1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _KERNEL_STATS_H
3 #define _KERNEL_STATS_H
5 #ifdef CONFIG_SCHEDSTATS
7 extern struct static_key_false sched_schedstats;
10 * Expects runqueue lock to be held for atomicity of update
13 rq_sched_info_arrive(struct rq *rq, unsigned long long delta)
16 rq->rq_sched_info.run_delay += delta;
17 rq->rq_sched_info.pcount++;
22 * Expects runqueue lock to be held for atomicity of update
25 rq_sched_info_depart(struct rq *rq, unsigned long long delta)
28 rq->rq_cpu_time += delta;
32 rq_sched_info_dequeue(struct rq *rq, unsigned long long delta)
35 rq->rq_sched_info.run_delay += delta;
37 #define schedstat_enabled() static_branch_unlikely(&sched_schedstats)
38 #define __schedstat_inc(var) do { var++; } while (0)
39 #define schedstat_inc(var) do { if (schedstat_enabled()) { var++; } } while (0)
40 #define __schedstat_add(var, amt) do { var += (amt); } while (0)
41 #define schedstat_add(var, amt) do { if (schedstat_enabled()) { var += (amt); } } while (0)
42 #define __schedstat_set(var, val) do { var = (val); } while (0)
43 #define schedstat_set(var, val) do { if (schedstat_enabled()) { var = (val); } } while (0)
44 #define schedstat_val(var) (var)
45 #define schedstat_val_or_zero(var) ((schedstat_enabled()) ? (var) : 0)
47 void __update_stats_wait_start(struct rq *rq, struct task_struct *p,
48 struct sched_statistics *stats);
50 void __update_stats_wait_end(struct rq *rq, struct task_struct *p,
51 struct sched_statistics *stats);
52 void __update_stats_enqueue_sleeper(struct rq *rq, struct task_struct *p,
53 struct sched_statistics *stats);
56 check_schedstat_required(void)
58 if (schedstat_enabled())
61 /* Force schedstat enabled if a dependent tracepoint is active */
62 if (trace_sched_stat_wait_enabled() ||
63 trace_sched_stat_sleep_enabled() ||
64 trace_sched_stat_iowait_enabled() ||
65 trace_sched_stat_blocked_enabled() ||
66 trace_sched_stat_runtime_enabled())
67 printk_deferred_once("Scheduler tracepoints stat_sleep, stat_iowait, stat_blocked and stat_runtime require the kernel parameter schedstats=enable or kernel.sched_schedstats=1\n");
70 #else /* !CONFIG_SCHEDSTATS: */
72 static inline void rq_sched_info_arrive (struct rq *rq, unsigned long long delta) { }
73 static inline void rq_sched_info_dequeue(struct rq *rq, unsigned long long delta) { }
74 static inline void rq_sched_info_depart (struct rq *rq, unsigned long long delta) { }
75 # define schedstat_enabled() 0
76 # define __schedstat_inc(var) do { } while (0)
77 # define schedstat_inc(var) do { } while (0)
78 # define __schedstat_add(var, amt) do { } while (0)
79 # define schedstat_add(var, amt) do { } while (0)
80 # define __schedstat_set(var, val) do { } while (0)
81 # define schedstat_set(var, val) do { } while (0)
82 # define schedstat_val(var) 0
83 # define schedstat_val_or_zero(var) 0
85 # define __update_stats_wait_start(rq, p, stats) do { } while (0)
86 # define __update_stats_wait_end(rq, p, stats) do { } while (0)
87 # define __update_stats_enqueue_sleeper(rq, p, stats) do { } while (0)
88 # define check_schedstat_required() do { } while (0)
90 #endif /* CONFIG_SCHEDSTATS */
92 #ifdef CONFIG_FAIR_GROUP_SCHED
93 struct sched_entity_stats {
94 struct sched_entity se;
95 struct sched_statistics stats;
96 } __no_randomize_layout;
99 static inline struct sched_statistics *
100 __schedstats_from_se(struct sched_entity *se)
102 #ifdef CONFIG_FAIR_GROUP_SCHED
103 if (!entity_is_task(se))
104 return &container_of(se, struct sched_entity_stats, se)->stats;
106 return &task_of(se)->stats;
110 void psi_task_change(struct task_struct *task, int clear, int set);
111 void psi_task_switch(struct task_struct *prev, struct task_struct *next,
113 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
114 void psi_account_irqtime(struct rq *rq, struct task_struct *curr, struct task_struct *prev);
116 static inline void psi_account_irqtime(struct rq *rq, struct task_struct *curr,
117 struct task_struct *prev) {}
118 #endif /*CONFIG_IRQ_TIME_ACCOUNTING */
120 * PSI tracks state that persists across sleeps, such as iowaits and
121 * memory stalls. As a result, it has to distinguish between sleeps,
122 * where a task's runnable state changes, and requeues, where a task
123 * and its state are being moved between CPUs and runqueues.
