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