Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1 | |
| 2 | #include <linux/sched.h> |
Clark Williams | cf4aebc2 | 2013-02-07 09:46:59 -0600 | [diff] [blame] | 3 | #include <linux/sched/sysctl.h> |
Clark Williams | 8bd75c7 | 2013-02-07 09:47:07 -0600 | [diff] [blame] | 4 | #include <linux/sched/rt.h> |
Frederic Weisbecker | 19d23dbf | 2016-09-26 02:29:20 +0200 | [diff] [blame] | 5 | #include <linux/u64_stats_sync.h> |
Dario Faggioli | aab03e0 | 2013-11-28 11:14:43 +0100 | [diff] [blame] | 6 | #include <linux/sched/deadline.h> |
Frederic Weisbecker | 034cd8a | 2017-01-31 04:09:32 +0100 | [diff] [blame] | 7 | #include <linux/kernel_stat.h> |
Steven Rostedt (Red Hat) | 3866e84 | 2016-02-22 16:26:51 -0500 | [diff] [blame] | 8 | #include <linux/binfmts.h> |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 9 | #include <linux/mutex.h> |
Johannes Weiner | 3df0e59 | 2018-10-26 15:06:27 -0700 | [diff] [blame] | 10 | #include <linux/psi.h> |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 11 | #include <linux/spinlock.h> |
| 12 | #include <linux/stop_machine.h> |
Steven Rostedt | b6366f0 | 2015-03-18 14:49:46 -0400 | [diff] [blame] | 13 | #include <linux/irq_work.h> |
Frederic Weisbecker | 9f3660c | 2013-04-20 14:35:09 +0200 | [diff] [blame] | 14 | #include <linux/tick.h> |
Mel Gorman | f809ca9 | 2013-10-07 11:28:57 +0100 | [diff] [blame] | 15 | #include <linux/slab.h> |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 16 | |
Peter Zijlstra | 391e43d | 2011-11-15 17:14:39 +0100 | [diff] [blame] | 17 | #include "cpupri.h" |
Juri Lelli | 6bfd6d7 | 2013-11-07 14:43:47 +0100 | [diff] [blame] | 18 | #include "cpudeadline.h" |
Li Zefan | 60fed78 | 2013-03-29 14:36:43 +0800 | [diff] [blame] | 19 | #include "cpuacct.h" |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 20 | |
Peter Zijlstra | 9148a3a | 2016-09-20 22:34:51 +0200 | [diff] [blame] | 21 | #ifdef CONFIG_SCHED_DEBUG |
| 22 | #define SCHED_WARN_ON(x) WARN_ONCE(x, #x) |
| 23 | #else |
| 24 | #define SCHED_WARN_ON(x) ((void)(x)) |
| 25 | #endif |
| 26 | |
Paul Gortmaker | 45ceebf | 2013-04-19 15:10:49 -0400 | [diff] [blame] | 27 | struct rq; |
Daniel Lezcano | 442bf3a | 2014-09-04 11:32:09 -0400 | [diff] [blame] | 28 | struct cpuidle_state; |
Paul Gortmaker | 45ceebf | 2013-04-19 15:10:49 -0400 | [diff] [blame] | 29 | |
Kirill Tkhai | da0c1e6 | 2014-08-20 13:47:32 +0400 | [diff] [blame] | 30 | /* task_struct::on_rq states: */ |
| 31 | #define TASK_ON_RQ_QUEUED 1 |
Kirill Tkhai | cca26e8 | 2014-08-20 13:47:42 +0400 | [diff] [blame] | 32 | #define TASK_ON_RQ_MIGRATING 2 |
Kirill Tkhai | da0c1e6 | 2014-08-20 13:47:32 +0400 | [diff] [blame] | 33 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 34 | extern __read_mostly int scheduler_running; |
| 35 | |
Paul Gortmaker | 45ceebf | 2013-04-19 15:10:49 -0400 | [diff] [blame] | 36 | extern unsigned long calc_load_update; |
| 37 | extern atomic_long_t calc_load_tasks; |
| 38 | |
Peter Zijlstra | 3289bdb | 2015-04-14 13:19:42 +0200 | [diff] [blame] | 39 | extern void calc_global_load_tick(struct rq *this_rq); |
Thomas Gleixner | d60585c | 2016-07-12 18:33:56 +0200 | [diff] [blame] | 40 | extern long calc_load_fold_active(struct rq *this_rq, long adjust); |
Peter Zijlstra | 3289bdb | 2015-04-14 13:19:42 +0200 | [diff] [blame] | 41 | |
| 42 | #ifdef CONFIG_SMP |
Frederic Weisbecker | cee1afc | 2016-04-13 15:56:50 +0200 | [diff] [blame] | 43 | extern void cpu_load_update_active(struct rq *this_rq); |
Joonwoo Park | 4f0693a | 2017-05-18 17:43:58 -0700 | [diff] [blame] | 44 | extern void check_for_migration(struct rq *rq, struct task_struct *p); |
Peter Zijlstra | 3289bdb | 2015-04-14 13:19:42 +0200 | [diff] [blame] | 45 | #else |
Frederic Weisbecker | cee1afc | 2016-04-13 15:56:50 +0200 | [diff] [blame] | 46 | static inline void cpu_load_update_active(struct rq *this_rq) { } |
Joonwoo Park | 4f0693a | 2017-05-18 17:43:58 -0700 | [diff] [blame] | 47 | static inline void check_for_migration(struct rq *rq, struct task_struct *p) { } |
Peter Zijlstra | 3289bdb | 2015-04-14 13:19:42 +0200 | [diff] [blame] | 48 | #endif |
Paul Gortmaker | 45ceebf | 2013-04-19 15:10:49 -0400 | [diff] [blame] | 49 | |
Greg Kroah-Hartman | a0a93e3 | 2017-07-19 09:58:49 +0200 | [diff] [blame] | 50 | #ifdef CONFIG_SCHED_SMT |
| 51 | extern void update_idle_core(struct rq *rq); |
| 52 | #else |
| 53 | static inline void update_idle_core(struct rq *rq) { } |
| 54 | #endif |
| 55 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 56 | /* |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 57 | * Helpers for converting nanosecond timing to jiffy resolution |
| 58 | */ |
| 59 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
| 60 | |
Li Zefan | cc1f4b1 | 2013-03-05 16:06:09 +0800 | [diff] [blame] | 61 | /* |
| 62 | * Increase resolution of nice-level calculations for 64-bit architectures. |
| 63 | * The extra resolution improves shares distribution and load balancing of |
| 64 | * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup |
| 65 | * hierarchies, especially on larger systems. This is not a user-visible change |
| 66 | * and does not change the user-interface for setting shares/weights. |
| 67 | * |
| 68 | * We increase resolution only if we have enough bits to allow this increased |
Peter Zijlstra | 2159197 | 2016-04-28 12:49:38 +0200 | [diff] [blame] | 69 | * resolution (i.e. 64bit). The costs for increasing resolution when 32bit are |
| 70 | * pretty high and the returns do not justify the increased costs. |
| 71 | * |
| 72 | * Really only required when CONFIG_FAIR_GROUP_SCHED is also set, but to |
| 73 | * increase coverage and consistency always enable it on 64bit platforms. |
Li Zefan | cc1f4b1 | 2013-03-05 16:06:09 +0800 | [diff] [blame] | 74 | */ |
Peter Zijlstra | 2159197 | 2016-04-28 12:49:38 +0200 | [diff] [blame] | 75 | #ifdef CONFIG_64BIT |
Yuyang Du | 172895e | 2016-04-05 12:12:27 +0800 | [diff] [blame] | 76 | # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT) |
Yuyang Du | 6ecdd74 | 2016-04-05 12:12:26 +0800 | [diff] [blame] | 77 | # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT) |
| 78 | # define scale_load_down(w) ((w) >> SCHED_FIXEDPOINT_SHIFT) |
Li Zefan | cc1f4b1 | 2013-03-05 16:06:09 +0800 | [diff] [blame] | 79 | #else |
Yuyang Du | 172895e | 2016-04-05 12:12:27 +0800 | [diff] [blame] | 80 | # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT) |
Li Zefan | cc1f4b1 | 2013-03-05 16:06:09 +0800 | [diff] [blame] | 81 | # define scale_load(w) (w) |
| 82 | # define scale_load_down(w) (w) |
| 83 | #endif |
| 84 | |
Yuyang Du | 6ecdd74 | 2016-04-05 12:12:26 +0800 | [diff] [blame] | 85 | /* |
Yuyang Du | 172895e | 2016-04-05 12:12:27 +0800 | [diff] [blame] | 86 | * Task weight (visible to users) and its load (invisible to users) have |
| 87 | * independent resolution, but they should be well calibrated. We use |
| 88 | * scale_load() and scale_load_down(w) to convert between them. The |
| 89 | * following must be true: |
| 90 | * |
| 91 | * scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD |
| 92 | * |
Yuyang Du | 6ecdd74 | 2016-04-05 12:12:26 +0800 | [diff] [blame] | 93 | */ |
Yuyang Du | 172895e | 2016-04-05 12:12:27 +0800 | [diff] [blame] | 94 | #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT) |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 95 | |
| 96 | /* |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 97 | * Single value that decides SCHED_DEADLINE internal math precision. |
| 98 | * 10 -> just above 1us |
| 99 | * 9 -> just above 0.5us |
| 100 | */ |
| 101 | #define DL_SCALE (10) |
| 102 | |
| 103 | /* |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 104 | * These are the 'tuning knobs' of the scheduler: |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 105 | */ |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 106 | |
| 107 | /* |
| 108 | * single value that denotes runtime == period, ie unlimited time. |
| 109 | */ |
| 110 | #define RUNTIME_INF ((u64)~0ULL) |
| 111 | |
Henrik Austad | 20f9cd2 | 2015-09-09 17:00:41 +0200 | [diff] [blame] | 112 | static inline int idle_policy(int policy) |
| 113 | { |
| 114 | return policy == SCHED_IDLE; |
| 115 | } |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 116 | static inline int fair_policy(int policy) |
| 117 | { |
| 118 | return policy == SCHED_NORMAL || policy == SCHED_BATCH; |
| 119 | } |
| 120 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 121 | static inline int rt_policy(int policy) |
| 122 | { |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 123 | return policy == SCHED_FIFO || policy == SCHED_RR; |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 124 | } |
| 125 | |
Dario Faggioli | aab03e0 | 2013-11-28 11:14:43 +0100 | [diff] [blame] | 126 | static inline int dl_policy(int policy) |
| 127 | { |
| 128 | return policy == SCHED_DEADLINE; |
| 129 | } |
Henrik Austad | 20f9cd2 | 2015-09-09 17:00:41 +0200 | [diff] [blame] | 130 | static inline bool valid_policy(int policy) |
| 131 | { |
| 132 | return idle_policy(policy) || fair_policy(policy) || |
| 133 | rt_policy(policy) || dl_policy(policy); |
| 134 | } |
Dario Faggioli | aab03e0 | 2013-11-28 11:14:43 +0100 | [diff] [blame] | 135 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 136 | static inline int task_has_rt_policy(struct task_struct *p) |
| 137 | { |
| 138 | return rt_policy(p->policy); |
| 139 | } |
| 140 | |
Dario Faggioli | aab03e0 | 2013-11-28 11:14:43 +0100 | [diff] [blame] | 141 | static inline int task_has_dl_policy(struct task_struct *p) |
| 142 | { |
| 143 | return dl_policy(p->policy); |
| 144 | } |
| 145 | |
Dario Faggioli | 2d3d891 | 2013-11-07 14:43:44 +0100 | [diff] [blame] | 146 | /* |
| 147 | * Tells if entity @a should preempt entity @b. |
| 148 | */ |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 149 | static inline bool |
| 150 | dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b) |
Dario Faggioli | 2d3d891 | 2013-11-07 14:43:44 +0100 | [diff] [blame] | 151 | { |
| 152 | return dl_time_before(a->deadline, b->deadline); |
| 153 | } |
| 154 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 155 | /* |
| 156 | * This is the priority-queue data structure of the RT scheduling class: |
| 157 | */ |
| 158 | struct rt_prio_array { |
| 159 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ |
| 160 | struct list_head queue[MAX_RT_PRIO]; |
| 161 | }; |
| 162 | |
| 163 | struct rt_bandwidth { |
| 164 | /* nests inside the rq lock: */ |
| 165 | raw_spinlock_t rt_runtime_lock; |
| 166 | ktime_t rt_period; |
| 167 | u64 rt_runtime; |
| 168 | struct hrtimer rt_period_timer; |
Peter Zijlstra | 4cfafd3 | 2015-05-14 12:23:11 +0200 | [diff] [blame] | 169 | unsigned int rt_period_active; |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 170 | }; |
Juri Lelli | a5e7be3 | 2014-09-19 10:22:39 +0100 | [diff] [blame] | 171 | |
| 172 | void __dl_clear_params(struct task_struct *p); |
| 173 | |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 174 | /* |
| 175 | * To keep the bandwidth of -deadline tasks and groups under control |
| 176 | * we need some place where: |
| 177 | * - store the maximum -deadline bandwidth of the system (the group); |
| 178 | * - cache the fraction of that bandwidth that is currently allocated. |
| 179 | * |
| 180 | * This is all done in the data structure below. It is similar to the |
| 181 | * one used for RT-throttling (rt_bandwidth), with the main difference |
| 182 | * that, since here we are only interested in admission control, we |
| 183 | * do not decrease any runtime while the group "executes", neither we |
| 184 | * need a timer to replenish it. |
| 185 | * |
| 186 | * With respect to SMP, the bandwidth is given on a per-CPU basis, |
| 187 | * meaning that: |
| 188 | * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU; |
| 189 | * - dl_total_bw array contains, in the i-eth element, the currently |
| 190 | * allocated bandwidth on the i-eth CPU. |
| 191 | * Moreover, groups consume bandwidth on each CPU, while tasks only |
| 192 | * consume bandwidth on the CPU they're running on. |
| 193 | * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw |
| 194 | * that will be shown the next time the proc or cgroup controls will |
| 195 | * be red. It on its turn can be changed by writing on its own |
| 196 | * control. |
| 197 | */ |
| 198 | struct dl_bandwidth { |
| 199 | raw_spinlock_t dl_runtime_lock; |
| 200 | u64 dl_runtime; |
| 201 | u64 dl_period; |
| 202 | }; |
| 203 | |
| 204 | static inline int dl_bandwidth_enabled(void) |
| 205 | { |
Peter Zijlstra | 1724813 | 2013-12-17 12:44:49 +0100 | [diff] [blame] | 206 | return sysctl_sched_rt_runtime >= 0; |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 207 | } |
| 208 | |
| 209 | extern struct dl_bw *dl_bw_of(int i); |
| 210 | |
| 211 | struct dl_bw { |
| 212 | raw_spinlock_t lock; |
| 213 | u64 bw, total_bw; |
| 214 | }; |
| 215 | |
Juri Lelli | 7f51412 | 2014-09-19 10:22:40 +0100 | [diff] [blame] | 216 | static inline |
| 217 | void __dl_clear(struct dl_bw *dl_b, u64 tsk_bw) |
| 218 | { |
| 219 | dl_b->total_bw -= tsk_bw; |
| 220 | } |
| 221 | |
| 222 | static inline |
| 223 | void __dl_add(struct dl_bw *dl_b, u64 tsk_bw) |
| 224 | { |
| 225 | dl_b->total_bw += tsk_bw; |
| 226 | } |
| 227 | |
| 228 | static inline |
| 229 | bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw) |
| 230 | { |
| 231 | return dl_b->bw != -1 && |
| 232 | dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw; |
| 233 | } |
