sched debug: BKL usage statistics
[linux-block.git] / kernel / sched_debug.c
CommitLineData
43ae34cb
IM
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
ef83a571
IM
31/*
32 * Ease the printing of nsec fields:
33 */
34static 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
46static 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
43ae34cb 56static void
a48da48b 57print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
43ae34cb
IM
58{
59 if (rq->curr == p)
60 SEQ_printf(m, "R");
61 else
62 SEQ_printf(m, " ");
63
ef83a571 64 SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
43ae34cb 65 p->comm, p->pid,
ef83a571 66 SPLIT_NS(p->se.vruntime),
43ae34cb 67 (long long)(p->nvcsw + p->nivcsw),
6f605d83 68 p->prio);
6cfb0d5d 69#ifdef CONFIG_SCHEDSTATS
c86da3a3 70 SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld\n",
ef83a571
IM
71 SPLIT_NS(p->se.vruntime),
72 SPLIT_NS(p->se.sum_exec_runtime),
73 SPLIT_NS(p->se.sum_sleep_runtime));
6cfb0d5d 74#else
ef83a571
IM
75 SEQ_printf(m, "%15Ld %15Ld %15Ld.%06ld %15Ld.%06ld %15Ld.%06ld\n",
76 0LL, 0LL, 0LL, 0L, 0LL, 0L, 0LL, 0L);
6cfb0d5d 77#endif
43ae34cb
IM
78}
79
a48da48b 80static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
43ae34cb
IM
81{
82 struct task_struct *g, *p;
83
84 SEQ_printf(m,
85 "\nrunnable tasks:\n"
c86da3a3
MG
86 " task PID tree-key switches prio"
87 " exec-runtime sum-exec sum-sleep\n"
1a75b94f 88 "------------------------------------------------------"
c86da3a3 89 "----------------------------------------------------\n");
43ae34cb
IM
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
a48da48b 97 print_task(m, rq, p);
43ae34cb
IM
98 } while_each_thread(g, p);
99
100 read_unlock_irq(&tasklist_lock);
101}
102
5cef9eca 103void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
43ae34cb 104{
86d9560c
IM
105 s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
106 spread, rq0_min_vruntime, spread0;
67e12eac
IM
107 struct rq *rq = &per_cpu(runqueues, cpu);
108 struct sched_entity *last;
109 unsigned long flags;
110
5167e75f 111 SEQ_printf(m, "\ncfs_rq\n");
43ae34cb 112
ef83a571
IM
113 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "exec_clock",
114 SPLIT_NS(cfs_rq->exec_clock));
67e12eac
IM
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;
86d9560c
IM
122 min_vruntime = rq->cfs.min_vruntime;
123 rq0_min_vruntime = per_cpu(runqueues, 0).cfs.min_vruntime;
67e12eac 124 spin_unlock_irqrestore(&rq->lock, flags);
ef83a571
IM
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));
67e12eac 131 spread = max_vruntime - MIN_vruntime;
ef83a571
IM
132 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread",
133 SPLIT_NS(spread));
86d9560c 134 spread0 = min_vruntime - rq0_min_vruntime;
ef83a571
IM
135 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread0",
136 SPLIT_NS(spread0));
545f3b18
SV
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);
b8efb561
IM
139 SEQ_printf(m, " .%-30s: %ld\n", "bkl_cnt",
140 rq->bkl_cnt);
43ae34cb
IM
141}
142
a48da48b 143static void print_cpu(struct seq_file *m, int cpu)
43ae34cb
IM
144{
145 struct rq *rq = &per_cpu(runqueues, cpu);
146
147#ifdef CONFIG_X86
148 {
149 unsigned int freq = cpu_khz ? : 1;
150
151 SEQ_printf(m, "\ncpu#%d, %u.%03u MHz\n",
152 cpu, freq / 1000, (freq % 1000));
153 }
154#else
155 SEQ_printf(m, "\ncpu#%d\n", cpu);
156#endif
157
158#define P(x) \
159 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x))
ef83a571
IM
160#define PN(x) \
161 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
43ae34cb
IM
162
163 P(nr_running);
164 SEQ_printf(m, " .%-30s: %lu\n", "load",
495eca49 165 rq->load.weight);
43ae34cb
IM
166 P(nr_switches);
167 P(nr_load_updates);
168 P(nr_uninterruptible);
169 SEQ_printf(m, " .%-30s: %lu\n", "jiffies", jiffies);
ef83a571 170 PN(next_balance);
43ae34cb 171 P(curr->pid);
ef83a571
IM
172 PN(clock);
173 PN(idle_clock);
174 PN(prev_clock_raw);
43ae34cb
IM
175 P(clock_warps);
176 P(clock_overflows);
2aa44d05 177 P(clock_deep_idle_events);
ef83a571 178 PN(clock_max_delta);
43ae34cb
IM
179 P(cpu_load[0]);
180 P(cpu_load[1]);
181 P(cpu_load[2]);
182 P(cpu_load[3]);
183 P(cpu_load[4]);
184#undef P
ef83a571 185#undef PN
43ae34cb 186
5cef9eca 187 print_cfs_stats(m, cpu);
43ae34cb 188
a48da48b 189 print_rq(m, rq, cpu);
43ae34cb
IM
190}
191
