Loading...
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 task_struct *t)
63{
64 unsigned long long now = task_rq(t)->clock, 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(task_rq(t), 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 task_struct *t)
81{
82 unsigned long long now = task_rq(t)->clock, 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(task_rq(t), 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 task_struct *t)
100{
101 if (unlikely(sched_info_on()))
102 if (!t->sched_info.last_queued)
103 t->sched_info.last_queued = task_rq(t)->clock;
104}
105
106/*
107 * Called when a process ceases being the active-running process, either
108 * voluntarily or involuntarily. Now we can calculate how long we ran.
109 * Also, if the process is still in the TASK_RUNNING state, call
110 * sched_info_queued() to mark that it has now again started waiting on
111 * the runqueue.
112 */
113static inline void sched_info_depart(struct task_struct *t)
114{
115 unsigned long long delta = task_rq(t)->clock -
116 t->sched_info.last_arrival;
117
118 rq_sched_info_depart(task_rq(t), delta);
119
120 if (t->state == TASK_RUNNING)
121 sched_info_queued(t);
122}
123
124/*
125 * Called when tasks are switched involuntarily due, typically, to expiring
126 * their time slice. (This may also be called when switching to or from
127 * the idle task.) We are only called when prev != next.
128 */
129static inline void
130__sched_info_switch(struct task_struct *prev, struct task_struct *next)
131{
132 struct rq *rq = task_rq(prev);
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(prev);
141
142 if (next != rq->idle)
143 sched_info_arrive(next);
144}
145static inline void
146sched_info_switch(struct task_struct *prev, struct task_struct *next)
147{
148 if (unlikely(sched_info_on()))
149 __sched_info_switch(prev, next);
150}
151#else
152#define sched_info_queued(t) do { } while (0)
153#define sched_info_reset_dequeued(t) do { } while (0)
154#define sched_info_dequeued(t) do { } while (0)
155#define sched_info_switch(t, next) do { } while (0)
156#endif /* CONFIG_SCHEDSTATS || CONFIG_TASK_DELAY_ACCT */
157
158/*
159 * The following are functions that support scheduler-internal time accounting.
160 * These functions are generally called at the timer tick. None of this depends
161 * on CONFIG_SCHEDSTATS.
162 */
163
164/**
165 * account_group_user_time - Maintain utime for a thread group.
166 *
167 * @tsk: Pointer to task structure.
168 * @cputime: Time value by which to increment the utime field of the
169 * thread_group_cputime structure.
170 *
171 * If thread group time is being maintained, get the structure for the
172 * running CPU and update the utime field there.
173 */
174static inline void account_group_user_time(struct task_struct *tsk,
175 cputime_t cputime)
176{
177 struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
178
179 if (!cputimer->running)
180 return;
181
182 raw_spin_lock(&cputimer->lock);
183 cputimer->cputime.utime += cputime;
184 raw_spin_unlock(&cputimer->lock);
185}
186
187/**
188 * account_group_system_time - Maintain stime for a thread group.
189 *
190 * @tsk: Pointer to task structure.
191 * @cputime: Time value by which to increment the stime field of the
192 * thread_group_cputime structure.
193 *
194 * If thread group time is being maintained, get the structure for the
195 * running CPU and update the stime field there.
196 */
197static inline void account_group_system_time(struct task_struct *tsk,
198 cputime_t cputime)
199{
200 struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
201
202 if (!cputimer->running)
203 return;
204
205 raw_spin_lock(&cputimer->lock);
206 cputimer->cputime.stime += cputime;
207 raw_spin_unlock(&cputimer->lock);
208}
209
210/**
211 * account_group_exec_runtime - Maintain exec runtime for a thread group.
212 *
213 * @tsk: Pointer to task structure.
214 * @ns: Time value by which to increment the sum_exec_runtime field
215 * of the thread_group_cputime structure.
216 *
217 * If thread group time is being maintained, get the structure for the
218 * running CPU and update the sum_exec_runtime field there.
