softlockup: fix false positives on nohz if CPU is 100% idle for more than 60 seconds
[deliverable/linux.git] / kernel / softlockup.c
CommitLineData
8446f1d3
IM
1/*
2 * Detect Soft Lockups
3 *
6687a97d 4 * started by Ingo Molnar, Copyright (C) 2005, 2006 Red Hat, Inc.
8446f1d3
IM
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 */
8446f1d3
IM
9#include <linux/mm.h>
10#include <linux/cpu.h>
82a1fcb9 11#include <linux/nmi.h>
8446f1d3
IM
12#include <linux/init.h>
13#include <linux/delay.h>
83144186 14#include <linux/freezer.h>
8446f1d3
IM
15#include <linux/kthread.h>
16#include <linux/notifier.h>
17#include <linux/module.h>
18
43581a10
IM
19#include <asm/irq_regs.h>
20
8446f1d3
IM
21static DEFINE_SPINLOCK(print_lock);
22
6687a97d
IM
23static DEFINE_PER_CPU(unsigned long, touch_timestamp);
24static DEFINE_PER_CPU(unsigned long, print_timestamp);
8446f1d3
IM
25static DEFINE_PER_CPU(struct task_struct *, watchdog_task);
26
90739081 27static int __read_mostly did_panic;
9383d967 28int __read_mostly softlockup_thresh = 60;
6687a97d 29
9c44bc03
IM
30/*
31 * Should we panic (and reboot, if panic_timeout= is set) when a
32 * soft-lockup occurs:
33 */
34unsigned int __read_mostly softlockup_panic =
35 CONFIG_BOOTPARAM_SOFTLOCKUP_PANIC_VALUE;
36
37static int __init softlockup_panic_setup(char *str)
38{
39 softlockup_panic = simple_strtoul(str, NULL, 0);
40
41 return 1;
42}
43__setup("softlockup_panic=", softlockup_panic_setup);
44
6687a97d
IM
45static int
46softlock_panic(struct notifier_block *this, unsigned long event, void *ptr)
8446f1d3
IM
47{
48 did_panic = 1;
49
50 return NOTIFY_DONE;
51}
52
53static struct notifier_block panic_block = {
54 .notifier_call = softlock_panic,
55};
56
966812dc
JF
57/*
58 * Returns seconds, approximately. We don't need nanosecond
59 * resolution, and we don't need to waste time with a big divide when
60 * 2^30ns == 1.074s.
61 */
a3b13c23 62static unsigned long get_timestamp(int this_cpu)
966812dc 63{
82a1fcb9 64 return cpu_clock(this_cpu) >> 30LL; /* 2^30 ~= 10^9 */
966812dc
JF
65}
66
8446f1d3
IM
67void touch_softlockup_watchdog(void)
68{
a3b13c23
IM
69 int this_cpu = raw_smp_processor_id();
70
71 __raw_get_cpu_var(touch_timestamp) = get_timestamp(this_cpu);
8446f1d3
IM
72}
73EXPORT_SYMBOL(touch_softlockup_watchdog);
74
04c9167f
JF
75void touch_all_softlockup_watchdogs(void)
76{
77 int cpu;
78
79 /* Cause each CPU to re-update its timestamp rather than complain */
80 for_each_online_cpu(cpu)
81 per_cpu(touch_timestamp, cpu) = 0;
82}
83EXPORT_SYMBOL(touch_all_softlockup_watchdogs);
84
8446f1d3
IM
85/*
86 * This callback runs from the timer interrupt, and checks
87 * whether the watchdog thread has hung or not:
88 */
6687a97d 89void softlockup_tick(void)
8446f1d3
IM
90{
91 int this_cpu = smp_processor_id();
6687a97d 92 unsigned long touch_timestamp = per_cpu(touch_timestamp, this_cpu);
966812dc 93 unsigned long print_timestamp;
43581a10 94 struct pt_regs *regs = get_irq_regs();
966812dc 95 unsigned long now;
8446f1d3 96
9383d967
DS
97 /* Is detection switched off? */
98 if (!per_cpu(watchdog_task, this_cpu) || softlockup_thresh <= 0) {
99 /* Be sure we don't false trigger if switched back on */
100 if (touch_timestamp)
101 per_cpu(touch_timestamp, this_cpu) = 0;
102 return;
103 }
104
04c9167f
JF
105 if (touch_timestamp == 0) {
106 touch_softlockup_watchdog();
966812dc 107 return;
04c9167f 108 }
966812dc
JF
109
110 print_timestamp = per_cpu(print_timestamp, this_cpu);
111
112 /* report at most once a second */
a115d5ca
IM
113 if ((print_timestamp >= touch_timestamp &&
114 print_timestamp < (touch_timestamp + 1)) ||
9383d967 115 did_panic) {
8446f1d3 116 return;
