Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 LT |
2 | * NMI watchdog support on APIC systems |
3 | * | |
4 | * Started by Ingo Molnar <mingo@redhat.com> | |
5 | * | |
6 | * Fixes: | |
7 | * Mikael Pettersson : AMD K7 support for local APIC NMI watchdog. | |
8 | * Mikael Pettersson : Power Management for local APIC NMI watchdog. | |
1da177e4 | 9 | * Mikael Pettersson : Pentium 4 support for local APIC NMI watchdog. |
1da177e4 LT |
10 | * Pavel Machek and |
11 | * Mikael Pettersson : PM converted to driver model. Disable/enable API. | |
12 | */ | |
13 | ||
c1eeb2de | 14 | #include <asm/genapic.h> |
3d1ba1da | 15 | |
bb81a09e | 16 | #include <linux/nmi.h> |
1da177e4 | 17 | #include <linux/mm.h> |
1da177e4 | 18 | #include <linux/delay.h> |
1da177e4 | 19 | #include <linux/interrupt.h> |
1da177e4 LT |
20 | #include <linux/module.h> |
21 | #include <linux/sysdev.h> | |
1da177e4 | 22 | #include <linux/sysctl.h> |
3e4ff115 | 23 | #include <linux/percpu.h> |
eddb6fb9 | 24 | #include <linux/kprobes.h> |
bb81a09e | 25 | #include <linux/cpumask.h> |
f8b5035b | 26 | #include <linux/kernel_stat.h> |
1eeb66a1 | 27 | #include <linux/kdebug.h> |
88ff0a47 | 28 | #include <linux/smp.h> |
1da177e4 | 29 | |
35542c5e MR |
30 | #include <asm/i8259.h> |
31 | #include <asm/io_apic.h> | |
1da177e4 | 32 | #include <asm/proto.h> |
6e908947 | 33 | #include <asm/timer.h> |
1da177e4 | 34 | |
553f265f | 35 | #include <asm/mce.h> |
1da177e4 | 36 | |
1164dd00 | 37 | #include <asm/mach_traps.h> |
e32ede19 | 38 | |
29cbc78b AK |
39 | int unknown_nmi_panic; |
40 | int nmi_watchdog_enabled; | |
29cbc78b | 41 | |
1714f9bf | 42 | static cpumask_t backtrace_mask = CPU_MASK_NONE; |
828f0afd | 43 | |
1da177e4 | 44 | /* nmi_active: |
f2802e7f DZ |
45 | * >0: the lapic NMI watchdog is active, but can be disabled |
46 | * <0: the lapic NMI watchdog has not been set up, and cannot | |
1da177e4 | 47 | * be enabled |
f2802e7f | 48 | * 0: the lapic NMI watchdog is disabled, but can be enabled |
1da177e4 | 49 | */ |
f2802e7f | 50 | atomic_t nmi_active = ATOMIC_INIT(0); /* oprofile uses this */ |
88ff0a47 | 51 | EXPORT_SYMBOL(nmi_active); |
1da177e4 | 52 | |
c376d454 | 53 | unsigned int nmi_watchdog = NMI_NONE; |
88ff0a47 HS |
54 | EXPORT_SYMBOL(nmi_watchdog); |
55 | ||
56 | static int panic_on_timeout; | |
1da177e4 | 57 | |
1798bc22 | 58 | static unsigned int nmi_hz = HZ; |
05cb007d | 59 | static DEFINE_PER_CPU(short, wd_enabled); |
88ff0a47 | 60 | static int endflag __initdata; |
92715e28 | 61 | |
fd5cea02 CG |
62 | static inline unsigned int get_nmi_count(int cpu) |
63 | { | |
1b437c8c | 64 | return per_cpu(irq_stat, cpu).__nmi_count; |
fd5cea02 CG |
65 | } |
66 | ||
67 | static inline int mce_in_progress(void) | |
68 | { | |
b8e0418b | 69 | #if defined(CONFIG_X86_64) && defined(CONFIG_X86_MCE) |
1798bc22 CG |
70 | return atomic_read(&mce_entry) > 0; |
71 | #endif | |
fd5cea02 CG |
72 | return 0; |
73 | } | |
74 | ||
75 | /* | |
