xref: /openbmc/linux/kernel/sched/stats.h (revision f68f2ff9)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 
3 #ifdef CONFIG_SCHEDSTATS
4 
5 extern struct static_key_false sched_schedstats;
6 
7 /*
8  * Expects runqueue lock to be held for atomicity of update
9  */
10 static inline void
11 rq_sched_info_arrive(struct rq *rq, unsigned long long delta)
12 {
13 	if (rq) {
14 		rq->rq_sched_info.run_delay += delta;
15 		rq->rq_sched_info.pcount++;
16 	}
17 }
18 
19 /*
20  * Expects runqueue lock to be held for atomicity of update
21  */
22 static inline void
23 rq_sched_info_depart(struct rq *rq, unsigned long long delta)
24 {
25 	if (rq)
26 		rq->rq_cpu_time += delta;
27 }
28 
29 static inline void
30 rq_sched_info_dequeue(struct rq *rq, unsigned long long delta)
31 {
32 	if (rq)
33 		rq->rq_sched_info.run_delay += delta;
34 }
35 #define   schedstat_enabled()		static_branch_unlikely(&sched_schedstats)
36 #define __schedstat_inc(var)		do { var++; } while (0)
37 #define   schedstat_inc(var)		do { if (schedstat_enabled()) { var++; } } while (0)
38 #define __schedstat_add(var, amt)	do { var += (amt); } while (0)
39 #define   schedstat_add(var, amt)	do { if (schedstat_enabled()) { var += (amt); } } while (0)
40 #define __schedstat_set(var, val)	do { var = (val); } while (0)
41 #define   schedstat_set(var, val)	do { if (schedstat_enabled()) { var = (val); } } while (0)
42 #define   schedstat_val(var)		(var)
43 #define   schedstat_val_or_zero(var)	((schedstat_enabled()) ? (var) : 0)
44 
45 void __update_stats_wait_start(struct rq *rq, struct task_struct *p,
46 			       struct sched_statistics *stats);
47 
48 void __update_stats_wait_end(struct rq *rq, struct task_struct *p,
49 			     struct sched_statistics *stats);
50 void __update_stats_enqueue_sleeper(struct rq *rq, struct task_struct *p,
51 				    struct sched_statistics *stats);
52 
53 static inline void
54 check_schedstat_required(void)
55 {
56 	if (schedstat_enabled())
57 		return;
58 
59 	/* Force schedstat enabled if a dependent tracepoint is active */
60 	if (trace_sched_stat_wait_enabled()    ||
61 	    trace_sched_stat_sleep_enabled()   ||
62 	    trace_sched_stat_iowait_enabled()  ||
63 	    trace_sched_stat_blocked_enabled() ||
64 	    trace_sched_stat_runtime_enabled())
65 		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");
66 }
67 
68 #else /* !CONFIG_SCHEDSTATS: */
69 
70 static inline void rq_sched_info_arrive  (struct rq *rq, unsigned long long delta) { }
71 static inline void rq_sched_info_dequeue(struct rq *rq, unsigned long long delta) { }
72 static inline void rq_sched_info_depart  (struct rq *rq, unsigned long long delta) { }
73 # define   schedstat_enabled()		0
74 # define __schedstat_inc(var)		do { } while (0)
75 # define   schedstat_inc(var)		do { } while (0)
76 # define __schedstat_add(var, amt)	do { } while (0)
77 # define   schedstat_add(var, amt)	do { } while (0)
78 # define __schedstat_set(var, val)	do { } while (0)
79 # define   schedstat_set(var, val)	do { } while (0)
80 # define   schedstat_val(var)		0
81 # define   schedstat_val_or_zero(var)	0
82 
83 # define __update_stats_wait_start(rq, p, stats)       do { } while (0)
84 # define __update_stats_wait_end(rq, p, stats)         do { } while (0)
85 # define __update_stats_enqueue_sleeper(rq, p, stats)  do { } while (0)
86 # define check_schedstat_required()                    do { } while (0)
87 
88 #endif /* CONFIG_SCHEDSTATS */
89 
90 #ifdef CONFIG_FAIR_GROUP_SCHED
91 struct sched_entity_stats {
92 	struct sched_entity     se;
93 	struct sched_statistics stats;
94 } __no_randomize_layout;
95 #endif
96 
97 static inline struct sched_statistics *
98 __schedstats_from_se(struct sched_entity *se)
99 {
100 #ifdef CONFIG_FAIR_GROUP_SCHED
101 	if (!entity_is_task(se))
102 		return &container_of(se, struct sched_entity_stats, se)->stats;
103 #endif
104 	return &task_of(se)->stats;
105 }
106 
107 #ifdef CONFIG_PSI
108 /*
109  * PSI tracks state that persists across sleeps, such as iowaits and
110  * memory stalls. As a result, it has to distinguish between sleeps,
111  * where a task's runnable state changes, and requeues, where a task
112  * and its state are being moved between CPUs and runqueues.
