sched: add fair-user scheduler
[linux-2.6.git] / 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 }
140
141 static void print_cpu(struct seq_file *m, int cpu)
142 {
143         struct rq *rq = &per_cpu(runqueues, cpu);
144
145 #ifdef CONFIG_X86
146         {
147                 unsigned int freq = cpu_khz ? : 1;
148
149                 SEQ_printf(m, "\ncpu#%d, %u.%03u MHz\n",
150                            cpu, freq / 1000, (freq % 1000));
151         }
152 #else
153         SEQ_printf(m, "\ncpu#%d\n", cpu);
154 #endif
155
156 #define P(x) \
157         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x))
158 #define PN(x) \
159         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
160
161         P(nr_running);
162         SEQ_printf(m, "  .%-30s: %lu\n", "load",
163                    rq->load.weight);
164         P(nr_switches);
165         P(nr_load_updates);
166         P(nr_uninterruptible);
167         SEQ_printf(m, "  .%-30s: %lu\n", "jiffies", jiffies);
168         PN(next_balance);
169         P(curr->pid);
170         PN(clock);
171         PN(idle_clock);
172         PN(prev_clock_raw);
173         P(clock_warps);
174         P(clock_overflows);
175         P(clock_deep_idle_events);
176         PN(clock_max_delta);
177         P(cpu_load[0]);
178         P(cpu_load[1]);
179         P(cpu_load[2]);
180         P(cpu_load[3]);
181         P(cpu_load[4]);
182 #undef P
183 #undef PN
184
185         print_cfs_stats(m, cpu);
186
187         print_rq(m, rq, cpu);
188 }
189
190 static int sched_debug_show(struct seq_file *m, void *v)
191 {
192         u64 now = ktime_to_ns(ktime_get());
193         int cpu;
194
195         SEQ_printf(m, "Sched Debug Version: v0.05-v20, %s %.*s\n",
196                 init_utsname()->release,
197                 (int)strcspn(init_utsname()->version, " "),
198                 init_utsname()->version);
199
200         SEQ_printf(m, "now at %Lu.%06ld msecs\n", SPLIT_NS(now));
201
202         for_each_online_cpu(cpu)
203                 print_cpu(m, cpu);
204
205         SEQ_printf(m, "\n");
206
207         return 0;
208 }
209
210 static void sysrq_sched_debug_show(void)
211 {
212         sched_debug_show(NULL, NULL);
213 }
214
215 #ifdef CONFIG_FAIR_USER_SCHED
216
217 static DEFINE_MUTEX(root_user_share_mutex);
218
219 static int
220 root_user_share_read_proc(char *page, char **start, off_t off, int count,
221                                  int *eof, void *data)
222 {
223         int len;
224
225         len = sprintf(page, "%d\n", init_task_grp_load);
226
227         return len;
228 }
229
230 static int
231 root_user_share_write_proc(struct file *file, const char __user *buffer,
232                                  unsigned long count, void *data)
233 {
234         unsigned long shares;
235         char kbuf[sizeof(unsigned long)+1];
236         int rc = 0;
237
238         if (copy_from_user(kbuf, buffer, sizeof(kbuf)))
239                 return -EFAULT;
240
241         shares = simple_strtoul(kbuf, NULL, 0);
242
243         if (!shares)
244                 shares = NICE_0_LOAD;
245
246         mutex_lock(&root_user_share_mutex);
247
248         init_task_grp_load = shares;
249         rc = sched_group_set_shares(&init_task_grp, shares);
250
251         mutex_unlock(&root_user_share_mutex);
252
253         return (rc < 0 ? rc : count);
254 }
255
256 #endif  /* CONFIG_FAIR_USER_SCHED */
257
258 static int sched_debug_open(struct inode *inode, struct file *filp)
259 {
260         return single_open(filp, sched_debug_show, NULL);
261 }
262
263 static struct file_operations sched_debug_fops = {
264         .open           = sched_debug_open,
265         .read           = seq_read,
266         .llseek         = seq_lseek,
267         .release        = single_release,
268 };
269
270 static int __init init_sched_debug_procfs(void)
271 {
272         struct proc_dir_entry *pe;
273
274         pe = create_proc_entry("sched_debug", 0644, NULL);
275         if (!pe)
276                 return -ENOMEM;
277
278         pe->proc_fops = &sched_debug_fops;
279
280 #ifdef CONFIG_FAIR_USER_SCHED
281         pe = create_proc_entry("root_user_share", 0644, NULL);
282         if (!pe)
283                 return -ENOMEM;
284
285         pe->read_proc = root_user_share_read_proc;
286         pe->write_proc = root_user_share_write_proc;
287 #endif
288
289         return 0;
290 }
291
292 __initcall(init_sched_debug_procfs);
293
294 void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
295 {
296         unsigned long flags;
297         int num_threads = 1;
298
299         rcu_read_lock();
300         if (lock_task_sighand(p, &flags)) {
301                 num_threads = atomic_read(&p->signal->count);
302                 unlock_task_sighand(p, &flags);
303         }
304         rcu_read_unlock();
305
306         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, p->pid, num_threads);
307         SEQ_printf(m, "----------------------------------------------\n");
308 #define P(F) \
309         SEQ_printf(m, "%-25s:%20Ld\n", #F, (long long)p->F)
310 #define PN(F) \
311         SEQ_printf(m, "%-25s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
312
313         PN(se.exec_start);
314         PN(se.vruntime);
315         PN(se.sum_exec_runtime);
316
317 #ifdef CONFIG_SCHEDSTATS
318         PN(se.wait_start);
319         PN(se.sleep_start);
320         PN(se.block_start);
321         PN(se.sleep_max);
322         PN(se.block_max);
323         PN(se.exec_max);
324         PN(se.slice_max);
325         PN(se.wait_max);
326 #endif
327         SEQ_printf(m, "%-25s:%20Ld\n",
328                    "nr_switches", (long long)(p->nvcsw + p->nivcsw));
329         P(se.load.weight);
330         P(policy);
331         P(prio);
332 #undef P
333 #undef PN
334
335         {
336                 u64 t0, t1;
337
338                 t0 = sched_clock();
339                 t1 = sched_clock();
340                 SEQ_printf(m, "%-25s:%20Ld\n",
341                            "clock-delta", (long long)(t1-t0));
342         }
343 }
344
345 void proc_sched_set_task(struct task_struct *p)
346 {
347 #ifdef CONFIG_SCHEDSTATS
348         p->se.sleep_max                 = 0;
349         p->se.block_max                 = 0;
350         p->se.exec_max                  = 0;
351         p->se.slice_max                 = 0;
352         p->se.wait_max                  = 0;
353 #endif
354         p->se.sum_exec_runtime          = 0;
355         p->se.prev_sum_exec_runtime     = 0;
356 }