76 | * Take the local apic timer and PIT/HPET into account. We don't | |
77 | * know which one is active, when we have highres/dyntick on | |
78 | */ | |
79 | static inline unsigned int get_timer_irqs(int cpu) | |
80 | { | |
81 | return per_cpu(irq_stat, cpu).apic_timer_irqs + | |
82 | per_cpu(irq_stat, cpu).irq0_irqs; | |
83 | } | |
1da177e4 | 84 | |
75152114 | 85 | #ifdef CONFIG_SMP |
1798bc22 CG |
86 | /* |
87 | * The performance counters used by NMI_LOCAL_APIC don't trigger when | |
75152114 AK |
88 | * the CPU is idle. To make sure the NMI watchdog really ticks on all |
89 | * CPUs during the test make them busy. | |
90 | */ | |
91 | static __init void nmi_cpu_busy(void *data) | |
1da177e4 | 92 | { |
366c7f55 | 93 | local_irq_enable_in_hardirq(); |
1798bc22 CG |
94 | /* |
95 | * Intentionally don't use cpu_relax here. This is | |
96 | * to make sure that the performance counter really ticks, | |
97 | * even if there is a simulator or similar that catches the | |
98 | * pause instruction. On a real HT machine this is fine because | |
99 | * all other CPUs are busy with "useless" delay loops and don't | |
100 | * care if they get somewhat less cycles. | |
101 | */ | |
92715e28 RT |
102 | while (endflag == 0) |
103 | mb(); | |
1da177e4 | 104 | } |
75152114 | 105 | #endif |
1da177e4 | 106 | |
8bb85190 IM |
107 | static void report_broken_nmi(int cpu, int *prev_nmi_count) |
108 | { | |
109 | printk(KERN_CONT "\n"); | |
110 | ||
111 | printk(KERN_WARNING | |
112 | "WARNING: CPU#%d: NMI appears to be stuck (%d->%d)!\n", | |
113 | cpu, prev_nmi_count[cpu], get_nmi_count(cpu)); | |
114 | ||
115 | printk(KERN_WARNING | |
116 | "Please report this to bugzilla.kernel.org,\n"); | |
117 | printk(KERN_WARNING | |
118 | "and attach the output of the 'dmesg' command.\n"); | |
119 | ||
120 | per_cpu(wd_enabled, cpu) = 0; | |
121 | atomic_dec(&nmi_active); | |
122 | } | |
123 | ||
7d5a78cd AR |
124 | static void __acpi_nmi_disable(void *__unused) |
125 | { | |
126 | apic_write(APIC_LVT0, APIC_DM_NMI | APIC_LVT_MASKED); | |
127 | } | |
128 | ||
416b7218 | 129 | int __init check_nmi_watchdog(void) |
1da177e4 | 130 | { |
29b70081 | 131 | unsigned int *prev_nmi_count; |
1da177e4 LT |
132 | int cpu; |
133 | ||
4de00436 | 134 | if (!nmi_watchdog_active() || !atomic_read(&nmi_active)) |
f2802e7f DZ |
135 | return 0; |
136 | ||
7496b606 | 137 | prev_nmi_count = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); |
416b7218 | 138 | if (!prev_nmi_count) |
35542c5e | 139 | goto error; |
1da177e4 | 140 | |
416b7218 | 141 | printk(KERN_INFO "Testing NMI watchdog ... "); |
ac6b931c | 142 | |
7554c3f0 | 143 | #ifdef CONFIG_SMP |
75152114 | 144 | if (nmi_watchdog == NMI_LOCAL_APIC) |
8691e5a8 | 145 | smp_call_function(nmi_cpu_busy, (void *)&endflag, 0); |
7554c3f0 | 146 | #endif |
1da177e4 | 147 | |
c8912599 | 148 | for_each_possible_cpu(cpu) |
fd5cea02 | 149 | prev_nmi_count[cpu] = get_nmi_count(cpu); |
1da177e4 | 150 | local_irq_enable(); |
