xref: /linux/kernel/sched/stats.h (revision 1dd0dd0b1fefd1e51cfaddf62316f759fde7de7d)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _KERNEL_STATS_H
3 #define _KERNEL_STATS_H
4 
5 #ifdef CONFIG_SCHEDSTATS
6 
7 extern struct static_key_false sched_schedstats;
8 
9 /*
10  * Expects runqueue lock to be held for atomicity of update
11  */
12 static inline void
13 rq_sched_info_arrive(struct rq *rq, unsigned long long delta)
14 {
15 	if (rq) {
16 		rq->rq_sched_info.run_delay += delta;
17 		rq->rq_sched_info.pcount++;
18 	}
19 }
20 
21 /*
22  * Expects runqueue lock to be held for atomicity of update
23  */
24 static inline void
25 rq_sched_info_depart(struct rq *rq, unsigned long long delta)
26 {
27 	if (rq)
28 		rq->rq_cpu_time += delta;
29 }
30 
31 static inline void
32 rq_sched_info_dequeue(struct rq *rq, unsigned long long delta)
33 {
34 	if (rq)
35 		rq->rq_sched_info.run_delay += delta;
36 }
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)
46 
47 void __update_stats_wait_start(struct rq *rq, struct task_struct *p,
48 			       struct sched_statistics *stats);
49 
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);
54 
55 static inline void
56 check_schedstat_required(void)
57 {
58 	if (schedstat_enabled())
59 		return;
60 
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");
68 }
69 
70 #else /* !CONFIG_SCHEDSTATS: */
71 
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
84 
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)
89 
90 #endif /* CONFIG_SCHEDSTATS */
91 
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;
97 #endif
98 
99 static inline struct sched_statistics *
100 __schedstats_from_se(struct sched_entity *se)
101 {
102 #ifdef CONFIG_FAIR_GROUP_SCHED
103 	if (!entity_is_task(se))
104 		return &container_of(se, struct sched_entity_stats, se)->stats;
105 #endif
106 	return &task_of(se)->stats;
107 }
108 
109 #ifdef CONFIG_PSI
110 /*
111  * PSI tracks state that persists across sleeps, such as iowaits and
112  * memory stalls. As a result, it has to distinguish between sleeps,
113  * where a task's runnable state changes, and requeues, where a task
114  * and its state are being moved between CPUs and runqueues.
115  */
116 static inline void psi_enqueue(struct task_struct *p, bool wakeup)
117 {
118 	int clear = 0, set = TSK_RUNNING;
119 
120 	if (static_branch_likely(&psi_disabled))
121 		return;
122 
123 	if (p->in_memstall)
124 		set |= TSK_MEMSTALL_RUNNING;
125 
126 	if (!wakeup || p->sched_psi_wake_requeue) {
127 		if (p->in_memstall)
128 			set |= TSK_MEMSTALL;
129 		if (p->sched_psi_wake_requeue)
130 			p->sched_psi_wake_requeue = 0;
131 	} else {
132 		if (p->in_iowait)
133 			clear |= TSK_IOWAIT;
134 	}
135 
136 	psi_task_change(p, clear, set);
137 }
138 
139 static inline void psi_dequeue(struct task_struct *p, bool sleep)
140 {
141 	int clear = TSK_RUNNING;
142 
143 	if (static_branch_likely(&psi_disabled))
144 		return;
145 
146 	/*
147 	 * A voluntary sleep is a dequeue followed by a task switch. To
148 	 * avoid walking all ancestors twice, psi_task_switch() handles
149 	 * TSK_RUNNING and TSK_IOWAIT for us when it moves TSK_ONCPU.
150 	 * Do nothing here.
151 	 */
152 	if (sleep)
153 		return;
154 
155 	if (p->in_memstall)
156 		clear |= (TSK_MEMSTALL | TSK_MEMSTALL_RUNNING);
157 
158 	psi_task_change(p, clear, 0);
159 }
160 
161 static inline void psi_ttwu_dequeue(struct task_struct *p)
162 {
163 	if (static_branch_likely(&psi_disabled))
164 		return;
165 	/*
166 	 * Is the task being migrated during a wakeup? Make sure to
167 	 * deregister its sleep-persistent psi states from the old
168 	 * queue, and let psi_enqueue() know it has to requeue.
