mm: compaction: add /proc trigger for memory compaction
[linux-2.6.git] / kernel / sched_debug.c
index 62965f0..87a330a 100644 (file)
                printk(x);                      \
  } while (0)
 
+/*
+ * Ease the printing of nsec fields:
+ */
+static long long nsec_high(unsigned long long nsec)
+{
+       if ((long long)nsec < 0) {
+               nsec = -nsec;
+               do_div(nsec, 1000000);
+               return -nsec;
+       }
+       do_div(nsec, 1000000);
+
+       return nsec;
+}
+
+static unsigned long nsec_low(unsigned long long nsec)
+{
+       if ((long long)nsec < 0)
+               nsec = -nsec;
+
+       return do_div(nsec, 1000000);
+}
+
+#define SPLIT_NS(x) nsec_high(x), nsec_low(x)
+
+#ifdef CONFIG_FAIR_GROUP_SCHED
+static void print_cfs_group_stats(struct seq_file *m, int cpu,
+               struct task_group *tg)
+{
+       struct sched_entity *se = tg->se[cpu];
+       if (!se)
+               return;
+
+#define P(F) \
+       SEQ_printf(m, "  .%-30s: %lld\n", #F, (long long)F)
+#define PN(F) \
+       SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
+
+       PN(se->exec_start);
+       PN(se->vruntime);
+       PN(se->sum_exec_runtime);
+#ifdef CONFIG_SCHEDSTATS
+       PN(se->statistics.wait_start);
+       PN(se->statistics.sleep_start);
+       PN(se->statistics.block_start);
+       PN(se->statistics.sleep_max);
+       PN(se->statistics.block_max);
+       PN(se->statistics.exec_max);
+       PN(se->statistics.slice_max);
+       PN(se->statistics.wait_max);
+       PN(se->statistics.wait_sum);
+       P(se->statistics.wait_count);
+#endif
+       P(se->load.weight);
+#undef PN
+#undef P
+}
+#endif
+
 static void
 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
 {
@@ -36,43 +95,47 @@ print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
        else
                SEQ_printf(m, " ");
 
-       SEQ_printf(m, "%15s %5d %15Ld %13Ld %13Ld %9Ld %5d ",
+       SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
                p->comm, p->pid,
-               (long long)p->se.fair_key,
-               (long long)(p->se.fair_key - rq->cfs.fair_clock),
-               (long long)p->se.wait_runtime,
+               SPLIT_NS(p->se.vruntime),
                (long long)(p->nvcsw + p->nivcsw),
                p->prio);
 #ifdef CONFIG_SCHEDSTATS
-       SEQ_printf(m, "%15Ld %15Ld %15Ld %15Ld %15Ld %15Ld\n",
-               (long long)p->se.vruntime,
-               (long long)p->se.sum_exec_runtime,
-               (long long)p->se.sum_wait_runtime,
-               (long long)p->se.sum_sleep_runtime,
-               (long long)p->se.wait_runtime_overruns,
-               (long long)p->se.wait_runtime_underruns);
+       SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
+               SPLIT_NS(p->se.vruntime),
+               SPLIT_NS(p->se.sum_exec_runtime),
+               SPLIT_NS(p->se.statistics.sum_sleep_runtime));
 #else
-       SEQ_printf(m, "%15Ld %15Ld %15Ld %15Ld %15Ld\n",
-               0LL, 0LL, 0LL, 0LL, 0LL);
+       SEQ_printf(m, "%15Ld %15Ld %15Ld.%06ld %15Ld.%06ld %15Ld.%06ld",
+               0LL, 0LL, 0LL, 0L, 0LL, 0L, 0LL, 0L);
+#endif
+
+#ifdef CONFIG_CGROUP_SCHED
+       {
+               char path[64];
+
+               rcu_read_lock();
+               cgroup_path(task_group(p)->css.cgroup, path, sizeof(path));
+               rcu_read_unlock();
+               SEQ_printf(m, " %s", path);
+       }
 #endif
+       SEQ_printf(m, "\n");
 }
 
