rt_period) { |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 767 | raw_spin_unlock(&iter->rt_runtime_lock); |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 768 | break; |
| 769 | } |
| 770 | } |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 771 | next: |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 772 | raw_spin_unlock(&iter->rt_runtime_lock); |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 773 | } |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 774 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 775 | } |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 776 | |
Peter Zijlstra | 78333cd | 2008-09-23 15:33:43 +0200 | [diff] [blame] | 777 | /* |
| 778 | * Ensure this RQ takes back all the runtime it lend to its neighbours. |
| 779 | */ |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 780 | static void __disable_runtime(struct rq *rq) |
| 781 | { |
| 782 | struct root_domain *rd = rq->rd; |
Cheng Xu | ec514c4 | 2011-05-14 14:20:02 +0800 | [diff] [blame] | 783 | rt_rq_iter_t iter; |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 784 | struct rt_rq *rt_rq; |
| 785 | |
| 786 | if (unlikely(!scheduler_running)) |
| 787 | return; |
| 788 | |
Cheng Xu | ec514c4 | 2011-05-14 14:20:02 +0800 | [diff] [blame] | 789 | for_each_rt_rq(rt_rq, iter, rq) { |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 790 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); |
| 791 | s64 want; |
| 792 | int i; |
| 793 | |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 794 | raw_spin_lock(&rt_b->rt_runtime_lock); |
| 795 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
Peter Zijlstra | 78333cd | 2008-09-23 15:33:43 +0200 | [diff] [blame] | 796 | /* |
| 797 | * Either we're all inf and nobody needs to borrow, or we're |
| 798 | * already disabled and thus have nothing to do, or we have |
| 799 | * exactly the right amount of runtime to take out. |
| 800 | */ |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 801 | if (rt_rq->rt_runtime == RUNTIME_INF || |
| 802 | rt_rq->rt_runtime == rt_b->rt_runtime) |
| 803 | goto balanced; |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 804 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 805 | |
Peter Zijlstra | 78333cd | 2008-09-23 15:33:43 +0200 | [diff] [blame] | 806 | /* |
| 807 | * Calculate the difference between what we started out with |
| 808 | * and what we current have, that's the amount of runtime |
| 809 | * we lend and now have to reclaim. |
| 810 | */ |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 811 | want = rt_b->rt_runtime - rt_rq->rt_runtime; |
| 812 | |
Peter Zijlstra | 78333cd | 2008-09-23 15:33:43 +0200 | [diff] [blame] | 813 | /* |
| 814 | * Greedy reclaim, take back as much as we can. |
| 815 | */ |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 816 | for_each_cpu(i, rd->span) { |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 817 | struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i); |
| 818 | s64 diff; |
| 819 | |
Peter Zijlstra | 78333cd | 2008-09-23 15:33:43 +0200 | [diff] [blame] | 820 | /* |
| 821 | * Can't reclaim from ourselves or disabled runqueues. |
| 822 | */ |
Peter Zijlstra | f1679d0 | 2008-08-14 15:49:00 +0200 | [diff] [blame] | 823 | if (iter == rt_rq || iter->rt_runtime == RUNTIME_INF) |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 824 | continue; |
| 825 | |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 826 | raw_spin_lock(&iter->rt_runtime_lock); |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 827 | if (want > 0) { |
| 828 | diff = min_t(s64, iter->rt_runtime, want); |
| 829 | iter->rt_runtime -= diff; |
| 830 | want -= diff; |
| 831 | } else { |
| 832 | iter->rt_runtime -= want; |
| 833 | want -= want; |
| 834 | } |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 835 | raw_spin_unlock(&iter->rt_runtime_lock); |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 836 | |
| 837 | if (!want) |
| 838 | break; |
| 839 | } |
| 840 | |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 841 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
Peter Zijlstra | 78333cd | 2008-09-23 15:33:43 +0200 | [diff] [blame] | 842 | /* |
| 843 | * We cannot be left wanting - that would mean some runtime |
| 844 | * leaked out of the system. |
| 845 | */ |
Ingo Molnar | 09348d7 | 2022-08-11 08:54:52 +0200 | [diff] [blame] | 846 | WARN_ON_ONCE(want); |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 847 | balanced: |
Peter Zijlstra | 78333cd | 2008-09-23 15:33:43 +0200 | [diff] [blame] | 848 | /* |
| 849 | * Disable all the borrow logic by pretending we have inf |
| 850 | * runtime - in which case borrowing doesn't make sense. |
| 851 | */ |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 852 | rt_rq->rt_runtime = RUNTIME_INF; |
Peter Boonstoppel | a4c96ae | 2012-08-09 15:34:47 -0700 | [diff] [blame] | 853 | rt_rq->rt_throttled = 0; |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 854 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
| 855 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
Kirill Tkhai | 99b6256 | 2014-06-25 12:19:48 +0400 | [diff] [blame] | 856 | |
| 857 | /* Make rt_rq available for pick_next_task() */ |
| 858 | sched_rt_rq_enqueue(rt_rq); |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 859 | } |
| 860 | } |
| 861 | |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 862 | static void __enable_runtime(struct rq *rq) |
| 863 | { |
Cheng Xu | ec514c4 | 2011-05-14 14:20:02 +0800 | [diff] [blame] | 864 | rt_rq_iter_t iter; |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 865 | struct rt_rq *rt_rq; |
| 866 | |
| 867 | if (unlikely(!scheduler_running)) |
| 868 | return; |
| 869 | |
Peter Zijlstra | 78333cd | 2008-09-23 15:33:43 +0200 | [diff] [blame] | 870 | /* |
| 871 | * Reset each runqueue's bandwidth settings |
| 872 | */ |
Cheng Xu | ec514c4 | 2011-05-14 14:20:02 +0800 | [diff] [blame] | 873 | for_each_rt_rq(rt_rq, iter, rq) { |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 874 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); |
| 875 | |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 876 | raw_spin_lock(&rt_b->rt_runtime_lock); |
| 877 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 878 | rt_rq->rt_runtime = rt_b->rt_runtime; |
| 879 | rt_rq->rt_time = 0; |
Zhang, Yanmin | baf2573 | 2008-09-09 11:26:33 +0800 | [diff] [blame] | 880 | rt_rq->rt_throttled = 0; |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 881 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
| 882 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 883 | } |
| 884 | } |
| 885 | |
Juri Lelli | 269b26a | 2015-09-02 11:01:36 +0100 | [diff] [blame] | 886 | static void balance_runtime(struct rt_rq *rt_rq) |
Peter Zijlstra | eff6549 | 2008-06-19 14:22:26 +0200 | [diff] [blame] | 887 | { |
Peter Zijlstra | 4a6184c | 2011-10-06 22:39:14 +0200 | [diff] [blame] | 888 | if (!sched_feat(RT_RUNTIME_SHARE)) |
Juri Lelli | 269b26a | 2015-09-02 11:01:36 +0100 | [diff] [blame] | 889 | return; |
Peter Zijlstra | 4a6184c | 2011-10-06 22:39:14 +0200 | [diff] [blame] | 890 | |
Peter Zijlstra | eff6549 | 2008-06-19 14:22:26 +0200 | [diff] [blame] | 891 | if (rt_rq->rt_time > rt_rq->rt_runtime) { |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 892 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
Juri Lelli | 269b26a | 2015-09-02 11:01:36 +0100 | [diff] [blame] | 893 | do_balance_runtime(rt_rq); |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 894 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
Peter Zijlstra | eff6549 | 2008-06-19 14:22:26 +0200 | [diff] [blame] | 895 | } |
Peter Zijlstra | eff6549 | 2008-06-19 14:22:26 +0200 | [diff] [blame] | 896 | } |
Dhaval Giani | 55e12e5 | 2008-06-24 23:39:43 +0530 | [diff] [blame] | 897 | #else /* !CONFIG_SMP */ |
Juri Lelli | 269b26a | 2015-09-02 11:01:36 +0100 | [diff] [blame] | 898 | static inline void balance_runtime(struct rt_rq *rt_rq) {} |
Dhaval Giani | 55e12e5 | 2008-06-24 23:39:43 +0530 | [diff] [blame] | 899 | #endif /* CONFIG_SMP */ |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 900 | |
| 901 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun) |
| 902 | { |
Peter Zijlstra | 42c62a5 | 2011-10-18 22:03:48 +0200 | [diff] [blame] | 903 | int i, idle = 1, throttled = 0; |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 904 | const struct cpumask *span; |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 905 | |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 906 | span = sched_rt_period_mask(); |
Mike Galbraith | e221d02 | 2012-08-07 10:02:38 +0200 | [diff] [blame] | 907 | #ifdef CONFIG_RT_GROUP_SCHED |
| 908 | /* |
| 909 | * FIXME: isolated CPUs should really leave the root task group, |
| 910 | * whether they are isolcpus or were isolated via cpusets, lest |
| 911 | * the timer run on a CPU which does not service all runqueues, |
| 912 | * potentially leaving other CPUs indefinitely throttled. If |
| 913 | * isolation is really required, the user will turn the throttle |
| 914 | * off to kill the perturbations it causes anyway. Meanwhile, |
| 915 | * this maintains functionality for boot and/or troubleshooting. |
| 916 | */ |
| 917 | if (rt_b == &root_task_group.rt_bandwidth) |
| 918 | span = cpu_online_mask; |
| 919 | #endif |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 920 | for_each_cpu(i, span) { |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 921 | int enqueue = 0; |
| 922 | struct rt_rq *rt_rq = sched_rt_period_rt_rq(rt_b, i); |
| 923 | struct rq *rq = rq_of_rt_rq(rt_rq); |
Hao Jia | 2679a83 | 2022-04-30 16:58:42 +0800 | [diff] [blame] | 924 | struct rq_flags rf; |
Dave Kleikamp | c249f25 | 2017-05-15 14:14:13 -0500 | [diff] [blame] | 925 | int skip; |
| 926 | |
| 927 | /* |
| 928 | * When span == cpu_online_mask, taking each rq->lock |
| 929 | * can be time-consuming. Try to avoid it when possible. |
| 930 | */ |
| 931 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
Hailong Liu | f3d133e | 2018-07-18 08:46:55 +0800 | [diff] [blame] | 932 | if (!sched_feat(RT_RUNTIME_SHARE) && rt_rq->rt_runtime != RUNTIME_INF) |
| 933 | rt_rq->rt_runtime = rt_b->rt_runtime; |
Dave Kleikamp | c249f25 | 2017-05-15 14:14:13 -0500 | [diff] [blame] | 934 | skip = !rt_rq->rt_time && !rt_rq->rt_nr_running; |
| 935 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
| 936 | if (skip) |
| 937 | continue; |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 938 | |
Hao Jia | 2679a83 | 2022-04-30 16:58:42 +0800 | [diff] [blame] | 939 | rq_lock(rq, &rf); |
Davidlohr Bueso | d29a206 | 2018-04-02 09:49:54 -0700 | [diff] [blame] | 940 | update_rq_clock(rq); |
| 941 | |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 942 | if (rt_rq->rt_time) { |
| 943 | u64 runtime; |
| 944 | |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 945 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
Peter Zijlstra | eff6549 | 2008-06-19 14:22:26 +0200 | [diff] [blame] | 946 | if (rt_rq->rt_throttled) |
| 947 | balance_runtime(rt_rq); |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 948 | runtime = rt_rq->rt_runtime; |
| 949 | rt_rq->rt_time -= min(rt_rq->rt_time, overrun*runtime); |
| 950 | if (rt_rq->rt_throttled && rt_rq->rt_time < runtime) { |
| 951 | rt_rq->rt_throttled = 0; |
| 952 | enqueue = 1; |
Mike Galbraith | 61eadef | 2011-04-29 08:36:50 +0200 | [diff] [blame] | 953 | |
| 954 | /* |
Peter Zijlstra | 9edfbfe | 2015-01-05 11:18:11 +0100 | [diff] [blame] | 955 | * When we're idle and a woken (rt) task is |
| 956 | * throttled check_preempt_curr() will set |
| 957 | * skip_update and the time between the wakeup |
| 958 | * and this unthrottle will get accounted as |
| 959 | * 'runtime'. |
Mike Galbraith | 61eadef | 2011-04-29 08:36:50 +0200 | [diff] [blame] | 960 | */ |
| 961 | if (rt_rq->rt_nr_running && rq->curr == rq->idle) |
Davidlohr Bueso | adcc8da | 2018-04-04 09:15:39 -0700 | [diff] [blame] | 962 | rq_clock_cancel_skipupdate(rq); |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 963 | } |
| 964 | if (rt_rq->rt_time || rt_rq->rt_nr_running) |
| 965 | idle = 0; |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 966 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
Balbir Singh | 0c3b916 | 2011-03-03 17:04:35 +0530 | [diff] [blame] | 967 | } else if (rt_rq->rt_nr_running) { |
Peter Zijlstra | 8a8cde1 | 2008-06-19 14:22:28 +0200 | [diff] [blame] | 968 | idle = 0; |
Balbir Singh | 0c3b916 | 2011-03-03 17:04:35 +0530 | [diff] [blame] | 969 | if (!rt_rq_throttled(rt_rq)) |
| 970 | enqueue = 1; |
| 971 | } |
Peter Zijlstra | 42c62a5 | 2011-10-18 22:03:48 +0200 | [diff] [blame] | 972 | if (rt_rq->rt_throttled) |
| 973 | throttled = 1; |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 974 | |
| 975 | if (enqueue) |
| 976 | sched_rt_rq_enqueue(rt_rq); |
Hao Jia | 2679a83 | 2022-04-30 16:58:42 +0800 | [diff] [blame] | 977 | rq_unlock(rq, &rf); |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 978 | } |
| 979 | |
Peter Zijlstra | 42c62a5 | 2011-10-18 22:03:48 +0200 | [diff] [blame] | 980 | if (!throttled && (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF)) |
| 981 | return 1; |
| 982 | |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 983 | return idle; |
| 984 | } |
| 985 | |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 986 | static inline int rt_se_prio(struct sched_rt_entity *rt_se) |
| 987 | { |
Peter Zijlstra | 052f1dc | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 988 | #ifdef CONFIG_RT_GROUP_SCHED |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 989 | struct rt_rq *rt_rq = group_rt_rq(rt_se); |
| 990 | |
| 991 | if (rt_rq) |
Gregory Haskins | e864c49 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 992 | return rt_rq->highest_prio.curr; |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 993 | #endif |
| 994 | |
| 995 | return rt_task_of(rt_se)->prio; |
| 996 | } |
| 997 | |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 998 | static int sched_rt_runtime_exceeded(struct rt_rq *rt_rq) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 999 | { |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 1000 | u64 runtime = sched_rt_runtime(rt_rq); |
Peter Zijlstra | fa85ae2 | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 1001 | |
Peter Zijlstra | fa85ae2 | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 1002 | if (rt_rq->rt_throttled) |
Peter Zijlstra | 23b0fdf | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 1003 | return rt_rq_throttled(rt_rq); |
Peter Zijlstra | fa85ae2 | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 1004 | |
Shan Hai | 5b680fd | 2011-11-29 11:03:56 +0800 | [diff] [blame] | 1005 | if (runtime >= sched_rt_period(rt_rq)) |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 1006 | return 0; |
| 1007 | |
Peter Zijlstra | b79f383 | 2008-06-19 14:22:25 +0200 | [diff] [blame] | 1008 | balance_runtime(rt_rq); |
| 1009 | runtime = sched_rt_runtime(rt_rq); |
| 1010 | if (runtime == RUNTIME_INF) |
| 1011 | return 0; |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 1012 | |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 1013 | if (rt_rq->rt_time > runtime) { |
Peter Zijlstra | 7abc63b | 2011-10-18 22:03:48 +0200 | [diff] [blame] | 1014 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); |
| 1015 | |
| 1016 | /* |
| 1017 | * Don't actually throttle groups that have no runtime assigned |
| 1018 | * but accrue some time due to boosting. |
| 1019 | */ |
| 1020 | if (likely(rt_b->rt_runtime)) { |
| 1021 | rt_rq->rt_throttled = 1; |
John Stultz | c224815 | 2014-06-04 16:11:41 -0700 | [diff] [blame] | 1022 | printk_deferred_once("sched: RT throttling activated\n"); |
Peter Zijlstra | 7abc63b | 2011-10-18 22:03:48 +0200 | [diff] [blame] | 1023 | } else { |
| 1024 | /* |
| 1025 | * In case we did anyway, make it go away, |
| 1026 | * replenishment is a joke, since it will replenish us |
| 1027 | * with exactly 0 ns. |
| 1028 | */ |
| 1029 | rt_rq->rt_time = 0; |
| 1030 | } |
| 1031 | |
Peter Zijlstra | 23b0fdf | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 1032 | if (rt_rq_throttled(rt_rq)) { |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 1033 | sched_rt_rq_dequeue(rt_rq); |
Peter Zijlstra | 23b0fdf | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 1034 | return 1; |
| 1035 | } |
Peter Zijlstra | fa85ae2 | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 1036 | } |
| 1037 | |
| 1038 | return 0; |
| 1039 | } |
| 1040 | |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1041 | /* |
| 1042 | * Update the current task's runtime statistics. Skip current tasks that |
| 1043 | * are not in our scheduling class. |
| 1044 | */ |
Alexey Dobriyan | a995744 | 2007-10-15 17:00:13 +0200 | [diff] [blame] | 1045 | static void update_curr_rt(struct rq *rq) |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1046 | { |
