ext3: Remove unnecessary quota functions
[pandora-kernel.git] / kernel / softlockup.c
1 /*
2  * Detect Soft Lockups
3  *
4  * started by Ingo Molnar, Copyright (C) 2005, 2006 Red Hat, Inc.
5  *
6  * this code detects soft lockups: incidents in where on a CPU
7  * the kernel does not reschedule for 10 seconds or more.
8  */
9 #include <linux/mm.h>
10 #include <linux/cpu.h>
11 #include <linux/nmi.h>
12 #include <linux/init.h>
13 #include <linux/delay.h>
14 #include <linux/freezer.h>
15 #include <linux/kthread.h>
16 #include <linux/lockdep.h>
17 #include <linux/notifier.h>
18 #include <linux/module.h>
19 #include <linux/sysctl.h>
20
21 #include <asm/irq_regs.h>
22
23 static DEFINE_SPINLOCK(print_lock);
24
25 static DEFINE_PER_CPU(unsigned long, touch_timestamp);
26 static DEFINE_PER_CPU(unsigned long, print_timestamp);
27 static DEFINE_PER_CPU(struct task_struct *, watchdog_task);
28
29 static int __read_mostly did_panic;
30 int __read_mostly softlockup_thresh = 60;
31
32 /*
33  * Should we panic (and reboot, if panic_timeout= is set) when a
34  * soft-lockup occurs:
35  */
36 unsigned int __read_mostly softlockup_panic =
37                                 CONFIG_BOOTPARAM_SOFTLOCKUP_PANIC_VALUE;
38
39 static int __init softlockup_panic_setup(char *str)
40 {
41         softlockup_panic = simple_strtoul(str, NULL, 0);
42
43         return 1;
44 }
45 __setup("softlockup_panic=", softlockup_panic_setup);
46
47 static int
48 softlock_panic(struct notifier_block *this, unsigned long event, void *ptr)
49 {
50         did_panic = 1;
51
52         return NOTIFY_DONE;
53 }
54
55 static struct notifier_block panic_block = {
56         .notifier_call = softlock_panic,
57 };
58
59 /*
60  * Returns seconds, approximately.  We don't need nanosecond
61  * resolution, and we don't need to waste time with a big divide when
62  * 2^30ns == 1.074s.
63  */
64 static unsigned long get_timestamp(int this_cpu)
65 {
66         return cpu_clock(this_cpu) >> 30LL;  /* 2^30 ~= 10^9 */
67 }
68
69 static void __touch_softlockup_watchdog(void)
70 {
71         int this_cpu = raw_smp_processor_id();
72
73         __raw_get_cpu_var(touch_timestamp) = get_timestamp(this_cpu);
74 }
75
76 void touch_softlockup_watchdog(void)
77 {
78         __raw_get_cpu_var(touch_timestamp) = 0;
79 }
80 EXPORT_SYMBOL(touch_softlockup_watchdog);
81
82 void touch_all_softlockup_watchdogs(void)
83 {
84         int cpu;
85
86         /* Cause each CPU to re-update its timestamp rather than complain */
87         for_each_online_cpu(cpu)
88                 per_cpu(touch_timestamp, cpu) = 0;
89 }
90 EXPORT_SYMBOL(touch_all_softlockup_watchdogs);
91
92 int proc_dosoftlockup_thresh(struct ctl_table *table, int write,
93                              struct file *filp, void __user *buffer,
94                              size_t *lenp, loff_t *ppos)
95 {
96         touch_all_softlockup_watchdogs();
97         return proc_dointvec_minmax(table, write, filp, buffer, lenp, ppos);
98 }
99
100 /*
101  * This callback runs from the timer interrupt, and checks
102  * whether the watchdog thread has hung or not:
103  */
104 void softlockup_tick(void)
105 {
106         int this_cpu = smp_processor_id();
107         unsigned long touch_timestamp = per_cpu(touch_timestamp, this_cpu);
108         unsigned long print_timestamp;
109         struct pt_regs *regs = get_irq_regs();
110         unsigned long now;
111
112         /* Is detection switched off? */
113         if (!per_cpu(watchdog_task, this_cpu) || softlockup_thresh <= 0) {
114                 /* Be sure we don't false trigger if switched back on */
115                 if (touch_timestamp)
116                         per_cpu(touch_timestamp, this_cpu) = 0;
117                 return;
118         }
119
120         if (touch_timestamp == 0) {
121                 __touch_softlockup_watchdog();
122                 return;
123         }
124
125         print_timestamp = per_cpu(print_timestamp, this_cpu);
126
127         /* report at most once a second */
128         if (print_timestamp == touch_timestamp || did_panic)
129                 return;
130
131         /* do not print during early bootup: */
132         if (unlikely(system_state != SYSTEM_RUNNING)) {
133                 __touch_softlockup_watchdog();
134                 return;
135         }
136
137         now = get_timestamp(this_cpu);
138
139         /*
140          * Wake up the high-prio watchdog task twice per
141          * threshold timespan.
