| 1 |
62 |
marcus.erl |
|
| 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,
|
| 56 |
|
|
" %u %u %u %u %u %u %u %u %u %u %u %u\n",
|
| 57 |
|
|
sd->alb_count, sd->alb_failed, sd->alb_pushed,
|
| 58 |
|
|
sd->sbe_count, sd->sbe_balanced, sd->sbe_pushed,
|
| 59 |
|
|
sd->sbf_count, sd->sbf_balanced, sd->sbf_pushed,
|
| 60 |
|
|
sd->ttwu_wake_remote, sd->ttwu_move_affine,
|
| 61 |
|
|
sd->ttwu_move_balance);
|
| 62 |
|
|
}
|
| 63 |
|
|
preempt_enable();
|
| 64 |
|
|
#endif
|
| 65 |
|
|
}
|
| 66 |
|
|
return 0;
|
| 67 |
|
|
}
|
| 68 |
|
|
|
| 69 |
|
|
static int schedstat_open(struct inode *inode, struct file *file)
|
| 70 |
|
|
{
|
| 71 |
|
|
unsigned int size = PAGE_SIZE * (1 + num_online_cpus() / 32);
|
| 72 |
|
|
char *buf = kmalloc(size, GFP_KERNEL);
|
| 73 |
|
|
struct seq_file *m;
|
| 74 |
|
|
int res;
|
| 75 |
|
|
|
| 76 |
|
|
if (!buf)
|
| 77 |
|
|
return -ENOMEM;
|
| 78 |
|
|
res = single_open(file, show_schedstat, NULL);
|
| 79 |
|
|
if (!res) {
|
| 80 |
|
|
m = file->private_data;
|
| 81 |
|
|
m->buf = buf;
|
| 82 |
|
|
m->size = size;
|
| 83 |
|
|
} else
|
| 84 |
|
|
kfree(buf);
|
| 85 |
|
|
return res;
|
| 86 |
|
|
}
|
| 87 |
|
|
|
| 88 |
|
|
const struct file_operations proc_schedstat_operations = {
|
| 89 |
|
|
.open = schedstat_open,
|
| 90 |
|
|
.read = seq_read,
|
| 91 |
|
|
.llseek = seq_lseek,
|
| 92 |
|
|
.release = single_release,
|
| 93 |
|
|
};
|
| 94 |
|
|
|
| 95 |
|
|
/*
|
| 96 |
|
|
* Expects runqueue lock to be held for atomicity of update
|
| 97 |
|
|
*/
|
| 98 |
|
|
static inline void
|
| 99 |
|
|
rq_sched_info_arrive(struct rq *rq, unsigned long long delta)
|
| 100 |
|
|
{
|
| 101 |
|
|
if (rq) {
|
| 102 |
|
|
rq->rq_sched_info.run_delay += delta;
|
| 103 |
|
|
rq->rq_sched_info.pcount++;
|
| 104 |
|
|
}
|
| 105 |
|
|
}
|
| 106 |
|
|
|
| 107 |
|
|
/*
|
| 108 |
|
|
* Expects runqueue lock to be held for atomicity of update
|
| 109 |
|
|
*/
|
| 110 |
|
|
static inline void
|
| 111 |
|
|
rq_sched_info_depart(struct rq *rq, unsigned long long delta)
|
| 112 |
|
|
{
|
| 113 |
|
|
if (rq)
|
| 114 |
|
|
rq->rq_sched_info.cpu_time += delta;
|
| 115 |
|
|
}
|
| 116 |
|
|
# define schedstat_inc(rq, field) do { (rq)->field++; } while (0)
|
| 117 |
|
|
# define schedstat_add(rq, field, amt) do { (rq)->field += (amt); } while (0)
|
| 118 |
|
|
# define schedstat_set(var, val) do { var = (val); } while (0)
|
| 119 |
|
|
#else /* !CONFIG_SCHEDSTATS */
|
| 120 |
|
|
static inline void
|
| 121 |
|
|
rq_sched_info_arrive(struct rq *rq, unsigned long long delta)
|
| 122 |
|
|
{}
|
| 123 |
|
|
static inline void
|
| 124 |
|
|
rq_sched_info_depart(struct rq *rq, unsigned long long delta)
|
| 125 |
|
|
{}
|
| 126 |
|
|
# define schedstat_inc(rq, field) do { } while (0)
|
| 127 |
|
|
# define schedstat_add(rq, field, amt) do { } while (0)
|
| 128 |
|
|
# define schedstat_set(var, val) do { } while (0)
|
| 129 |
|
|
#endif
|
| 130 |
|
|
|
| 131 |
|
|
#if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT)
|
| 132 |
|
|
/*
|
| 133 |
|
|
* Called when a process is dequeued from the active array and given
|
| 134 |
|
|
* the cpu. We should note that with the exception of interactive
|
| 135 |
|
|
* tasks, the expired queue will become the active queue after the active
|
| 136 |
|
|
* queue is empty, without explicitly dequeuing and requeuing tasks in the
|
| 137 |
|
|
* expired queue. (Interactive tasks may be requeued directly to the
|
| 138 |
|
|
* active queue, thus delaying tasks in the expired queue from running;
|
| 139 |
|
|
* see scheduler_tick()).