| 1047 | struct task_struct *curr = rq->curr; |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1048 | struct sched_rt_entity *rt_se = &curr->rt; |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1049 | u64 delta_exec; |
Wen Yang | a771160 | 2018-02-06 09:53:28 +0800 | [diff] [blame] | 1050 | u64 now; |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1051 | |
Peter Zijlstra | 06c3bc6 | 2011-02-02 13:19:48 +0100 | [diff] [blame] | 1052 | if (curr->sched_class != &rt_sched_class) |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1053 | return; |
| 1054 | |
Wen Yang | a771160 | 2018-02-06 09:53:28 +0800 | [diff] [blame] | 1055 | now = rq_clock_task(rq); |
Wen Yang | e7ad203 | 2018-02-05 11:18:41 +0800 | [diff] [blame] | 1056 | delta_exec = now - curr->se.exec_start; |
Kirill Tkhai | fc79e24 | 2013-01-30 16:50:36 +0400 | [diff] [blame] | 1057 | if (unlikely((s64)delta_exec <= 0)) |
| 1058 | return; |
Ingo Molnar | 6cfb0d5 | 2007-08-02 17:41:40 +0200 | [diff] [blame] | 1059 | |
Yafang Shao | ceeadb8 | 2021-09-05 14:35:41 +0000 | [diff] [blame] | 1060 | schedstat_set(curr->stats.exec_max, |
| 1061 | max(curr->stats.exec_max, delta_exec)); |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1062 | |
Yafang Shao | ed7b564 | 2021-09-05 14:35:44 +0000 | [diff] [blame] | 1063 | trace_sched_stat_runtime(curr, delta_exec, 0); |
| 1064 | |
Shang XiaoJing | 5531ecf | 2022-08-24 16:28:56 +0800 | [diff] [blame] | 1065 | update_current_exec_runtime(curr, now, delta_exec); |
Peter Zijlstra | fa85ae2 | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 1066 | |
Peter Zijlstra | 0b148fa | 2008-08-19 12:33:04 +0200 | [diff] [blame] | 1067 | if (!rt_bandwidth_enabled()) |
| 1068 | return; |
| 1069 | |
Dhaval Giani | 354d60c | 2008-04-19 19:44:59 +0200 | [diff] [blame] | 1070 | for_each_sched_rt_entity(rt_se) { |
Giedrius Rekasius | 0b07939 | 2014-05-25 15:23:31 +0100 | [diff] [blame] | 1071 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); |
Li Hua | 9b58e97 | 2021-12-03 03:36:18 +0000 | [diff] [blame] | 1072 | int exceeded; |
Dhaval Giani | 354d60c | 2008-04-19 19:44:59 +0200 | [diff] [blame] | 1073 | |
Peter Zijlstra | cc2991c | 2008-08-19 12:33:03 +0200 | [diff] [blame] | 1074 | if (sched_rt_runtime(rt_rq) != RUNTIME_INF) { |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 1075 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
Peter Zijlstra | cc2991c | 2008-08-19 12:33:03 +0200 | [diff] [blame] | 1076 | rt_rq->rt_time += delta_exec; |
Li Hua | 9b58e97 | 2021-12-03 03:36:18 +0000 | [diff] [blame] | 1077 | exceeded = sched_rt_runtime_exceeded(rt_rq); |
| 1078 | if (exceeded) |
Kirill Tkhai | 8875125 | 2014-06-29 00:03:57 +0400 | [diff] [blame] | 1079 | resched_curr(rq); |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 1080 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
Li Hua | 9b58e97 | 2021-12-03 03:36:18 +0000 | [diff] [blame] | 1081 | if (exceeded) |
| 1082 | do_start_rt_bandwidth(sched_rt_bandwidth(rt_rq)); |
Peter Zijlstra | cc2991c | 2008-08-19 12:33:03 +0200 | [diff] [blame] | 1083 | } |
Dhaval Giani | 354d60c | 2008-04-19 19:44:59 +0200 | [diff] [blame] | 1084 | } |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1085 | } |
| 1086 | |
Kirill Tkhai | f4ebcbc | 2014-03-15 02:15:00 +0400 | [diff] [blame] | 1087 | static void |
Nicolas Saenz Julienne | 5c66d1b | 2022-06-28 11:22:59 +0200 | [diff] [blame] | 1088 | dequeue_top_rt_rq(struct rt_rq *rt_rq, unsigned int count) |
Kirill Tkhai | f4ebcbc | 2014-03-15 02:15:00 +0400 | [diff] [blame] | 1089 | { |
| 1090 | struct rq *rq = rq_of_rt_rq(rt_rq); |
| 1091 | |
| 1092 | BUG_ON(&rq->rt != rt_rq); |
| 1093 | |
| 1094 | if (!rt_rq->rt_queued) |
| 1095 | return; |
| 1096 | |
| 1097 | BUG_ON(!rq->nr_running); |
| 1098 | |
Nicolas Saenz Julienne | 5c66d1b | 2022-06-28 11:22:59 +0200 | [diff] [blame] | 1099 | sub_nr_running(rq, count); |
Kirill Tkhai | f4ebcbc | 2014-03-15 02:15:00 +0400 | [diff] [blame] | 1100 | rt_rq->rt_queued = 0; |
Peter Zijlstra | 8f111bc | 2017-12-20 16:26:12 +0100 | [diff] [blame] | 1101 | |
Kirill Tkhai | f4ebcbc | 2014-03-15 02:15:00 +0400 | [diff] [blame] | 1102 | } |
| 1103 | |
| 1104 | static void |
| 1105 | enqueue_top_rt_rq(struct rt_rq *rt_rq) |
| 1106 | { |
| 1107 | struct rq *rq = rq_of_rt_rq(rt_rq); |
| 1108 | |
| 1109 | BUG_ON(&rq->rt != rt_rq); |
| 1110 | |
| 1111 | if (rt_rq->rt_queued) |
| 1112 | return; |
Vincent Guittot | 296b2ff | 2018-06-26 15:53:22 +0200 | [diff] [blame] | 1113 | |
| 1114 | if (rt_rq_throttled(rt_rq)) |
Kirill Tkhai | f4ebcbc | 2014-03-15 02:15:00 +0400 | [diff] [blame] | 1115 | return; |
| 1116 | |
Vincent Guittot | 296b2ff | 2018-06-26 15:53:22 +0200 | [diff] [blame] | 1117 | if (rt_rq->rt_nr_running) { |
| 1118 | add_nr_running(rq, rt_rq->rt_nr_running); |
| 1119 | rt_rq->rt_queued = 1; |
| 1120 | } |
Peter Zijlstra | 8f111bc | 2017-12-20 16:26:12 +0100 | [diff] [blame] | 1121 | |
| 1122 | /* Kick cpufreq (see the comment in kernel/sched/sched.h). */ |
| 1123 | cpufreq_update_util(rq, 0); |
Kirill Tkhai | f4ebcbc | 2014-03-15 02:15:00 +0400 | [diff] [blame] | 1124 | } |
| 1125 | |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 1126 | #if defined CONFIG_SMP |
Gregory Haskins | e864c49 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 1127 | |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 1128 | static void |
| 1129 | inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) |
Steven Rostedt | 63489e4 | 2008-01-25 21:08:03 +0100 | [diff] [blame] | 1130 | { |
Gregory Haskins | 4d98427 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 1131 | struct rq *rq = rq_of_rt_rq(rt_rq); |
Gregory Haskins | 4d98427 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 1132 | |
Kirill Tkhai | 757dfca | 2013-11-27 19:59:13 +0400 | [diff] [blame] | 1133 | #ifdef CONFIG_RT_GROUP_SCHED |
| 1134 | /* |
| 1135 | * Change rq's cpupri only if rt_rq is the top queue. |
| 1136 | */ |
| 1137 | if (&rq->rt != rt_rq) |
| 1138 | return; |
| 1139 | #endif |
Steven Rostedt | 5181f4a4 | 2011-06-16 21:55:23 -0400 | [diff] [blame] | 1140 | if (rq->online && prio < prev_prio) |
| 1141 | cpupri_set(&rq->rd->cpupri, rq->cpu, prio); |
Steven Rostedt | 63489e4 | 2008-01-25 21:08:03 +0100 | [diff] [blame] | 1142 | } |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1143 | |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 1144 | static void |
| 1145 | dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) |
Steven Rostedt | 63489e4 | 2008-01-25 21:08:03 +0100 | [diff] [blame] | 1146 | { |
Gregory Haskins | 4d98427 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 1147 | struct rq *rq = rq_of_rt_rq(rt_rq); |
Gregory Haskins | 6e0534f | 2008-05-12 21:21:01 +0200 | [diff] [blame] | 1148 | |
Kirill Tkhai | 757dfca | 2013-11-27 19:59:13 +0400 | [diff] [blame] | 1149 | #ifdef CONFIG_RT_GROUP_SCHED |
| 1150 | /* |
| 1151 | * Change rq's cpupri only if rt_rq is the top queue. |
| 1152 | */ |
| 1153 | if (&rq->rt != rt_rq) |
| 1154 | return; |
| 1155 | #endif |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 1156 | if (rq->online && rt_rq->highest_prio.curr != prev_prio) |
| 1157 | cpupri_set(&rq->rd->cpupri, rq->cpu, rt_rq->highest_prio.curr); |
| 1158 | } |
| 1159 | |
| 1160 | #else /* CONFIG_SMP */ |
| 1161 | |
| 1162 | static inline |
| 1163 | void inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {} |
| 1164 | static inline |
| 1165 | void dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {} |
| 1166 | |
| 1167 | #endif /* CONFIG_SMP */ |
| 1168 | |
Steven Rostedt | 63489e4 | 2008-01-25 21:08:03 +0100 | [diff] [blame] | 1169 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 1170 | static void |
| 1171 | inc_rt_prio(struct rt_rq *rt_rq, int prio) |
| 1172 | { |
| 1173 | int prev_prio = rt_rq->highest_prio.curr; |
Steven Rostedt | 63489e4 | 2008-01-25 21:08:03 +0100 | [diff] [blame] | 1174 | |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 1175 | if (prio < prev_prio) |
| 1176 | rt_rq->highest_prio.curr = prio; |
| 1177 | |
| 1178 | inc_rt_prio_smp(rt_rq, prio, prev_prio); |
| 1179 | } |
| 1180 | |
| 1181 | static void |
| 1182 | dec_rt_prio(struct rt_rq *rt_rq, int prio) |
| 1183 | { |
| 1184 | int prev_prio = rt_rq->highest_prio.curr; |
| 1185 | |
| 1186 | if (rt_rq->rt_nr_running) { |
| 1187 | |
| 1188 | WARN_ON(prio < prev_prio); |
Gregory Haskins | e864c49 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 1189 | |
| 1190 | /* |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 1191 | * This may have been our highest task, and therefore |
| 1192 | * we may have some recomputation to do |
Gregory Haskins | e864c49 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 1193 | */ |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 1194 | if (prio == prev_prio) { |
Gregory Haskins | e864c49 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 1195 | struct rt_prio_array *array = &rt_rq->active; |
| 1196 | |
| 1197 | rt_rq->highest_prio.curr = |
Steven Rostedt | 764a9d6 | 2008-01-25 21:08:04 +0100 | [diff] [blame] | 1198 | sched_find_first_bit(array->bitmap); |
Gregory Haskins | e864c49 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 1199 | } |
| 1200 | |
Peter Zijlstra | 934fc33 | 2020-10-14 21:06:49 +0200 | [diff] [blame] | 1201 | } else { |
| 1202 | rt_rq->highest_prio.curr = MAX_RT_PRIO-1; |
| 1203 | } |
Gregory Haskins | 73fe6aae | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1204 | |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 1205 | dec_rt_prio_smp(rt_rq, prio, prev_prio); |
| 1206 | } |
Gregory Haskins | 1f11eb6a | 2008-06-04 15:04:05 -0400 | [diff] [blame] | 1207 | |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 1208 | #else |
| 1209 | |
| 1210 | static inline void inc_rt_prio(struct rt_rq *rt_rq, int prio) {} |
| 1211 | static inline void dec_rt_prio(struct rt_rq *rt_rq, int prio) {} |
| 1212 | |
| 1213 | #endif /* CONFIG_SMP || CONFIG_RT_GROUP_SCHED */ |
| 1214 | |
Gregory Haskins | 73fe6aae | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1215 | #ifdef CONFIG_RT_GROUP_SCHED |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 1216 | |
| 1217 | static void |
| 1218 | inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
| 1219 | { |
Gregory Haskins | 73fe6aae | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1220 | if (rt_se_boosted(rt_se)) |
Steven Rostedt | 764a9d6 | 2008-01-25 21:08:04 +0100 | [diff] [blame] | 1221 | rt_rq->rt_nr_boosted++; |
Peter Zijlstra | 052f1dc | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 1222 | |
Peter Zijlstra | 23b0fdf | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 1223 | if (rt_rq->tg) |
| 1224 | start_rt_bandwidth(&rt_rq->tg->rt_bandwidth); |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 1225 | } |
| 1226 | |
| 1227 | static void |
| 1228 | dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
| 1229 | { |
Peter Zijlstra | 23b0fdf | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 1230 | if (rt_se_boosted(rt_se)) |
| 1231 | rt_rq->rt_nr_boosted--; |
| 1232 | |
| 1233 | WARN_ON(!rt_rq->rt_nr_running && rt_rq->rt_nr_boosted); |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 1234 | } |
| 1235 | |
| 1236 | #else /* CONFIG_RT_GROUP_SCHED */ |
| 1237 | |
| 1238 | static void |
| 1239 | inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
| 1240 | { |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1241 | start_rt_bandwidth(&def_rt_bandwidth); |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 1242 | } |
| 1243 | |
| 1244 | static inline |
| 1245 | void dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) {} |
| 1246 | |
| 1247 | #endif /* CONFIG_RT_GROUP_SCHED */ |
| 1248 | |
| 1249 | static inline |
Kirill Tkhai | 22abdef | 2014-03-15 02:14:49 +0400 | [diff] [blame] | 1250 | unsigned int rt_se_nr_running(struct sched_rt_entity *rt_se) |
| 1251 | { |
| 1252 | struct rt_rq *group_rq = group_rt_rq(rt_se); |
| 1253 | |
| 1254 | if (group_rq) |
| 1255 | return group_rq->rt_nr_running; |
| 1256 | else |
| 1257 | return 1; |
| 1258 | } |
| 1259 | |
| 1260 | static inline |
Frederic Weisbecker | 01d36d0 | 2015-11-04 18:17:10 +0100 | [diff] [blame] | 1261 | unsigned int rt_se_rr_nr_running(struct sched_rt_entity *rt_se) |
| 1262 | { |
| 1263 | struct rt_rq *group_rq = group_rt_rq(rt_se); |
| 1264 | struct task_struct *tsk; |
| 1265 | |
| 1266 | if (group_rq) |
| 1267 | return group_rq->rr_nr_running; |
| 1268 | |
| 1269 | tsk = rt_task_of(rt_se); |
| 1270 | |
| 1271 | return (tsk->policy == SCHED_RR) ? 1 : 0; |
| 1272 | } |
| 1273 | |
| 1274 | static inline |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 1275 | void inc_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
| 1276 | { |
| 1277 | int prio = rt_se_prio(rt_se); |
| 1278 | |
| 1279 | WARN_ON(!rt_prio(prio)); |
Kirill Tkhai | 22abdef | 2014-03-15 02:14:49 +0400 | [diff] [blame] | 1280 | rt_rq->rt_nr_running += rt_se_nr_running(rt_se); |
Frederic Weisbecker | 01d36d0 | 2015-11-04 18:17:10 +0100 | [diff] [blame] | 1281 | rt_rq->rr_nr_running += rt_se_rr_nr_running(rt_se); |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 1282 | |
| 1283 | inc_rt_prio(rt_rq, prio); |
| 1284 | inc_rt_migration(rt_se, rt_rq); |
| 1285 | inc_rt_group(rt_se, rt_rq); |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1286 | } |
| 1287 | |
Peter Zijlstra | 23b0fdf | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 1288 | static inline |
| 1289 | void dec_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
| 1290 | { |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1291 | WARN_ON(!rt_prio(rt_se_prio(rt_se))); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1292 | WARN_ON(!rt_rq->rt_nr_running); |
Kirill Tkhai | 22abdef | 2014-03-15 02:14:49 +0400 | [diff] [blame] | 1293 | rt_rq->rt_nr_running -= rt_se_nr_running(rt_se); |
Frederic Weisbecker | 01d36d0 | 2015-11-04 18:17:10 +0100 | [diff] [blame] | 1294 | rt_rq->rr_nr_running -= rt_se_rr_nr_running(rt_se); |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1295 | |
Gregory Haskins | 398a153 | 2009-01-14 09:10:04 -0500 | [diff] [blame] | 1296 | dec_rt_prio(rt_rq, rt_se_prio(rt_se)); |
| 1297 | dec_rt_migration(rt_se, rt_rq); |
| 1298 | dec_rt_group(rt_se, rt_rq); |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1299 | } |
| 1300 | |
Peter Zijlstra | ff77e46 | 2016-01-18 15:27:07 +0100 | [diff] [blame] | 1301 | /* |
| 1302 | * Change rt_se->run_list location unless SAVE && !MOVE |
| 1303 | * |
| 1304 | * assumes ENQUEUE/DEQUEUE flags match |
| 1305 | */ |
| 1306 | static inline bool move_entity(unsigned int flags) |
| 1307 | { |
| 1308 | if ((flags & (DEQUEUE_SAVE | DEQUEUE_MOVE)) == DEQUEUE_SAVE) |
| 1309 | return false; |
| 1310 | |
| 1311 | return true; |
| 1312 | } |
| 1313 | |
| 1314 | static void __delist_rt_entity(struct sched_rt_entity *rt_se, struct rt_prio_array *array) |
| 1315 | { |
| 1316 | list_del_init(&rt_se->run_list); |
| 1317 | |
| 1318 | if (list_empty(array->queue + rt_se_prio(rt_se))) |
| 1319 | __clear_bit(rt_se_prio(rt_se), array->bitmap); |
| 1320 | |
| 1321 | rt_se->on_list = 0; |
| 1322 | } |
| 1323 | |
Yafang Shao | 57a5c2d | 2021-09-05 14:35:45 +0000 | [diff] [blame] | 1324 | static inline struct sched_statistics * |
| 1325 | __schedstats_from_rt_se(struct sched_rt_entity *rt_se) |
| 1326 | { |
| 1327 | #ifdef CONFIG_RT_GROUP_SCHED |
| 1328 | /* schedstats is not supported for rt group. */ |
| 1329 | if (!rt_entity_is_task(rt_se)) |
| 1330 | return NULL; |
| 1331 | #endif |
| 1332 | |
| 1333 | return &rt_task_of(rt_se)->stats; |
| 1334 | } |
| 1335 | |
| 1336 | static inline void |
| 1337 | update_stats_wait_start_rt(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se) |
| 1338 | { |
| 1339 | struct sched_statistics *stats; |
| 1340 | struct task_struct *p = NULL; |
| 1341 | |
| 1342 | if (!schedstat_enabled()) |
| 1343 | return; |
| 1344 | |
| 1345 | if (rt_entity_is_task(rt_se)) |
| 1346 | p = rt_task_of(rt_se); |
| 1347 | |
| 1348 | stats = __schedstats_from_rt_se(rt_se); |
| 1349 | if (!stats) |
| 1350 | return; |
| 1351 | |
| 1352 | __update_stats_wait_start(rq_of_rt_rq(rt_rq), p, stats); |
| 1353 | } |
| 1354 | |
| 1355 | static inline void |
| 1356 | update_stats_enqueue_sleeper_rt(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se) |
| 1357 | { |
| 1358 | struct sched_statistics *stats; |
| 1359 | struct task_struct *p = NULL; |
| 1360 | |
| 1361 | if (!schedstat_enabled()) |
| 1362 | return; |
| 1363 | |
| 1364 | if (rt_entity_is_task(rt_se)) |
| 1365 | p = rt_task_of(rt_se); |
| 1366 | |
| 1367 | stats = __schedstats_from_rt_se(rt_se); |
| 1368 | if (!stats) |
| 1369 | return; |
| 1370 | |
| 1371 | __update_stats_enqueue_sleeper(rq_of_rt_rq(rt_rq), p, stats); |
| 1372 | } |
| 1373 | |
| 1374 | static inline void |