1798bc22 | 151 | mdelay((20 * 1000) / nmi_hz); /* wait 20 ticks */ |
1da177e4 | 152 | |
394e3902 | 153 | for_each_online_cpu(cpu) { |
05cb007d | 154 | if (!per_cpu(wd_enabled, cpu)) |
f2802e7f | 155 | continue; |
8bb85190 IM |
156 | if (get_nmi_count(cpu) - prev_nmi_count[cpu] <= 5) |
157 | report_broken_nmi(cpu, prev_nmi_count); | |
1da177e4 | 158 | } |
416b7218 | 159 | endflag = 1; |
f2802e7f | 160 | if (!atomic_read(&nmi_active)) { |
416b7218 | 161 | kfree(prev_nmi_count); |
f2802e7f | 162 | atomic_set(&nmi_active, -1); |
35542c5e | 163 | goto error; |
f2802e7f | 164 | } |
1da177e4 LT |
165 | printk("OK.\n"); |
166 | ||
1798bc22 CG |
167 | /* |
168 | * now that we know it works we can reduce NMI frequency to | |
169 | * something more reasonable; makes a difference in some configs | |
170 | */ | |
05cb007d AK |
171 | if (nmi_watchdog == NMI_LOCAL_APIC) |
172 | nmi_hz = lapic_adjust_nmi_hz(1); | |
1da177e4 | 173 | |
416b7218 | 174 | kfree(prev_nmi_count); |
1da177e4 | 175 | return 0; |
35542c5e | 176 | error: |
7d5a78cd AR |
177 | if (nmi_watchdog == NMI_IO_APIC) { |
178 | if (!timer_through_8259) | |
179 | disable_8259A_irq(0); | |
180 | on_each_cpu(__acpi_nmi_disable, NULL, 1); | |
181 | } | |
182 | ||
5b4d2386 MR |
183 | #ifdef CONFIG_X86_32 |
184 | timer_ack = 0; | |
185 | #endif | |
35542c5e | 186 | return -1; |
1da177e4 LT |
187 | } |
188 | ||
867ab545 | 189 | static int __init setup_nmi_watchdog(char *str) |
1da177e4 | 190 | { |
2b6addad | 191 | unsigned int nmi; |
1da177e4 | 192 | |
d1b946b9 | 193 | if (!strncmp(str, "panic", 5)) { |
1da177e4 LT |
194 | panic_on_timeout = 1; |
195 | str = strchr(str, ','); | |
196 | if (!str) | |
197 | return 1; | |
198 | ++str; | |
199 | } | |
200 | ||
b062f841 CG |
201 | if (!strncmp(str, "lapic", 5)) |
202 | nmi_watchdog = NMI_LOCAL_APIC; | |
203 | else if (!strncmp(str, "ioapic", 6)) | |
204 | nmi_watchdog = NMI_IO_APIC; | |
205 | else { | |
206 | get_option(&str, &nmi); | |
207 | if (nmi >= NMI_INVALID) | |
208 | return 0; | |
209 | nmi_watchdog = nmi; | |
210 | } | |
f2802e7f | 211 | |
1da177e4 LT |
212 | return 1; |
213 | } | |
1da177e4 LT |
214 | __setup("nmi_watchdog=", setup_nmi_watchdog); |
215 | ||
1798bc22 CG |
216 | /* |
217 | * Suspend/resume support | |
218 | */ | |
1da177e4 LT |
219 | #ifdef CONFIG_PM |
220 | ||
221 | static int nmi_pm_active; /* nmi_active before suspend */ | |
222 | ||
829ca9a3 | 223 | static int lapic_nmi_suspend(struct sys_device *dev, pm_message_t state) |
1da177e4 | 224 | { |
4038f901 | 225 | /* only CPU0 goes here, other CPUs should be offline */ |
f2802e7f | 226 | nmi_pm_active = atomic_read(&nmi_active); |
4038f901 SL |
227 | stop_apic_nmi_watchdog(NULL); |
228 | BUG_ON(atomic_read(&nmi_active) != 0); | |
1da177e4 LT |
229 | return 0; |
230 | } | |
231 | ||
232 | static int lapic_nmi_resume(struct sys_device *dev) | |
233 | { | |
4038f901 SL |
234 | /* only CPU0 goes here, other CPUs should be offline */ |
