xref: /linux/kernel/sched/stats.h (revision 17cfcb68af3bc7d5e8ae08779b1853310a2949f3)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 
3 #ifdef CONFIG_SCHEDSTATS
4 
5 /*
6  * Expects runqueue lock to be held for atomicity of update
7  */
8 static inline void
9 rq_sched_info_arrive(struct rq *rq, unsigned long long delta)
10 {
11 	if (rq) {
12 		rq->rq_sched_info.run_delay += delta;
13 		rq->rq_sched_info.pcount++;
14 	}
15 }
16 
17 /*
18  * Expects runqueue lock to be held for atomicity of update
19  */
20 static inline void
21 rq_sched_info_depart(struct rq *rq, unsigned long long delta)
22 {
23 	if (rq)
24 		rq->rq_cpu_time += delta;
25 }
26 
27 static inline void
28 rq_sched_info_dequeued(struct rq *rq, unsigned long long delta)
29 {
30 	if (rq)
31 		rq->rq_sched_info.run_delay += delta;
32 }
33 #define   schedstat_enabled()		static_branch_unlikely(&sched_schedstats)
34 #define __schedstat_inc(var)		do { var++; } while (0)
35 #define   schedstat_inc(var)		do { if (schedstat_enabled()) { var++; } } while (0)
36 #define __schedstat_add(var, amt)	do { var += (amt); } while (0)
37 #define   schedstat_add(var, amt)	do { if (schedstat_enabled()) { var += (amt); } } while (0)
38 #define __schedstat_set(var, val)	do { var = (val); } while (0)
39 #define   schedstat_set(var, val)	do { if (schedstat_enabled()) { var = (val); } } while (0)
40 #define   schedstat_val(var)		(var)
41 #define   schedstat_val_or_zero(var)	((schedstat_enabled()) ? (var) : 0)
42 
43 #else /* !CONFIG_SCHEDSTATS: */
44 static inline void rq_sched_info_arrive  (struct rq *rq, unsigned long long delta) { }
45 static inline void rq_sched_info_dequeued(struct rq *rq, unsigned long long delta) { }
46 static inline void rq_sched_info_depart  (struct rq *rq, unsigned long long delta) { }
47 # define   schedstat_enabled()		0
48 # define __schedstat_inc(var)		do { } while (0)
49 # define   schedstat_inc(var)		do { } while (0)
50 # define __schedstat_add(var, amt)	do { } while (0)
51 # define   schedstat_add(var, amt)	do { } while (0)
52 # define __schedstat_set(var, val)	do { } while (0)
53 # define   schedstat_set(var, val)	do { } while (0)
54 # define   schedstat_val(var)		0
55 # define   schedstat_val_or_zero(var)	0
56 #endif /* CONFIG_SCHEDSTATS */
57 
58 #ifdef CONFIG_PSI
59 /*
60  * PSI tracks state that persists across sleeps, such as iowaits and
61  * memory stalls. As a result, it has to distinguish between sleeps,
62  * where a task's runnable state changes, and requeues, where a task
63  * and its state are being moved between CPUs and runqueues.
64  */
65 static inline void psi_enqueue(struct task_struct *p, bool wakeup)
66 {
67 	int clear = 0, set = TSK_RUNNING;
68 
69 	if (static_branch_likely(&psi_disabled))
70 		return;
71 
72 	if (!wakeup || p->sched_psi_wake_requeue) {
73 		if (p->flags & PF_MEMSTALL)
74 			set |= TSK_MEMSTALL;
75 		if (p->sched_psi_wake_requeue)
76 			p->sched_psi_wake_requeue = 0;
77 	} else {
78 		if (p->in_iowait)
79 			clear |= TSK_IOWAIT;
80 	}
81 
82 	psi_task_change(p, clear, set);
83 }
84 
85 static inline void psi_dequeue(struct task_struct *p, bool sleep)
86 {
87 	int clear = TSK_RUNNING, set = 0;
88 
89 	if (static_branch_likely(&psi_disabled))
90 		return;
91 
92 	if (!sleep) {
93 		if (p->flags & PF_MEMSTALL)
94 			clear |= TSK_MEMSTALL;
95 	} else {
96 		if (p->in_iowait)
97 			set |= TSK_IOWAIT;
98 	}
99 
100 	psi_task_change(p, clear, set);
101 }
102 
103 static inline void psi_ttwu_dequeue(struct task_struct *p)
104 {
105 	if (static_branch_likely(&psi_disabled))
106 		return;
107 	/*
108 	 * Is the task being migrated during a wakeup? Make sure to
109 	 * deregister its sleep-persistent psi states from the old
110 	 * queue, and let psi_enqueue() know it has to requeue.