125 static inline void psi_enqueue(struct task_struct *p, bool wakeup)
127 int clear = 0, set = TSK_RUNNING;
129 if (static_branch_likely(&psi_disabled))
133 set |= TSK_MEMSTALL_RUNNING;
143 psi_task_change(p, clear, set);
146 static inline void psi_dequeue(struct task_struct *p, bool sleep)
148 if (static_branch_likely(&psi_disabled))
152 * A voluntary sleep is a dequeue followed by a task switch. To
153 * avoid walking all ancestors twice, psi_task_switch() handles
154 * TSK_RUNNING and TSK_IOWAIT for us when it moves TSK_ONCPU.
160 psi_task_change(p, p->psi_flags, 0);
163 static inline void psi_ttwu_dequeue(struct task_struct *p)
165 if (static_branch_likely(&psi_disabled))
168 * Is the task being migrated during a wakeup? Make sure to
169 * deregister its sleep-persistent psi states from the old
170 * queue, and let psi_enqueue() know it has to requeue.
172 if (unlikely(p->psi_flags)) {
176 rq = __task_rq_lock(p, &rf);
177 psi_task_change(p, p->psi_flags, 0);
178 __task_rq_unlock(rq, &rf);
182 static inline void psi_sched_switch(struct task_struct *prev,
183 struct task_struct *next,
186 if (static_branch_likely(&psi_disabled))
189 psi_task_switch(prev, next, sleep);
192 #else /* CONFIG_PSI */
193 static inline void psi_enqueue(struct task_struct *p, bool wakeup) {}
194 static inline void psi_dequeue(struct task_struct *p, bool sleep) {}
195 static inline void psi_ttwu_dequeue(struct task_struct *p) {}
196 static inline void psi_sched_switch(struct task_struct *prev,
197 struct task_struct *next,
199 static inline void psi_account_irqtime(struct rq *rq, struct task_struct *curr,
200 struct task_struct *prev) {}
201 #endif /* CONFIG_PSI */
203 #ifdef CONFIG_SCHED_INFO
205 * We are interested in knowing how long it was from the *first* time a
206 * task was queued to the time that it finally hit a CPU, we call this routine
207 * from dequeue_task() to account for possible rq->clock skew across CPUs. The
208 * delta taken on each CPU would annul the skew.
210 static inline void sched_info_dequeue(struct rq *rq, struct task_struct *t)
212 unsigned long long delta = 0;
214 if (!t->sched_info.last_queued)
217 delta = rq_clock(rq) - t->sched_info.last_queued;
218 t->sched_info.last_queued = 0;
219 t->sched_info.run_delay += delta;
221 rq_sched_info_dequeue(rq, delta);
225 * Called when a task finally hits the CPU. We can now calculate how
226 * long it was waiting to run. We also note when it began so that we
227 * can keep stats on how long its time-slice is.
229 static void sched_info_arrive(struct rq *rq, struct task_struct *t)
231 unsigned long long now, delta = 0;
233 if (!t->sched_info.last_queued)
237 delta = now - t->sched_info.last_queued;
238 t->sched_info.last_queued = 0;
239 t->sched_info.run_delay += delta;
240 t->sched_info.last_arrival = now;
241 t->sched_info.pcount++;
243 rq_sched_info_arrive(rq, delta);
247 * This function is only called from enqueue_task(), but also only updates
248 * the timestamp if it is already not set. It's assumed that
249 * sched_info_dequeue() will clear that stamp when appropriate.
251 static inline void sched_info_enqueue(struct rq *rq, struct task_struct *t)
253 if (!t->sched_info.last_queued)
254 t->sched_info.last_queued = rq_clock(rq);
258 * Called when a process ceases being the active-running process involuntarily
259 * due, typically, to expiring its time slice (this may also be called when
260 * switching to the idle task). Now we can calculate how long we ran.
261 * Also, if the process is still in the TASK_RUNNING state, call
262 * sched_info_enqueue() to mark that it has now again started waiting on
265 static inline void sched_info_depart(struct rq *rq, struct task_struct *t)
267 unsigned long long delta = rq_clock(rq) - t->sched_info.last_arrival;
269 rq_sched_info_depart(rq, delta);
271 if (task_is_running(t))
272 sched_info_enqueue(rq, t);
276 * Called when tasks are switched involuntarily due, typically, to expiring
277 * their time slice. (This may also be called when switching to or from
278 * the idle task.) We are only called when prev != next.
281 sched_info_switch(struct rq *rq, struct task_struct *prev, struct task_struct *next)
284 * prev now departs the CPU. It's not interesting to record
285 * stats about how efficient we were at scheduling the idle
288 if (prev != rq->idle)
289 sched_info_depart(rq, prev);
291 if (next != rq->idle)
292 sched_info_arrive(rq, next);
295 #else /* !CONFIG_SCHED_INFO: */
296 # define sched_info_enqueue(rq, t) do { } while (0)
297 # define sched_info_dequeue(rq, t) do { } while (0)
298 # define sched_info_switch(rq, t, next) do { } while (0)
299 #endif /* CONFIG_SCHED_INFO */
301 #endif /* _KERNEL_STATS_H */