| 234 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 235 | extern struct mutex sched_domains_mutex; |
| 236 | |
| 237 | #ifdef CONFIG_CGROUP_SCHED |
| 238 | |
| 239 | #include <linux/cgroup.h> |
Johannes Weiner | 3df0e59 | 2018-10-26 15:06:27 -0700 | [diff] [blame] | 240 | #include <linux/psi.h> |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 241 | |
| 242 | struct cfs_rq; |
| 243 | struct rt_rq; |
| 244 | |
Mike Galbraith | 35cf4e5 | 2012-08-07 05:00:13 +0200 | [diff] [blame] | 245 | extern struct list_head task_groups; |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 246 | |
| 247 | struct cfs_bandwidth { |
| 248 | #ifdef CONFIG_CFS_BANDWIDTH |
| 249 | raw_spinlock_t lock; |
| 250 | ktime_t period; |
| 251 | u64 quota, runtime; |
Zhihui Zhang | 9c58c79 | 2014-09-20 21:24:36 -0400 | [diff] [blame] | 252 | s64 hierarchical_quota; |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 253 | u64 runtime_expires; |
| 254 | |
Peter Zijlstra | 4cfafd3 | 2015-05-14 12:23:11 +0200 | [diff] [blame] | 255 | int idle, period_active; |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 256 | struct hrtimer period_timer, slack_timer; |
| 257 | struct list_head throttled_cfs_rq; |
| 258 | |
| 259 | /* statistics */ |
| 260 | int nr_periods, nr_throttled; |
| 261 | u64 throttled_time; |
Phil Auld | bc1fccc | 2018-10-08 10:36:40 -0400 | [diff] [blame] | 262 | |
| 263 | bool distribute_running; |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 264 | #endif |
| 265 | }; |
| 266 | |
| 267 | /* task group related information */ |
| 268 | struct task_group { |
| 269 | struct cgroup_subsys_state css; |
| 270 | |
| 271 | #ifdef CONFIG_FAIR_GROUP_SCHED |
| 272 | /* schedulable entities of this group on each cpu */ |
| 273 | struct sched_entity **se; |
| 274 | /* runqueue "owned" by this group on each cpu */ |
| 275 | struct cfs_rq **cfs_rq; |
| 276 | unsigned long shares; |
| 277 | |
Alex Shi | fa6bdde | 2013-06-20 10:18:46 +0800 | [diff] [blame] | 278 | #ifdef CONFIG_SMP |
Waiman Long | b036762 | 2015-12-02 13:41:49 -0500 | [diff] [blame] | 279 | /* |
| 280 | * load_avg can be heavily contended at clock tick time, so put |
| 281 | * it in its own cacheline separated from the fields above which |
| 282 | * will also be accessed at each tick. |
| 283 | */ |
| 284 | atomic_long_t load_avg ____cacheline_aligned; |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 285 | #endif |
Alex Shi | fa6bdde | 2013-06-20 10:18:46 +0800 | [diff] [blame] | 286 | #endif |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 287 | |
| 288 | #ifdef CONFIG_RT_GROUP_SCHED |
| 289 | struct sched_rt_entity **rt_se; |
| 290 | struct rt_rq **rt_rq; |
| 291 | |
| 292 | struct rt_bandwidth rt_bandwidth; |
| 293 | #endif |
| 294 | |
| 295 | struct rcu_head rcu; |
| 296 | struct list_head list; |
| 297 | |
| 298 | struct task_group *parent; |
| 299 | struct list_head siblings; |
| 300 | struct list_head children; |
| 301 | |
| 302 | #ifdef CONFIG_SCHED_AUTOGROUP |
| 303 | struct autogroup *autogroup; |
| 304 | #endif |
| 305 | |
| 306 | struct cfs_bandwidth cfs_bandwidth; |
| 307 | }; |
| 308 | |
| 309 | #ifdef CONFIG_FAIR_GROUP_SCHED |
| 310 | #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD |
| 311 | |
| 312 | /* |
| 313 | * A weight of 0 or 1 can cause arithmetics problems. |
| 314 | * A weight of a cfs_rq is the sum of weights of which entities |
| 315 | * are queued on this cfs_rq, so a weight of a entity should not be |
| 316 | * too large, so as the shares value of a task group. |
| 317 | * (The default weight is 1024 - so there's no practical |
| 318 | * limitation from this.) |
| 319 | */ |
| 320 | #define MIN_SHARES (1UL << 1) |
| 321 | #define MAX_SHARES (1UL << 18) |
| 322 | #endif |
| 323 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 324 | typedef int (*tg_visitor)(struct task_group *, void *); |
| 325 | |
| 326 | extern int walk_tg_tree_from(struct task_group *from, |
| 327 | tg_visitor down, tg_visitor up, void *data); |
| 328 | |
| 329 | /* |
| 330 | * Iterate the full tree, calling @down when first entering a node and @up when |
| 331 | * leaving it for the final time. |
| 332 | * |
| 333 | * Caller must hold rcu_lock or sufficient equivalent. |
| 334 | */ |
| 335 | static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
| 336 | { |
| 337 | return walk_tg_tree_from(&root_task_group, down, up, data); |
| 338 | } |
| 339 | |
| 340 | extern int tg_nop(struct task_group *tg, void *data); |
| 341 | |
| 342 | extern void free_fair_sched_group(struct task_group *tg); |
| 343 | extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent); |
Peter Zijlstra | 8663e24 | 2016-06-22 14:58:02 +0200 | [diff] [blame] | 344 | extern void online_fair_sched_group(struct task_group *tg); |
Peter Zijlstra | 6fe1f34 | 2016-01-21 22:24:16 +0100 | [diff] [blame] | 345 | extern void unregister_fair_sched_group(struct task_group *tg); |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 346 | extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
| 347 | struct sched_entity *se, int cpu, |
| 348 | struct sched_entity *parent); |
| 349 | extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b); |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 350 | |
| 351 | extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b); |
Peter Zijlstra | 77a4d1a | 2015-04-15 11:41:57 +0200 | [diff] [blame] | 352 | extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b); |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 353 | extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq); |
| 354 | |
| 355 | extern void free_rt_sched_group(struct task_group *tg); |
| 356 | extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent); |
| 357 | extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
| 358 | struct sched_rt_entity *rt_se, int cpu, |
| 359 | struct sched_rt_entity *parent); |
| 360 | |
Li Zefan | 25cc7da | 2013-03-05 16:07:33 +0800 | [diff] [blame] | 361 | extern struct task_group *sched_create_group(struct task_group *parent); |
| 362 | extern void sched_online_group(struct task_group *tg, |
| 363 | struct task_group *parent); |
| 364 | extern void sched_destroy_group(struct task_group *tg); |
| 365 | extern void sched_offline_group(struct task_group *tg); |
| 366 | |
| 367 | extern void sched_move_task(struct task_struct *tsk); |
| 368 | |
| 369 | #ifdef CONFIG_FAIR_GROUP_SCHED |
| 370 | extern int sched_group_set_shares(struct task_group *tg, unsigned long shares); |
Byungchul Park | ad936d8 | 2015-10-24 01:16:19 +0900 | [diff] [blame] | 371 | |
| 372 | #ifdef CONFIG_SMP |
| 373 | extern void set_task_rq_fair(struct sched_entity *se, |
| 374 | struct cfs_rq *prev, struct cfs_rq *next); |
| 375 | #else /* !CONFIG_SMP */ |
| 376 | static inline void set_task_rq_fair(struct sched_entity *se, |
| 377 | struct cfs_rq *prev, struct cfs_rq *next) { } |
| 378 | #endif /* CONFIG_SMP */ |
| 379 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
Li Zefan | 25cc7da | 2013-03-05 16:07:33 +0800 | [diff] [blame] | 380 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 381 | #else /* CONFIG_CGROUP_SCHED */ |
| 382 | |
| 383 | struct cfs_bandwidth { }; |
| 384 | |
| 385 | #endif /* CONFIG_CGROUP_SCHED */ |
| 386 | |
| 387 | /* CFS-related fields in a runqueue */ |
| 388 | struct cfs_rq { |
| 389 | struct load_weight load; |
Peter Zijlstra | c82513e | 2012-04-26 13:12:27 +0200 | [diff] [blame] | 390 | unsigned int nr_running, h_nr_running; |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 391 | |
| 392 | u64 exec_clock; |
| 393 | u64 min_vruntime; |
| 394 | #ifndef CONFIG_64BIT |
| 395 | u64 min_vruntime_copy; |
| 396 | #endif |
| 397 | |
| 398 | struct rb_root tasks_timeline; |
| 399 | struct rb_node *rb_leftmost; |
| 400 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 401 | /* |
| 402 | * 'curr' points to currently running entity on this cfs_rq. |
| 403 | * It is set to NULL otherwise (i.e when none are currently running). |
| 404 | */ |
| 405 | struct sched_entity *curr, *next, *last, *skip; |
| 406 | |
| 407 | #ifdef CONFIG_SCHED_DEBUG |
| 408 | unsigned int nr_spread_over; |
| 409 | #endif |
| 410 | |
Paul Turner | 2dac754 | 2012-10-04 13:18:30 +0200 | [diff] [blame] | 411 | #ifdef CONFIG_SMP |
| 412 | /* |
Yuyang Du | 9d89c25 | 2015-07-15 08:04:37 +0800 | [diff] [blame] | 413 | * CFS load tracking |
Paul Turner | 2dac754 | 2012-10-04 13:18:30 +0200 | [diff] [blame] | 414 | */ |
Yuyang Du | 9d89c25 | 2015-07-15 08:04:37 +0800 | [diff] [blame] | 415 | struct sched_avg avg; |
Yuyang Du | 1396223 | 2015-07-15 08:04:41 +0800 | [diff] [blame] | 416 | u64 runnable_load_sum; |
| 417 | unsigned long runnable_load_avg; |
Yuyang Du | 9d89c25 | 2015-07-15 08:04:37 +0800 | [diff] [blame] | 418 | #ifdef CONFIG_FAIR_GROUP_SCHED |
| 419 | unsigned long tg_load_avg_contrib; |
Vincent Guittot | 96956e2 | 2016-11-08 10:53:44 +0100 | [diff] [blame] | 420 | unsigned long propagate_avg; |
Yuyang Du | 9d89c25 | 2015-07-15 08:04:37 +0800 | [diff] [blame] | 421 | #endif |
| 422 | atomic_long_t removed_load_avg, removed_util_avg; |
| 423 | #ifndef CONFIG_64BIT |
| 424 | u64 load_last_update_time_copy; |
| 425 | #endif |
Alex Shi | 141965c | 2013-06-26 13:05:39 +0800 | [diff] [blame] | 426 | |
Paul Turner | c566e8e | 2012-10-04 13:18:30 +0200 | [diff] [blame] | 427 | #ifdef CONFIG_FAIR_GROUP_SCHED |
Paul Turner | 8295836 | 2012-10-04 13:18:31 +0200 | [diff] [blame] | 428 | /* |
| 429 | * h_load = weight * f(tg) |
| 430 | * |
| 431 | * Where f(tg) is the recursive weight fraction assigned to |
| 432 | * this group. |
| 433 | */ |
| 434 | unsigned long h_load; |
Vladimir Davydov | 6852079 | 2013-07-15 17:49:19 +0400 | [diff] [blame] | 435 | u64 last_h_load_update; |
| 436 | struct sched_entity *h_load_next; |
| 437 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
Paul Turner | 8295836 | 2012-10-04 13:18:31 +0200 | [diff] [blame] | 438 | #endif /* CONFIG_SMP */ |
| 439 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 440 | #ifdef CONFIG_FAIR_GROUP_SCHED |
| 441 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
| 442 | |
| 443 | /* |
| 444 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in |
| 445 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
| 446 | * (like users, containers etc.) |
| 447 | * |
| 448 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This |
| 449 | * list is used during load balance. |
| 450 | */ |
| 451 | int on_list; |
| 452 | struct list_head leaf_cfs_rq_list; |
| 453 | struct task_group *tg; /* group that "owns" this runqueue */ |
| 454 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 455 | #ifdef CONFIG_CFS_BANDWIDTH |
| 456 | int runtime_enabled; |
| 457 | u64 runtime_expires; |
| 458 | s64 runtime_remaining; |
| 459 | |
Paul Turner | f1b1728 | 2012-10-04 13:18:31 +0200 | [diff] [blame] | 460 | u64 throttled_clock, throttled_clock_task; |
| 461 | u64 throttled_clock_task_time; |
Peter Zijlstra | 55e16d3 | 2016-06-22 15:14:26 +0200 | [diff] [blame] | 462 | int throttled, throttle_count; |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 463 | struct list_head throttled_list; |
Satya Durga Srinivasu Prabhala | a56900fa | 2017-11-08 12:13:47 -0800 | [diff] [blame] | 464 | #ifdef CONFIG_SCHED_WALT |
| 465 | u64 cumulative_runnable_avg; |
| 466 | #endif /* CONFIG_SCHED_WALT */ |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 467 | #endif /* CONFIG_CFS_BANDWIDTH */ |
| 468 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
| 469 | }; |
| 470 | |
| 471 | static inline int rt_bandwidth_enabled(void) |
| 472 | { |
| 473 | return sysctl_sched_rt_runtime >= 0; |
| 474 | } |
| 475 | |
Steven Rostedt | b6366f0 | 2015-03-18 14:49:46 -0400 | [diff] [blame] | 476 | /* RT IPI pull logic requires IRQ_WORK */ |
Steven Rostedt (Red Hat) | 1c37ff7 | 2017-10-06 14:05:04 -0400 | [diff] [blame] | 477 | #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP) |
Steven Rostedt | b6366f0 | 2015-03-18 14:49:46 -0400 | [diff] [blame] | 478 | # define HAVE_RT_PUSH_IPI |
| 479 | #endif |
| 480 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 481 | /* Real-Time classes' related field in a runqueue: */ |
| 482 | struct rt_rq { |
| 483 | struct rt_prio_array active; |
Peter Zijlstra | c82513e | 2012-04-26 13:12:27 +0200 | [diff] [blame] | 484 | unsigned int rt_nr_running; |
Frederic Weisbecker | 01d36d0 | 2015-11-04 18:17:10 +0100 | [diff] [blame] | 485 | unsigned int rr_nr_running; |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 486 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
| 487 | struct { |
| 488 | int curr; /* highest queued rt task prio */ |
| 489 | #ifdef CONFIG_SMP |
| 490 | int next; /* next highest */ |
| 491 | #endif |
| 492 | } highest_prio; |
| 493 | #endif |
| 494 | #ifdef CONFIG_SMP |
| 495 | unsigned long rt_nr_migratory; |
| 496 | unsigned long rt_nr_total; |
| 497 | int overloaded; |
| 498 | struct plist_head pushable_tasks; |
Steven Rostedt | b6366f0 | 2015-03-18 14:49:46 -0400 | [diff] [blame] | 499 | #endif /* CONFIG_SMP */ |
Kirill Tkhai | f4ebcbc | 2014-03-15 02:15:00 +0400 | [diff] [blame] | 500 | int rt_queued; |
| 501 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 502 | int rt_throttled; |
| 503 | u64 rt_time; |
| 504 | u64 rt_runtime; |
| 505 | /* Nests inside the rq lock: */ |
| 506 | raw_spinlock_t rt_runtime_lock; |
| 507 | |
| 508 | #ifdef CONFIG_RT_GROUP_SCHED |
| 509 | unsigned long rt_nr_boosted; |
| 510 | |
| 511 | struct rq *rq; |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 512 | struct task_group *tg; |