111 	 */
112 	if (unlikely(p->in_iowait || (p->flags & PF_MEMSTALL))) {
113 		struct rq_flags rf;
114 		struct rq *rq;
115 		int clear = 0;
116 
117 		if (p->in_iowait)
118 			clear |= TSK_IOWAIT;
119 		if (p->flags & PF_MEMSTALL)
120 			clear |= TSK_MEMSTALL;
121 
122 		rq = __task_rq_lock(p, &rf);
123 		psi_task_change(p, clear, 0);
124 		p->sched_psi_wake_requeue = 1;
125 		__task_rq_unlock(rq, &rf);
126 	}
127 }
128 
129 static inline void psi_task_tick(struct rq *rq)
130 {
131 	if (static_branch_likely(&psi_disabled))
132 		return;
133 
134 	if (unlikely(rq->curr->flags & PF_MEMSTALL))
135 		psi_memstall_tick(rq->curr, cpu_of(rq));
136 }
137 #else /* CONFIG_PSI */
138 static inline void psi_enqueue(struct task_struct *p, bool wakeup) {}
139 static inline void psi_dequeue(struct task_struct *p, bool sleep) {}
140 static inline void psi_ttwu_dequeue(struct task_struct *p) {}
141 static inline void psi_task_tick(struct rq *rq) {}
142 #endif /* CONFIG_PSI */
143 
144 #ifdef CONFIG_SCHED_INFO
145 static inline void sched_info_reset_dequeued(struct task_struct *t)
146 {
147 	t->sched_info.last_queued = 0;
148 }
149 
150 /*
151  * We are interested in knowing how long it was from the *first* time a
152  * task was queued to the time that it finally hit a CPU, we call this routine
153  * from dequeue_task() to account for possible rq->clock skew across CPUs. The
154  * delta taken on each CPU would annul the skew.
155  */
156 static inline void sched_info_dequeued(struct rq *rq, struct task_struct *t)
157 {
158 	unsigned long long now = rq_clock(rq), delta = 0;
159 
160 	if (sched_info_on()) {
161 		if (t->sched_info.last_queued)
162 			delta = now - t->sched_info.last_queued;
163 	}
164 	sched_info_reset_dequeued(t);
165 	t->sched_info.run_delay += delta;
166 
167 	rq_sched_info_dequeued(rq, delta);
168 }
169 
170 /*
171  * Called when a task finally hits the CPU.  We can now calculate how
172  * long it was waiting to run.  We also note when it began so that we
173  * can keep stats on how long its timeslice is.
174  */
175 static void sched_info_arrive(struct rq *rq, struct task_struct *t)
176 {
177 	unsigned long long now = rq_clock(rq), delta = 0;
178 
179 	if (t->sched_info.last_queued)
180 		delta = now - t->sched_info.last_queued;
181 	sched_info_reset_dequeued(t);
182 	t->sched_info.run_delay += delta;
183 	t->sched_info.last_arrival = now;
184 	t->sched_info.pcount++;
185 
186 	rq_sched_info_arrive(rq, delta);
187 }
188 
189 /*
190  * This function is only called from enqueue_task(), but also only updates
191  * the timestamp if it is already not set.  It's assumed that
192  * sched_info_dequeued() will clear that stamp when appropriate.
193  */
194 static inline void sched_info_queued(struct rq *rq, struct task_struct *t)
195 {
196 	if (sched_info_on()) {
197 		if (!t->sched_info.last_queued)
198 			t->sched_info.last_queued = rq_clock(rq);
199 	}
200 }
201 
202 /*
203  * Called when a process ceases being the active-running process involuntarily
204  * due, typically, to expiring its time slice (this may also be called when
205  * switching to the idle task).  Now we can calculate how long we ran.
206  * Also, if the process is still in the TASK_RUNNING state, call
207  * sched_info_queued() to mark that it has now again started waiting on
208  * the runqueue.
209  */
210 static inline void sched_info_depart(struct rq *rq, struct task_struct *t)
211 {
212 	unsigned long long delta = rq_clock(rq) - t->sched_info.last_arrival;
213 
214 	rq_sched_info_depart(rq, delta);
215 
216 	if (t->state == TASK_RUNNING)
217 		sched_info_queued(rq, t);
218 }
219 
220 /*
221  * Called when tasks are switched involuntarily due, typically, to expiring
222  * their time slice.  (This may also be called when switching to or from
223  * the idle task.)  We are only called when prev != next.
224  */
225 static inline void
226 __sched_info_switch(struct rq *rq, struct task_struct *prev, struct task_struct *next)
227 {
228 	/*
229 	 * prev now departs the CPU.  It's not interesting to record
230 	 * stats about how efficient we were at scheduling the idle
231 	 * process, however.
232 	 */
233 	if (prev != rq->idle)
234 		sched_info_depart(rq, prev);
235 
236 	if (next != rq->idle)
237 		sched_info_arrive(rq, next);
238 }
239 
240 static inline void
241 sched_info_switch(struct rq *rq, struct task_struct *prev, struct task_struct *next)
242 {
243 	if (sched_info_on())
244 		__sched_info_switch(rq, prev, next);
245 }
246 
247 #else /* !CONFIG_SCHED_INFO: */
248 # define sched_info_queued(rq, t)	do { } while (0)
249 # define sched_info_reset_dequeued(t)	do { } while (0)
250 # define sched_info_dequeued(rq, t)	do { } while (0)
251 # define sched_info_depart(rq, t)	do { } while (0)
252 # define sched_info_arrive(rq, next)	do { } while (0)
253 # define sched_info_switch(rq, t, next)	do { } while (0)
254 #endif /* CONFIG_SCHED_INFO */
255