113  */
114 static inline void psi_enqueue(struct task_struct *p, bool wakeup)
115 {
116 	int clear = 0, set = TSK_RUNNING;
117 
118 	if (static_branch_likely(&psi_disabled))
119 		return;
120 
121 	if (p->in_memstall)
122 		set |= TSK_MEMSTALL_RUNNING;
123 
124 	if (!wakeup || p->sched_psi_wake_requeue) {
125 		if (p->in_memstall)
126 			set |= TSK_MEMSTALL;
127 		if (p->sched_psi_wake_requeue)
128 			p->sched_psi_wake_requeue = 0;
129 	} else {
130 		if (p->in_iowait)
131 			clear |= TSK_IOWAIT;
132 	}
133 
134 	psi_task_change(p, clear, set);
135 }
136 
137 static inline void psi_dequeue(struct task_struct *p, bool sleep)
138 {
139 	int clear = TSK_RUNNING;
140 
141 	if (static_branch_likely(&psi_disabled))
142 		return;
143 
144 	/*
145 	 * A voluntary sleep is a dequeue followed by a task switch. To
146 	 * avoid walking all ancestors twice, psi_task_switch() handles
147 	 * TSK_RUNNING and TSK_IOWAIT for us when it moves TSK_ONCPU.
148 	 * Do nothing here.
149 	 */
150 	if (sleep)
151 		return;
152 
153 	if (p->in_memstall)
154 		clear |= (TSK_MEMSTALL | TSK_MEMSTALL_RUNNING);
155 
156 	psi_task_change(p, clear, 0);
157 }
158 
159 static inline void psi_ttwu_dequeue(struct task_struct *p)
160 {
161 	if (static_branch_likely(&psi_disabled))
162 		return;
163 	/*
164 	 * Is the task being migrated during a wakeup? Make sure to
165 	 * deregister its sleep-persistent psi states from the old
166 	 * queue, and let psi_enqueue() know it has to requeue.
167 	 */
168 	if (unlikely(p->in_iowait || p->in_memstall)) {
169 		struct rq_flags rf;
170 		struct rq *rq;
171 		int clear = 0;
172 
173 		if (p->in_iowait)
174 			clear |= TSK_IOWAIT;
175 		if (p->in_memstall)
176 			clear |= TSK_MEMSTALL;
177 
178 		rq = __task_rq_lock(p, &rf);
179 		psi_task_change(p, clear, 0);
180 		p->sched_psi_wake_requeue = 1;
181 		__task_rq_unlock(rq, &rf);
182 	}
183 }
184 
185 static inline void psi_sched_switch(struct task_struct *prev,
186 				    struct task_struct *next,
187 				    bool sleep)
188 {
189 	if (static_branch_likely(&psi_disabled))
190 		return;
191 
192 	psi_task_switch(prev, next, sleep);
193 }
194 
195 #else /* CONFIG_PSI */
196 static inline void psi_enqueue(struct task_struct *p, bool wakeup) {}
197 static inline void psi_dequeue(struct task_struct *p, bool sleep) {}
198 static inline void psi_ttwu_dequeue(struct task_struct *p) {}
199 static inline void psi_sched_switch(struct task_struct *prev,
200 				    struct task_struct *next,
201 				    bool sleep) {}
202 #endif /* CONFIG_PSI */
203 
204 #ifdef CONFIG_SCHED_INFO
205 /*
206  * We are interested in knowing how long it was from the *first* time a
207  * task was queued to the time that it finally hit a CPU, we call this routine
208  * from dequeue_task() to account for possible rq->clock skew across CPUs. The
209  * delta taken on each CPU would annul the skew.