235 | if (nmi_pm_active > 0) { | |
236 | setup_apic_nmi_watchdog(NULL); | |
237 | touch_nmi_watchdog(); | |
238 | } | |
1da177e4 LT |
239 | return 0; |
240 | } | |
241 | ||
242 | static struct sysdev_class nmi_sysclass = { | |
af5ca3f4 | 243 | .name = "lapic_nmi", |
1da177e4 LT |
244 | .resume = lapic_nmi_resume, |
245 | .suspend = lapic_nmi_suspend, | |
246 | }; | |
247 | ||
248 | static struct sys_device device_lapic_nmi = { | |
416b7218 | 249 | .id = 0, |
1da177e4 LT |
250 | .cls = &nmi_sysclass, |
251 | }; | |
252 | ||
253 | static int __init init_lapic_nmi_sysfs(void) | |
254 | { | |
255 | int error; | |
256 | ||
1798bc22 CG |
257 | /* |
258 | * should really be a BUG_ON but b/c this is an | |
f2802e7f DZ |
259 | * init call, it just doesn't work. -dcz |
260 | */ | |
261 | if (nmi_watchdog != NMI_LOCAL_APIC) | |
262 | return 0; | |
263 | ||
416b7218 | 264 | if (atomic_read(&nmi_active) < 0) |
1da177e4 LT |
265 | return 0; |
266 | ||
267 | error = sysdev_class_register(&nmi_sysclass); | |
268 | if (!error) | |
269 | error = sysdev_register(&device_lapic_nmi); | |
270 | return error; | |
271 | } | |
1798bc22 | 272 | |
1da177e4 LT |
273 | /* must come after the local APIC's device_initcall() */ |
274 | late_initcall(init_lapic_nmi_sysfs); | |
275 | ||
276 | #endif /* CONFIG_PM */ | |
277 | ||
416b7218 HS |
278 | static void __acpi_nmi_enable(void *__unused) |
279 | { | |
593f4a78 | 280 | apic_write(APIC_LVT0, APIC_DM_NMI); |
416b7218 HS |
281 | } |
282 | ||
283 | /* | |
284 | * Enable timer based NMIs on all CPUs: | |
285 | */ | |
286 | void acpi_nmi_enable(void) | |
287 | { | |
288 | if (atomic_read(&nmi_active) && nmi_watchdog == NMI_IO_APIC) | |
15c8b6c1 | 289 | on_each_cpu(__acpi_nmi_enable, NULL, 1); |
416b7218 HS |
290 | } |
291 | ||
416b7218 HS |
292 | /* |
293 | * Disable timer based NMIs on all CPUs: | |
294 | */ | |
295 | void acpi_nmi_disable(void) | |
296 | { | |
297 | if (atomic_read(&nmi_active) && nmi_watchdog == NMI_IO_APIC) | |
15c8b6c1 | 298 | on_each_cpu(__acpi_nmi_disable, NULL, 1); |
416b7218 HS |
299 | } |
300 | ||
b3e15bde AR |
301 | /* |
302 | * This function is called as soon the LAPIC NMI watchdog driver has everything | |
303 | * in place and it's ready to check if the NMIs belong to the NMI watchdog | |
304 | */ | |
305 | void cpu_nmi_set_wd_enabled(void) | |
306 | { | |
307 | __get_cpu_var(wd_enabled) = 1; | |
308 | } | |
309 | ||
f2802e7f DZ |
310 | void setup_apic_nmi_watchdog(void *unused) |
311 | { | |
416b7218 | 312 | if (__get_cpu_var(wd_enabled)) |
4038f901 SL |
313 | return; |
314 | ||
315 | /* cheap hack to support suspend/resume */ | |
316 | /* if cpu0 is not active neither should the other cpus */ | |
1798bc22 | 317 | if (smp_processor_id() != 0 && atomic_read(&nmi_active) <= 0) |
4038f901 SL |
318 | return; |
319 | ||
05cb007d AK |
320 | switch (nmi_watchdog) { |
321 | case NMI_LOCAL_APIC: | |
05cb007d AK |
322 | if (lapic_watchdog_init(nmi_hz) < 0) { |
323 | __get_cpu_var(wd_enabled) = 0; | |