a115d5ca 117 }
8446f1d3 118
6687a97d
IM
119 /* do not print during early bootup: */
120 if (unlikely(system_state != SYSTEM_RUNNING)) {
121 touch_softlockup_watchdog();
8446f1d3 122 return;
6687a97d 123 }
8446f1d3 124
a3b13c23 125 now = get_timestamp(this_cpu);
966812dc 126
ed50d6cb
PZ
127 /* Wake up the high-prio watchdog task every second: */
128 if (now > (touch_timestamp + 1))
129 wake_up_process(per_cpu(watchdog_task, this_cpu));
130
82a1fcb9 131 /* Warn about unreasonable delays: */
c4f3b63f 132 if (now <= (touch_timestamp + softlockup_thresh))
43581a10 133 return;
8446f1d3 134
43581a10
IM
135 per_cpu(print_timestamp, this_cpu) = touch_timestamp;
136
137 spin_lock(&print_lock);
c4f3b63f
RT
138 printk(KERN_ERR "BUG: soft lockup - CPU#%d stuck for %lus! [%s:%d]\n",
139 this_cpu, now - touch_timestamp,
ba25f9dc 140 current->comm, task_pid_nr(current));
43581a10
IM
141 if (regs)
142 show_regs(regs);
143 else
6687a97d 144 dump_stack();
43581a10 145 spin_unlock(&print_lock);
9c44bc03
IM
146
147 if (softlockup_panic)
148 panic("softlockup: hung tasks");
8446f1d3
IM
149}
150
82a1fcb9
IM
151/*
152 * Have a reasonable limit on the number of tasks checked:
153 */
90739081 154unsigned long __read_mostly sysctl_hung_task_check_count = 1024;
82a1fcb9
IM
155
156/*
157 * Zero means infinite timeout - no checking done:
158 */
90739081 159unsigned long __read_mostly sysctl_hung_task_timeout_secs = 120;
82a1fcb9 160
90739081 161unsigned long __read_mostly sysctl_hung_task_warnings = 10;
82a1fcb9
IM
162
163/*
164 * Only do the hung-tasks check on one CPU:
165 */
166static int check_cpu __read_mostly = -1;
167
168static void check_hung_task(struct task_struct *t, unsigned long now)
169{
170 unsigned long switch_count = t->nvcsw + t->nivcsw;
171
172 if (t->flags & PF_FROZEN)
173 return;
174
175 if (switch_count != t->last_switch_count || !t->last_switch_timestamp) {
176 t->last_switch_count = switch_count;
177 t->last_switch_timestamp = now;
178 return;
179 }
180 if ((long)(now - t->last_switch_timestamp) <
181 sysctl_hung_task_timeout_secs)
182 return;
183 if (sysctl_hung_task_warnings < 0)
184 return;
185 sysctl_hung_task_warnings--;
186
187 /*
188 * Ok, the task did not get scheduled for more than 2 minutes,
189 * complain:
190 */
191 printk(KERN_ERR "INFO: task %s:%d blocked for more than "
192 "%ld seconds.\n", t->comm, t->pid,
193 sysctl_hung_task_timeout_secs);
194 printk(KERN_ERR "\"echo 0 > /proc/sys/kernel/hung_task_timeout_secs\""
195 " disables this message.\n");
196 sched_show_task(t);
197 __debug_show_held_locks(t);
198
199 t->last_switch_timestamp = now;
200 touch_nmi_watchdog();
9c44bc03
IM
201
202 if (softlockup_panic)
203 panic("softlockup: blocked tasks");
82a1fcb9
IM
204}
205
206/*
207 * Check whether a TASK_UNINTERRUPTIBLE does not get woken up for
208 * a really long time (120 seconds). If that happens, print out
209 * a warning.
210 */
211static void check_hung_uninterruptible_tasks(int this_cpu)
212{
213 int max_count = sysctl_hung_task_check_count;
214 unsigned long now = get_timestamp(this_cpu);
215 struct task_struct *g, *t;
216
217 /*
218 * If the system crashed already then all bets are off,
219 * do not report extra hung tasks:
220 */
221 if ((tainted & TAINT_DIE) || did_panic)
222 return;
223
224 read_lock(&tasklist_lock);
225 do_each_thread(g, t) {
226 if (!--max_count)
ed50d6cb 227 goto unlock;
82a1fcb9
IM
228 if (t->state & TASK_UNINTERRUPTIBLE)
229 check_hung_task(t, now);
230 } while_each_thread(g, t);
ed50d6cb 231 unlock:
82a1fcb9
IM
232 read_unlock(&tasklist_lock);
233}
234
8446f1d3
IM
235/*
236 * The watchdog thread - runs every second and touches the timestamp.