| 513 | #endif |
| 514 | }; |
| 515 | |
Dario Faggioli | aab03e0 | 2013-11-28 11:14:43 +0100 | [diff] [blame] | 516 | /* Deadline class' related fields in a runqueue */ |
| 517 | struct dl_rq { |
| 518 | /* runqueue is an rbtree, ordered by deadline */ |
| 519 | struct rb_root rb_root; |
| 520 | struct rb_node *rb_leftmost; |
| 521 | |
| 522 | unsigned long dl_nr_running; |
Juri Lelli | 1baca4c | 2013-11-07 14:43:38 +0100 | [diff] [blame] | 523 | |
| 524 | #ifdef CONFIG_SMP |
| 525 | /* |
| 526 | * Deadline values of the currently executing and the |
| 527 | * earliest ready task on this rq. Caching these facilitates |
| 528 | * the decision wether or not a ready but not running task |
| 529 | * should migrate somewhere else. |
| 530 | */ |
| 531 | struct { |
| 532 | u64 curr; |
| 533 | u64 next; |
| 534 | } earliest_dl; |
| 535 | |
| 536 | unsigned long dl_nr_migratory; |
Juri Lelli | 1baca4c | 2013-11-07 14:43:38 +0100 | [diff] [blame] | 537 | int overloaded; |
| 538 | |
| 539 | /* |
| 540 | * Tasks on this rq that can be pushed away. They are kept in |
| 541 | * an rb-tree, ordered by tasks' deadlines, with caching |
| 542 | * of the leftmost (earliest deadline) element. |
| 543 | */ |
| 544 | struct rb_root pushable_dl_tasks_root; |
| 545 | struct rb_node *pushable_dl_tasks_leftmost; |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 546 | #else |
| 547 | struct dl_bw dl_bw; |
Juri Lelli | 1baca4c | 2013-11-07 14:43:38 +0100 | [diff] [blame] | 548 | #endif |
Dario Faggioli | aab03e0 | 2013-11-28 11:14:43 +0100 | [diff] [blame] | 549 | }; |
| 550 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 551 | #ifdef CONFIG_SMP |
| 552 | |
Dietmar Eggemann | bbb138b | 2015-09-26 18:19:54 +0100 | [diff] [blame] | 553 | struct max_cpu_capacity { |
| 554 | raw_spinlock_t lock; |
| 555 | unsigned long val; |
| 556 | int cpu; |
| 557 | }; |
| 558 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 559 | /* |
| 560 | * We add the notion of a root-domain which will be used to define per-domain |
| 561 | * variables. Each exclusive cpuset essentially defines an island domain by |
| 562 | * fully partitioning the member cpus from any other cpuset. Whenever a new |
| 563 | * exclusive cpuset is created, we also create and attach a new root-domain |
| 564 | * object. |
| 565 | * |
| 566 | */ |
| 567 | struct root_domain { |
| 568 | atomic_t refcount; |
| 569 | atomic_t rto_count; |
| 570 | struct rcu_head rcu; |
| 571 | cpumask_var_t span; |
| 572 | cpumask_var_t online; |
| 573 | |
Tim Chen | 4486edd | 2014-06-23 12:16:49 -0700 | [diff] [blame] | 574 | /* Indicate more than one runnable task for any CPU */ |
| 575 | bool overload; |
| 576 | |
Morten Rasmussen | a562dfc | 2015-05-09 16:49:57 +0100 | [diff] [blame] | 577 | /* Indicate one or more cpus over-utilized (tipping point) */ |
| 578 | bool overutilized; |
| 579 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 580 | /* |
Juri Lelli | 1baca4c | 2013-11-07 14:43:38 +0100 | [diff] [blame] | 581 | * The bit corresponding to a CPU gets set here if such CPU has more |
| 582 | * than one runnable -deadline task (as it is below for RT tasks). |
| 583 | */ |
| 584 | cpumask_var_t dlo_mask; |
| 585 | atomic_t dlo_count; |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 586 | struct dl_bw dl_bw; |
Juri Lelli | 6bfd6d7 | 2013-11-07 14:43:47 +0100 | [diff] [blame] | 587 | struct cpudl cpudl; |
Juri Lelli | 1baca4c | 2013-11-07 14:43:38 +0100 | [diff] [blame] | 588 | |
Steven Rostedt (Red Hat) | 1c37ff7 | 2017-10-06 14:05:04 -0400 | [diff] [blame] | 589 | #ifdef HAVE_RT_PUSH_IPI |
| 590 | /* |
| 591 | * For IPI pull requests, loop across the rto_mask. |
| 592 | */ |
| 593 | struct irq_work rto_push_work; |
| 594 | raw_spinlock_t rto_lock; |
| 595 | /* These are only updated and read within rto_lock */ |
| 596 | int rto_loop; |
| 597 | int rto_cpu; |
| 598 | /* These atomics are updated outside of a lock */ |
| 599 | atomic_t rto_loop_next; |
| 600 | atomic_t rto_loop_start; |
| 601 | #endif |
Juri Lelli | 1baca4c | 2013-11-07 14:43:38 +0100 | [diff] [blame] | 602 | /* |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 603 | * The "RT overload" flag: it gets set if a CPU has more than |
| 604 | * one runnable RT task. |
| 605 | */ |
| 606 | cpumask_var_t rto_mask; |
| 607 | struct cpupri cpupri; |
Dietmar Eggemann | cd92bfd | 2016-08-01 19:53:35 +0100 | [diff] [blame] | 608 | |
Dietmar Eggemann | bbb138b | 2015-09-26 18:19:54 +0100 | [diff] [blame] | 609 | /* Maximum cpu capacity in the system. */ |
| 610 | struct max_cpu_capacity max_cpu_capacity; |
Dietmar Eggemann | 14774e7 | 2017-01-08 16:16:59 +0000 | [diff] [blame] | 611 | |
| 612 | /* First cpu with maximum and minimum original capacity */ |
| 613 | int max_cap_orig_cpu, min_cap_orig_cpu; |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 614 | }; |
| 615 | |
| 616 | extern struct root_domain def_root_domain; |
Steven Rostedt (VMware) | a384e54 | 2018-01-23 20:45:38 -0500 | [diff] [blame] | 617 | extern void sched_get_rd(struct root_domain *rd); |
| 618 | extern void sched_put_rd(struct root_domain *rd); |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 619 | |
Steven Rostedt (Red Hat) | 1c37ff7 | 2017-10-06 14:05:04 -0400 | [diff] [blame] | 620 | #ifdef HAVE_RT_PUSH_IPI |
| 621 | extern void rto_push_irq_work_func(struct irq_work *work); |
| 622 | #endif |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 623 | #endif /* CONFIG_SMP */ |
| 624 | |
| 625 | /* |
| 626 | * This is the main, per-CPU runqueue data structure. |
| 627 | * |
| 628 | * Locking rule: those places that want to lock multiple runqueues |
| 629 | * (such as the load balancing or the thread migration code), lock |
| 630 | * acquire operations must be ordered by ascending &runqueue. |
| 631 | */ |
| 632 | struct rq { |
| 633 | /* runqueue lock: */ |
| 634 | raw_spinlock_t lock; |
| 635 | |
| 636 | /* |
| 637 | * nr_running and cpu_load should be in the same cacheline because |
| 638 | * remote CPUs use both these fields when doing load calculation. |
| 639 | */ |
Peter Zijlstra | c82513e | 2012-04-26 13:12:27 +0200 | [diff] [blame] | 640 | unsigned int nr_running; |
Peter Zijlstra | 0ec8aa0 | 2013-10-07 11:29:33 +0100 | [diff] [blame] | 641 | #ifdef CONFIG_NUMA_BALANCING |
| 642 | unsigned int nr_numa_running; |
| 643 | unsigned int nr_preferred_running; |
| 644 | #endif |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 645 | #define CPU_LOAD_IDX_MAX 5 |
| 646 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; |
Morten Rasmussen | 4c6a824 | 2016-02-25 12:47:54 +0000 | [diff] [blame] | 647 | unsigned int misfit_task; |
Frederic Weisbecker | 3451d02 | 2011-08-10 23:21:01 +0200 | [diff] [blame] | 648 | #ifdef CONFIG_NO_HZ_COMMON |
Frederic Weisbecker | 9fd81dd | 2016-04-19 17:36:51 +0200 | [diff] [blame] | 649 | #ifdef CONFIG_SMP |
| 650 | unsigned long last_load_update_tick; |
| 651 | #endif /* CONFIG_SMP */ |
Suresh Siddha | 1c792db | 2011-12-01 17:07:32 -0800 | [diff] [blame] | 652 | unsigned long nohz_flags; |
Frederic Weisbecker | 9fd81dd | 2016-04-19 17:36:51 +0200 | [diff] [blame] | 653 | #endif /* CONFIG_NO_HZ_COMMON */ |
Frederic Weisbecker | 265f22a | 2013-05-03 03:39:05 +0200 | [diff] [blame] | 654 | #ifdef CONFIG_NO_HZ_FULL |
| 655 | unsigned long last_sched_tick; |
| 656 | #endif |
Joseph Lo | 7750186 | 2013-04-22 14:39:18 +0800 | [diff] [blame] | 657 | |
| 658 | #ifdef CONFIG_CPU_QUIET |
| 659 | /* time-based average load */ |
| 660 | u64 nr_last_stamp; |
| 661 | u64 nr_running_integral; |
| 662 | seqcount_t ave_seqcnt; |
| 663 | #endif |
| 664 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 665 | /* capture load from *all* tasks on this cpu: */ |
| 666 | struct load_weight load; |
| 667 | unsigned long nr_load_updates; |
| 668 | u64 nr_switches; |
| 669 | |
| 670 | struct cfs_rq cfs; |
| 671 | struct rt_rq rt; |
Dario Faggioli | aab03e0 | 2013-11-28 11:14:43 +0100 | [diff] [blame] | 672 | struct dl_rq dl; |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 673 | |
| 674 | #ifdef CONFIG_FAIR_GROUP_SCHED |
| 675 | /* list of leaf cfs_rq on this cpu: */ |
| 676 | struct list_head leaf_cfs_rq_list; |
Vincent Guittot | 96956e2 | 2016-11-08 10:53:44 +0100 | [diff] [blame] | 677 | struct list_head *tmp_alone_branch; |
Peter Zijlstra | a35b646 | 2012-08-08 21:46:40 +0200 | [diff] [blame] | 678 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
| 679 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 680 | /* |
| 681 | * This is part of a global counter where only the total sum |
| 682 | * over all CPUs matters. A task can increase this counter on |
| 683 | * one CPU and if it got migrated afterwards it may decrease |
| 684 | * it on another CPU. Always updated under the runqueue lock: |
| 685 | */ |
| 686 | unsigned long nr_uninterruptible; |
| 687 | |
| 688 | struct task_struct *curr, *idle, *stop; |
| 689 | unsigned long next_balance; |
| 690 | struct mm_struct *prev_mm; |
| 691 | |
Peter Zijlstra | 9edfbfe | 2015-01-05 11:18:11 +0100 | [diff] [blame] | 692 | unsigned int clock_skip_update; |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 693 | u64 clock; |
| 694 | u64 clock_task; |
| 695 | |
| 696 | atomic_t nr_iowait; |
| 697 | |
| 698 | #ifdef CONFIG_SMP |
| 699 | struct root_domain *rd; |
| 700 | struct sched_domain *sd; |
| 701 | |
Nicolas Pitre | ced549f | 2014-05-26 18:19:38 -0400 | [diff] [blame] | 702 | unsigned long cpu_capacity; |
Vincent Guittot | ca6d75e | 2015-02-27 16:54:09 +0100 | [diff] [blame] | 703 | unsigned long cpu_capacity_orig; |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 704 | |
Peter Zijlstra | e3fca9e | 2015-06-11 14:46:37 +0200 | [diff] [blame] | 705 | struct callback_head *balance_callback; |
| 706 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 707 | unsigned char idle_balance; |
| 708 | /* For active balancing */ |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 709 | int active_balance; |
| 710 | int push_cpu; |
Srivatsa Vaddagiri | 70e14af | 2014-03-31 10:34:41 -0700 | [diff] [blame] | 711 | struct task_struct *push_task; |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 712 | struct cpu_stop_work active_balance_work; |
| 713 | /* cpu of this runqueue: */ |
| 714 | int cpu; |
| 715 | int online; |
| 716 | |
Peter Zijlstra | 367456c | 2012-02-20 21:49:09 +0100 | [diff] [blame] | 717 | struct list_head cfs_tasks; |
| 718 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 719 | u64 rt_avg; |
| 720 | u64 age_stamp; |
| 721 | u64 idle_stamp; |
| 722 | u64 avg_idle; |
Jason Low | 9bd721c | 2013-09-13 11:26:52 -0700 | [diff] [blame] | 723 | |
| 724 | /* This is used to determine avg_idle's max value */ |
| 725 | u64 max_idle_balance_cost; |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 726 | #endif |
| 727 | |
Srivatsa Vaddagiri | 26c2154 | 2016-05-31 09:08:38 -0700 | [diff] [blame] | 728 | #ifdef CONFIG_SCHED_WALT |
Srivatsa Vaddagiri | 26c2154 | 2016-05-31 09:08:38 -0700 | [diff] [blame] | 729 | u64 cumulative_runnable_avg; |
Srivatsa Vaddagiri | 26c2154 | 2016-05-31 09:08:38 -0700 | [diff] [blame] | 730 | u64 window_start; |
| 731 | u64 curr_runnable_sum; |
| 732 | u64 prev_runnable_sum; |
| 733 | u64 nt_curr_runnable_sum; |
| 734 | u64 nt_prev_runnable_sum; |
Srinath Sridharan | 3a73c96 | 2016-07-22 13:21:15 +0100 | [diff] [blame] | 735 | u64 cur_irqload; |
| 736 | u64 avg_irqload; |
| 737 | u64 irqload_ts; |
Joonwoo Park | 7f17fff | 2017-02-03 11:15:31 -0800 | [diff] [blame] | 738 | u64 cum_window_demand; |
Srivatsa Vaddagiri | 26c2154 | 2016-05-31 09:08:38 -0700 | [diff] [blame] | 739 | #endif /* CONFIG_SCHED_WALT */ |
| 740 | |
| 741 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 742 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
| 743 | u64 prev_irq_time; |
| 744 | #endif |
| 745 | #ifdef CONFIG_PARAVIRT |
| 746 | u64 prev_steal_time; |
| 747 | #endif |
| 748 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING |
| 749 | u64 prev_steal_time_rq; |
| 750 | #endif |
| 751 | |
| 752 | /* calc_load related fields */ |
| 753 | unsigned long calc_load_update; |
| 754 | long calc_load_active; |
| 755 | |
| 756 | #ifdef CONFIG_SCHED_HRTICK |
| 757 | #ifdef CONFIG_SMP |
| 758 | int hrtick_csd_pending; |
| 759 | struct call_single_data hrtick_csd; |
| 760 | #endif |
| 761 | struct hrtimer hrtick_timer; |
| 762 | #endif |
| 763 | |
| 764 | #ifdef CONFIG_SCHEDSTATS |
| 765 | /* latency stats */ |
| 766 | struct sched_info rq_sched_info; |
| 767 | unsigned long long rq_cpu_time; |
| 768 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ |
| 769 | |
| 770 | /* sys_sched_yield() stats */ |
| 771 | unsigned int yld_count; |
| 772 | |
| 773 | /* schedule() stats */ |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 774 | unsigned int sched_count; |
| 775 | unsigned int sched_goidle; |
| 776 | |
| 777 | /* try_to_wake_up() stats */ |
| 778 | unsigned int ttwu_count; |
| 779 | unsigned int ttwu_local; |
Chris Redpath | 385dcec | 2017-06-03 15:03:03 +0100 | [diff] [blame] | 780 | #ifdef CONFIG_SMP |
Dietmar Eggemann | af88a16 | 2017-03-22 18:23:13 +0000 | [diff] [blame] | 781 | struct eas_stats eas_stats; |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 782 | #endif |
Chris Redpath | 385dcec | 2017-06-03 15:03:03 +0100 | [diff] [blame] | 783 | #endif |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 784 | |
| 785 | #ifdef CONFIG_SMP |
| 786 | struct llist_head wake_list; |
| 787 | #endif |
Daniel Lezcano | 442bf3a | 2014-09-04 11:32:09 -0400 | [diff] [blame] | 788 | |
