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 /* Wake up the high-prio watchdog task every second: */
105 if (now > (touch_timestamp + 1))
106 wake_up_process(per_cpu(watchdog_task, this_cpu));
107
108 /* Warn about unreasonable delays: */
109 if (now <= (touch_timestamp + softlockup_thresh))
110 return;
111
112 per_cpu(print_timestamp, this_cpu) = touch_timestamp;
113
114 spin_lock(&print_lock);
115 printk(KERN_ERR "BUG: soft lockup - CPU#%d stuck for %lus! [%s:%d]\n",
116 this_cpu, now - touch_timestamp,
117 current->comm, task_pid_nr(current));
118 if (regs)
119 show_regs(regs);
120 else
121 dump_stack();
122 spin_unlock(&print_lock);
123 }
124
125 /*
126 * Have a reasonable limit on the number of tasks checked:
127 */
128 unsigned long __read_mostly sysctl_hung_task_check_count = 1024;
129
130 /*
131 * Zero means infinite timeout - no checking done:
132 */
133 unsigned long __read_mostly sysctl_hung_task_timeout_secs = 120;
134
135 unsigned long __read_mostly sysctl_hung_task_warnings = 10;
136
137 /*
138 * Only do the hung-tasks check on one CPU:
139 */
140 static int check_cpu __read_mostly = -1;
141
142 static void check_hung_task(struct task_struct *t, unsigned long now)
143 {
144 unsigned long switch_count = t->nvcsw + t->nivcsw;
145
146 if (t->flags & PF_FROZEN)
147 return;
148
149 if (switch_count != t->last_switch_count || !t->last_switch_timestamp) {
150 t->last_switch_count = switch_count;
151 t->last_switch_timestamp = now;
152 return;
153 }
154 if ((long)(now - t->last_switch_timestamp) <
155 sysctl_hung_task_timeout_secs)
156 return;
157 if (sysctl_hung_task_warnings < 0)
158 return;
159 sysctl_hung_task_warnings--;
160
161 /*
162 * Ok, the task did not get scheduled for more than 2 minutes,
163 * complain:
164 */
165 printk(KERN_ERR "INFO: task %s:%d blocked for more than "
166 "%ld seconds.\n", t->comm, t->pid,
167 sysctl_hung_task_timeout_secs);
168 printk(KERN_ERR "\"echo 0 > /proc/sys/kernel/hung_task_timeout_secs\""
169 " disables this message.\n");
170 sched_show_task(t);
171 __debug_show_held_locks(t);
172
173 t->last_switch_timestamp = now;
174 touch_nmi_watchdog();
175 }
176
177 /*
178 * Check whether a TASK_UNINTERRUPTIBLE does not get woken up for
179 * a really long time (120 seconds). If that happens, print out
180 * a warning.
181 */
182 static void check_hung_uninterruptible_tasks(int this_cpu)
183 {
184 int max_count = sysctl_hung_task_check_count;
185 unsigned long now = get_timestamp(this_cpu);
186 struct task_struct *g, *t;
187
188 /*
189 * If the system crashed already then all bets are off,
190 * do not report extra hung tasks:
191 */
192 if ((tainted & TAINT_DIE) || did_panic)
193 return;
194
195 read_lock(&tasklist_lock);
196 do_each_thread(g, t) {
197 if (!--max_count)
198 goto unlock;
199 if (t->state & TASK_UNINTERRUPTIBLE)
200 check_hung_task(t, now);
201 } while_each_thread(g, t);
202 unlock:
203 read_unlock(&tasklist_lock);
204 }
205
206 /*
207 * The watchdog thread - runs every second and touches the timestamp.
208 */
209 static int watchdog(void *__bind_cpu)
210 {
211 struct sched_param param = { .sched_priority = MAX_RT_PRIO-1 };
212 int this_cpu = (long)__bind_cpu;
213
214 sched_setscheduler(current, SCHED_FIFO, ¶m);
215
216 /* initialize timestamp */
217 touch_softlockup_watchdog();
218
219 set_current_state(TASK_INTERRUPTIBLE);
220 /*
221 * Run briefly once per second to reset the softlockup timestamp.
