kernel/sched/stats.h

   1
   2 #ifdef CONFIG_SCHEDSTATS
   3
   4 /*
   5  * Expects runqueue lock to be held for atomicity of update
   6  */
   7 static inline void
   8 rq_sched_info_arrive(struct rq *rq, unsigned long long delta)
   9 {
  10         if (rq) {
  11                 rq->rq_sched_info.run_delay += delta;
  12                 rq->rq_sched_info.pcount++;
  13         }
  14 }
  15
  16 /*
  17  * Expects runqueue lock to be held for atomicity of update
  18  */
  19 static inline void
  20 rq_sched_info_depart(struct rq *rq, unsigned long long delta)
  21 {
  22         if (rq)
  23                 rq->rq_cpu_time += delta;
  24 }
  25
  26 static inline void
  27 rq_sched_info_dequeued(struct rq *rq, unsigned long long delta)
  28 {
  29         if (rq)
  30                 rq->rq_sched_info.run_delay += delta;
  31 }
  32 # define schedstat_inc(rq, field)       do { (rq)->field++; } while (0)
  33 # define schedstat_add(rq, field, amt)  do { (rq)->field += (amt); } while (0)
  34 # define schedstat_set(var, val)        do { var = (val); } while (0)
  35 #else /* !CONFIG_SCHEDSTATS */
  36 static inline void
  37 rq_sched_info_arrive(struct rq *rq, unsigned long long delta)
  38 {}
  39 static inline void
  40 rq_sched_info_dequeued(struct rq *rq, unsigned long long delta)
  41 {}
  42 static inline void
  43 rq_sched_info_depart(struct rq *rq, unsigned long long delta)
  44 {}
  45 # define schedstat_inc(rq, field)       do { } while (0)
  46 # define schedstat_add(rq, field, amt)  do { } while (0)
  47 # define schedstat_set(var, val)        do { } while (0)
  48 #endif
  49
  50 #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT)
  51 static inline void sched_info_reset_dequeued(struct task_struct *t)
  52 {
  53         t->sched_info.last_queued = 0;
  54 }
  55
  56 /*
  57  * We are interested in knowing how long it was from the *first* time a
  58  * task was queued to the time that it finally hit a cpu, we call this routine
  59  * from dequeue_task() to account for possible rq->clock skew across cpus. The
  60  * delta taken on each cpu would annul the skew.
  61  */
  62 static inline void sched_info_dequeued(struct task_struct *t)
  63 {
  64         unsigned long long now = task_rq(t)->clock, delta = 0;
  65
  66         if (unlikely(sched_info_on()))
  67                 if (t->sched_info.last_queued)
  68                         delta = now - t->sched_info.last_queued;
  69         sched_info_reset_dequeued(t);
  70         t->sched_info.run_delay += delta;
  71
  72         rq_sched_info_dequeued(task_rq(t), delta);
  73 }
  74
  75 /*
  76  * Called when a task finally hits the cpu.  We can now calculate how
  77  * long it was waiting to run.  We also note when it began so that we
  78  * can keep stats on how long its timeslice is.
  79  */
  80 static void sched_info_arrive(struct task_struct *t)
  81 {
  82         unsigned long long now = task_rq(t)->clock, delta = 0;
  83
  84         if (t->sched_info.last_queued)
  85                 delta = now - t->sched_info.last_queued;
  86         sched_info_reset_dequeued(t);
  87         t->sched_info.run_delay += delta;
  88         t->sched_info.last_arrival = now;
  89         t->sched_info.pcount++;
  90
  91         rq_sched_info_arrive(task_rq(t), delta);
  92 }
  93
  94 /*
  95  * This function is only called from enqueue_task(), but also only updates
  96  * the timestamp if it is already not set.  It's assumed that
  97  * sched_info_dequeued() will clear that stamp when appropriate.
  98  */
  99 static inline void sched_info_queued(struct task_struct *t)
 100 {
 101         if (unlikely(sched_info_on()))
 102                 if (!t->sched_info.last_queued)
 103                         t->sched_info.last_queued = task_rq(t)->clock;
 104 }
 105
 106 /*
 107  * Called when a process ceases being the active-running process, either
 108  * voluntarily or involuntarily.  Now we can calculate how long we ran.
 109  * Also, if the process is still in the TASK_RUNNING state, call
 110  * sched_info_queued() to mark that it has now again started waiting on
 111  * the runqueue.
 112  */
 113 static inline void sched_info_depart(struct task_struct *t)
 114 {
 115         unsigned long long delta = task_rq(t)->clock -
 116                                         t->sched_info.last_arrival;
 117
 118         rq_sched_info_depart(task_rq(t), delta);
 119
 120         if (t->state == TASK_RUNNING)
 121                 sched_info_queued(t);
 122 }
 123
 124 /*
 125  * Called when tasks are switched involuntarily due, typically, to expiring
 126  * their time slice.  (This may also be called when switching to or from
 127  * the idle task.)  We are only called when prev != next.