210  */
211 static inline void sched_info_dequeue(struct rq *rq, struct task_struct *t)
212 {
213 	unsigned long long delta = 0;
214 
215 	if (!t->sched_info.last_queued)
216 		return;
217 
218 	delta = rq_clock(rq) - t->sched_info.last_queued;
219 	t->sched_info.last_queued = 0;
220 	t->sched_info.run_delay += delta;
221 
222 	rq_sched_info_dequeue(rq, delta);
223 }
224 
225 /*
226  * Called when a task finally hits the CPU.  We can now calculate how
227  * long it was waiting to run.  We also note when it began so that we
228  * can keep stats on how long its timeslice is.
229  */
230 static void sched_info_arrive(struct rq *rq, struct task_struct *t)
231 {
232 	unsigned long long now, delta = 0;
233 
234 	if (!t->sched_info.last_queued)
235 		return;
236 
237 	now = rq_clock(rq);
238 	delta = now - t->sched_info.last_queued;
239 	t->sched_info.last_queued = 0;
240 	t->sched_info.run_delay += delta;
241 	t->sched_info.last_arrival = now;
242 	t->sched_info.pcount++;
243 
244 	rq_sched_info_arrive(rq, delta);
245 }
246 
247 /*
248  * This function is only called from enqueue_task(), but also only updates
249  * the timestamp if it is already not set.  It's assumed that
250  * sched_info_dequeue() will clear that stamp when appropriate.
251  */
252 static inline void sched_info_enqueue(struct rq *rq, struct task_struct *t)
253 {
254 	if (!t->sched_info.last_queued)
255 		t->sched_info.last_queued = rq_clock(rq);
256 }
257 
258 /*
259  * Called when a process ceases being the active-running process involuntarily
260  * due, typically, to expiring its time slice (this may also be called when
261  * switching to the idle task).  Now we can calculate how long we ran.
262  * Also, if the process is still in the TASK_RUNNING state, call
263  * sched_info_enqueue() to mark that it has now again started waiting on
264  * the runqueue.
265  */
266 static inline void sched_info_depart(struct rq *rq, struct task_struct *t)
267 {
268 	unsigned long long delta = rq_clock(rq) - t->sched_info.last_arrival;
269 
270 	rq_sched_info_depart(rq, delta);
271 
272 	if (task_is_running(t))
273 		sched_info_enqueue(rq, t);
274 }
275 
276 /*
277  * Called when tasks are switched involuntarily due, typically, to expiring
278  * their time slice.  (This may also be called when switching to or from
279  * the idle task.)  We are only called when prev != next.
280  */
281 static inline void
282 sched_info_switch(struct rq *rq, struct task_struct *prev, struct task_struct *next)
283 {
284 	/*
285 	 * prev now departs the CPU.  It's not interesting to record
286 	 * stats about how efficient we were at scheduling the idle
287 	 * process, however.
288 	 */
289 	if (prev != rq->idle)
290 		sched_info_depart(rq, prev);
291 
292 	if (next != rq->idle)
293 		sched_info_arrive(rq, next);
294 }
295 
296 #else /* !CONFIG_SCHED_INFO: */
297 # define sched_info_enqueue(rq, t)	do { } while (0)
298 # define sched_info_dequeue(rq, t)	do { } while (0)
299 # define sched_info_switch(rq, t, next)	do { } while (0)
300 #endif /* CONFIG_SCHED_INFO */
301