 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
 {
        struct task_struct *g, *p;
+       unsigned long flags;
 
        SEQ_printf(m,
        "\nrunnable tasks:\n"
-       "            task   PID        tree-key         delta       waiting"
-       "  switches  prio"
-       "    exec-runtime        sum-exec        sum-wait       sum-sleep"
-       "    wait-overrun   wait-underrun\n"
-       "------------------------------------------------------------------"
-       "--------------------------------"
-       "------------------------------------------------"
-       "--------------------------------\n");
+       "            task   PID         tree-key  switches  prio"
+       "     exec-runtime         sum-exec        sum-sleep\n"
+       "------------------------------------------------------"
+       "----------------------------------------------------\n");
 
-       read_lock_irq(&tasklist_lock);
+       read_lock_irqsave(&tasklist_lock, flags);
 
        do_each_thread(g, p) {
                if (!p->se.on_rq || task_cpu(p) != rq_cpu)
@@ -81,82 +144,108 @@ static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
                print_task(m, rq, p);
        } while_each_thread(g, p);
 
-       read_unlock_irq(&tasklist_lock);
+       read_unlock_irqrestore(&tasklist_lock, flags);
 }
 
-static void
-print_cfs_rq_runtime_sum(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
+#if defined(CONFIG_CGROUP_SCHED) && \
+       (defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED))
+static void task_group_path(struct task_group *tg, char *buf, int buflen)
 {
-       s64 wait_runtime_rq_sum = 0;
-       struct task_struct *p;
-       struct rb_node *curr;
-       unsigned long flags;
-       struct rq *rq = &per_cpu(runqueues, cpu);
-
-       spin_lock_irqsave(&rq->lock, flags);
-       curr = first_fair(cfs_rq);
-       while (curr) {
-               p = rb_entry(curr, struct task_struct, se.run_node);
-               wait_runtime_rq_sum += p->se.wait_runtime;
-
-               curr = rb_next(curr);
+       /* may be NULL if the underlying cgroup isn't fully-created yet */
+       if (!tg->css.cgroup) {
+               buf[0] = '\0';
+               return;
        }
-       spin_unlock_irqrestore(&rq->lock, flags);
-
-       SEQ_printf(m, "  .%-30s: %Ld\n", "wait_runtime_rq_sum",
-               (long long)wait_runtime_rq_sum);
+       cgroup_path(tg->css.cgroup, buf, buflen);
 }
+#endif
 
 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
 {
        s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
                spread, rq0_min_vruntime, spread0;
-       struct rq *rq = &per_cpu(runqueues, cpu);
+       struct rq *rq = cpu_rq(cpu);
        struct sched_entity *last;
        unsigned long flags;
 
-       SEQ_printf(m, "\ncfs_rq\n");
+#if defined(CONFIG_CGROUP_SCHED) && defined(CONFIG_FAIR_GROUP_SCHED)
+       char path[128];
+       struct task_group *tg = cfs_rq->tg;
 
-#define P(x) \
-       SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(cfs_rq->x))
+       task_group_path(tg, path, sizeof(path));
 
-       P(fair_clock);
-       P(exec_clock);
+       SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, path);
+#else
+       SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
+#endif
+       SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
+                       SPLIT_NS(cfs_rq->exec_clock));
 
