History

[PATCH] clockevents: Fix suspend/resume to disk hangs

I finally found a dual core box, which survives suspend/resume without crashing in the middle of nowhere. Sigh, I never figured out from the code and the bug reports what's going on. The observed hangs are caused by a stale state transition of the clock event devices, which keeps the RCU synchronization away from completion, when the non boot CPU is brought back up. The suspend/resume in oneshot mode needs the similar care as the periodic mode during suspend to RAM. My assumption that the state transitions during the different shutdown/bringups of s2disk would go through the periodic boot phase and then switch over to highres resp. nohz mode were simply wrong. Add the appropriate suspend / resume handling for the non periodic modes.

Thomas Gleixner committed 19 years ago

Commit cd05a1f1 parent 24c4ac04 files · +51 −12Browse files

Showing the changes to kernel (4 of 4 files). Show all

Unified · Split

4 files
  1. kernel/time/tick-broadcast.c +22 −5
  2. kernel/time/tick-common.c +7 −6
  3. kernel/time/tick-internal.h +10 −1
  4. kernel/time/tick-oneshot.c +12 −0
kernel/time/tick-broadcast.c+22−5
@@ -307,12 +307,19 @@ int tick_resume_broadcast(void) · 306 unmodified lines
307307 spin_lock_irqsave(&tick_broadcast_lock, flags);
308308
309309 bc = tick_broadcast_device.evtdev;
310− if (bc) {
311− if (tick_broadcast_device.mode == TICKDEV_MODE_PERIODIC &&
312− !cpus_empty(tick_broadcast_mask))
313− tick_broadcast_start_periodic(bc);
314310
315− broadcast = cpu_isset(smp_processor_id(), tick_broadcast_mask);
311+ if (bc) {
312+ switch (tick_broadcast_device.mode) {
313+ case TICKDEV_MODE_PERIODIC:
314+ if(!cpus_empty(tick_broadcast_mask))
315+ tick_broadcast_start_periodic(bc);
316+ broadcast = cpu_isset(smp_processor_id(),
317+ tick_broadcast_mask);
318+ break;
319+ case TICKDEV_MODE_ONESHOT:
320+ broadcast = tick_resume_broadcast_oneshot(bc);
321+ break;
322+ }
316323 }
317324 spin_unlock_irqrestore(&tick_broadcast_lock, flags);
318325
@@ -347,6 +354,16 @@ static int tick_broadcast_set_event(ktime_t expires, int force) · 28 unmodified lines
347354 }
348355 }
349356
357+int tick_resume_broadcast_oneshot(struct clock_event_device *bc)
358+{
359+ clockevents_set_mode(bc, CLOCK_EVT_MODE_ONESHOT);
360+
361+ if(!cpus_empty(tick_broadcast_oneshot_mask))
362+ tick_broadcast_set_event(ktime_get(), 1);
363+
364+ return cpu_isset(smp_processor_id(), tick_broadcast_oneshot_mask);
365+}
366+
350367 /*
351368 * Reprogram the broadcast device:
352369 *
kernel/time/tick-common.c+7−6
@@ -298,18 +298,17 @@ static void tick_shutdown(unsigned int *cpup) · 297 unmodified lines
298298 spin_unlock_irqrestore(&tick_device_lock, flags);
299299 }
300300
301−static void tick_suspend_periodic(void)
301+static void tick_suspend(void)
302302 {
303303 struct tick_device *td = &__get_cpu_var(tick_cpu_device);
304304 unsigned long flags;
305305
306306 spin_lock_irqsave(&tick_device_lock, flags);
307− if (td->mode == TICKDEV_MODE_PERIODIC)
308− clockevents_set_mode(td->evtdev, CLOCK_EVT_MODE_SHUTDOWN);
307+ clockevents_set_mode(td->evtdev, CLOCK_EVT_MODE_SHUTDOWN);
309308 spin_unlock_irqrestore(&tick_device_lock, flags);
