]> git.proxmox.com Git - mirror_ubuntu-eoan-kernel.git/blame - kernel/softlockup.c
softlockup: fix watchdog task wakeup frequency
[mirror_ubuntu-eoan-kernel.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 15#include <linux/kthread.h>
8d5be7f4 16#include <linux/lockdep.h>
8446f1d3
IM
17#include <linux/notifier.h>
18#include <linux/module.h>
19
43581a10
IM
20#include <asm/irq_regs.h>
21
8446f1d3
IM
22static DEFINE_SPINLOCK(print_lock);
23
6687a97d
IM
24static DEFINE_PER_CPU(unsigned long, touch_timestamp);
25static DEFINE_PER_CPU(unsigned long, print_timestamp);
8446f1d3
IM
26static DEFINE_PER_CPU(struct task_struct *, watchdog_task);
27
90739081 28static int __read_mostly did_panic;
9383d967 29int __read_mostly softlockup_thresh = 60;
6687a97d 30
9c44bc03
IM
31/*
32 * Should we panic (and reboot, if panic_timeout= is set) when a
33 * soft-lockup occurs:
34 */
35unsigned int __read_mostly softlockup_panic =
36 CONFIG_BOOTPARAM_SOFTLOCKUP_PANIC_VALUE;
37
38static int __init softlockup_panic_setup(char *str)
39{
40 softlockup_panic = simple_strtoul(str, NULL, 0);
41
42 return 1;
43}
44__setup("softlockup_panic=", softlockup_panic_setup);
45
6687a97d
IM
46static int
47softlock_panic(struct notifier_block *this, unsigned long event, void *ptr)
8446f1d3
IM
48{
49 did_panic = 1;
50
51 return NOTIFY_DONE;
52}
53
54static struct notifier_block panic_block = {
55 .notifier_call = softlock_panic,
56};
57
966812dc
JF
58/*
59 * Returns seconds, approximately. We don't need nanosecond
60 * resolution, and we don't need to waste time with a big divide when
61 * 2^30ns == 1.074s.
62 */
a3b13c23 63static unsigned long get_timestamp(int this_cpu)
966812dc 64{
82a1fcb9 65 return cpu_clock(this_cpu) >> 30LL; /* 2^30 ~= 10^9 */
966812dc
JF
66}
67
8c2238ea 68static void __touch_softlockup_watchdog(void)
8446f1d3 69{
a3b13c23
IM
70 int this_cpu = raw_smp_processor_id();
71
72 __raw_get_cpu_var(touch_timestamp) = get_timestamp(this_cpu);
8446f1d3 73}
8c2238ea
JW
74
75void touch_softlockup_watchdog(void)
76{
77 __raw_get_cpu_var(touch_timestamp) = 0;
78}
8446f1d3
IM
79EXPORT_SYMBOL(touch_softlockup_watchdog);
80
04c9167f
JF
81void touch_all_softlockup_watchdogs(void)
82{
83 int cpu;
84
85 /* Cause each CPU to re-update its timestamp rather than complain */
86 for_each_online_cpu(cpu)
87 per_cpu(touch_timestamp, cpu) = 0;
88}
89EXPORT_SYMBOL(touch_all_softlockup_watchdogs);
90
8446f1d3
IM
91/*
92 * This callback runs from the timer interrupt, and checks
93 * whether the watchdog thread has hung or not:
94 */
6687a97d 95void softlockup_tick(void)
8446f1d3
IM
96{
97 int this_cpu = smp_processor_id();
6687a97d 98 unsigned long touch_timestamp = per_cpu(touch_timestamp, this_cpu);
966812dc 99 unsigned long print_timestamp;
43581a10 100 struct pt_regs *regs = get_irq_regs();
966812dc 101 unsigned long now;
8446f1d3 102
9383d967
DS
103 /* Is detection switched off? */
104 if (!per_cpu(watchdog_task, this_cpu) || softlockup_thresh <= 0) {
105 /* Be sure we don't false trigger if switched back on */
106 if (touch_timestamp)
107 per_cpu(touch_timestamp, this_cpu) = 0;
108 return;
109 }
110
04c9167f 111 if (touch_timestamp == 0) {
8c2238ea 112 __touch_softlockup_watchdog();
966812dc 113 return;
04c9167f 114 }
966812dc
JF
115
116 print_timestamp = per_cpu(print_timestamp, this_cpu);
117
118 /* report at most once a second */
a115d5ca
IM
119 if ((print_timestamp >= touch_timestamp &&
120 print_timestamp < (touch_timestamp + 1)) ||
9383d967 121 did_panic) {
8446f1d3 122 return;
a115d5ca 123 }
8446f1d3 124
6687a97d
IM
125 /* do not print during early bootup: */
126 if (unlikely(system_state != SYSTEM_RUNNING)) {
8c2238ea 127 __touch_softlockup_watchdog();
8446f1d3 128 return;
6687a97d 129 }
8446f1d3 130
a3b13c23 131 now = get_timestamp(this_cpu);
966812dc 132
dd7a1e56
JW
133 /*
134 * Wake up the high-prio watchdog task twice per
135 * threshold timespan.