| 789 | #ifdef CONFIG_CPU_IDLE |
| 790 | /* Must be inspected within a rcu lock section */ |
| 791 | struct cpuidle_state *idle_state; |
Morten Rasmussen | 0691064 | 2015-01-27 13:48:07 +0000 | [diff] [blame] | 792 | int idle_state_idx; |
Daniel Lezcano | 442bf3a | 2014-09-04 11:32:09 -0400 | [diff] [blame] | 793 | #endif |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 794 | }; |
| 795 | |
| 796 | static inline int cpu_of(struct rq *rq) |
| 797 | { |
| 798 | #ifdef CONFIG_SMP |
| 799 | return rq->cpu; |
| 800 | #else |
| 801 | return 0; |
| 802 | #endif |
| 803 | } |
| 804 | |
Pranith Kumar | 8b06c55 | 2014-08-13 13:28:12 -0400 | [diff] [blame] | 805 | DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 806 | |
Peter Zijlstra | 518cd62 | 2011-12-07 15:07:31 +0100 | [diff] [blame] | 807 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) |
Christoph Lameter | 4a32fea | 2014-08-17 12:30:27 -0500 | [diff] [blame] | 808 | #define this_rq() this_cpu_ptr(&runqueues) |
Peter Zijlstra | 518cd62 | 2011-12-07 15:07:31 +0100 | [diff] [blame] | 809 | #define task_rq(p) cpu_rq(task_cpu(p)) |
| 810 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) |
Christoph Lameter | 4a32fea | 2014-08-17 12:30:27 -0500 | [diff] [blame] | 811 | #define raw_rq() raw_cpu_ptr(&runqueues) |
Peter Zijlstra | 518cd62 | 2011-12-07 15:07:31 +0100 | [diff] [blame] | 812 | |
Johannes Weiner | 089605b | 2018-10-26 15:06:19 -0700 | [diff] [blame] | 813 | extern void update_rq_clock(struct rq *rq); |
| 814 | |
Peter Zijlstra | cebde6d | 2015-01-05 11:18:10 +0100 | [diff] [blame] | 815 | static inline u64 __rq_clock_broken(struct rq *rq) |
| 816 | { |
Jason Low | 316c1608d | 2015-04-28 13:00:20 -0700 | [diff] [blame] | 817 | return READ_ONCE(rq->clock); |
Peter Zijlstra | cebde6d | 2015-01-05 11:18:10 +0100 | [diff] [blame] | 818 | } |
| 819 | |
Frederic Weisbecker | 78becc2 | 2013-04-12 01:51:02 +0200 | [diff] [blame] | 820 | static inline u64 rq_clock(struct rq *rq) |
| 821 | { |
Peter Zijlstra | cebde6d | 2015-01-05 11:18:10 +0100 | [diff] [blame] | 822 | lockdep_assert_held(&rq->lock); |
Frederic Weisbecker | 78becc2 | 2013-04-12 01:51:02 +0200 | [diff] [blame] | 823 | return rq->clock; |
| 824 | } |
| 825 | |
| 826 | static inline u64 rq_clock_task(struct rq *rq) |
| 827 | { |
Peter Zijlstra | cebde6d | 2015-01-05 11:18:10 +0100 | [diff] [blame] | 828 | lockdep_assert_held(&rq->lock); |
Frederic Weisbecker | 78becc2 | 2013-04-12 01:51:02 +0200 | [diff] [blame] | 829 | return rq->clock_task; |
| 830 | } |
| 831 | |
Peter Zijlstra | 9edfbfe | 2015-01-05 11:18:11 +0100 | [diff] [blame] | 832 | #define RQCF_REQ_SKIP 0x01 |
| 833 | #define RQCF_ACT_SKIP 0x02 |
| 834 | |
| 835 | static inline void rq_clock_skip_update(struct rq *rq, bool skip) |
| 836 | { |
| 837 | lockdep_assert_held(&rq->lock); |
| 838 | if (skip) |
| 839 | rq->clock_skip_update |= RQCF_REQ_SKIP; |
| 840 | else |
| 841 | rq->clock_skip_update &= ~RQCF_REQ_SKIP; |
| 842 | } |
| 843 | |
Matt Fleming | 5a91d73 | 2016-09-21 14:38:10 +0100 | [diff] [blame] | 844 | struct rq_flags { |
| 845 | unsigned long flags; |
| 846 | struct pin_cookie cookie; |
| 847 | }; |
| 848 | |
| 849 | static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf) |
| 850 | { |
| 851 | rf->cookie = lockdep_pin_lock(&rq->lock); |
| 852 | } |
| 853 | |
| 854 | static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf) |
| 855 | { |
| 856 | lockdep_unpin_lock(&rq->lock, rf->cookie); |
| 857 | } |
| 858 | |
| 859 | static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf) |
| 860 | { |
| 861 | lockdep_repin_lock(&rq->lock, rf->cookie); |
| 862 | } |
| 863 | |
Johannes Weiner | 089605b | 2018-10-26 15:06:19 -0700 | [diff] [blame] | 864 | struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
| 865 | __acquires(rq->lock); |
| 866 | |
| 867 | struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
| 868 | __acquires(p->pi_lock) |
| 869 | __acquires(rq->lock); |
| 870 | |
| 871 | static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf) |
| 872 | __releases(rq->lock) |
| 873 | { |
| 874 | rq_unpin_lock(rq, rf); |
| 875 | raw_spin_unlock(&rq->lock); |
| 876 | } |
| 877 | |
| 878 | static inline void |
| 879 | task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf) |
| 880 | __releases(rq->lock) |
| 881 | __releases(p->pi_lock) |
| 882 | { |
| 883 | rq_unpin_lock(rq, rf); |
| 884 | raw_spin_unlock(&rq->lock); |
| 885 | raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags); |
| 886 | } |
| 887 | |
| 888 | static inline void |
| 889 | rq_lock_irqsave(struct rq *rq, struct rq_flags *rf) |
| 890 | __acquires(rq->lock) |
| 891 | { |
| 892 | raw_spin_lock_irqsave(&rq->lock, rf->flags); |
| 893 | rq_pin_lock(rq, rf); |
| 894 | } |
| 895 | |
| 896 | static inline void |
| 897 | rq_lock_irq(struct rq *rq, struct rq_flags *rf) |
| 898 | __acquires(rq->lock) |
| 899 | { |
| 900 | raw_spin_lock_irq(&rq->lock); |
| 901 | rq_pin_lock(rq, rf); |
| 902 | } |
| 903 | |
| 904 | static inline void |
| 905 | rq_lock(struct rq *rq, struct rq_flags *rf) |
| 906 | __acquires(rq->lock) |
| 907 | { |
| 908 | raw_spin_lock(&rq->lock); |
| 909 | rq_pin_lock(rq, rf); |
| 910 | } |
| 911 | |
| 912 | static inline void |
| 913 | rq_relock(struct rq *rq, struct rq_flags *rf) |
| 914 | __acquires(rq->lock) |
| 915 | { |
| 916 | raw_spin_lock(&rq->lock); |
| 917 | rq_repin_lock(rq, rf); |
| 918 | } |
| 919 | |
| 920 | static inline void |
| 921 | rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf) |
| 922 | __releases(rq->lock) |
| 923 | { |
| 924 | rq_unpin_lock(rq, rf); |
| 925 | raw_spin_unlock_irqrestore(&rq->lock, rf->flags); |
| 926 | } |
| 927 | |
| 928 | static inline void |
| 929 | rq_unlock_irq(struct rq *rq, struct rq_flags *rf) |
| 930 | __releases(rq->lock) |
| 931 | { |
| 932 | rq_unpin_lock(rq, rf); |
| 933 | raw_spin_unlock_irq(&rq->lock); |
| 934 | } |
| 935 | |
| 936 | static inline void |
| 937 | rq_unlock(struct rq *rq, struct rq_flags *rf) |
| 938 | __releases(rq->lock) |
| 939 | { |
| 940 | rq_unpin_lock(rq, rf); |
| 941 | raw_spin_unlock(&rq->lock); |
| 942 | } |
| 943 | |
Johannes Weiner | 6a99eb1 | 2018-10-26 15:06:23 -0700 | [diff] [blame] | 944 | static inline struct rq * |
| 945 | this_rq_lock_irq(struct rq_flags *rf) |
| 946 | __acquires(rq->lock) |
| 947 | { |
| 948 | struct rq *rq; |
| 949 | |
| 950 | local_irq_disable(); |
| 951 | rq = this_rq(); |
| 952 | rq_lock(rq, rf); |
| 953 | return rq; |
| 954 | } |
| 955 | |
Rik van Riel | 9942f79 | 2014-10-17 03:29:49 -0400 | [diff] [blame] | 956 | #ifdef CONFIG_NUMA |
Rik van Riel | e3fe70b | 2014-10-17 03:29:50 -0400 | [diff] [blame] | 957 | enum numa_topology_type { |
| 958 | NUMA_DIRECT, |
| 959 | NUMA_GLUELESS_MESH, |
| 960 | NUMA_BACKPLANE, |
| 961 | }; |
| 962 | extern enum numa_topology_type sched_numa_topology_type; |
Rik van Riel | 9942f79 | 2014-10-17 03:29:49 -0400 | [diff] [blame] | 963 | extern int sched_max_numa_distance; |
| 964 | extern bool find_numa_distance(int distance); |
| 965 | #endif |
| 966 | |
Mel Gorman | f809ca9 | 2013-10-07 11:28:57 +0100 | [diff] [blame] | 967 | #ifdef CONFIG_NUMA_BALANCING |
Iulia Manda | 44dba3d | 2014-10-31 02:13:31 +0200 | [diff] [blame] | 968 | /* The regions in numa_faults array from task_struct */ |
| 969 | enum numa_faults_stats { |
| 970 | NUMA_MEM = 0, |
| 971 | NUMA_CPU, |
| 972 | NUMA_MEMBUF, |
| 973 | NUMA_CPUBUF |
| 974 | }; |
Peter Zijlstra | 0ec8aa0 | 2013-10-07 11:29:33 +0100 | [diff] [blame] | 975 | extern void sched_setnuma(struct task_struct *p, int node); |
Mel Gorman | e6628d5 | 2013-10-07 11:29:02 +0100 | [diff] [blame] | 976 | extern int migrate_task_to(struct task_struct *p, int cpu); |
Peter Zijlstra | ac66f54 | 2013-10-07 11:29:16 +0100 | [diff] [blame] | 977 | extern int migrate_swap(struct task_struct *, struct task_struct *); |
Mel Gorman | f809ca9 | 2013-10-07 11:28:57 +0100 | [diff] [blame] | 978 | #endif /* CONFIG_NUMA_BALANCING */ |
| 979 | |
Peter Zijlstra | 518cd62 | 2011-12-07 15:07:31 +0100 | [diff] [blame] | 980 | #ifdef CONFIG_SMP |
| 981 | |
Peter Zijlstra | e3fca9e | 2015-06-11 14:46:37 +0200 | [diff] [blame] | 982 | static inline void |
| 983 | queue_balance_callback(struct rq *rq, |
| 984 | struct callback_head *head, |
| 985 | void (*func)(struct rq *rq)) |
| 986 | { |
| 987 | lockdep_assert_held(&rq->lock); |
| 988 | |
| 989 | if (unlikely(head->next)) |
| 990 | return; |
| 991 | |
| 992 | head->func = (void (*)(struct callback_head *))func; |
| 993 | head->next = rq->balance_callback; |
| 994 | rq->balance_callback = head; |
| 995 | } |
| 996 | |
Peter Zijlstra | e3baac4 | 2014-06-04 10:31:18 -0700 | [diff] [blame] | 997 | extern void sched_ttwu_pending(void); |
| 998 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 999 | #define rcu_dereference_check_sched_domain(p) \ |
| 1000 | rcu_dereference_check((p), \ |
| 1001 | lockdep_is_held(&sched_domains_mutex)) |
| 1002 | |
| 1003 | /* |
| 1004 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. |
| 1005 | * See detach_destroy_domains: synchronize_sched for details. |
| 1006 | * |
| 1007 | * The domain tree of any CPU may only be accessed from within |
| 1008 | * preempt-disabled sections. |
| 1009 | */ |
| 1010 | #define for_each_domain(cpu, __sd) \ |
Peter Zijlstra | 518cd62 | 2011-12-07 15:07:31 +0100 | [diff] [blame] | 1011 | for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \ |
| 1012 | __sd; __sd = __sd->parent) |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1013 | |
Suresh Siddha | 77e8136 | 2011-11-17 11:08:23 -0800 | [diff] [blame] | 1014 | #define for_each_lower_domain(sd) for (; sd; sd = sd->child) |
| 1015 | |
Peter Zijlstra | 518cd62 | 2011-12-07 15:07:31 +0100 | [diff] [blame] | 1016 | /** |
| 1017 | * highest_flag_domain - Return highest sched_domain containing flag. |
| 1018 | * @cpu: The cpu whose highest level of sched domain is to |
| 1019 | * be returned. |
| 1020 | * @flag: The flag to check for the highest sched_domain |
| 1021 | * for the given cpu. |
| 1022 | * |
| 1023 | * Returns the highest sched_domain of a cpu which contains the given flag. |
| 1024 | */ |
| 1025 | static inline struct sched_domain *highest_flag_domain(int cpu, int flag) |
| 1026 | { |
| 1027 | struct sched_domain *sd, *hsd = NULL; |
| 1028 | |
| 1029 | for_each_domain(cpu, sd) { |
| 1030 | if (!(sd->flags & flag)) |
| 1031 | break; |
| 1032 | hsd = sd; |
| 1033 | } |
| 1034 | |
| 1035 | return hsd; |
| 1036 | } |
| 1037 | |
Mel Gorman | fb13c7e | 2013-10-07 11:29:17 +0100 | [diff] [blame] | 1038 | static inline struct sched_domain *lowest_flag_domain(int cpu, int flag) |
| 1039 | { |
| 1040 | struct sched_domain *sd; |
| 1041 | |
| 1042 | for_each_domain(cpu, sd) { |
| 1043 | if (sd->flags & flag) |
| 1044 | break; |
| 1045 | } |
| 1046 | |
| 1047 | return sd; |
| 1048 | } |
| 1049 | |
Peter Zijlstra | 518cd62 | 2011-12-07 15:07:31 +0100 | [diff] [blame] | 1050 | DECLARE_PER_CPU(struct sched_domain *, sd_llc); |
Peter Zijlstra | 7d9ffa8 | 2013-07-04 12:56:46 +0800 | [diff] [blame] | 1051 | DECLARE_PER_CPU(int, sd_llc_size); |
Peter Zijlstra | 518cd62 | 2011-12-07 15:07:31 +0100 | [diff] [blame] | 1052 | DECLARE_PER_CPU(int, sd_llc_id); |
Peter Zijlstra | 0e369d7 | 2016-05-09 10:38:01 +0200 | [diff] [blame] | 1053 | DECLARE_PER_CPU(struct sched_domain_shared *, sd_llc_shared); |
Mel Gorman | fb13c7e | 2013-10-07 11:29:17 +0100 | [diff] [blame] | 1054 | DECLARE_PER_CPU(struct sched_domain *, sd_numa); |
Preeti U Murthy | 37dc6b5 | 2013-10-30 08:42:52 +0530 | [diff] [blame] | 1055 | DECLARE_PER_CPU(struct sched_domain *, sd_asym); |
Morten Rasmussen | 30786a0 | 2015-01-02 17:08:52 +0000 | [diff] [blame] | 1056 | DECLARE_PER_CPU(struct sched_domain *, sd_ea); |
Morten Rasmussen | 61bf625 | 2014-12-18 14:47:18 +0000 | [diff] [blame] | 1057 | DECLARE_PER_CPU(struct sched_domain *, sd_scs); |
Peter Zijlstra | 518cd62 | 2011-12-07 15:07:31 +0100 | [diff] [blame] | 1058 | |
Nicolas Pitre | 63b2ca3 | 2014-05-26 18:19:37 -0400 | [diff] [blame] | 1059 | struct sched_group_capacity { |
Li Zefan | 5e6521e | 2013-03-05 16:06:23 +0800 | [diff] [blame] | 1060 | atomic_t ref; |
| 1061 | /* |
Yuyang Du | 172895e | 2016-04-05 12:12:27 +0800 | [diff] [blame] | 1062 | * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity |
Nicolas Pitre | 63b2ca3 | 2014-05-26 18:19:37 -0400 | [diff] [blame] | 1063 | * for a single CPU. |
Li Zefan | 5e6521e | 2013-03-05 16:06:23 +0800 | [diff] [blame] | 1064 | */ |
Morten Rasmussen | 5cdeb5f | 2016-02-25 12:43:49 +0000 | [diff] [blame] | 1065 | unsigned long capacity; |
| 1066 | unsigned long max_capacity; /* Max per-cpu capacity in group */ |
Morten Rasmussen | 3d8cb90 | 2016-10-14 14:41:09 +0100 | [diff] [blame] | 1067 | unsigned long min_capacity; /* Min per-CPU capacity in group */ |
Li Zefan | 5e6521e | 2013-03-05 16:06:23 +0800 | [diff] [blame] | 1068 | unsigned long next_update; |
Nicolas Pitre | 63b2ca3 | 2014-05-26 18:19:37 -0400 | [diff] [blame] | 1069 | int imbalance; /* XXX unrelated to capacity but shared group state */ |
Li Zefan | 5e6521e | 2013-03-05 16:06:23 +0800 | [diff] [blame] | 1070 | |
| 1071 | unsigned long cpumask[0]; /* iteration mask */ |
| 1072 | }; |
| 1073 | |
| 1074 | struct sched_group { |
| 1075 | struct sched_group *next; /* Must be a circular list */ |
| 1076 | atomic_t ref; |
| 1077 | |
| 1078 | unsigned int group_weight; |
Nicolas Pitre | 63b2ca3 | 2014-05-26 18:19:37 -0400 | [diff] [blame] | 1079 | struct sched_group_capacity *sgc; |