142          */
143         if (now > touch_timestamp + softlockup_thresh/2)
144                 wake_up_process(per_cpu(watchdog_task, this_cpu));
145
146         /* Warn about unreasonable delays: */
147         if (now <= (touch_timestamp + softlockup_thresh))
148                 return;
149
150         per_cpu(print_timestamp, this_cpu) = touch_timestamp;
151
152         spin_lock(&print_lock);
153         printk(KERN_ERR "BUG: soft lockup - CPU#%d stuck for %lus! [%s:%d]\n",
154                         this_cpu, now - touch_timestamp,
155                         current->comm, task_pid_nr(current));
156         print_modules();
157         print_irqtrace_events(current);
158         if (regs)
159                 show_regs(regs);
160         else
161                 dump_stack();
162         spin_unlock(&print_lock);
163
164         if (softlockup_panic)
165                 panic("softlockup: hung tasks");
166 }
167
168 /*
169  * Have a reasonable limit on the number of tasks checked:
170  */
171 unsigned long __read_mostly sysctl_hung_task_check_count = 1024;
172
173 /*
174  * Zero means infinite timeout - no checking done:
175  */
176 unsigned long __read_mostly sysctl_hung_task_timeout_secs = 480;
177
178 unsigned long __read_mostly sysctl_hung_task_warnings = 10;
179
180 /*
181  * Only do the hung-tasks check on one CPU:
182  */
183 static int check_cpu __read_mostly = -1;
184
185 static void check_hung_task(struct task_struct *t, unsigned long now)
186 {
187         unsigned long switch_count = t->nvcsw + t->nivcsw;
188
189         if (t->flags & PF_FROZEN)
190                 return;
191
192         if (switch_count != t->last_switch_count || !t->last_switch_timestamp) {
193                 t->last_switch_count = switch_count;
194                 t->last_switch_timestamp = now;
195                 return;
196         }
197         if ((long)(now - t->last_switch_timestamp) <
198                                         sysctl_hung_task_timeout_secs)
199                 return;
200         if (!sysctl_hung_task_warnings)
201                 return;
202         sysctl_hung_task_warnings--;
203
204         /*
205          * Ok, the task did not get scheduled for more than 2 minutes,
206          * complain:
207          */
208         printk(KERN_ERR "INFO: task %s:%d blocked for more than "
209                         "%ld seconds.\n", t->comm, t->pid,
210                         sysctl_hung_task_timeout_secs);
211         printk(KERN_ERR "\"echo 0 > /proc/sys/kernel/hung_task_timeout_secs\""
212                         " disables this message.\n");
213         sched_show_task(t);
214         __debug_show_held_locks(t);
215
216         t->last_switch_timestamp = now;
217         touch_nmi_watchdog();
218
219         if (softlockup_panic)
220                 panic("softlockup: blocked tasks");
221 }
222
223 /*
224  * Check whether a TASK_UNINTERRUPTIBLE does not get woken up for
225  * a really long time (120 seconds). If that happens, print out
226  * a warning.
227  */
228 static void check_hung_uninterruptible_tasks(int this_cpu)
229 {
230         int max_count = sysctl_hung_task_check_count;
231         unsigned long now = get_timestamp(this_cpu);
232         struct task_struct *g, *t;
233
234         /*
235          * If the system crashed already then all bets are off,
236          * do not report extra hung tasks:
237          */
238         if (test_taint(TAINT_DIE) || did_panic)
239                 return;
240
241         read_lock(&tasklist_lock);
242         do_each_thread(g, t) {
243                 if (!--max_count)
244                         goto unlock;
245                 /* use "==" to skip the TASK_KILLABLE tasks waiting on NFS */
246                 if (t->state == TASK_UNINTERRUPTIBLE)
247                         check_hung_task(t, now);
248         } while_each_thread(g, t);
249  unlock:
250         read_unlock(&tasklist_lock);
251 }
252
253 /*
254  * The watchdog thread - runs every second and touches the timestamp.
