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         int mask_len = NR_CPUS/32 * 9;
  13         char *mask_str = kmalloc(mask_len, GFP_KERNEL);
  14
  15         if (mask_str == NULL)
  16                 return -ENOMEM;
  17
  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);
  22 #ifdef CONFIG_SMP
  23                 struct sched_domain *sd;
  24                 int dcount = 0;
  25 #endif
  26
  27                 /* runqueue-specific stats */
  28                 seq_printf(seq,
  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);
  36
  37                 seq_printf(seq, "\n");
  38
  39 #ifdef CONFIG_SMP
  40                 /* domain-specific stats */
  41                 preempt_disable();
  42                 for_each_domain(cpu, sd) {
  43                         enum cpu_idle_type itype;
  44
  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;
  48                                         itype++) {
  49                                 seq_printf(seq, " %u %u %u %u %u %u %u %u",
  50                                     sd->lb_count[itype],
  51                                     sd->lb_balanced[itype],
  52                                     sd->lb_failed[itype],
  53                                     sd->lb_imbalance[itype],
  54                                     sd->lb_gained[itype],
  55                                     sd->lb_hot_gained[itype],
  56                                     sd->lb_nobusyq[itype],
  57                                     sd->lb_nobusyg[itype]);
  58                         }
  59                         seq_printf(seq,
  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);
  66                 }
  67                 preempt_enable();
  68 #endif
  69         }
  70         kfree(mask_str);
  71         return 0;
  72 }
  73
  74 static int schedstat_open(struct inode *inode, struct file *file)
  75 {
  76         unsigned int size = PAGE_SIZE * (1 + num_online_cpus() / 32);
  77         char *buf = kmalloc(size, GFP_KERNEL);
  78         struct seq_file *m;
  79         int res;
  80
  81         if (!buf)
  82                 return -ENOMEM;
  83         res = single_open(file, show_schedstat, NULL);
  84         if (!res) {
  85                 m = file->private_data;
  86                 m->buf = buf;
  87                 m->size = size;
  88         } else
  89                 kfree(buf);
  90         return res;
  91 }
  92
  93 const struct file_operations proc_schedstat_operations = {
  94         .open    = schedstat_open,
  95         .read    = seq_read,
  96         .llseek  = seq_lseek,
  97         .release = single_release,
  98 };
  99
 100 /*
 101  * Expects runqueue lock to be held for atomicity of update
 102  */
 103 static inline void
 104 rq_sched_info_arrive(struct rq *rq, unsigned long long delta)
 105 {
 106         if (rq) {
 107                 rq->rq_sched_info.run_delay += delta;
 108                 rq->rq_sched_info.pcount++;
 109         }
 110 }
 111
 112 /*
 113  * Expects runqueue lock to be held for atomicity of update
 114  */
 115 static inline void
 116 rq_sched_info_depart(struct rq *rq, unsigned long long delta)
 117 {
 118         if (rq)
 119                 rq->rq_sched_info.cpu_time += delta;
 120 }
 121 # define schedstat_inc(rq, field)       do { (rq)->field++; } while (0)
 122 # define schedstat_add(rq, field, amt)  do { (rq)->field += (amt); } while (0)
 123 # define schedstat_set(var, val)        do { var = (val); } while (0)
 124 #else /* !CONFIG_SCHEDSTATS */
 125 static inline void
 126 rq_sched_info_arrive(struct rq *rq, unsigned long long delta)
 127 {}
 128 static inline void
 129 rq_sched_info_depart(struct rq *rq, unsigned long long delta)
 130 {}
 131 # define schedstat_inc(rq, field)       do { } while (0)
 132 # define schedstat_add(rq, field, amt)  do { } while (0)
 133 # define schedstat_set(var, val)        do { } while (0)
 134 #endif
 135
 136 #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT)
 137 /*
 138  * Called when a process is dequeued from the active array and given
 139  * the cpu.  We should note that with the exception of interactive
 140  * tasks, the expired queue will become the active queue after the active
 141  * queue is empty, without explicitly dequeuing and requeuing tasks in the
 142  * expired queue.  (Interactive tasks may be requeued directly to the
 143  * active queue, thus delaying tasks in the expired queue from running;
 144  * see scheduler_tick()).