Greg Hackmann | 2a3c6e6 | 2017-03-07 10:37:56 -0800 | [diff] [blame] | 1080 | const struct sched_group_energy *sge; |
Li Zefan | 5e6521e | 2013-03-05 16:06:23 +0800 | [diff] [blame] | 1081 | |
| 1082 | /* |
| 1083 | * The CPUs this group covers. |
| 1084 | * |
| 1085 | * NOTE: this field is variable length. (Allocated dynamically |
| 1086 | * by attaching extra space to the end of the structure, |
| 1087 | * depending on how many CPUs the kernel has booted up with) |
| 1088 | */ |
| 1089 | unsigned long cpumask[0]; |
| 1090 | }; |
| 1091 | |
| 1092 | static inline struct cpumask *sched_group_cpus(struct sched_group *sg) |
| 1093 | { |
| 1094 | return to_cpumask(sg->cpumask); |
| 1095 | } |
| 1096 | |
| 1097 | /* |
| 1098 | * cpumask masking which cpus in the group are allowed to iterate up the domain |
| 1099 | * tree. |
| 1100 | */ |
| 1101 | static inline struct cpumask *sched_group_mask(struct sched_group *sg) |
| 1102 | { |
Nicolas Pitre | 63b2ca3 | 2014-05-26 18:19:37 -0400 | [diff] [blame] | 1103 | return to_cpumask(sg->sgc->cpumask); |
Li Zefan | 5e6521e | 2013-03-05 16:06:23 +0800 | [diff] [blame] | 1104 | } |
| 1105 | |
| 1106 | /** |
| 1107 | * group_first_cpu - Returns the first cpu in the cpumask of a sched_group. |
| 1108 | * @group: The group whose first cpu is to be returned. |
| 1109 | */ |
| 1110 | static inline unsigned int group_first_cpu(struct sched_group *group) |
| 1111 | { |
| 1112 | return cpumask_first(sched_group_cpus(group)); |
| 1113 | } |
| 1114 | |
Peter Zijlstra | c117487 | 2012-05-31 14:47:33 +0200 | [diff] [blame] | 1115 | extern int group_balance_cpu(struct sched_group *sg); |
| 1116 | |
Steven Rostedt (Red Hat) | 3866e84 | 2016-02-22 16:26:51 -0500 | [diff] [blame] | 1117 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
| 1118 | void register_sched_domain_sysctl(void); |
| 1119 | void unregister_sched_domain_sysctl(void); |
| 1120 | #else |
| 1121 | static inline void register_sched_domain_sysctl(void) |
| 1122 | { |
| 1123 | } |
| 1124 | static inline void unregister_sched_domain_sysctl(void) |
| 1125 | { |
| 1126 | } |
| 1127 | #endif |
| 1128 | |
Peter Zijlstra | e3baac4 | 2014-06-04 10:31:18 -0700 | [diff] [blame] | 1129 | #else |
| 1130 | |
| 1131 | static inline void sched_ttwu_pending(void) { } |
| 1132 | |
Peter Zijlstra | 518cd62 | 2011-12-07 15:07:31 +0100 | [diff] [blame] | 1133 | #endif /* CONFIG_SMP */ |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1134 | |
Peter Zijlstra | 391e43d | 2011-11-15 17:14:39 +0100 | [diff] [blame] | 1135 | #include "stats.h" |
| 1136 | #include "auto_group.h" |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1137 | |
| 1138 | #ifdef CONFIG_CGROUP_SCHED |
| 1139 | |
| 1140 | /* |
| 1141 | * Return the group to which this tasks belongs. |
| 1142 | * |
Tejun Heo | 8af01f5 | 2013-08-08 20:11:22 -0400 | [diff] [blame] | 1143 | * We cannot use task_css() and friends because the cgroup subsystem |
| 1144 | * changes that value before the cgroup_subsys::attach() method is called, |
| 1145 | * therefore we cannot pin it and might observe the wrong value. |
Peter Zijlstra | 8323f26 | 2012-06-22 13:36:05 +0200 | [diff] [blame] | 1146 | * |
| 1147 | * The same is true for autogroup's p->signal->autogroup->tg, the autogroup |
| 1148 | * core changes this before calling sched_move_task(). |
| 1149 | * |
| 1150 | * Instead we use a 'copy' which is updated from sched_move_task() while |
| 1151 | * holding both task_struct::pi_lock and rq::lock. |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1152 | */ |
| 1153 | static inline struct task_group *task_group(struct task_struct *p) |
| 1154 | { |
Peter Zijlstra | 8323f26 | 2012-06-22 13:36:05 +0200 | [diff] [blame] | 1155 | return p->sched_task_group; |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1156 | } |
| 1157 | |
| 1158 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ |
| 1159 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) |
| 1160 | { |
| 1161 | #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED) |
| 1162 | struct task_group *tg = task_group(p); |
| 1163 | #endif |
| 1164 | |
| 1165 | #ifdef CONFIG_FAIR_GROUP_SCHED |
Byungchul Park | ad936d8 | 2015-10-24 01:16:19 +0900 | [diff] [blame] | 1166 | set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]); |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1167 | p->se.cfs_rq = tg->cfs_rq[cpu]; |
| 1168 | p->se.parent = tg->se[cpu]; |
| 1169 | #endif |
| 1170 | |
| 1171 | #ifdef CONFIG_RT_GROUP_SCHED |
| 1172 | p->rt.rt_rq = tg->rt_rq[cpu]; |
| 1173 | p->rt.parent = tg->rt_se[cpu]; |
| 1174 | #endif |
| 1175 | } |
| 1176 | |
| 1177 | #else /* CONFIG_CGROUP_SCHED */ |
| 1178 | |
| 1179 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } |
| 1180 | static inline struct task_group *task_group(struct task_struct *p) |
| 1181 | { |
| 1182 | return NULL; |
| 1183 | } |
| 1184 | |
| 1185 | #endif /* CONFIG_CGROUP_SCHED */ |
| 1186 | |
| 1187 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
| 1188 | { |
| 1189 | set_task_rq(p, cpu); |
| 1190 | #ifdef CONFIG_SMP |
| 1191 | /* |
| 1192 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be |
| 1193 | * successfuly executed on another CPU. We must ensure that updates of |
| 1194 | * per-task data have been completed by this moment. |
| 1195 | */ |
| 1196 | smp_wmb(); |
Andy Lutomirski | c65eacb | 2016-09-13 14:29:24 -0700 | [diff] [blame] | 1197 | #ifdef CONFIG_THREAD_INFO_IN_TASK |
| 1198 | p->cpu = cpu; |
| 1199 | #else |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1200 | task_thread_info(p)->cpu = cpu; |
Andy Lutomirski | c65eacb | 2016-09-13 14:29:24 -0700 | [diff] [blame] | 1201 | #endif |
Peter Zijlstra | ac66f54 | 2013-10-07 11:29:16 +0100 | [diff] [blame] | 1202 | p->wake_cpu = cpu; |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1203 | #endif |
| 1204 | } |
| 1205 | |
| 1206 | /* |
| 1207 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: |
| 1208 | */ |
| 1209 | #ifdef CONFIG_SCHED_DEBUG |
Ingo Molnar | c5905af | 2012-02-24 08:31:31 +0100 | [diff] [blame] | 1210 | # include <linux/static_key.h> |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1211 | # define const_debug __read_mostly |
| 1212 | #else |
| 1213 | # define const_debug const |
| 1214 | #endif |
| 1215 | |
| 1216 | extern const_debug unsigned int sysctl_sched_features; |
| 1217 | |
| 1218 | #define SCHED_FEAT(name, enabled) \ |
| 1219 | __SCHED_FEAT_##name , |
| 1220 | |
| 1221 | enum { |
Peter Zijlstra | 391e43d | 2011-11-15 17:14:39 +0100 | [diff] [blame] | 1222 | #include "features.h" |
Peter Zijlstra | f8b6d1c | 2011-07-06 14:20:14 +0200 | [diff] [blame] | 1223 | __SCHED_FEAT_NR, |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1224 | }; |
| 1225 | |
| 1226 | #undef SCHED_FEAT |
| 1227 | |
Peter Zijlstra | f8b6d1c | 2011-07-06 14:20:14 +0200 | [diff] [blame] | 1228 | #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL) |
Peter Zijlstra | f8b6d1c | 2011-07-06 14:20:14 +0200 | [diff] [blame] | 1229 | #define SCHED_FEAT(name, enabled) \ |
Ingo Molnar | c5905af | 2012-02-24 08:31:31 +0100 | [diff] [blame] | 1230 | static __always_inline bool static_branch_##name(struct static_key *key) \ |
Peter Zijlstra | f8b6d1c | 2011-07-06 14:20:14 +0200 | [diff] [blame] | 1231 | { \ |
Jason Baron | 6e76ea8 | 2014-07-02 15:52:41 +0000 | [diff] [blame] | 1232 | return static_key_##enabled(key); \ |
Peter Zijlstra | f8b6d1c | 2011-07-06 14:20:14 +0200 | [diff] [blame] | 1233 | } |
| 1234 | |
| 1235 | #include "features.h" |
| 1236 | |
| 1237 | #undef SCHED_FEAT |
| 1238 | |
Ingo Molnar | c5905af | 2012-02-24 08:31:31 +0100 | [diff] [blame] | 1239 | extern struct static_key sched_feat_keys[__SCHED_FEAT_NR]; |
Peter Zijlstra | f8b6d1c | 2011-07-06 14:20:14 +0200 | [diff] [blame] | 1240 | #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x])) |
| 1241 | #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */ |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1242 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) |
Peter Zijlstra | f8b6d1c | 2011-07-06 14:20:14 +0200 | [diff] [blame] | 1243 | #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */ |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1244 | |
Srikar Dronamraju | 2a59572 | 2015-08-11 21:54:21 +0530 | [diff] [blame] | 1245 | extern struct static_key_false sched_numa_balancing; |
Mel Gorman | cb25176 | 2016-02-05 09:08:36 +0000 | [diff] [blame] | 1246 | extern struct static_key_false sched_schedstats; |
Peter Zijlstra | cbee9f8 | 2012-10-25 14:16:43 +0200 | [diff] [blame] | 1247 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1248 | static inline u64 global_rt_period(void) |
| 1249 | { |
| 1250 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; |
| 1251 | } |
| 1252 | |
| 1253 | static inline u64 global_rt_runtime(void) |
| 1254 | { |
| 1255 | if (sysctl_sched_rt_runtime < 0) |
| 1256 | return RUNTIME_INF; |
| 1257 | |
| 1258 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; |
| 1259 | } |
| 1260 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1261 | static inline int task_current(struct rq *rq, struct task_struct *p) |
| 1262 | { |
| 1263 | return rq->curr == p; |
| 1264 | } |
| 1265 | |
| 1266 | static inline int task_running(struct rq *rq, struct task_struct *p) |
| 1267 | { |
| 1268 | #ifdef CONFIG_SMP |
| 1269 | return p->on_cpu; |
| 1270 | #else |
| 1271 | return task_current(rq, p); |
| 1272 | #endif |
| 1273 | } |
| 1274 | |
Kirill Tkhai | da0c1e6 | 2014-08-20 13:47:32 +0400 | [diff] [blame] | 1275 | static inline int task_on_rq_queued(struct task_struct *p) |
| 1276 | { |
| 1277 | return p->on_rq == TASK_ON_RQ_QUEUED; |
| 1278 | } |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1279 | |
Kirill Tkhai | cca26e8 | 2014-08-20 13:47:42 +0400 | [diff] [blame] | 1280 | static inline int task_on_rq_migrating(struct task_struct *p) |
| 1281 | { |
| 1282 | return p->on_rq == TASK_ON_RQ_MIGRATING; |
| 1283 | } |
| 1284 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1285 | #ifndef prepare_arch_switch |
| 1286 | # define prepare_arch_switch(next) do { } while (0) |
| 1287 | #endif |
Catalin Marinas | 01f23e1 | 2011-11-27 21:43:10 +0000 | [diff] [blame] | 1288 | #ifndef finish_arch_post_lock_switch |
| 1289 | # define finish_arch_post_lock_switch() do { } while (0) |
| 1290 | #endif |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1291 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1292 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
| 1293 | { |
| 1294 | #ifdef CONFIG_SMP |
| 1295 | /* |
| 1296 | * We can optimise this out completely for !SMP, because the |
| 1297 | * SMP rebalancing from interrupt is the only thing that cares |
| 1298 | * here. |
| 1299 | */ |
| 1300 | next->on_cpu = 1; |
| 1301 | #endif |
| 1302 | } |
| 1303 | |
| 1304 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
| 1305 | { |
| 1306 | #ifdef CONFIG_SMP |
| 1307 | /* |
| 1308 | * After ->on_cpu is cleared, the task can be moved to a different CPU. |
| 1309 | * We must ensure this doesn't happen until the switch is completely |
| 1310 | * finished. |
Peter Zijlstra | 95913d9 | 2015-09-29 14:45:09 +0200 | [diff] [blame] | 1311 | * |
Peter Zijlstra | b75a225 | 2015-10-06 14:36:17 +0200 | [diff] [blame] | 1312 | * In particular, the load of prev->state in finish_task_switch() must |
| 1313 | * happen before this. |
| 1314 | * |
Peter Zijlstra | 1f03e8d | 2016-04-04 10:57:12 +0200 | [diff] [blame] | 1315 | * Pairs with the smp_cond_load_acquire() in try_to_wake_up(). |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1316 | */ |
Peter Zijlstra | 95913d9 | 2015-09-29 14:45:09 +0200 | [diff] [blame] | 1317 | smp_store_release(&prev->on_cpu, 0); |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1318 | #endif |
| 1319 | #ifdef CONFIG_DEBUG_SPINLOCK |
| 1320 | /* this is a valid case when another task releases the spinlock */ |
| 1321 | rq->lock.owner = current; |
| 1322 | #endif |
| 1323 | /* |
| 1324 | * If we are tracking spinlock dependencies then we have to |
| 1325 | * fix up the runqueue lock - which gets 'carried over' from |
| 1326 | * prev into current: |
| 1327 | */ |
| 1328 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); |
| 1329 | |
| 1330 | raw_spin_unlock_irq(&rq->lock); |
| 1331 | } |
| 1332 | |
Li Zefan | b13095f | 2013-03-05 16:06:38 +0800 | [diff] [blame] | 1333 | /* |
| 1334 | * wake flags |
| 1335 | */ |
| 1336 | #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */ |
| 1337 | #define WF_FORK 0x02 /* child wakeup after fork */ |
| 1338 | #define WF_MIGRATED 0x4 /* internal use, task got migrated */ |
| 1339 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1340 | /* |
| 1341 | * To aid in avoiding the subversion of "niceness" due to uneven distribution |
| 1342 | * of tasks with abnormal "nice" values across CPUs the contribution that |
| 1343 | * each task makes to its run queue's load is weighted according to its |
| 1344 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
| 1345 | * scaled version of the new time slice allocation that they receive on time |
| 1346 | * slice expiry etc. |
| 1347 | */ |
| 1348 | |
| 1349 | #define WEIGHT_IDLEPRIO 3 |
| 1350 | #define WMULT_IDLEPRIO 1431655765 |
| 1351 | |
Andi Kleen | ed82b8a | 2015-11-29 20:59:43 -0800 | [diff] [blame] | 1352 | extern const int sched_prio_to_weight[40]; |
| 1353 | extern const u32 sched_prio_to_wmult[40]; |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1354 | |
Peter Zijlstra | ff77e46 | 2016-01-18 15:27:07 +0100 | [diff] [blame] | 1355 | /* |