| 1375 | update_stats_enqueue_rt(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se, |
| 1376 | int flags) |
| 1377 | { |
| 1378 | if (!schedstat_enabled()) |
| 1379 | return; |
| 1380 | |
| 1381 | if (flags & ENQUEUE_WAKEUP) |
| 1382 | update_stats_enqueue_sleeper_rt(rt_rq, rt_se); |
| 1383 | } |
| 1384 | |
| 1385 | static inline void |
| 1386 | update_stats_wait_end_rt(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se) |
| 1387 | { |
| 1388 | struct sched_statistics *stats; |
| 1389 | struct task_struct *p = NULL; |
| 1390 | |
| 1391 | if (!schedstat_enabled()) |
| 1392 | return; |
| 1393 | |
| 1394 | if (rt_entity_is_task(rt_se)) |
| 1395 | p = rt_task_of(rt_se); |
| 1396 | |
| 1397 | stats = __schedstats_from_rt_se(rt_se); |
| 1398 | if (!stats) |
| 1399 | return; |
| 1400 | |
| 1401 | __update_stats_wait_end(rq_of_rt_rq(rt_rq), p, stats); |
| 1402 | } |
| 1403 | |
| 1404 | static inline void |
| 1405 | update_stats_dequeue_rt(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se, |
| 1406 | int flags) |
| 1407 | { |
| 1408 | struct task_struct *p = NULL; |
| 1409 | |
| 1410 | if (!schedstat_enabled()) |
| 1411 | return; |
| 1412 | |
| 1413 | if (rt_entity_is_task(rt_se)) |
| 1414 | p = rt_task_of(rt_se); |
| 1415 | |
| 1416 | if ((flags & DEQUEUE_SLEEP) && p) { |
| 1417 | unsigned int state; |
| 1418 | |
| 1419 | state = READ_ONCE(p->__state); |
| 1420 | if (state & TASK_INTERRUPTIBLE) |
| 1421 | __schedstat_set(p->stats.sleep_start, |
| 1422 | rq_clock(rq_of_rt_rq(rt_rq))); |
| 1423 | |
| 1424 | if (state & TASK_UNINTERRUPTIBLE) |
| 1425 | __schedstat_set(p->stats.block_start, |
| 1426 | rq_clock(rq_of_rt_rq(rt_rq))); |
| 1427 | } |
| 1428 | } |
| 1429 | |
Peter Zijlstra | ff77e46 | 2016-01-18 15:27:07 +0100 | [diff] [blame] | 1430 | static void __enqueue_rt_entity(struct sched_rt_entity *rt_se, unsigned int flags) |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1431 | { |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1432 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); |
| 1433 | struct rt_prio_array *array = &rt_rq->active; |
| 1434 | struct rt_rq *group_rq = group_rt_rq(rt_se); |
Dmitry Adamushko | 20b6331 | 2008-06-11 00:58:30 +0200 | [diff] [blame] | 1435 | struct list_head *queue = array->queue + rt_se_prio(rt_se); |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1436 | |
Peter Zijlstra | ad2a3f1 | 2008-06-19 09:06:57 +0200 | [diff] [blame] | 1437 | /* |
| 1438 | * Don't enqueue the group if its throttled, or when empty. |
| 1439 | * The latter is a consequence of the former when a child group |
| 1440 | * get throttled and the current group doesn't have any other |
| 1441 | * active members. |
| 1442 | */ |
Peter Zijlstra | ff77e46 | 2016-01-18 15:27:07 +0100 | [diff] [blame] | 1443 | if (group_rq && (rt_rq_throttled(group_rq) || !group_rq->rt_nr_running)) { |
| 1444 | if (rt_se->on_list) |
| 1445 | __delist_rt_entity(rt_se, array); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1446 | return; |
Peter Zijlstra | ff77e46 | 2016-01-18 15:27:07 +0100 | [diff] [blame] | 1447 | } |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1448 | |
Peter Zijlstra | ff77e46 | 2016-01-18 15:27:07 +0100 | [diff] [blame] | 1449 | if (move_entity(flags)) { |
| 1450 | WARN_ON_ONCE(rt_se->on_list); |
| 1451 | if (flags & ENQUEUE_HEAD) |
| 1452 | list_add(&rt_se->run_list, queue); |
| 1453 | else |
| 1454 | list_add_tail(&rt_se->run_list, queue); |
| 1455 | |
| 1456 | __set_bit(rt_se_prio(rt_se), array->bitmap); |
| 1457 | rt_se->on_list = 1; |
| 1458 | } |
| 1459 | rt_se->on_rq = 1; |
Peter Zijlstra | 78f2c7d | 2008-01-25 21:08:27 +0100 | [diff] [blame] | 1460 | |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1461 | inc_rt_tasks(rt_se, rt_rq); |
| 1462 | } |
| 1463 | |
Peter Zijlstra | ff77e46 | 2016-01-18 15:27:07 +0100 | [diff] [blame] | 1464 | static void __dequeue_rt_entity(struct sched_rt_entity *rt_se, unsigned int flags) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1465 | { |
| 1466 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); |
| 1467 | struct rt_prio_array *array = &rt_rq->active; |
| 1468 | |
Peter Zijlstra | ff77e46 | 2016-01-18 15:27:07 +0100 | [diff] [blame] | 1469 | if (move_entity(flags)) { |
| 1470 | WARN_ON_ONCE(!rt_se->on_list); |
| 1471 | __delist_rt_entity(rt_se, array); |
| 1472 | } |
| 1473 | rt_se->on_rq = 0; |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1474 | |
| 1475 | dec_rt_tasks(rt_se, rt_rq); |
| 1476 | } |
| 1477 | |
| 1478 | /* |
| 1479 | * Because the prio of an upper entry depends on the lower |
| 1480 | * entries, we must remove entries top - down. |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1481 | */ |
Peter Zijlstra | ff77e46 | 2016-01-18 15:27:07 +0100 | [diff] [blame] | 1482 | static void dequeue_rt_stack(struct sched_rt_entity *rt_se, unsigned int flags) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1483 | { |
Peter Zijlstra | ad2a3f1 | 2008-06-19 09:06:57 +0200 | [diff] [blame] | 1484 | struct sched_rt_entity *back = NULL; |
Nicolas Saenz Julienne | 5c66d1b | 2022-06-28 11:22:59 +0200 | [diff] [blame] | 1485 | unsigned int rt_nr_running; |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1486 | |
Peter Zijlstra | 58d6c2d | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 1487 | for_each_sched_rt_entity(rt_se) { |
| 1488 | rt_se->back = back; |
| 1489 | back = rt_se; |
| 1490 | } |
| 1491 | |
Nicolas Saenz Julienne | 5c66d1b | 2022-06-28 11:22:59 +0200 | [diff] [blame] | 1492 | rt_nr_running = rt_rq_of_se(back)->rt_nr_running; |
Kirill Tkhai | f4ebcbc | 2014-03-15 02:15:00 +0400 | [diff] [blame] | 1493 | |
Peter Zijlstra | 58d6c2d | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 1494 | for (rt_se = back; rt_se; rt_se = rt_se->back) { |
| 1495 | if (on_rt_rq(rt_se)) |
Peter Zijlstra | ff77e46 | 2016-01-18 15:27:07 +0100 | [diff] [blame] | 1496 | __dequeue_rt_entity(rt_se, flags); |
Peter Zijlstra | ad2a3f1 | 2008-06-19 09:06:57 +0200 | [diff] [blame] | 1497 | } |
Nicolas Saenz Julienne | 5c66d1b | 2022-06-28 11:22:59 +0200 | [diff] [blame] | 1498 | |
| 1499 | dequeue_top_rt_rq(rt_rq_of_se(back), rt_nr_running); |
Peter Zijlstra | ad2a3f1 | 2008-06-19 09:06:57 +0200 | [diff] [blame] | 1500 | } |
| 1501 | |
Peter Zijlstra | ff77e46 | 2016-01-18 15:27:07 +0100 | [diff] [blame] | 1502 | static void enqueue_rt_entity(struct sched_rt_entity *rt_se, unsigned int flags) |
Peter Zijlstra | ad2a3f1 | 2008-06-19 09:06:57 +0200 | [diff] [blame] | 1503 | { |
Kirill Tkhai | f4ebcbc | 2014-03-15 02:15:00 +0400 | [diff] [blame] | 1504 | struct rq *rq = rq_of_rt_se(rt_se); |
| 1505 | |
Yafang Shao | 57a5c2d | 2021-09-05 14:35:45 +0000 | [diff] [blame] | 1506 | update_stats_enqueue_rt(rt_rq_of_se(rt_se), rt_se, flags); |
| 1507 | |
Peter Zijlstra | ff77e46 | 2016-01-18 15:27:07 +0100 | [diff] [blame] | 1508 | dequeue_rt_stack(rt_se, flags); |
Peter Zijlstra | ad2a3f1 | 2008-06-19 09:06:57 +0200 | [diff] [blame] | 1509 | for_each_sched_rt_entity(rt_se) |
Peter Zijlstra | ff77e46 | 2016-01-18 15:27:07 +0100 | [diff] [blame] | 1510 | __enqueue_rt_entity(rt_se, flags); |
Kirill Tkhai | f4ebcbc | 2014-03-15 02:15:00 +0400 | [diff] [blame] | 1511 | enqueue_top_rt_rq(&rq->rt); |
Peter Zijlstra | ad2a3f1 | 2008-06-19 09:06:57 +0200 | [diff] [blame] | 1512 | } |
| 1513 | |
Peter Zijlstra | ff77e46 | 2016-01-18 15:27:07 +0100 | [diff] [blame] | 1514 | static void dequeue_rt_entity(struct sched_rt_entity *rt_se, unsigned int flags) |
Peter Zijlstra | ad2a3f1 | 2008-06-19 09:06:57 +0200 | [diff] [blame] | 1515 | { |
Kirill Tkhai | f4ebcbc | 2014-03-15 02:15:00 +0400 | [diff] [blame] | 1516 | struct rq *rq = rq_of_rt_se(rt_se); |
| 1517 | |
Yafang Shao | 57a5c2d | 2021-09-05 14:35:45 +0000 | [diff] [blame] | 1518 | update_stats_dequeue_rt(rt_rq_of_se(rt_se), rt_se, flags); |
| 1519 | |
Peter Zijlstra | ff77e46 | 2016-01-18 15:27:07 +0100 | [diff] [blame] | 1520 | dequeue_rt_stack(rt_se, flags); |
Peter Zijlstra | ad2a3f1 | 2008-06-19 09:06:57 +0200 | [diff] [blame] | 1521 | |
| 1522 | for_each_sched_rt_entity(rt_se) { |
| 1523 | struct rt_rq *rt_rq = group_rt_rq(rt_se); |
| 1524 | |
| 1525 | if (rt_rq && rt_rq->rt_nr_running) |
Peter Zijlstra | ff77e46 | 2016-01-18 15:27:07 +0100 | [diff] [blame] | 1526 | __enqueue_rt_entity(rt_se, flags); |
Peter Zijlstra | 58d6c2d | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 1527 | } |
Kirill Tkhai | f4ebcbc | 2014-03-15 02:15:00 +0400 | [diff] [blame] | 1528 | enqueue_top_rt_rq(&rq->rt); |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1529 | } |
| 1530 | |
| 1531 | /* |
| 1532 | * Adding/removing a task to/from a priority array: |
| 1533 | */ |
Thomas Gleixner | ea87bb7 | 2010-01-20 20:58:57 +0000 | [diff] [blame] | 1534 | static void |
Peter Zijlstra | 371fd7e | 2010-03-24 16:38:48 +0100 | [diff] [blame] | 1535 | enqueue_task_rt(struct rq *rq, struct task_struct *p, int flags) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1536 | { |
| 1537 | struct sched_rt_entity *rt_se = &p->rt; |
| 1538 | |
Peter Zijlstra | 371fd7e | 2010-03-24 16:38:48 +0100 | [diff] [blame] | 1539 | if (flags & ENQUEUE_WAKEUP) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1540 | rt_se->timeout = 0; |
| 1541 | |
Yafang Shao | 57a5c2d | 2021-09-05 14:35:45 +0000 | [diff] [blame] | 1542 | check_schedstat_required(); |
| 1543 | update_stats_wait_start_rt(rt_rq_of_se(rt_se), rt_se); |
| 1544 | |
Peter Zijlstra | ff77e46 | 2016-01-18 15:27:07 +0100 | [diff] [blame] | 1545 | enqueue_rt_entity(rt_se, flags); |
Peter Zijlstra | c09595f | 2008-06-27 13:41:14 +0200 | [diff] [blame] | 1546 | |
Ingo Molnar | 4b53a34 | 2017-02-05 15:41:03 +0100 | [diff] [blame] | 1547 | if (!task_current(rq, p) && p->nr_cpus_allowed > 1) |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 1548 | enqueue_pushable_task(rq, p); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1549 | } |
| 1550 | |
Peter Zijlstra | 371fd7e | 2010-03-24 16:38:48 +0100 | [diff] [blame] | 1551 | static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int flags) |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1552 | { |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1553 | struct sched_rt_entity *rt_se = &p->rt; |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1554 | |
| 1555 | update_curr_rt(rq); |
Peter Zijlstra | ff77e46 | 2016-01-18 15:27:07 +0100 | [diff] [blame] | 1556 | dequeue_rt_entity(rt_se, flags); |
Peter Zijlstra | c09595f | 2008-06-27 13:41:14 +0200 | [diff] [blame] | 1557 | |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 1558 | dequeue_pushable_task(rq, p); |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1559 | } |
| 1560 | |
| 1561 | /* |
Richard Weinberger | 6068631 | 2011-11-12 18:07:57 +0100 | [diff] [blame] | 1562 | * Put task to the head or the end of the run list without the overhead of |
| 1563 | * dequeue followed by enqueue. |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1564 | */ |
Dmitry Adamushko | 7ebefa8 | 2008-07-01 23:32:15 +0200 | [diff] [blame] | 1565 | static void |
| 1566 | requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se, int head) |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1567 | { |
Ingo Molnar | 1cdad71 | 2008-06-19 09:09:15 +0200 | [diff] [blame] | 1568 | if (on_rt_rq(rt_se)) { |
Dmitry Adamushko | 7ebefa8 | 2008-07-01 23:32:15 +0200 | [diff] [blame] | 1569 | struct rt_prio_array *array = &rt_rq->active; |
| 1570 | struct list_head *queue = array->queue + rt_se_prio(rt_se); |
| 1571 | |
| 1572 | if (head) |
| 1573 | list_move(&rt_se->run_list, queue); |
| 1574 | else |
| 1575 | list_move_tail(&rt_se->run_list, queue); |
Ingo Molnar | 1cdad71 | 2008-06-19 09:09:15 +0200 | [diff] [blame] | 1576 | } |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1577 | } |
| 1578 | |
Dmitry Adamushko | 7ebefa8 | 2008-07-01 23:32:15 +0200 | [diff] [blame] | 1579 | static void requeue_task_rt(struct rq *rq, struct task_struct *p, int head) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1580 | { |
| 1581 | struct sched_rt_entity *rt_se = &p->rt; |
| 1582 | struct rt_rq *rt_rq; |
| 1583 | |
| 1584 | for_each_sched_rt_entity(rt_se) { |
| 1585 | rt_rq = rt_rq_of_se(rt_se); |
Dmitry Adamushko | 7ebefa8 | 2008-07-01 23:32:15 +0200 | [diff] [blame] | 1586 | requeue_rt_entity(rt_rq, rt_se, head); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1587 | } |
| 1588 | } |
| 1589 | |
| 1590 | static void yield_task_rt(struct rq *rq) |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1591 | { |
Dmitry Adamushko | 7ebefa8 | 2008-07-01 23:32:15 +0200 | [diff] [blame] | 1592 | requeue_task_rt(rq, rq->curr, 0); |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1593 | } |
| 1594 | |
Gregory Haskins | e7693a3 | 2008-01-25 21:08:09 +0100 | [diff] [blame] | 1595 | #ifdef CONFIG_SMP |
Gregory Haskins | 318e089 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 1596 | static int find_lowest_rq(struct task_struct *task); |
| 1597 | |
Peter Zijlstra | 0017d73 | 2010-03-24 18:34:10 +0100 | [diff] [blame] | 1598 | static int |
Valentin Schneider | 3aef1551 | 2020-11-02 18:45:13 +0000 | [diff] [blame] | 1599 | select_task_rq_rt(struct task_struct *p, int cpu, int flags) |
Gregory Haskins | e7693a3 | 2008-01-25 21:08:09 +0100 | [diff] [blame] | 1600 | { |
Peter Zijlstra | 7608dec | 2011-04-05 17:23:46 +0200 | [diff] [blame] | 1601 | struct task_struct *curr; |
| 1602 | struct rq *rq; |
Qais Yousef | 804d402 | 2019-10-09 11:46:11 +0100 | [diff] [blame] | 1603 | bool test; |
Steven Rostedt | c37495f | 2011-06-16 21:55:22 -0400 | [diff] [blame] | 1604 | |
| 1605 | /* For anything but wake ups, just return the task_cpu */ |
Valentin Schneider | 3aef1551 | 2020-11-02 18:45:13 +0000 | [diff] [blame] | 1606 | if (!(flags & (WF_TTWU | WF_FORK))) |
Steven Rostedt | c37495f | 2011-06-16 21:55:22 -0400 | [diff] [blame] | 1607 | goto out; |
| 1608 | |
Peter Zijlstra | 7608dec | 2011-04-05 17:23:46 +0200 | [diff] [blame] | 1609 | rq = cpu_rq(cpu); |
| 1610 | |
| 1611 | rcu_read_lock(); |
Jason Low | 316c1608d | 2015-04-28 13:00:20 -0700 | [diff] [blame] | 1612 | curr = READ_ONCE(rq->curr); /* unlocked access */ |
Peter Zijlstra | 7608dec | 2011-04-05 17:23:46 +0200 | [diff] [blame] | 1613 | |
Gregory Haskins | 318e089 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 1614 | /* |
Peter Zijlstra | 7608dec | 2011-04-05 17:23:46 +0200 | [diff] [blame] | 1615 | * If the current task on @p's runqueue is an RT task, then |
Steven Rostedt | e1f47d8 | 2008-01-25 21:08:12 +0100 | [diff] [blame] | 1616 | * try to see if we can wake this RT task up on another |
| 1617 | * runqueue. Otherwise simply start this RT task |
| 1618 | * on its current runqueue. |
| 1619 | * |
Steven Rostedt | 43fa546 | 2010-09-20 22:40:03 -0400 | [diff] [blame] | 1620 | * We want to avoid overloading runqueues. If the woken |
| 1621 | * task is a higher priority, then it will stay on this CPU |
| 1622 | * and the lower prio task should be moved to another CPU. |
| 1623 | * Even though this will probably make the lower prio task |
| 1624 | * lose its cache, we do not want to bounce a higher task |
| 1625 | * around just because it gave up its CPU, perhaps for a |
| 1626 | * lock? |
| 1627 | * |
| 1628 | * For equal prio tasks, we just let the scheduler sort it out. |
Peter Zijlstra | 7608dec | 2011-04-05 17:23:46 +0200 | [diff] [blame] | 1629 | * |
Gregory Haskins | 318e089 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 1630 | * Otherwise, just let it ride on the affined RQ and the |
| 1631 | * post-schedule router will push the preempted task away |
Peter Zijlstra | 7608dec | 2011-04-05 17:23:46 +0200 | [diff] [blame] | 1632 | * |
| 1633 | * This test is optimistic, if we get it wrong the load-balancer |
| 1634 | * will have to sort it out. |
Qais Yousef | 804d402 | 2019-10-09 11:46:11 +0100 | [diff] [blame] | 1635 | * |
| 1636 | * We take into account the capacity of the CPU to ensure it fits the |
| 1637 | * requirement of the task - which is only important on heterogeneous |
| 1638 | * systems like big.LITTLE. |
Gregory Haskins | 318e089 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 1639 | */ |
Qais Yousef | 804d402 | 2019-10-09 11:46:11 +0100 | [diff] [blame] | 1640 | test = curr && |
| 1641 | unlikely(rt_task(curr)) && |
| 1642 | (curr->nr_cpus_allowed < 2 || curr->prio <= p->prio); |
| 1643 | |
| 1644 | if (test || !rt_task_fits_capacity(p, cpu)) { |
Peter Zijlstra | 7608dec | 2011-04-05 17:23:46 +0200 | [diff] [blame] | 1645 | int target = find_lowest_rq(p); |
| 1646 | |
Tim Chen | 80e3d87 | 2014-12-12 15:38:12 -0800 | [diff] [blame] | 1647 | /* |
Qais Yousef | b28bc1e | 2020-03-02 13:27:17 +0000 | [diff] [blame] | 1648 | * Bail out if we were forcing a migration to find a better |
| 1649 | * fitting CPU but our search failed. |
| 1650 | */ |