75152114 | 324 | return; |
f2802e7f | 325 | } |
05cb007d AK |
326 | /* FALL THROUGH */ |
327 | case NMI_IO_APIC: | |
328 | __get_cpu_var(wd_enabled) = 1; | |
329 | atomic_inc(&nmi_active); | |
f2802e7f | 330 | } |
f2802e7f | 331 | } |
75152114 | 332 | |
4038f901 | 333 | void stop_apic_nmi_watchdog(void *unused) |
f2802e7f DZ |
334 | { |
335 | /* only support LOCAL and IO APICs for now */ | |
4de00436 | 336 | if (!nmi_watchdog_active()) |
1798bc22 | 337 | return; |
05cb007d | 338 | if (__get_cpu_var(wd_enabled) == 0) |
4038f901 | 339 | return; |
05cb007d AK |
340 | if (nmi_watchdog == NMI_LOCAL_APIC) |
341 | lapic_watchdog_stop(); | |
6f290b4e AR |
342 | else |
343 | __acpi_nmi_disable(NULL); | |
05cb007d | 344 | __get_cpu_var(wd_enabled) = 0; |
f2802e7f | 345 | atomic_dec(&nmi_active); |
1da177e4 LT |
346 | } |
347 | ||
348 | /* | |
349 | * the best way to detect whether a CPU has a 'hard lockup' problem | |
350 | * is to check it's local APIC timer IRQ counts. If they are not | |
351 | * changing then that CPU has some problem. | |
352 | * | |
353 | * as these watchdog NMI IRQs are generated on every CPU, we only | |
354 | * have to check the current processor. | |
1da177e4 LT |
355 | * |
356 | * since NMIs don't listen to _any_ locks, we have to be extremely | |
357 | * careful not to rely on unsafe variables. The printk might lock | |
358 | * up though, so we have to break up any console locks first ... | |
1798bc22 | 359 | * [when there will be more tty-related locks, break them up here too!] |
1da177e4 LT |
360 | */ |
361 | ||
75152114 AK |
362 | static DEFINE_PER_CPU(unsigned, last_irq_sum); |
363 | static DEFINE_PER_CPU(local_t, alert_counter); | |
364 | static DEFINE_PER_CPU(int, nmi_touch); | |
1da177e4 | 365 | |
567f3e42 | 366 | void touch_nmi_watchdog(void) |
1da177e4 | 367 | { |
4de00436 | 368 | if (nmi_watchdog_active()) { |
99019e91 | 369 | unsigned cpu; |
1da177e4 | 370 | |
99019e91 | 371 | /* |
f784946d | 372 | * Tell other CPUs to reset their alert counters. We cannot |
99019e91 JB |
373 | * do it ourselves because the alert count increase is not |
374 | * atomic. | |
375 | */ | |
567f3e42 AM |
376 | for_each_present_cpu(cpu) { |
377 | if (per_cpu(nmi_touch, cpu) != 1) | |
378 | per_cpu(nmi_touch, cpu) = 1; | |
379 | } | |
99019e91 | 380 | } |
8446f1d3 | 381 | |
8446f1d3 IM |
382 | /* |
383 | * Tickle the softlockup detector too: | |
384 | */ | |
416b7218 | 385 | touch_softlockup_watchdog(); |
1da177e4 | 386 | } |
416b7218 | 387 | EXPORT_SYMBOL(touch_nmi_watchdog); |
1da177e4 | 388 | |
5deb45e3 SR |
389 | notrace __kprobes int |
390 | nmi_watchdog_tick(struct pt_regs *regs, unsigned reason) | |
1da177e4 | 391 | { |
1da177e4 LT |
392 | /* |
393 | * Since current_thread_info()-> is always on the stack, and we | |
394 | * always switch the stack NMI-atomically, it's safe to use | |
395 | * smp_processor_id(). | |
396 | */ | |
b791ccef | 397 | unsigned int sum; |
