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 = DIV_ROUND_UP(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_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,
  46                                           sched_domain_span(sd));
  47                         seq_printf(seq, "domain%d %s", dcount++, mask_str);
  48                         for (itype = CPU_IDLE; itype < CPU_MAX_IDLE_TYPES;
  49                                         itype++) {
  50                                 seq_printf(seq, " %u %u %u %u %u %u %u %u",
  51                                     sd->lb_count[itype],
  52                                     sd->lb_balanced[itype],
  53                                     sd->lb_failed[itype],
  54                                     sd->lb_imbalance[itype],
  55                                     sd->lb_gained[itype],
  56                                     sd->lb_hot_gained[itype],
  57                                     sd->lb_nobusyq[itype],
  58                                     sd->lb_nobusyg[itype]);
  59                         }
  60                         seq_printf(seq,
  61                                    " %u %u %u %u %u %u %u %u %u %u %u %u\n",
  62                             sd->alb_count, sd->alb_failed, sd->alb_pushed,
  63                             sd->sbe_count, sd->sbe_balanced, sd->sbe_pushed,
  64                             sd->sbf_count, sd->sbf_balanced, sd->sbf_pushed,
  65                             sd->ttwu_wake_remote, sd->ttwu_move_affine,
  66                             sd->ttwu_move_balance);
  67                 }
  68                 preempt_enable();
  69 #endif
  70         }
  71         kfree(mask_str);
  72         return 0;
  73 }
  74
  75 static int schedstat_open(struct inode *inode, struct file *file)
  76 {
  77         unsigned int size = PAGE_SIZE * (1 + num_online_cpus() / 32);
  78         char *buf = kmalloc(size, GFP_KERNEL);
  79         struct seq_file *m;
  80         int res;
  81
  82         if (!buf)
  83                 return -ENOMEM;
  84         res = single_open(file, show_schedstat, NULL);
  85         if (!res) {
  86                 m = file->private_data;
  87                 m->buf = buf;
  88                 m->size = size;
  89         } else
  90                 kfree(buf);
  91         return res;
  92 }
  93
  94 static const struct file_operations proc_schedstat_operations = {
  95         .open    = schedstat_open,
  96         .read    = seq_read,
  97         .llseek  = seq_lseek,
  98         .release = single_release,
  99 };
 100
 101 static int __init proc_schedstat_init(void)
 102 {
 103         proc_create("schedstat", 0, NULL, &proc_schedstat_operations);
 104         return 0;
 105 }
 106 module_init(proc_schedstat_init);
 107
 108 /*
 109  * Expects runqueue lock to be held for atomicity of update
 110  */
 111 static inline void
 112 rq_sched_info_arrive(struct rq *rq, unsigned long long delta)
 113 {
 114         if (rq) {
 115                 rq->rq_sched_info.run_delay += delta;
 116                 rq->rq_sched_info.pcount++;
 117         }
 118 }
 119
 120 /*
 121  * Expects runqueue lock to be held for atomicity of update
 122  */
 123 static inline void
 124 rq_sched_info_depart(struct rq *rq, unsigned long long delta)
 125 {
 126         if (rq)
 127                 rq->rq_cpu_time += delta;
 128 }
 129
 130 static inline void
 131 rq_sched_info_dequeued(struct rq *rq, unsigned long long delta)
 132 {
 133         if (rq)
 134                 rq->rq_sched_info.run_delay += delta;
 135 }
 136 # define schedstat_inc(rq, field)       do { (rq)->field++; } while (0)
 137 # define schedstat_add(rq, field, amt)  do { (rq)->field += (amt); } while (0)
 138 # define schedstat_set(var, val)        do { var = (val); } while (0)
 139 #else /* !CONFIG_SCHEDSTATS */
 140 static inline void
 141 rq_sched_info_arrive(struct rq *rq, unsigned long long delta)
 142 {}
 143 static inline void
 144 rq_sched_info_dequeued(struct rq *rq, unsigned long long delta)
 145 {}
 146 static inline void
 147 rq_sched_info_depart(struct rq *rq, unsigned long long delta)
 148 {}
 149 # define schedstat_inc(rq, field)       do { } while (0)
 150 # define schedstat_add(rq, field, amt)  do { } while (0)
 151 # define schedstat_set(var, val)        do { } while (0)
 152 #endif
 153
 154 #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT)
 155 static inline void sched_info_reset_dequeued(struct task_struct *t)
 156 {
 157         t->sched_info.last_queued = 0;
 158 }
 159
 160 /*
 161  * Called when a process is dequeued from the active array and given
 162  * the cpu.  We should note that with the exception of interactive
 163  * tasks, the expired queue will become the active queue after the active
 164  * queue is empty, without explicitly dequeuing and requeuing tasks in the
 165  * expired queue.  (Interactive tasks may be requeued directly to the
 166  * active queue, thus delaying tasks in the expired queue from running;
 167  * see scheduler_tick()).
