xref: /openbmc/linux/kernel/sched/stats.h (revision 0a907292)
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 (!wakeup || p->sched_psi_wake_requeue) {
122 		if (p->in_memstall)
123 			set |= TSK_MEMSTALL;
124 		if (p->sched_psi_wake_requeue)
125 			p->sched_psi_wake_requeue = 0;
126 	} else {
127 		if (p->in_iowait)
128 			clear |= TSK_IOWAIT;
129 	}
130 
131 	psi_task_change(p, clear, set);
132 }
133 
134 static inline void psi_dequeue(struct task_struct *p, bool sleep)
135 {
136 	int clear = TSK_RUNNING;
137 
138 	if (static_branch_likely(&psi_disabled))
139 		return;
140 
141 	/*
142 	 * A voluntary sleep is a dequeue followed by a task switch. To
143 	 * avoid walking all ancestors twice, psi_task_switch() handles
144 	 * TSK_RUNNING and TSK_IOWAIT for us when it moves TSK_ONCPU.
145 	 * Do nothing here.
146 	 */
147 	if (sleep)
148 		return;
149 
150 	if (p->in_memstall)
151 		clear |= TSK_MEMSTALL;
152 
153 	psi_task_change(p, clear, 0);
154 }
155 
156 static inline void psi_ttwu_dequeue(struct task_struct *p)
157 {
158 	if (static_branch_likely(&psi_disabled))
159 		return;
160 	/*
161 	 * Is the task being migrated during a wakeup? Make sure to
162 	 * deregister its sleep-persistent psi states from the old
163 	 * queue, and let psi_enqueue() know it has to requeue.
164 	 */
165 	if (unlikely(p->in_iowait || p->in_memstall)) {
166 		struct rq_flags rf;
167 		struct rq *rq;
168 		int clear = 0;
169 
170 		if (p->in_iowait)
171 			clear |= TSK_IOWAIT;
172 		if (p->in_memstall)
173 			clear |= TSK_MEMSTALL;
174 
175 		rq = __task_rq_lock(p, &rf);
176 		psi_task_change(p, clear, 0);
177 		p->sched_psi_wake_requeue = 1;
178 		__task_rq_unlock(rq, &rf);
179 	}
180 }
181 
182 static inline void psi_sched_switch(struct task_struct *prev,
183 				    struct task_struct *next,
184 				    bool sleep)
185 {
186 	if (static_branch_likely(&psi_disabled))
187 		return;
188 
189 	psi_task_switch(prev, next, sleep);
190 }
191 
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,
198 				    bool sleep) {}
199 #endif /* CONFIG_PSI */
200 
201 #ifdef CONFIG_SCHED_INFO
202 /*
203  * We are interested in knowing how long it was from the *first* time a
204  * task was queued to the time that it finally hit a CPU, we call this routine
205  * from dequeue_task() to account for possible rq->clock skew across CPUs. The
206  * delta taken on each CPU would annul the skew.
207  */
208 static inline void sched_info_dequeue(struct rq *rq, struct task_struct *t)
209 {
210 	unsigned long long delta = 0;
211 
212 	if (!t->sched_info.last_queued)
213 		return;
214 
215 	delta = rq_clock(rq) - t->sched_info.last_queued;
216 	t->sched_info.last_queued = 0;
217 	t->sched_info.run_delay += delta;
218 
219 	rq_sched_info_dequeue(rq, delta);
220 }
221 
222 /*
223  * Called when a task finally hits the CPU.  We can now calculate how
224  * long it was waiting to run.  We also note when it began so that we
225  * can keep stats on how long its timeslice is.
226  */
227 static void sched_info_arrive(struct rq *rq, struct task_struct *t)
228 {
229 	unsigned long long now, delta = 0;
230 
231 	if (!t->sched_info.last_queued)
232 		return;
233 
234 	now = rq_clock(rq);
235 	delta = now - t->sched_info.last_queued;
236 	t->sched_info.last_queued = 0;
237 	t->sched_info.run_delay += delta;
238 	t->sched_info.last_arrival = now;
239 	t->sched_info.pcount++;
240 
241 	rq_sched_info_arrive(rq, delta);
242 }
243 
244 /*
245  * This function is only called from enqueue_task(), but also only updates
246  * the timestamp if it is already not set.  It's assumed that
247  * sched_info_dequeue() will clear that stamp when appropriate.
248  */
249 static inline void sched_info_enqueue(struct rq *rq, struct task_struct *t)
250 {
251 	if (!t->sched_info.last_queued)
252 		t->sched_info.last_queued = rq_clock(rq);
253 }
254 
255 /*
256  * Called when a process ceases being the active-running process involuntarily
257  * due, typically, to expiring its time slice (this may also be called when
258  * switching to the idle task).  Now we can calculate how long we ran.
259  * Also, if the process is still in the TASK_RUNNING state, call
260  * sched_info_enqueue() to mark that it has now again started waiting on
261  * the runqueue.
262  */
263 static inline void sched_info_depart(struct rq *rq, struct task_struct *t)
264 {
265 	unsigned long long delta = rq_clock(rq) - t->sched_info.last_arrival;
266 
267 	rq_sched_info_depart(rq, delta);
268 
269 	if (task_is_running(t))
270 		sched_info_enqueue(rq, t);
271 }
272 
273 /*
274  * Called when tasks are switched involuntarily due, typically, to expiring
275  * their time slice.  (This may also be called when switching to or from
276  * the idle task.)  We are only called when prev != next.
277  */
278 static inline void
279 sched_info_switch(struct rq *rq, struct task_struct *prev, struct task_struct *next)
280 {
281 	/*
282 	 * prev now departs the CPU.  It's not interesting to record
283 	 * stats about how efficient we were at scheduling the idle
284 	 * process, however.
285 	 */
286 	if (prev != rq->idle)
287 		sched_info_depart(rq, prev);
288 
289 	if (next != rq->idle)
290 		sched_info_arrive(rq, next);
291 }
292 
293 #else /* !CONFIG_SCHED_INFO: */
294 # define sched_info_enqueue(rq, t)	do { } while (0)
295 # define sched_info_dequeue(rq, t)	do { } while (0)
296 # define sched_info_switch(rq, t, next)	do { } while (0)
297 #endif /* CONFIG_SCHED_INFO */
298