xref: /openbmc/linux/kernel/sched/stats.h (revision 349f631d)
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_dequeue(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_dequeue(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->in_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;
88 
89 	if (static_branch_likely(&psi_disabled))
90 		return;
91 
92 	/*
93 	 * A voluntary sleep is a dequeue followed by a task switch. To
94 	 * avoid walking all ancestors twice, psi_task_switch() handles
95 	 * TSK_RUNNING and TSK_IOWAIT for us when it moves TSK_ONCPU.
96 	 * Do nothing here.
97 	 */
98 	if (sleep)
99 		return;
100 
101 	if (p->in_memstall)
102 		clear |= TSK_MEMSTALL;
103 
104 	psi_task_change(p, clear, 0);
105 }
106 
107 static inline void psi_ttwu_dequeue(struct task_struct *p)
108 {
109 	if (static_branch_likely(&psi_disabled))
110 		return;
111 	/*
112 	 * Is the task being migrated during a wakeup? Make sure to
113 	 * deregister its sleep-persistent psi states from the old
114 	 * queue, and let psi_enqueue() know it has to requeue.
115 	 */
116 	if (unlikely(p->in_iowait || p->in_memstall)) {
117 		struct rq_flags rf;
118 		struct rq *rq;
119 		int clear = 0;
120 
121 		if (p->in_iowait)
122 			clear |= TSK_IOWAIT;
123 		if (p->in_memstall)
124 			clear |= TSK_MEMSTALL;
125 
126 		rq = __task_rq_lock(p, &rf);
127 		psi_task_change(p, clear, 0);
128 		p->sched_psi_wake_requeue = 1;
129 		__task_rq_unlock(rq, &rf);
130 	}
131 }
132 
133 static inline void psi_sched_switch(struct task_struct *prev,
134 				    struct task_struct *next,
135 				    bool sleep)
136 {
137 	if (static_branch_likely(&psi_disabled))
138 		return;
139 
140 	psi_task_switch(prev, next, sleep);
141 }
142 
143 #else /* CONFIG_PSI */
144 static inline void psi_enqueue(struct task_struct *p, bool wakeup) {}
145 static inline void psi_dequeue(struct task_struct *p, bool sleep) {}
146 static inline void psi_ttwu_dequeue(struct task_struct *p) {}
147 static inline void psi_sched_switch(struct task_struct *prev,
148 				    struct task_struct *next,
149 				    bool sleep) {}
150 #endif /* CONFIG_PSI */
151 
152 #ifdef CONFIG_SCHED_INFO
153 /*
154  * We are interested in knowing how long it was from the *first* time a
155  * task was queued to the time that it finally hit a CPU, we call this routine
156  * from dequeue_task() to account for possible rq->clock skew across CPUs. The
157  * delta taken on each CPU would annul the skew.
158  */
159 static inline void sched_info_dequeue(struct rq *rq, struct task_struct *t)
160 {
161 	unsigned long long delta = 0;
162 
163 	if (!t->sched_info.last_queued)
164 		return;
165 
166 	delta = rq_clock(rq) - t->sched_info.last_queued;
167 	t->sched_info.last_queued = 0;
168 	t->sched_info.run_delay += delta;
169 
170 	rq_sched_info_dequeue(rq, delta);
171 }
172 
173 /*
174  * Called when a task finally hits the CPU.  We can now calculate how
175  * long it was waiting to run.  We also note when it began so that we
176  * can keep stats on how long its timeslice is.
177  */
178 static void sched_info_arrive(struct rq *rq, struct task_struct *t)
179 {
180 	unsigned long long now, delta = 0;
181 
182 	if (!t->sched_info.last_queued)
183 		return;
184 
185 	now = rq_clock(rq);
186 	delta = now - t->sched_info.last_queued;
187 	t->sched_info.last_queued = 0;
188 	t->sched_info.run_delay += delta;
189 	t->sched_info.last_arrival = now;
190 	t->sched_info.pcount++;
191 
192 	rq_sched_info_arrive(rq, delta);
193 }
194 
195 /*
196  * This function is only called from enqueue_task(), but also only updates
197  * the timestamp if it is already not set.  It's assumed that
198  * sched_info_dequeue() will clear that stamp when appropriate.
199  */
200 static inline void sched_info_enqueue(struct rq *rq, struct task_struct *t)
201 {
202 	if (!t->sched_info.last_queued)
203 		t->sched_info.last_queued = rq_clock(rq);
204 }
205 
206 /*
207  * Called when a process ceases being the active-running process involuntarily
208  * due, typically, to expiring its time slice (this may also be called when
209  * switching to the idle task).  Now we can calculate how long we ran.
210  * Also, if the process is still in the TASK_RUNNING state, call
211  * sched_info_enqueue() to mark that it has now again started waiting on
212  * the runqueue.
213  */
214 static inline void sched_info_depart(struct rq *rq, struct task_struct *t)
215 {
216 	unsigned long long delta = rq_clock(rq) - t->sched_info.last_arrival;
217 
218 	rq_sched_info_depart(rq, delta);
219 
220 	if (task_is_running(t))
221 		sched_info_enqueue(rq, t);
222 }
223 
224 /*
225  * Called when tasks are switched involuntarily due, typically, to expiring
226  * their time slice.  (This may also be called when switching to or from
227  * the idle task.)  We are only called when prev != next.
228  */
229 static inline void
230 sched_info_switch(struct rq *rq, struct task_struct *prev, struct task_struct *next)
231 {
232 	/*
233 	 * prev now departs the CPU.  It's not interesting to record
234 	 * stats about how efficient we were at scheduling the idle
235 	 * process, however.
236 	 */
237 	if (prev != rq->idle)
238 		sched_info_depart(rq, prev);
239 
240 	if (next != rq->idle)
241 		sched_info_arrive(rq, next);
242 }
243 
244 #else /* !CONFIG_SCHED_INFO: */
245 # define sched_info_enqueue(rq, t)	do { } while (0)
246 # define sched_info_dequeue(rq, t)	do { } while (0)
247 # define sched_info_switch(rq, t, next)	do { } while (0)
248 #endif /* CONFIG_SCHED_INFO */
249