222 * If this gets delayed for more than 60 seconds then the
223 * debug-printout triggers in softlockup_tick().
224 */
225 while (!kthread_should_stop()) {
226 touch_softlockup_watchdog();
227 schedule();
228
229 if (kthread_should_stop())
230 break;
231
232 if (this_cpu == check_cpu) {
233 if (sysctl_hung_task_timeout_secs)
234 check_hung_uninterruptible_tasks(this_cpu);
235 }
236
237 set_current_state(TASK_INTERRUPTIBLE);
238 }
239 __set_current_state(TASK_RUNNING);
240
241 return 0;
242 }
243
244 /*
245 * Create/destroy watchdog threads as CPUs come and go:
246 */
247 static int __cpuinit
248 cpu_callback(struct notifier_block *nfb, unsigned long action, void *hcpu)
249 {
250 int hotcpu = (unsigned long)hcpu;
251 struct task_struct *p;
252
253 switch (action) {
254 case CPU_UP_PREPARE:
255 case CPU_UP_PREPARE_FROZEN:
256 BUG_ON(per_cpu(watchdog_task, hotcpu));
257 p = kthread_create(watchdog, hcpu, "watchdog/%d", hotcpu);
258 if (IS_ERR(p)) {
259 printk(KERN_ERR "watchdog for %i failed\n", hotcpu);
260 return NOTIFY_BAD;
261 }
262 per_cpu(touch_timestamp, hotcpu) = 0;
263 per_cpu(watchdog_task, hotcpu) = p;
264 kthread_bind(p, hotcpu);
265 break;
266 case CPU_ONLINE:
267 case CPU_ONLINE_FROZEN:
268 check_cpu = any_online_cpu(cpu_online_map);
269 wake_up_process(per_cpu(watchdog_task, hotcpu));
270 break;
271 #ifdef CONFIG_HOTPLUG_CPU
272 case CPU_DOWN_PREPARE:
273 case CPU_DOWN_PREPARE_FROZEN:
274 if (hotcpu == check_cpu) {
275 cpumask_t temp_cpu_online_map = cpu_online_map;
276
277 cpu_clear(hotcpu, temp_cpu_online_map);
278 check_cpu = any_online_cpu(temp_cpu_online_map);
279 }
280 break;
281
282 case CPU_UP_CANCELED:
283 case CPU_UP_CANCELED_FROZEN:
284 if (!per_cpu(watchdog_task, hotcpu))
285 break;
286 /* Unbind so it can run. Fall thru. */
287 kthread_bind(per_cpu(watchdog_task, hotcpu),
288 any_online_cpu(cpu_online_map));
289 case CPU_DEAD:
290 case CPU_DEAD_FROZEN:
291 p = per_cpu(watchdog_task, hotcpu);
292 per_cpu(watchdog_task, hotcpu) = NULL;
293 kthread_stop(p);
294 break;
295 #endif /* CONFIG_HOTPLUG_CPU */
296 }
297 return NOTIFY_OK;
298 }
299
300 static struct notifier_block __cpuinitdata cpu_nfb = {
301 .notifier_call = cpu_callback
302 };
303
304 __init void spawn_softlockup_task(void)
305 {
306 void *cpu = (void *)(long)smp_processor_id();
307 int err = cpu_callback(&cpu_nfb, CPU_UP_PREPARE, cpu);
308
309 BUG_ON(err == NOTIFY_BAD);
310 cpu_callback(&cpu_nfb, CPU_ONLINE, cpu);
311 register_cpu_notifier(&cpu_nfb);
312
313 atomic_notifier_chain_register(&panic_notifier_list, &panic_block);
314 }
315
|
This page was automatically generated by the
LXR engine.
|