xref: /openbmc/linux/kernel/sched/stats.h (revision 6a551c11)
1 
2 #ifdef CONFIG_SCHEDSTATS
3 
4 /*
5  * Expects runqueue lock to be held for atomicity of update
6  */
7 static inline void
8 rq_sched_info_arrive(struct rq *rq, unsigned long long delta)
9 {
10 	if (rq) {
11 		rq->rq_sched_info.run_delay += delta;
12 		rq->rq_sched_info.pcount++;
13 	}
14 }
15 
16 /*
17  * Expects runqueue lock to be held for atomicity of update
18  */
19 static inline void
20 rq_sched_info_depart(struct rq *rq, unsigned long long delta)
21 {
22 	if (rq)
23 		rq->rq_cpu_time += delta;
24 }
25 
26 static inline void
27 rq_sched_info_dequeued(struct rq *rq, unsigned long long delta)
28 {
29 	if (rq)
30 		rq->rq_sched_info.run_delay += delta;
31 }
32 # define schedstat_enabled()		static_branch_unlikely(&sched_schedstats)
33 # define schedstat_inc(rq, field)	do { if (schedstat_enabled()) { (rq)->field++; } } while (0)
34 # define schedstat_add(rq, field, amt)	do { if (schedstat_enabled()) { (rq)->field += (amt); } } while (0)
35 # define schedstat_set(var, val)	do { if (schedstat_enabled()) { var = (val); } } while (0)
36 # define schedstat_val(rq, field)	((schedstat_enabled()) ? (rq)->field : 0)
37 
38 #else /* !CONFIG_SCHEDSTATS */
39 static inline void
40 rq_sched_info_arrive(struct rq *rq, unsigned long long delta)
41 {}
42 static inline void
43 rq_sched_info_dequeued(struct rq *rq, unsigned long long delta)
44 {}
45 static inline void
46 rq_sched_info_depart(struct rq *rq, unsigned long long delta)
47 {}
48 # define schedstat_enabled()		0
49 # define schedstat_inc(rq, field)	do { } while (0)
50 # define schedstat_add(rq, field, amt)	do { } while (0)
51 # define schedstat_set(var, val)	do { } while (0)
52 # define schedstat_val(rq, field)	0
53 #endif
54 
55 #ifdef CONFIG_SCHED_INFO
56 static inline void sched_info_reset_dequeued(struct task_struct *t)
57 {
58 	t->sched_info.last_queued = 0;
59 }
60 
61 /*
62  * We are interested in knowing how long it was from the *first* time a
63  * task was queued to the time that it finally hit a cpu, we call this routine
64  * from dequeue_task() to account for possible rq->clock skew across cpus. The
65  * delta taken on each cpu would annul the skew.
66  */
67 static inline void sched_info_dequeued(struct rq *rq, struct task_struct *t)
68 {
69 	unsigned long long now = rq_clock(rq), delta = 0;
70 
71 	if (unlikely(sched_info_on()))
72 		if (t->sched_info.last_queued)
73 			delta = now - t->sched_info.last_queued;
74 	sched_info_reset_dequeued(t);
75 	t->sched_info.run_delay += delta;
76 
77 	rq_sched_info_dequeued(rq, delta);
78 }
79 
80 /*
81  * Called when a task finally hits the cpu.  We can now calculate how
82  * long it was waiting to run.  We also note when it began so that we
83  * can keep stats on how long its timeslice is.
84  */
85 static void sched_info_arrive(struct rq *rq, struct task_struct *t)
86 {
87 	unsigned long long now = rq_clock(rq), delta = 0;
88 
89 	if (t->sched_info.last_queued)
90 		delta = now - t->sched_info.last_queued;
91 	sched_info_reset_dequeued(t);
92 	t->sched_info.run_delay += delta;
93 	t->sched_info.last_arrival = now;
94 	t->sched_info.pcount++;
95 
96 	rq_sched_info_arrive(rq, delta);
97 }
98 
99 /*
100  * This function is only called from enqueue_task(), but also only updates
101  * the timestamp if it is already not set.  It's assumed that
102  * sched_info_dequeued() will clear that stamp when appropriate.
103  */
104 static inline void sched_info_queued(struct rq *rq, struct task_struct *t)
105 {
106 	if (unlikely(sched_info_on()))
107 		if (!t->sched_info.last_queued)
108 			t->sched_info.last_queued = rq_clock(rq);
109 }
110 
111 /*
112  * Called when a process ceases being the active-running process involuntarily
113  * due, typically, to expiring its time slice (this may also be called when
114  * switching to the idle task).  Now we can calculate how long we ran.
115  * Also, if the process is still in the TASK_RUNNING state, call
116  * sched_info_queued() to mark that it has now again started waiting on
117  * the runqueue.