| 1651 | if (!test && target != -1 && !rt_task_fits_capacity(p, target)) |
| 1652 | goto out_unlock; |
| 1653 | |
| 1654 | /* |
Tim Chen | 80e3d87 | 2014-12-12 15:38:12 -0800 | [diff] [blame] | 1655 | * Don't bother moving it if the destination CPU is |
| 1656 | * not running a lower priority task. |
| 1657 | */ |
| 1658 | if (target != -1 && |
| 1659 | p->prio < cpu_rq(target)->rt.highest_prio.curr) |
Peter Zijlstra | 7608dec | 2011-04-05 17:23:46 +0200 | [diff] [blame] | 1660 | cpu = target; |
| 1661 | } |
Qais Yousef | b28bc1e | 2020-03-02 13:27:17 +0000 | [diff] [blame] | 1662 | |
| 1663 | out_unlock: |
Peter Zijlstra | 7608dec | 2011-04-05 17:23:46 +0200 | [diff] [blame] | 1664 | rcu_read_unlock(); |
| 1665 | |
Steven Rostedt | c37495f | 2011-06-16 21:55:22 -0400 | [diff] [blame] | 1666 | out: |
Peter Zijlstra | 7608dec | 2011-04-05 17:23:46 +0200 | [diff] [blame] | 1667 | return cpu; |
Gregory Haskins | e7693a3 | 2008-01-25 21:08:09 +0100 | [diff] [blame] | 1668 | } |
Dmitry Adamushko | 7ebefa8 | 2008-07-01 23:32:15 +0200 | [diff] [blame] | 1669 | |
| 1670 | static void check_preempt_equal_prio(struct rq *rq, struct task_struct *p) |
| 1671 | { |
Wanpeng Li | 308a623 | 2014-10-31 06:39:31 +0800 | [diff] [blame] | 1672 | /* |
| 1673 | * Current can't be migrated, useless to reschedule, |
| 1674 | * let's hope p can move out. |
| 1675 | */ |
Ingo Molnar | 4b53a34 | 2017-02-05 15:41:03 +0100 | [diff] [blame] | 1676 | if (rq->curr->nr_cpus_allowed == 1 || |
Qais Yousef | a1bd02e | 2020-03-02 13:27:18 +0000 | [diff] [blame] | 1677 | !cpupri_find(&rq->rd->cpupri, rq->curr, NULL)) |
Dmitry Adamushko | 7ebefa8 | 2008-07-01 23:32:15 +0200 | [diff] [blame] | 1678 | return; |
| 1679 | |
Wanpeng Li | 308a623 | 2014-10-31 06:39:31 +0800 | [diff] [blame] | 1680 | /* |
| 1681 | * p is migratable, so let's not schedule it and |
| 1682 | * see if it is pushed or pulled somewhere else. |
| 1683 | */ |
Qais Yousef | 804d402 | 2019-10-09 11:46:11 +0100 | [diff] [blame] | 1684 | if (p->nr_cpus_allowed != 1 && |
Qais Yousef | a1bd02e | 2020-03-02 13:27:18 +0000 | [diff] [blame] | 1685 | cpupri_find(&rq->rd->cpupri, p, NULL)) |
Dmitry Adamushko | 7ebefa8 | 2008-07-01 23:32:15 +0200 | [diff] [blame] | 1686 | return; |
| 1687 | |
Dmitry Adamushko | 7ebefa8 | 2008-07-01 23:32:15 +0200 | [diff] [blame] | 1688 | /* |
Ingo Molnar | 97fb7a0 | 2018-03-03 14:01:12 +0100 | [diff] [blame] | 1689 | * There appear to be other CPUs that can accept |
| 1690 | * the current task but none can run 'p', so lets reschedule |
| 1691 | * to try and push the current task away: |
Dmitry Adamushko | 7ebefa8 | 2008-07-01 23:32:15 +0200 | [diff] [blame] | 1692 | */ |
| 1693 | requeue_task_rt(rq, p, 1); |
Kirill Tkhai | 8875125 | 2014-06-29 00:03:57 +0400 | [diff] [blame] | 1694 | resched_curr(rq); |
Dmitry Adamushko | 7ebefa8 | 2008-07-01 23:32:15 +0200 | [diff] [blame] | 1695 | } |
| 1696 | |
Peter Zijlstra | 6e2df05 | 2019-11-08 11:11:52 +0100 | [diff] [blame] | 1697 | static int balance_rt(struct rq *rq, struct task_struct *p, struct rq_flags *rf) |
| 1698 | { |
| 1699 | if (!on_rt_rq(&p->rt) && need_pull_rt_task(rq, p)) { |
| 1700 | /* |
| 1701 | * This is OK, because current is on_cpu, which avoids it being |
| 1702 | * picked for load-balance and preemption/IRQs are still |
| 1703 | * disabled avoiding further scheduler activity on it and we've |
| 1704 | * not yet started the picking loop. |
| 1705 | */ |
| 1706 | rq_unpin_lock(rq, rf); |
| 1707 | pull_rt_task(rq); |
| 1708 | rq_repin_lock(rq, rf); |
| 1709 | } |
| 1710 | |
| 1711 | return sched_stop_runnable(rq) || sched_dl_runnable(rq) || sched_rt_runnable(rq); |
| 1712 | } |
Gregory Haskins | e7693a3 | 2008-01-25 21:08:09 +0100 | [diff] [blame] | 1713 | #endif /* CONFIG_SMP */ |
| 1714 | |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1715 | /* |
| 1716 | * Preempt the current task with a newly woken task if needed: |
| 1717 | */ |
Peter Zijlstra | 7d47872 | 2009-09-14 19:55:44 +0200 | [diff] [blame] | 1718 | static void check_preempt_curr_rt(struct rq *rq, struct task_struct *p, int flags) |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1719 | { |
Gregory Haskins | 45c01e8 | 2008-05-12 21:20:41 +0200 | [diff] [blame] | 1720 | if (p->prio < rq->curr->prio) { |
Kirill Tkhai | 8875125 | 2014-06-29 00:03:57 +0400 | [diff] [blame] | 1721 | resched_curr(rq); |
Gregory Haskins | 45c01e8 | 2008-05-12 21:20:41 +0200 | [diff] [blame] | 1722 | return; |
| 1723 | } |
| 1724 | |
| 1725 | #ifdef CONFIG_SMP |
| 1726 | /* |
| 1727 | * If: |
| 1728 | * |
| 1729 | * - the newly woken task is of equal priority to the current task |
| 1730 | * - the newly woken task is non-migratable while current is migratable |
| 1731 | * - current will be preempted on the next reschedule |
| 1732 | * |
| 1733 | * we should check to see if current can readily move to a different |
| 1734 | * cpu. If so, we will reschedule to allow the push logic to try |
| 1735 | * to move current somewhere else, making room for our non-migratable |
| 1736 | * task. |
| 1737 | */ |
Hillf Danton | 8dd0de8b | 2011-06-14 18:36:24 -0400 | [diff] [blame] | 1738 | if (p->prio == rq->curr->prio && !test_tsk_need_resched(rq->curr)) |
Dmitry Adamushko | 7ebefa8 | 2008-07-01 23:32:15 +0200 | [diff] [blame] | 1739 | check_preempt_equal_prio(rq, p); |
Gregory Haskins | 45c01e8 | 2008-05-12 21:20:41 +0200 | [diff] [blame] | 1740 | #endif |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1741 | } |
| 1742 | |
Peter Zijlstra | a0e813f | 2019-11-08 14:16:00 +0100 | [diff] [blame] | 1743 | static inline void set_next_task_rt(struct rq *rq, struct task_struct *p, bool first) |
Muchun Song | ff1cdc9 | 2018-10-26 21:17:43 +0800 | [diff] [blame] | 1744 | { |
Yafang Shao | 57a5c2d | 2021-09-05 14:35:45 +0000 | [diff] [blame] | 1745 | struct sched_rt_entity *rt_se = &p->rt; |
| 1746 | struct rt_rq *rt_rq = &rq->rt; |
| 1747 | |
Muchun Song | ff1cdc9 | 2018-10-26 21:17:43 +0800 | [diff] [blame] | 1748 | p->se.exec_start = rq_clock_task(rq); |
Yafang Shao | 57a5c2d | 2021-09-05 14:35:45 +0000 | [diff] [blame] | 1749 | if (on_rt_rq(&p->rt)) |
| 1750 | update_stats_wait_end_rt(rt_rq, rt_se); |
Muchun Song | ff1cdc9 | 2018-10-26 21:17:43 +0800 | [diff] [blame] | 1751 | |
| 1752 | /* The running task is never eligible for pushing */ |
| 1753 | dequeue_pushable_task(rq, p); |
Peter Zijlstra | f95d4ea | 2019-05-29 20:36:40 +0000 | [diff] [blame] | 1754 | |
Peter Zijlstra | a0e813f | 2019-11-08 14:16:00 +0100 | [diff] [blame] | 1755 | if (!first) |
| 1756 | return; |
| 1757 | |
Peter Zijlstra | f95d4ea | 2019-05-29 20:36:40 +0000 | [diff] [blame] | 1758 | /* |
| 1759 | * If prev task was rt, put_prev_task() has already updated the |
| 1760 | * utilization. We only care of the case where we start to schedule a |
| 1761 | * rt task |
| 1762 | */ |
| 1763 | if (rq->curr->sched_class != &rt_sched_class) |
| 1764 | update_rt_rq_load_avg(rq_clock_pelt(rq), rq, 0); |
| 1765 | |
| 1766 | rt_queue_push_tasks(rq); |
Muchun Song | ff1cdc9 | 2018-10-26 21:17:43 +0800 | [diff] [blame] | 1767 | } |
| 1768 | |
Dietmar Eggemann | 821aecd | 2022-03-02 19:34:33 +0100 | [diff] [blame] | 1769 | static struct sched_rt_entity *pick_next_rt_entity(struct rt_rq *rt_rq) |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1770 | { |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1771 | struct rt_prio_array *array = &rt_rq->active; |
| 1772 | struct sched_rt_entity *next = NULL; |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1773 | struct list_head *queue; |
| 1774 | int idx; |
| 1775 | |
| 1776 | idx = sched_find_first_bit(array->bitmap); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1777 | BUG_ON(idx >= MAX_RT_PRIO); |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1778 | |
| 1779 | queue = array->queue + idx; |
Pietro Borrello | 7c4a5b8 | 2023-02-06 22:33:54 +0000 | [diff] [blame] | 1780 | if (SCHED_WARN_ON(list_empty(queue))) |
| 1781 | return NULL; |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1782 | next = list_entry(queue->next, struct sched_rt_entity, run_list); |
Dmitry Adamushko | 326587b | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 1783 | |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1784 | return next; |
| 1785 | } |
| 1786 | |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 1787 | static struct task_struct *_pick_next_task_rt(struct rq *rq) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1788 | { |
| 1789 | struct sched_rt_entity *rt_se; |
Peter Zijlstra | 606dba2 | 2012-02-11 06:05:00 +0100 | [diff] [blame] | 1790 | struct rt_rq *rt_rq = &rq->rt; |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1791 | |
| 1792 | do { |
Dietmar Eggemann | 821aecd | 2022-03-02 19:34:33 +0100 | [diff] [blame] | 1793 | rt_se = pick_next_rt_entity(rt_rq); |
Pietro Borrello | 7c4a5b8 | 2023-02-06 22:33:54 +0000 | [diff] [blame] | 1794 | if (unlikely(!rt_se)) |
| 1795 | return NULL; |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1796 | rt_rq = group_rt_rq(rt_se); |
| 1797 | } while (rt_rq); |
| 1798 | |
Muchun Song | ff1cdc9 | 2018-10-26 21:17:43 +0800 | [diff] [blame] | 1799 | return rt_task_of(rt_se); |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 1800 | } |
| 1801 | |
Peter Zijlstra | 21f56ffe | 2020-11-17 18:19:32 -0500 | [diff] [blame] | 1802 | static struct task_struct *pick_task_rt(struct rq *rq) |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 1803 | { |
Peter Zijlstra | 606dba2 | 2012-02-11 06:05:00 +0100 | [diff] [blame] | 1804 | struct task_struct *p; |
Peter Zijlstra | 606dba2 | 2012-02-11 06:05:00 +0100 | [diff] [blame] | 1805 | |
Peter Zijlstra | 6e2df05 | 2019-11-08 11:11:52 +0100 | [diff] [blame] | 1806 | if (!sched_rt_runnable(rq)) |
Peter Zijlstra | 606dba2 | 2012-02-11 06:05:00 +0100 | [diff] [blame] | 1807 | return NULL; |
| 1808 | |
Peter Zijlstra | 606dba2 | 2012-02-11 06:05:00 +0100 | [diff] [blame] | 1809 | p = _pick_next_task_rt(rq); |
Peter Zijlstra | 21f56ffe | 2020-11-17 18:19:32 -0500 | [diff] [blame] | 1810 | |
| 1811 | return p; |
| 1812 | } |
| 1813 | |
| 1814 | static struct task_struct *pick_next_task_rt(struct rq *rq) |
| 1815 | { |
| 1816 | struct task_struct *p = pick_task_rt(rq); |
| 1817 | |
| 1818 | if (p) |
| 1819 | set_next_task_rt(rq, p, true); |
| 1820 | |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1821 | return p; |
| 1822 | } |
| 1823 | |
Peter Zijlstra | 6e2df05 | 2019-11-08 11:11:52 +0100 | [diff] [blame] | 1824 | 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] | 1825 | { |
Yafang Shao | 57a5c2d | 2021-09-05 14:35:45 +0000 | [diff] [blame] | 1826 | struct sched_rt_entity *rt_se = &p->rt; |
| 1827 | struct rt_rq *rt_rq = &rq->rt; |
| 1828 | |
| 1829 | if (on_rt_rq(&p->rt)) |
| 1830 | update_stats_wait_start_rt(rt_rq, rt_se); |
| 1831 | |
Ingo Molnar | f1e14ef | 2007-08-09 11:16:48 +0200 | [diff] [blame] | 1832 | update_curr_rt(rq); |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 1833 | |
Vincent Guittot | 2312729 | 2019-01-23 16:26:53 +0100 | [diff] [blame] | 1834 | update_rt_rq_load_avg(rq_clock_pelt(rq), rq, 1); |
Vincent Guittot | 371bf42 | 2018-06-28 17:45:05 +0200 | [diff] [blame] | 1835 | |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 1836 | /* |
| 1837 | * The previous task needs to be made eligible for pushing |
| 1838 | * if it is still active |
| 1839 | */ |
Ingo Molnar | 4b53a34 | 2017-02-05 15:41:03 +0100 | [diff] [blame] | 1840 | if (on_rt_rq(&p->rt) && p->nr_cpus_allowed > 1) |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 1841 | enqueue_pushable_task(rq, p); |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1842 | } |
| 1843 | |
Peter Williams | 681f3e6 | 2007-10-24 18:23:51 +0200 | [diff] [blame] | 1844 | #ifdef CONFIG_SMP |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 1845 | |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1846 | /* Only try algorithms three times */ |
| 1847 | #define RT_MAX_TRIES 3 |
| 1848 | |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1849 | static int pick_rt_task(struct rq *rq, struct task_struct *p, int cpu) |
| 1850 | { |
Peter Zijlstra | 0b9d46f | 2022-09-06 12:33:04 +0200 | [diff] [blame] | 1851 | if (!task_on_cpu(rq, p) && |
Peter Zijlstra | 95158a8 | 2020-10-01 16:05:39 +0200 | [diff] [blame] | 1852 | cpumask_test_cpu(cpu, &p->cpus_mask)) |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1853 | return 1; |
Ingo Molnar | 97fb7a0 | 2018-03-03 14:01:12 +0100 | [diff] [blame] | 1854 | |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 1855 | return 0; |
| 1856 | } |
| 1857 | |
Kirill Tkhai | e23ee74 | 2013-06-07 15:37:43 -0400 | [diff] [blame] | 1858 | /* |
| 1859 | * Return the highest pushable rq's task, which is suitable to be executed |
Ingo Molnar | 97fb7a0 | 2018-03-03 14:01:12 +0100 | [diff] [blame] | 1860 | * on the CPU, NULL otherwise |
Kirill Tkhai | e23ee74 | 2013-06-07 15:37:43 -0400 | [diff] [blame] | 1861 | */ |
| 1862 | static struct task_struct *pick_highest_pushable_task(struct rq *rq, int cpu) |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1863 | { |
Kirill Tkhai | e23ee74 | 2013-06-07 15:37:43 -0400 | [diff] [blame] | 1864 | struct plist_head *head = &rq->rt.pushable_tasks; |
| 1865 | struct task_struct *p; |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1866 | |
Kirill Tkhai | e23ee74 | 2013-06-07 15:37:43 -0400 | [diff] [blame] | 1867 | if (!has_pushable_tasks(rq)) |
| 1868 | return NULL; |
Peter Zijlstra | 3d07467 | 2010-03-10 17:07:24 +0100 | [diff] [blame] | 1869 | |
Kirill Tkhai | e23ee74 | 2013-06-07 15:37:43 -0400 | [diff] [blame] | 1870 | plist_for_each_entry(p, head, pushable_tasks) { |
| 1871 | if (pick_rt_task(rq, p, cpu)) |
| 1872 | return p; |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1873 | } |
| 1874 | |
Kirill Tkhai | e23ee74 | 2013-06-07 15:37:43 -0400 | [diff] [blame] | 1875 | return NULL; |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1876 | } |
| 1877 | |
Rusty Russell | 0e3900e | 2008-11-25 02:35:13 +1030 | [diff] [blame] | 1878 | static DEFINE_PER_CPU(cpumask_var_t, local_cpu_mask); |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1879 | |
Gregory Haskins | 6e1254d | 2008-01-25 21:08:11 +0100 | [diff] [blame] | 1880 | static int find_lowest_rq(struct task_struct *task) |
| 1881 | { |
| 1882 | struct sched_domain *sd; |
Christoph Lameter | 4ba2968 | 2014-08-26 19:12:21 -0500 | [diff] [blame] | 1883 | struct cpumask *lowest_mask = this_cpu_cpumask_var_ptr(local_cpu_mask); |
Gregory Haskins | 6e1254d | 2008-01-25 21:08:11 +0100 | [diff] [blame] | 1884 | int this_cpu = smp_processor_id(); |
| 1885 | int cpu = task_cpu(task); |
Qais Yousef | a1bd02e | 2020-03-02 13:27:18 +0000 | [diff] [blame] | 1886 | int ret; |
Gregory Haskins | 6e1254d | 2008-01-25 21:08:11 +0100 | [diff] [blame] | 1887 | |
Steven Rostedt | 0da938c | 2011-06-14 18:36:25 -0400 | [diff] [blame] | 1888 | /* Make sure the mask is initialized first */ |
| 1889 | if (unlikely(!lowest_mask)) |
| 1890 | return -1; |
| 1891 | |
Ingo Molnar | 4b53a34 | 2017-02-05 15:41:03 +0100 | [diff] [blame] | 1892 | if (task->nr_cpus_allowed == 1) |
Gregory Haskins | 6e0534f | 2008-05-12 21:21:01 +0200 | [diff] [blame] | 1893 | return -1; /* No other targets possible */ |
| 1894 | |
Qais Yousef | a1bd02e | 2020-03-02 13:27:18 +0000 | [diff] [blame] | 1895 | /* |
| 1896 | * If we're on asym system ensure we consider the different capacities |
| 1897 | * of the CPUs when searching for the lowest_mask. |
| 1898 | */ |
Dietmar Eggemann | 740cf8a | 2022-07-29 13:13:03 +0200 | [diff] [blame] | 1899 | if (sched_asym_cpucap_active()) { |
Qais Yousef | a1bd02e | 2020-03-02 13:27:18 +0000 | [diff] [blame] | 1900 | |
| 1901 | ret = cpupri_find_fitness(&task_rq(task)->rd->cpupri, |
| 1902 | task, lowest_mask, |
| 1903 | rt_task_fits_capacity); |
| 1904 | } else { |
| 1905 | |
| 1906 | ret = cpupri_find(&task_rq(task)->rd->cpupri, |
| 1907 | task, lowest_mask); |
| 1908 | } |
| 1909 | |
| 1910 | if (!ret) |
Gregory Haskins | 06f90db | 2008-01-25 21:08:13 +0100 | [diff] [blame] | 1911 | return -1; /* No targets found */ |
| 1912 | |
| 1913 | /* |
Ingo Molnar | 97fb7a0 | 2018-03-03 14:01:12 +0100 | [diff] [blame] | 1914 | * At this point we have built a mask of CPUs representing the |
Gregory Haskins | 6e1254d | 2008-01-25 21:08:11 +0100 | [diff] [blame] | 1915 | * lowest priority tasks in the system. Now we want to elect |
| 1916 | * the best one based on our affinity and topology. |
| 1917 | * |
Ingo Molnar | 97fb7a0 | 2018-03-03 14:01:12 +0100 | [diff] [blame] | 1918 | * We prioritize the last CPU that the task executed on since |
Gregory Haskins | 6e1254d | 2008-01-25 21:08:11 +0100 | [diff] [blame] | 1919 | * it is most likely cache-hot in that location. |
| 1920 | */ |
Rusty Russell | 96f874e2 | 2008-11-25 02:35:14 +1030 | [diff] [blame] | 1921 | if (cpumask_test_cpu(cpu, lowest_mask)) |
