1
2#ifdef CONFIG_SCHEDSTATS
3
4/*
5 * Expects runqueue lock to be held for atomicity of update
6 */
7static inline void
8rq_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 */
19static inline void
20rq_sched_info_depart(struct rq *rq, unsigned long long delta)
21{
22	if (rq)
23		rq->rq_cpu_time += delta;
24}
25
26static inline void
27rq_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_inc(rq, field)	do { (rq)->field++; } while (0)
33# define schedstat_add(rq, field, amt)	do { (rq)->field += (amt); } while (0)
34# define schedstat_set(var, val)	do { var = (val); } while (0)
35#else /* !CONFIG_SCHEDSTATS */
36static inline void
37rq_sched_info_arrive(struct rq *rq, unsigned long long delta)
38{}
39static inline void
40rq_sched_info_dequeued(struct rq *rq, unsigned long long delta)
41{}
42static inline void
43rq_sched_info_depart(struct rq *rq, unsigned long long delta)
44{}
45# define schedstat_inc(rq, field)	do { } while (0)
46# define schedstat_add(rq, field, amt)	do { } while (0)
47# define schedstat_set(var, val)	do { } while (0)
48#endif
49
50#if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT)
51static inline void sched_info_reset_dequeued(struct task_struct *t)
52{
53	t->sched_info.last_queued = 0;
54}
55
56/*
57 * We are interested in knowing how long it was from the *first* time a
58 * task was queued to the time that it finally hit a cpu, we call this routine
59 * from dequeue_task() to account for possible rq->clock skew across cpus. The
60 * delta taken on each cpu would annul the skew.
61 */
62static inline void sched_info_dequeued(struct rq *rq, struct task_struct *t)
63{
64	unsigned long long now = rq_clock(rq), delta = 0;
65
66	if (unlikely(sched_info_on()))
67		if (t->sched_info.last_queued)
68			delta = now - t->sched_info.last_queued;
69	sched_info_reset_dequeued(t);
70	t->sched_info.run_delay += delta;
71
72	rq_sched_info_dequeued(rq, delta);
73}
74
75/*
76 * Called when a task finally hits the cpu.  We can now calculate how
77 * long it was waiting to run.  We also note when it began so that we
78 * can keep stats on how long its timeslice is.
79 */
80static void sched_info_arrive(struct rq *rq, struct task_struct *t)
81{
82	unsigned long long now = rq_clock(rq), delta = 0;
83
84	if (t->sched_info.last_queued)
85		delta = now - t->sched_info.last_queued;
86	sched_info_reset_dequeued(t);
87	t->sched_info.run_delay += delta;
88	t->sched_info.last_arrival = now;
89	t->sched_info.pcount++;
90
91	rq_sched_info_arrive(rq, delta);
92}
93
94/*
95 * This function is only called from enqueue_task(), but also only updates
96 * the timestamp if it is already not set.  It's assumed that
97 * sched_info_dequeued() will clear that stamp when appropriate.
98 */
99static inline void sched_info_queued(struct rq *rq, struct task_struct *t)
100{
101	if (unlikely(sched_info_on()))
102		if (!t->sched_info.last_queued)
103			t->sched_info.last_queued = rq_clock(rq);
104}
105
106/*
107 * Called when a process ceases being the active-running process involuntarily
108 * due, typically, to expiring its time slice (this may also be called when
109 * switching to the idle task).  Now we can calculate how long we ran.
110 * Also, if the process is still in the TASK_RUNNING state, call
111 * sched_info_queued() to mark that it has now again started waiting on
112 * the runqueue.
113 */
114static inline void sched_info_depart(struct rq *rq, struct task_struct *t)
115{
116	unsigned long long delta = rq_clock(rq) -
117					t->sched_info.last_arrival;
118
119	rq_sched_info_depart(rq, delta);
120
121	if (t->state == TASK_RUNNING)
122		sched_info_queued(rq, t);
123}
124
125/*
126 * Called when tasks are switched involuntarily due, typically, to expiring
127 * their time slice.  (This may also be called when switching to or from
128 * the idle task.)  We are only called when prev != next.
129 */
130static inline void
131__sched_info_switch(struct rq *rq,
132		    struct task_struct *prev, struct task_struct *next)
133{
134	/*
135	 * prev now departs the cpu.  It's not interesting to record
136	 * stats about how efficient we were at scheduling the idle
137	 * process, however.