| 1356 | * {de,en}queue flags: |
| 1357 | * |
| 1358 | * DEQUEUE_SLEEP - task is no longer runnable |
| 1359 | * ENQUEUE_WAKEUP - task just became runnable |
| 1360 | * |
| 1361 | * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks |
| 1362 | * are in a known state which allows modification. Such pairs |
| 1363 | * should preserve as much state as possible. |
| 1364 | * |
| 1365 | * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location |
| 1366 | * in the runqueue. |
| 1367 | * |
| 1368 | * ENQUEUE_HEAD - place at front of runqueue (tail if not specified) |
| 1369 | * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline) |
Peter Zijlstra | 59efa0b | 2016-05-10 18:24:37 +0200 | [diff] [blame] | 1370 | * ENQUEUE_MIGRATED - the task was migrated during wakeup |
Peter Zijlstra | ff77e46 | 2016-01-18 15:27:07 +0100 | [diff] [blame] | 1371 | * |
| 1372 | */ |
| 1373 | |
| 1374 | #define DEQUEUE_SLEEP 0x01 |
| 1375 | #define DEQUEUE_SAVE 0x02 /* matches ENQUEUE_RESTORE */ |
| 1376 | #define DEQUEUE_MOVE 0x04 /* matches ENQUEUE_MOVE */ |
| 1377 | |
Peter Zijlstra | 1de6444 | 2015-09-30 17:44:13 +0200 | [diff] [blame] | 1378 | #define ENQUEUE_WAKEUP 0x01 |
Peter Zijlstra | ff77e46 | 2016-01-18 15:27:07 +0100 | [diff] [blame] | 1379 | #define ENQUEUE_RESTORE 0x02 |
| 1380 | #define ENQUEUE_MOVE 0x04 |
| 1381 | |
| 1382 | #define ENQUEUE_HEAD 0x08 |
| 1383 | #define ENQUEUE_REPLENISH 0x10 |
Li Zefan | c82ba9f | 2013-03-05 16:06:55 +0800 | [diff] [blame] | 1384 | #ifdef CONFIG_SMP |
Peter Zijlstra | 59efa0b | 2016-05-10 18:24:37 +0200 | [diff] [blame] | 1385 | #define ENQUEUE_MIGRATED 0x20 |
Li Zefan | c82ba9f | 2013-03-05 16:06:55 +0800 | [diff] [blame] | 1386 | #else |
Peter Zijlstra | 59efa0b | 2016-05-10 18:24:37 +0200 | [diff] [blame] | 1387 | #define ENQUEUE_MIGRATED 0x00 |
Li Zefan | c82ba9f | 2013-03-05 16:06:55 +0800 | [diff] [blame] | 1388 | #endif |
Juri Lelli | 43aac89 | 2015-06-26 12:14:23 +0100 | [diff] [blame] | 1389 | #define ENQUEUE_WAKEUP_NEW 0x40 |
Li Zefan | c82ba9f | 2013-03-05 16:06:55 +0800 | [diff] [blame] | 1390 | |
Peter Zijlstra | 37e117c | 2014-02-14 12:25:08 +0100 | [diff] [blame] | 1391 | #define RETRY_TASK ((void *)-1UL) |
| 1392 | |
Li Zefan | c82ba9f | 2013-03-05 16:06:55 +0800 | [diff] [blame] | 1393 | struct sched_class { |
| 1394 | const struct sched_class *next; |
| 1395 | |
| 1396 | void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags); |
| 1397 | void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags); |
| 1398 | void (*yield_task) (struct rq *rq); |
| 1399 | bool (*yield_to_task) (struct rq *rq, struct task_struct *p, bool preempt); |
| 1400 | |
| 1401 | void (*check_preempt_curr) (struct rq *rq, struct task_struct *p, int flags); |
| 1402 | |
Peter Zijlstra | 606dba2 | 2012-02-11 06:05:00 +0100 | [diff] [blame] | 1403 | /* |
| 1404 | * It is the responsibility of the pick_next_task() method that will |
| 1405 | * return the next task to call put_prev_task() on the @prev task or |
| 1406 | * something equivalent. |
Peter Zijlstra | 37e117c | 2014-02-14 12:25:08 +0100 | [diff] [blame] | 1407 | * |
| 1408 | * May return RETRY_TASK when it finds a higher prio class has runnable |
| 1409 | * tasks. |
Peter Zijlstra | 606dba2 | 2012-02-11 06:05:00 +0100 | [diff] [blame] | 1410 | */ |
| 1411 | struct task_struct * (*pick_next_task) (struct rq *rq, |
Peter Zijlstra | e7904a2 | 2015-08-01 19:25:08 +0200 | [diff] [blame] | 1412 | struct task_struct *prev, |
Matt Fleming | 5a91d73 | 2016-09-21 14:38:10 +0100 | [diff] [blame] | 1413 | struct rq_flags *rf); |
Li Zefan | c82ba9f | 2013-03-05 16:06:55 +0800 | [diff] [blame] | 1414 | void (*put_prev_task) (struct rq *rq, struct task_struct *p); |
| 1415 | |
| 1416 | #ifdef CONFIG_SMP |
Peter Zijlstra | ac66f54 | 2013-10-07 11:29:16 +0100 | [diff] [blame] | 1417 | int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags); |
xiaofeng.yan | 5a4fd03 | 2015-09-23 14:55:59 +0800 | [diff] [blame] | 1418 | void (*migrate_task_rq)(struct task_struct *p); |
Li Zefan | c82ba9f | 2013-03-05 16:06:55 +0800 | [diff] [blame] | 1419 | |
Li Zefan | c82ba9f | 2013-03-05 16:06:55 +0800 | [diff] [blame] | 1420 | void (*task_woken) (struct rq *this_rq, struct task_struct *task); |
| 1421 | |
| 1422 | void (*set_cpus_allowed)(struct task_struct *p, |
| 1423 | const struct cpumask *newmask); |
| 1424 | |
| 1425 | void (*rq_online)(struct rq *rq); |
| 1426 | void (*rq_offline)(struct rq *rq); |
| 1427 | #endif |
| 1428 | |
| 1429 | void (*set_curr_task) (struct rq *rq); |
| 1430 | void (*task_tick) (struct rq *rq, struct task_struct *p, int queued); |
| 1431 | void (*task_fork) (struct task_struct *p); |
Dario Faggioli | e6c390f | 2013-11-07 14:43:35 +0100 | [diff] [blame] | 1432 | void (*task_dead) (struct task_struct *p); |
Li Zefan | c82ba9f | 2013-03-05 16:06:55 +0800 | [diff] [blame] | 1433 | |
Kirill Tkhai | 67dfa1b | 2014-10-27 17:40:52 +0300 | [diff] [blame] | 1434 | /* |
| 1435 | * The switched_from() call is allowed to drop rq->lock, therefore we |
| 1436 | * cannot assume the switched_from/switched_to pair is serliazed by |
| 1437 | * rq->lock. They are however serialized by p->pi_lock. |
| 1438 | */ |
Li Zefan | c82ba9f | 2013-03-05 16:06:55 +0800 | [diff] [blame] | 1439 | void (*switched_from) (struct rq *this_rq, struct task_struct *task); |
| 1440 | void (*switched_to) (struct rq *this_rq, struct task_struct *task); |
| 1441 | void (*prio_changed) (struct rq *this_rq, struct task_struct *task, |
| 1442 | int oldprio); |
| 1443 | |
| 1444 | unsigned int (*get_rr_interval) (struct rq *rq, |
| 1445 | struct task_struct *task); |
| 1446 | |
Stanislaw Gruszka | 6e99891 | 2014-11-12 16:58:44 +0100 | [diff] [blame] | 1447 | void (*update_curr) (struct rq *rq); |
| 1448 | |
Vincent Guittot | ea86cb4 | 2016-06-17 13:38:55 +0200 | [diff] [blame] | 1449 | #define TASK_SET_GROUP 0 |
| 1450 | #define TASK_MOVE_GROUP 1 |
| 1451 | |
Li Zefan | c82ba9f | 2013-03-05 16:06:55 +0800 | [diff] [blame] | 1452 | #ifdef CONFIG_FAIR_GROUP_SCHED |
Vincent Guittot | ea86cb4 | 2016-06-17 13:38:55 +0200 | [diff] [blame] | 1453 | void (*task_change_group) (struct task_struct *p, int type); |
Li Zefan | c82ba9f | 2013-03-05 16:06:55 +0800 | [diff] [blame] | 1454 | #endif |
| 1455 | }; |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1456 | |
Peter Zijlstra | 3f1d2a3 | 2014-02-12 10:49:30 +0100 | [diff] [blame] | 1457 | static inline void put_prev_task(struct rq *rq, struct task_struct *prev) |
| 1458 | { |
| 1459 | prev->sched_class->put_prev_task(rq, prev); |
| 1460 | } |
| 1461 | |
Peter Zijlstra | b2bf6c3 | 2016-09-20 22:00:38 +0200 | [diff] [blame] | 1462 | static inline void set_curr_task(struct rq *rq, struct task_struct *curr) |
| 1463 | { |
| 1464 | curr->sched_class->set_curr_task(rq); |
| 1465 | } |
| 1466 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1467 | #define sched_class_highest (&stop_sched_class) |
| 1468 | #define for_each_class(class) \ |
| 1469 | for (class = sched_class_highest; class; class = class->next) |
| 1470 | |
| 1471 | extern const struct sched_class stop_sched_class; |
Dario Faggioli | aab03e0 | 2013-11-28 11:14:43 +0100 | [diff] [blame] | 1472 | extern const struct sched_class dl_sched_class; |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1473 | extern const struct sched_class rt_sched_class; |
| 1474 | extern const struct sched_class fair_sched_class; |
| 1475 | extern const struct sched_class idle_sched_class; |
| 1476 | |
| 1477 | |
| 1478 | #ifdef CONFIG_SMP |
| 1479 | |
Patrick Bellasi | 2178e84 | 2016-07-22 11:35:59 +0100 | [diff] [blame] | 1480 | extern void init_max_cpu_capacity(struct max_cpu_capacity *mcc); |
Nicolas Pitre | 63b2ca3 | 2014-05-26 18:19:37 -0400 | [diff] [blame] | 1481 | extern void update_group_capacity(struct sched_domain *sd, int cpu); |
Li Zefan | b719203 | 2013-03-07 10:00:26 +0800 | [diff] [blame] | 1482 | |
Daniel Lezcano | 7caff66 | 2014-01-06 12:34:38 +0100 | [diff] [blame] | 1483 | extern void trigger_load_balance(struct rq *rq); |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1484 | |
Peter Zijlstra | c5b2803 | 2015-05-15 17:43:35 +0200 | [diff] [blame] | 1485 | extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask); |
| 1486 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1487 | #endif |
| 1488 | |
Daniel Lezcano | 442bf3a | 2014-09-04 11:32:09 -0400 | [diff] [blame] | 1489 | #ifdef CONFIG_CPU_IDLE |
| 1490 | static inline void idle_set_state(struct rq *rq, |
| 1491 | struct cpuidle_state *idle_state) |
| 1492 | { |
| 1493 | rq->idle_state = idle_state; |
| 1494 | } |
| 1495 | |
| 1496 | static inline struct cpuidle_state *idle_get_state(struct rq *rq) |
| 1497 | { |
Peter Zijlstra | 9148a3a | 2016-09-20 22:34:51 +0200 | [diff] [blame] | 1498 | SCHED_WARN_ON(!rcu_read_lock_held()); |
Daniel Lezcano | 442bf3a | 2014-09-04 11:32:09 -0400 | [diff] [blame] | 1499 | return rq->idle_state; |
| 1500 | } |
Morten Rasmussen | 0691064 | 2015-01-27 13:48:07 +0000 | [diff] [blame] | 1501 | |
| 1502 | static inline void idle_set_state_idx(struct rq *rq, int idle_state_idx) |
| 1503 | { |
| 1504 | rq->idle_state_idx = idle_state_idx; |
| 1505 | } |
| 1506 | |
| 1507 | static inline int idle_get_state_idx(struct rq *rq) |
| 1508 | { |
| 1509 | WARN_ON(!rcu_read_lock_held()); |
| 1510 | return rq->idle_state_idx; |
| 1511 | } |
Daniel Lezcano | 442bf3a | 2014-09-04 11:32:09 -0400 | [diff] [blame] | 1512 | #else |
| 1513 | static inline void idle_set_state(struct rq *rq, |
| 1514 | struct cpuidle_state *idle_state) |
| 1515 | { |
| 1516 | } |
| 1517 | |
| 1518 | static inline struct cpuidle_state *idle_get_state(struct rq *rq) |
| 1519 | { |
| 1520 | return NULL; |
| 1521 | } |
Morten Rasmussen | 0691064 | 2015-01-27 13:48:07 +0000 | [diff] [blame] | 1522 | |
| 1523 | static inline void idle_set_state_idx(struct rq *rq, int idle_state_idx) |
| 1524 | { |
| 1525 | } |
| 1526 | |
| 1527 | static inline int idle_get_state_idx(struct rq *rq) |
| 1528 | { |
| 1529 | return -1; |
| 1530 | } |
Daniel Lezcano | 442bf3a | 2014-09-04 11:32:09 -0400 | [diff] [blame] | 1531 | #endif |
| 1532 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1533 | extern void sysrq_sched_debug_show(void); |
| 1534 | extern void sched_init_granularity(void); |
| 1535 | extern void update_max_interval(void); |
Juri Lelli | 1baca4c | 2013-11-07 14:43:38 +0100 | [diff] [blame] | 1536 | |
| 1537 | extern void init_sched_dl_class(void); |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1538 | extern void init_sched_rt_class(void); |
| 1539 | extern void init_sched_fair_class(void); |
| 1540 | |
Kirill Tkhai | 8875125 | 2014-06-29 00:03:57 +0400 | [diff] [blame] | 1541 | extern void resched_curr(struct rq *rq); |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1542 | extern void resched_cpu(int cpu); |
| 1543 | |
| 1544 | extern struct rt_bandwidth def_rt_bandwidth; |
| 1545 | extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime); |
| 1546 | |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 1547 | extern struct dl_bandwidth def_dl_bandwidth; |
| 1548 | extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime); |
Dario Faggioli | aab03e0 | 2013-11-28 11:14:43 +0100 | [diff] [blame] | 1549 | extern void init_dl_task_timer(struct sched_dl_entity *dl_se); |
| 1550 | |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 1551 | unsigned long to_ratio(u64 period, u64 runtime); |
| 1552 | |
Yuyang Du | 540247f | 2015-07-15 08:04:39 +0800 | [diff] [blame] | 1553 | extern void init_entity_runnable_average(struct sched_entity *se); |
Yuyang Du | 2b8c41d | 2016-03-30 04:30:56 +0800 | [diff] [blame] | 1554 | extern void post_init_entity_util_avg(struct sched_entity *se); |
Alex Shi | a75cdaa | 2013-06-20 10:18:47 +0800 | [diff] [blame] | 1555 | |
Frederic Weisbecker | 76d92ac | 2015-07-17 22:25:49 +0200 | [diff] [blame] | 1556 | #ifdef CONFIG_NO_HZ_FULL |
| 1557 | extern bool sched_can_stop_tick(struct rq *rq); |
| 1558 | |
| 1559 | /* |
| 1560 | * Tick may be needed by tasks in the runqueue depending on their policy and |
| 1561 | * requirements. If tick is needed, lets send the target an IPI to kick it out of |
| 1562 | * nohz mode if necessary. |
| 1563 | */ |
| 1564 | static inline void sched_update_tick_dependency(struct rq *rq) |
| 1565 | { |
| 1566 | int cpu; |
| 1567 | |
| 1568 | if (!tick_nohz_full_enabled()) |
| 1569 | return; |
| 1570 | |
| 1571 | cpu = cpu_of(rq); |
| 1572 | |
| 1573 | if (!tick_nohz_full_cpu(cpu)) |
| 1574 | return; |
| 1575 | |
| 1576 | if (sched_can_stop_tick(rq)) |
| 1577 | tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED); |
| 1578 | else |
| 1579 | tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED); |
| 1580 | } |
| 1581 | #else |
| 1582 | static inline void sched_update_tick_dependency(struct rq *rq) { } |
| 1583 | #endif |
| 1584 | |
Joseph Lo | 7750186 | 2013-04-22 14:39:18 +0800 | [diff] [blame] | 1585 | static inline void __add_nr_running(struct rq *rq, unsigned count) |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1586 | { |
Kirill Tkhai | 7246544 | 2014-05-09 03:00:14 +0400 | [diff] [blame] | 1587 | unsigned prev_nr = rq->nr_running; |
| 1588 | |
| 1589 | rq->nr_running = prev_nr + count; |
Frederic Weisbecker | 9f3660c | 2013-04-20 14:35:09 +0200 | [diff] [blame] | 1590 | |
Kirill Tkhai | 7246544 | 2014-05-09 03:00:14 +0400 | [diff] [blame] | 1591 | if (prev_nr < 2 && rq->nr_running >= 2) { |
Tim Chen | 4486edd | 2014-06-23 12:16:49 -0700 | [diff] [blame] | 1592 | #ifdef CONFIG_SMP |