Gregory Haskins | 6e1254d | 2008-01-25 21:08:11 +0100 | [diff] [blame] | 1922 | return cpu; |
| 1923 | |
| 1924 | /* |
| 1925 | * Otherwise, we consult the sched_domains span maps to figure |
Ingo Molnar | 97fb7a0 | 2018-03-03 14:01:12 +0100 | [diff] [blame] | 1926 | * out which CPU is logically closest to our hot cache data. |
Gregory Haskins | 6e1254d | 2008-01-25 21:08:11 +0100 | [diff] [blame] | 1927 | */ |
Rusty Russell | e2c8806 | 2009-11-03 14:53:15 +1030 | [diff] [blame] | 1928 | if (!cpumask_test_cpu(this_cpu, lowest_mask)) |
| 1929 | this_cpu = -1; /* Skip this_cpu opt if not among lowest */ |
Gregory Haskins | 6e1254d | 2008-01-25 21:08:11 +0100 | [diff] [blame] | 1930 | |
Xiaotian Feng | cd4ae6a | 2011-04-22 18:53:54 +0800 | [diff] [blame] | 1931 | rcu_read_lock(); |
Rusty Russell | e2c8806 | 2009-11-03 14:53:15 +1030 | [diff] [blame] | 1932 | for_each_domain(cpu, sd) { |
| 1933 | if (sd->flags & SD_WAKE_AFFINE) { |
| 1934 | int best_cpu; |
Gregory Haskins | 6e1254d | 2008-01-25 21:08:11 +0100 | [diff] [blame] | 1935 | |
Rusty Russell | e2c8806 | 2009-11-03 14:53:15 +1030 | [diff] [blame] | 1936 | /* |
| 1937 | * "this_cpu" is cheaper to preempt than a |
| 1938 | * remote processor. |
| 1939 | */ |
| 1940 | if (this_cpu != -1 && |
Xiaotian Feng | cd4ae6a | 2011-04-22 18:53:54 +0800 | [diff] [blame] | 1941 | cpumask_test_cpu(this_cpu, sched_domain_span(sd))) { |
| 1942 | rcu_read_unlock(); |
Rusty Russell | e2c8806 | 2009-11-03 14:53:15 +1030 | [diff] [blame] | 1943 | return this_cpu; |
Xiaotian Feng | cd4ae6a | 2011-04-22 18:53:54 +0800 | [diff] [blame] | 1944 | } |
Gregory Haskins | 6e1254d | 2008-01-25 21:08:11 +0100 | [diff] [blame] | 1945 | |
Peter Zijlstra | 14e292f8 | 2020-10-01 15:54:14 +0200 | [diff] [blame] | 1946 | best_cpu = cpumask_any_and_distribute(lowest_mask, |
| 1947 | sched_domain_span(sd)); |
Xiaotian Feng | cd4ae6a | 2011-04-22 18:53:54 +0800 | [diff] [blame] | 1948 | if (best_cpu < nr_cpu_ids) { |
| 1949 | rcu_read_unlock(); |
Rusty Russell | e2c8806 | 2009-11-03 14:53:15 +1030 | [diff] [blame] | 1950 | return best_cpu; |
Xiaotian Feng | cd4ae6a | 2011-04-22 18:53:54 +0800 | [diff] [blame] | 1951 | } |
Gregory Haskins | 6e1254d | 2008-01-25 21:08:11 +0100 | [diff] [blame] | 1952 | } |
| 1953 | } |
Xiaotian Feng | cd4ae6a | 2011-04-22 18:53:54 +0800 | [diff] [blame] | 1954 | rcu_read_unlock(); |
Gregory Haskins | 6e1254d | 2008-01-25 21:08:11 +0100 | [diff] [blame] | 1955 | |
| 1956 | /* |
| 1957 | * And finally, if there were no matches within the domains |
| 1958 | * just give the caller *something* to work with from the compatible |
| 1959 | * locations. |
| 1960 | */ |
Rusty Russell | e2c8806 | 2009-11-03 14:53:15 +1030 | [diff] [blame] | 1961 | if (this_cpu != -1) |
| 1962 | return this_cpu; |
| 1963 | |
Peter Zijlstra | 14e292f8 | 2020-10-01 15:54:14 +0200 | [diff] [blame] | 1964 | cpu = cpumask_any_distribute(lowest_mask); |
Rusty Russell | e2c8806 | 2009-11-03 14:53:15 +1030 | [diff] [blame] | 1965 | if (cpu < nr_cpu_ids) |
| 1966 | return cpu; |
Ingo Molnar | 97fb7a0 | 2018-03-03 14:01:12 +0100 | [diff] [blame] | 1967 | |
Rusty Russell | e2c8806 | 2009-11-03 14:53:15 +1030 | [diff] [blame] | 1968 | return -1; |
Gregory Haskins | 07b4032 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 1969 | } |
| 1970 | |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1971 | /* Will lock the rq it finds */ |
Ingo Molnar | 4df64c0 | 2008-01-25 21:08:15 +0100 | [diff] [blame] | 1972 | 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] | 1973 | { |
| 1974 | struct rq *lowest_rq = NULL; |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1975 | int tries; |
Ingo Molnar | 4df64c0 | 2008-01-25 21:08:15 +0100 | [diff] [blame] | 1976 | int cpu; |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1977 | |
| 1978 | for (tries = 0; tries < RT_MAX_TRIES; tries++) { |
Gregory Haskins | 07b4032 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 1979 | cpu = find_lowest_rq(task); |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1980 | |
Gregory Haskins | 2de0b46 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 1981 | if ((cpu == -1) || (cpu == rq->cpu)) |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1982 | break; |
| 1983 | |
Gregory Haskins | 07b4032 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 1984 | lowest_rq = cpu_rq(cpu); |
| 1985 | |
Tim Chen | 80e3d87 | 2014-12-12 15:38:12 -0800 | [diff] [blame] | 1986 | if (lowest_rq->rt.highest_prio.curr <= task->prio) { |
| 1987 | /* |
| 1988 | * Target rq has tasks of equal or higher priority, |
| 1989 | * retrying does not release any lock and is unlikely |
| 1990 | * to yield a different result. |
| 1991 | */ |
| 1992 | lowest_rq = NULL; |
| 1993 | break; |
| 1994 | } |
| 1995 | |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1996 | /* if the prio of this runqueue changed, try again */ |
Gregory Haskins | 07b4032 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 1997 | if (double_lock_balance(rq, lowest_rq)) { |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 1998 | /* |
| 1999 | * We had to unlock the run queue. In |
| 2000 | * the mean time, task could have |
| 2001 | * migrated already or had its affinity changed. |
| 2002 | * Also make sure that it wasn't scheduled on its rq. |
Schspa Shi | feffe5b | 2022-08-29 01:03:02 +0800 | [diff] [blame] | 2003 | * It is possible the task was scheduled, set |
| 2004 | * "migrate_disabled" and then got preempted, so we must |
| 2005 | * check the task migration disable flag here too. |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 2006 | */ |
Gregory Haskins | 07b4032 | 2008-01-25 21:08:10 +0100 | [diff] [blame] | 2007 | if (unlikely(task_rq(task) != rq || |
Peter Zijlstra | 95158a8 | 2020-10-01 16:05:39 +0200 | [diff] [blame] | 2008 | !cpumask_test_cpu(lowest_rq->cpu, &task->cpus_mask) || |
Peter Zijlstra | 0b9d46f | 2022-09-06 12:33:04 +0200 | [diff] [blame] | 2009 | task_on_cpu(rq, task) || |
Xunlei Pang | 13b5ab0 | 2016-05-09 12:11:31 +0800 | [diff] [blame] | 2010 | !rt_task(task) || |
Schspa Shi | feffe5b | 2022-08-29 01:03:02 +0800 | [diff] [blame] | 2011 | is_migration_disabled(task) || |
Kirill Tkhai | da0c1e6 | 2014-08-20 13:47:32 +0400 | [diff] [blame] | 2012 | !task_on_rq_queued(task))) { |
Ingo Molnar | 4df64c0 | 2008-01-25 21:08:15 +0100 | [diff] [blame] | 2013 | |
Peter Zijlstra | 7f1b439 | 2012-05-17 21:19:46 +0200 | [diff] [blame] | 2014 | double_unlock_balance(rq, lowest_rq); |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 2015 | lowest_rq = NULL; |
| 2016 | break; |
| 2017 | } |
| 2018 | } |
| 2019 | |
| 2020 | /* If this rq is still suitable use it. */ |
Gregory Haskins | e864c49 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 2021 | if (lowest_rq->rt.highest_prio.curr > task->prio) |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 2022 | break; |
| 2023 | |
| 2024 | /* try again */ |
Peter Zijlstra | 1b12bbc | 2008-08-11 09:30:22 +0200 | [diff] [blame] | 2025 | double_unlock_balance(rq, lowest_rq); |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 2026 | lowest_rq = NULL; |
| 2027 | } |
| 2028 | |
| 2029 | return lowest_rq; |
| 2030 | } |
| 2031 | |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 2032 | static struct task_struct *pick_next_pushable_task(struct rq *rq) |
| 2033 | { |
| 2034 | struct task_struct *p; |
| 2035 | |
| 2036 | if (!has_pushable_tasks(rq)) |
| 2037 | return NULL; |
| 2038 | |
| 2039 | p = plist_first_entry(&rq->rt.pushable_tasks, |
| 2040 | struct task_struct, pushable_tasks); |
| 2041 | |
| 2042 | BUG_ON(rq->cpu != task_cpu(p)); |
| 2043 | BUG_ON(task_current(rq, p)); |
Ingo Molnar | 4b53a34 | 2017-02-05 15:41:03 +0100 | [diff] [blame] | 2044 | BUG_ON(p->nr_cpus_allowed <= 1); |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 2045 | |
Kirill Tkhai | da0c1e6 | 2014-08-20 13:47:32 +0400 | [diff] [blame] | 2046 | BUG_ON(!task_on_rq_queued(p)); |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 2047 | BUG_ON(!rt_task(p)); |
| 2048 | |
| 2049 | return p; |
| 2050 | } |
| 2051 | |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 2052 | /* |
| 2053 | * If the current CPU has more than one RT task, see if the non |
| 2054 | * running task can migrate over to a CPU that is running a task |
| 2055 | * of lesser priority. |
| 2056 | */ |
Peter Zijlstra | a7c8155 | 2020-09-28 17:06:07 +0200 | [diff] [blame] | 2057 | static int push_rt_task(struct rq *rq, bool pull) |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 2058 | { |
| 2059 | struct task_struct *next_task; |
| 2060 | struct rq *lowest_rq; |
Hillf Danton | 311e800 | 2011-06-16 21:55:20 -0400 | [diff] [blame] | 2061 | int ret = 0; |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 2062 | |
Gregory Haskins | a22d7fc | 2008-01-25 21:08:12 +0100 | [diff] [blame] | 2063 | if (!rq->rt.overloaded) |
| 2064 | return 0; |
| 2065 | |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 2066 | next_task = pick_next_pushable_task(rq); |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 2067 | if (!next_task) |
| 2068 | return 0; |
| 2069 | |
Peter Zijlstra | 4924627 | 2010-10-17 21:46:10 +0200 | [diff] [blame] | 2070 | retry: |
Valentin Schneider | 49bef33 | 2022-01-27 15:40:59 +0000 | [diff] [blame] | 2071 | /* |
| 2072 | * It's possible that the next_task slipped in of |
| 2073 | * higher priority than current. If that's the case |
| 2074 | * just reschedule current. |
| 2075 | */ |
| 2076 | if (unlikely(next_task->prio < rq->curr->prio)) { |
| 2077 | resched_curr(rq); |
| 2078 | return 0; |
| 2079 | } |
| 2080 | |
Peter Zijlstra | a7c8155 | 2020-09-28 17:06:07 +0200 | [diff] [blame] | 2081 | if (is_migration_disabled(next_task)) { |
| 2082 | struct task_struct *push_task = NULL; |
| 2083 | int cpu; |
| 2084 | |
| 2085 | if (!pull || rq->push_busy) |
| 2086 | return 0; |
| 2087 | |
Valentin Schneider | 49bef33 | 2022-01-27 15:40:59 +0000 | [diff] [blame] | 2088 | /* |
| 2089 | * Invoking find_lowest_rq() on anything but an RT task doesn't |
| 2090 | * make sense. Per the above priority check, curr has to |
| 2091 | * be of higher priority than next_task, so no need to |
| 2092 | * reschedule when bailing out. |
| 2093 | * |
| 2094 | * Note that the stoppers are masqueraded as SCHED_FIFO |
| 2095 | * (cf. sched_set_stop_task()), so we can't rely on rt_task(). |
| 2096 | */ |
| 2097 | if (rq->curr->sched_class != &rt_sched_class) |
| 2098 | return 0; |
| 2099 | |
Peter Zijlstra | a7c8155 | 2020-09-28 17:06:07 +0200 | [diff] [blame] | 2100 | cpu = find_lowest_rq(rq->curr); |
| 2101 | if (cpu == -1 || cpu == rq->cpu) |
| 2102 | return 0; |
| 2103 | |
| 2104 | /* |
| 2105 | * Given we found a CPU with lower priority than @next_task, |
| 2106 | * therefore it should be running. However we cannot migrate it |
| 2107 | * to this other CPU, instead attempt to push the current |
| 2108 | * running task on this CPU away. |
| 2109 | */ |
| 2110 | push_task = get_push_task(rq); |
| 2111 | if (push_task) { |
Peter Zijlstra | 5cb9eaa | 2020-11-17 18:19:31 -0500 | [diff] [blame] | 2112 | raw_spin_rq_unlock(rq); |
Peter Zijlstra | a7c8155 | 2020-09-28 17:06:07 +0200 | [diff] [blame] | 2113 | stop_one_cpu_nowait(rq->cpu, push_cpu_stop, |
| 2114 | push_task, &rq->push_work); |
Peter Zijlstra | 5cb9eaa | 2020-11-17 18:19:31 -0500 | [diff] [blame] | 2115 | raw_spin_rq_lock(rq); |
Peter Zijlstra | a7c8155 | 2020-09-28 17:06:07 +0200 | [diff] [blame] | 2116 | } |
| 2117 | |
| 2118 | return 0; |
| 2119 | } |
| 2120 | |
Yangtao Li | 9ebc605 | 2018-11-03 13:26:02 -0400 | [diff] [blame] | 2121 | if (WARN_ON(next_task == rq->curr)) |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 2122 | return 0; |
| 2123 | |
Gregory Haskins | 697f0a4 | 2008-01-25 21:08:09 +0100 | [diff] [blame] | 2124 | /* We might release rq lock */ |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 2125 | get_task_struct(next_task); |
| 2126 | |
| 2127 | /* find_lock_lowest_rq locks the rq if found */ |
Gregory Haskins | 697f0a4 | 2008-01-25 21:08:09 +0100 | [diff] [blame] | 2128 | lowest_rq = find_lock_lowest_rq(next_task, rq); |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 2129 | if (!lowest_rq) { |
| 2130 | struct task_struct *task; |
| 2131 | /* |
Hillf Danton | 311e800 | 2011-06-16 21:55:20 -0400 | [diff] [blame] | 2132 | * find_lock_lowest_rq releases rq->lock |
Gregory Haskins | 1563513 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 2133 | * so it is possible that next_task has migrated. |
| 2134 | * |
| 2135 | * We need to make sure that the task is still on the same |
| 2136 | * run-queue and is also still the next task eligible for |
| 2137 | * pushing. |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 2138 | */ |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 2139 | task = pick_next_pushable_task(rq); |
Byungchul Park | de16b91 | 2017-05-12 10:05:43 +0900 | [diff] [blame] | 2140 | if (task == next_task) { |
Gregory Haskins | 1563513 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 2141 | /* |
Hillf Danton | 311e800 | 2011-06-16 21:55:20 -0400 | [diff] [blame] | 2142 | * The task hasn't migrated, and is still the next |
| 2143 | * eligible task, but we failed to find a run-queue |
| 2144 | * to push it to. Do not retry in this case, since |
Ingo Molnar | 97fb7a0 | 2018-03-03 14:01:12 +0100 | [diff] [blame] | 2145 | * other CPUs will pull from us when ready. |
Gregory Haskins | 1563513 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 2146 | */ |
Gregory Haskins | 1563513 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 2147 | goto out; |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 2148 | } |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 2149 | |
Gregory Haskins | 1563513 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 2150 | if (!task) |
| 2151 | /* No more tasks, just exit */ |
| 2152 | goto out; |
| 2153 | |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 2154 | /* |
Gregory Haskins | 1563513 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 2155 | * Something has shifted, try again. |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 2156 | */ |
Gregory Haskins | 1563513 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 2157 | put_task_struct(next_task); |
| 2158 | next_task = task; |
| 2159 | goto retry; |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 2160 | } |
| 2161 | |
Gregory Haskins | 697f0a4 | 2008-01-25 21:08:09 +0100 | [diff] [blame] | 2162 | deactivate_task(rq, next_task, 0); |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 2163 | set_task_cpu(next_task, lowest_rq->cpu); |
| 2164 | activate_task(lowest_rq, next_task, 0); |
Peter Zijlstra | a7c8155 | 2020-09-28 17:06:07 +0200 | [diff] [blame] | 2165 | resched_curr(lowest_rq); |
Hillf Danton | 311e800 | 2011-06-16 21:55:20 -0400 | [diff] [blame] | 2166 | ret = 1; |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 2167 | |
Peter Zijlstra | 1b12bbc | 2008-08-11 09:30:22 +0200 | [diff] [blame] | 2168 | double_unlock_balance(rq, lowest_rq); |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 2169 | out: |
| 2170 | put_task_struct(next_task); |
| 2171 | |
Hillf Danton | 311e800 | 2011-06-16 21:55:20 -0400 | [diff] [blame] | 2172 | return ret; |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 2173 | } |
| 2174 | |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 2175 | static void push_rt_tasks(struct rq *rq) |
| 2176 | { |
| 2177 | /* push_rt_task will return true if it moved an RT */ |
Peter Zijlstra | a7c8155 | 2020-09-28 17:06:07 +0200 | [diff] [blame] | 2178 | while (push_rt_task(rq, false)) |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 2179 | ; |
| 2180 | } |
| 2181 | |
Steven Rostedt | b6366f0 | 2015-03-18 14:49:46 -0400 | [diff] [blame] | 2182 | #ifdef HAVE_RT_PUSH_IPI |
Steven Rostedt | b6366f0 | 2015-03-18 14:49:46 -0400 | [diff] [blame] | 2183 | |
Steven Rostedt (VMware) | 3e777f9 | 2017-02-28 15:50:30 -0500 | [diff] [blame] | 2184 | /* |
| 2185 | * When a high priority task schedules out from a CPU and a lower priority |
| 2186 | * task is scheduled in, a check is made to see if there's any RT tasks |
| 2187 | * on other CPUs that are waiting to run because a higher priority RT task |
| 2188 | * is currently running on its CPU. In this case, the CPU with multiple RT |
| 2189 | * tasks queued on it (overloaded) needs to be notified that a CPU has opened |
| 2190 | * up that may be able to run one of its non-running queued RT tasks. |
| 2191 | * |
Steven Rostedt (Red Hat) | 4bdced5 | 2017-10-06 14:05:04 -0400 | [diff] [blame] | 2192 | * All CPUs with overloaded RT tasks need to be notified as there is currently |
| 2193 | * no way to know which of these CPUs have the highest priority task waiting |
| 2194 | * to run. Instead of trying to take a spinlock on each of these CPUs, |
