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_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}
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 */
8static inline void
9rq_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 */
20static inline void
21rq_sched_info_depart(struct rq *rq, unsigned long long delta)
22{
23 if (rq)
24 rq->rq_cpu_time += delta;
25}
26
27static inline void
28rq_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: */
44static inline void rq_sched_info_arrive (struct rq *rq, unsigned long long delta) { }
45static inline void rq_sched_info_dequeue(struct rq *rq, unsigned long long delta) { }
46static 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 */
65static 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
85static 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
107static 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
133static 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 */
144static inline void psi_enqueue(struct task_struct *p, bool wakeup) {}
145static inline void psi_dequeue(struct task_struct *p, bool sleep) {}
146static inline void psi_ttwu_dequeue(struct task_struct *p) {}
147static 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 */
159static 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 */
178static 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 */
200static 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 */
214static 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 */
229static inline void
230sched_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 */