Loading...
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_dequeued(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_dequeued(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->flags & PF_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, set = 0;
88
89 if (static_branch_likely(&psi_disabled))
90 return;
91
92 if (!sleep) {
93 if (p->flags & PF_MEMSTALL)
94 clear |= TSK_MEMSTALL;
95 } else {
96 if (p->in_iowait)
97 set |= TSK_IOWAIT;
98 }
99
100 psi_task_change(p, clear, set);
101}
102
103static inline void psi_ttwu_dequeue(struct task_struct *p)
104{
105 if (static_branch_likely(&psi_disabled))
106 return;
107 /*
108 * Is the task being migrated during a wakeup? Make sure to
109 * deregister its sleep-persistent psi states from the old
110 * queue, and let psi_enqueue() know it has to requeue.
111 */
112 if (unlikely(p->in_iowait || (p->flags & PF_MEMSTALL))) {
113 struct rq_flags rf;
114 struct rq *rq;
115 int clear = 0;
116
117 if (p->in_iowait)
118 clear |= TSK_IOWAIT;
119 if (p->flags & PF_MEMSTALL)
120 clear |= TSK_MEMSTALL;
121
122 rq = __task_rq_lock(p, &rf);
123 psi_task_change(p, clear, 0);
124 p->sched_psi_wake_requeue = 1;
125 __task_rq_unlock(rq, &rf);
126 }
127}
128
129static inline void psi_task_tick(struct rq *rq)
130{
131 if (static_branch_likely(&psi_disabled))
132 return;
133
134 if (unlikely(rq->curr->flags & PF_MEMSTALL))
135 psi_memstall_tick(rq->curr, cpu_of(rq));
136}
137#else /* CONFIG_PSI */
138static inline void psi_enqueue(struct task_struct *p, bool wakeup) {}
139static inline void psi_dequeue(struct task_struct *p, bool sleep) {}
140static inline void psi_ttwu_dequeue(struct task_struct *p) {}
141static inline void psi_task_tick(struct rq *rq) {}
142#endif /* CONFIG_PSI */
143
144#ifdef CONFIG_SCHED_INFO
145static inline void sched_info_reset_dequeued(struct task_struct *t)
146{
147 t->sched_info.last_queued = 0;
148}
149
150/*
151 * We are interested in knowing how long it was from the *first* time a
152 * task was queued to the time that it finally hit a CPU, we call this routine
153 * from dequeue_task() to account for possible rq->clock skew across CPUs. The
154 * delta taken on each CPU would annul the skew.
155 */
156static inline void sched_info_dequeued(struct rq *rq, struct task_struct *t)
157{
158 unsigned long long now = rq_clock(rq), delta = 0;
159
160 if (sched_info_on()) {
161 if (t->sched_info.last_queued)
162 delta = now - t->sched_info.last_queued;
163 }
164 sched_info_reset_dequeued(t);
165 t->sched_info.run_delay += delta;
166
167 rq_sched_info_dequeued(rq, delta);
168}
169
170/*
171 * Called when a task finally hits the CPU. We can now calculate how
172 * long it was waiting to run. We also note when it began so that we
173 * can keep stats on how long its timeslice is.
174 */
175static void sched_info_arrive(struct rq *rq, struct task_struct *t)
176{
177 unsigned long long now = rq_clock(rq), delta = 0;
178
179 if (t->sched_info.last_queued)
180 delta = now - t->sched_info.last_queued;
181 sched_info_reset_dequeued(t);
182 t->sched_info.run_delay += delta;
183 t->sched_info.last_arrival = now;
184 t->sched_info.pcount++;
185
186 rq_sched_info_arrive(rq, delta);
187}
188
189/*
190 * This function is only called from enqueue_task(), but also only updates
191 * the timestamp if it is already not set. It's assumed that
192 * sched_info_dequeued() will clear that stamp when appropriate.
193 */
194static inline void sched_info_queued(struct rq *rq, struct task_struct *t)
195{
196 if (sched_info_on()) {
197 if (!t->sched_info.last_queued)
198 t->sched_info.last_queued = rq_clock(rq);
199 }
200}
201
202/*
203 * Called when a process ceases being the active-running process involuntarily
204 * due, typically, to expiring its time slice (this may also be called when
205 * switching to the idle task). Now we can calculate how long we ran.
206 * Also, if the process is still in the TASK_RUNNING state, call
207 * sched_info_queued() to mark that it has now again started waiting on
208 * the runqueue.
209 */
210static inline void sched_info_depart(struct rq *rq, struct task_struct *t)
211{
212 unsigned long long delta = rq_clock(rq) - t->sched_info.last_arrival;
213
214 rq_sched_info_depart(rq, delta);
215
216 if (t->state == TASK_RUNNING)
217 sched_info_queued(rq, t);
218}
219
220/*
221 * Called when tasks are switched involuntarily due, typically, to expiring
222 * their time slice. (This may also be called when switching to or from
223 * the idle task.) We are only called when prev != next.
224 */
225static inline void
226__sched_info_switch(struct rq *rq, struct task_struct *prev, struct task_struct *next)
227{
228 /*
229 * prev now departs the CPU. It's not interesting to record
230 * stats about how efficient we were at scheduling the idle
231 * process, however.
232 */
233 if (prev != rq->idle)
234 sched_info_depart(rq, prev);
235
236 if (next != rq->idle)
237 sched_info_arrive(rq, next);
238}
239
240static inline void
241sched_info_switch(struct rq *rq, struct task_struct *prev, struct task_struct *next)
242{
243 if (sched_info_on())
244 __sched_info_switch(rq, prev, next);
245}
246
247#else /* !CONFIG_SCHED_INFO: */
248# define sched_info_queued(rq, t) do { } while (0)
249# define sched_info_reset_dequeued(t) do { } while (0)
250# define sched_info_dequeued(rq, t) do { } while (0)
251# define sched_info_depart(rq, t) do { } while (0)
252# define sched_info_arrive(rq, next) do { } while (0)
253# define sched_info_switch(rq, t, next) do { } while (0)
254#endif /* CONFIG_SCHED_INFO */
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}