255  */
256 static int watchdog(void *__bind_cpu)
257 {
258         struct sched_param param = { .sched_priority = MAX_RT_PRIO-1 };
259         int this_cpu = (long)__bind_cpu;
260
261         sched_setscheduler(current, SCHED_FIFO, &param);
262
263         /* initialize timestamp */
264         __touch_softlockup_watchdog();
265
266         set_current_state(TASK_INTERRUPTIBLE);
267         /*
268          * Run briefly once per second to reset the softlockup timestamp.
269          * If this gets delayed for more than 60 seconds then the
270          * debug-printout triggers in softlockup_tick().
271          */
272         while (!kthread_should_stop()) {
273                 __touch_softlockup_watchdog();
274                 schedule();
275
276                 if (kthread_should_stop())
277                         break;
278
279                 if (this_cpu == check_cpu) {
280                         if (sysctl_hung_task_timeout_secs)
281                                 check_hung_uninterruptible_tasks(this_cpu);
282                 }
283
284                 set_current_state(TASK_INTERRUPTIBLE);
285         }
286         __set_current_state(TASK_RUNNING);
287
288         return 0;
289 }
290
291 /*
292  * Create/destroy watchdog threads as CPUs come and go:
293  */
294 static int __cpuinit
295 cpu_callback(struct notifier_block *nfb, unsigned long action, void *hcpu)
296 {
297         int hotcpu = (unsigned long)hcpu;
298         struct task_struct *p;
299
300         switch (action) {
301         case CPU_UP_PREPARE:
302         case CPU_UP_PREPARE_FROZEN:
303                 BUG_ON(per_cpu(watchdog_task, hotcpu));
304                 p = kthread_create(watchdog, hcpu, "watchdog/%d", hotcpu);
305                 if (IS_ERR(p)) {
306                         printk(KERN_ERR "watchdog for %i failed\n", hotcpu);
307                         return NOTIFY_BAD;
308                 }
309                 per_cpu(touch_timestamp, hotcpu) = 0;
310                 per_cpu(watchdog_task, hotcpu) = p;
311                 kthread_bind(p, hotcpu);
312                 break;
313         case CPU_ONLINE:
314         case CPU_ONLINE_FROZEN:
315                 check_cpu = cpumask_any(cpu_online_mask);
316                 wake_up_process(per_cpu(watchdog_task, hotcpu));
317                 break;
318 #ifdef CONFIG_HOTPLUG_CPU
319         case CPU_DOWN_PREPARE:
320         case CPU_DOWN_PREPARE_FROZEN:
321                 if (hotcpu == check_cpu) {
322                         /* Pick any other online cpu. */
323                         check_cpu = cpumask_any_but(cpu_online_mask, hotcpu);
324                 }
325                 break;
326
327         case CPU_UP_CANCELED:
328         case CPU_UP_CANCELED_FROZEN:
329                 if (!per_cpu(watchdog_task, hotcpu))
330                         break;
331                 /* Unbind so it can run.  Fall thru. */
332                 kthread_bind(per_cpu(watchdog_task, hotcpu),
333                              cpumask_any(cpu_online_mask));
334         case CPU_DEAD:
335         case CPU_DEAD_FROZEN:
336                 p = per_cpu(watchdog_task, hotcpu);
337                 per_cpu(watchdog_task, hotcpu) = NULL;
338                 kthread_stop(p);
339                 break;
340 #endif /* CONFIG_HOTPLUG_CPU */
341         }
342         return NOTIFY_OK;
343 }
344
345 static struct notifier_block __cpuinitdata cpu_nfb = {
346         .notifier_call = cpu_callback
347 };
348
349 static int __initdata nosoftlockup;
350
351 static int __init nosoftlockup_setup(char *str)
352 {
353         nosoftlockup = 1;
354         return 1;
355 }
356 __setup("nosoftlockup", nosoftlockup_setup);
357
358 static int __init spawn_softlockup_task(void)
359 {
360         void *cpu = (void *)(long)smp_processor_id();
361         int err;
362
363         if (nosoftlockup)
364                 return 0;
365
366         err = cpu_callback(&cpu_nfb, CPU_UP_PREPARE, cpu);
367         if (err == NOTIFY_BAD) {
368                 BUG();
369                 return 1;
370         }
371         cpu_callback(&cpu_nfb, CPU_ONLINE, cpu);
372         register_cpu_notifier(&cpu_nfb);
373
374         atomic_notifier_chain_register(&panic_notifier_list, &panic_block);
375
376         return 0;
377 }
378 early_initcall(spawn_softlockup_task);