75152114 | 398 | int touched = 0; |
bb81a09e | 399 | int cpu = smp_processor_id(); |
05cb007d | 400 | int rc = 0; |
f2802e7f DZ |
401 | |
402 | /* check for other users first */ | |
403 | if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) | |
404 | == NOTIFY_STOP) { | |
3adbbcce | 405 | rc = 1; |
f2802e7f DZ |
406 | touched = 1; |
407 | } | |
1da177e4 | 408 | |
fd5cea02 CG |
409 | sum = get_timer_irqs(cpu); |
410 | ||
75152114 AK |
411 | if (__get_cpu_var(nmi_touch)) { |
412 | __get_cpu_var(nmi_touch) = 0; | |
413 | touched = 1; | |
414 | } | |
f2802e7f | 415 | |
bb81a09e AM |
416 | if (cpu_isset(cpu, backtrace_mask)) { |
417 | static DEFINE_SPINLOCK(lock); /* Serialise the printks */ | |
418 | ||
419 | spin_lock(&lock); | |
88ff0a47 | 420 | printk(KERN_WARNING "NMI backtrace for cpu %d\n", cpu); |
bb81a09e AM |
421 | dump_stack(); |
422 | spin_unlock(&lock); | |
423 | cpu_clear(cpu, backtrace_mask); | |
424 | } | |
425 | ||
fd5cea02 CG |
426 | /* Could check oops_in_progress here too, but it's safer not to */ |
427 | if (mce_in_progress()) | |
553f265f | 428 | touched = 1; |
1da177e4 | 429 | |
f8b5035b | 430 | /* if the none of the timers isn't firing, this cpu isn't doing much */ |
75152114 | 431 | if (!touched && __get_cpu_var(last_irq_sum) == sum) { |
1da177e4 LT |
432 | /* |
433 | * Ayiee, looks like this CPU is stuck ... | |
434 | * wait a few IRQs (5 seconds) before doing the oops ... | |
435 | */ | |
75152114 | 436 | local_inc(&__get_cpu_var(alert_counter)); |
1798bc22 | 437 | if (local_read(&__get_cpu_var(alert_counter)) == 5 * nmi_hz) |
748f2edb GA |
438 | /* |
439 | * die_nmi will return ONLY if NOTIFY_STOP happens.. | |
440 | */ | |
ddca03c9 | 441 | die_nmi("BUG: NMI Watchdog detected LOCKUP", |
d1b946b9 | 442 | regs, panic_on_timeout); |
1da177e4 | 443 | } else { |
75152114 AK |
444 | __get_cpu_var(last_irq_sum) = sum; |
445 | local_set(&__get_cpu_var(alert_counter), 0); | |
1da177e4 | 446 | } |
f2802e7f DZ |
447 | |
448 | /* see if the nmi watchdog went off */ | |
05cb007d AK |
449 | if (!__get_cpu_var(wd_enabled)) |
450 | return rc; | |
451 | switch (nmi_watchdog) { | |
452 | case NMI_LOCAL_APIC: | |
453 | rc |= lapic_wd_event(nmi_hz); | |
454 | break; | |
455 | case NMI_IO_APIC: | |
1798bc22 CG |
456 | /* |
457 | * don't know how to accurately check for this. | |
05cb007d AK |
458 | * just assume it was a watchdog timer interrupt |
459 | * This matches the old behaviour. | |
460 | */ | |
461 | rc = 1; | |
462 | break; | |
75152114 | 463 | } |
3adbbcce | 464 | return rc; |
1da177e4 LT |
465 | } |
466 | ||
1da177e4 LT |
467 | #ifdef CONFIG_SYSCTL |
468 | ||
6f290b4e AR |
469 | static void enable_ioapic_nmi_watchdog_single(void *unused) |
470 | { | |
471 | __get_cpu_var(wd_enabled) = 1; | |
472 | atomic_inc(&nmi_active); | |
473 | __acpi_nmi_enable(NULL); | |
474 | } | |
475 | ||
476 | static void enable_ioapic_nmi_watchdog(void) | |
477 | { | |
478 | on_each_cpu(enable_ioapic_nmi_watchdog_single, NULL, 1); | |
