sched debug: print settings
[safe/jmp/linux-2.6] / kernel / sched_debug.c
index 29f2c21..d79e1ec 100644 (file)
                printk(x);                      \
  } while (0)
 
+/*
+ * Ease the printing of nsec fields:
+ */
+static long long nsec_high(long long nsec)
+{
+       if (nsec < 0) {
+               nsec = -nsec;
+               do_div(nsec, 1000000);
+               return -nsec;
+       }
+       do_div(nsec, 1000000);
+
+       return nsec;
+}
+
+static unsigned long nsec_low(long long nsec)
+{
+       if (nsec < 0)
+               nsec = -nsec;
+
+       return do_div(nsec, 1000000);
+}
+
+#define SPLIT_NS(x) nsec_high(x), nsec_low(x)
+
 static void
-print_task(struct seq_file *m, struct rq *rq, struct task_struct *p, u64 now)
+print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
 {
        if (rq->curr == p)
                SEQ_printf(m, "R");
        else
                SEQ_printf(m, " ");
 
-       SEQ_printf(m, "%15s %5d %15Ld %13Ld %13Ld %9Ld %5d "
-                     "%15Ld %15Ld %15Ld %15Ld %15Ld\n",
+       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,
-               (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);
+               p->prio);
+#ifdef CONFIG_SCHEDSTATS
+       SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld\n",
+               SPLIT_NS(p->se.vruntime),
+               SPLIT_NS(p->se.sum_exec_runtime),
+               SPLIT_NS(p->se.sum_sleep_runtime));
+#else
+       SEQ_printf(m, "%15Ld %15Ld %15Ld.%06ld %15Ld.%06ld %15Ld.%06ld\n",
+               0LL, 0LL, 0LL, 0L, 0LL, 0L, 0LL, 0L);
+#endif
 }
 
-static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu, u64 now)
+static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
 {
        struct task_struct *g, *p;
 
        SEQ_printf(m,
        "\nrunnable tasks:\n"
-       "            task   PID        tree-key         delta       waiting"
-       "  switches  prio"
-       "        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);
 
@@ -72,54 +94,55 @@ static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu, u64 now)
                if (!p->se.on_rq || task_cpu(p) != rq_cpu)
                        continue;
 
-               print_task(m, rq, p, now);
+               print_task(m, rq, p);
        } while_each_thread(g, p);
 
        read_unlock_irq(&tasklist_lock);
 }
 
-static void
-print_cfs_rq_runtime_sum(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
+void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
 {
-       s64 wait_runtime_rq_sum = 0;
-       struct task_struct *p;
-       struct rb_node *curr;
-       unsigned long flags;
+       s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
+               spread, rq0_min_vruntime, spread0;
        struct rq *rq = &per_cpu(runqueues, cpu);
+       struct sched_entity *last;
+       unsigned long flags;
 
-       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);
-       }
-       spin_unlock_irqrestore(&rq->lock, flags);
-
-       SEQ_printf(m, "  .%-30s: %Ld\n", "wait_runtime_rq_sum",
-               (long long)wait_runtime_rq_sum);
-}
-
-void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq, u64 now)
-{
-       SEQ_printf(m, "\ncfs_rq %p\n", cfs_rq);
+       SEQ_printf(m, "\ncfs_rq\n");
 
-#define P(x) \
-       SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(cfs_rq->x))
-
-       P(fair_clock);
-       P(exec_clock);
-       P(wait_runtime);
-       P(wait_runtime_overruns);
-       P(wait_runtime_underruns);
-       P(sleeper_bonus);
-#undef P
+       SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
+                       SPLIT_NS(cfs_rq->exec_clock));
 
-       print_cfs_rq_runtime_sum(m, cpu, cfs_rq);
+       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.%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.%06ld\n", "spread",
+                       SPLIT_NS(spread));
+       spread0 = min_vruntime - rq0_min_vruntime;
+       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);
+#ifdef CONFIG_SCHEDSTATS
+       SEQ_printf(m, "  .%-30s: %ld\n", "bkl_cnt",
+                       rq->bkl_cnt);
+#endif
 }
 
-static void print_cpu(struct seq_file *m, int cpu, u64 now)
+static void print_cpu(struct seq_file *m, int cpu)
 {
        struct rq *rq = &per_cpu(runqueues, cpu);
 
@@ -136,34 +159,36 @@ static void print_cpu(struct seq_file *m, int cpu, u64 now)
 
 #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",
-                  rq->ls.load.weight);
-       P(ls.delta_fair);
-       P(ls.delta_exec);
+                  rq->load.weight);
        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(prev_clock_raw);
+       PN(clock);
+       PN(idle_clock);
+       PN(prev_clock_raw);
        P(clock_warps);
        P(clock_overflows);
-       P(clock_unstable_events);
-       P(clock_max_delta);
+       P(clock_deep_idle_events);
+       PN(clock_max_delta);
        P(cpu_load[0]);
        P(cpu_load[1]);
        P(cpu_load[2]);
        P(cpu_load[3]);
        P(cpu_load[4]);
 #undef P
+#undef PN
 
-       print_cfs_stats(m, cpu, now);
+       print_cfs_stats(m, cpu);
 
