kernel/sched_stats.h

   1
   2 #ifdef CONFIG_SCHEDSTATS
   3 /*
   4  * bump this up when changing the output format or the meaning of an existing
   5  * format, so that tools can adapt (or abort)
   6  */
   7 #define SCHEDSTAT_VERSION 14
   8
   9 static int show_schedstat(struct seq_file *seq, void *v)
  10 {
  11         int cpu;
  12
  13         seq_printf(seq, "version %d\n", SCHEDSTAT_VERSION);
  14         seq_printf(seq, "timestamp %lu\n", jiffies);
  15         for_each_online_cpu(cpu) {
  16                 struct rq *rq = cpu_rq(cpu);
  17 #ifdef CONFIG_SMP
  18                 struct sched_domain *sd;
  19                 int dcnt = 0;
  20 #endif
  21
  22                 /* runqueue-specific stats */
  23                 seq_printf(seq,
  24                     "cpu%d %lu %lu %lu %lu %lu %lu %lu %lu %lu %llu %llu %lu",
  25                     cpu, rq->yld_both_empty,
  26                     rq->yld_act_empty, rq->yld_exp_empty, rq->yld_cnt,
  27                     rq->sched_switch, rq->sched_cnt, rq->sched_goidle,
  28                     rq->ttwu_cnt, rq->ttwu_local,
  29                     rq->rq_sched_info.cpu_time,
  30                     rq->rq_sched_info.run_delay, rq->rq_sched_info.pcnt);
  31
  32                 seq_printf(seq, "\n");
  33
  34 #ifdef CONFIG_SMP
  35                 /* domain-specific stats */
  36                 preempt_disable();
  37                 for_each_domain(cpu, sd) {
  38                         enum cpu_idle_type itype;
  39                         char mask_str[NR_CPUS];
  40
  41                         cpumask_scnprintf(mask_str, NR_CPUS, sd->span);
  42                         seq_printf(seq, "domain%d %s", dcnt++, mask_str);
  43                         for (itype = CPU_IDLE; itype < CPU_MAX_IDLE_TYPES;
  44                                         itype++) {
  45                                 seq_printf(seq, " %lu %lu %lu %lu %lu %lu %lu "
  46                                                 "%lu",
  47                                     sd->lb_cnt[itype],
  48                                     sd->lb_balanced[itype],
  49                                     sd->lb_failed[itype],
  50                                     sd->lb_imbalance[itype],
  51                                     sd->lb_gained[itype],
  52                                     sd->lb_hot_gained[itype],
  53                                     sd->lb_nobusyq[itype],
  54                                     sd->lb_nobusyg[itype]);
  55                         }
  56                         seq_printf(seq, " %lu %lu %lu %lu %lu %lu %lu %lu %lu"
  57                             " %lu %lu %lu\n",
  58                             sd->alb_cnt, sd->alb_failed, sd->alb_pushed,
  59                             sd->sbe_cnt, sd->sbe_balanced, sd->sbe_pushed,
  60                             sd->sbf_cnt, sd->sbf_balanced, sd->sbf_pushed,
  61                             sd->ttwu_wake_remote, sd->ttwu_move_affine,
  62                             sd->ttwu_move_balance);
  63                 }
  64                 preempt_enable();
  65 #endif
  66         }
  67         return 0;
  68 }
  69
  70 static int schedstat_open(struct inode *inode, struct file *file)
  71 {
  72         unsigned int size = PAGE_SIZE * (1 + num_online_cpus() / 32);
  73         char *buf = kmalloc(size, GFP_KERNEL);
  74         struct seq_file *m;
  75         int res;
  76
  77         if (!buf)
  78                 return -ENOMEM;
  79         res = single_open(file, show_schedstat, NULL);
  80         if (!res) {
  81                 m = file->private_data;
  82                 m->buf = buf;
  83                 m->size = size;
  84         } else
  85                 kfree(buf);
  86         return res;
  87 }
  88
  89 const struct file_operations proc_schedstat_operations = {
  90         .open    = schedstat_open,
  91         .read    = seq_read,
  92         .llseek  = seq_lseek,
  93         .release = single_release,
  94 };
  95
  96 /*
  97  * Expects runqueue lock to be held for atomicity of update
  98  */
  99 static inline void
 100 rq_sched_info_arrive(struct rq *rq, unsigned long long delta)
 101 {
 102         if (rq) {
 103                 rq->rq_sched_info.run_delay += delta;
 104                 rq->rq_sched_info.pcnt++;
 105         }
 106 }
 107
 108 /*
 109  * Expects runqueue lock to be held for atomicity of update
 110  */
 111 static inline void
 112 rq_sched_info_depart(struct rq *rq, unsigned long long delta)
 113 {
 114         if (rq)
 115                 rq->rq_sched_info.cpu_time += delta;
 116 }
 117 # define schedstat_inc(rq, field)       do { (rq)->field++; } while (0)
 118 # define schedstat_add(rq, field, amt)  do { (rq)->field += (amt); } while (0)
 119 # define schedstat_set(var, val)        do { var = (val); } while (0)
 120 #else /* !CONFIG_SCHEDSTATS */
 121 static inline void
 122 rq_sched_info_arrive(struct rq *rq, unsigned long long delta)
 123 {}
 124 static inline void
 125 rq_sched_info_depart(struct rq *rq, unsigned long long delta)
 126 {}
 127 # define schedstat_inc(rq, field)       do { } while (0)
 128 # define schedstat_add(rq, field, amt)  do { } while (0)
 129 # define schedstat_set(var, val)        do { } while (0)
 130 #endif
 131
 132 #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT)
 133 /*
 134  * Called when a process is dequeued from the active array and given
 135  * the cpu.  We should note that with the exception of interactive
 136  * tasks, the expired queue will become the active queue after the active
 137  * queue is empty, without explicitly dequeuing and requeuing tasks in the
 138  * expired queue.  (Interactive tasks may be requeued directly to the
 139  * active queue, thus delaying tasks in the expired queue from running;
 140  * see scheduler_tick()).