237 */
a5f2ce3c 238static int watchdog(void *__bind_cpu)
8446f1d3 239{
02fb6149 240 struct sched_param param = { .sched_priority = MAX_RT_PRIO-1 };
82a1fcb9 241 int this_cpu = (long)__bind_cpu;
8446f1d3
IM
242
243 sched_setscheduler(current, SCHED_FIFO, &param);
8446f1d3 244
966812dc
JF
245 /* initialize timestamp */
246 touch_softlockup_watchdog();
247
7be2a03e 248 set_current_state(TASK_INTERRUPTIBLE);
8446f1d3 249 /*
6687a97d 250 * Run briefly once per second to reset the softlockup timestamp.
82a1fcb9 251 * If this gets delayed for more than 60 seconds then the
6687a97d 252 * debug-printout triggers in softlockup_tick().
8446f1d3
IM
253 */
254 while (!kthread_should_stop()) {
8446f1d3 255 touch_softlockup_watchdog();
ed50d6cb
PZ
256 schedule();
257
258 if (kthread_should_stop())
259 break;
82a1fcb9 260
7be2a03e
DA
261 if (this_cpu == check_cpu) {
262 if (sysctl_hung_task_timeout_secs)
263 check_hung_uninterruptible_tasks(this_cpu);
264 }
ed50d6cb 265
7be2a03e 266 set_current_state(TASK_INTERRUPTIBLE);
8446f1d3 267 }
7be2a03e 268 __set_current_state(TASK_RUNNING);
8446f1d3
IM
269
270 return 0;
271}
272
273/*
274 * Create/destroy watchdog threads as CPUs come and go:
275 */
8c78f307 276static int __cpuinit
8446f1d3
IM
277cpu_callback(struct notifier_block *nfb, unsigned long action, void *hcpu)
278{
279 int hotcpu = (unsigned long)hcpu;
280 struct task_struct *p;
281
282 switch (action) {
283 case CPU_UP_PREPARE:
8bb78442 284 case CPU_UP_PREPARE_FROZEN:
8446f1d3
IM
285 BUG_ON(per_cpu(watchdog_task, hotcpu));
286 p = kthread_create(watchdog, hcpu, "watchdog/%d", hotcpu);
287 if (IS_ERR(p)) {
a5f2ce3c 288 printk(KERN_ERR "watchdog for %i failed\n", hotcpu);
8446f1d3
IM
289 return NOTIFY_BAD;
290 }
a5f2ce3c
IM
291 per_cpu(touch_timestamp, hotcpu) = 0;
292 per_cpu(watchdog_task, hotcpu) = p;
8446f1d3 293 kthread_bind(p, hotcpu);
a5f2ce3c 294 break;
8446f1d3 295 case CPU_ONLINE:
8bb78442 296 case CPU_ONLINE_FROZEN:
82a1fcb9 297 check_cpu = any_online_cpu(cpu_online_map);
8446f1d3
IM
298 wake_up_process(per_cpu(watchdog_task, hotcpu));
299 break;
300#ifdef CONFIG_HOTPLUG_CPU
82a1fcb9
IM
301 case CPU_DOWN_PREPARE:
302 case CPU_DOWN_PREPARE_FROZEN:
303 if (hotcpu == check_cpu) {
304 cpumask_t temp_cpu_online_map = cpu_online_map;
305
306 cpu_clear(hotcpu, temp_cpu_online_map);
307 check_cpu = any_online_cpu(temp_cpu_online_map);
308 }
309 break;
ed50d6cb
PZ
310
311 case CPU_UP_CANCELED:
312 case CPU_UP_CANCELED_FROZEN:
313 if (!per_cpu(watchdog_task, hotcpu))
314 break;
315 /* Unbind so it can run. Fall thru. */
316 kthread_bind(per_cpu(watchdog_task, hotcpu),
317 any_online_cpu(cpu_online_map));
8446f1d3 318 case CPU_DEAD:
8bb78442 319 case CPU_DEAD_FROZEN:
8446f1d3
IM
320 p = per_cpu(watchdog_task, hotcpu);
321 per_cpu(watchdog_task, hotcpu) = NULL;
322 kthread_stop(p);
323 break;
324#endif /* CONFIG_HOTPLUG_CPU */
a5f2ce3c 325 }
8446f1d3
IM
326 return NOTIFY_OK;
327}
328
8c78f307 329static struct notifier_block __cpuinitdata cpu_nfb = {
8446f1d3
IM
330 .notifier_call = cpu_callback
331};
332
333__init void spawn_softlockup_task(void)
334{
335 void *cpu = (void *)(long)smp_processor_id();
07dccf33 336 int err = cpu_callback(&cpu_nfb, CPU_UP_PREPARE, cpu);
8446f1d3 337
07dccf33 338 BUG_ON(err == NOTIFY_BAD);
8446f1d3
IM
339 cpu_callback(&cpu_nfb, CPU_ONLINE, cpu);
340 register_cpu_notifier(&cpu_nfb);
341
e041c683 342 atomic_notifier_chain_register(&panic_notifier_list, &panic_block);
8446f1d3 343}
This page took 0.358112 seconds and 5 git commands to generate.