kernel/sched_debug.c

   1 /*
   2  * kernel/time/sched_debug.c
   3  *
   4  * Print the CFS rbtree
   5  *
   6  * Copyright(C) 2007, Red Hat, Inc., Ingo Molnar
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 as
  10  * published by the Free Software Foundation.
  11  */
  12
  13 #include <linux/proc_fs.h>
  14 #include <linux/sched.h>
  15 #include <linux/seq_file.h>
  16 #include <linux/kallsyms.h>
  17 #include <linux/utsname.h>
  18
  19 /*
  20  * This allows printing both to /proc/sched_debug and
  21  * to the console
  22  */
  23 #define SEQ_printf(m, x...)                     \
  24  do {                                           \
  25         if (m)                                  \
  26                 seq_printf(m, x);               \
  27         else                                    \
  28                 printk(x);                      \
  29  } while (0)
  30
  31 /*
  32  * Ease the printing of nsec fields:
  33  */
  34 static long long nsec_high(long long nsec)
  35 {
  36         if (nsec < 0) {
  37                 nsec = -nsec;
  38                 do_div(nsec, 1000000);
  39                 return -nsec;
  40         }
  41         do_div(nsec, 1000000);
  42
  43         return nsec;
  44 }
  45
  46 static unsigned long nsec_low(long long nsec)
  47 {
  48         if (nsec < 0)
  49                 nsec = -nsec;
  50
  51         return do_div(nsec, 1000000);
  52 }
  53
  54 #define SPLIT_NS(x) nsec_high(x), nsec_low(x)
  55
  56 static void
  57 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
  58 {
  59         if (rq->curr == p)
  60                 SEQ_printf(m, "R");
  61         else
  62                 SEQ_printf(m, " ");
  63
  64         SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
  65                 p->comm, p->pid,
  66                 SPLIT_NS(p->se.vruntime),
  67                 (long long)(p->nvcsw + p->nivcsw),
  68                 p->prio);
  69 #ifdef CONFIG_SCHEDSTATS
  70         SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld\n",
  71                 SPLIT_NS(p->se.vruntime),
  72                 SPLIT_NS(p->se.sum_exec_runtime),
  73                 SPLIT_NS(p->se.sum_sleep_runtime));
  74 #else
  75         SEQ_printf(m, "%15Ld %15Ld %15Ld.%06ld %15Ld.%06ld %15Ld.%06ld\n",
  76                 0LL, 0LL, 0LL, 0L, 0LL, 0L, 0LL, 0L);
  77 #endif
  78 }
  79
  80 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
  81 {
  82         struct task_struct *g, *p;
  83
  84         SEQ_printf(m,
  85         "\nrunnable tasks:\n"
  86         "            task   PID         tree-key  switches  prio"
  87         "     exec-runtime         sum-exec        sum-sleep\n"
  88         "------------------------------------------------------"
  89         "----------------------------------------------------\n");
  90
  91         read_lock_irq(&tasklist_lock);
  92
  93         do_each_thread(g, p) {
  94                 if (!p->se.on_rq || task_cpu(p) != rq_cpu)
  95                         continue;
  96
  97                 print_task(m, rq, p);
  98         } while_each_thread(g, p);
  99
 100         read_unlock_irq(&tasklist_lock);
 101 }
 102
 103 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
 104 {
 105         s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
 106                 spread, rq0_min_vruntime, spread0;
 107         struct rq *rq = &per_cpu(runqueues, cpu);
 108         struct sched_entity *last;
 109         unsigned long flags;
 110
 111         SEQ_printf(m, "\ncfs_rq\n");
 112
 113         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
 114                         SPLIT_NS(cfs_rq->exec_clock));
 115
 116         spin_lock_irqsave(&rq->lock, flags);
 117         if (cfs_rq->rb_leftmost)
 118                 MIN_vruntime = (__pick_next_entity(cfs_rq))->vruntime;
 119         last = __pick_last_entity(cfs_rq);
 120         if (last)
 121                 max_vruntime = last->vruntime;
 122         min_vruntime = rq->cfs.min_vruntime;
 123         rq0_min_vruntime = per_cpu(runqueues, 0).cfs.min_vruntime;
