softlockup: fix signedness
[linux-2.6.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/notifier.h>
17 #include <linux/module.h>
18
19 #include <asm/irq_regs.h>
20
21 static DEFINE_SPINLOCK(print_lock);
22
23 static DEFINE_PER_CPU(unsigned long, touch_timestamp);
24 static DEFINE_PER_CPU(unsigned long, print_timestamp);
25 static DEFINE_PER_CPU(struct task_struct *, watchdog_task);
26
27 static int __read_mostly did_panic;
28 unsigned long __read_mostly softlockup_thresh = 60;
29
30 static int
31 softlock_panic(struct notifier_block *this, unsigned long event, void *ptr)
32 {
33         did_panic = 1;
34
35         return NOTIFY_DONE;
36 }
37
38 static struct notifier_block panic_block = {
39         .notifier_call = softlock_panic,
40 };
41
42 /*
43  * Returns seconds, approximately.  We don't need nanosecond
44  * resolution, and we don't need to waste time with a big divide when
45  * 2^30ns == 1.074s.
46  */
47 static unsigned long get_timestamp(int this_cpu)
48 {
49         return cpu_clock(this_cpu) >> 30LL;  /* 2^30 ~= 10^9 */
50 }
51
52 void touch_softlockup_watchdog(void)
53 {
54         int this_cpu = raw_smp_processor_id();
55
56         __raw_get_cpu_var(touch_timestamp) = get_timestamp(this_cpu);
57 }
58 EXPORT_SYMBOL(touch_softlockup_watchdog);
59
60 void touch_all_softlockup_watchdogs(void)
61 {
62         int cpu;
63
64         /* Cause each CPU to re-update its timestamp rather than complain */
65         for_each_online_cpu(cpu)
66                 per_cpu(touch_timestamp, cpu) = 0;
67 }
68 EXPORT_SYMBOL(touch_all_softlockup_watchdogs);
69
70 /*
71  * This callback runs from the timer interrupt, and checks
72  * whether the watchdog thread has hung or not:
73  */
74 void softlockup_tick(void)
75 {
76         int this_cpu = smp_processor_id();
77         unsigned long touch_timestamp = per_cpu(touch_timestamp, this_cpu);
78         unsigned long print_timestamp;
79         struct pt_regs *regs = get_irq_regs();
80         unsigned long now;
81
82         if (touch_timestamp == 0) {
83                 touch_softlockup_watchdog();
84                 return;
85         }
86
87         print_timestamp = per_cpu(print_timestamp, this_cpu);
88
89         /* report at most once a second */
90         if ((print_timestamp >= touch_timestamp &&
91                         print_timestamp < (touch_timestamp + 1)) ||
92                         did_panic || !per_cpu(watchdog_task, this_cpu)) {
93                 return;
94         }
95
96         /* do not print during early bootup: */
97         if (unlikely(system_state != SYSTEM_RUNNING)) {
98                 touch_softlockup_watchdog();
99                 return;
100         }
101
102         now = get_timestamp(this_cpu);
103
104         /* Warn about unreasonable delays: */
105         if (now <= (touch_timestamp + softlockup_thresh))
106                 return;
107
108         per_cpu(print_timestamp, this_cpu) = touch_timestamp;
109
110         spin_lock(&print_lock);
111         printk(KERN_ERR "BUG: soft lockup - CPU#%d stuck for %lus! [%s:%d]\n",
112                         this_cpu, now - touch_timestamp,
113                         current->comm, task_pid_nr(current));
114         if (regs)
115                 show_regs(regs);
116         else
117                 dump_stack();
118         spin_unlock(&print_lock);
119 }
120
121 /*
122  * Have a reasonable limit on the number of tasks checked:
123  */
124 unsigned long __read_mostly sysctl_hung_task_check_count = 1024;
125
126 /*
127  * Zero means infinite timeout - no checking done:
128  */
129 unsigned long __read_mostly sysctl_hung_task_timeout_secs = 120;
130
131 unsigned long __read_mostly sysctl_hung_task_warnings = 10;
132
133 /*
134  * Only do the hung-tasks check on one CPU:
135  */
136 static int check_cpu __read_mostly = -1;
137
138 static void check_hung_task(struct task_struct *t, unsigned long now)
139 {
140         unsigned long switch_count = t->nvcsw + t->nivcsw;
141
142         if (t->flags & PF_FROZEN)
143                 return;
144
145         if (switch_count != t->last_switch_count || !t->last_switch_timestamp) {
146                 t->last_switch_count = switch_count;
147                 t->last_switch_timestamp = now;
148                 return;
149         }
150         if ((long)(now - t->last_switch_timestamp) <
151                                         sysctl_hung_task_timeout_secs)
152                 return;
153         if (sysctl_hung_task_warnings < 0)
154                 return;
155         sysctl_hung_task_warnings--;
156
157         /*
158          * Ok, the task did not get scheduled for more than 2 minutes,
159          * complain:
160          */
161         printk(KERN_ERR "INFO: task %s:%d blocked for more than "
162                         "%ld seconds.\n", t->comm, t->pid,
163                         sysctl_hung_task_timeout_secs);
164         printk(KERN_ERR "\"echo 0 > /proc/sys/kernel/hung_task_timeout_secs\""
165                         " disables this message.\n");
166         sched_show_task(t);
167         __debug_show_held_locks(t);
168
169         t->last_switch_timestamp = now;
170         touch_nmi_watchdog();
171 }
172
173 /*
174  * Check whether a TASK_UNINTERRUPTIBLE does not get woken up for
175  * a really long time (120 seconds). If that happens, print out
176  * a warning.