219 */
220static inline void account_group_exec_runtime(struct task_struct *tsk,
221 unsigned long long ns)
222{
223 struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
224
225 if (!cputimer->running)
226 return;
227
228 raw_spin_lock(&cputimer->lock);
229 cputimer->cputime.sum_exec_runtime += ns;
230 raw_spin_unlock(&cputimer->lock);
231}
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_enabled() static_branch_unlikely(&sched_schedstats)
33#define schedstat_inc(var) do { if (schedstat_enabled()) { var++; } } while (0)
34#define schedstat_add(var, amt) do { if (schedstat_enabled()) { var += (amt); } } while (0)
35#define schedstat_set(var, val) do { if (schedstat_enabled()) { var = (val); } } while (0)
36#define schedstat_val(var) (var)
37#define schedstat_val_or_zero(var) ((schedstat_enabled()) ? (var) : 0)
38
39#else /* !CONFIG_SCHEDSTATS */
40static inline void
41rq_sched_info_arrive(struct rq *rq, unsigned long long delta)
42{}
43static inline void
44rq_sched_info_dequeued(struct rq *rq, unsigned long long delta)
45{}
46static inline void
47rq_sched_info_depart(struct rq *rq, unsigned long long delta)
48{}
49#define schedstat_enabled() 0
50#define schedstat_inc(var) do { } while (0)
51#define schedstat_add(var, amt) do { } while (0)
52#define schedstat_set(var, val) do { } while (0)
53#define schedstat_val(var) 0
54#define schedstat_val_or_zero(var) 0
55#endif /* CONFIG_SCHEDSTATS */
56
57#ifdef CONFIG_SCHED_INFO
58static inline void sched_info_reset_dequeued(struct task_struct *t)
59{
60 t->sched_info.last_queued = 0;
61}
62
63/*
64 * We are interested in knowing how long it was from the *first* time a
65 * task was queued to the time that it finally hit a cpu, we call this routine
66 * from dequeue_task() to account for possible rq->clock skew across cpus. The
67 * delta taken on each cpu would annul the skew.
68 */
69static inline void sched_info_dequeued(struct rq *rq, struct task_struct *t)
70{
71 unsigned long long now = rq_clock(rq), delta = 0;
72
73 if (unlikely(sched_info_on()))
74 if (t->sched_info.last_queued)
75 delta = now - t->sched_info.last_queued;
76 sched_info_reset_dequeued(t);
77 t->sched_info.run_delay += delta;
78
79 rq_sched_info_dequeued(rq, delta);
80}
81
82/*
83 * Called when a task finally hits the cpu. We can now calculate how
84 * long it was waiting to run. We also note when it began so that we
85 * can keep stats on how long its timeslice is.
86 */
87static void sched_info_arrive(struct rq *rq, struct task_struct *t)
88{
89 unsigned long long now = rq_clock(rq), delta = 0;
90
91 if (t->sched_info.last_queued)
92 delta = now - t->sched_info.last_queued;
93 sched_info_reset_dequeued(t);
94 t->sched_info.run_delay += delta;
95 t->sched_info.last_arrival = now;
96 t->sched_info.pcount++;
97
98 rq_sched_info_arrive(rq, delta);
99}
100
101/*
102 * This function is only called from enqueue_task(), but also only updates
103 * the timestamp if it is already not set. It's assumed that
104 * sched_info_dequeued() will clear that stamp when appropriate.
105 */
106static inline void sched_info_queued(struct rq *rq, struct task_struct *t)
107{
108 if (unlikely(sched_info_on()))
109 if (!t->sched_info.last_queued)
110 t->sched_info.last_queued = rq_clock(rq);
111}
112
113/*
114 * Called when a process ceases being the active-running process involuntarily
115 * due, typically, to expiring its time slice (this may also be called when
116 * switching to the idle task). Now we can calculate how long we ran.
117 * Also, if the process is still in the TASK_RUNNING state, call
118 * sched_info_queued() to mark that it has now again started waiting on
119 * the runqueue.
120 */
121static inline void sched_info_depart(struct rq *rq, struct task_struct *t)
122{
123 unsigned long long delta = rq_clock(rq) -
124 t->sched_info.last_arrival;
125
126 rq_sched_info_depart(rq, delta);
127
128 if (t->state == TASK_RUNNING)
129 sched_info_queued(rq, t);
130}
131
132/*
133 * Called when tasks are switched involuntarily due, typically, to expiring
134 * their time slice. (This may also be called when switching to or from
135 * the idle task.) We are only called when prev != next.