| 2195 | * which has shown to cause large latency when done on machines with many |
| 2196 | * CPUs, sending an IPI to the CPUs to have them push off the overloaded |
| 2197 | * RT tasks waiting to run. |
Steven Rostedt (VMware) | 3e777f9 | 2017-02-28 15:50:30 -0500 | [diff] [blame] | 2198 | * |
Steven Rostedt (Red Hat) | 4bdced5 | 2017-10-06 14:05:04 -0400 | [diff] [blame] | 2199 | * Just sending an IPI to each of the CPUs is also an issue, as on large |
| 2200 | * count CPU machines, this can cause an IPI storm on a CPU, especially |
| 2201 | * if its the only CPU with multiple RT tasks queued, and a large number |
| 2202 | * of CPUs scheduling a lower priority task at the same time. |
Steven Rostedt (VMware) | 3e777f9 | 2017-02-28 15:50:30 -0500 | [diff] [blame] | 2203 | * |
Steven Rostedt (Red Hat) | 4bdced5 | 2017-10-06 14:05:04 -0400 | [diff] [blame] | 2204 | * Each root domain has its own irq work function that can iterate over |
| 2205 | * all CPUs with RT overloaded tasks. Since all CPUs with overloaded RT |
Ingo Molnar | 3b03706 | 2021-03-18 13:38:50 +0100 | [diff] [blame] | 2206 | * task must be checked if there's one or many CPUs that are lowering |
Steven Rostedt (Red Hat) | 4bdced5 | 2017-10-06 14:05:04 -0400 | [diff] [blame] | 2207 | * their priority, there's a single irq work iterator that will try to |
| 2208 | * push off RT tasks that are waiting to run. |
Steven Rostedt (VMware) | 3e777f9 | 2017-02-28 15:50:30 -0500 | [diff] [blame] | 2209 | * |
Steven Rostedt (Red Hat) | 4bdced5 | 2017-10-06 14:05:04 -0400 | [diff] [blame] | 2210 | * When a CPU schedules a lower priority task, it will kick off the |
| 2211 | * irq work iterator that will jump to each CPU with overloaded RT tasks. |
| 2212 | * As it only takes the first CPU that schedules a lower priority task |
| 2213 | * to start the process, the rto_start variable is incremented and if |
| 2214 | * the atomic result is one, then that CPU will try to take the rto_lock. |
| 2215 | * This prevents high contention on the lock as the process handles all |
| 2216 | * CPUs scheduling lower priority tasks. |
Steven Rostedt (VMware) | 3e777f9 | 2017-02-28 15:50:30 -0500 | [diff] [blame] | 2217 | * |
Steven Rostedt (Red Hat) | 4bdced5 | 2017-10-06 14:05:04 -0400 | [diff] [blame] | 2218 | * All CPUs that are scheduling a lower priority task will increment the |
| 2219 | * rt_loop_next variable. This will make sure that the irq work iterator |
| 2220 | * checks all RT overloaded CPUs whenever a CPU schedules a new lower |
| 2221 | * priority task, even if the iterator is in the middle of a scan. Incrementing |
| 2222 | * the rt_loop_next will cause the iterator to perform another scan. |
Steven Rostedt (VMware) | 3e777f9 | 2017-02-28 15:50:30 -0500 | [diff] [blame] | 2223 | * |
Steven Rostedt (VMware) | 3e777f9 | 2017-02-28 15:50:30 -0500 | [diff] [blame] | 2224 | */ |
Steven Rostedt (VMware) | ad0f1d9 | 2018-01-23 20:45:37 -0500 | [diff] [blame] | 2225 | static int rto_next_cpu(struct root_domain *rd) |
Steven Rostedt | b6366f0 | 2015-03-18 14:49:46 -0400 | [diff] [blame] | 2226 | { |
Steven Rostedt (Red Hat) | 4bdced5 | 2017-10-06 14:05:04 -0400 | [diff] [blame] | 2227 | int next; |
Steven Rostedt | b6366f0 | 2015-03-18 14:49:46 -0400 | [diff] [blame] | 2228 | int cpu; |
| 2229 | |
Steven Rostedt (Red Hat) | 4bdced5 | 2017-10-06 14:05:04 -0400 | [diff] [blame] | 2230 | /* |
| 2231 | * When starting the IPI RT pushing, the rto_cpu is set to -1, |
| 2232 | * rt_next_cpu() will simply return the first CPU found in |
| 2233 | * the rto_mask. |
| 2234 | * |
Ingo Molnar | 97fb7a0 | 2018-03-03 14:01:12 +0100 | [diff] [blame] | 2235 | * If rto_next_cpu() is called with rto_cpu is a valid CPU, it |
Steven Rostedt (Red Hat) | 4bdced5 | 2017-10-06 14:05:04 -0400 | [diff] [blame] | 2236 | * will return the next CPU found in the rto_mask. |
| 2237 | * |
| 2238 | * If there are no more CPUs left in the rto_mask, then a check is made |
| 2239 | * against rto_loop and rto_loop_next. rto_loop is only updated with |
| 2240 | * the rto_lock held, but any CPU may increment the rto_loop_next |
| 2241 | * without any locking. |
| 2242 | */ |
| 2243 | for (;;) { |
| 2244 | |
| 2245 | /* When rto_cpu is -1 this acts like cpumask_first() */ |
| 2246 | cpu = cpumask_next(rd->rto_cpu, rd->rto_mask); |
| 2247 | |
| 2248 | rd->rto_cpu = cpu; |
| 2249 | |
| 2250 | if (cpu < nr_cpu_ids) |
| 2251 | return cpu; |
| 2252 | |
| 2253 | rd->rto_cpu = -1; |
| 2254 | |
| 2255 | /* |
| 2256 | * ACQUIRE ensures we see the @rto_mask changes |
| 2257 | * made prior to the @next value observed. |
| 2258 | * |
| 2259 | * Matches WMB in rt_set_overload(). |
| 2260 | */ |
| 2261 | next = atomic_read_acquire(&rd->rto_loop_next); |
| 2262 | |
| 2263 | if (rd->rto_loop == next) |
| 2264 | break; |
| 2265 | |
| 2266 | rd->rto_loop = next; |
Steven Rostedt | b6366f0 | 2015-03-18 14:49:46 -0400 | [diff] [blame] | 2267 | } |
| 2268 | |
Steven Rostedt (Red Hat) | 4bdced5 | 2017-10-06 14:05:04 -0400 | [diff] [blame] | 2269 | return -1; |
| 2270 | } |
Steven Rostedt | b6366f0 | 2015-03-18 14:49:46 -0400 | [diff] [blame] | 2271 | |
Steven Rostedt (Red Hat) | 4bdced5 | 2017-10-06 14:05:04 -0400 | [diff] [blame] | 2272 | static inline bool rto_start_trylock(atomic_t *v) |
| 2273 | { |
| 2274 | return !atomic_cmpxchg_acquire(v, 0, 1); |
| 2275 | } |
| 2276 | |
| 2277 | static inline void rto_start_unlock(atomic_t *v) |
| 2278 | { |
| 2279 | atomic_set_release(v, 0); |
| 2280 | } |
| 2281 | |
| 2282 | static void tell_cpu_to_push(struct rq *rq) |
| 2283 | { |
| 2284 | int cpu = -1; |
| 2285 | |
| 2286 | /* Keep the loop going if the IPI is currently active */ |
| 2287 | atomic_inc(&rq->rd->rto_loop_next); |
| 2288 | |
| 2289 | /* Only one CPU can initiate a loop at a time */ |
| 2290 | if (!rto_start_trylock(&rq->rd->rto_loop_start)) |
Steven Rostedt | b6366f0 | 2015-03-18 14:49:46 -0400 | [diff] [blame] | 2291 | return; |
| 2292 | |
Steven Rostedt (Red Hat) | 4bdced5 | 2017-10-06 14:05:04 -0400 | [diff] [blame] | 2293 | raw_spin_lock(&rq->rd->rto_lock); |
Steven Rostedt | b6366f0 | 2015-03-18 14:49:46 -0400 | [diff] [blame] | 2294 | |
Steven Rostedt (Red Hat) | 4bdced5 | 2017-10-06 14:05:04 -0400 | [diff] [blame] | 2295 | /* |
Ingo Molnar | 97fb7a0 | 2018-03-03 14:01:12 +0100 | [diff] [blame] | 2296 | * The rto_cpu is updated under the lock, if it has a valid CPU |
Steven Rostedt (Red Hat) | 4bdced5 | 2017-10-06 14:05:04 -0400 | [diff] [blame] | 2297 | * then the IPI is still running and will continue due to the |
| 2298 | * update to loop_next, and nothing needs to be done here. |
| 2299 | * Otherwise it is finishing up and an ipi needs to be sent. |
| 2300 | */ |
| 2301 | if (rq->rd->rto_cpu < 0) |
Steven Rostedt (VMware) | ad0f1d9 | 2018-01-23 20:45:37 -0500 | [diff] [blame] | 2302 | cpu = rto_next_cpu(rq->rd); |
Steven Rostedt (Red Hat) | 4bdced5 | 2017-10-06 14:05:04 -0400 | [diff] [blame] | 2303 | |
| 2304 | raw_spin_unlock(&rq->rd->rto_lock); |
| 2305 | |
| 2306 | rto_start_unlock(&rq->rd->rto_loop_start); |
| 2307 | |
Steven Rostedt (VMware) | 364f566 | 2018-01-23 20:45:38 -0500 | [diff] [blame] | 2308 | if (cpu >= 0) { |
| 2309 | /* Make sure the rd does not get freed while pushing */ |
| 2310 | sched_get_rd(rq->rd); |
Steven Rostedt (Red Hat) | 4bdced5 | 2017-10-06 14:05:04 -0400 | [diff] [blame] | 2311 | irq_work_queue_on(&rq->rd->rto_push_work, cpu); |
Steven Rostedt (VMware) | 364f566 | 2018-01-23 20:45:38 -0500 | [diff] [blame] | 2312 | } |
Steven Rostedt | b6366f0 | 2015-03-18 14:49:46 -0400 | [diff] [blame] | 2313 | } |
| 2314 | |
| 2315 | /* Called from hardirq context */ |
Steven Rostedt (Red Hat) | 4bdced5 | 2017-10-06 14:05:04 -0400 | [diff] [blame] | 2316 | void rto_push_irq_work_func(struct irq_work *work) |
Steven Rostedt | b6366f0 | 2015-03-18 14:49:46 -0400 | [diff] [blame] | 2317 | { |
Steven Rostedt (VMware) | ad0f1d9 | 2018-01-23 20:45:37 -0500 | [diff] [blame] | 2318 | struct root_domain *rd = |
| 2319 | container_of(work, struct root_domain, rto_push_work); |
Steven Rostedt (Red Hat) | 4bdced5 | 2017-10-06 14:05:04 -0400 | [diff] [blame] | 2320 | struct rq *rq; |
Steven Rostedt | b6366f0 | 2015-03-18 14:49:46 -0400 | [diff] [blame] | 2321 | int cpu; |
| 2322 | |
Steven Rostedt (Red Hat) | 4bdced5 | 2017-10-06 14:05:04 -0400 | [diff] [blame] | 2323 | rq = this_rq(); |
Steven Rostedt | b6366f0 | 2015-03-18 14:49:46 -0400 | [diff] [blame] | 2324 | |
Steven Rostedt (Red Hat) | 4bdced5 | 2017-10-06 14:05:04 -0400 | [diff] [blame] | 2325 | /* |
| 2326 | * We do not need to grab the lock to check for has_pushable_tasks. |
| 2327 | * When it gets updated, a check is made if a push is possible. |
| 2328 | */ |
Steven Rostedt | b6366f0 | 2015-03-18 14:49:46 -0400 | [diff] [blame] | 2329 | if (has_pushable_tasks(rq)) { |
Peter Zijlstra | 5cb9eaa | 2020-11-17 18:19:31 -0500 | [diff] [blame] | 2330 | raw_spin_rq_lock(rq); |
Peter Zijlstra | a7c8155 | 2020-09-28 17:06:07 +0200 | [diff] [blame] | 2331 | while (push_rt_task(rq, true)) |
| 2332 | ; |
Peter Zijlstra | 5cb9eaa | 2020-11-17 18:19:31 -0500 | [diff] [blame] | 2333 | raw_spin_rq_unlock(rq); |
Steven Rostedt | b6366f0 | 2015-03-18 14:49:46 -0400 | [diff] [blame] | 2334 | } |
| 2335 | |
Steven Rostedt (VMware) | ad0f1d9 | 2018-01-23 20:45:37 -0500 | [diff] [blame] | 2336 | raw_spin_lock(&rd->rto_lock); |
Steven Rostedt (Red Hat) | 4bdced5 | 2017-10-06 14:05:04 -0400 | [diff] [blame] | 2337 | |
Steven Rostedt | b6366f0 | 2015-03-18 14:49:46 -0400 | [diff] [blame] | 2338 | /* Pass the IPI to the next rt overloaded queue */ |
Steven Rostedt (VMware) | ad0f1d9 | 2018-01-23 20:45:37 -0500 | [diff] [blame] | 2339 | cpu = rto_next_cpu(rd); |
Steven Rostedt | b6366f0 | 2015-03-18 14:49:46 -0400 | [diff] [blame] | 2340 | |
Steven Rostedt (VMware) | ad0f1d9 | 2018-01-23 20:45:37 -0500 | [diff] [blame] | 2341 | raw_spin_unlock(&rd->rto_lock); |
Steven Rostedt | b6366f0 | 2015-03-18 14:49:46 -0400 | [diff] [blame] | 2342 | |
Steven Rostedt (VMware) | 364f566 | 2018-01-23 20:45:38 -0500 | [diff] [blame] | 2343 | if (cpu < 0) { |
| 2344 | sched_put_rd(rd); |
Steven Rostedt | b6366f0 | 2015-03-18 14:49:46 -0400 | [diff] [blame] | 2345 | return; |
Steven Rostedt (VMware) | 364f566 | 2018-01-23 20:45:38 -0500 | [diff] [blame] | 2346 | } |
Steven Rostedt | b6366f0 | 2015-03-18 14:49:46 -0400 | [diff] [blame] | 2347 | |
Steven Rostedt | b6366f0 | 2015-03-18 14:49:46 -0400 | [diff] [blame] | 2348 | /* Try the next RT overloaded CPU */ |
Steven Rostedt (VMware) | ad0f1d9 | 2018-01-23 20:45:37 -0500 | [diff] [blame] | 2349 | irq_work_queue_on(&rd->rto_push_work, cpu); |
Steven Rostedt | b6366f0 | 2015-03-18 14:49:46 -0400 | [diff] [blame] | 2350 | } |
| 2351 | #endif /* HAVE_RT_PUSH_IPI */ |
| 2352 | |
Peter Zijlstra | 8046d68 | 2015-06-11 14:46:40 +0200 | [diff] [blame] | 2353 | static void pull_rt_task(struct rq *this_rq) |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 2354 | { |
Peter Zijlstra | 8046d68 | 2015-06-11 14:46:40 +0200 | [diff] [blame] | 2355 | int this_cpu = this_rq->cpu, cpu; |
| 2356 | bool resched = false; |
Peter Zijlstra | a7c8155 | 2020-09-28 17:06:07 +0200 | [diff] [blame] | 2357 | struct task_struct *p, *push_task; |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 2358 | struct rq *src_rq; |
Steven Rostedt | f73c52a | 2017-12-02 13:04:54 -0500 | [diff] [blame] | 2359 | int rt_overload_count = rt_overloaded(this_rq); |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 2360 | |
Steven Rostedt | f73c52a | 2017-12-02 13:04:54 -0500 | [diff] [blame] | 2361 | if (likely(!rt_overload_count)) |
Peter Zijlstra | 8046d68 | 2015-06-11 14:46:40 +0200 | [diff] [blame] | 2362 | return; |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 2363 | |
Peter Zijlstra | 7c3f2ab | 2013-10-15 12:35:07 +0200 | [diff] [blame] | 2364 | /* |
| 2365 | * Match the barrier from rt_set_overloaded; this guarantees that if we |
| 2366 | * see overloaded we must also see the rto_mask bit. |
| 2367 | */ |
| 2368 | smp_rmb(); |
| 2369 | |
Steven Rostedt | f73c52a | 2017-12-02 13:04:54 -0500 | [diff] [blame] | 2370 | /* If we are the only overloaded CPU do nothing */ |
| 2371 | if (rt_overload_count == 1 && |
| 2372 | cpumask_test_cpu(this_rq->cpu, this_rq->rd->rto_mask)) |
| 2373 | return; |
| 2374 | |
Steven Rostedt | b6366f0 | 2015-03-18 14:49:46 -0400 | [diff] [blame] | 2375 | #ifdef HAVE_RT_PUSH_IPI |
| 2376 | if (sched_feat(RT_PUSH_IPI)) { |
| 2377 | tell_cpu_to_push(this_rq); |
Peter Zijlstra | 8046d68 | 2015-06-11 14:46:40 +0200 | [diff] [blame] | 2378 | return; |
Steven Rostedt | b6366f0 | 2015-03-18 14:49:46 -0400 | [diff] [blame] | 2379 | } |
| 2380 | #endif |
| 2381 | |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 2382 | for_each_cpu(cpu, this_rq->rd->rto_mask) { |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 2383 | if (this_cpu == cpu) |
| 2384 | continue; |
| 2385 | |
| 2386 | src_rq = cpu_rq(cpu); |
Gregory Haskins | 74ab8e4 | 2008-12-29 09:39:50 -0500 | [diff] [blame] | 2387 | |
| 2388 | /* |
| 2389 | * Don't bother taking the src_rq->lock if the next highest |
| 2390 | * task is known to be lower-priority than our current task. |
| 2391 | * This may look racy, but if this value is about to go |
| 2392 | * logically higher, the src_rq will push this task away. |
| 2393 | * And if its going logically lower, we do not care |
| 2394 | */ |
| 2395 | if (src_rq->rt.highest_prio.next >= |
| 2396 | this_rq->rt.highest_prio.curr) |
| 2397 | continue; |
| 2398 | |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 2399 | /* |
| 2400 | * We can potentially drop this_rq's lock in |
| 2401 | * double_lock_balance, and another CPU could |
Gregory Haskins | a872894 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 2402 | * alter this_rq |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 2403 | */ |
Peter Zijlstra | a7c8155 | 2020-09-28 17:06:07 +0200 | [diff] [blame] | 2404 | push_task = NULL; |
Gregory Haskins | a872894 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 2405 | double_lock_balance(this_rq, src_rq); |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 2406 | |
| 2407 | /* |
Kirill Tkhai | e23ee74 | 2013-06-07 15:37:43 -0400 | [diff] [blame] | 2408 | * We can pull only a task, which is pushable |
| 2409 | * on its rq, and no others. |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 2410 | */ |
Kirill Tkhai | e23ee74 | 2013-06-07 15:37:43 -0400 | [diff] [blame] | 2411 | p = pick_highest_pushable_task(src_rq, this_cpu); |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 2412 | |
| 2413 | /* |
| 2414 | * Do we have an RT task that preempts |
| 2415 | * the to-be-scheduled task? |
| 2416 | */ |
Gregory Haskins | a872894 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 2417 | if (p && (p->prio < this_rq->rt.highest_prio.curr)) { |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 2418 | WARN_ON(p == src_rq->curr); |
Kirill Tkhai | da0c1e6 | 2014-08-20 13:47:32 +0400 | [diff] [blame] | 2419 | WARN_ON(!task_on_rq_queued(p)); |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 2420 | |
| 2421 | /* |
| 2422 | * There's a chance that p is higher in priority |
Ingo Molnar | 97fb7a0 | 2018-03-03 14:01:12 +0100 | [diff] [blame] | 2423 | * than what's currently running on its CPU. |
Ingo Molnar | 3b03706 | 2021-03-18 13:38:50 +0100 | [diff] [blame] | 2424 | * This is just that p is waking up and hasn't |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 2425 | * had a chance to schedule. We only pull |
| 2426 | * p if it is lower in priority than the |
Gregory Haskins | a872894 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 2427 | * current task on the run queue |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 2428 | */ |
Gregory Haskins | a872894 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 2429 | if (p->prio < src_rq->curr->prio) |
Mike Galbraith | 614ee1f | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 2430 | goto skip; |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 2431 | |
Peter Zijlstra | a7c8155 | 2020-09-28 17:06:07 +0200 | [diff] [blame] | 2432 | if (is_migration_disabled(p)) { |
| 2433 | push_task = get_push_task(src_rq); |
| 2434 | } else { |
| 2435 | deactivate_task(src_rq, p, 0); |
| 2436 | set_task_cpu(p, this_cpu); |
| 2437 | activate_task(this_rq, p, 0); |
| 2438 | resched = true; |
| 2439 | } |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 2440 | /* |
| 2441 | * We continue with the search, just in |
| 2442 | * case there's an even higher prio task |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 2443 | * in another runqueue. (low likelihood |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 2444 | * but possible) |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 2445 | */ |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 2446 | } |