 168  *
 169  * Though we are interested in knowing how long it was from the *first* time a
 170  * task was queued to the time that it finally hit a cpu, we call this routine
 171  * from dequeue_task() to account for possible rq->clock skew across cpus. The
 172  * delta taken on each cpu would annul the skew.
 173  */
 174 static inline void sched_info_dequeued(struct task_struct *t)
 175 {
 176         unsigned long long now = task_rq(t)->clock, delta = 0;
 177
 178         if (unlikely(sched_info_on()))
 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
 184         rq_sched_info_dequeued(task_rq(t), delta);
 185 }
 186
 187 /*
 188  * Called when a task finally hits the cpu.  We can now calculate how
 189  * long it was waiting to run.  We also note when it began so that we
 190  * can keep stats on how long its timeslice is.
 191  */
 192 static void sched_info_arrive(struct task_struct *t)
 193 {
 194         unsigned long long now = task_rq(t)->clock, delta = 0;
 195
 196         if (t->sched_info.last_queued)
 197                 delta = now - t->sched_info.last_queued;
 198         sched_info_reset_dequeued(t);
 199         t->sched_info.run_delay += delta;
 200         t->sched_info.last_arrival = now;
 201         t->sched_info.pcount++;
 202
 203         rq_sched_info_arrive(task_rq(t), delta);
 204 }
 205
 206 /*
 207  * Called when a process is queued into either the active or expired
 208  * array.  The time is noted and later used to determine how long we
 209  * had to wait for us to reach the cpu.  Since the expired queue will
 210  * become the active queue after active queue is empty, without dequeuing
 211  * and requeuing any tasks, we are interested in queuing to either. It
 212  * is unusual but not impossible for tasks to be dequeued and immediately
 213  * requeued in the same or another array: this can happen in sched_yield(),
 214  * set_user_nice(), and even load_balance() as it moves tasks from runqueue
 215  * to runqueue.
 216  *
 217  * This function is only called from enqueue_task(), but also only updates
 218  * the timestamp if it is already not set.  It's assumed that
 219  * sched_info_dequeued() will clear that stamp when appropriate.
 220  */
 221 static inline void sched_info_queued(struct task_struct *t)
 222 {
 223         if (unlikely(sched_info_on()))
 224                 if (!t->sched_info.last_queued)
 225                         t->sched_info.last_queued = task_rq(t)->clock;
 226 }
 227
 228 /*
 229  * Called when a process ceases being the active-running process, either
 230  * voluntarily or involuntarily.  Now we can calculate how long we ran.
 231  * Also, if the process is still in the TASK_RUNNING state, call
 232  * sched_info_queued() to mark that it has now again started waiting on
 233  * the runqueue.
 234  */
 235 static inline void sched_info_depart(struct task_struct *t)
 236 {
 237         unsigned long long delta = task_rq(t)->clock -
 238                                         t->sched_info.last_arrival;
 239
 240         rq_sched_info_depart(task_rq(t), delta);
 241
 242         if (t->state == TASK_RUNNING)
 243                 sched_info_queued(t);
 244 }
 245
 246 /*
 247  * Called when tasks are switched involuntarily due, typically, to expiring
 248  * their time slice.  (This may also be called when switching to or from
 249  * the idle task.)  We are only called when prev != next.