192static int sched_debug_show(struct seq_file *m, void *v)
193{
194 u64 now = ktime_to_ns(ktime_get());
195 int cpu;
196
6cfb0d5d 197 SEQ_printf(m, "Sched Debug Version: v0.05-v20, %s %.*s\n",
43ae34cb
IM
198 init_utsname()->release,
199 (int)strcspn(init_utsname()->version, " "),
200 init_utsname()->version);
201
ef83a571 202 SEQ_printf(m, "now at %Lu.%06ld msecs\n", SPLIT_NS(now));
43ae34cb
IM
203
204 for_each_online_cpu(cpu)
a48da48b 205 print_cpu(m, cpu);
43ae34cb
IM
206
207 SEQ_printf(m, "\n");
208
209 return 0;
210}
211
f3373461 212static void sysrq_sched_debug_show(void)
43ae34cb
IM
213{
214 sched_debug_show(NULL, NULL);
215}
216
24e377a8
SV
217#ifdef CONFIG_FAIR_USER_SCHED
218
219static DEFINE_MUTEX(root_user_share_mutex);
220
221static int
222root_user_share_read_proc(char *page, char **start, off_t off, int count,
223 int *eof, void *data)
224{
225 int len;
226
227 len = sprintf(page, "%d\n", init_task_grp_load);
228
229 return len;
230}
231
232static int
233root_user_share_write_proc(struct file *file, const char __user *buffer,
234 unsigned long count, void *data)
235{
236 unsigned long shares;
237 char kbuf[sizeof(unsigned long)+1];
238 int rc = 0;
239
240 if (copy_from_user(kbuf, buffer, sizeof(kbuf)))
241 return -EFAULT;
242
243 shares = simple_strtoul(kbuf, NULL, 0);
244
245 if (!shares)
246 shares = NICE_0_LOAD;
247
248 mutex_lock(&root_user_share_mutex);
249
250 init_task_grp_load = shares;
251 rc = sched_group_set_shares(&init_task_grp, shares);
252
253 mutex_unlock(&root_user_share_mutex);
254
255 return (rc < 0 ? rc : count);
256}
257
258#endif /* CONFIG_FAIR_USER_SCHED */
259
43ae34cb
IM
260static int sched_debug_open(struct inode *inode, struct file *filp)
261{
262 return single_open(filp, sched_debug_show, NULL);
263}
264
265static struct file_operations sched_debug_fops = {
266 .open = sched_debug_open,
267 .read = seq_read,
268 .llseek = seq_lseek,
5ea473a1 269 .release = single_release,
43ae34cb
IM
270};
271
272static int __init init_sched_debug_procfs(void)
273{
274 struct proc_dir_entry *pe;
275
276 pe = create_proc_entry("sched_debug", 0644, NULL);
277 if (!pe)
278 return -ENOMEM;
279
280 pe->proc_fops = &sched_debug_fops;
281
24e377a8
SV
282#ifdef CONFIG_FAIR_USER_SCHED
283 pe = create_proc_entry("root_user_share", 0644, NULL);
284 if (!pe)
285 return -ENOMEM;
286
287 pe->read_proc = root_user_share_read_proc;
288 pe->write_proc = root_user_share_write_proc;
289#endif
290
43ae34cb
IM
291 return 0;
292}
293
294__initcall(init_sched_debug_procfs);
295
296void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
297{
298 unsigned long flags;
299 int num_threads = 1;
300
301 rcu_read_lock();
302 if (lock_task_sighand(p, &flags)) {
303 num_threads = atomic_read(&p->signal->count);
304 unlock_task_sighand(p, &flags);
305 }
306 rcu_read_unlock();
307
308 SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, p->pid, num_threads);
309 SEQ_printf(m, "----------------------------------------------\n");
310#define P(F) \
311 SEQ_printf(m, "%-25s:%20Ld\n", #F, (long long)p->F)
ef83a571
IM
312#define PN(F) \
313 SEQ_printf(m, "%-25s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
43ae34cb 314
ef83a571
IM
315 PN(se.exec_start);
316 PN(se.vruntime);
317 PN(se.sum_exec_runtime);
6cfb0d5d
IM
318
319#ifdef CONFIG_SCHEDSTATS
ef83a571
IM
320 PN(se.wait_start);
321 PN(se.sleep_start);
322 PN(se.block_start);
323 PN(se.sleep_max);
324 PN(se.block_max);
325 PN(se.exec_max);
326 PN(se.slice_max);
327 PN(se.wait_max);
b8efb561 328 P(sched_info.bkl_cnt);
6cfb0d5d 329#endif
43ae34cb
IM
330 SEQ_printf(m, "%-25s:%20Ld\n",
331 "nr_switches", (long long)(p->nvcsw + p->nivcsw));
332 P(se.load.weight);
333 P(policy);
334 P(prio);
335#undef P
ef83a571 336#undef PN
43ae34cb
IM
337
338 {
339 u64 t0, t1;
340
341 t0 = sched_clock();
342 t1 = sched_clock();
343 SEQ_printf(m, "%-25s:%20Ld\n",
344 "clock-delta", (long long)(t1-t0));
345 }
346}
347
348void proc_sched_set_task(struct task_struct *p)
349{
6cfb0d5d 350#ifdef CONFIG_SCHEDSTATS
a4b29ba2
IM
351 p->se.sleep_max = 0;
352 p->se.block_max = 0;
353 p->se.exec_max = 0;
eba1ed4b 354 p->se.slice_max = 0;
a4b29ba2 355 p->se.wait_max = 0;
b8efb561 356 p->sched_info.bkl_cnt = 0;
6cfb0d5d 357#endif
a4b29ba2 358 p->se.sum_exec_runtime = 0;
2491b2b8 359 p->se.prev_sum_exec_runtime = 0;
43ae34cb 360}