| 1593 | if (!rq->rd->overload) |
| 1594 | rq->rd->overload = true; |
| 1595 | #endif |
Tim Chen | 4486edd | 2014-06-23 12:16:49 -0700 | [diff] [blame] | 1596 | } |
Frederic Weisbecker | 76d92ac | 2015-07-17 22:25:49 +0200 | [diff] [blame] | 1597 | |
| 1598 | sched_update_tick_dependency(rq); |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1599 | } |
| 1600 | |
Joseph Lo | 7750186 | 2013-04-22 14:39:18 +0800 | [diff] [blame] | 1601 | static inline void __sub_nr_running(struct rq *rq, unsigned count) |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1602 | { |
Kirill Tkhai | 7246544 | 2014-05-09 03:00:14 +0400 | [diff] [blame] | 1603 | rq->nr_running -= count; |
Frederic Weisbecker | 76d92ac | 2015-07-17 22:25:49 +0200 | [diff] [blame] | 1604 | /* Check if we still need preemption */ |
| 1605 | sched_update_tick_dependency(rq); |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1606 | } |
| 1607 | |
Joseph Lo | 7750186 | 2013-04-22 14:39:18 +0800 | [diff] [blame] | 1608 | #ifdef CONFIG_CPU_QUIET |
| 1609 | #define NR_AVE_SCALE(x) ((x) << FSHIFT) |
| 1610 | static inline u64 do_nr_running_integral(struct rq *rq) |
| 1611 | { |
| 1612 | s64 nr, deltax; |
| 1613 | u64 nr_running_integral = rq->nr_running_integral; |
| 1614 | |
| 1615 | deltax = rq->clock_task - rq->nr_last_stamp; |
| 1616 | nr = NR_AVE_SCALE(rq->nr_running); |
| 1617 | |
| 1618 | nr_running_integral += nr * deltax; |
| 1619 | |
| 1620 | return nr_running_integral; |
| 1621 | } |
| 1622 | |
| 1623 | static inline void add_nr_running(struct rq *rq, unsigned count) |
| 1624 | { |
| 1625 | write_seqcount_begin(&rq->ave_seqcnt); |
| 1626 | rq->nr_running_integral = do_nr_running_integral(rq); |
| 1627 | rq->nr_last_stamp = rq->clock_task; |
| 1628 | __add_nr_running(rq, count); |
| 1629 | write_seqcount_end(&rq->ave_seqcnt); |
| 1630 | } |
| 1631 | |
| 1632 | static inline void sub_nr_running(struct rq *rq, unsigned count) |
| 1633 | { |
| 1634 | write_seqcount_begin(&rq->ave_seqcnt); |
| 1635 | rq->nr_running_integral = do_nr_running_integral(rq); |
| 1636 | rq->nr_last_stamp = rq->clock_task; |
| 1637 | __sub_nr_running(rq, count); |
| 1638 | write_seqcount_end(&rq->ave_seqcnt); |
| 1639 | } |
| 1640 | #else |
| 1641 | #define add_nr_running __add_nr_running |
| 1642 | #define sub_nr_running __sub_nr_running |
| 1643 | #endif |
| 1644 | |
Frederic Weisbecker | 265f22a | 2013-05-03 03:39:05 +0200 | [diff] [blame] | 1645 | static inline void rq_last_tick_reset(struct rq *rq) |
| 1646 | { |
| 1647 | #ifdef CONFIG_NO_HZ_FULL |
| 1648 | rq->last_sched_tick = jiffies; |
| 1649 | #endif |
| 1650 | } |
| 1651 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1652 | extern void activate_task(struct rq *rq, struct task_struct *p, int flags); |
| 1653 | extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags); |
| 1654 | |
| 1655 | extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags); |
| 1656 | |
| 1657 | extern const_debug unsigned int sysctl_sched_time_avg; |
| 1658 | extern const_debug unsigned int sysctl_sched_nr_migrate; |
| 1659 | extern const_debug unsigned int sysctl_sched_migration_cost; |
| 1660 | |
| 1661 | static inline u64 sched_avg_period(void) |
| 1662 | { |
| 1663 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; |
| 1664 | } |
| 1665 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1666 | #ifdef CONFIG_SCHED_HRTICK |
| 1667 | |
| 1668 | /* |
| 1669 | * Use hrtick when: |
| 1670 | * - enabled by features |
| 1671 | * - hrtimer is actually high res |
| 1672 | */ |
| 1673 | static inline int hrtick_enabled(struct rq *rq) |
| 1674 | { |
| 1675 | if (!sched_feat(HRTICK)) |
| 1676 | return 0; |
| 1677 | if (!cpu_active(cpu_of(rq))) |
| 1678 | return 0; |
| 1679 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
| 1680 | } |
| 1681 | |
| 1682 | void hrtick_start(struct rq *rq, u64 delay); |
| 1683 | |
Mike Galbraith | b39e66e | 2011-11-22 15:20:07 +0100 | [diff] [blame] | 1684 | #else |
| 1685 | |
| 1686 | static inline int hrtick_enabled(struct rq *rq) |
| 1687 | { |
| 1688 | return 0; |
| 1689 | } |
| 1690 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1691 | #endif /* CONFIG_SCHED_HRTICK */ |
| 1692 | |
| 1693 | #ifdef CONFIG_SMP |
| 1694 | extern void sched_avg_update(struct rq *rq); |
Peter Zijlstra | dfbca41 | 2015-03-23 14:19:05 +0100 | [diff] [blame] | 1695 | |
| 1696 | #ifndef arch_scale_freq_capacity |
| 1697 | static __always_inline |
| 1698 | unsigned long arch_scale_freq_capacity(struct sched_domain *sd, int cpu) |
| 1699 | { |
| 1700 | return SCHED_CAPACITY_SCALE; |
| 1701 | } |
| 1702 | #endif |
Vincent Guittot | b5b4860 | 2015-02-27 16:54:08 +0100 | [diff] [blame] | 1703 | |
Dietmar Eggemann | da6833cf | 2017-07-13 09:48:42 +0100 | [diff] [blame] | 1704 | #ifndef arch_scale_max_freq_capacity |
| 1705 | static __always_inline |
| 1706 | unsigned long arch_scale_max_freq_capacity(struct sched_domain *sd, int cpu) |
| 1707 | { |
| 1708 | return SCHED_CAPACITY_SCALE; |
| 1709 | } |
| 1710 | #endif |
| 1711 | |
Ionela Voinescu | 33550ef | 2017-08-30 16:43:11 +0100 | [diff] [blame] | 1712 | #ifndef arch_scale_min_freq_capacity |
| 1713 | static __always_inline |
| 1714 | unsigned long arch_scale_min_freq_capacity(struct sched_domain *sd, int cpu) |
| 1715 | { |
| 1716 | /* |
| 1717 | * Multiplied with any capacity value, this scale factor will return |
| 1718 | * 0, which represents an un-capped state |
| 1719 | */ |
| 1720 | return 0; |
| 1721 | } |
| 1722 | #endif |
| 1723 | |
Morten Rasmussen | 8cd5601 | 2015-08-14 17:23:10 +0100 | [diff] [blame] | 1724 | #ifndef arch_scale_cpu_capacity |
| 1725 | static __always_inline |
| 1726 | unsigned long arch_scale_cpu_capacity(struct sched_domain *sd, int cpu) |
| 1727 | { |
Dietmar Eggemann | e3279a2 | 2015-08-15 00:04:41 +0100 | [diff] [blame] | 1728 | if (sd && (sd->flags & SD_SHARE_CPUCAPACITY) && (sd->span_weight > 1)) |
Morten Rasmussen | 8cd5601 | 2015-08-14 17:23:10 +0100 | [diff] [blame] | 1729 | return sd->smt_gain / sd->span_weight; |
| 1730 | |
| 1731 | return SCHED_CAPACITY_SCALE; |
| 1732 | } |
| 1733 | #endif |
| 1734 | |
Steve Muckle | 608d494 | 2015-06-25 14:12:33 +0100 | [diff] [blame] | 1735 | #ifdef CONFIG_SMP |
| 1736 | static inline unsigned long capacity_of(int cpu) |
| 1737 | { |
| 1738 | return cpu_rq(cpu)->cpu_capacity; |
| 1739 | } |
| 1740 | |
| 1741 | static inline unsigned long capacity_orig_of(int cpu) |
| 1742 | { |
| 1743 | return cpu_rq(cpu)->cpu_capacity_orig; |
| 1744 | } |
| 1745 | |
Srivatsa Vaddagiri | 26c2154 | 2016-05-31 09:08:38 -0700 | [diff] [blame] | 1746 | extern unsigned int sysctl_sched_use_walt_cpu_util; |
| 1747 | extern unsigned int walt_ravg_window; |
Vikram Mulukutla | 44310bf | 2017-08-10 17:26:20 -0700 | [diff] [blame] | 1748 | extern bool walt_disabled; |
Srivatsa Vaddagiri | 26c2154 | 2016-05-31 09:08:38 -0700 | [diff] [blame] | 1749 | |
Steve Muckle | 608d494 | 2015-06-25 14:12:33 +0100 | [diff] [blame] | 1750 | /* |
| 1751 | * cpu_util returns the amount of capacity of a CPU that is used by CFS |
| 1752 | * tasks. The unit of the return value must be the one of capacity so we can |
| 1753 | * compare the utilization with the capacity of the CPU that is available for |
| 1754 | * CFS task (ie cpu_capacity). |
| 1755 | * |
| 1756 | * cfs_rq.avg.util_avg is the sum of running time of runnable tasks plus the |
| 1757 | * recent utilization of currently non-runnable tasks on a CPU. It represents |
| 1758 | * the amount of utilization of a CPU in the range [0..capacity_orig] where |
| 1759 | * capacity_orig is the cpu_capacity available at the highest frequency |
| 1760 | * (arch_scale_freq_capacity()). |
| 1761 | * The utilization of a CPU converges towards a sum equal to or less than the |
| 1762 | * current capacity (capacity_curr <= capacity_orig) of the CPU because it is |
| 1763 | * the running time on this CPU scaled by capacity_curr. |
| 1764 | * |
| 1765 | * Nevertheless, cfs_rq.avg.util_avg can be higher than capacity_curr or even |
| 1766 | * higher than capacity_orig because of unfortunate rounding in |
| 1767 | * cfs.avg.util_avg or just after migrating tasks and new task wakeups until |
| 1768 | * the average stabilizes with the new running time. We need to check that the |
| 1769 | * utilization stays within the range of [0..capacity_orig] and cap it if |
| 1770 | * necessary. Without utilization capping, a group could be seen as overloaded |
| 1771 | * (CPU0 utilization at 121% + CPU1 utilization at 80%) whereas CPU1 has 20% of |
| 1772 | * available capacity. We allow utilization to overshoot capacity_curr (but not |
| 1773 | * capacity_orig) as it useful for predicting the capacity required after task |
| 1774 | * migrations (scheduler-driven DVFS). |
| 1775 | */ |
| 1776 | static inline unsigned long __cpu_util(int cpu, int delta) |
| 1777 | { |
| 1778 | unsigned long util = cpu_rq(cpu)->cfs.avg.util_avg; |
| 1779 | unsigned long capacity = capacity_orig_of(cpu); |
| 1780 | |
Srivatsa Vaddagiri | 26c2154 | 2016-05-31 09:08:38 -0700 | [diff] [blame] | 1781 | #ifdef CONFIG_SCHED_WALT |
Joonwoo Park | cc10342 | 2017-01-20 11:10:15 -0800 | [diff] [blame] | 1782 | if (!walt_disabled && sysctl_sched_use_walt_cpu_util) |
| 1783 | util = div64_u64(cpu_rq(cpu)->cumulative_runnable_avg, |
| 1784 | walt_ravg_window >> SCHED_CAPACITY_SHIFT); |
Srivatsa Vaddagiri | 26c2154 | 2016-05-31 09:08:38 -0700 | [diff] [blame] | 1785 | #endif |
Steve Muckle | 608d494 | 2015-06-25 14:12:33 +0100 | [diff] [blame] | 1786 | delta += util; |
| 1787 | if (delta < 0) |
| 1788 | return 0; |
| 1789 | |
| 1790 | return (delta >= capacity) ? capacity : delta; |
| 1791 | } |
| 1792 | |
| 1793 | static inline unsigned long cpu_util(int cpu) |
| 1794 | { |
| 1795 | return __cpu_util(cpu, 0); |
| 1796 | } |
| 1797 | |
Joonwoo Park | 8b34bba | 2016-12-08 16:12:12 -0800 | [diff] [blame] | 1798 | static inline unsigned long cpu_util_freq(int cpu) |
| 1799 | { |
| 1800 | unsigned long util = cpu_rq(cpu)->cfs.avg.util_avg; |
| 1801 | unsigned long capacity = capacity_orig_of(cpu); |
| 1802 | |
| 1803 | #ifdef CONFIG_SCHED_WALT |
Joonwoo Park | cc10342 | 2017-01-20 11:10:15 -0800 | [diff] [blame] | 1804 | if (!walt_disabled && sysctl_sched_use_walt_cpu_util) |
| 1805 | util = div64_u64(cpu_rq(cpu)->prev_runnable_sum, |
| 1806 | walt_ravg_window >> SCHED_CAPACITY_SHIFT); |
Joonwoo Park | 8b34bba | 2016-12-08 16:12:12 -0800 | [diff] [blame] | 1807 | #endif |
| 1808 | return (util >= capacity) ? capacity : util; |
| 1809 | } |
| 1810 | |
Steve Muckle | 608d494 | 2015-06-25 14:12:33 +0100 | [diff] [blame] | 1811 | #endif |
| 1812 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1813 | static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) |
| 1814 | { |
Vincent Guittot | b5b4860 | 2015-02-27 16:54:08 +0100 | [diff] [blame] | 1815 | rq->rt_avg += rt_delta * arch_scale_freq_capacity(NULL, cpu_of(rq)); |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1816 | } |
| 1817 | #else |
| 1818 | static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { } |
| 1819 | static inline void sched_avg_update(struct rq *rq) { } |
| 1820 | #endif |
| 1821 | |
Patrick Bellasi | d248900 | 2016-07-28 18:44:40 +0100 | [diff] [blame] | 1822 | extern struct rq *lock_rq_of(struct task_struct *p, struct rq_flags *flags); |
| 1823 | extern void unlock_rq_of(struct rq *rq, struct task_struct *p, struct rq_flags *flags); |
| 1824 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1825 | #ifdef CONFIG_SMP |
| 1826 | #ifdef CONFIG_PREEMPT |
| 1827 | |
| 1828 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2); |
| 1829 | |
| 1830 | /* |
| 1831 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
| 1832 | * way at the expense of forcing extra atomic operations in all |
| 1833 | * invocations. This assures that the double_lock is acquired using the |
| 1834 | * same underlying policy as the spinlock_t on this architecture, which |
| 1835 | * reduces latency compared to the unfair variant below. However, it |
| 1836 | * also adds more overhead and therefore may reduce throughput. |
| 1837 | */ |
| 1838 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
| 1839 | __releases(this_rq->lock) |
| 1840 | __acquires(busiest->lock) |
| 1841 | __acquires(this_rq->lock) |
| 1842 | { |
| 1843 | raw_spin_unlock(&this_rq->lock); |
| 1844 | double_rq_lock(this_rq, busiest); |
| 1845 | |
| 1846 | return 1; |
| 1847 | } |
| 1848 | |
| 1849 | #else |
| 1850 | /* |
| 1851 | * Unfair double_lock_balance: Optimizes throughput at the expense of |
| 1852 | * latency by eliminating extra atomic operations when the locks are |
| 1853 | * already in proper order on entry. This favors lower cpu-ids and will |
| 1854 | * grant the double lock to lower cpus over higher ids under contention, |
| 1855 | * regardless of entry order into the function. |
| 1856 | */ |
| 1857 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
| 1858 | __releases(this_rq->lock) |
| 1859 | __acquires(busiest->lock) |
| 1860 | __acquires(this_rq->lock) |
| 1861 | { |
| 1862 | int ret = 0; |
| 1863 | |
| 1864 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { |
| 1865 | if (busiest < this_rq) { |
| 1866 | raw_spin_unlock(&this_rq->lock); |
| 1867 | raw_spin_lock(&busiest->lock); |
| 1868 | raw_spin_lock_nested(&this_rq->lock, |
| 1869 | SINGLE_DEPTH_NESTING); |
| 1870 | ret = 1; |
| 1871 | } else |
| 1872 | raw_spin_lock_nested(&busiest->lock, |
| 1873 | SINGLE_DEPTH_NESTING); |
| 1874 | } |
| 1875 | return ret; |