-       spin_lock_irqsave(&rq->lock, flags);
+       raw_spin_lock_irqsave(&rq->lock, flags);
        if (cfs_rq->rb_leftmost)
                MIN_vruntime = (__pick_next_entity(cfs_rq))->vruntime;
        last = __pick_last_entity(cfs_rq);
        if (last)
                max_vruntime = last->vruntime;
-       min_vruntime = rq->cfs.min_vruntime;
-       rq0_min_vruntime = per_cpu(runqueues, 0).cfs.min_vruntime;
-       spin_unlock_irqrestore(&rq->lock, flags);
-       SEQ_printf(m, "  .%-30s: %Ld\n", "MIN_vruntime",
-                       (long long)MIN_vruntime);
-       SEQ_printf(m, "  .%-30s: %Ld\n", "min_vruntime",
-                       (long long)min_vruntime);
-       SEQ_printf(m, "  .%-30s: %Ld\n", "max_vruntime",
-                       (long long)max_vruntime);
+       min_vruntime = cfs_rq->min_vruntime;
+       rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
+       raw_spin_unlock_irqrestore(&rq->lock, flags);
+       SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
+                       SPLIT_NS(MIN_vruntime));
+       SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
+                       SPLIT_NS(min_vruntime));
+       SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
+                       SPLIT_NS(max_vruntime));
        spread = max_vruntime - MIN_vruntime;
-       SEQ_printf(m, "  .%-30s: %Ld\n", "spread",
-                       (long long)spread);
+       SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
+                       SPLIT_NS(spread));
        spread0 = min_vruntime - rq0_min_vruntime;
-       SEQ_printf(m, "  .%-30s: %Ld\n", "spread0",
-                       (long long)spread0);
+       SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
+                       SPLIT_NS(spread0));
+       SEQ_printf(m, "  .%-30s: %ld\n", "nr_running", cfs_rq->nr_running);
+       SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
+
+       SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
+                       cfs_rq->nr_spread_over);
+#ifdef CONFIG_FAIR_GROUP_SCHED
+#ifdef CONFIG_SMP
+       SEQ_printf(m, "  .%-30s: %lu\n", "shares", cfs_rq->shares);
+#endif
+       print_cfs_group_stats(m, cpu, cfs_rq->tg);
+#endif
+}
+
+void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
+{
+#if defined(CONFIG_CGROUP_SCHED) && defined(CONFIG_RT_GROUP_SCHED)
+       char path[128];
+       struct task_group *tg = rt_rq->tg;
+
+       task_group_path(tg, path, sizeof(path));
+
+       SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, path);
+#else
+       SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
+#endif
 
-       P(wait_runtime);
-       P(wait_runtime_overruns);
-       P(wait_runtime_underruns);
-       P(sleeper_bonus);
-#undef P
 
-       print_cfs_rq_runtime_sum(m, cpu, cfs_rq);
+#define P(x) \
+       SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
+#define PN(x) \
+       SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
+
+       P(rt_nr_running);
+       P(rt_throttled);
+       PN(rt_time);
+       PN(rt_runtime);
+
+#undef PN
+#undef P
 }
 
 static void print_cpu(struct seq_file *m, int cpu)
 {
-       struct rq *rq = &per_cpu(runqueues, cpu);
+       struct rq *rq = cpu_rq(cpu);
 
 #ifdef CONFIG_X86
        {
@@ -171,6 +260,8 @@ static void print_cpu(struct seq_file *m, int cpu)
 
 #define P(x) \
        SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x))
+#define PN(x) \
+       SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
 
        P(nr_running);
        SEQ_printf(m, "  .%-30s: %lu\n", "load",
@@ -178,39 +269,77 @@ static void print_cpu(struct seq_file *m, int cpu)
        P(nr_switches);
        P(nr_load_updates);
        P(nr_uninterruptible);
-       SEQ_printf(m, "  .%-30s: %lu\n", "jiffies", jiffies);
-       P(next_balance);
+       PN(next_balance);
        P(curr->pid);
-       P(clock);
-       P(idle_clock);
-       P(prev_clock_raw);
-       P(clock_warps);
-       P(clock_overflows);
-       P(clock_deep_idle_events);
-       P(clock_max_delta);
+       PN(clock);
        P(cpu_load[0]);
        P(cpu_load[1]);
        P(cpu_load[2]);
        P(cpu_load[3]);
        P(cpu_load[4]);
 #undef P
+#undef PN
+
+#ifdef CONFIG_SCHEDSTATS
+#define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, rq->n);
+#define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
+
+       P(yld_count);
+
+       P(sched_switch);
+       P(sched_count);
+       P(sched_goidle);
+#ifdef CONFIG_SMP
+       P64(avg_idle);
+#endif
 