136 */
137 if (now > touch_timestamp + softlockup_thresh/2)
ed50d6cb
PZ
138 wake_up_process(per_cpu(watchdog_task, this_cpu));
139
82a1fcb9 140 /* Warn about unreasonable delays: */
c4f3b63f 141 if (now <= (touch_timestamp + softlockup_thresh))
43581a10 142 return;
8446f1d3 143
43581a10
IM
144 per_cpu(print_timestamp, this_cpu) = touch_timestamp;
145
146 spin_lock(&print_lock);
c4f3b63f
RT
147 printk(KERN_ERR "BUG: soft lockup - CPU#%d stuck for %lus! [%s:%d]\n",
148 this_cpu, now - touch_timestamp,
ba25f9dc 149 current->comm, task_pid_nr(current));
688c9175 150 print_modules();
8d5be7f4 151 print_irqtrace_events(current);
43581a10
IM
152 if (regs)
153 show_regs(regs);
154 else
6687a97d 155 dump_stack();
43581a10 156 spin_unlock(&print_lock);
9c44bc03
IM
157
158 if (softlockup_panic)
159 panic("softlockup: hung tasks");
8446f1d3
IM
160}
161
82a1fcb9
IM
162/*
163 * Have a reasonable limit on the number of tasks checked:
164 */
90739081 165unsigned long __read_mostly sysctl_hung_task_check_count = 1024;
82a1fcb9
IM
166
167/*
168 * Zero means infinite timeout - no checking done:
169 */
90739081 170unsigned long __read_mostly sysctl_hung_task_timeout_secs = 120;
82a1fcb9 171
90739081 172unsigned long __read_mostly sysctl_hung_task_warnings = 10;
82a1fcb9
IM
173
174/*
175 * Only do the hung-tasks check on one CPU:
176 */
177static int check_cpu __read_mostly = -1;
178
179static void check_hung_task(struct task_struct *t, unsigned long now)
180{
181 unsigned long switch_count = t->nvcsw + t->nivcsw;
182
183 if (t->flags & PF_FROZEN)
184 return;
185
186 if (switch_count != t->last_switch_count || !t->last_switch_timestamp) {
187 t->last_switch_count = switch_count;
188 t->last_switch_timestamp = now;
189 return;
190 }
191 if ((long)(now - t->last_switch_timestamp) <
192 sysctl_hung_task_timeout_secs)
193 return;
194 if (sysctl_hung_task_warnings < 0)
195 return;
196 sysctl_hung_task_warnings--;
197
198 /*
199 * Ok, the task did not get scheduled for more than 2 minutes,
200 * complain:
201 */
202 printk(KERN_ERR "INFO: task %s:%d blocked for more than "
203 "%ld seconds.\n", t->comm, t->pid,
204 sysctl_hung_task_timeout_secs);
205 printk(KERN_ERR "\"echo 0 > /proc/sys/kernel/hung_task_timeout_secs\""
206 " disables this message.\n");
207 sched_show_task(t);
208 __debug_show_held_locks(t);
209
210 t->last_switch_timestamp = now;
211 touch_nmi_watchdog();
9c44bc03
IM
212
213 if (softlockup_panic)
214 panic("softlockup: blocked tasks");
82a1fcb9
IM
215}
216
217/*
218 * Check whether a TASK_UNINTERRUPTIBLE does not get woken up for
219 * a really long time (120 seconds). If that happens, print out
220 * a warning.
221 */
222static void check_hung_uninterruptible_tasks(int this_cpu)
223{
224 int max_count = sysctl_hung_task_check_count;
225 unsigned long now = get_timestamp(this_cpu);
226 struct task_struct *g, *t;
227
228 /*
229 * If the system crashed already then all bets are off,
230 * do not report extra hung tasks:
231 */
232 if ((tainted & TAINT_DIE) || did_panic)
233 return;
234
235 read_lock(&tasklist_lock);
236 do_each_thread(g, t) {
237 if (!--max_count)
ed50d6cb 238 goto unlock;
82a1fcb9
IM
239 if (t->state & TASK_UNINTERRUPTIBLE)
240 check_hung_task(t, now);
241 } while_each_thread(g, t);
ed50d6cb 242 unlock:
82a1fcb9
IM
243 read_unlock(&tasklist_lock);
244}
245
8446f1d3
IM
246/*
247 * The watchdog thread - runs every second and touches the timestamp.