 124         spin_unlock_irqrestore(&rq->lock, flags);
 125         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
 126                         SPLIT_NS(MIN_vruntime));
 127         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
 128                         SPLIT_NS(min_vruntime));
 129         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
 130                         SPLIT_NS(max_vruntime));
 131         spread = max_vruntime - MIN_vruntime;
 132         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
 133                         SPLIT_NS(spread));
 134         spread0 = min_vruntime - rq0_min_vruntime;
 135         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
 136                         SPLIT_NS(spread0));
 137         SEQ_printf(m, "  .%-30s: %ld\n", "nr_running", cfs_rq->nr_running);
 138         SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
 139 #ifdef CONFIG_SCHEDSTATS
 140         SEQ_printf(m, "  .%-30s: %ld\n", "bkl_cnt",
 141                         rq->bkl_cnt);
 142 #endif
 143 }
 144
 145 static void print_cpu(struct seq_file *m, int cpu)
 146 {
 147         struct rq *rq = &per_cpu(runqueues, cpu);
 148
 149 #ifdef CONFIG_X86
 150         {
 151                 unsigned int freq = cpu_khz ? : 1;
 152
 153                 SEQ_printf(m, "\ncpu#%d, %u.%03u MHz\n",
 154                            cpu, freq / 1000, (freq % 1000));
 155         }
 156 #else
 157         SEQ_printf(m, "\ncpu#%d\n", cpu);
 158 #endif
 159
 160 #define P(x) \
 161         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x))
 162 #define PN(x) \
 163         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
 164
 165         P(nr_running);
 166         SEQ_printf(m, "  .%-30s: %lu\n", "load",
 167                    rq->load.weight);
 168         P(nr_switches);
 169         P(nr_load_updates);
 170         P(nr_uninterruptible);
 171         SEQ_printf(m, "  .%-30s: %lu\n", "jiffies", jiffies);
 172         PN(next_balance);
 173         P(curr->pid);
 174         PN(clock);
 175         PN(idle_clock);
 176         PN(prev_clock_raw);
 177         P(clock_warps);
 178         P(clock_overflows);
 179         P(clock_deep_idle_events);
 180         PN(clock_max_delta);
 181         P(cpu_load[0]);
 182         P(cpu_load[1]);
 183         P(cpu_load[2]);
 184         P(cpu_load[3]);
 185         P(cpu_load[4]);
 186 #undef P
 187 #undef PN
 188
 189         print_cfs_stats(m, cpu);
 190
 191         print_rq(m, rq, cpu);
 192 }
 193
 194 static int sched_debug_show(struct seq_file *m, void *v)
 195 {
 196         u64 now = ktime_to_ns(ktime_get());
 197         int cpu;
 198
 199         SEQ_printf(m, "Sched Debug Version: v0.05-v20, %s %.*s\n",
 200                 init_utsname()->release,
 201                 (int)strcspn(init_utsname()->version, " "),
 202                 init_utsname()->version);
 203
 204         SEQ_printf(m, "now at %Lu.%06ld msecs\n", SPLIT_NS(now));
 205
 206 #define P(x) \
 207         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(x))
 208 #define PN(x) \
 209         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 210         PN(sysctl_sched_latency);
 211         PN(sysctl_sched_min_granularity);
 212         PN(sysctl_sched_wakeup_granularity);
 213         PN(sysctl_sched_batch_wakeup_granularity);
 214         PN(sysctl_sched_child_runs_first);
 215         P(sysctl_sched_features);
 216 #undef PN
 217 #undef P
 218
 219         for_each_online_cpu(cpu)
 220                 print_cpu(m, cpu);
 221
 222         SEQ_printf(m, "\n");
 223
 224         return 0;
 225 }
 226
 227 static void sysrq_sched_debug_show(void)
 228 {
 229         sched_debug_show(NULL, NULL);
 230 }
 231
 232 #ifdef CONFIG_FAIR_USER_SCHED
 233
 234 static DEFINE_MUTEX(root_user_share_mutex);
 235
 236 static int
 237 root_user_share_read_proc(char *page, char **start, off_t off, int count,
 238                                  int *eof, void *data)
 239 {
 240         int len;
 241
 242         len = sprintf(page, "%d\n", init_task_grp_load);
 243
 244         return len;
 245 }
 246
 247 static int