-       print_rq(m, rq, cpu, now);
+       print_rq(m, rq, cpu);
 }
 
 static int sched_debug_show(struct seq_file *m, void *v)
@@ -171,26 +196,82 @@ 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, %s %.*s\n",
+       SEQ_printf(m, "Sched Debug Version: v0.05-v20, %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, "  .%-30s: %Ld\n", #x, (long long)(x))
+#define PN(x) \
+       SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(x))
+       PN(sysctl_sched_latency);
+       PN(sysctl_sched_min_granularity);
+       PN(sysctl_sched_wakeup_granularity);
+       PN(sysctl_sched_batch_wakeup_granularity);
+       PN(sysctl_sched_child_runs_first);
+       P(sysctl_sched_features);
+#undef PN
+#undef P
 
        for_each_online_cpu(cpu)
-               print_cpu(m, cpu, now);
+               print_cpu(m, cpu);
 
        SEQ_printf(m, "\n");
 
        return 0;
 }
 
-void sysrq_sched_debug_show(void)
+static void sysrq_sched_debug_show(void)
 {
        sched_debug_show(NULL, NULL);
 }
 
+#ifdef CONFIG_FAIR_USER_SCHED
+
+static DEFINE_MUTEX(root_user_share_mutex);
+
+static int
+root_user_share_read_proc(char *page, char **start, off_t off, int count,
+                                int *eof, void *data)
+{
+       int len;
+
+       len = sprintf(page, "%d\n", init_task_grp_load);
+
+       return len;
+}
+
+static int
+root_user_share_write_proc(struct file *file, const char __user *buffer,
+                                unsigned long count, void *data)
+{
+       unsigned long shares;
+       char kbuf[sizeof(unsigned long)+1];
+       int rc = 0;
+
+       if (copy_from_user(kbuf, buffer, sizeof(kbuf)))
+               return -EFAULT;
+
+       shares = simple_strtoul(kbuf, NULL, 0);
+
+       if (!shares)
+               shares = NICE_0_LOAD;
+
+       mutex_lock(&root_user_share_mutex);
+
+       init_task_grp_load = shares;
+       rc = sched_group_set_shares(&init_task_grp, shares);
+
+       mutex_unlock(&root_user_share_mutex);
+
+       return (rc < 0 ? rc : count);
+}
+
+#endif /* CONFIG_FAIR_USER_SCHED */
+
 static int sched_debug_open(struct inode *inode, struct file *filp)
 {
        return single_open(filp, sched_debug_show, NULL);
@@ -200,7 +281,7 @@ static struct file_operations sched_debug_fops = {
        .open           = sched_debug_open,
        .read           = seq_read,
        .llseek         = seq_lseek,
-       .release        = seq_release,
+       .release        = single_release,
 };
 
 static int __init init_sched_debug_procfs(void)
@@ -213,6 +294,15 @@ static int __init init_sched_debug_procfs(void)
 
        pe->proc_fops = &sched_debug_fops;
 
+#ifdef CONFIG_FAIR_USER_SCHED
+       pe = create_proc_entry("root_user_share", 0644, NULL);
+       if (!pe)
+               return -ENOMEM;
+
+       pe->read_proc = root_user_share_read_proc;
+       pe->write_proc = root_user_share_write_proc;
+#endif
+
        return 0;
 }
 
@@ -234,28 +324,31 @@ void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
        SEQ_printf(m, "----------------------------------------------\n");
 #define P(F) \
        SEQ_printf(m, "%-25s:%20Ld\n", #F, (long long)p->F)
-
-       P(se.wait_start);
-       P(se.wait_start_fair);
-       P(se.exec_start);
-       P(se.sleep_start);
-       P(se.sleep_start_fair);
-       P(se.block_start);
-       P(se.sleep_max);
-       P(se.block_max);
-       P(se.exec_max);
-       P(se.wait_max);
-       P(se.wait_runtime);
-       P(se.wait_runtime_overruns);
-       P(se.wait_runtime_underruns);
-       P(se.sum_wait_runtime);
-       P(se.sum_exec_runtime);
+#define PN(F) \
+       SEQ_printf(m, "%-25s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
+
+       PN(se.exec_start);
+       PN(se.vruntime);
+       PN(se.sum_exec_runtime);
+
+#ifdef CONFIG_SCHEDSTATS
+       PN(se.wait_start);
+       PN(se.sleep_start);
+       PN(se.block_start);
+       PN(se.sleep_max);
+       PN(se.block_max);
+       PN(se.exec_max);
+       PN(se.slice_max);
+       PN(se.wait_max);
+       P(sched_info.bkl_cnt);
+#endif
        SEQ_printf(m, "%-25s:%20Ld\n",
                   "nr_switches", (long long)(p->nvcsw + p->nivcsw));
        P(se.load.weight);
        P(policy);
        P(prio);
 #undef P
+#undef PN
 
        {
                u64 t0, t1;
@@ -269,7 +362,14 @@ void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
 
 void proc_sched_set_task(struct task_struct *p)
 {
-       p->se.sleep_max = p->se.block_max = p->se.exec_max = p->se.wait_max = 0;
-       p->se.wait_runtime_overruns = p->se.wait_runtime_underruns = 0;
-       p->se.sum_exec_runtime = 0;
+#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->sched_info.bkl_cnt           = 0;
+#endif
+       p->se.sum_exec_runtime          = 0;
+       p->se.prev_sum_exec_runtime     = 0;
 }