136 */
137static inline void
138__sched_info_switch(struct rq *rq,
139 struct task_struct *prev, struct task_struct *next)
140{
141 /*
142 * prev now departs the cpu. It's not interesting to record
143 * stats about how efficient we were at scheduling the idle
144 * process, however.
145 */
146 if (prev != rq->idle)
147 sched_info_depart(rq, prev);
148
149 if (next != rq->idle)
150 sched_info_arrive(rq, next);
151}
152static inline void
153sched_info_switch(struct rq *rq,
154 struct task_struct *prev, struct task_struct *next)
155{
156 if (unlikely(sched_info_on()))
157 __sched_info_switch(rq, prev, next);
158}
159#else
160#define sched_info_queued(rq, t) do { } while (0)
161#define sched_info_reset_dequeued(t) do { } while (0)
162#define sched_info_dequeued(rq, t) do { } while (0)
163#define sched_info_depart(rq, t) do { } while (0)
164#define sched_info_arrive(rq, next) do { } while (0)
165#define sched_info_switch(rq, t, next) do { } while (0)
166#endif /* CONFIG_SCHED_INFO */
167
168/*
169 * The following are functions that support scheduler-internal time accounting.
170 * These functions are generally called at the timer tick. None of this depends
171 * on CONFIG_SCHEDSTATS.
172 */
173
174/**
175 * cputimer_running - return true if cputimer is running
176 *
177 * @tsk: Pointer to target task.
178 */
179static inline bool cputimer_running(struct task_struct *tsk)
180
181{
182 struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
183
184 /* Check if cputimer isn't running. This is accessed without locking. */
185 if (!READ_ONCE(cputimer->running))
186 return false;
187
188 /*
189 * After we flush the task's sum_exec_runtime to sig->sum_sched_runtime
190 * in __exit_signal(), we won't account to the signal struct further
191 * cputime consumed by that task, even though the task can still be
192 * ticking after __exit_signal().
193 *
194 * In order to keep a consistent behaviour between thread group cputime
195 * and thread group cputimer accounting, lets also ignore the cputime
196 * elapsing after __exit_signal() in any thread group timer running.
197 *
198 * This makes sure that POSIX CPU clocks and timers are synchronized, so
199 * that a POSIX CPU timer won't expire while the corresponding POSIX CPU
200 * clock delta is behind the expiring timer value.
201 */
202 if (unlikely(!tsk->sighand))
203 return false;
204
205 return true;
206}
207
208/**
209 * account_group_user_time - Maintain utime for a thread group.
210 *
211 * @tsk: Pointer to task structure.
212 * @cputime: Time value by which to increment the utime field of the
213 * thread_group_cputime structure.
214 *
215 * If thread group time is being maintained, get the structure for the
216 * running CPU and update the utime field there.
217 */
218static inline void account_group_user_time(struct task_struct *tsk,
219 cputime_t cputime)
220{
221 struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
222
223 if (!cputimer_running(tsk))
224 return;
225
226 atomic64_add(cputime, &cputimer->cputime_atomic.utime);
227}
228
229/**
230 * account_group_system_time - Maintain stime for a thread group.
231 *
232 * @tsk: Pointer to task structure.
233 * @cputime: Time value by which to increment the stime field of the
234 * thread_group_cputime structure.
235 *
236 * If thread group time is being maintained, get the structure for the
237 * running CPU and update the stime field there.
238 */
239static inline void account_group_system_time(struct task_struct *tsk,
240 cputime_t cputime)
241{
242 struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
243
244 if (!cputimer_running(tsk))
245 return;
246
247 atomic64_add(cputime, &cputimer->cputime_atomic.stime);
248}
249
250/**
251 * account_group_exec_runtime - Maintain exec runtime for a thread group.
252 *
253 * @tsk: Pointer to task structure.
254 * @ns: Time value by which to increment the sum_exec_runtime field
255 * of the thread_group_cputime structure.
256 *
257 * If thread group time is being maintained, get the structure for the
258 * running CPU and update the sum_exec_runtime field there.
259 */
260static inline void account_group_exec_runtime(struct task_struct *tsk,
261 unsigned long long ns)
262{
263 struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
264
265 if (!cputimer_running(tsk))
266 return;
267
268 atomic64_add(ns, &cputimer->cputime_atomic.sum_exec_runtime);
269}