 145  *
 146  * This function is only called from sched_info_arrive(), rather than
 147  * dequeue_task(). Even though a task may be queued and dequeued multiple
 148  * times as it is shuffled about, we're really interested in knowing how
 149  * long it was from the *first* time it was queued to the time that it
 150  * finally hit a cpu.
 151  */
 152 static inline void sched_info_dequeued(struct task_struct *t)
 153 {
 154         t->sched_info.last_queued = 0;
 155 }
 156
 157 /*
 158  * Called when a task finally hits the cpu.  We can now calculate how
 159  * long it was waiting to run.  We also note when it began so that we
 160  * can keep stats on how long its timeslice is.
 161  */
 162 static void sched_info_arrive(struct task_struct *t)
 163 {
 164         unsigned long long now = task_rq(t)->clock, delta = 0;
 165
 166         if (t->sched_info.last_queued)
 167                 delta = now - t->sched_info.last_queued;
 168         sched_info_dequeued(t);
 169         t->sched_info.run_delay += delta;
 170         t->sched_info.last_arrival = now;
 171         t->sched_info.pcount++;
 172
 173         rq_sched_info_arrive(task_rq(t), delta);
 174 }
 175
 176 /*
 177  * Called when a process is queued into either the active or expired
 178  * array.  The time is noted and later used to determine how long we
 179  * had to wait for us to reach the cpu.  Since the expired queue will
 180  * become the active queue after active queue is empty, without dequeuing
 181  * and requeuing any tasks, we are interested in queuing to either. It
 182  * is unusual but not impossible for tasks to be dequeued and immediately
 183  * requeued in the same or another array: this can happen in sched_yield(),
 184  * set_user_nice(), and even load_balance() as it moves tasks from runqueue
 185  * to runqueue.
 186  *
 187  * This function is only called from enqueue_task(), but also only updates
 188  * the timestamp if it is already not set.  It's assumed that
 189  * sched_info_dequeued() will clear that stamp when appropriate.
 190  */
 191 static inline void sched_info_queued(struct task_struct *t)
 192 {
 193         if (unlikely(sched_info_on()))
 194                 if (!t->sched_info.last_queued)
 195                         t->sched_info.last_queued = task_rq(t)->clock;
 196 }
 197
 198 /*
 199  * Called when a process ceases being the active-running process, either
 200  * voluntarily or involuntarily.  Now we can calculate how long we ran.
 201  * Also, if the process is still in the TASK_RUNNING state, call
 202  * sched_info_queued() to mark that it has now again started waiting on
 203  * the runqueue.
 204  */
 205 static inline void sched_info_depart(struct task_struct *t)
 206 {
 207         unsigned long long delta = task_rq(t)->clock -
 208                                         t->sched_info.last_arrival;
 209
 210         t->sched_info.cpu_time += delta;
 211         rq_sched_info_depart(task_rq(t), delta);
 212
 213         if (t->state == TASK_RUNNING)
 214                 sched_info_queued(t);
 215 }
 216
 217 /*
 218  * Called when tasks are switched involuntarily due, typically, to expiring
 219  * their time slice.  (This may also be called when switching to or from
 220  * the idle task.)  We are only called when prev != next.
 221  */
 222 static inline void
 223 __sched_info_switch(struct task_struct *prev, struct task_struct *next)
 224 {
 225         struct rq *rq = task_rq(prev);
 226
 227         /*
 228          * prev now departs the cpu.  It's not interesting to record
 229          * stats about how efficient we were at scheduling the idle
 230          * process, however.
 231          */
 232         if (prev != rq->idle)
 233                 sched_info_depart(prev);
 234
 235         if (next != rq->idle)
 236                 sched_info_arrive(next);
 237 }
 238 static inline void
 239 sched_info_switch(struct task_struct *prev, struct task_struct *next)
 240 {
 241         if (unlikely(sched_info_on()))
 242                 __sched_info_switch(prev, next);
 243 }
 244 #else
 245 #define sched_info_queued(t)            do { } while (0)
 246 #define sched_info_switch(t, next)      do { } while (0)
 247 #endif /* CONFIG_SCHEDSTATS || CONFIG_TASK_DELAY_ACCT */
 248