118  */
119 static inline void sched_info_depart(struct rq *rq, struct task_struct *t)
120 {
121 	unsigned long long delta = rq_clock(rq) -
122 					t->sched_info.last_arrival;
123 
124 	rq_sched_info_depart(rq, delta);
125 
126 	if (t->state == TASK_RUNNING)
127 		sched_info_queued(rq, t);
128 }
129 
130 /*
131  * Called when tasks are switched involuntarily due, typically, to expiring
132  * their time slice.  (This may also be called when switching to or from
133  * the idle task.)  We are only called when prev != next.
134  */
135 static inline void
136 __sched_info_switch(struct rq *rq,
137 		    struct task_struct *prev, struct task_struct *next)
138 {
139 	/*
140 	 * prev now departs the cpu.  It's not interesting to record
141 	 * stats about how efficient we were at scheduling the idle
142 	 * process, however.
143 	 */
144 	if (prev != rq->idle)
145 		sched_info_depart(rq, prev);
146 
147 	if (next != rq->idle)
148 		sched_info_arrive(rq, next);
149 }
150 static inline void
151 sched_info_switch(struct rq *rq,
152 		  struct task_struct *prev, struct task_struct *next)
153 {
154 	if (unlikely(sched_info_on()))
155 		__sched_info_switch(rq, prev, next);
156 }
157 #else
158 #define sched_info_queued(rq, t)		do { } while (0)
159 #define sched_info_reset_dequeued(t)	do { } while (0)
160 #define sched_info_dequeued(rq, t)		do { } while (0)
161 #define sched_info_depart(rq, t)		do { } while (0)
162 #define sched_info_arrive(rq, next)		do { } while (0)
163 #define sched_info_switch(rq, t, next)		do { } while (0)
164 #endif /* CONFIG_SCHED_INFO */
165 
166 /*
167  * The following are functions that support scheduler-internal time accounting.
168  * These functions are generally called at the timer tick.  None of this depends
169  * on CONFIG_SCHEDSTATS.
170  */
171 
172 /**
173  * cputimer_running - return true if cputimer is running
174  *
175  * @tsk:	Pointer to target task.
176  */
177 static inline bool cputimer_running(struct task_struct *tsk)
178 
179 {
180 	struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
181 
182 	/* Check if cputimer isn't running. This is accessed without locking. */
183 	if (!READ_ONCE(cputimer->running))
184 		return false;
185 
186 	/*
187 	 * After we flush the task's sum_exec_runtime to sig->sum_sched_runtime
188 	 * in __exit_signal(), we won't account to the signal struct further
189 	 * cputime consumed by that task, even though the task can still be
190 	 * ticking after __exit_signal().
191 	 *
192 	 * In order to keep a consistent behaviour between thread group cputime
193 	 * and thread group cputimer accounting, lets also ignore the cputime
194 	 * elapsing after __exit_signal() in any thread group timer running.
195 	 *
196 	 * This makes sure that POSIX CPU clocks and timers are synchronized, so
197 	 * that a POSIX CPU timer won't expire while the corresponding POSIX CPU
198 	 * clock delta is behind the expiring timer value.
199 	 */
200 	if (unlikely(!tsk->sighand))
201 		return false;
202 
203 	return true;
204 }
205 
206 /**
207  * account_group_user_time - Maintain utime for a thread group.
208  *
209  * @tsk:	Pointer to task structure.
210  * @cputime:	Time value by which to increment the utime field of the
211  *		thread_group_cputime structure.
212  *
213  * If thread group time is being maintained, get the structure for the
214  * running CPU and update the utime field there.
215  */
216 static inline void account_group_user_time(struct task_struct *tsk,
217 					   cputime_t cputime)
218 {
219 	struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
220 
221 	if (!cputimer_running(tsk))
222 		return;
223 
224 	atomic64_add(cputime, &cputimer->cputime_atomic.utime);
225 }
226 
227 /**
228  * account_group_system_time - Maintain stime for a thread group.
229  *
230  * @tsk:	Pointer to task structure.
231  * @cputime:	Time value by which to increment the stime field of the
232  *		thread_group_cputime structure.
233  *
234  * If thread group time is being maintained, get the structure for the
235  * running CPU and update the stime field there.
236  */
237 static inline void account_group_system_time(struct task_struct *tsk,
238 					     cputime_t cputime)
239 {
240 	struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
241 
242 	if (!cputimer_running(tsk))
243 		return;
244 
245 	atomic64_add(cputime, &cputimer->cputime_atomic.stime);
246 }
247 
248 /**
249  * account_group_exec_runtime - Maintain exec runtime for a thread group.
250  *
251  * @tsk:	Pointer to task structure.
252  * @ns:		Time value by which to increment the sum_exec_runtime field
253  *		of the thread_group_cputime structure.
254  *
255  * If thread group time is being maintained, get the structure for the
256  * running CPU and update the sum_exec_runtime field there.
257  */
258 static inline void account_group_exec_runtime(struct task_struct *tsk,
259 					      unsigned long long ns)
260 {
261 	struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
262 
263 	if (!cputimer_running(tsk))
264 		return;
265 
266 	atomic64_add(ns, &cputimer->cputime_atomic.sum_exec_runtime);
267 }
268