sched: scheduler debugging, core
authorIngo Molnar <mingo@elte.hu>
Mon, 9 Jul 2007 16:52:00 +0000 (18:52 +0200)
committerIngo Molnar <mingo@elte.hu>
Mon, 9 Jul 2007 16:52:00 +0000 (18:52 +0200)
scheduler debugging core: implement /proc/sched_debug and
/proc/<PID>/sched files for scheduler debugging.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
fs/proc/base.c
include/linux/sched.h
kernel/sched_debug.c [new file with mode: 0644]

index 0f40e82..46ea5d5 100644 (file)
@@ -929,6 +929,69 @@ static const struct file_operations proc_fault_inject_operations = {
 };
 #endif
 
+#ifdef CONFIG_SCHED_DEBUG
+/*
+ * Print out various scheduling related per-task fields:
+ */
+static int sched_show(struct seq_file *m, void *v)
+{
+       struct inode *inode = m->private;
+       struct task_struct *p;
+
+       WARN_ON(!inode);
+
+       p = get_proc_task(inode);
+       if (!p)
+               return -ESRCH;
+       proc_sched_show_task(p, m);
+
+       put_task_struct(p);
+
+       return 0;
+}
+
+static ssize_t
+sched_write(struct file *file, const char __user *buf,
+           size_t count, loff_t *offset)
+{
+       struct inode *inode = file->f_path.dentry->d_inode;
+       struct task_struct *p;
+
+       WARN_ON(!inode);
+
+       p = get_proc_task(inode);
+       if (!p)
+               return -ESRCH;
+       proc_sched_set_task(p);
+
+       put_task_struct(p);
+
+       return count;
+}
+
+static int sched_open(struct inode *inode, struct file *filp)
+{
+       int ret;
+
+       ret = single_open(filp, sched_show, NULL);
+       if (!ret) {
+               struct seq_file *m = filp->private_data;
+
+               m->private = inode;
+       }
+       return ret;
+}
+
+static const struct file_operations proc_pid_sched_operations = {
+       .open           = sched_open,
+       .read           = seq_read,
+       .write          = sched_write,
+       .llseek         = seq_lseek,
+       .release        = seq_release,
+};
+
+#endif
+
 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
 {
        struct inode *inode = dentry->d_inode;
@@ -1963,6 +2026,9 @@ static const struct pid_entry tgid_base_stuff[] = {
        INF("environ",    S_IRUSR, pid_environ),
        INF("auxv",       S_IRUSR, pid_auxv),
        INF("status",     S_IRUGO, pid_status),
+#ifdef CONFIG_SCHED_DEBUG
+       REG("sched",      S_IRUGO|S_IWUSR, pid_sched),
+#endif
        INF("cmdline",    S_IRUGO, pid_cmdline),
        INF("stat",       S_IRUGO, tgid_stat),
        INF("statm",      S_IRUGO, pid_statm),
@@ -2247,6 +2313,9 @@ static const struct pid_entry tid_base_stuff[] = {
        INF("environ",   S_IRUSR, pid_environ),
        INF("auxv",      S_IRUSR, pid_auxv),
        INF("status",    S_IRUGO, pid_status),
+#ifdef CONFIG_SCHED_DEBUG
+       REG("sched",     S_IRUGO|S_IWUSR, pid_sched),
+#endif
        INF("cmdline",   S_IRUGO, pid_cmdline),
        INF("stat",      S_IRUGO, tid_stat),
        INF("statm",     S_IRUGO, pid_statm),
index c9d6573..785ec84 100644 (file)
@@ -132,6 +132,26 @@ extern unsigned long nr_active(void);
 extern unsigned long nr_iowait(void);
 extern unsigned long weighted_cpuload(const int cpu);
 
+struct seq_file;
+struct cfs_rq;
+#ifdef CONFIG_SCHED_DEBUG
+extern void proc_sched_show_task(struct task_struct *p, struct seq_file *m);
+extern void proc_sched_set_task(struct task_struct *p);
+extern void
+print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq, u64 now);
+#else
+static inline void
+proc_sched_show_task(struct task_struct *p, struct seq_file *m)
+{
+}
+static inline void proc_sched_set_task(struct task_struct *p)
+{
+}
+static inline void
+print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq, u64 now)
+{
+}
+#endif
 