248 */
a5f2ce3c 249static int watchdog(void *__bind_cpu)
8446f1d3 250{
02fb6149 251 struct sched_param param = { .sched_priority = MAX_RT_PRIO-1 };
82a1fcb9 252 int this_cpu = (long)__bind_cpu;
8446f1d3
IM
253
254 sched_setscheduler(current, SCHED_FIFO, &param);
8446f1d3 255
966812dc 256 /* initialize timestamp */
8c2238ea 257 __touch_softlockup_watchdog();
966812dc 258
7be2a03e 259 set_current_state(TASK_INTERRUPTIBLE);
8446f1d3 260 /*
6687a97d 261 * Run briefly once per second to reset the softlockup timestamp.
82a1fcb9 262 * If this gets delayed for more than 60 seconds then the
6687a97d 263 * debug-printout triggers in softlockup_tick().
8446f1d3
IM
264 */
265 while (!kthread_should_stop()) {
8c2238ea 266 __touch_softlockup_watchdog();
ed50d6cb
PZ
267 schedule();
268
269 if (kthread_should_stop())
270 break;
82a1fcb9 271
7be2a03e
DA
272 if (this_cpu == check_cpu) {
273 if (sysctl_hung_task_timeout_secs)
274 check_hung_uninterruptible_tasks(this_cpu);
275 }
ed50d6cb 276
7be2a03e 277 set_current_state(TASK_INTERRUPTIBLE);
8446f1d3 278 }
7be2a03e 279 __set_current_state(TASK_RUNNING);
8446f1d3
IM
280
281 return 0;
282}
283
284/*
285 * Create/destroy watchdog threads as CPUs come and go:
286 */
8c78f307 287static int __cpuinit
8446f1d3
IM
288cpu_callback(struct notifier_block *nfb, unsigned long action, void *hcpu)
289{
290 int hotcpu = (unsigned long)hcpu;
291 struct task_struct *p;
292
293 switch (action) {
294 case CPU_UP_PREPARE:
8bb78442 295 case CPU_UP_PREPARE_FROZEN:
8446f1d3
IM
296 BUG_ON(per_cpu(watchdog_task, hotcpu));
297 p = kthread_create(watchdog, hcpu, "watchdog/%d", hotcpu);
298 if (IS_ERR(p)) {
a5f2ce3c 299 printk(KERN_ERR "watchdog for %i failed\n", hotcpu);
8446f1d3
IM
300 return NOTIFY_BAD;
301 }
a5f2ce3c
IM
302 per_cpu(touch_timestamp, hotcpu) = 0;
303 per_cpu(watchdog_task, hotcpu) = p;
8446f1d3 304 kthread_bind(p, hotcpu);
a5f2ce3c 305 break;
8446f1d3 306 case CPU_ONLINE:
8bb78442 307 case CPU_ONLINE_FROZEN:
82a1fcb9 308 check_cpu = any_online_cpu(cpu_online_map);
8446f1d3
IM
309 wake_up_process(per_cpu(watchdog_task, hotcpu));
310 break;
311#ifdef CONFIG_HOTPLUG_CPU
82a1fcb9
IM
312 case CPU_DOWN_PREPARE:
313 case CPU_DOWN_PREPARE_FROZEN:
314 if (hotcpu == check_cpu) {
315 cpumask_t temp_cpu_online_map = cpu_online_map;
316
317 cpu_clear(hotcpu, temp_cpu_online_map);
318 check_cpu = any_online_cpu(temp_cpu_online_map);
319 }
320 break;
ed50d6cb
PZ
321
322 case CPU_UP_CANCELED:
323 case CPU_UP_CANCELED_FROZEN:
324 if (!per_cpu(watchdog_task, hotcpu))
325 break;
326 /* Unbind so it can run. Fall thru. */
327 kthread_bind(per_cpu(watchdog_task, hotcpu),
328 any_online_cpu(cpu_online_map));
8446f1d3 329 case CPU_DEAD:
8bb78442 330 case CPU_DEAD_FROZEN:
8446f1d3
IM
331 p = per_cpu(watchdog_task, hotcpu);
332 per_cpu(watchdog_task, hotcpu) = NULL;
333 kthread_stop(p);
334 break;
335#endif /* CONFIG_HOTPLUG_CPU */
a5f2ce3c 336 }
8446f1d3
IM
337 return NOTIFY_OK;
338}
339
8c78f307 340static struct notifier_block __cpuinitdata cpu_nfb = {
8446f1d3
IM
341 .notifier_call = cpu_callback
342};
343
344__init void spawn_softlockup_task(void)
345{
346 void *cpu = (void *)(long)smp_processor_id();
07dccf33 347 int err = cpu_callback(&cpu_nfb, CPU_UP_PREPARE, cpu);
8446f1d3 348
07dccf33 349 BUG_ON(err == NOTIFY_BAD);
8446f1d3
IM
350 cpu_callback(&cpu_nfb, CPU_ONLINE, cpu);
351 register_cpu_notifier(&cpu_nfb);
352
e041c683 353 atomic_notifier_chain_register(&panic_notifier_list, &panic_block);
8446f1d3 354}