+       P(ttwu_count);
+       P(ttwu_local);
+
+       P(bkl_count);
+
+#undef P
+#endif
        print_cfs_stats(m, cpu);
+       print_rt_stats(m, cpu);
 
        print_rq(m, rq, cpu);
 }
 
+static const char *sched_tunable_scaling_names[] = {
+       "none",
+       "logaritmic",
+       "linear"
+};
+
 static int sched_debug_show(struct seq_file *m, void *v)
 {
        u64 now = ktime_to_ns(ktime_get());
        int cpu;
 
-       SEQ_printf(m, "Sched Debug Version: v0.05-v20, %s %.*s\n",
+       SEQ_printf(m, "Sched Debug Version: v0.09, %s %.*s\n",
                init_utsname()->release,
                (int)strcspn(init_utsname()->version, " "),
                init_utsname()->version);
 
-       SEQ_printf(m, "now at %Lu nsecs\n", (unsigned long long)now);
+       SEQ_printf(m, "now at %Lu.%06ld msecs\n", SPLIT_NS(now));
+
+#define P(x) \
+       SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
+#define PN(x) \
+       SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
+       P(jiffies);
+       PN(sysctl_sched_latency);
+       PN(sysctl_sched_min_granularity);
+       PN(sysctl_sched_wakeup_granularity);
+       PN(sysctl_sched_child_runs_first);
+       P(sysctl_sched_features);
+#undef PN
+#undef P
+
+       SEQ_printf(m, "  .%-40s: %d (%s)\n", "sysctl_sched_tunable_scaling",
+               sysctl_sched_tunable_scaling,
+               sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
 
        for_each_online_cpu(cpu)
                print_cpu(m, cpu);
@@ -230,7 +359,7 @@ static int sched_debug_open(struct inode *inode, struct file *filp)
        return single_open(filp, sched_debug_show, NULL);
 }
 
-static struct file_operations sched_debug_fops = {
+static const struct file_operations sched_debug_fops = {
        .open           = sched_debug_open,
        .read           = seq_read,
        .llseek         = seq_lseek,
@@ -241,12 +370,9 @@ static int __init init_sched_debug_procfs(void)
 {
        struct proc_dir_entry *pe;
 
-       pe = create_proc_entry("sched_debug", 0644, NULL);
+       pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
        if (!pe)
                return -ENOMEM;
-
-       pe->proc_fops = &sched_debug_fops;
-
        return 0;
 }
 
@@ -254,54 +380,105 @@ __initcall(init_sched_debug_procfs);
 
 void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
 {
+       unsigned long nr_switches;
        unsigned long flags;
        int num_threads = 1;
 
-       rcu_read_lock();
        if (lock_task_sighand(p, &flags)) {
                num_threads = atomic_read(&p->signal->count);
                unlock_task_sighand(p, &flags);
        }
-       rcu_read_unlock();
 
        SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, p->pid, num_threads);
-       SEQ_printf(m, "----------------------------------------------\n");
+       SEQ_printf(m,
+               "---------------------------------------------------------\n");
+#define __P(F) \
+       SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)F)
 #define P(F) \
-       SEQ_printf(m, "%-25s:%20Ld\n", #F, (long long)p->F)
+       SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)p->F)
+#define __PN(F) \
+       SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
+#define PN(F) \
+       SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
+
+       PN(se.exec_start);
+       PN(se.vruntime);
+       PN(se.sum_exec_runtime);
 
-       P(se.wait_runtime);
-       P(se.wait_start_fair);
-       P(se.exec_start);
-       P(se.sleep_start_fair);
-       P(se.vruntime);
-       P(se.sum_exec_runtime);
+       nr_switches = p->nvcsw + p->nivcsw;
 