310309 }
311310
312−static void tick_resume_periodic(void)
311+static void tick_resume(void)
313312 {
314313 struct tick_device *td = &__get_cpu_var(tick_cpu_device);
315314 unsigned long flags;
@@ -317,6 +316,8 @@ static void tick_resume_periodic(void) · 1 unmodified line
317316 spin_lock_irqsave(&tick_device_lock, flags);
318317 if (td->mode == TICKDEV_MODE_PERIODIC)
319318 tick_setup_periodic(td->evtdev, 0);
319+ else
320+ tick_resume_oneshot();
320321 spin_unlock_irqrestore(&tick_device_lock, flags);
321322 }
322323
@@ -348,13 +349,13 @@ static int tick_notify(struct notifier_block *nb, unsigned long reason, · 25 unmodified lines
348349 break;
349350
350351 case CLOCK_EVT_NOTIFY_SUSPEND:
351− tick_suspend_periodic();
352+ tick_suspend();
352353 tick_suspend_broadcast();
353354 break;
354355
355356 case CLOCK_EVT_NOTIFY_RESUME:
356357 if (!tick_resume_broadcast())
357− tick_resume_periodic();
358+ tick_resume();
358359 break;
359360
360361 default:
kernel/time/tick-internal.h+10−1
@@ -19,12 +19,13 @@ extern void tick_setup_oneshot(struct clock_event_device *newdev, · 18 unmodified lines
1919 extern int tick_program_event(ktime_t expires, int force);
2020 extern void tick_oneshot_notify(void);
2121 extern int tick_switch_to_oneshot(void (*handler)(struct clock_event_device *));
22−
22+extern void tick_resume_oneshot(void);
2323 # ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
2424 extern void tick_broadcast_setup_oneshot(struct clock_event_device *bc);
2525 extern void tick_broadcast_oneshot_control(unsigned long reason);
2626 extern void tick_broadcast_switch_to_oneshot(void);
2727 extern void tick_shutdown_broadcast_oneshot(unsigned int *cpup);
28+extern int tick_resume_broadcast_oneshot(struct clock_event_device *bc);
2829 # else /* BROADCAST */
2930 static inline void tick_broadcast_setup_oneshot(struct clock_event_device *bc)
3031 {
@@ -43,6 +44,10 @@ void tick_setup_oneshot(struct clock_event_device *newdev, · 12 unmodified lines
4344 {
4445 BUG();
4546 }
47+static inline void tick_resume_oneshot(void)
48+{
49+ BUG();
50+}
4651 static inline int tick_program_event(ktime_t expires, int force)
4752 {
4853 return 0;
@@ -54,6 +59,10 @@ static inline void tick_broadcast_setup_oneshot(struct clock_event_device *bc) · 5 unmodified lines
5459 }
5560 static inline void tick_broadcast_oneshot_control(unsigned long reason) { }
5661 static inline void tick_shutdown_broadcast_oneshot(unsigned int *cpup) { }
62+static inline int tick_resume_broadcast_oneshot(struct clock_event_device *bc)
63+{
64+ return 0;
65+}
5766 #endif /* !TICK_ONESHOT */
5867
5968 /*
kernel/time/tick-oneshot.c+12−0
@@ -40,6 +40,18 @@ int tick_program_event(ktime_t expires, int force) · 39 unmodified lines
4040 }
4141 }
4242
43+/**
44+ * tick_resume_onshot - resume oneshot mode
45+ */
46+void tick_resume_oneshot(void)
47+{
48+ struct tick_device *td = &__get_cpu_var(tick_cpu_device);
49+ struct clock_event_device *dev = td->evtdev;
50+
51+ clockevents_set_mode(dev, CLOCK_EVT_MODE_ONESHOT);
52+ tick_program_event(ktime_get(), 1);
53+}
54+
4355 /**
4456 * tick_setup_oneshot - setup the event device for oneshot mode (hres or nohz)
4557 */