| 1876 | } |
| 1877 | |
| 1878 | #endif /* CONFIG_PREEMPT */ |
| 1879 | |
| 1880 | /* |
| 1881 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. |
| 1882 | */ |
| 1883 | static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest) |
| 1884 | { |
| 1885 | if (unlikely(!irqs_disabled())) { |
| 1886 | /* printk() doesn't work good under rq->lock */ |
| 1887 | raw_spin_unlock(&this_rq->lock); |
| 1888 | BUG_ON(1); |
| 1889 | } |
| 1890 | |
| 1891 | return _double_lock_balance(this_rq, busiest); |
| 1892 | } |
| 1893 | |
| 1894 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
| 1895 | __releases(busiest->lock) |
| 1896 | { |
Todd Kjos | a31778a | 2016-07-04 15:04:45 +0100 | [diff] [blame] | 1897 | if (this_rq != busiest) |
| 1898 | raw_spin_unlock(&busiest->lock); |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1899 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); |
| 1900 | } |
| 1901 | |
Peter Zijlstra | 7460231 | 2013-10-10 20:17:22 +0200 | [diff] [blame] | 1902 | static inline void double_lock(spinlock_t *l1, spinlock_t *l2) |
| 1903 | { |
| 1904 | if (l1 > l2) |
| 1905 | swap(l1, l2); |
| 1906 | |
| 1907 | spin_lock(l1); |
| 1908 | spin_lock_nested(l2, SINGLE_DEPTH_NESTING); |
| 1909 | } |
| 1910 | |
Mike Galbraith | 60e69ee | 2014-04-07 10:55:15 +0200 | [diff] [blame] | 1911 | static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2) |
| 1912 | { |
| 1913 | if (l1 > l2) |
| 1914 | swap(l1, l2); |
| 1915 | |
| 1916 | spin_lock_irq(l1); |
| 1917 | spin_lock_nested(l2, SINGLE_DEPTH_NESTING); |
| 1918 | } |
| 1919 | |
Peter Zijlstra | 7460231 | 2013-10-10 20:17:22 +0200 | [diff] [blame] | 1920 | static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2) |
| 1921 | { |
| 1922 | if (l1 > l2) |
| 1923 | swap(l1, l2); |
| 1924 | |
| 1925 | raw_spin_lock(l1); |
| 1926 | raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING); |
| 1927 | } |
| 1928 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1929 | /* |
| 1930 | * double_rq_lock - safely lock two runqueues |
| 1931 | * |
| 1932 | * Note this does not disable interrupts like task_rq_lock, |
| 1933 | * you need to do so manually before calling. |
| 1934 | */ |
| 1935 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) |
| 1936 | __acquires(rq1->lock) |
| 1937 | __acquires(rq2->lock) |
| 1938 | { |
| 1939 | BUG_ON(!irqs_disabled()); |
| 1940 | if (rq1 == rq2) { |
| 1941 | raw_spin_lock(&rq1->lock); |
| 1942 | __acquire(rq2->lock); /* Fake it out ;) */ |
| 1943 | } else { |
| 1944 | if (rq1 < rq2) { |
| 1945 | raw_spin_lock(&rq1->lock); |
| 1946 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); |
| 1947 | } else { |
| 1948 | raw_spin_lock(&rq2->lock); |
| 1949 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); |
| 1950 | } |
| 1951 | } |
| 1952 | } |
| 1953 | |
| 1954 | /* |
| 1955 | * double_rq_unlock - safely unlock two runqueues |
| 1956 | * |
| 1957 | * Note this does not restore interrupts like task_rq_unlock, |
| 1958 | * you need to do so manually after calling. |
| 1959 | */ |
| 1960 | static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
| 1961 | __releases(rq1->lock) |
| 1962 | __releases(rq2->lock) |
| 1963 | { |
| 1964 | raw_spin_unlock(&rq1->lock); |
| 1965 | if (rq1 != rq2) |
| 1966 | raw_spin_unlock(&rq2->lock); |
| 1967 | else |
| 1968 | __release(rq2->lock); |
| 1969 | } |
| 1970 | |
| 1971 | #else /* CONFIG_SMP */ |
| 1972 | |
| 1973 | /* |
| 1974 | * double_rq_lock - safely lock two runqueues |
| 1975 | * |
| 1976 | * Note this does not disable interrupts like task_rq_lock, |
| 1977 | * you need to do so manually before calling. |
| 1978 | */ |
| 1979 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) |
| 1980 | __acquires(rq1->lock) |
| 1981 | __acquires(rq2->lock) |
| 1982 | { |
| 1983 | BUG_ON(!irqs_disabled()); |
| 1984 | BUG_ON(rq1 != rq2); |
| 1985 | raw_spin_lock(&rq1->lock); |
| 1986 | __acquire(rq2->lock); /* Fake it out ;) */ |
| 1987 | } |
| 1988 | |
| 1989 | /* |
| 1990 | * double_rq_unlock - safely unlock two runqueues |
| 1991 | * |
| 1992 | * Note this does not restore interrupts like task_rq_unlock, |
| 1993 | * you need to do so manually after calling. |
| 1994 | */ |
| 1995 | static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
| 1996 | __releases(rq1->lock) |
| 1997 | __releases(rq2->lock) |
| 1998 | { |
| 1999 | BUG_ON(rq1 != rq2); |
| 2000 | raw_spin_unlock(&rq1->lock); |
| 2001 | __release(rq2->lock); |
| 2002 | } |
| 2003 | |
| 2004 | #endif |
| 2005 | |
| 2006 | extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq); |
| 2007 | extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq); |
Srikar Dronamraju | 6b55c96 | 2015-06-25 22:51:41 +0530 | [diff] [blame] | 2008 | |
| 2009 | #ifdef CONFIG_SCHED_DEBUG |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 2010 | extern void print_cfs_stats(struct seq_file *m, int cpu); |
| 2011 | extern void print_rt_stats(struct seq_file *m, int cpu); |
Wanpeng Li | acb3213 | 2014-10-31 06:39:33 +0800 | [diff] [blame] | 2012 | extern void print_dl_stats(struct seq_file *m, int cpu); |
Srikar Dronamraju | 6b55c96 | 2015-06-25 22:51:41 +0530 | [diff] [blame] | 2013 | extern void |
| 2014 | print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq); |
Srikar Dronamraju | 397f237 | 2015-06-25 22:51:43 +0530 | [diff] [blame] | 2015 | |
| 2016 | #ifdef CONFIG_NUMA_BALANCING |
| 2017 | extern void |
| 2018 | show_numa_stats(struct task_struct *p, struct seq_file *m); |
| 2019 | extern void |
| 2020 | print_numa_stats(struct seq_file *m, int node, unsigned long tsf, |
| 2021 | unsigned long tpf, unsigned long gsf, unsigned long gpf); |
| 2022 | #endif /* CONFIG_NUMA_BALANCING */ |
| 2023 | #endif /* CONFIG_SCHED_DEBUG */ |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 2024 | |
| 2025 | extern void init_cfs_rq(struct cfs_rq *cfs_rq); |
Abel Vesa | 07c54f7 | 2015-03-03 13:50:27 +0200 | [diff] [blame] | 2026 | extern void init_rt_rq(struct rt_rq *rt_rq); |
| 2027 | extern void init_dl_rq(struct dl_rq *dl_rq); |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 2028 | |
Ben Segall | 1ee14e6 | 2013-10-16 11:16:12 -0700 | [diff] [blame] | 2029 | extern void cfs_bandwidth_usage_inc(void); |
| 2030 | extern void cfs_bandwidth_usage_dec(void); |
Suresh Siddha | 1c792db | 2011-12-01 17:07:32 -0800 | [diff] [blame] | 2031 | |
Frederic Weisbecker | 3451d02 | 2011-08-10 23:21:01 +0200 | [diff] [blame] | 2032 | #ifdef CONFIG_NO_HZ_COMMON |
Suresh Siddha | 1c792db | 2011-12-01 17:07:32 -0800 | [diff] [blame] | 2033 | enum rq_nohz_flag_bits { |
| 2034 | NOHZ_TICK_STOPPED, |
| 2035 | NOHZ_BALANCE_KICK, |
| 2036 | }; |
| 2037 | |
| 2038 | #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags) |
Thomas Gleixner | 20a5c8c | 2016-03-10 12:54:20 +0100 | [diff] [blame] | 2039 | |
| 2040 | extern void nohz_balance_exit_idle(unsigned int cpu); |
| 2041 | #else |
| 2042 | static inline void nohz_balance_exit_idle(unsigned int cpu) { } |
Suresh Siddha | 1c792db | 2011-12-01 17:07:32 -0800 | [diff] [blame] | 2043 | #endif |
Frederic Weisbecker | 73fbec6 | 2012-06-16 15:57:37 +0200 | [diff] [blame] | 2044 | |
| 2045 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
Frederic Weisbecker | 19d23dbf | 2016-09-26 02:29:20 +0200 | [diff] [blame] | 2046 | struct irqtime { |
Frederic Weisbecker | aee0c9e | 2017-04-25 16:10:48 +0200 | [diff] [blame] | 2047 | u64 total; |
Frederic Weisbecker | 034cd8a | 2017-01-31 04:09:32 +0100 | [diff] [blame] | 2048 | u64 tick_delta; |
Frederic Weisbecker | 19d23dbf | 2016-09-26 02:29:20 +0200 | [diff] [blame] | 2049 | u64 irq_start_time; |
| 2050 | struct u64_stats_sync sync; |
| 2051 | }; |
Frederic Weisbecker | 73fbec6 | 2012-06-16 15:57:37 +0200 | [diff] [blame] | 2052 | |
Frederic Weisbecker | 19d23dbf | 2016-09-26 02:29:20 +0200 | [diff] [blame] | 2053 | DECLARE_PER_CPU(struct irqtime, cpu_irqtime); |
Frederic Weisbecker | 73fbec6 | 2012-06-16 15:57:37 +0200 | [diff] [blame] | 2054 | |
Frederic Weisbecker | aee0c9e | 2017-04-25 16:10:48 +0200 | [diff] [blame] | 2055 | /* |
| 2056 | * Returns the irqtime minus the softirq time computed by ksoftirqd. |
| 2057 | * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime |
| 2058 | * and never move forward. |
| 2059 | */ |
Frederic Weisbecker | 73fbec6 | 2012-06-16 15:57:37 +0200 | [diff] [blame] | 2060 | static inline u64 irq_time_read(int cpu) |
| 2061 | { |
Frederic Weisbecker | 19d23dbf | 2016-09-26 02:29:20 +0200 | [diff] [blame] | 2062 | struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu); |
| 2063 | unsigned int seq; |
| 2064 | u64 total; |
Frederic Weisbecker | 73fbec6 | 2012-06-16 15:57:37 +0200 | [diff] [blame] | 2065 | |
| 2066 | do { |
Frederic Weisbecker | 19d23dbf | 2016-09-26 02:29:20 +0200 | [diff] [blame] | 2067 | seq = __u64_stats_fetch_begin(&irqtime->sync); |
Frederic Weisbecker | aee0c9e | 2017-04-25 16:10:48 +0200 | [diff] [blame] | 2068 | total = irqtime->total; |
Frederic Weisbecker | 19d23dbf | 2016-09-26 02:29:20 +0200 | [diff] [blame] | 2069 | } while (__u64_stats_fetch_retry(&irqtime->sync, seq)); |
Frederic Weisbecker | 73fbec6 | 2012-06-16 15:57:37 +0200 | [diff] [blame] | 2070 | |
Frederic Weisbecker | 19d23dbf | 2016-09-26 02:29:20 +0200 | [diff] [blame] | 2071 | return total; |
Frederic Weisbecker | 73fbec6 | 2012-06-16 15:57:37 +0200 | [diff] [blame] | 2072 | } |
Frederic Weisbecker | 73fbec6 | 2012-06-16 15:57:37 +0200 | [diff] [blame] | 2073 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
Rafael J. Wysocki | adaf9fc | 2016-03-10 20:44:47 +0100 | [diff] [blame] | 2074 | |
| 2075 | #ifdef CONFIG_CPU_FREQ |
| 2076 | DECLARE_PER_CPU(struct update_util_data *, cpufreq_update_util_data); |
| 2077 | |
| 2078 | /** |
| 2079 | * cpufreq_update_util - Take a note about CPU utilization changes. |
Rafael J. Wysocki | 12bde33 | 2016-08-10 03:11:17 +0200 | [diff] [blame] | 2080 | * @rq: Runqueue to carry out the update for. |
Rafael J. Wysocki | 58919e8 | 2016-08-16 22:14:55 +0200 | [diff] [blame] | 2081 | * @flags: Update reason flags. |
Rafael J. Wysocki | adaf9fc | 2016-03-10 20:44:47 +0100 | [diff] [blame] | 2082 | * |
Rafael J. Wysocki | 58919e8 | 2016-08-16 22:14:55 +0200 | [diff] [blame] | 2083 | * This function is called by the scheduler on the CPU whose utilization is |
| 2084 | * being updated. |
Rafael J. Wysocki | adaf9fc | 2016-03-10 20:44:47 +0100 | [diff] [blame] | 2085 | * |
| 2086 | * It can only be called from RCU-sched read-side critical sections. |
Rafael J. Wysocki | adaf9fc | 2016-03-10 20:44:47 +0100 | [diff] [blame] | 2087 | * |
| 2088 | * The way cpufreq is currently arranged requires it to evaluate the CPU |
| 2089 | * performance state (frequency/voltage) on a regular basis to prevent it from |
| 2090 | * being stuck in a completely inadequate performance level for too long. |
| 2091 | * That is not guaranteed to happen if the updates are only triggered from CFS, |
| 2092 | * though, because they may not be coming in if RT or deadline tasks are active |
| 2093 | * all the time (or there are RT and DL tasks only). |
| 2094 | * |
| 2095 | * As a workaround for that issue, this function is called by the RT and DL |
| 2096 | * sched classes to trigger extra cpufreq updates to prevent it from stalling, |
| 2097 | * but that really is a band-aid. Going forward it should be replaced with |
| 2098 | * solutions targeted more specifically at RT and DL tasks. |
| 2099 | */ |
Rafael J. Wysocki | 12bde33 | 2016-08-10 03:11:17 +0200 | [diff] [blame] | 2100 | static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) |
Rafael J. Wysocki | adaf9fc | 2016-03-10 20:44:47 +0100 | [diff] [blame] | 2101 | { |
Rafael J. Wysocki | 58919e8 | 2016-08-16 22:14:55 +0200 | [diff] [blame] | 2102 | struct update_util_data *data; |
| 2103 | |
| 2104 | data = rcu_dereference_sched(*this_cpu_ptr(&cpufreq_update_util_data)); |
| 2105 | if (data) |
Rafael J. Wysocki | 12bde33 | 2016-08-10 03:11:17 +0200 | [diff] [blame] | 2106 | data->func(data, rq_clock(rq), flags); |
| 2107 | } |
| 2108 | |
| 2109 | static inline void cpufreq_update_this_cpu(struct rq *rq, unsigned int flags) |
| 2110 | { |
| 2111 | if (cpu_of(rq) == smp_processor_id()) |
| 2112 | cpufreq_update_util(rq, flags); |
Rafael J. Wysocki | adaf9fc | 2016-03-10 20:44:47 +0100 | [diff] [blame] | 2113 | } |
| 2114 | #else |
Rafael J. Wysocki | 12bde33 | 2016-08-10 03:11:17 +0200 | [diff] [blame] | 2115 | static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {} |
| 2116 | static inline void cpufreq_update_this_cpu(struct rq *rq, unsigned int flags) {} |
Rafael J. Wysocki | adaf9fc | 2016-03-10 20:44:47 +0100 | [diff] [blame] | 2117 | #endif /* CONFIG_CPU_FREQ */ |
Linus Torvalds | be53f58 | 2016-03-24 09:42:50 -0700 | [diff] [blame] | 2118 | |
Joonwoo Park | 7f17fff | 2017-02-03 11:15:31 -0800 | [diff] [blame] | 2119 | #ifdef CONFIG_SCHED_WALT |
| 2120 | |
| 2121 | static inline bool |
| 2122 | walt_task_in_cum_window_demand(struct rq *rq, struct task_struct *p) |
| 2123 | { |
| 2124 | return cpu_of(rq) == task_cpu(p) && |
| 2125 | (p->on_rq || p->last_sleep_ts >= rq->window_start); |
| 2126 | } |
| 2127 | |
| 2128 | #endif /* CONFIG_SCHED_WALT */ |
| 2129 | |
Rafael J. Wysocki | 9bdcb44 | 2016-04-02 01:09:12 +0200 | [diff] [blame] | 2130 | #ifdef arch_scale_freq_capacity |
| 2131 | #ifndef arch_scale_freq_invariant |
| 2132 | #define arch_scale_freq_invariant() (true) |
| 2133 | #endif |
| 2134 | #else /* arch_scale_freq_capacity */ |
| 2135 | #define arch_scale_freq_invariant() (false) |
| 2136 | #endif |