|
| 140 |
|
|
*
|
| 141 |
|
|
* This function is only called from sched_info_arrive(), rather than
|
| 142 |
|
|
* dequeue_task(). Even though a task may be queued and dequeued multiple
|
| 143 |
|
|
* times as it is shuffled about, we're really interested in knowing how
|
| 144 |
|
|
* long it was from the *first* time it was queued to the time that it
|
| 145 |
|
|
* finally hit a cpu.
|
| 146 |
|
|
*/
|
| 147 |
|
|
static inline void sched_info_dequeued(struct task_struct *t)
|
| 148 |
|
|
{
|
| 149 |
|
|
t->sched_info.last_queued = 0;
|
| 150 |
|
|
}
|
| 151 |
|
|
|
| 152 |
|
|
/*
|
| 153 |
|
|
* Called when a task finally hits the cpu. We can now calculate how
|
| 154 |
|
|
* long it was waiting to run. We also note when it began so that we
|
| 155 |
|
|
* can keep stats on how long its timeslice is.
|
| 156 |
|
|
*/
|
| 157 |
|
|
static void sched_info_arrive(struct task_struct *t)
|
| 158 |
|
|
{
|
| 159 |
|
|
unsigned long long now = task_rq(t)->clock, delta = 0;
|
| 160 |
|
|
|
| 161 |
|
|
if (t->sched_info.last_queued)
|
| 162 |
|
|
delta = now - t->sched_info.last_queued;
|
| 163 |
|
|
sched_info_dequeued(t);
|
| 164 |
|
|
t->sched_info.run_delay += delta;
|
| 165 |
|
|
t->sched_info.last_arrival = now;
|
| 166 |
|
|
t->sched_info.pcount++;
|
| 167 |
|
|
|
| 168 |
|
|
rq_sched_info_arrive(task_rq(t), delta);
|
| 169 |
|
|
}
|
| 170 |
|
|
|
| 171 |
|
|
/*
|
| 172 |
|
|
* Called when a process is queued into either the active or expired
|
| 173 |
|
|
* array. The time is noted and later used to determine how long we
|
| 174 |
|
|
* had to wait for us to reach the cpu. Since the expired queue will
|
| 175 |
|
|
* become the active queue after active queue is empty, without dequeuing
|
| 176 |
|
|
* and requeuing any tasks, we are interested in queuing to either. It
|
| 177 |
|
|
* is unusual but not impossible for tasks to be dequeued and immediately
|
| 178 |
|
|
* requeued in the same or another array: this can happen in sched_yield(),
|
| 179 |
|
|
* set_user_nice(), and even load_balance() as it moves tasks from runqueue
|
| 180 |
|
|
* to runqueue.
|
| 181 |
|
|
*
|
| 182 |
|
|
* This function is only called from enqueue_task(), but also only updates
|
| 183 |
|
|
* the timestamp if it is already not set. It's assumed that
|
| 184 |
|
|
* sched_info_dequeued() will clear that stamp when appropriate.
|
| 185 |
|
|
*/
|
| 186 |
|
|
static inline void sched_info_queued(struct task_struct *t)
|
| 187 |
|
|
{
|
| 188 |
|
|
if (unlikely(sched_info_on()))
|
| 189 |
|
|
if (!t->sched_info.last_queued)
|
| 190 |
|
|
t->sched_info.last_queued = task_rq(t)->clock;
|
| 191 |
|
|
}
|
| 192 |
|
|
|
| 193 |
|
|
/*
|
| 194 |
|
|
* Called when a process ceases being the active-running process, either
|
| 195 |
|
|
* voluntarily or involuntarily. Now we can calculate how long we ran.
|
| 196 |
|
|
*/
|
| 197 |
|
|
static inline void sched_info_depart(struct task_struct *t)
|
| 198 |
|
|
{
|
| 199 |
|
|
unsigned long long delta = task_rq(t)->clock -
|
| 200 |
|
|
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 |
|
|
|