2 #ifdef CONFIG_SCHEDSTATS
4 * bump this up when changing the output format or the meaning of an existing
5 * format, so that tools can adapt (or abort)
7 #define SCHEDSTAT_VERSION 14
9 static int show_schedstat(struct seq_file *seq, void *v)
12 int mask_len = NR_CPUS/32 * 9;
13 char *mask_str = kmalloc(mask_len, GFP_KERNEL);
18 seq_printf(seq, "version %d\n", SCHEDSTAT_VERSION);
19 seq_printf(seq, "timestamp %lu\n", jiffies);
20 for_each_online_cpu(cpu) {
21 struct rq *rq = cpu_rq(cpu);
23 struct sched_domain *sd;
27 /* runqueue-specific stats */
29 "cpu%d %u %u %u %u %u %u %u %u %u %llu %llu %lu",
30 cpu, rq->yld_both_empty,
31 rq->yld_act_empty, rq->yld_exp_empty, rq->yld_count,
32 rq->sched_switch, rq->sched_count, rq->sched_goidle,
33 rq->ttwu_count, rq->ttwu_local,
34 rq->rq_sched_info.cpu_time,
35 rq->rq_sched_info.run_delay, rq->rq_sched_info.pcount);
37 seq_printf(seq, "\n");
40 /* domain-specific stats */
42 for_each_domain(cpu, sd) {
43 enum cpu_idle_type itype;
45 cpumask_scnprintf(mask_str, mask_len, sd->span);
46 seq_printf(seq, "domain%d %s", dcount++, mask_str);
47 for (itype = CPU_IDLE; itype < CPU_MAX_IDLE_TYPES;
49 seq_printf(seq, " %u %u %u %u %u %u %u %u",
51 sd->lb_balanced[itype],
53 sd->lb_imbalance[itype],
55 sd->lb_hot_gained[itype],
56 sd->lb_nobusyq[itype],
57 sd->lb_nobusyg[itype]);
60 " %u %u %u %u %u %u %u %u %u %u %u %u\n",
61 sd->alb_count, sd->alb_failed, sd->alb_pushed,
62 sd->sbe_count, sd->sbe_balanced, sd->sbe_pushed,
63 sd->sbf_count, sd->sbf_balanced, sd->sbf_pushed,
64 sd->ttwu_wake_remote, sd->ttwu_move_affine,
65 sd->ttwu_move_balance);
73 static int schedstat_open(struct inode *inode, struct file *file)
75 unsigned int size = PAGE_SIZE * (1 + num_online_cpus() / 32);
76 char *buf = kmalloc(size, GFP_KERNEL);
82 res = single_open(file, show_schedstat, NULL);
84 m = file->private_data;
92 const struct file_operations proc_schedstat_operations = {
93 .open = schedstat_open,
96 .release = single_release,
100 * Expects runqueue lock to be held for atomicity of update
103 rq_sched_info_arrive(struct rq *rq, unsigned long long delta)
106 rq->rq_sched_info.run_delay += delta;
107 rq->rq_sched_info.pcount++;
112 * Expects runqueue lock to be held for atomicity of update
115 rq_sched_info_depart(struct rq *rq, unsigned long long delta)
118 rq->rq_sched_info.cpu_time += delta;
120 # define schedstat_inc(rq, field) do { (rq)->field++; } while (0)
121 # define schedstat_add(rq, field, amt) do { (rq)->field += (amt); } while (0)
122 # define schedstat_set(var, val) do { var = (val); } while (0)
123 #else /* !CONFIG_SCHEDSTATS */
125 rq_sched_info_arrive(struct rq *rq, unsigned long long delta)
128 rq_sched_info_depart(struct rq *rq, unsigned long long delta)
130 # define schedstat_inc(rq, field) do { } while (0)
131 # define schedstat_add(rq, field, amt) do { } while (0)
132 # define schedstat_set(var, val) do { } while (0)
135 #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT)
137 * Called when a process is dequeued from the active array and given
138 * the cpu. We should note that with the exception of interactive
139 * tasks, the expired queue will become the active queue after the active
140 * queue is empty, without explicitly dequeuing and requeuing tasks in the
141 * expired queue. (Interactive tasks may be requeued directly to the
142 * active queue, thus delaying tasks in the expired queue from running;
143 * see scheduler_tick()).
145 * This function is only called from sched_info_arrive(), rather than
146 * dequeue_task(). Even though a task may be queued and dequeued multiple
147 * times as it is shuffled about, we're really interested in knowing how
148 * long it was from the *first* time it was queued to the time that it
151 static inline void sched_info_dequeued(struct task_struct *t)
153 t->sched_info.last_queued = 0;
157 * Called when a task finally hits the cpu. We can now calculate how
158 * long it was waiting to run. We also note when it began so that we
159 * can keep stats on how long its timeslice is.
161 static void sched_info_arrive(struct task_struct *t)
163 unsigned long long now = task_rq(t)->clock, delta = 0;
165 if (t->sched_info.last_queued)
166 delta = now - t->sched_info.last_queued;
167 sched_info_dequeued(t);
168 t->sched_info.run_delay += delta;
169 t->sched_info.last_arrival = now;
170 t->sched_info.pcount++;
172 rq_sched_info_arrive(task_rq(t), delta);
176 * Called when a process is queued into either the active or expired
177 * array. The time is noted and later used to determine how long we
178 * had to wait for us to reach the cpu. Since the expired queue will
179 * become the active queue after active queue is empty, without dequeuing
180 * and requeuing any tasks, we are interested in queuing to either. It
181 * is unusual but not impossible for tasks to be dequeued and immediately
182 * requeued in the same or another array: this can happen in sched_yield(),
183 * set_user_nice(), and even load_balance() as it moves tasks from runqueue
186 * This function is only called from enqueue_task(), but also only updates
187 * the timestamp if it is already not set. It's assumed that
188 * sched_info_dequeued() will clear that stamp when appropriate.
190 static inline void sched_info_queued(struct task_struct *t)
192 if (unlikely(sched_info_on()))
193 if (!t->sched_info.last_queued)
194 t->sched_info.last_queued = task_rq(t)->clock;
198 * Called when a process ceases being the active-running process, either
199 * voluntarily or involuntarily. Now we can calculate how long we ran.
201 static inline void sched_info_depart(struct task_struct *t)
203 unsigned long long delta = task_rq(t)->clock -
204 t->sched_info.last_arrival;
206 t->sched_info.cpu_time += delta;
207 rq_sched_info_depart(task_rq(t), delta);
211 * Called when tasks are switched involuntarily due, typically, to expiring
212 * their time slice. (This may also be called when switching to or from
213 * the idle task.) We are only called when prev != next.
216 __sched_info_switch(struct task_struct *prev, struct task_struct *next)
218 struct rq *rq = task_rq(prev);
221 * prev now departs the cpu. It's not interesting to record
222 * stats about how efficient we were at scheduling the idle
225 if (prev != rq->idle)
226 sched_info_depart(prev);
228 if (next != rq->idle)
229 sched_info_arrive(next);
232 sched_info_switch(struct task_struct *prev, struct task_struct *next)
234 if (unlikely(sched_info_on()))
235 __sched_info_switch(prev, next);
238 #define sched_info_queued(t) do { } while (0)
239 #define sched_info_switch(t, next) do { } while (0)
240 #endif /* CONFIG_SCHEDSTATS || CONFIG_TASK_DELAY_ACCT */