 248 root_user_share_write_proc(struct file *file, const char __user *buffer,
 249                                  unsigned long count, void *data)
 250 {
 251         unsigned long shares;
 252         char kbuf[sizeof(unsigned long)+1];
 253         int rc = 0;
 254
 255         if (copy_from_user(kbuf, buffer, sizeof(kbuf)))
 256                 return -EFAULT;
 257
 258         shares = simple_strtoul(kbuf, NULL, 0);
 259
 260         if (!shares)
 261                 shares = NICE_0_LOAD;
 262
 263         mutex_lock(&root_user_share_mutex);
 264
 265         init_task_grp_load = shares;
 266         rc = sched_group_set_shares(&init_task_grp, shares);
 267
 268         mutex_unlock(&root_user_share_mutex);
 269
 270         return (rc < 0 ? rc : count);
 271 }
 272
 273 #endif  /* CONFIG_FAIR_USER_SCHED */
 274
 275 static int sched_debug_open(struct inode *inode, struct file *filp)
 276 {
 277         return single_open(filp, sched_debug_show, NULL);
 278 }
 279
 280 static struct file_operations sched_debug_fops = {
 281         .open           = sched_debug_open,
 282         .read           = seq_read,
 283         .llseek         = seq_lseek,
 284         .release        = single_release,
 285 };
 286
 287 static int __init init_sched_debug_procfs(void)
 288 {
 289         struct proc_dir_entry *pe;
 290
 291         pe = create_proc_entry("sched_debug", 0644, NULL);
 292         if (!pe)
 293                 return -ENOMEM;
 294
 295         pe->proc_fops = &sched_debug_fops;
 296
 297 #ifdef CONFIG_FAIR_USER_SCHED
 298         pe = create_proc_entry("root_user_share", 0644, NULL);
 299         if (!pe)
 300                 return -ENOMEM;
 301
 302         pe->read_proc = root_user_share_read_proc;
 303         pe->write_proc = root_user_share_write_proc;
 304 #endif
 305
 306         return 0;
 307 }
 308
 309 __initcall(init_sched_debug_procfs);
 310
 311 void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
 312 {
 313         unsigned long flags;
 314         int num_threads = 1;
 315
 316         rcu_read_lock();
 317         if (lock_task_sighand(p, &flags)) {
 318                 num_threads = atomic_read(&p->signal->count);
 319                 unlock_task_sighand(p, &flags);
 320         }
 321         rcu_read_unlock();
 322
 323         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, p->pid, num_threads);
 324         SEQ_printf(m, "----------------------------------------------\n");
 325 #define P(F) \
 326         SEQ_printf(m, "%-25s:%20Ld\n", #F, (long long)p->F)
 327 #define PN(F) \
 328         SEQ_printf(m, "%-25s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
 329
 330         PN(se.exec_start);
 331         PN(se.vruntime);
 332         PN(se.sum_exec_runtime);
 333
 334 #ifdef CONFIG_SCHEDSTATS
 335         PN(se.wait_start);
 336         PN(se.sleep_start);
 337         PN(se.block_start);
 338         PN(se.sleep_max);
 339         PN(se.block_max);
 340         PN(se.exec_max);
 341         PN(se.slice_max);
 342         PN(se.wait_max);
 343         P(sched_info.bkl_cnt);
 344 #endif
 345         SEQ_printf(m, "%-25s:%20Ld\n",
 346                    "nr_switches", (long long)(p->nvcsw + p->nivcsw));
 347         P(se.load.weight);
 348         P(policy);
 349         P(prio);
 350 #undef P
 351 #undef PN
 352
 353         {
 354                 u64 t0, t1;
 355
 356                 t0 = sched_clock();
 357                 t1 = sched_clock();
 358                 SEQ_printf(m, "%-25s:%20Ld\n",
 359                            "clock-delta", (long long)(t1-t0));
 360         }
 361 }
 362
 363 void proc_sched_set_task(struct task_struct *p)
 364 {
 365 #ifdef CONFIG_SCHEDSTATS
 366         p->se.sleep_max                 = 0;
 367         p->se.block_max                 = 0;
 368         p->se.exec_max                  = 0;
 369         p->se.slice_max                 = 0;
 370         p->se.wait_max                  = 0;
 371         p->sched_info.bkl_cnt           = 0;
 372 #endif
 373         p->se.sum_exec_runtime          = 0;
 374         p->se.prev_sum_exec_runtime     = 0;
 375 }