 250  */
 251 static inline void
 252 __sched_info_switch(struct task_struct *prev, struct task_struct *next)
 253 {
 254         struct rq *rq = task_rq(prev);
 255
 256         /*
 257          * prev now departs the cpu.  It's not interesting to record
 258          * stats about how efficient we were at scheduling the idle
 259          * process, however.
 260          */
 261         if (prev != rq->idle)
 262                 sched_info_depart(prev);
 263
 264         if (next != rq->idle)
 265                 sched_info_arrive(next);
 266 }
 267 static inline void
 268 sched_info_switch(struct task_struct *prev, struct task_struct *next)
 269 {
 270         if (unlikely(sched_info_on()))
 271                 __sched_info_switch(prev, next);
 272 }
 273 #else
 274 #define sched_info_queued(t)                    do { } while (0)
 275 #define sched_info_reset_dequeued(t)    do { } while (0)
 276 #define sched_info_dequeued(t)                  do { } while (0)
 277 #define sched_info_switch(t, next)              do { } while (0)
 278 #endif /* CONFIG_SCHEDSTATS || CONFIG_TASK_DELAY_ACCT */
 279
 280 /*
 281  * The following are functions that support scheduler-internal time accounting.
 282  * These functions are generally called at the timer tick.  None of this depends
 283  * on CONFIG_SCHEDSTATS.
 284  */
 285
 286 /**
 287  * account_group_user_time - Maintain utime for a thread group.
 288  *
 289  * @tsk:        Pointer to task structure.
 290  * @cputime:    Time value by which to increment the utime field of the
 291  *              thread_group_cputime structure.
 292  *
 293  * If thread group time is being maintained, get the structure for the
 294  * running CPU and update the utime field there.
 295  */
 296 static inline void account_group_user_time(struct task_struct *tsk,
 297                                            cputime_t cputime)
 298 {
 299         struct task_cputime *times;
 300         struct signal_struct *sig;
 301
 302         /* tsk == current, ensure it is safe to use ->signal */
 303         if (unlikely(tsk->exit_state))
 304                 return;
 305
 306         sig = tsk->signal;
 307         times = &sig->cputime.totals;
 308
 309         spin_lock(&times->lock);
 310         times->utime = cputime_add(times->utime, cputime);
 311         spin_unlock(&times->lock);
 312 }
 313
 314 /**
 315  * account_group_system_time - Maintain stime for a thread group.
 316  *
 317  * @tsk:        Pointer to task structure.
 318  * @cputime:    Time value by which to increment the stime field of the
 319  *              thread_group_cputime structure.
 320  *
 321  * If thread group time is being maintained, get the structure for the
 322  * running CPU and update the stime field there.
 323  */
 324 static inline void account_group_system_time(struct task_struct *tsk,
 325                                              cputime_t cputime)
 326 {
 327         struct task_cputime *times;
 328         struct signal_struct *sig;
 329
 330         /* tsk == current, ensure it is safe to use ->signal */
 331         if (unlikely(tsk->exit_state))
 332                 return;
 333
 334         sig = tsk->signal;
 335         times = &sig->cputime.totals;
 336
 337         spin_lock(&times->lock);
 338         times->stime = cputime_add(times->stime, cputime);
 339         spin_unlock(&times->lock);
 340 }
 341
 342 /**
 343  * account_group_exec_runtime - Maintain exec runtime for a thread group.
 344  *
 345  * @tsk:        Pointer to task structure.
 346  * @ns:         Time value by which to increment the sum_exec_runtime field
 347  *              of the thread_group_cputime structure.
 348  *
 349  * If thread group time is being maintained, get the structure for the
 350  * running CPU and update the sum_exec_runtime field there.
 351  */
 352 static inline void account_group_exec_runtime(struct task_struct *tsk,
 353                                               unsigned long long ns)
 354 {
 355         struct task_cputime *times;
 356         struct signal_struct *sig;
 357
 358         sig = tsk->signal;
 359         /* see __exit_signal()->task_rq_unlock_wait() */
 360         barrier();
 361         if (unlikely(!sig))
 362                 return;
 363
 364         times = &sig->cputime.totals;
 365
 366         spin_lock(&times->lock);
 367         times->sum_exec_runtime += ns;
 368         spin_unlock(&times->lock);
 369 }