177  */
178 static void check_hung_uninterruptible_tasks(int this_cpu)
179 {
180         int max_count = sysctl_hung_task_check_count;
181         unsigned long now = get_timestamp(this_cpu);
182         struct task_struct *g, *t;
183
184         /*
185          * If the system crashed already then all bets are off,
186          * do not report extra hung tasks:
187          */
188         if ((tainted & TAINT_DIE) || did_panic)
189                 return;
190
191         read_lock(&tasklist_lock);
192         do_each_thread(g, t) {
193                 if (!--max_count)
194                         break;
195                 if (t->state & TASK_UNINTERRUPTIBLE)
196                         check_hung_task(t, now);
197         } while_each_thread(g, t);
198
199         read_unlock(&tasklist_lock);
200 }
201
202 /*
203  * The watchdog thread - runs every second and touches the timestamp.
204  */
205 static int watchdog(void *__bind_cpu)
206 {
207         struct sched_param param = { .sched_priority = MAX_RT_PRIO-1 };
208         int this_cpu = (long)__bind_cpu;
209
210         sched_setscheduler(current, SCHED_FIFO, &param);
211
212         /* initialize timestamp */
213         touch_softlockup_watchdog();
214
215         /*
216          * Run briefly once per second to reset the softlockup timestamp.
217          * If this gets delayed for more than 60 seconds then the
218          * debug-printout triggers in softlockup_tick().
219          */
220         while (!kthread_should_stop()) {
221                 touch_softlockup_watchdog();
222                 msleep_interruptible(10000);
223
224                 if (this_cpu != check_cpu)
225                         continue;
226
227                 if (sysctl_hung_task_timeout_secs)
228                         check_hung_uninterruptible_tasks(this_cpu);
229         }
230
231         return 0;
232 }
233
234 /*
235  * Create/destroy watchdog threads as CPUs come and go:
236  */
237 static int __cpuinit
238 cpu_callback(struct notifier_block *nfb, unsigned long action, void *hcpu)
239 {
240         int hotcpu = (unsigned long)hcpu;
241         struct task_struct *p;
242
243         switch (action) {
244         case CPU_UP_PREPARE:
245         case CPU_UP_PREPARE_FROZEN:
246                 BUG_ON(per_cpu(watchdog_task, hotcpu));
247                 p = kthread_create(watchdog, hcpu, "watchdog/%d", hotcpu);
248                 if (IS_ERR(p)) {
249                         printk(KERN_ERR "watchdog for %i failed\n", hotcpu);
250                         return NOTIFY_BAD;
251                 }
252                 per_cpu(touch_timestamp, hotcpu) = 0;
253                 per_cpu(watchdog_task, hotcpu) = p;
254                 kthread_bind(p, hotcpu);
255                 break;
256         case CPU_ONLINE:
257         case CPU_ONLINE_FROZEN:
258                 check_cpu = any_online_cpu(cpu_online_map);
259                 wake_up_process(per_cpu(watchdog_task, hotcpu));
260                 break;
261 #ifdef CONFIG_HOTPLUG_CPU
262         case CPU_UP_CANCELED:
263         case CPU_UP_CANCELED_FROZEN:
264                 if (!per_cpu(watchdog_task, hotcpu))
265                         break;
266                 /* Unbind so it can run.  Fall thru. */
267                 kthread_bind(per_cpu(watchdog_task, hotcpu),
268                              any_online_cpu(cpu_online_map));
269         case CPU_DOWN_PREPARE:
270         case CPU_DOWN_PREPARE_FROZEN:
271                 if (hotcpu == check_cpu) {
272                         cpumask_t temp_cpu_online_map = cpu_online_map;
273
274                         cpu_clear(hotcpu, temp_cpu_online_map);
275                         check_cpu = any_online_cpu(temp_cpu_online_map);
276                 }
277                 break;
278         case CPU_DEAD:
279         case CPU_DEAD_FROZEN:
280                 p = per_cpu(watchdog_task, hotcpu);
281                 per_cpu(watchdog_task, hotcpu) = NULL;
282                 kthread_stop(p);
283                 break;
284 #endif /* CONFIG_HOTPLUG_CPU */
285         }
286         return NOTIFY_OK;
287 }
288
289 static struct notifier_block __cpuinitdata cpu_nfb = {
290         .notifier_call = cpu_callback
291 };
292
293 __init void spawn_softlockup_task(void)
294 {
295         void *cpu = (void *)(long)smp_processor_id();
296         int err = cpu_callback(&cpu_nfb, CPU_UP_PREPARE, cpu);
297
298         BUG_ON(err == NOTIFY_BAD);
299         cpu_callback(&cpu_nfb, CPU_ONLINE, cpu);
300         register_cpu_notifier(&cpu_nfb);
301
302         atomic_notifier_chain_register(&panic_notifier_list, &panic_block);
303 }