 128  */
 129 static inline void
 130 __sched_info_switch(struct task_struct *prev, struct task_struct *next)
 131 {
 132         struct rq *rq = task_rq(prev);
 133
 134         /*
 135          * prev now departs the cpu.  It's not interesting to record
 136          * stats about how efficient we were at scheduling the idle
 137          * process, however.
 138          */
 139         if (prev != rq->idle)
 140                 sched_info_depart(prev);
 141
 142         if (next != rq->idle)
 143                 sched_info_arrive(next);
 144 }
 145 static inline void
 146 sched_info_switch(struct task_struct *prev, struct task_struct *next)
 147 {
 148         if (unlikely(sched_info_on()))
 149                 __sched_info_switch(prev, next);
 150 }
 151 #else
 152 #define sched_info_queued(t)                    do { } while (0)
 153 #define sched_info_reset_dequeued(t)    do { } while (0)
 154 #define sched_info_dequeued(t)                  do { } while (0)
 155 #define sched_info_switch(t, next)              do { } while (0)
 156 #endif /* CONFIG_SCHEDSTATS || CONFIG_TASK_DELAY_ACCT */
 157
 158 /*
 159  * The following are functions that support scheduler-internal time accounting.
 160  * These functions are generally called at the timer tick.  None of this depends
 161  * on CONFIG_SCHEDSTATS.
 162  */
 163
 164 /**
 165  * cputimer_running - return true if cputimer is running
 166  *
 167  * @tsk:        Pointer to target task.
 168  */
 169 static inline bool cputimer_running(struct task_struct *tsk)
 170
 171 {
 172         struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
 173
 174         if (!cputimer->running)
 175                 return false;
 176
 177         /*
 178          * After we flush the task's sum_exec_runtime to sig->sum_sched_runtime
 179          * in __exit_signal(), we won't account to the signal struct further
 180          * cputime consumed by that task, even though the task can still be
 181          * ticking after __exit_signal().
 182          *
 183          * In order to keep a consistent behaviour between thread group cputime
 184          * and thread group cputimer accounting, lets also ignore the cputime
 185          * elapsing after __exit_signal() in any thread group timer running.
 186          *
 187          * This makes sure that POSIX CPU clocks and timers are synchronized, so
 188          * that a POSIX CPU timer won't expire while the corresponding POSIX CPU
 189          * clock delta is behind the expiring timer value.
 190          */
 191         if (unlikely(!tsk->sighand))
 192                 return false;
 193
 194         return true;
 195 }
 196
 197 /**
 198  * account_group_user_time - Maintain utime for a thread group.
 199  *
 200  * @tsk:        Pointer to task structure.
 201  * @cputime:    Time value by which to increment the utime field of the
 202  *              thread_group_cputime structure.
 203  *
 204  * If thread group time is being maintained, get the structure for the
 205  * running CPU and update the utime field there.
 206  */
 207 static inline void account_group_user_time(struct task_struct *tsk,
 208                                            cputime_t cputime)
 209 {
 210         struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
 211
 212         if (!cputimer_running(tsk))
 213                 return;
 214
 215         raw_spin_lock(&cputimer->lock);
 216         cputimer->cputime.utime += cputime;
 217         raw_spin_unlock(&cputimer->lock);
 218 }
 219
 220 /**
 221  * account_group_system_time - Maintain stime for a thread group.
 222  *
 223  * @tsk:        Pointer to task structure.
 224  * @cputime:    Time value by which to increment the stime field of the
 225  *              thread_group_cputime structure.
 226  *
 227  * If thread group time is being maintained, get the structure for the
 228  * running CPU and update the stime field there.
 229  */
 230 static inline void account_group_system_time(struct task_struct *tsk,
 231                                              cputime_t cputime)
 232 {
 233         struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
 234
 235         if (!cputimer_running(tsk))
 236                 return;
 237
 238         raw_spin_lock(&cputimer->lock);
 239         cputimer->cputime.stime += cputime;
 240         raw_spin_unlock(&cputimer->lock);
 241 }
 242
 243 /**
 244  * account_group_exec_runtime - Maintain exec runtime for a thread group.
 245  *
 246  * @tsk:        Pointer to task structure.
 247  * @ns:         Time value by which to increment the sum_exec_runtime field
 248  *              of the thread_group_cputime structure.
 249  *
 250  * If thread group time is being maintained, get the structure for the
 251  * running CPU and update the sum_exec_runtime field there.
 252  */
 253 static inline void account_group_exec_runtime(struct task_struct *tsk,
 254                                               unsigned long long ns)
 255 {
 256         struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
 257
 258         if (!cputimer_running(tsk))
 259                 return;
 260
 261         raw_spin_lock(&cputimer->lock);
 262         cputimer->cputime.sum_exec_runtime += ns;
 263         raw_spin_unlock(&cputimer->lock);
 264 }