Peter Zijlstra | 4924627 | 2010-10-17 21:46:10 +0200 | [diff] [blame] | 2447 | skip: |
Peter Zijlstra | 1b12bbc | 2008-08-11 09:30:22 +0200 | [diff] [blame] | 2448 | double_unlock_balance(this_rq, src_rq); |
Peter Zijlstra | a7c8155 | 2020-09-28 17:06:07 +0200 | [diff] [blame] | 2449 | |
| 2450 | if (push_task) { |
Peter Zijlstra | 5cb9eaa | 2020-11-17 18:19:31 -0500 | [diff] [blame] | 2451 | raw_spin_rq_unlock(this_rq); |
Peter Zijlstra | a7c8155 | 2020-09-28 17:06:07 +0200 | [diff] [blame] | 2452 | stop_one_cpu_nowait(src_rq->cpu, push_cpu_stop, |
| 2453 | push_task, &src_rq->push_work); |
Peter Zijlstra | 5cb9eaa | 2020-11-17 18:19:31 -0500 | [diff] [blame] | 2454 | raw_spin_rq_lock(this_rq); |
Peter Zijlstra | a7c8155 | 2020-09-28 17:06:07 +0200 | [diff] [blame] | 2455 | } |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 2456 | } |
| 2457 | |
Peter Zijlstra | 8046d68 | 2015-06-11 14:46:40 +0200 | [diff] [blame] | 2458 | if (resched) |
| 2459 | resched_curr(this_rq); |
Steven Rostedt | f65eda4 | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 2460 | } |
| 2461 | |
Gregory Haskins | 8ae121a | 2008-04-23 07:13:29 -0400 | [diff] [blame] | 2462 | /* |
| 2463 | * If we are not running and we are not going to reschedule soon, we should |
| 2464 | * try to push tasks away now |
| 2465 | */ |
Peter Zijlstra | efbbd05 | 2009-12-16 18:04:40 +0100 | [diff] [blame] | 2466 | static void task_woken_rt(struct rq *rq, struct task_struct *p) |
Steven Rostedt | 4642daf | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 2467 | { |
Peter Zijlstra | 0b9d46f | 2022-09-06 12:33:04 +0200 | [diff] [blame] | 2468 | bool need_to_push = !task_on_cpu(rq, p) && |
Qais Yousef | 804d402 | 2019-10-09 11:46:11 +0100 | [diff] [blame] | 2469 | !test_tsk_need_resched(rq->curr) && |
| 2470 | p->nr_cpus_allowed > 1 && |
| 2471 | (dl_task(rq->curr) || rt_task(rq->curr)) && |
| 2472 | (rq->curr->nr_cpus_allowed < 2 || |
| 2473 | rq->curr->prio <= p->prio); |
| 2474 | |
Qais Yousef | d94a9df | 2020-03-02 13:27:20 +0000 | [diff] [blame] | 2475 | if (need_to_push) |
Steven Rostedt | 4642daf | 2008-01-25 21:08:07 +0100 | [diff] [blame] | 2476 | push_rt_tasks(rq); |
| 2477 | } |
| 2478 | |
Ingo Molnar | bdd7c81 | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 2479 | /* Assumes rq->lock is held */ |
Gregory Haskins | 1f11eb6a | 2008-06-04 15:04:05 -0400 | [diff] [blame] | 2480 | static void rq_online_rt(struct rq *rq) |
Ingo Molnar | bdd7c81 | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 2481 | { |
| 2482 | if (rq->rt.overloaded) |
| 2483 | rt_set_overload(rq); |
Gregory Haskins | 6e0534f | 2008-05-12 21:21:01 +0200 | [diff] [blame] | 2484 | |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 2485 | __enable_runtime(rq); |
| 2486 | |
Gregory Haskins | e864c49 | 2008-12-29 09:39:49 -0500 | [diff] [blame] | 2487 | cpupri_set(&rq->rd->cpupri, rq->cpu, rq->rt.highest_prio.curr); |
Ingo Molnar | bdd7c81 | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 2488 | } |
| 2489 | |
| 2490 | /* Assumes rq->lock is held */ |
Gregory Haskins | 1f11eb6a | 2008-06-04 15:04:05 -0400 | [diff] [blame] | 2491 | static void rq_offline_rt(struct rq *rq) |
Ingo Molnar | bdd7c81 | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 2492 | { |
| 2493 | if (rq->rt.overloaded) |
| 2494 | rt_clear_overload(rq); |
Gregory Haskins | 6e0534f | 2008-05-12 21:21:01 +0200 | [diff] [blame] | 2495 | |
Peter Zijlstra | 7def2be | 2008-06-05 14:49:58 +0200 | [diff] [blame] | 2496 | __disable_runtime(rq); |
| 2497 | |
Gregory Haskins | 6e0534f | 2008-05-12 21:21:01 +0200 | [diff] [blame] | 2498 | cpupri_set(&rq->rd->cpupri, rq->cpu, CPUPRI_INVALID); |
Ingo Molnar | bdd7c81 | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 2499 | } |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 2500 | |
| 2501 | /* |
| 2502 | * When switch from the rt queue, we bring ourselves to a position |
| 2503 | * that we might want to pull RT tasks from other runqueues. |
| 2504 | */ |
Peter Zijlstra | da7a735 | 2011-01-17 17:03:27 +0100 | [diff] [blame] | 2505 | static void switched_from_rt(struct rq *rq, struct task_struct *p) |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 2506 | { |
| 2507 | /* |
| 2508 | * If there are other RT tasks then we will reschedule |
| 2509 | * and the scheduling of the other RT tasks will handle |
| 2510 | * the balancing. But if we are the last RT task |
| 2511 | * we may need to handle the pulling of RT tasks |
| 2512 | * now. |
| 2513 | */ |
Kirill Tkhai | da0c1e6 | 2014-08-20 13:47:32 +0400 | [diff] [blame] | 2514 | if (!task_on_rq_queued(p) || rq->rt.rt_nr_running) |
Kirill Tkhai | 1158ddb | 2012-11-23 00:02:15 +0400 | [diff] [blame] | 2515 | return; |
| 2516 | |
Ingo Molnar | 02d8ec9 | 2018-03-03 16:27:54 +0100 | [diff] [blame] | 2517 | rt_queue_pull_task(rq); |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 2518 | } |
Rusty Russell | 3d8cbdf | 2008-11-25 09:58:41 +1030 | [diff] [blame] | 2519 | |
Li Zefan | 11c785b | 2014-02-08 14:17:45 +0800 | [diff] [blame] | 2520 | void __init init_sched_rt_class(void) |
Rusty Russell | 3d8cbdf | 2008-11-25 09:58:41 +1030 | [diff] [blame] | 2521 | { |
| 2522 | unsigned int i; |
| 2523 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 2524 | for_each_possible_cpu(i) { |
Yinghai Lu | eaa9584 | 2009-06-06 14:51:36 -0700 | [diff] [blame] | 2525 | zalloc_cpumask_var_node(&per_cpu(local_cpu_mask, i), |
Mike Travis | 6ca09df | 2008-12-31 18:08:45 -0800 | [diff] [blame] | 2526 | GFP_KERNEL, cpu_to_node(i)); |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 2527 | } |
Rusty Russell | 3d8cbdf | 2008-11-25 09:58:41 +1030 | [diff] [blame] | 2528 | } |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 2529 | #endif /* CONFIG_SMP */ |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 2530 | |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 2531 | /* |
| 2532 | * When switching a task to RT, we may overload the runqueue |
| 2533 | * with RT tasks. In this case we try to push them off to |
| 2534 | * other runqueues. |
| 2535 | */ |
Peter Zijlstra | da7a735 | 2011-01-17 17:03:27 +0100 | [diff] [blame] | 2536 | static void switched_to_rt(struct rq *rq, struct task_struct *p) |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 2537 | { |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 2538 | /* |
Vincent Donnefort | fecfcbc | 2021-06-21 11:37:51 +0100 | [diff] [blame] | 2539 | * If we are running, update the avg_rt tracking, as the running time |
| 2540 | * will now on be accounted into the latter. |
| 2541 | */ |
| 2542 | if (task_current(rq, p)) { |
| 2543 | update_rt_rq_load_avg(rq_clock_pelt(rq), rq, 0); |
| 2544 | return; |
| 2545 | } |
| 2546 | |
| 2547 | /* |
| 2548 | * If we are not running we may need to preempt the current |
| 2549 | * running task. If that current running task is also an RT task |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 2550 | * then see if we can move to another run queue. |
| 2551 | */ |
Vincent Donnefort | fecfcbc | 2021-06-21 11:37:51 +0100 | [diff] [blame] | 2552 | if (task_on_rq_queued(p)) { |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 2553 | #ifdef CONFIG_SMP |
Qais Yousef | d94a9df | 2020-03-02 13:27:20 +0000 | [diff] [blame] | 2554 | if (p->nr_cpus_allowed > 1 && rq->rt.overloaded) |
Ingo Molnar | 02d8ec9 | 2018-03-03 16:27:54 +0100 | [diff] [blame] | 2555 | rt_queue_push_tasks(rq); |
Sebastian Andrzej Siewior | 619bd4a | 2017-01-24 15:40:06 +0100 | [diff] [blame] | 2556 | #endif /* CONFIG_SMP */ |
Paul E. McKenney | 2fe2582 | 2017-10-13 17:00:18 -0700 | [diff] [blame] | 2557 | if (p->prio < rq->curr->prio && cpu_online(cpu_of(rq))) |
Kirill Tkhai | 8875125 | 2014-06-29 00:03:57 +0400 | [diff] [blame] | 2558 | resched_curr(rq); |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 2559 | } |
| 2560 | } |
| 2561 | |
| 2562 | /* |
| 2563 | * Priority of the task has changed. This may cause |
| 2564 | * us to initiate a push or pull. |
| 2565 | */ |
Peter Zijlstra | da7a735 | 2011-01-17 17:03:27 +0100 | [diff] [blame] | 2566 | static void |
| 2567 | prio_changed_rt(struct rq *rq, struct task_struct *p, int oldprio) |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 2568 | { |
Kirill Tkhai | da0c1e6 | 2014-08-20 13:47:32 +0400 | [diff] [blame] | 2569 | if (!task_on_rq_queued(p)) |
Peter Zijlstra | da7a735 | 2011-01-17 17:03:27 +0100 | [diff] [blame] | 2570 | return; |
| 2571 | |
Hui Su | 65bcf07 | 2020-10-31 01:32:23 +0800 | [diff] [blame] | 2572 | if (task_current(rq, p)) { |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 2573 | #ifdef CONFIG_SMP |
| 2574 | /* |
| 2575 | * If our priority decreases while running, we |
| 2576 | * may need to pull tasks to this runqueue. |
| 2577 | */ |
| 2578 | if (oldprio < p->prio) |
Ingo Molnar | 02d8ec9 | 2018-03-03 16:27:54 +0100 | [diff] [blame] | 2579 | rt_queue_pull_task(rq); |
Peter Zijlstra | fd7a4be | 2015-06-11 14:46:41 +0200 | [diff] [blame] | 2580 | |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 2581 | /* |
| 2582 | * If there's a higher priority task waiting to run |
Peter Zijlstra | fd7a4be | 2015-06-11 14:46:41 +0200 | [diff] [blame] | 2583 | * then reschedule. |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 2584 | */ |
Peter Zijlstra | fd7a4be | 2015-06-11 14:46:41 +0200 | [diff] [blame] | 2585 | if (p->prio > rq->rt.highest_prio.curr) |
Kirill Tkhai | 8875125 | 2014-06-29 00:03:57 +0400 | [diff] [blame] | 2586 | resched_curr(rq); |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 2587 | #else |
| 2588 | /* For UP simply resched on drop of prio */ |
| 2589 | if (oldprio < p->prio) |
Kirill Tkhai | 8875125 | 2014-06-29 00:03:57 +0400 | [diff] [blame] | 2590 | resched_curr(rq); |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 2591 | #endif /* CONFIG_SMP */ |
| 2592 | } else { |
| 2593 | /* |
| 2594 | * This task is not running, but if it is |
| 2595 | * greater than the current running task |
| 2596 | * then reschedule. |
| 2597 | */ |
| 2598 | if (p->prio < rq->curr->prio) |
Kirill Tkhai | 8875125 | 2014-06-29 00:03:57 +0400 | [diff] [blame] | 2599 | resched_curr(rq); |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 2600 | } |
| 2601 | } |
| 2602 | |
Nicolas Pitre | b18b6a9 | 2017-01-21 00:09:08 -0500 | [diff] [blame] | 2603 | #ifdef CONFIG_POSIX_TIMERS |
Peter Zijlstra | 78f2c7d | 2008-01-25 21:08:27 +0100 | [diff] [blame] | 2604 | static void watchdog(struct rq *rq, struct task_struct *p) |
| 2605 | { |
| 2606 | unsigned long soft, hard; |
| 2607 | |
Jiri Slaby | 78d7d40 | 2010-03-05 13:42:54 -0800 | [diff] [blame] | 2608 | /* max may change after cur was read, this will be fixed next tick */ |
| 2609 | soft = task_rlimit(p, RLIMIT_RTTIME); |
| 2610 | hard = task_rlimit_max(p, RLIMIT_RTTIME); |
Peter Zijlstra | 78f2c7d | 2008-01-25 21:08:27 +0100 | [diff] [blame] | 2611 | |
| 2612 | if (soft != RLIM_INFINITY) { |
| 2613 | unsigned long next; |
| 2614 | |
Ying Xue | 57d2aa0 | 2012-07-17 15:03:43 +0800 | [diff] [blame] | 2615 | if (p->rt.watchdog_stamp != jiffies) { |
| 2616 | p->rt.timeout++; |
| 2617 | p->rt.watchdog_stamp = jiffies; |
| 2618 | } |
| 2619 | |
Peter Zijlstra | 78f2c7d | 2008-01-25 21:08:27 +0100 | [diff] [blame] | 2620 | next = DIV_ROUND_UP(min(soft, hard), USEC_PER_SEC/HZ); |
Thomas Gleixner | 3a245c0 | 2019-08-21 21:09:06 +0200 | [diff] [blame] | 2621 | if (p->rt.timeout > next) { |
| 2622 | posix_cputimers_rt_watchdog(&p->posix_cputimers, |
| 2623 | p->se.sum_exec_runtime); |
| 2624 | } |
Peter Zijlstra | 78f2c7d | 2008-01-25 21:08:27 +0100 | [diff] [blame] | 2625 | } |
| 2626 | } |
Nicolas Pitre | b18b6a9 | 2017-01-21 00:09:08 -0500 | [diff] [blame] | 2627 | #else |
| 2628 | static inline void watchdog(struct rq *rq, struct task_struct *p) { } |
| 2629 | #endif |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 2630 | |
Frederic Weisbecker | d84b313 | 2018-02-21 05:17:27 +0100 | [diff] [blame] | 2631 | /* |
| 2632 | * scheduler tick hitting a task of our scheduling class. |
| 2633 | * |
| 2634 | * NOTE: This function can be called remotely by the tick offload that |
| 2635 | * goes along full dynticks. Therefore no local assumption can be made |
| 2636 | * and everything must be accessed through the @rq and @curr passed in |
| 2637 | * parameters. |
| 2638 | */ |
Peter Zijlstra | 8f4d37e | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 2639 | 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] | 2640 | { |
Colin Cross | 454c799 | 2012-05-16 21:34:23 -0700 | [diff] [blame] | 2641 | struct sched_rt_entity *rt_se = &p->rt; |
| 2642 | |
Peter Zijlstra | 67e2be0 | 2007-12-20 15:01:17 +0100 | [diff] [blame] | 2643 | update_curr_rt(rq); |
Vincent Guittot | 2312729 | 2019-01-23 16:26:53 +0100 | [diff] [blame] | 2644 | update_rt_rq_load_avg(rq_clock_pelt(rq), rq, 1); |
Peter Zijlstra | 67e2be0 | 2007-12-20 15:01:17 +0100 | [diff] [blame] | 2645 | |
Peter Zijlstra | 78f2c7d | 2008-01-25 21:08:27 +0100 | [diff] [blame] | 2646 | watchdog(rq, p); |
| 2647 | |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 2648 | /* |
| 2649 | * RR tasks need a special form of timeslice management. |
| 2650 | * FIFO tasks have no timeslices. |
| 2651 | */ |
| 2652 | if (p->policy != SCHED_RR) |
| 2653 | return; |
| 2654 | |
Peter Zijlstra | fa71706 | 2008-01-25 21:08:27 +0100 | [diff] [blame] | 2655 | if (--p->rt.time_slice) |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 2656 | return; |
| 2657 | |
Clark Williams | ce0dbbb | 2013-02-07 09:47:04 -0600 | [diff] [blame] | 2658 | p->rt.time_slice = sched_rr_timeslice; |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 2659 | |
Dmitry Adamushko | 98fbc79 | 2007-08-24 20:39:10 +0200 | [diff] [blame] | 2660 | /* |
Li Bin | e9aa39b | 2013-10-21 20:15:43 +0800 | [diff] [blame] | 2661 | * Requeue to the end of queue if we (and all of our ancestors) are not |
| 2662 | * the only element on the queue |
Dmitry Adamushko | 98fbc79 | 2007-08-24 20:39:10 +0200 | [diff] [blame] | 2663 | */ |
Colin Cross | 454c799 | 2012-05-16 21:34:23 -0700 | [diff] [blame] | 2664 | for_each_sched_rt_entity(rt_se) { |
| 2665 | if (rt_se->run_list.prev != rt_se->run_list.next) { |
| 2666 | requeue_task_rt(rq, p, 0); |
Kirill Tkhai | 8aa6f0e | 2014-09-22 22:36:43 +0400 | [diff] [blame] | 2667 | resched_curr(rq); |
Colin Cross | 454c799 | 2012-05-16 21:34:23 -0700 | [diff] [blame] | 2668 | return; |
| 2669 | } |
Dmitry Adamushko | 98fbc79 | 2007-08-24 20:39:10 +0200 | [diff] [blame] | 2670 | } |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 2671 | } |
| 2672 | |
H Hartley Sweeten | 6d686f4 | 2010-01-13 20:21:52 -0700 | [diff] [blame] | 2673 | static unsigned int get_rr_interval_rt(struct rq *rq, struct task_struct *task) |
Peter Williams | 0d721ce | 2009-09-21 01:31:53 +0000 | [diff] [blame] | 2674 | { |
| 2675 | /* |
| 2676 | * Time slice is 0 for SCHED_FIFO tasks |
| 2677 | */ |
| 2678 | if (task->policy == SCHED_RR) |
Clark Williams | ce0dbbb | 2013-02-07 09:47:04 -0600 | [diff] [blame] | 2679 | return sched_rr_timeslice; |
Peter Williams | 0d721ce | 2009-09-21 01:31:53 +0000 | [diff] [blame] | 2680 | else |
| 2681 | return 0; |
| 2682 | } |
| 2683 | |
Hao Jia | 530bfad | 2023-03-16 16:18:06 +0800 | [diff] [blame] | 2684 | #ifdef CONFIG_SCHED_CORE |
| 2685 | static int task_is_throttled_rt(struct task_struct *p, int cpu) |
| 2686 | { |
| 2687 | struct rt_rq *rt_rq; |
| 2688 | |
| 2689 | #ifdef CONFIG_RT_GROUP_SCHED |
| 2690 | rt_rq = task_group(p)->rt_rq[cpu]; |
| 2691 | #else |
| 2692 | rt_rq = &cpu_rq(cpu)->rt; |
| 2693 | #endif |
| 2694 | |
| 2695 | return rt_rq_throttled(rt_rq); |
| 2696 | } |
| 2697 | #endif |
| 2698 | |
Peter Zijlstra | 43c31ac | 2020-10-21 15:45:33 +0200 | [diff] [blame] | 2699 | DEFINE_SCHED_CLASS(rt) = { |
| 2700 | |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 2701 | .enqueue_task = enqueue_task_rt, |
| 2702 | .dequeue_task = dequeue_task_rt, |
| 2703 | .yield_task = yield_task_rt, |
| 2704 | |
| 2705 | .check_preempt_curr = check_preempt_curr_rt, |
| 2706 | |
| 2707 | .pick_next_task = pick_next_task_rt, |
| 2708 | .put_prev_task = put_prev_task_rt, |
Peter Zijlstra | 03b7fad | 2019-05-29 20:36:41 +0000 | [diff] [blame] | 2709 | .set_next_task = set_next_task_rt, |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 2710 | |
Peter Williams | 681f3e6 | 2007-10-24 18:23:51 +0200 | [diff] [blame] | 2711 | #ifdef CONFIG_SMP |
Peter Zijlstra | 6e2df05 | 2019-11-08 11:11:52 +0100 | [diff] [blame] | 2712 | .balance = balance_rt, |
Peter Zijlstra | 21f56ffe | 2020-11-17 18:19:32 -0500 | [diff] [blame] | 2713 | .pick_task = pick_task_rt, |
Li Zefan | 4ce72a2 | 2008-10-22 15:25:26 +0800 | [diff] [blame] | 2714 | .select_task_rq = select_task_rq_rt, |
Peter Zijlstra | 6c37067 | 2015-05-15 17:43:36 +0200 | [diff] [blame] | 2715 | .set_cpus_allowed = set_cpus_allowed_common, |