138	 */
139	if (prev != rq->idle)
140		sched_info_depart(rq, prev);
141
142	if (next != rq->idle)
143		sched_info_arrive(rq, next);
144}
145static inline void
146sched_info_switch(struct rq *rq,
147		  struct task_struct *prev, struct task_struct *next)
148{
149	if (unlikely(sched_info_on()))
150		__sched_info_switch(rq, prev, next);
151}
152#else
153#define sched_info_queued(rq, t)		do { } while (0)
154#define sched_info_reset_dequeued(t)	do { } while (0)
155#define sched_info_dequeued(rq, t)		do { } while (0)
156#define sched_info_depart(rq, t)		do { } while (0)
157#define sched_info_arrive(rq, next)		do { } while (0)
158#define sched_info_switch(rq, t, next)		do { } while (0)
159#endif /* CONFIG_SCHEDSTATS || CONFIG_TASK_DELAY_ACCT */
160
161/*
162 * The following are functions that support scheduler-internal time accounting.
163 * These functions are generally called at the timer tick.  None of this depends
164 * on CONFIG_SCHEDSTATS.
165 */
166
167/**
168 * cputimer_running - return true if cputimer is running
169 *
170 * @tsk:	Pointer to target task.
171 */
172static inline bool cputimer_running(struct task_struct *tsk)
173
174{
175	struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
176
177	if (!cputimer->running)
178		return false;
179
180	/*
181	 * After we flush the task's sum_exec_runtime to sig->sum_sched_runtime
182	 * in __exit_signal(), we won't account to the signal struct further
183	 * cputime consumed by that task, even though the task can still be
184	 * ticking after __exit_signal().
185	 *
186	 * In order to keep a consistent behaviour between thread group cputime
187	 * and thread group cputimer accounting, lets also ignore the cputime
188	 * elapsing after __exit_signal() in any thread group timer running.
189	 *
190	 * This makes sure that POSIX CPU clocks and timers are synchronized, so
191	 * that a POSIX CPU timer won't expire while the corresponding POSIX CPU
192	 * clock delta is behind the expiring timer value.
193	 */
194	if (unlikely(!tsk->sighand))
195		return false;
196
197	return true;
198}
199
200/**
201 * account_group_user_time - Maintain utime for a thread group.
202 *
203 * @tsk:	Pointer to task structure.
204 * @cputime:	Time value by which to increment the utime field of the
205 *		thread_group_cputime structure.
206 *
207 * If thread group time is being maintained, get the structure for the
208 * running CPU and update the utime field there.
209 */
210static inline void account_group_user_time(struct task_struct *tsk,
211					   cputime_t cputime)
212{
213	struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
214
215	if (!cputimer_running(tsk))
216		return;
217
218	raw_spin_lock(&cputimer->lock);
219	cputimer->cputime.utime += cputime;
220	raw_spin_unlock(&cputimer->lock);
221}
222
223/**
224 * account_group_system_time - Maintain stime for a thread group.
225 *
226 * @tsk:	Pointer to task structure.
227 * @cputime:	Time value by which to increment the stime field of the
228 *		thread_group_cputime structure.
229 *
230 * If thread group time is being maintained, get the structure for the
231 * running CPU and update the stime field there.
232 */
233static inline void account_group_system_time(struct task_struct *tsk,
234					     cputime_t cputime)
235{
236	struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
237
238	if (!cputimer_running(tsk))
239		return;
240
241	raw_spin_lock(&cputimer->lock);
242	cputimer->cputime.stime += cputime;
243	raw_spin_unlock(&cputimer->lock);
244}
245
246/**
247 * account_group_exec_runtime - Maintain exec runtime for a thread group.
248 *
249 * @tsk:	Pointer to task structure.
250 * @ns:		Time value by which to increment the sum_exec_runtime field
251 *		of the thread_group_cputime structure.
252 *
253 * If thread group time is being maintained, get the structure for the
254 * running CPU and update the sum_exec_runtime field there.
255 */
256static inline void account_group_exec_runtime(struct task_struct *tsk,
257					      unsigned long long ns)
258{
259	struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
260
261	if (!cputimer_running(tsk))
262		return;
263
264	raw_spin_lock(&cputimer->lock);
265	cputimer->cputime.sum_exec_runtime += ns;
266	raw_spin_unlock(&cputimer->lock);
267}
268