169 	 */
170 	if (unlikely(p->in_iowait || p->in_memstall)) {
171 		struct rq_flags rf;
172 		struct rq *rq;
173 		int clear = 0;
174 
175 		if (p->in_iowait)
176 			clear |= TSK_IOWAIT;
177 		if (p->in_memstall)
178 			clear |= TSK_MEMSTALL;
179 
180 		rq = __task_rq_lock(p, &rf);
181 		psi_task_change(p, clear, 0);
182 		p->sched_psi_wake_requeue = 1;
183 		__task_rq_unlock(rq, &rf);
184 	}
185 }
186 
187 static inline void psi_sched_switch(struct task_struct *prev,
188 				    struct task_struct *next,
189 				    bool sleep)
190 {
191 	if (static_branch_likely(&psi_disabled))
192 		return;
193 
194 	psi_task_switch(prev, next, sleep);
195 }
196 
197 #else /* CONFIG_PSI */
198 static inline void psi_enqueue(struct task_struct *p, bool wakeup) {}
199 static inline void psi_dequeue(struct task_struct *p, bool sleep) {}
200 static inline void psi_ttwu_dequeue(struct task_struct *p) {}
201 static inline void psi_sched_switch(struct task_struct *prev,
202 				    struct task_struct *next,
203 				    bool sleep) {}
204 #endif /* CONFIG_PSI */
205 
206 #ifdef CONFIG_SCHED_INFO
207 /*
208  * We are interested in knowing how long it was from the *first* time a
209  * task was queued to the time that it finally hit a CPU, we call this routine
210  * from dequeue_task() to account for possible rq->clock skew across CPUs. The
211  * delta taken on each CPU would annul the skew.
212  */
213 static inline void sched_info_dequeue(struct rq *rq, struct task_struct *t)
214 {
215 	unsigned long long delta = 0;
216 
217 	if (!t->sched_info.last_queued)
218 		return;
219 
220 	delta = rq_clock(rq) - t->sched_info.last_queued;
221 	t->sched_info.last_queued = 0;
222 	t->sched_info.run_delay += delta;
223 
224 	rq_sched_info_dequeue(rq, delta);
225 }
226 
227 /*
228  * Called when a task finally hits the CPU.  We can now calculate how
229  * long it was waiting to run.  We also note when it began so that we
230  * can keep stats on how long its timeslice is.
231  */
232 static void sched_info_arrive(struct rq *rq, struct task_struct *t)
233 {
234 	unsigned long long now, delta = 0;
235 
236 	if (!t->sched_info.last_queued)
237 		return;
238 
239 	now = rq_clock(rq);
240 	delta = now - t->sched_info.last_queued;
241 	t->sched_info.last_queued = 0;
242 	t->sched_info.run_delay += delta;
243 	t->sched_info.last_arrival = now;
244 	t->sched_info.pcount++;
245 
246 	rq_sched_info_arrive(rq, delta);
247 }
248 
249 /*
250  * This function is only called from enqueue_task(), but also only updates
251  * the timestamp if it is already not set.  It's assumed that
252  * sched_info_dequeue() will clear that stamp when appropriate.
253  */
254 static inline void sched_info_enqueue(struct rq *rq, struct task_struct *t)
255 {
256 	if (!t->sched_info.last_queued)
257 		t->sched_info.last_queued = rq_clock(rq);
258 }
259 
260 /*
261  * Called when a process ceases being the active-running process involuntarily
262  * due, typically, to expiring its time slice (this may also be called when
263  * switching to the idle task).  Now we can calculate how long we ran.
264  * Also, if the process is still in the TASK_RUNNING state, call
265  * sched_info_enqueue() to mark that it has now again started waiting on
266  * the runqueue.
267  */
268 static inline void sched_info_depart(struct rq *rq, struct task_struct *t)
269 {
270 	unsigned long long delta = rq_clock(rq) - t->sched_info.last_arrival;
271 
272 	rq_sched_info_depart(rq, delta);
273 
274 	if (task_is_running(t))
275 		sched_info_enqueue(rq, t);
276 }
277 
278 /*
279  * Called when tasks are switched involuntarily due, typically, to expiring
280  * their time slice.  (This may also be called when switching to or from
281  * the idle task.)  We are only called when prev != next.
282  */
283 static inline void
284 sched_info_switch(struct rq *rq, struct task_struct *prev, struct task_struct *next)
285 {
286 	/*
287 	 * prev now departs the CPU.  It's not interesting to record
288 	 * stats about how efficient we were at scheduling the idle
289 	 * process, however.
290 	 */
291 	if (prev != rq->idle)
292 		sched_info_depart(rq, prev);
293 
294 	if (next != rq->idle)
295 		sched_info_arrive(rq, next);
296 }
297 
298 #else /* !CONFIG_SCHED_INFO: */
299 # define sched_info_enqueue(rq, t)	do { } while (0)
300 # define sched_info_dequeue(rq, t)	do { } while (0)
301 # define sched_info_switch(rq, t, next)	do { } while (0)
302 #endif /* CONFIG_SCHED_INFO */
303 
304 #endif /* _KERNEL_STATS_H */
305