Gregory Haskins | 1f11eb6a | 2008-06-04 15:04:05 -0400 | [diff] [blame] | 2716 | .rq_online = rq_online_rt, |
| 2717 | .rq_offline = rq_offline_rt, |
Peter Zijlstra | efbbd05 | 2009-12-16 18:04:40 +0100 | [diff] [blame] | 2718 | .task_woken = task_woken_rt, |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 2719 | .switched_from = switched_from_rt, |
Peter Zijlstra | a7c8155 | 2020-09-28 17:06:07 +0200 | [diff] [blame] | 2720 | .find_lock_rq = find_lock_lowest_rq, |
Peter Williams | 681f3e6 | 2007-10-24 18:23:51 +0200 | [diff] [blame] | 2721 | #endif |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 2722 | |
| 2723 | .task_tick = task_tick_rt, |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 2724 | |
Peter Williams | 0d721ce | 2009-09-21 01:31:53 +0000 | [diff] [blame] | 2725 | .get_rr_interval = get_rr_interval_rt, |
| 2726 | |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 2727 | .prio_changed = prio_changed_rt, |
| 2728 | .switched_to = switched_to_rt, |
Stanislaw Gruszka | 6e99891 | 2014-11-12 16:58:44 +0100 | [diff] [blame] | 2729 | |
| 2730 | .update_curr = update_curr_rt, |
Patrick Bellasi | 982d9cd | 2019-06-21 09:42:10 +0100 | [diff] [blame] | 2731 | |
Hao Jia | 530bfad | 2023-03-16 16:18:06 +0800 | [diff] [blame] | 2732 | #ifdef CONFIG_SCHED_CORE |
| 2733 | .task_is_throttled = task_is_throttled_rt, |
| 2734 | #endif |
| 2735 | |
Patrick Bellasi | 982d9cd | 2019-06-21 09:42:10 +0100 | [diff] [blame] | 2736 | #ifdef CONFIG_UCLAMP_TASK |
| 2737 | .uclamp_enabled = 1, |
| 2738 | #endif |
Ingo Molnar | bb44e5d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 2739 | }; |
Peter Zijlstra | ada18de | 2008-06-19 14:22:24 +0200 | [diff] [blame] | 2740 | |
Nicolas Pitre | 8887cd9 | 2017-06-21 14:22:02 -0400 | [diff] [blame] | 2741 | #ifdef CONFIG_RT_GROUP_SCHED |
| 2742 | /* |
| 2743 | * Ensure that the real time constraints are schedulable. |
| 2744 | */ |
| 2745 | static DEFINE_MUTEX(rt_constraints_mutex); |
| 2746 | |
Nicolas Pitre | 8887cd9 | 2017-06-21 14:22:02 -0400 | [diff] [blame] | 2747 | static inline int tg_has_rt_tasks(struct task_group *tg) |
| 2748 | { |
Konstantin Khlebnikov | b4fb015 | 2020-01-25 17:50:38 +0300 | [diff] [blame] | 2749 | struct task_struct *task; |
| 2750 | struct css_task_iter it; |
| 2751 | int ret = 0; |
Nicolas Pitre | 8887cd9 | 2017-06-21 14:22:02 -0400 | [diff] [blame] | 2752 | |
| 2753 | /* |
| 2754 | * Autogroups do not have RT tasks; see autogroup_create(). |
| 2755 | */ |
| 2756 | if (task_group_is_autogroup(tg)) |
| 2757 | return 0; |
| 2758 | |
Konstantin Khlebnikov | b4fb015 | 2020-01-25 17:50:38 +0300 | [diff] [blame] | 2759 | css_task_iter_start(&tg->css, 0, &it); |
| 2760 | while (!ret && (task = css_task_iter_next(&it))) |
| 2761 | ret |= rt_task(task); |
| 2762 | css_task_iter_end(&it); |
Nicolas Pitre | 8887cd9 | 2017-06-21 14:22:02 -0400 | [diff] [blame] | 2763 | |
Konstantin Khlebnikov | b4fb015 | 2020-01-25 17:50:38 +0300 | [diff] [blame] | 2764 | return ret; |
Nicolas Pitre | 8887cd9 | 2017-06-21 14:22:02 -0400 | [diff] [blame] | 2765 | } |
| 2766 | |
| 2767 | struct rt_schedulable_data { |
| 2768 | struct task_group *tg; |
| 2769 | u64 rt_period; |
| 2770 | u64 rt_runtime; |
| 2771 | }; |
| 2772 | |
| 2773 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
| 2774 | { |
| 2775 | struct rt_schedulable_data *d = data; |
| 2776 | struct task_group *child; |
| 2777 | unsigned long total, sum = 0; |
| 2778 | u64 period, runtime; |
| 2779 | |
| 2780 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
| 2781 | runtime = tg->rt_bandwidth.rt_runtime; |
| 2782 | |
| 2783 | if (tg == d->tg) { |
| 2784 | period = d->rt_period; |
| 2785 | runtime = d->rt_runtime; |
| 2786 | } |
| 2787 | |
| 2788 | /* |
| 2789 | * Cannot have more runtime than the period. |
| 2790 | */ |
| 2791 | if (runtime > period && runtime != RUNTIME_INF) |
| 2792 | return -EINVAL; |
| 2793 | |
| 2794 | /* |
Konstantin Khlebnikov | b4fb015 | 2020-01-25 17:50:38 +0300 | [diff] [blame] | 2795 | * Ensure we don't starve existing RT tasks if runtime turns zero. |
Nicolas Pitre | 8887cd9 | 2017-06-21 14:22:02 -0400 | [diff] [blame] | 2796 | */ |
Konstantin Khlebnikov | b4fb015 | 2020-01-25 17:50:38 +0300 | [diff] [blame] | 2797 | if (rt_bandwidth_enabled() && !runtime && |
| 2798 | tg->rt_bandwidth.rt_runtime && tg_has_rt_tasks(tg)) |
Nicolas Pitre | 8887cd9 | 2017-06-21 14:22:02 -0400 | [diff] [blame] | 2799 | return -EBUSY; |
| 2800 | |
| 2801 | total = to_ratio(period, runtime); |
| 2802 | |
| 2803 | /* |
| 2804 | * Nobody can have more than the global setting allows. |
| 2805 | */ |
| 2806 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) |
| 2807 | return -EINVAL; |
| 2808 | |
| 2809 | /* |
| 2810 | * The sum of our children's runtime should not exceed our own. |
| 2811 | */ |
| 2812 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
| 2813 | period = ktime_to_ns(child->rt_bandwidth.rt_period); |
| 2814 | runtime = child->rt_bandwidth.rt_runtime; |
| 2815 | |
| 2816 | if (child == d->tg) { |
| 2817 | period = d->rt_period; |
| 2818 | runtime = d->rt_runtime; |
| 2819 | } |
| 2820 | |
| 2821 | sum += to_ratio(period, runtime); |
| 2822 | } |
| 2823 | |
| 2824 | if (sum > total) |
| 2825 | return -EINVAL; |
| 2826 | |
| 2827 | return 0; |
| 2828 | } |
| 2829 | |
| 2830 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
| 2831 | { |
| 2832 | int ret; |
| 2833 | |
| 2834 | struct rt_schedulable_data data = { |
| 2835 | .tg = tg, |
| 2836 | .rt_period = period, |
| 2837 | .rt_runtime = runtime, |
| 2838 | }; |
| 2839 | |
| 2840 | rcu_read_lock(); |
| 2841 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); |
| 2842 | rcu_read_unlock(); |
| 2843 | |
| 2844 | return ret; |
| 2845 | } |
| 2846 | |
| 2847 | static int tg_set_rt_bandwidth(struct task_group *tg, |
| 2848 | u64 rt_period, u64 rt_runtime) |
| 2849 | { |
| 2850 | int i, err = 0; |
| 2851 | |
| 2852 | /* |
| 2853 | * Disallowing the root group RT runtime is BAD, it would disallow the |
| 2854 | * kernel creating (and or operating) RT threads. |
| 2855 | */ |
| 2856 | if (tg == &root_task_group && rt_runtime == 0) |
| 2857 | return -EINVAL; |
| 2858 | |
| 2859 | /* No period doesn't make any sense. */ |
| 2860 | if (rt_period == 0) |
| 2861 | return -EINVAL; |
| 2862 | |
Huaixin Chang | d505b8a | 2020-04-25 18:52:48 +0800 | [diff] [blame] | 2863 | /* |
| 2864 | * Bound quota to defend quota against overflow during bandwidth shift. |
| 2865 | */ |
| 2866 | if (rt_runtime != RUNTIME_INF && rt_runtime > max_rt_runtime) |
| 2867 | return -EINVAL; |
| 2868 | |
Nicolas Pitre | 8887cd9 | 2017-06-21 14:22:02 -0400 | [diff] [blame] | 2869 | mutex_lock(&rt_constraints_mutex); |
Nicolas Pitre | 8887cd9 | 2017-06-21 14:22:02 -0400 | [diff] [blame] | 2870 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
| 2871 | if (err) |
| 2872 | goto unlock; |
| 2873 | |
| 2874 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
| 2875 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
| 2876 | tg->rt_bandwidth.rt_runtime = rt_runtime; |
| 2877 | |
| 2878 | for_each_possible_cpu(i) { |
| 2879 | struct rt_rq *rt_rq = tg->rt_rq[i]; |
| 2880 | |
| 2881 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
| 2882 | rt_rq->rt_runtime = rt_runtime; |
| 2883 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
| 2884 | } |
| 2885 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
| 2886 | unlock: |
Nicolas Pitre | 8887cd9 | 2017-06-21 14:22:02 -0400 | [diff] [blame] | 2887 | mutex_unlock(&rt_constraints_mutex); |
| 2888 | |
| 2889 | return err; |
| 2890 | } |
| 2891 | |
| 2892 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
| 2893 | { |
| 2894 | u64 rt_runtime, rt_period; |
| 2895 | |
| 2896 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
| 2897 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; |
| 2898 | if (rt_runtime_us < 0) |
| 2899 | rt_runtime = RUNTIME_INF; |
Konstantin Khlebnikov | 1a010e2 | 2019-02-27 11:10:17 +0300 | [diff] [blame] | 2900 | else if ((u64)rt_runtime_us > U64_MAX / NSEC_PER_USEC) |
| 2901 | return -EINVAL; |
Nicolas Pitre | 8887cd9 | 2017-06-21 14:22:02 -0400 | [diff] [blame] | 2902 | |
| 2903 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
| 2904 | } |
| 2905 | |
| 2906 | long sched_group_rt_runtime(struct task_group *tg) |
| 2907 | { |
| 2908 | u64 rt_runtime_us; |
| 2909 | |
| 2910 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
| 2911 | return -1; |
| 2912 | |
| 2913 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
| 2914 | do_div(rt_runtime_us, NSEC_PER_USEC); |
| 2915 | return rt_runtime_us; |
| 2916 | } |
| 2917 | |
| 2918 | int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us) |
| 2919 | { |
| 2920 | u64 rt_runtime, rt_period; |
| 2921 | |
Konstantin Khlebnikov | 1a010e2 | 2019-02-27 11:10:17 +0300 | [diff] [blame] | 2922 | if (rt_period_us > U64_MAX / NSEC_PER_USEC) |
| 2923 | return -EINVAL; |
| 2924 | |
Nicolas Pitre | 8887cd9 | 2017-06-21 14:22:02 -0400 | [diff] [blame] | 2925 | rt_period = rt_period_us * NSEC_PER_USEC; |
| 2926 | rt_runtime = tg->rt_bandwidth.rt_runtime; |
| 2927 | |
| 2928 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
| 2929 | } |
| 2930 | |
| 2931 | long sched_group_rt_period(struct task_group *tg) |
| 2932 | { |
| 2933 | u64 rt_period_us; |
| 2934 | |
| 2935 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); |
| 2936 | do_div(rt_period_us, NSEC_PER_USEC); |
| 2937 | return rt_period_us; |
| 2938 | } |
| 2939 | |
Baisong Zhong | 28f152c | 2022-03-18 10:54:17 +0800 | [diff] [blame] | 2940 | #ifdef CONFIG_SYSCTL |
Nicolas Pitre | 8887cd9 | 2017-06-21 14:22:02 -0400 | [diff] [blame] | 2941 | static int sched_rt_global_constraints(void) |
| 2942 | { |
| 2943 | int ret = 0; |
| 2944 | |
| 2945 | mutex_lock(&rt_constraints_mutex); |
Nicolas Pitre | 8887cd9 | 2017-06-21 14:22:02 -0400 | [diff] [blame] | 2946 | ret = __rt_schedulable(NULL, 0, 0); |
Nicolas Pitre | 8887cd9 | 2017-06-21 14:22:02 -0400 | [diff] [blame] | 2947 | mutex_unlock(&rt_constraints_mutex); |
| 2948 | |
| 2949 | return ret; |
| 2950 | } |
Baisong Zhong | 28f152c | 2022-03-18 10:54:17 +0800 | [diff] [blame] | 2951 | #endif /* CONFIG_SYSCTL */ |
Nicolas Pitre | 8887cd9 | 2017-06-21 14:22:02 -0400 | [diff] [blame] | 2952 | |
| 2953 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) |
| 2954 | { |
| 2955 | /* Don't accept realtime tasks when there is no way for them to run */ |
| 2956 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) |
| 2957 | return 0; |
| 2958 | |
| 2959 | return 1; |
| 2960 | } |
| 2961 | |
| 2962 | #else /* !CONFIG_RT_GROUP_SCHED */ |
Baisong Zhong | 28f152c | 2022-03-18 10:54:17 +0800 | [diff] [blame] | 2963 | |
| 2964 | #ifdef CONFIG_SYSCTL |
Nicolas Pitre | 8887cd9 | 2017-06-21 14:22:02 -0400 | [diff] [blame] | 2965 | static int sched_rt_global_constraints(void) |
| 2966 | { |
| 2967 | unsigned long flags; |
| 2968 | int i; |
| 2969 | |
| 2970 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
| 2971 | for_each_possible_cpu(i) { |
| 2972 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; |
| 2973 | |
| 2974 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
| 2975 | rt_rq->rt_runtime = global_rt_runtime(); |
| 2976 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
| 2977 | } |
| 2978 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
| 2979 | |
| 2980 | return 0; |
| 2981 | } |
Baisong Zhong | 28f152c | 2022-03-18 10:54:17 +0800 | [diff] [blame] | 2982 | #endif /* CONFIG_SYSCTL */ |
Nicolas Pitre | 8887cd9 | 2017-06-21 14:22:02 -0400 | [diff] [blame] | 2983 | #endif /* CONFIG_RT_GROUP_SCHED */ |
| 2984 | |
Baisong Zhong | 28f152c | 2022-03-18 10:54:17 +0800 | [diff] [blame] | 2985 | #ifdef CONFIG_SYSCTL |
Nicolas Pitre | 8887cd9 | 2017-06-21 14:22:02 -0400 | [diff] [blame] | 2986 | static int sched_rt_global_validate(void) |
| 2987 | { |
| 2988 | if (sysctl_sched_rt_period <= 0) |
| 2989 | return -EINVAL; |
| 2990 | |
| 2991 | if ((sysctl_sched_rt_runtime != RUNTIME_INF) && |
Huaixin Chang | d505b8a | 2020-04-25 18:52:48 +0800 | [diff] [blame] | 2992 | ((sysctl_sched_rt_runtime > sysctl_sched_rt_period) || |
| 2993 | ((u64)sysctl_sched_rt_runtime * |
| 2994 | NSEC_PER_USEC > max_rt_runtime))) |
Nicolas Pitre | 8887cd9 | 2017-06-21 14:22:02 -0400 | [diff] [blame] | 2995 | return -EINVAL; |
| 2996 | |
| 2997 | return 0; |
| 2998 | } |
| 2999 | |
| 3000 | static void sched_rt_do_global(void) |
| 3001 | { |
Li Hua | 9b58e97 | 2021-12-03 03:36:18 +0000 | [diff] [blame] | 3002 | unsigned long flags; |
| 3003 | |
| 3004 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
Nicolas Pitre | 8887cd9 | 2017-06-21 14:22:02 -0400 | [diff] [blame] | 3005 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); |
| 3006 | def_rt_bandwidth.rt_period = ns_to_ktime(global_rt_period()); |
Li Hua | 9b58e97 | 2021-12-03 03:36:18 +0000 | [diff] [blame] | 3007 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
Nicolas Pitre | 8887cd9 | 2017-06-21 14:22:02 -0400 | [diff] [blame] | 3008 | } |
| 3009 | |
Zhen Ni | d9ab0e6 | 2022-02-15 19:45:59 +0800 | [diff] [blame] | 3010 | static int sched_rt_handler(struct ctl_table *table, int write, void *buffer, |
Christoph Hellwig | 3292739 | 2020-04-24 08:43:38 +0200 | [diff] [blame] | 3011 | size_t *lenp, loff_t *ppos) |
Nicolas Pitre | 8887cd9 | 2017-06-21 14:22:02 -0400 | [diff] [blame] | 3012 | { |
| 3013 | int old_period, old_runtime; |
| 3014 | static DEFINE_MUTEX(mutex); |
| 3015 | int ret; |
| 3016 | |
| 3017 | mutex_lock(&mutex); |
| 3018 | old_period = sysctl_sched_rt_period; |
| 3019 | old_runtime = sysctl_sched_rt_runtime; |
| 3020 | |
| 3021 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
| 3022 | |
| 3023 | if (!ret && write) { |
| 3024 | ret = sched_rt_global_validate(); |
| 3025 | if (ret) |
| 3026 | goto undo; |
| 3027 | |
| 3028 | ret = sched_dl_global_validate(); |
| 3029 | if (ret) |
| 3030 | goto undo; |
| 3031 | |
| 3032 | ret = sched_rt_global_constraints(); |
| 3033 | if (ret) |
| 3034 | goto undo; |
| 3035 | |
| 3036 | sched_rt_do_global(); |
| 3037 | sched_dl_do_global(); |
| 3038 | } |
| 3039 | if (0) { |
| 3040 | undo: |
| 3041 | sysctl_sched_rt_period = old_period; |
| 3042 | sysctl_sched_rt_runtime = old_runtime; |
| 3043 | } |
| 3044 | mutex_unlock(&mutex); |
| 3045 | |
| 3046 | return ret; |
| 3047 | } |
| 3048 | |
Zhen Ni | dafd7a9 | 2022-02-15 19:46:01 +0800 | [diff] [blame] | 3049 | static int sched_rr_handler(struct ctl_table *table, int write, void *buffer, |
Christoph Hellwig | 3292739 | 2020-04-24 08:43:38 +0200 | [diff] [blame] | 3050 | size_t *lenp, loff_t *ppos) |
Nicolas Pitre | 8887cd9 | 2017-06-21 14:22:02 -0400 | [diff] [blame] | 3051 | { |
| 3052 | int ret; |
| 3053 | static DEFINE_MUTEX(mutex); |
| 3054 | |
| 3055 | mutex_lock(&mutex); |
| 3056 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
| 3057 | /* |
| 3058 | * Make sure that internally we keep jiffies. |
| 3059 | * Also, writing zero resets the timeslice to default: |
| 3060 | */ |
| 3061 | if (!ret && write) { |
| 3062 | sched_rr_timeslice = |
| 3063 | sysctl_sched_rr_timeslice <= 0 ? RR_TIMESLICE : |
| 3064 | msecs_to_jiffies(sysctl_sched_rr_timeslice); |
Cyril Hrubis | c1fc648 | 2023-08-02 17:19:06 +0200 | [diff] [blame] | 3065 | |
| 3066 | if (sysctl_sched_rr_timeslice <= 0) |
| 3067 | sysctl_sched_rr_timeslice = jiffies_to_msecs(RR_TIMESLICE); |
Nicolas Pitre | 8887cd9 | 2017-06-21 14:22:02 -0400 | [diff] [blame] | 3068 | } |
| 3069 | mutex_unlock(&mutex); |
Ingo Molnar | 97fb7a0 | 2018-03-03 14:01:12 +0100 | [diff] [blame] | 3070 | |
Nicolas Pitre | 8887cd9 | 2017-06-21 14:22:02 -0400 | [diff] [blame] | 3071 | return ret; |
| 3072 | } |
Baisong Zhong | 28f152c | 2022-03-18 10:54:17 +0800 | [diff] [blame] | 3073 | #endif /* CONFIG_SYSCTL */ |
Nicolas Pitre | 8887cd9 | 2017-06-21 14:22:02 -0400 | [diff] [blame] | 3074 | |
Peter Zijlstra | ada18de | 2008-06-19 14:22:24 +0200 | [diff] [blame] | 3075 | #ifdef CONFIG_SCHED_DEBUG |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 3076 | void print_rt_stats(struct seq_file *m, int cpu) |
Peter Zijlstra | ada18de | 2008-06-19 14:22:24 +0200 | [diff] [blame] | 3077 | { |
Cheng Xu | ec514c4 | 2011-05-14 14:20:02 +0800 | [diff] [blame] | 3078 | rt_rq_iter_t iter; |
Peter Zijlstra | ada18de | 2008-06-19 14:22:24 +0200 | [diff] [blame] | 3079 | struct rt_rq *rt_rq; |
| 3080 | |
| 3081 | rcu_read_lock(); |
Cheng Xu | ec514c4 | 2011-05-14 14:20:02 +0800 | [diff] [blame] | 3082 | for_each_rt_rq(rt_rq, iter, cpu_rq(cpu)) |
Peter Zijlstra | ada18de | 2008-06-19 14:22:24 +0200 | [diff] [blame] | 3083 | print_rt_rq(m, cpu, rt_rq); |
| 3084 | rcu_read_unlock(); |
| 3085 | } |
Dhaval Giani | 55e12e5 | 2008-06-24 23:39:43 +0530 | [diff] [blame] | 3086 | #endif /* CONFIG_SCHED_DEBUG */ |