479 | touch_nmi_watchdog(); | |
480 | } | |
481 | ||
482 | static void disable_ioapic_nmi_watchdog(void) | |
483 | { | |
484 | on_each_cpu(stop_apic_nmi_watchdog, NULL, 1); | |
485 | } | |
486 | ||
e3a61b0a | 487 | static int __init setup_unknown_nmi_panic(char *str) |
8f4e956b | 488 | { |
e3a61b0a SA |
489 | unknown_nmi_panic = 1; |
490 | return 1; | |
8f4e956b | 491 | } |
e3a61b0a | 492 | __setup("unknown_nmi_panic", setup_unknown_nmi_panic); |
1da177e4 LT |
493 | |
494 | static int unknown_nmi_panic_callback(struct pt_regs *regs, int cpu) | |
495 | { | |
496 | unsigned char reason = get_nmi_reason(); | |
497 | char buf[64]; | |
498 | ||
2fbe7b25 | 499 | sprintf(buf, "NMI received for unknown reason %02x\n", reason); |
6c8decdf | 500 | die_nmi(buf, regs, 1); /* Always panic here */ |
1da177e4 LT |
501 | return 0; |
502 | } | |
503 | ||
407984f1 DZ |
504 | /* |
505 | * proc handler for /proc/sys/kernel/nmi | |
506 | */ | |
507 | int proc_nmi_enabled(struct ctl_table *table, int write, struct file *file, | |
508 | void __user *buffer, size_t *length, loff_t *ppos) | |
509 | { | |
510 | int old_state; | |
511 | ||
512 | nmi_watchdog_enabled = (atomic_read(&nmi_active) > 0) ? 1 : 0; | |
513 | old_state = nmi_watchdog_enabled; | |
514 | proc_dointvec(table, write, file, buffer, length, ppos); | |
515 | if (!!old_state == !!nmi_watchdog_enabled) | |
516 | return 0; | |
517 | ||
4de00436 | 518 | if (atomic_read(&nmi_active) < 0 || !nmi_watchdog_active()) { |
1798bc22 CG |
519 | printk(KERN_WARNING |
520 | "NMI watchdog is permanently disabled\n"); | |
e33e89ab | 521 | return -EIO; |
407984f1 DZ |
522 | } |
523 | ||
e33e89ab | 524 | if (nmi_watchdog == NMI_LOCAL_APIC) { |
407984f1 DZ |
525 | if (nmi_watchdog_enabled) |
526 | enable_lapic_nmi_watchdog(); | |
527 | else | |
528 | disable_lapic_nmi_watchdog(); | |
6f290b4e AR |
529 | } else if (nmi_watchdog == NMI_IO_APIC) { |
530 | if (nmi_watchdog_enabled) | |
531 | enable_ioapic_nmi_watchdog(); | |
532 | else | |
533 | disable_ioapic_nmi_watchdog(); | |
407984f1 | 534 | } else { |
1798bc22 | 535 | printk(KERN_WARNING |
407984f1 DZ |
536 | "NMI watchdog doesn't know what hardware to touch\n"); |
537 | return -EIO; | |
538 | } | |
539 | return 0; | |
540 | } | |
541 | ||
1798bc22 | 542 | #endif /* CONFIG_SYSCTL */ |
1da177e4 | 543 | |
a062bae9 LZ |
544 | int do_nmi_callback(struct pt_regs *regs, int cpu) |
545 | { | |
546 | #ifdef CONFIG_SYSCTL | |
547 | if (unknown_nmi_panic) | |
548 | return unknown_nmi_panic_callback(regs, cpu); | |
549 | #endif | |
550 | return 0; | |
551 | } | |
552 | ||
bb81a09e AM |
553 | void __trigger_all_cpu_backtrace(void) |
554 | { | |
555 | int i; | |
556 | ||
557 | backtrace_mask = cpu_online_map; | |
558 | /* Wait for up to 10 seconds for all CPUs to do the backtrace */ | |
559 | for (i = 0; i < 10 * 1000; i++) { | |
560 | if (cpus_empty(backtrace_mask)) | |
561 | break; | |
562 | mdelay(1); | |
563 | } | |
564 | } |