 #ifdef CONFIG_SCHEDSTATS
-       P(se.wait_start);
-       P(se.sleep_start);
-       P(se.block_start);
-       P(se.sleep_max);
-       P(se.block_max);
-       P(se.exec_max);
-       P(se.slice_max);
-       P(se.wait_max);
-       P(se.wait_runtime_overruns);
-       P(se.wait_runtime_underruns);
-       P(se.sum_wait_runtime);
+       PN(se.statistics.wait_start);
+       PN(se.statistics.sleep_start);
+       PN(se.statistics.block_start);
+       PN(se.statistics.sleep_max);
+       PN(se.statistics.block_max);
+       PN(se.statistics.exec_max);
+       PN(se.statistics.slice_max);
+       PN(se.statistics.wait_max);
+       PN(se.statistics.wait_sum);
+       P(se.statistics.wait_count);
+       PN(se.statistics.iowait_sum);
+       P(se.statistics.iowait_count);
+       P(sched_info.bkl_count);
+       P(se.nr_migrations);
+       P(se.statistics.nr_migrations_cold);
+       P(se.statistics.nr_failed_migrations_affine);
+       P(se.statistics.nr_failed_migrations_running);
+       P(se.statistics.nr_failed_migrations_hot);
+       P(se.statistics.nr_forced_migrations);
+       P(se.statistics.nr_wakeups);
+       P(se.statistics.nr_wakeups_sync);
+       P(se.statistics.nr_wakeups_migrate);
+       P(se.statistics.nr_wakeups_local);
+       P(se.statistics.nr_wakeups_remote);
+       P(se.statistics.nr_wakeups_affine);
+       P(se.statistics.nr_wakeups_affine_attempts);
+       P(se.statistics.nr_wakeups_passive);
+       P(se.statistics.nr_wakeups_idle);
+
+       {
+               u64 avg_atom, avg_per_cpu;
+
+               avg_atom = p->se.sum_exec_runtime;
+               if (nr_switches)
+                       do_div(avg_atom, nr_switches);
+               else
+                       avg_atom = -1LL;
+
+               avg_per_cpu = p->se.sum_exec_runtime;
+               if (p->se.nr_migrations) {
+                       avg_per_cpu = div64_u64(avg_per_cpu,
+                                               p->se.nr_migrations);
+               } else {
+                       avg_per_cpu = -1LL;
+               }
+
+               __PN(avg_atom);
+               __PN(avg_per_cpu);
+       }
 #endif
-       SEQ_printf(m, "%-25s:%20Ld\n",
-                  "nr_switches", (long long)(p->nvcsw + p->nivcsw));
+       __P(nr_switches);
+       SEQ_printf(m, "%-35s:%21Ld\n",
+                  "nr_voluntary_switches", (long long)p->nvcsw);
+       SEQ_printf(m, "%-35s:%21Ld\n",
+                  "nr_involuntary_switches", (long long)p->nivcsw);
+
        P(se.load.weight);
        P(policy);
        P(prio);
+#undef PN
+#undef __PN
 #undef P
+#undef __P
 
        {
+               unsigned int this_cpu = raw_smp_processor_id();
                u64 t0, t1;
 
-               t0 = sched_clock();
-               t1 = sched_clock();
-               SEQ_printf(m, "%-25s:%20Ld\n",
+               t0 = cpu_clock(this_cpu);
+               t1 = cpu_clock(this_cpu);
+               SEQ_printf(m, "%-35s:%21Ld\n",
                           "clock-delta", (long long)(t1-t0));
        }
 }
@@ -309,14 +486,6 @@ void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
 void proc_sched_set_task(struct task_struct *p)
 {
 #ifdef CONFIG_SCHEDSTATS
-       p->se.sleep_max                 = 0;
-       p->se.block_max                 = 0;
-       p->se.exec_max                  = 0;
-       p->se.slice_max                 = 0;
-       p->se.wait_max                  = 0;
-       p->se.wait_runtime_overruns     = 0;
-       p->se.wait_runtime_underruns    = 0;
+       memset(&p->se.statistics, 0, sizeof(p->se.statistics));
 #endif
-       p->se.sum_exec_runtime          = 0;
-       p->se.prev_sum_exec_runtime     = 0;
 }