 /*
  * Task state bitmask. NOTE! These bits are also
diff --git a/kernel/sched_debug.c b/kernel/sched_debug.c
new file mode 100644 (file)
index 0000000..1baf87c
--- /dev/null
@@ -0,0 +1,275 @@
+/*
+ * kernel/time/sched_debug.c
+ *
+ * Print the CFS rbtree
+ *
+ * Copyright(C) 2007, Red Hat, Inc., Ingo Molnar
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include <linux/proc_fs.h>
+#include <linux/sched.h>
+#include <linux/seq_file.h>
+#include <linux/kallsyms.h>
+#include <linux/utsname.h>
+
+/*
+ * This allows printing both to /proc/sched_debug and
+ * to the console
+ */
+#define SEQ_printf(m, x...)                    \
+ do {                                          \
+       if (m)                                  \
+               seq_printf(m, x);               \
+       else                                    \
+               printk(x);                      \
+ } while (0)
+
+static void
+print_task(struct seq_file *m, struct rq *rq, struct task_struct *p, u64 now)
+{
+       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",
+               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,
+               (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);
+}
+
+static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu, u64 now)
+{
+       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");
+
+       read_lock_irq(&tasklist_lock);
+
+       do_each_thread(g, p) {
+               if (!p->se.on_rq || task_cpu(p) != rq_cpu)
+                       continue;
+
+               print_task(m, rq, p, now);
+       } 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)
+{
+       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);
+       }
+       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);
+
+#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
+
+       print_cfs_rq_runtime_sum(m, cpu, cfs_rq);
+}
+
+static void print_cpu(struct seq_file *m, int cpu, u64 now)
+{
+       struct rq *rq = &per_cpu(runqueues, cpu);
+
+#ifdef CONFIG_X86
+       {
+               unsigned int freq = cpu_khz ? : 1;
+
+               SEQ_printf(m, "\ncpu#%d, %u.%03u MHz\n",
+                          cpu, freq / 1000, (freq % 1000));
+       }
+#else
+       SEQ_printf(m, "\ncpu#%d\n", cpu);
+#endif
+
+#define P(x) \
+       SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x))
+
+       P(nr_running);
+       SEQ_printf(m, "  .%-30s: %lu\n", "load",
+                  rq->ls.load.weight);
+       P(ls.delta_fair);
+       P(ls.delta_exec);
+       P(nr_switches);
+       P(nr_load_updates);
+       P(nr_uninterruptible);
+       SEQ_printf(m, "  .%-30s: %lu\n", "jiffies", jiffies);
+       P(next_balance);
+       P(curr->pid);
+       P(clock);
+       P(prev_clock_raw);
+       P(clock_warps);
+       P(clock_overflows);
+       P(clock_unstable_events);
+       P(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
+
+       print_cfs_stats(m, cpu, now);
+
+       print_rq(m, rq, cpu, now);
+}
+
+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: v0.04, cfs-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);
+
+       for_each_online_cpu(cpu)
+               print_cpu(m, cpu, now);
+
+       SEQ_printf(m, "\n");
+
+       return 0;
+}
+
+void sysrq_sched_debug_show(void)
+{
+       sched_debug_show(NULL, NULL);
+}
+
+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 = {
+       .open           = sched_debug_open,
+       .read           = seq_read,
+       .llseek         = seq_lseek,
+       .release        = seq_release,
+};
+
+static int __init init_sched_debug_procfs(void)
+{
+       struct proc_dir_entry *pe;
+
+       pe = create_proc_entry("sched_debug", 0644, NULL);
+       if (!pe)
+               return -ENOMEM;
+
+       pe->proc_fops = &sched_debug_fops;
+
+       return 0;
+}
+
+__initcall(init_sched_debug_procfs);
+
+void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
+{
+       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");
+#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);
+       SEQ_printf(m, "%-25s:%20Ld\n",
+                  "nr_switches", (long long)(p->nvcsw + p->nivcsw));
+       P(se.load.weight);
+       P(policy);
+       P(prio);
+#undef P
+
+       {
+               u64 t0, t1;
+
+               t0 = sched_clock();
+               t1 = sched_clock();
+               SEQ_printf(m, "%-25s:%20Ld\n",
+                          "clock-delta", (long long)(t1-t0));
+       }
+}
+
+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;
+}