 141  *
 142  * This function is only called from sched_info_arrive(), rather than
 143  * dequeue_task(). Even though a task may be queued and dequeued multiple
 144  * times as it is shuffled about, we're really interested in knowing how
 145  * long it was from the *first* time it was queued to the time that it
 146  * finally hit a cpu.
 147  */
 148 static inline void sched_info_dequeued(struct task_struct *t)
 149 {
 150         t->sched_info.last_queued = 0;
 151 }
 152
 153 /*
 154  * Called when a task finally hits the cpu.  We can now calculate how
 155  * long it was waiting to run.  We also note when it began so that we
 156  * can keep stats on how long its timeslice is.
 157  */
 158 static void sched_info_arrive(struct task_struct *t)
 159 {
 160         unsigned long long now = sched_clock(), delta = 0;
 161
 162         if (t->sched_info.last_queued)
 163                 delta = now - t->sched_info.last_queued;
 164         sched_info_dequeued(t);
 165         t->sched_info.run_delay += delta;
 166         t->sched_info.last_arrival = now;
 167         t->sched_info.pcnt++;
 168
 169         rq_sched_info_arrive(task_rq(t), delta);
 170 }
 171
 172 /*
 173  * Called when a process is queued into either the active or expired
 174  * array.  The time is noted and later used to determine how long we
 175  * had to wait for us to reach the cpu.  Since the expired queue will
 176  * become the active queue after active queue is empty, without dequeuing
 177  * and requeuing any tasks, we are interested in queuing to either. It
 178  * is unusual but not impossible for tasks to be dequeued and immediately
 179  * requeued in the same or another array: this can happen in sched_yield(),
 180  * set_user_nice(), and even load_balance() as it moves tasks from runqueue
 181  * to runqueue.
 182  *
 183  * This function is only called from enqueue_task(), but also only updates
 184  * the timestamp if it is already not set.  It's assumed that
 185  * sched_info_dequeued() will clear that stamp when appropriate.
 186  */
 187 static inline void sched_info_queued(struct task_struct *t)
 188 {
 189         if (unlikely(sched_info_on()))
 190                 if (!t->sched_info.last_queued)
 191                         t->sched_info.last_queued = sched_clock();
 192 }
 193
 194 /*
 195  * Called when a process ceases being the active-running process, either
 196  * voluntarily or involuntarily.  Now we can calculate how long we ran.
 197  */
 198 static inline void sched_info_depart(struct task_struct *t)
 199 {
 200         unsigned long long delta = sched_clock() - t->sched_info.last_arrival;
 201
 202         t->sched_info.cpu_time += delta;
 203         rq_sched_info_depart(task_rq(t), delta);
 204 }
 205
 206 /*
 207  * Called when tasks are switched involuntarily due, typically, to expiring
 208  * their time slice.  (This may also be called when switching to or from
 209  * the idle task.)  We are only called when prev != next.
 210  */
 211 static inline void
 212 __sched_info_switch(struct task_struct *prev, struct task_struct *next)
 213 {
 214         struct rq *rq = task_rq(prev);
 215
 216         /*
 217          * prev now departs the cpu.  It's not interesting to record
 218          * stats about how efficient we were at scheduling the idle
 219          * process, however.
 220          */
 221         if (prev != rq->idle)
 222                 sched_info_depart(prev);
 223
 224         if (next != rq->idle)
 225                 sched_info_arrive(next);
 226 }
 227 static inline void
 228 sched_info_switch(struct task_struct *prev, struct task_struct *next)
 229 {
 230         if (unlikely(sched_info_on()))
 231                 __sched_info_switch(prev, next);
 232 }
 233 #else
 234 #define sched_info_queued(t)            do { } while (0)
 235 #define sched_info_switch(t, next)      do { } while (0)
 236 #endif /* CONFIG_SCHEDSTATS || CONFIG_TASK_DELAY_ACCT */
 237